Nucl - Phys.B v.695

Nuclear Physics B 695 (2004) 337
www.elsevier.com/locate/npe
Exclusive electroproduction of J / mesons

at HERA
ZEUS Collaboration
S. Chekanov, M. Derrick, J.H. Loizides 1 , S. Magill, S. Miglioranzi 1 ,
B. Musgrave, J. Repond, R. Yoshida
Argonne National Laboratory, Argonne, IL 60439-4815, USA 40
M.C.K. Mattingly
Andrews University, Berrien Springs, MI 49104-0380, USA
N. Pavel
Institut fr Physik der Humboldt-Universitt zu Berlin, Berlin, Germany
P. Antonioli, G. Bari, M. Basile, L. Bellagamba, D. Boscherini,

A. Bruni, G. Bruni, G. Cara Romeo, L. Cifarelli, F. Cindolo,
A. Contin, M. Corradi, S. De Pasquale, P. Giusti, G. Iacobucci,
A. Margotti, A. Montanari, R. Nania, F. Palmonari, A. Pesci,
L. Rinaldi, G. Sartorelli, A. Zichichi
University and INFN Bologna, Bologna, Italy 31
G. Aghuzumtsyan, D. Bartsch, I. Brock, S. Goers, H. Hartmann,

E. Hilger, P. Irrgang, H.-P. Jakob, O. Kind, U. Meyer, E. Paul 2 ,
J. Rautenberg, R. Renner, A. Stifutkin, J. Tandler 3 , K.C. Voss,
M. Wang
Physikalisches Institut der Universitt Bonn, Bonn, Germany 28
D.S. Bailey 4 , N.H. Brook, J.E. Cole, G.P. Heath, T. Namsoo,

S. Robins, M. Wing
H.H. Wills Physics Laboratory, University of Bristol, Bristol, United Kingdom 39
0550-3213/$ see front matter 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2004.06.034
RAPID COMMUNICATION
ZEUS Collaboration / Nuclear Physics B 695 (2004) 337
M. Capua, A. Mastroberardino, M. Schioppa, G. Susinno

Calabria University, Physics Department and INFN, Cosenza, Italy 31
J.Y. Kim, I.T. Lim, K.J. Ma, M.Y. Pac 5

Chonnam National University, Kwangju, South Korea 33
M. Helbich, Y. Ning, Z. Ren, W.B. Schmidke, F. Sciulli

Nevis Laboratories, Columbia University, Irvington on Hudson, NY 10027, USA 41
J. Chwastowski, A. Eskreys, J. Figiel, A. Galas, K. Olkiewicz,

P. Stopa, L. Zawiejski
Institute of Nuclear Physics, Cracow, Poland 35
L. Adamczyk, T. Bod, I. Grabowska-Bod 6 , D. Kisielewska,

A.M. Kowal, M. Kowal, J. ukasik, M. Przybycien, L. Suszycki,
D. Szuba, J. Szuba 7
Faculty of Physics and Nuclear Techniques, AGH-University of Science and Technology, Cracow, Poland 42
A. Kotanski 8 , W. Sominski
Department of Physics, Jagellonian University, Cracow, Poland
V. Adler, U. Behrens, I. Bloch, K. Borras, V. Chiochia,

D. Dannheim 9 , G. Drews, J. Fourletova, U. Fricke, A. Geiser,
P. Gttlicher 10 , O. Gutsche, T. Haas, W. Hain, S. Hillert 11 , C. Horn,
B. Kahle, U. Ktz, H. Kowalski, G. Kramberger, H. Labes, D. Lelas,
H. Lim, B. Lhr, R. Mankel, I.-A. Melzer-Pellmann, C.N. Nguyen,
D. Notz, A.E. Nuncio-Quiroz, A. Polini, A. Raval, L. Rurua,
U. Schneekloth, U. Stsslein, G. Wolf, C. Youngman, W. Zeuner
Deutsches Elektronen-Synchrotron DESY, Hamburg, Germany
S. Schlenstedt
DESY Zeuthen, Zeuthen, Germany
G. Barbagli, E. Gallo, C. Genta, P.G. Pelfer

University and INFN, Florence, Italy 31
RAPID COMMUNICATION
A. Bamberger, A. Benen, F. Karstens, D. Dobur, N.N. Vlasov

Fakultt fr Physik der Universitt Freiburg i.Br., Freiburg i.Br., Germany 28
M. Bell, P.J. Bussey, A.T. Doyle, J. Ferrando, J. Hamilton,

S. Hanlon, D.H. Saxon, I.O. Skillicorn
Department of Physics and Astronomy, University of Glasgow, Glasgow, United Kingdom 39
I. Gialas
Department of Engineering in Management and Finance, University of Aegean, Greece
T. Carli, T. Gosau, U. Holm, N. Krumnack, E. Lohrmann, M. Milite,

H. Salehi, P. Schleper, T. Schrner-Sadenius, S. Stonjek 11 ,
K. Wichmann, K. Wick, A. Ziegler, Ar. Ziegler
Hamburg University, Institute of Experimental Physics, Hamburg, Germany 28
C. Collins-Tooth, C. Foudas, R. Gonalo 12 , K.R. Long,

A.D. Tapper
Imperial College London, High Energy Nuclear Physics Group, London, United Kingdom 39
P. Cloth, D. Filges
Forschungszentrum Jlich, Institut fr Kernphysik, Jlich, Germany
M. Kataoka 13 , K. Nagano, K. Tokushuku 14 , S. Yamada,

Y. Yamazaki
Institute of Particle and Nuclear Studies, KEK, Tsukuba, Japan 32
A.N. Barakbaev, E.G. Boos, N.S. Pokrovskiy, B.O. Zhautykov

Institute of Physics and Technology of Ministry of Education and Science of Kazakhstan, Almaty, Kazakhstan
D. Son
Kyungpook National University, Center for High Energy Physics, Daegu, South Korea 33
K. Piotrzkowski
Institut de Physique Nuclaire, Universit Catholique de Louvain, Louvain-la-Neuve, Belgium
RAPID COMMUNICATION
F. Barreiro, C. Glasman 15 , O. Gonzlez, L. Labarga, J. del Peso,

E. Tassi, J. Terrn, M. Zambrana
Departamento de Fsica Terica, Universidad Autnoma de Madrid, Madrid, Spain 38
M. Barbi, F. Corriveau, S. Gliga, J. Lainesse, S. Padhi, D.G. Stairs,

R. Walsh
Department of Physics, McGill University, Montral, PQ, H3A 2T8 Canada 27
T. Tsurugai
Meiji Gakuin University, Faculty of General Education, Yokohama, Japan 32
A. Antonov, P. Danilov, B.A. Dolgoshein, D. Gladkov,

V. Sosnovtsev, S. Suchkov
Moscow Engineering Physics Institute, Moscow, Russia 36
R.K. Dementiev, P.F. Ermolov, I.I. Katkov, L.A. Khein,

I.A. Korzhavina, V.A. Kuzmin, B.B. Levchenko, O.Yu. Lukina,
A.S. Proskuryakov, L.M. Shcheglova, S.A. Zotkin
Moscow State University, Institute of Nuclear Physics, Moscow, Russia 37
I. Abt, C. Bttner, A. Caldwell, X. Liu, J. Sutiak

Max-Planck-Institut fr Physik, Mnchen, Germany
N. Coppola, S. Grijpink, E. Koffeman, P. Kooijman, E. Maddox,

A. Pellegrino, S. Schagen, H. Tiecke, M. Vzquez, L. Wiggers,
E. de Wolf
NIKHEF and University of Amsterdam, Amsterdam, Netherlands 34
N. Brmmer, B. Bylsma, L.S. Durkin, T.Y. Ling

Physics Department, Ohio State University, Columbus, OH 43210, USA 40
A.M. Cooper-Sarkar, A. Cottrell, R.C.E. Devenish, B. Foster,

G. Grzelak, C. Gwenlan 16 , T. Kohno, S. Patel, P.B. Straub,
R. Walczak
Department of Physics, University of Oxford, Oxford, United Kingdom 39
RAPID COMMUNICATION
A. Bertolin, R. Brugnera, R. Carlin, F. Dal Corso, S. Dusini,

A. Garfagnini, S. Limentani, A. Longhin, A. Parenti, M. Posocco,
L. Stanco, M. Turcato
Dipartimento di Fisica dell Universit and INFN, Padova, Italy 31
E.A. Heaphy, F. Metlica, B.Y. Oh, J.J. Whitmore 17

Department of Physics, Pennsylvania State University, University Park, PA 16802, USA 41
Y. Iga
Polytechnic University, Sagamihara, Japan 32
G. DAgostini, G. Marini, A. Nigro

Dipartimento di Fisica, Universit La Sapienza and INFN, Rome, Italy 31
C. Cormack 18 , J.C. Hart, N.A. McCubbin

Rutherford Appleton Laboratory, Chilton, Didcot, Oxon, United Kingdom 39
C. Heusch
University of California, Santa Cruz, CA 95064, USA 40
I.H. Park
Department of Physics, Ewha Womans University, Seoul, South Korea
H. Abramowicz, A. Gabareen, S. Kananov, A. Kreisel, A. Levy

Raymond and Beverly Sackler Faculty of Exact Sciences, School of Physics, Tel-Aviv University,
Tel-Aviv, Israel 30
M. Kuze
Department of Physics, Tokyo Institute of Technology, Tokyo, Japan 32
T. Fusayasu, S. Kagawa, T. Tawara, T. Yamashita

Department of Physics, University of Tokyo, Tokyo, Japan 32
R. Hamatsu, T. Hirose 2 , M. Inuzuka, H. Kaji, S. Kitamura 19 ,

K. Matsuzawa
Tokyo Metropolitan University, Department of Physics, Tokyo, Japan 32
RAPID COMMUNICATION
M. Costa, M.I. Ferrero, V. Monaco, R. Sacchi, A. Solano

Universit di Torino and INFN, Torino, Italy 31
M. Arneodo, M. Ruspa
Universit del Piemonte Orientale, Novara, and INFN, Torino, Italy 31
T. Koop, J.F. Martin, A. Mirea

Department of Physics, University of Toronto, Toronto, ON, M5S 1A7 Canada 27
J.M. Butterworth 20 , R. Hall-Wilton, T.W. Jones, M.S. Lightwood,

M.R. Sutton 4 , C. Targett-Adams
Physics and Astronomy Department, University College London, London, United Kingdom 39
J. Ciborowski 21 , R. Ciesielski 22 , P. uzniak 23 , R.J. Nowak,

J.M. Pawlak, J. Sztuk 24 , T. Tymieniecka, A. Ukleja, J. Ukleja 25 ,
A.F. Zarnecki
Warsaw University, Institute of Experimental Physics, Warsaw, Poland 43
M. Adamus, P. Plucinski
Institute for Nuclear Studies, Warsaw, Poland 43
Y. Eisenberg, D. Hochman, U. Karshon, M. Riveline

Department of Particle Physics, Weizmann Institute, Rehovot, Israel 29
A. Everett, L.K. Gladilin 26 , D. Kira, S. Lammers, L. Li,

D.D. Reeder, M. Rosin, P. Ryan, A.A. Savin, W.H. Smith
Department of Physics, University of Wisconsin, Madison, WI 53706, USA 40
S. Dhawan
Department of Physics, Yale University, New Haven, CT 06520-8121, USA 40
S. Bhadra, C.D. Catterall, S. Fourletov, G. Hartner, S. Menary,

M. Soares, J. Standage
Department of Physics, York University, Ontario, M3J 1P3 Canada 27
RAPID COMMUNICATION
Received 2 June 2004; accepted 11 June 2004

E-mail address: yoshida@mail.desy.de (R. Yoshida).
1 Also affiliated with University College London, London, UK.
2 Retired.
3 Self-employed.
4 PPARC Advanced Fellow.
5 Now at Dongshin University, Naju, South Korea.
6 Partly supported by Polish Ministry of Scientific Research and Information Technology, grant No. 2P03B
12225.
7 Partly supported by Polish Ministry of Scientific Research and Information Technology, grant No. 2P03B
12625.
8 Supported by the Polish State Committee for Scientific Research, grant No. 2P03B 09322.
9 Now at Columbia University, NY, USA.
10 Now at DESY group FEB.
11 Now at University of Oxford, Oxford, UK.
12 Now at Royal Holoway University of London, London, UK.
13 Also at Nara Womens University, Nara, Japan.
14 Also at University of Tokyo, Tokyo, Japan.
15 Ramn y Cajal Fellow.
16 PPARC Postdoctoral Research Fellow.
17 On leave of absence at The National Science Foundation, Arlington, VA, USA.
18 Now at University of London, Queen Mary College, London, UK.
19 Present address: Tokyo Metropolitan University of Health Sciences, Tokyo 116-8551, Japan.
20 Also at University of Hamburg, Alexander von Humboldt Fellow.
21 Also at dz University, Poland.
22 Supported by the Polish State Committee for Scientific Research, grant No. 2P03B 07222.
23 dz University, Poland.
24 dz University, Poland, supported by the KBN grant No. 2P03B 12925.
25 Supported by the KBN grant No. 2P03B 12725.
26 On leave from MSU, partly supported by the Weizmann Institute via the USIsrael BSF.
27 Supported by the Natural Sciences and Engineering Research Council of Canada (NSERC).
28 Supported by the German Federal Ministry for Education and Research (BMBF), under contract Nos.
HZ1GUA 2, HZ1GUB 0, HZ1PDA 5, HZ1VFA 5.
29 Supported by the MINERVA Gesellschaft fr Forschung GmbH, the Israel Science Foundation, the US
Israel Binational Science Foundation and the Benozyio Center for High Energy Physics.
30 Supported by the GermanIsraeli Foundation and the Israel Science Foundation.
31 Supported by the Italian National Institute for Nuclear Physics (INFN).
32 Supported by the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT) and its
grants for Scientific Research.
33 Supported by the Korean Ministry of Education and Korea Science and Engineering Foundation.
34 Supported by the Netherlands Foundation for Research on Matter (FOM).
35 Supported by the Polish State Committee for Scientific Research, grant No. 620/E-77/SPB/DESY/P03/DZ117/2003-2005.
36 Partially supported by the German Federal Ministry for Education and Research (BMBF).
37 Supported by RF President grant No. 1685.2003.2 for the leading scientific schools and by the Russian
Ministry of Industry, Science and Technology through its grant for Scientific Research on High Energy Physics.
38 Supported by the Spanish Ministry of Education and Science through funds provided by CICYT.
39 Supported by the Particle Physics and Astronomy Research Council, UK.
40 Supported by the US Department of Energy.
41 Supported by the US National Science Foundation.
RAPID COMMUNICATION
10
Abstract
The exclusive electroproduction of J / mesons, ep epJ /, has been studied with the ZEUS
detector at HERA for virtualities of the exchanged photon in the ranges 0.15 < Q2 < 0.8 GeV2 and
2 < Q2 < 100 GeV2 using integrated luminosities of 69 and 83 pb1 , respectively. The photon
proton centre-of-mass energy was in the range 30 < W < 220 GeV and the squared four-momentum
transfer at the proton vertex |t| < 1 GeV2 . The cross sections and decay angular distributions are
presented as functions of Q2 , W and t. The effective parameters of the pomeron trajectory are in
agreement with those found in J / photoproduction. The spin-density matrix elements, calculated
from the decay angular distributions, are consistent with the hypothesis of s-channel helicity
conservation. The ratio of the longitudinal to transverse cross sections, L /T , grows with Q2 ,
whilst no dependence on W or t is observed. The results are in agreement with perturbative QCD
calculations and exhibit a strong sensitivity to the gluon distribution in the proton.
2004 Elsevier B.V. All rights reserved.
1. Introduction
The exclusive electroproduction of light (, , ) and heavy (J /, , ) vector
mesons, ep eVp, has been investigated at HERA [19]. The increased precision of
the recent data allows the study of the dependence of this process on the different scales
involved: the mass squared of the vector meson, MV2 , the square of the centre-of-mass
energy of the photonproton system, W 2 , the exchanged-photon virtuality, Q2 , and the
four-momentum transfer squared at the proton vertex, t.
Exclusive electroproduction of vector mesons involving a sufficiently large scale is
calculable perturbatively because of the QCD factorisation theorem [10]. QCD-based
models of this process assume that the exchanged virtual photon, seen from the proton
rest-frame, fluctuates into a quarkantiquark pair which interacts with the proton via
the exchange of two gluons in a colour-singlet configuration. After the interaction, the
q q pair becomes a bound state. The cross section is proportional to the generalised
parton distribution functions (GPDs) [1113] of the proton, which contain information
on the momentum distributions of the partons in the proton and their correlations. At the
leading-order approximation in ln(1/x) and vanishing t, the generalised gluon distribution
can be approximated by the usual gluon distribution. The gluon density is probed at
x (Q2 + MV2 )/W 2 and at a scale 2 Q2 + MV2 [14]. The cross section is thus expected
to rise steeply with W , a reflection of the steep rise of the gluon density as x decreases.
Data from exclusive production [13] show that the cross section ( p p) rises
with W as W , with increasing with Q2 from about 0.2 at Q2 = 0 (photoproduction) to
about 0.8 at Q2 30 GeV2 . However, in the case of exclusive J / production the cross
42 Supported by the Polish Ministry of Scientific Research and Information Technology, grant No. 112/E356/SPUB/DESY/P-03/DZ116/2003-2005.
43 Supported by the Polish State Committee for Scientific Research, grant No. 115/E-343/SPUB-M/DESY/P03/DZ121/2001-2002, 2P03B 07022.
RAPID COMMUNICATION
11
section rises steeply with W even for photoproduction [9]. It is therefore interesting to
investigate J / production at larger values of Q2 .
This paper presents measurements of the exclusive electroproduction of J / mesons.
Cross sections are given as functions of W , Q2 and t. The W dependence is also studied
as a function of t. The helicity structure of the J / has been investigated to test s-channel
helicity conservation (SCHC) and to extract the ratio of the cross sections for longitudinally
(L ) and transversely (T ) polarised virtual photons, R = L /T , as a function of W , Q2
and t. The results are compared to perturbative QCD (pQCD) model calculations.
The data cover the kinematic range 30 < W < 220 GeV and |t| < 1 GeV2 for two
ranges of photon virtuality: 0.15 < Q2 < 0.8 GeV2 (low-Q2 sample) and 2 < Q2 <
100 GeV2 (high-Q2 sample). The low-Q2 sample was measured in the e+ e decay
channel and the high-Q2 sample in both e+ e and + channels. The low-Q2 range
has been measured for the first time. The high-Q2 sample represents more than an order
of magnitude increase in statistics compared to the previous ZEUS results [1], and extends
both the W and Q2 ranges of the measurement.
2. Experimental set-up
The data used for this measurement were taken at the HERA ep collider using the
ZEUS detector in 19982000. During this period, HERA operated with a proton energy
of 920 GeV and an electron44 energy of 27.5 GeV. The data correspond to integrated
luminosities of 69 pb1 for the low-Q2 sample and 83 pb1 for the high-Q2 sample.
A detailed description of the ZEUS detector can be found elsewhere [15]. A brief outline
of the components that are most relevant for this analysis is given below.
Charged particles were reconstructed in the central tracking detector (CTD) [1618]
covering the polar-angle45 region 15 < < 164 . The transverse-momentum resolution
for full-length tracks is (pT )/pT = 0.0058pT 0.0065 0.0014/pT , with pT in GeV.
The high-resolution uranium-calorimeter (CAL) [1922] consists of three parts: the
forward (FCAL), the barrel (BCAL) and the rear (RCAL) calorimeters. Each part is
subdivided transversely into towers and longitudinally into an electromagnetic section
(EMC) and either one (RCAL) or two (FCAL and BCAL) hadronic sections. The CAL
covers 99.7% of the total solid angle.
The energy resolution obtained from test-beam
E in the electromagnetic sections and (E)/E =
measurements
was
(E)/E
=
0.18/
0.35/ E in the hadronic sections, with E in GeV.

The forward plug calorimeter (FPC) [23] was a lead-scintillator sandwich calorimeter
with readout via wavelength shifter fibres. It was installed in the beamhole of the FCAL
and extended the pseudorapidity coverage of the forward calorimeter from 4 to 5.
The beampipe calorimeter (BPC) [24] was a tungsten-scintillator sampling calorimeter
installed to measure scattered electrons at small angles, in the range 1.15 < 180 <
44 Hereafter, both e+ and e are referred to as electrons, unless explicitly stated otherwise.
45 The ZEUS coordinate system is a right-handed Cartesian system, with the Z axis pointing in the proton
beam direction, referred to as the forward direction, and the X axis pointing left towards the centre of HERA.
The coordinate origin is at the nominal interaction point.
RAPID COMMUNICATION
12
2.30 .The energy resolution, as measured under test-beam conditions, was (E)/E =
0.17/ E, with E in GeV. The impact position of the scattered electron was measured
with an accuracy of about 0.5 mm corresponding to an angular resolution of 0.17 mrad.
The small-angle rear tracking detector (SRTD) [25] consists of two planes of scintillator
strips read out via optical fibres and photomultiplier tubes. It is attached to the front face
of the RCAL and covers an angular range between 4 and 18 around the beampipe. The
SRTD provides a transverse position resolution of 0.3 cm [9] corresponding to an angular
resolution of 2 mrad.
The hadronelectron separator installed in the RCAL (RHES) [15] consists of silicon
diodes placed at a longitudinal depth of three radiation lengths. The RHES provides an
electron position resolution of 0.9 cm for a single hit and 0.5 cm if the shower spans at
least two adjacent pads [26].
The luminosity was determined from the rate of the bremsstrahlung process ep ep,
where the photon was measured with a lead-scintillator calorimeter [27] located at Z =
107 m in the HERA tunnel.
3. Kinematics and cross sections

The following kinematic variables are used to describe exclusive J / production,
e(k)p(P ) e(k )J /(v)p(P ), where k, k , P , P and v are, respectively, the fourmomenta of the incident electron, scattered electron, incident proton, scattered proton and
J /:
Q2 = q 2 = (k k )2 , the negative four-momentum squared of the virtual photon;
W 2 = (q + P )2 , the squared invariant mass of the photonproton system;
y = (P q)/(P k), the fraction of the electron energy transferred to the proton in the
proton rest frame;
x = Q2 /(2P q), the Bjorken variable;
t = (P P )2 , the squared four-momentum transfer at the proton vertex.
The kinematic variables were reconstructed with the constrained method [1] which
uses the momentum of the J / and the polar and azimuthal angles of the scattered
electron.
The ep cross section can be expressed in terms of the transverse, T , and longitudinal,
L , virtual photoproduction cross sections as

d 2 epeJ /p
= T y, Q2 (T + L ),
2
dy dQ
where T is the flux of transverse virtual photons [28] and is the ratio of longitudinal and
transverse virtual photon fluxes, given by = 2(1 y)/(1 + (1 y)2 ). In the kinematic
range studied here, is in the range 0.8 < < 1, with an average value of 0.99.
RAPID COMMUNICATION
13
The virtual photonproton cross section, pJ /p T + L , can be used to

pJ /p
evaluate the total exclusive cross section, tot
T + L , through the relation
pJ /p
tot
1 + R pJ /p
,
1 + R
where R = L /T is the ratio of the cross sections for longitudinal and transverse photons.
The helicity structure of the J / production is used to determine R as described in
Section 7.6.1.
4. Reconstruction and selection of the events

The signature of exclusive J / electroproduction, ep eJ /p, consists of the
scattered electron and two charged leptons from the J / decay, e+ e or + . The
scattered proton is deflected through a small angle and escapes undetected down the
beampipe.
The events were selected online by a three-level trigger [29,30]. For the low-Q2 sample,
the trigger [9] for J / photoproduction events with decay to the e+ e final state was used,
while for the high-Q2 sample, the trigger required a scattered electron in the CAL with
energy greater than 4 GeV.
The following criteria were applied offline to reconstruct and select the events [3133]:
the energy and position of the scattered electron were measured in the BPC for the lowQ2 sample and in the CAL for the high-Q2 sample. The energy was required to satisfy
Ee > 10 GeV. The position measurement of the CAL was improved using the SRTD
(88% of the events) and the RHES (10% of the events). To ensure full containment
of the electromagnetic shower, fiducial cuts were applied to the impact position of the
electron on face of the RCAL;
the J / mesons were reconstructed from the decay leptons. Two tracks of opposite
charge, well-reconstructed in the CTD with pT > 0.2 GeV, were selected (two-track
events). In the case of the electron decay channel, events were also selected by
requiring one well-reconstructed CTD track and one CAL energy cluster [34] not
related to the track (one-track events). In addition:
the two-track events were required to have the higher-momentum track matched to
a calorimeter energy cluster for which the fraction of the energy deposited in the
EMC was consistent with that of an electron or a muon;
the one-track events were accepted if, in addition to the measured CTD track
associated with a CAL cluster, the second cluster lay in the angular range outside the
CTD acceptance with energy between 2 and 10 GeV. Both clusters were required to
have a fraction of energy deposited in the EMC consistent with that of an electron.
For both types of events, one additional CTD track was allowed. If present, this track was
required to match the scattered electron. Events with further tracks were rejected:
RAPID COMMUNICATION
14
the position of the reconstructed vertex was required to be compatible with that of an
ep collision;
to remove events with large initial-state radiation and to reduce the background
from
photoproduction, the requirement 45 < < 65 GeV was imposed, where =
Ei (1 cos i ), Ei is the energy of the ith calorimeter cell, i its polar angle and the
sum runs over the cells associated to the final-state leptons;
to suppress non-exclusive events, the energy of each CAL cluster not associated to any
of the final-state leptons was required to be less than 0.3 or 0.4 GeV, depending on
the CAL section; these thresholds were set to be above the noise level of the CAL. To
suppress further the contamination from proton-dissociative events, ep eJ /Y , the
energy in the FPC was required to be less than 1 GeV and the sum of the energy in the
FCAL cells surrounding the beamhole to be less than 0.5 GeV. These cuts restrict the
mass of the proton-dissociated system, Y , to MY 3.0 GeV.
Unless otherwise stated, the results are quoted in the following kinematic range:
|t| < 1 GeV2 , 30 < W < 220 GeV for the electron channel and 45 < W < 160 GeV for the
muon channel. The larger W range for the electron channel was achieved by the inclusion
of the one-track events. The Q2 range was 0.15 < Q2 < 0.8 GeV2 for the low-Q2 sample
and 2 < Q2 < 100 GeV2 for the high-Q2 sample.
The final high-Q2 sample contains 728 events in the muon channel and 955 events in the
electron channel, 275 of which are reconstructed using only one track. The final low-Q2
sample contains 137 events in the electron channel, 16 of which are reconstructed using
only one track. The distribution of the events in the xQ2 plane is shown in Fig. 1.
5. Monte Carlo simulation

The acceptance and the effects of the detector response were determined using samples
of Monte Carlo (MC) events. All generated events were passed through the standard ZEUS
detector simulation, based on the G EANT 3.13 programme [35], and the ZEUS trigger
simulation package.
The exclusive process ep eJ /p was modelled using the Z EUSVM [36] MC
generator interfaced to H ERACLES 4.6.1 [37,38] to account for first-order QED radiative
effects. The p J /p cross section was parametrised as W eb|t | (MJ2/ + Q2 )n .
The parameter values n = 2.5(2.3), = 0.75(0.7) and b = 4.5(4.5) GeV2 were used to
describe the high-Q2 (low-Q2 ) data. The leptonic decay of the J / was simulated by
the P HOTOS programme [39] which includes final-state radiation from the decay leptons.
This generator assumes SCHC and that the ratio of the cross sections for longitudinal and
transverse photons is R = 0.5(Q2/MJ2/ ).
Proton-dissociative events, ep eJ /Y , were modelled using the generator E PSOFT
[40,41]. The p J /Y cross section was parametrised as

n
d 2 pJ /Y
W eb|t | MJ2/ + Q2
MY
dt dM 2
(1)
RAPID COMMUNICATION
15
Fig. 1. The distribution of the events in the muon and electron channels in the kinematic plane of Bjorken-x and
Q2 . The events reconstructed using one and two measured tracks are shown separately.
with the parameters n = 2.5, = 0.75, b = 0.81 GeV2 and = 2.57 chosen as described
in Section 6.2.
The QED background stemming from two-photon lepton-pair production l + l ,
where the virtual photon originates from the electron vertex and the second photon is
radiated off the proton, was simulated using the L PAIR [42] generator at low Q2 and the
G RAPE -D ILEPTON 1.1 [43] generator at high Q2 . The QED-Compton-like processes with
internal photon conversion were also generated with G RAPE.
6. Extraction of the J / signal

Fig. 2 shows the invariant-mass distributions of the muon and electron pairs, obtained
after the selection described in Section 4. The MC distributions for exclusive J /
production and the QED background are also shown. The width of the resonance is
dominated by the detector resolution, which deteriorates at low and high values of W .
RAPID COMMUNICATION
16
Fig. 2. Invariant mass distributions of the lepton pairs for (a) the low-Q2 sample and (b)(e) the high-Q2 sample.
The shaded histograms are the QED MC distributions and the open histograms the sum of the J / and QED
MC events. The small excess of data at low mass is due to background from pions. The error bars indicate the
statistical uncertainties.
RAPID COMMUNICATION
17
6.1. Non-resonant background

The non-resonant background was estimated from the MC distributions of the QEDbackground processes: two-photon lepton-pair production and Compton scattering. For
the low-Q2 sample, the normalisation of the QED-background was estimated from a twoparameter fit of the signal and the background MC distributions to the invariant mass
spectra of the data. For the high-Q2 sample, the normalisation was based on the known
cross sections and the integrated luminosity of the data. After subtraction of the nonresonant distributions, the J / signal was determined by counting the events in the mass
windows 2.8 < M+ < 3.4 GeV for the muon channel and 2.6 < Me+ e < 3.4 GeV for
the electron channel. The lower limit on Me+ e was chosen to include events with reduced
invariant mass due to bremsstrahlung. The contribution of the non-resonant background in
the signal range is typically 22% for the electron channel and 14% for the muon channel.
For the high-Q2 sample, additional background from pions misidentified as electrons
or muons was studied using a sample of events with two tracks, neither of which were
identified as a muon or an electron. The contribution was (2.7 0.6)% for the electron
channel and (0.8 0.3)% for the muon channel and was subtracted bin-by-bin for the t
and decay-angle distributions and globally for the W and Q2 distributions.
Events from exclusive (2S) production contribute to the J / sample through two
different decay channels: (i) (2S) J / + neutrals (branching ratio (23.9 1.2)%
[44]), where the neutrals are not detected in the CAL, and (ii) (2S) l + l (branching
ratios (7.30.4)103 for the electron channel and (7.00.9)103 for the muon decay
channel [44]), because of the limited resolution in the reconstruction of the invariant mass.
The contribution from both these processes to the J / sample was determined using MC
samples under the assumption that ((2S))/ (J /) = 0.166 0.013 [8]. A contribution
of (1.8 0.2)% was subtracted.
6.2. Proton-dissociative background
The remaining source of background consists of J / production accompanied by
proton dissociation, ep eJ /Y , where the particles from the breakup of the proton
are not detected.
Proton-dissociative events were studied using a sample of diffractive events selected as
described in Section 4, with the following exceptions:
the elasticity criterion (last criterion in Section 4) was not applied to the FPC and to a
region of FCAL of approximately 50 cm radius around the beampipe;
events with decay-lepton tracks at angles smaller than 30 with respect to the outgoing
proton direction were removed to ensure a rapidity gap between the J / and the
system Y .
Proton-dissociative events were selected by requiring an energy larger than 1 GeV in the
FPC. The sample of data tagged by the FPC contained 100 events for |t| < 3 GeV2 in
the kinematic range 45 < W < 160 GeV and Q2 > 2 GeV2 . The parameters (see Eq. (1))
that best describe the Q2 , W and t dependences are n = 2.57 0.09, = 0.61 0.40
RAPID COMMUNICATION
18
and b = 0.81 0.25 GeV2 . The MC distribution of MY2 was tuned to describe the FPC
energy distribution, yielding = 2.57 0.67. The values for n and are in agreement
with those described in Section 7.3. The latter are more precise and were used in E PSOFT.
The values of b and are in agreement with those found for proton-dissociative J /
photoproduction [9].
The fraction of proton-dissociative events in the elastic sample, fp-diss , was determined
data (1/ 1), where f data denotes the fraction of the
from the relation fp-diss = fFPC
FPC
proton-dissociative sample tagged by the FPC and = 32% is the FPC tagging
efficiency, estimated using E PSOFT. The fraction of proton-dissociative events in the final
sample, averaged over t for |t| < 1 GeV2 , was fp-diss = (14.2 2.0(stat.)+6.8
3.6 (syst.))%,
independent of W and Q2 . The systematic uncertainty was dominated by the uncertainty
on . The fraction increases from 4% for 0 < |t| < 0.1 GeV2 to 20% for 0.2 < |t| <
1.0 GeV2 . The cross sections presented in the next sessions were corrected for this
background in bins of t, and globally in W and Q2 .
7. Results
7.1. Measurement of cross sections
In each bin of a kinematic variable, the ep cross section was extracted for each decay
channel using the formula
(Ndata Nbgd )(1 fp-diss)
,
ABL
where Ndata is the number of events in the data and Nbgd is the number of events from the
non-resonant background (QED processes and pionic background) and (2S) production.
The overall acceptance is denoted as A, B accounts for the J / decay branching ratios
[44], (5.93 0.10)% and (5.88 0.10)% for the electron and muon channels, respectively,
and L is the integrated luminosity.
The total exclusive photonproton cross section was calculated as
epeJ /p =

d 2 epeJ /p
,
Q2 , W = 1/ Q2 , W
dQ2 dW
where the effective photon flux [45] contains the corrections for bin-centring and R,
both estimated from the MC simulation. The final cross section was the error-weighted
average of the cross sections for each decay channel.
The cross sections are quoted at the QED-Born level. The radiative corrections range
from 1% to 10% (on average 5%), depending on the kinematic region.
The cross sections were measured for |t| < 1 GeV2 . Assuming d/dt eb|t | , with
b = 4.5 GeV2 , the correction factor needed to extrapolate to the cross section integrated
over the full t range is 1.012. In addition, for x > 0.01 both the acceptance and the
expected cross section are small, and the measurement in this region therefore involves
an extrapolation, made in order to quote the measurement in bins of W and Q2 . The
uncertainty introduced by this extrapolation, as evaluated from the MC simulation, is
negligible.
pJ /p
tot
RAPID COMMUNICATION
19
7.2. Systematic uncertainties

The systematic uncertainties on the measured cross sections were determined by varying
the selection cuts and by modifying the analysis procedure. The sources of systematic
uncertainties considered were similar to those of previously published analyses [9,46].
For the low-Q2 sample, the main contribution arises from the uncertainty of 1 mm in
the position of the BPC, leading to a 10% uncertainty in the cross section.
For the high-Q2 sample, the dominant sources of uncertainty are as follows. The
corresponding average uncertainties are given in parentheses:
the trigger efficiency (+2.8
1.2 %);
the fiducial volume cut on the electron position was changed by 1 cm; the SRTD
alignment was changed by 2 mm along the Y axis (+5.5
3.5 %). A maximum change of
11% was observed in the lowest Q2 bin;
the mass window used for signal extraction was extended by 0.1 GeV (1.7%);
the normalisation of the QED background was changed by 10% (2.4%); the
maximum effect of 5% was found for the lowest t bin.
The uncertainty due to the subtraction of proton-dissociative background has been
discussed in Section 6.2. Additional contributions come from the uncertainties on the
integrated luminosity, 2.25%, and on the branching ratios, 1.7%. Uncertainties from
the minimum energy requirement of the scattered electron (0.7%), the elasticity cut
(+0.2
1.9 %), the selection of the electron and muon samples (1.2%) and the dependence on
the MC parametrisations (0.7%) were also estimated. The total systematic uncertainty
was determined by adding the individual contributions in quadrature. The correlated and
+7.4
uncorrelated systematic uncertainties were evaluated separately and were +5
8 % and 6.4 %,
respectively.
7.3. Dependence on W and Q2
pJ /p
The cross section tot

, measured as a function of W and Q2 for |t| < 1 GeV2 ,
is given in Tables 1 and 2. The same cross section, extrapolated to the full t range, is shown
in Fig. 3 together with the H1 [5] measurements46 as well as the ZEUS measurement of
exclusive J / photoproduction [9]. The H1 measurements are systematically lower than
the ZEUS data; however they are compatible when the normalisation uncertainties of both
measurements are considered.
The functional form W was fitted to the ZEUS data; the results of the fit are
shown in Fig. 3(a) and in Table 4. No significant variation of with Q2 is seen. The mean
value of is 0.73 0.11(stat.)+0.04
0.08 (syst.). It is consistent with the values found for J /
photoproduction [1] and for electroproduction at high Q2 [2].
46 In Fig. 3(a), the H1 cross sections, measured at Q2 values of 3.5, 10.1 and 33.6 GeV2 , have been rescaled
to the Q2 values of 3.1, 6.8 and 16 GeV2 using the Q2 dependence of the data measured by H1. The systematic
uncertainties due to this extrapolation were negligible.
0.150.8
510
10100
0.4
3.1
6.8
16.0
W (GeV)
W (GeV)
Nee
Aee
3065
49
32
0.031
65105
86
55
0.044
105220
167
3045
217 53+12
19
257 46+18
17
50
0.021
37
29.2
0.111
4570
57
51.5
53.2
0.180
0.173
7090
80
36.7
60.0
0.204
0.224
90112
101
61.7
37.5
0.221
0.223
112145
128
51.2
46.4
0.197
0.167
145220
180
71.6
0.154
3050
40
27.8
0.215
5074
62
48.7
45.8
0.403
0.383
7496
85
39.6
52.4
0.439
0.486
96120
108
37.1
46.4
0.479
0.514
120150
135
33.9
49.0
0.475
0.395
150220
183
58.4
0.343
3055
42
16.1
0.235
5578
66
27.7
37.8
0.555
0.659
78100
89
31.0
43.6
0.673
0.728
100124
112
37.5
36.2
0.645
0.704
124160
141
39.6
41.3
0.563
0.591
160220
189
51.0
0.361
epJ /p (pb)
498 89+37
38
41.5 8.4+5.6
6.6
48.8 5.2+3.1
3.9
36.4 4.1+10.5
3.0
35.4 4.0+3.0
4.5
44.7 5.0+9.0
4.3
76.5 10.3+11.5
5.1
19.6 4.1+3.9
1.9
19.3 2.2+2.9
1.3
15.6 1.8+1.6
1.4
13.5 1.7+1.1
0.7
14.9 1.9+1.1
1.3
27.9 4.1+4.5
1.4
10.9 3.1+0.8
1.0
8.4 1.2+1.4
0.4
8.6 1.1+0.9
1.4
8.4 1.1+0.4
1.2
10.8 1.4+2.1
0.8
25.1 3.8+1.7
1.2
pJ /p
tot
(nb)
39.2 9.6+2.2
3.4
75.713.5+5.2
4.9
118.021.0+8.8
9.1
24.8 5.0+3.3
3.9
27.4 3.0+1.8
2.2
36.7 4.2+10.6
3.0
43.0 4.8+3.7
5.4
48.8 5.5+9.8
4.7
61.1 8.2+9.2
4.1
12.7 2.7+2.5
1.2
16.6 1.9+2.5
1.1
20.7 2.3+2.2
1.9
21.9 2.7+1.7
1.2
25.8 3.3+1.9
2.3
33.2 4.9+5.3
1.6
3.3 0.9+0.2
0.3
4.5 0.6+0.7
0.2
6.7 0.9+0.7
1.1
7.9 1.0+0.3
1.1
9.3 1.2+1.7
0.7
20.8 3.1+1.4
1.0
25
Q2 (GeV2 )
RAPID COMMUNICATION
Q2 (GeV2 )
20
Table 1
The cross sections for the reaction p J /p measured as a function of W in bins of Q2 and for |t| < 1 GeV2 : W and Q2 are the mean values in the indicated
ranges; Nee and N are the number of events in the signal region after non-resonant background subtraction of the electron and muon pairs, respectively; Aee and A
are the corresponding acceptances. The first uncertainty of the cross sections is statistical and the second systematic. An overall normalisation uncertainty of +5%
8% was
not included
Q2 (GeV2 )
Q2 (GeV2 )
Nee
epJ /p
Aee
ee
(pb)
ee: 30 < W < 220 GeV

0.4
137.0
0.029
954 108+63
74
23.2
2.5
141.3
0.156
150 14+53
8
3.25
3.9
160.6
0.202
57
5.9
123.1
0.327
8.4
epJ /p
(pb)
122.5
0.466
12
106.7
0.485
1540
22
84.3
0.473
40100
54
12.0
0.423
59.9 6.1+5.5
3.6
42.6 4.3+4.7
5.0
36.7 4.0+1.4
2.9
29.3 3.7+2.0
4.7
4.5 1.5+0.5
1.1
(nb)
72.6 8.2+4.8
5.6
90.5
0.159
118.6
0.217
100.4
0.336
104.1
0.529
87.2
0.607
71.6
0.651
7.4
0.554
96 11+5
14
39.7 2.9+5.9
2.9
48.7 5.2+1.2
2.5
24.3 1.8+1.1
1.0
91.6 8.9+12.2
6.6
32.5 3.4+4.4
2.4
24.1 2.8+1.2
1.6
18.4 2.4+0.9
1.4
2.2 0.9+0.4
0.6
38.7 2.5+3.3
3.6
15.8 1.2+1.5
0.7
11.6 1.0+0.4
0.6
4.0 0.4+0.2
0.3
0.65 0.17+0.08
0.16
21
RAPID COMMUNICATION
1015
132 12+8
17
pJ /p
tot
W = 90 GeV
: 45 < W < 160 GeV
0.150.8
710
Table 2
The cross sections for the reaction p J /p measured as a function of Q2 , for a mean value W = 90 GeV and for |t| < 1 GeV2 : Q2 indicates the mean value
in the Q2 range considered; Nee and N are the numbers of events in the signal region after non-resonant background subtraction of the electron and muon pairs,
respectively; Aee and A are the corresponding acceptances. The first uncertainty of the cross sections is statistical and the second systematic. An overall normalisation
uncertainty of +5%
8% was not included
RAPID COMMUNICATION
22
Fig. 3. Exclusive J / electroproduction cross section (a) as a function of W for four values of Q2 and (b) as
a function of Q2 at W = 90 GeV. ZEUS photoproduction and H1 electroproduction cross sections are also
2
shown. The full lines are fits to the ZEUS data of the form (a) W (Q ) and (b) (Q2 + MJ2/ )n . The
inner error bars represent the statistical uncertainties, the outer bars are the statistical and systematic uncertainties
added in quadrature. An overall normalisation uncertainty of +5%
8% was not included.
RAPID COMMUNICATION
23
The function = 0 (MJ2/ /(Q2 + MJ2/ ))n , fitted to the ZEUS data including the
photoproduction point, is shown in Fig. 3(b). The resulting parameters are 0 = 77 3 nb
and n = 2.44 0.08, with 2 /ndf = 4.1/7. The fit, which takes both the statistical and
uncorrelated systematic uncertainties into account, describes the data well over the full Q2
range.
7.3.1. Comparison to model predictions
Models based on QCD are able to describe exclusive vector meson production at HERA.
In such models, in the frame where the proton is at rest, the photon emitted from the
electron fluctuates into a q q state, this q q pair subsequently interacts with the proton
through the exchange of gluons in a colour-singlet configuration and eventually forms a
bound meson state. The transverse size of the q q pair depends on Q2 and on the quark
mass; for Q2 > O(10) GeV2 or for heavy quarks, it is assumed to be considerably smaller
than the size of the proton. At such distances, the QCD coupling is small and perturbation
theory can be applied. The QCD factorisation theorem for hard exclusive electroproduction
of mesons [10] predicts that, in the limit of large Q2 and fixed x, the cross section can be
estimated from a hard interaction part calculable in pQCD, the q q wave function of the
meson and the generalised parton distributions (GPDs) [1113] which contain information
about the correlations of the partons inside the proton and their momentum distribution.
A rapid rise in the cross section with W is predicted which is related to the fast increase of
the gluon density inside the proton at small values of x. Recently, NLO calculations have
been completed for the exclusive photoproduction of heavy vector mesons [47]. A selection
of the available models is compared to the data and discussed below. A more complete
discussion on the available models is given elsewhere [14].
Frankfurt, Koepf and Strikman (FKS) [48,49] have proposed a model based on the
leading-order approximation s ln(Q2 ). The usual parton distribution functions (PDFs) are
used. The J / wave function is estimated in the non-relativistic approximation.
In the model of Martin, Ryskin and Teubner (MRT) [50], the calculations are also
performed at the leading order, s ln(Q2 ). Assuming partonhadron duality, the component
of the cc pair which has the correct spin-parity for the J / is used instead of the J / wave
function. The cross section is integrated over the J / mass range. The GPDs are estimated
using the conventional next-to-leading (NLO) gluon distributions.
Gotsman et al. (GLLMN) [51] have presented a dipole model where the cross section
is expressed as the convolution of the wave function of the virtual photon, the dipole
scattering amplitude and the J / wave function. The dipole scattering amplitude is
estimated at leading order, s ln(1/x), as the solution of the BalitskyKovchegov [52,53]
evolution equation, including both the linear BFKL terms due to parton splitting and nonlinear terms due to recombination of partons in the high-density region at low x. The J /
wave function is estimated in the non-relativistic approximation.
The W and Q2 dependence of the cross sections measured by ZEUS are compared to
the QCD predictions in Fig. 4. As the full NLO corrections have not yet been estimated, all
the models have significant normalisation uncertainties. Therefore the normalisation was
fixed using the ZEUS photoproduction data at W = 90 GeV; the different normalisation
factors are indicated in Fig. 4. The gluon PDFs ZEUS-S [54] for MRT and CTEQ4L [55]
for FKS were used. The Q2 dependence of is compared in the insert in Fig. 4(a). All
RAPID COMMUNICATION
24
a function of Q2 at W = 90 GeV. ZEUS photoproduction results are also shown. The curves represent the
predictions of the QCD models MRT, FKS and GLLMN (see text) normalised to the ZEUS photoproduction
point at W = 90 GeV. The insert shows the parameter as a function of Q2 . The inner error bars represent
the statistical uncertainties, the outer bars are the statistical and systematic uncertainties added in quadrature. An
overall normalisation uncertainty of +5%
RAPID COMMUNICATION
25
models predict a rise of the cross section with increasing W and have a Q2 dependence
similar to that of the data.
7.3.2. Comparison to model predictions for different gluon parametrisations
The MRT model was used to test three different gluon distributions: MRST02 [56],
CTEQ6M [57] and ZEUS-S [54], obtained from NLO DGLAP analyses of structure
function data. In deriving the GPDs from the PDFs, sensitivity to the gluon distribution
at very low x is introduced. Again, the predictions were normalised to the ZEUS
photoproduction measurement at W = 90 GeV.
Fig. 5 compares the data with the predictions. While CTEQ6M describes the W and
Q2 dependence of the data, MRST02 has the wrong shape in W , particularly at low Q2 .
ZEUS-S describes the W dependence but falls too quickly with increasing Q2 .
The data exhibit a strong sensitivity to the gluon distribution in the proton. However,
full NLO calculations are needed in order to use these data in global fits to constrain the
gluon density.
7.4. Dependence on t
The differential cross section, d pJ /p /dt, measured as a function of t in the range

|t| < 1 GeV2 , is shown in Table 3 and Fig. 6(a)(d) for the high Q2 sample as well as
for three Q2 intervals. A function of the form d/dt = d/dt|t =0 eb|t | was fitted to
the data and the results of the fit are given in Table 4. The slope parameter b is shown
in Fig. 6(e) as a function of Q2 and is compared to the ZEUS photoproduction and H1
electroproduction values. No significant Q2 dependence in b is seen over the measured
range of Q2 . This behaviour is different from that of exclusive electroproduction, where
the b slope strongly decreases with increasing Q2 , reaching the value of that of the J / at
Q2 30 GeV2 [2].
In QCD-based models, at high Q2 , the size of the q q pair in the direction transverse
to the reaction axis decreases as 1/Q and the t dependence should reach a universal limit,
independent of the flavour of the quark constituents of the meson [58]. Hence, in this
limit, the t dependence is given solely by the GPDs of the nucleon. Following this idea,
the differential cross section was also fitted using an elastic form factor for two-gluon
exchange, d/dt (1 t/m22g )4 , where m22g is the square of the two-gluon invariant
mass, as suggested by Frankfurt and Strikman [59]. The fit, including both statistical and
systematic uncertainties, yields m22g = 0.55 0.02 GeV2 and is shown in Fig. 6(a).
7.5. Pomeron trajectory
Soft diffractive processes are described by Regge phenomenology [60] in terms of the
exchange of a pomeron trajectory. In hard interactions, where Regge phenomenology may
not be applicable, an effective pomeron trajectory may nevertheless be extracted. The
high-Q2 sample was analysed to determine the effective pomeron trajectory. In the Regge
formalism, the differential cross section can be expressed as
d
W 4(P (t )1),
dt
(2)
RAPID COMMUNICATION
26
a function of Q2 at W = 90 GeV. ZEUS photoproduction results are also shown. The data are compared to
the MRT predictions (see text) obtained with different parametrisations of the gluon density and normalised to
the ZEUS photoproduction point at W = 90 GeV. The insert shows the parameter as a function of Q2 . The
inner error bars represent the statistical uncertainties, the outer bars are the statistical and systematic uncertainties
added in quadrature. An overall normalisation uncertainty of +5%
RAPID COMMUNICATION
27
Table 3
The differential cross sections for the reaction p J /p measured as a function of t in bins of Q2 for a
mean value W = 90 GeV. The first uncertainty is statistical and the second systematic
Q2 (GeV2 )
2100
25
Q2 (GeV2 )
6.8
3.1
510
6.8
10100
16
d pJ /p /dt (nb/GeV2 )
|t| (GeV2 )
|t| (GeV2 )
0.00.1
0.05
0.10.2
0.15
0.20.4
0.29
0.41.0
0.58
0.00.1
0.05
0.10.2
0.15
0.20.4
0.29
0.41.0
0.58
10.7 1.1+0.9
0.7
0.00.1
0.05
75.6 8.3+5.0
9.7
0.10.2
0.15
0.20.4
0.29
0.41.0
0.58
6.5 0.6+0.6
0.3
0.00.1
0.05
28.0 3.4+2.6
2.8
0.10.2
0.15
0.20.4
0.29
0.41.0
0.58
79.2 5.0+6.1
6.5
43.9 3.0+2.8
1.9
25.8 1.4+1.9
1.5
6.0 0.4+0.4
0.4
148 15+22
14
86.9 9.6+12.5
8.1
49.2 4.3+6.6
2.8
39.6 4.9+2.0
2.4
23.9 2.4+2.1
1.1
15.6 2.0+0.9
2.0
9.4 1.0+0.6
1.4
1.8 0.2+0.1
0.2
where the trajectory P is usually parametrised as

P (t) = P (0) + P t.
The effective pomeron trajectory was determined by fitting Eq. (2) to the differential cross
sections at different t values. The fit was performed in four t bins at Q2 = 6.8 GeV2 . Since
the proton-dissociative process has the same W dependence as the exclusive process, the
extraction of P is not sensitive to this background contribution, which populates the high-t
region. Therefore the analysis was extended up to |t| = 2 GeV2 . The fit results are shown
in Fig. 7 and in Table 5. The parameters of the trajectory, determined from the linear fit
are:
P (0) = 1.20 0.03(stat.)+0.01
0.02 (syst.),
2
P = 0.07 0.05(stat.)+0.03
0.04 (syst.) GeV .
These values are in good agreement with the ZEUS results from J / photoproduction [9]
which are also shown in Fig. 7. They are also in agreement with expectations of pQCDbased models [61,62], but are not consistent with the trajectory measured in soft diffractive
processes, P = 1.08 + 0.25t [63,64].
RAPID COMMUNICATION
28
Fig. 6. Differential cross sections d/dt (a) over the entire Q2 range and (b)(d) for three bins of Q2 , for
30 < W < 220 GeV and |t| < 1 GeV2 . The full lines are the results of a fit to the form d/dt = d/dt|t=0 eb|t|
and the dashed line is the result of a fit using an elastic form factor assuming two-gluon exchange:
d/dt (1 t/m22g )4 . (e) The slope b, as a function of Q2 , compared to the ZEUS photoproduction and
H1 results. The mean value of b is indicated by the horizontal line. The inner error bars represent the statistical
uncertainty, the outer bars the statistical and systematic uncertainties added in quadrature.
RAPID COMMUNICATION
29
Table 4
2
The parameters , b and d
dt |t=0 measured as a function of Q in the range 30 < W < 220 GeV and 45 < W <
160 GeV for the electron and muon channels, respectively, and |t| < 1 GeV2 . The first uncertainty is statistical
and the second systematic

d
b|t| (GeV2 )
2
Q2 (GeV2 )
Q2 (GeV2 )
( W )
b d
dt e
dt t=0 (nb/GeV )
0.150.8
0.4
0.87 0.22+0.04
0.01
25
3.1
0.65 0.17+0.16
0.05
4.85 0.24+0.26
0.19
510
6.8
0.60 0.18+0.04
0.10
4.44 0.26+0.12
0.27
84.7 7.9+7.3
9.6
1.12 0.20+0.03
0.16
5.06 0.27+0.29
0.17
35.5 3.4+2.9
3.5
0.73 0.11+0.04
0.08
4.72 0.15+0.12
0.12
95.2 4.9+8.1
7.9
10100
16
2100
6.8
185 15+30
21
7.6. Decay angular distributions

The study of the angular distributions of the decay of the J / provides information
about the photon and J / polarisation states. In the helicity frame [65], the production
and decay of the J / can be described in terms of three angles (see Fig. 8): h , the
angle between the J / production plane and the lepton scattering plane; h , the polar
angle, and h , the azimuthal angle of the positively charged decay lepton. Under the
assumption of SCHC, the normalised angular distribution depends only on two angles,
h and h = h h , and can be expressed in the form
2

1 dN
3
04
04
cos h ,
= 1 + r00
+ 1 3r00
N d cos h 8

1 dN
1
1
=
cos 2h .
1 r11
N dh 2
(3)
(4)
04 represents the probability that the J / is produced in

The spin-density matrix element r00
the helicity-0 state from a virtual photon of helicity 0 or 1. The spin-density matrix element
1
r11
gives the probability for the J / to be produced in the helicity-1 state from a virtual
photon of helicity 1 or 1. Assuming SCHC and natural spin-parity exchange (NPE) [65],
04 and r 1
the matrix elements r00
11 are related by
1
=
r11

1
04
1 r00
.
2
(5)
The cross sections at W = 90 GeV are shown in Fig. 9(a)(f) for three intervals of Q2 .
04
Eqs. (3) and (4) were fitted to the data. The values of the spin-density matrix elements r00
1
1
and r11 , determined from the fits, are given in Table 6. The measured values of r11 are
consistent with those obtained from Eq. (5), also shown in Table 6, supporting the SCHC
and NPE hypotheses.
Figs. 10 and 11 show the cross sections in bins of W and t, respectively. They are
quoted at the reference value Q2 = 6.8 GeV2 . Eq. (3) was fitted to the data. The values of
04 , given in Tables 7 and 8, are consistent with no W or t dependence.
r00
RAPID COMMUNICATION
30
Fig. 7. (a)(d) Differential cross sections d/dt as a function of W for fixed ranges of t; the full lines are fits
to W 4(P (t)1) . (e) Pomeron trajectory: the lines are linear fits to the data at Q2 = 6.8 GeV2 (full) and to the
J / photoproduction results (dashed); the dotted line is the soft pomeron trajectory [66]. The inner error bars
represent the statistical uncertainty, the outer bars the statistical and systematic uncertainties added in quadrature.
RAPID COMMUNICATION
31
Table 5
The pomeron trajectory P (t) measured in four t bins, in the range 2 < Q2 < 100 GeV2 at a mean value
Q2 = 6.8 GeV2 . The first uncertainty is statistical and the second systematic
|t| (GeV2 )
|t| (GeV2 )
P (t)
0.00.1
0.046
1.22 0.04+0.03
0.04
0.10.3
0.186
0.30.9
0.483
0.92.0
1.123
1.17 0.04+0.02
0.02
1.17 0.03+0.02
0.04
1.13 0.04+0.03
0.04
Fig. 8. Schematic diagrams of (a) the process ep eJ /p in the p centre-of-mass system, and (b) the decay
of the J / in its rest frame. Three angles suffice to describe the reaction: the azimuthal angle between the
scattering plane and the production plane, h ; and the two J / decay angles, h , the azimuthal angle between
the production and decay planes, defined in either the p system or in the J / rest frame; and h , which is the
polar angle of the positively-charged decay product defined with respect to the direction of the J / momentum
vector in the p system, or, equivalently, the direction opposite to the momentum-vector of the final-state proton
in the rest frame of the J / meson. This choice of the spin-quantisation axis defines the helicity frame.
7.6.1. Longitudinal and transverse cross sections

The ratio of the longitudinal to transverse cross section, R = L /T , was calculated as
04 according to the relation
a function of Q2 , W and t from r00
RAPID COMMUNICATION
32
Fig. 9. (a)(f) Distributions of cos h and h in three Q2 bins; the curves are the fits to Eqs. (3) and (4). (g) Ratio
R = L /T as a function of Q2 ; the full curve is the result of the fit to the ZEUS data while the dashed and
dotted curves are the predictions of the MRT and GLLMN models, respectively. The inner error bars represent
the statistical uncertainty, the outer bars the statistical and systematic uncertainties added in quadrature.
R=
04
1 r00
,
04
1 r00
which is valid under the assumption of SCHC.

The values of R as a function of Q2 are given in Table 6 and compared with the
H1 results [5] in Fig. 9(g). The expression R = (Q2 /MJ2/ ) was fitted to the ZEUS
data yielding = 0.52 0.16(stat.). In QCD-based models, the scale that controls the
RAPID COMMUNICATION
33
Table 6
04 and r 1 , the ratio of cross sections of longitudinally and transversely
The spin-density matrix elements, r00
11
04 ) measured in bins of Q2 . The first uncertainty is
1
polarised photons, R, and the quantity r11
12 (1 r00
statistical and the second systematic
Q2 (GeV2 )
Q2 (GeV2 )
25
3.1
510
6.8
10100
16
04
r00
1
r11
R = L /T
0.12 0.08+0.13
0.15
0.25 0.09+0.10
0.06
0.54 0.10+0.06
0.03
0.34 0.09+0.03
0.06
0.44 0.09+0.06
0.07
0.26 0.09+0.09
0.04
0.13 0.11+0.09
0.16
0.33 0.16+0.19
0.11
1.19 0.51+0.28
0.14
1
04 )
r11
12 (1 r00
0.10 0.09+0.08
0.06
0.06 0.10+0.08
0.06
0.03 0.11+0.07
0.02
Fig. 10. (a)(e) Distributions of cos h for five bins of W ; the curves are the fits to Eq. (3). (f) Ratio R = L /T
as a function of W ; the dashed line is the MRT prediction and the full line is the result of a one-parameter fit. The
error bars are statistical (inner) and total (outer).
RAPID COMMUNICATION
34
Fig. 11. (a)(d) Distributions of cos h for four bins of t; the curves are the fits to Eq. (3). (e) Ratio R = L /T
as a function of t; the full line is the result of a one-parameter fit. The error bars are statistical (inner) and total
(outer).
transverse size of the q q fluctuation of the photon may behave differently for L and T .
However, in the MRT model, L and T have the same W dependence, dictated by the
gluon distribution. Therefore the ratio is constant. This model correctly describes the rising
behaviour of R with Q2 whereas the GLLMN prediction somewhat overestimates it.
The values of R as a function of W and t are given in Tables 7 and 8 and shown in
Figs. 10(f) and 11(g), respectively.
8. Summary
The exclusive electroproduction of J / mesons, ep eJ /p, has been measured with
the ZEUS detector at HERA for photon virtualities in the ranges 0.15 < Q2 < 0.8 GeV2
RAPID COMMUNICATION
35
Table 7
04 and the ratio of cross sections of longitudinally and transversely polarised
The spin density matrix element r00
photons, R, measured in bins of W , in the range 2 < Q2 < 100 GeV2 at a mean value Q2 = 6.8 GeV2 . The
first uncertainty is statistical and the second systematic
W (GeV)
W (GeV)
3055
43.5
5580
68.1
80120
95.6
120160
128.1
160220
184.4
04
r00
R = L /T
0.21 0.16+0.32
0.18
0.24 0.13+0.10
0.10
0.25 0.09+0.09
0.05
0.12 0.11+0.11
0.05
0.36 0.16+0.12
0.10
0.27 0.26+0.45
0.17
0.31 0.23+0.26
0.22
0.33 0.16+0.15
0.07
0.14 0.15+0.12
0.05
0.56 0.40+0.23
0.16
Table 8
04 and the ratio of cross sections of longitudinally and transversely polarised
The spin density matrix element r00
photons, R, measured in bins of |t|, in the range 2 < Q2 < 100 GeV2 at a mean value Q2 = 6.8 GeV2 . The
first uncertainty is statistical and the second systematic
|t| (GeV2 )
|t| (GeV2 )
0.00.1
0.046
0.10.2
0.146
0.20.4
0.285
0.41.0
0.579
04
r00
R = L /T
0.24 0.11+0.12
0.06
0.36 0.13+0.08
0.11
0.19 0.10+0.07
0.12
0.16 0.10+0.05
0.05
0.31 0.19+0.22
0.10
0.56 0.30+0.17
0.20
0.23 0.15+0.11
0.16
0.19 0.14+0.08
0.08
and 2 < Q2 < 100 GeV2 , for photonproton centre-of-mass energies in the range 30 <
W < 220 GeV and for four-momentum-transfer squared in the range |t| < 1 GeV2 .
The cross section of the process p J /p rises with W as W , with a slope
parameter of about 0.7. This parameter does not change significantly with Q2 and is
consistent with that observed in J / photoproduction.
The cross section at W = 90 GeV and over the whole Q2 range is described by the
function (Q2 + MJ2/ )n , with n = 2.44 0.08.
The t distribution, measured for |t| < 1 GeV2 , is well described by an exponential
dependence over the range 2 < Q2 < 100 GeV2 . The slope parameter, b, is consistent with
being constant in this range. The mean value is b = 4.72 0.15(stat.)0.12(syst.) GeV2 ,
consistent with that observed in J / photoproduction.
An analysis of the cross sections in the framework of Regge phenomenology yields an
effective pomeron trajectory consistent with that measured in J / photoproduction.
04
1
and r00
are consistent with s-channel-helicity
The spin-density matrix elements r11
conservation. The ratio of the cross sections for longitudinally and transversely polarised
photons, R, increases with Q2 , but is independent of W and t, within the measured range.
The J / electroproduction data can be qualitatively described within the framework
of pQCD that successfully describes J / photoproduction data. The data exhibit a
strong sensitivity to the gluon distribution in the proton. Full next-to-leading-order QCD
calculations would allow these data to be used in global QCD fits to constrain the gluon
density function in the proton.
RAPID COMMUNICATION
36
Acknowledgements
We thank the DESY directorate for their strong support and encouragement. The special
effort of the HERA group is gratefully acknowledged. We are grateful for the support of
the DESY computing and network services. The design, construction and installation of the
ZEUS detector has been made possible by the efforts and ingenuity of many people who
are not listed as authors. It is a pleasure to thank E. Naftali and T. Teubner for providing us
with model predictions.
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]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
ZEUS Collaboration, J. Breitweg, et al., Eur. Phys. J. C 6 (1999) 603.

H1 Collaboration, C. Adloff, et al., Eur. Phys. J. C 13 (2000) 371.
H1 Collaboration, C. Adloff, et al., Phys. Lett. B 539 (2002) 25.
ZEUS Collaboration, J. Breitweg, et al., Eur. Phys. J. C 12 (2000) 393.
H1 Collaboration, C. Adloff, et al., Eur. Phys. J. C 10 (1999) 373.
ZEUS Collaboration, J. Breitweg, et al., Phys. Lett. B 437 (1998) 432.
ZEUS Collaboration, S. Chekanov, et al., Eur. Phys. J. C 24 (2002) 345.
J.C. Collins, L. Frankfurt, M. Strikman, Phys. Rev. D 56 (1997) 2982.
D. Muller, et al., Fortschr. Phys. 42 (1994) 101.
X.D. Ji, Phys. Rev. Lett. 78 (1997) 610.
A.V. Radyushkin, Phys. Rev. D 56 (1997) 5524.
V. Barone, E. Predazzi, High-Energy Particle Diffraction, Springer-Verlag, Berlin, 2002, and references
therein.
ZEUS Collaboration, U. Holm (Ed.), The ZEUS Detector, Status Report (unpublished), DESY (1993),
available on http://www-zeus.desy.de/bluebook/bluebook.html.
N. Harnew, et al., Nucl. Instrum. Methods A 279 (1989) 290.
B. Foster, et al., Nucl. Phys. B (Proc. Suppl.) 32 (1993) 181.
B. Foster, et al., Nucl. Instrum. Methods A 338 (1994) 254.
M. Derrick, et al., Nucl. Instrum. Methods A 309 (1991) 77.
A. Andresen, et al., Nucl. Instrum. Methods A 309 (1991) 101.
A. Caldwell, et al., Nucl. Instrum. Methods A 321 (1992) 356.
A. Bernstein, et al., Nucl. Instrum. Methods A 336 (1993) 23.
A. Bamberger, et al., Nucl. Instrum. Methods A 450 (2000) 235.
ZEUS Collaboration, J. Breitweg, et al., Phys. Lett. B 407 (1997) 432.
A. Bamberger, et al., Nucl. Instrum. Methods A 401 (1997) 63.
A. Dwurazny, et al., Nucl. Instrum. Methods A 277 (1989) 176.
J. Andruszkow, et al., Acta Phys. Pol. B 32 (2001) 2025.
L. Hand, Phys. Rev. 129 (1963) 1834.
W.H. Smith, K. Tokushuku, L.W. Wiggers, in: C. Verkerk, W. Wojcik (Eds.), Proc. Computing in HighEnergy Physics (CHEP), Annecy, France, September 1992, CERN, Geneva, Switzerland, 1992, p. 222, also
in preprint DESY 92-150B.
W.H. Smith, et al., Nucl. Instrum. Methods A 355 (1995) 278.
J. Tandler, PhD thesis, Universitt Bonn, Germany, Report BONN-IR-2003-06 ISSN-0172-8741, 2003.
R. Ciesielski, PhD thesis, Warsaw University, Poland, 2004.
K. Voss, PhD thesis, Universitt Bonn, Germany, 2004.
G.M. Briskin, PhD thesis, Tel Aviv University, Report DESY-THESIS-1999-036, DESY, 1999.
R. Brun, et al., G EANT 3, Technical Report CERN-DD/EE/84-1, CERN, 1987.
K. Muchorowski, PhD thesis, Warsaw University, Poland, 1996, unpublished.
RAPID COMMUNICATION
37
[37] A. Kwiatkowski, H. Spiesberger, H.-J. Mhring, Comput. Phys. Commun. 69 (1992) 155;
A. Kwiatkowski, H. Spiesberger, H.-J. Mhring, in: W. Buchmller, G. Ingelman (Eds.), Proc. Workshop
Physics at HERA, vol. 3, DESY, Hamburg, 1991, p. 1419.
[38] H. Spiesberger, An Event Generator for ep Interactions at HERA Including Radiative Processes (Version
4.6), 1996, available on http://www.desy.de/~hspiesb/heracles.html.
[39] E. Barberio, Z. Was, Comput. Phys. Commun. 79 (1994) 291.
[40] L. Adamczyk, PhD thesis, University of Mining and Metallurgy, Cracow, Poland, Report DESY-THESIS1999-045, DESY, 1999.
[41] M. Kasprzak, PhD thesis, Warsaw University, Warsaw, Poland, Report DESY lF35D-96-16, DESY, 1996.
[42] S.P. Baranov, et al., in: W. Buchmller, G. Ingelman (Eds.), Proc. Workshop on Physics at HERA, vol. 3,
DESY, Hamburg, Germany, 1991, p. 1478.
[43] T. Abe, Comput. Phys. Commun. 136 (2001) 126.
[44] Particle Data Group, K. Hagiwara, et al., Phys. Rev. D 66 (2002) 010001.
[45] V.M. Budnev, et al., Phys. Rep. C 15 (1974) 181.
[46] ZEUS Collaboration, S. Chekanov, et al., Eur. Phys. J. C 21 (2001) 443.
[47] D.Y. Ivanov, et al., hep-ph/0401131.
[48] L. Frankfurt, W. Koepf, M. Strikman, Phys. Rev. D 54 (1996) 3194.
[49] L. Frankfurt, W. Koepf, M. Strikman, Phys. Rev. D 57 (1998) 512.
[50] A.D. Martin, M.G. Ryskin, T. Teubner, Phys. Rev. D 62 (2000) 14022.
[51] E. Gotsman, et al., Acta Phys. Pol. B 34 (2003) 3255.
[52] I. Balitsky, Nucl. Phys. B 463 (1996) 99.
[53] Yu. Kovchegov, Phys. Rev. D 60 (2000) 034008.
[54] ZEUS Collaboration, S. Chekanov, et al., Phys. Rev. D 67 (2003) 12007.
[55] H.L. Lai, et al., Phys. Rev. D 55 (1997) 1280.
[56] A.D. Martin, et al., Eur. Phys. J. C 23 (2002) 73.
[57] J. Pumplin, et al., JHEP 0207 (2002) 012.
[58] S.J. Brodsky, et al., Phys. Rev. D 50 (1994) 3134.
[59] L. Frankfurt, M. Strikman, Phys. Rev. D 66 (2002) 031502.
[60] P.D.B. Collins, An Introduction to Regge Theory and High Energy Physics, Cambridge Univ. Press, 1977.
[61] S.J. Brodsky, et al., JETP Lett. 70 (1999) 155.
[62] L. Frankfurt, M. McDermott, M. Strikman, JHEP 0103 (2001) 45.
[63] A. Donnachie, P.V. Landshoff, Phys. Lett. B 348 (1995) 213.
[64] G.A. Jaroszkiewicz, P.V. Landshoff, Phys. Rev. D 10 (1974) 170.
[65] K. Schilling, G. Wolf, Nucl. Phys. B 61 (1973) 381.
[66] A. Donnachie, P.V. Landshoff, Nucl. Phys. B 244 (1984) 322.

Classification of the chiral Z2 Z2 fermionic

models in the heterotic superstring
A.E. Faraggi a , C. Kounnas b , S.E.M. Nooij a , J. Rizos b,c
a Theoretical Physics Department, University of Oxford, Oxford OX1 3NP, UK
b Laboratoire de Physique Theorique, Ecole Normale Suprieure, 24 rue Lhomond,
F-75231 Paris cedex 05, France

c Department of Physics, University of Ioannina, GR45110 Ioannina, Greece
Received 26 March 2004; accepted 11 June 2004
Abstract
The first particle physics observable whose origin may be sought in string theory is the triple
replication of the matter generations. The class of Z2 Z2 orbifolds of six-dimensional compactified
tori, that have been most widely studied in the free fermionic formulation, correlate the family
triplication with the existence of three twisted sectors in this class. In this work we seek an
improved understanding of the geometrical origin of the three generation free fermionic models.
Using fermionic and orbifold techniques we classify the Z2 Z2 orbifold with symmetric shifts
on six-dimensional compactified internal manifolds. We show that perturbative three generation
models are not obtained in the case of Z2 Z2 orbifolds with symmetric shifts on complex tori,
and that the perturbative three generation models in this class necessarily employ an asymmetric
shift. We present a class of three generation models in which the SO(10) gauge symmetry cannot be
broken perturbatively, while preserving the Standard Model matter content. We discuss the potential
implications of the asymmetric shift for strongweak coupling duality and moduli stabilization. We
show that the freedom in the modular invariant phases in the N = 1 vacua that control the chiral
content, can be interpreted as vacuum expectation values of background fields of the underlying
N = 4 theory, whose dynamical components are projected out by the Z2 -fermionic projections. In
this class of vacua the chiral content of the models is determined by the underlying N = 4 mother
theory.
PACS: 11.25.Mj; 12.10.Dm
E-mail address: faraggi@thphys.ox.ac.uk (A.E. Faraggi).

doi:10.1016/j.nuclphysb.2004.06.030
42
A.E. Faraggi et al. / Nuclear Physics B 695 (2004) 4172
1. Introduction
String theory is in a precarious state of affairs. On the one hand the theory shows
great promise in its ability to provide a consistent framework for perturbative quantum
gravity, while at the same time giving rise to the gauge and matter structures that are
observed experimentally. However, the existence of a multitude of possible string vacua
has led some authors to lose all hope and to advocate resorting to anthropic principles
as the possible resolution for the contrived set of parameters that seem to govern our
universe [1].
Our point of view is different. Ultimately the search for the principles that underly
string theory and the vacuum selection will entail the conceptual resolution of the quantum
gravity synthesis, and the fundamental understanding of quantum mechanics with its
probabilistic interpretation when applied to the spacetime arena.
A more pragmatic view of string theory suggests that the basic properties of the low
energy data, as well as the basic properties of string theories should be utilized in trying to
isolate vacua, or classes of vacua, that look most promising. From the low energy data point
of view we may hypothesize that the viable string theory vacua should accommodate two
pivotal ingredients: the existence of three generations and their embedding in an underlying
SO(10), or E6 grand unified group structure. From the string theory point of view the basic
properties that may serve as guides are the various T- and S-duality symmetries. In this
respect it is also plausible that the self-dual points under these dualities may play a role in
the vacuum selection principle.
A given set of string vacua that exhibits compelling properties must then be investigated
in depth. In the least these can be viewed as case examples providing the concrete
laboratories to study how the properties of the observed data may arise from quantum
gravity, and to develop the tools to relate between the theory and experiment. However,
there also exist the possibility that certain case examples capture some properties of
the true string vacuum that may eventually prove relevant for the understanding of the
low energy data. In any case, it is clear that different approaches must be pursued
for better understanding of string theory and its possible connection with experimental
data.
The first sector among the low energy experimental observables whose origin we
may seek in a theory of quantum gravity is the flavor sector. In the context of the
quantum field theories underlying the Standard Model of particle physics, this group
of parameters does not arise from any physical principle, like the gauge principle. It
is then encouraging that already from the early days of superstring phenomenology, it
was observed that the flavor replication is related a topological property of the string
compactifications, namely the Euler characteristic [2]. However, this observation does not
yet explain the existence of three generations. The first particle physics observable whose
origin we may seek to relate to string theory is therefore the replication of the three matter
generations.
Among the most advanced string models to date are the three generation heterotic string
models [38], constructed in the free-fermion formulation [9]. These models have been
the subject of detailed studies, showing that they can, at least in principle, account for
desirable physical features including the observed fermion mass spectrum, the longevity of
43
the proton, small neutrino masses, the consistency of gauge-coupling unification with the
experimental data from LEP and elsewhere, and the universality of the soft supersymmetrybreaking parameters [10]. An important property of the fermionic construction is the
standard SO(10) embedding of the Standard Model spectrum, which ensures natural
consistency with the experimental values for s (MZ ) and sin2 W (MZ ). Furthermore,
this class of models yielded the only known string model that reproduces in the low
energy effective field theory solely the spectrum of the minimal supersymmetric standard
model [11].
A vital property of the realistic fermionic models is their underlying Z2 Z2 orbifold
structure. Many of the encouraging phenomenological characteristics of these models are
rooted in this structure. In particular, the emergence of the three chiral generations in a large
class of fermionic constructions is correlated with the existence of three twisted sectors in
the Z2 Z2 orbifold of the six-dimensional internal manifold. Each twisted sector produces
exactly one of the light chiral generations and there is no additional chiral matter. Thus,
the fermionic construction offers a plausible and compelling explanation to the existence
of three generations in nature.
To see more precisely the orbifold correspondence of the fermionic construction, we
recall that the free-fermion models are generated by a set of basis vectors which define the
transformation properties of the world-sheet fermions as they are transported around noncontractible loops of the string world sheet. A set of realistic fermionic models contains
a subset of boundary conditions, the so-called extended NAHE-set, which can be seen
to correspond to Z2 Z2 orbifold compactification with the standard embedding of the
gauge connection [12]. The fermionic model constructed just with the basis vectors of the
extended NAHE-set gives rise to 24 generations from the twisted sectors, as well as three
additional generation/antigeneration pairs from the untwisted sector. At the N = 4 level
the fermionic point in the moduli space corresponds to an SO(12) enhancement of the
internal lattice. The induced Z2 Z2 action gives rise to a model with (h11 , h21 ) = (27, 3),
matching the data of the free-fermion model. We note that the data of this model differs
from the Z2 Z2 orbifold at a generic point in the moduli space, which has (h11 , h21 ) =
(51, 3). Alternatively, we can start with the Z2 Z2 orbifold at a generic point and produce
the one at the free fermionic point by adding a freely acting shift on the internal lattice
[13,14].
The above remarks make apparent the need to understand better the general structure
of the realistic free fermionic models, and, in particular, the geometrical structure that
underlies the three generation models.
In the framework of the fermionic construction the three generations are obtained by
adding three, or four, additional boundary condition basis vectors beyond the minimal
NAHE-set. The basis vectors reduce the number of generations to three generations, one
from each of the twisted sectors of the Z2 Z2 orbifold.
In this paper we observe in some of the concrete quasi-realistic three generation
models [6] that the action of two of the additional boundary condition basis vectors
correspond to symmetric shifts on the internal coordinates, whereas the third corresponds
to a fully asymmetric shift. We then proceed to classify all possible Z2 Z2 orbifolds
with symmetric shifts, and demonstrate that three generations cannot be obtained solely
with symmetric shifts on complex tori. This is one of the main results of the analysis and it
44
reveals, at least in the context of the three generation models, that the geometrical structures
that underly these models may not be simple CalabiYau manifolds, but it corresponds to
geometries that are yet to be defined. This observation may eventually prove important for
the issue of moduli stabilization.
Additionally, we will demonstrate the existence of three generation models with a
perturbatively unbroken SO(10)/E6 gauge group, in which the internal manifold is reduced
to a product of six circles. This again demonstrates the possibility that the geometries
relating to the viable vacua may not correspond to the complex geometries that have been
more prevalent in the literature. Some of phenomenological difficulties that have been
associated with symmetric compactifications, like supersymmetry breaking and moduli
stabilization, may therefore be cured in the viable geometries. This class of models, while
not viable with respect to perturbative phenomenology, produces one generation from a
single fixed point in each twisted sector. Hence, realizing the Z2 Z2 geometric picture
of the three chiral generations. Our classification demonstrates additionally that in a large
class of N = 1 models the freedom in the phases appearing in the N = 1 partition function
can be understood as the vacuum expectation value (VEV) of background fields of the
N = 4 underlying theory, whose dynamical components are projected out by the extra
Z2 Z2 projections. Thus, the information on the chiral content of the N = 1 models is
already contained at the N = 4 level. Examples of this phenomenon are already noted
in the case of the Z2 Z2 orbifold on SO(12) versus SO(4)3 lattices, as discussed
above.
Our paper is organized as follows: in Section 2 we discuss the general structure of
the models based on the fermionic construction. In a concrete model we show that the
additional boundary vectors beyond the NAHE-set can be regarded as two symmetric shifts
plus one fully asymmetric shift. The main aim of this section is to establish the connection
of the analysis to follow with the phenomenological three generation models.
In Section 3 we present the setup of our analysis. We present the most general free
fermionic model describing the heterotic string on a Z2 Z2 orbifold. In Section 4 we
present our method to classify all possible symmetric shifts and proceed to perform the
complete classification for gauge groups that descend from the N = 4 mother theory.
We find that down to six generations the perturbative models can be described in
terms of symmetric shifts and hence possess a geometrical interpretation in terms of
Z2 Z2 symmetric orbifolds. However, the three generation perturbative models are
not admitted in this classification and entail an additional shift which is necessarily
asymmetric between the left and the right-movers. We demonstrate the existence of a
class of three twisted generation models in which the GUT symmetry group cannot be
broken perturbatively, while preserving complete twisted matter multiplets. Additionally,
in this class of models the six-dimensional internal lattice is reduced to a product
of six circles. Hence, one of the main conclusions of the analysis is that in the
framework of Z2 Z2 orbifolds, three generations models are not obtained solely with
symmetric shifts on complex tori, and suggests that the geometrical objects underlying
the quasi-realistic free fermionic models are more esoteric than ordinary Z2 Z2
CalabiYau manifolds. In Section 5 we present our results and Section 6 concludes our
paper.
45
2. General structure of realistic free fermionic models

In this section we recapitulate the main structure of the realistic free fermionic models.
The notation and details of the construction of these models are given elsewhere [36,11,
15,16]. In the free fermionic formulation of the heterotic string in four dimensions all the
world-sheet degrees of freedom required to cancel the conformal anomaly are represented
in terms of free world-sheet fermions [9]. In the light-cone gauge the world-sheet field
content consists of two transverse left- and right-moving spacetime coordinate bosons,
X1,2 and X 1,2 , and their left-moving fermionic superpartners 1,2 , and additional 62 purely
internal MajoranaWeyl fermions, of which 18 are left-moving, and 44 are right-moving.
In the supersymmetric sector the world-sheet supersymmetry is realized non-linearly and
the world-sheet supercurrent [17] is given by
TF = X + i I y I I
(I = 1, . . . , 6).
(2.1)
The { I , y I , I } (I = 1, . . . , 6) are 18 real free fermions transforming as the adjoint

representation of SU(2)6 . Under parallel transport around a non-contractible loop on
the toroidal world-sheet the fermionic fields pick up a phase, f ei(f ) f , (f )
(1, +1]. Each set of specified phases for all world-sheet fermions, around all the noncontractible loops is called the spin structure of the model. Such spin structures are usually
given in the form of 64-dimensional boundary condition vectors, with each element of the
vector specifying the phase of the corresponding world-sheet fermion. The basis vectors
are constrained by string consistency requirements and completely determine the vacuum
structure of the model. The physical spectrum is obtained by applying the generalized GSO
projections [9].
The boundary condition basis defining a typical realistic free fermionic heterotic string
model is constructed in two stages. The first stage consists of the NAHE set, which is a
set of five boundary condition basis vectors, {1, S, b1 , b2 , b3 } [15,18]. The gauge group
induced by the NAHE set is SO(10) SO(6)3 E8 with N = 1 supersymmetry. The
spacetime vector bosons that generate the gauge group arise from the NeveuSchwarz
sector and from the sector 2 1 + b1 + b2 + b3 . The NeveuSchwarz sector produces
the generators of SO(10) SO(6)3 SO(16). The 2 -sector produces the spinorial 128
of SO(16) and completes the hidden gauge group to E8 . The NAHE set divides the
internal world-sheet fermions in the following way: 1,...,8 generate the hidden E8 gauge
group, 1,...,5 generate the SO(10) gauge group, and {y 3,...,6 , 1 }, {y 1 , y 2 , 5 , 6 , 2 },
{ 1,...,4 , 3 } generate the three horizontal SO(6) symmetries. The left-moving {y, } states
are divided into {y 3,...,6 }, {y 1 , y 2 , 5 , 6 }, {1,...,4 } and 12 , 34 , 56 generate the leftmoving N = 2 world-sheet supersymmetry. At the level of the NAHE set the sectors b1 ,
b2 and b3 produce 48 multiplets, 16 from each, in the 16 representation of SO(10). The
states from the sectors bj are singlets of the hidden E8 gauge group and transform under
the horizontal SO(6)j (j = 1, 2, 3) symmetries. This structure is common to all known
realistic free fermionic models.
The second stage of the construction consists of adding to the NAHE set three (or
four) additional basis vectors. These additional vectors reduce the number of generations
to three, one from each of the sectors b1 , b2 and b3 , and simultaneously break the fourdimensional gauge group. The assignment of boundary conditions to { 1,...,5 } breaks
46
SO(10) to one of its subgroups SU (5) U (1) [3], SO(6) SO(4) [5], SU(3) SU(2)
U (1)2 [4,6,11], SU(3) SU(2)2 U (1) [16] or SU(4) SU(2) U (1) [20]. Similarly, the
hidden E8 symmetry is broken to one of its subgroups, and the flavor SO(6)3 symmetries
are broken to U (1)n , with 3 n 9. For details and phenomenological studies of these
three generation string models we refer interested readers to the original literature and
review articles [10].
The correspondence of the free fermionic models with the orbifold construction is
illustrated by extending the NAHE set, {1, S, b1, b2 , b3 }, by at least one additional
boundary condition basis vector [12]
1 = (0, . . . , 0| 1, . . . , 1 , 0, . . . , 0).

(2.2)
1,...,5 , 1,2,3
With a suitable choice of the GSO projection coefficients the model possesses an SO(4)3
E6 U (1)2 E8 gauge group and N = 1 spacetime supersymmetry. The matter fields
include 24 generations in the 27 representation of E6 , eight from each of the sectors
b1 b1 + 1 , b2 b2 + 1 and b3 b3 + 1 . Three additional 27 and 27 pairs are obtained
from the NeveuSchwarz 1 sector.
To construct the model in the orbifold formulation one starts with the compactification
on a torus with nontrivial background fields [19]. The subset of basis vectors
{1, S, 1 , 2 }
(2.3)
generates a toroidally-compactified model with N = 4 spacetime supersymmetry and

SO(12) E8 E8 gauge group. The same model is obtained in the geometric (bosonic)
language by tuning the background fields to the values corresponding to the SO(12) lattice.
The metric of the six-dimensional compactified manifold is then the Cartan matrix of
SO(12), while the antisymmetric tensor is given by
i > j,
Gij ,
i = j,
Bij = 0,
(2.4)
G , i < j.
ij
When all the radii of the six-dimensional compactified manifold are fixed at RI = 2,
I
it is seen that the left- and right-moving momenta PR,L
= [mi 12 (Bij Gij )nj ]eiI
reproduce the massless root vectors in the lattice of SO(12). Here ei = {eiI } are six linearly
independent vielbeins normalized so that (ei )2 = 2. The eiI are dual to the ei , with
ei ej = ij .
Adding the two basis vectors b1 and b2 to the set (2.3) corresponds to the Z2 Z2
orbifold model with standard embedding. Starting from the N = 4 model with SO(12)
E8 E8 symmetry [19], and applying the Z2 Z2 twist on the internal coordinates,
reproduces the spectrum of the free-fermion model with the six-dimensional basis set
{1, S, 1 , 2 , b1 , b2 }. The Euler characteristic of this model is 48 with h11 = 27 and h21 = 3.
It is noted that the effect of the additional basis vector 1 of Eq. (2.2), is to separate
the gauge degrees of freedom, spanned by the world-sheet fermions { 1,...,5 , 1 , 2 , 3 ,
1,...,8 }, from the internal compactified degrees of freedom {y, |y,
}
1,...,6 . In the realistic
47
free fermionic models this is achieved by the vector 2 [12], with

2 = (0, . . . , 0|
1, . . . , 1

, 0, . . . , 0),
(2.5)
1,...,5 , 1,2,3 1,...,4
which breaks the E8 E8 symmetry to SO(16) SO(16). The Z2 Z2 twist induced

by b1 and b2 breaks the gauge symmetry to SO(4)3 SO(10) U (1)3 SO(16). The
orbifold still yields a model with 24 generations, eight from each twisted sector, but now
the generations are in the chiral 16 representation of SO(10), rather than in the 27 of E6 .
The same model can be realized with the set {1, S, 1 , 2 , b1 , b2 }, by projecting out the
16 16 from the 1 -sector taking

1
c 1 .
c
(2.6)
2
2
This choice also projects out the massless vector bosons in the 128 of SO(16) in the
hidden-sector E8 gauge group, thereby breaking the E6 E8 symmetry to SO(10)
U (1) SO(16). We can define two N = 4 models generated by the set (2.3), Z+ and
Z , depending on the sign in Eq. (2.6). The first, say Z+ , produces the E8 E8 model,
whereas the second, say Z , produces the SO(16) SO(16) model. However, the Z2 Z2
twist acts identically in the two models, and their physical characteristics differ only due
to the discrete torsion Eq. (2.6).
This analysis confirms that the Z2 Z2 orbifold on the SO(12) lattice is at the core of
the realistic free fermionic models. To illustrate how the chiral generations are generated
in the free fermionic models we consider the E6 model which is produced by the extended
NAHE-set {1, S, 1 , 2 , b1 , b2 }.
b
The chirality of the states from a twisted sector bj is determined by the free phase c bj .
i
Since we have a freedom in the choice of the sign of this free phase, we can get from the
sector (bi ) either the 27 or the 27. Which of those we obtain in the physical spectrum
b
depends on the sign of the free phase. The free phases c bj also fix the total number
i
of chiral generations. Since there are two bi projections for each sector bj , i = j we can
use one projections to project out the states with one chirality and the other projection to
project out the states with the other chirality. Thus, the total number of generations with
this set of basis vectors is given by

b1

b2

b3
c b2 + c bb13
c b1 + c bb23
c b1 + c bb31
8
+8
+8
.
2
2
2
b
Since the modular invariance rules fix c bj = c bji we get that the total number of
i
generations is either 24 or 8. Thus, to reduce the number of generation further it is
necessary to introduce additional basis vectors.
To illustrate the reduction to three generations in the realistic free fermionic models we
consider the model in Table 1.
Here the vector 1 (2.2) is replaced by the vector 2 (2.5). At the level of the NAHE set
we have 48 generations. One half of the generations is projected by the vector 2 . Each of
the three vectors in Table 1 acts non-trivially on the degenerate vacuum of the sectors b1 ,
48
Table 1
12 34 56
1,...,5
0
0
0
0
0
0
1
1
1
1
1
1
0
0
1
2
1
2
1
2
1
2
0
0
0
0
0
0
1 2 3
1,...,8
0
0
0
0
0
0
0
0
1
1
1
2
1
2
1
2
1
2
1
2
1
1
0
1
1
1
1
1
1
0
0
0
0
0
0
1
2
1
2
1
2
0
0
0
y3y6
y 4 y 4
y 5 y 5
y 3 y 6
y 1 5
y 2 y 2
6 6
y 1 5
2 4
1 1
3 3
2 4
1
0
0
0
0
1
0
1
0
0
1
1
0
1
0
0
0
1
1
0
0
1
0
1
0
0
1
0
1
0
1
0
0
1
1
0
Table 2
+
+
+ +
y 3 y 6 y 4 y 4 y 5 y 5 y 3 y 6
y 1 5 y 2 y 2 6 6 y 1 5
2 4 1 1 3 3 2 4
1
0
1
1
1
1
0
1
1
0
1
1
1
1
1
1
0
0
0
1
1
1
0
1
1
1
0
1
1
1
1
0
1
0
1
0
b2 and b3 and reduces the number of generations in each step by a half. Thus, we obtain
one generation from each sector b1 , b2 and b3 .
The geometrical interpretation of the basis vectors beyond the NAHE set is facilitated
by taking combinations of the basis vectors in Table 1, which entails choosing another set
to generate the same vacuum. The combinations + , + , + + produce the
following boundary conditions under the set of internal real fermions.
It is noted that the two combinations + and + are fully symmetric between
the left and right movers, whereas the third, + + , is asymmetric. The action of
the first two combinations on the compactified bosonic coordinates translates therefore to
symmetric shifts. Thus, we see that reduction of the number of generations is obtained by
further action of symmetric shifts.
Due to the presence of the third combination the situation, however, is more
complicated. The third combination in Table 2 is asymmetric between the left and right
movers and therefore does not have an obvious geometrical interpretation. Below we
perform a complete classification of all the possible NAHE-based Z2 Z2 orbifold models
with symmetric shifts on the complex tori, which reveals that three generations are not
obtained in this manner. Three generations are obtained in the free fermionic models by
the inclusion of the asymmetric shift in Table 2. This outcome has profound implications
on the type of geometries that may be related to the realistic string vacua, as well as on the
issue of moduli stabilization.
3. N = 1 heterotic orbifold constructions
In this section we revise the Z2 Z2 heterotic orbifold construction and relate this
to the free fermionic construction. We isolate the individual conformal blocks that will
facilitate the classification of the models and set up a procedure to analyse all possible
49
N = 1 heterotic Z2 Z2 models. We start by describing the procedure to descend from

N = 4 to N = 1 supersymmetric heterotic vacua.
3.1. The N = 4 models
The partition function for any heterotic model via the fermionic construction is
1
20 12
44
a 1
1
1
aj 2
ai
.
Z=
bi
bj
b 2M
2 12 24
a,b
i=1
(3.1)
j =1
In the above equation M is the number of basis vectors and the parameters in the functions represent the action of the vectors. In order to obtain a supersymmetric model
we need at least two basis vectors {1, S}.

1 = 1,2 , 1,...,6 , y 1,...,6 , 1,...,6 y 1,...,6 , 1,...,6 , 1,...,6 , 1,2,3, 1,...,8 ,
(3.2)
1,2 1,...,6
.
S = ,
(3.3)
The supersymmetric GSO projection is induced by the set S for any choice of the GSO
coefficient

S
c
= 1.
(3.4)
1
The corresponding partition function has a factorized left-moving contribution coming
from the sector S,
a 4
1
6,6+16 [SO(44)]
1 1
a+b+ab b
(1)
Z1,S =
(3.5)
4
4
2 || 2
6 22
a,b=0
where
1
6,6+16 SO(44) =
2
c,d
and
=
c 6
c 22
d d
,
6
22
(3.6)

1
S
1c
1
2
defines the chirality of N = 4 supersymmetry. Therefore, the role of the boundary condition
vector S is to factorize the left-moving contribution,

3
1
1
a
a
L
ZN=4
(3.7)
=
(1)a+b+ab
(v)
(0) v 4
b
b
2
a,b=0
which is zero with the multiplicity of N = 4 supersymmetry.

The above partition function gives rise to an SO(44) right-moving gauge group and
is the maximally symmetric point in the moduli space of the Narain 6,6+16 lattice. The
general 6,6+16 lattice depends on 6 22 moduli, the metric Gij and the antisymmetric
50
tensor Bij of the six-dimensional internal space, as well as the Wilson lines YiI that appear
in the 2d-world-sheet.

1
1
S=
d 2 gg ab Gij a Xi b Xj +
d 2 ab Bij a Xi b Xj
4
4

1
I

i I .
+
+ YiI X
(3.8)
d 2 g
4
I
Here i runs over the internal coordinates and I runs over the extra 16 right-moving degrees
of freedom described by I .
The compactified sector of the partition function is given by 6,6+16

j

Tij
i
(det G)3
i
j
m
6,6+16 =
exp
+
n
m
+
n
2
23
m,n
16

1
exp ini mj + nj YiI YjI
2
, I =1

I i

Yi m + ni | ,
(3.9)
where Tij = Gij + Bij .

Eq. (3.9) is the winding mode representation of the partition function. Using a Poisson
resummation we can put it in the momentum representation form:

i
i ij
Pi Gij Pj
exp
6,22 =
(3.10)
Pi G Pj i Q I Q I ,
2
2
P ,P ,Q
with
1
Pi = mi + Bij nj + YiI YjI nj + YiI QI + Gij nj ,
2
1
Pi = mi + Bij nj + YiI YjI nj + YiI QI Gij nj ,
2
Q I = QI + YiI ni .
(3.11)
(3.12)
(3.13)
The charge momenta QI are induced by the right-moving fermions I which appear
explicitly in the -functions
I
(QI )2
bI
I
a
q 2 e2i(v 2 )Q ,
I =
(3.14)
b
nZ
where the charge momentum QI = (n a2 ).

For generic Gij , Bij and for vanishing values for Wilson lines, YiI = 0 one obtains an
N = 4 model with a gauge group U (1)6 SO(32). The U (1)6 can be extended to SO(12)
by fixing the moduli of the internal manifold [12].
The N = 4 fermionic construction based on {1, S} (3.5) has an extended gauge group,
SO(44). From the lattice construction point of view, an N = 4 model with a gauge group
51
G SO(44) can be generated by switching on Wilson lines and fine tune the moduli of the
internal manifold. Moving from the SO(44) to U (1)6 SO(32) heterotic point as well as
to the U (1)6 E8 E8 point can be realized continuously [23]. The partition function at
the U (1)6 E8 E8 point takes a simple factorized form
6,6+16 =

Tij
i
m + ni mj + nj
exp
2
m,n
(det G)3
23

1 8 1 +h 8
+g
2
2
,
(3.15)
3.2. The N = 1 models

To break the number of supersymmetries down from N = 4 to N = 1 in the fermionic
formulation we need to introduce the vectors b1 and b2 .

b1 = 3,4 , 5,6 , y 3,4, y 5,6 | . . . ,

b2 = 1,2 , 5,6 , y 1,2, y 5,6 | . . . .
(3.16)
(3.17)
The b1 twists the second and third complex planes (3, 4) and (5, 6) while b2 twists the
first and third (1, 2) and (5, 6) ones. Thus, b1 , b2 separate the internal lattice into the three
complex planes: (1, 2), (3, 4) and (5, 6).
The action of the bi -twists fully determines the fermionic content for the left-moving
sector. The dots . . . in b1 , b2 stand for the n1 , n2 right-moving fermions. To generate a
modular invariant model we can distinguish four options. ni are either 8, 16, 24 or 32 real
right-moving fermions in the basis vector bi .
Defining the basis vectors with 8 real right-moving fermions leads to massless states
in the spectrum in vectorial representations of the gauge groups; 16 real right-moving
fermions give rise to spinorial representations on each plane. Adding either 24 or 32 rightmoving fermions would produce massive states in the spectrum. We therefore discard the
last two options. We thus need to introduce 16 real fermions (8 complex) in the vectors
b1 , b2 for the existence of spinorial representations on the first and second plane.
A suitable choice is for instance,

b1 = 3,4 , 5,6 , y 3,4, y 5,6 |y 3,4, y 5,6 , 1 , 1,...,5 ,

b2 = 1,2 , 5,6 , y 1,2, y 5,6 |y 1,2, y 5,6 , 2 , 1,...,5 .
(3.18)
(3.19)
We define the vectors x = {0, . . . | 1,...,5 , 1,2,3 }, and b1,2 = b1,2 + x. The N = 1 partition
function based on {1, S, b 1 , b2 } takes the following form:
52
1 1 i(a+b+ab)
e
2 ||4 2
,

a+h2
a+h1
ah1 h2
1 ab b+g2 b+g1 bg1 g2
h1 ,h2 ,g1 ,g2
,h1 ,h2
,h1 ,h2
13
1 6,6 ,g1 ,g2 Z ,g1 ,g2 Z26
eiL ,
6 6
3
,

2
2

+ h1
, h1 , h2
+ h2

6,6
=
, g1 , g2
+ g2
+ g1

2

h1 h2

,
g1 g2

, h1 , h2
+ h2 + h1 h1 h2
Z
=
,
, g1 , g2
+ g2
+ h2
g1 g2

13
Z26
.
=
ZN=1 =
(3.20)
(3.21)
(3.22)
(3.23)
In Eq. (3.21) the internal manifold is twisted and thereby separated explicitly into three
planes. The above model is the minimal Z2 Z2 with N = 1 supersymmetry and massless
spinorial representations in the same SO(10) group coming from the first and/or from
the second plane. The number of families depends on the choice of the phase L . The
freedom of this
arises from the different possible choices of the modular invariance

v phase
coefficients c vji . The maximal number of the families for this model is 32. Introducing
internal shifts, associated to L , can reduce this number as we will discuss below.
We could have chosen the boundary conditions for different right-moving fermions.
This would lead to spinorial representations on each plane, but the group to which they
would belong would differ in each plane. As we require spinors in the same group we
have discarded this option. Choosing an overlap with more than 6 complex fermions in the
right-moving sector between the vectors b1 and b2 leads to a SO(14) gauge group, which
does not have chiral fermions.
In order to have spinors in the spectrum on all three planes we need to separate at least
an SO(16) (or E8 ) from the 6,22 lattice. We therefore need to introduce the additional
vector

z = 1,...,8
(3.24)
to the set. With this vector the partition function for the gauge sector (3.23) modifies to

1 5 + hz 8
, hz
Z26
(3.25)
=
.
, gz
+ gz
2
hz ,gz
We can further separate out the internal 6,6 lattice by introducing the additional vector,
e = {y1,...,6 , 1,...,6 |y1,...,6 , 1,...,6 },
(3.26)
53
which modifies the 6,6 in (3.21) by

6,6

2
1 + he
, h1 , h2
+ he + h2
=
+ ge
, g1 , g2
+ ge + g2
2
he ,ge

2
+ he
+ he + h1

+ ge
+ ge + g1

2
+ he
+ he h1 h2

.
+ ge
+ ge g1 g2
(3.27)
In the above {1, S, b1 , b2 , e} construction the gauge group of the observable sector becomes
either SO(10) U (1)3 or E6 U (1)2 and the hidden sector necessarily is SO(16) or E8
depending on the generalized GSO coefficients, (the choice of the phase L ), while the
gauge group from the 6,6 lattice becomes GL = SO(6) U (1)3 .
So far the construction of the N = 1 models is generic. The only requirement
we are imposing is the presence of spinors on all three planes. We call this the S 3
subclass of models. In a general N = 1 model the spinors could be replaced by vectorial
representations of the observable gauge group. This replacement gives rise to three
additional classes of models which we denote by S 2 V , SV 2 and V 3 . In this work we
will focus on the S 3 class and we will deal with the other classes in a future work. The
condition of spinorial representations arising from each one of the Z2 Z2 orbifold planes
together with the complete separation of the internal manifold is synonymous to having a
well-defined hidden gauge group.
3.3. The general S 3 N = 1 model
In the class of Z2 Z2 orbifold models, the internal manifold is broken into three
planes. The hidden gauge group is necessarily E8 or SO(16) broken to any subgroup by
Wilson lines (at the N = 4 level). In order to classify all possible S 3 models, it is necessary
to consider all possible basis vectors consistent with modular invariance. Namely:

z1 = 1,...,4 ,

z2 = 5,...,8 ,
ei = {yi , i |yi , i },
(3.28)
(3.29)
i {1, 2, 3, 4, 5, 6}.
(3.30)
The z1 , z2 vectors allow for a breaking of hidden E8 or SO(16) to SO(8) SO(8)

depending on the modular coefficients. As we discuss below this splitting of the hidden
gauge group has important consequences in the classification of the S 3 class of models by
the number of generations. The introduction of ei vectors is necessary in order to obtain
all possible internal shifts which also induces all possible modification to the number of
generations.
54
The general N = 1, S 3 model based on {1, S, ei , z1 , z2 , b1 , b2 } is

1 1 i(a+b+ab)
e
2 ||4 2
,

a+h1
a+h2
ah1 h2
1 ab b+g1 b+g2 bg1 g2
h1 ,h2 ,g1 ,g2
h1 |p1 ,p2
h2 |p3 ,p4
h1 h2 |p5 ,p6
1 2,2 g1 |q1 ,q2 2,2 g2 |q3 ,q4 2,2 g1 g2 |q5 ,q6
6
2 p ,q
2 2
2 2
2 2
i i
,h1 ,h2

Z16 ,,

Z ,g
Z
1
10
,g
1 2
eiL ,
8
3
5
8

, ,,, ,

, h1 , h2
+ h2 + h1 h1 h2
Z
=
,
, g1 , g2
+ g2
+ h2
g1 g2

5
Z10
=
,

4
4
,
+
Z16
=
.
,
+
ZN=1 =
(3.31)
(3.32)
(3.33)
(3.34)
The 6,6 lattice of N = 4 is twisted by hi , gi , thus in the N = 1 case separated into three
(2, 2) planes. The contribution of each of these planes in N = 1 partition function is written
in terms of twisted by hi , gi and shifted by pi , qi 2,2 lattice. The expressions of those
lattices at the self-dual point (fermionic construction point) is

h|pi , pj
2,2
g|qi , qj f.p

a
1
a
a +h
a + h
j j
,
=
ei1 +i2 i i
(3.35)
bi
bj
bi + g
bj + g
4
ai ,bi ,aj ,bj
where the phases

i = ai qi + bi pi + qi pi ,
j = aj qj + bj pj + qj pj
define the two shifts of the 2,2 lattice. At the generic point of the moduli space the
shifted 2,2 lattice depends on the moduli (T , U ), keeping however identical modular
transformation properties as those of the fermionic point.
For non-zero twist, (h, g) = (0, 0), 2,2 is independent of the moduli T , U and thus it
is identical to that of (3.35) constructed at the fermionic point [21,22]. Thus for non-zero
twist, (h, g) = (0, 0),

h|pi , pj
h|pi , pj

2,2
(3.36)
= 2,2
.
g|qi , qj (T ,U )(h,g)=(0,0)
g|qi , qj f.p
55
For zero twist, (h, g) = (0, 0), the momentum and winding modes are moduli dependent
and are shifted by qi , qj and pi , pj ,

0|pi , pj
2,2
0|qi , qj (T ,U )

pj
pi
i{m1 qi +m2 qj }
+ m2 n2 +
=
e
exp 2i m1 n1 +
2
2

m,
n Z

pj 2
2
pi
m
(3.37)
U
m
+
T
n
+
+
.
+
T
U
n
1
2
1
2
T 2 U2
2
2
The phase L is determined by the chirality of the supersymmetry as well as by the other
modular coefficients

1
v
L (a, b) =
(3.38)
1c i
i j ,
vj
2
i,j
where i and j are the upper- and lower-arguments in -functions corresponding to the
boundary conditions in the two directions of the world sheet torus and which are associated
to the basis vectors vi and vj of the fermionic construction. The only freedom which
remains in the general S
3 N = 1 model is therefore the choice of the generalized GSO
v
projection coefficients c vji = 1. The space of models is classified according to that
choice
determines at the end the phase L . We have in total 55 independent choices

vwhich
for c vji that can take the values 1. Thus, the total number of models in this restricted
class of N = 1 models is 255 . Latter, we will classify all these models according to the
values of the GSO coefficients.
The so-called NAHE models is a small subclass of the general S 3 , N = 1 deformed
fermionic N = 1 model. More precisely we can write the NAHE set basis vectors as a
linear combination of basis vectors {1, S, ei , z1 , z2 , b1 , b2 } which define the general S 3
N = 1 model:
b1NAHE = S + b1 ,
b2NAHE
b3NAHE
(3.39)
= S + b 2 + e5 + e6 ,
(3.40)
= 1 + b1 + b2 + e5 + e6 + z1 + z2 .
(3.41)
We see that the NAHE set is included in these models as mentioned in Section 3.2.
3.4. The N = 4 gauge group
We describe the gauge configuration of the models defined by the basis vectors
{1, S, ei , z1 , z2 , b1 , b2 }. For this purpose we start with a simplification and separate out
the internal manifold using Eq. (3.26). As the twisting vectors b1 and b2 are used to break
the SO(16) SO(10) U (1)3 we will firstly describe the configuration without these
vectors. The gauge group induced by the vectors {1, S, e, z1 , z2 } without enhancements is
G = SO(16) SO(8)1 SO(8)2 SO(12),
(3.42)
56
Table 3
The configuration of the gauge group of the N = 4 theory. We have separated a priori the internal and the hidden
and observable gauge group using the vectors e and zi . Introducing the other vectors ei and bi only induce
breaking of these groups
e
c z12
c z1
c z2
Gauge group G
+
+
+
+
+
+
+
+
E8 SO(28)
SO(24) SO(20)
SO(24) SO(20)
SO(32) SO(12)
SO(16) SO(16) SO(12)
SO(16) SO(16) SO(12)
SO(16) SO(16) SO(12)
E8 E8 SO(12)
where the internal manifold is described by SO(12) and the hidden sector by SO(8)SO(8)
and the observable by SO(16). By choosing the GSO coefficients the SO(16) can enhance
either to E8 or mix with the other sectors producing either SO(24) or SO(32). Similarly
the SO(8) SO(8) can enhance either to SO(16) or E8 or mix with the observable or
internal manifold gauge group. This leads to enhancements of
theform
SO(20)

or SO(24).
The exact form depends only on the three GSO coefficients c ze1 , c ze2 , c zz12 . We have
shown the results in Table 3.
Proceeding to the complete model {1, S, ei , z1 , z2 , b1 , b2 } we break these gauge groups
to their subgroups. Imposing the shifts ei we can break the internal gauge group down to
its Cartan generators by a suitable choice of the coefficients. By a suitable choice we can
break SO(20) SO(8) U (1)6 .
When we also include the twists we break SO(16) SO(10) U (1)3 and E8
E6 U (1)2 . Similarly we can break SO(24) SO(10) U (1)3 SO(8) and SO(32)
SO(10) U (1)3 SO(8) SO(8). Enhancements can subsequently occur of the form
SO(8) U (1) SO(32) SO(10) or SO(8) SO(8) U (1) SO(32) SO(18). We
find possible enhancements of the form SO(10)
SO(8) SO(32) SO(18).

In Table 3 we notice that the coefficient c zz12 distinguishes between the SO(32) models
and the E8 E8 models. Since we require complete separation of the gauge
group into a
well-defined observable and hidden gauge group, we set the coefficient c zz12 = 1 in the
classification.
4. Generic Z2 Z2 model in the free fermionic formulation
4.1. General formalism
In the free fermionic formulation of the heterotic superstring, a model is determined
by a set of basis vectors, associated with the phases picked up by the fermions when
parallelly transported along non-trivial loops and a set of coefficients associated with
GSO projections. The free fermions in the light-cone gauge in the traditional notation are:
, i , y i , i , i = 1, . . . , 6 (left movers) and y i , i , i = 1, . . . , 6, A , A = 1, . . . , 5, ,
= 1, 8 (right movers). The class of models under consideration is generated by a set of
57
12 basis vectors
B = {v1 , v2 , . . . , v12 },
where

v1 = 1 = , 1,...,6 , y 1,...,6 , 1,...,6 |y 1,...,6 , 1,...,6 , 1,2,3, 1,...,5 , 1,...,8 ,

v2 = S = , 1,...,6 ,

v2+i = ei = y i , i |y i , i , i = 1, . . . , 6,

v9 = b1 = 34 , 56 , y 34 , y 56 |y 34, y 56 , 1 , 1,...,5 ,

v10 = b2 = 12 , 56 , y 12 , y 56|y 12 , y 56 , 2 , 1,...,5 ,

v11 = z1 = 1,...,4 ,

v12 = z2 = 5,...,8 .
(4.1)
The vectors 1, S generate an N = 4 supersymmetric model. The vectors ei , i = 1, . . . , 6

give rise to all possible symmetric shifts of internal fermions (y i , i , y i , i ) while b1 and
b2 stand for the Z2 Z2 orbifold twists. The remaining fermions not affected by the action
of the previous vectors are i , i = 1, . . . , 8 which normally give rise to the hidden sector
gauge group. The vectors z1 , z2 divide these eight fermions in two sets of four which in the
Z2 Z2 case is the maximum consistent partition function [9]. This is the most general
basis, with symmetric shifts for the internal fermions, that is compatible with KacMoody
level one SO(10) embedding.
v
j = 1, . . . , 12 and
can
The associated projection coefficients are denoted by c vji
, i,
v
v
take the values 1. They are related by modular invariance c vji = exp{i 2 vi vj }c vji
vi

v
and c vi = exp{i 4 vi vi }c 1j leaving 266 independent coefficients. Out of them, the

requirement of N = 1 supersymmetric spectrum fixes (up to a phase convection) all c vSi ,

i = 1, . . . , 12. Moreover, without loss of generality we can set c 11 = 1, and leave the
rest 55 coefficients free. Therefore, a simple counting gives 255 (that is approximately
1016.6 ) distinct models in the class under consideration. In the following we study this
class of models by deriving analytic formulas for the gauge group and the spectrum and
then using these formulas for the classification.
4.2. The gauge group
Gauge bosons arise from the following four sectors:
G = {0, z1 , z2 , z1 + z2 , x},
where
x =1+S+
6

i=1
ei +
2

zk = 123 , 12345 .
k=1
The 0 sector gauge bosons give rise to the gauge group

SO(10) U (1)3 SO(8)2 .
(4.2)
58
Table 4
Typical enhanced gauge groups and associated projection coefficients for a generic model generated by the basis
(4.1) (coefficients not included equal to +1 except those fixed by spacetime supersymmetry and conventions)
e
c z12
c bz1
c bz2
c bz1
c bz2
c z11
c z22
c e12
Gauge group
1
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
SO(10) SO(18) U (1)2

SO(10) SO(9)2 U (1)3
SO(10)2 SO(9) U (1)2
SO(10)3 U (1)
SO(26) U (1)3
E6 U (1)2 E8
E6 U (1)2 SO(16)
E6 U (1)2 SO(8) SO(8)
SO(10) U (1)3 E8
SO(10) U (1)3 SO(16)
The x gauge bosons when present lead to enhancements of the traditionally called
observable sector (the sector that includes SO(10)) while the z1 + z2 sector can enhance the
hidden sector (SO(8)2 ). However, the z1 , z2 sectors accept oscillators that can also give rise
to mixed type gauge bosons and completely reorganize
gauge group.
The
appearance
the
of mixed states is in general controlled by the phase c zz12 . The choice c zz12 = +1 allows
for mixed gauge
bosons
and leads to the gauge groups presented in Table 4.

The choice c zz12 = 1 eliminates all mixed gauge bosons and there are a few possible
enhancements: SO(10) U (1) E6 and/or SO(8)2 {SO(16), E8 }. The x sector gauge
bosons survive only when
6

(ei |ej ) +
j =1,i=j
6

2

(ei |zk ) = 0 mod 2,
i = 1, . . . , 6,
(4.3)
k=1
(ej |zk ) = 0 mod 2,
k = 1, 2,
(4.4)
j =1
where we have introduced the notation

v
c i = ei(vi |vj ) , (vi |vj ) = 0, 1
vj
(4.5)
and one of the constraints in (4.3), (4.4) can be dropped because is linearly independent
with the rest.
As far as the SO(8)2 is concerned we have the following possibilities:
(i)
(ei |z1 ) = (ba |z1 ) = 0 i = 1, . . . , 6, a = 1, 2,
(4.6)
(ii)
(ei |z2 ) = (ba |z2 ) = 0 i = 1, . . . , 6, a = 1, 2,
(4.7)
(iii)
(ei |z1 + z2 ) = (ba |z1 + z2 ) = 0
(4.8)
i = 1, . . . , 6, a = 1, 2.
Depending on which of the above equations are true the enhancement is

both (i) and (ii) E8 ,
(4.9)
one of (i) or (ii) or (iii) SO(16),
59
(4.10)
none of (i) or (ii) or (iii) SO(8) SO(8).

(4.11)
z1
In the sequel we will restrict to the case
c z2 = 1 as this is the more promising
phenomenologically, we intent to examine c zz12 = +1 in detail in a future publication.
4.3. Observable matter spectrum
The untwisted sector matter is common to all models and consists of six vectorials
of SO(10) and 12 non-Abelian gauge group singlets. In models where the gauge group
enhances to E6 extra matter comes from the x sector giving rise to six E6 fundamental
reps (27).
Chiral twisted matter arise from the following sectors
(1)
Bpqrs
= S + b1 + pe3 + qe4 + re5 + se6 + (x),
(2)
Bpqrs
= S + b2 + pe1 + qe2 + re5 + se6 + (x),
(3)
Bpqrs
= S + b3 + pe1 + qe2 + re3 + se4 + (x),
(4.12)
where b3 = b1 + b2 + x. These are 48 sectors (16 sectors per orbifold plane) and we choose
to label them using the plane number i (upper index) and the integers pi , qi , ri , si = {0, 1}
(lower index) corresponding to the coefficients of the appropriate shift vectors. Note that
for a particular orbifold plane i only four shift vectors can be added to the twist vector bi
(the ones that have non empty intersection) the other two give rise to massive states. Each
of the above sectors (4.12) can produce a single spinorial of SO(10) (or fundamental of E6
in the case of enhancement). Since the E6 model spectrum is in one to one correspondence
with the SO(10) spectrum in the following we use the name spinorial meaning the 16 of
SO(10) and in the case of enhancement the 27 of E6 .
One of the advantages of our formulation is that it allows to extract generic formulas
regarding the number and the chirality of each spinorial. This is important because it allow
an algebraic treatment of the entire class of models without deriving each model explicitly.
The number of surviving spinorials per sector (4.12) is given by

ei
zm
1
(1)
Ppqrs
(4.13)
=
1c
1
c
,
(1)
(1)
16
B
B
pqrs
pqrs
i=1,2
m=1,2

e
zm
1
i
(2)
Ppqrs =
(4.14)
1c
1c
,
(2)
(2)
16
Bpqrs
Bpqrs
i=3,4
m=1,2

ei
zm
1
(3)
=
Ppqrs
(4.15)
1c
1
c
,
(3)
(3)
16
Bpqrs
Bpqrs
i=5,6
m=1,2
i
where Ppqrs
is a projector that takes values {0, 1}. The chirality of the surviving spinorials
is given by

b2 + (1 r)e5 + (1 s)e6
(1)
,
Xpqrs
(4.16)
=c
(1)
Bpqrs
60

b1 + (1 r)e5 + (1 s)e6
,
(2)
Bpqrs

b1 + (1 r)e3 + (1 s)e4
(3)
Xpqrs
=c
,
(3)
Bpqrs
(2)
Xpqrs
=c
(4.17)
(4.18)
i
i
= + corresponds to a 16 of SO(10) (or 27 in the case of E6 ) and Xpqrs
=
where Xpqrs
corresponds to a 16 (or 27) and we have chosen the spacetime chirality C( ) = +1.
The net number of spinorials and thus the net number of families is given by
NF =
3

1

(i)
(i)
Xpqrs
Ppqrs
.
(4.19)
i=1 p,q,r,s=0
Similar formulas can be easily derived for the number of vectorials and the number of
singlets and can be extended to the U (1) charges but in this work we will restrict to the
spinorial calculation.
Formulas (4.13)(4.15) allow us to identify the mechanism of spinorial reduction, or in
other words the fixed point reduction, in the fermionic language. For a particular sector
(i)
(Bpqrs ) of the orbifold plane i there exist two shift vectors (e2i1 , e2i ) and the two zeta
(i)
. Setting the relative projection
vectors (z1 , z2 ) that have no common elements with Bpqrs
coefficients (4.15) to 1 each of the above four vectors acts as a projector that cuts the
number of fixed points in the associated sector by a factor of two. Since four such projectors
are available for each sector the number of fixed points can be reduced from 16 to 1 per
plane.
The projector action (4.13)(4.15) can be expanded and written in a simpler form
(i) W (i) = Y (i) ,
where
(e1 |e3 )
(e |e )
(1) = 2 3
(z1 |e3 )
(z2 |e3 )
(e3 |e1 )
(e |e )
(2) = 4 1
(z1 |e1 )
(z2 |e1 )
(e5 |e1 )
(e |e )
(3) = 6 1
(z1 |e1 )
(z2 |e1 )
and
pi
q
Wi = i .
ri
si
(4.20)
(e1 |e4 )
(e2 |e4 )
(z1 |e4 )
(z2 |e4 )
(e3 |e2 )
(e4 |e2 )
(z1 |e2 )
(z2 |e2 )
(e5 |e2 )
(e6 |e2 )
(z1 |e2 )
(z2 |e2 )
(e1 |e5 )
(e2 |e5 )
(z1 |e5 )
(z2 |e5 )
(e3 |e5 )
(e4 |e5 )
(z1 |e5 )
(z2 |e5 )
(e5 |e3 )
(e6 |e3 )
(z1 |e3 )
(z2 |e3 )
(e1 |e6 )
(e2 |e6 )
,
(z1 |e6 )
(z2 |e6 )
(e3 |e6 )
(e4 |e6 )
,
(z1 |e6 )
(z2 |e6 )
(e5 |e4 )
(e6 |e4 )
,
(z1 |e4 )
(z2 |e4 )
(e1 |b1 )
(e |b )
Y (1) = 2 1 ,
(z1 |b1 )
(z2 |b1 )
(e3 |b2 )
(e |b )
Y (2) = 4 2 ,
(z1 |b2 )
(z2 |b2 )
(e5 |b3 )
(e |b )
Y (3) = 6 3
(z1 |b3 )
(z2 |b3 )
(4.21)
(4.22)
61
They form three systems of equations of the form i W i = Y i (one for each orbifolds
plane). Each system contains 4 unknowns pi , qi , ri , si which correspond to the labels of
surviving spinorials in the plane i. We call the set of solutions of each system i . The net
number of families (4.19) can be written as
3

NF =
(i)
Xpqrs
.
(4.23)
i=1 p,q,r,si
(i)
The chiralities (4.16)(4.18) can be further expanded in the exponential form Xpqrs =
(i)
)
exp(ipqrs
(1)
pqrs
= 1 + (b1 |b2 ) + (1 r)(e5 |b1 ) + (1 s)(e6 |b1 ) + p(e3 |b2 ) + q(e4 |b2 )
+ r(e5 |b2 ) + s(e6 |b2 ) + p(1 r)(e3 |e5 ) + p(1 s)(e3 |e6 )
+ q(1 r)(e4 |e5 ) + q(1 s)(e4 |e6 ) + (r + s)(e5 |e6 ) mod 2,
(2)
pqrs
(4.24)
= 1 + (b1 |b2 ) + (1 r)(e5 |b2 ) + (1 s)(e6 |b2 ) + p(e1 |b1 ) + q(e2 |b1 )
+ r(e5 |b1 ) + s(e6 |b1 ) + p(1 r2 )(e1 |e5 ) + q(1 r)(e2 |e5 )
+ p(1 s)(e1 |e6 ) + q(1 s)(e2 |e6 ) + (r + s)(e5 |e6 ) mod 2,
(3)
pqrs
(4.25)
= 1 + (b1 |b2 ) + (1 p)(e1 |b1 ) + (1 q)(e2 |b1 ) + (e5 + e6 |b1 )

+ (1 r)(e3 |b2 ) + (1 s)(e4 |b2 ) + (1 r)(1 p)(e3 |e1 )
+ (1 r)(1 q)(e3 |e2 ) + (1 r)(e3 |e5 ) + (1 r)(e3 |e6 )
+ (1 s)(e4 |e6 ) + (1 r)(e3 |z1 + z2 ) + (1 s)(e4 |z1 + z2 )
+ (b1 |z1 + z2 ) mod 2.
We remark here that the projection coefficient c

that our equations depend only on
(ei |ej ),
(ei |bA ),
(ei |zn ),
i = 1, . . . , 6, A = 1, 2, n = 1, 2.
b1
b2
(4.26)
simply fixes the overall chirality and
(zn |bA ),
(4.27)
However, the following six parameters do not appear in the expressions (e1 |e2 ), (e3 |e4 ),
(e3 |b1 ), (e4 |b1 ), (e1 |b2 ), (e2 |b2 ) and thus a generic model depends on 37 discrete parameters.
5. Results
5.1. Models
We apply here the formalism developed above in order to derive sample models in the
free fermionic formulation.
62
5.1.1. The Z2 Z2 symmetric orbifold

The simplest example is the symmetric Z2 Z2 orbifold. Here we
all the free GSO
v set
phases (4.27) to zero. The full GSO phase matrix takes the form (c vji = exp[i(vi |vj )])
1
S
e1
e2
e3
e
(vi |vj ) = 4
e5
e6
b1
b2
z1
z2
1 S
1
1
1
1
1
1
1
1
1
1
0
0
1
1
e1
e2
e3
e4
e5
e6
b1
b2
z1
z2
1
1
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
0
0
0
0
1
1
0
0
0
0
0
0
1
1
0
0
1
1
0
0
0
0
0
0
1
1
0
0
1
1
0
0
0
0
0
0
0
0
1
1
1
1
0
0
0
0
0
0
0
0
1
1
With the above choice (i) = W (i) = 0 in Eq. (4.20). All projectors become inactive and
thus the number of surviving twisted sector spinorials takes its maximum value which is
48 with all chiralities positive according to (4.24)(4.26). Moreover three spinorials and
three antispinorials arise from the untwisted sector. Following (4.3), (4.4) the gauge group
enhances to E6 U (1)2 E8 and the spinorials of SO(10) combine with vectorials and
singlets to produce 48 + 3 = 51 families (27) and 3 antifamilies (27) of E6 .
5.1.2. A three generation E6 model
We can obtain a three family E6 model by choosing the following set of projection
coefficients
1
S
e1
e2
e3
e
(vi |vj ) = 4
e5
e6
b1
b2
z1
z2
1 S
1
1
1
1
1
1
1
1
1
1
0
0
1
1
e1
e2
e3
e4
e5
e6
b1
b2
z1
z2
1
1
0
0
1
0
0
1
0
0
0
0
1
1
0
0
0
0
1
0
0
0
0
1
1
1
1
0
0
0
1
0
0
0
0
0
1
1
0
0
0
0
1
0
0
0
1
0
1
1
0
1
1
1
0
1
0
0
0
0
1
1
1
0
0
0
1
0
0
0
1
1
1
1
0
0
0
0
0
0
1
0
0
0
1
1
0
0
0
0
0
0
0
1
0
0
1
1
0
0
0
1
0
1
0
0
1
1
1
1
0
1
0
0
0
1
0
0
1
1
63
The full gauge group is here E6 U (1)2 SO(8)2 . Three families (27), one from each
plane, arise from the sectors S + bi + (x), i = 1, 2, 3. Another set of three families and
three antifamilies arise from the untwisted sector. The hidden sector consists of nine 8-plets
under each SO(8). In addition there exist a number of non-Abelian gauge group singlets.
The model could be phenomenologically acceptable provided one finds a way of breaking
E6 . Since it is not possible to generate the E6 adjoint (not in KacMoody level one), we
need to realize the breaking by an additional Wilson-line like vector. However, a detailed
investigation of acceptable basis vectors, shows that the E6 breaking is accompanied by
truncation of the fermion families. Thus this kind of perturbative E6 breaking is not
compatible with the presence of three generations. It would be interesting to utilize string
dualities in order to study the non-perturbative aspects of such models.
5.1.3. A six generation E6 model
Similarly a six family E6 U (1)2 E8 model can be obtained using the following
projection coefficients
1
S
e1
e2
e3
e
(vi |vj ) = 4
e5
e6
b1
b2
z1
z2
1 S
1
1
1
1
1
1
1
1
1
1
0
0
1
1
e1
e2
e3
e4
e5
e6
b1
b2
z1
z2
1
1
0
0
0
0
1
1
0
0
0
0
1
1
0
0
1
0
0
1
0
0
0
0
1
1
0
1
0
0
0
1
0
0
0
0
1
1
0
0
0
0
1
0
0
0
0
1
1
1
1
0
0
1
0
0
0
0
0
0
1
1
1
1
1
0
0
0
0
0
0
1
1
1
0
0
0
0
0
0
1
0
0
0
1
1
0
0
0
0
0
0
0
1
0
0
1
1
0
0
0
0
0
0
0
0
1
1
1
1
0
0
0
1
0
1
0
0
1
1
In this model we have six families from the twisted sector, two from each plane together
with three families and three antifamilies from the untwisted sector, accompanied by a
number of singlets and 8-plets of both hidden SO(8)s.
5.2. N = 4 liftable vacua
In the models considered above we have managed to separate the orbifold twist
action (represented here by b1 , b2 ) from the shifts (represented by ei ) and the Wilson
lines (z1 , z2 ).
However,
these actions are further correlated through the GSO projection
v
coefficients c vji . Nevertheless, we remark that the twist action can be decoupled from
the other two in the case

b
b
c n = c m = +1, i = 1, . . . , 6, k = 1, 2, m, n = 1, 2, 3.
(5.1)
zk
ei
64
The above relation defines a subclass of N = 1 four-dimensional vacua with interesting

phenomenological properties and includes three generation models. Due to the decoupling
of the orbifold twist action these vacua are direct descendants of N = 4 vacua so we
will refer to these models as N = 4 liftable models. In this subclass of models some
important phenomenological properties of the vacuum, as the number of generations, are
predetermined at the N = 4 level as it is related to the (ei |ej ) and (zi |ej ) phases. The
orbifold action reduces the supersymmetries and the gauge group and makes chirality
apparent, however the number of generations is selected by the N = 4 vacuum structure.
At the N = 1 level this is understood as follows: the Z2 Z2 orbifold has 48 fix points.
Switching on some of the above phases correspond to a free action that removes some
of the fixed points and thus reduces the number of spinorials. Moreover, in this case, the
chirality of the surviving spinorials is again related as seen by (4.24)(4.26) to the (ei |ej )
and (ei |zk ) coefficients, which are all fixed at the N = 4 level. The observable gauge group
of liftable models is always E6 and this can be easily seen by applying (5.1) to (4.3), (4.4).
Typical examples of such vacua are the three and six generation E6 U (1)2 SO(8)2
models presented in Section 5.1. A careful counting, taking into account some symmetries
among the coefficients, shows that this class of models consists of 220 models, or 221
if we include (b1 |b2 ). These vacua are interesting because they can admit a geometrical
interpretation.
From the orbifold description we learn that all breakings of the hidden and observable
gauge group are induced using Wilson lines. From the 4D point of view the internal gauge
group is broken in a similar fashion using Wilson lines. The twisted planes in Eq. (3.36)
describe the removal of the free moduli using twists. When a group is broken using Wilson
lines the field corresponding to this Wilson line obtains a non-zero VEV. The fixing of the
moduli using twists can be interpreted as the removal of the quantum fluctuations of the
fields identified with the Wilson lines. These Wilson lines become discrete Wilson lines
and the VEV becomes a fixed value.
5.3. Classification
As we discussed above, the free GSO phases of the N = 1 partition function control
the number of chiral generations in a given model. In Section 3 we have given analytic
formulas that enable the calculation of the number of generations for any given set of
phases. To gain an insight into the structure of this class of vacua we can proceed with
a computer evaluation of these formulas and thus classify the space of these vacua with
respect to the number of generations. This also allows a detailed examination of the
structure of these vacua and in particular how the generations are distributed among the
three orbifold planes. The main obstacle to this approach is the huge number of vacua
under consideration. As a first step in this direction we restrict ourselves to the class of
liftable vacua that is physically appealing and contains representative models with the
right number of generations. As stated above this class consists in principle of 221 models
and their complete classification takes a few minutes on a personal computer using an
appropriate computer program. The program analyses all different options for the free
GSO coefficients. The different configurations are then used to calculate the number of
generations using formulae (4.13)(4.19). For the analysis of the gauge group we use
65
Table 5
Inequivalent realistic liftable models with a E6 U (1)2 SO(8) SO(8) gauge group. The chiral content of
each model is listed per plane and numbered, + lists all the positive chiral states per plane while lists all
the negative states per plane. The total sum of all the planes is then listed and subsequently the net total number
of chiral states. The list is ordered by the total net number of chiral states
No.
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
30
Total
Net
16
8
8
8
8
12
8
6
4
8
4
4
6
4
4
4
4
6
6
10
6
3
3
2
4
4
4
6
3
2
0
0
0
0
0
4
0
2
0
0
0
0
2
4
0
0
0
2
2
6
2
1
1
0
0
0
0
2
1
0
8
8
8
6
4
4
8
4
4
2
4
4
4
4
4
3
2
3
2
2
4
3
2
2
2
2
1
1
3
2
0
0
0
2
0
0
0
0
0
0
0
0
0
0
0
1
0
1
0
0
0
1
0
0
2
0
0
0
1
0
8
8
4
4
4
4
4
4
4
2
2
3
2
4
2
2
2
2
2
2
2
2
2
2
2
1
1
1
1
1
0
0
0
0
0
0
4
0
0
0
0
1
0
0
2
0
0
0
0
0
2
0
0
0
0
1
0
0
0
0
32
24
20
18
16
20
20
14
12
12
10
11
12
12
10
9
8
11
10
14
12
8
7
6
8
7
6
8
7
5
0
0
0
2
0
4
4
2
0
0
0
1
2
4
2
1
0
3
2
6
4
2
1
0
2
1
0
2
2
0
32
24
20
16
16
16
16
12
12
12
10
10
10
8
8
8
8
8
8
8
8
6
6
6
6
6
6
6
5
5
No.
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
Total
Net
3
3
2
4
4
3
2
2
3
1
2
3
1
2
2
2
1
2
4
1
1
1
2
2
2
4
4
1
2
1
4
1
1
2
4
0
1
0
0
1
1
0
1
0
0
0
2
1
2
4
1
1
1
2
0
2
4
4
1
2
3
4
2
2
2
2
2
2
2
1
1
3
1
1
1
1
2
2
1
1
1
1
1
0
2
2
1
2
1
1
2
1
4
0
0
0
0
2
0
0
0
0
1
0
1
0
1
0
0
0
0
0
1
0
1
2
2
1
2
1
1
2
0
4
1
2
2
2
2
1
1
1
1
3
1
1
1
1
1
1
1
1
1
3
1
3
2
1
1
2
1
1
1
1
4
0
2
0
0
2
1
1
0
0
1
1
0
0
1
3
1
0
0
0
1
1
1
2
3
1
2
1
1
1
0
4
6 1
7 3
6 2
8 4
8 4
6 2
5 1
4 0
5 1
7 3
4 1
5 2
3 0
4 2
5 3
5 3
3 1
4 2
6 4
5 3
3 2
4 3
6 6
5 5
4 4
8 8
6 6
3 3
5 5
3 3
12 12
5
4
4
4
4
4
4
4
4
4
3
3
3
2
2
2
2
2
2
2
1
1
0
0
0
0
0
0
0
0
0
formulae (4.3)(4.8). The results are presented in Tables 57. In these tables we list the
number of generations coming from the twisted sectors. They are listed per plane. The
number of positive chiral generations is separated from the number of negative chiral
generations on each plane. The total number is then listed before listing the total net number
of generations. As the sign of the chirality is determined by the coefficient (b1 |b2 ) (see
(4.24)(4.26)) we have included models that have a positive net number of generations. In
order to maintain a complete separation of the hidden gauge group we have set (z1 |z2 ) = 1.
The tables are ordered by the total net number of chiral states.
We find that there are no liftable models with a SO(10) observable gauge group, which
is always extended to E6 , and the states from the vector x are not projected out. Since
66
Table 6
Inequivalent realistic liftable models with a E6 U (1)2 SO(16) gauge group. The chiral content of each model
is listed per plane and numbered, + lists all the positive chiral states per plane while lists all the negative
states per plane. The total sum of all the planes is then listed and subsequently the net total number of chiral
states. The list is ordered by the total net number of chiral states
No.
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
30
31
32
33
34
35
Total
Net
16
8
8
8
12
8
6
4
4
4
4
4
6
6
10
6
3
3
2
3
2
4
4
3
2
1
2
1
2
2
2
4
4
1
4
0
0
0
0
4
0
2
0
4
0
0
0
2
2
6
2
1
1
0
1
2
4
0
1
0
1
0
1
2
0
2
4
4
1
4
8
8
6
4
4
8
4
4
4
4
3
2
3
2
2
4
3
2
2
2
2
2
2
2
2
3
1
1
2
2
1
2
1
1
4
0
0
2
0
0
0
0
0
0
0
1
0
1
0
0
0
1
0
0
0
0
0
2
0
0
1
1
1
2
2
1
2
1
1
4
8
8
4
4
4
4
4
4
4
2
2
2
2
2
2
2
2
2
2
2
2
2
2
1
1
3
1
3
2
1
1
2
1
1
4
0
0
0
0
0
4
0
0
0
2
0
0
0
0
0
2
0
0
0
2
0
0
2
1
1
1
1
1
2
3
1
2
1
1
4
32
24
18
16
20
20
14
12
12
10
9
8
11
10
14
12
8
7
6
7
6
8
8
6
5
7
4
5
6
5
4
8
6
3
12
0
0
2
0
4
4
2
0
4
2
1
0
3
2
6
4
2
1
0
3
2
4
4
2
1
3
2
3
6
5
4
8
6
3
12
32
24
16
16
16
16
12
12
8
8
8
8
8
8
8
8
6
6
6
4
4
4
4
4
4
4
2
2
0
0
0
0
0
0
0
the models admit a geometrical interpretation, it means that they must descend from the
ten-dimensional E8 E8 heterotic-string on Z2 Z2 CalabiYau threefold.
In 3% of all the models the hidden gauge group is enhanced to SO(8) SO(8)
SO(16). We find that in total 1024 liftable models are enhanced to SO(8) SO(8) E8 .
We find that the 24 generations NAHE model as explained in Section 2 is present in Table 5.
The problem of a detailed investigation of the full class of vacua will be considered further
in a future publication.
67
Table 7
Inequivalent realistic liftable models with a E6 U (1)2 E8 gauge group. The chiral content of each model is
listed per plane and numbered, + lists all the positive chiral states per plane while lists all the negative states
per plane. The total sum of all the planes is then listed and subsequently the net total number of chiral states. The
list is ordered by the total net number of chiral states
No.
1
2
3
4
5
6
7
8
9
10
11
Total
Net
16
12
8
10
6
6
4
3
4
4
2
0
4
0
6
2
2
0
1
4
4
2
16
8
8
4
6
4
4
3
2
4
2
0
0
0
0
2
0
0
1
2
4
2
16
8
8
4
4
4
4
3
2
4
2
0
0
0
0
0
0
0
1
2
4
2
48
28
24
18
16
14
12
9
8
12
6
0
4
0
6
4
2
0
3
8
12
6
48
24
24
12
12
12
12
6
0
0
0
5.4. General properties

In Section 3 we discussed a direct translation between the bosonic formulation and
the fermionic formulation of the heterotic string compactifications. Z2 Z2 orbifold
compactifications are relevant for our class of models. These orbifolds contain three twisted
sectors, or three twisted planes. A priori we may have the possibility that all three twisted
planes produce spinorial SO(10) representations. We refer to this subclass of models
as S 3 models. The alternatives are models in which spinorial representations may be
obtained from only two, one, or none, twisted planes, and the others produce vectorial
representations. We refer to these cases as S 2 V , SV 2 and V 3 models, respectively. The
focus of the analysis in this paper is on the S 3 subclass of models, which also contains the
NAHE-based three generation models. The S 3 subclass allows, depending on the one-loop
GSO projection coefficients, the possibility of spinorials on each plane. In specific models
in this subclass each Standard Model family is obtained from a distinct orbifold plane.
Such models therefore produce three generation models and may be phenomenologically
interesting. The only other phenomenologically viable option can come from the subclass
S 2 V models as this class of models may contain a model with for example 2 generations
coming from the first plane and 1 generation coming form the second plane and none from
the third. The SV 2 class of models cannot produce a physical model because it is not
possible to reduce the number of families to 3 as they would have to be coming from one
plane and 3 cannot be written as a power of 2. Similarly the V 3 subclass of models will
not contain phenomenologically interesting models.
5.4.1. 3 generations realized only with twisted and shifted real manifolds
Since the projectors are constructed using the complete separation of the internal
manifold we see that three generation models are only possible when
3
6
1,1
.
6,6 = 2,2
(5.2)
68
These 1,1 internal parts do not describe a complex manifold. They describe internal
3 , and using only
real circles. If we use solely complex manifolds, of the type 6,6 = 2,2
symmetric shifts, we find that there are no 3 generation models. We therefore conclude
that the net number of generations can never be equal to three in the framework of Z2 Z2
CalabiYau compactification. This implies the necessity of non-zero torsion in CY Z2 Z2
compactifications in order to obtain semi-realistic three generation models.
In the realistic free fermionic models the reduction of the number of families together
with the breaking of the observable SO(10) is realized by isolating full multiplets at two
fixed points on the internal manifold. In reducing the number of families down to one,
different component of each family are attached to the two distinct fixed points. We remove
one full multiplet and simultaneously break the SO(10) symmetry. We therefore keep a full
multiplet on each twisted plane. In the SO(10) models described here a whole 16 or 16 of
SO(10) is attached to a fixed point. We are therefore not able to break the SO(10), and
simultaneously preserve the full Standard Model multiplets. For this reason we find that
the observable SO(10) cannot be broken perturbatively in this class of three generation
models, and may only be broken non-perturbatively. It is therefore not possible to reduce
both the number of families down to 3 and break the observable gauge group SO(10) down
to its subgroups perturbatively.
We conclude that there is a method to reduce the number of generations from 48 to 3.
Since we need 4 projectors we need to separate the hidden gauge group using SO(8)
characters
0,8 0,4 0,4
(5.3)
and we need to break the internal complex manifold to an internal real manifold
6,6 [1,11,1 ]3 .
(5.4)
If we reduce the number of generations to 3 we cannot break the SO(10) observable group
to its subgroups, while maintaining a full multiplet. The SO(10) observable gauge group
cannot therefore be broken perturbatively. We can reduce the number of generations from
48 to 6 using 3 projectors. This entails that we can choose either to separate the hidden
gauge group using SO(16) characters, or to leave the internal manifold complex.
We argued above that we cannot break SO(10) down to a subgroup perturbatively,
while reducing the number of generations to 3. If we want to break the SO(10) symmetry
perturbatively, and keep a full SO(10) multiplet from a given twisted plane, we can only
reduce the number of generations to 6. This can be achieved if we define three different
projectors like the ones defined in Eqs. (4.13)(4.15). We are therefore left with two
options.
We can use SO(16) characters for the separation of the hidden gauge group. We have
then constructed only one projector which leaves us no other option than to break the
complex structure using symmetric shifts
6
.
6,6 1,1
(5.5)
We can use SO(8) characters for the separation of the hidden gauge group. In doing so
we have constructed two projectors. The third can be realized by the symmetric shifts
69
that leave the complex structure of the internal manifold intact

3
6,6 2,2
.
(5.6)
6. Discussion and conclusions

String theory duly attracts wide interest. It provides a consistent approach to perturbative quantum gravity, while at the same time incorporating the gauge and matter structures
that are relevant for particle physics phenomenology. However, the multitude of vacua that
the theory admits and the lack of a dynamical principle to choose among them, hinders
the prospects that the theory will yield unique experimental predictions. This has led some
authors to advocate the anthropic principle as a possible resolution for understanding the
contrived set of parameters that seem to govern our world.
The approach pursued in our work is different. In our view the understanding of the
dynamical principles that underly quantum gravity and the vacuum selection must await
the better conceptual understanding of the quantum gravity synthesis. It may well be that at
the end of the day the probabilistic nature of quantum mechanics will emerge as a derived
property rather than a fundamental property of quantum gravity. In this respect we should
regard the string theories as merely providing a perturbative glimpse into the underlying
properties of quantum gravity, and how it may relate to the gauge and matter observables.
In this context we must utilize both the low energy data as well as the basic properties of
string theory to isolate promising string vacua and develop the tools to discern between the
experimental predictions of different classes. An example, is the SO(10) embedding of the
Standard Model spectrum, which is viable in the heterotic limit of M-theory, but not in its
type I limit.
Given the Standard Model properties, we may hypothesize that the true string vacuum
should accommodate two pivotal ingredients. One is the existence of three generations and
the second is their embedding in an underlying SO(10) or E6 grand unified gauge group. In
this context, the replication of the matter generations is the first particle physics observable
whose origin may be sought in string theory. This follows from the fact that the flavor
sector of the Standard Model does not arise from any physical principle, like the gauge
principle, as well as from the fact that in certain classes of string compactifications the
number of generations is related to a topological number of compact manifolds, the Euler
characteristic.
A class of string compactifications that admit three generations, as well as their
embedding in an underlying SO(10) group structure are the NAHE-based free fermionic
models. A subset of the boundary condition basis vectors that span these models can be
seen to correspond to Z2 Z2 orbifold compactifications at special points in the moduli
space. However, the geometrical understanding of the full three generation models is still
lacking. The aim of the current work is therefore to advance the geometrical understanding
of the NAHE-based free fermionic models. In this paper we showed that two of the
boundary condition basis vectors beyond those that correspond to the Z2 Z2 orbifold
correspond to symmetric shifts on the compact tori, whereas the third correspond to an
asymmetric shift. We then proceeded to classify all possible symmetric shifts on complex
70
tori and demonstrated that three generation models do not arise in this manner. Three
generation models that realize the Z2 Z2 orbifold picture of the three chiral generations
were found. In these cases the SO(10) gauge group cannot be broken perturbatively, while
preserving the full Standard Model matter content. Additionally in these cases the internal
6 , i.e., to a product of six circles. In this class of models each of the
lattice is broken to 1,1
chiral generations is attached to a single fixed point in each of the twisted orbifold planes.
This should be contrasted with the case of the three generation free fermionic models in
which the SO(10) symmetry is broken perturbatively by Wilson lines. In those cases, each
generation is obtained from a separate orbifold plane, but different components of each
generation are attached to different fixed points of the corresponding twisted sector.
Additionally, we demonstrated in this paper that for a wide range of models, for which
a geometrical origin is understood, there exist an interpretation of the phases that appear in
the N = 1 partition function, in terms of vacuum expectation values of background fields
of the N = 4 vacua. In these cases the dynamical components of the background fields
are projected out, but their vacuum expectation value is retained and takes the form of the
free GSO phases of the N = 1 partition function. These phases also control the chirality
of the models. Thus, we have the situation in which the chirality of the models is already
determined by the VEVs of the background fields of the N = 4 vacuum. In effect, the chiral
content of the N = 1 vacua in these cases is determined by the Narain N = 4 lattice. An
example of this phenomenon was already seen in the case of Z2 Z2 on SO(12) lattice
that yields 24 generations versus the Z2 Z2 orbifold on SO(4)3 lattice that yields 48
generations. The interpretation of the chiral content of the N = 1 models in terms of the
N = 4 vacua will be especially instrumental when seeking the strong coupling duals of the
N = 1 models, due to the fact that the N = 4 duals can be obtained with relative ease. The
understanding of the N = 1 duals will then entail the understanding of the corresponding
Z2 Z2 operation on the dual side.
We discovered in this paper that the three generation free fermionic models necessarily
employ an asymmetric shift on the internal compactified space. This observation has
profound implications. In the first place, since the asymmetric shift can act only at
enhanced symmetry points in the moduli space, it implies that some moduli are fixed and
frozen. In fact in some cases it is seen that all the geometrical moduli are projected out.
In those cases the geometrical moduli may be interchanged with twisted moduli which
are much more difficult to identify, and hence their moduli spaces are more intricate.
Additionally, the necessity of incorporating an asymmetric shift has important implications
in the context of non-perturbative dualities. In the case of the duals of the heterotic
models, a geometric moduli is interchanged with the dilaton. Hence, the fact that the
geometric moduli are fixed around their self-dual value on the heterotic side implies
that on the dual side the dilaton has to be fixed around its self-dual point. This is a
fascinating possibility that we will return to in future work. However, we note that the
low energy phenomenological data may point in the direction of esoteric compactifications
that would have otherwise been overlooked. The results show that, in the framework of
Z2 Z2 CalabiYau compactification, the net number of generation can never be equal to
three. This implies the necessity of non-zero torsion in CY compactifications in order to
obtain semi-realistic three generation models. Additionally, the necessity to incorporate an
asymmetric shift in the reduction to three generations, has profound implications for the
71
issues of moduli stabilization and vacuum selection. The reason being that it can only
be implemented at enhanced symmetry points in the moduli space. In this context we
envision that the self-dual point under T-duality plays a special role. In the context of nonperturbative dualities the dilaton and moduli are interchanged, with potentially important
implications for the problem of dilaton stabilization. We will report on these aspects in
future publications.
Acknowledgements
This work was supported in part by the Pieter Langerhuizen Lambertuszoon Fund of the
Royal Holland Society of Sciences and Humanities, the Noorthey Academy and the VSB
foundation (S.N.); by the European Union under the contracts HPRN-CT-2000-00122,
HPRN-CT-2000-00131, HPRN-CT-2000-00148, HPRN-CT-2000-00152 and HPMF-CT2002-01898; by the PPARC and by the Royal Society.
References
[1] L. Susskind, hep-th/0302219;
M.R. Douglas, JHEP 0305 (2003) 046;
T. Banks, M. Dine, E. Gorbatov, hep-th/0309170;
J.F. Donoghue, hep-th/0310203.
[2] P. Candelas, G.T. Horowitz, A. Strominger, E. Witten, Nucl. Phys. B 258 (1985) 46.
[3] I. Antoniadis, J. Ellis, J. Hagelin, D.V. Nanopoulos, Phys. Lett. B 231 (1989) 65.
[4] A.E. Faraggi, D.V. Nanopoulos, K. Yuan, Nucl. Phys. B 335 (1990) 347.
[5] I. Antoniadis, G.K. Leontaris, J. Rizos, Phys. Lett. B 245 (1990) 161.
[6] A.E. Faraggi, Phys. Lett. B 278 (1992) 131;
A.E. Faraggi, Phys. Lett. B 274 (1992) 47;
A.E. Faraggi, Nucl. Phys. B 387 (1992) 239;
A.E. Faraggi, Nucl. Phys. B 403 (1993) 101.
[7] S. Chaudhuri, G. Hockney, J.D. Lykken, Nucl. Phys. B 469 (1996) 357;
G. Cleaver, M. Cvetic, J.R. Espinosa, L. Everett, P. Langacker, Nucl. Phys. B 545 (1999) 47.
[8] G.K. Leontaris, Phys. Lett. B 372 (1996) 212;
G.K. Leontaris, J. Rizos, Nucl. Phys. B 554 (1999) 3.
[9] I. Antoniadis, C. Bachas, C. Kounnas, Nucl. Phys. B 289 (1987) 87;
H. Kawai, D.C. Lewellen, S.H.-H. Tye, Nucl. Phys. B 288 (1987) 1;
I. Antoniadis, C. Bachas, Nucl. Phys. B 298 (1988) 586.
[10] For reviews and references, see, e.g., J. Lykken, hep-ph/9511456;
J.L. Lopez, hep-ph/9601208;
A.E. Faraggi, hep-ph/9707311;
A.E. Faraggi, hep-th/0208125;
A.E. Faraggi, hep-th/0307037.
[11] G.B. Cleaver, A.E. Faraggi, D.V. Nanopoulos, Phys. Lett. B 455 (1999) 135;
G.B. Cleaver, A.E. Faraggi, D.V. Nanopoulos, J.W. Walker, Nucl. Phys. B 593 (2001) 471.
[12] A.E. Faraggi, Phys. Lett. B 326 (1994) 62;
J. Ellis, A.E. Faraggi, D.V. Nanopoulos, Phys. Lett. B 419 (1998) 123.
[13] P. Berglund, J. Ellis, A.E. Faraggi, D.V. Nanopoulos, Z. Qiu, Phys. Lett. B 433 (1998) 269;
P. Berglund, J. Ellis, A.E. Faraggi, D.V. Nanopoulos, Z. Qiu, Int. J. Mod. Phys. A 15 (2000) 1345.
[14] A.E. Faraggi, Phys. Lett. B 544 (2002) 207.
72
[15] A.E. Faraggi, D.V. Nanopoulos, Phys. Rev. D 48 (1993) 3288;

A.E. Faraggi, Int. J. Mod. Phys. A 14 (1999) 1663.
[16] G.B. Cleaver, A.E. Faraggi, C. Savage, Phys. Rev. D 63 (2001) 066001;
G.B. Cleaver, D.J. Clements, A.E. Faraggi, Phys. Rev. D 65 (2002) 106003.
[17] I. Antoniadis, C. Bachas, C. Kounnas, P. Windey, Phys. Lett. B 171 (1986) 51.
[18] S. Ferrara, L. Girardello, C. Kounnas, M. Porrati, Phys. Lett. B 192 (1987) 368;
S. Ferrara, L. Girardello, C. Kounnas, M. Porrati, Phys. Lett. B 194 (1987) 356;
S. Ferrara, C. Kounnas, M. Porrati, F. Zwirner, Phys. Lett. B 194 (1987) 366;
S. Ferrara, C. Kounnas, M. Porrati, F. Zwirner, Nucl. Phys. B 318 (1989) 75;
E. Kiritsis, C. Kounnas, P.M. Petropoulos, J. Rizos, Nucl. Phys. B 483 (1997) 141;
E. Kiritsis, C. Kounnas, P.M. Petropoulos, J. Rizos, Nucl. Phys. B 540 (1999) 87.
[19] K. Narain, Phys. Lett. B 169 (1986) 41;
K.S. Narain, M.H. Sarmadi, E. Witten, Nucl. Phys. B 279 (1987) 369.
[20] G.B. Cleaver, A.E. Faraggi, S.E.M. Nooij, Nucl. Phys. B 672 (2003) 64.
[21] A. Gregori, C. Kounnas, J. Rizos, Nucl. Phys. B 549 (1999) 16.
[22] A. Gregori, C. Kounnas, P.M. Petropoulos, Nucl. Phys. B 553 (1999) 108.
[23] E. Kiritsis, C. Kounnas, Nucl. Phys. B 503 (1997) 117, hep-th/9703059.

Non-linear YangMills instantons from strings are

-stable D-branes
H. Enger, C.A. Ltken
Theory Group, Department of Physics, University of Oslo,
P.O. Box 1048 Blindern, NO-0316 Oslo, Norway
Received 15 April 2004; received in revised form 1 June 2004; accepted 25 June 2004
Abstract
We show that B-type -stable D-branes do not in general reduce to the (Gieseker-) stable holomorphic vector bundles used in mathematics to construct moduli spaces. We show that solutions of
the almost Hermitian YangMills equations for the non-linear deformations of YangMills instantons
that appear in the low-energy geometric limit of strings exist iff they are -stable, a geometric large
volume version of -stability. This shows that -stability is the correct physical stability concept.
We speculate that this string-canonical choice of stable objects, which is encoded in and derived from
the central charge of the string-algebra, should find applications to algebraic geometry where there
is no canonical choice of stable geometrical objects.
PACS: 02.40.-k; 11.15.-q; 11.25.-w
Keywords: D-branes; Pi-stability; Gauge theory; Instantons
1. Introduction and summary

Stability-ideas play a central role in both physics and mathematics, but it is far from
obvious what relationship, if any, there is between these apparently quite distinct concepts.
E-mail addresses: hakon.enger@fys.uio.no (H. Enger), lutken@fys.uio.no (C.A. Ltken).

doi:10.1016/j.nuclphysb.2004.06.051
74
H. Enger, C.A. Ltken / Nuclear Physics B 695 (2004) 7383
Mathematicians want to classify bundles and sheaves, and find useful compactifications
of moduli spaces that can assist them in this and other tasks. Their starting point is the observation that all bundles can be derived from the small subset of stable bundles. A bundle
over a curve is called Mumford-(-)stable if it satisfies a certain topological constraint. The
condition is that the ratio of the degree to the rank of every sub-bundle E is smaller than
this ratio for the bundle E itself:
deg(E)
.
E < E :=
(1)
rk(E)
The ratio E is called the slope of the bundle. Equivalently, if we to each bundle associate
a complex number
Z(E) := i rk(E) deg(E) = i rk(E) c1 (E)
(2)
this is a condition on the phase of the sub-bundles:

E < E := Arg Z(E) = Im Log Z(E) = Arc cot
deg(E)
.
rk(E)
(3)
The classification problem is now reduced to finding all the stable bundles, whose moduli space1 is compactified by gluing in the marginally stable cases, which are called
strictly semi-stable.
This is somewhat reminiscent of the physical problem of identifying irreducible constituents from which all modes can be built, but it is not clear how conservation laws and
field equations transmute into these topological conditions, and in fact it is not even clear
from mathematics which topological stability condition is appropriate.
In the presence of sufficient supersymmetry the bridge between the mathematical and
physical stabilities is provided by the central extension of the superalgebra, called the
central- or BPS-charge (Z) in physics. In the simplest low-energy, large volume limit of
string where conventional geometry is recovered, some D-branes E should reduce to holomorphic bundles E, and indeed one finds that Z(E) Z(E) in this limit in the case where
the target space is an elliptic curve. In this simple case -stability can be immediately identified as conservation of energy and charge of the brane.
While the topological interpretation of Z cannot be maintained in general, the BPScharge is a fully stringy concept, and it is not unreasonable to expect that the condition on
the phase of this complex charge (i.e., Arg Z Arg Z) can be retained as the fundamental
stability criterion [2]. This so-called -stability encodes in a simple and physically transparent manner the physics of charge and energy conservation, and therefore suggests that
this is a fundamental concept delivered to us by string theory.
It is encouraging that -stability reduces to -stability in the one-dimensional case,
but it is not clear that this is the case in general. In one (complex) dimension -(semi-)
stability appears to be the relevant property for classifying holomorphic bundles, but in
higher dimensions it seems that a more refined concept of semi-stability is needed. While
there does not appear to be a canonical way to choose this refinement, it is an empirical
observation that so-called Gieseker- ( -)stability (defined and discussed below) provides
1 Some features of moduli spaces of -stable bundles were reviewed by Sharpe [1].
75
less singular compactifications and is thus more useful for bundles on higher-dimensional
manifolds [3]. It is natural to conjecture that this is the appropriate concept that will emerge
in the low-energy geometric limit of strings.
More precisely, in the limit where string solutions like D-branes are well approximated
by gauge (vector-) bundles supported on cycles of CalabiYau manifolds, one would expect (marginally) -stable D-branes to emerge as -(semi-)stable bundles. While this
appears to be true in two dimensions (and therefore automatically in one dimension where
-stability reduces to -stability), we shall see in Section 3 that this is not case for the
physically relevant case of three-dimensional CY-manifolds.
-stability is believed to be the physically correct condition in the full quantum theory, and we investigate its consequences in the low-energy (large volume) limit. While
-stability works on the triangulated derived category of sheaves, the geometric large volume version of -stability (here called -stability) is applicable directly to bundles and
sheaves. We show that - or -stability is not enough to capture the spectrum of stable
D6-branes in the large volume limit where the branes may be represented by bundles.
The purpose of this investigation two-fold: to establish beyond reasonable doubt that stability is the physically correct stability concept for CY-filling bundles, and to discover
to what it corresponds in the low-energy algebrao-geometric limit. Our strategy is to solve
the second problem first by arguing in Section 4 for a specific form of the non-linear deformations of the self-dual Hermitian YangMills (YM) equations from strings, and then use
this to prove that -stability is the correct physical concept by showing that the solutions of
these equations, the non-linear YM-instantons, exist in the limit of a large enough volume
for the CY iff they are -stable.
This result on -stability is analogous to the DonaldsonUhlenbeckYau (DUY) theorem relating solutions of the HermitianYM (HermitianEinstein) equations to -stability,
and it reduces to this (as it must) when the string-induced non-linear deformations are neglected. Our approach is similar to Leungs construction [4] of non-linear YM equations,
and we follow his conventions. Leung calls these equations the almost HermitianEinstein
equations, whose solutions by construction are -stable bundles. The deformed YM equations we consider, on the other hand, are canonically given to us by physics (strings), and
they turn out to be different from Leungs deformations. Stringy instantons will therefore
not in general be -stable, and our task is to use these equations to construct the physically
correct topological stability criterion. This it turns out is precisely the geometrical limit of
-stability, thus closing the circle of ideas and providing a convincing consistency check
on these.
2. Mathematical (- and -) stability

All stability criteria degenerate to Mumford- (-)stability when the base manifold is a
curve (one dimension). By the DUY theorem [5,6] a vector bundle over an n-dimensional
Khler space is -stable,2 which means that all sub-sheaves E E have E < E , iff
2 The general n-dimensional -stability is known as MumfordTakemoto-stability.
76
the HermitianYM equation

n1 F =
E
I
(n 1)!
(4)
has a unique solution, where is the Khler form and I is an identity matrix in colour
space.
A vector bundle is -semistable if all sub-sheaves E E have E E . This definition includes the stable bundles among the semistable. A bundle which is semistable but
not stable is called strictly semistable.
In higher dimensions Gieseker ( )-stability is the commonly used topological stability
condition because it is better behaved. The -stability condition on a bundle E over a
Khler manifold X is phrased in terms the normalised Hilbert polynomial with respect
to the Khler form [3], which may be calculated using the HirzebruchRiemannRoch
theorem (see, e.g., [7]):

1
E (t) =
(5)
et ch(E) Td(X).
rk E
The Chern character is defined by ch(E) = tr eF , where F is the field strength. A vector
bundle is -stable if all sub-sheaves E E have E (t) < E (t) when t . Using
Eq. (5) one may show that -stability can be expressed in terms of certain topological
(k)
invariants E which we will call generalised slopes. They are defined by

1
nk
(k)
(6)
,
:=
chk (E)
E
rk(E)
(n k)!
X
(1)
so that E E .
The stability condition now takes the form of a series of inequalities. A bundle is stable if
(1)
(1)
E < E
for all E E,
(7)
or, if E = E for some E E and there is no E E with E > E ,

(1)
(2)
(1)
(2)
E < E
(1)
(1)
for all E E,
(8)
or, if E = E for some E E and there is no E E with E > E ,

(2)
(2)
(3)
(3)
E < E
(2)
for all E E,
(2)
(9)
and so on. The first of these inequalities implies -stability, a -stable bundle is always stable. It also follows that a -semistable bundle is always -semistable, since a -stable
(1)

bundle must have (1)
E E for all E E, but the converse does not hold.
A result analogous to the DUY theorem exists for -stability. Leung [4] has found a
non-linear deformation of the self-dual YM equation,
t +F
(2n)
e
Td(X)
= E (t)I,
77
which he calls the almost HermitianEinstein equation, whose solutions exist iff the instanton bundle is -stable. The supscript (2n) means to take the 2n-form part of the expression.
Leung showed that when t 0 (i.e., for all t > T , where T depends on the bundle) this
equation has a solution iff the bundle is -stable, which means that E (t) < E (t) for
all sub-sheaves E E. This equation is a deformation of the HermitianYM equation in
the sense that the highest order term in t is identical to the HermitianYM equation. Note
that the equation is perturbative in the sense that it is only valid for sufficiently large t.
Furthermore, the equation becomes trivial when taking the trace and integrating.
3. Physical ( -) stability
A quantum field theory with N = 2 supersymmetry contains 2 spinorial symmetry generators QA
, A = 1, 2 satisfying
A B
Q , Q = 2 AB P 2i AB Z ,
(10)
where is a Dirac matrix. Z is the central charge of the algebra and can be regarded
as a complex number. The mass m of a state is given by m |Z|, which is called the
BogomolnyiPrasadSommerfield (BPS) bound. If m = |Z|, the particle is in a short
(BPS) representation of the algebra, which preserves half of the supersymmetry of the
theory. The HermitianYM equation (4) is known to be related to the BPS condition in
N = 2 YM theory [8].
Douglas et al. [2] have proposed a stringy analogue of -stability called -stability.
A brane is -stable iff, for all sub-branes E E one has E < E := Arg Z(E). In the
geometric or large volume limit (t ) the central charge is given by [8,9]3

Z(E) Z(E) = eit ch(E) Td(X),
(11)
X
where ch(E) is the Chern character of the vector bundle E, and Td(X) is the Todd class
of the CalabiYau space X. When X is topologically non-trivial the Todd form mixes
the electric and magnetic charges of the brane, a phenomenon dubbed the geometric
Witten effect in Ref. [8], by analogy with the fact, discovered by Witten [12], that magnetic
monopoles are dyons on topologically non-trivial spaces.
Using the low-energy expression Eq. (11) for the central charge, one may obtain a hierarchy of inequalities similar to the conditions derived from -stability. The calculation is
in this case a bit more involved, and it turns out that the resulting inequalities depend on
the Todd form of the CalabiYau,
as opposed to the Gieseker case. For a torus T 6 , we get

(using the normalisation n /n! = 1)
(1)
(1)
E < E ,
3 The full expression for the central charge [10,11] for a D-brane on a sub-manifold S X contains the field
FNS /2
strength of the normal bundle and the A-roof genus, e

when S = X.
S ), but this factor reduces to Td(X)

S )/A(N
A(T
78

(1) (2)
(3)
(2) (3)
(2) (3)
(1) (2)
(3)
E E E < E E E ,
E E < E E .
For a quintic hypersurface in P4 , the inequalities become [13]
(1)
(1)
E < E ,
(1) (2)
(3)
(1) (2)
(3)
E E E < E E E ,
5 (3) 5 (1) (2)
5 (3) 5 (1) (2)
(3)
(2) (3)
E + E E + (2)
E E < E + E E + E E .
2
6
2
6
It is clear from these expressions that -stability does not reduce to -stability in the geometric limit, although this has been speculated in the literature. Note also that as in the
case of -stability, a -stable bundle is always -stable, but a -stable bundle is not necessarily -stable. Furthermore, a -semistable bundle does not have to be -semistable. It
is therefore not strictly correct to say that -stability always reduces to -stability in the
large volume limit.
Leungs almost HermitianYM equation can therefore not be the relevant instanton
equation for string theory, and the question arises whether there exists another deformation
of the HermitianYM equation whose solutions correspond exactly to -stable bundles.
The rest of the paper is devoted to this problem.
4. Stringy instantons
In the geometric limit the conditions for a D-brane (wrapped on a compact CalabiYau
manifold) to be BPS may be stated in terms of geometrical equations. The BPS condition
gives equations both for the cycle on which the brane is wrapped [14] and for the gauge
field strength F and background B-field. The equation for the gauge field strength was
found by Mario et al. [15] for the case of a D6-brane wrapping a Khler 3-fold with
B = 04

1 3
1 3
2
2
F F = cot E F ,
(12)
3
3
where E is a constant determined by the topology of the gauge vector bundle E. The value
of cot E may be found by integrating Eq. (12). This equation was recently confirmed by a
world-sheet approach [16].
Eq. (12) does not include
all instanton corrections [17], and there has been specula
tion [18] that the factor Td(X) should be included in this equation. Our starting point is a
generalisation of the above equation, introducing this factor5 and the volume parameter t:

(2n)

(2n)

= cot E (t) Im eit +F Td(X)
,
Re eit +F Td(X)
(13)
4 Our notation differs slightly in the definition of from [15]. Products are exterior products.
E
5 For a bundle on a proper sub-manifold S X, the factor Td(X) should be replaced as in Footnote 3.
79
where E (t) is the phase of the central charge given by Eq. (11). By comparing Eqs. (12)
and (13) one sees that in the case of the torus T 6 where Td(X) = 1, E = E (setting
t = 1). The condition for -stability is therefore in this case equivalent to E < E . We
will show that Eq. (13) has a solution iff the bundle E is -stable in the limit t 0. An
additional technical assumption we must make [4] is that the solution of Eq. (13) is wellbehaved in the limit, in the sense that limt |F |/t = 0. The necessity of this assumption
will become clear below.
Our strategy is to adapt Leungs proof for the case of -stability to this case. The main
technical tool is geometric invariant theory (GIT). We will not here repeat those parts of
the proof which coincide with Leungs case, but summarise the most important points and
emphasise the parts that are different in the two cases. We will restrict ourselves to the case
when the D-brane is filling a 3-dimensional CalabiYau space, but the theorem should
be straight forward to generalise to other branes. One should also note that this procedure is not restricted to Eq. (13), but may be used for a rather generic deformation of the
HermitianYM equations, as long as the stability condition is deformed correspondingly.
Our first claim is that Eq. (13) implies that the bundle E is -stable. Let S be any subbundle of E. The key point is to decompose the connection on E into connections on S
and Q := E/S,

AS B
A=
(14)
,
B AQ
so the field strength FE may be written

B
FS + BB
FE =
.

B
FQ + B B
(15)
In the case n = 3, Eq. (13) is expanded as

3

1 3
t 3
2
tFE t Td2 (X) .
t FE + FE + FE Td2 (X) = cot E (t)
3
3
2 2
By taking the trace over S only, this becomes

trS [t 2 2 FE + FE3 /3 + FE Td2 (X)]
cot E (t) =
trS [t 3 3 /3 tFE2 t Td2 (X)]

1
= cot S (t)
trS 2 BB + ,
t rk(S)

=
(16)
(17)
trS [(FE3 FS3 )/3 + BB Td2 (X)]
trS [t 3 3 /3 tFE2 t Td2 (X)]

trS [tFE2 t Td2 (X)]
trS 2 BB

+
t rk(S)
trS [t 3 3 /3 tFE2 t Td2 (X)]

trS [t 2 2 FS + FS3 /3 + FS Td2 (X)] trS [t(FS2 FE2 )]

.
trS [t 3 3 /3 tFE2 t Td2 (X)] trS [t 3 3 /3 tFS2 t Td2 (X)]
80

The term trS 2 BB is positive, see, e.g., [19, p. 79]. It is therefore enough to show
that || < | trS 2 BB |/t rk(S). Leung shows that if Tk is a term containing a product
B , F
of k + 1 or fewer terms of the type B B, B,
S and FQ , then the assumption
limt |F |/t = 0 leads to the conclusion that Tk /t k |B|2 .
We can choose a t large enough so that the denominators of the fractions in are
arbitrarily close to t 3 rk(S), and
by Leungs arguments, each of these terms is then not
greater than a small fraction of |B|2 . Thus, since we get cot E (t) cot S (t) < 0 for all
sub-bundles S E, as t we have S (t) < E (t), which proves the first part of the
proposition.
For the converse we must show that there exists a solution of Eq. (16) for any -stable
bundle E, so assume now that the bundle E is -stable. We expand the topological parameter cot E (t) as a series in 1/t. Defining E (t) = t cot E (t), we may write Eq. (16)
as

1
2 F + t 2 1 F 2 F 3 + 1 Td2 (X)
3
(2)

3
(2)
+ t 4 E F 2 + E Td2 (X) + O t 6 = E (t) ,
(18)
3
where

(1)
(2)
(3)
E (t) = E + t 2 E + t 4 E + O t 6
(1) (2)
(1)
(3)
(1)
= E + t 2 E E E + E

(2) (1) (2)
(3)
(1)
+ t 4 E E E E + E + O t 6 ,
(19)

and = Td2 (X). For 0 < E < , -stability is now equivalent to E (t) < E (t),
and we must show that there is a solution to Eq. (18) in this case.
An important property of (strictly) -semistable bundles is that they may be deconstructed into a sequence (or filtration) of -stable bundles. This is the JordanHlder
theorem: If E is a -semistable vector bundle, there is a filtration E E1 Ek 0
such that each Qj := Ej /Ej +1 is -stable and 1 (Qj ) = 1 (E) for each j .
Since the quotients Qj are -stable, there exist connections A(j ) on each of them sat(1)
isfying the HermitianYM equation nj 1 F (j ) = E jn /nj !. Therefore, there exists a
(Hermitian) connection A on E,
B (k1,k)
B (1,k)
A(k)
(k1,k)
(k1)
(1,k1)
A
B
B
.
..
..
..
..
A=
(20)
.
.
.
.
B (1,k)
B (2,k)
A(k)
The off-diagonal parts B (j,) are uniquely given [4] up to a scalar factor from the extension
0 Ej +1 Ej Qj 0. B (j,) is the component in Hom(Qj , Q ) of the representative of the extension class of this extension.
The JordanHlder theorem is applicable to our case since a -stable bundle is always
-semistable. Assume now that a bundle has a connection satisfying Eq. (13). Then E is
-semistable. This may be proven by noting that there exists a metric on E for all positive
81
constants such that its curvature satisfies | R 1 I | < , where R = (n1 R). This
follows since there is a connection satisfying Eq. (13) for any t sufficiently large. Since
E is -semistable, we may then use the JordanHlder theorem to create a filtration E
E1 Ek 0.
Following Leung we use the simplest case, in which the bundle has a short Jordan
Hlder filtration, as an example: E S 0. We define Q := E/S. By the JordanHlder
(1)
(1)
(1)
theorem we then have Q = S = E . We will also assume for now that E is -stable
at second order, i.e., that E(1) = E(1) and E(2) < E(2) for all sub-bundles E .
Since we want to find a solution to Eq. (18) which is a perturbation of the DUY equation,
we now perturb the connection (20) with corrections proportional to t 1 and t 2 , and let

2
)/t 2
AS + (h1
B/t (
S )hS /t
AE =
(21)
2 ,
B /t + /t 2
AQ + (h1
Q )hQ /t
where we will determine the proper normalisation of B later. Eq. (18) now becomes
1

hS hS BB 2

3
1 3
1
(2)
2
,
F
= E
+ 1 Td2 (X) + O
(22)
1 F0S
3
3 0S
t
1

hQ hQ B B 2

3
1 3
1
(2)
2
1 F0Q
= E
F0Q
+ 1 Td2 (X) + O
(23)
,
3
3
t

1
2
.
=O
(24)
t
The following theorem [4,6] tells us when these equations have solutions: let E be a stable bundle
over a compact Khler manifold X. Suppose H0 is an endomorphism of E
such that trE H0 = 0. Then there exists a unique positive self-adjoint endomorphism h of
1 h) = H0 .
E with det h = 1, which solves (h
This is where that the requirement for -stability enters. In order to fulfil the requirement for the theorem, we must have

(2)
(2)
0 |B|2 = E S rk(S).
(25)

(2)
(2)
Eq. (25) can only be satisfied when S E . When this is the case, we have trS H0 =
0. Therefore, Eq. (22) with t always has a solution when E is -stable.
This can be extended to t < by rescaling B by a parameter
depending on hS , hQ
and . One can show that there is a rescaling parameter such that trS H = 0, where H is
given by Eq. (12), thus the above theorem also holds for t < .
Following Leung, this example may be generalised to
(n)
(n)
S = E
(n = 1, 2, . . . , m),
(m+1)
(m+1)
< E
Using the above approach order by order in t, there will exist a connection
(n1)
2n
n /t 2
+ h1
B/t m
AS
(n)S hS /t
A(n) =
(n1)
2n
B /t m n /t 2 AQ
+ h1
(n)S hS /t
(26)
82
where n = m/2+1 /t m+2 + + n /t 2n for n > m/2, and n = 0 otherwise. This connection will solve Eq. (12) to order t 2m . For higher orders, one must again rescale B.
Finally, assume that E is a general -stable bundle, which then has a JordanHlder
filtration E = E0 E1 Ek Ek+1 = 0, and the bundles Qj := Ej /Ej +1 are (1)
stable and have the same slope E as E. The connection on E will now be
(k,k)
A
A(k,k1)
A(k,1)
(k1,k)
(k1,k1)
(k1,1)
A
A
A
,
..
..
..
..
A=
.
.
.
.
A(1,k)
A(1,k1)
A(1,1)
with
1 1
1
1
)
h hj , A(i,j
= B(i,j ) 2
(i,j ) ,
t2 j
t
t
1
1
)
= B(j,i)
+ 2 (j,i) , i < j,
A(i,j
t
t
(j )
)
A(j,j
= A0 +
i > j,
with one B(i,j ) and one (i,j ) for each pair (i, j ), i > j . The above procedure can then be
repeated in this general case, and we may finally conclude that Eq. (13) has a solution iff
the bundle E is -stable.
5. Conclusion and outlook

We have related -stability of a D-brane in the large volume limit to a geometric condition on the field strength of the gauge bundle, Eq. (13), which is a deformation of the
well-known HermitianYM equation. -stability is the correct stability condition in the
full (quantum corrected) theory and we propose Eq. (13) to be the correct instanton equation for describing BPS D-branes geometrically (in the large volume limit), generalising
the HermitianYM equation which describes BPS instantons in (undeformed) YM theory.
The string-deformed low-energy YM field equations are just the fossilised low-energy
remains of a huge framework of great beauty, consistency and subtlety whose low-energy
implications have in the past had significant impact on algebraic geometry (e.g., mirror
symmetry). It is therefore not unreasonable to hope that geometric -stability will be of
use as well. While -stability is a refinement of -stability, a finer sieve used to select
semi-stable bundles, -stability is quite distinct from -stability in three dimensions, and
may therefore be expected to give rise to genuinely new and interesting compactifications
of the moduli spaces studied in algebraic geometry.
The large volume limit has in this paper been taken with B = 0, which is the traditional
geometric limit. However, it would be interesting to see the consequences of including
also a B-field in the geometric picture. Shifting the B-field by an integral form is equivalent
to tensoring the bundle with a line bundle, which in the case of -stability does not affect
stability. For -stability however, an integral shift of the B-field may affect stability, as
found in [20]. This means that including the B-field may extend the moduli space of stable bundles in a non-trivial way.
83
Acknowledgements
We are grateful to M. Douglas, W. Lerche and G. Ellingsrud for helpful discussions, and
especially to C. Rmelsberger for patiently explaining -stability and for collaboration at
an early stage of this project. This work was supported by the Research Council of Norway
and was made possible by the generosity of the CERN Theory Group.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
E.R. Sharpe, Adv. Theor. Math. Phys. 2 (1999) 1441, hep-th/9810064.

M.R. Douglas, B. Fiol, C. Rmelsberger, hep-th/0002037.
R. Friedman, Algebraic Surfaces and Holomorphic Vector Bundles, Springer, New York, 1998.
N.C. Leung, J. Differential Geom. 45 (1997) 514.
S.K. Donaldson, Proc. London Math. Soc. 50 (1985) 1.
K. Uhlenbeck, S.T. Yau, Commun. Pure Appl. Math. 39 (1986) S257.
R. Hartshorne, Algebraic Geometry, Springer, New York, 1977.
J.A. Harvey, G.W. Moore, Commun. Math. Phys. 197 (1998) 489, hep-th/9609017.
D.E. Diaconescu, C. Rmelsberger, Nucl. Phys. B 574 (2000) 245, hep-th/9910172.
S. Katz, E. Sharpe, Adv. Theor. Math. Phys. 6 (2003) 979, hep-th/0208104.
D.S. Freed, E. Witten, hep-th/9907189.
E. Witten, Phys. Lett. B 86 (1979) 283.
C. Rmelsberger, private communication.
K. Becker, M. Becker, A. Strominger, Nucl. Phys. B 456 (1995) 130, hep-th/9507158.
M. Marino, et al., JHEP 0001 (2000) 005, hep-th/9911206.
A. Kapustin, Y. Li, hep-th/0311101.
M.R. Douglas, math.AG/0009209.
R. Minasian, A. Tomasiello, Nucl. Phys. B 631 (2002) 43, hep-th/0104041.
P. Griffiths, J. Harris, Principles of Algebraic Geometry, Wiley, New York, 1978.
M.R. Douglas, B. Fiol, C. Rmelsberger, hep-th/0003263.

Supersymmetric quenching
of the Toda lattice equation
K. Splittorff a , J.J.M. Verbaarschot b
a Nordita, Blegdamsvej 17, DK-2100, Copenhagen , Denmark
b Department of Physics and Astronomy, SUNY, Stony Brook, NY 11794, USA
Abstract
The average of the ratio of powers of the spectral determinants of the Dirac operator in the -regime
of QCD is shown to satisfy a Toda lattice equation. The quenched limit of this Toda lattice equation
is obtained using the supersymmetric method. This supersymmetric approach is then shown to be
equivalent to taking the replica limit of the Toda lattice equation. Among others, the factorization of
the microscopic spectral correlation functions of the QCD Dirac operator into fermionic and bosonic
partition functions follows naturally from both approaches. While the replica approach relies on an
analytic continuation in the number of flavors no such assumptions are made in the present approach
where the numbers of flavors in the Toda lattice equation are strictly integer.
1. Introduction
A quenched theory is one in which a determinant has been removed from weight
in the partition function. Removing a determinant from the partition function does not
necessarily lead to an unphysical theory. The determinant can act as a source term for
Greens functions, and, once a derivative of the partition function has been taken, this
source is removed. Two widely used analytical tools to remove determinants are the replica
trick and the supersymmetric method. For the replica trick, one introduces n identical
copies of the determinant, performs the appropriate derivatives, and finally takes the limit
E-mail addresses: split@alf.nbi.dk (K. Splittorff), verbaarschot@tonic.physics.sunysb.edu
(J.J.M. Verbaarschot).
doi:10.1016/j.nuclphysb.2004.06.022
K. Splittorff, J.J.M. Verbaarschot / Nuclear Physics B 695 (2004) 84102
85
n 0. For the supersymmetric method, a ratio of two determinants with different values of
the sources is introduced in the partition function. Derivatives are performed with respect
to the sources and the determinants are removed by setting the sources equal. This paper
deals with the equivalence of these two approaches.
Originally, both the replica trick [1] and the supersymmetric method [2] where invented
for the study of disordered systems. Applications of the replica trick soon proved to be
delicate, see, e.g., [3], the trouble being in the choice of the dominant saddle points and the
analytic continuation in the replica index [4]. The supersymmetric method [2] does not rely
on such analytic continuation, and never faces this problem. At the perturbative level, one
only finds a polynomial dependence on the number of replicas and both approaches yield
identical results (see, for example, [46]). The replica trick has made its mark in diverse
areas of theoretical physics. Some examples are the proof of the linked cluster theorem
(see [7]) and the study of vacua in supersymmetric gauge theories [8].
In disordered systems, which also includes the Dirac operator in a gauge field distributed
according to the YangMills action, the standard approach both for the replica trick and the
supersymmetric method is to rewrite the bosonic and/or fermionic integrals in terms of a
non-linear -model. Such a -model describes the Goldstone modes associated with the
spontaneous breaking of the global symmetries of the original model. Both its symmetry
breaking term and its kinetic term are determined by the pattern of global symmetry
breaking and Lorentz invariance. In the theory of disordered systems, these Goldstone
modes are known as diffusons whereas in QCD they are known as the pseudoscalar mesons
(for reviews see [2,9]). For QCD, the -model is known as a chiral Lagrangian and has been
studied extensively in relation to the low-energy properties of the strong interactions [10].
In a parameter domain where the kinetic term is important, in most cases, only a
perturbative treatment of the theory is feasible. However, in a domain where the kinetic
term can be ignored, which is know as the ergodic domain or the domain, the -model
becomes zero dimensional and can be solved exactly in many cases. The zero-dimensional
-models are equivalent to random matrix theories [2,1114]. Using this connection, the
problems with the replica trick where shown to be of a more serious nature [15]. This leads
to the general consensus that the replica trick was only reliable for perturbative calculations
and could generally not be used to obtain exact non-perturbative results. A first hint that
non-perturbative results could be obtained from the replica trick came from [16,17] where
the oscillatory factors in the asymptotic expansion of the two-point correlations function
were reproduced. However, only in cases where the perturbative expansion terminates [18],
exact analytical results could be obtained, and, as was pointed out in [18], a systematic and
exact replica method remained illusive in these works. Recently, replica methods were
introduced that went beyond these limitations [1922]. Most significantly, exact analytical
results were obtained in cases where other methods have failed. While the formulation of
the replica trick introduced in [20] has been shown to be widely applicable, it still relies on
an assumed analytical continuation in the replica index. The central purpose of this paper
is to justify this analytical continuation.
The new development which allowed for an exact analytic evaluation of the spectral
correlation functions is the link between zero-dimensional -models and an exactly
solvable system know as the Toda lattice. This system is one-dimensional and consists
of massive particles interacting through an exponential potential [23]. The Hamilton
86
equations of motion for the position of the masses as a function of time are known as the
Toda lattice equations. The system is integrable, and its solution for appropriate boundary
conditions is given by the partition function of a zero-dimensional -model.
Previously, we have shown [20,21] that the replica limit of the Toda lattice equation
not only gives the exact analytic results but also explains the factorization of the
universal microscopic correlation functions into a product of a bosonic and a fermionic
partition function. Though this factorization suggests a direct relationship with the
supersymmetric generating functionals, it has remained a miracle why the final result
factorizes within the supersymmetric method. In this paper we show this factorization
based on explicit expressions for the ratio of different powers of spectral determinants
[24]. The key observation is that this family of partition functions also satisfies Toda
lattice equations. Quenching these Toda equations as in the supersymmetric method, that
is setting bosonic and fermionic masses equal, automatically reveals the factorization of
correlation functions. As stressed above, the replica limit of the Toda lattice equation relies
on an analytic continuation in the number of flavors from integer values to real values.
Although the replica limit of the Toda lattice equation, in cases that have been studied up
to now, has given the correct analytic result, no definite arguments have been given that
the usual problems of the replica limit do not manifest themselves in this approach. As
we will show in this paper the replica limit of the Toda lattice equation is equivalent to
quenching the Toda lattice equation of a family of supersymmetric partition functions with
a strictly integer number of bosonic and fermionic flavors. This result directly justifies the
interpretation of the partition function with zero flavors in the usual Toda lattice equation
as the supersymmetric partition function.
Throughout this paper we consider the zero-dimensional limit of the -model relevant
for QCD. In the random matrix formulation this corresponds to the chiral unitary ensemble
(chUE). We expect that similar results can be obtained for the unitary ensemble.
This paper is organized as follows. First we describe the replica method and the
supersymmetric method. Then, in Section 3, we discuss Toda lattice equations relevant for
the one-point function and the two-point function. In Section 4 the quenching of the Toda
lattice equations is discussed. This automatically yields the quenched correlation functions
in the factorized form. The equivalence of the replica method and the supersymmetric
approach is discussed in Section 5. Finally, in Section 6, we summarize our results and
comment on possible extensions of our work. In two appendices we derive the Toda lattice
equations for the supersymmetric partition functions.
2. Definitions of replica and supersymmetric quenching

In this section we set our notation and introduce the supersymmetric method and the
replica method. We illustrate the general definitions with generating functions for the
microscopic limit of the one-point functions of the chUE.
The essential ingredient in the partition functions are the fermionic and bosonic
determinants which will be written explicitly whereas all other parts of the weight are
included in the average denoted by . The fermionic replicated partition function is
87
defined by

ZNf (x) = detNf (D + x) ,
(1)
where D is an operator, the mass x is a number, and, as just mentioned, denotes the
average for the specific theory (note that x 1 = 0). Analogously, the bosonic replicated
partition function is given by

ZNb (y) = detNb (D + y) .
(2)
The replica trick for removing the determinants is to take the limit Nf 0 or Nb 0.
The supersymmetric partition function is defined by the average ratio of powers of the
spectral determinants

ZNf ,Nb (x|y) = detNf (D + x)detNb (D + y)
(3)
which can be removed by taking the limit y x. For example,
lim Z4,2 (x|y) = Z2 (x).
yx
(4)
Obviously, it is not particularly interesting to introduce and remove determinants unless

we differentiate with respect to the masses first. The average resolvent of the operator D is
defined as

1
.
G(x) Tr
(5)
D+x
In the fermionic replica limit it is given by
1
x log ZNf (x),
Nf 0 Nf
G(x) = lim
(6)
in the bosonic replica limit we have that

1
x log ZNb (x),
Nb 0 Nb
G(x) = lim
(7)
and the supersymmetric method is based on the identity

G(x) = lim x log Z1,1 (x|y).
yx
(8)
While the average in Eq. (1) appears to be well defined for any real value of Nf it is
often not possible to evaluate the average unless Nf is an integer. For this reason, taking
the limit Nf 0 involves an analytic continuation in Nf .
In this paper we focus on the topological trivial sector of the QCD partition function
which is the average of a fermionic determinant (with D = i + A )
Nf

det(i D mf ) eSYM (A)
ZNf {mf } = [dA]0
(9)
f =1
over gauge field configurations with trivial topology weighted by the YangMills action
SYM . The quark masses {mf } m1 , . . . , mNf may take different or identical values. In
88
particular, we are interested in spectral correlation functions of the Dirac operator D

in the phase where chiral symmetry is spontaneously broken. The order parameter for the
spontaneous breakdown of chiral symmetry is the chiral condensate denoted by . At
low energies, the effective degrees of freedom are the Goldstone modes associated with
this spontaneously broken symmetry, and the effective theory is a -model [25]. The zerodimensional limit of this -model allows for an analytical evaluation of the correlation
functions on microscopic scales. The microscopic scale is set by the inverse of V ,
where V is volume of the system. The domain in which the zero momentum modes of this
effective theory decouple from the non-zero modes is known as the -regime of QCD [10].
In the topologically trivial sector, = 0, the fermionic partition functions are given by the
unitary, and hence compact, integral [2628]

1
ZNf {xf } =
(10)
dU e 2 Tr[M U +MU ] ,
U U (Nf )
m1 , . . . , V
mNf ) is the rescaled mass
where M diag(x1 , . . . , xNf ) diag(V
matrix. The bosonic partition functions are given by a non-compact integral over positive
definite matrices [13,29]

1
1
ZNb {yb } =
(11)
dQ e 2 Tr[M Q+MQ ] ,
QGl(Nb )/U (Nb )
where the dependence on the bosonic quark masses, the chiral condensate, and the volume
is through the products yb mb V appearing in the rescaled quark mass matrix
M diag(y1 , . . . , yNb ). The supersymmetric partition function of QCD in the -regime
is given by an integral over a supergroup [12,13]

1
1
dQ e 2 Str[M Q+MQ ] ,
ZNf ,Nb {xf }{yb } =
(12)
f |Nb )
QGl(N
where M = diag(x1 , . . . , xNf , y1 , . . . , yNb ) is the rescaled mass matrix. The bosonboson
f |Nb ) is
f |Nb ) is Gl(Nb )/U (Nb ) and the fermionfermion block of Gl(N
block of Gl(N
U (Nf ). All three integrals are fixed uniquely by the symmetries.
Much has been learned about these partition functions over the past few years. Explicit
expressions in terms of determinants of Bessel functions are known in all cases [13,20,24,
27,2939]. For ZNf and ZNb the expressions have been derived directly from the group
integrals (10) and (11) while the derivation of the general expression for ZNf ,Nb [24,34
36] goes through the random matrix representation. For all the partition functions Nf and
Nb must take integer values and the analytic continuation in Nf and Nb is not uniquely
defined.
As mentioned in the introduction, the latest developments include a link between such
matrix integrals and an exactly solvable one-dimensional system known as the Toda lattice
[32,33,40]. The Hamiltonian equation of motion of the Toda lattice is known as the Toda
lattice equation. The remarkable link is that the QCD partition functions also obey such
Toda lattice equation. This equation connects partition functions with different numbers of
89
flavors, much in the same way as recursion relations for orthogonal polynomials connect
polynomials with different indices. It turns out that the Toda lattice equation for fermionic
flavors is also valid for bosonic flavors. Both in the bosonic case and the fermionic case
the Toda lattice equation terminates for zero flavors. Based on the correct analytical result
for the replica limit of these hierarchies, it was conjectured in [20] that both hierarchies are
connected by the supersymmetric partition function. This assertion is proved in this paper
starting from the observation that the supersymmetric partition functions also obey a Toda
lattice equation.
3. The Toda lattice equations for the supersymmetric partition function

The Toda lattice equations considered in the literature (see for instance [4042]) connect
partition functions with a different number of fermionic flavors or a different number
of bosonic flavors. That is, the Toda lattice equations deal with semi-infinite hierarchies
with either positive or negative index. In [20,21] it was found that these two semi-infinite
hierarchies are connected. The connection was realized as the replica limit of the index in
either the positive or the negative semi-infinite hierarchy. Here, we consider Toda lattice
equations for the partition functions with both fermions and bosons, i.e., supersymmetric
partition functions. Referring to the graded symmetry of the supersymmetric partition
functions we call these graded Toda lattice equations. In the next section we will derive
the quenched resolvents from these graded Toda lattice equations.
3.1. A graded Toda lattice equation affecting one flavor
Starting from the expression for the partition function (12) with Nf fermionic flavors
and Nb bosonic flavors [20,3436]1
j 1

det[z Jj 1 (zi )]i,j =1,...,Nf +Nb
ZNf ,Nb {xf }{yb } = N i
,

f
2 x 2 ) Nb
2 y 2)
(x
(y
i
i
j >i=1 j
j >i=1 j
(13)
where zi = xi for i = 1, . . . , Nf , zNf +i = yi for i = 1, . . . , Nb , Jj 1 (zi ) Ij 1 (xi )

for i = 1, . . . , Nf , and Jj 1 (zNf +i ) (1)j 1 Kj 1 (yi ) for i = 1, . . . , Nb , we prove in
Appendix A that the partition function (12) satisfies the graded Toda lattice relation

Nf +Nb xi log ZNf ,Nb {xf }{yb }
ZNf +1,Nb (xi , {xf }|{yb })ZNf 1,Nb ({xf }f =i |{yb })
= 2nxi2
(14)
.
[ZNf ,Nb ({xk }|{yb })]2
1 Notation. The vertical bar in the argument of Z separates the masses {x } x , . . . , x
f
Nf of the fermionic
1
quarks from the masses {yb } y1 , . . . , yNb of the bosonic quarks. For degenerate masses we either repeat the
mass in the argument of Z according to the degeneracy or put an additional subscript on Z indicating the
degeneracy, e.g., Z3,1,2 (x1 , x2 |y) Z4,2 (x1 , x1 , x1 , x2 |y, y). For bosonic masses, the subscritpt of Z that
denotes the number of flavors will contain and explicit minus sign. If all flavors are of the same kind the vertical
bar will be omitted.
90
Here, n is the number of fermionic flavors with mass xi in ZNf ,Nb ({xf }|{yb }). The
derivation x is defined by
x xx
(15)
and
Nf +Nb
Nf

f =1
x f +
Nb

y b .
(16)
b=1
For degenerate masses the sum only extends over the different masses.
This is the generalization of the Toda lattice equation for fermions obtained in [42],
to include both fermions and bosons. Below, in Section 4.1, we quench this graded Toda
lattice equation as in the supersymmetric method and show that the factorization of the
quenched resolvent follows naturally.
3.2. A graded Toda lattice equation affecting two flavors
The key ingredient in understanding the factorization property [21] of the two-point
correlation function within the supersymmetric approach is a graded Toda lattice equation
where two indices are raised and lowered simultaneously. In [21] it was shown that for m
degenerate fermionic flavors with mass x1 and n degenerate fermionic flavors with mass
x2 , the partition function (10) satisfies the Toda lattice equation (m and n are positive
integers)
x1 x2 log Zm,n (x1 , x2 ) = 4mnx12 x22
Zm+1,n+1 (x1 , x2 )Zm1,n1 (x1 , x2 )

.
[Zm,n (x1 , x2 )]2
(17)
Using the explicit expressions for the partition function given in (13) we have shown that
x1 x2 log Z2,2 (x1 , x2 |y1 , y2 ) = 4x12 x22
Z4,2 (x1 , x1 , x2 , x2 |y1 , y2 )Z2 (y1 , y2 )

[Z2,2 (x1 , x2 |y1 , y2 ))]2
(18)
and a similar equation for other small numerical values of the number of flavors (see
Appendix B).
We conjecture that these results can be generalized to any integer number of fermions,
Nf , and any integer number of bosons, Nb (further support for this conjecture will be given
in Appendix B):

xi xj log ZNf ,Nb {xf }{yb }
ZNf +2,Nb (xi , xj , {xf }|{yb })ZNf 2,Nb ({xf }f =i,j |{yb })
,
= 4nmxi2 xj2
(19)
[ZNf ,Nb ({xf }|{yb })]2
where m and n are the number of flavors with mass xi and xj in ZNf ,Nb ({xf }|{yb }),
respectively. A similar equation holds for differentiation with respect to two bosonic
masses. In that case the fermionic flavors are passive. Furthermore, we have shown for
a number of values of Nf and Nb (see Appendix B) that a corresponding equation is valid
when we differentiate with respect to one fermionic and one bosonic mass. Without proof
91
we assert that the general result in this case is given by

xi yj log ZNf ,Nb {xf }{yb }
= 4nmxi2 yj2
ZNf +1,Nb +1 (xi , {xf }|{yb }b=j )ZNf 1,Nb 1 ({xf }f =i |yj , {yb })
[ZNf ,Nb ({xf }|{yb })]2
(20)
where m and n are the number of flavors with mass xi and yj in ZNf ,Nb ({xf }|{yb }). These
graded Toda lattice equations go beyond previous results [21] in two respects: (1) they have
been extended to supersymmetric partition functions, and (2) the masses in these equations
can be either degenerate or non-degenerate.
4. Supersymmetric quenching of the graded Toda lattice equation

In the supersymmetric approach the partially quenched or fully quenched correlation
functions are obtained by differentiating before taking the limit of equal fermionic and
bosonic masses. In this section we show how the graded Toda lattice equations of the
previous section automatically give the quenched one-point and two-point functions. The
results are in exact agreement with expressions obtained previously. The important new
result of this section is that the factorization of correlation functions into products of
partition functions follows naturally without assuming an analytical continuation in the
number of flavors.
4.1. The one point function
In this section we derive the quenched spectral one point function by quenching the
graded Toda lattice equation (14) using the supersymmetric method. For this it is sufficient
to have one bosonic and one fermionic flavor, and the graded Toda lattice equation then
reads
x (x + y ) log Z1,1 (x|y) = 2x 2
Z2,1(x, x|y)Z1(y)
.
[Z1,1 (x|y)]2
(21)
The limit y x of the l.h.s. of this equation is given by

1
1
y Tr
lim x (x + y ) log Z1,1 (x|y) = lim x x Tr
yx
yx
D + x x,y
D + y x,y
= x xG(x).
Here, we have used the notation

det(D + x)
x,y =
.
det(D + y)
(22)
(23)
Taking also the limit y x of the r.h.s. of (21) we find

x xG(x) = 2x 2Z1 (x)Z1 (x).
(24)
92
This result is in complete agreement with known results for the resolvent and the partition
functions. For completeness we mention that they are given by

G(x) = x K0 (x)I0 (x) + K1 (x)I1 (x) ,
Z1 (x) = I0 (x),
Z1 (x) = K0 (x).
(25)
Clearly, this is a rather cumbersome way of deriving the resolvent from Z1,1 . Let us
therefore again stress that the important new result here is that the factorization in Eq. (24)
is proved on the basis of the Toda lattice hierarchy for the supersymmetric partition
functions. We also stress that this approach does not rely on an analytic continuation in
the number of flavors.
Following the same strategy as above one can also obtain the partially quenched
resolvents. As we seek to stress a matter of principle rather than reproducing existing
results we instead turn to the two-point function.
4.2. The two-point function
The two-point disconnected susceptibility contains two derivatives just as in the Toda
lattice equation. This property simplifies the derivation of two point-functions from the
Toda lattice equation for ratios of spectral determinants.
The fully quenched disconnected susceptibility in the chiral unitary ensemble with
identical arguments is defined as
(x, x) lim x y log Z1,1 (x|y).
yx
(26)
It follows from (20) for Nf = Nb = 1 at equal values of the arguments that

(x, x) lim x y log Z1,1 (x|y) = 4x 2Z2 (x, x)Z2(x, x).
yx
(27)
In order to get the quenched disconnected susceptibility for different values of the
arguments, we start from the partition function describing two fermions and two bosons
all with different masses. We now have a choice: we can differentiate with respect to two
fermionic masses, with respect to two bosonic masses, or with respect to one fermionic
mass and one bosonic mass, before pairing up the fermion and boson masses. If we choose
the first option and use (18) we find
(x1 , x2 )
lim
y1 x1 ,y2 x2
x1 x2 log Z2,2 (x1 , x2 |y1 , y2 )
Z4,2 (x1 , x1 , x2 , x2 |y1 , y2 )Z2 (y1 , y2 )

y1 x1 ,y2 x2
[Z2,2 (x1 , x2 |y1 , y2 )]2
= 4x1 x2 Z2 (x1 , x2 )Z2 (x1 , x2 ).
lim
4x1x2
(28)
The derivation when differentiating with respect to the two bosonic masses is analogous
and leads to exactly the same quenched correlation function. If we choose to differentiate
with respect to one fermionic mass and one bosonic mass and use (20) for Nf = Nb = 2,
x2 y1 log Z2,2 (x1 , x2 |y1 , y2 )
Z3,1 (x1 , x2 , x2 |y2 )Z1,3 (x1 |y1 , y1 , y2 )
,
= 4x2 y1
[Z2,2 (x1 , x2 |y1 , y2 )]2
(29)
93
(which we have also verified explicitly) we obtain

(x1 , x2 )
lim
y1 x1 ,y2 x2
x2 y1 log Z2,2 (x1 , x2 |y1 , y2 )
Z3,1 (x1 , x2 , x2 |y2 )Z1,3 (x1 |y1 , y1 , y2 )

y1 x1 ,y2 x2
[Z2,2(x1 , x2 |y1 , y2 )]2
= 4x1 x2 Z2 (x1 , x2 )Z2 (x1 , x2 ).
lim
4x2y1
(30)
This result coincides with (28) and is in agreement with the result obtained from a
supersymmetric calculation [43] and with the result obtained from the replica limit of the
Toda lattice equation of the fermionic hierarchy [21]. Most importantly, it is clear that the
factorization property of the quenched two-point function in the supersymmetric approach
is a direct consequence of the structure in the graded Toda lattice equation. No assumptions
on the analytic continuation in the flavor index have to be made in this case.
5. Equivalence of replica and supersymmetric quenching

In this section we show that the replica limit of the Toda lattice equation necessarily
gives the same answer as the supersymmetrically quenched Toda lattice equation. This
directly justifies the central assumption of [20,21], namely that the bosonic and fermionic
semi-infinite Toda lattice hierarchies are connected by the supersymmetric partition
function.
5.1. Equivalence for the one point function
The QCD partition function (10) with Nf mass degenerate fermionic quarks satisfies
the Toda lattice equation [42]
x2 log ZNf (x) = 2Nf x 2
ZNf +1 (x)ZNf 1 (x)

[ZNf (x)]2
(31)
(Note that this is obtained from (14), satisfied by the supersymmetric partition function,
setting n = Nf and Nb = 0.) In the replica approach we would like to take the Nf 0
limit of this equation. Using the representation (1) of the partition function, which depends
continuously on Nf , we get [20]
1 2
log ZNf (x) = x xG(x),
Nf x
ZNf +1 (x)ZNf 1 (x)
= 2x 2Z1 (x)Z1 (x).
lim 2x 2
Nf 0
[ZNf (x)]2
lim
Nf 0
(32)
(33)
However, the Toda lattice equation (31) was established starting from (10), which is only
well defined for positive integer values of Nf , and the boundary condition ZNf =0 (x) = 1.
The Toda lattice equation (31) therefore is not necessarily valid for (1) when Nf is noninteger. However, as we will now show, equating (32) to (33) is identical to quenching the
Toda lattice equation using the supersymmetric method, since the supersymmetric partition
94
function satisfies the Toda lattice equation. To this end we write (31) as
ZNf +1 (x)ZNf 1 (x)
1 2
x log ZNf (x) = 2x 2
Nf
[ZNf (x)]2
ZNf +2,1 (x|y)ZNf ,1 (x|y)
= lim 2x 2
.
yx
[ZNf +1,1 (x|y)]2
(34)
These equalities are valid for positive integer values of Nf . However, the r.h.s. of the
second equality is also well defined for Nf = 0. If we take this as the definition of the
Nf 0 limit of the middle term of this equation we get 2x 2 Z1 (x)Z1 (x) exactly as in
(33), but now without ambiguities in the interpretation of Z1 (x). Moreover, in Appendix
A it is shown that the supersymmetric partition function entering the r.h.s. satisfies the
same Toda lattice equation
x (x + y ) log ZNf +1,1 (x|y) = 2(Nf + 1)x 2
ZNf +2,1 (x|y)ZNf ,1 (x|y)

[ZNf +1,1 (x|y)]2
(35)
This equation is also well-defined for Nf = 0 and in this case the limit y x is (cf. (22))
lim x (x + y ) log Z1,1 (x|y) = x xG(x).
yx
(36)
This is precisely what we obtained in (32) using the representation (1) of the fermionic
partition function. This shows that the replica limit of the Toda lattice equation is
necessarily identical to the quenched limit of the supersymmetric Toda lattice equation.
5.2. Equivalence for the two-point function
We now consider the replica limit of the Toda lattice equation affecting two indices
simultaneously. In [21] it was shown that for integer n > 0

Zn+1,n+1 (x1 , x2 )Zn1,n1 (x1 , x2 )
x1 x2 log Zn,n (x1 , x2 ) = 4n2 x1 x2
(37)
.
[Zn,n (x1 , x2 )]2
Making the analytic continuation in n it was assumed [21] that the limit n 0 connects
this semi-infinite hierarchy to the semi-infinite hierarchy for Zn,n , viz.

1
lim 2 x1 x2 log Zn,n (x1 , x2 ) = 4x1x2 Z1,1(x1 , x2 )Z1,1 (x1 , x2 ).
(38)
n0 n
While the exact agreement of the resulting quenched two-point correlation function with
the known result [43] vindicates this replica limit, the underlying assumption again could
not be proved.
We now prove the assumption for the two-point function using similar arguments as for
the one-point function. Again we establish the troublesome connection between positive
and negative integer values of n by quenching determinants as in the supersymmetric
method. The ratio of partition functions in the r.h.s. of (37) can be written as
Zn+1,n+1 (x1 , x2 )Zn1,n1 (x1 , x2 )
[Zn,n (x1 , x2 )]2
Zn+2,n+2,1,1 (x1 , x2 |y1 , y2 )Zn,n,1,1 (x1 , x2 |y1 , y2 )
=
lim
.
y1 x1 ,y2 x2
[Zn+1,n+1,1,1 (x1 , x2 |y1 , y2 )]2
(39)
95
As discussed in Section 4.2 the ratio on the r.h.s. satisfies a Toda recursion relation even
before taking the limit of degenerate masses
Zn+2,n+2,1,1 (x1 , x2 |y1 , y2 )Zn,n,1,1 (x1 , x2 |y1 , y2 )
[Zn+1,n+1,1,1 (x1 , x2 |y1 , y2 )]2

= x1 x2 log Zn+1,n+1,1,1 (x1 , x2 |y1 , y2 ) .
4(n + 1)2 x1 x2
(40)
In this relation we can safely replace n by zero so that

1
x1 x2 log Zn,n (x1 , x2 )
2
n0 n

x1 x2 log Z1,1,1,1(x1 , x2 |y1 , y2 ) .
=
lim
lim
y1 x1 ,y2 x2
(41)
This shows that the replica limit of the Toda lattice equation necessarily gives the same
answer as the supersymmetrically quenched Toda lattice equation, thus justifying the
analytic continuation in n.
6. Conclusion
The QCD partition function in the regime is equivalent to the microscopic limit of a
chiral random matrix theory with the same global symmetries. This equivalence holds for
any number of bosonic and fermionic quarks, and each partition function in this family
can be expressed as a determinant of Bessel functions. In this paper we have shown that
this family of partition functions satisfies a Toda lattice equation which we have called the
graded Toda lattice equation. Based on this result we have shown that the semi-infinite
hierarchy of bosonic or fermionic partition functions can be continued to the limit of
zero flavors which is given by the supersymmetric partition function. This was the central
assumption when taking the replica limit of the Toda lattice equation. By proving this
assumption we have thus shown the equivalence of the replica limit of the Toda lattice
equation and the supersymmetric method. Quenching the partition functions in the graded
Toda lattice equation shows that the resolvent of the supersymmetric partition function
factorizes into the product of a bosonic and a fermionic partition function.
In its simplest form, the graded Toda lattice equation relates derivatives of the
supersymmetric partition function to partition functions with one more flavor or with one
flavor less. This case is relevant for the analysis of the one-point function and has been
proved in all its generality. For the analysis of the two-point function we need a graded
Toda lattice equation in which derivatives of the partition function are related to partition
functions with two more flavors or two less flavors. In addition to the cases that were
required for our analysis, this Toda lattice equation was proved for several other cases
with a fixed number of flavors. Based on these results and known results for exclusively
fermionic flavors, we have conjectured that this equation can be generalized to an arbitrary
number of bosonic and fermionic flavors.
96
It is not yet clear whether variants of the Toda lattice equation are suitable for the
investigation of the replica limit of real symmetric (D = 1) and quaternion real (D = 4)
random matrix ensembles. However, our work can be extended in several other directions.
First, our results can be extended in a straightforward way to partition functions with a nonzero topological charge. The reason that we considered only zero topological charge in this
paper is that it somewhat simplifies our notation. Second, in the formulation of Kanzieper,
the replica limit is taken of the Painlev equation rather than the Toda lattice equations.
It would be very interesting to understand if the supersymmetric partition functions can
also be incorporated in this framework. Third, the spectral density of the quenched Dirac
operator at non-zero chemical potential in the limit of weak non-hermiticity was obtained
from the replica limit of a Toda lattice equation. We expect that the supersymmetric
generating functional at non-zero chemical potential will satisfy a similar Toda lattice
equation so that the replica limit can also be justified in this case. Fourth, the two-point
function of the unitary ensemble can also be obtained from the replica limit of a Toda lattice
equation. Also in this case we believe that a generalization of the Toda lattice equation to
supersymmetric partition functions is possible. Finally, we have no doubt that a similar
arguments can be applied to QCD in 3 dimensions.
Acknowledgements
We wish to thank Eugene Kanzieper for useful correspondence on Painlev equations
and Poul Henrik Damgaard for discussions of the replica trick. This work was supported
in part by US DOE Grant No. DE-FG-88ER40388.
Appendix A. The graded Toda lattice equation where one index is affected
In this appendix we prove the graded Toda lattice equation (14) relevant for the partially
quenched one point functions, that is the relation where one flavor is added and subtracted.
We start by slightly rewriting the general expression for the supersymmetric partition
function in (13) [20,3436]
ZNf ,Nb
j 1

det[zi J (zi )]i,j =1,...,Nf +Nb

.
{xf } {yb } = N
Nb
f
2
2
2
2
j >i=1 (xj xi ) j >i=1 (yj yi )
(A.1)
Here, zi = xi for i = 1, . . . , Nf , and zNf +i = yi for i = 1, . . . , Nb . Furthermore, the

symbol J (zi ) I0 (xi ) for i = 1, . . . , Nf and J (zNf +i ) K0 (yi ) for i = 1, . . . , Nb . In
order to keep track of the indices let us consider the last two fermionic flavors as mass
degenerate, that is xNf = xNf +1 . In this case the partition function is equal to

ZNf +1,Nb xNf , {xf }f =1,...,Nf {yb }
det[ANf +1,Nb (xNf , {xf }f =1,...,Nf |{yb })]
,
Nf 1 2
Nb
2
2
2
2
2
j >i=1 (xj xi ) i=1 (xNf xi )2xNf
j >i=1 (yj yi )
= N
f
(A.2)
97
where we have introduced the (Nf + 1 + Nb ) (Nf + 1 + Nb ) matrix

ANf +1,Nb xNf , {xf }f =1,...,Nf {yb }
N +N
x1 I0 (x1 )
x1f b I0 (x1 )
I0 (x1 )
..
..
.
.
Nf +Nb
I0 (xNf )
xNf
I0 (xNf )
Nf +Nb
.
xN I0 (xNf )
I
(x
)
x
0
N
x
N
f
Nf
f
f
N +N
K0 (y1 )
y1 K0 (y1 )
y1f b K0 (y1 )
..
..
.
.
Nf +Nb
K0 (yNb )
yNb
K0 (yNb )
(A.3)
The proof of the graded Toda lattice equation follows from the Sylvester identity [44] valid
for the determinant of any matrix
Ci,j Cp,q Ci,q Cp,j = det(A)Ci,j ;p,q ,
(A.4)
where Ci,j is the cofactor of matrix element ij

Ci,j
det(A)
,
Aij
(A.5)
and Ci,j ;p,q is the double cofactor of matrix elements ij and pq

Ci,j ;p,q
2 det(A)
.
Aij Apq
(A.6)
Setting i = Nf + 1, j = Nf + 1 + Nb , p = Nf , and q = Nf + Nb we have

CNf +1,Nf +1+Nb CNf ,Nf +Nb CNf +1,Nf +Nb CNf ,Nf +1+Nb
= det(ANf +1,Nb )CNf +1,Nf +1+Nb ;Nf ,Nf +Nb .
(A.7)
From the structure (A.3) of the matrix ANf +1,Nb it follows that (Nf +Nb is defined in (16))
CNf +1,Nf +1+Nb = (1)Nb det ANf ,Nb ,
CNf ,Nf +Nb = (1)Nb
1
xNf
xNf Nf +Nb det ANf ,Nb ,
CNf +1,Nf +Nb = (1)Nb +1 Nf +Nb det ANf ,Nb ,

CNf ,Nf +1+Nb = (1)Nb +1
1
xNf
xNf det ANf ,Nb ,
CNf +1,Nf +1+Nb ;Nf ,Nf +Nb = det ANf 1,Nb .
(A.8)
98
For completeness we write out
I0 (x1 )
..
I (x )

0 Nf
ANf ,Nb {xf }{yb }
K0 (y1 )
..
x1 I0 (x1 )
y1 K0 (y1 )
K0 (yNb )
N 1+Nb
x1f
I0 (x1 )
Nf 1+Nb
xNf
I0 (xNf )
. (A.9)
Nf 1+Nb
y 1
K0 (y1 )
..
.
..
.
N 1+Nb
yNfb
K0 (yNb )
Inserting this in (A.7) we get
[det ANf ,Nb ]2 xNf Nf +Nb log[det ANf ,Nb ]

= xNf det ANf +1,Nb det ANf 1,Nb .
(A.10)
Finally, using the relation (A.2) between the partition functions and the matrices A we
obtain the desired graded Toda lattice equation

Nf +Nb xi log ZNf ,Nb {xf }{yb }
= 2xi2
ZNf +1,Nb (xi , {xf }|{yb })ZNf 1,Nb ({xf }f =i |{yb })
[ZNf ,Nb ({xk }|{yb })]2
(A.11)
Note that the Vandermonde determinants, ({zi2 }), in the squared masses do not contribute
to the left-hand side of the Toda lattice equation since Nf +Nb log[({zi2})] is a number.
In the case where n flavors have mass xNf the proof is similar. The partition function is
now given by

Zn,Nf 1,Nb xNf , {xf }f =Nf {yb }
det[An,Nf 1,Nb (xNf , {xf }f =Nf |{yb })]
,
Nf 1 2

2
2
2 n1 (2x )n(n1)/2 Nb
2
2
Nf
j >i=1 (xj xi ) i=1 (xNf xi )
j >i=1 (yj yi )
(A.12)
= N
f
where we have introduced the (Nf + n 1 + Nb ) (Nf + n 1 + Nb ) matrix

An,Nf 1,Nb xNf , {xf }f =Nf {yb }
I0 (x1 )
x1 I0 (x1 )
.
.
I0 (xNf )
1
1! xN I0 (xNf )
f
..
.
1
n1
(n1)! xNf I0 (xNf )
y1 K0 (y1 )
K0 (y1 )
..
K0 (yNb )
N +n2+Nb
x1f
99
I0 (x1 )
Nf +n2+Nb
xNf
I0 (xNf )
N
+n2+N
f
b
1
I0 (xNf )
1! xNf xNf
..
.
.
N
+n2+N
f
b
1
n1
I0 (xNf )
(n1)! xNf xNf
Nf +n2+Nb
y 1
K0 (y1 )
..
.
..
.
N +n2+Nb
yNfb
K0 (yNb )
(A.13)
Using again the Sylvester identity, (A.4), this time with i = Nf + n 1, j = Nf + n 1 +
Nb , p = Nf + n 2, and q = Nf + n 2 + Nb we find the graded Toda lattice equation

Nf +Nb xi log Zn,Nf 1,Nb xi , {xf }f =i {yb }
= 2nxi2
Zn+1,Nf 1,Nb (xi , {xf }f =i |{yb })Zn1,Nf 1,Nb (xi , {xf }f =i |{yb })
[Zn,Nf 1,Nb (xi , {xk }f =i |{yb })]2
(A.14)
Appendix B. The graded Toda lattice equation where two indices are affected
In this appendix we consider the graded Toda lattice equations (19) and (20) relevant for
the two-point functions. Although we have been able to check these equations in a number
of cases including those relevant for quenching the Toda lattice equation we do not have a
general proof.
The simplest possible example is
x y log Z1,1 (x|y) = 4xy
Z2 (x)Z2(y)
.
[Z1,1 (x|y)]2
(B.1)
This result follows by rewriting the l.h.s. of this equation as

2
Z1,1 (x|y)
x Z1,1 (x|y)
Z1,1 (x|y) x y log Z1,1 (x|y) = det
, (B.2)
y Z1,1 (x|y) y x Z1,1 (x|y)
and using the identity

x Z1,1 (x|y)
Z1,1 (x|y)
y Z1,1 (x|y) y x Z1,1 (x|y)

1
0 1
Z1 (y) y Z1 (y)
Z1 (x) x Z1 (x)
=
,
y Z1 (y) y2 Z1 (y)
1 0
2 x Z1 (x) x2 Z1 (x)
(B.3)
100
Table 1
Table 2
Eq. (B.6)
Nf
Nb
Eq. (B.7)
Nf
Nb
Case 1
Case 2
Case 3
Case 4
Case 5
Case 6
1
2
2
2
3
1
1
2
2
2
1
3
1
1
2
2
3
1
1
1
1
2
1
3
Case 1
Case 2
Case 3
Case 4
Case 5
any
2
3
4
4
0
2
1
1
1
any
1
2
2
3
any
1
1
2
1
which valid if the partition function is of the form

1
Z1,1 (x|y) = Z1 (x)y Z1 (y) + x Z1 (x)Z1 (y) .
(B.4)
2
This structure of the supersymmetric partition function easily follows by integrating the
bosonic variables before the Grassmann variables,

Z1,1 (x|y) = d d I0 x(1 + /2) K0 y(1 /2)
=

1
I0 (x)y K0 (y) + x I0 (x)K0 (y) .
2
(B.5)
The general case. Our conjecture is that the generalization of the above graded Toda
lattice equation is given by

xi yj log ZNf ,Nb {xf }{yb }
ZNf +1,Nb +1 (xi , {xf }|{yb }b=j )ZNf 1,Nb 1 ({xf }f =i |yj , {yb })
= 4nmxi2 yj2
,
[ZNf ,Nb ({xf }|{yb })]2
(B.6)
where n is the degeneracy of xi and m is the degeneracy of yj in ZNf ,Nb ({xf }|{yb }),
respectively. We have explicitly checked this equation for the cases listed in Table 1.
A similar equation holds when both derivatives are with respect to fermionic masses. It
is conjectured to be given by

xi xj log ZNf ,Nb {xf }{yb }
ZNf +2,Nb (xi , xj , {xf }|{yb })ZNf 2,Nb ({xf }f =i,j |{yb })
= 4nmxi2 xj2
(B.7)
,
[ZNf ,Nb ({xf }|{yb })]2
where n is the degeneracy of xi and m is the degeneracy of xj in ZNf ,Nb ({xf }|{yb }), in
this order. We have verified this identity for the following cases, see Table 2.
References
[1] S.F. Edwards, P.W. Anderson, J. Phys. F 5 (1975) 965.
[2] K.B. Efetov, Phys. Rev. Lett. 79 (1997) 491;
K.B. Efetov, Adv. Phys. 32 (1983) 53;
K.B. Efetov, Supersymmetry in Disorder and Chaos, Cambridge Univ. Press, Cambridge, 1997.
[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]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
101
M. Mzard, G. Parisi, M.A. Virasoro, Spin Glass Theory and Beyond, World Scientific, Singapore, 1987.
J.J.M. Verbaarschot, M.R. Zirnbauer, Ann. Phys. 158 (1984) 78.
F.J. Wegner, Z. Phys. 49 (1983) 297.
P.H. Damgaard, K. Splittorff, Nucl. Phys. B 572 (2000) 478;
P.H. Damgaard, K. Splittorff, Phys. Rev. D 62 (2000) 054509;
P.H. Damgaard, Phys. Lett. B 476 (2000) 465.
J.W. Negele, H. Orland, Quantum Many-Particle Systems, AddisonWesley, Reading, MA, 1998.
R. Boels, J. de Boer, R. Duivenvoorden, J. Wijnhout, JHEP 0403 (2004) 009.
H. Leutwyler, Phys. Rev. D 49 (1994) 3033.
J. Gasser, H. Leutwyler, Ann. Phys. 158 (1984) 142;
J. Gasser, H. Leutwyler, Nucl. Phys. B 250 (1985) 465;
H. Leutwyler, Ann. Phys. 235 (1994) 165.
J.J.M. Verbaarschot, H.A. Weidenmller, M.R. Zirnbauer, Phys. Rep. 129 (1985) 367.
M.R. Zirnbauer, J. Math. Phys. 37 (1996) 4986.
J.C. Osborn, D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 540 (1999) 317.
P.H. Damgaard, J.C. Osborn, D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 547 (1999) 305.
J.J.M. Verbaarschot, M.R. Zirnbauer, J. Phys. A 18 (1985) 1093.
A. Kamenev, M. Mzard, J. Phys. A 32 (1999) 4373;
A. Kamenev, M. Mzard, Phys. Rev. B 60 (1999) 3944.
I.V. Yurkevich, I.V. Lerner, Phys. Rev. B 60 (1999) 3955.
M.R. Zirnbauer, cond-mat/9903338.
E. Kanzieper, Phys. Rev. Lett. 89 (2002) 250201.
K. Splittorff, J.J.M. Verbaarschot, Phys. Rev. Lett. 90 (2003) 041601.
K. Splittorff, J.J.M. Verbaarschot, Nucl. Phys. B 683 (2004) 467.
E. Kanzieper, cond-mat/0312006.
M. Toda, Theory of Nonlinear Lattices, second ed., in: Springer Series in Solid-State Sciences, vol. 20,
Springer, Berlin, 1989.
Y.V. Fyodorov, Nucl. Phys. B 621 (2002) 643.
S. Weinberg, Phys. Rev. 166 (1968) 1568.
J. Gasser, H. Leutwyler, Phys. Lett. B 188 (1987) 477.
H. Leutwyler, A. Smilga, Phys. Rev. D 46 (1992) 5607.
E.V. Shuryak, J.J.M. Verbaarschot, Nucl. Phys. A 560 (1993) 306;
J.J.M. Verbaarschot, Phys. Rev. Lett. 72 (1994) 2531.
D. Dalmazi, J.J.M. Verbaarschot, Nucl. Phys. B 592 (2001) 419.
R.C. Brower, P. Rossi, C.-I. Tan, Nucl. Phys. B 190 (1981) 699;
R.C. Brower, M. Nauenberg, Nucl. Phys. B 180 (1981) 221.
J.B. Kogut, M. Snow, M. Stone, Nucl. Phys. B 200 (1982) 211.
R. Dijkgraaf, H. Verlinde, E. Verlinde, Nucl. Phys. B 348 (1991) 435.
A. Mironov, A. Morozov, G. Semenoff, Int. J. Mod. Phys. A 11 (1996) 5031.
Y.V. Fyodorov, E. Strahov, Nucl. Phys. B 647 (2002) 581.
Y.V. Fyodorov, E. Strahov, J. Phys. A 36 (2003) 3203.
Y.V. Fyodorov, G. Akemann, JETP Lett. 77 (2003) 438.
A.D. Jackson, M.K. Sener, J.J.M. Verbaarschot, Phys. Lett. B 387 (1996) 355.
T. Guhr, T. Wettig, Nucl. Phys. B 506 (1997) 589.
A.B. Balantekin, Phys. Rev. D 62 (2000) 085017.
P. Forrester, Log-gases and random matrices, Web Bookavailable at http://www.ms.unimelb.edu.
au/~matpjf/matpjf.html.
K. Okamoto, Funkcial. Ekvac. 30 (1987) 305;
P.J. Forrester, N.S. Witte, Commun. Math. Phys. 219 (2001) 357;
P.J. Forrester, N.S. Witte, Commun. Pure Appl. Math. 55 (2002) 679;
P.J. Forrester, N.S. Witte, math-ph/0204008;
M. Adler, P. van Moerbeke, Ann. Math. 153 (2001) 149;
M. Adler, P. van Moerbeke, Commun. Pure Appl. Math. 54 (2001) 153;
P. van Moerbeke, Random Matrices and their Applications, MSRI-publication #40, Cambridge Univ. Press,
2001.
102
[42] S. Kharchev, A. Marshakov, A. Mironov, A. Morozov, A. Zabrodin, Nucl. Phys. B 380 (1992) 181.
[43] D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 603 (2001) 343.
[44] J.J. Sylvester, C. R. Acad. Sci. 54 (1862) 129.

Topological masses from broken supersymmetry

Ignatios Antoniadis a,1 , Tomasz R. Taylor b
a CERN Theory Division, CH-1211 Geneva 23, Switzerland
b Department of Physics, Northeastern University, Boston, MA 02115, USA
Received 6 April 2004; accepted 11 June 2004
Abstract
We develop a formalism for computing one-loop gravitational corrections to the effective action
of D-branes. In particular, we study bulk to brane mediation of supersymmetry breaking in models
where supersymmetry is broken at the tree-level in the closed string sector (bulk) by Scherk
Schwarz boundary conditions, while it is realized on a collection of D-branes in a linear or non-linear
way. We compute the gravitational corrections to the fermion masses m1/2 (gauginos or goldstino)
induced from the exchange of closed strings, which are non-vanishing for world-sheets with Euler
characteristic 1 (genus 3/2) due to a string diagram with one handle and one hole. We show that
the corrections have a topological origin and that in general, for a small gravitino mass, the induced
mass behaves as m1/2 g 4 m3/2 , with g the gauge coupling. In generic orbifold compactifications
however, this leading term vanishes as a consequence of cancellations caused by discrete symmetries,
and the remainder is exponentially suppressed by a factor of exp(1/ m23/2 ).
1. Introduction
There are many important reasons for studying gravitational corrections to the effective
actions describing D-brane excitations. One of the phenomenologically-oriented goals
presenting quite a challenging theoretical problem is to understand how supersymmetry
breaking in the bulk can possibly be communicated to the gauge theory of supersymmetric
branes. The world-sheet configuration that mediates such a supersymmetry breaking is
a bordered Riemann surface with one hole and one handle (or two cross-caps in the
E-mail address: taylor@neu.edu (T.R. Taylor).
1 On leave from CPHT Ecole Polytechnique (UMR du CNRS 7644) F-91128, Palaiseau.
doi:10.1016/j.nuclphysb.2004.06.025
104
I. Antoniadis, T.R. Taylor / Nuclear Physics B 695 (2004) 103131
presence of orientifolds) and has Euler characteristic 1: we call it a genus 3/2 surface.
In this work, we develop a formalism for such computations and study the (Majorana)
fermion masses on branes generated from gravitational corrections in type II string
models with supersymmetry breaking realized via ScherkSchwarz boundary conditions
[15]. The gravitino mass is then proportional to the compactification scale 1/R, while
supersymmetry remains unbroken locally on the world-volume of branes transverse to the
ScherkSchwarz (SS) direction [5].
Actually, in type I theory, there is also a pair of orientifold (O) and antiorientifold (O)
planes formed, respectively, at the two endpoints of the compactification interval, requiring
in general the presence of branes and antibranes for canceling the RamondRamond (RR)
O
charge. One can then construct two types of D-brane configurations [6]: DO or D
preserving linearly half of the bulk supersymmetry on the brane world-volume [5], and DO
or DO with non-supersymmetric spectra but realizing non-linear supersymmetry [7,8]. Our
study of gravitational corrections to Majorana fermion masses applies both to gauginos of
the supersymmetric configurations, as well as to the goldstino of the non-supersymmetric
models.
Since the mediation of supersymmetry breaking from the bulk is a local phenomenon,
the possible existence of distinct branes and orientifolds far away in the bulk can only
lead to a secondary subleading effect. Thus, for simplicity and without loss of generality,
we will restrict our analysis to the oriented string case, the generalization in the presence
of orientifolds being straightforward. Moreover, for the same reason, we will ignore the
possible existence of distinct antibranes which are required by RR charge conservation
in the compact case. Note that the braneantibrane system is not affected by the SS
deformation. In any case, their presence can be avoided if some other direction of the
transverse space is non-compact, without affecting our results.
Now consider a fermion mass term generated by the two-point function at zero
momentum of two (tree-level) massless gauginos (or goldstinos) of the same (fourdimensional) chirality. The generic oriented string diagram that contributes to this
amplitude has g handles and h holes corresponding to D-brane boundaries. Obviously, one
must have h 1 in order to insert the gaugino vertices. The power of the string couplings
is determined by the Euler characteristic = 2 2g h. A simple inspection of the
internal N = 2 world-sheet superconformal charge, of which each gaugino vertex carries
3/2 units, shows that one needs at least three world-sheet supercurrent insertions to ensure
charge conservation. This implies that independently of the source of bulk supersymmetry
breaking, the lowest order diagram that can give non-vanishing contribution has = 1
and genus g + h/2 = 3/2. In the oriented string case, there are two such diagrams: one
with three boundaries (g = 0, h = 3) and one with a handle and a boundary (g = 1, h = 1).
In the framework we described above, all boundaries are of the same type and thus
the first diagram is supersymmetric and can be ignored. We are therefore left with the
(g = 1, h = 1) surface that contains the information about gravitational interactions. This
world-sheet configuration will be studied in this work in great detail.
The main result of our analysis is that the gaugino mass is determined by an amplitude
closely related to the topological partition function F2 [9,10]. Furthermore, for smooth
compactifications with at most N = 2 supersymmetry on the branes, we obtain, in the
limit of low-energy supersymmetry breaking, a fermion mass proportional to the gravitino
105
mass m3/2 . On the other hand, within the effective field theory, the one loop gravitational
correction leads obviously to two inverse powers of the Planck mass which can be canceled
only if the momentum integral is quadratically divergent. Thus the leading contribution
computes precisely the coefficient of this divergence, cutoff by the string scale. However,
in the case of orbifolds, this term vanishes as a result of cancellations caused by discrete
symmetries, and the remainder is suppressed exponentially as exp(1/ m23/2).
The paper is organized as follows. In Section 2, we describe the moduli space of the
relevant genus 3/2 string diagram (g = 1, h = 1), obtained by an appropriate involution
of the genus 2 (g = 2, h = 0) surface. We also obtain the left-over modular group and the
corresponding fundamental domain of integration. In Section 3, we derive the partition
function modified by the SS deformation and study the degeneration regions associated
to the large radius limit. In Section 4, we compute the amplitude that determines the
gaugino mass. We point out that it is closely related to the amplitude corresponding to the
topological F2 term in type II theory. To simplify our discussion, we assume that the action
of the SS twist is limited to the two-torus part of a T 2 K3 compactification manifold.
Before discussing the general K3 , we consider orbifolds and find that the amplitude is
zero in such compactification limits. We also derive a general expression valid for all K3
compactifications. In Section 5, we discuss the world-sheet degeneration limit relevant to
large radius compactifications and present explicit expressions for theta functions, primeform etc. As a first step toward constructing more explicit and calculable examples,
and in order to gain more insight into the general case, we reconsider orbifolds in
Section 6, but now in the presence of non-vanishing VEVs for the blowing-up modes. In
Section 7, we overcome another obstaclea cancellation due to the Z2 symmetry of the SS
mechanism, which occurs in the simplest form of SS compactificationsby considering its
ZN generalization la Kounnas and Porrati [3]. This allows extracting the leading radius
dependence of the gaugino mass and thus its behavior for small gravitino mass. Finally,
Section 7 contains our conclusions and outlook. Some technical details are relegated to
Appendices AC.
2. Bordered (g = 1, h = 1) Riemann surface from the involution of (g = 2, h = 0)
In this section, we study the genus 3/2 Riemann surface with one handle and one
boundary, see Fig. 1. The cycles a and b shown in Fig. 1 are the canonical homology
basis; the boundary c is homologous to ab1 a1 b. Such a surface can be obtained from
the double torus of genus 2 by applying the world-sheet involution I that exchanges left
and right movers, according to Fig. 2. Its action consists of a reflection with respect to the
plane of the dividing geodesics c, which in the canonical homology basis interchanges a
and b cycles, while simultaneously reverts the orientation of the later [11]: a1 a2 and
b1 b2 . It follows that the involution symplectic matrix I , acting in order on blocks of
a and b cycles, takes the form:

1
0
0
I
(2.1)
,
=
C
0 1
where is a symmetric matrix of integers with 2 = 1. In our case, it is given by the
2 2 Pauli matrix 1 . A period matrix invariant under the involution (2.1) satisfies [11]:
106
Fig. 1. Bordered g = 1 surface with h = 1 boundary.
Fig. 2. The same (g = 1, h = 1) surface obtained by a mirror involution of g = 2.
= I () = (C ) = 1 1 . It can thus be put in the form

il
=
,
il
(2.2)
where = 1 + i2 ; it depends on three real parameters 1 , 2 and l.

The relative modular group, defined by the modular transformations Sp(4, Z) that
preserve the involution, was studied in general in Ref. [12]. In the case under consideration,
it consists of the usual genus-one transformations that act simultaneously on the two
tori of the double cover. These are generated by the familiar S and T transformations;
S interchanges a with b cycles, up to a sign dictated by the involution: a1 b1 and
a2 b2 ; on the other hand, T leaves invariant the a cycles and shifts the b cycles by
b1 b1 + a1 and b2 b2 a2 . In matrix notation, they are given by

0 3
1 0
S=
(2.3)
;
T=
,
3 0
3 1
which, when applied to the period matrix (2.2), yield:
T : + 1, l l;
S :
, l 2
,
2
l
| | l 2
| |2
(2.4)
107
Fig. 3. The shaded region is the (g = 1, h = 1) fundamental domain.
or equivalently:
T : + 3;
S : 3 1 3 .
(2.5)
= | |2
Positivity of the period matrix implies that 2 and det

are positive.
The fundamental domain of integration can be easily derived from (2.4). In the -plane,
for fixed
l, it consists of the part of the strip 1/2 < 1 < 1/2 which lies above the circle of
radius 1 + l 2 , such that | |2 > 1 + l 2 (see Fig. 3). Obviously, the parameter l determines
the size of the hole. In the limit l 0, the surface becomes a regular torus and the modular
transformations (2.4) are reduced to the usual SL(2, Z) transformations + 1 and
1/ with their familiar fundamental domain. Note that (2.4) preserve the sign of l
which remains undetermined. Actually, before the involution, there is no constraint on the
sign of the non-diagonal element of the period matrix 12 . However, after the involution,
we show below that the sign of l is fixed to be positive [13].
We use the basis of two holomorphic differentials i on the genus-two surface ,
normalized to the a cycles:

i = ij ;
i = ij ;
i j = Im ij .
(2.6)
aj
l2
bj
Consider
now the surface 1 obtained from by the involution (2.1), and evaluate
i
j using the identity [14]:
1

z
i
(2.7)
i j =
i j (a1 b1 ) + i (z) j .
2
1
a1
b1
From the involution (2.1), one deduces that on the boundary i = ij j and thus
1,2 (x) = 2,1 (x),
x c.
(2.8)
108
Using this property, one has

i j = i j j1k k
1
i
=
2

i
a1

j j1k k (a1 b1 ) ,
(2.9)
b1
since the boundary contribution vanishes. Using now the relations (2.6) and (2.2), we find:

l
l
l
(2.10)
1 1 = 2 ;
2 2 = ;
1 2 = .
2
2
2
1
From the second relation above, it then follows that l is positive.
3. Partition functions
In this section, we compute bosonic and fermionic partition functions, as well as the SS
deformation, on the genus 3/2 (g = 1, h = 1) Riemann surface that was described above.
Following the analysis of Ref. [11], the bosonic determinants can be obtained by taking the
appropriate square root of the corresponding expression on the genus 2 double cover, up
to a correction factor R that depends on the involution and on the boundary conditions,
Neumann (N) or Dirichlet (D):2
g = 2: (det Im )1/2|Z1 |1
1/2
(g = 1, h = 1): (R1 det Im )1/4Z1
(3.1)
where Z1 is the chiral part of the determinant, and obviously after the involution the period
matrix has the reduced form (2.2). The correction factor is given by

D 1
1+
2 + l
1
N
1
R1 = R1
(3.2)
Im +
(Im )
,
=
= det
2
2
2 l
where is defined from the form of the involution (2.1) and equals 1 in our case. As a
1/2
result, the bosonic determinant on 1 becomes (2 l)1/2 Z1 , where the plus (minus)
sign corresponds to Neumann (Dirichlet) boundary conditions. Note that the correction
factor (3.2) is invariant under the modular transformations (2.4). It will be also useful to
define

1 + i 22 l 2 ,
(3.3)
which has the usual form of SL(2, Z) transformations, + 1 and 1/ .
2 Note the difference in the definition of R between the two references of [11]; the one is inverse of the
other. Here, we use the definition of Bianchi and Sagnotti.
109
The generalization to a compact boson is straightforward. The classical action (see

Appendix A) is:
T

R 2
(m
+ n )(Im
)1 m
+ n T
2

R 2 2
=
m + 2mn1 + n2 | |2 l 2 ,
(2 l)
Scl =
(3.4)
= m(1, 1) and n = n(1, 1)

where R is the compactification radius in units, and m
are the winding numbers around the b and a cycles, respectively. Their form is fixed by the
involution, while the two signs correspond to N and D boundary conditions. Combining
(3.4) with the quantum determinant (3.1), and summing over all classical solutions, one
finds the total contribution to the partition function of a compact boson:
1/2
ZB = Z1
N,D
Zm,n
,
N,D
Zm,n
=
m,n
(m+n1 )
R
R 2
2 l
e
1/2
(2 l)
+n2 (2 l)
(3.5)
N,D
After performing a Poisson resummation in m, one can bring the partition function Zm,n
N,D
into the Hamiltonian form Zm,n
:
N,D
=e
Zm,n
i
2
2
2 (pL pR )
pL,R =
R N,D
nR N,D
with
R N,D
R
N,D 1/4
(R
)
1
(3.6)
where was defined in (3.3). Thus, the lattice partition function takes the familiar form
in terms of an effective radius modified by the correction factor (3.2). Obviously, in the
limit where the size of the hole vanishes, l 0, the correction factor becomes unity and
the partition function is reduced to its toroidal form. From the expression (3.6), it is also
easy to see that T-duality, R 1/R, exchanges N and D boundary conditions. Note that
R 1/R is also formally equivalent to l l.
3.1. Fermions and theta functions
The method of taking the square root from the double cover can be also applied to the
fermionic determinants giving rise to theta-functions. Each complex fermion leads to 16
spin structures corresponding to four boundary conditions a = (a1 , a2 ) and b = (b1 , b2 )
along the two non-trivial cycles a and b of 1 for the left and right movers:

T
3 T
a
() =
(3.7)
ei( n+ a)( n+ a ) +2i( n+ a ) b .
b
n =(n1 ,n2 )
The insertion of 3 in the last phase was chosen for convenience, so that in the toroidal
limit l 0:

a
a1
a2
()
(3.8)
( )
( ).
b
b1
b2
110
It follows that under the modular transformations (2.5), transforms as the product of the
1-loop thetas :

a
a
3
ia1 (a1 +1)+ia2 (a2 +1)
(),

T: + =e
b + a + 12
b

a
b
S: 3 3 = det e2i(a1b1 a2 b2 )
(3.9)
().
b
a
The
partition function involves a sum over spin structures with appropriate coefficients
c ab , determined at one loop level. As usual, at higher loops the corresponding
coefficients are determined by the factorization properties of the vacuum amplitude. In
our case, it is sufficient to consider the l 0 limit, to deduce in a shorthand notation:

a
a1
a2
c = cL
(3.10)
c
,
b
b1 R b2
where the subscripts L, R denote the left and right movers in the torus amplitude.
3.2. The ScherkSchwarz deformation and the large radius limit
Here, we will consider the breaking of bulk supersymmetry via SS boundary conditions
along a direction transverse to the D-brane stack [15]. In the closed string sector,
it amounts to deform the partition function by coupling the lattice momenta to the
fermionic spin structures, imposing antiperiodic boundary conditions to spacetime
fermions consistently with world-sheet modular invariance:

a + n1
a
a
2in(b1 b2 )
SS

= s ,

e
b + m1
b
b
sSS = e2i[m(a1 a2 )n(b1 b2 )] ,
(3.11)
of (3.5). The choice

where 1 (1, 1), and m, n are the winding numbers appearing in
of Dirichlet conditions follows from the transversality of the SS direction to the boundary.
Thus, the SS deformation consists of inserting the phase sSS (3.11) in the partition function.
In the zero winding sector, n = 0, upon Poisson resummation in m, it amounts to shift the
KaluzaKlein momenta m
in (3.6) of spacetime fermions (a1 a2 = 1/2) by 1/2, giving
a mass to the gravitino 1/2R. In the odd winding n sector, on the other hand, there is a
change in the GSO projection due to the additional phase.
We consider now the large radius limit R , where supersymmetry is restored.
D , corresponding to the lower sign of (3.5). Since
The relevant partition function is Zm,n
at the origin of the lattice, m = n = 0, the SS phase becomes trivial, the contribution to
the gaugino mass should vanish by supersymmetry. Thus, for finite moduli (2 and l) the
result is exponentially suppressed in R. To avoid exponential suppression, n should vanish
and 2 should go to infinity. This limit corresponds to the handle degeneration, describing
a massless closed string exchange in the loop, and thus gravitational corrections in the
effective field theory.
Note that for Neumann boundary conditions, one should get an additional contribution
from all winding numbers n = 0 in the limit 2 , l with 2 l 0, so that 22 l 2
D
Zm,n
111
remains fixed. This degeneration describes the non-gravitational two-loop non-planar

diagram in the gauge theory of the D-branes. Obviously, this limit is not relevant for our
purposes, since it does not contain any information on the mediation of supersymmetry
breaking from the bulk to the boundary.
4. Gaugino mass as a descendant of a topological amplitude

In this section, we evaluate the two-point function involving two gauginos at zero
momentum coupled to the boundary of a (g = 1, h = 1) surface 1 . We consider the
case with N = 2, D = 4 supersymmetry on the boundary. Thus the D = 10 type II
theory is compactified on the six-dimensional internal space K3 T 2 , where T 2 contains
the ScherkSchwarz circle transverse to the boundary direction. The K3 manifold is
described by a (4, 4) superconformal field theory, yielding N = 4 supersymmetry in the
bulk, spontaneously broken by the SS boundary conditions. As it becomes clear in the
following, the computation of the gaugino mass is closely related to the computation of the
genus 2 topological amplitude F2 that determines the gravitational coupling W 4 in Calabi
Yau compactifications of type II theory [10]. Before a full-fledged discussion of K3, it is
very instructive to consider its orbifold limit, with K3 realized as T 4 /H , where H = Z2
reflection (z z) or another ZN symmetry of T 4 .
4.1. The orbifold case
The gaugino vertex operator of definite chirality , at zero momentum, in the canonical
1/2 ghost picture, reads:
V(1/2)(x) = :e/2 S Sint :,
(4.1)
where x is a position on the boundary of the world-sheet, is the scalar bosonizing

the superghost system, and S (Sint ) is the spacetime (internal) spin field. Upon
complexification of the four spacetime fermionic coordinates, I for I = 1, 2, and
introducing the bosonized scalars I = eiI , one has
i
S = e 2 (1 +2 ) ,
= .
Similarly, bosonizing the fermionic coordinates of

3 and 4,5 , respectively, one has:
i
Sint = e 2 (3 +4 +5 ) .
(4.2)
T2
and K3, by introducing the scalars

(4.3)
In order to compute the amplitude involving two fermions (4.1) at the boundary of a
(g = 1, h = 1) surface, one has to insert three picture changing operators e TF , where TF
is the world-sheet supercurrent. We choose to insert all these operators on the boundary c.
One of them is needed to change the ghost-picture of one gaugino to +1/2, while the other
two arise from the integration over the supermoduli:

d()
dx dy A,
m1/2 = gs2
F (1 )
112

A=
(1/2)
(1/2)
V+
(x)V
(y)
3

e TF (za ) ,
(4.4)
a=1
where gs is the string coupling. Its square includes gs associated to Euler characteristic
1 as well as an extra gs from the normalization of gaugino kinetic terms on the disk.
The moduli integration is over the fundamental domain F (1 ), shown in Fig. 3, with the
appropriate measure d(). The dependence on the positions za of the picture changing
operators is gauge artifact and should disappear from the physical amplitude.
By internal charge conservation, since both gauginos carry charge +1/2 for 3 , 4
and 5 , only the internal part of the world-sheet supercurrents, TFint contributes; each
TF should
provide 1 charge for 3 , 4 and 5 , respectively. In the orbifold case,
TFint = 5I =3 I XI + c.c., where X3 and X4,5 are the complexified coordinates of T 2
and K3, respectively. The later transform under an element h of the orbifold group H as
X4 hX4 and X5 h1 X5 . The amplitude (4.4) then becomes:
2

5

eiI /2 (x)eiI /2 (y)
e (zI )
A = e/2 (x)e/2 (y)
I =3
I =1
5

i /2

e I (x)eiI /2 (y)eiI (zI ) ,
(4.5)
I =3
where {zI }, I = 3, 4, 5, is a permutation of {za }, a = 1, 2, 3, and an implicit summation

over all permutations is understood.
Performing the contractions for a given spin structure s, one finds:

s2 ( 12 (x y)) 5I =3 s,hI ( 12 (x + y) zI )XhI (zI )
As =

s ( 12 (x + y) 5I =3 zI + 2)
Z2
(x) (y)
3,4,5
(4.6)
sSS Zlat ,
5
5
4
2
Z1 I =3 Z1,hI
I <J E(zI , zJ ) I =3 (zI )
where s is the genus-two theta-function of spin structure s, E is the prime form, is a onedifferential with no zeros or poles and is the Riemann -constant [15,16]. Z1,hI is the
(chiral) non-zero mode determinants of the hI -twisted (1, 0) system, with h3 = 1, h4 =
h, h5 = h1 , while Z1 Z1,1 , and Z2 is the chiral non-zero mode determinant of the
(2, 1) bc ghost system. Finally, Zlat stands for all zero-mode parts of spacetime and
internal coordinates. In our case, it is given by:
Zlat =
1
Z D Zrest ,
(2 + l)2 m,n
(4.7)
where an implicit summation over m, n should be performed, taking into account also sSS
and the XI factors in (4.6). Recall that sSS is the phase (3.11) of the SS deformation,
D is the momentum lattice of the SS circle (3.5), satisfying Dirichlet conditions at the
Zm,n
boundary, while Zrest denotes the remaining zero-mode contribution of the K3 lattice,
together with the additional (non-SS) direction of T 2 .
The large radius limit along the SS direction was described in the previous section.
In this limit, the amplitude is exponentially suppressed unless 2 with l fixed,
113
corresponding to the effective field theory gravitational loop exchange. Moreover, the
momentum sum is restricted to vanishing winding number n. Then as

sSS = (1)2m a1 ,
R :
Zlat
1
Z D Zrest (2 ).
(2 + l)2 m,0
(4.8)
Note that for m even, the SS deformation becomes trivial hence, as shown below, the result
vanishes by supersymmetry; the SS lattice sum is in fact restricted to the odd winding
numbers m.
In order to perform an explicit sum over spin structures, we choose the positions of the
picture-changing operators satisfying the condition [10]
I =5
zI = y + 2.
(4.9)
I =3
Then the spin structure-dependent part of the amplitudes (4.6) simplifies, yielding the
following sum:
SO =

5
1
1
a
a

(x y)
(x + y) zI ,
(1)2m a1
hI
b
b
2
2
(4.10)
I =3
a ,b
where the theta-function subscripts hI denote additional shifts in their characteristics

associated to the respective orbifold group elements. By using the Riemann identity [16],
we obtain

SO =

5

0
0
(x
1
m
m (zI ).
hI
21
21
(4.11)
I =3
Note that as anticipated, for m even, the above result vanishes by supersymmetry, due to
the Riemann vanishing theorem, (z ) = 0 z . For m odd, the amplitude becomes

A=

(x) (y)
0

1 (x ) 3,4,5
E(zI , zJ ) 5
21
5

I =3

h1
I
I <J
I =3
2 (z )
I

0
1 (zI )XhI (zI ) Zlat .
21
Z14
Z2
5
I =3 Z1,hI
(4.12)
As mentioned before, the amplitude (4.12) contains an implicit sum over 6 permutations
{zI } of the supercurrent insertion points {za }. Due to complete
antisymmetry of the

prefactor, this leads to the antisymmetrization of the product 3a=1 h1 (za )XhI (za )
I
in {za }. Recall that h3 = 1, h4 = h, h5 = h1 . For a T 4 /Z2 orbifold, h4 = h5 , hence the

amplitude vanishes after summing over all permutations. In fact, as shown in Section 6,
it also vanishes for all T 4 /ZN orbifolds. This cancellation is clearly due to discrete
symmetries, therefore it is not expected to hold for a generic K3 compactification.
114
4.2. The general K3 case

The case of a general K3 T 2 compactification can be discussed by slightly modifying
most of previous computations. The underlying N = 4 superconformal field theory (SCFT)
of central charge c = 4 is characterized by an SU(2) affine algebra at level one. This can
be realized in terms of a free boson compactified at the self-dual radius, that couples
also to the spin structure s. The internal part (4.3) of the gaugino vertex operator (4.1)
becomes:
Sint = ei3 /2+i/
(4.13)
while the internal part of the world-sheet supercurrent reads:

TFint = 3 X3 + ei/
+ c.c.,
G
(4.14)
has no singular operator product expansion with the U (1) current and carwhere G
ries a conformal dimension 5/4. The internal part of the amplitude corresponding to the
last product of contractions in (4.5) then becomes:

Aint = ei3 /2 (x)ei3 /2 (y)ei3 (z3 ) ei/ 2 (x)ei/ 2 (y)ei/ 2 (z4 )ei/ 2 (z5 )

(z5 ) ,
(z4 )G
G
(4.15)
giving rise to the following spin structure dependent amplitude:

As =
s2 ( 12 (x y))s ( 12 (x + y) z3 )Chr,s (x + y z4 z5 )

s ( 12 (x + y) 5I =3 zI + 2)
E 1/2(z4 , z5 ) (x) (y)
3,4,5
5
2
I <J E(zI , zJ ) I =3 (zI )
Z2
(z4 )G
(z5 )r .
4 sSS X3 (z3 )Zlat T 2 G
Z1
(4.16)
Here, the label r = 0, 1/2 denotes the two different types of spectral flow of the N = 4
SCFT, and a summation over r is implicit. Zlat T 2 contains the zero-mode parts in the partition function of spacetime and T 2 coordinates, while the K3 contribution is included in
the last factor which is also independent of the spin structure. Finally, Chr,s () is the spin
structure dependent SU(2) level one character:

T
3
a
T
1/2
v ) = Z1
Chr (
e2i( n+ r + a)( n+ r + a) +2i( n+ r + a) ( v+2b)
b
n

r + a

1/2
= Z1 (1)4b( r + a )
(2, v ).
(4.17)
0
In order to perform an explicit sum over spin structures, we choose here the positions
of the picture-changing operators satisfying the same condition as in (4.9). Then the
spin structure-dependent part of the amplitude (4.16) simplifies, yielding the following
sum:
SK3 =
115

(1)2m a1 (1)4b( r + a )
a ,b

1
1
a
a
(x y)
(x + y) z3

b
b
2
2

r + a
(2, x + y z4 z5 ),
0
(4.18)
where we used the SS factor sSS = (1)2m a1 relevant to the R limit, Eq. (4.8). Now
the summation over b can be performed by using the addition theorem [16]:

1
+
+
4

(1)
(u
+
u
)
(u1 u2 )
1
2
+

2g

+
=
(4.19)
(2, 2u1)
(2, 2u2),

with the result

r + a
r + a
SS
SK3 =
s
(2, x + y z4 z5 )
(2, x z3 )
0
0
a

r
(2, y z3 ).
0
Finally, after using the inversion of (4.19) [16], we obtain

0
0

SK3 = (1)2m r 1 m (x ) m (z3 )
21
21

r
(2, z4 + z5 2).
0
(4.20)
(4.21)
As expected, the above result vanishes by supersymmetry for m even, due to the
Riemann vanishing theorem, (z ) = 0 z . For m odd, the amplitude becomes

r

(z4 )G
(z5 ) E 1/2 (z4 , z5 ) (z4 ) (z5 )
A = (1)2 r 1
(2, z4 + z5 2) G
r
0
X3 (z3 )(z
3 ) 3,4,5
I <J
where
(x)
2i
Z2
(x)
(y)
9/2 Zlat T 2 ,
5
3
E(zI , zJ ) I =3 (zI ) Z1

0
1 (x ) (x)
21
(4.22)
(4.23)
is the holomorphic one-differential twisted by (1) along the two b-cycles of the genus 2
double-cover, i.e., twisted by (1) along the b-cycle of the bordered surface. In Eq. (4.22),
116
and in most cases below, we neglect some purely numerical (i.e., position- and moduliindependent) factors.
At this point, the amplitude (4.22) still seems to depend of the positions {za } of the
supercurrent insertions. However, it contains an implicit sum over 6 permutations {zI } of
these points. As a result, the factor

r

B(zI ) (1)2 r 1
(2, z4 + z5 2)
0

(z4 )G
(z5 ) E 1/2 (z4 , z5 ) (z4 ) (z5 )
G
r
X3 (z3 )(z
3)
(4.24)
yields a completely antisymmetric combination B(za ) that transforms as a quadratic

differential in each za , twisted by (1) along the two b-cycles, with first order zeroes
as za zb . This implies that
B(za ) = B det h a (zb ),
(4.25)
where B is constant (position independent) and h a are the zero modes of the twisted
(2, 1) bc system. For the latter, we can use the bosonization formula [15,17]
1,2,3

a=3
a=3

0
1/2
za 3
E(za , zb )
3 (za )Z1
= det h a (zb )Z 2 ,
1
(4.26)
1
2
a=1
a<b
a=1
where Z 2 is the non-zero-mode determinant of the twisted bc system. Finally, after using
all these relations, together with (4.9), we obtain:
A=
Z2
BZlat T 2 (x)
(y),
Z 2 Z15
(4.27)
which does not depend on the supercurrent insertion points.

In order to compute the gaugino mass, the amplitude (4.27) should be integrated
over the positions of the gaugino vertex operators and over the moduli of 1 . Since
x, y c ab1 a1 b,

(4.28)
(x)
= 2 (x).
We show in the appendix that, with the twisted differential normalized as in (4.23),

lim
(4.29)
(x)
= 1.
2
In this way, we obtain

Z2
BZlat T 2 ,
m1/2 = gs2 d()
Z 2Z13
(4.30)
which still remains to be integrated over the moduli of 1 . In the next section and in
Appendices A, B and C, we discuss the degeneration limit 2 , and derive explicit
expressions for the integration measure and the determinants.
117
5. Degeneration limit 2 , l fixed

In the limit 2 , l fixed, the two a-cycles of the double cover are pinched along
the handles, see Fig. 4. The limiting curve is a sphere with two pairs of points, (a, b)
identified as the punctures. This curve can be constructed from a two-sphere
and (a,
b),
by cutting out small disks on its surface (with diameters of order e2 ), and applying
the plumbing procedure for attaching the handles [16]. On the projective plane P1 , the
see Fig. 5. The boundary of 1
handles extend symmetrically from a to b and from a to b,
is mapped to the real axis, Im z = 0. The a-cycle winds around point a while the b-cycle
runs from b to a.
On , the normalized basis of holomorphic one-forms is given by [16]:
1 = (z; a, b),
a),
2 = (z; b,
Fig. 4. The 2 , l fixed degeneration limit of the double-cover.
Fig. 5. Degenerate surface on the projective plane.
(5.1)
118
where the SL(2, C)-covariant one-form

1
1
1
.
(z; p1 , p2 )
2i z p1 z p2
(5.2)
Note that on the real axis, 1,2 = 2,1 , as required by the involution (2.1). Furthermore,
Eq. (5.1) holds up to the order O(e2 ). Thus in the leading O(1) order, the generalized
Jacobi map becomes
za
z b
c1 (p), e2iz2 =
c2 (p),
(5.3)
zb
z a
where the multiplicative constants c1 and c2 depend on the base point p. They will cancel
inside all theta function arguments, therefore from now on we can set c1 (p) = c2 (p) = 1.
The off-diagonal period matrix element is determined by the SL(2, C)-invariant cross-ratio
of the four puncture points:
z P1
e2iz1 =
=
e2l = (a, b; a , b)
(a b)(b
a)
(a a)(b
b)
(5.4)
Note that as expected, l > 0, since a and b lie on the same half-plane of P1 .
Naively, the 2 limit of the theta function seems to be trivial [ (, z) 1],
unless one considers an argument that diverges in this limit. Indeed, the arguments of the
form z that we are interested in, have this property because i ii /2 . It is
convenient to write
z = z + (),
where the half-period
1
1
() = 1 =
2
2
il
il
(5.5)

(5.6)
and
= + ()
(5.7)
is the finite Riemann constant associated to the spin structure . In the leading order of the
theta series expansion,
(z + ) = 1 + el e2iz1 + el e2iz2 + e2i(z1 +z2 ) ,
with a remainder of order
O(e2 ).
(5.8)
With the Jacobian map given in (5.3), we obtain
zb
z a
+ el e2i2
za
z b
(z
b)(z
a)
.
+ e2i(1 +2 )
(z a)(z b)
(z ) = 1 + el e2i1
(5.9)
Then the Riemann vanishing theorem, (z ) = 0, z P1 , dictates

e2i1 = el
a b
,
b b
e2i2 = el
a b
,
a a
(5.10)
119
with l determined by Eq. (5.4).

For the prime-form, the limit can be taken in the same way as for the theta function, and
yields the obvious result:
E(x, y) = x y.
(5.11)
The differential (z) is more subtle. One can start from the quotient [15]
(z p1 p2 + ) E(w, p1 )E(w, p2 )
(z)
=
,
(w) (w p1 p2 + ) E(z, p1 )E(z, p2 )
(5.12)
where p1 and p2 are arbitrary points, and choose

p2 as the point associated to one of
0
the odd spin structures, say p2 = + 1/2 [18]. Now it is straightforward to take
a limit similar to the one that led to (5.9). It does not depend on p1 , and one finds
(z) (z b)1 (z a)
1 . The constant factor is fixed by SL(2, C) covariance, hence
(z) = (z; a,
b).
(5.13)
In Appendix B, we take into account Beltrami differentials dual to the integration

measure
d() = d1 d2 dl,
(5.14)
and we show that in the degeneration limit, the quantum determinants appearing in
Eq. (4.30) satisfy:
1
Z2
= 2l
.
3
e 1
Z 2 Z1
This leads to the gaugino mass

BZ 2
m1/2 = g 4
d1 d2 dl 2llat T ,
e 1
(5.15)
(5.16)
F ( )
where we replaced the string coupling by the gauge coupling g, with gs = g 2 .

6. Double degeneration limit l 0 on blown-up orbifolds
It is clear from Eq. (5.16) that the dominant contribution to the gaugino mass comes
from the region of small l. The l 0 limit corresponds to the double degeneration of
the surface, with a b on P1 . Then the surface splits into a small neighborhood of a
and b, of radius O(l), which comprises the degenerate torus, and the surrounding annulus
extending from the torus to the boundary; in the l 0 limit the annulus degenerates into a
disk with a puncture at z = a. As mentioned before, l cannot become smaller than e2 ,
which is the cutoff provided by the diameter of the degenerate handle.
The final result for the gaugino mass (5.16) depends critically on the small l behavior
of the constant B. If B vanishes as any positive power of l, then the integral over l
yields a 2 -independent constant. On the other hand, if B is constant, the integral depends
logarithmically on the cutoff, bringing a factor of 2 which, as we will see at the end,
120
will be converted to R 2 . Recall that B was introduced in Eq. (4.25), in the context of
K3 compactifications, and refers to a specific basis of two-differentials (C.3) given in
Appendix C.
Since the analysis of a general K3 turns out to be quite elaborate, we defer it to another
place and return to the orbifold limit. As mentioned before, the orbifold amplitude (4.12)
vanishes due to symmetry with respect to the permutations of the supercurrent insertions.
The simplest way to move away from the orbifold point is by inserting a vertex operator
for a massless blowing-up mode. However, first we explain why the amplitude is zero
for any ZN orbifold.
6.1. Vanishing amplitude on ZN orbifolds
After using the explicit parameterization of the bosonic zero modes, Eqs. (A.1) and
(A.5), and summing over all permutations of the supercurrent insertion points and over all
twisted sectors,3 one finds that the amplitude (4.12) contains the factor
BO (za ) =
N1
Lk LNk Li Bk,i (za ),
(6.1)
k=1 (ak ,bk ) i=1,2
where k labels the group element e2ik/N , (ak , bk ) labels the associated twists along the aand b-cycles, and

1 ), k Nk (z2 ), Nk k (z3 ) .
Bk,i (za ) = det i (z
(6.2)
First, consider twists that are non-trivial only along the b-cycle, as discussed explicitly
in Appendix A. Then the determinant (6.2) is exactly the same as the determinant
computed in Appendix C. For i = D it is zero, while for i = N it is the determinant
(C.4), with a k-dependent prefactor Im(h) = sin(2ik/N). Although Neumann boundary
conditions in the SS direction are beyond the scope of the present discussion, it is worth
mentioning that, even in that case, the final result would vanish, because the contribution
of the kth group element in (6.1) would be canceled by the (N k)th. For N even, the
remaining contribution of k = N/2 vanishes by the same argument as for the Z2 orbifold
mentioned in Section 4. Next, consider twists that are non-trivial only along the a-cycle.
The corresponding twisted differentials are easy to construct: they have the property that
k Nk Nk k , therefore the determinant (6.2) vanishes also in this case. Similar
property holds for twists that are non-trivial along both a- and b-cycles. Hence
BO (za ) = 0,
where the above equation holds to the same
holomorphic differentials.
(6.3)
O(e2 )
accuracy as the expansions of the
6.2. Blowing-up mode insertion

We now insert in the amplitude a blowing-up mode B at zero momentum, associated to
the twisted sector (h, h1 ) with h = e2i and = k/N . Its vertex in the 1 ghost picture
3 The contribution of the untwisted sector drops out after symmetrization.
121
is:
(1,1)
VB
(, ) = :ei(4 4 )+i(1)(5 5 ) 4 5 :,
(6.4)
where I is the corresponding twist field of conformal dimension (1 )/2 in both
left and right movers. Using the N = 2 world-sheet supercurrent, one finds the blowing-up
vertex operator in the 0-ghost picture:
(0,0)
VB
(, ) = :ei(1)(45 4 +5 ) 4++ 5 + ei(1)(45 )i(4 5 ) 4+ 5+
+ ei(4 5 )+i(1)(4 5 ) 4+ 5+
+ ei(4 5 4 +5 ) 4 5++ :,
(6.5)
where we used the short distance (OPE) expansions [19]: I (z, z )XI (w) (z
(w) (z w)
1+ I+ .4 Going from the to the +
w)1+ I+ and I (z, z )X
I
component of the twist field, its conformal dimension is increased by the corresponding
twist ( or 1 ).

(0,0)
After inserting the blowing-up vertex d 2 VB (, ), the K3 part of the amplitude,
corresponding to the last factor of Eq. (4.5), becomes:
i /2

e 4 (x)ei4 /2 (y)ei4 (z4 )ei(1)4 ( )ei(1)4 ( ) s X4 (z4 )4++ (, )

ei5 /2 (x)ei5 /2 (y)ei5 (z5 )ei(1)5 ( )ei(1)5 ( ) s X5 (z5 )5 (, )

+ ei4 /2 (x)ei4 /2 (y)ei4 (z4 )ei4 ( )ei 4 ( ) s X4 (z4 )4 (, )

ei5 /2 (x)ei5 /2 (y)ei5 (z5 )ei5 ( )ei 5 ( ) s X5 (z5 )5++ (, )

1
s,h4 (x + y) z4 (1 )( )
2

1
s,h5 (x + y) z5 + (1 )( )
2

E(z4 , )E(z5 , ) 1
1
E(z4 , )E(z5 , )
E(, )2(1)

X4 (z4 )4++ (, ) X5 (z5 )5 (, )

1
1
(x + y) z4 + ( ) s,h5
(x + y) z5 ( )
+ s,h4
2
2

E(z4 , )E(z5 , )
1
E(z4 , )E(z5 , ) E(, )2 2

X4 (z4 )4 (, ) X5 (z5 )5++ (, ) ,
(6.6)
where the expression after the
symbol replaces the product of the last two theta-functions
in Eq. (4.6), together with the I =4,5 XI factors from the two supercurrent insertions.
4 The leftright symmetric twist field ++ is called in Ref. [19].
122
The spin structure sum can be performed using the same gauge condition (4.9) with the
result:

0
0
h1 m z4 + ( ) h1 m z5 ( )
4
5
21
21

E(z4 , )E(z5 , )
1
X4 (z4 )4 (, ) h X5 (z5 )5++ (, ) h

2
4
5
E(z4 , )E(z5 , ) E(, )2
+ (4 5, 1 ),
(6.7)
which has to replace the last two terms in the product of Eq. (4.12).
We want to show that after integrating Eq. (6.7) over the vertex position , the result is
not symmetric under the transposition z4 z5 ; this will be true even in the untwisted
sector with h4 = h5 = 1. To that end, we can go directly to the double degeneration
limit l 0. Recall that the vertex position is to be integrated over the whole half-plane,
Im > 0. If one is interested only in terms that are non-vanishing in the l 0 limit, one
can neglect the small (radius O(l)) neighborhood of the puncture points. This means that
the contribution of the factorized torus is negligible and that the vertex is de facto inserted
on a hemisphere with one puncture point a. Then one can set = 0 in the arguments
of theta functions. Furthermore, the twist field correlators should be evaluated on such a
hemisphere. Due to the ZN symmetry of the lattice, only the quantum parts of these
correlators [20] contribute to the amplitude.5 Their form is completely determined by OPE
and SL(2, R) invariance [19,20]. Without going into much detail, it is clear that Eq. (6.7)
is not symmetric in z4 , z5 : in the last line, z4 z5 is accompanied by 1 , and this
asymmetry should survive the -integration. With the twist correlators non-vanishing on a
punctured hemisphere, the constant B(l = 0) will be non-zero on blown-up orbifolds.
7. Radius dependence of the gaugino mass

We return to the gaugino mass (5.16), now turning our attention to the dependence of the
amplitude on the string configuration on T 2 . Eq. (4.24) contains the respective zero mode,
X3 = X + iX , where X is the coordinate in the SS direction and X is the non-SS
coordinate on T 2 . For the zero-mode configurations parameterized as in Eq. (A.1),
X(z) = 2R
(7.1)
Li i (z),
i=1,2
where the constants Li depend on the winding numbers m and n, see Eq. (A.3). The SS
coordinate satisfies D boundary conditions and, as explained before, in the 2 limit
all n = 0 modes are exponentially suppressed by the partition function, hence
X =
mR
D (z),
i(2 l)
D (z) = 1 (z) + 2 (z).
(7.2)
5 Actually, the vanishing due to the Z symmetry can be avoided by turning on a non-diagonal modulus
N
mixing the two planes. However, the resulting contribution would be exponentially suppressed in the 2
limit.
123
Furthermore, non-trivial X modes are also exponentially suppressed in this limit, unless
the corresponding radius blows up in the same rate as R. Since the final result is the same,
we will not consider here such a fine-tuned scenario. In this way, we obtain
B m
R
,
2 l
(7.3)
where we ignored the factors associated to K3 which had already been discussed in the
previous section.
The zero modes are weighted by the torus partition function. After combining Eq. (7.3)
with (4.8) and (3.5), we obtain

m
R2
m2 R 2
.
BZlat T 2
(7.4)
exp
2 l
(2 l)3/2 (2 + l)2
It is clear that the amplitude vanishes after summing over the windings, due to the Z2
symmetry m m. Since this symmetry is absent in the presence of Wilson lines, one
could try to insert into the amplitude (4.27) a zero-momentum vertex for the Wilson line
modulus :

x
x
(x)
(y)
du X3 (u) (y)
(x)
(y)
(x)

(x)
(y)
D (x) D (y) ,
du D (u)
y
(7.5)
where in the second line we replaced X3 (u) by its zero mode (7.2), since it cannot be
contracted with other operators, while in the third line we used the property that D (u) is
a closed one-form to write D (u) dD (u). The above result obviously vanishes again
upon integration over x and y.
7.1. Generalized ScherkSchwarz deformation
To avoid the vanishing of the summation over windings of the SS circle, we consider
a ZN SS deformation which is a straightforward generalization of the Z2 case [3]. For
instance, consider compactifications on the product space of a K3 orbifold times T 2 , and
define the internal rotation current J , in one of the two complex coordinates of K3:
J = 4 4 + X4 X4 .
(7.6)
Obviously, only discrete rotations remain thatcorrespond to symmetries of the compactification lattice, generated by Q = exp{2ie J } with e = 1/N for a ZN symmetry. A
generalized SS deformation can then me obtained by coupling the rotation charges with
the winding numbers of a circle from T 2 (or with the whole T 2 momentum lattice in general). The corresponding partition function is obtained by replacing:

ab
a +ne1
e2ien(b1 b2 ) b+me1 ,
(7.7)

Z1 hhba
Z1 ha +ne1
hb +me1
124
where Z1 denotes the contribution of the (ha , hb ) twisted sector. Eq. (7.7) generalizes
(3.11), which is recovered by setting e = 1, corresponding to a 2 rotation. In this
case, the bosonic part is not modified and the deformation amounts to inserting the
phase sSS .
Following the reasoning of Section 3, in the large radius limit, only vanishing windings
n = 0 may give corrections that escape exponential suppression in the degeneration limit
2 . Thus, only the boundary conditions along the b cycle are modified in (7.7),
and the spin structure-dependent part of the amplitude (4.10) (with the picture-changing
insertion points satisfying the same gauge condition (4.9)) becomes:

a 1
1
a
(x y)
(x + y) z3

SO
b
b
2
2
a ,b

1
a
(x + y) z4
h
b + me1
2

1
a
(x + y) z5 ,
h1
b
2
(7.8)
where we defined, as in Section 4.1, h4 h and thus h3 = 1 and h5 = h1 for an internal

space T 2 T 4 /H . The spin structure sum can be performed using the Riemann identity
with the result:

0
0
0
SO m (x ) m (z3 )h1
(z4 )
m2 e1
2 e1
2 e1

0
h m (z5 ).
(7.9)
2 e1
After incorporating the remaining factors, one finds that the amplitude (4.12) becomes:

Z2
(x) (y)
0
A m (x ) 3,4,5
5
5
e
1
2
Z1 Z1,hme Z1,h1
2
I <J E(zI , zJ ) I =3 (zI )

0
0
0
m (z3 )h1
)
(z
4
h m (z5 )
m2 e1
2 e1
2 e1
4
X3 (z3 )Xhme
(z4 )Xh51 (z5 ) Zlat ,
(7.10)
where the symbol h me denotes the addition of me to the twist h (ha , hb ) along the
b cycle only: h me (ha , hb e2ime ). Note that the b-cycle periodicity conditions for all
zI s, including z4 , and x are twisted by the same Z2N group element eike = eik/N for
m = k mod 2N . The contribution of the k = 0 class of windings disappears by Riemann
vanishing theorem (supersymmetry), while the class of k = N has the same structure
as odd windings in the standard SS mechanism, and disappears after summation as a
consequence of m m symmetry. Since the zero modes of X3 are linear in these
winding numbers, the remaining contributions of m = k mod 2N come with opposite sign
to those of m = 2N k mod 2N ; however, as explained below, they are weighted by
different factors.
Starting from the amplitude (7.10) and repeating the same steps that led us in Section 4
to the final expression (4.27): antisymmetrization, bosonization formula, etc., for each
125
class of k mod 2N windings, one obtains a similar expression, but now with the tilde
representing a twist by eik/N along the b cycle, i.e.,
(x)
(y)
k (x),
k (y).
(7.11)
The integration over the vertex positions yields

2
k (x)
2

= 1 eik/N .
(7.12)
Note that the class of 2N k mod 2N windings (which come with the opposite sign) is
weighted by the complex conjugate of the above factor. Now it is clear that the sum over
the SS winding numbers does not vanish. However, one still needs to blow up singularities
as in Section 6, in order to avoid cancellations due to similar symmetries of the K3 orbifold
sector.
7.2. Determination of the gaugino mass
At this point, we can exclude any accidental cancellation and return to Eq. (7.4),
assuming that the sum over m is restricted to one of the m = k mod 2N (k = 1, . . . , N 1)
class of windings. Then Eq. (5.16) gives

m1/2 g 4

d2

mR 2
dl
m2 R 2
1
.
exp
2 l
(e2l 1) (2 l)3/2 m (2 + l)2
(7.13)
The integral over l has a logarithmic divergence at l = 0 which, as explained in Section 6,

is regulated by e2 . This brings a (large) factor of 2 , while l can be set zero everywhere
else in the integrand, so that

m1/2 g
d2
5/2

m2 R 2
mR exp
.
2
2
(7.14)
After rescaling 2 2 R 2 , the above expression yields

m1/2
g4
g 4 m3/2 .
R
(7.15)
The mass (7.15) can be understood within the effective field theory by looking at a
generic one-loop graph involving a gravitational exchange. Each vertex brings one power
of the Plank mass in the denominator, and thus m1/2 m33/2 /MP2 if the loop momentum
integral is convergent [21]. However, if as navely expected, the momentum integral is
quadratically divergent, then one obtains m1/2 m3/2 2UV /MP2 . Since the ultraviolet
cutoff UV MP , the result (7.15) confirms this expectation. Note, however, that in the
126

1/ m23/2
case of orbifold compactifications, for which m1/2 = O(e

fails because it ignores the effects of discrete symmetries.
), such a crude argument
8. Conclusions and outlook

In this work, we developed a formalism for computing one-loop gravitational corrections to the effective action of D-branes. Furthermore, we have shown that the genus 3/2
amplitude responsible for communicating the bulk supersymmetry breaking to open string
fermions is closely related to the well-known genus 2 topological term F2 . For models with
large extra dimensions and low-energy supersymmetry breaking la ScherkSchwarz, this
mass is proportional to the gravitino mass. However, the result is zero in the case of the
standard, Z2 -symmetric, ScherkSchwarz mechanism, hence one is forced to consider its
ZN generalization in order to generate a non-vanishing gaugino mass.
In fact, it is not easy to find explicit examples of such a mass generation in calculable
orbifold models. Orbifolds have discrete symmetries that generically lead to further
cancellations. These, in turn, can be avoided by blowing up the orbifold singularities,
which can be accomplished by switching on non-zero VEVs of certain twisted fields. This
mass generation mechanism clearly needs a deeper explanation. On one hand, one should
find a more profound connection to the underlying world-sheet theory. On the other hand,
it would be very useful to have a more precise field-theoretical description in terms of
effective interactions and Feynman diagrams. Both the topological origin as well as the
relation to quadratic divergences strongly suggest that the mass is generated by anomalies,
but more work is needed to identify a direct link.
One could think that the gaugino masses are due to the so-called anomaly-mediation
[22] of the four-dimensional (super)conformal anomaly. However, such a mechanism
predicts one-loop masses proportional to g 2 and not to g 4 as we find. Furthermore,
the anomaly-mediated gaugino masses are proportional to the beta functions, therefore
they would require the presence of at least one more world-sheet boundary. Thus the
mismatch of the powers of gauge coupling and of the group-theoretical coefficients hints
against superconformal anomaly mediation. In fact, our results raise an interesting question
whether the anomaly mediation mechanism can be implemented at all in any theory with a
sensible ultraviolet completion.
Acknowledgements
This work would have never been completed without the help of Narain, who had
assisted us at its crucial stages. We are deeply indebted to him. We are grateful to Pierre
Vanhove and Frank Ferrari for their help in resolving several technical problems. We
acknowledge illuminating discussions with Louis Alvarez-Gaum, Carlo Angelantonj, Edi
Gava, Stephan Stieberger and Cumrun Vafa. T.R.T. is grateful to CERN Theory Division
for its hospitality during the sabbatical leave of 2003, when most of this work was done.
His work is supported in part by the National Science Foundation under grant PHY-0242834. The work of I.A. is supported in part by the European Commission under the RTN
Contract HPRN-CT-2000-00148.
127
Appendix A. Bosonic zero modes and twisted differentials

The amplitude that determines the gaugino mass includes three supercurrent insertions,
with one of them effectively coupled to the SS torus T 2 and the other two to the K3 surface.
The T 2 contribution involves an explicit factor of the (untwisted) zero mode X3 ; in the
case of orbifold compactifications, the other two insertions involve also the zero modes
associated to the twisted sectors, see respectively Eqs. (4.16) and (4.6).
In the case of untwisted direction, it is convenient to consider separately the two real
components, each of the form
P

P
Li i (z) + Li i (z) ,
X(P ) = 2R
(A.1)
i=1,2
where O is an arbitrary base point. Due to the reality property (2.8), the Neumann and
Dirichlet boundary conditions read, respectively,

+ for N,
L 1,2 = L2,1
(A.2)
for D.
By considering the boundary conditions for a string winding n times around the a-cycle
and m times around the b-cycle, we obtain
L1 =
m + n[1 i(2 l)]

,
2i(2 l)
L2 = L 1 .
(A.3)

The classical action, Scl = (1/(2 )) 1 z Xz X, can be computed by using Eq. (2.10),
with the result written in Eq. (3.4). As explained in Section 3, in the degeneration limit
2 , l fixed, the n = 0 modes are exponentially suppressed by the partition function,
therefore the relevant zero modes are given by
z XN,D =
mR
N,D ,
i(2 l)
N,D = 1 (z) 2 (z).
(A.4)
In order to construct the zero modes twisted by a ZN group element h, we can start from
the g = 2 double cover , with the twisted complex instanton configuration [23]
P
P

Xh (P ) = 2R Lh h (z) + Lh h (z) ,
(A.5)
O
where h (z) is the (unique) holomorphic one-differential twisted by (hai , hbi ), i = 1, 2,

along the a- and b-cycles of , respectively. The constants Lh and L h are determined
by the boundary conditions. Since we are interested in the instanton solutions on 1 ,
we identify ha1 = ha , hb1 = hb . For our purposes, it is completely sufficient to consider
configurations with a non-trivial twist only along the b-cycle: ha = 1, hb h.
The simplest way to obtain the 2 limits of twisted differentials is by working
directly in the degeneration limit of the punctured hemisphere. The (unique) SL(2, C)covariant, holomorphic differential twisted by h along the b-cycle is given by:
h (z) = (z; a, a ) h(z; b, b),
(A.6)
128
where (z; p1 , p2 ) is defined in (5.1). If an additional twist by (1) along the b-cycle is
present, then
h (z) = (z; a, a ) + h(z; b, b).
(A.7)
It is easy to see that (z)
= h=1 (z) given by the above formula does indeed appear in the
2 limit of Eq. (4.23). This also proves Eq. (4.29). Note that at the boundary,
h (x) = h (x),
x 1 ,
(A.8)
therefore the Neumann and Dirichlet boundary conditions read, respectively,

+ for N,
L h = Lh
(A.9)
for D.
Since the amplitude considered in the paper vanishes for orbifold compactifications, we
do not pursue the discussion of twisted instanton configurations any further. We will make
use, however, of the twisted differentials in the following computation of Z 2 .
Appendix B. Moduli integration measure

The integration measure over the moduli of the (g = 1, h = 1) Riemann surface under
consideration is induced by the chiral measure of its g = 2 double-cover. It involves 3
insertions of the reparametrization ghosts b necessary to soak up the corresponding zero
modes:

p=3

1 a1 a2 a3
d() = d d dl Z2
(B.1)
dwp ap (wp )b(wp ) ,
p=1
where ap , p = 1, 2, 3, are the Beltrami differentials dual to the moduli (d = d11 , d =

d22 , dl = d12 ), respectively. The correlation function on the r.h.s.,

b(w1 )b(w2 )b(w3) Z2 (w1 , w2 , w3 ) = Z2 det i j (wp ),
(B.2)
where the constant factor Z2 can be considered as the (oscillator part of the) partition
function of the (2, 1) bc system. This determinant has already been included in the
amplitude (4.6), therefore in order to avoid double counting we inserted the factor Z21 in
(B.1). The three quadratic differentials, 1 1 , 1 2 and 2 2 form a basis of the b-ghost
zero modes. After inserting (B.2) into the r.h.s. of (B.1), the position integration can be
performed explicitly; since the Beltrami differentials are dual to dij ,
d() = d1 d2 dl.
(B.3)
In order to determine the large 2 behavior of Z2 , it is convenient to use the bosonization

formula [15]:
p=3
1,2,3
p=3

1/2
wp 3
E(wp , wq )
3 (wp ),
Z2 (w1 , w2 , w3 ) = Z1
(B.4)
p=1
p<q
p=1
129
1/2
where the factor Z1

is the (oscillator part of the) partition function of a chiral scalar
field. The argument of the theta function can be rewritten as:
p=3
wp 3 = + 3,
p=1
p=3
wp 3 .
(B.5)
p=1
Since is finite in the 2 limit, the leading term of the theta series is

p=3
wp 3 = e4(2l) f ( ),
(B.6)
p=1
where

f ( ) = e2(1+2 ) 1 + el e2i1 + el e2i2 + e2i(1 +2 ) .
(B.7)
It is a matter of straightforward but quite tedious algebra to show that for the differentials
(5.1),
det i j (wp ) = f ( )(e2l 1)2 e6l
1,2,3

E(wp , wq )
p<q
p=3

3 (wp ).
(B.8)
p=1
Hence, after comparing Eqs. (B.2) and (B.4), we obtain

1/2 e
Z2 = Z1
4(2 l) e 6l
(B.9)
.
(e2l 1)2
At this point, it remains to determine the large 2 limit of the chiral boson partition
function Z1 . To that end, it is convenient to use the chiral bosonization formula [15]:
3/2
Z1
(u1 + u2 v )
E(u1 , u2 )
(u1 ) (u2 )
,
det i (uj )
E(u1 , v)E(u2 , v)
(v)
(B.10)
where u1 , u2 and v are arbitrary. After taking the limit in a similar way as before, we obtain
Z1 = 1.
(B.11)
In order to compare with the existing literature, we note that, according to our results,
the two-loop cosmological constant of D = 26 bosonic string,
Z2
Z113
d() = d1 d2 dl
e4(2l) e6l
,
(e2l 1)2
(B.12)
where we used Eqs. (B.3), (B.9) and (B.11). This quantity should be compared with the
formal square root of the well-known result [24,25]:
dij
Z2
,
d()
=
(B.13)
|10 ()|
Z113
ij

where 10 is the weight 10 generator of the Isgusa ring, 10 () = 2 [ ](, 0), with
the product including all 10 even spin structures. Indeed, Eq. (B.12) follows in the 2
(l fixed) degeneration limit of 10 . The double pole appearing in the factorization limit
12 = l = 0 can be attributed to the tachyon [25].
130
Appendix C. Twisted determinant Z 2

In order to find the 2 limit of the determinant Z 2 , we will compare the
degeneration limits of the left- and right-hand sides of Eq. (4.26).
On the l.h.s., we have a theta function series, with the leading term

a=3
a=3
0
za 3 = e4(2l) f( ), =
za 3 ,
1
(C.1)
21
a=1
a=1
where

f( ) = e2(1+2 ) 1 el e2i1 el e2i2 + e2i(1 +2 ) .
(C.2)
Furthermore, the degeneration limits of the prime form and the -differential are given by
Eqs. (5.11) and (5.13), respectively, and Z1 = 1, as shown in Appendix B.
On the r.h.s., we have the determinant det h a (zb ) involving twisted two-differentials h a ,
a = 1, 2, 3. We can compute this determinant explicitly by choosing the basis
h 1 = N ,
h 2 = h h1 ,
h 3 = h1 h ,
(C.3)
with the -differentials listed in Eqs. (A.4), (A.6), (A.7) and (5.1). After a straightforward
but tedious computation, we find that, up to a numerical factor,
1,2,3
a=3

E(za , zb )
3 (za ).
det h a (zb ) = Im(h) f( ) e2l 1 e6l
a<b
(C.4)
a=1
By comparing the two sides of Eq. (4.26), we find:

e4(2l) e6l
Z 2 =
(C.5)
.
e2l 1
Then Eq. (5.15) follows after combining the above result with Eqs. (B.9) and (B.11). The
Im(h) factor, which reflects antisymmetry of the determinant under h h1 for a specific
choice of basis (C.3), is irrelevant to the present derivation, however it can be important
when summing over twisted sectors, as pointed out in Section 6.
References
[1] J. Scherk, J.H. Schwarz, Phys. Lett. B 82 (1979) 60.
[2] R. Rohm, Nucl. Phys. B 237 (1984) 553;
H. Itoyama, T.R. Taylor, Phys. Lett. B 186 (1987) 129.
[3] C. Kounnas, M. Porrati, Nucl. Phys. B 310 (1988) 355.
[4] S. Ferrara, C. Kounnas, M. Porrati, F. Zwirner, Nucl. Phys. B 318 (1989) 75;
C. Kounnas, B. Rostand, Nucl. Phys. B 341 (1990) 641;
I. Antoniadis, Phys. Lett. B 246 (1990) 377.
[5] I. Antoniadis, E. Dudas, A. Sagnotti, Nucl. Phys. B 544 (1999) 469, hep-th/9807011.
[6] I. Antoniadis, K. Benakli, A. Laugier, T. Maillard, Nucl. Phys. B 662 (2003) 40, hep-ph/0211409.
[7] S. Sugimoto, Prog. Theor. Phys. 102 (1999) 685, hep-th/9905159;
I. Antoniadis, E. Dudas, A. Sagnotti, Phys. Lett. B 464 (1999) 38, hep-th/9908023;
G. Aldazabal, A.M. Uranga, JHEP 9910 (1999) 024, hep-th/9908072.
131
[8] E. Dudas, J. Mourad, Phys. Lett. B 514 (2001) 173, hep-th/0012071;

G. Pradisi, F. Riccioni, Nucl. Phys. B 615 (2001) 33, hep-th/0107090.
[9] M. Bershadsky, S. Cecotti, H. Ooguri, C. Vafa, Commun. Math. Phys. 165 (1994) 311, hep-th/9309140.
[10] I. Antoniadis, E. Gava, K.S. Narain, T.R. Taylor, Nucl. Phys. B 413 (1994) 162, hep-th/9307158.
[11] S.K. Blau, S. Carlip, M. Clements, S. Della Pietra, V. Della Pietra, Nucl. Phys. B 301 (1988) 285;
M. Bianchi, A. Sagnotti, Phys. Lett. B 211 (1988) 407.
[12] M. Bianchi, A. Sagnotti, Phys. Lett. B 231 (1989) 389.
[13] I. Antoniadis, K.S. Narain, T.R. Taylor, in preparation.
[14] H. Ooguri, C. Vafa, hep-th/0302109.
[15] E. Verlinde, H. Verlinde, Nucl. Phys. B 288 (1987) 357.
[16] J.D. Fay, Theta Functions on Riemann Surfaces, in: Lecture Notes in Mathematics, Springer-Verlag, New
York, 1973.
[17] R. Russo, S. Sciuto, Nucl. Phys. B 669 (2003) 207, hep-th/0306129;
R. Russo, S. Sciuto, Fortschr. Phys. 52 (2004) 678, hep-th/0312205.
[18] E. DHoker, D.H. Phong, Nucl. Phys. B 639 (2002) 129, hep-th/0111040.
[19] L. Dixon, D. Friedan, M. Martinec, S. Shenker, Nucl. Phys. B 282 (1987) 13.
[20] J.J. Atick, L.J. Dixon, P.A. Griffin, Nucl. Phys. B 298 (1988) 1.
[21] I. Antoniadis, M. Quiros, Nucl. Phys. B 505 (1997) 109, hep-th/9705037;
R. Rattazzi, C.A. Scrucca, A. Strumia, Nucl. Phys. B 674 (2003) 171, hep-th/0305184.
[22] L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79, hep-th/9810155;
G. Giudice, M.A. Luty, H. Murayama, R. Rattazzi, JHEP 9812 (1998) 027, hep-ph/9810442.
[23] K.S. Narain, M.H. Sarmadi, C. Vafa, Nucl. Phys. B 356 (1991) 163.
[24] A.A. Belavin, V. Knizhnik, A. Morozov, A. Perelomov, JETP Lett. 43 (1986) 411, Phys. Lett. B 177 (1986)
324.
[25] G. Moore, Phys. Lett. B 176 (1986) 369.

On the construction of asymmetric orbifold models

Kenichiro Aoki a , Eric DHoker b , D.H. Phong c
a Hiyoshi Department of Physics, Keio University, Yokohama 223-8521, Japan
b Department of Physics and Astronomy, University of California, Los Angeles, CA 90095, USA
c Department of Mathematics, Columbia University, New York, NY 10027, USA
Abstract
Various asymmetric orbifold models based on chiral shifts and chiral reflections are investigated.
Special attention is devoted to the consistency of the models with two fundamental principles for
asymmetric orbifolds: modular invariance and the existence of a proper Hilbert space formulation
for states and operators. The interplay between these two principles is non-trivial. It is shown, for
example, that their simultaneous requirement forces the order of a chiral reflection to be 4, instead
of the naive 2. A careful explicit construction is given of the associated one-loop partition functions.
At higher loops, the partition functions of asymmetric orbifolds are built from the chiral blocks of
associated symmetric orbifolds, whose pairings are determined by degenerations to one-loop.
1. Introduction
String theories are built of independent left and right movers on the worldsheet, and both
type II [1] and heterotic [2] superstring theories are inherently chiral on the worldsheet. As
such, they admit asymmetric compactifications for left and right degrees of freedom. This
fact is basic to the construction of the heterotic string [2] and to other models such as
the Narain compactification [3]. Compactifications by orbifolds of flat spacetime ([4,5];
for reviews see [68]), are particularly important because their conformal field theories
Research supported in part by a Grant-in-Aid from the Ministry of Education, Science, Sports and Culture
and grants from Keio University (K.A.), and by National Science Foundation grants PHY-01-40151 (E.D.) and
DMS-02-45371 (D.P.).
E-mail addresses: ken@phys-h.keio.ac.jp (K. Aoki), dhoker@physics.ucla.edu (E. DHoker),
phong@cpw.math.columbia.edu (D.H. Phong).
doi:10.1016/j.nuclphysb.2004.06.038
K. Aoki et al. / Nuclear Physics B 695 (2004) 132168
133
are explicitly solvable from free field theories via group theoretic methods, and yet their
spacetime properties are non-trivial.
For symmetric orbifolds, the orbifold group acts identically on left and right movers.
The associated worldsheet conformal field theory admits a consistent construction
equivalently via functional integral or operator methods [911], and the functional integral
formulation naturally guarantees modular invariance.
For asymmetric orbifolds, the orbifold group acts differently on left and right movers.
This circumstance renders functional integral formulations problematic. Therefore, modular invariance and the validity of a Hilbert space interpretation cannot be taken for granted.
Some general principles for compactifications on asymmetric orbifolds have been formulated in [12], but relatively few examples have been studied explicitly (see for example [13]). (At special compactification radii, asymmetric orbifold conformal field theories
may be reformulated in terms of free fermionic degrees of freedom alone. The action of the
orbifold group on chiral fermions is well understood and systematic studies of asymmetric
free fermion models are available in [14,15].)
Asymmetric orbifolds are clearly needed when orbifolding the heterotic string (for
example, see [16]). More recently, asymmetric orbifold models for type II strings
generated by chiral reflections and shifts were also proposed [1720]. These Kachru
KumarSilverstein (KKS) models are of particular interest, since they exhibit spacetime
supersymmetry breaking and vanishing cosmological constant at one-loop order. Initially,
there were hopes that the cosmological constant would continue to vanish to two loops [17,
18] (see also [21]), via an independent cancellation of the sums over spin structures for left
and right movers, but this is now known not to be the case [22].
In the present paper, we take the opportunity of the study of the KKS models to examine
the principles of the construction of asymmetric orbifold models based on chiral shifts and
twists in greater detail, and to add to the hitherto relatively short list of examples worked
out explicitly. In view of the methods of chiral splitting [2325] (see also [26,27]) and
two-loop superstring perturbation theory developed over the past few years in [2832] (see
also [33,34]), it is natural to postulate that the partition function for an asymmetric orbifold
is obtained by suitably pairing the chiral blocks of symmetric orbifold theories. Thus,
the main problem of asymmetry is the determination of the corresponding pairing matrix
(see [22] and Section 6). It can be determined in principle by degenerations to one-loop.
Thus, the key problem reduces to finding all the one-loop traces Z g h = TrHh (gq L0 q L0 ),
where g, h runs over all the elements of the orbifold group, and Hh is the Hilbert space of
the sector twisted by h.
Even for the relatively simple asymmetric orbifold models based on chiral shifts and
twists, there are important subtleties in the construction: (1) chiral operators g, h are
defined only up to phases, which can affect their orders and change the whole structure of
the theory; (2) it is unclear how to incorporate in the construction of the Hilbert space Hh
for the asymmetric theory the ground state degeneracies of the corresponding symmetric
orbifold theories; (3) one has to extend to Hh the operator g which was originally defined
only on the untwisted Hilbert space.
We find that for the asymmetric orbifold models based on chiral shifts and twists, the
combined requirements of modular covariance and Hilbert space interpretation constrain
the orders, and hence the phases, of the elements of the orbifold group. The degeneracies in
134
the Hilbert spaces for the twisted asymmetric theory can be handled by suitable selection
rules on the larger set of ground states coming from the symmetric theory. And upon this
construction, the operators of the theory admit consistent extensions to the twisted sectors.
It can be hoped that similar considerations will apply to more general asymmetric orbifold
models.
1.1. Principles of orbifold constructions
Let G be an orbifold group acting on a flat torus Tn . The orbifold group G will be
taken to be finite and Abelian for simplicity.1 The following principles will be taken as the
starting point for orbifold constructions on the worldsheet of a torus with modulus ;
The existence of a consistent Hilbert space formulation
The partition function ZG of the G-orbifold theory is given by a summation over partition
traces Z g h ,

1 g

Z h ( ), Z g h ( ) TrHh gq L0 q L0 ,
ZG ( ) =
(1.1)
|G|
g,hG
where q exp{2i } and |G| is the order of G. The key assumption in this principle is
that a consistent Hilbert space formulation exists in which Hh is the Hilbert space of the
sector twisted by h, and the group elements g G have a consistent operator realization in
each of these twisted Hilbert spaces.
Modular covariance of partition traces
Modular invariance of the partition function ZG is guaranteed by the stronger condition of
modular covariance of each of the partition traces,
1
Z g h ( + 1) = Z gh
Z g h (1/ ) = Z
h1
h ( ),
g ( )
or more generally,

a + b
d b
g
Z h
= Z g h g c ha ( ),
c + d
(1.2)
a
c
b
d

SL(2, Z).
(1.3)
Modular covariance of the partition traces implies that Z g h ( + nh ) = Z g h ( ) for any

element h of order nh . This relation, in turn, is equivalent to the familiar requirement of
level matching,
1
L0 L 0 Z
nh
(1.4)
1 For a more detailed summary of symmetric and asymmetric orbifold constructions, see [22]. The orbifold
construction may be carried out either by coseting flat Euclidean space Rn by a full orbifold group including
the action of translations, or by coseting the flat torus Tn by the point group PG . For symmetric orbifolds, these
procedures are equivalent, as is demonstrated in some simple cases in Appendix B. For asymmetric orbifolds, the
coseting procedure starting from the torus is taken as a definition.
135
when acting on the Hilbert space Hh of states twisted by an element h of order nh . Finally,
the transformation in (1.3) given by a = d = 1 and b = c = 0 belongs to the center
of SL(2, Z), reverses the orientation of both A and B cycles, and corresponds to charge
conjugation symmetry.
When the orbifold group G acts on genuine well-defined fields, and a functional integral
formulation is available, the above principles may be deduced from the functional integral
formulation using standard quantum field theory methods. In particular, a simple change
of variables in the functional integral will imply the modular covariance relations for the
partition traces expressed above.
For general asymmetric orbifolds, a proper functional integral formulation may be
lacking and it may not be possible to derive the above relations from first principles. In
such theories, the above principles will simply be postulated as necessary and sufficient
conditions for the existence of physically viable asymmetric orbifold models. It has been
argued in [9] that level matching suffices to insure the full modular covariance (1.2) of
the theory. This condition is not always sufficient to guarantee the full modular covariance
of the partition traces. In this paper some examples will be presented where this lack of
modular covariance conflicts with the existence of a proper Hilbert space interpretation.
1.2. Algorithm for the construction of asymmetric orbifolds
In this paper, asymmetric orbifold models based on chiral shifts and reflections (such as
arise in the constructions of [17]) are examined in detail. Special attention is devoted to a
proper construction of states and operators in the untwisted and twisted sectors of Hilbert
space and to the modular covariance of the partition traces. Models with only chiral shifts
are constructed first and shown to satisfy modular covariance, at least in some critical
dimensions. In models involving chiral reflections, difficulties are found with modular
covariance, even though the level matching condition (1.4) is satisfied. These problems
are traced to the precise definition of the chiral reflection operators in Hilbert space. In
particular, it is shown that a chiral reflection of order 2 leads to conflicts with the modular
transformation 1/ , but that a chiral reflection of order 4 is consistent with modular
invariance.
Generally, in models satisfying level matching, it had been expected that the modular
orbit method would yield all partition traces Z g h ( ) from applying the modular covariance
condition (1.2) to the partition traces in the untwisted sector Z g 1 ( ). The models with
chiral reflections examined here show that the method must be applied with some care:
modular transformations may not generate all partition traces from the untwisted sector,
and even when they do, the objects they generate may not have a proper interpretation as
partition traces in a twisted sector.
While our discussions later in the paper will be in terms of specific examples based
on shifts and twists, the methods developed there may be recast in terms of a recursive
algorithm, which we shall now summarize.
136
(1) It is assumed that all the group elements g G in the orbifold group have well-defined
operator realizations on the untwisted Hilbert space H1 . The partition traces Z g 1 in
the untwisted sector for all g G are then well-defined and may be calculated.
(2) Modular transformations applied to the partition traces Z h 1 yield Z 1 h . The Hilbert
space representation of Z 1 h ( ) = TrHh q L0 q L0 may be viewed as a spectral density

function for the Hilbert space Hh , which yields the conformal spectrum, including the
multiplicities for all states in Hh . This essentially determines Hh .
n
(3) The further application of modular transformations to Z 1 h yields Z h h , from which
the action of the operators hn on Hh may be deduced.
(4) Combining the knowledge of the action of operators hn on Hh with the properties
of the symmetric operators in the associated symmetric orbifold theories has allowed
us, in all cases considered here, to construct also the action of elements g on Hh ,
even when g is not a power of h. This last step permits us to calculate the remaining
partition traces Z g h .
The twisted Hilbert space Hh which emerges from the above construction is of the
following form. Let G be the asymmetric orbifold group whose elements consist of pairs
h = (hL ; hR ), where hL and hR act respectively on the left and the right sectors. The
left and right elements themselves span groups GL and GR and G may be viewed as
a subgroup of the product GL GR . The starting point is the two symmetric orbifold
theories with symmetric orbifold groups GL and GR consisting of pairs (hL ; hL ) and
(hR ; hR ), respectively. Let HhL (pL ) and HhR (pR ) be the Hilbert spaces of the sectors
twisted by (hL , hL ) and (hR , hR ) in the symmetric theories. Here, we denote by pL ML
and pR MR any additional labels of the blocks, such as for example the internal loop
momenta. ML and MR may be viewed as the degeneracies of the ground states in the
symmetric theories. Then the Hilbert space of the h-twisted sector in the asymmetric theory
is of the form

HhL (pL ) HhR (pR ) .
Hh =
(1.5)
(pL ,pR )I
Here, I is a pairing or selection rule, and we keep only chiral oscillators in HhL (pL )
and HhR (pR ). The set I is usually strictly smaller than the set of all ML MR possible
choices for (pL , pR ). It is determined by modular transformations from the asymmetric
traces in the untwisted sector. In [12], it was proposed to obtain the partition function of
the asymmetric theory by taking square roots of suitably extended symmetric theories. Our
prescription is a significant departure from this, since it avoids square roots altogether and
replaces them by selection rules after a suitable chiral splitting.
The remainder of this paper is organized as follows. In Section 2, a brief summary
is given of the circle theory. In Section 3, orbifolds generated by chiral shifts only
are constructed. In Section 4, orbifolds with chiral reflections only are investigated and
subtle issues of modular invariance are addressed and solved. In Section 5, the results of
Sections 3 and 4 are combined and orbifolds with both chiral shifts and reflections are
solved, and the effects of adding worldsheet fermions are included. In Section 6, general
rules for the partition function in higher genus are formulated in terms of chiral splitting
and a summation over pairings of chiral blocks. Finally, in Appendix A, the construction of
137
symmetric orbifolds is briefly reviewed; in Appendix B, the equivalence is demonstrated

for symmetric orbifolds between orbifolding the full line by the full orbifold group and
orbifolding the circle by the point group; and in Appendix C, useful -function identities
are collected.
2. The circle theory

We begin by recalling some basic facts about the S 1 theory, mainly for normalizations
and conventions. The torus with modulus = 1 +i2 , 2 > 0, is parametrized by z, with
the identifications z z + 1 and z z + . It is equipped with the metric ds 2 = 21 dz d z .
The action S[x] for a theory of a single bosonic scalar field x is normalized to be

1
d 2 z x x,
S[x] =
(2.1)
42
where is the string scale.2 In the S 1 theory, the field x(z) takes values in a circle of radius
R, and we have the identification x x + 2R. The evaluation of the functional integral
over all instanton sectors defined by the boundary conditions x(z + 1) = x(z) + 2m1 R
and x(z + ) = x(z) + 2m2 R gives the well-known partition function

R
R 2
2
ZS 1 ( ) =
(2.2)
exp
|m
m
|
.
1
2
R
222
22 |( )|2
m1 ,m2 Z
Here, the argument of the exponential is the action of the corresponding instanton solution,
and the prefactor combines the contributions of the quantum fluctuations and the zero mode
integration. The partition function ZS 1 can be recast in Hamiltonian language by a Poisson
R
resummation in m2

1 2 1 2
1
ZS 1 ( ) =
q 2 pL q 2 pR , q e2i ,
(2.3)
2
R
|( )|
m1 ,m2 Z
where the left and the right (dimensionless) momenta pL and pR are given by
pL =

R
m2 m1 ,
R
2
pR =

R
m2 + m1 ,
R
2
m1 , m2 Z.
(2.4)
2 p2 = 2m m 2Z.
In particular (pL , pR ) belongs to an even Lorentzian lattice pL
1 2
R
(Note that the dimensionful momenta are given by pL /2 and pR /2, so that the total
dimensionful momentum is (pL + pR )/2 = m2 /R, as expected.)
Henceforth, we consider the S 1 theory at the self-dual radius R 2 = 22 . The left and
right momenta simplify to
(2.5)
2pL = m2 m1 = nL ,
2pR = m2 + m1 = nR ,
2 The parameter is related to the Regge slope parameter by 22 = . Customary conventions used in the
literature are as follows. In [16], the Regge slope parameter is exhibited explicitly; in [24] and [7], one sets
= 1; in [6], one sets = 1/2 instead.
138
with nL , nR Z and nL + nR 2Z. The momentum summations may be carried out in

terms of -functions. At the self-dual radius, it is convenient to cast the -functions results
in terms of the -function, defined by
1 (n+)2 2in [|](0, 2 ) 2i
e
q
e
=
.
[|]( )
(2.6)
( )
( )
nZ
Various transformation properties of its arguments are listed in Appendix C. One finds
2
1
ZS 1 ( ) = [0|0]( ) +
0 ( )
,
(2.7)
2
R= 2
where the first term arises from the contributions with nL , nR 2Z and the second term
from contributions with nL , nR 2Z + 1.
The Hilbert space H of the circle theory can be built from chiral sectors as follows

H=
xn1 xnk |nL L
xn1 xnk |nR R .
nL +nR 2Z
nj =1
nj =1
(2.8)
Here, the states |nL L , |nR R are momentum eigenstates, characterized by their quantum
numbers nL , nR as defined in (2.5), and xnj , xnj are oscillators for each chiral sector. Then
ZS 1 becomes
R

ZS 1 ( ) = TrH q L0 q L0
(2.9)
with the following definitions of the Virasoro generators,
L0 =
1 2
1
+ pL
+
nxn xn ,
24 2
n=1
1 2
1
L 0 = + pR
+
nxn xn .
24 2
(2.10)
n=1

Finally, at any radius R, the circle theory has a chiral U
(1)L U
(1)R symmetry, generated
3
3
At the self-dual radius R 2 = 22 , the
by the chiral currents JL = x and JR = x.
R , which arises due to the existence of the extra
L SU(2)
symmetry is enhanced
to SU(2)
currents JL = exp{i 2x+ } and JR = exp{i 2x }, where x denote the chiral parts
of the field x. The latter symmetry will be exploited in Section 4.2.
3. Orbifolds defined by a chiral shift

The operator s realizing a shift x x + R by a half-circumference commutes with
the oscillators xn and xn and acts on the momentum ground states by s|nL L |nR R =
ei(pL +pR )R/(2)|nL L |nR R . At the self-dual radius R 2 = 22 , this becomes simply
s|nL L |nR R = ei(nL +nR )/2 |nL L |nR R .
(3.1)
This form naturally permits chiral splitting into actions of left and right chiral shift
operators sL and sR , which also commute with the oscillators xn and xn . Their action
139
on the ground states is defined by

1
sL |nL L |nR R = e 2 inL |nL L |nR R ,

1
sR |nL L |nR R = e 2 inR |nL L |nR R .
(3.2)
Clearly these definitions reproduce s = sL sR . Note that each operator sL and sR has order
4 while their product s has order 2 in view of the fact that nL + nR 2Z. In this section,
orbifolds generated by either sR or sR2 will be considered.
3.1. The sR models
Recall that the partition function of the orbifold theory defined by a finite Abelian
group G is given (1.1), where |G| is the order of G, and Hh is the Hilbert space of the
twisted sector defined by h. As is reviewed in Appendix A, in the case of the symmetric
shift s = sL sR , the Hilbert space in the twisted sector corresponding to an element s a is
generated by oscillators from the ground states |nL + a2 |nR a2 .
In the asymmetric orbifold theory generated by sR , the Hilbert space of the sector
twisted by the element (sR )a is defined to be the space generated by applying the
oscillators xn and xn to the ground states of the form
, nL + nR 2Z.
|nL L
nR
(3.3)
2 R
The untwisted sector may simply be viewed as the sector twisted by (sR )a for a = 0, an
observation that permits us to treat all sectors at once. This construction is consistent with
the fact that sR is of order 4, since a shift a a + 4 may be compensated by relabeling
the states by nR nR + 2, a transformation that preserves the condition nL + nR 2Z.
The action of (sR )b on a state in the sector twisted by (sR )a is given by

a
ib
a
a
b
= exp
(sR ) |nL L
nR
(3.4)
nR
|nL L
nR
2 R
2
2
2 R
with, as always, the constraint that nL + nR 2Z.
The partition traces for a single scalar field are easily calculated. The oscillator
contributions are all equal to those of the untwisted sector and result in the familiar
-function factors. The momentum dependence may be read off from the momentum
assignments of the states. Abbreviating Z g h = Z b a when h = (sR )a and g = (sR )b , the
partition traces become
Z b a ( ) =
1
|( )|2
eib 2 (nR 2 ) q 4 nL q 4 (nR 2 ) .
1 2
a 2
(3.5)
nL +nR 2Z
The sums over nL and nR are conveniently expressed in terms of the -function defined
in (2.6). The constraint nL + nR 2Z in (3.5) may be solved explicitly and gives rise to
two parts: the first in which nL and nR are independent even integers (first term on rhs)
140
and the second in which they are independent odd integers (second term on rhs)

1
a
b
Z b a ( ) =
[0|0](0,
2
)
(0, 2 )
2
|( )|
4
2

a 1
b
1
+
(0, 2 )
+
0 (0, 2 )
2
4 2
2

i ab
a
b
1
a 1
b
+
( ) .
= e 4 [0|0]( )
( ) + eib 2
0 ( )
4 2
2
4 2 2
(3.6)
From the partition traces of the sR -twisted theory of a single boson it is straightforward
to obtain the partition traces of a theory of d bosons taking values in a square torus of all
self-dual radii and where sR acts as a right chiral shift simultaneously on all directions.
As the d bosons in this model are independent of one another, the corresponding partition
traces are simply given by (Z b a )d .
3.1.1. Interpretation in terms of chiral blocks constructions
With the extension to higher loops in mind, it is useful to interpret the previous formulas
in the following manner. We would like to view the traces Z b a ( ) of the sL theory as built
from the chiral blocks of the symmetric s = (sL , sR ) theory and the untwisted theory, with
a suitable pairing. Now the partition function for the symmetric s theory at the self-radius
R 2 = 22 is simply given by the partition function for the circle theory at R 2 = 2 /2 and
hence given by [10]
1
ZRcircle
(3.7)
Z ( ),
2 =2 /2 ( ) =
2
,
where [|] runs over all half-characteristics, and the trace Z ( ) is given by
+
(0, 2 )
.
Z ( ) =
2
|( )|
2
1
(3.8)
{0, 2 }
On the other hand, the partition function of the self-dual circle model in arbitrary genus h
is given by

[ |0](0, 2 )
2 ,
ZRcircle
(3.9)
2 =22 ( ) =
2
|( )|
1
{0, 2 }h
where is now viewed as the period matrix and ( ) as the chiral boson partition function
at genus h. The formula (3.6) suggests that, for fixed b, a Z4 = {0, 1, 2, 3}, the chiral
blocks of the s theory and the untwisted theory can be defined respectively as
1
s-theory
+ a
(0, 2 ),
ZR 2 =2 /2 :
( )
4
1
[ |0](0, 2 ),
ZRcircle
(3.10)
2 =2 /2 :
( )
where {0, 12 } and we have expressed b as

1
b = 2( + c), 0, , c {0, 1}.
2
141
(3.11)
With this choice of phases for the chiral blocks arising from the symmetric theories, the
blocks Z b a ( ) of the asymmetric sL theory arise by pairing them with the matrix
K = e2ic
(3.12)
so that we indeed recover the earlier partition traces,

1
1
b
2ic
Z a ( ) =
e
+ a
(0, 2 )[ |0](0, 2 ).
4
|( )|2
1
(3.13)
{0, 2 }
The full partition function for a theory of d bosons is then given by

1
Z( ) =
4
a,bZ4
d

1
1
2ic
e
+ a
(0, 2 )[ |0](0, 2 ) .
|( )|2
4
1
{0, 2 }
(3.14)
The partition function for the sR theory for d bosons is of course obtained by complex
conjugation.
3.1.2. Modular invariance of the sR model for d 16N bosons
The modular covariance properties (1.2) of the partition traces follow from the modular
transformation laws (C.11) for the functions [|]( ) and the relations (C.9) and (C.12).
One finds
Z b a ( + 1) = eia
2 /8
Z a+b a ( ),
Z b a (1/ ) = eiab/4 Z a b ( ).
(3.15)
Note that the exponent a 2 /16 is as expected in the + 1 transformation law because
the ground state in the sector twisted by (sR )a has conformal dimension a 2 /16 mod 1.
As a result, the operator sR in the twisted sector actually has order 16 with sR8 = 1. To
properly restore the order of sR to be 4 as it was in the untwisted sector, the dimension of
the torus (see the last paragraph of the preceding section) must be divisible by 4.
More critically, the orbifold model for a single boson does not satisfy the modular
covariance requirement of (1.2). Partition traces with full modular invariance are obtained
only when the dimension of the torus satisfies d 16N.
3.1.3. Asymmetry of the sR partition function for d = 16 bosons
Often, an orbifold constructed from the action of an asymmetric orbifold group still
possesses a symmetric partition function, i.e., Z( ) = Z( ). It will be verified below that
the orbifold theory constructed from 16 bosons and the asymmetric group action of sR is
in fact asymmetric and satisfies Z( )
= Z( ). The starting point is the partition function
142
defined by the partition traces of (3.6)

Z( ) =
1
4
Z b a ( )
16
(3.16)
a,b=0,1,2,3
The most convenient study of the asymmetry issue is in terms of -functions of modulus .
To this end, Z is first expressed in terms of -functions of modulus 2 , which are then
converted into -functions with modulus by using the doubling formulas. To simplify
this calculation, note that a shift b b + 2 produces a relative sign between the two terms
in the parenthesis of (3.6). It is convenient to restrict the range of b = 0, 1 and isolate the
summation over the shifts b b + 2 using a new variable c = 0, 1. One obtains
b+2c
a ( )
2

1
a
b
1
1 a
b 2
c
+
[0|0]
+ ()
0
=
(0, 2 )
|( )|4
4 2
2
2 4
2
(3.17)
the partition function may be expressed as follows, To compute the above partition traces
in terms of -functions with argument , i (0, ), with i = 2, 3, 4, we use the doubling
formula for -functions given in Appendix C. One finds (here a, c = 0, 1)

1
|3 |4 + ()a |4 |4 + ()c |2 |4 ,
2||4
1+2c
2

1 2
Z
3 + ()a 42 3 4 ,
2a ( ) =
4
2||
2c
2 2c
2

1 2
Z 1 ( ) = Z 3 ( ) =
3 + ()c 22 2 3 ,
4
2||
1+2c

2 1+2c
2
1
Z
i42 + ()c 22 2 4 .
1 ( ) = Z
3 ( ) =
4
2||
Z 2c 2a ( )
2
(3.18)
Combining all contributions, one obtains

Z=
1
210 |( )|32

|3 |4 + |4 |4 + |2 |4
8
8

+ |3 |4 + |4 |4 |2 |4

8
8
+ |3 |4 |4 |4 + |2 |4 + |3 |4 |4 |4 |2 |4
8
8

+ 2 32 + 42 + 32 42 ( 3 4 )8
8
8

+ 2 32 + 22 + 32 22 ( 3 2 )8
8
8

+ 2 22 i42 + 22 + i42 ( 2 4 )8 .
(3.19)
Using (C.4), it is straightforward to verify that Z is modular invariant.

The partition function Z for the 16-dimensional orbifold by (1, sR ) is asymmetric.
To see this, it suffices to show that Z Z
= 0. The contribution of the first 4 terms
The remaining terms may be
in Z is manifestly symmetric and cancels out of Z Z.
4
4
4
simplified using the Jacobi identity 3 = 2 + 4 , and expressed as a function of a single
143
homogeneous variable t 44 /34 ,

15|3 |32
(t t)(1 t t)(1 t t)(t + t t t).
Z Z = 7
2 |( )|32
(3.20)
Manifestly, this is not zero and the model has asymmetric conformal weights.
3.2. The sR2 models
The sR2 can be easily derived as a subsector of the sR models, consisting of the blocks
a ( ) with a, b = 0, 2. The Hilbert space interpretation of all the blocks is a direct
consequence of the Hilbert space interpretation provided for the sR theory. Because fewer
blocks appear in the sR2 theory, the requirement of modular invariance block by block is
less restrictive. Explicitly,
Zb
2
1

2
Z 0 0 ( ) =
[0|0]
+
,
2
2
1

Z 2 0 ( ) =
[0|0]

,
2

1
0
Z 2 ( ) = [0|0]
0 +
0 [0|0],
2
2

1
2
i
[0|0]
0
0 [0|0] .
Z 2 ( ) = e
2
2
(3.21)
Level matching requires the number d of bosonic fields to be a multiple of 4, in which case
we have modular covariance of the partition traces. The resulting partition function is

2 d
2
1
Z( ) =
[0|0] +
2
2
=1

d
1
1
1
[0|0]
0 +
0 [0|0]
2
2
2
(3.22)
=1
for d 4N. We observe that, although the orbifold group consists of asymmetric elements,
the resulting partition function Z turns out to be symmetric in this case. For the minimal
dimension of d = 4, one may alternatively express the partition function in terms of function with modulus . The result is
Z( ) =

1
|2 |8 + |3 |8 + |4 |8 .
2||8
(3.23)
It is easy to see that this partition function equals the partition function for 8 decoupled
Majorana fermions with SO(8) invariant action.
144
4. Orbifolds defined by a chiral reflection

The symmetric reflection r : x(z) x(z) is realized as a unitary operator on the
untwisted Hilbert space H by r|nL L |nR R = |nL L |nR R , and on the
oscillators by rxn = xn r, r xn = xn r. A chiral reflection rL should reflect nL
while preserving nR , while a chiral reflection rR should reflect nR while preserving nL .
Although these actions of the chiral operators rL and rR are natural and unique, the
splitting r = rL rR generally leaves some phases undetermined. Truly chiral operators will
be obtained only if the phases associated with rL only depend upon nL (and the phases
associated to rR only depend on nR ). It may not always be possible to achieve this, in
which case the splitting of the operator r = rL rR may be viewed as anomalous. Here,
models will be sought which are anomaly free. Thus, the following action of rL and rR
will be postulated
rL |nL L |nR R = L (nL )| nL L |nR R ,
rR |nL L |nR R = R (nR )|nL L | nR R ,
rL xn = xn rL ,
rR xn = xn rR ,
rL xn = xn rL ,
rR xn = xn rR .
(4.1)
where L and R are phases which may depend on nL and nR respectively. The
requirement that r = rL rR is a further constraint on the phases: L (nL )R (nR ) = 1
whenever nL + nR 2Z. Furthermore, note that the choice of L (nL ) and R (nR ) dictates
the order of the chiral operators rL and rR . It will be shown next thatcontrary to naive
expectationthe order of the operators rL and rR cannot be 2, but must be larger.
4.1. The order of the chiral operator rL
In order for the expression (1.1) to correspond to a trace of projections onto invariant
subspaces, the operators rL and rR should have the same order in all twisted sectors.
Therefore, it suffices to rule out the order 2 in the untwisted sector. In this sector, the
Hilbert space is well known, and the partition traces for the rL theory easily computed.
The partition trace Z 1 1 is, of course, the same as Z 0 0 in (3.6), and Z rL 1 may be computed
using the definition of rL given above
Z 1 1 ( ) =
[0|0]
( ) +
( ),
2

1
Z rL 1 ( ) = L (0) 0
( ) [0|0]( ).
(4.2)
2
In the second line above, the insertion of rL in the trace causes only the states generated
from nL = 0 to contribute, and produces a factor ()k when acting on the state
xn1 xnk |nL L . Note that the traces in the untwisted sector only involve L (0). Under
+ 1, the partition trace Z rL 1 is invariant, as expected.
It will be assumed that the partition traces satisfy modular covariance (possibly when
raised to some critical power d), and it will be shown that the assumption rL2 = 1 leads to
145
a contradiction. Indeed, if rL2 = 1, only the traces Z 1 rL and Z rL rL in the sector twisted by
rL remain to be determined. Modular covariance gives,

1
Z 1 rL ( ) = Z rL 1 (1/ ) = L (0)
0 [0|0] +
0 ( ),
4
2

1
1
1
rL1
1
i/8
i/2

[0|0] + e

0 ( ).
Z rL ( ) = Z rL ( + 1) = L (0)e
4 2
2
(4.3)
rL1
= 1 and thus
= rL ) implies that, rL ( + 2) =
1
1
i/4
L (0)e
[ 4 |0]( [0|0] [ 2 |0])( ). But this result is clearly distinct from Z 1 rL ( ), and
Assuming that rL is of order two,
(rL2
Z1
hence violates modular invariance and in particular the level matching condition of (1.4),
irrespective of the dimension d. Therefore, the order of rL cannot be 2. On the other hand,
an analogous calculation yields Z 1 rL ( + 4) = ei/2 Z 1 rL ( ). Therefore, rL is of order 4
provided that additionally the dimension d is a multiple of 4 and that L (0)4 = 1.
4.2. Chiral reflections of order 4
Henceforth, it will be assumed that rL is of order 4, and that the dimension d is a
multiple of 4. In the preceding subsection, it was shown that these conditions are necessary
in order to have modular invariance and a consistent Hilbert space formulation. It was also
necessary for the factor L (0) to be a 4th root of unity. Since all effects of L (0) will
disappears from all the partition traces in all dimensions d which are multiples of 4, we may
simply set L (0) = 1 without loss of generality. The traces Z 1 1 and Z rL 1 are unchanged
from (4.2) but are now considered with L (0) = 1. Using modular covariance and charge
conjugation symmetry, one has the following traces,

1
rL
rL3
Z 1 ( ) = Z 1 ( ) = 0
( ) [0|0]( ),
2

1
Z 1 rL ( ) = Z 1 r 3 ( ) = Z rL 1 (1/ ) =
0 [0|0] +
0 ( ),
L
4
2
b
rL ( ) = Z
rLb
+ b)

1
b
1
ib/8
ib/2
=e

[0|0] + e

0 ( ),
4 2
2

1
3
1
2
Z rL r 2 ( ) = Z rL r 2 ( ) = Z rL rL (1/ ) = ei/4 0
0 ( ).
L
L
2
2
Z rL
1
rL3 ( ) = Z rL (
(4.4)
Note that the partition traces in the untwisted sector and in the sector twisted by rL2
indeed yield different results.
4.2.1. The sector twisted by rL
The next key step is to construct a Hilbert space HrL for the sector twisted by rL which
can reproduce the traces Z 1 rL and Z rL rL . Here, we encounter the key difficulty of how
146
to determine the pairing between left and right movers in the twisted sector. Specifically,
the chiral operator rL produces two twisted ground states, which may both be taken to be
eigenstates, rL | L | L . Therefore, the Hilbert space in the sector twisted by rL should
be a subspace of the full direct product space of left and right movers

H rL = xn1 + 1 xnk + 1 |+ L xn1 x nk |nR R , nR Z
2
2

xn1 + 1 xnk + 1 | L x n1 xnk |nR R , nR Z .
(4.5)
2
To determine which subspace HrL is the appropriate choice and how rL should act on this
space, use will be made of the partition traces Z 1 rL and Z rL rL , which were obtained earlier
by modular covariance.
The partition trace Z 1 rL plays the role of a spectral density, as each state contributes
a factor q L0 q L0 times its multiplicity factor. The spectral density (for a theory of bosons
only) suffers no accidental cancellations and may be used as an accurate guide for how left
and right chiralities are to be paired against one another. The expansion of Z 1 rL in powers
of q and q suggests defining HrL by the following pairing

HrL = xn1 + 1 xnk + 1 |+ L xn1 x nk |nR R , nR 2Z
2
2

xn1 + 1 xnk + 1 | L x n1 xnk |nR R , nR 2Z + 1 .
(4.6)
2
The operator theoretic trace TrHrL (q L0 q L0 ) agrees then with Z 1 rL ( ), up to a factor of

ei/8 . Taking the dimension d to be a multiple of 16, complete agreement is obtained.
b
More generally, the partition traces Z rL rL , with b Z play the role of a spectral
density weighed by the eigenvalues of the operator rLb on the subspace HrL , as each state
contributes a factor rLb q L0 q L0 times its multiplicity factor. Clearly, when b is not a multiple
of 4, this quantity may suffer from accidental cancellations between states with identical
conformal weight but different rLb eigenvalue. Therefore, this partition trace will not, in
general, determine the action of rLb is a unique manner, and one will have to restrict to
searching for internally consistent choices. Taking the operator rL to have order 4 on HrL ,
one has the following natural assignment
rL |+ L |nR R = |+ L |nR R ,
rL | L |nR R = e
12 i
nR 2Z,
| L |nR R ,
nR 2Z + 1
(4.7)
and the usual commutation relations with the oscillators. With this, the evaluation of all
traces in the sector twisted by rL may be completed, and one finds for b Z,

1
b
1
b
Z rL rL ( ) =
(4.8)
[0|0] + eib/2
0 ( ).
4 2
2
For b = 0, 1, this formula indeed reproduces the result of (4.3), (which was obtained by
modular transformations from partition traces in the untwisted sector) up to an overall
factor of eib/8 . Therefore, the Hilbert space formulation of the operator rL in the
sector twisted by rL , given above, will be consistent with modular covariance only if the
dimension d is a multiple of 16.
147
4.2.2. The sector twisted by rL3

Charge conjugation symmetry, represented by the operator C = S 2 , ensures that the
partition traces behave naturally under the reversal of orientation of the homology cycles,
1
and implies the following relation on the partition traces, Z g h ( ) = Z g h1 ( ). As the
dimension is always assumed to be even, the factor is immaterial and will be omitted.
The charge conjugation relation may be used to investigate the partition traces and the
Hilbert space in the sector twisted by rL3 = rL1 . Applying charge conjugation to (4.8), it is
readily derived that for b Z,

1
b
1
rLb
ib/2
Z r 3 ( ) =
(4.9)
[0|0] + e

0 ( ).
L
4 2
2
Notice the difference in the relative phase factor between the two terms in the parenthesis
above when b is odd. As a result of this phase difference the structure of the Hilbert space
Hr 3 is very close to that of HrL but not identical, as the operator rL has a different action
L
on both spaces. Therefore, extra care is needed in constructing also this space correctly.
The two degenerate ground states in the sector twisted by rL3 will be denoted by | CL ,
where the superscript C stands for charge conjugation. The Hilbert space is

Hr 3 = xn1 + 1 xnk + 1 |+ CL xn1 x nk |nR R , nR 2Z
L
2
2

xn1 + 1 xnk + 1 | CL x n1 xnk |nR R , nR 2Z + 1
(4.10)
2
and the analysis of the partition traces in (4.9) leads to the following definition of the action
of the operator rL on Hr 3 ,
L
rL |+ CL
rL | CL
|nR R = |+ CL |nR R ,
|nR R = e
+ 12 i
nR 2Z,
| CL |nR R ,
nR 2Z + 1
(4.11)
and the usual commutation relations with the oscillators. With these assignments, all
partition traces in Hr 3 are consistent with modular covariance and the values of the
L
remaining partition traces Z rL 1 = Z rL 1 .

4.2.3. The sector twisted by rL2
It remains to analyze the sector twisted by rL2 . Since rL is not of order 2 but rather 4, the
Hilbert space Hr 2 in the sector twisted by rL2 is expected to be distinct from the untwisted
L
Hilbert space. The starting point is the partition trace with rL2 inserted in the untwisted
sector. The eigenvalues of rL2 in the untwisted sector are deduced from (4.1),
rL2 |nL L |nR R = L (nL )L (nL )|nL L |nR R .
(4.12)
Since
= 1, the combination L (nL )L (nL ) can take eigenvalues 1 only. There are
two natural assignments,

I L (nL )L (nL ) = 1
for all nL Z,
(4.13)
II L (nL )L (nL ) = (1)nL for all nL Z.
rL4
148
In the next subsection, it will be shown that these are not only natural assignments, but in
R symmetry of the theory.
L SU(2)
fact the only ones that are compatible with the SU(2)
2
In case I, rL acts as the identity in the untwisted sector and modular covariance requires
2
2
1
Z 1 = Z rL 1 = Z rL r 2 = Z 1 r 2 . The full partition function is therefore
L

d

d
d
2
1

2 d 1
1
1
ZI = [0|0] +

0
+ 0
[0|0] + 0
0
2
2
2
2
2
2

d

d
3

1
1
b
1
1
+

[0|0] + e 2 ib
0
.
(4.14)
2
4 2
2
b=0
This theory appears to have the undesirable property that the order of rL is 4 in the sector
twisted by rL but that the order is only 2 in the untwisted sector. The counting of the low
lying states suggests that the theory is inconsistent.
In case II, we have the following partition trace in the untwisted sector
2
1

2
2
Z rL 1 =
[0|0]

.
(4.15)
2
Modular covariance may be used to derive the remaining partition traces

1
1
Z r 2 = [0|0]
0 + [0|0]
0 ,
L
2
2

1
3
1
Z rL r 2 = Z rL r 2 = 0
0 ,
L
L
2
2

1
rL2
Z r 2 = [0|0]
0 [0|0]
0 .
L
2
2
(4.16)
Proceeding as in the other twisted sectors, the Hilbert space Hr 2 is obtained from Z 1 r 2 ,
L
L

Hr 2 = xn1 xnk xn1 xnk |nL L |nR 1 R , nL + nR 2Z .
(4.17)
L
The trace over this Hilbert space of the operator q L0 q L0 indeed reproduces Z 1 r 2 . The
L
action of rL on Hr 2 is inferred from Z rL r 2 . The following phase assignment will be

L
L
assumed, consistently with the fact that the chiral parts of Hr 2 are identical to the chiral
L
parts of the untwisted sector,
rL |nL L |nR 1 R = L (nL )| nL L |nR 1 R .
(4.18)
We continue to assume that L (0) = 1 without loss of generality. As only states with
3
nL = 0 contribute to Z rL r 2 = Z rL r 2 , these partition traces are readily reproduced by the
L
above assignments. From the expression for Z rL r 2 , it is clear that rL2 has eigenvalue +1
L
when nL is even, and 1 when nL is odd. Therefore, (nL ) must satisfy
L (nL )(nL ) = (1)nL
(4.19)
149
consistently with the fact that rL4 = 1. There is no canonical choice L satisfying this
relation; a convenient choice is L (nL ) = exp{i|nL |}, which has the additional property
that L (nL )4 = 1 for all nL Z.
Altogether, the rL theory with d 16N fields is modular covariant, possesses a
consistent Hilbert space interpretation and has the following full partition function

2 d
2
1
ZII =
[0|0] +
4
2
=1

d

1
1
+
[0|0]
0 + [0|0]
0
4
2
2
=1

d

d
d
1
1
1
1
+ 0
[0|0] + 0
0
2
2
2
2
2
d
d

3
1 1
b
1
1
ib
+

[0|0] + e 2
0
.
2
4 2
2
(4.20)
b=0
Term by term comparison reveals that this partition function is identical to (the complex
conjugate of) the partition function of the model twisted by sR , given by combining (3.6)
with (3.16) for dimension d = 16. The significance of this identification will be elucidated
in the next section.
R
L SU(2)
4.3. The symmetry SU(2)
At the self-dual radius, R 2 = 22 , the conformal field theory of a single boson possesses
L SU(2)
R . The generators of SU(2)
L are the holomorphic currents
the symmetry SU(2)
J + = 1 (J 1 + iJ 2 ) = e+i 2x+ ,
L
L
L
2
JL3 = x+ ,
(4.21)
J = 1 (J 1 iJ 2 ) = ei 2x+ ,
L
L
L
2
where x+ denotes the left chiralpart of the field x. Using the fact that under sL , one has the
transformation x+ x+ + / 2, we are in a position to obtain the transformation laws
of these generators under the chiral operators (see, e.g., [7])

a
rL JL1 rL = +JL1 ,
rL JL2,3 rL = JL2,3 ,
JL , rR = 0,
a

JL , sR = 0,
(4.22)
sL JL3 sL = +JL3 ,
sL JL1,2 sL = JL1,2 .
Thus, rL and sL both correspond to SU(2)L rotations by 180 degrees, rL about the 1-axis
and sL about the 3-axis. This observation explains (1) why the orders of rL and sL must be
equal to 4; (2) why the critical dimension of the asymmetric orbifold models constructed
from them coincide and are multiples of 16; (3) why their partition functions coincide.
The presence of this enhanced symmetry also has consequences for the phase
assignments L (nL ) associated with rL . The operator rL2 corresponds to a 360 degree
rotations and, as expected, commutes with the currents JLa and must thus assume constant
L . In the untwisted sector,
eigenvalue throughout any irreducible representation of SU(2)
150
two representations occur, one of spin 0 corresponding to nL even, and one of spin 1/2
corresponding to nL odd. The first representation of rL2 would be the trivial (identity)
representation, but this was ruled out for being inconsistent with modular invariance. The
only other non-trivial representation is then
rL2 |nL = ()nL |nL ,
rL |nR = |nR ,
rR2 |nR = ()nR |nR ,
rR |nL = |nL .
(4.23)
2
in the
Notice that these assignments are in complete analogy with the action of sL,R
untwisted sector. For sL,R , however, the phase assignment of their squares dictated their
own action. For rL,R , the condition is weaker as the invariant generator JL1 is not diagonal
in the basis |nL L . It simply amounts to the relation L (nL )(nL ) = (1)nL , already
derived in (4.19).
4.3.1. Interpretation in terms of chiral blocks constructions

As we did earlier for the asymmetric sL theory, we can give an interpretation for the
asymmetric rL theory in terms of chiral blocks of the symmetric r = (rL , rR ) theory. Now
the chiral operator rL is of higher order than the symmetric operator r = (rL , rR ), and at
first sight, there seems to be no way of differentiating between the conformal blocks of, say,
rLb and rLb+2 . But the one-loop case shows that the blocks come with a specific assignment
of phases, and it is these phases which distinguish between the blocks of rLb and rLb+2 .
More precisely, let b, a Z4 as before, with b expressed uniquely in terms of {0, 12 }
and c {0, 1} as in (3.11). In [10], it was shown that the orbifold of the circle theory at
self-dual radius by r = (rL , rR ) coincides with the circle theory at half the self-dual radius.
Thus the chiral blocks of the r = (rL , rR ) theory and the chiral blocks of the untwisted
theory are given as before by (3.10). This time, our formulas obtained earlier by Hilbert
space methods show that they are paired differently in order to give the partial traces of the
asymmetric rL theory. In fact, the pairing matrix is K = e2i(c+2) , and the partial
traces given by

1
1
2i(c+2)
Z a ( ) =
e
+ a
(0, 2 )[ |0](0, 2 ).
4
|( )|2
1
b
(4.24)
{0, 2 }
Hence the partition function of the rL theory for d bosons is given by

d

1
1
1
2i(c+2)
Z( ) =
e
+ a
(0, 2 )[ |0](0, 2 ) .
2
4
4
g |( )|
1
a,bZ4
2Z
(4.25)
As noted before, the partition functions of the sL and the rL theories are the same, although
the blocks for the assignment sLb and sLa on the B and A cycles differ by phases from the
blocks for the assignment of rLb and rLa on the same cycles.
151
5. Orbifolds defined by chiral shifts and reflections

The next step towards analyzing the models of KachruKumarSilverstein is the
consideration of asymmetric orbifold models which involve both chiral shifts sL,R and
chiral reflections rL,R . First, a detailed study is presented of a purely bosonic model with
only a single chiral generator f (rL , sR ); second, the model is extended to include two
chiral generators f and g (sL , rR ); third, fermions are included as well and the full KKS
model is analyzed.
5.1. The f = (rL , sR ) model
The proper definition of the chiral operator f involves some of the same delicate issues
that arose when defining rL . First, in the untwisted sector, f will involve the phases
L (nL ),
f |nL L |nR R = L (nL )einR /2 |nL L |nR R ,
f 2 |nL L |nR R = L (nL )L (nL )(1)nR |nL L |nR R .
(5.1)
Depending on whether L (nL )L (nL ) equals 1 or (1)nL , the order of rL
is 2 or 4, while
the order of f will be 4 or 2, respectively (the latter since nL + nR 2Z). Which of these
choice (if any) leads to a consistent f -orbifold is analyzed below.
5.1.1. Ruling out the case of trivial phases L (nL )
It is first shown that the naive definition of rL , for which L (nL )L (nL ) = 1 for
all nL Z, and f is of order 4, is incompatible with the dual requirements of modular
covariance and a proper Hilbert space interpretation. With this naive choice, one has
f 2 = (1, sR2 )
= 1. The 4 partition traces in the untwisted sector are readily evaluated to
be
2
1

2
Z 1 1 =
[0|0]
+
,
Z f 1 =
,
2
2
2
1

2
3
Z f 1 =
[0|0]

,
(5.2)
Z f 1 =
.
2
2
2
Note that the relative sign in Z f 1 appears due to the presence of sR2 in f 2 . Assuming
modular covariance, all other partition traces may be derived from (5.2); they are
1
1
f2
f2
Z f = Z f 3 = Z f = Z f 3 = 2
,
4
1
1
3
3
Z f f = Z f f 3 = Z f f = Z f f 3 = 2
(5.3)
4 2
in the sectors twisted by f and f 3 , and

1
Z 1 f 2 = [0|0]
0 + [0|0]
0 ,
2
2
Z f f 2 ( ) =
,
2
152
f2
f2
=e
i/2

1
[0|0]
0 [0|0]
0 ,
2
2
f3
f 2 ( ) =
0
(5.4)
in the sector twisted by f 2 . Since f 2 = (1, sR2 ), the Hilbert space of the sector twisted by
f 2 may easily be constructed and one has

Hf 2 = xn1 xnk |nL L xn1 xnk |nR 1 R , nL + nR 2Z .
(5.5)
This allows an independent evaluation of Z f f 2 ( ) as an operator trace; one finds

TrHf 2 f q L0 q L0 =
[ 12 | 12 ](0, )
1

1/24
n=1 (1 + q
n)
( )
= 0.
(5.6)
This contradicts the earlier, non-vanishing, formula for Z f f 2 ( ) obtained from modular
covariance
0
= Z f 1 Z 1 f Z f f Z f
Z f f 2
= 0.
(5.7)
Therefore, the assumption that the phase assignment of rL is trivial cannot lead to a
consistent orbifold theory.
5.1.2. Consistency of the case of non-trivial phases L (nL )
For the non-trivial phase assignment, L (nL )L (nL ) = (1)nL , one has f 2 = 1 in the
untwisted sector and the theory truncates. The traces Z 1 1 and Z f 1 in the untwisted sector
do not depend on the choice of L (nL ) (having set L (0) = 1 without loss of generality),
and are still given by their expressions in (5.2). Since the traces Z 1 f and Z f f follow by
modular covariance, they are also given by their expressions in (5.3).
The only issue that remains to be verified is that the traces in the sector twisted by
f can indeed be realized as partition traces in a twisted Hilbert space Hf , where f is
realized as an operator of order 2. It is well known [6,7] that the chiral sector twisted
by rL contains two degenerate ground states, which we shall denote by | L . Also, the
chiral sector twisted by sR contains two sectors |nR 1/2 R with nR either even or odd.
Following the method of Section 4, the Hilbert space is found to be

1
Hf = (xn1 + 1 xnk + 1 |+ L xn1 xnk
nR
, nR 2Z
2
2
2 R

(xn1 + 1 xnk + 1 | L xn1 xnk

nR , nR 2Z + 1 . (5.8)
2
2
2
Note that the pairing between the | L and |nR 1/2 >R provides a highly non-trivial
piece of information on how the left and right chiral Hilbert spaces need to be combined.
The pairing indeed implies that out of the 4 ground states arising from a full tensor product
of left and right, only 2 are to be retained in the asymmetric orbifold theory. The same
pairing issue arose already in the theory twisted by the symmetric operator r = (rL , rR ),
153
where also only 2 states out of a total of 4 should be retained.3 It is readily verified that
L L
1
0
0
Z f = TrHf q q
(5.9)
= 2
4
as desired. To obtain Z f f from Hf as well, it suffices to define rL on the momenta ground
states in Hf as

1
1
rL |+ L
nR
= |+ L
nR
, nR 2Z,
2 R
2 R

1
1
= i| L
nR
, nR 2Z + 1
rL | L
nR
(5.10)
2 R
2 R
and similarly for excited states. With this construction, TrHf (q L0 q L0 ) is indeed found to
agree with the expression found earlier for Z f f by modular covariance.
Altogether, the theory obtained this way is modular covariant and admits a Hilbert space
realization in terms of twisted sectors. Its partition function in dimension d 4N is given
by

2d

2
1

2 d
Z=
+
[0|0]
+
2
2
2

2d

2d
1
1
+ 2d
+ 2d
(5.11)
4
4 2
which coincides with that of the d-dimensional circle theory orbifolded by r = (rL , rR ).
5.2. Orbifolds generated by f and g
Next, we consider the theory of d bosonic scalar fields moded out by the group G
generated by the two elements f = (rL , sR ) and g = (rR , sL ). This group is Abelian;
indeed, in view of (5.1), one has in the untwisted sector
f |nL L |nR R = L (nL )einR /2 |nL L |nR R ,
g|nL L |nR R = R (nR )einL /2 |nL L | nR R
(5.12)
and therefore
gf |nL L |nR R = ei(nR nL ) fg|nL L |nR R .
(5.13)
In view of the fact that in the untwisted sector nL + nR 2Z, the phase factor on the rhs is
just 1, and one has fg = gf . Since also f 2 = g 2 = 1, the orbifold group generated by f, g
is just the group Z2 Z2 . As usual, it taken for granted that the group composition laws
remain the same in all sectors.
3 Although the issue is well known in the symmetric theory [6], it has nonetheless given rise to confusion
when the proper pairing was not taken into account.
154
5.2.1. The sectors twisted by f and g

n
n
The partition traces Z f 1 , (and by interchanging left and right movers also Z g 1 ),
have already been determined in (5.2) and we have e.g. Z f 1 = Z g 1 = | [0| 12 ]|2 . By
modular transformation, the partition traces are found in the twisted sectors, Z 1 f = Z 1 g =
2| [ 14 |0]|2 . The Hilbert space Hf in the sector twisted by f , (and by interchanging left and
right movers also the Hilbert space Hg in the sector twisted by g) was given in (5.8).
The novel partition traces to be calculated in Hf are Z g f and Z fg f . They belong to a
modular orbit that does not include any partition trace evaluated in the untwisted sector,
and is thus not derivable from preceding results. To evaluate Z g f , it suffices to find g on
the ground states of Hf ,

1
1
= sL | L rR
nR
g| L
nR
2 R
2 R

1
.
sL | L
nR +
(5.14)
2 R
Here, the sign has been used, because a phase factor may arise when applying rR to | R
states. The half-integer moding of the | R states alone guarantees that the operator g has
no diagonal matrix elements in Hf , and must therefore have vanishing trace, Z g f = 0.
Modular invariance then implies that all partition traces in the same orbit must vanish
Z g f = Z f g = Z fg f = Z f fg = Z fg g = Z g fg = 0.
(5.15)
In the sector Hf , the vanishing of the partition trace

f holds for the same reasons as
the vanishing of Z g f did, namely that fg has vanishing diagonal matrix elements in Hf .
The verification of the traces Z f fg and Z g fg will require the construction of the Hilbert
space Hfg in the sector twisted by fg, to be given below.
Finally, it is worth noting that although fg = gf when f , g are viewed as elements
of the point group, they do not commute when they are viewed as elements of the full
orbifold group, since fgf 1 g 1 = (sL2 , sR2 ). For asymmetric orbifold groups, we are
aware of no arguments based on first principles that would guarantee the vanishing of
the torus contribution for twists f, g that do not commute in the full orbifold group (but do
commute in the point group). Nonetheless, this vanishing actually does take place in the
model considered here.
Z fg
5.2.2. The action of sL,R in the sector twisted by rL,R

For the sake of completeness, as well as for later use, we briefly discuss the action of sL
on the twisted states | L , which enters (5.14) but is not actually needed for the evaluation
of Z g f . In the leftright symmetric conformal field theory of a single scalar field, twisted
by the symmetric reflection operator r, the symmetric ground states | are associated
with the fixed points of r. Applying r to the circle of radius R, there are two solutions
to the fixed point equation r(x) x (mod 2R), namely x = 0 and x = R. Under r,
each fixed point is mapped into itself, which is tantamount to r|+ |+ and r| | ,
and the state |+ may be associated with the fixed point 0 while | is associated with
the fixed point R. On the other hand, the operator s acts by shifts s(x) = x + R and
thus interchanges the two twisted ground states, s|+ | and s| |+ . It is natural
155
and internally consistent to induce analogous actions on the chiral twisted states, namely
rL | L | L and sL | L | L , and rR | R | R and sR | R | R .
5.2.3. The sector twisted by fg
The only partition trace in the untwisted sector which has not yet been determined is
Z fg 1 ( ). Since the product fg involves reflection operators on both left and right movers,
only states with nL = nR = 0 will contribute in the untwisted sector and the partition trace
manifestly reduces to that of the symmetric operator r = rL rR ; it is readily computed and
one finds
fg
Z 1 =
.
(5.16)
2
Modular covariance implies
Z 1 fg = 2
,
4
1
1
Z fg fg = 2
.
4 2
The first result yields the Hilbert space of the sector twisted by fg

Hfg = xn1 + 1 xnk + 1 xn1 + 1 xn + 1 | .
2
(5.17)
(5.18)
Here, | are two ground states in the sector twisted by fg = (rL sL , sR rR ), both of which
are of conformal weight 1/16.
If the presence of the translation operators sL and sR were ignored, fg would coincide
with the non-chiral reflection r, and | = | would just be the ground states in the sector
twisted by r. The problem with this simplified picture is that the vanishing of Z f fg does
not allow for a chiral formulation of the states | and the operator fg consistent with the
relations rL |+ L = |+ L and rL | L = i| L , used successfully in other sectors, such
as (4.7), (4.11), (5.10).
If the presence of the translation operators sL and sR is carefully taken into account,
a consistent chiral formulation of the twisted states | and the action thereupon by
the chiral operators rL,R and sL,R does exist and reproduces the corresponding partition
traces predicted from modular covariance. The key observation is that the states twisted by
(rL , rR ) and by fg = (rL sL , sR rR ) are different but isomorphic to one another. Following
the spirit of Section 5.2.2, the twisted states may be associated with geometrical fixed
points in the symmetric theory. While the fixed points of (rL , rR ) on the circle are 0 and
R, those of fg = (rL sL , sR rR ) are shifted by R/2, i.e., they are +R/2 and R/2.
Denoting associated twisted states by |+ and | respectively, it is now clear that one
should expect to have r|+ | , r| |+ , as well as s|+ | , s| |+ .
On the other hand, the states | are eigenstates of the operators rs.
The above geometry-inspired picture may be realized concretely in terms of chiral states
and operators. We start with the chiral ground states | L,R of the sectors twisted by rL
and rR and the action of the chiral reflection operators on these states
rL |+ L = |+ L ,
rL | L = i| L ,
rR |+ R = |+ R ,
rR | R = +i| R .
(5.19)
156
Using the standard relations sLa rL sLa = rL and sRa rR sRa = rR for any a R, one deduces
(rL sL )|+ L = |+ L ,
(rL sL )| L = i| L ,
(sR rR )|+ R = |+ R ,
(sR rR )| R = +i| R ,
| L sL 2 | L ,
+1
| R sR 2 | R .
(5.20)
| L
Thus, the action of the operator rL sL on the twisted states

is isomorphic to the
action of rL on the twisted states | L . The action of the chiral operators rL,R on | L,R ,
however, is now calculable from the above definitions, from sLa rL sLa = rL , and from the
fact that sL,R | L,R | L,R , and we find
rL | L | L ,
rR | R | R .
(5.21)
This relation readily guarantees that f and g have vanishing diagonal matrix elements in
Hfg , so that Z f fg = Z g fg = 0.
The final relation to be implemented is the partition trace Z fg fg = 2| [ 14 | 12 ]|2 ,
which effectively requires that fg be the identity on both states | . In view of fg =
(rL sL , sR rR ) and (5.20), this provides with a unique correspondence between the chiral
and non-chiral twisted states and we have
|+ = |+ L |+ R ,
| = | L | R .
(5.22)
With this construction, all the partition traces of (5.17) and (5.15) are indeed reproduced
in a chiral fashion. The partition function for the orbifold generated by f and g is, in
dimension d 4N is therefore given by

2d

2
1

2 d
Z=
[0|0]
+
+ 3
4
2
2

2d

2d
1
1
+ 3 2d
+ 3 2d
.
(5.23)
4
4 2
Notice that this partition function involves the same blocks as that of the orbifold by f
alone, but the relative proportions of the three non-vanishing modular orbits is different.
5.3. Including worldsheet fermions
It is straightforward to include worldsheet fermions in models twisted by shifts s and

reflections r. Given a canonical homology basis of 1-cycles A, B, the worldsheet fermions
are defined with a spin structure = ( | ). The models of greatest interest are those with
worldsheet supersymmetry [35]. Invariance of the matter supercurrent Sm = 1/2+ z x
thus forces to undergo the same transformation as the bosonic field z x. Shifts do
not act on . Reflections may be parametrized by a half-characteristic = ( | ), with
, = 0, 12 , such that is double-valued around the cycle D = A + B. The

combined boundary conditions due to the spin structure and the reflection are,
+2
(z + 1) = ()2 +2 (z),
(z + ) = ()2
(z).
(5.24)
157
The chiral partition function is given by

F [; ] =
[ + ](0, )
.
( )
(5.25)
When + equals ( 12 | 12 ), the chiral partition function vanishes, F [; ] = 0.

Notice that while the bosonic models twisted by r and by s are identical, once fermions
are included, they will differ, since the fermions are twisted by r but not by s.
6. Asymmetric orbifolds via chiral splitting

The operator and Hilbert space methods used earlier in this paper to construct the
partition traces and full partition functions for asymmetric orbifolds on a torus worldsheet
do not easily generalize to higher loop order. The construction of symmetric orbifolds is, of
course, well-understood, both in the functional integral and the operator formulations (see,
for example, [57,10,11]). The construction of the chiral blocks has also been extensively
investigated using operator methods, the chiral OPE, and the implications of modular
invariance [10,11]. Furthermore, in recent work [22], the full Z2 -twisted chiral blocks were
calculated in the presence of non-trivial supermoduli and simple expressions in terms of functions were derived. It is the construction of the full partition functions for asymmetric
orbifolds that remains much less well-understood.
By chiral splitting, it is clear that the correct blocks of an asymmetric orbifold theory
should be defined as the products of the blocks of the left and right chiral halves. To
be concrete, the asymmetric orbifold group G is viewed as a subgroup of the product
GL GR of left and right groups GL and GR , so that elements of G may be labeled as
pairs (gL ; gR ), with gL GL and gR GR . The starting point of the proposed construction
is two symmetric orbifolds, one with group GL , the other with group GR . For simplicity,
the discussion will be carried out here for the torus, with boundary conditions around a
single A and a single B cycle; the generalization to higher genus are analogous.
By chiral splitting, the blocks are given by |F gL hL (pL ; )|2 and |F gR hR (pR ; )|2
respectively, i.e., the left and right chiral blocks are complex conjugates of one another.
Here, pL , pR stand for any extra labels of the chiral blocks, such as, for example,
dependence on the internal loop momentum. A block Z g h of the asymmetric orbifold with
h = (hL ; hR ) and g = (gL ; gR ) is then defined as the product of the corresponding chiral
blocks; as illustrated by the diagram
Symm
F gL hL
Symm
F gR hR
g
F L hL

F gL hL

Asymm F gL hL F gR hR
F gR hR

F gR hR
158
and may be expressed by the following formula,

Z (gL ;gR ) (hL ;hR ) (pL , pR ; , ) = F gL hL (pL ; )F gR hR (pR ; ).
(6.1)
The key difficulty with asymmetric orbifolds resides in the precise pairing between the left
and right chiral blocks, i.e., in the superposition of the blocks Z (gL ;gR ) (hL ;hR ) .
Following [22], the superposition of the blocks may be expressed in terms of a pairing
matrix on left and right chiral blocks, which is denoted K. The full partition function is
then given by4
ZG (, ) =
1
K(pL , gL , hL ; pR , gR , hR )F hL gL (pL ; )
|G|
g,hG pL ,pR
h

F R gR (pR ; ) .
(6.2)
Here the summations are over g = (gL ; gR ) and h = (hL ; hR ). Clearly, the matrix K
cannot depend upon moduli , as this would violate chiral splitting. Many models
admit the same symmetrized theories, and hence the same chiral blocks. It is then the
pairing matrix K which differentiates between different models, and the key issue is its
determination.
Modular invariance places strong constraints on K, which may be derived as follows.
The modular transformations of the blocks of the symmetric theory are as in (1.2). Using
chiral splitting, it follows that the transformations of the chiral blocks themselves are
known up to phases and mixing coefficients S, which are independent of moduli,
1
F g h (pL ; + 1) = ei(g,h) F gh h (pL ; ),

1

F g h (pL ; 1/ ) =
S(pL , pL
)F h g (pL
; ).
(6.3)

pL
A sufficient condition for modular invariance of the full partition function is the
requirement that K satisfy the following relations

1
K pL , gL h1
L , hL ; pR , gR hR , hR
= ei(gL ,hL )i(gR ,hR ) K(pL , gL , hL ; pR , gR , hR ),

1
K pL ; h1
L , gL ; pR , hR , gR

=
S(pL , pL
)S(pR , pR
) K(pL , gL , hL ; pR , gR , hR ).
(6.4)
,p
pL
R
These conditions are also necessary when all the chiral block-functions F g h are linearly
independent. If the block-functions exhibit non-trivial linear dependences, the matrix K
4 It is useful to note that, when considering the compactification and orbifolding of several dimensions d > 1,
the group elements gL , hL , gR , hR represent the action of the orbifold group on all dimensions and the chiral
blocks F represent the chiral blocks for d dimensions. When the d dimensions are orthogonal, as has been the
case in this paper, the blocks F themselves are the product of d one-dimensional blocks and the action of a group
element g accordingly decomposes into d group elements each acting on a one-dimensional block.
159
can be reduced to pair only the linearly independent bock-functions; modular invariance
may then be expressed on this reduced K matrix just as in (6.4).
Modular invariance cannot, in general, determine K completely, even when all the
block-functions are linearly independent. The action of modular transformations
(a + b)(c + d)1 (with a, b, c, d Z and ad bc = 1) induces the following
transformations on pairs of group elements

(gL , hL ) gLa hbL , gLc hdL ,

(gR , hR ) gRa hbR , gRc hdR .
(6.5)
Generally, this action will decompose into a several disjoint modular orbits. Clearly, the
modular transformation laws of K in (6.4) cannot serve to relate K on different orbits; this
information must come from physical input, derived from the Hilbert space formulation
and the proper actions of the chiral group elements gL , hL .
Since K is independent of moduli, its value may be determined by taking various
degeneration limits of moduli space in which the corresponding chiral blocks do not all
vanish. For example, in the case of genus two, the separating degeneration will produce two
tori with prescribed twist sectors. On each of these tori, the matrix K is known. Therefore,
the genus two matrix K will be known for the subset of twists for which the genus one
limits of the chiral blocks are non-vanishing (and linearly independent). We expect that the
matching of the Hilbert space construction with the expression of (6.2) produces a unique
pairing matrix K to higher genus.
Acknowledgements
We are happy to acknowledge helpful conversations with Constantin Bachas, Michael
Dine, Michael Gutperle, and Per Kraus. We are especially grateful to Shamit Kachru and
Eva Silverstein for helpful discussions and correspondence on asymmetric orbifolds.
Appendix A. Orbifolding by symmetric orbifold groups

In this appendix, some concrete orbifold partition functions for symmetric orbifolds are
constructed. The general prescription in terms of conjugacy classes is reviewed. Finally,
simple examples of symmetric orbifolds based on shifts and reflections are worked out.
A.1. The functional integral and conjugacy class prescription
Consider a scalar field x with classical action S[x] on a torus worldsheet with modulus
subject to the following twisted boundary conditions B on the A- and B-cycles,
corresponding to z z + 1 and z z + respectively

x(z + 1) = hx(z),
B
(A.1)
x(z + ) = gx(z).
160
The functional integral with boundary conditions B is denoted by Z g h ( ) and given by

g
Z h ( ) Dx eS[x] .
(A.2)
B
It is well known that when gh

= hg, the above boundary conditions have no solutions,
and such sectors do not contribute to the functional integral. Since the boundary conditions
have also the following two equivalent representations

x(z + 1) = hx(z),
x(z + ) = g x(z),
B
(A.3)
x(z + + 1) = ghx(z)
x(z 1) = h1 x(z)
the following modular identities immediately result from the functional integral representation
1
Z g h ( ) = Z gh h ( + 1) = Z h
g (1/ )
(A.4)
which, in turn, are equivalent to the modular transformation laws of (1.2).

The general prescription for the one-loop orbifold partition function ZG defined by the
group G is given in terms of a summation over all possible twisted boundary conditions
g, h G. The following two formulas are equivalent
ZG =
1
1 g
Z h=
Z Ci h .
|G|
|Ni |
g,hG
hg=gh
(A.5)
hNi
The second expression above is in terms of a summation over all conjugacy classes of G,
which are indexed by i and any representative is denoted by Ci in the above formula. Also,
Ni is the stabilizer of Ci , i.e., the subgroup of all h G such that g commutes with every
element of Ci . The equivalence of both expressions is readily established by using the fact
1
that Z g h = Z ugu uhu1 for all g, h, u G, and by breaking up the sum over all g G
into a sum over g and u with g = ugu1 .
A.2. The Poisson resummation formula
The reformulation of momentum and winding mode summations is carried out using
the Poisson resummation formula. Let A be an invertible n n symmetric matrix and B, C
two n-column vectors

exp Aij mi + C i mj + C j + 2iBi mi + C i
m1 ,...,mn Z
1
(det A)1/2
ij

exp A1 (mi Bi )(mj Bj ) + 2iC i (mi Bi ) . (A.6)
m1 ,,mn Z
The derivation of this formula is standard [24].
161
A.3. Symmetric Abelian orbifolds defined by ZN shifts

Consider the symmetric orbifold generated by ZN shifts, sx = x + 2R/N . In this
subsection, the theory SR1 /s is worked out using the conjugacy class prescription and it is
1
verified to coincides with the theory SR/N
, as should be expected on geometric grounds.
The prescription of the preceding subsection for this case gives
N1
s
1 g
Z h.
ZS 1 /s =
R
N
(A.7)
g,h=1
By definition, Z 1 1 = ZS 1 is the untwisted partition function.

R
General shifted boundary conditions may be specified by the characteristics = ( , )
where , = 0, 1/N, 2/N, . . . , (N 1)/N and the following correspondence with the

group elements, h = s N and g = s N ,

x 1 + 1, 2 = x 1 , 2 + 2R (mod 2R),

x 1 , 2 + 1 = x 1 , 2 + 2R (mod 2R).
(A.8)
The associated instanton solutions and action are given by
( ) = 2R 1 (m1 + ) + 2R 2 (m2 + ),
xm
1 ,m2
2

R 2
S xm
= 2
(m1 + ) (m2 + )
.
1 ,m2
2 2
(A.9)
( ) is now a doubly periodic scalar function. Its functional

The field y( ) x( ) xm
1 ,m2
integral produces the well-known factor produces a factor Det = 2 |( )|4 from the
1
non-zero modes and a factor of (8 2 ) 2 2R/ = R/ 2 from the zero mode of y.

Assembling all contributions, one finds (still with h = s N and g = s N )
R
Z h ( ) =
22 |( )|2
m1 ,m2 Z
2

R 2

exp 2
(m1 + ) (m2 + )
. (A.10)
2 2
Under the modular transformations + 1, one has and , i.e.,

h h and g gh1 in accord with (1.2). Under the modular transformations 1/ ,
one has and , i.e., h g and g h1 , also in accord with (1.2).
The formula (A.10) obtained above has manifest modular transformation properties but
does not exhibit the chiral block structure of the theory. To make the chiral block structure
manifest, Poisson resummation in m2 is carried out, using (A.6) for n = 1, A = R 2 /22 2 ,
B = 0 and C = 1 (m1 + ),
Z g h ( ) =
1
|( )|2

(pL ,pR )R
e2i
(p
R
L +pR ) 2
q 2 pL q 2 pR .
(A.11)
162

Here, the momenta (pL , pR ) span the lattice R , defined by

R

R

R

m2 (m1 + ), m2 + (m1 + ) ,
(pL , pR ) =
R
2
R
2

m1 , m2 Z .
(A.12)
The modular transformation properties of (A.11) of course follow from those of (A.10)
which were discussed after (A.10). But they may also be derived directly from (A.11).
This is manifest for + 1, while for = 1/ , it is achieved by applying the
Poisson resummation formula (A.6) in both m1 and m2 for n = 2 and
R2

2 i
0
2
A = 2
(A.13)
,
B
=
,
C
=

R2
0
i 2 2 2
and one recovers results in accord with (1.2).
R
to be an integer multiple of N , so
The summation over forces m2 = (pL + pR ) 2
that m2 = Nn2 with n2 Z, and eliminates the overall factor of N1 . Next, the summation
over is reproduced by replacing m1 with N1 n1 , where n1 Z. Combining all, one finds
ZS 1 /s =
R
1
|( )|2
q 2 pL q 2 pR .
(A.14)
0
(pL ,pR )R/N
This expression clearly coincides with ZS 1 R/N . The above formula can be understood in
the Hamiltonian picture as a trace over a Hilbert space Hs a twisted by s a with the insertion
of the bth power of the translation operator s of shifts by 2R/N ,

b p2 /2 p2 /2
R
g
L
R
Z h = Trs a s q
(A.15)
, s exp 2i(pL + pR )
q
.
2N
Appendix B. The point group versus by the full orbifold group

In the construction of an orbifold theory of flat space Rn , the orbifold group G (or
more generally the space group of G) is a discrete subgroup of the Euclidean group of
Rn , consisting of elements g = (Rg , vg ), where Rg O(n) is a rotation and vg Rn a
translation. The maximal subgroup of pure translations is denoted G while the subgroup
of all elements Rg is the point group PG . The normal subgroup PG G/G is isomorphic
to PG , while the coset Rn /G = TG n is a torus. Symmetric orbifolds may be constructed
either coseting Rn by the full G or coseting Tn by PG ,
Rn /G = TG n /PG .
(B.1)
In this appendix, the simplest non-trivial such case when n = 1 and G = Z2 Z will be
shown to yield the same partition function when treated either way.
163
B.1. The functional integral from S 1 /Z2

Here, the field x takes values in the circle SR1 . The point group P = Z2 =
{(1, 0), (1, 0)} is Abelian, and hence each element forms a conjugacy class by itself. The
centralizer of each element is always the full P , so that the cardinality of the centralizer is
always 2, producing a factor 12 . In the twisted sectors, consider all solutions to the twisted
boundary conditions. For example, when the twist is placed on the B-cycle,

x 1 + 1, 2 = +x 1 , 2 mod (2R),

x 1 , 2 + 1 = x 1 , 2 mod (2R)
(B.2)
the general solution is a combination of oscillating solutions, plus a constant,

1 2
1 2
1
.
xm1 ,m2 , = x0 + exp 2i m1 + m2 +
2
(B.3)
The constant may be interpreted as the center of mass of the string, as usual. It must satisfy
x0 = x0 mod (2R), which produces two solutions or fixed points x0 = 0 and x0 = R.
Both fixed points produce equal contributions to the functional integral, whence a factor
of 2. Putting all together, one finds

( )
Dx eS[x] = ZS 1 +
(B.4)
(0, )
.
2 R
i
i=2,3,4
This formula agrees with [6].

B.2. The functional integral from R/(Z2 Z)
Quotienting by the full orbifold group, the conjugacy classes are

N1 = G,
C1 = (1, 0) ,

N2 = (1, n), n Z ,
C2 = (1, m), m > 0 ,

N3 = (1, 0), (1, 0) ,
C3 = (1, 0) ,

N4 = (1, 0), (1, 1) .
C4 = (1, 0) ,
(B.5)
For the classes C3 and C4 , the centralizers have cardinality 2, resulting in an overall factor
of 1/2 in the partition function. On the other hand, the contributions from C3 and C4 are
identical, therefore canceling the factor of 1/2. Their contributions result in the twisted
functional integrals with i for i = 3, 4. For the classes S1 and S2 , the cardinality of the
centralizers are
#(N1 ) = 2#(R ),
#(N2 ) = #(R ).
(B.6)
Recall that R = {2nR, n Z}. The factor of 2 arises because N1 = Z2 R .

The elements in N1 of the type (1, n) produce #(R ) identical copies of the twisting
(1, 0), which yields the contribution with 2 . To derive the precise weight of this
164
contribution is a little tricky. For a given element (1, n), the boundary conditions are

x 1 + 1, 2 = +x 1 , 2 ,

x 1 , 2 + 1 = x 1 , 2 + 2nR.
(B.7)
The general solution of these equations is

1
1
2
xm1 ,m2 1 , 2 = nR + e2i((m1 + 2 ) +m2 ) .
(B.8)
Here, n Z specifies the center of mass of the string at nR. These positions may be
viewed as translates of one another by the lattice 1 R . The original shifts in the boundary
2
conditions were 2nR, which may be viewed as translates in the lattice R . Although
both quantities are infinite, their ratio is well-defined, and given by #( 1 R ) = 2#(R ). As
2
a result, the factor of #( 1 R ) cancels the factor of 1/#(N1 ) and the contribution involving
2
2 is recovered with coefficient 1. The elements in N1 of the type (1, n) combine with all
of N2 to produce half of the untwisted partition function.
Appendix C. -function identities

Throughout, the following correspondence of -function notations is being used,
(see [36]),

1
1
1 (z, ) =
(z, ),
3 (z, ) = [0|0](z, ),
2 2

1
1
4 (z, ) = 0
(z, ),
2 (z, ) =
0 (z, ),
(C.1)
2
2
where the -functions with half-characteristics are defined by

2
ei (n+) +2i(n+)(z+) .
[|](z, )
(C.2)
nZ
The -constants are defined by setting z = 0. The Dedekind -function is defined by

( ) ei/12

1 e2in
(C.3)
n=1
and satisfies 2( )3 = 2 (0, )3 (0, )4 (0, ). The modular transformations of the constants and -function are given by
2 (0, + 1) = ei/4 2 (0, ),

2 (0, 1/ ) = i 4 (0, ),
3 (0, 1/ ) = i 3 (0, ),
3 (0, + 1) = 4 (0, ),
4 (0, 1/ ) = i 2 (0, ),
4 (0, + 1) = 3 (0, ),
(1/ ) = i ( ).
( + 1) = ei/12 ( ),
(C.4)
165
The Jacobi -identity is 2 (0, )4 + 4 (0, )4 = 3 (0, )4 . As a result, the following

combinations of 4 may be expressed in terms of 8th powers of ,
224 34 = 28 + 38 48 ,
234 44 = 38 + 48 28 ,
244 24 = 28 + 48 38 .
(C.5)
The following doubling identities, which relate -constants with modulus 2 to those of
modulus , will be needed. For general characteristics,

1
1
1
2
2i
[0|0] [2|] + e
(0, ).
[|](0, 2 ) =
0
2
+
2
2
2
(C.6)
For some of the characteristics needed here, for example,

1
1
3
0 (0, 2 )2 = 2 3 (0, ),

0 (0, 2 )2 = 2 3 (0, ),
4
2
4
2

1 1
i
3 1
i

(0, 2 )2 = + 2 4 (0, ),

(0, 2 )2 = 2 4 (0, ). (C.7)

4 2
2
4 2
2
C.1. The function [|]( )
The definition of the function [|]( ) may be given either in terms of -functions for
modulus 2 , or directly in terms of momentum summations (with q e2i )
1 (n+)2 2in
[|](0, 2 )
=
q
e
.
[|]( ) e2i
(C.8)
( )
( )
nZ
The following periodicity properties are readily derived

[| ]( ) = [|]( ),
[ + 1|]( ) = e2i [|]( ),
[| + 1]( ) = [|]( ).
(C.9)
Modular transformations act as follows

2
[|]( + 1) = e2i [| + 2]( ),

1
1
1
[|]

2 ( ) , (C.10)
= e2i
2 ( ) + e2i
2
2 2
2
where = exp{i/12}. Actually, it is useful to spell out the modular transformations on
the characteristics that are needed in the orbifold constructions in this paper

1
1
[0|0]( + 1) = [0|0]( ),
[0|0](1/ ) = [0|0] +
0 ( ),
2
2

1
1
0
( + 1) = 0
( ),
0
(1/ ) = 2
0 ( ),
2
2
2
4
166
i/2
0
(
+
1)
=
e
0 ( ),
2
2

1
1
1
0 (1/ ) = [0|0]
0 ( ),
2
2
2

1
1
1
1
1

0 ( + 1) = ei/8
( ),

0 (1/ ) = 0
( ),
4
4 2
4
2
2

1 1
1 1
1 1
1

(1/ ) =
( ). (C.11)

( + 1) = ei/8
0 ( ),
4 2
4
4 2
4 2
Another useful fact is that for a 2Z + 1,

a 1
b
a
b
= ei(a+1)b/2
4 2 2
4 2
(C.12)
both of which are proportional (with a factor) to [ 14 | b2 ]. Other useful identities are

1
q 1/24
1
b
q 1/48

=
( ). (C.13)
= 0
( ),

1
n
n
2
4 2
b
2)
n=1 (1 + q )
n=1 (1 () q
References
[1] M.B. Green, J.H. Schwarz, Supersymmetrical string theories, Phys. Lett. B 109 (1982) 444.
[2] D.J. Gross, J.A. Harvey, E.J. Martinec, R. Rohm, Heterotic string theory, II: the interacting heterotic string,
Nucl. Phys. B 267 (1986) 75.
[3] K.S. Narain, New heterotic string theories in uncompactified dimensions < 10, Phys. Lett. B 169 (1986) 41;
K.S. Narain, M.H. Sarmadi, E. Witten, A Note on toroidal compactification of heterotic string theory, Nucl.
Phys. B 279 (1987) 369.
[4] L. Dixon, J.A. Harvey, C. Vafa, E. Witten, Strings on orbifolds, Nucl. Phys. B 261 (1985) 678;
L. Dixon, J.A. Harvey, C. Vafa, E. Witten, Strings on orbifolds, II, Nucl. Phys. B 274 (1986) 285.
[5] L. Dixon, D. Friedan, E. Martinec, S. Shenker, The conformal field theory of orbifolds, Nucl. Phys. B 282
(1987) 13.
[6] P. Ginsparg, Applied conformal field theory, in: E. Brzin, J. Zinn-Justin (Eds.), Fields Strings and Critical
Phenomena, Session XLIX, Les Houches, 1988, North-Holland, 1990.
[7] Ph. Di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Springer, 1997.
[8] E. Kiritsis, Introduction to string theory, hep-th/9709062;
W. Lerche, A.N. Schellekens, N.P. Warner, Lattices and strings, Phys. Rep. 177 (1989) 1.
[9] C. Vafa, Modular Invariance and discrete torsion on orbifolds, Nucl. Phys. B 273 (1986) 592;
D.S. Freed, C. Vafa, Global anomalies on orbifolds, Commun. Math. Phys. 110 (1987) 349.
[10] R. Dijkgraaf, E. Verlinde, H. Verlinde, c = 1 conformal field theories on Riemann surfaces, Commun. Math.
Phys. 115 (1988) 649690.
[11] R. Dijkgraaf, C. Vafa, E. Verlinde, H. Verlinde, The operator algebra of orbifold models, Commun. Math.
Phys. 123 (1989) 485.
[12] K.S. Narain, M.H. Sarmidi, C. Vafa, Asymmetric orbifolds, Nucl. Phys. B 288 (1987) 551;
K.S. Narain, M.H. Sarmidi, C. Vafa, Asymmetric orbifolds: path integral and operator formulations, Nucl.
Phys. B 356 (1991) 163.
[13] A.H. Chamseddine, J.-P. Derendinger, M. Quiros, Modular invariant twisted strings, Phys. Lett. B 227 (1989)
41;
C.M. Hull, Chiral conformal field theory and asymmetric string compactification, Phys. Lett. B 212 (1988)
437;
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
167
A. Font, L.E. Ibanez, F. Quevedo, ZN ZM orbifolds and discrete torsion, Phys. Lett. B 217 (1989) 272;
I. Senda, A. Sugamoto, Orbifold models and modular transformations, Nucl. Phys. B 302 (1988) 291.
H. Kawai, D.C. Lewellen, S.-H. Tye, Construction of four-dimensional fermionic string models, Phys. Rev.
Lett. 57 (1986) 1832;
H. Kawai, D.C. Lewellen, S.-H. Tye, Construction of fermionic string models in four dimensions, Nucl.
Phys. B 288 (1986) 1;
Z. Kakushadze, G. Shiu, S.-H. Tye, Asymmetric non-Abelian orbifolds and model building, Phys. Rev. D 54
(1996) 7545.
I. Antoniadis, C.P. Bachas, C. Kounnas, Four-dimensional superstrings, Nucl. Phys. B 289 (1987) 87;
I. Antoniadis, C. Bachas, 4d fermionic superstrings with arbitrary twists, Nucl. Phys. B 298 (1988) 586.
J. Polchinski, String Theory, vols. I, II, Cambridge Univ. Press, 1997.
S. Kachru, J. Kumar, E. Silverstein, Vacuum energy cancellation in a nonsupersymmetric string, Phys. Rev.
D 59 (1999) 106004, hep-th/9807076;
S. Kachru, E. Silverstein, Self-dual nonsupersymmetric type II string compactifications, JHEP 9811 (1998)
001, hep-th/9808056.
S. Kachru, E. Silverstein, On vanishing two loop cosmological constants in nonsupersymmetric strings,
JHEP 9901 (1999) 004, hep-th/9810129.
J. Harvey, String duality and non-supersymmetric strings, Phys. Rev. D 59 (1999) 026002, hep-th/9807213.
Z. Kakushadze, S.H.H. Tye, Brane world, Nucl. Phys. B 548 (1999) 180, hep-th/9809147;
G. Shiu, S.H.H. Tye, BoseFermi degeneracy and duality in non-supersymmetric strings, Nucl. Phys. B 542
(1999) 45, hep-th/9808095.
R. Iengo, C.J. Zhu, Evidence for non-vanishing cosmological constant in non-SUSY superstring models,
JHEP 0004 (2000) 028, hep-th/9912074.
K. Aoki, E. DHoker, D.H. Phong, Two-loop superstrings on orbifold compactifications, Nucl. Phys. B 688
(2004) 3, hep-th/0312181.
E. DHoker, D.H. Phong, Superholomorphic anomalies and supermoduli space, Nucl. Phys. B 292 (1987)
317.
E. DHoker, D.H. Phong, The geometry of string perturbation theory, Rev. Mod. Phys. 60 (1988) 9171065.
E. DHoker, D.H. Phong, Conformal scalar fields and chiral splitting on super-Riemann surfaces, Commun.
Math. Phys. 125 (1989) 469513.
E. Verlinde, H. Verlinde, Chiral bosonization, determinants and the string partition function, Nucl. Phys.
B 288 (1987) 357;
E. Verlinde, H. Verlinde, Superstring perturbation theory, in: M. Green, et al. (Eds.), Superstrings 88, World
Scientific, Singapore, 1989, pp. 222240.
E. Verlinde, H. Verlinde, Multiloop calculations in covariant superstring theory, Phys. Lett. B 192 (1987)
95102;
H. Verlinde, A note on the integral over fermionic supermoduli, Utrecht Preprint No. THU-87/26 (1987),
unpublished.
E. DHoker, D.H. Phong, Two-loop superstrings, I: main formulas, Phys. Lett. B 529 (2002) 241255, hepth/0110247.
E. DHoker, D.H. Phong, Two-loop superstrings, II: the chiral measure on moduli space, Nucl. Phys. B 636
(2002) 360, hep-th/0110283.
E. DHoker, D.H. Phong, Two-loop superstrings, III: slice independence and absence of ambiguities, Nucl.
Phys. B 636 (2002) 6179, hep-th/0111016.
E. DHoker, D.H. Phong, Two-loop superstrings, IV: the cosmological constant and modular forms, Nucl.
Phys. B 639 (2002) 129181, hep-th/0111040.
E. DHoker, D.H. Phong, Lectures on two-loop superstrings, hep-th/0211111.
Z.J. Zheng, J.B. Wu, C.J. Zhu, Two-loop superstrings in hyperelliptic language, I: the main results, Phys.
Lett. B 559 (2003) 89, hep-th/0212191;
Z.J. Zheng, J.B. Wu, C.J. Zhu, Two-loop superstrings in hyperelliptic language, II: the vanishing of the
cosmological constant and the non-renormalization theorem, Nucl. Phys. B 663 (2003) 79, hep-th/0212198;
Z.J. Zheng, J.B. Wu, C.J. Zhu, Two-loop superstrings in hyperelliptic language, III: the four-particle
amplitude, Nucl. Phys. B 663 (2003) 95, hep-th/0212219.
168
[34] C.J. Zhu, Two-loop computation in superstring theory, hep-th/0301018;

W.J. Bao, C.J. Zhu, Comments on two-loop four-particle amplitude in superstring theory, JHEP 0305 (2003)
056, hep-th/0303152.
[35] L. Dixon, P. Ginsparg, J.A. Harvey, c = 1 superconformal field theory, Nucl. Phys. B 306 (1988) 470.
[36] A. Erdelyi (Ed.), Higher Transcendental Functions, Bateman Manuscript Project, vol. II, Krieger, 1981.

Constraints on neutrino masses from leptogenesis

models
Thomas Hambye a , Yin Lin b , Alessio Notari b , Michele Papucci b ,
Alessandro Strumia c
a Department of Physics, Theoretical Physics, University of Oxford, Oxford OX1 3NP, UK
b Scuola Normale Superiore, and INFN, Piazza dei Cavalieri 7, I-56126 Pisa, Italy
c Dipartimento di Fisica dellUniversit di Pisa and INFN, Italy
Received 3 February 2004; accepted 11 June 2004
Abstract
Upper bounds on the CP asymmetry relevant for leptogenesis are reexamined and found weaker
than in previous literature, both for hierarchical and for quasi-degenerate right-handed neutrinos.
Successful leptogenesis implies the usual lower bound on right-handed neutrino masses, and an
upper bound on left-handed neutrino masses (which we obtain to be 0.15 eV at 3 ) only if
right-handed neutrinos are assumed to be much more hierarchical than left-handed neutrinos.
Otherwise both bounds can be considerably relaxed. The constraint on light neutrino masses varies
assuming different interpretations of why neutrinos should be quasi-degenerate. With conservative
assumptions, we find that a mild quasi-degeneracy allows neutrinos heavier than an eV compatibly
with leptogenesis.
We also extend computations of thermal leptogenesis to an alternative model of neutrino mass
mediated by fermion triplets which was never considered so far for leptogenesis. Leptogenesis can
be successful despite the effect of gauge interactions, resulting in only slightly stronger constraints
on neutrino masses.
2004 Published by Elsevier B.V.
Assuming that neutrino masses are generated by tree level exchange of right-handed
neutrinos, that the observed baryon asymmetry is produced via thermal leptogenesis
[1] and that right-handed neutrinos are hierarchical, one can derive interesting constraints [24]: right-handed neutrinos must be heavier than about 108 GeV and left-handed
E-mail address: astrumia@difi.unipi.it (A. Strumia).
0550-3213/$ see front matter 2004 Published by Elsevier B.V.
doi:10.1016/j.nuclphysb.2004.06.027
170
T. Hambye et al. / Nuclear Physics B 695 (2004) 169191
neutrinos must be lighter than about 0.1 eV [3,4]. Since the former constraint leads to a
possible conflict between leptogenesis and gravitino overproduction, and since the later
one is stronger than present experimental bounds and is close to the mass scale suggested
by atmospheric oscillations (i.e., m3 ? 0.05 eV), in this article we reconsider these constraints in details. To this end we adopt the results of [4] for a precise computation of the
dynamics of thermal leptogenesis, and we reexamine the upper bound on CP violation in
right-handed neutrino decays.
In Section 1 we consider a hierarchical spectrum of right-handed neutrinos. We show
that the lower bound above on their masses can be significantly evaded dropping the
assumption (made in [2]) that the hierarchy among right-handed neutrinos is much larger
than the observed one among left-handed neutrinos. Moreover, even under this assumption,
we derive a precise upper bound on the CP-asymmetry and find it weaker than in previous
literature [3], leading to a slightly higher neutrino mass bound.
Since this constraint is based on the doubtful assumption that quasi-degenerate neutrinos
be produced by hierarchical right-handed neutrinos, in Section 2 we study what happens
allowing quasi-degenerate right-handed neutrinos. As well known, the asymmetry can be
resonantly enhanced [5,6]; with respect to the analysis of [3] we find other effects that relax
the upper bound on the CP-asymmetry so that our constraint on neutrino masses is much
weaker. We discuss how the result depends on possible reasons that naturally give rise to
quasi-degenerate left-handed neutrinos.
In Section 3 we study alternative mechanisms of leptogenesis and discuss the
corresponding constraints on neutrino masses. Neutrino masses can be produced by
tree-level exchange of three different kinds of particles: (a) right-handed neutrinos [7];
(b) fermion SU(2)L triplets [8,9]; (c) one or more scalar triplets [10]. Special emphasis
is put on the case (b) which has never been considered for leptogenesis. We show that it
is efficient, even if in this case the gauge interactions can keep triplets close to thermal
equilibrium.
Results are summarized in Section 4.
1. Hierarchical right-handed neutrinos

If neutrino masses are produced by the see-saw model described by the following
Lagrangian

ij
MN
ij i j
L = LSM + Ni i/Ni + N L H +
Ni Nj + h.c. ,
2
(1)
the most generic high energy parameters that give rise to the desired neutrino masses
m3 > m2 > m1 0 and mixings V can be written as [11]
MN = diag(M1 , M2 , M3 ),
1 1/2
= MN R diag(m1 , m2 , m3 )1/2 V .
v
(2)
One can always work in the mass eigenstate basis of right-handed neutrinos, and choose
M3 > M2 > M1 0. R is an arbitrary complex orthogonal matrix (i.e., R T R = 1), that can
171
be written in terms of 3 complex mixing angles as

R = diag(1, 1, 1)R (23)(z23 )R (13) (z13 )R (12) (z12 ),
(3)
R (ij )
where
is a rotation in the (ij ) plane with complex angles zij . This parameterization
explicitly shows that the see-saw model has 12 real and 6 complex parameters beyond ones
already present in the SM: 6 + 3 can be measured by low energy experiments (m1 , m2 , m3
and the three complex mixing angles in V ) while the remaining 6 + 3 (M1 , M2 , M3 and
the three complex mixing angles in R: z12 , z23 , z13 ) cannot.
1.1. The CP asymmetry
One important ingredient that determines the baryon asymmetry produced by thermal
leptogenesis is the CP asymmetry i in decays of right-handed neutrinos Ni . Since we will
later study the generic case where right-handed neutrinos can be quasi-degenerate, it is
useful to decompose the CP-asymmetry in N1 decays (and similarly for N2,3 decays) as
the sum of a V ertex contribution and of a Self-energy contribution (with the self energy
contribution as given in [6])
3 M1 j 2Sj + Vj
,
Ij
1 =
(4)
2 Mj Mj
3
j =2,3
where
Ij =
Im[( )21j ]
| |11 | |jj
m
j Mj
j
| |jj
=
,
Mj
8
8v 2
(5)
and where
Sj =
2
Mj2 M1j
,
2
2 )2 + M 2
(M1j
1 j
Vj = 2

Mj2
M12
1
+
log
1
+
1
,
M12
M12
Mj2
Mj2
(6)
are loop factors, with Mij2 = Mj2 Mi2 . In the parameterization of Eq. (2), the light
neutrino mixing matrix V does not affect 1 and CP-violation in leptogenesis arises
from R. While Vj 1, the factor Sj is resonantly enhanced, up to Sj Mj /j , when
Mj M1 j . We normalized the resonance factors Sj and the vertex factors Vj in such
a way that S2,3 = 1 and V2,3 = 1 (so that (2Sj + Vj )/3 = 1) in the hierarchical limit
M2,3 /M1 . In this limit, inserting the parameterization of Eq. (2) into Eq. (5) gives

2
3 M1 Im[( )21j ]
3 M1 i m2i Im R1i

1 =
(7)
=
,
2
16 Mj | |11
16 v 2
i mi |R1i |
j =2,3
which leads to the DavidsonIbarra (DI) bound [2]

3 M1
(8)
(m3 m1 ),
16 v 2
where m3 (m1 ) is the mass of the heaviest (lightest) neutrino. The DI upper bound is plotted
in Fig. 1, together with a random sampling that turns out to find points above it. In fact,
DI
|1 | max
=
172
(a)
(b)
(c)
Fig. 1. The DavidsonIbarra bound on the CP asymmetry 1 was derived and holds for M2,3 /M1 = . The
random samplings of the parameter space (performed assuming different finite
hierarchies among right-handed
neutrinos) confirms that |1 | can be above the DI bound. For all points |ij | < 4 , m2atm = 2 103 eV2 ,
M1 = 108 GeV. As usual, the density of points depends on arbitrary details of the sampling procedure and carries
no information about the likelihood of different regions.
the DI bound is derived and holds for M2,3 /M1 = while in Fig. 1 we assumed a finite
hierarchy. One expects that for M2,3 M1 the DI bound remains approximatively valid,
up to small corrections of relative order (M1 /M2,3 )2 . We now explain why this is not the
case.
In the infinitely hierarchical limit 1 is given by a sum over right-handed neutrinos
weighted by 1/Mj exactly like neutrino masses: the resulting simple expression has special
properties, e.g., in this limit CP violation in leptogenesis disappears when light neutrinos
173
are degenerate. The extra terms suppressed by (M1 /M2,3 )2 do not share this property:
e.g., they do not vanish when neutrinos are degenerate. Moreover, even with hierarchical
neutrinos, these extra terms can be enhanced by m
2,3 /m2,3 . One gets points above the DI
bound when the enhancements overcompensate the (M1 /M2,3 )2 suppression.
This observation can be relevant for leptogenesis, since this enhancement of 1 can
be achieved without introducing significant wash-out factors in the dynamics of thermal
leptogenesis. In fact, L = 2 washout scatterings mediated by off-shell N1,2,3 exchange
are controlled by neutrino masses and do not depend on m
2,3 (and scatterings mediated by
on-shell N2,3 are Boltzmann suppressed). On the contrary L = 2 scatterings mediated by
1 , and the efficiency of leptogenesis is maximal
on-shell N1 exchange are controlled by m
for a relatively small value of m
1 103 eV [3,4]. Therefore we must show that the
enhancement under discussion is possible for small m
1 . We show this analytically and
obtain simple estimates by focusing on the simple limit m1 = m2 = 0. (Our final result
remains valid in the more complicated realistic case with small non-zero m1 and m2 ).
Inserting in the parameterization (2) z23 = 0 + iy23 and z13 = x13 + iy13 with cos 2x13 =
1/ cosh 2y13 we get

3 M1 m3
2
2
F
,
cosh
y
F
sinh
y
3
23
2
23
16 v 2
cosh 2y13 cos 2x13
.
m
1 = m3 | sin z13 |2 = m3
(9)
2
Choosing a small x13 allows to keep m
1 arbitrarily small without affecting 1 . In the fully
hierarchical limit the loop functions Fi = (2Si + Vi )/3 satisfy F2 = F3 = 1 and the last
max . For finite M /M
term in 1 simplifies to 1, giving |1 | = DI
1
2,3 one instead has F2 = F3
and 1 can be enhanced by choosing a large y23 . The maximal value is limited only by
perturbativity of the largest neutrino Yukawa coupling. This violation of the DI bound does
not correspond to a local maximum of 1 and therefore was missedin analyses that
tried to
maximize 1 by imposing d1 /dzij = 0. Imposing 33 = cosh y23 M3 m3 /v 4 gives

M13
DI
|1 | max
(10)
, max .
M3 M22
1 =
This estimate is confirmed by the random sampling in Fig. 1, performed for M1 = 108 GeV
8
which is probably the most interesting choice, as the DI bound implies M
1 ? 10 GeV (see,
1
e.g., [4]). Allowing only neutrino Yukawa couplings smaller than
4 and assuming
M3 /M2 = M2 /M1 = 10n the new configuration under discussion allows to reach |1 |
104n . The DI bound was derived and holds in the limit n . For n = 3 the DI
bound is still an excellent approximation (Fig. 1(a)). For n = 2 the DI bound starts
failing only when neutrinos are quasi-degenerate (Fig. 1(b)). For n = 1 the DI bound
can be significantly evaded (Fig. 1(c)). This is possibly the most realistic case, as solar
and atmospheric oscillations indicate that there is at most a mild hierarchy between lefthanded neutrinos: m3 /m2 6. Allowing right-handed neutrinos to have a similarly mild
1 In supersymmetric models where sleptons acquire SUSY-breaking mass terms at high scale the nonobservation of e implies a somewhat stronger bound, | |e 1011 .
174
hierarchy, M2 /M1 10, Eq. (10) shows that successful leptogenesis with hierarchical Ni
is possible even for N1 much lighter than 108 GeV, as possibly needed in supersymmetric
models in order to avoid overproduction of gravitinos [12]. Thermal leptogenesis can be
successful even if right-handed neutrinos are light and hierarchical. We can make a more
quantitative statement by assuming a 103 efficiency (this is a reasonably conservative
value, see, e.g., [4]): Eq. (10) then requires M3 M22 /M13 104 and consequently allows
some hierarchy among M1,2,3 .
However, a CP-asymmetry above the DI bound is realized for m
2,3 m2,3 and
therefore involves apparently unlikely cancellations, as it needs that N2 and N3 each
gives a large contribution to neutrino masses, but they cancel among each others. This
configuration seems justifiable in a natural way, by, e.g., building models where N2 and
N3 form a quasi-Dirac couple with quasi-chiral Yukawa couplings. See also the appendix
of [13].
In the rest of this section we assume that M3 M22 /M13 is large enough for the DI upper
bound to hold and precisely compute, under all stated assumptions, the maximal value of
neutrino masses compatible with thermal leptogenesis. We find that previous bounds must
be weakened for a different reason, which has no relation with the above discussion.
1.2. Bound on neutrino masses
The DI bound becomes more stringent if neutrinos are quasi-degenerate since in this
case m3 m1 m2atm /2m3 in Eq. (8) decreases. Moreover in this case the efficiency
factor is smaller because larger neutrino masses need larger neutrino Yukawa couplings
and therefore implies larger wash-out scattering rates. In fact the out-of equilibrium
condition H (M1 ) means m
1 103v 2 /MPl 103 eV where

v2

v2
m
1 11
= 81 2 =
mi |R1i |2
M1
M1
i
is always larger than m1 . The minimum value m

1 = m1 implies R12 = R13 = 0 and
therefore a vanishing CP-asymmetry (in the M2,3 /M1 = limit). Conversely, the DI
bound is saturated for large values of m
1 , when is strongly suppressed. As a result the
maximal baryon asymmetry is reached for m
1 larger than m1 m3 but rather close to it [3].
In order to compute the leptogenesis constraint on neutrino masses, one must compute
the maximal value of 1 at given m
1 , and next maximize the baryon asymmetry nB /n
0.011 with respect to remaining free parameters, essentially m
1 and M1 . To determine
the bound on 1 at fixed m
1 we can neglect m2sun
m2atm , so that m1 = m2 and we end
up with a 2 neutrino case: rotations in the (12) plane do not have physical effects.2 We can
2 One can explicitly verify that they do not affect . From its explicit expression in Eq. (7) the matrix R in
1
12
T coming from T . This holds for the numerator of
commutes with diag(m1 , m2 , m3 ) and cancels out with R12
Eq. (7) because we assumed M2,3 /M1 = and will be no longer true when we will relax this assumption. The
denominator has a dependence on Im(z12 ), but only as c + c cosh[Im(z12 )] with c, c positive. The maximization
of 1 naturally leads to Im(z12 ) = 0, hence the validity of what stated.
write R as
175

cos z13 0 sin z13
,
R = R (z13 ) =
0
1
0
sin z13 0 cos z13
because 1 does not depend on z23 . Here z13 = x + iy is a complex mixing angle. The
condition m
1 = m1 |R11 |2 + m3 |R13 |2 fixes x as function of y:
(13)
2m
1 (m1 + m3 ) cosh 2y
,
m3 m1
allowing to write the CP asymmetry as
cos 2x =

3 M1 m23 m21
2
Im R11
2
16 v
m
1

2
3 M1 m3 m21
2m
1 (m1 + m3 ) cosh 2y 2
=
sinh
2y
1
.
16 v 2 2m
1
m3 m1
|1 | =
Maximizing with respect to y gives our bound

|1 | max

max

2

2
DI
=
1 (1 a)m
1/(m3 m1 )
(1 + a)2 (m3 + m1 )/m
1 ,
2
where

1/3

(m21 + m23 + m
21 )3
m1 m3 1/3
a = 2 Re
1
> 0.
1
i
m
21
27m21m23 m
21
In the hierarchical and quasi-degenerate light neutrino limits it simplifies to

1 m1 /m
1
if m1
m3 ,
max
|1 | max DI
1 m21 /m
21 if m1 m3 .
(11)
(12)
Fig. 2. Maximal value of 1 assuming a large hierarchy at fixed m

1 . The dashed line shows the attempt in [3].
The random sampling was performed as an additional check that the analytical bound of Eq. (11) is correct.
176
(a)
(b)
Fig. 3. Leptogenesis constraint on neutrino masses assuming M2,3 M1 . The plot shows the measured baryon
asymmetry (horizontal line) compared with the maximal leptogenesis value as function of the heaviest neutrino
mass m3 , renormalized at low energy. Error bars are at 3 .
The continuous line in Fig. 2 (plotted assuming m1 = 0.12 eV) is our upper bound. The
dashed line shows the result of [3], that first attempted to compute max . Their bound in
max (1 m2 /m
2
the quasi-degenerate limit reduces to DI
1 1 ) and would be stronger than our
bound, but does not hold as confirmed by the numerical scanning.3 In the hierarchical
limit m1
m3 the difference becomes less relevant. In the MSSM the CP-asymmetry, and
consequently its upper limit, is 2 times larger than in the SM. The MSSM result in Fig. 3(b)
is obtained using the Boltzmann equations of [4].
Combining our revised bound on the CP-asymmetry with the revised computation of
the efficiency of leptogenesis of [4]4 we get
m3 < 0.15 eV
(at 3 )
(13)
as illustrated in Fig. 3. This constraint holds under the doubtful assumption that hierarchical
right-handed neutrinos give quasi-degenerate left-handed neutrinos.
2. Quasi-degenerate right-handed neutrinos

Successful thermal leptogenesis implies interesting restrictions on the masses of quasidegenerate neutrinos under the hypothesis of hierarchical right-handed neutrinos. This is
a crucial but doubtful assumption. In fact, one expects that quasi-degenerate neutrinos be
more naturally produced by quasi-degenerate right-handed neutrinos (rather than by an
3 We explain analytically the reason of the disagreement. Ref. [3] made unjustified assumptions on the
elements of R. In their notation variables xi and yi were used (with i = 1, 2, 3), together with the correspondence:
2 m /( ) x + iy . The maximal | | is reached for maximal y . In order to find this maximum they
R1i
i
i
i
11
1
3
assume x2 = y2 = x3 = 0. The first two assumptions are correct in the limit m2sun
m2atm and correspond to
2 = 0) is incorrect, and doing
our z12 = 0. But the extra assumption x3 = 0 (which would corresponds to Re R13
so one does not get the true maximum. In our numerical example with quasi-degenerate neutrinos the maximum
is reached for x3 0.4.
4 In Appendix A we explain why, when computing such constraints, the full network of Boltzmann equations
can be approximated with a single equation for the total B L asymmetry, as tacitly assumed in previous analyses.
177
interplay between heavier N2,3 with bigger Yukawa couplings and lighter N1 with smaller
couplings). In absence of simple predictive models one might naively expect that lefthanded and right-handed neutrinos show similar levels of degeneracy.
In this section we study how much the constraint on neutrino masses gets relaxed
when we make this kind of reasonable assumptions. We think this is an interesting
but qualitative issue. Therefore (unlike in the hierarchical case) we do not attempt to
derive a precise absolute constraint. Our results should however be a qualitatively correct
approximation to it. A fully precise discussion is anyhow prevented by the fact that, in
a generic see-saw model, quasi-degenerate neutrinos are not even stable under radiative
corrections.
We start studying simple particular cases and later show that they catch the new relevant
effects that we need to consider.
2.1. The conservative case
We first study what happens for M3 M2 M1 . Using Eq. (4), in this limit 1 (and
similarly 2 ) can be approximated as
M1 2
1 M1
1
S2 |I2 | <
|I2 | |I2 |,
(14)
M2 M2
2 M2
2
where the inequality is obtained by taking the resonance condition M2 M1 = 2 /2 which
maximizes the resonance factor S2 and gives S2 = M2 /22 . It is useful to compare Eq. (14)
with Eq. (7). The DI upper bound in Eq. (8) can be rewritten schematically as the product
of two suppression factors,
|1 | =
3 M1 j m3 m1
2 Mj Mj
m
j
(15)
present for different physical reasons. The first factor is related to the fact that heavy
particles with small couplings give small effects and it is just the result of a naive estimation
of 1 performed dropping the flavour indices
1
2 2
3 M1 j 1 3 M1 j
.
16 Mj 21
2 Mj Mj
(16)
The second factor comes from a flavour subtlety. In the hierarchical limit, due to the
orthogonality of the R matrix in Eq. (2), the 2j in the numerator of the asymmetry is
j ) as in Eq. (16) but to the m3 m1 mass difference. This
not proportional to j (or to m
j suppression of the asymmetry, which is significant when
results in the extra (m3 m1 )/m
neutrinos are quasi-degenerate. For example for m1 1 eV this suppression is at least of
j 103 .
order [m2atm/(m3 + m1 )]/m
Eq. (14) shows that, when also right-handed neutrinos are quasi-degenerate, none of
these two suppressions are there. The first one can be completely compensated by the
resonance factor S2 which is 1 in the hierarchical case and M2 /22 at the resonance. The
second suppression disappears because 1 is no longer directly related to neutrino masses,
so that it no longer vanishes when neutrinos are degenerate. More technically, I2 (unlike
Eq. (7)) is not suppressed by an orthogonality relation coming from the R matrix.
178
The net result is that the CP asymmetry can be of order unity independently on the
magnitude of the neutrino masses. Whether it is of order unity is controlled by the size of
I2 in Eq. (14). As shown in details in Appendix B, I2 can be easily of order unity except if
m
i are very close to their minimum values (where 1 vanishes).
A formula that correctly estimates the maximal CP asymmetry not only in the quasidegenerate case we are considering but also in the hierarchical limit and in intermediate
cases is obtained by multiplying our bound max of Eq. (11) valid in the hierarchical limit
with an appropriate rescaling factor
|1 | max
S2 m3 m1
.
m3 m1
(17)
This bound reduces to Eq. (11) in the hierarchical limit m3 m1 . In the quasi-degenerate
limit, m3 m1 , it reproduces Eq. (14) up to an order-one factor. This is the maximal
CP-asymmetry possible if a quasi-degeneracy in neutrino masses, m1 m2 m3 m
1
m
2 m
3 arises accidentally, as can happen if no flavour symmetry acts on neutrinos. In
such a case one expects that the three m
i are equal only up to order-one factors. The
resulting bound on m is given by the solid line in Fig. 4(a). A numerical sampling
performed including O(1) factors reveals that (17) can be reached, and in fact the most
conservative bound would be even somewhat weaker. Already for a 10% degeneracy
between N2 and N1 successful leptogenesis can occur for neutrinos heavier than 1 eV.
2.2. A simple special limit
Next we consider the case where the 3 right-handed neutrinos are quasi-degenerate. It is
interesting to consider first the case where two neutrinos are exactly degenerate and quasidegenerate to a third one, i.e., M1 M2 = M3 and assuming 2 = 3 . In such a special
case S2 = S3 and the asymmetry reduces to
|1 | =
M1 2
1 M1
S2 |I2 + I3 | <
(m3 m1 )S2 .
M2 M2
8 v 2
(18)
At the resonance this gives

|1 | =
1 M1
1
1 (m3 m1 )
|I2 + I3 | |I2 + I3 | <
.
2 M2
2
2
m
2
(19)
As a result the asymmetry is smaller than in the previous 2 quasi degenerate right-handed
neutrino case by one power of the orthogonality factor that suppresses I2 + I3 . In other
words 1 is enhanced by the resonance factor but still suppressed when neutrinos are quasidegenerate, by the (m3 m1 )/m
2 = m2atm /m
2 (m3 + m1 ) factor already discussed above.
The constraint on neutrino masses obtained in this special case is shown by the dashed
line in Fig. 4(a) and is significantly stronger than the one obtained in the previous case
(continuous line). This special case, which does not correspond to the most conservative
situation, roughly corresponds to what is obtained in [3]. In this reference, the hierarchical
asymmetry has been enhanced by the resonance factor but was still suppressed by the
orthogonality factor (which is 103 for m1 1 eV).
179
(a)
(b)
Fig. 4. In (a) we estimate how much the maximal value of neutrino mass compatible with thermal leptogenesis
increases when right-handed neutrinos are allowed to be quasi-degenerate. The dashed line includes only the
resonant enhancement of CP-violation, Eq. (19), while the continuous line includes all effects. A numerical
sampling confirms that these constraints can be saturated and even slightly exceeded. (b) Holds in models where
everything is as degenerate as neutrinos, see Eqs. (20). The parameter n quantifies how much m
i are assumed to
be close to neutrino masses mi . As n increases our assumptions get relaxed, and therefore the constraint on m3
becomes weaker.
180
2.3. The most realistic case

Since the leptogenesis constraint on neutrino masses is relevant only for a quasidegenerate spectrum of light neutrinos, one can wonder which is the most natural righthanded neutrino mass spectrum that produces quasi-degenerate neutrinos. Presumably
the answer is: three quasi-degenerate right-handed neutrinos. In fact, other spectra (e.g.,
hierarchical right-handed neutrinos) can give rise to quasi-degenerate neutrinos only in
presence of an appropriate precise correlation between the Yukawa couplings and the righthanded neutrino masses. It is difficult to find a theoretical reason that can justify this kind
of correlation among different objects.
With three quasi-degenerate right-handed neutrinos, no qualitatively new effect appears
with respect to the two neutrino quasi-degenerate case of Eq. (14) we discussed above.
An orthogonality suppression similar to the one of the special case above is generically not
present. As a consequence the constraint on neutrino masses is again well estimated by the
continuous line in Fig. 4(a).
There exists one specific pattern, which is probably the most realistic one, which leads
to more stringent bounds. In fact, if neutrinos were quasi-degenerate, the degeneracy would
presumably not be accidental but due to some reason: a broken SO(3) flavour symmetry
is probably the simplest possibility. One expects that in such framework all quantities, and
not only neutrino masses, are close to the ideal limit where three degenerate right-handed
neutrinos give equal masses (with equal CP phases) to three orthogonal combinations of
j mi mj and
left-handed neutrinos. Therefore one expects something like m
i m
2
1 m2 m1 m2sun
2 m
M2 M1 m
4 eV
0.5
10
,
(20a)
M1
m1
m1
m2
2m21
2
2 m2atm
3m
M3 M2 m3 m2 m
3 eV
10
.
(20b)
M2
m2
m2
m2
2m23
Right-handed neutrinos can be more degenerate than in the above estimates if only the
neutrino Yukawa couplings deviate from the symmetric limit, and can be less degenerate
only if there are accidental cancellations between non-universal Yukawa couplings
and non-degenerate M1,2,3 in the see-saw prediction for neutrino masses. Within the
parameterization of Eq. (2), our reasonable assumption means |yij | m2 /m2 .
To calculate the bound on the neutrino mass considering the realistic example of
Eqs. (20), it is an excellent approximation to take in Eq. (4) (and similar equation for N2,3 )
all Mi equal to the same value M everywhere except in the resonance S factors, and to take
all decay widths equal to the same value everywhere except in the Yukawa coupling
I factors. It is also an excellent approximation to calculate the wash-out effects in the
2 =m
3 m.
Making these approximations (and assuming
symmetric limits with m
1 =m
M1,2,3 1010 GeV, so that interaction rates induced by charged lepton Yukawa couplings
can be neglected) the complicated set of Boltzmann equations for the lepton asymmetries
generated by N1,2,3 decays splits into three independent Boltzmann equations. The net
result is that the efficiency is the same as in one flavour approximation, with the
CP-asymmetry now given by = 1 + 2 + 3 (which can be rewritten in a rephasinginvariant way as a trace of an appropriate matrix function of and MN , see [13]). In
181
this case in the limit M1 = M2 the asymmetry reduces to the one of the special case
above which is orthogonality suppressed by factors (m3 m1 )/m
i . Therefore the only
terms which are not orthogonality suppressed and which are dominant are the terms
involving the M22 M12 splitting. Neglecting higher order terms in the splitting parameters
(M3 M1 )/M1 , (M2 M1 )/M1 and (M2 M1 )/(M3 M1 ) these terms are
2
M2 M1 (M3 M1 )2 2 /4
S2 I2 + 2 S3 I3
.
M
M
M3 M1 (M3 M1 )2 + 2 /4
(21)
The first term is the asymmetry of the 2 quasi degenerate case of Eq. (14) (from both 1
and 2 ) in which N3 has a negligible effect. The second term comes from the difference
between the N1 N3 and N2 N3 mass splittings in the diagram involving these righthanded neutrinos. It is easy to check that although this term can have large effects
for large M,5 it has a completely negligible effect on the bound which is obtained for
smaller M. The asymmetry relevant for the determination of the bound reduces therefore
to the 2 quasi-degenerate case above of Eq. (14) summed on both 1 and 2 . In addition
to the fact that it is not suppressed by any orthogonality relation factor, it turns out to be
j mi
little suppressed by the resonance factor S2 . For values of m1 around eV and with m
the factor S2 is naturally at the resonance or close to it. From Eq. (20a), the resonance
condition 2(M2 M1 ) gives
3
eV
.
M 1011 GeV
(22)
mi
i close to m1 .
The only large suppression can come from the factor I2 for values of the m
From the bound on I2 given in Appendix B, this suppression goes like
|max | = I2max (1 m1 /m)
3/2 ,
(23)
where for simplicity in the last equality we have taken all m

i equal to the same value m.
Using this bound, in Fig. 4(b) we give the baryon asymmetry we obtain as a function of m3
for values of m
= m1 + n(m3 m1 ) with n = {1/2, 1, 2, 4}. Taking m
= m3 (n = 1), as the
generic example for the case that the m
would be precisely of order the neutrino masses,
gives the constraint
m3 < 0.6 eV
(24)
which is stronger than in the conservative case because we are now assuming smaller m,
close to neutrino masses. Taking larger values of m

leads rapidly to larger bounds. For
example taking n = 4 (which starts be fine-tuned) gives m3 < 1 eV. The dependence of the
bound on M1 is quite sensitive to the exact value of the splitting we take for the righthanded neutrinos because this determines the position of the resonance. For Eqs. (20)
the bound is obtained for values close to where the N1 /N2 resonance occurs, i.e., around
M 1011 GeV. Without a predictive flavour model which would show how the correlations
between the see-saw parameters at the origin of the degenerate spectrum occur, these
5 At large M (e.g., above 101314 GeV), unlike for smaller M, the sum of both terms is suppressed by
i factors, and the corresponding neutrino mass bound is therefore suppressed.
orthogonality (m3 m1 )/m
182
bounds obtained from Eqs. (20) are only indicative of what happens and in order to have a
safe bound we must consider the conservative case of Fig. 4(a) (where n was left as a free
parameter in order to maximize the asymmetry, so that m
j can differ from mi by factors
of order one). Even in a very constrained situation the neutrino masses can be as large as
0.6 eV, Eq. (24).6
3. Leptogenesis in alternative minimal models of neutrino masses

Generic neutrino masses can be mediated by tree-level exchange of:
(a) At least three fermion singlets (right-handed neutrinos), described by the see-saw
Lagrangian of Eq. (1).
(b) At least three fermion SU(2)L triplets with zero hypercharges: the Lagrangian keeps
the same structure as in the singlet case, but with different contractions of the SU(2)L
indices that we explicitly show

ij
M
a
L = LSM + N i i/
(25)
DNi + ij
Nia Lj H + N Nia Nja + h.c. .
2
The index a runs over {1, 2, 3}, , over {1, 2} and a are the Pauli matrices.
(c) One scalar (Higgs) triplet T with appropriate hypercharge, such that the most generic
renormalizable Lagrangian is
ij

L = LSM + |D T |2 MT2 |T |2 + T Li Lj T + M H H T + h.c. .
(26)
Neutrino masses can be also mediated by combinations of the above possibilities, among
which it is interesting to consider:
(d) Two or more scalar triplets T with similar interactions.
(e) One scalar triplet and fermion singlets.
Model (c) has the minimal number of beyond-the-SM parameters (8 + 3, while (a) and (b)
have both 12 + 6 extra parameters) but does not lead to a large enough lepton asymmetry.7
Adding to (c) other scalar triplets as in (d) or fermion singlets as in (e) allows successful
leptogenesis8 at the price of introducing more unknown parameters than in (a), (b) and (c).
6 Stronger constraints will arise if supersymmetry exists and if right-handed neutrinos lighter than in Eq. (22)
will be needed to avoid gravitino overproduction [12]. Furthermore, making extra reasonable assumptions about
the flavour structure of Yukawa couplings one gets a smaller CP-asymmetry (suppressed by 2 powers of the quasidegeneracy factor, rather than by 3/2 powers) and consequently a slightly stronger constraint on neutrino masses.
7 One expects that a CP asymmetry in the total triplet decay rate, (T LL) = (T L
L)
arises at two
or more loops. Taking into account how T and the Yukawa couplings of charged leptons break U(3)L U(3)E
flavour rotations and proceeding along the lines of [15] we find that a non zero CP asymmetry needs four powers
of and two powers of , and is therefore too small (unless enhanced by IR effects, which might give only a
mild logarithmic GIM-like suppression).
8 Leptogenesis in case (d) has been studied in [16,17]. We do not consider further this possibility here.
183
Fig. 5. Fermion triplet leptogenesis. Feynman diagrams that give the new interaction rate A .
Theoretically, case (a) is preferred because singlets nicely marry with grand unification
(which is maybe the most promising speculation that we have today). The combination
(e) can also find theoretical support because 3 singlets and a Higgs triplet are naturally
present in renormalizable SO(10) models (as well as the in underlying leftright models)
and can play an important role in leptogenesis [1820]. This possibility does not lead to
relevant constraints on neutrino mass. Cases (b) and (d) seem to be less natural within a
grand unified scheme. E.g., fermion triplets with the needed Yukawa couplings can arise
from adjoints of SU(5). However we believe that it is worth to study case (b) because it
is the only possibility which, with as few parameters as the singlet model, can lead to
successful leptogenesis. This is what we show in the following.
Neutrino singlets trivially allow thermal leptogenesis: not having gauge interactions
they easily satisfy the out-of-equilibrium Sacharov condition for baryogenesis. Fermion
triplets (as scalar triplets) have gauge interactions so that it is more difficult to have
a non-thermal abundancy. We now study thermal leptogenesis in decays of charged
particles, finding that gauge interactions, rather than preventing baryogenesis, make it more
predictive. The point is that gauge interactions involve two particles (see Fig. 5) and are
therefore doubly Boltzmann suppressed at temperatures below their mass (Fig. 6(a) shows
an explicit example), so that they cannot wash-out the lepton asymmetry in an efficient
way.9 On the contrary gauge interactions are efficient at higher temperatures and thermalize
the initial abundancy, so that the final baryon asymmetry almost never depends on it (unlike
what happens in the singlet case).
With fermion triplets neutrino masses are still given by the usual see-saw formula,
m = v 2 T MN1 , without changing any O(1) factor. Using the same notations as in the
singlet case (see Eq. (4) for a more precise discussion) the CP asymmetry is now given by
3 M1 j Vj 2Sj
1 =
(27)
,
Ij
2 Mj Mj
3
j =2,3
and is therefore 3 times smaller in the hierarchical limit. The final amount of baryon
asymmetry is given by the CP-asymmetry times the efficiency factor times a numerical
9 A similar result was found for scalar triplets in [17].
184
(a)
(b)
Fig. 6. Fermion triplet leptogenesis. (a) Interaction rates | /H n | for M1 = 1010 GeV and m
1 = 0.06 eV.
(b) Contour-levels of the efficiency . Successful leptogenesis with (infinitely) hierarchical triplets is possible
inside the green area. (For interpretation of the references to colour in this figure legend, the reader is referred to
the web version of this article.)
185
coefficient which is 3 times bigger than in the singlet case because now N1 has three
components:
nB
= 0.0291.
(28)
n
The N1 decay width is given by the same expression as in the singlet case, so that the
thermally averaged decay rate D becomes 3 times bigger (again because N1 now has 3
components). The on-shell part of L = 2 scattering rates, equal to D /4 [4], becomes
therefore also 3 times bigger, and the off-shell part is affected in a different way. We find

Ns LH LH

( )211
2 + xDs2sub + (2 3x ) Re Ds + 3(x 2)
=
4

2 ln(1 + x)
1 (Re Ds 3 )(1 + x) ,
(29a)
x

Nt LL H H

( )211 3x 2
2
3
=
(29b)
+
+ 3
ln(1 + x) ,
2
2
1+x
2+x
1 . It is defined
where x = s/M12 . A natural value of the parameter is = m3 /m
as follows: the amplitude of N2,3 -mediated scatterings is written as times the value
computed assuming that N2,3 give the same neutrino masses as N1 . In order to deal with
this issue in a more precise way one should know the flavour structure of N1 couplings
and solve Boltzmann equations for the asymmetries in the various flavours. Our simplified
approach is justified by the fact that has a minor impact in most of the reasonable
parameter space. The reduced cross sections are related to the corresponding interaction
rates as summarized in [4], that also explains how to perform a proper subtraction of the
s-channel propagator Ds .
We computed gauge scatterings A (see Fig. 5) summing over the 12 SM fermionic
doublets D = {L1,2,3 , Q1,2,3 } and neglecting scatterings into Higgs doublets (since they
are not enhanced by a large number of final states and since the threshold behavior at
s M12 is the same). At s M12 the NN AA cross section is enhanced by IR effects.
We find
AA)
A (N1 N1 D D,

4
2g 4
6g
2
17
4
4
1+r
r + 2 r 4 +
+ 3 1 + 2 ln
,
= 2 1+
(29c)
x
x x
1r
where r = 1 4/x. Symmetry factors for initial and final state particles are included in
the reduced cross sections. For simplicity we neglected L = 1 scatterings, three body
decays, one loop and thermal corrections.
The Boltzmann equation involve the new gauge term, A :

2

YN1
YN
dYN1
= eq1 1 D
1
A ,
sH z
(30a)
2eq
dz
YN1
YN1
186
sH z

dYBL
YBL D
YN1
sub
= D N1
+
2
,
eq
eq
N
dz
2
YN1
YL
(30b)
where s is the entropy of SM particles, H is the Hubble constant at temperature T .

The efficiency factor (m
1 , M1 , ), computed solving numerically the Boltzmann
equations, is shown in Fig. 6(b), assuming the reasonable value of (other reasonable
values would give minor differences). The result is qualitatively different from the
analogous result in the singlet case, shown in Fig. 8 of [4]:
At m
1
103 eV only gauge interactions drive the N1 abundancy close to thermal
equilibrium (unless M1 ? 1015 GeV). A relic fraction of N1 survives to gauge
annihilations and later decays generating the baryon asymmetry with efficiency
M1 /1013 GeV ( is larger if m
1 103 eV because some N1 decay during the
annihilation stage). Gauge interactions give a stronger suppression at smaller M1 ,
because at low temperatures the expansion of the universe is slower, H T 2 /MPl .
This can be contrasted to what happens in the R case: since it has no gauge
interactions the efficiency only has a minor dependence on M1 .
At m
1 103 eV neutrino Yukawa interactions drive the N1 abundancy close to
thermal equilibrium. As a consequence there are only O(1) differences between singlet
and triplet leptogenesis.
Assuming a sufficiently huge hierarchy, M1 M2,3 we can derive a region where thermal
leptogenesis can be successful using our maximal CP-asymmetry (which is 1/3 of what
obtained in Eq. (11) in the singlet case). The lower bound on the N1 mass and the upper
bound on neutrino masses are at 3
M1 ? 1.5 1010 GeV,
m3 < 0.12 eV.
(31)
case,10
These bounds are slightly stronger than in the right-handed neutrino

and are
subject to all the caveats discussed in that case. In particular quasi-degenerate Ni allow
leptogenesis at the TeV-scale: Fig. 6(b) shows that, despite gauge interactions, the
efficiency remains large enough. This case is testable at collider, where Ni triplets can
be produced and detected (while Ni singlets cannot, because have too low cross sections).
Only O(1) factors are modified if supersymmetry is introduced in the usual minimal
way.
4. Conclusions
Atmospheric oscillations suggest that the heaviest neutrino mass m3 is larger than
about 0.05 eV. Various techniques could reach the necessary sensitivity and presently give
the following 95% C.L. bounds: m3 < 2.2 eV from -decay [21], m3 < 1.0h eV from
neutrino-less double-beta decay [22] (assuming Majorana masses; h 1 renormalizes
10 Within one order of magnitude, our bound on M is in agreement with the estimated bound in [17] for scalar
1
triplets (see also [19]).
187
the uncertain nuclear matrix element), m3 < 0.23 eV from cosmology [23] (assuming
a minimal model). A stronger constraint, m3 0.1 eV [3], is obtained assuming
thermal leptogenesis within see-saw models with hierarchical right-handed neutrinos. We
reanalyzed this leptogenesis constraint, merging the revised computation of dynamics of
leptogenesis of [4] with a revised bound on the CP-asymmetry (see Eq. (11)). It is weaker
than previous bounds, and its validity needs extra assumptions to discard a special (but
non-necessarily fine-tuned) choice of parameters that can give a much larger asymmetry
(see Eq. (10) and Fig. 1). Furthermore in Appendix A we explained why and which single
Boltzmann equation for the total B L asymmetry is a good approximation in the region
where the constraint on neutrino masses is saturated.
From this revised analysis we obtain that if neutrinos turn out to be lighter than
mi < 0.15 eV
(32)
thermal leptogenesis can generically produce the observed baryon abundancy. This critical
value is the present 3 bound (see Fig. 3) and can mildly shift with more accurate
measurements of m2atm , of nB , or if supersymmetry will be discovered.
We studied what happens dropping the dubious assumption that hierarchical righthanded neutrinos give quasi-degenerate neutrino masses. If neutrinos are heavier than
0.15 eV quasi-degenerate R would be suggested by good taste and allow to weaken largely
the leptogenesis constraint. How much depends on why neutrinos are quasi-degenerate. We
consider two possible classes of interpretations:
(a) Neutrinos are not controlled by any flavour symmetry: this naturally gives large mixing
angles and comparable neutrino masses, which might accidentally show some mild
level of quasi-degeneracy. If we therefore assume mi m
j we find that a mild
degeneracy, (M2 M1 )/M1 0.1 is sufficient to push the leptogenesis constraint
above 1 eV (continuous line in Fig. 4(a)). This happens for two different reasons:
the CP asymmetry can be resonantly enhanced and is no longer suppressed by one
power of the orthogonality suppression factor m2atm /m23 , see Eq. (15). A sub-eV
leptogenesis constraint would survive only if one of these two suppressions were
present (dashed line in Fig. 4(a) and previous analyses [3]), but this has no reason
to generically occur.
(b) Some flavour symmetry (e.g., SO(3)) keeps left and right-handed neutrinos quasidegenerate giving m
j very close to neutrino masses mi . We find that this gives a
CP-asymmetry suppressed by 3/2 powers of the quasi degeneracy factor 1 m1 /m
j
m2atm /m23 , see Eq. (23), resulting in a constraint m3 0.6 eV (which can be largely
relaxed if the m
j are slightly larger than the mi , see Fig. 4(b)).
In the last section we studied leptogenesis in alternative minimal models. Neutrino masses
can be mediated by tree-level exchange of right-handed neutrinos, or of fermion SU(2)L
triplets or of scalar triplets. We find that in the last two cases leptogenesis can proceed
enough out-of-equilibrium, despite the new effect of gauge interactions. The reason is that
their rates are strongly Boltzmann suppressed in the last stages of decay processes. While
fermion triplets lead to successful leptogenesis (giving only slightly stronger constraints
188
than with singlets, Eq. (31)) using only a single scalar triplet it seems impossible to achieve
a sufficiently large CP-asymmetry.
Acknowledgements
This work has been partially supported by the EU under TMR contracts HPRN-CT2000-00148, HPRN-CT-2000-00152 and for T.H. by the EU Marie Curie contract HPMFCT-01765. We thank Sacha Davidson for useful suggestions and painful criticisms.
Appendix A. Boltzmann equations with flavour

We here explain how the full network of Boltzmann equations can be approximated with
a single equation for the total B L asymmetry when computing the constraint on quasidegenerate neutrino masses. In the standard approximation one writes one Boltzmann
equation for the total asymmetry, without caring about how it is shared among different
lepton doublet flavours. In simple cases this can be a good approximation [14] if done
properly, as the following example shows.
Let us suppose that N1 decays generate a
lepton asymmetry in1 = ( + )/ 2 and that there are wash-out interactions acting
on 2 = ( )/ 2: one can wonder if they are weighted by (a) |1 |, |2 = 1/2
or by (b) |1 |2 |2 = 0? The answer is (a) when scatterings induced by the Yukawa
coupling are much faster than the expansion of the universe (because they convert 1 into a
incoherent mixture of and ) and (b) when they are much slower (because 1 remains
a coherent superposition of and ). Around the values T M1 1011 GeV for which
the leptogenesis constraint is saturated, all SM lepton Yukawa couplings can be neglected
(case (b) and the Boltzmann equation for leptogenesis is [14]

d
YN
d
{N , }
= zsH
= D 1 eq1
(A.1)
eq ,
dt
dz
+
YN1
8YL
where is the 3 3 matrix density that fully describes how the 3 flavours share the
B L asymmetry. ( ) is the projector over the lepton (anti-lepton) flavour to which N1
decays with decay width ( ). At tree level = = 1 /2 and ij = ij = (1 )ij
1i 1j /| |11 . At one-loop N1 decays into leptons and anti-leptons with different rates
(giving the total CP-asymmetry 1 = ( )/( + )) and into different flavours
( = ). N is the 3 3 flavour matrix of interaction rates of L = 2 scatterings. It can
be decomposed as N = 4D 1 + Nsub , where the first term takes into account resonant
scatterings mediated by on-shell N1 , and Nsub describes off-shell scatterings mediated by
N1,2,3 . For all the other symbols we adopted the notations of [4] (e.g., z = M1 /T , D is
the decay interaction rate, . . .).
In general, without making approximations the matrix equation (A.1) cannot be
replaced by a single equation for the total asymmetry YBL = Tr, nor by three equations
for the diagonal components of (in some flavour basis). In the present case, taking into
account that for quasi-degenerate neutrinos Nsub is a linear combination of 1 and of
189
1 1 ,11 it is non-trivial to verify that at leading order in 1 the solution to (A.1) is

= YBL 1 +
,
21
where YBL satisfies the Boltzmann equation of [4] in one flavour approximation,

YN
YBL D
dYBL
= 1 D 1 eq1 eq
+ 2Nsub ,
zsH
dz
2
YN
YL
1
that is therefore adequate for studying the heaviest neutrino mass compatible with
leptogenesis. This is the equation we used to calculate the constraint on neutrino masses.
Appendix B. Maximal CP asymmetry with 2 quasi-degenerate R

In the following we compute the maximum value of I2 for fixed value of max(m
1, m
2 ),
which allows to give a bound on the asymmetry both in the 2 quasi-degenerate case,
Eq. (14), and in the most realistic case of Eq. (21). Using the parameterization in Eq. (2),
I2 can be written as
I2 =
2

1
Im R diag(m1 , m2 , m3 )R 12 ,
m
1m
2
(B.1)
with

m
1 = R diag(m1 , m2 , m3 )R 11 ,

m
2 = R diag(m1 , m2 , m3 )R 22 .
(B.2)
Neglecting the solar mass splitting, I2 depends on 5 real parameters: the complex angles
z13 and z23 and the imaginary part of the angle z12 (the real part of z12 cancels out
because m1 = m2 ). However, numerical inspection shows that the maximal value of I2
can be obtained for any value of Re z13 and of Re z23 : we can therefore put them to zero,
simplifying the expression and reducing the number of the free parameters to 3. Moreover,
for fixed value of max(m
1, m
2 ) the bound is obtained for m
1 =m
2 m.
For fixed m
1 the
maximum of 1 + 2 is also obtained for m
1 =m
2 . All this together with Eq. (B.2) gives

m21 1/2
m
m1 1/2
m1
max
1 2
1
z
I2 = 2 max z 1 z 1
z
m
m
+ m1
m

m
m1 1/2
m
m1 1/2
1+
z
z
,
(B.3)
1
m
+ m3
m
+ m3
where z = [cosh(2 Im z23 ) 1](m
+ m3 )/2(m
m1 ) can vary in the interval [0, 1]. We
now need to maximize the previous expression with respect to z. This can be done
analytically, but gives a lengthy expression. In the limit m
mi the maximum is reached
11 A less accurate approximation is ( sub ) sub : when left-handed neutrinos are quasi-degenerate sub
N ij
N ij
N
is controlled by the average squared neutrino mass (rather than by their sum, which is 3 times larger).
190
for z = 1/ 2, while in the opposite limit m

m1 it is reached for z = 2/3. Since these two
values are close (and since functions are almost flat around their maximum), an excellent
approximation is obtained by setting z to any of these two numbers. Assuming quasidegenerate neutrinos, m1 m3 , one finds a simple expression that interpolates between
these two numbers, giving

1 + (1 + 8r)3/2
5
(1 m1 /m)
3/2
I2max = 1 +
(B.4)
2r
8r 2
where r = m
2 /m21 and the last simple approximation is accurate to better than 10%. So far
3 =m
1,2 also gives Eq. (B.4)
we did not put any restriction on m
3 . Requiring in addition m
with an accuracy better than 10%. We used therefore Eq. (B.4) for all numerical analyses.
Note that I2max reaches unity asymptotically for large m.
References
[1] M. Fukugita, T. Yanagida, Phys. Lett. B 174 (1986) 45.
[2] S. Davidson, A. Ibarra, hep-ph/0202239;
The DI value of 1 was present also in K. Hamaguchi, H. Murayama, T. Yanagida, hep-ph/0109030. An
earlier work ([14] around Eq. (5.1)) presented the DI bound as valid only up to cancellations, in agreement
with our present findings.
[3] W. Buchmuller, P. Di Bari, M. Plmacher, Nucl. Phys. B 665 (2003) 445, hep-ph/0302092.
[4] G.F. Giudice, A. Notari, M. Raidal, A. Riotto, A. Strumia, hep-ph/0310123.
[5] M. Flanz, E.A. Paschos, U. Sarkar, Phys. Lett. B 345 (1995) 248;
M. Flanz, E.A. Paschos, U. Sarkar, J. Weiss, Phys. Lett. B 389 (1996) 693;
L. Covi, E. Roulet, F. Vissani, Phys. Lett. B 384 (1996) 169;
A. Pilaftsis, Phys. Rev. D 56 (1997) 5431;
A. Pilaftsis, T.E.J. Underwood, hep-ph/0309342;
T. Hambye, Nucl. Phys. B 633 (2002) 171.
[6] A. Pilaftsis, Nucl. Phys. B 504 (1997) 61.
[7] M. Gell-Mann, P. Ramond, R. Slansky, in: P. van Nieuwenhuizen, D. Freedman (Eds.), Supergravity, NorthHolland, 1979, p. 315;
S.L. Glashow, in: M. Lvy, et al. (Eds.), Quarks and Leptons, Cargse, Plenum, New-York, 1980, p. 707;
T. Yanagida, in: O. Sawada, A. Sugamoto Proceedings of the Workshop on the Unified Theory and the
Baryon Number in the Universe, KEK Report No. 79-18, Tsukuba, 1979, p. 95;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
[8] R. Foot, H. Lew, X.-G. He, G.C. Joshi, Z. Phys. C 44 (1989) 441.
[9] E. Ma, Phys. Rev. Lett. 81 (1998) 1171;
E. Ma, D.P. Roy, Nucl. Phys. B 644 (2002) 290.
[10] G. Lazarides, Q. Shafi, C. Wetterich, Nucl. Phys. B 181 (1981) 287;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. D 23 (1981) 165;
C. Wetterich, Nucl. Phys. B 187 (1981) 343.
[11] J.A. Casas, A. Ibarra, hep-ph/0103065.
[12] J.R. Ellis, J. Kim, D.V. Nanopoulos, Phys. Lett. B 145 (1984) 181;
L.M. Krauss, Nucl. Phys. B 227 (1983) 556;
M.Yu. Khlopov, A.D. Linde, Phys. Lett. B 138 (1984) 265;
J.R. Ellis, D.V. Nanopoulos, K.A. Olive, S.-J. Rey, Astropart. Phys. 4 (1996) 371;
M. Bolz, A. Brandenburg, W. Buchmuller, Nucl. Phys. B 606 (2001) 518;
R.H. Cyburt, J. Ellis, B.D. Fields, K.A. Olive, Phys. Rev. D 67 (2003) 103521;
For a review see M.Yu. Khlopov, Cosmoparticle Physics, World Scientific, 1999.
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
191
S. Davidson, R. Kitano, hep-ph/0312007.

R. Barbieri, P. Creminelli, A. Strumia, N. Tetradis, Nucl. Phys. B 575 (2000) 61, hep-ph/9911315.
A. Romanino, A. Strumia, Nucl. Phys. B 490 (1997) 204, hep-ph/9610485.
E. Ma, U. Sarkar, Phys. Rev. Lett. 80 (1998) 5716.
T. Hambye, E. Ma, U. Sarkar, Nucl. Phys. B 602 (2001) 23.
P. ODonnell, U. Sarkar, Phys. Rev. D 49 (1994) 2118.
T. Hambye, G. Senjanovic, hep-ph/0307237, Phys. Lett. B, in press.
See also G. Lazarides, Q. Shafi, Phys. Rev. D 58 (1998) 071702;
For a related model based on inflation see also T. Dent, G. Lazarides, R. Ruiz de Austri, hep-ph/0312033.
[21] Mainz Collaboration, Nucl. Phys. B (Proc. Suppl.) 91 (2001) 273;
Troitsk Collaboration, Nucl. Phys. B (Proc. Suppl.) 91 (2001) 280.
[22] The 02 bound on neutrino masses is obtained by combining 02 data, HeidelbergMoscow Collaboration, H.V. Klapdor-Kleingrothaus, et al., Eur. Phys. J. A 12 (2001) 147, hep-ph/0103062;
With oscillation data: see F. Feruglio, A. Strumia, F. Vissani, Nucl. Phys. B 637 (2002) 345, hep-ph/0201291,
and references therein.
[23] The cosmological bound on neutrino masses is obtained by combining WMAP data, C.L. Bennett, et al.,
astro-ph/0302207;
With large-scale structure data: see D.N. Spergel, et al., astro-ph/0302209, and references therein;
This is similar to pre-WMAP analyses, A. Lewis, S. Bridle, Phys. Rev. D 66 (2002) 103511;
Weaker more conservative bounds are obtained in S. Hannestad, JCAP 0305 (2003) 004, astro-ph/0303076.

In search of the chiral regime

Silas R. Beane a,b
a Department of Physics, University of New Hampshire, Durham, NH 03824-3568, USA
b Jefferson Laboratory, 12000 Jefferson Avenue, Newport News, VA 23606, USA
Abstract
A critical appraisal is given of a recent analysis of the quark-mass and finite-size dependence of
unquenched lattice QCD data for the nucleon mass. We use this forum to estimate the boundary of
the chiral regime for nucleon properties.
PACS: 11.15.Ha
How low in quark masses must lattice QCD simulations of baryon properties go in
order to reach the regime where chiral perturbation theory ( PT) is an extrapolation tool
with controlled errors? In interesting recent work [1,2] unquenched lattice QCD data for
the nucleon mass at large pion masses, m > 500 MeV, has been analyzed using baryon
PT to O(m2q ) (without including the as an explicit degree of freedom). These papers
fit the quark-mass dependence of unquenched lattice QCD data for the nucleon mass
(including the physical point) with natural values of the strong-interaction parameters
and then successfully predict (in Ref. [2]) the finite-size dependence. These results are
quite remarkable. However, little attempt is made in Refs. [1,2] to gauge the uncertainties
associated with the quark-mass extrapolation procedure. Given the provocative results that
are found and the rather large quark masses that are simulated, it is essential to perform an
error analysis. We will therefore do so in this note. And, in so doing, we will provide an
estimate of the quark-mass boundary of the chiral regime for nucleon properties [3].
E-mail address: silas@physics.unh.edu (S.R. Beane).

doi:10.1016/j.nuclphysb.2004.06.002
S.R. Beane / Nuclear Physics B 695 (2004) 192198
193
Table 1
Fit parameters. The scale-dependent parameter e1 is evaluated at = 1 GeV
Fit 1 (IR)
Fit 2 (IR)
Fit 1 (DR)
Fit 2 (DR)
f (MeV)
gA
M0 (GeV)
c1 (GeV1 )
e1 (GeV3 )
131
124
131
124
1.267
1.520
1.267
1.520
0.880
0.872
0.885
0.879
0.95
1.22
0.92
1.07
2.44
4.90
3.61
6.79
The nucleon mass at O(m2q ) in the chiral expansion, evaluated using infrared
regularization [4], is given by [1]

2
3gA
m2
m2
2
3
MN = M0 4c1 m
+ arctan
m 1
8 2 f 2
4M02 2
4M 2 m2 m4

+ e1 () +
2
gA
3
c2
+
2
2
32 f
M0
2

2

gA
m
3
m4 .
8c1 + c2 + 4c3 ln
16 2 f 2 M0
(1)
Following Refs. [1,2] we choose the parameter set gA = 1.267, f = 131 MeV, c2 =
3.2 GeV1 and c3 = 3.4 GeV1 . One then easily fits the remaining low-energy constants
to unquenched lattice QCD data [2,57] over a wide range of pion masses [1,2]. For
instance, in Table 1 (fit 1 (IR)) we list a set of parameter values1 which give rise to the
solid curve in Fig. 1 (left panel). While one sees an excellent description of the data with
natural values of the strong-interaction parameters (which are consistent with N phase
shift analyses), one must quantify the errors associated with omitted higher-order effects
in order to estimate the reliability of perturbation theory over the range of fit pion masses.
There are many ways to estimate the reliability of the chiral expansion and Refs. [1,2]
suggest one robust method. The difference between a matrix element at a given order in
the chiral expansion evaluated with the physical values of gA and f and evaluated with
(0)
their chiral-limit values gA and f (0) is a measure of the importance of higher-order
(0)
effects. Ref. [2] estimate gA
1.2 and f (0) 124 MeV and find little variation in the
fit curve. However, this is no surprise as these chiral-limit values leave the ratio gA /f
unchanged. Here we will consider how variation of the chiral-limit parameters affects the
chiral expansion.
(0)
It is worth considering what is known experimentally about gA and f (0) . At one-loop
order in the chiral expansion one has the well-known formulas [8,9],
(0)
gA = gA

(0)
(g )2 m2
4m2
1 A2 (0) 2 + (0) d16 ,
8 (f )
gA

f = f (0) 1 +

m2
l
4 ,
8 2 (f (0) )2
(2)
1 We fit the full expression in Eq. (1), while Refs. [1,2] fit to an expanded form; hence the small differences in
the values of the parameters.
194
Fig. 1. Left panel: the nucleon mass in baryon PT (computed using infrared regularization) at O(m2q )
(parameters given in Table 1 (fit 1 (IR))) vs. the pion mass squared. Symbols are unquenched lattice QCD data
taken from Refs. [2,57]. The solid curve is the fit curve and the dashed curves give maximal variations due to
the spread in chiral limit values of f and gA , as explained in the text. The gray region corresponds to the error
associated with omitted higher orders. Right panel: same but with a second fit, (parameters given in Table 1 (fit
2 (IR))) with corresponding spread in the chiral limit values of f and gA . (Note that there are two curves going
through the data.)
where d16 and l 4 are (scale-independent) low-energy constants.2

The parameter l 4 is determined from K3 and K2 decays [9] and from a two-loop
analysis of scattering [12]. For simplicity, we will follow Ref. [9] and fix f (0) =
124 MeV, keeping in mind that not accounting for variation of f (0) will necessarily lead
to an underestimate of the final error. An analysis [8,13] of the process N N
at one-loop order in the chiral expansion provides three distinct determinations of d16 :
0.91 0.74, 1.01 0.72 and 1.76 0.85 GeV2 . Taking the 1 limits of the
(0)
three determinations gives 2.61 GeV2 < d16 < 0.17, or gA
= 1.42 0.10. A recent
analysis suggests that the (which is not included as an explicit degree of freedom in
Refs. [8,13]) plays a crucial role in the determination of d16 and suggests a lower central
(0)
value: gA
= 1.20 0.10 [14]. Taking 2.61 GeV2 < d16 < 2.43 GeV2 encompasses
(0)
both analyses, i.e., gA
= 1.101.52. As naturalness suggests |d16| 1, this range
of values does not introduce anomalously large low-energy constants into the chiral
expansion. This experimental/theoretical uncertainty therefore provides an estimate of the
importance of neglected higher orders in the chiral expansion.3 In Fig. 1 (left panel), we
illustrate (dashed curves) the spread of the fit curve when one replaces f and gA by their
chiral limit values, including the range of d16 values.
It is important to realize that all curves encompassed by the gray region of Fig. 1 (left
5/2
panel), between the dashed curves, differ only by terms that are higher order, O(mq ), in
the chiral expansion. The spread of curves in Fig. 1 (left panel) therefore suggests a 50%
2 The barred constants contain chiral logarithms and therefore the formulas in Eq. (2) are useful only at the
physical value of the pion mass. Extrapolation formulas are available in Refs. [10,11].
3 We emphasize that even if d (and therefore g (0) ) were known with high precision, varying d over a
16
16
A
range of natural values would continue to be a legitimate way of estimating errors due to omitted higher-order
effects.
195
Fig. 2. Left panel: the nucleon mass in baryon PT (computed using dim reg with MS) at O(m2q ) (parameters
given in Table 1 (fit 1 (DR))) vs. the pion mass squared. Symbols are unquenched lattice QCD data taken from
Refs. [2,57]. The solid curve is the fit curve and the dashed curves give maximal variations due to the spread
in chiral limit values of f and gA , as explained in the text. The gray region corresponds to the error associated
with omitted higher orders. Right panel: same but with a second fit (parameters given in Table 1 (fit 2 (DR))) with
corresponding spread in the chiral limit values of f and gA .
error associated with neglected higher orders in the chiral expansion at the lattice point
with the lowest pion mass. We also perform a second fit with select chiral-limit values of f
and gA , see Table 1 (fit 2 (IR)), and again considering variation in the chiral limit values.4
The first and second fit are shown in Fig. 1 (right panel), together with the spread (dashed
and dotted curves) of the fit curves when one replaces f and gA by their chiral limit values,
including the range of d16 values. The spread of curves in Fig. 1 (right panel), suggests an
80% error at the lattice point with the lowest pion mass.
It is interesting to compare the error analysis performed using infrared regularization
with the same analysis performed using dimensional regularization (dim reg) with MS; as
physics must be independent of the regulator, so must the results of the error analysis.5 The
nucleon mass in dim reg with MS at O(m2q ) in the chiral expansion is given by [16]
2
3gA
m3
16f 2

2
gA
3
c2
+ e1 () +
+
32 2 f 2
M0
2
2

gA
m
3
8c
+
c
+
4c
ln
m4 .
1
2
3
16 2 f 2 M0
MN = M0 4c1 m2
(3)
Again one easily finds a fit to unquenched lattice QCD data over a wide range of pion
masses. In Table 1 (fit 1 (DR)) we list a set of parameter values which give rise to the solid
4 Notice that although the parameter e (1 GeV) appears rather large, we have defined e as in Refs. [1,2]
1
1
which has absorbed a factor of 4 into its definition as compared to the parameter to which naturalness arguments
should be applied [15]. Hence, all values of e1 in Table 1 are technically quite natural.
5 Refs. [1,2] use the nomenclature relativistic PT to describe infrared regularization; we dislike this
appellation as it may be misinterpreted to suggest that there is physics in the regulator.
196
Fig. 3. Errors in the nucleon mass extrapolation curves vs. the pion mass squared. The solid lines correspond to
(0)
the errors abstracted from the left panels of Figs. 1 and 2; they each arise from the effect of gA and f (0) on
a single fit (Fit 1), performed with the physical values of gA and f . The dashed lines correspond to the errors
(0)
abstracted from the right panels of Figs. 1 and 2; same as above but supplemented with the effect of gA and
(0)
on a second fit (Fit 2), performed with select chiral-limit values of gA and f . The horizontal dotted lines
f
indicate 10% and 20% errors. The star indicates the physical pion mass.
curve in Fig. 2 (left panel). Again we consider variations in the chiral limit values of f and
gA (gray region) and perform another fit (see Table 1 (fit 2 (DR)) and Fig. 2 (right panel)).
We find very little difference in the error analyses for the two regularization schemes.
Our results are summarized in Fig. 3, which plots the errors in the nucleon mass
extrapolation curves taken from Figs. 1 and 2; the curves take into account the error
associated with the gray regions in Figs. 1 and 2. Clearly the chiral expansion is nicely
convergent in the vicinity of the physical pion mass. In our view the dashed curves
(right panels of Figs. 1 and 2) give a conservative estimate of the errors. If one is
willing to tolerate a 20% error associated with neglected higher-order terms for an O(m2q )
calculation, then the error analysis would indicate that one is in the chiral regime for
m 300 MeV. If one instead takes the errors given by the solid curves (left panels
of Figs. 1 and 2), willingness to tolerate a 20% error would indicate that one is in the
chiral regime for m 400 MeV. It may appear odd that our estimate of the boundary
of the chiral regime is less than the kaon mass, which governs the convergence of SU(3)
baryon PT. We stress that the convergence of the chiral expansion is process and flavor
dependent; for instance, some observables in SU(3) baryon PT converge well and others
do not.
The method used here to estimate errors is, of course, only one method among many to
quantify the reliability of perturbation theory. For instance, by comparing various orders in
the chiral expansion, Ref. [17] finds that one is in the chiral regime for m 500 MeV. An
interesting finding of this work is that the O(m2q ) correction remains a small perturbation
3/2
on the O(mq ) result for m < 600 MeV. It is important to emphasize that error analyses
197
such as that presented in this note and in Ref. [17] are merely indicative. We believe that
we have given a conservative estimate of the errors, as they presently stand. Naively, one
way of weakening the strength of higher-dimensional operators, and thereby reducing the
error, is to include the as an explicit degree of freedom in PT.6 It may also be possible
to reduce the extrapolation error at larger pion masses by computing MN , gA and f in the
same lattice simulation in order to fix the lattice values of d16 7 and l4 . It is clear from the
results presented here that a definitive determination of the boundary of the chiral regime
will not be possible until these strong-interaction parameters are well determined. Finally,
it is clear that by imposing various prejudices on our model-independent analysis, one may
shrink the gray regions of Figs. 1 and 2 to any desired size.
In this note we have provided evidence that the quark-mass dependenceand by association, the finite-size and lattice-spacing dependenceof currently-extant unquenched
lattice QCD data for the nucleon mass are not presently described by a perturbative effective field theory with a controlled error estimate. Hence these lattice data cannot be reliably
extrapolated to predict nucleon properties. In our view, a symbiotic relationship between
lattice QCD and baryon chiral perturbation theory which will lead to first-principles predictions of baryon properties with meaningful errors must await smaller lattice quark masses.
Acknowledgements
I would like to thank Maarten Golterman, Thomas Hemmert, Ulf Meiner and
Martin Savage for helpful comments/discussions. This work was partly supported by
DOE contract DE-AC05-84ER40150, under which the Southeastern Universities Research
Association (SURA) operates the Thomas Jefferson National Accelerator Facility.
References
[1] M. Procura, T.R. Hemmert, W. Weise, Phys. Rev. D 69 (2004) 034505.
[2] A. Ali Khan, et al., QCDSF-UKQCD Collaboration, hep-lat/0312030.
[3] For a recent discussion, see C. Bernard, S. Hashimoto, D.B. Leinweber, P. Lepage, E. Pallante, S.R. Sharpe,
H. Wittig, Nucl. Phys. B (Proc. Suppl.) 119 (2003) 170.
[4] T. Becher, H. Leutwyler, Eur. Phys. J. C 9 (1999) 643.
[5] C.R. Allton, et al., UKQCD Collaboration, Phys. Rev. D 65 (2002) 054502.
[6] A. Ali Khan, et al., CP-PACS Collaboration, Phys. Rev. D 65 (2002) 054505;
A. Ali Khan, et al., CP-PACS Collaboration, Phys. Rev. D 67 (2003) 059901, Erratum.
[7] S. Aoki, et al., JLQCD Collaboration, Phys. Rev. D 68 (2003) 054502.
[8] N. Fettes, V. Bernard, U.-G. Meiner, Nucl. Phys. A 669 (2000) 269.
[9] J. Gasser, H. Leutwyler, Ann. Phys. 158 (1984) 142.
[10] S.R. Beane, M.J. Savage, Nucl. Phys. A 717 (2003) 91.
[11] E. Epelbaum, U.-G. Meiner, W. Glockle, Nucl. Phys. A 714 (2003) 535.
6 As very-few quantities have been computed including the at non-trivial orders in baryon PT, it is
presently unclear whether or not this is the case.
7 This parameter also provides a major source of uncertainty in the quark-mass dependence of the deuteron
binding energy [10,11].
198
[12]
[13]
[14]
[15]
[16]
[17]
G. Colangelo, J. Gasser, H. Leutwyler, Nucl. Phys. B 603 (2001) 125.

N. Fettes, JUL-3814, PhD Thesis.
T.R. Hemmert, M. Procura, W. Weise, Phys. Rev. D 68 (2003) 075009.
N. Fettes, U.-G. Meiner, M. Mojzis, S. Steininger, Ann. Phys. 283 (2000) 273.
S. Steininger, U.-G. Meiner, N. Fettes, JHEP 9809 (1998) 008.
V. Bernard, T.R. Hemmert, U.-G. Meiner, Nucl. Phys. A 732 (2004) 149.

Relation between bottom-quark MS Yukawa

coupling and pole mass
Bernd A. Kniehl, Jan H. Piclum, Matthias Steinhauser
II. Institut fr Theoretische Physik, Universitt Hamburg, Luruper Chaussee 149, 22761 Hamburg, Germany
Abstract
We calculate the O(s ) corrections to the relationships between the MS Yukawa couplings
and the pole masses of the first five quark flavours in the standard model. We also present the
corresponding relationships between the MS and pole masses, which emerge as by-products of our
main analysis. The occurring self-energies are evaluated using the method of asymptotic expansion.
1. Introduction
One of the most intriguing puzzles in contemporary elementary particle physics is
related to the deciphering of the seemingly arbitrary pattern of the fermion masses
and the elements of the CabibboKobayashiMaskawa (CKM) quark mixing matrix of
the standard model (SM) in the framework of some more fundamental theory. It is
generally believed that grand unified theories (GUTs), possibly realized in the context
of supersymmetry, are able to provide a key to the understanding of this fundamental
problem [1]. The crucial idea is that a judiciously chosen set of independent parameters,
appropriately evolved to the GUT scale, obey simple relationships. Essential ingredients
of this formalism include renormalization group equations, which determine the scale
dependence of the running parameters, and threshold relations, which relate the running
parameters at the scales of the elementary-particle physics experiments to the physical
parameters, e.g., pole masses and CKM matrix elements, extracted from the latter. It has
E-mail address: bernd.kniehl@desy.de (B.A. Kniehl).
doi:10.1016/j.nuclphysb.2004.06.036
200
B.A. Kniehl et al. / Nuclear Physics B 695 (2004) 199216
become customary to define the running parameters in the modified minimal-subtraction

MS scheme.
The CallanSymanzik beta functions of the SM and its most popular extensions are
well established at one loop and beyond [2]. As for the threshold relations of the Yukawa
couplings and CKM matrix elements the current status is as follows. The relationships
between the MS and on-shell definitions of the CKM matrix elements are known at one
loop [3]. The relationships between the MS and pole definitions of the quark masses have
been elaborated at one [4], two [5], and three [6] loops in quantum chromodynamics
(QCD). These pure QCD corrections readily carry over to the relationships between the
MS Yukawa couplings and the pole masses of the quarks, which are actually relevant for
GUT analyses. However, the situation becomes more involved if electroweak quantum
corrections are taken into account. Then, the relationships between the MS definitions of
the fermion masses and the Yukawa couplings become nontrivial [7,8]. The full one-loop
corrections to the relationships between the MS Yukawa couplings and pole masses of the
SM fermions were derived in Ref. [7]. They were found to be gauge-parameter independent
and devoid of tadpole contributions. However, in that reference, also a gauge-parameterindependent MS definition of fermion mass was presented, and tadpole contributions were
found to be essential ingredients for that.
In this paper, we take the next step and evaluate the mixed two-loop corrections,
involving one power of Sommerfelds fine-structure constant and one power of the
strong-coupling constant s , to the relationships between the MS Yukawa couplings and
the pole masses of the first five quark flavours in the SM. For simplicity, the calculation
is performed in the limit where the third quark generation does not mix with the first two,
i.e., where Vub = Vcb = Vt d = Vt s = 0, which is approximately realized in nature [9]. As
in Ref. [7], we pay special attention to the tadpole contributions. Again, they are found to
cancel in the relationships between the MS Yukawa couplings and the pole masses, while
they are indispensable to render the MS definitions of quark mass independent of the gauge
parameters.
As for the top quark, the O(s ) correction to the relationship between the MS and
pole masses was recently evaluated in Ref. [10] retaining the full mass dependence. Also
there, the tadpole contribution was found to be necessary in order to get rid of the gaugeparameter dependence. An alternative definition of MS top-quark mass, without inclusion
of the tadpole contribution, was proposed in Ref. [11]. Since Ref. [11] is dealing with
leading heavy-top-quark effects, the gauge-parameter dependence does not yet show up.
However, we caution the reader that it will once subleading corrections are to be included
in this formalism. Then, the formulas that express physical observables in terms of the sodefined MS top-quark mass will explicitly depend on gauge parameters as will the value
of this mass. Although this is not forbidden by first principles, it is cumbersome in practice
because, whenever a value of this mass is to be quoted, the underlying choice of gauge
needs to be specified as well [7].
This paper is organized as follows. In Section 2, we set up the theoretical framework and
derive a master formula for the relationship between the MS Yukawa coupling and the pole
mass of a quark through O(s ). In Section 3, we briefly recall the method of asymptotic
expansion and apply it to recover the O() results for the first five quark flavours. We then
illustrate the goodness of the resulting expansion for the case of bottom by comparison with
201
the analytic O() result of Ref. [7]. The O(s ) results for the first five quark flavours are
presented in Section 4. As a by-product of our analysis, we also find the relationships
between the MS and pole masses of the first five quark flavours to this order. They can be
found in Section 5. Finally, we conclude with a summary in Section 6.
2. Deriving the relations

The relationship between the MS Yukawa coupling h f () at renormalization scale
and the pole mass Mf of a fermion f can generically be written as

1/2
h f () = 23/4 GF Mf 1 + f () ,
(1)
where GF is Fermis constant. Here, we compute the corrections f () for the first five
quark flavours to O(), O(s ), and O(s ).
In the SM, the fermion masses are generated from the Yukawa interactions through
the Higgs mechanism of spontaneous electroweak symmetry breaking. In terms of bare
parameters, which are henceforth labelled by the subscript zero, this is manifested through
the identity
v0
mf,0 = hf,0 ,
(2)
2
where v is the vacuum expectation value of the Higgs field. Next, we relate the bare
quantities in Eq. (2) with their renormalized counterparts in Eq. (1).
The pole mass is defined as the zero of the inverse propagator. The inverse fermion
propagator can be written as
iSf (q)1 = q/ mf,0 + f (q),
(3)
where
2
2
(l) m2f,0
mf,0
mf,0
(l)
(l)
q/ f,V
+ q/ 5 f,A
+ mf,0 f,S
f (q) =
2
2
q
q
q2
(4)
is the self-energy of fermion f . In Eq. (4), the subscripts V , A, and S denote the vector,
axial-vector, and scalar components, respectively, and the sum runs over all contributions
through the desired order. Here, we consider the one-loop contributions of O() and
O(s ) and the two-loop ones of O(s ), which we denote by the superscripts (1) and
(2), respectively. Some sample Feynman diagrams are depicted in Fig. 1(a)(c). Inserting
Eq. (4) into Eq. (3) and setting the latter to zero leads to [5]
(1)

(1) 2f(1) (1) ,
Mf = mf,0 1 f(1) (1) f(2) (1) + f(1) (1) f,V
(5)
which is valid through the two-loop order. Here, the prime indicates differentiation with
(l)
(l)
(l)
respect to m2f,0 /q 2 , and the notation f = f,V + f,S has been introduced. Note that,
while deriving Eq. (5), an expansion of mf,0 about Mf has been performed in order to
(l)
obtain unity as the argument in the self-energies. Furthermore, we remark that f,A does
not appear in Eq. (5). For later convenience, we also provide the inverted relation, which
202
Fig. 1. Sample diagrams which contribute to the bottom-quark self-energy.
reads

(1)

mf,0 = Mf 1 + f(1) (1) + f(2) (1) + f(1) (1) f,S
(1) + 2f(1) (1) .
(6)
By construction, the mass of fermion f which appears in the self-energies on the r.h.s.
of Eqs. (5) and (6) is the pole mass. These equations become manifestly gauge-parameter
independent if one includes the tadpole contributions Tf in the self-energies [7], i.e., if
(l)
(l)
(l)
the tadpole-free self-energies f in Eqs. (5) and (6) are replaced by f + Tf . Some
sample tadpole diagrams are shown in Fig. 1(d)(f).
The Fermi constant GF is not a basic parameter of the SM Lagrangian density. This
means that it has to be expressed in terms of such parameters. To do this, one calculates a
process in the SM and in the Fermi theory, and equates the two results. This was first done
in a classical paper by Sirlin [12], who considered the muon lifetime. Adapting his result
for corrections of O(s ), we have

(1)
(2)
(1)
(2)
W W (0) + W W (0) TW W + TW W
1
GF = 2 1 +
(7)
+
+E .
2
2
MW
MW
2v0
Here, W W (q 2 ) is the tadpole-free W -boson self-energy at four-momentum q, and TW W
denotes the corresponding tadpole contribution. The quantity E contains those wavefunction renormalization, vertex and box corrections to the muon decay width which the
SM introduces on top of the Fermi model improved by quantum electrodynamics (QED).
At O() and in t HooftFeynman gauge, it may be written as

2
7
4
2
+
4
ln
+
6
ln
c
+
6
,
E=
(8)
w
4sw2
2sw2
MZ2
2 = 1 s 2 = M 2 /M 2 . Here and in the following,
where we have used the abbreviation cw
w
W
Z
we use dimensional regularization, with d = 4 2 spacetime dimensions and t Hooft
mass scale , adopt a MS-like notation where the typical combination E ln(4) is
suppressed, and do not display terms of O(). There are no corrections of O(s ) to this
quantity. At the one-loop level, the W -boson self-energy can be split into a bosonic and a
203
fermionic part as
2
bos
fer
W W q 2 = W
W q + W W q ,
(9)
and, in the t HooftFeynman gauge, we have (see, e.g., Ref. [7])

2
MW
2
17
1
1
1
bos
2
+
ln
W
(0)
=
2
+
+
2
+
ln cw
W
2
2
2
4sw2
cw

cw
4sw2
MW
2
2
MW
M2
MH
7
17
3
+ 2 H2
,
ln
2 M2
2
4
4 MW
8cw 8MW
MH
H

1
2
Nc
2
1
fer
2
+ ln 2 +
m + mb,0
W W (0) =
,
8sw2 t,0

mt,0 2
(10)
(11)
where Nc = 3 denotes the number of quark colours and we have omitted terms quartic in
fer (0) receives O( ) corrections.
mb,0 in Eq. (11). Note that only W
s
W
Finally, the relation between the bare and the MS-renormalized Yukawa couplings is
given by
hf,0 = h f () + hf
(1)
(2)
= h f () 1 + f,CT + f,CT ,
(1)
(12)
(2)
where hf , f,CT , and f,CT denote the appropriate MS counterterms.

Inserting Eqs. (6), (7), and (12) in Eq. (2) and comparing the result with Eq. (1), where
(1)
(2)
we decompose f () = f () + f (), we obtain the master equations
(1)
W W (0)
E
(1)
f,CT ,
2
2
2MW

(2)
(0)
(2)
(2)
(1)
(1)
(1)
f () = f (1) W W2 + f (1) f,S (1) + 2f (1)
2MW
(2)
(1)
(1)
f,CT f,CT f ().
(1)
(1)
f () = f (1)
(13)
(1)
W W (0)
2
2MW
E
2
(14)
Here, all self-energies are tadpole-free and it is understood that all terms of higher orders
in the electroweak couplings are discarded. We emphasize again that Eqs. (6) and (7) have
to include tadpole contributions to ensure their gauge-parameter independence. However,
in Eqs. (13) and (14), these contributions cancel thanks to the identities
(1)
TW W
2
2MW
(2)
(1)
= Tf ,
TW W
2
2MW
(2)
(1)
= Tf + Tf

(1)
(1)
f,S (1) + 2f (1) ,
(15)
which relate the tadpole contributions to the fermion and boson self-energies at O()
and O(s ), respectively. We have verified Eq. (15) for arbitrary values of the gauge
parameters. Note that Eqs. (13) and (14) are finite and gauge-parameter independent. The
later follows immediately from the gauge-parameter independence of Eqs. (6) and (7). It
should also be remarked that the O() terms of the quantities W W (0) and E are not
needed explicitly, as they drop out in the combination contained in Eq. (14).
204
Through O(s ), f () may be decomposed as

f () =
()
s
s
( )
( )
f () + CF f s () + 2 CF f s (),
(16)
where CF = (Nc2 1)/(2Nc ) is the eigenvalue of the Casimir operator of the fundamental
representation of SU(3)c . The well-known one-loop QCD contribution reads:
f(s ) () = 1
3 2
ln
.
4 Mf2
(17)
The two-loop [5] and three-loop [6] QCD contributions are also known. The O()
contribution is given in Ref. [7]. The evaluation of the O(s ) one is the main purpose
of this paper. It is outlined in Section 4. Due to the presence of many different mass scales,
an exact calculation is rather cumbersome. Thus, we adopt the concept of asymptotic
expansion, which we recall and apply to the one-loop case in the next section.
3. Asymptotic expansion
The method of asymptotic expansion [13] allows for the evaluation of Feynman
integrals which depend on various parameters of different sizes. This is achieved by
expanding the integrand in small ratios of these parameters. As a result, one obtains an
infinite series in powers and logarithms of the expansion parameters.
In our case, the so-called hard-mass procedure is used, where one mass M is much
larger than all other masses {mi } and external momenta {qi }. In this limit, the expansion
can be performed in a diagrammatic way, according to the prescription
M
/ (q, m) T(q ,m ) (M, m , q ),
(M, m, q)
(18)
where is the initial diagram and the operator T(q ,m ) performs a Taylor expansion in all
small parameters of the subdiagram . The sum runs over all subdiagrams that contain the
large mass M and are one-particle-irreducible in all connected parts after contracting all
lines with the large mass.
Let us first consider the case of bottom. In terms of the inherent mass scales, the bottomquark self-energy diagrams can be divided into two classes:
(1) those involving the exchange of charged bosons (W, ); and
(2) those involving the exchange of neutral bosons (Z, , H ).
Exploiting the fact that M = W MW and M = Z MZ depend on the arbitrary gauge

parameters W and Z , we can arrange for the masses involved in these diagrams to fulfil
the following hierarchies:
Mb MW , M Mt ,
(19)
Mb MZ , M , MH .
(20)
205
Since our final results are gauge-parameter independent, we can replace Eq. (19) by the
alternative hierarchy1
Mb MW Mt M .
(21)
This changes the intermediate results for some of the diagrams, but not the final result for
b ().
In the case of the light quarks q = u, d, s, c, the situation is much simpler. Since we
neglect their masses mq , except for the linear appearances in front of q,S , all diagrams
are either reduced to vacuum integrals where the scale is given by a boson mass or to
on-shell integrals which can be adopted from the bottom-quark case.
Similarly, in the case of the fermionic part of the W -boson self-energy, which has to be
evaluated for vanishing external momentum, we only have the hierarchy Mb Mt . As for
the bosonic part of the W -boson self-energy, we use the exact expression given in Eq. (10).
The calculation of the self-energy diagrams can be performed in a completely automated
way. This is achieved by the successive use of the computer programs QGRAF [14], q2e
[15], exp [16], and MATAD [17] (for a review, see Ref. [18]). First, QGRAF is used to
generate the Feynman diagrams. Its output is then rewritten by q2e to be understandable
by exp. This program performs an asymptotic expansion of the diagrams, if necessary,
in an iterated fashion. Finally, the FORM-based [19] program MATAD is used to perform
the actual calculations and expansions in . For the calculation of certain diagrams, it is
supplemented with the package ON-SHELL2 [20].
In a first step, we explicitly check the gauge-parameter independence of the combination
(1)
(1)
f (1) + Tf appearing in Eq. (6). We refrain from listing the corresponding results,
which can be extracted from the expressions given in Section 5, where the relation between
the MS and pole masses is discussed.
()
Let us now move on to b . Since E is only known in t HooftFeynman gauge, we
(1)
(1)
have to choose this gauge also for f (1) and W W (0). We determine the counterterm
(1)
f,CT in Eq. (12) by requiring that Eq. (13) be finite and thus find
(1)
b,CT =

31
s
3
1
3
3
3
CF
+
+
vb2 Q2b 2
2
4
4
4
sw 64cw
16

2
m2b,0 1 3
mt,0 1
Nc
Nc
1
3
+ 2 2
+
.
+ 2 2 +
32 16

MW sw
MW sw 32 16
(22)
Inserting all the above ingredients into Eq. (13), we obtain

()
b () =

2
MH
1
3
Mt2 Nc 1
Mt2 1
5
1 1
+
l
l
+
+
t
t
2
2
2
2
2
64 32
MW sw 32 16
MW sw
MW sw2 64
1 In principle, we could also choose M , M M . In this case, however, Eq. (18) could not be applied, since
b
the external momentum is on the bottom-quark mass shell, which requires different rules of calculation [13].
206

3
1 3
1
5
1
lW H + 2 lW + 2 2
+ 2
2
2
s
32
s
32
16
s
MW MH w
w
w cw

3
1 3
3
5 3
11
2
lZ + 4
ln cw

+ Q2b 1 lb + vb2 lZ
128 64
4
8 4
sw 32

6
2
4
M 1 3
M 1 3
M 1
3
3
+ W2 2
+ lW t + W4 2
+ lW t + W6 2
Mt sw 32 32
Mt sw 32 16
Mt sw

8
10
MW 1 3
MW 1 3
9
3
15
3
+ lW t + 8 2
+ lW t + 10 2
+ lW t
32 32
Mt sw 32 8
Mt sw 32 32

Mb2 1 Nc
11
Nc
1
1
1
3
+
+
l
l
l
l
l
+ 2
t
b+
t +
Z+
H
16
32
32
32
MW sw2 32 192 16

M2
2
(23)
+ W2 vb2 lbZ + .
3
MZ
2
MH
12 /M 12 ), O(M 2 M 2 /M 4 ), or
Here and in the following, ellipses stand for terms of O(MW
t
t
W b
4 ). As we will see below, they are very small and can safely be neglected. Note
O(Mb4 /MW
that there are no terms proportional to Mb2 /Mt2 in Eq. (23). Here and in the following, Qf
is the fractional electric charge of fermion f , I3,f is the third component of weak isospin
of its left-handed component, vf = (If3 2sw2 Qf )/(2cw sw ) and af = If3 /(2cw sw ) are its
vector and axial-vector couplings to the Z boson, respectively, and we use the abbreviations
li = ln(2 /Mi2 ) and lij = ln(Mi2 /Mj2 ),
()
Fig. 2. Comparison of our result for (/ )b (Mb ) with the exact result. The latter is shown as a continuous line.
2 only. The contributions
The long-dashed line indicates the leading-order contribution proportional to Mt2 and MH
given by successively adding the subleading terms are indicated by the short-dashed lines. The horizontal line
marks the actual value of MW /Mt . Note that for this plot MW is fixed to its default value and Mt is varied.
207
For the light quarks q = u, d, s, c, we have

q() () =

2
2
MH
MH
1
1 1
1 3
Mt2 Nc 1
+
l
+
lW H
+
t
2
2
2
2
2
2
MW sw 32 16
MW sw 64 MW MH sw2 32

3
3
1
1
1
11
1 3
2
lZ + 4
ln cw
+ 2 lW + 2 2
sw
4 16
sw cw
128 16
sw 32

M 2 Nc 1
3
5 3
1
+ Q2q 1 lq + vq2 lZ + 2b 2
+ lt + .
4
8 4
MW sw 32 16
(24)
Let us now discuss the quality of our approximation by comparing b() () of Eq. (23)
with the exact result obtained in Ref. [7]. As input values for our numerical analysis,
we use = 1/137.035, Mb = 4.5 GeV, Mt = 174.3 GeV, MW = 80.423 GeV, MZ =
91.1876 GeV [9], and MH = 120 GeV. Furthermore, we choose = Mb . Fig. 2 displays
our result as well as the exact one as functions of MW /Mt . To illustrate the convergence
of the expansion, we show the leading-order contribution separately and then successively
2 /M 2 converges
add the subleading terms. It is obvious from this figure that the series in MW
t
very fast. Indeed, in the range MW /Mt 0.8, the sum of all calculated terms can barely be
10
distinguished from the exact result. The approximation including the O(MW
/Mt10 ) terms
actually provides an excellent approximation also for MW /Mt 1.5, whereas the lowerorder approximations start to exhibit significant deviations at MW /Mt 1.
4. Two-loop result and numerical analysis

In order to obtain f () to O(s ), one has to evaluate the respective two-loop
corrections to f (q) and W W (0), as can be seen from Eq. (14). The result for the latter
is well known (see, e.g., Ref. [21]). Including terms quadratic in mb , one has
(2)
W W (0) =

3 1
2
3
2 1
s CF Nc
1
2
ln
+ +
mt,0 + mb,0
2 sw2
32 2
64 16 m2t,0

2
(2)
3 2 2
1
7
ln 2 +
ln 2 m2t,0
+
32 mt,0 16
128
32
mt,0

5
15
+ (2) + ,
m2b,0
128 32
(25)
where denotes Riemanns zeta function, with the value (2) = 2 /6. We again refrain
from listing explicitly our results for the fermion self-energies, but provide the results
(2)
(2)
for f (), from which the expressions for f (1) can be obtained if the result for the
(2)
counterterm f,CT introduced in Eq. (12) is given. Similarly to the one-loop case discussed
(2)
in the previous section, f,CT
is determined by requiring that Eq. (14) be finite upon
208
renormalization of the top- and bottom-quark masses, and we find

1
9
2
9
s
9
1
(2)
b,CT
= 2 CF
+
Qb + vb2 + 2
2
16
sw 256cw 64 2

m2t,0 1
21
3
2
9
Qb + vb2 + ab2 +
+
+
2 s2
32
32
128sw2
MW
w

2
mb,0 1 3
5
5
1
3
Nc + 2 2
+
Nc
.
+
32 128

MW sw 32 128
(26)
Note that there are no poles proportional to m4t,0 or m4b,0 , although such terms do appear in
certain tadpole diagrams. However, the tadpole contributions cancel in the computation of
b (), as we have explained in Section 2. As an additional check, we can exploit the fact
that the nonlocal logarithmic terms have to cancel as well.
Finally, the O(s ) correction to the relationship between the MS Yukawa coupling
and the pole mass of the bottom quark turns out to be

(2)
3
Mt2 Nc 21
7
3
3 2
(s )
lb lt lb lt lt
b () = 2 2
32
128
64
64
64
MW sw 256

2
3
15
Mt 1
3
9
9 2
13
lb + lt +
lb lt +
l
+ 2 2 + (2) +
64 16
256
32
128
128 t
MW sw

2
2
MH
MH
3
9
1
lW H
1
3
+ 2 2
lb + 2
lb

2 s2
64 256
32 128
MW sw
MW MH
w

3
15
1
9
3
9
71
+ (2) +
lb + lt + lW + lb lW
+ 2
sw
256 16
128
64
16
64

1
33
9
33
9
135
ln c2
3
+ 2 2
+
lb +
lZ +
lb lZ + 4 w
lb
sw cw 1024 512
256
256
sw
32 128

15
3
21
9
7
+ 6 (2) ln 2 (2) (3) + lb + lb2
+ Q2b
64
4
2
16
16

15
9
9
2 23
+ lb + lZ + lb lZ
+ vb
64 32
16
16

2
M 1
21
9
3
9
81
+ (2)
lb
lW t
lb lW t
+ W2 2
128 64
128
128
128
Mt sw

4
M 1
9
5
9
305 15
+ (2)
lb lW t lb lW t
+ W4 2
384 32
128
64
64
Mt sw

6
MW 1
9
5
27
377 39
+ (2)
lb lW t
lb lW t
+ 6 2
384 64
128
64
128
Mt sw

8
MW
9
22669 3
1
13
9
+ (2)
lb
lW t lb lW t
+ 8 2
19200 4
128
320
32
Mt sw
209

10
MW
9
33
45
106451 57
1
+ (2)
lb +
lW t
lb lW t
+ 10 2
76800
64
128
1280
128
Mt sw

Mb2 Nc 29
(2)
9
1
9
3 2
+
lb lt lb lt + lt
+ 2
32
128
16
64
64
MW sw2 256

1 77
89
19
3
(2)
229
+ 2
+
lb
lt +
lZ +
lH
24
2304
1152
1152
128
sw 216
9
17
27
7 2
3 2
lb lt +
lb lZ
lb lH +
l +
l
128
384
128
192 b 128 t

13 2
9 2
l +
l
384 Z 128 H

2
MW
8
25
13
11 2 1 2
25
2
+ 2 vb + (2) lb + lZ + lb lZ lb lZ
18
9
18
12
12
6
MZ

2
M 1
5
13 (2)
+ lbt + .
+ W2 2 +
(27)
96
16
96
Mt sw
We should mention that the QED portion of the O(s ) contribution can also be inferred
from the two-loop O(s2 ) one [5] by replacing s2 CF2 with 2s CF Q2f and setting all the
( )
other colour structures to zero. In order to obtain the result for b s () in Eq. (27), we
have used the tadpole-free bottom-quark self-energy and adopted the t HooftFeynman
gauge, the one in which E is given. The gauge-parameter independence of the full selfenergy is discussed in the next section.
In the case of the light quarks q = u, d, s, c, we obtain the compact expression
q(s ) () =

(2)
3
Mt2 Nc 21
7
3
3 2
lq lt lq lt lt
2 s 2 256
32
128
64
64
64
MW
w

2
2
MH
MH
3
9
1
3
1
lW H
lq + 2
lb
+ 2 2

2 s2
64 256
32 128
MW sw
MW MH
w

3
1 79
21
9
+ lq + lW + lq lW
+ 2
64
64
sw 256 16

33
33
9
1
135
+
lq +
lZ +
lq lZ
+ 2 2
sw cw 1024 512
256
256

9
ln c2
3
lq
+ 4w
sw
32 128

15
3
21
9 2
2 7
+ 6(2) ln 2 (2) (3) + lq + lq
+ Qq
64
4
2
16
16

9
9
23 15
+ vq2
+ lq + lZ + lq lZ
64 32
16
16
210

Mb2 Nc 29
(2)
3
3
1
3
+
lq lb lt lq lt
+ 2 2
s
256
32
128
64
16
64
MW w

3
3
lb lt + lt2 + .
32
64
(28)
In order to demonstrate the good convergence properties of our result, we show in

Fig. 3 the O(s ) contribution to b (Mb ) as a function of MW /Mt for MH = 120 GeV. In
addition to the parameters specified after Eq. (24), we use s (Mb ) = 0.1905 appropriate for
(5)
six active quark flavours, which we evaluate from the present world average s (MZ ) =
0.1172 [9] using the program RunDec [22]. Again, we observe an excellent convergence
up to MW /Mt 0.8, which is more than sufficient for our purpose. Since it is not possible
to distinguish between the different lines in the physically interesting region in Fig. 3, we
give explicit numbers for the subleading terms at MW /Mt 0.46. For this purpose, we
write the series as
2 k

MW
M2
M2
b(s ) (Mb ) = Cl,1 2t + Cl,2 H
(29)
+
C
k
2
2
MW
MW
M
t
k=0
and thus obtain
5

n=0
5
2 /M 2 )n
Cn (MW
t
2
2 m
m=0 Cm (MW /Mt )
= 0.997925 + 0.004058 0.001451 0.000448

0.000078 0.000006.
( )
(30)
Fig. 3. (s / 2 )CF b s (Mb ) as a function of MW /Mt for MH = 120 GeV. The long-dashed line indicates
the leading-order contribution only. The contributions given by successively adding the subleading terms are
indicated by the short-dashed lines. The horizontal line marks the actual value of MW /Mt . Note that for this plot
MW is fixed to its default value and Mt is varied.
211
Fig. 4. b (Mb ) as a function of MH . The dash-dotted and full lines indicate the O() and O(s ) contributions,
respectively. For comparison, also the pure QCD contributions of O(s ) (short-dashed line), O(s2 ) (long-dashed
line), and O(s3 ) (dotted line) are shown. The latter are, of course, independent of MH .
0
This demonstrates that, aside from the leading-order terms, the O(MW
/Mt0 ) term
yields the largest contribution by far, while all other subleading terms are rather small.
Furthermore, we wish to mention that the inclusion of the corrections quadratic in Mb
have no visible effect in Fig. 3.
Our final numerical results are presented in Figs. 4 and 5. Fig. 4 shows the O() and
O(s ) contributions to b (Mb ) as functions of MH . For comparison, also the pure QCD
ones of O(s ), O(s2 ), and O(s3 ) are plotted. We observe that the O(s ) contribution
exhibits a rather weak dependence on MH . Futhermore, for MH 500 GeV, it becomes
comparable in size to the O() contribution, whose magnitude decreases with increasing
value of MH . From Fig. 4, we also see that our new O(s ) contribution is of the same
order of magnitude as the O(s3 ) one.
Fig. 5 shows b () for MH = 120 GeV as a function of in the range Mb < <
1000 GeV. We observe that the O(s ) contribution takes positive values in the range
20 600 GeV. It is comparable in size to the O(s3 ) one only in the lower range.
5. Relationship between MS and pole masses

As a by-product of our calculation, we obtain the O(s ) correction to the relationship
between the MS and pole definitions of mass for the first five quark flavours. As discussed
in Section 2, it is necessary to include the tadpole contributions in order to obtain
gauge-parameter-independent results. This has been checked explicitly by considering the
hierarchies discussed in Eqs. (19), (20), and (21).
212
Fig. 5. b () as a function of for MH = 120 GeV. The same coding as in Fig. 4 is adopted.
The starting point for the derivation of the MS to on-shell relationship is Eq. (6), where
we replace the bare quark mass by the MS one. The corresponding relation is obtained in
analogy to Eq. (26), with the only difference that we also have to allow for terms quartic in
mb,0 and mt,0 because of the tadpole contribution. By requiring finiteness of the resulting
relation, we obtain

2

MZ2
mb,0
3
31
s
3 MH
3
Q2b + vb2 ab2
+
= 1 CF
2 s2 M 2
2
m
b
4
4
16cw
32sw2 MW
w
H

2b 1
2t
m
4t
3 M2
3 m
Nc
3 m
2 W
+
+
2
2
2 M2
2
8sw MH
32sw2 MW
4sw2 MW
32sw2 MW
H

2

MZ2
9
s
9 MH
9
2
Qb + vb2 ab2 +
+ 2 CF
+
2 s2 M 2
2
16
64cw
128sw2 MW
w
H

2
2b 1
2t
m
4t
9 MW
9 m
9Nc
9 m
+
+
2
2
2 M2
2 2
32sw2 MH
64sw2 MW
16sw2 MW
64sw2 MW
H

21
3
2
9
Qb + vb2 + ab2 +
+
32
32
128sw2

2b 1
2t
m
4t
3 m
Nc
3 m
+ .
+
+
(31)
2
2 M2
2
32sw2 MW
8sw2 MW
32sw2 MW
H
If the tadpole contributions were omitted here, then there would be no terms quartic in m
b
and m
t and no terms involving MH . In turn, gauge-parameter dependence would appear in
the nonleading terms.
213
The relationship between the MS and pole masses of the bottom quark through O(s )
reads:

Mt4 Nc 1 1
3
m
b ()
s
+
l
= 1 + CF 1 lb +
t
2 M 2 s2 4
Mb
4
MW
4
H w

2
M 1
M2 1
3
3
5
3
3
2
+ 2t 2 lt + H
l
l
+
Q
1
H
b
b
2 s2
64 32
32 32
4
MW sw
MW
w

M2 2
5 3
1 3
+ vb2 lZ + ab2 + lZ + Z
a
1
3l
Z
2 b
8 4
8 4
MH

2
2
MW
MW
3
1
1 3
1 3
+ lW t
+ 2 2 lW + 2 2
8 8
MH sw
Mt sw 32 32

6
4
MW
MW
3
9
1 3
1 3
+ lW t + 6 2
+ lW t
+ 4 2
Mt sw 32 16
Mt sw 32 32

8
10
MW
MW
3
15
1 3
1 3
+ 8 2
+ lW t + 10 2
+ lW t
Mt sw 32 8
Mt sw 32 32

Mb2 1 11
1
1
1
3
lb + lt + lZ + lH
+ 2
32
32
32
MW sw2 192 16

M2
2
+ W2 vb2 lbZ
3
MZ

s
3
Mt4 Nc
1
3
3 2
+ 2 CF
lb lt lb lt lt
2 M 2 s2
8 16
16
8
MW
H w

2
3
15
3
9
9 2
Mt 1
13
lb + lt +
lb lt +
l
+ 2 2 + (2) +
64 16
256
32
128
128 t
MW sw

2
MH
9
3
9
3
1
111
1
+
lb + 2
+ (2) + lt
+ 2 2 (1 + lH )
32 128
sw
256 16
64
MW sw

15
3
21
7
9
+ 6 (2) ln 2 (2) (3) + lb + lb2
+ Q2b
64
4
2
16
16

15
9
9
2 23
+ lb + lZ + lb lZ
+ vb
64 32
16
16

3
9
9
55
+ lb + lZ lb lZ
+ ab2
64 32
16
16

2
2
MW
MZ 1
3
3
1
1
1
+ lb + 2 2 (1 + 3lW ) + lb
+ 2 2 2 (1 + 3lZ )
16 64
8 32
MH sw cw
MH sw

2
MW 1
21
9
81
3
9
+ (2)
lb
lW t
lb lW t
+ 2 2
128 64
128
128
128
Mt sw
214

4
MW
9
5
9
305 15
1
+ (2)
lb lW t lb lW t
+ 4 2
384 32
128
64
64
Mt sw

M6 1
9
5
27
377 39
+ (2)
lb lW t
lb lW t
+ W6 2
384 64
128
64
128
Mt sw

8
M 1
9
13
9
22669 3
+ (2)
lb
lW t lb lW t
+ W8 2
19200 4
128
320
32
Mt sw

10
M 1
33
45
106451 57
9
+ W
+
+
(2)
l
l
l
b
W
t
b
W
t
76800
64
128
1280
128
Mt10 sw2

2
M
(2)
229
89
19
3
1 77
+
lb
lt +
lZ +
lH
+ 2b
2
24
2304
1152
1152
128
MW sw 216

9
17
27
7 2
3 2
13 2
9 2
lb lt +
lb lZ
lb lH +
lb +
lt
lZ +
lH
128
384
128
192
128
384
128

2
M
8
25
25
13
11
1
+ W2 vb2 + (2) lb + lZ + lb lZ lb2 lZ2
18
9
18
12
12
6
MZ

2
M 1
5
13 (2)
+ lbt
+ W2 2 +
(32)
+ .
96
16
96
Mt sw
At O(s ), the term quadratic in Mt agrees with the result of Ref. [23].
We conclude this section by presenting the relationship through O(s ) between the
MS and pole masses of the first four quark flavours. It reads:

m
q ()
Mt4 Nc 1 1
3
s
+ lt
= 1 + CF 1 lq +
2 M 2 s2 4
Mq
4
MW
4
H w

2
MH 1
3
3
1 3
3
5 3
2
2
+ Qq 1 lq + vq lZ
+ 2 2 lH 2
32 32
sw 32
4
8 4
MW sw

2
2
M 1
M 2
1 3
1 3
+ aq2 + lZ + Z
lW
aq [1 3lZ ] + W
2
2
2
8 4
8 8
MH
MH sw

4
3
Mt Nc
1
3
3 2
s
lq lt lq lt lt
+ 2 CF
2 M 2 s2
8 16
16
8
MW
H w

2
MH
9
9
1
3
9
1 39
+
lq + 2
+
lq + lW
+ 2 2 (1 + lH )
32 128
sw 256 128
64
MW sw

15
3
21
7
9
+ 6 (2) ln 2 (2) (3) + lq + lq2
+ Q2q
64
4
2
16
16

9
9
15
23
+ vq2
+ lq + lZ + lq lZ
64 32
16
16

MZ2 2
3
9
9
3
2 55
+ lq + lZ lq lZ + 2 aq (1 + 3lZ ) 1 + lq
+ aq
64 32
16
16
4
MH

2
MW
3
1
1
+ 2 2 (1 + 3lW ) + lq + .
8 32
MH sw
215
(33)
6. Summary and conclusion

We calculated the O(s ) corrections to the relationships between the MS Yukawa
couplings and the pole masses of the first five quark flavours in the SM. We demonstrated
that these relationships are gauge-parameter independent and free of tadpole contributions.
The method of asymptotic expansion was used in the calculation, and the results were
found as polynomials in small mass ratios with coefficients that contain logarithms of
these mass ratios. The goodness of this method was tested in two ways. On the one hand,
our O() expression for b (Mb ) was found to reproduce the analytic result of Ref. [7]
extremely well. On the other hand, our O(s ) result turned out to converge very fast
2 /M 2 were included. These two observations reassure us of the
as higher powers of MW
t
reliability of our result, which is equivalent to the exact analytic result for all practical
purposes.
As for the phenomenological significance of our results, our numerical analysis revealed
that, at MH = 100 GeV, the O(s ) contribution to b (Mb ) amounts to roughly one half
of the O() one. These two contributions depend very differently on the value of MH , and
they even cross over at MH 500 GeV. Obviously, the O(s ) contribution cannot be
neglected against the O() one.
We also calculated the O(s ) corrections to the relationships between the MS and
pole masses of the first five quark flavours. We showed that these relationships are gaugeparameter independent if the tadpole contributions are properly included in the contributing
self-energies.
References
[1] M. Bando, T. Kugo, N. Maekawa, H. Nakano, Mod. Phys. Lett. A 7 (1992) 3379;
M. Carena, M. Olechowski, S. Pokorski, C.E.M. Wagner, Nucl. Phys. B 426 (1994) 269;
B.C. Allanach, S.F. King, Phys. Lett. B 328 (1994) 360;
C.F. Kolda, L. Roszkowski, J.D. Wells, G.L. Kane, Phys. Rev. D 50 (1994) 3498;
B. Schrempp, Phys. Lett. B 344 (1995) 193;
J. Kubo, M. Mondragon, G. Zoupanos, Prog. Theor. Phys. Suppl. 123 (1996) 127;
B. Brahmachari, Phys. Rev. D 59 (1999) 035009;
S. Komine, M. Yamaguchi, Phys. Rev. D 65 (2002) 075013;
K. Tobe, J.D. Wells, Nucl. Phys. B 663 (2003) 123.
[2] H. Arason, D. Castao, B. Keszthelyi, S. Mikaelian, E. Piard, P. Ramond, B. Wright, Phys. Rev. Lett. 67
(1991) 2933;
H. Arason, D.J. Castao, B. Keszthelyi, S. Mikaelian, E.J. Piard, P. Ramond, B.D. Wright, Phys. Rev. D 46
(1992) 3945;
H. Arason, D.J. Castao, E.J. Piard, P. Ramond, Phys. Rev. D 47 (1993) 232;
M. Luo, Y. Xiao, Phys. Rev. Lett. 90 (2003) 011601;
M. Luo, H. Wang, Y. Xiao, Phys. Rev. D 67 (2003) 065019;
M. Luo, Y. Xiao, Phys. Lett. B 555 (2003) 279.
[3] B.A. Kniehl, F. Madricardo, M. Steinhauser, Phys. Rev. D 62 (2000) 073010;
216
K.-P.O. Diener, B.A. Kniehl, Nucl. Phys. B 617 (2001) 291.

[4] R. Tarrach, Nucl. Phys. B 183 (1981) 384;
E. Braaten, J.P. Leveille, Phys. Rev. D 22 (1980) 715.
[5] N. Gray, D.J. Broadhurst, W. Grafe, K. Schilcher, Z. Phys. C 48 (1990) 673;
D.J. Broadhurst, N. Gray, K. Schilcher, Z. Phys. C 52 (1991) 111.
[6] K.G. Chetyrkin, M. Steinhauser, Phys. Rev. Lett. 83 (1999) 4001;
K.G. Chetyrkin, M. Steinhauser, Nucl. Phys. B 573 (2000) 617;
K. Melnikov, T.V. Ritbergen, Phys. Lett. B 482 (2000) 99.
[7] R. Hempfling, B.A. Kniehl, Phys. Rev. D 51 (1995) 1386.
[8] A.I. Bochkarev, R.S. Willey, Phys. Rev. D 51 (1995) 2049.
[9] Particle Data Group, K. Hagiwara, et al., Phys. Rev. D 66 (2002) 010001.
[10] F. Jegerlehner, M.Y. Kalmykov, Nucl. Phys. B 676 (2004) 365;
F. Jegerlehner, M.Y. Kalmykov, Acta Phys. Pol. B 34 (2003) 5335.
[11] M. Faisst, J.H. Khn, T. Seidensticker, O. Veretin, Nucl. Phys. B 665 (2003) 649;
M. Faisst, J.H. Khn, O. Veretin, Report No. TTP 04-04, SFB/CPP-04-06, hep-ph/0403026.
[12] A. Sirlin, Phys. Rev. D 22 (1980) 971.
[13] V.A. Smirnov, Applied Asymptotic Expansions in Momenta and Masses, Springer-Verlag, Berlin, 2001.
[14] P. Nogueira, J. Comput. Phys. 105 (1993) 279.
[15] T. Seidensticker, unpublished.
[16] T. Seidensticker, Report No. TTP99-22, hep-ph/9905298;
R. Harlander, T. Seidensticker, M. Steinhauser, Phys. Lett. B 426 (1998) 125.
[17] M. Steinhauser, Comput. Phys. Commun. 134 (2001) 335.
[18] R. Harlander, M. Steinhauser, Prog. Part. Nucl. Phys. 43 (1999) 167.
[19] J.A.M. Vermaseren, Symbolic Manipulation with FORM, Computer Algebra Netherlands, Amsterdam,
1991.
[20] J. Fleischer, M.Y. Kalmykov, Comput. Phys. Commun. 128 (2000) 531.
[21] A. Djouadi, P. Gambino, Phys. Rev. D 49 (1994) 3499;
A. Djouadi, P. Gambino, Phys. Rev. D 53 (1996) 4111, Erratum.
[22] K.G. Chetyrkin, J.H. Khn, M. Steinhauser, Comput. Phys. Commun. 133 (2000) 43.
[23] A. Kwiatkowski, M. Steinhauser, Report No. LBNL-37881, TTP 95-35, hep-ph/9510456.

Neutrino oscillation physics with a higher- -beam

J. Burguet-Castell a , D. Casper b , J.J. Gmez-Cadenas a ,
P. Hernndez a , F. Snchez c
a Departamento de Fsica Terica and IFIC, Universidad de Valncia, E-46100 Burjassot, Spain
b Department of Physics and Astronomy, University of California, Irvine, CA 92697-4575, USA
c IFAE, Universidad Autonoma de Barcelona, E-08193 Bellaterra, Barcelona, Spain
Received 15 December 2003; received in revised form 9 April 2004; accepted 11 June 2004
Abstract
The precision measurement and discovery potential of a neutrino factory based on boosted
radioactive ions in a storage ring (-beam) is re-examined. In contrast with past designs, which
assume ion factors of 100 and baselines of L = 130 km, we emphasize the advantages of
boosting the ions to higher and increasing the baseline proportionally. In particular, we consider a
medium- scenario ( 500, L 730 km) and a high- scenario ( 2000, L 3000 km).
The increase in statistics, which grow linearly with the average beam energy, the ability to exploit
the energy dependence of the signal and the sizable matter effects at this longer baseline all increase
the discovery potential of such a machine very significantly.
1. Introduction
The spectacular results in atmospheric [1], solar [2], reactor [3] and long-baseline [4]
neutrino experiments in recent years can be economically accommodated in the Standard
Model (SM) with neutrino masses and a three-neutrino mixing matrix [5]. In this case, the
lepton sector of the SM closely resembles that of the quarks and there are a number of
new physical parameters that can be measured at low energies: the three neutrino masses,
mi (i = 1, 2, 3), three mixing angles, ij (i = j = 1, 2, 3), and a CP violating phase, .
E-mail addresses: jordi.burguet-castell@cern.ch (J. Burguet-Castell), dcasper@uci.edu (D. Casper),
gomez@mail.cern.ch (J.J. Gmez-Cadenas), pilar.hernandez@cern.ch (P. Hernndez), fsanchez@ifae.es
(F. Snchez).
doi:10.1016/j.nuclphysb.2004.06.021
218
J. Burguet-Castell et al. / Nuclear Physics B 695 (2004) 217240
In contrast with the quark sector two additional phases could be present if neutrinos are
Majorana. Of all these new parameters, present experiments have determined just two
neutrino mass-square differences and two mixing angles: (|m223| 2.5 103 eV2 ,
23 45 ) which mostly drive the atmospheric oscillation and (m212 7 105 eV2 ,
12 35 ) which mostly drive the solar one. The third angle, 13 , as well as the CPviolating phases (, and possible Majorana phases) remain undetermined. Another essential
piece of information needed to clarify the low-energy structure of the lepton flavor sector
of the SM is the neutrino mass hierarchy. This is related to the sign of the largest masssquare difference (m223 ), which determines if the spectrum is hierarchical (if the two most
degenerate neutrinos are lighter than the third one) or degenerate (if they are heavier).
The measurement of these parameters requires, for the first time, high-precision
neutrino-oscillation experiments. A number of possible experimental setups to significantly
improve the present sensitivity to 13 , and the sign of m223 have been discussed in the
literature: neutrino factories (neutrino beams from boosted-muon decays) [6], superbeams
(very intense conventional neutrino beams) [79] improved reactor experiments [10] and
more recently -beams (neutrinos from boosted-ion decays) [11,12]. These are quite
different in terms of systematics but all face a fundamental problem which limits the
reach of each individual experiment significantly, namely the problem of correlations and
degeneracies between parameters [1319]; 13 and must be measured simultaneously,
and other oscillation parameters are not known with infinite precision.
To resolve these degeneracies it is important to measure as many independent channels
as possible and to exploit the energy and/or baseline dependence of the oscillation signals
and matter effects in neutrino propagation. In many cases, the best way to do this is by
combining different experiments; indeed the synergies between some combinations of the
setups mentioned above have been shown to be very large.
The neutrino factory is thought to provide ultimate sensitivity to leptonic CP violation,
and thus has been considered for a long time as the last step on a long-term road map to
reveal the lepton-flavor sector of the SM. In this road map there are important intermediate
milestones, such as superbeams or improved reactor experiments, that can mitigate the
problem of degeneracies. In contrast, the present conception of the -beam, which shares
many of the good properties of the neutrino factory, has been shown to provide a rather
limited sensitivity.
The purpose of this paper is to show that, in fact, a -beam running at a higher
than previously considered (and longer baselines), in combination with a massive water
detector, can reach sensitivity to leptonic CP-violation and the sign(m223) that competes
with the neutrino factory. On the other hand, the R&D effort required to increase the
factor for the -beam setup is probably much less than required to realize a high-energy,
high-luminosity neutrino factory. Thus, an optimized -beam may turn out to be a serious
competitor to the neutrino factory as the ultimate machine to search for CP violation.
The paper is organized as follows. In Section 2 we recall a few facts about the present
design for the -beam and discuss the advantages of increasing the ion factor and the
baseline. We also discuss the expected fluxes and event rates for three reference setups
that will be considered for comparison, at low, medium and high . In Section 3, the
performance of a large water Cherenkov apparatus, proposed as the optimal detector for
219
the low and medium setups is discussed in detail. Section 4 compares the physics results
of the three reference setups. In Section 6 we present our outlook and conclusions.
2. The -beam
The -beam concept was first introduced in [11]. It involves producing a beam of unstable heavy ions, accelerating them to some reference energy, and allowing them to
decay in the straight section of a storage ring, resulting in a very intense neutrino beam.
Two ions have been identified as ideal candidates: 6 He, to produce a pure e beam, and
18 Ne, to produce a beam. The golden subleading transitions and can
e
e
be measured through the appearance of muons in a distant detector.

As for the neutrino factory muon-induced neutrino beams, the -beam offers the unique
features of being pure (e.g., only one neutrino species, in contrast to a conventional superbeam where contamination of other neutrino species is inevitable) and virtually systematics
free, since the spectrum can be calculated exactly (again, in contrast with a conventional
beam, where knowledge of the spectrum always involves a sizable systematic uncertainty).
One of the most attractive features of the -beam is that it leverages existing CERN
facilities. The present design, whose feasibility with existing technology has been recently
demonstrated [20], envisions a low- scenario, in which ions are produced by a new
facility (EURISOL), accelerated by the present SPS to 150, and stored in a storage
ring (also a new facility) with straight sections pointing to the experimental area. An
underground location where a very massive neutrino detector could be located has been
identified in the Frjus tunnel, roughly 130 km from CERN. This baseline is ideal for
exploring the first peak of the atmospheric oscillation, the optimal environment to search
for CP-violating effects. A new, very large cavern excavated in the Frjus tunnel would
host a megaton water Cherenkov detector a l UNO. The capabilities of such a detector are
well-matched to this energy range, and the low neutrino energies produced by the low-
option (in the range of a few hundred MeV) require a very large mass to compensate for
the tiny cross sections.
Furthermore, the existing design calls for a conventional low-energy super-beam
based on the proposed SPL proton driver [21], that would deliver a total power-on-target of
about 4 MW, resulting in a very intense neutrino beam. The physics reach of such a superbeam has been studied in detail, both alone [9], and together with a low- -beam [12,22].
The results of these studies can be summarized as follows:
Neither the SPL super-beam nor the low- -beam, by themselves, result in fullysatisfactory performance, especially compared to other proposed facilities such as
JHF [7]. The performance is limited, in spite of the large detector mass, by (relatively)
small rates (due to the small cross sections), by systematics due to the SPL superbeam backgrounds (both beam- and detector-related) and by the intrinsic degeneracies
identified in [14].
A combination of the super-beam and -beam would explore a large range of the
parameter space (13 , ). A detailed study of the systematics involved (in particular
the absolute flux normalization of the SPL super-beam and uncertainties in background
220
Fig. 1. P (e ) and P (e ) as a function of the baseline L in km, at a neutrino energy

E/L = |m223 |/2 and for 13 = 8 and = 0 (solid) and 90 (dashed).
subtraction) remains to be done, however. On-going studies in the context of the JHF
project suggest that these systematics are not small.
The sign of the atmospheric m223 cannot be determined because matter effects are
negligible.
Here, instead, we explore increasing the energy of the -beam, with a corresponding
increase in baseline to keep E /L approximately constant. There are three reasons to
expect an improvement of sensitivity in this case:
The rates (and thus sensitivity to 13 if the backgrounds are kept under control) increase
linearly with E at fixed E /L.
At longer baselines, measurement of the neutrino mass hierarchy becomes possible,
as matter effects become more sizable. This is illustrated in Fig. 1, which shows
the e oscillation probability for neutrinos and anti-neutrinos, as a function
of the baseline, for neutrino energy constrained to the first atmospheric peak, i.e.,
E/L = |m223|/2 . The difference between the neutrino and anti-neutrino oscillation
probabilities induced by matter effects becomes comparable to that due to CP-violation
for L = O(1000) km.
Increased neutrino energy enhances the energy dependence of oscillation signals.
This is extremely useful in resolving the correlations and degeneracies in parameter
space [13,14]. Energy dependence is particularly helpful for energies close to the peak
221
Fig. 2. P (e ) and P (e ) as a function of the energy in GeV, at L = 730 km, for 13 = 6 and
(E ), (E ) with E = 1.5 GeV (dashed).
= 40 (solid) and for the degenerate solution at 13
of the atmospheric oscillation [18], which is precisely the regime under consideration.
Fig. 2 compares the vacuum probabilities for e and e with 13 = 6
and = 40 to those of the intrinsic-degenerate solution at E( ) . The neutrino and
anti-neutrino probabilities cross at the E( ) , but are quite different at other energies.
Thus spectral information can definitely help in disentangling them.
From a technical point of view, designs aiming at higher factors are conceivable by
direct extrapolation of existing technology, and would not require a long R&D program.
A medium- scenario ( 600) could be realized at CERN by accelerating ions in a
refurbished SPS with super-conducting magnets, or in LHC (up to 2488 for 6 He and
4158 for 18 Ne) [23]. Another candidate would be Fermilab, where a combination of the
existing Main-Injector and Tevatron could accelerate ions to factors of a few hundred.
The possibility of an associated super-beam1 will not be considered here, because
the systematics are very different. To gain a clear view of the relative merits of each
-beam scenario, it is best to first compare their stand-alone performance. Three scenarios
for the -beam alone are considered, and a future paper will address comparisons and
combinations with other facilities. We define the following three setups:
1 Note that it is also possible to increase the energy of the super-beam in the present design, by increasing the
energy of the SPL.
222
Setup I, low energy: = 60 for 6 He and = 100 for 18 Ne, with L = 130 km (CERN
Frjus) as in [12,22].2
Setup II, medium energy: = 350 for 6 He and = 580 for 18 Ne, with L = 732 km
(e.g., CERNGran Sasso with a refurbished SPS or with the LHC, FNALSoudan).
Setup III, high energy: = 1500 for 6 He and = 2500 for 18 Ne, with L = 3000 km
(e.g., CERNCanary islands with the LHC).
In all cases we assume the same number of ions as the existing design, that is,
2.9 1018 6 He and 1.1 1018 18 Ne decays per year.3 This seems reasonable, as one does
not expect loses with a refurbished SPS (to extrapolate, for instance, setup I to setup II at
CERN). For the LHC one could compensate for injection loses due to the different optics
with a different acceleration scheme with more or longer bunches (thus more ions) [24].
2.1. Neutrino fluxes
Neglecting small Coulomb corrections, the electron energy spectrum produced by an
allowed nuclear -decay at rest is described by
dN rest
pe2 (Ee E0 )2 ,
dpe
(2.1)
where E0 is the electron end-point energy and Ee and pe are the electron energy and
momentum. For 6 He, E0 = 3.5 MeV + me , while for 18 Ne, E0 = 3.4 MeV + me .
We are interested instead in the neutrino spectrum resulting from ion decays after they
are boosted by some fixed . In the ion rest frame the spectrum of the neutrinos is

dN rest
E2 (E0 E ) (E E0 )2 m2e .
(2.2)
d cos dE
After performing the boost and normalizing the total number of ion decays to be N per
year, the neutrino flux per solid angle in a detector located at a distance L aligned with the
straight sections of the storage ring is

N 2 2
d lab
y

(1
y)
(1 y)2 ye2 ,
(2.3)
dS dy 0 L2 g(ye )
E
2 E0
1 ye , ye = me /E0 and

1
ye

g(ye )
.
1 ye2 2 9ye2 8ye4 + 15ye4 Log
60
1 1 ye2
where 0 y =
(2.4)
Note the similarity of this expression and the electron neutrino fluxes at a neutrino
factory [6]. Another similarity is that the fluxes are known very accurately and the ( )
2 Different for 6 He and 18 Ne are required to allow simultaneous acceleration and storage of both ions in
the same ring, reducing the necessary running time by a factor of two. The different ion charge:mass ratios imply
a 1.67 ratio of factors [23].
3 Although the originally-projected intensity of 18 Ne was a factor three less, it was recently proposed [23]
that three bunches could be accommodated.
223
appearance signal has no background from contamination of the beam. The latter is true
for the neutrino factory only to the extent that the charge of final-state leptons can be
determined, which requires, therefore, a magnetized device (thus, in particular it prevents
the use of massive water detectors).
Fig. 3 shows these fluxes at the three reference setups as a function of the neutrino
energy. Although integrated fluxes for all the three setups are nearly identical, about
O(1011 )e /e m2 year1 , setups II and III have the advantage that the appearance signals
energy dependence should be more significant, while at low energy the neutrino energy
resolution is degraded by the Fermi motion.
It is instructive to compare with the neutrino factory flux of 1012 e /e m2 year1 at
the optimum baseline of 3000 km. This is a factor 10 higher than setup II, but E /L for
setup II and III is matched to the atmospheric splitting, while the neutrino factory is not.
The oscillation probabilities are thus smaller in the latter.
2.2. Cross-sections
The (anti-)neutrino cross-sections relevant in the three setups are quite different.
While in the low energy option quasi-elastic events are dominant and the cross-section
grows rapidly with energy, in the highest-energy option samples are mostly deep-inelastic
scattering and the growth is linear in the neutrino energy. For the medium-energy option,
there is a sizable contribution from both types of events, as well as resonant channels. Fig. 4
shows the cross sections per nucleon and per neutrino energy used in this analysis.
Asymptotically the number of events grows linearly with . Table 1 shows the number
of charged-current events expected in one year, per kiloton. The number of events is
somewhat sensitive to the isotopic composition of the target, as free protons in water
contribute significantly to the anti-neutrino event rates.
Interestingly, the detector mass can be reduced linearly with without changing the
number of events. This offers the possibility of moving from the large water Cherenkov
detector required in the lowest-energy option to a less massive but more granular detector
in the higher-energy one.
3. Detectors: efficiencies and backgrounds

The signature for the golden subleading transitions e and e in a -beam
is the appearance of prompt muons which must be separated from the main background
of prompt electrons due to the bulk e /e charged interactions. Since there is only
one neutrino species in the beam, no charge identification is required. Furthermore, to
compensate the small cross sections, specially for setup I, very massive detectors are
needed. In addition, good energy resolution is required in order to resolve degeneracies.
As demonstrated by Super-Kamiokande [1], massive water detectors are capable
of offering simultaneously excellent particle identification and good energy resolution,
particularly in the range of few hundred MeV to about 1 GeV, where most of the
interactions are quasi elastic, yielding simple event topologies (a typical QE interaction
is characterized by a single ring from the final muon, the scattered proton being below
224
Fig. 3. Fluxes for the three setups as function of the neutrino energy for e (solid) and e (dashed).
225
Fig. 4. Cross section per nucleon for an isoscalar target, divided by neutrino energy in GeV.
Table 1
Number of charged-current events without oscillations per kton-year for the three reference setups. Also shown
is the average neutrino energy
L (km)
e CC
e CC
E (GeV)
60/100
350/580
1500/2500
130
730
3000
4.7
57.5
282.7
32.8
224.7
993.1
0.23/0.37
1.35/2.18
5.80/9.39
Cherenkov threshold thus invisible). As energy increases, deep inelastic processes start to
dominate the cross section and the event topology becomes more complicated. The turnover region is about 1.5 GeV. The neutrino spectra in setup II extend all the way up to
4 GeV. Nevertheless, as it will be shown below, water is still the best option in this range.
For neutrinos energies up to 10 GeV, as considered in setup III, deep inelastic CC and
NC events are dominant and water is no longer suitable. Massive tracking calorimeters are
the best option in this range [13,25].
3.1. Signal selection and background suppression in water
We have considered a Megaton-class water detector, as proposed by the UNO
Collaboration [26] with a fiducial mass of 400 kiloton, for both setups I and II. The
226
response of the detector was studied using the NUANCE [27] neutrino physics generator
and detector simulation and realistic reconstruction algorithms as described in [9].
Particle identification in water exploits the difference in the Cherenkov patterns
produced by showering (e-like) and non-showering (-like) particles. Besides, for the
energies of interest the difference in Cherenkov opening angle between an electron and
a muon can also be exploited. Furthermore, muons which stop and decay (100% of +
and 78% of ) produce a detectable delayed electron signature that can be used as an
additional handle for background rejection.
For this study, we have used a particle identification criteria similar to the one used
by the Super-Kamiokande Collaboration, which is based on a maximum likelihood fit of
both -like and e-like hypotheses. Fig. 5 shows the particle identification estimator Pid , for
electron-like events (solid line) and for muon-like events (dashed line). The normalization
is arbitrary. A cut at Pid > 0 (PID cut) separates optimally the e-like and -like
populations. Since most events are followed by a muon-decay signature, the background
is further reduced by accepting only events with a delayed coincidence.
To summarize, the event selection for both setup I and setup II requires:
The event must be fully contained in the fiducial volume. This is necessary to guarantee
a good measurement of the energy as well as to exploit the muon-decay signature.
A single ring in the event.
Fig. 5. Rejection of eCC background in a water Cherenkov detector. The particle ID estimator Pid (in arbitrary
units) is shown for the electron-like background (left, solid line) and muon-like signal (right, dashed).
227
The PID estimator must be muon-like P > 0.

Event must show a delayed coincidence (muon decay signature).
3.2. Signal and backgrounds in setup I
The PID cut eliminates almost completely the electron background, leaving a residual
background due to eNC and diffractive events in which a single pion is confused with a
muon. The low energy of setup I (particularly in the case of the anti-neutrinos) results
in negligible backgrounds for 6 He. In the case of 18 Ne, diffractive events result in an
integrated background fraction below 102 . The efficiency is rather large but drops
dramatically below 300 MeV. Background ratio and efficiency as a function of the energy
in setup I are shown in Fig. 6.
A major drawback of setup I is that no energy binning is possible, since the neutrino
energy is of the order of the Fermi motion. This is illustrated in Fig. 7 where the
reconstructed neutrino energy is plotted against the true energy. As it can be seen the events
Fig. 6. Background fraction (top) and efficiency (bottom) as a function of the true neutrino energies for 6 He and
18 Ne in water in setup I.
228
Fig. 7. Reconstructed versus true neutrino energy for 18 Ne. The lack of correlation shows clearly that the event
energy information is completely washed out by Fermi motion.
are almost uncorrelated. Therefore, spectral information cannot be used and one has to
make do with the integrated signal, which, alas, cannot resolve the intrinsic degeneracies.
3.3. Signal and backgrounds in setup II
Fig. 8 shows the reconstructed energy spectrum for both signal and background in
setup II. Notice that, as for setup I, the backgrounds are smaller for 6 He than for 18 Ne,
and that both neutrino and anti-neutrino backgrounds tend to cluster at low energies. A cut
demanding that the reconstructed energy be larger than 500 MeV suppresses most of the
residual backgrounds at a modest cost for the efficiency.
Fig. 9 shows the reconstructed energy (solid line), the QE component (dotted line)
and the non-QE component (dashed line) for events passing the selection criteria. Notice
that the non-QE contamination is high, specially for neutrino events. This spoils sizably
the resolution, since the neutrino energy is reconstructed under the hypothesis that the
interaction was QE. The effect is illustrated in Fig. 10, which shows the reconstructed
versus true energy for anti-neutrinos in setup II for QE events only (top) and all events
(QE and non-QE) that pass the selection criteria (bottom). Notice the excellent correlation
between reconstructed and true energy in the case of QE events, which is, however, spoiled
by the non-QE events. We take into account this effect by computing a matrix that describes
the migrations between the true and the reconstructed neutrino energy. Migration matrices
229
Fig. 8. Reconstructed energy for signal and background in setup II (the absolute normalization is arbitrary) for
6 He (top) and 18 Ne (bottom).
230
Fig. 9. Reconstructed energy (solid line), the QE component (dotted line) and the non-QE component (dashed
line) for signal-like events (arbitrary normalization) for 6 He (top) and 18 Ne (bottom).
231
Fig. 10. Reconstructed versus true energy for anti-neutrinos in setup II for QE events on the top and for signal-like
events (QE and non-QE events) on the bottom.
232
Fig. 11. Efficiencies as a function of the true and reconstructed neutrino energies for 6 He in water in setup II.
233
Fig. 12. Efficiencies as a function of the true and reconstructed neutrino energies for 18 Ne in water for the case
of setup II.
234
have also been computed for the backgrounds. Figs. 11 and 12 show those matrices (in the
form of lego-plots) for 6 He and 18 Ne, respectively. The integrated efficiencies are quite
high ( 3050%) for background fractions below 3 103 .
One possibility to control still better the backgrounds would be to have a tunable .
In this way one could characterize the signal at a given energy reducing the background
coming from higher energy events maximally. The optimization of such a design (how
many years one should run at each ) is an interesting one that will be considered
elsewhere.
3.4. Setup III
The neutrino spectra for setup III extends up to a few GeV, well in the CC regime,
where a tracking calorimeter (possibly a massive version of Minos [28]) could offer better
performance than water. The performance of such a device (a 40 kiloton magnetized
calorimeter) for the case of the neutrino factory has been extensively studied [13,25].
However, the neutrino energy for that setup was higher (mean energy of about 25 GeV
to compare with mean energy of about 5 GeV here) and the charge of the muon had to be
measured, which is not the case here. We have not yet optimized the detector characteristics
for the high energy option of the -beam, but we expect a similar performance, with
efficiencies better than 30% and background fractions better than 104 . This will be the
subject of a forthcoming study. In this paper, we will assume these numbers. We also
assume that, as it was the case in [13], the neutrino energy can be reconstructed also for
CC events. Energy bins of 1 GeV will be considered and we discard events with neutrino
energies below 1 GeV.
4. Determination of 13 and
The simultaneous measurement of 13 and is affected by correlations [13] and the socalled intrinsic degeneracy [14], which results in either a proliferation of disconnected
regions of parameter space, where the oscillation probabilities are very similar to be
distinguished, or artificially large uncertainties in both parameters when these regions
overlap.
As has been discussed before, these degeneracies can be resolved either combining
different experiments with different E /L or matter effects, or exploiting, whenever this is
possible, the energy dependence of the signal within one experiment.
One of the main advantages of going to higher energies and longer baselines in the beam is precisely to have some significant energy resolution which allows to resolve these
degeneracies.
In Fig. 13 we compare the reach concerning CP-violation on the plane (13, ), i.e., the
range of parameters where it is possible to distinguish with a 99% CL, from 0 or 180
for the different setups and 10 years of running in each case. We assume that both ions are
bunched and accelerated simultaneously. We thus include the results from the measurement
of both polarities. The remaining oscillation parameters are fixed close to their best fit
values: m223 = 2.5 103 eV2 , 23 = 45 , m212 = 7 104 eV2 , 12 = 35 .
235
Fig. 13. Region where can be distinguished from = 0 or = 180 with a 99% CL for setup I (solid), setup II
with the UNO-type detector of 400 kton described in Section 3.1 (dashed) and with the same detector with a
factor 10 smaller mass (dashed-dotted) and setup III (dotted) with a 40 kton tracking calorimeter described in
Section 3.4. In all cases we consider 10 years of running time for both polarities (since they are accelerated
simultaneously).
The solid line corresponds to the setup I, which has been studied before [12,22]. The
dashed lines correspond to setup II for the UNO detector described in the previous section
and for a detector scaled down by a factor of 10. Surprisingly the small water Cherenkov
in setup II performs similarly to the UNO detector in setup I, while the performance of the
latter in setup II is spectacular and clearly comparable with the neutrino factory. One of the
reasons for this improvement is precisely the resolution of correlations. This can be seen in
Fig. 14, where we compare the result of the fits for setups I, II and III. While the intrinsic
degeneracies are present for the low-energy option, they tend to get resolved in the higher
one, even with the smaller detector.
Notice that the necessity to suppress backgrounds due to charge misidentification forces
a very stringent cut in the momentum of the muon when searching for wrong sign muons
at the neutrino factory [13]. This cut translates in practice in throwing away neutrino
energies below 5 GeV, thus loosing precious spectral information. In that respect, the
presence of two neutrino species in the neutrino factory is a disadvantage, compared
236
Fig. 14. 1,2,3 contours on the plane (13 , ) in the setups I, II for the 40 and 400 kton detectors and for setup III
in 10 years of running time. The input values of the parameters are indicated by a star.
with the -beam, where one has only one neutrino species and the ability to identify low
energy muons in water, separating them clearly from backgrounds (as opposed to a tracking
calorimeter, where a muon of momentum less than about 1 GeV cannot be easily separated
from the pion background).
Comparing setups II and III we realize that no improvement is gained at the highest
as regards CP violation. The reason for this is probably the fact that matter effects become
dominant at this longer baselines (see Fig. 1) and hide the genuine CP violation. Although
an optimization of the detector design has not been performed in this case, it is clear that
no gain will result with respect to setup II.
Although other systematic errors, such as the knowledge of the flux or the error on the
backgrounds and efficiencies have not been included in this study, they are very unlikely
to change the conclusion concerning the comparison of the three setups of the -beam,
since they would affect them in a similar manner. However all systematic errors should be
included in a fair comparison of the -beam and the neutrino factory, since they might be
quite different in both machines.
237
5. Determination of the sign(m223 )

The sign(m223 ) is an essential missing piece of information to determine the structure
of the neutrino mass matrix. The measurement of this quantity cannot be done from the
measurement of neutrino oscillations in vacuum, so matter effects need to be sizable. In
setup I, matter effects are too small to allow the determination of this unknown, however
this is no longer the case for the intermediate baseline setup.
In Fig. 15 we show the exclusion plot for the sign of m223 on the plane (13 , ) at
99% CL. The measurement of the sign is possible in a very significant region of parameter
space. In particular for the largest detector, it can be measured for 13 4 , simultaneously
with 13 and ! On the other hand, in setup III the sign can be measured in a larger region
of parameter space. As discussed in the previous section this comes with the price that the
sensitivity to CP violation gets spoiled.
Fig. 15. Regions where the true sign(m223 ) = +1 can be measured at 99% CL (i.e., no solution at this level
of confidence exists for the opposite sign). The lines correspond to setup II with the 400 kton water Cherenkov
(solid), the 40 kton one (dashed) and to setup III (dotted) in 10 years of running time.
238
6. Conclusions and outlook

In this paper we have studied the physics potential of a higher -beam and compared
it with that of the present design ( 100, L = 130 km). From a technical point of view,
there does not seem to be any major difficulty in increasing the of the -beam.
From the physics point of view the increase in is advantageous for three reasons:
at the first atmospheric peak (which fixes /L) the rates increase linearly with , the
increase in the baseline gives sizable matter effect, which allows to distinguish the sign of
m223 , and finally the energy dependence of the oscillation signal is easier to measure and
gives precious information to resolve correlations and degeneracies between oscillation
parameters.
From the experimental point of view, a very important advantage of the low-energy
option was the possibility to use a very massive water Cherenkov a la UNO, remains true if
the is increased to 500: the result of our realistic simulation shows that backgrounds
can be reduced to a very small level with rather high efficiencies. At even higher energies
water Cherenkovs are no longer optimal, but one could give up the increase in statistics
by decreasing the detector mass, in such a way that a different technology, such as a
calorimeter, can be used.
We have thus considered two reference higher options: 500 and a water
Cherenkov a la UNO at 730 km and 2000 with a 10 times smaller calorimeter (which
is still to be optimized) at 3000 km. Our results show that the intermediate option is
spectacularly better than the low option previously considered, both in terms of the reach
in CP violation as in the possibility to measure the neutrino mass hierarchy. The highest
option instead has an intermediate performance as regards the reach in CP violation.
This is due to the fact that the baseline is so large that matter effects become dominant and
hide to a large extent the genuine CP violation. On the other hand, the determination of the
neutrino mass hierarchy is possible in a much enlarged region of parameter space.
In this study we have compared the different options alone. No combinations with
possible super-beams or among themselves have been considered. Neither the interesting
possibility to get the silver transition e in the highest energy options has been
studied. The combination of golden and silver transitions has been shown to be extremely
powerful [19] in resolving degeneracies so this is an interesting question that will be
addressed elsewhere.
In summary, we have shown that the perspectives as regards the physics reach of the
-beam are much more promising than previously thought, to the extent that it can become
competitive with the neutrino factory.4 In the light of these results, we believe that the
high- option of a -beam design deserves careful consideration. Note in particular that a
-beam, irrespective of the , does not require a proton driver with so much power as that
required by the SPL super-beam or the neutrino factory.
Clearly a finer optimization of the (s) and baseline(s) to define an optimal roadmap for
the -beam complex requires a more detailed study (and also the better knowledge of the
atmospheric and solar parameters that will be achieved in the next generation of neutrino
4 A fair comparison between the -beam and the neutrino factory should however include a proper treatment
of all systematic errors.
239
experiments), but our results show that a in the range of O(500) with a megaton water
Cherenkov at a distance O(1000 km) will be hard to beat.
Acknowledgements
We warmly thank A. Donini and S. Rigolin for usefull discussions and also M. Lindroos
for his ideas on the feasibility of a higher energy -beam and his encouragement. This work
has been partially supported by FPA2002-00612 and FPA2001-1910/C03-02 of the CICYT
and GV00-054-1.
References
[1] Y. Fukuda, et al., Phys. Rev. Lett. 81 (1998) 1562.
[2] S. Fukuda, et al., Phys. Rev. Lett. 86 (2001) 5651;
Q.R. Ahmad, et al., Phys. Rev. Lett. 89 (2002) 011301, 011302.
[3] M. Apollonio, et al., Eur. Phys. J. C 27 (2003) 331;
K. Eguchi, et al., Phys. Rev. Lett. 90 (2003) 021802.
[4] M.H. Ahn, Phys. Rev. Lett. 90 (2003) 041801.
[5] Z. Maki, M. Nakagawa, S. Sakata, Prog. Theor. Phys. 28 (1962) 970.
[6] S. Geer, Phys. Rev. D 57 (1998) 6989;
A. De Rjula, M.B. Gavela, P. Hernndez, Nucl. Phys. B 547 (1999) 21;
For recent reviews see M. Apollonio, et al., hep-ex/0210192;
J.J. Gmez-Cadenas, D.A. Harris, Annu. Rev. Nucl. Part. Sci. 52 (2002) 253.
[7] Y. Itow, et al., Nucl. Phys. B (Proc. Suppl.) 111 (2002) 146, hep-ex/0106019.
[8] A. Para, M. Szleper, hep-ex/0110032;
D. Ayres, et al., hep-ex/0210005.
[9] CERN Working Group on Super Beams Collaboration, J.J. Gmez-Cadenas, et al., hep-ph/0105297.
[10] V. Martemyanov, et al., hep-ex/0211070;
H. Minakata, et al., hep-ph/0211111;
P. Huber, et al., hep-ph/0303232;
O. Yasuda, hep-ph/0309333.
[11] P. Zucchelli, Phys. Lett. B 532 (2002) 166.
[12] M. Mezzetto, J. Phys. G 29 (2003) 1771, 1781.
[13] A. Cervera, et al., Nucl. Phys. B 579 (2000) 17, hep-ph/0002108.
[14] J. Burguet-Castell, et al., Nucl. Phys. B 608 (2001) 301, hep-ph/0103258.
[15] H. Minakata, H. Nunokawa, JHEP 0110 (2001) 1, hep-ph/0108085.
[16] V. Barger, D. Marfatia, K. Whisnant, Phys. Rev. D 65 (2002) 073023, hep-ph/0112119.
[17] P. Huber, M. Lindner, W. Winter, Nucl. Phys. B 645 (2002) 3, hep-ph/0204352.
[18] J. Burguet-Castell, et al., Nucl. Phys. B 646 (2002) 301, hep-ph/0207080.
[19] A. Donini, D. Meloni, P. Migliozzi, Nucl. Phys. B 646 (2002) 321, hep-ph/0206034;
D. Autiero, et al., hep-ph/0305185.
[20] B. Autin, et al., CERN/PS 2002-078 (OP), in: Proceedings of Nufact02 (London 2002), J. Phys. G, in press.
For more references see also http://beta-beam.web.cern.ch/beta-beam/.
[21] B. Autin, et al, CERN/PS 2002-012.
[22] J. Bouchez, M. Lindroos, M. Mezzetto, hep-ex/0310059.
[23] M. Lindroos, talk at the Moriond Workshop on Radioactive Beams for Nuclear Physics and Neutrino
Physics, http://moriond.in2p3.fr/radio/.
[24] M. Lindroos, private communication.
[25] A. Cervera, F. Dydak, J.J. Gmez-Cadenas, Nucl. Instrum. Methods A 451 (2000) 123.
240
[26] M. Goodman, et al., in: D. Casper, C.K. Jung, C. McGrew, C. Yanagisawa (Eds.), Physics Potential and
Feasibility of UNO, SBHEP01-3, July 2001.
[27] D. Casper, Nucl. Phys. B (Proc. Suppl.) 112 (2002) 161, hep-ph/0208030.
[28] See http://www-numi.fnal.gov/.

On supersymmetric solutions of type IIB

supergravity with general fluxes
Gianguido DallAgata
Humboldt Universitt zu Berlin, Institut fr Physik, Newtonstrasse 15, D-12489 Berlin, Germany
Abstract
We propose a general spinor ansatz to find supersymmetric configurations preserving 4dimensional Poincar invariance in the context of type IIB supergravity in the presence of general
fluxes. We show how this removes the imaginary-selfduality (ISD) constraint on the 3-form flux and
present a simple example with non-vanishing (1, 2) + (0, 3) flux. To characterize the geometrical
properties of such configurations we will use the tool of SU(2) structures on the internal space,
which are naturally linked to the form of the ansatz we propose.
1. Introduction
The analysis of supersymmetric backgrounds in the presence of fluxes has a prominent
role in addressing various longstanding problems of string theory like moduli stabilization.
Vacua preserving 4-dimensional Poincar invariance are especially interesting because
they can be used either in string compactifications or in the context of the gauge/gravity
correspondence.
In this paper we will focus on N = 1 solutions of type IIB supergravity and give a
geometrical characterization of these type of backgrounds for generic configurations of
fluxes. In order to do so we are going to use the tool of group structures on the internal
manifold. This powerful method gives a systematic way to translate supersymmetry
conditions in terms of differential constraints on some structures on the internal manifold,
which then define the metric and flux. This type of analysis, first introduced in [1], has
E-mail address: dallagat@physik.hu-berlin.de (G. DallAgata).
doi:10.1016/j.nuclphysb.2004.06.037
244
G. DallAgata / Nuclear Physics B 695 (2004) 243266
been very fruitful in constructing and classifying new solutions in the context of string
theory [213] (for reviews on string theory solutions with fluxes see also [14,15]). It has
been especially emphasized in [7] that the choice of an appropriate spinor ansatz is strictly
related to the possible group structure existing on a manifold and may affect the form of
the resulting solutions.
Looking for N = 1 solutions of type IIB theory, three different types of supersymmetry
parameters were used. All these assume that there exist one globally defined spinor on
the internal manifold (with its complex conjugate + ) so that the 10-dimensional
supersymmetry parameter reads
Type A:
= a + a + ,
(1.1)
Type B:
= a ,
(1.2)
Type C:
= a + b + ,
(1.3)
with a and b complex functions, is the 4-dimensional supersymmetry parameter and

are normalized to 1. Type A ansatz was first introduced by Strominger in the context of the
heterotic theory [16], applied to type IIB in [17] and recently also considered in [18,3,5,
19,20]. Type B, named after [21], was utilized in the type IIB context in [2226]. Finally,
a spinor of the form (1.3) was proposed in [27] as a way to construct solutions interpolating
between A and B types. We name it type C, being the third different type of ansatz used so
far, through this should not be confused with the S-dual of type A which is named in the
same way in [27].
Based on the existence of a single globally defined spinor, (1.1)(1.3) imply the
existence of an SU(3) structure on the internal manifold. This structure is characterized
by an almost complex structure J and a 3-form which are naturally given in terms
of :
T
Jmn = i
mn + ,
T
mnp = +
mnp + .
(1.4)
The supersymmetry conditions following from (1.1)(1.3) using a metric and a 5-form
preserving Poincar symmetry, impose that the allowed supersymmetric 3-form fluxes have
to be only of type (2, 1) or (1, 2) with respect to the natural complex structure J . For type B
the requirement is even stronger, because only fluxes which are of type (2, 1) and primitive
with respect to J are allowed [25,26].
There are however solutions in the literature which hint to the possibility of supersymmetric N = 1 solutions of type IIB string theory in which the flux may contain also (3, 0)
and (0, 3) contributions [24,28]. For this reason one should consider a more general ansatz
for the form of the supersymmetry parameter which could provide these solutions. A general ansatz which gives the desired result is
Type D:
= a + (b+ + c+ ),
(1.5)
where a new spinor , orthogonal to , is introduced. This imposes some strong restriction
on the possible choice of internal manifold. One is indeed allowed to employ such an ansatz
only if the internal manifold has an SU(2) structure and therefore admits the existence
245
of two globally defined spinors.1 Of course, since an SU(2) structure can be embedded
in different ways inside an SU(3) structure there is no more the notion of a natural
complex structure with respect to which one defines the Hodge decomposition of forms.
There is actually a U (1)-worth of different almost complex structures, which therefore can
be parameterized by a phase. Anyway, once a choice is made, one can see that it is possible
to combine (3, 0) and/or (0, 3) fluxes with supersymmetry. Also the statements about the
integrability of such a structure will depend on the phase choice.
Let us stress here that although we look for N = 1 solutions, the configurations we
obtain may preserve more supersymmetry. Already in the case of a strict SU(3) structure,
namely an internal manifold allowing for just one globally defined spinor, supersymmetric
configurations may preserve N = 2. This happens for instance in the degenerate case
of zero fluxes, where the group structure is identified with the holonomy of the internal
manifold, which is then CalabiYau. This, of course, is even more true in the case of SU(2)
structures.
It is interesting to notice that, due to the SU(2) structure, the internal manifold will
always admit an almost product structure which allows us to discuss them as fibrations
of 2-manifolds on 4-manifolds. In simple cases this structure is even integrable, therefore
simplifying the analysis and allowing to exhibit solutions in closed form. In this paper
we will discuss the form of the SU(2) structures allowed by supersymmetry when the
type D ansatz (1.5) is used for the spinor parameter. We will show that the generic internal
manifold may be non-complex and indeed allows for (3,0) or (0,3) fluxes. As an application
we will provide a simple solution with (1, 2) + (0, 3) flux and holomorphic dilaton. This
is a warped product of Minkowski spacetime with an internal manifold which is an R2
fibration over a K3.
After this introduction, in Section 2 we will discuss in detail the spinor ansatz and
its relation to both SU(3) and SU(2) structures. Then, in Section 3, we will perform
a detailed analysis of the supersymmetry conditions reinterpreting them in terms of the
SU(2) structures. Finally, we will show how to produce simple solutions with (3, 0)
and (0, 3) fluxes in Section 4 and conclude with some comments in Section 5. We also
give an appendix with more details on conventions, notations and SU(2) structures in 6
dimensions.
2. Supersymmetry ansatz and group structures

2.1. Preliminaries
Type IIB supersymmetry transformations [29,30] read

1
i
1
1
i
M1 ...M5 FM1 ...M5 M M G GM ,
DM QM +
M =
2
480
16
8
(2.1)
1 It should be remembered that although the two spinors are orthogonal they are not independent, but, as we
will show later, they are related through the SU(2) structure tensors. This also explains why seemingly different
anstze like in [28] fall instead in our classification.
246
i M
i
(2.2)
PM G,
4
where we used the contraction G 1/6GMNP MNP . Here and in the following we will
use the conventions of [30], where the supersymmetry conditions and covariant equations
of motion for type IIB supergravity were first derived.2 The definition of the dilatonaxion
curvature PM = f 2 M B and U (1) connection QM = f 2 Im(BM B ) can be related to the
2
2
standard stringy quantities = C + ie by using B = 1+i
1i and f = 1/(1 |B| ). The
3-form field strength G = f (F3 BF3 ), where F3 = dA2 , follows from a complex 2-form
A2 which is related to the usual string NS and RR forms as A2 = g(BNS + iBRR ). The
Bianchi identities for these forms are
=
dP = 2iQ P ,
(2.3)
dG = iQ G + G P ,
(2.4)
dF5 = i G G,
(2.5)
8
together with the connection which satisfies dQ = i P P . All the spinors are complex
Weyl spinors satisfying 11 M = M , 11 = and 11 = .
As said in the introduction, we are looking for solutions which preserve 4-dimensional
Poincar invariance. This type of solutions can be used for compactifications of type IIB
supergravity as well as for the gauge/gravity correspondence. It should indeed be noticed
that AdS5 can be written as an R foliation of 4-dimensional Minkowski spacetime and
that the domain-wall solutions describing holographic renormalization group flows are
also warped products of 4-dimensional Minkowski space with R and some 5-dimensional
compact space which, together, define a non-compact 6-dimensional internal manifold. For
the 10-dimensional metric we will then take a warped product of 4-dimensional Minkowski
space and some internal 6-dimensional euclidean manifold [26]

1
dx dx + Z(y)gmn (y) dy m dy n .
ds 2 =
(2.6)
Z(y)
In order to preserve 4-dimensional Poincar invariance on the full solution we also ask that
the RR 5-form satisfies
F0123m = m h(y),
(2.7)
as well as the self-duality property F5 = F5 . The dilaton and the 3-form flux may have
non-vanishing values on the internal manifold.
Before solving the supersymmetry conditions one needs to further specify an ansatz for
the spinor parameter and therefore we will now discuss more in detail how we arrived to
the (1.5) ansatz.
2.2. Spinor ansatz
Since we are looking for solutions preserving 4-dimensional Poincar invariance, we
will use this property to perform a 4 + 6 splitting of the supersymmetry spinor as well
2 For a discussion and derivation of a covariant action see [31,32].
247
as of the 10-dimensional matrices. For generic N = 1 solutions this implies that the
10-dimensional supersymmetry parameter gets factorized

(x, y) = (x) 1 (y) + (x) 2 (y) ,
(2.8)
where is a Weyl spinor on Minkowski spacetime and 1 , 2 are generic 6-dimensional
chiral spinors on the internal manifold.
In order to explicitly solve the supersymmetry equations it is important to establish
the properties and relations between these spinors. First one should note that due to
the chirality properties of the 10- and 4-dimensional spinors 11 (x, y) = (x, y),
5 (x) = (x) and the definition of the 10-dimensional matrices in terms of the lower
dimensional ones (see Appendix A for more details), i should have a definite chirality
7 i = i . More restrictions can then follow by assuming that these 2 spinors are not
unrelated, as in [17,26,27]. Generically, a globally well-defined supersymmetric solution
implies that the aforementioned spinors are also globally defined and, accordingly, that the
structure group of the tangent bundle over the internal manifold is reduced. The group of
transformations G required to patch the tangent bundle over the manifold is not the generic
SO(6) group of a Riemannian six-manifold, but a subgroup of it, G SO(6). The spinors
which are globally defined, those which define our supersymmetry parameters, must not
transform under G and therefore are singlets under the SO(6) G decomposition. When
1 and 2 are chosen parallel, or one of the two is vanishing, then the internal manifold
displays an SU(3) structure. Decomposing SO(6) SU(4) SU(3), we see that the
spinorial representation admits one singlet 4 3 + 1. This implies that an SU(3) structure
admits just one globally defined complex chiral spinor which is to be identified with the
surviving supersymmetry parameter . Since this latter is globally defined, we can always
normalize it to 1, or extract from it the normalized spinor, say . The subscript refers to the
6-dimensional chirality 7 = . The general form of the supersymmetry parameter
for solutions with an internal manifold preserving an SU(3) structure is then
(x, y) = (x) a(y) (y) + (x) b(y)+ (y),
(2.9)
This includes all the solutions presented

where a, b are complex functions and + =
in the paper [27] and more because we allow for the norms of a and b to be unrelated
before supersymmetry is imposed. For a = b we recover the type A ansatz (1.1) and for
b = 0 we recover the B ansatz (1.2). The functions a and b need not be related, though.
The 10-dimensional supersymmetry parameter is a Weyl spinor, which can be written as
the sum of 2 real MajoranaWeyl spinors: = 1 + i2 . Having an SU(3) structure means
that we can write them as
i = fi + fi + ,
(2.10)
where fi (y) are arbitrary complex functions. This finally implies that a = f1 + if2 and
b = f1 + if2 which therefore are generically unrelated complex functions too.
The SU(3) structure can also be described [33] by an almost complex structure J
and a globally defined 3-form . These are also singlets under the SO(6) SU(3)
decomposition and arise in the product of the fundamental invariant spinors:
T
mn + ,
Jmn = i
(2.11)
248

T
mnp = +
mnp + .
(2.12)
By using these tensors in the analysis of the supersymmetry conditions, once contracted
with , one can extract quite easily the conditions
G = 0 = G.
(2.13)
This means that having an SU(3) structure on the internal space, which forces the (2.9)
spinor ansatz, one can only use 3-form fluxes which are of (2, 1) + (1, 2) type with respect
to the natural complex structure defined by the covariantly constant spinor . Any (3, 0)
or (0, 3) flux with respect to J defined by (2.11) breaks supersymmetry.
A possible wayout is to make a stronger requirement on the structure of the internal
manifold and ask for an SU(2) structure. The existence of an SU(2) structure on the
internal manifold implies the possibility of having an additional globally defined spinor ,
which may be used to modify (2.9). There are indeed 2 singlets in the decomposition of
the fundamental representation of SU(4) under SU(2). Taking again to be canonically
normalized and using the same conventions as for , we can generically write the
10-dimensional spinor ansatz as

(x, y) = (x) f1 (y) (y) + f2 (y) (y)

+ (x) f3 (y)+ (y) + f4 (y)+ (y) ,
(2.14)
with fi C generic functions to be determined. It is useful to notice however that one can
always simplify the above ansatz, by removing one of the functions by a field redefinition.
f1 +f2
has norm one and defines an
This is easily proven if one notices that
2
2
|f1 | +|f2 |
f f
SU(2) structure together with the orthogonal combination 2 2 1 2 . We remind
|f1 | +|f2 |
that for the spinor to be well-behaved everywhere on the solution, we also have to require
that the fi functions are globally defined.
We have therefore argued that the most general spinor ansatz for an SU(2) structure is
the ansatz D (1.5) given in the introduction:

(x, y) = a(y)(x) (y) + (x) b(y)+ (y) + c(y)+ (y) .
(2.15)
One could think about having even more restricting structures on the tangent bundle,
of course. The result however would be that the internal space factorizes, often leading to
solutions preserving more supersymmetries, and we will therefore not consider it further
here.
2.3. SU(3) and SU(2) structures
We have just seen how the existence of different group structures allows for different
spinor anstze which may lead to new supersymmetric solutions. Since in what follows
we are going to make an extensive use of them, we will give some more details on the
introduction of a G-structure, for G = SU(3) or G = SU(2). Some parts of what follows
are also covered in [5,11,13].
In the previous section we have learned how globally defined spinors imply a group
structure on the tangent bundle of a 6-manifold. Alternatively, one can characterize a
249
G-structure by invariant forms, i.e., tensors which are singlets in the decomposition
under G. For an SU(3) structure one needs a 2-form J and a complex 3-form , as defined
in (2.11), (2.12), which satisfy the compatibility constraints
3
J J J = i .
(2.16)
4
An SU(2) structure requires the existence of a triplet of 2-forms J i and a complex
1-form w. The 1-form must lie on an orthogonal space with respect to the one spanned
by the 2-forms
J = 0,
wJ i = 0
(2.17)
and these latter give the volume of a 4-dimensional subspace

J i J j = 0,
for i
= j,
J i J i = Vol4 .
(2.18)
One can also use the metric to raise one of the indices of the 2-forms and obtain a triplet
of almost complex structures on the base satisfying
J i J j = ij ij k J k .
(2.19)
These can also be combined again into a 2-form, a complex 3-form and a complex 1-form.
As for the SU(3) structures, one can build the SU(2) structure tensors from combinations of the invariant spinors. We could provide directly the expression analogous to (2.11)
and (2.12), but it is instructive to derive the SU(2) tensors starting from an existing SU(3)
structure and imposing the existence of some additional independent spinor. In this way it
will be clear the difference between the previous analyses of the supersymmetry conditions
and the one presented here.
With respect to SU(3) structures, the SU(2) ones are characterized by the existence of
an additional globally defined complex vector w. One can therefore construct them from
existing SU(3) structures by introducing new globally defined spinors as
1
1
= w m m + ,
+ = wm m ,
(2.20)
2
2
where it is clear that = . Assuming that the spinors and the globally defined
= 1 and wm w m = 2, one obtains more

vector w are canonically normalized, i.e.,
orthogonality relations
= 1,
= 0,
(2.21)
where we also used that w is holomorphic with respect to the almost complex structure
defined by (2.11) Jm n wn = iwm .
Combining the information coming from the definition of the SU(3) structure (2.11),
(2.12), the definition of the SU(2) spinors (2.20) and the orthogonality properties (2.21)
we can then obtain the definitions of all the invariant tensors in terms of the fundamental
spinors:
T m
+ = w m ,
+
T
mn = Kmn ,
+
mn + = iJmn ,
T
mn + = K mn ,
T mn + = 2w[m w n] iJmn ,
T
+
mnp + = mnp ,
(2.22)
250
where we have introduced the combination K J 2 + iJ 3 . Moreover, using the identities

on the 6-dimensional matrices and the spinors and reported in the appendix, it can
also be checked that J = J 1 2i w w and = K w.
In doing so, we have made a choice for the embedding of the SU(2) structure inside
the SU(3) one. This was done in a somewhat natural way because we have constructed the
SU(2) structure starting from an existing SU(3) one. However, it should be clear that given
2 globally defined spinors there is no preferred choice for the one describing the SU(3)
structure. Let us then establish how the SU(2) structures are embedded into the SU(3)
ones generically. Considering always normalized spinors, there is a U (1) degeneracy of
possible spinors defining the SU(3) structure starting from the 2 globally defined spinors
describing the SU(2) one:
= cos + sin .
(2.23)
Using this normalized spinor one can then define the SU(3) structure as given in (2.11) and
(2.12), now with replacing :

i
i
J = 1 2 sin2 J 1 w w + sin 2(K K),

2
2
= cos2 K w + sin2 K w + i sin 2J 1 w.
(2.24)
(2.25)
It is now clear that the properties of the SU(3) structure depend on the choice of the
embedding angle. Among these, the integrability of the complex structure. Although this
will not change the fact that the solution be supersymmetric or not, it may change the
physical interpretation as there are instances in which there is a natural choice of complex
structure which specifies . In the following we will consider for definiteness = 0, i.e.,
= as done previously.
3. Analysis of supersymmetry conditions

In this section we are going to interpret the supersymmetry equations as conditions on
the intrinsic torsion of the internal manifold. This means that we are going to specify some
differential constraints on the SU(2) structure tensors defined in (2.22). At the same time
we will show that there are some constraints on the 3-form fluxes and dilaton/axion, though
they will not be as restrictive as those obtained previously using the anstze (1.1)(1.3). For
this reason, we list here the general expansion of the fluxes in terms of the SU(2) structures,
so that statements on the single modules can be made more precise. The dilaton/axion can
be decomposed in 3 pieces
P = p1 w + p2 w + ,
(3.1)
where w = 0, for a containing both a (1, 0) and a (0, 1) component. The 3-form flux
is then
G = g30 K w + g21 K w + g21 J 1 w + J 1 V1 + w w V2
+ w T1 + w T2 + g12 K w + g 12 J 1 w + g03 K w,
(3.2)
251
where the flux components further satisfy

J i T1,2 = 0,
wT1,2 = 0,
wV1,2 = 0,
1,2 = 0.
wT
1,2 = 0 and wV
(3.3)
We can also combine V1 and V2 so that the primitive and non-primitive parts with respect to
(V1 2iV2).
J are explicit: J 1 V1 + w w V2 = 12 J (V1 + 2iV2 ) + 12 (J 1 + 2i w w)
We also notice that g12 , g21 refer to primitive fluxes with respect to J and g21 , g 12 to nonprimitive ones. Restriction to ISD fluxes can be made by setting
g30 = g12 = g 21 = 0,
T2 = 0,
(1 + iJ )(V1 2iV2 ) = 0 = (1 iJ )(V1 + 2iV2 ).
(3.4)
The supersymmetry equations (2.1), (2.2) are not written in this language. In order to
extract the information we want in terms of SU(2) structure conditions we should perform
a projection on the full basis of independent spinors and use the relations between the
spinor bilinears and the SU(2) structure tensors (2.22). For the case at hand, one should
m...
,
, up to 3 matrices, and the same for the
project on the full basis given by
spinors . Luckily, it is not necessary to consider all these projections, because only a
subset of them gives independent conditions. As a first fact one can notice that only the
projections along are necessary due to the relation (2.20) between and . Then,
using group theory, from the SU(4) SU(3) SU(2) decomposition one learns that the
only independent objects one can build are and m . Therefore we will just consider
these projections.
We will start analyzing first the dilatino equation (2.2) and then the gravitino
equation (2.1) for the free index in the external space and then in the internal space. The
first two sets of equations will give us conditions on the fluxes as well as determine the
warp-factor and 5-form flux in terms of the 3-form flux. Then, from the internal gravitino
we will get the proper differential constraints on the SU(2) structure.
Let us start with the analysis of the dilatino equation. Making explicit use of the D
spinor ansatz, (2.2) reads
= 0

m Pm a + + b + c

= mnp (a ) (b+ + c+ ) Gmnp .

24 Z
(3.5)
T and T give us two constraints on the

Simple contractions of this equation with
m
fluxes
2bg03 + icg 12 = 0,
2b g30 ic g21 = 0,
(3.6)
(3.7)
and equations which determine the dilaton/axion in terms of the components of the 3-form
flux as specified in (3.2)

Zc p1 = ag30 ,
(3.8)

2 Za p2 = (2cg21 + ib g 12),
(3.9)
252
a (1 + iJ )

= i b(1 + iJ ) V 1 + 2iV 2 cK V 1 2iV 2 ,
4 Z

b (1 iJ ) + c K = ia(1 iJ ) V 1 + 2iV 2 .
4 Z
(3.10)
(3.11)
From now on, when J , J 1 or K are not separated by wedge products from other forms
we understand the associated structures, i.e., J V ea Ja b Vb . It should be noted from
(3.8), (3.9), using (3.6), that the (3, 0) and (0, 3) components of the 3-form flux drive a
holomorphic and anti-holomorphic part of the dilaton, respectively. Moreover, as noted
also in [24], the appearance of a (0, 3) flux must be accompanied by a non-primitive (1, 2)
component.
As a second step one has to analyze the gravitino equation (2.1), when the free index
is on the external space M = . Using that the 4-dimensional spinor on Minkowski space
can be chosen to be constant (x) = 0, the gravitino equation reads

1
m m log Z (a ) (b+ + c+ )
8 Z

1 m
+
Z m h (a ) + (b+ + c+ )
2

1
+
mnp Gmnp a + + b + c
16 6Z
= 0,
(3.12)
where we also made explicit the connection of the warped metric, expressing everything in
terms of the quantities on the unwarped spaces.
T and T , we obtain further constraints on the fluxes and
Again, projecting along
m
determine the warp factor and 5-form flux in terms of the 3-form flux. We also decompose
the derivative of the warp factor and the 5-form flux to make explicit the irreducible SU(2)
modules
m log Z = wm + w m + m ,
m h = wm + w m + Hm ,
(3.13)
(3.14)
where w = wH = 0 and and H are real. Plugging these expressions into (3.12) and
performing the appropriate projections, not only we get the same flux constraint as (3.7),
but also some equations determining the warp factor and the 5-form flux

b
c
4Z =
(3.15)
2 g12 i g 21 ,
a
a
Z
2a
+ 4Z = g03 ,
(3.16)
Zc
(1 iJ )[ 4ZH ]

b
c
= i (1 iJ ) V 1 + 2iV 2 + i K V 1 2iV 2 ,
(3.17)
2a Z
2a Z
c
+ 4ZH )
(1 + iJ )[ + 4ZH ] + K(
b

a
= i (1 + iJ ) V 1 + 2iV 2 .
2b Z
253
(3.18)
It should be noted that restriction to ISD fluxes imposes a precise relation between the
warp-factor and 5-form flux. Indeed, using (3.4), the right-hand side of both (3.15) and
(3.17) vanishes and therefore we get that d log Z = 4Z dh. This relation, using different
conventions, was pointed out in [23,25].
The last, and more demanding, step is given by computing the differential constraints
on the SU(2) structure from the analysis of the variation of the internal gravitino m .
Since the various structures (2.22) are defined in terms of and , in order to obtain
differential constraints on them we have to extract m and m from m = 0. This
can be done by separating the contributions proportional to (x) from those proportional
to (x) and, in the case of , by taking appropriate linear combinations of them. The
resulting expressions are

i
1
m = m log a + Qm m n n log Z n m Zn h
2
8
2

b
3
1
r
Gpqr
pqr m + pq m
+
16 6
32
Z a

c
3
1
r
+
(3.19)
Gpqr
pqr m + pq m
,
16 6
32
Z a
and

i
b
1
b
m = m log c + Qm + m log a m b i Qm
2
c
c
c
1 n
n
b n
m n log Z + m Zn h + m Zn h
8
2

c
1
b
3 pq r
pqr
Gpqr
m + m
16 6
32
Z a

2
2
pqr
(b ) Gpqr a G
3 pq r
1
pqr
m + m .
(3.20)
c a
16 6
32
Z
We can now compute dw and dJ i , but from the orthogonality properties of the and
spinors we can also obtain some differential equations specifying the behavior of the
norms a, b and c appearing in (1.5) in terms of the fluxes. First of all, by recalling that
d(+ ) = 0 = d(+ ) we get differential conditions on the absolute value of the a and
c functions:

b
b
c
c
2 d log |a| = Z dh w i g21 2i g12
2 g12 4 g03
a
a
a
a
4 Z
1

c
b

(J 3i) V + 2iV 2 K(2V2 3iV1 ) + c.c. ,
a
16 Z a
(3.21)
254

b
b
2 d log |c| = Z dh + Z K + K dh
c
c

a
(J + 3i) V 1 2iV 2 + K(2V2 + 3iV1)
c
16 Z a
2
(b )
K(2V2 + 3iV1 )
ac

a
a
b
b2
w 2 g03 + 4 g12 2i g 12 4 g21 + c.c. . (3.22)
c
c
a
ca
4 Z
Then, from d[+ ] = 0 we get an extra equation for the other function
b
1
db d log a
c
c
b
b
= i Q + Z(1 iJ ) dh Z K dh
c
c

c

a
+
(J + 3i) V 1 + 2iV 2 + (J + 3i) V 1 2iV 2
a
16 Z c
1

b
b2
b 2
1
(J + 3i) V 2i V 2 + K(2V
2 + 3iV1 ) + K 2V + 3i V
a c
a
a

b
a
b
w 4 g03
2i g21 + 2i g12
a
c
a
c
4 Z

b
b
c
b2
g 12 + i g21 .
w 2 g12 + 4 g21 i 2 +
(3.23)
a
a
a
c
a c
4 Z
Let us then turn to the computation of the SU(2) torsion classes. Of course, one could
express the various components of the intrinsic torsion completely in terms of the 3-form
flux. However, we prefer to use where possible also the warp-factor, the 5-form flux and
the spinor norms a, b and c, so as to make the final expressions a bit more concise. In
some of the following formulae we will also make use of the almost complex structure
J = J 1 2i w w and of the (3, 0)-form = K w, to make more intuitive the meaning
of some of the various contributions.
The exterior derivative on the globally defined 1-form w is given by

b
1
1
dw = d log a d log c d log Z + 2iZJ dh + 2 ZK dh w
4
c
2

i
c
a

K(V1 2i V2) + K V 1 + 2iV2
a
c
8 Z
b
c
b
2 (V1 2iV2) + K(V1 + 2iV2) + 4i (1 + iJ )V2 w.
(3.24)
a
a
a
Performing a straightforward calculation, one obtains in (3.24) other terms proportional to
However, by using (3.6), (3.7) and (3.16), (3.17) one can show
the 2-forms J 1 , K and K.
+ 2bg ,
that all such terms vanish. For instance, the K term comes multiplied by icg 21
30
which is zero using (3.7). The term proportional to the 2-form K comes multiplied by
2 bc Z +
1
)
(i b g i ac g12 + 2 ab g12
4 Z a c 21
and this is again vanishing because of (3.17),
(3.18). There is also a contribution proportional to J 1 . This is 2Zi + i

a
c g03
255
c
a g12 ),
( b g
2 Z ca 30
which vanishes again using the same equations. Similar simplifications

can also be applied to the w w terms to arrive at the final form presented above.
The result in (3.24) imposes already some strong restrictions on the general form of
the internal space. The group structures determine the metric of the internal space. For an
SU(2) structure defined by J i and w the metric can always be written as
ds62 = ds42 (y) + w w,
(3.25)
where both ds42 and w will generically have legs on all possible directions of the cotangent
space. However, since from the above result we see that dw is proportional to w itself, we
can always define the coordinate differentials so that
w = eA(y) dy5 + ieB(y) dy6 ,
(3.26)
for A and B complex functions depending on all the coordinates. The 4-dimensional part
ds42 will still generically depend also on dy 5 and dy 6 . If, however, the almost product
structure ab defined by w and its dual vector as ab = (wa w b + w a wb ab ) is integrable
then the 6-dimensional metric is further reduced to a block diagonal form
ds62 =
gmn (y) dy m dy n + w w.
(3.27)
m,n=1
Unfortunately, (3.24) is not enough to show that is integrable, unless further constraints
are imposed.
We can now complete the analysis by computing the intrinsic torsion contributions
coming from the exterior differential on the triplet of 2-forms. Although we should really
compute dJ i , we will present in the following dJ and dK. It is then obvious how to
extract dJ 1 . We just show dJ because it is easier to compute and also because it gives
direct information on the integrability of the associated almost complex structure J . The
expression for its exterior differential is

1
dJ = J 2 d log |a| + d log Z Z dh
2

b
i
+
J (8g12 w 12g21 w 9V1 + 5iJ V1 2iV2 + 6J V2 )
a
16 Z
1
4iw w (1 iJ )(V1 2iV2) 16T w 16g21K w

c
1
+i
w w n K(V1 10iV2) + 8g12K w
a
2

1
J K(2V2 5iV1 ) J (16g03w + 24g12w) c.c. .
(3.28)
It is worth noting that although there are in principle non-vanishing (3, 0) or (0, 3)
components, their coefficient vanish identically using (3.6), (3.7). Finally, we can also
compute
256

1
dK = K d log c a iQ d log Z
2

3
c
i b
a
+ Z 3/2 + g03
+
g
g
3a
6 a 12 2c 03
Z

c
3
i b
i b
g 12
+ K w Z 3/2 + g12
+
g
21
2a
4 a
6 a
Z
a
(b )2
g12
+ g21
3c
3c a

c
b
b
a
1 (b )2
2i g21
+ i g12 g12
+
+ J w g12
21
a
a
a
2c
2 c a
Z
b
b
ZJ (1 iJ ) dh i ZJ K dh
c
c

(b )2
a
a
+ J 6 (1 iJ )V1 + 6 (1 iJ )V1 4i (1 iJ )V2
c
ac
c
16 Z
+ 2i
c
1 4i (b ) (1 iJ )V2 + 8 b K V1 2 b K(V1 2iV2)
V
iJ
(1
)
a
ac
a
a

ib
3 b
+ w w 3Z 3/2K dh
K(V1 + 2iV2 ) + i K(V1 + 2i V2 )
4 a
8 a
2 Z

i a
(b )2 i
3 c
(V1 2iV2 )
(V1 + 2i V2 ) +
+ (1 iJ ) i (V1 + 2i V2 )
8 a
2 c
ac 2

c
a 2 (b )2 1
w T
T +
w T 1
c
ac
Z
Z a
c 1
+
J (1 + iJ )(V1 + 2i V2).
(3.29)
4 Z a
+8
Also this expression has been simplified using the constraints on the fluxes coming from
the dilatino and external gravitino. Terms of the form J w,
which are present in the
straightforward calculation, vanish using (3.23) and (3.6), (3.7).
Having the complete form of the intrinsic torsion for the SU(2) structure, we can make
some comments on the integrability of the complex structure J . It is known that given
the SU(3) structure defined by J and = K w, the torsion classes which describe the
integrability of J are W1 and W2 , as can be read from
3
1 ) + W3 + J W4 ,
dJ = i(W1 W
4
d = W1 J J + J W2 + W5 .
(3.30)
(3.31)
It is already clear from dJ that W1 = 0, since there is no (0, 3) or (3, 0) form in (3.28).
Moreover, dw is always proportional to w and therefore the only obstruction to the
integrability of J can only be given by the (1, 2) contributions coming from dK. By
inspection of (3.29) it is easy to see that only its last line contains objects of (1, 2) type.
From there we conclude that

c 1
8i T w (1 + iJ )(V1 + 2i V2 ) ,
W2 =
4 Z a
257
(3.32)
and therefore J is an integrable complex structure only if the following additional

constraints on the 3-form flux are imposed:
T 1 = 0,
(1 iJ )(V1 2iV2 ) = 0.
(3.33)
It should be noted that these constraints are somehow orthogonal to the ISD requirement
(3.4) and imply that if such condition is imposed the only primitive flux allowed is g21 . Of
course, this is just one of the possible almost complex structures described by the SU(2)
structure. It is anyway possible to show that for any choice of phase in (2.24) W1 remains
zero. For W2 we get an involved expression depending on T1 , T2 , V1 and V2 , which does
not seem to vanish for any choice of phase. As noted also by Frey, in the c = 0 limit (3.32)
vanishes, i.e., W1 = W2 = 0 and therefore the complex structure is always integrable.
4. Solutions
In principle, from the expressions we obtained in the previous section for the SU(2)
structure, one should be able to provide classes of manifolds having the right properties
to be supersymmetric backgrounds to type IIB supergravity. Unfortunately, as it is
clear by inspection of (3.24), (3.28) and (3.29), although dw and dJ i give us some
interesting information on the general properties of the internal manifold, they are not very
illuminating on how to reproduce them in explicit examples. We leave for the future a more
detailed analysis of these conditions as well as the presentation of more examples. Here
we will anyway provide some very simple classes of manifolds which have the right group
structures. First, we will make contact with the known results obtained for the type A (1.1)
and type B ansatz (1.2). Then we will provide a new class of supersymmetric configurations
which admit also (0, 3) flux.
4.1. Known limits: type B and type A solutions
In order to make contact with the general solution using the type B ansatz provided
in [26] we should set b = c = 0. Of course, since in this case there is just one spinor on
the internal manifold, we cannot discuss anymore constraints on the SU(2) structure but
only on the SU(3) one. This means that it is not correct to use the form of P and G as
in (3.1) and (3.2). We can however assume that form locally, so that we can check our
previous conditions. Moreover one should be careful in setting b = c = 0 in the differential
conditions derived from the internal gravitino, since it would not be consistent, but one
should rather reconsider (2.1) in the appropriate conventions.
From the dilatino and external gravitino conditions one gets that
(1 + iJ ) = p2 = 0,
= 4Z,
g03 = g30 = g12 = g 21 = (V1 + 2iV2),
(1 iJ ) ( 4ZH ) = 0.
(4.1)
(4.2)
258
The first condition means that an holomorphic dilaton is allowed, whereas its antiholomorphic part should vanish P (0,1) = 0. It is also evident what mentioned in the
introduction that such ansatz does not allow for any (3, 0) or (0, 3) fluxes. Moreover, the
3-form flux is also constrained so that the non-primitive (1, 2)NP and (2, 1)NP parts are
vanishing as well. Due to reality of both the warp factor Z and the 5-form flux h, the
second equation tells us that
1
,
(4.3)
4Z
which is a mentioned feature of all ISD backgrounds. From the internal gravitino then one
gets that
d log Z = 4Z dh
2 d log |a| = Z dh
h=
|a| = Z 1/8 .
(4.4)
If we now decompose the function in front of the spinor into norm and phase, a = |a|eia ,
and use (4.4), the differential condition on the spinor which is left is
i
(4.5)
(Qm 2m a ) ,
2
which is equivalent to (2.19) of [26]. This now implies the differential constraints on the
SU(3) structure, which read
m =
dJ = 0,
d = i (Q 2da ) .
(4.6)
For consistency from (2.1) we also obtain that

G(1,2) = 0,
(4.7)
which finally leaves only a (2, 1) flux and primitive. Since Q is a U (1) connection, (4.6)
implies that the internal space is a Khler manifold with vanishing first Chern class, i.e., a
CalabiYau manifold, but equipped with a metric which is not the usual Ricci-flat one. If
we want to use the standard Ricci-flat metric, we can reabsorb the extra connection in the
phase a , by imposing

1
da = J d log 1 |B|2 .
(4.8)
4
This is possible because in (4.6) only the anti-holomorphic part of Q appears and this
latter can be integrated to i/2 log(1 |B|2 ). Once (4.8) is imposed, we obtain the usual
CalabiYau conditions, namely having a Khler form and a closed holomorphic 3-form
dJ = d = 0.
(4.9)
The final result is the known fact that to preserve supersymmetry with the B ansatz (1.2)
one needs a (2, 1)P flux and the internal manifold must be conformally CalabiYau.
It should be noted that by choosing c
= 0 and a = 0 one gets an anti-holomorphic
dilaton, instead of holomorphic, and the flux is (1, 2)P . Moreover the signs in (4.3), (4.4)
are reversed.
For what concerns the type A ansatz, we will briefly show that we can match the
conditions coming from the analysis of the common sector of all supergravity theories
259
[3,5]. For this reason we assume here that the axion is constant, which means that P = 12 d
and Q = 0, that the 5-form flux is vanishing, h = 0, and that the only non-vanishing 3-form
= G. In the spinor ansatz we have to impose that
is the NS one, which also means that G
c = 0 and b = a , which means again that we have to be very careful in taking the limit of
our analysis, since only an SU(3) structure is allowed.
The dilatino equation kills immediately the (3, 0) and (0, 3) parts of the flux as expected,
g30 = g03 = 0. Then we obtain that the non-primitive
(2, 1) + (1, 2) components of G
are related to the dilaton as G(2,1)NP+(1,2)NP = 2 Z J J P . From the external gravitino

we obtain the link between warp-factor and the same components of the flux, which
then translate into a relation between dilaton and warp-factor d log Z = d. Finally, from
the internal gravitino we obtain a relation between the norm of the spinor and again the
(2, 1)NP + (1, 2)NP parts of the flux, d log |a| = 18 d, and the differential conditions on
the group structures
dJ = J d G,
1
d = d.
2
(4.10)
(4.11)
In (4.10) the dual of the 3-form flux can be reconstructed by using the duality properties of
the various flux components as discussed in Appendix A (A.22)(A.24).
4.2. A simple supersymmetric configuration with (0, 3) flux
We now work out a simple example of a configuration which preserves N = 1
supersymmetry and has a non-vanishing (0, 3) component of the 3-form flux. To simplify
at most the framework in which we work we will assume that b = 0 and that the flux is
given only by
G = g03 K w + g12 K w.
(4.12)
Of course, the Hodge decomposition is done with respect to the complex structure (2.11).
However, with respect to the general definitions (2.25) one sees that the 3-form flux
contains a non-vanishing (0, 3) flux for any phase such that cos
= 0 and in that case
it contains a non-vanishing (3, 0) part. We will also assume that the 5-form is vanishing,
i.e., h = 0. From its Bianchi identity (2.5) we get that

= i |g03 |2 |g12 |2 = 0,
dF5 = i G G
8
16
(4.13)
and this implies that

|g03 | = |g12 |.
(4.14)
The conditions following from the dilatino equation give

p1 = p2 = 0,
P
(0,1)
= 0,
(1 + iJ ) = 0 = K
wP
= 0.
(4.15)
260
This means that the dilaton admits a non-trivial (1, 0) component on the base space, and
possibly it is holomorphic with respect to the complex coordinates defined by the natural
complex structure when this latter is integrable.
From the external components of the gravitino, one gets (using the fact that d log Z is
real)
2c
2a
g03 = g12 ,
=
Zc
Za
and H = 0 as requires compatibility with our assumptions. Therefore

2 c
a
d log Z =
g12 w + g12
w .
c
Z a
(4.16)
(4.17)
From the internal gravitino we finally get conditions on the normalizations of the spinors
defining the SU(2) structure:

1 1 c 2
2 d log |a| =
(4.18)
d log Z,
4 2 a

1 1 a 2
d log Z.
2 d log |c| =
(4.19)
4 2 c
A possible solution to such conditions is given by choosing a and c real and
a = c = Z 1/8 .
(4.20)
The differential conditions defining the torsion classes are then

1
d log Z w,
2
dJ = 0,
(4.22)
dK = iK Q,
(4.23)
dw =
(4.21)
where we used (4.19) and (4.20) and the inherited SU(3) structure satisfies dJ = 0,
d = 12 (d log Z + 2iQ). For such examples the product structure ab = (wa w b +
w a wb ab ) is integrable and Q depends only on the coordinates of the base. The total
space is therefore a direct product of a CalabiYau 2-fold and R2 . Finally, we impose the
G Bianchi identity (2.4). A solution, which ensures also the equations of motion at least
for constant dilaton, is given by
= Z 1/2 f (y5 + iy6 ),

g12 = g03
(4.24)
where f is an arbitrary holomorphic function depending only on the fiber coordinates.

From (4.22) one can see that the almost complex structure is actually integrable and
that therefore the axiondilaton P is holomorphic in the appropriate complex coordinates
defined by J .
For the case of a constant dilaton/axion the solution then reads

2
2
= Z 1/2 dx02 dx12 dx22 dx32 Z 1/2 dsK3
(y1...4 )
ds10

2
2
Z dy5 + dy6 ,
(4.25)
261

G = Z 1/2 f (y)
+ f (y)K w ,

10 = Z 1/8 + + .
(4.27)
Consistency further imposes that

2

2
f (y) dy + f (y)
d y ,
Z(y5 , y6 ) =
(4.28)
(4.26)
where y y5 + iy6 . It is worth noting that the warp factor is a function which cannot be
consistently defined on a torus and this implies that the form presented above is correct
only for a non-compact background. This agrees with the fact that without adding sources
one cannot obtain compact solutions for non-trivial fluxes [34,35].
5. Comments
The conditions on the fluxes and on the SU(2) structures of the internal manifold
given here followed simply by the analysis of the supersymmetry conditions. The Bianchi
identities and the equations of motion will further restrict the possible solutions. Under
quite general assumptions it has been shown in [25] that for compact backgrounds the
allowed 3-form fluxes are of the ISD type. So far only solutions with (2, 1) and primitive
flux have been given when also supersymmetry is required. From the analysis we have
presented, though, it should be possible to compactify on 6-manifolds also in the presence
of (0, 3) and (1, 2) non-primitive fluxes, still preserving supersymmetry.
When the internal manifold is not compact, no further restriction on the Hodge type
of the 3-form fluxes follows from the equations of motion. Holographic descriptions of
renormalization group flows are in this class of solutions and indeed, by the analysis of
[24,36,28], simple perturbations of the AdS5 S 5 metric by 3-form fluxes generically turn
on all possible types of the same flux. The flow presented in [37] is especially interesting.
There a confining N = 1 gauge theory is obtained by adding a mass perturbation to N = 4
YangMills theory. It was argued that such solution could be described on the dual side
by a type IIB configuration which interpolates between a vacuum with only D3 branes and
one where the D3 branes are polarized into D5/NS5 branes wrapped on a 2-sphere of the
internal space. The supersymmetry conditions for this flow have been studied up to second
order in the perturbation parameter [24] (see also [26,36,27]). Again, the standard A, B or
C supersymmetry Anstze are not enough to describe it and it can be argued that it must
then fall into our description. A detailed analysis of the flow is currently under investigation
[38] and it shows that the internal manifold presents an SU(2) structure and the solution of
the supersymmetry equations involves again the D type ansatz (1.5).
The possibility of having general types of fluxes compatible with supersymmetry is very
important in the context of the gauge/gravity correspondence. It has been shown [39,40,25]
that the superpotential of gauge theories dualto IIB strings on CalabiYau manifolds in the
presence of 3-form fluxes is given by W = G . From the field theory point of view
then there is no reason for W to vanish even for supersymmetric solutions. Supersymmetric
configurations are only specified by I W = 0, where I runs on the moduli of the theory.
This however cannot happen if the standard solutions of type IIB theory are used, since
262
the flux contains only (2, 1) fluxes and therefore W = 0. The framework presented here
on the other hand allows for non-vanishing (0, 3) fluxes, so that G
= 0 and therefore
it should be employed to obtain dual descriptions of those gauge theories which allow for
non-vanishing W .
Recently, in [28], Pilch and Warner have discussed a class of dielectric N = 1
solutions of type IIB supergravity which include the dual of the LeighStrassler flow. Their
approach is similar to the one presented here, though they impose from the beginning a
definite ansatz for the metric and set to zero the dilaton/axion. Having a purely holomorphic
2-form potential, it is obvious that it may have non-vanishing (3, 0) components. It is
therefore interesting to understand how their solution can fit into our discussion. Following
the explanations given in the introduction, their spinor ansatz must not be of the A, B, or
C type and indeed, as they notice, it does not fit in those schemes. It can instead be put in
the form of the type D ansatz (1.5). The projector on the 10-dimensional spinor presented
in [28] is

1
1 + i 0 1 2 3 4 cos ei sin 7 10 = ,
(5.1)
2
where the star denotes complex conjugation and we used their numbering for the
matrices. It is a simple exercise to show that (1.5) satisfies this constraint assuming that
a = A(1 + cos ), b = 0, c = Aei sin and + = 7 10 + , for a so far undetermined
function A. It can also be shown that their choice of complex structure is not the standard
one built from . We will return elsewhere on the precise embedding of the full solution
given in [28] into the torsion conditions provided in this paper as well as on other solutions
which fit into our scheme [41].
Acknowledgements
We would like to thank G.L. Cardoso, G. Curio, D. Lst and N. Prezas for enlightening
discussions. This work is partially supported by the European Communitys Human
Potential Programme under contract HPRN-CT-2000-00131 Quantum Spacetime.
Appendix A. Useful formulae
For the 10-dimensional notations and conventions we follow [30]. This means that
the 10-dimensional -matrices satisfy the Clifford algebra { M , N } = 2MN where
= diag{+ } and the duality relation is
1
A1 ...A10 Ap+1 ...A10 11 ,
(A.1)
(10 p)!
where we denote by A1 , . . . , A10 flat 10-dimensional indices. In a concrete representation
one can choose 0 anti-symmetric and imaginary, with the others symmetric and
imaginary. 11 is then symmetric and real. When performing the reduction on the solution,
the 10-dimensional matrices get split as
A1 ...Ap = (1)p(p1)/2
= 1,
a = 5 a,
11 = 5 7 ,
(A.2)
263
where we have used the flat indices on Minkowski spacetime = 0, . . . , 3 and on the
6-dimensional internal space ds62 , a = 1, . . . , 6. We are also assuming that 5 = i 0123
and then the 6-dimensional duality relation is
k(k1)
m1 ...mk
(1) 2
m1 ...m6 mk+1 ...m6 7 .
= i
(6 k)!
(A.3)
Again, in a concrete representation the 6-dimensional matrices are anti-Hermitian

( m ) = m and purely real ( m ) = m .
In the text we have assumed that the globally defined spinors defining the SU(3) and
= . This happens for instance in a concrete
SU(2) structures are chiral and satisfy
representation where

0 i
i
i
7
=
(A.4)
,
+ =
,
=
,
i 0
for = an arbitrary 4-dimensional spinor. It may be useful to have an explicit

representation of the matrices satisfying all the above properties. One possibility is for
instance given by
1
1
1
1
18
= 23
= 54
= 76
= 1,
2
41
3
17
4
31
5
25
6
12
2
= 72
3
= 42
4
= 28
5
= 43
6
= 38
2
= 36
3
= 53
4
= 64
5
= 61
6
= 65
2
= 58
3
= 68
4
= 57
5
= 78
6
= 74
(A.5)
= 1,
(A.6)
= 1,
(A.7)
= 1,
(A.8)
= 1,
(A.9)
= 1.
(A.10)
Useful rearrangements when computing the conditions deriving from supersymmetry are

i
1
1 7 Jmn mn 1 7 ,
8
16
(A.11)
and
1
1
T
mnp mnp ,

= mnp mnp .
(A.12)
48
48
We list here all the useful relations (transpose is understood where needed to build the
appropriate tensor combination) for the 1-forms:
T
=
+ +
+ m + = wm ,
+ m + = wm ,
(A.13)
= w ,
= w ,
(A.14)
2-forms:
mn + = iJmn ,
+ mn = Kmn ,
+ mn = iJmn ,
mn + = K mn ,
mn + = 2w[m w n] iJmn ,
+ mn = 2w[m w n] + iJmn ,
(A.15)
(A.16)
(A.17)
264
3-forms:
+ mnp + = mnp ,
mnp = mnp ,
mnp = 3K[mn w p] ,
mnp = 3iJ[mn w p] ,
+ mnp + = 3K [mn wp] ,

+ mnp + = 3iJ[mn wp] ,
(A.18)
(A.19)
(A.20)
4-forms:
mnpq + = + mnpq = 3J[mn Jp]q .
(A.21)
It is also interesting to point out the following duality relations

ij k abc abc = 6i ij k ,
ij k abc abc = 6iij k ,
abc mnp Kmn w p = 6iK[ab w c] ,
abc mnp K mn wp = 6i K [ab wc] ,
abc
mnp
Jmn wp = 6iJ[ab wc] ,
abc
mnp
Jmn w p = 6iJ[ab w c] .
(A.22)
(A.23)
(A.24)
These equations show that given a 3-form, the (3, 0) + (1, 2)P + (2, 1)NP parts are selfdual
whereas the (0, 3) + (2, 1)P + (1, 2)NP are anti-selfdual (here P stands for primitive with
respect to J and NP for non-primitive).
A.1. SU(2) structure and Dirac spinors
A further alternative possibility to define the SU(2) structure is by means of a single
Dirac spinor

1
1
1
m
= ( + + ) = 1 + wm .
(A.25)
2
2
2
This spinor does not have a definite chirality and is independent from . The structure
then follows from
1 = ,
wm = m
1
Jmn

1 + ,
7
= i mn ,
Kmn = mn .
(A.27)
(A.28)
w[m w n] = mn ,
(A.26)
(A.29)
(A.30)
The other combinations vanish if they cannot be obtained by the above ones by complex
conjugation.
References
[1] J.P. Gauntlett, D. Martelli, S. Pakis, D. Waldram, G-structures and wrapped NS5-branes, hep-th/0205050.
[2] S. Gurrieri, J. Louis, A. Micu, D. Waldram, Mirror symmetry in generalized CalabiYau compactifications,
Nucl. Phys. B 654 (2003) 61113, hep-th/0211102.
[3] G.L. Cardoso, G. Curio, G. DallAgata, D. Lst, P. Manousselis, G. Zoupanos, Non-Kaehler string
backgrounds and their five torsion classes, Nucl. Phys. B 652 (2003) 534, hep-th/0211118.
265
[4] J.P. Gauntlett, S. Pakis, The geometry of D = 11 Killing spinors, JHEP 0304 (2003) 039, hep-th/0212008.
[5] J.P. Gauntlett, D. Martelli, D. Waldram, Superstrings with intrinsic torsion, hep-th/0302158.
[6] P. Kaste, R. Minasian, A. Tomasiello, Supersymmetric M-theory compactifications with fluxes on sevenmanifolds and G-structures, JHEP 0307 (2003) 004, hep-th/0303127.
[7] D. Martelli, J. Sparks, G-structures, fluxes and calibrations in M-theory, Phys. Rev. D 68 (2003) 085014,
hep-th/0306225.
[8] J.P. Gauntlett, J.B. Gutowski, S. Pakis, The geometry of D = 11 null Killing spinors, JHEP 0312 (2003) 049,
hep-th/0311112.
[9] K. Behrndt, C. Jeschek, Fluxes in M-theory on 7-manifolds: G-structures and superpotential, hepth/0311119.
[10] S. Fidanza, R. Minasian, A. Tomasiello, Mirror symmetric SU(3)-structure manifolds with NS fluxes, hepth/0311122.
[11] G. DallAgata, N. Prezas, N = 1 geometries for M-theory and type IIA strings with fluxes, hep-th/0311146.
[12] K. Behrndt, M. Cvetic, General N = 1 supersymmetric flux vacua of (massive) type IIA string theory, hepth/0403049.
[13] J.P. Gauntlett, D. Martelli, J. Sparks, D. Waldram, upersymmetric AdS5 solutions of M-theory, hepth/0402153.
[14] A.R. Frey, Warped strings: self-dual flux and contemporary compactifications, PhD thesis, Department of
Physics University of California, Santa Barbara, 2003, hep-th/0308156.
[15] G. DallAgata, String compactifications with fluxes, Class. Quantum Grav. 21 (2004) S1479S1499.
[16] A. Strominger, Superstrings with torsion, Nucl. Phys. B 274 (1986) 253.
[17] G. Papadopoulos, A.A. Tseytlin, Complex geometry of conifolds and 5-brane wrapped on 2- sphere, Class.
Quantum Grav. 18 (2001) 13331354, hep-th/0012034.
[18] S. Kachru, M.B. Schulz, P.K. Tripathy, S.P. Trivedi, New supersymmetric string compactifications,
JHEP 0303 (2003) 061, hep-th/0211182.
[19] K. Becker, M. Becker, K. Dasgupta, P.S. Green, Compactifications of heterotic theory on non-Kaehler
complex manifolds: I, JHEP 0304 (2003) 052, hep-th/0301161.
[20] K. Becker, M. Becker, P.S. Green, K. Dasgupta, E. Sharpe, Compactifications of heterotic strings on nonKaehler complex manifolds: II, Nucl. Phys. B 678 (2004) 19100, hep-th/0310058.
[21] K. Becker, M. Becker, M-theory on eight-manifolds, Nucl. Phys. B 477 (1996) 155167, hep-th/9605053.
[22] A. Kehagias, New type IIB vacua and their F-theory interpretation, Phys. Lett. B 435 (1998) 337342, hepth/9805131.
[23] S.S. Gubser, Supersymmetry and F-theory realization of the deformed conifold with three-form flux, hepth/0010010.
[24] M. Grana, J. Polchinski, Supersymmetric three-form flux perturbations on AdS5 , Phys. Rev. D 63 (2001)
026001, hep-th/0009211.
[25] S.B. Giddings, S. Kachru, J. Polchinski, Hierarchies from fluxes in string compactifications, Phys. Rev. D 66
(2002) 106006, hep-th/0105097.
[26] M. Grana, J. Polchinski, Gauge/gravity duals with holomorphic dilaton, Phys. Rev. D 65 (2002) 126005,
hep-th/0106014.
[27] A.R. Frey, M. Grana, Type IIB solutions with interpolating supersymmetries, Phys. Rev. D 68 (2003)
106002, hep-th/0307142.
[28] K. Pilch, N.P. Warner, N = 1 supersymmetric solutions of IIB supergravity, hep-th/0403005.
[29] J.H. Schwarz, P.C. West, Symmetries and transformations of chiral N = 2, D = 10 supergravity, Phys. Lett.
B 126 (1983) 301.
[30] J.H. Schwarz, Covariant field equations of chiral N = 2, D = 10 supergravity, Nucl. Phys. B 226 (1983)
269.
[31] G. DallAgata, K. Lechner, D.P. Sorokin, Covariant actions for the bosonic sector of D = 10 IIB
supergravity, Class. Quantum Grav. 14 (1997) L195L198, hep-th/9707044.
[32] G. DallAgata, K. Lechner, M. Tonin, D = 10, N = I I B supergravity: Lorentz-invariant actions and duality,
JHEP 9807 (1998) 017, hep-th/9806140.
[33] S. Chiossi, S. Salamon, The intrinsic torsion of SU(3) and G2 structures, in: Differential Geometry, Valencia,
2001, World Scientific, River Edge, NJ, 2002, pp. 115133, math.DG/0202282.
266
[34] B. de Wit, D.J. Smit, N.D. Hari Dass, Residual supersymmetry of compactified d = 10 supergravity, Nucl.
Phys. B 283 (1987) 165.
[35] J.M. Maldacena, C. Nunez, Supergravity description of field theories on curved manifolds and a no go
theorem, Int. J. Mod. Phys. A 16 (2001) 822855, hep-th/0007018.
[36] O. Aharony, B. Kol, S. Yankielowicz, On exactly marginal deformations of N = 4 SYM and type IIB
supergravity on AdS5 S 5 , JHEP 0206 (2002) 039, hep-th/0205090.
[37] J. Polchinski, M.J. Strassler, The string dual of a confining four-dimensional gauge theory, hep-th/0003136.
[38] G.L. Cardoso, G. Curio, G. DallAgata, D. Lst, in preparation.
[39] S. Gukov, C. Vafa, E. Witten, CFTs from CalabiYau four-folds, Nucl. Phys. B 584 (2000) 69108, hepth/9906070.
[40] S. Gukov, Solitons, superpotentials and calibrations, Nucl. Phys. B 574 (2000) 169188, hep-th/9911011.
[41] J.F.G. Cascales, A.M. Uranga, M5-brane geometries, T-duality and fluxes, JHEP 0401 (2004) 021, hepth/0307156.

Confinement/deconfinement transition of large N

gauge theories with Nf fundamentals: nf /n finite
Howard J. Schnitzer
Martin Fisher School of Physics, Brandeis University, Waltham, MA 02454, USA
Received 22 April 2004; received in revised form 19 May 2004; accepted 25 June 2004
Abstract
We consider large N zero-coupling d-dimensional U(N) gauge theories, with Nf matter fields in
the fundamental representation on a compact spatial manifold S d1 time, with Nf /N finite. The
Gauss law constraint induces interactions among the fields, in spite of the zero-coupling. This class
of theories undergo a 3rd order deconfinement phase transition at a temperature Tc proportional to
the inverse length scale of the compact manifold. The low-temperature phase has a free-energy of
O(Nf2 ), interpreted as that of a gas of (color singlet) mesons and glueballs. The high-temperature
(deconfinement) phase has a free energy of order N 2 f (Nf /N, T ), which is interpreted as that of a
gas of gluons and of fundamental and anti-fundamental matter states. This suggests the existence of
a dual string theory, and a transition to a black hole at high temperature.
1. Introduction
There has been steady interest in trying to find string duals to large N gauge theories
at weak-coupling [1]. Although this is a challenging problem, information keeps accumulating which keeps the subject progressing [28]. Large N gauge theories are believed to
undergo a deconfinement phase-transition at sufficiently high temperature, which might
possibly be related to Hagedorn behavior in the dual string theory [9]. By considering
gauge theories on a compact space [10,11], one obtains an additional parameter R to
E-mail address: schnitzr@brandeis.edu (H.J. Schnitzer).
doi:10.1016/j.nuclphysb.2004.06.057
268
H.J. Schnitzer / Nuclear Physics B 695 (2004) 267282
vary which may be tuned to weak coupling, where R is the size of the compact space, and
is the dynamical scale of the gauge theory. Because the gauge theory is on a compact
manifold, one must impose a Gauss law constraint, which induces interactions among the
gluons and the matter multiplets. Aharony et al. [11], have provided a general framework,
which we apply to the issues of our concern.
We are particularly interested in U(N) gauge theories with Nf matter multiplets in the
fundamental representation of the gauge group, with Nf /N finite in the large N limit, as
we have previously discussed such theories [12] and proposed [7] for d = 4 a possible
infinite spin representation, bulk/boundary correspondence, and ( )1 expansion. In this
paper we show that these theories on S d1 time have two phases, separated by a thirdorder phase transition at temperature Tc . The free energy in the low-temperature phase
behaves as
F /T Nf2 f1 (T ),
0 T Tc
and in the high-energy phase as

Nf
, T , T Tc .
F /T N 2 f2
N
(1.1)
(1.2)
This is attributed to a gas of glueballs and (color singlet) mesons in the low-energy phase,
and a phase-transition to a deconfinement transition, with the free-energy that of a gas of
gluons and fundamental and anti-fundamental matter states. It is speculated that the lowtemperature phase has a string dual, with a high-temperature transition to a black hole.
In Section 2 we specialize the work of Aharony et al. [11], to the models of our interest.
Section 3 explores the phase-structure of these models. The d = 4 gauged vector models
[12] and their N = 1 supersymmetric cousins [7] are presented in Section 4 as concrete examples of the theories we are considering. Section 5 summarizes our principle findings, and
argues for the possible existence of string duals for the class of theories studied throughout
this paper.
2. U(N ) gauge theory on S d1 R at large N
Aharony et al. [11], showed that the partition function on a compact space for free U(N)
gauge theory, at large N , with free matter multiplets, subject only to Gauss law constraint
is

1
n
R n
n+1 R n
.
Z(x) = [dU ] exp
(2.1)
z x + (1) zF x R U
n B
R
R (x)
In (2.1) U is a N N matrix, R (U ) is the character for the representation R, zB
R
and zF (x) are the single-particle bosonic and fermionic partition functions for each representation, and x = e = e1/T at temperature T . (See Refs. [10,11] for details of the
derivation, and for the explicit single-particle partition functions.) It is important to note
that the Gauss law constraint induces interactions between the matter fields and the gauge
sector, even though the theory is at zero-coupling.
269
2.1. U(N) gauge theory with Nf /N finite

Consider U(N) gauge theories with Nf matter multiplets in the fundamental representation of the gauge group, with Nf /N finite as N . Our motivation for discussing
such theories is to explore the possible bulk/boundary duality for such field theories in the
weak-coupling limit. The putative bulk string theory, considered in an ( )1 expansion,
would correspond to the field theory at or near the UV fixed-point. It is speculated that
the bulk theory has stringy behavior, and thus studying the free-field theory on a compact
manifold may give additional insights for this issue.
With these considerations in mind, we restrict (2.1) to the adjoint and fundamental representations. Let zB (x) and zF (x) be the single-particle partition function for the adjoint
representation, and ZB (x) and ZF (x) that for the fundamental representation. Then (2.1)
becomes

1

Z(x) = [dU ] exp
ZB x n + (1)n+1 ZF x n tr U n + tr U +n
n
n=1

n n +n
1 n
n+1
zB x + (1) zF x tr U tr U
.
+
(2.2)
n
n=1
It is to be emphasized that the same matrix U appears for the adjoint and fundamental matter, as the Gauss law constraint has induced interactions between these two sectors even
at zero-coupling. One can write (2.2) in terms of the eigenvalues of U ; {exp ii } ( <
i ) in the standard way. That is,

Z(x) =
[di ] exp
VA (i j ) +
i=j
VF (i )
(2.3)
where [11]

1 n
VA ( ) = ln sin(/2)
zB x + (1)n+1 zF x n cos n
n
(2.4a)
n=1
= ln 2 +

1
n=1

1 zB x n (1)n+1 zF x n cos n
(2.4b)
and
VF ( ) =

2
n=1

ZB x n + (1)n+1 ZF x n cos n.
(2.5)
The second term in (2.4a), as well as (2.5) serve to bring the eigenvalues closer. Thus,
the presence of matter in the fundamental representation increases the clustering of the
eigenvalues.
270
Introduce the eigenvalue distribution ( ) proportional to the density of eigenvalues ei

of U at , with everywhere non-negative subject to the choice of normalization

d ( ) = 1.
(2.6)
Define the moment

n = d ( ) cos n,
(2.7)
where the eigenvalue distribution is assumed to be symmetric about = 0, as VA ( ) =

VA ( ) and VF ( ) = VF ( ). Then, for Nf multiplets in the fundamental representation,
the effective action for the eigenvalues becomes

N2
Nf
S ( ) =
(2.8)
(n )2 VnA (T ) + 2
n VnF (T ) ,
N
n=1
where
VnA (T ) =

1 zB x n (1)n+1 zF x n
n
(2.9)
and

n
ZB x + (1)n+1 ZF x n
n
for the adjoint and fundamental representations, respectively. That is
2

Nf (VnF )2
Z=
exp
(VnA )
n=1
F 2

Nf
Vn
N 2 A
,
[dn ]
exp
Vn n +
N
VnA
VnF (T ) =
(2.10)
(2.11)
n=1
which can be written as

2

Nf (VnF )2
N2
exp
]
exp
(n )2 VnA (T ),
Z=
[d
n
(VnA )
n=1
where
(2.12)
n=1
Nf
n = n +
N

VnF
.
VnA
(2.13)
The integral in (2.12) is analogous to that for purely adjoint matter, but with n instead
of n .
A saddle-point solution of (2.11) or (2.12) implies that

F
Nf
Vn
n =
(2.14)
N
VnA
271
or equivalently
n = 0.
(2.15)
From (2.10) we observe that

V1F (T ) < 0
(2.16)
always, and
V1A (T ) > 0,
(2.17a)
if

zB (x) + zF (x) < 1,
(2.17b)
so that n 0 if (2.17) is satisfied. One might think that the solution n = 0 is valid as
long as (2.17) is satisfied. However, (2.7) implies
|n | 1.
If we formally define TH (the Hagedorn temperature) by

zB (xT ) + zF (xT ) = 1,
(2.18)
(2.19)
then a necessary condition imposed by (2.14)(2.18) is

1 1 > 0,
(2.20)
which occurs for 0 T < TH . The Hagedorn temperature is never reached by the saddlepoint solution, as (2.20) implies that

Nf
ZB (x) + ZF (x) 1,
zB (x) + zF (x) +
(2.21)
N
so that (2.19) is not compatible with (2.21). We shall see in the next section that a thirdorder phase transition to a new phase occurs even before the upper-limit in (2.21) is
attained.
Also noteworthy is that the free-energy
F = T ln Z Nf2 ,
(2.22)
when the saddle-point solution is valid, as one observes from (2.12) with n = 0, contrary
to the low-temperature phase for models with matter only in the adjoint representation,
where F O(1).
3. Phase structure
Let us consider the phase-structure of this class of models in more detail. The solution
for the eigenvalue distribution ( ) can be obtained from the equilibrium conditions for a
272
matrix model with action

Nf
1
an n +
bn tr U n + tr U +n
S =N
n
N
(3.1)
n=1
which generalizes (5.22) of Ref. [11], where we have defined

an = zB x n + (1)n+1 zF x n
and

bn = ZB x n + (1)n+1 ZF x n .
(3.2)
(3.3)
This model, with action

S =N

1 n
cn tr U + tr U +n
2n
(3.4)
n=1
has been solved in Ref. [13], where here

Nf
cn = 2 an n +
(3.5)
bn .
N
(The notation in Ref. [13] is n instead of cn .) The method used by Aharony et al. [11], to
treat n and ( ) independently, and solve for n and ( ) self-consistently, using

n =
d ( ) cos n
(3.6)
works here as well.

There are two phases for the model under consideration, denoted by A0 and A1 in the
notation of Ref. [13]. The phase A0 is characterized by a distribution of eigenvalues which
covers the circle < without gaps, while in the phase A1 the circle has a single
gap with zero density.
3.1. The A0 phase
The eigenvalue distribution in phase A0 is not constant, but is

1
1+
A0 =
cn cos n
2
(3.7)
n=1
for a density chosen to be symmetric about = 0. Self-consistency requires

Nf
1
bn .
n = cn = an n +
2
N
That is

Nf
bn
(n )A0 =
N 1 an

Nf (VnF )
,
=
N (VnA )
(3.8)
(3.9a)
(3.9b)
273
where we have used (2.9) and (2.10). The eigenvalue distribution in phase A0 is identical
to that obtained from the saddle-point solution, i.e., (2.14), so that n = 0 in phase A0 . The
free-energy for this phase is given by (2.11) and (2.22).
3.2. The A1 phase
The A1 phase can be described by a generalization of Aharony et al. [11], Eqs. (5.23).
Consider a solution for ( ) where ( ) = 0 for 0 < < 0 , and ( ) otherwise, where
0 = at the (A0 , A1 ) phase boundary. The solution for ( ) in the A1 phase is

1
1 2
2
s sin
Qn cos n
( ) =
(3.10)
,
2
2
n=1
where s = sin(0 /2), and

Qn =
cn+ P (cos 0 )
(3.11)
=0
where P (x) is the Legendre polynomial. From Eq. (5.15) of Ref. [13], we have (where
M = 1 for the A1 phase)
Q = Q0 + 2.
(3.12)
That is (since c0 = 0)

Nf
b P1 1 2s 2 P 1 2s 2 = 1.
a +
N
(3.13)
=1
Define

G = a P1 1 2s 2 P 1 2s 2
(3.14)

D = b P1 1 2s 2 P 1 2s 2 .
(3.15)
and
Then (3.13) becomes

= 1
G
Nf
D .
N
(3.16)
=1
Define
1
B n1/2 s 2 =

0
d
0
s2

1
cos n
.
sin
2
2
2
(3.17)
274
Therefore

1 n+k1/2 2
Qk B
s + B |nk+1/2| s 2
2
k=1

Nf
b B n+k1/2 s 2 + B |nk+1/2| s 2
=
a +
N
n =
(3.18)
=0 k=1
Pk (1 2s),
where the
polynomials B n+1/2
B n+1/2 (x)zn =
n=0
(3.19)
are defined by

1
(1 z)2 + 4zx + (z 1) .
2z
(3.20)
Let
Rn = a

B n+k1/2 s 2 + B |nk+1/2| s 2 Pk 1 2s 2
(3.21)

B n+k1/2 s 2 + B |nk+1/2| s 2 Pk 1 2s 2 .
(3.22)
k=1
and
Cn = b

k=1
Thus (3.18) becomes

Nf
n =
Cn .
Rn n +
N
(3.23)
=1
That is
Nf
Cn .
R =
N
(3.24)
=1
The eigenvalue moments and angle 0 in the A1 phase are determined by (3.16) and (3.24),
with solution
Nf

(R I )1 C,
=
(3.25)
N

where C =
=1 Cn .
It is difficult to solve these equations in general, so we consider the simplified model
with an = bn = 0 for n 2, which should give a good qualitative description of the model.
In that case consider the pair of equations
1 = R11 1 +
Nf
C11
N
(3.26)
and
G1 1 = 1
Nf
D1 .
N
(3.27)
275
Explicitly

R11 = a1 2s 2 s 4 ,

C11 = b1 2s 2 s 4
and

D1 = b1 2s 2 ,

G1 = a1 2s 2 .
Combining (3.25)(3.30) we find in the A1 phase

s2
1 2 0
1 = 1
= 1 sin
,
2
2
2
(3.28)
(3.29)
(3.30)
(3.31)
(3.32)
with
(1 )A1 =
1
2
(3.33)
and
2Nf
b1 = 1
N
at the (A0 , A1 ) boundary. However, in the A0 phase

Nf
b1
(1 )A0 =
.
N 1 a1
a1 +
(3.34)
(3.35)
Since the single-particle partition functions are continuous, (3.34) holds at the boundary of
the A0 phase, and thus
1
at the boundary.
(3.36)
2
Hence 1 is continuous across the (A0 , A1 ) phase boundary, as it must. Eqs. (3.33) or
(3.36) imply that the (A0 , A1 ) phase transition occurs before the bound (2.21) is reached.
The temperature Tc at which the phase transition takes place satisfies (3.34), i.e.,

2Nf
a1 (xc ) +
(3.37)
b1 (xc ) = 1.
N
(1 )A0 =
Since (3.37) implies a1 (xc ) < 1, so that Tc < TH , as expected.

One can solve for ( ), and s = sin2 20 for the A1 phase in the simplified model. From
(3.27) and (3.30)(3.32) we obtain

Nf b1
Nf b1 2
1 1/2
2
2 0
= 1+
s = sin
(3.38)
1+
2
N a1
N a1
a1
A1
with s 2 = 1 at the (A0 , A1 ) boundary, and 0 s 2 1 throughout the A1 phase. From
(3.10), with

Nf
1
b1 = 2
Q1 = 2 a1 1 +
(3.39)
N
s
276
and Qn = 0 for n 2, one obtains

1/2

1
cos
( ) = 2 s 2 sin2
2
2
s
(3.40)
and
0
n =
d ( ) cos n.
(3.41)
For the simplified model, the free-energy in phase A0 , obtained from (2.19) and (3.9),
is
Nf2 b12
1
(F )A0 =
T
(1 a1 )
N2
(1 a1 )
bdy 4

N 2 Nf
b1 ,
=
2
N
where we have used (3.34). On the other hand, in the A1 phase

Nf
1
1
1
1
2
(F )A1 = N 2
ln
s
+
b
1
1
T
2s 2 2
2
N

4
3
s
s4
1
= N 2 ln s 2 + s 2 +
+ (1 a1 ) 1 s 2 +
,
2
4
4
4
where we used (3.32) and (3.39). Since s 1 at the boundary,

N2
N 2 Nf
1
(F )A1
(1 a1 ) =
b1 .
bdy 4
T
2
N
(3.42)
(3.43a)
(3.43b)
(3.44a)
(3.44b)
(3.45)
Therefore, the free-energy is continuous across the boundary as required. One may verify
from (3.42) and (3.44) that the phase-transition, of the GrossWitten type [14], is thirdorder. Alternately, the arguments (and Fig. 1) of Ref. [13] arrive at the same conclusion.
N
Consider (3.44b) in conjunction with (3.38). With ( Nf ) fixed for a given model, the
N
free-energy FA1 N 2 f2 ( Nf , T ), while recall that FA0 Nf2 f1 (T ), in the A0 phase, with
a smooth third-order transition at the (A0 , A1 ) boundary as indicated by (3.43), due to the
relation between a1 and b1 at the boundary.
3.3. Is there a Hagedorn transition?
Formally define the Hagedorn temperature TH by (2.19), or equivalently by
a1 (xH ) = 1,
(3.46)
277
noting from (3.37), that TH > Tc for our class of models. From (3.38), we observe that
0 < s 2 < 1 at xH . Further both (3.38) and (3.44) are non-singular at a1 = 1. Recall that
x = e1/T , so that a1 increases for T > TH , with a1 as x 1. But from (3.38)
a1 implies s 0, with (F )A1 < 0 in this limit.
From Fig. 1 of Ref. [13], with 2 = 0, corresponding to our simplified model, observe
that there is no additional phase-transition for 1 > 1. (The n in that figure is equivalent
to our cn , cf. (3.5).) One might consider 2 and 1 = 0, i.e., c2 and c1 = 0, with

Nf 2
b2 x .
c2 = 2 a2 x 2 2 +
(3.47)
N
We show that c2 (x) c1 (x) for the explicit models we shall consider. Then (7.4) of
Ref. [13] applies, and no further phase-transitions are anticipated, as is also seen in Fig. 1
of Ref. [13].
The Gaussian fluctuations to the saddle-point solution of (2.11)(2.12) provide corrections which are O(1), i.e., 1/N 2 corrections to the O(Nf2 ) free-energy in phase A0 . These
Gaussian fluctuations to the free energy are of the form
1
F1 ln V1A (T ) ln(1 a1 ),
(3.48)
T
which becomes
1
F1 ln b1
(3.49)
T
at the (A0 , A1 ) boundary. The restriction (3.37) means that these fluctuations do not diverge, contrary to models with adjoint matter only.
Therefore, there does not appear to be Hagedorn phase-transition in these models, by
which we mean a divergent partition function, and accompanying phase-transition.
3.4. High-temperature behavior
From Eqs. (5.17) and (B.12) of Aharony et al. [11], one has

zi (x) 2Ni T d1 + O T d2
(3.50)
at very high temperatures, where Ni is the number of physical polarizations of the fields
of the theory, with spacetime dimension d. The potentials VnF and VnA become strongly
attractive in the large temperature limit, and 0 0 in that limit, i.e., s 2 0. Thus, at
very high temperatures n 1 to leading order, and ( ) approaches a delta-function.
Therefore, for the class of models we are considering,

2Nf
f
F (x 1) = 2N 2 T d (d) NBA +
NB
N

2Nf
1
f
NF ,
+ 1 d1 NFA +
(3.51)
N
2
f
where NBA (NB ) and NFA (NF ) are the bosonic and fermionic degrees of freedom for the
adjoint (fundamental) fields respectively, so that free-energy behaves as N 2 for the adjoint
and NNf for the fundamental representations.
278
4. Specific models
One of our primary interests is to examine the thermodynamic structure of large N
gauged vector models in four dimensions [12], as these theories are candidates [7] for an
( )1 expansion in AdS5 , and a conjectured AdS5 /CFT correspondence. These theories
have both UV and IR fixed points, with the considerations of this paper relevant to the UV
fixed-point, while the IR fixed-points are also within the perturbative domain for appropriately chosen (Nf /N).
The single-particle partition functions for d = 4 are
x2 + x
,
(1 x)3
6x 2 + 2x 3
zV =
(1 x)3
zS =
(4.1)
(4.2)
and
zF =
8x 3/2
(1 x)3
(4.3)
for scalar, vector, and Dirac fermions, respectively. Identical single-particle partition functions apply to ZS (x), ZV (x), and ZF (x) as well.
4.1. Gauged vector model
The gauged U(N) vector model [12] has a scalar and Nf fermions in the fundamental
representation, taken in the large N limit, with Nf /N finite. (That is, the gauged vector
model is coupled to the BanksZaks model [15].) There is a window [12],

11
3 3
+ 1 Nf /N
3.6
(4.4)
2
2
for which there is an IR fixed-point, which is more restrictive than that of the BanksZaks
model.
For both the gauged vector-model, and the BanksZaks model

an (x) = zV x n
(4.5)
and

bn (x) = (1)n+1 ZF x n
(4.6)
in the large N limit. The (A0 , A1 ) phase-transition occurs when

2Nf
b1 (xc ) = 1
N
according to (3.34), or
a1 (xc ) +
zV (xc ) +
2Nf
ZF (xc ) = 1.
N
(4.7)
(4.8)
279
Choosing (Nf /N) = 5, as in the figure of Ref. [7], gives

xc 0.048.
(4.9)
Eq. (4.9) justifies the use of the simplified model of (3.2).

4.2. Supersymmetric gauged vector model
Consider N = 1 supersymmetric QCD with gauge group SU(N), Nf chiral multiplets
in the fundamental representation, Q i in the anti-fundamental representation (i, i = 1
to Nf ), and a massless chiral superfield , which is a color and flavor singlet [7]. The
chiral superfields interact by means of the superpotential
Qi

W=
Nf
i
Q Qi .
N
(4.10)
i=1
The model with = 0 was studied extensively by Seiberg [16]. For our discussion, we
restrict consideration to the non-Abelian Coulomb phase with 3N/2 < Nf < 3N , which is
the conformal window.
For this model the (A0 , A1 ) phase-transition occurs when (4.7) is satisfied. Here
a1 (x) = zV (x) + zF (x)
(4.11)
b1 (x) = 2ZS (x) + ZF (x),
(4.12)
and
as each chiral multiplet has a complex scalar and a Weyl fermion. For example, choosing
(Nf /N) = 2, the (A0 , A1 ) phase-transition takes place when
xc 0.051.
(4.13)
Once again the simplified model is justified.
5. Discussion
We have considered the thermodynamic phase structure for free U(N) gauge theory
together with Nf free matter multiplets, at large N on a compact manifold (in particular
S d1 time); where the Gauss law constraint induces interactions between the gluons and
the matter multiplets. Two phases were found, with the low-temperature phase exhibiting
a free energy which behaves as
F /T Nf2 f1 (T ),
0 T Tc .
There is a third-order phase transition to the second phase, for which

Nf
2
, T , T Tc
F /T N f2
N
(5.1)
(5.2)
280
for two different functions f1 and f2 of the indicated variables, subject to

Nf
2
2
Nf f1 (Tc ) = N f2
, Tc .
N
(5.3)
At very high temperatures, the limiting behavior of the free energy is given by (3.51), as
appropriate for a gas of deconfined states.
The low-energy phase of this class of theories is that of a gas of glueballs and (color
singlet) mesons Mab (a, b, = 1 to Nf ), since the glueball contribution gives an O(1)
contribution to the free energy, while the mesons have O(Nf2 ) degrees of freedom. The
phase-transition at Tc is a (third-order) deconfining transition, where for T > Tc one has
a gas of gluons, and fundamental and anti-fundamental matter states, since gluons have
N 2 degrees of freedom, and Nf fundamental matter states contribute O(NNf ) to the freeenergy. Our computation of the free energies in phases A0 and A1 supports this picture.
The challenge is to find string duals for weakly coupled field theories. It has been argued
that large N deconfined phases (our A1 phase) should be associated with black holes [9],
even for weakly coupled gauge theories [11]. That as, for weakly coupled gauge theories
in the high temperature phase, one may search for a bulk dual, with radius of curvature of
the order of the string scale s where

p
(5.4)
1.
s
Standard arguments [9,11] show that the entropy behaves as N 2 in the high temperature
phase.
Is there evidence of stringy behavior in the low-temperature (A0 ) phase? In Eq. (3.48)
we argue that the O(1) Gaussian fluctuations give a 1/N 2 correction to free-energy which
behaves as
a
F1 ln V1A (T ) ln(1 a1 ).
(5.5)
T
We interpret this to mean that the glueballs contribute to the density of states which goes
as
1
adj (E) eE/T H , 0 T Tc < TH
(5.6)
E
which is stringy. However, the limiting temperature is never reached, as Tc < TH ,
since there is a deconfinement phase-transition at Tc . We associate the glueball states with
closed strings. On the other hand, one should associate the gas of meson states Mab in
the low-temperature phase and the gas of fundamentals and anti-fundamentals in the high
temperature phase, with open strings.
Notice that the specific models discussed in Section 4 all have IR fixed-points, which
prevent extrapolation to arbitrary large t Hooft couplings, and thus exclude the limit of
RAdS
s , which is the case of the better understood Maldacena limit [17]. For d = 4
one might be tempted to consider a IIB model with N D3 branes and Nf D7 branes, with
Nf /N of O(1) in the large N limit, but this is not possible, as the number of allowed
D7 branes is limited. In short, a specific string or brane picture eludes us. (See however
Ref. [7] for conjectured infinite spin-representations for examples in d = 4.)
281
In conclusion, there is significant evidence that conformal theories at weak coupling

have string duals, but there is a great deal that must be done to make this more concrete.
Note added
Earlier work related to this paper is [18], where one considers a SU(N) colored, quark
gluon gas partition function for d = 3 in the large N limit, with Nf fundamentals, taking
into account the colored-single constraint. It was shown that the first-order phase transition
of the pure gluon gas changes to a third-order phase transition when Nf /N is finite, but
making use of Boltzmann statistics only. The general set-up is presented in [19]. Extensions
to a conserved U(1) baryon number current in [20] may also be of interest. We thank
Professor Skagerstam for bringing these papers to our attention.
We also acknowledge the helpful suggestions of the referee.
Acknowledgements
This research was supported in part by the DOE under grant DE-FG02-92ER40706.
I wish to thank Shiraz Minwalla, Anton Ryzhov, and especially Albion Lawrence for very
useful conversations. I am grateful to the string theory group and Physics Department of
Harvard University for their hospitality extended over a long period of time.
References
[1] E.S. Fradkin, M.A. Vasiliev, Ann. Phys. (N.Y.) 177 (1987) 63;
M.A. Vasiliev, hep-th/9910096;
M.A. Vasiliev, Phys. Lett. B 567 (2003) 139, hep-th/0304049;
E. Sezgin, P. Sundell, JHEP 0109 (2001) 036, hep-th/0105001;
E. Sezgin, P. Sundell, Nucl. Phys. B 644 (2002) 303, hep-th/0205131;
E. Sezgin, P. Sundell, Nucl. Phys. B 660 (2003) 403, Erratum;
C. Fronsdal, Phys. Rev. D 20 (1979) 848;
M. Flato, C. Fronsdal, Lett. Math. Phys. 8 (1984) 159;
S. Ferrara, C. Fronsdal, Class. Quantum Grav. 15 (1998) 2153, hep-th/9712239;
M. Gnaydin, D. Minic, Nucl. Phys. B 523 (1998) 145, hep-th/9802047;
M. Gnaydin, Nucl. Phys. B 528 (1998) 432, hep-th/9803138;
M. Gnaydin, D. Minic, M. Zagermann, Nucl. Phys. B 534 (1998) 96, hep-th/9806042;
M. Gnaydin, D. Minic, M. Zagermann, Nucl. Phys. B 538 (1999) 531, Erratum;
G. Mack, Commun. Math. Phys. 55 (1977) 1;
V.K. Dobrev, V.B. Petkova, Phys. Lett. B 162 (1985) 127;
S. Ferrara, A. Zaffaroni, Phys. Lett. B 431 (1998) 49, hep-th/9803060;
S. Ferrara, A. Zaffaroni, hep-th/9807090;
D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, Adv. Theor. Math. Phys. 3 (1999) 363, hep-th/9904017;
A. Ceresole, G. DallAgata, R. DAuria, S. Ferrara, Phys. Rev. D 61 (2000) 066001, hep-th/9905226.
282
[2] W. Muck, K.S. Viswanathan, Phys. Rev. D 60 (1999) 046003, hep-th/9903194;

L. Girardello, M. Porrati, A. Zaffaroni, Phys. Lett. B 561 (2003) 289, hep-th/0212181;
N.V. Suryanarayana, JHEP 0306 (2003) 036, hep-th/0304208;
R.G. Leigh, A.C. Petkou, JHEP 0306 (2003) 011, hep-th/0304217;
R.G. Leigh, A.C. Petkou, hep-th/0309177;
E. Sezgin, P. Sundell, hep-th/0305040.
[3] A.C. Petkou, JHEP 0303 (2003) 049, hep-th/0302063;
T. Leonhardt, A. Megiane, W. Ruhl, Phys. Lett. B 555 (2003) 271, hep-th/0211092;
T. Leonhardt, W. Ruhl, hep-th/0308111.
[4] P. Haggi-Mani, B. Sundborg, JHEP 0004 (2000) 031, hep-th/0002189;
B. Sundborg, Nucl. Phys. B (Proc. Suppl.) 16 (2001) 113, hep-th/0103247;
A. Mikhailov, hep-th/0201019;
E. Witten, talk at the John Schwartz 60th Birthday Symposium, November 2001, http://theory.caltech.edu/
jhs60/witten/1.html;
U. Lindstrm, M. Zabzine, hep-th/0305098;
G. Bonelli, hep-th/0309222.
[5] G. Gopakumar, hep-th/0308184.
[6] S.R. Das, A. Jevicki, Phys. Rev. D 68 (2003) 044011, hep-th/0304093.
[7] H.J. Schnitzer, hep-th/0310210.
[8] E. Witten, hep-th/0312171.
[9] E. Witten, Adv. Theor. Math. Phys. 2 (1998) 505, hep-th/9803131.
[10] B. Sundborg, Nucl. Phys. B 573 (2000) 349.
[11] O. Aharony, J. Marsano, S. Minwalla, K. Papadodinas, M. Van Ramsdonk, hep-th/0310285.
[12] D. Olmsted, H.J. Schnitzer, Nucl. Phys. B 512 (1998) 237, hep-th/9602069;
H. Rhedin, H.J. Schnitzer, Nucl. Phys. B 537 (1999) 516, hep-th/9804008.
[13] I. Jurkiewicz, K. Zalewski, Nucl. Phys. B 220 (1983) 167.
[14] D.J. Gross, E. Witten, Phys. Rev. D 21 (1980) 446.
[15] T. Banks, A. Zaks, Nucl. Phys. B 196 (1982) 189.
[16] N. Seiberg, Nucl. Phys. B 435 (1995) 129, hep-th/9411149.
[17] J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231;
J.M. Maldacena, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200;
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109;
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[18] B.-S. Skagerstam, Z. Phys. C 24 (1984) 97.
[19] B.-S. Skagerstam, J. Phys. A: Math. Gen. 18 (1985) 1.
[20] S.I. Azakov, P. Salmonsson, B.-S. Skagerstam, Phys. Rev. D 36 (1987) 2137.

Gauged vector models and higher-spin

representations in AdS5
Howard J. Schnitzer
Martin A. Fisher School of Physics, Brandeis University, Waltham, MA 02254, USA
Abstract
Motivated by the work of Klebanov and Polyakov [hep-th/0210114] on the relationship of the large
N O(N) vector model in three-dimensions to AdS4 and higher spin representations, we attempt to
find analogous connections for AdS5 . Since the usual O(N) vector model in four-dimensions is inconsistent, we consider the (consistent) large N gauged vector model and a N = 1 supersymmetric
analogue in four-dimensions. Both these theories have UV and IR fixed points, and are candidates
for a ( )1 expansion in AdS5 , a conjectured AdS5 /CFT correspondence and higher-spin representations in the bulk theory.
1. Introduction
The most common and productive application of the AdSd /CFT correspondence is that
of a duality between a conformal field theory (CFT) on the boundary of the AdSd , and
a supergravity approximation to a string theory [1]. This is attained in the large N limit,
with 1, which implies that the AdS radius is large compared to the string scale, and
is dual to the CFT in a strong coupling regime as well. On the other hand, one might
attempt to interpret any CFT as a theory of quantum gravity in an asymptotically AdS
spacetime, with a radius of curvature of the order of the Planck mass or string scale.
In terms of string theory, this suggests an AdS/CFT correspondence to a CFT with an
E-mail address: schnitzr@brandeis.edu (H.J. Schnitzer).
doi:10.1016/j.nuclphysb.2004.06.055
284
ultraviolet (UV) fixed point, beginning with = , together with 1/ corrections. This
would correspond to a CFT in a weak coupling regime, with the expansion around the
UV fixed point of the CFT in contrast to the usual examples of AdS/CFT. More interesting
would be examples where the CFT also had an infrared (IR) fixed point at which the theory
was also conformal, perhaps even in the perturbative domain of the CFT. Then one could
consider the renormalization group (RG) flow in the ( )1 expansion of the bulk theory
from the UVIR as well as in the boundary theory. Such a study might give new insights
into the role of string theory in the AdS/CFT correspondence.
In this context Klebanov and Polyakov (KP) [2] discussed an example for AdS4 ; the
large N limit of the O(N) vector model in three-dimensions, which has both a UV and
IR fixed point, with the theory being conformal at the IR fixed point. They emphasize the
advantage of having a CFT with matter in the fundamental representation of O(N), rather
than the adjoint, as it makes available for the bulk theory the work of Vasiliev and Fradkin
[3] and others [4,5] on higher-spin representations.
This paper is an exploratory project aimed at generalizing in an appropriate way the
work of KP to four-dimensions. The four-dimensional, large N , O(N) vector model will
not work because it is inconsistent [6], and parenthetically lacks an IR fixed point. A large
N gauged vector model coupled to fermions in the fundamental representation does provide a consistent model in four-dimensions [7,8], as do N = 1 supersymmetric analogues.
Both classes of models have both UV and IR fixed points, so are candidates for consideration of the kinds of questions raised by KP, but now in four-dimensions. The nonsupersymmetric gauged vector model [7,8] is presented in Section 2. However, it is not
known whether this theory is conformal at the IR fixed point. An N = 1 cousin of this
model is presented in Section 3, where supersymmetry ensures conformal invariance at
the IR fixed point. This theory has Seiberg duality [9], so that there are both electric and
magnetic descriptions of the theory. Which one is more useful depends on where one is in
the conformal window.
We use the model of Section 3 to revisit the issues raised by KP, now in the context of
AdS5 /CFT. In Section 4 we discuss a possible AdS5 /CFT correspondence for the scalar
currents of the theory, and then in Section 5 extend this to possible higher-spin representations. At the IR fixed point, the fundamental fields have anomalous dimensions, so that a
generalization of the representation theory of Vasiliev and Fradkin [3] and of Sezgin and
Sundell [4], though not yet available, is called for as one expects the higher-spin currents
to have definite but anomalous dimensions.
Although this work is speculative, it already points to a number of issues worthy of
further study.
2. The gauged vector model in D = 4
The O(N) vector model in four-dimensions has a long history [6,7,10]. However, in
contrast to the vector model in three-dimensions, the four-dimensional model is unsuitable
for study of the issues raised by KlebanovPolyakov [2] for (at least) two reasons; (i) the
effective potential has no lowest energy bound as , and (ii) there is an absence of
an infrared (IR) fixed point.
285
However, both these difficulties are overcome by the gauged U(N) vector model [7,8],
coupled to Nf fermions in the fundamental representation. This model can be studied in the
large N limit, with Nf /N finite. (That is, the gauged vector model is coupled to the Banks
Zaks model [11].) The restrictions of asymptotic freedom, and the reality of the coupling
constants throughout the renormalization flows places important restrictions on Nf /N . For
massless mesons, these conditions are sufficiently restrictive to imply the existence of an IR
fixed point (g , ) in both gauge and 4 couplings. This is a consistent massless theory
which is scale invariant at the IR fixed point, and is in a non-Abelian Coulomb phase. The
Lagrangian density of the massless model is [7,8]
N 1 L = | + igA |2 +
1 2
2
N
f

1
|| Tr F F + i
( i Di ).
4
(2.1)
i=1
It is important tonote that the coupling

constants and fields have been rescaled (g 2 N g 2 ;
N , N; A N A ) so that N is an overall factor of the Lagrangian, and

hence N 1 is a suitable expansion parameter. Thus in (2.1), g and are t Hooft couplings.
The fields and i transform in the fundamental representation of U(N), the constraint
field is a U(N) singlet, and D the covariant derivative. Note the absence of a Yukawa
coupling between and , as both are in the fundamental representation. Study of the
model [7,8] indicates that there is a zero-mass scalar bound state exactly at the IR fixed
point, which appears in () scattering. (This state evolves from the tachyon which is
present in the 4d vector model with g = 0 [6,10].)
The renormalized gauge and scalar (t Hooft) coupling constants g(M) and (M) depend on an arbitrary mass-scale M as a result of the renormalization process. They satisfy
the RG equations
g = M
dg
dM
(2.2a)
and
d
,
dM
where in the large N limit, with Nf /N fixed is [7,8,11,12]

Nf
4 Nf
4 g5
2
3 22
+ ,
34 13
16 g = g
3
3 N
3 (4)2
N
and
= M
= a0 2 a1 g 2 + a2 g 4 ,
(2.2b)
(2.3)
(2.4)
g2
dependent coefficients are [7,8,12]

where the
g 2
2
(4) a0 = 2 + 16
4 + ,
N
g 2
(4)2 a1 = 12 + 13 256 40 Nf
+ ,
4

N
2
g
+ .
(4)2 a2 = 6 + 13 304 64 Nf
4
(2.5)
286
Let us review the solution of the coupled RG equations (2.2)(2.5) [8], as the method will
be applicable to the next section. Since g(M) in (2.3) does not depend on , one may solve
for it first. Define
t = ln M,
(2.6a)
x(t) = g (M),
(2.6b)
then from (2.2a)(2.5)

dx
= b0 x 2 + b1 x 3 + .
dt
(2.7a)
11
34 Nf
<
< ,
13
N
2
(2.7b)
If
then g 2 (M) is asymptotically free, and there is an IR fixed point for g 2 , given by [11]
2
Nf
( 11
g
2 N )
=
(2.8)
.
N
4
13( Nf 34
)
13
To solve for the flow of , use the explicit solution for g 2 (M), and make the change of
variables
ds = x(t) dt.
Then
(2.9)

b1 1
,
x(s) = A exp(sb0 ) +
b0
where
A=
1
b1
x0 b0
(2.10)
with
x(s = 0) x0 = g 2 (s = 0).
(2.11)
The variables t and s are related by

b1
A
(2.12)
exp(sb0 ) 1 + s,
b0
b0
so that t versus s is single-valued, where s when t , with integration constants chosen so that t = 0 implies s = 0.
Define
t=
y(s) = (s)/g 2 (s).
(2.13)
Then the RG equation for becomes

dy
= a0 y y+ (s) y y (s) ,
ds
(2.14)
287
where the s-dependent coefficients a0 , a1 and a2 are given by (2.5), b0 , b1 by (2.3), (2.7a),
and

a1 b0 + b1 x(s)
a1 b0 + b1 x(s) 2 a2 1/2
> 0,
y (s) =
(2.15)
2a0
2a0
a0
where reality of the coefficients is required for all values of s. The curve y+ (s) separates the
asymptotically free from the non-asymptotic free phase. Flows for y(s) > y+ (s) grow in
the UV to y(s) + , which is not a consistent phase of the model. Flows for y(s) <
s+
y (s) evolve in the IR to y(s) , which is excluded, as negative couplings are
s
not allowed. Therefore y (s) y(s) y+ (s) is required for consistency, which describes
a non-Abelian Coulomb phase of the theory.
The phase boundary near the UV fixed point is

2
1/2

4 Nf
2 Nf
1 +
1 3
y+ () =
(2.16)
,
3 N
9 N
which gives the upper-bound of the couplings in the UV,

2
1/2

4 Nf
2 Nf
2
/g
1 +
1 3
.
3 N
9 N
(2.17)
Reality of the coupling constants in the UV requires that (2.17) be real, which when combined with (2.7b) gives

3 3
11
3.6
(2.18)
+ 1 Nf /N ,
2
2
which is more restrictive than (2.7a).
The RG flow described by (2.13)(2.15) in the consistent phase of the theory has a
UV and IR fixed point in the large N limit. As an example, the RG flow for Nf /N = 5
is shown in Fig. 1, as extracted from Ref. [8]. We emphasize that the couplings in the
figure are t Hooft couplings. Note that since Nf /N is close to the upper-bound (2.18),
these t Hooft couplings are small, justifying perturbation theory in the large N limit. The
RG flow is already present in the BanksZaks model [11], to which we have coupled the
gauged vector model. It is not known whether the IR fixed point of our model is stable
under 1/N corrections.
Notice that because a2 = 0 in (2.5) and (2.15) (which comes from purely two or more
gauge boson exchange in the scalar scattering), the requirement that couplings be real is
a non-trivial constraint on Nf /N , as seen in (2.17). This means that the lower phaseboundary is y (s) = 0, which implies that in the large N limit, the RG flow is disjoint
from the case where 0, which has RG flow along the real axis (y (s) = 0) of Fig. 1.
Thus, the g 4 contribution to (2.4) leads to the discontinuous behavior for 0.
An important issue is whether the theory at the IR fixed is conformal, and not just
scale invariant. This question is difficult to resolve, and needs additional study. Since we
are interested in the possible infinite spin representations for conformal theories, we turn in
the next section to a N = 1 supersymmetric cousin of the gauged vector model described
in this section, where dimensions of chiral operators are protected. If one could show that
288
Fig. 1. Graph of the renormalization group flow for Nf /N = 5, where ultraviolet to infrared flow progressing
from left to right. The upper and lower dashed lines, y+ and y respectively, brackets flows consistent with
asymptotic freedom and stability of the theory. The vertical dashed line at the right marks the value of the infrared
fixed point g .
the theory of this section is conformal at the IR fixed point, one could analyze it in analogy
to Sections 4 and 5.
3. Supersymmetric gauged vector model

Consider N = 1 supersymmetric QCD with gauge group SU(N), Nf flavors of quarks
in the fundamental representation, Q in the anti-fundamental representation (i, =
1 to Nf ), and in addition a massless chiral superfield which is a color and flavor singlet.
The chiral superfields interact by means of the superpotential
Qi

W=
Nf
i
Q Q .
(3.1)
i=1
The coupling in (3.1) has been scaled so that it is the t Hooft coupling, with the same
convention as Section 2. Thus the chiral superfield plays a role analogous to in (2.1),
except that here we keep it as a propagating degree of freedom. (If were to be integrated
out, one would have a N = 1 theory analogous to that studied in Section 2.) This model
with = 0 was studied extensively by Seiberg [9], while for = 0 one encounters some is-
289
sues reminiscent of those considered by Leigh and Strassler [13]. In our discussion we only
consider the non-Abelian Coulomb phase with 3N/2 < Nf < 3N , which is the conformal
window.
3.1. RG flow
The RG equations [11,21] in the N , N/Nf finite limit are determined by the
-functions (where g and are t Hooft couplings)

Nf
Nf
g3
g5
g =
(3.2)
3
4
+
6 +
16 2
N
(16 2 )2
N
and [14]
= [ + 2Q ]

2

Nf
g
= 2
2
+ ,
N
16 2
16 2
(3.3)
(3.4)
where and Q are the anomalous dimensions of and Q, respectively. Note, that due
to the N = 1 supersymmetry, only the anomalous dimensions determine , as there is
no vertex renormalization. In contrast to (2.4) and (2.5), we see that a2 0, which has
consequences for the RG flow. Asymptotic freedom of the non-Abelian Coulomb phase
requires
3N
< Nf < 3N
2
and

0<2

N
g2 .
Nf
(3.5)
A necessary condition for an IR fixed point, with = 0, is + 2Q = 0 at the fixed point.

The solution of the RG equations (3.2)(3.4) proceeds exactly as described in Section 2.
There is a non-trivial IR fixed point for g 2 , which for Nf /N = 3 , with 1, is
g2 =

8 2
+ O 2 .
3
(3.6)
If 0 < (M) < 2(N/Nf )g 2 (M), the RG flow for g 2 (M) drives the flow for (M) as in
Section 2. Here we have
= a0 2 a1 g 2
with
Nf
(4)2 a0 = 2 N + ,
(4 2 )a1 = 4 + ,
(4 2 )a2 0.
(3.7)
290
Then defining y(s) = (s)/g 2 (s) as before,

dy
= a0 y y+ (s) y y (s) = a0 y y y+ (s) ,
ds
where
a1 b0 + b1 x(s)
y+ (s) =
a0
(3.8)
and
y (s) = 0.
(3.9)
Thus, the flow is as in Fig. 1, except the lower phase boundary is y (s) = 0. Hence, here
may be smoothly turned off, in contrast to the model of Section 2. Therefore, we have
an IR fixed point g2 and , which, in analogy with the argument of Seiberg [9], should
extend throughout the range 3N/2 < Nf < 3N .
Given such an IR fixed point, one can use the superconformal algebra to relate exact
results for the dimensions D of operators, with the R-symmetry charge R. They satisfy
D 32 |R|, with D = 32 R for chiral operators. The anomaly free global symmetry is
SU(Nf ) SU(Nf ) U(1)B U(1)R ,
where the quark chiral superfields transform as1

Nf N
,
Q:
Nf , 1, 1,
Nf

Nf N
Q:
1, N f , 1,
Nf
and
:
(3.10)

2N
.
1, 1, 0,
Nf
This implies that the gauge invariant operator QQ has dimensions

Nf N
D(QQ) = 3
Nf
= 2 /3
for 1.
(3.11a)
(3.11b)
Since W at the fixed point has R = 2,

3N
Nf
= 1 + /3
D( ) =
(3.12a)
for 1.
(3.12b)
1 Although we call the chiral superfields quarks, Q = + + 2 F , so that the bosonic component behaves
a
as in the vector model.
291
Note that D( ) > 1 as is required for an interacting gauge invariant chiral operator. At the
other end of the conformal window

D(QQ)
1, as N /N 3/2.
(3.13)
f
D( ) 2,
3.2. Seiberg duality
Section 3.1 describes the theory in electric variables. Following Seiberg [9], one also
expects a dual magnetic description of the same theory, even when = 0, where again
there is an IR fixed point in these variables. The dual group is SU(N ) = SU(Nf N), with
Nf flavors of quarks q and q,
which are not elementary, but non-polynomial functions of
Q and Q, with the magnetic description more natural in the range 3N/2 Nf < 2N . In
j
addition there is a gauge singlet meson Mi which appears in the dual description, as well
as the dual of , denoted by . The assignments of the quantum numbers of the global
symmetry group are

N
N
,
q:
Nf , 1,
,
Nf N Nf

N
N
,
q:
1, Nf ,
,
Nf N Nf

Nf N
,
M:
Nf , Nf , 0, 2
Nf

Nf N
:
1, 1, 0, 2
(3.14)
.
Nf
The gauge singlets have the superpotentials

Nf

q qi
W=
N
i=1
and

W =
i
M qi q ,
N
(3.15)
where N = (Nf N) and (g)

2 , and are the t Hooft couplings in the magnetic description.
The -functions of the magnetic description, in the limit N and Nf /N fixed, are

Nf
Nf
Nf
1 g 3
2g 5
g =
(3.16)
3
+
1+
+
2 16 2
N
(16 2 )2
N
N
and

= 2
Nf
Nf N
2

2
g
+
2
16 2
16 2
(3.17)
292
From an argument due to Leigh and Strassler [13], the flows for and are proportional
to each other. Therefore the RG flow is entirely analogous to that described in Section 3.1,
2
with asymptotic freedom requiring 3N

2 < Nf < 3N and 0 < < 2(1 N/Nf )g . The RG
flow may again be visualized as in Fig. 1, with the lower phase-boundary y (s) = 0. If
Nf
3
N = 2 + , with 1, the flow in g and is in the domain of perturbation theory. Thus,

the addition of (3.1) to the theory does not break Seiberg duality.
4. AdS5 and gauged vector models

Klebanov and Polyakov (KP) [2] suggested a general relation between theories with
an infinite number of higher-spin massless gauge fields in AdSd + 1 and large N conformal theories in d-dimensions containing N -component fields with an infinite number of
conserved currents. In particular they focused on the singlet sector of the 3d O(N) vector
model, and proposed that in the large N limit, the vector model was dual to the minimal
bosonic theory in AdS4 containing massless gauge fields of even spin [2,15]. This proposal
has been generalized by extending the discussion to N = 1 supersymmetry [16], as well
as to a consistency check of the idea [17].
The study of massless higher-spin theories has evolved over many years, beginning
with the work of Fronsdal [5], and of Fradkin and Vasiliev [3]. This program has been
generalized in several different ways [18,19]. The ideas of KP involve the holographic
duals of massless higher-spin theories in AdS4 . Little is known about such theories so far,
and their representations are in turn simpler than those in AdS5 . Therefore our interest in
the relationship of gauged vector models in 4-dimensions to higher-spin massless gauge
theories is quite speculative. Nevertheless it is interesting to explore these issues, despite
our concerns.
Motivated by the work of KP, we wish to find duals to large N conformal theories
containing N component fields rather than N N matrix fields. As discussed in the introduction, the O(N) invariant vector model in d = 4 has many difficulties [6,10], as well as
the absence of an IR fixed point. For these reasons we considered the gauged vector model
in Section 2, and an N = 1 extension thereof in Section 3. Since we are not assured that
the non-supersymmetric model is conformally invariant at the IR fixed point, we will now
confine our discussion to the N = 1 example of Section 3. If it can eventually be shown
that the model in Section 2 is conformal at the IR fixed point, then that theory could be
analyzed as well by the methods of this section.
The N = 1 theory has a spin-zero current J = Qi Q , which is a flavor and color singlet.
(This is called spin-zero as its lowest component is the bosonic spin-zero current.) The
dimension of the current is

N
D(J ) = 3 1
(4.1)
.
Nf
Further in the large N limit, the anomalous dimension D(J ) is known in perturbation
theory, i.e.,
D(J ) = 2

g2
+ O g4 .
8 2
(4.2)
293
At the IR fixed point, for 1, one has

g2

+ O g4 = 2 + O 2
2
3
8
while near the UV fixed point
DIR (J ) = 2
(4.3)

g2
(4.4)
+ O g4 .
2
8
Following KP, we conjecture that the correlation functions of the singlet currents at the
conformal UV or IR fixed points may be obtained from AdS5 from an AdS/CFT prescription. The dimensions of scalar fields in AdS5 is

= 2 4 + (mL)2 ,
(4.5)
DUV (J ) = 2
where L is the radius of AdS5 . Near the UV (IR) fixed point we identify
DUV (J ) = ( )UV
and
DIR (J ) = ( )IR
(4.6)
which implies, from (4.3)(4.5), that

2

2
mL UV = 4 + g 2 /8 2 +
and
mL2

IR
2

= 4 + g2 /8 2 +
for 1. Further, from the RG flow, one has 0 g 2 g2 , so that

0 < m2 L2 IR < m2 L2 UV ,
(4.7)
(4.8)
which is consistent with the RG flow in the bulk. Suppose that the current J is dual to a
scalar field h in AdS5 [1], with action

N

S(h) =
(4.9)
d 5 x g ( h)2 + m2 h2 + ,
2
where m2 = 4/L2 + O(g 4 ), and L is the AdS5 radius. This identification required us
to choose the in Eqs. (4.5)(4.8) for the CFT, since the anomalous dimension of the
operator D(J ) is negative. (We do not have an interpretation of + in the CFT analogous
to that of KP [2].) There is another independent singlet scalar current present if is a
dynamical chiral superfield. Define
= ( )
(4.10)
whose dimension at the IR fixed point can be expressed in terms of + , (4.5), but with
a perturbative correction. At the IR fixed point of , (3.3) implies that the anomalous
dimensions satisfy
+ J = 0.
(4.11)
294
This means that
g2
DIR () = 2DIR ( ) = 2 J + = 2 +
8 2
Nf
+
N

16 2

+ .
(4.12)
Then (4.3) and (4.5) give

(+ )IR = 2 +
4 + (mL)2
g2
=2+
8 2
Hence
DIR () = (+ )IR + = (+ )IR +
Nf
N
+ .

16 2
(4.13)

+ .
Similarly, near the UV fixed point

Nf
+ .
DUV () = (+ )UV +
N
16 2
If one associates to a dual field H in AdS5 with mass , satisfying

D() = 2 + 4 + (L)2
(4.14)
(4.15)
(4.16)
then

2 = 4/L2 + O g 4 , g 2 , 2
(4.17)
and so

2 = m2 + O g 4 , g 2 , 2 .
(4.18)
As in (4.8), we also have

0 < 2 L2 IR < 2 L2 UV
consistent with RG flow in the bulk. The total action for the scalar fields h and H in AdS5
is then conjectured to be

N

S=
(4.19)
d 5 x g ( h)2 + ( H )2 + m2 h2 + 2 H 2 + ,
2
where the interaction terms should include mixing between h and H , as well as selfinteractions.
The discussion of this section has focused on the electric description of the theory. For
Nf /N = 3/2 + , with 1, the magnetic description should be more useful. Define the
scalar superfields for the magnetic description,
J = q i q
and
= .
(4.20)
295
Then

4
D(J) = 3N/Nf = 2 + O 2
3
(4.21)
and

4
D( ) = 3(1 N/Nf ) = 1 + + O 2 .
(4.22)
3
In a discussion entirely parallel to the electric description, one associates the scalar fields
with an action of the form of (4.19), involving h and H , and
h and H in AdS5 to J and ,
masses m
and ,
where

m
2 = 4/L2 + O g 4 ,

m
2 = 2 + O g 4 , g 2 , 2 .
(4.23)
If Nf /N is not at either end of the conformal window 3/2N < Nf < 3N , then a perturba is not likely to be rapidly converging, so that
tive expansion in terms of (g 2 , ) or (g 2 , )
the description in terms of scalar fields in AdS5 is likely to involve the omitted interactions in (4.19) in an essential way. Therefore, it may be difficult to give a suitable explicit
completion of (4.19) valid outside the perturbative end-points of the conformal window.
5. Infinite spin representations

The theories described in Sections 2 and 3 have a class of U(N) or SU(N), respectively,
gauge and flavor singlet conserved currents at the UV fixed point (g = = 0),
J(1 ...s ) = a D (1 ... D s ) a + fermion terms
(5.1)
for each spin s, where D is the gauge covariant derivative. The model of Section 3 has an
additional class of conserved currents of even spins only,
(1 ...s ) = (1 ... s ) + fermion terms,
(5.2)
where in (5.2) is the bosonic component of the chiral superfield . Following closely KP
[2], and earlier workers [18], one conjectures that the correlation functions of these singlet
currents in the free 4d theories can be obtained from a bulk action in AdS5 , through the
AdS/CFT property with relates the boundary values of fields with that of sources h(1 ...s )
in the dual field theory. That is

( ... )
exp d 4 x h0 1 s J(1 ...s ) = eS[h0 ] ,
(5.3)
where S[h0 ] is the action of a high-spin gauge theory in AdS5 , given in terms of the boundary values h0 of fields. (For spin zero, this action is given by (4.19).)
In order to understand the action for the bulk fields h(1 ...s ) one needs an appropriate
representation theory for an infinite tower of higher-spin massless-representations in AdS5 .
For our purposes, the work of Sezgin and Sundell (SS) [4] seems most suitable. Representations of N = 1 supermultiplets in AdS5 are classified by SU(2, 2|1), while the truncation
296
to the bosonic components of the currents (5.1) and (5.2) transform according to SO(4, 2),
whose group theory has been analyzed by SS [4].
Denote the bosonic components of the chiral superfields Q and of Section 3 as
and , respectively. At the UV fixed point, i.e., free field limit, the dimensions of and
are
() = () = 1 = 1 + j
with j = 0.
(5.4)
Thus and are a pair of doubleton representations of SO(4, 2) [4,5], denoted by

D (0, 0; 1) and D (0, 0; 1), respectively. As SS [4] show, one obtains states with even
spins s = jL + jR = 0, 2, 4, . . . from the symmetric tensor product of spin-zero doubleton
representations, and odd spins from the anti-symmetric product. The maximal compact
subgroup of SO(4, 2) is SU(2)L SU(2)R U(1)R whose weight spaces are labeled
D(jL , jR ; ), with lowest weight states |jL , jR ; , where = 32 R for chiral operators.
Then

s s

D(0, 0; 1) D(0, 0; 1) S =
(5.5)
D , ;s + 2 ,
2 2
s even

s s

D(0, 0; 1) D(0, 0; 1) A =
(5.6)
D , ;s + 2 .
2 2
s odd
We conjecture that the bosonic components of the currents (5.1) and (5.2) are classified
by
D (0, 0; 1) D (0, 0; 1)
and

D (0, 0; 1) D (0, 0; 1) S ,
(5.7)
(5.8)
respectively, with the gauge and flavor singlets being selected in (5.7). Since is real,
we only keep the symmetric product of (5.6) in constructing (5.8), while (5.7) requires
both (5.5) and (5.6). Notice that s = 0, = 2 in (5.5) which implies for the bulk states
m2 = 2 = 4/L2 , corresponding to the spin-zero currents J and of Section 3.
Sezgin and Sundell [4] build their representations from the four-component SO(4, 1)
Dirac spinor y , and its conjugate y . They then extend this representation to SU(2, 2|1)
by introducing an additional set of Grassman odd complex oscillators , forming a Clifford
algebra and an odd supercharge Q = y . This generates the supersymmetric extension
of hs(2, 2) to hs(2, 2|1) by means of the N = 1 version of the product (5.5) and (5.6). We
do not present the details here.
The above discussion is applicable to the UV fixed point, where g = = 0. Now consider the gauge singlet currents (5.1) and (5.2) for g 2 = 0, = 0 (not at either the UV or
IR fixed point). The gauge and flavor singlet bosonic current J() has a dependence on
the gauge field given by
J() = D ( D )

( )
+ g 2 A A {T , T }

( )
=
=
g2
A A + g 2 d A A (T ),
N
(5.9)
297
where g 2 is the t Hooft coupling. In (5.9) T is a generator of the gauge group SU(N) in
the fundamental representation, satisfying
[T , T ] = if T
and
1
(5.10)
+ d T .
N
This can be easily generalized to all of the currents of (5.1) where the current (5.9) and its
generalization to higher spins is defined with normal ordered fields. Therefore, the gauge
field contribution to the singlet part of the current is suppressed by a factor of 1/N . (Recall the rescaling of fields in (2.1), and in Section 3.) Thus, neglect of the gauge fields in
the gauge singlet, flavor singlet part of the currents appears to be a consistent truncation,
correct to leading order in 1/N .
There are a very large number of other operators in the CFT involving gauge field
strengths, which are gauge and flavor singlets. Among these are
{T , T } =
D 2 D s1 Fs )
tr F(
1
which have dimensions = 2 + s ( 4). Another example is

2

F
F = F
(5.11)
(5.12)
which is dual to the dilation, which has = 4 and AdS5 mass M 2 = 0. There is also a
double-trace operator
2

F J + Fermi terms,
F J = F
(5.13)
with dimensions (F 2 J ) = 6 at the UV fixed point, corresponding to M 2 = 4/L2 . There
are infinite number which generalize (5.12) or (5.13) involving additional field strengths.
Other operators are typified by

F (T ) + Fermi terms,
d F
(5.14)
with D = 6, but are not gauge singlets in the matter fields, so are omitted in our truncation to gauge, flavor singlets. Similar remarks apply to the higher-spin fields analogous to
(5.1), i.e., operators such as tr(F 2 )J(1 ...s ) , etc. Operators (5.11)(5.14) and their generalizations cannot be represented by (5.5) or (5.6). In order to include them one will need
to consider appropriate generalizations of the representation theory. This is work for the
future.
If one is not at the UV fixed point, the currents (5.1) and (5.2) are not conserved. Rather
their divergence can be expressed as a power series in g 2 and , with related to g 2 on the
RG flow. This suggest a Higgs-like mechanism for the bulk gauge fields h(1 ...s ) . For even
spins, the currents (5.1) and (5.2) will mix, as will their bulk gauge field duals. The details
of the relevant Higgs mechanism remains to be worked out, and is an unsolved problem.
At the IR fixed point the theory is superconformal. A generic superconformal primary
satisfies unitarity bounds given in [20]. The unitarity thresholds are satisfied by massless
fields, and conserved tensor fields. However, due to the anomalous dimensions at the IR
298
fixed point, this is not the case for the currents (5.1) or (5.2). Since the N = 1 currents (5.1)
and (5.2) have anomalous dimensions at the IR fixed point, they are not conserved, and are
no longer the products of fields which saturate the unitarity lower-bound. This then opens
into question the group theory of (5.5) and (5.8), as the currents are no longer the products
of doubleton representations at the IR fixed point. Perhaps a straightforward generalization
of (5.5)(5.8) for currents with anomalous dimensions is possible. This is a project for the
future.
6. Concluding remarks
We have presented the large N gauged vector model in four-dimensions, and an N = 1
supersymmetric analogue, and studied the RG flow between the UV and IR fixed points of
the theories. It was speculated that AdS5 duals to the gauge and flavor singlet currents of
the model at the UV fixed point could be characterized by the higher-spin representations
of a type proposed by Vasiliev and Fradkin [3], and studied in detail for AdS5 by Sezgin
and Sundell [4]. The higher-spin currents at the UV fixed point are conjectured to be dual
to direct products of doubleton representations of SU(2, 2|1). The infinite spin gauge invariance is broken as one moves away from the UV fixed point. The higher-spin currents
are no longer conserved, but are possibly higgsed, with their divergences proportional to a
power series in the t Hooft couplings, as appropriate to a small radius expansion of AdS5
(1/ 1). At the IR fixed point the theory is conformal, though the currents have anomalous dimensions. Therefore the representation theory will require a generalization of the
work of SS [4], which is not available as yet.
The discussion of Sections 4 and 5 is speculative, and requires more work to clarify
the status of those ideas. Further, we do not have an understanding from the point of view
of dual bulk representations of the non-singlet currents of the model. Clearly additional
attention to these issues is required, even if the overall picture presented here is correct.
Acknowledgements
We would like to thank Antal Jevicki, Albion Lawrence, and Niclas Wyllard for stimulating discussions. We have benefited from the program of the Simons Institute of SUNY,
Stony Brook during August 2003, where part of this paper was written. We wish to
acknowledge the hospitality of the string group at the Harvard University Physics Department, which has been extended to me over a long period of time. H.J.S. was supported
in part by the DOE under grant DE-FG02-92ER40706.
References
[1] J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231;
J.M. Maldacena, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200;
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109;
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
299
[2] I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 550 (2002) 213, hep-th/0210114.
[3] E.S. Fradkin, M.A. Vasiliev, Ann. Phys. (N.Y.) 177 (1987) 63;
M.A. Vasiliev, Phys. Lett. B 567 (2003) 139, hep-th/0304049.
[4] E. Sezgin, P. Sundell, JHEP 0109 (2001) 036, hep-th/0105001;
E. Sezgin, P. Sundell, Nucl. Phys. B 644 (2002) 303, hep-th/0205131;
E. Sezgin, P. Sundell, Nucl. Phys. B 660 (2003) 403, Erratum.
[5] C. Fronsdal, Phys. Rev. D 20 (1979) 848;
S. Ferrara, C. Fronsdal, Class. Quantum Grav. 15 (1998) 2153, hep-th/9712239;
M. Gnaydin, D. Minic, Nucl. Phys. B 523 (1998) 145, hep-th/9802047;
M. Gnaydin, Nucl. Phys. B 528 (1998) 432, hep-th/9803138;
M. Gnaydin, D. Minic, M. Zagermann, Nucl. Phys. B 534 (1998) 96, hep-th/9806042;
M. Gnaydin, D. Minic, M. Zagermann, Nucl. Phys. B 538 (1999) 531, Erratum.
[6] L.F. Abbott, J.S. Kang, H.J. Schnitzer, Phys. Rev. D 13 (1976) 2212.
[7] D. Olmsted, H.J. Schnitzer, Nucl. Phys. B 512 (1998) 237, hep-th/9602069.
[8] H. Rhedin, H.J. Schnitzer, Nucl. Phys. B 537 (1999) 516, hep-th/9804008.
[9] N. Seiberg, Nucl. Phys. B 435 (1995) 129, hep-th/9411149.
[10] L. Dolan, R. Jackiw, Phys. Rev. D 9 (1974) 3320;
H.J. Schnitzer, Phys. Rev. D 10 (1974) 1800;
H.J. Schnitzer, Phys. Rev. D 10 (1974) 2042;
S. Coleman, R. Jackiw, H.D. Politzer, Phys. Rev. D 10 (1974) 2491;
M. Kobayashi, T. Kugo, Prog. Theor. Phys. 54 (1975) 1537;
R.G. Root, Phys. Rev. D 10 (1974) 3322;
H.J. Schnitzer, Nucl. Phys. B 109 (1976) 297.
[11] T. Banks, A. Zaks, Nucl. Phys. B 196 (1982) 189.
[12] M. Machacek, M.T. Vaughn, Nucl. Phys. B 222 (1983) 83;
M. Machacek, M.T. Vaughn, Nucl. Phys. B 249 (1985) 70.
[13] R.G. Leigh, M.J. Strassler, Nucl. Phys. B 447 (1995) 95, hep-th/9503121.
[14] P.M. Ferreria, I. Jack, D.R.T. Jones, Phys. Lett. B 399 (1997) 258, hep-ph/9702304.
[15] S.R. Das, A. Jevicki, Phys. Rev. D 68 (2003) 044011, hep-th/0304093.
[16] W. Muck, K.S. Viswanathan, Phys. Rev. D 60 (1999) 046003, hep-th/9903194;
L. Girardello, M. Porrati, A. Zaffaroni, Phys. Lett. B 561 (2003) 289, hep-th/0212181;
N.V. Suryanarayana, JHEP 0306 (2003) 036, hep-th/0304208;
R.G. Leigh, A.C. Petkou, JHEP 0306 (2003) 011, hep-th/0304217;
R.G. Leigh, A.C. Petkou, hep-th/0309177;
E. Sezgin, P. Sundell, hep-th/0305040.
[17] A.C. Petkou, JHEP 0303 (2003) 049, hep-th/0302063;
T. Leonhardt, A. Megiane, W. Ruhl, Phys. Lett. B 555 (2003) 271, hep-th/0211092;
T. Leonhardt, W. Ruhl, hep-th/0308111.
[18] P. Haggi-Moni, B. Sundborg, JHEP 0004 (2000) 031, hep-th/0002189;
B. Sundborg, Nucl. Phys. B (Proc. Suppl.) 16 (2001) 113, hep-th/0103247;
A. Mikhailov, hep-th/0201019;
E. Witten, talk at the John Schwartz 60th Birthday Symposium, November 2001, http://theory.caltech.edu/
jhs60/witten/1.html;
U. Lindstrm, M. Zabzine, hep-th/0305098;
G. Bonelli, hep-th/0309222.
[19] G. Gopakumar, hep-th/0308184.
[20] G. Mack, Commun. Math. Phys. 55 (1977) 1;
V.K. Dobrev, V.B. Petkova, Phys. Lett. B 162 (1985) 127;
300
S. Ferrara, A. Zaffaroni, Phys. Lett. B 431 (1998) 49, hep-th/9803060;

S. Ferrara, A. Zaffaroni, hep-th/9807090;
D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, Adv. Theor. Math. Phys. 3 (1999) 363, hep-th/9904017;
A. Ceresole, G. DallAgata, R. DAuria, S. Ferrara, Phys. Rev. D 61 (2000) 066001, hep-th/9905226.
[21] V. Novikov, M. Shifman, A. Vainshtein, V. Zakharov, Nucl. Phys. B 229 (1983) 381;
M. Shifman, A. Vainshtein, Nucl. Phys. B 277 (1986) 456;
M. Shifman, A. Vainshtein, Nucl. Phys. B 359 (1991) 571.

Brane-induced gravity in warped backgrounds

and the absence of the radion
Mikhail N. Smolyakov
Physics Department, Moscow State University, Vorobevy Gory, 119992 Moscow, Russia
Abstract
The RandallSundrum model with brane-localized curvature terms is considered. It is shown that
this model possesses some interesting features, in particular, the radion field is absent in it. Although
there is no modification of gravity at long distances, the model predicts deviations from Newtons law
at short distances. This effect can be observed in the experiments for testing gravity at sub-millimeter
scales.
PACS: 04.50.+h
Keywords: KaluzaKlein theories; Branes; Induced gravity
1. Introduction
Models with brane-localized curvature terms have been widely discussed in the literature during the last few years. In paper [1] it was argued that matter on the brane can
induce a brane-localized curvature term via the quantum corrections, which appears in the
low-energy effective action. An interesting feature of this model is a modification of gravity at ultra-large scales, which can be very interesting from the cosmological point of view.
But later it was shown [2,3] that there exists a strong coupling effect in this model, which
makes it unacceptable. There were attempts to merge the DGP-proposal and models with
E-mail address: smolyakov@theory.sinp.msu.ru (M.N. Smolyakov).
doi:10.1016/j.nuclphysb.2004.06.048
302
M.N. Smolyakov / Nuclear Physics B 695 (2004) 301312
warped backgrounds to get a long-distance modification of gravity. But it turns out that
such models must be rejected for some reasons (see, for example, [2,4,5]). For example,
the RandallSundrum model with brane-localized curvature terms admits a long-distance
modification of gravity, but in this case either the radion, the graviton or both fields become
ghosts.
It would be interesting to consider such models from another point of view. A modification of gravity at large distances is not the only interesting effect, which can arise in the
models. For example, in papers [6,7] the spectrum of KaluzaKlein gravitons in the RS
background with brane-localized terms for different values of parameters was studied and
some experimental constraints were found (for example, for collider experiments), but the
radion field was not taken into account. As it was noted above, this field plays an important
role in the spectrum of gravitational fluctuations and its existence can change some parameters of the model considerably to make it acceptable from the phenomenological point
of view. One can recall the original RandallSundrum model with two branes [8], in which
it is necessary to stabilize the size of extra dimension and to make the radion field massive
(see [9,10]), for example, with the help of the GoldbergerWise mechanism [11].
In the present paper we study a model with brane-localized curvature terms, which is
based on the RandallSundrum solution for the background metric. We will show that, with
appropriate parameters, the model reproduces 4-dimensional gravity on the brane, does not
contradict the known experimental data and provides some interesting consequences.
2. The setup
Let us choose the action of the model in the following form
S = Sg + S1 + S2 ,
where Sg , S1 and S2

1
Sg =
16 G
E

1
S1 =
16 G
E

2
S2 =
16 G
(1)
are given by
(R ) g d 4 x dy,

g(
R 1 )(y) d 4 x dy,

g(
R 2 )(y R) d 4 x dy.
(2)
Here g is the induced metric on the branes and the subscripts 1 and 2 label the branes.
The model possesses the usual Z2 orbifold symmetry. We also note that the signature of
the metric gMN is chosen to be (, +, +, +, +). Obviously, the model admits the Randall
Sundrum solution for the metric, which has the form
ds 2 = MN dx M dx N = dx dx + dy 2 ,
(3)
where =
is the Minkowski metric and the function (y) = k|y| in
the interval R y R. The parameter k is positive and has the dimension of mass, the
e2 (y) ,
303
parameters and 1,2 , 1,2 are related to it as follows:

= k1 1 = k2 2 = 12k 2.
(4)
The function has the properties

4 = k sign(y),

42 = 2k (y) (y R) 2k .
(5)
The parameters 1 and 2 are not specified by the solution, and their possible values
will be discussed below.

where G
is the five-dimensional gravitational constant, and
We denote = 16 G,
parameterize the metric gMN as
gMN = MN + h
MN ,
(6)
hMN being the metric fluctuations. In papers [9,10] the second variation Lagrangian for
the fluctuations of metric in the RandallSundrum model was obtained. In the case under
consideration the presence of the brane-localized curvature terms changes this Lagrangian,
and the addition can be easily calculated. But even with this addition the corresponding
action is invariant under the gauge transformations

hMN (x, y) = hMN (x, y) M N (x, y) + N M (x, y) ,
(7)
where M is the covariant derivative with respect to the background metric MN , and the
functions N (x, y) satisfy the orbifold symmetry conditions
(x, y) = (x, y),
4 (x, y) = 4 (x, y).
(8)
With the help of these gauge transformations we can impose the gauge
h4 = 0,
h44 = h44 (x) (x),
(9)
which will be called the unitary gauge (see [9]). We would like to emphasize once again
that the branes remain straight in this gauge, i.e., we do not use the bent-brane formulation,
which allegedly destroys the structure of the model (this problem was discussed in [12]).
First, let us consider the case, where there is no matter on the branes. In this case the
equations of motion for different components of the metric fluctuations in the unitary gauge
take the form:
(1) -component

1
1
1
h h h + 42 h 2k 2 h + h +
2
2
2

1
+ h h 4 h 44 4 h + 12k
2

i
+ [2kh 3k ] + i h h h + h
2

+ h h = 0,
(10)
where i = 1, 2, 1 = (y), 2 = (y R);
304
(2) 4-component

4 h h 34 = 0,
(11)
which plays the role of a constraint;

(3) 44-component
3
1
h h 4 4 h + 6k 2 = 0,
2
2
where h = h .
(12)
In what follows, we will also use an auxiliary equation, which is obtained by multiplying
the equation for 44-component by 2 and subtracting it from the contracted equation for component. This equation contains h and only and has the form

2
42 h + 24 4 h 8k 2 + 8k + i i h h = 0. (13)
3
Eq. (4) suggests that there exist preferred values of the parameters i . Namely, if we
choose these parameters to be
1
1
2 = ,
1 = ,
(14)
k
k
the values of the cosmological constants on the branes i coincide with the cosmological
constant in the bulk . For our choice of the parameters brane 1 has a positive energy density, whereas brane 2 has a negative one. We note that one does not need to worry about the
negative sign of the parameter 1 : we will see that the model is stable and does not contain tachyons or ghosts. In fact, our choice of i does not introduce any new dimensional
parameter and does not contradict the naturalness condition.
An interesting observation is that, with conditions (14), Eqs. (10)(13) possess an additional symmetry under the transformations

1
1
(x),
h (x, y) h (x, y) + (x) + 2 +
(15)
2k
2
(x) (x) + (x),
(16)
which do not belong to the gauge transformations (7). It is evident that with the help of
these transformations we can impose the condition (x) 0. At the first glance this symmetry seems to be rather strange. But let us have a look at Eq. (12). It implies that the
second variation Lagrangian of the theory does not contain the kinetic term for the field
h44 = (x) (even in the original RandallSundrum model). It means that the radion field
can be regarded as an auxiliary field (recall the supersymmetry). Therefore, there is no
contradiction that for some values of parameters of the model this field can be totally eliminated from the theory. It should be noted that the symmetry (15) and (16) of the linearized
equations of motion can correspond to some general symmetry of the action (1), which
is not evident at first sight (but this is not necessarily so). As we will see later, if there is
matter on the brane, there appears a scalar field due to the existence of the extra dimension,
which cannot be identified with the 44-component of the metric fluctuations.
305
It should be noted, that the elimination of the radion in the RandallSundrum model with
brane-localized curvature terms was discussed earlier, for example, in [2,5] (it is evident
that since there exists some range of parameters for which the coefficient in front of the
kinetic term for the radion could be either positive or negative, there exist some values
of the parameters for which the radion is absent at all). But in this paper the equations
of motion for the model with warped background are treated much more thoroughly, a
convenient gauge is used and these equation are solved exactly.
Thus, we can consider the equations of motion without the radion. With the help of the
regularization
1

1
4 42 = 4 (4 )2 = 4 k 2 = 0,
2
2
we get from Eqs. (12) and (13)
4 h = const e2 .
(17)
(18)
Let us consider Fourier expansion of all terms of Eq. (18) with respect to coordinate y.
Since the term with the derivative 4 has no zero mode, this equation implies that

4 e2 h = 0,
(19)
where h = h . The residual gauge transformations are sufficient to impose the
transverse-traceless gauge on the field h (see [9])
h = 0,
h = 0.
(20)
Thus, the -equation takes the form (2 = )

1 2
1 2
e
2h + 42 h 2k 2 h + 2k h
2h = 0.
e
2
2k
(21)
It is not difficult to solve this equation. The zero mode has the form
h0 = (x)e2 ,
(22)
whereas the massive modes have the form

m
m
m
m
2b
(x) = m2 b
(x),
hm
= b (x) (y),

m
m
e
e
m (y) = AJ2
+ BN2
,
k
k
(23)
J2 (t) and N2 (t) being the Bessel and Neumann functions.

The term with -functions can be taken into account by imposing the boundary condition

m
m
AJ0
e
e
+ BN0
=0
k
k
306
at y = 0 and y = R. The first boundary condition can be satisfied by an appropriate choice

of the coefficients A and B:

m
m
m
m
m
J2
e
N2
e
J0
,
(y) = Nm N0
(24)
k
k
k
k
where Nm is the norm of the eigenfunction. The second boundary condition, at y = R,
defines the mass spectrum of the theory and can be rewritten as

m
m kR
m
m kR
N0
(25)
J0
= 0.
J0
e
N0
e
k
k
k
k
One can see, that it is analogous to the one obtained in [9]. There exists a theorem about
such combinations of products of Bessel and Neumann functions, which asserts that for
ekR > 1 this combination is an even function of m/k and its zeros are real and simple [13].
Thus, one does not need to worry about the stability of the system: there are no tachyons.
The normalized functions m (y) satisfy the equation (see, for example, [7])

1
1
2
1 (y) + (y R) m n = mn .
dy e
(26)
k
k
Let us calculate the norm of the zero mode eigenfunction 0 (y) = N0 e2 . Substituting
it into (26) one can find that 0 (y) cannot be normalized, because the left part of Eq. (26)
is equal to zero for arbitrary N0 . In other words, the second variation Lagrangian of the
theory does not contain the kinetic term for the massless graviton, i.e., it is absent in this
model. One can ask, why do we consider the model, which does not contain long-range
gravity? But as we will see in the next section, the situation is rather different, if we place
matter on the brane.
3. Matter on the brane

Let us suppose that there is matter on one of the branes (we will specify, which brane to
choose later). Following the DGP-proposal [1], this matter induces a brane-localized term

2

ind
Sind =
(27)
d 4 x g R,
k16 G
where ind is a dimensionless parameter.
Now let us discuss, on which brane we can put the matter (it is evident that the term
(27) on a brane leads to the redefinition of the corresponding parameter i , for example,
2 1)). The problem is that the additional
for brane 1 one gets 1 = 1k 1 = 1k (ind
term (27) changes Eqs. (21) and (25) for the eigenvalues, and there may appear tachyonic
modes. This situation was discussed in detail in paper [14] (see also [15]). It was shown
in these papers that the gravitational tachyons can be avoided, at least if (in our notations
used in (1))
1 0,
2 0
(28)
307
(it was shown above that tachyons are absent in the case 1 = 1k and 2 = 1k too). So if
ind 1 and the term (27) arises on the brane 1 (at y = 0), the conditions (28) are satisfied
and we do not need to worry about the stability of the model. We would also like to note
that coordinates x are Galilean on brane 1, so that all the results obtained in coordinates x
for brane 1 are correct from the physical point of view.
One can easily check that the term (27) does not violate the symmetry (15), (16) (because of the fact that |y=0 = 0), but only if it arises on brane 1. Thus, the radion is absent
in this model, and we can forget about the radion as a ghost. On the contrary, the term
(27) on brane 2 (at y = R) violates the symmetry (15), (16). Moreover, the existence of the
induced terms (27) on both branes makes the radion to be a ghost (see [5,14]). One can also
consider the case = kR k|y| and the existence of matter on brane 2 only (in this case
coordinates x are Galilean on brane 2). The radion can be eliminated in this case too (since
|y=R = 0), but since 1 = 1k < 0 and 2 > 1k there may appear gravitational tachyons
(see [14,15]). Thus, the only physically relevant case, in which tachyons and ghost are absent (and the symmetry (15), (16) is preserved) is when the matter (and the induced term
(27)) exists on brane 1 only. Brane 2 can be interpreted as a naked brane, i.e., a brane
without matter on it.
Taking into consideration (27), we get new equations of motion (in the case (x) 0):
(1) -component

1
1
h h h + 42 h 2k 2 h + h + 2kh
2
2

1
2
+ h h 4 h 44 4 h
2

1
h h h + h + h h
2k

2
+ ind (y) h h h + h
2k

+ h h
= (y)t (x);
(29)
2
(2) 4-component

4 h h = 0;
(30)
(3) 44-component
3
1
h h 4 4 h = 0,
2
2
(31)
and

2
2 2
(y) h h
42 h + 24 4 h + h h ind
3k
3k
= (y)t (x),
(32)
3
308
where t is the energymomentum tensor of matter on the brane, for example, of a static
point-like mass. Since |y=0 = 0, the existence of matter on brane 1 does not violate the
symmetry (15), (16).
To solve these equations, it is convenient to make the following substitution
1
f (x).
(33)
2k 2
We note that the second term in substitution (33) is a pure gauge from the four-dimensional
point of view from the brane.
Let us take Eq. (32). Multiplying it by e2 , using Eq. (31), taking into account regularization (17) and the fact that the terms with derivative have no zero mode, we get
h (x, y) = b (x, y) +
b b =
t ,
2
2ind
(34)
b b + 3 f = 0,
(35)
4 b = 0,
(36)
and
as it was made in Section 2. With the help of the 4-component of the equations one can
impose the de Donder gauge on the field b (see also [10])

1
b b = 0.
(37)
2
It follows from Eqs. (34), (35) and (37) that
2f =
t
2
6ind
(38)
2b =
t,
2
ind
(39)
and
where t = t . We would also like to note that it is not correct to identify the field
f with the radion field , which is the 44-component of the metric fluctuations. But it is
apparent that this scalar part of -component of the metric fluctuations exists due to the
existence of the extra dimension and appears only if there is some matter on the brane. It
is similar to what happens in the RS2 model, where the radion field appears, if we place
matter on the brane (see [16]).
Substituting (33) into (29), we get

1
1 2
2 b b + 42 b 2k 2 b + 2kb
e
2
2

2
ind
1
1
1
2
2
2 b b +
2 b b
(y) e
e
2k
2
2k
2

= (y)t + 1
(40)
[ f 2f ],
2
k
309
where f is defined by (38).

It would be interesting to get the equation for the zero mode of h . As it was noted
above, the zero mode has the form h0 = e2 . Let us multiply Eq. (40) by e2 and
integrate it over coordinate y. Using the orthonormality conditions, which is modified by
the term (27) to be

2
ind
1
1
2
(y) m n = mn ,
1 (y) + (y R) +
dy e
(41)
k
k
k
we get
1
k
= 2 t.
2
ind
(42)
An analogous procedure was made in the case of RS1 model in paper [10].
A very interesting thing happens: the massless graviton reappears in the model. It looks
as if the matter produces the massless gravity via the induced term (which appears if
there is matter on the brane) for itself. Thus, the gravity on the brane in the zero mode
approximation is defined by h0 |y=0 = . The four-dimensional gravitational constant
is defined by the parameter ind (i.e., by the induced term) instead of the factor e2kR in the
original RS1 model.
4. Massive modes
Thus, we have found a solution for the gravity in the zero mode approximation. Now
let us estimate the effects, which can be produced by the massive modes. We will not solve
Eq. (40), as it was made in [10] for the RS1 model. Let us estimate the masses of the lowest
modes and their wave functions.
An analogue of Eq. (21) in the presence of the term (27) has the form

1 2
1 2
e
2h + 42 h 2k 2 h + 2k h
2h
e
2
2k
2
+ ind (y)e2 2h = 0.
2k
Following in the footsteps of Section 2, we arrive at the following relations:

m kR
m
m kR
m
m
(y) = Nm N0
e
e
e
e
J2
J0
N2
,
k
k
k
k
where Nm is a normalization constant, and

m kR
m
m kR
m
e
e
N0
J0
N0
J0
k
k
k
k

m
m
m
m
2
kR
kR
e
J0
e
= 0.
J2
N2
+ ind N0
k
k
k
k
(43)
(44)
(45)
Let us choose kR such that ekR is of the order 102 103 . We can make this assumption,
since it is not necessary to solve the hierarchy problem with the help of the factor ekR the
310
four-dimensional Planck mass is defined by ind . Since ind is assumed to be much larger
than 1 (we want to have a small five-dimensional Planck mass), the masses of the lowest
modes are defined by

m kR
e
= 0.
J0
(46)
k
Thus, mlow kekR .
Now let us estimate the normalization constants of the lowest modes. Using the fact
that the Bessel and Neumann functions are of the order 1 for the masses, which are the
solutions of (46), from (41) we get
2
1 2

ind
1 2kR 2
1
2kR
2
Z
e
Zm
Re
(0)
+
Z
(R)
+
(0),
m
Nm2
k m
k
k
where m (y) = Nm Zm (y). Since ind e2kR , we get
k
.
Nm
ind Zm (0)
(47)
(48)
Now it is not difficult to calculatethe coupling constants of the massless and massive
k
modes to matter on the brane (N0 = ind
):

k
1
k
4
(49)
.
d x
(x)t +
b (x)t
2
ind
ind
m
brane
Identifying the combination ind with the four-dimensional Planck mass MPl , we get
k

1
1
1 m
4
(50)
d x
(x)t +
b (x)t
.
2
MPl
MPl m
brane
We see that the coupling constant of the lowest massive modes is the same, as of the
massless graviton.
Now let us discuss the parameters of the model, which can give some interesting experimental consequences. Taking M k in the eV range (M is the five-dimensional Planck
mass) and kR 5 we get the size of the extra dimension of the order of 105 cm. The lowest modes have the masses of the order of 102 eV, which correspond to the corrections
to Newtons law with the strength of the massless graviton at the distances of the order of
103 cm, which is the micrometer scale. It can be interesting for experiments on testing
gravity at sub-millimeter scales.
5. Conclusion
In this paper a model with brane-localized terms, based on the RandallSundrum background solution for the metric, was considered. The original RandallSundrum scenario
was used not for the solution of hierarchy problem, but as a constructor of the branes
311
with tension. Thus, the proper gravitational field of the branes was taken into account. We
showed that this model is free from tachyons and ghosts and provides some interesting effects, such as the absence of the radion and inducing the long-range gravity by matter on
the brane. With an appropriate choice of the parameters, this model can lead to interesting
experimental consequences.
One may ask: why not to choose the original RandallSundrum model with the DGP
2 ), for example, on brane 1 only? According to [5,14] there are no gravitational
term ( ind
tachyons and ghosts in this case (but the radion field exists because of the absence of the
symmetry (15), (16)). But the answer to the question about the radion as a ghost can be
obtained only after the exact solving the equations of motion for the linearized gravity (as it
was made, for example, in [9] for the RS1 model). Nevertheless, if we retain strong gravity
in the bulk (small M , otherwise all the constructions make no sense) and ind e2kR ,
the term (27) on brane 1 makes the radion to be a ghost (all the reasonings are the same as
in paper [2] for the TeV brane in the RandallSundrum model)! But even if for some range
of parameters the radion is not a ghost (we remind that this issue can become clear after
solving the corresponding equations), the possible bounds on the coupling constant of the
radion may create an additional restrictions on the choice of the parameters k and R. At
the same time the case of 2 = 1k provides the absence of the radion and more freedom in
our choice of parameters.
We would also like to add that the accuracy in treating the equations of motion can
provide some interesting results, which are not evident at the first sight.
It would be interesting to calculate corrections to Newtons potential by solving directly
Eq. (40), as it was made in [10] for the RS1 model, but this problem deserves a further
investigation.
Acknowledgements
The author is grateful to D. Levkov and especially to I.P. Volobuev for valuable discussions. The work was supported by the grant UR.02.03.002 of the scientific program Universities of Russia, by the RFBR grant 04-02-16476 and by the grant NS-1685.2003.2 of
the Russian Federal Agency for Science.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
G.R. Dvali, G. Gabadadze, M. Porrati, Phys. Lett. B 485 (2000) 208.

M.A. Luty, M. Porrati, R. Rattazzi, JHEP 0309 (2003) 029.
V.A. Rubakov, Strong coupling in brane-induced gravity in five dimensions, hep-th/0303125.
S.L. Dubovsky, M.V. Libanov, JHEP 0311 (2003) 038.
A. Padilla, Ghost free braneworld bigravity, Class Quantum Grav. 21 (2004) 2899.
E. Kiritsis, N. Tetradis, T.N. Tomaras, JHEP 0203 (2002) 019.
H. Davoudiasl, J.L. Hewett, T.G. Rizzo, JHEP 0308 (2003) 034.
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370.
E.E. Boos, Yu.A. Kubyshin, M.N. Smolyakov, I.P. Volobuev, Theor. Math. Phys. 131 (2) (2002) 629.
I. Volobuev, M. Smolyakov, Theor. Math. Phys. 139 (1) (2004) 458.
W.D. Goldberger, M.B. Wise, Phys. Rev. Lett. 83 (1999) 4922.
312
[12] I.Ya. Arefeva, M.G. Ivanov, W. Mck, K.S. Viswanathan, I.V. Volovich, Nucl. Phys. B 590 (2000) 273.
[13] H. Bateman, A. Erdlyi, Higher Transcendental Functions, vol. 2, McGrawHill, 1953.
[14] C. Charmousis, J.-F. Dufaux, GaussBonnet gravity renders negative tension braneworlds unstable, hepth/0311267.
[15] Y. Shtanov, A. Viznyuk, Linearized gravity on the RandallSundrum two-brane background with curvature
terms in the action for the branes, hep-th/0312261.
[16] M.N. Smolyakov, I.P. Volobuev, Eur. J. Phys. 2 (1) (2004) 25.

Flavor violation and extra dimensions

S. Khalil a,b , R. Mohapatra c
a IPPP, Physics Department, Durham University, DH1 3LE, Durham, UK
b Ain Shams University, Faculty of Science, Cairo 11566, Egypt
c Department of Physics, University of Maryland, College Park, MD 20742, USA
Abstract
We analyze new sources of flavor violation in models with extra dimensions. We focus on three
major classes of five-dimensional models: models with universal extra dimension, models with split
fermions, and models with warped extra dimension. We study the implications of these new sources
on the associate CP violating asymmetries to the rare B-decays. We show that among these models
only the split fermions scenario may accommodate the recent experimental deviation between the
CP asymmetry of Bd KS and sin 2.
PACS: 11.30.Hv; 13.25.Hw
1. Introduction
It is a common belief held by many theorists for a long time that our world is not
limited to just four dimensions but it can extend to as many as eleven. More recently, extra
dimensions have been proposed as an alternative way to address the origin of the TeV scale
[13]. This has led to a new surge of activities to look for their implications for particle
phenomenology.
There are many variations on the basic theme of extra dimensions in order to discuss
different puzzles of the standard model; for instance, there have been attempts to underE-mail address: shaaban.khalil@durham.ac.uk (S. Khalil).
doi:10.1016/j.nuclphysb.2004.06.042
314
S. Khalil, R. Mohapatra / Nuclear Physics B 695 (2004) 313327
stand geometrically the origin of the fermion mass hierarchy [4,5], CP violation [6] etc. A
generic feature of this class of models is that they contain new sources of flavor violation,
due to the coupling of zero-mode fermions to the KaluzaKlien (KK) excitations of the
gauge and Higgs bosons, which might imply severe constraints on the string scale. This
raises the possibility of a conflict between the new physics scale preferred by the solution
of the hierarchy problem (MS O(1) TeV) and the one needed to satisfy the flavor constraints. Thus the issue of flavor violation in extra dimension models is an important one
and has only recently begun to be addressed. In particular, in several recent papers [7] this
has been discussed in the context of the universal extra dimension models [8]. Our goal in
this paper is to pursue this in the context of different extra dimension models and to isolate
possible flavor violation signals.
We are specially motivated by the recent experimental measurements [9,10] of the CP
asymmetries of Bd KS (SKS ) and Bd KS (S KS ), which seem to indicate a
deviation from standard model predictions by more than two standard deviations. This has
been considered as perhaps a hint for new CP and/or flavor violating sources beyond the
SM. There have been attempts to interpret this within the framework of supersymmetric
theories with a new flavor structure beyond the Yukawa matrices, i.e., in the squark mass
matrices. These results can be easily accommodated [11,12] even if the phase in the CKM
mixing matrix is the only source of CP violation [13]. If this anomaly is confirmed it would
be natural to look for the type of new physics that can generate such deviation between
SKS and SJ /KS and to ask if models with extra dimensions can accommodate this result.
In this paper we study the implications of models with extra dimensions on these
processes. We analyze three major classes of five-dimensional models. We start with the
models with universal extra dimension, then we consider the models with split fermions,
and finally we deal with scenario of warped extra dimension. We discuss possible new
sources of flavor and CP violations that these models could have beyond those in the CKM.
We also study the impact of these new sources on the associated CP violating asymmetries
to the rare B decays. We show that in the case of universal extra dimension, the flavor and
the CP violation are given by the CKM matrix as in the SM and no significant deviations
in the results of the CP asymmetries of the B decays are expected. We also emphasize that
with split fermions and models with warped extra dimensions, where fermions and gauge
fields live in the bulk, the KK contributions to the Bd B d and Bd KS occur at the tree
level. The MBd experimental limit then implies that the compactification scale is of order
104 GeV. With such a large scale, we find that only the split fermion models can lead to a
significant deviation between the CP asymmetries SKS and SJ /KS and the recent results
by Belle can be accommodated.
The paper is organized as follows. In the next section, we present model independent
expressions for the flavor violating interactions due to the KK-excitations of the gauge
fields in models with large extra dimensions. In Section 3, we study the effect of the new
flavor on the rare B-processes in the models with universal extra dimension. Section 4 is
devoted for analyzing the impact of the new sources of flavor in split fermions scenarios in
Bd B d mixing and CP asymmetry of Bd KS . In Section 5, we carry the same analysis
in the models with warped extra dimension. Our conclusions are presented in Section 6.
315
2. New source of flavor violation

We start our analysis by examining the impact of the infinite towers of the Kaluza
Klein (KK) modes that occur in extra dimensions on the B = 2 processes (B here stands
not for baryon number but the b quark number); in particular we study these effects for
Bd B d mixing and the CP asymmetry SJ /KS , and the B = 1 processes, like Bd KS .
Generally MBd and SJ /KS can be calculated via

B=2
MBd = 2Bd |Heff
|Bd ,
1
B=2
SJ /KS = sin 2eff , where eff = argBd |Heff
(1)
|Bd ,
2
B=2 is the effective Hamiltonian for the transition B = 2. In the framework of
where Heff
the SM,

(VCKM )cd (VCKM

)cb
.
SJSM
(2)
=
sin
2,
=
arg
/KS
(VCKM )t d (VCKM
)t b
2 and
The effect of the KK contribution can be described by a dimensionless parameter rKK
a phase 2KK defined as
2
e2iKK =
rKK
M12 (Bd )
SM (B )
M12
d
(3)
SM
KK (B ), and M KK is the new contribution arising from
where M12 (Bd ) = M12
(Bd ) + M12
d
12
SM
2 and the
(Bd )|rKK
the exchange of KK-modes. Thus MBd is given by MBd = 2|M12
CP asymmetry of Bd J /KS is given as
SJ /KS = sin 2eff = sin(2 + 2KK),
(4)
where 2KK = arg(1 +

The SM contribution is known at NLO accuracy in
QCD and it is given by

6/23
GF 2 2
SM
M12 =
MW (Vt d Vtb )2 S0 (xt )B s (b )
4

s (b )
4
2
1+
(5)
J5
mB f B1 () ,
4
3 d Bd
KK /M SM ).
M12
12
where fBd is the B-meson decay constant, B1 () = 0.86, and xt = (mt /mW )2 . The renormalization group evolution factors B and J5 are given by B = 0.551 and J5 = 1.627.
The InamiLim function S0 (x) is given by
S0 (x) =
4x 11x 2 + x 3
3x 3 ln x
.
2
4(1 x)
2(1 x)3
(6)
Now we turn to the effect of the KK contributions to the CP asymmetries of the Bd

KS . The direct and the mixing CP asymmetry are respectively given by [11]
CKS =
2
|(K
S )| 1
,
2
|(K
S )| + 1
SKS =
2 Im[(q/p)(K
S )]
,
2
|(K
S )| + 1
(7)
316
where (K
S ) = A(K
S )/A(KS ). The A(K
S ) and A(KS ) are the decay amplitudes
of B 0 and B 0 mesons, which can be written as
d
B=1 0

B ,
A(K
S ) = KS Heff

B=1 0

B .
A(KS ) = KS Heff
(8)
The mixing parameters q/p is given by

(i/2)
M12
q
12
=
,
p
M12 (i/2)12
(9)
where

B=2 0
B M12 i 12 .
B 0 Heff
2
The B = 1 effective Hamiltonian, including the KK-mediation, is given by

B=1
CiSM + CiKK Qi + (L R),
=
Heff
(10)
(11)
where Ci are the Wilson coefficients and Qi are the operators represent the b s s s transition. Thus, one obtains
A(KS ) = ASM (KS ) + AKK (KS ).
(12)
In this respect and following the parametrization of the SM and the KK as in Ref. [11], the
CP asymmetries of Bd KS are given by
SKS =
+ 2) + R 2 sin(2 + 2)
sin 2 + 2RKK cos KK sin(KK
KK
KK
CKS =
+ R2
1 + 2RKK cos KK cos KK
KK

2RKK sin KK sin KK
+ R2
1 + 2RKK cos KK cos KK
KK
(13)
(14)
= arg(AKK /ASM ), and

where RKK = |AKK /ASM |, KK
KK is the strong phase.
3. Flavor violation in universal extra dimension scenario

Recently there has been a growing interest concerning the models with one universal
extra dimension (UED), as an alternative view of gauge hierarchy problem of the SM
[8]. In this class of models, the compactification scale is the only additional free parameter
relative to the SM. Also, the tree level KK contributions to low energy processes are absent,
however, their one loop contributions are important as have emphasized in Ref. [7,8]. It was
shown that for 1/R 400 GeV, the KK impacts could be very significant. Nonetheless,
the electroweak precision impose a lower bound on 1/R to be larger than 300 GeV.
Concerning the flavor violation and CP violation in this model, it is, as in the SM, given
by the CKM matrix only. Therefore, one would not expect a significant deviation from
the SM results in the CP asymmetries of the B decays. In fact, within the UED scenario,
the main effect of the KK modes on these processes is the modification of the Inami
Lim one loop functions, as was found for other processes in Ref. [7]. We will show that
317
this modification is quite limited and cannot explain the 2.7 deviation from sin 2 in the
process Bd KS announced by Belle and BaBar Collaborations [9,10].
In the models with UED, the fifth dimension is compactified on the orbifold S1 /Z2 to
produce chiral fermion in four dimensions. There are infinite KK modes of the SM particle
with universal masses
n2
,
(15)
R2
where m0 is the mass of the zero mode, which is the ordinary SM particles. It was noticed
2
that the ratio xi(n) = m2i(n) /MW
(n) is a natural variable that enter the InamiLim functions
[7], where mi(n) are the masses of the fermionic KK modes and mW (n) are the masses of
the W boson KK modes. The effective Hamiltonian for the B = 2 transition in the UED
is given by [7]

6/23
G2
1 (5)
B=2
2
s (b )
= F 2 MW
(Vtb Vt d )2 B S xt ,
Heff
R
16

(5)
s (b )
J5 Q(B = 2) + h.c.,
1+
(16)
4
m2(n) = m20 +
where S(xt , 1/R) is defined as

1
= S0 (xt ) +
S xt ,
Sn (xt , xn ).
R
(17)
n=1
The KK contributions are represented by the function Sn (xt , xn ). A lower bound on the
compactification scale 1/R 165 GeV has been obtained by requiring the KK contribution
does not exceed the experimental central value MBd < 0.484 (ps)1 [18].
B=2 , the phase
It is worth mentioning that with the above expression of Heff
KK defined
in the previous section is identically zero, hence there is no new source of CP violation in
this class of models and the CP asymmetry of Bd J /KS would be given by the SM
value, sin 2, which is consistent with the experimental measurements.
Now we analyze the KK contributions to the CP asymmetry Bd KS decay in this
class of models with UED. The effective Hamiltonian for the B = 1 transitions through
the dominant gluon and chromomagnetic penguins is given by

6

G
F
B=1
Heff
(18)
= Vt b Vts
Ci Oi + Cg Og + (L R) ,
2
i=3
where Oi and Og are the relevant local operators, which are given in Ref. [19]. The corresponding Wilson coefficients are given as follows:

s
1
,
C3 (MW ) = 3C4 (MW ) = C5 (MW ) = 3C6 (MW ) =
(19)
E xt ,
24
R
and

1
1
Cg (MW ) = E xt ,
.
2
R
(20)
318
t , 1/R) and E (xt , 1/R) are given by

The functions E(x

1
E,
= F0 (xt ) +
Fn (xt , xn ),
F E,
F xt ,
R
n=1
are given in Ref. [7]. Taking into account the lower bound obtained on
where E 0,n and E0,n
the compactification scale from MBd measurements, one finds that the ratio RKK 0.05
0.02. Clearly these values are much smaller than the required values
and the phase KK
mentioned in Ref. [11] in order to deviate SKS from sin 2. Indeed we find that, in this
case, the total SKS is given by SKS 0.72. Thus, one can conclude that an experimental
confirmation of a deviation of SKS from sin 2 will disfavor models with UED.
We point out that the original UED models do not address the neutrino mass problem
and an extended version of this model based on the gauge groups SU(2)L U (1)I3R
U (1)BL and SU(2)L SU(2)R U (1)BL which solves the neutrino mass problem has
been discussed in [14]. In the four-dimensional leftright models [15], there is the well
known WL WR exchange box graph [16] which makes a large contribution to flavor changing hadronic processes such as KK mixing. One might therefore suspect that there will be
similar contributions in the 5D leftright model which will then lead to tighter constraints
on the compactfication scale. However, it turns out that since the lightest WR mode in 5D
LR models of the type discussed in Ref. [14] is a KK mode with a specific Z2 Z2 quantum number (+, ), it only connects the known quarks and leptons (which are zero modes
of the 5D fermion field) to KK excitations of the quarks with Z2 Z2 quantum numbers
(+, ). Therefore there is no box graph with WR and WL exchange that contributes to
FCNC processes. So only flavor changing diagrams arise from exchange of KK modes of
two WL s in a manner very similar to the UED models with SU(2)L U (1)Y gauge group.
This implies that the standard model prediction for Bd KS CP asymmetry is unaffected in the 5D LR models. Furthermore, there are no WL WR mixing contributions to
penguin type graphs since the same Z2 Z2 symmetry forbids WL WR mixing. This is to
be contrasted with the 4-dimensional leftright model case, where the WL WR loop gluon
penguin is indeed present and gives a large new CP violating contribution to Bd KS
and lead to significant deviation from the standard model predictions [17].
Thus our first conclusion is that in all the UED type models constructed so far, no
significant deviation from the standard model prediction for Bd KS CP asymmetry is
to be expected.
4. Flavor violation in split fermions scenario

Another class of extra dimensions which have drawn a lot of attention is the so called
split fermion scenario [4]. The idea of split fermion is that having different fermion localization along the fifth direction may provide a geometrical way to look at the fermion mass
hierarchy. As in the UED type models, in these models, all particles are considered to be
in all five dimensions; but in contrast with the UED models, the SM fermions i are lo2
2
calized at different points in the fifth dimension with Gaussian profiles i e(yyi ) / .
Here yi is the position of the quark in the fifth dimension and is the width of the fermion
319
wave functions with R. In order to avoid introducing another hierarchy, one takes
/R 0.1. The quark mass matrices arise from the interaction of fermions and the vacuum expectation value (VEV) of Higgs zero mode and are given by
(Mu )ij =
v0 (u )ij
e
2
(u
ij )
4 2
d 2
v0 (d )ij (ij2 )
e 4 ,
(Md )ij =
2
(21)
where ij = |yi yj | is the distance between flavor i and j . The parameters (u,d )ij are
the 5D Yukawa couplings, which are in general arbitrary matrices. In order to relate the
hierarchy of the fermion masses to the locations of different fermion families, one assumes
that these couplings are of order unity. However, in order to avoid the factorizable form
of the quark mass matrices which has always two vanishing eigenvalues, the u,d matrices
cannot be unit matrices. Thus, in this class of models the number of free parameters is
larger than the number of the observed fermion masses and mixings and it is very easy
to accommodate different types of Yukawa textures with hierarchical or non-hierarchical
features. Also with complex ij one can get the SM phase in the CKM of order /2.
Examples of hierarchical Yukawa couplings have been obtained in Ref. [2022], which
fit all the quark and lepton masses and mixing angles. For instance, the solution of Ref. [21]
leads to the following position for the up and down quarks in the fifth dimension:

0
yQL 14.2349 ,
8.20333

6.13244
yuR 20.092 .
9.64483

ydR

19.4523
5.15818 ,
10.1992
(22)
As we will explain below, in these models and due to the non-universal couplings with
KK-gluon, both left- and right-handed rotations VL,R that diagonalize the mass matrix
are observable. This is unlike the case in the SM where only VL rotations are physical
+
(VCKM = VLd VLu ) and VR is completely decoupled. In general, VR matrix has six phases,
these new phases might play an important rule in the CP violation of the B system and this
what we are going to examine below. It is also worth mentioning that for a hierarchical VR ,
this new KK effect is suppressed, and therefore, in our analysis we choose the parameters
ij such that VR is a non-hierarchical matrix.
The fields living in the bulk can be defined to be even or odd under the Z2 parity.
Thus we will assume that the gauge boson in 5D bulk are even under the Z2 and have the
following Fourier expansion
1
A (x, y) = A(0)
+
R
2
ny
cos A(n)
(x),
R
R
n=1
(23)
320
Fig. 1. The KK-modes gluon contributions to the Bd B d mixing.
where R is the size of the extra dimension. The relevant terms of the effective 4D Lagrangian as far as flavor is concerned are given by
L = dR Md
diag
diag
dL + u R Mu
g
uL + W u L VCKM dL
2

d(n)
d(n)
dL T A UL dL + dR T A UR dR + (d u) + h.c.
2gs GA(n)
n=1
(24)
In obtaining Eq. (24), the fifth dimension has been integrated out and due to the Gaussian
nature of the quark wave functions, only the interaction of the KK-gluon, GA(n)
, with zero
mode mass eigenstate quarks at the points yi , given in Eq. (22), are picked. The matrices
d(n)
UL,R
are defined as
d(n)
d(n) d
d
= VL(R)
CL(R)
VL(R),
UL(R)
(25)
d
where VL,R
are the unitary matrices that diagonalize the down quark mass matrix. The
d(n)
non-universal couplings CL(R) are given by

d(n)
CL(R) = diag cos(nydL(R) /R), cos(nysL(R) /R), cos(nybL(R) /R) .
(26)
The non-universality in the position of the different families at y-direction leads to a

q(n)
new source of flavor in UL,R in addition to the usual VCKM . It is clear that this flavor
violation can be mediated at tree level by the KK modes of the gluon which makes it quite
dangerous and strong bounds on the compactification scale have been obtained [23]. Of
course, there are similar contributions from the KK modes of the , Z and W bosons;
however, one expects that the largest effect is due to the KK gluon.
Unlike the SM and UED, the contribution from the KK-modes in split fermion models
to the Bd B d mixing is at the tree level. The KK-gluon, shown in Fig. 1, gives the dominant
contributions to this process. However, since the KK-modes of the gluino has no coupling
between the b and c quarks at tree level, its contribution to the Bd J /KS decay takes
place at loop level. Therefore, the main source of CP violation in this process is through
the oscillation as in the SM.
321
Table 1
The lower bounds on the compactification scale from the experimental measurements of MBd and MK as
function of the parameter /R. The unit of the mass MS is in GeV
/R
Lower bounds on MS from Exp. MBd
0.1
0.05
0.01
Lower bounds on MS from Exp. MK

8 105
3 105
8 104
104
8 103
5 103
In the models with split fermions, the B = 2 effective Lagrangian for the KK-gluon
exchange is given by
n=1
i,j =L,R
2 1
= gs2
LB=2
KK
3
Mn2

d(n) d(n)
Ui(bd) Uj (db) bi di bj dj .
(27)
The hadronic matrix elements B d |(bi di )(bj dj )|Bd , i, j = L, R are given by

1
B d | bL dL bL dL |Bd = mBd fB2d B1 (),
3

2

1
m Bd
mBd fB2d B4 (),

Bd | bL dL bR dR |Bd =
4 mb () + md ()
(28)
(29)
where mBd is the mass of the Bd meson and mb and md are the masses the b and d
quarks at the scale . In our analysis, we assume that mb (mb ) = 4.6 GeV and md (mb ) =
0.0054 GeV. The B-parameters are given by: B1 (mb ) = 0.87 and B4 (mb ) = 1.16.
The other operators which obtained by exchanging L R from the above ones, have
KK =
the same matrix elements, since the strong interaction preserve parity. Thus, M12
B=2
Bd |Heff |Bd can be written as

KK
M12
=
3
2 gs2
3 Mc2
B d | bi di bj dj |Bd
k,m=1 i,j =L,R

cos(n(yi )k /R) cos(n(uj )m /R)
Vid k3 Vid k1 Vjd m1 Vjd m3
. (30)
n2
n
For the example of 5D split fermion discussed above, the experimental limit MBd <
3.21013 GeV leads to a lower bound on the compactification scale of order 104 GeV. As
can be seen from Table 1, the bounds on MS , for /R = 0.1, 0.05, 0.01, derived from the
experimental measurement of MK 3.5 1015 GeV are about one order of magnitude
larger than those obtained from the experimental limit on MBd . However, in the KK
mixing, there is a significant uncertainty due to the low QCD corrections which make
this constraint is unreliable. Therefore, we will follow the conservative approach and will
consider, through our analysis, the Bd B d mixing to constrain the compactification scale.
Applying these constraints on the compactification scale, one finds that the values of
the phase KK are very small ( 102 ), hence the CP asymmetry SJ /KS remains equal
to the SM prediction. It is worth stressing that, the KK contribution to Bd B d mixing
(30) is proportional to the transition factor between the first and third generations which
322
Fig. 2. The SM and the KK-modes gluon contributions to the Bd KS decay.
is typically very small. This counts as an extra suppression factor for the phase KK , in
addition to the large compactification scale. Thus, the chance of having a significant effect
on the CP asymmetry SJ /KS is reduced.
Now we consider the KK contribution to the CP asymmetry SKS . The tree level
KK-modes gluon contributions to the Bd KS decay is shown in Fig. 2. The decay
amplitude of B d0 KS (employing the naive factorization approximation) is given by
B=1 0
A(K
|Bd
S ) = KS |Heff
n=1
i,j =L,R
2 1
= gs2
3
Mn2

d(n) d(n)
Ui(bs)
Uj (ss) KS bi si (sj sj )B d0 .
(31)
The matrix elements KS |(bi si )(sj sj )|B d0 can be found in Ref. [11]. It is remarkd U d in A(K
d
d
able that the coupling U(bs)

S ) is typically larger than the coupling U(bd) U(db)
(ss)
of the Bd B d mixing, with one or two orders of magnitude. The size of this deviation
d(n)
strongly depends on the non-universality among the parameters CL,R
and also on the flad
vor structure of the rotational matrices VL,R that diagonalize the down quark mass matrix.
This implies that it is quite natural to have significant KK contributions to the CP asymmetry of Bd KS process and negligible one to the CP asymmetry of Bd J /KS .
As shown in Eq. (13), the deviation of SKS from sin 2 is governed by the size of the
. In this class of models, there is a new source of flavor and CP
ratio RKK and the phase KK
violation due to the impact of the right-handed rotation VRd . We find that the size of the RKK
strongly depend on the favor structure of V d and the non-hierarchical form of V d
and KK
R
R
is favored to enhance their size and hence increase deviation between SKS and SJ /KS . It
is also worth noting that the required form of VRd can be obtained by tuning the arbitrary
5D Yukawa couplings dij . We have explicitly studied different examples and found that
RKK can vary from O(0.01) with VRd O(VCKM ) to RKK 0.3 for non-hierarchical VRd .

Moreover, the phase KK
is also quite sensitive to the structure of VRd and it could be of
order O(1).
In Fig. 3 we present the predicted values of the CP asymmetry of SKS as function of
the position of the right-handed bottom quark in the fifth dimension (which is one of the
relevant parameters for SKS ). The other quark positions have been fixed as in Eq. (22).
323
Fig. 3. The figure shows the predictions for SKS as a function of ybR / for various choice of the ratio /R.
The various values of /R are 0.1 for the solid line, 0.05 for dashed line and 0.01 for dash-dotted line.
Also the matrix VR has been chosen to be non-hierarchical and three values of /R have
used as in Table 1. As can be seen from this figure, it is possible to deviate SKS significantly from SJ /KS . Moreover, with a larger mixing in VR (specially between the 2nd and
3rd generations) or with different set of positions from those given in Eq. (22), one gets
smaller and even negative values for SKS .
Thus in the split fermions scenario, the free parameters can be easily adjusted in order
to accommodate the anomaly SKS and sin 2.
5. Flavor violation in warped extra dimensions

Finally we consider the five-dimensional models with warped geometry where the SM
fermions and gauge bosons correspond to bulk fields [24]. The warped geometry has been
proposed as solution of the hierarchy problem. In the original model, the SM fields were
localized to one of the boundaries and gravity is allowed to propagate in the bulk. However, it was realized that the scenarios of SM gauge bosons and fermions in the bulk may
lead to a new flavor and possible geometrical interpretation for the hierarchy of quark and
lepton masses. The Higgs field has to be confined to the TeV brane in order to obtain the
observable masses of the W and Z gauge bosons.
We will consider the scenario of Ref. [3], based on the metric
ds 2 = e2 (y) dx dx + dy 2,
(32)
where (y) = |y| with MP is the curvature scale determined by the negative cosmological constant in the five-dimensional bulk. The fermion fields reside in the bulk of this
324
non-factorizable geometry can be decomposed as

(x, y) =
1
2R
(n) (x)e2 (y)fn (y).
(33)
n=0
Here R is the radius of the compactified fifth dimension on an orbifold S1 /Z2 so that the
bulk is a slice of AdS5 space between two four-dimensional boundaries. The zero mode
wave function is given by [5]
ec (y)
(34)
,
N0
where c = m / and m is the bulk mass term. Using the orthonormal condition:
R
(1/2R) R dy e f0 (y)f0 (y) = 1, one finds that N0 is given by

eR(12c) 1
.
N0 =
(35)
R(1 2c)
f0 (y) =
The tower of fermion KK excited states is not relevant to our discussion here. Also the
massless gauge fields that propagate in this curved background can be decomposed as [5]
A (x, y) =
2R n=0
A
A(n)
(x)fn (y),
(36)
with fnA is given as

fnA (y) =

(y)
mn (y)
mn (y)
e
e
J1
+ bA (mn )Y1
,
Nn
(37)
where J1 and Y1 are Bessel function of order one and bA (mn ) = J0 (mn /)/Y0 (mn /).
The coupling of the gauge KK modes to the fermion is given by
g (5)
gij n =
(2R)3/2
R
e (y)fi (y)fj (y)fnA (y) dy.
(38)
As in the split fermion scenario, the non-universal parameters ci lead to non-universal couplings to the KK state of the gluon. In the basis of mass eigenstates we have the following
flavor dependent couplings
+
d(n)
d
d
g00n VL(R)
,
UL(R) = VL(R)
(39)
where the g00n is given by

R
mn y
mn y
e
e
+ bA (mn )Y1
.
e(12c)y J1
0
(40)
The gauge coupling g is defined as g = g5 / 2R, where g5 is the 5D gauge coupling.

d diagonalize the down quark mass matrix which is given
Finally, the unitary matrices VL,R
1 2c
g00n = g R(12c)
e
1 Nn
325
in this model as
lij
d
f d (R)f0j
Yijd =
R (R).
R 0iL
(41)
(5)
(5)
The dimensionless parameters lij are defined as ij where ij are the 5D Yukawa
couplings which are free parameters to be fixed by the observable masses and mixing.
Therefore, this class of model with warped geometry is similar to the models with split
fermions on large extra dimensions where the number of free parameters is larger than the
number of the quark masses and mixings. In this respect, any type of Yukawa textures can
be obtained in this models by tuning the free parameters ci and lij . For instance, in order
to get the non-hierarchical Yukawa textures that we have used in the previous section:

0.006 0.018 0.018
0.001 0.0017 0.019
|Yd | 0.012 0.056 0.037 ,
|Yu | 0.0037 0.0069 0.0002 ,
0.0189 0.005 0.99
0.012 0.041 0.976
(42)
one can start with any reasonable choice of ci which is consistent with the observed hierarchy between different generations and between up and down type quarks and find the
corresponding couplings (5)
ij that lead to these textures. This choice is not unique, for
example, we can use the parameters ci listed in Ref. [26]:
cQ1 = 0.643,
cQ2 = 0.583,
cQ3 = 0.317,
cD1 = 0.643,
cD2 = 0.601,
cD3 = 0.601,
cU1 = 0.671,
cU2 = 0.528,
cU3 = 0.460,
with the following values of (5)

ij :

1.36 1.08 0.72
d
l 0.45 0.58 0.39 ,
ij
0.03 0.002 0.42
u
l =
ij
28
0.8
0.66
17.7 0.57 0.0014
2.19 0.13 0.23
(43)

(44)
to get the Yukawa couplings in Eq. (42).

As advocated above, this class of model leads to a flavor violation at the tree level
similar to the split fermion scenario. Eqs. (27) and (31) are still valid for the effective
d(n)
Hamiltonians of the B = 2 and B = 1 processes, but after substituting UL(R) with its
value from Eq. (39). Having fixed the Yukawa couplings in both cases, thus the difference
between the flavor predictions of these two models is mainly due to the difference between
d(n)
(in split fermions) and g00n (in warped geometry). In
the non-universal couplings: cL,R
d(n)
fact, it is easy to note that the non-universality of cL,R
is much stronger than the nonuniversality of g00n . For instance, with our above assumption, the g001 which gives the
dominant contribution is given by g001 g diag{0.04, 0.03, 4.7 107 } while CLd(1) is
given by CLd(1) diag{1, 0.14, 0.68}. Such large non-universality in split fermions scenario
has been proved to be very useful in order to get a significant effect on the CP asymmetry
of Bd KS process, even with compactification scale of order 104 GeV. This makes
the chance of saturating the new results of the CP asymmetry possible within this class of
model.
326
In warped geometry, the compactification scale is constrained by the electroweak corrections to be MS 10 TeV [25]. It was shown that with this value of MS most of the flavor
violating processes induced by the KK excitations of the gauge boson are suppressed and
quite below their experimental limits [26]. We have also verified that the KK contribution
to MBd is quite negligible respect to the SM value. The same is true for the CP asymmetry SJ /KS which is essentially given by the SM value sin 2. Finally, we comment on
the KK contributions to Bd KS in the warped extra dimensions. We find that the ratio
of the KK amplitude to the SM amplitude is very small (RKK O(102 ), which implies
that it is not possible in this class of models to deviate the value of SKS from the value of
SJ /KS .
6. Conclusion
To summarize, we have searched for new flavor violating CP asymmetric effects in three
different classes of extra dimension models that could significantly alter the predictions of
the standard model and have found that the only models where measurable deviations can
occur are those based on the split fermion hypothesis. In other models such CP violating
effects are suppressed because of the constraints from the CP conserving sector. While
experimental confirmation of such deviations will not necessarily be an evidence for such
models, any lack of deviation from standard model will impose constraints on the parameters of the model. For instance, agreement with standard model would mean that the
Yukawa couplings in the split fermion models have specific patterns. It may also impose
constraints on the location of the flavors in the fifth dimension in specific models.
Acknowledgements
The work of R.N.M. was supported by the US National Science Foundation Grant No.
PHY-0099544 and the work of S.K. was supported by PPARC. S.K. would like to thank
V. Sanz for enlightening discussion.
References
[1] I. Antoniadis, Phys. Lett. B 246 (1990) 377;
J.D. Lykken, Phys. Rev. D 54 (1996) 3693;
E. Witten, Nucl. Phys. B 471 (1996) 135;
P. Horava, E. Witten, Nucl. Phys. B 475 (1996) 94;
P. Horava, E. Witten, Nucl. Phys. B 460 (1996) 506;
E. Caceres, V.S. Kaplunovsky, I.M. Mandelberg, Nucl. Phys. B 493 (1997) 73.
[2] N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 429 (1998) 263;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 436 (1998) 257;
N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Rev. D 59 (1999) 086004.
[3] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370.
[4] N. Arkani-Hamed, M. Schmaltz, Phys. Rev. D 61 (2000) 033005.
327
[5] T. Gherghetta, A. Pomarol, Nucl. Phys. B 586 (2000) 141;

S.J. Huber, Q. Shafi, Phys. Lett. B 498 (2001) 256.
[6] D. Chang, R.N. Mohapatra, Phys. Rev. Lett. 87 (2001) 211601;
D. Chang, W. Keung, R.N. Mohapatra, Phys. Lett. B 515 (2001) 431;
N. Cosme, J.-M. Frere, L. Lopez Honorez, Phys. Rev. D 68 (2003) 096001;
D. Dooling, D.A. Easson, K. Kang, JHEP 0207 (2002) 036;
B. Grzadkowski, J. Wudka, hep-ph/0401232.
[7] A.J. Buras, M. Spranger, A. Weiler, Nucl. Phys. B 660 (2003) 225;
A.J. Buras, A. Poschenrieder, M. Spranger, A. Weiler, Nucl. Phys. B 678 (2004) 455.
[8] T. Appelquist, H.C. Cheng, B.A. Dobrescu, Phys. Rev. D 64 (2001) 035002.
[9] Belle Collaboration, K. Abe, et al., hep-ex/0308035.
[10] BaBar Collaboration, S.M. Spanier, BABAR-PLOT-0056, Contributed to 21st International Symposium on
Lepton and Photon Interactions at High Energies (LP 03), Batavia, IL, 1116 August 2003.
[11] S. Khalil, E. Kou, Phys. Rev. D 67 (2003) 055009;
S. Khalil, E. Kou, Phys. Rev. Lett. 91 (2003) 241602;
S. Khalil, E. Kou, eConf C0304052 (2003) WG305.
[12] E. Lunghi, D. Wyler, Phys. Lett. B 521 (2001) 320;
T. Moroi, Phys. Lett. B 493 (2000) 366;
D. Chang, A. Masiero, H. Murayama, Phys. Rev. D 67 (2003) 075013;
R. Harnik, D.T. Larson, H. Murayama, A. Pierce, hep-ph/0212180;
M. Ciuchini, E. Franco, A. Masiero, L. Silvestrini, Phys. Rev. D 67 (2003) 075016;
M. Ciuchini, E. Franco, A. Masiero, L. Silvestrini, Phys. Rev. D 68 (2003) 079901, Erratum;
S. Baek, Phys. Rev. D 67 (2003) 096004;
K. Agashe, C.D. Carone, Phys. Rev. D 68 (2003) 035017;
G.L. Kane, P. Ko, H.B. Wang, C. Kolda, J.H. Park, L.T. Wang, Phys. Rev. Lett. 90 (2003) 141803;
D. Chakraverty, E. Gabrielli, K. Huitu, S. Khalil, Phys. Rev. D 68 (2003) 095004;
J.F. Cheng, C.S. Huang, X.H. Wu, hep-ph/0306086;
T. Goto, Y. Okada, Y. Shimizu, T. Shindou, M. Tanaka, hep-ph/0306093;
R. Arnowitt, B. Dutta, B. Hu, Phys. Rev. D 68 (2003) 075008.
[13] S. Khalil, V. Sanz, Phys. Lett. B 576 (2003) 107.
[14] R.N. Mohapatra, A. Perez-Lorenzana, Phys. Rev. D 67 (2003) 075015.
[15] J.C. Pati, A. Salam, Phys. Rev. D 10 (1974) 275;
R.N. Mohapatra, J.C. Pati, Phys. Rev. D 11 (1975) 566, 2558;
G. Senjanovic, R.N. Mohapatra, Phys. Rev. D 12 (1975) 1502.
[16] G. Beall, M. Bander, A. Soni, Phys. Rev. Lett. 48 (1982) 848.
[17] M. Raidal, Phys. Rev. Lett. 89 (2002) 231803.
[18] D. Chakraverty, K. Huitu, A. Kundu, Phys. Lett. B 558 (2003) 173.
[19] G. Buchalla, A.J. Buras, M.E. Lautenbacher, Rev. Mod. Phys. 68 (1996) 1125.
[20] Y. Grossman, G. Perez, Phys. Rev. D 67 (2003) 015011.
[21] E.A. Mirabelli, M. Schmaltz, Phys. Rev. D 61 (2000) 113011.
[22] W.F. Chang, J.N. Ng, JHEP 0212 (2002) 077.
[23] B. Lillie, J.L. Hewett, Phys. Rev. D 68 (2003) 116002;
A. Delgado, A. Pomarol, M. Quiros, JHEP 0001 (2000) 030;
S.A. Abel, M. Masip, J. Santiago, JHEP 0304 (2003) 057;
G.C. Branco, A. de Gouvea, M.N. Rebelo, Phys. Lett. B 506 (2001) 115, hep-ph/0012289.
[24] W.D. Goldberger, M.B. Wise, Phys. Rev. D 60 (1999) 107505;
H. Davoudiasl, J.L. Hewett, T.G. Rizzo, Phys. Lett. B 473 (2000) 43;
A. Pomarol, Phys. Lett. B 486 (2000) 153;
S. Chang, J. Hisano, H. Nakano, N. Okada, M. Yamaguchi, Phys. Rev. D 62 (2000) 084025;
Y. Grossman, M. Neubert, Phys. Lett. B 474 (2000) 361.
[25] J.L. Hewett, F.J. Petriello, T.G. Rizzo, JHEP 0209 (2002) 030;
S.J. Huber, C.A. Lee, Q. Shafi, Phys. Lett. B 531 (2002) 112.
[26] S.J. Huber, Nucl. Phys. B 666 (2003) 269;
G. Burdman, hep-ph/0310144.

CUMULATIVE AUTHOR INDEX B691B695
Abramowicz, H.
Abt, I.
Adamczyk, L.
Adamus, M.
Adler, V.
Aghuzumtsyan, G.
Aharony, O.
Akemann, G.
Alishahiha, M.
Alkalaev, K.B.
Antoniadis, I.
Antonioli, P.
Antonov, A.
Aoki, K.
Arneodo, M.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B691 (2004) 3
B694 (2004) 59
B691 (2004) 111
B692 (2004) 363
B695 (2004) 103
B695 (2004) 3
B695 (2004) 3
B695 (2004) 132
B695 (2004) 3
Bailey, D.S.
Bamberger, A.
Barakbaev, A.N.
Barbagli, G.
Barbi, M.
Bari, G.
Barreiro, F.
Bartsch, D.
Basile, M.
Beane, S.R.
Becker, M.
Behrens, U.
Behrndt, K.
Bell, M.
Bellagamba, L.
Bellucci, S.
Beneke, M.
Benen, A.
Bertolin, A.
Bhadra, S.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 192
B693 (2004) 223
B695 (2004) 3
B694 (2004) 99
B695 (2004) 3
B695 (2004) 3
B693 (2004) 51
B692 (2004) 232
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
0550-3213/2004 Published by Elsevier B.V.

doi:10.1016/S0550-3213(04)00539-5
Bigazzi, F.
Binoth, T.
Bloch, I.
Bod, T.
Bonanno, A.
Boos, E.G.
Borras, K.
Boscherini, D.
Brihaye, Y.
Brock, I.
Brook, N.H.
Brower, R.
Brugnera, R.
Brmmer, N.
Bruni, A.
Bruni, G.
Buchbinder, I.L.
Burguet-Castell, J.
Bussey, P.J.
Butterworth, J.M.
Bttner, C.
Bylsma, B.
B694 (2004) 3
B693 (2004) 134
B695 (2004) 3
B695 (2004) 3
B693 (2004) 36
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B691 (2004) 79
B695 (2004) 3
B695 (2004) 3
B693 (2004) 149
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B693 (2004) 51
B695 (2004) 217
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
Caldwell, A.
Callan Jr., C.G.
Capua, M.
Cara Romeo, G.
Carli, T.
Carlin, R.
Casper, D.
Catterall, C.D.
Chandrasekharan, S.
Chekanov, S.
Chiochia, V.
Chwastowski, J.
Ciborowski, J.
Ciesielski, R.
B695 (2004) 3
B694 (2004) 115
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 217
B695 (2004) 3
B693 (2004) 149
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
330
Cifarelli, L.
Cindolo, F.
Cloth, P.
Cole, J.E.
Collins-Tooth, C.
Contin, A.
Cooper-Sarkar, A.M.
Coppola, N.
Cormack, C.
Corradi, M.
Corriveau, F.
Costa, M.
Cotrone, A.L.
Cottrell, A.
Curio, G.
Curio, G.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 3
B695 (2004) 3
B693 (2004) 195
B693 (2004) 223
DAgostini, G.
Dal Corso, F.
DallAgata, G.
Danilov, P.
Dannheim, D.
DAuria, R.
del Peso, J.
Dementiev, R.K.
De Pasquale, S.
Dereli, T.
Derendinger, J.-P.
Derrick, M.
Descotes-Genon, S.
Devenish, R.C.E.
de Wolf, E.
Dhawan, S.
DHoker, E.
Dobur, D.
Dolgoshein, B.A.
Donagi, R.
Doyle, A.T.
Drews, G.
Dubovsky, S.L.
Durkin, L.S.
Dusini, S.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 243
B695 (2004) 3
B695 (2004) 3
B693 (2004) 261
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B691 (2004) 223
B691 (2004) 233
B695 (2004) 3
B693 (2004) 103
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 132
B695 (2004) 3
B695 (2004) 3
B694 (2004) 187
B695 (2004) 3
B695 (2004) 3
B691 (2004) 91
B695 (2004) 3
B695 (2004) 3
Eisenberg, Y.
Enger, H.
Ermolov, P.F.
Eskreys, A.
Everett, A.
Eynard, B.
B695 (2004) 3
B695 (2004) 73
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 443
Faraggi, A.E.
Faraggi, A.E.
Felder, G.
Ferrando, J.
B694 (2004) 187

B695 (2004) 41
B691 (2004) 251
B695 (2004) 3
Ferrara, S.
Ferrero, M.I.
Figiel, J.
Filges, D.
Fischbacher, T.
Foster, B.
Foudas, C.
Fourletov, S.
Fourletova, J.
Fr, P.
Fricke, U.
Fuchs, J.
Fusayasu, T.
Fyodorov, Y.V.
B693 (2004) 261

B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 525
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 239
B695 (2004) 3
B694 (2004) 277
B695 (2004) 3
B694 (2004) 59
Gabareen, A.
Gaberdiel, M.R.
Gade, R.M.
Galas, A.
Gallo, E.
Garca-Compen, H.
Garfagnini, A.
Gasperini, M.
Gattringer, C.
Gehrmann, T.
Gehrmann-De Ridder, A.
Geiser, A.
Genta, C.
Ghodsi, A.
Gialas, I.
Giovannini, M.
Giusti, P.
Giveon, A.
Gladilin, L.K.
Gladkov, D.
Glasman, C.
Gliga, S.
Glover, E.W.N.
Gckeler, M.
Goers, S.
Gmez-Cadenas, J.J.
Gonalo, R.
Gonzlez, O.
Gosau, T.
Gttlicher, P.
Grabowska-Bod, I.
Grijpink, S.
Grzelak, G.
Gutjahr, P.
Gutsche, O.
Gwenlan, C.
B695 (2004) 3
B693 (2004) 281
B694 (2004) 354
B695 (2004) 3
B695 (2004) 3
B694 (2004) 405
B695 (2004) 3
B694 (2004) 206
B694 (2004) 170
B691 (2004) 195
B691 (2004) 195
B695 (2004) 3
B695 (2004) 3
B691 (2004) 111
B695 (2004) 3
B694 (2004) 206
B695 (2004) 3
B691 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B691 (2004) 195
B694 (2004) 170
B695 (2004) 3
B695 (2004) 217
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B692 (2004) 110
B695 (2004) 3
B695 (2004) 3
Haas, T.
Hadasz, L.
B695 (2004) 3
B694 (2004) 493
Hain, W.
Hall-Wilton, R.
Hamatsu, R.
Hambye, T.
Hamilton, J.
Hanlon, S.
Hart, J.C.
Hartmann, B.
Hartmann, H.
Hartner, G.
Heaphy, E.A.
Heath, G.P.
Heinrich, G.
Helbich, M.
Hernndez, P.
Heusch, C.
Hilger, E.
Hillert, S.
Hirose, T.
Hochman, D.
Holland, K.
Holm, U.
Horn, C.
Horsley, R.
Huber, P.
Hung, P.Q.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 169
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B691 (2004) 79
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B693 (2004) 134
B695 (2004) 3
B695 (2004) 217
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 35
B695 (2004) 3
B695 (2004) 3
B693 (2004) 3
B694 (2004) 170
B692 (2004) 83
Iacobucci, G.
Iga, Y.
Inuzuka, M.
Irrgang, P.
Ivanov, E.A.
Ivanov, R.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 473
B694 (2004) 509
Jakob, H.-P.
Jasklski, Z.
Jeschek, C.
Jones, T.W.
B695 (2004) 3
B694 (2004) 493
B694 (2004) 99
B695 (2004) 3
Kagawa, S.
Kahle, B.
Kaji, H.
Kananov, S.
Karshon, U.
Karstens, F.
Kataoka, M.
Katkov, I.I.
Kira, D.
Kenna, R.
Khalil, S.
Khein, L.A.
Kim, J.Y.
Kimura, Y.
Kind, O.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B691 (2004) 292
B695 (2004) 313
B695 (2004) 3
B695 (2004) 3
B692 (2004) 394
B695 (2004) 3
331
Kisielewska, D.
Kitamura, S.
Kiyo, Y.
Klemm, H.
Kniehl, B.A.
Koffeman, E.
Kohno, T.
Kokotov, A.
Koma, M.
Koma, Y.
Kooijman, P.
Koop, T.
Korotkin, D.
Korzhavina, I.A.
Kotanski, A.
Ktz, U.
Kounnas, C.
Kounnas, C.
Kowal, A.M.
Kowal, M.
Kowalski, H.
Kramberger, G.
Krause, A.
Krause, A.
Kreisel, A.
Kruczenski, M.
Krumnack, N.
Krykhtin, V.A.
Kutasov, D.
Kuze, M.
Kuzmin, V.A.
B695 (2004) 3
B695 (2004) 3
B692 (2004) 232
B693 (2004) 281
B695 (2004) 199
B695 (2004) 3
B695 (2004) 3
B694 (2004) 443
B692 (2004) 209
B692 (2004) 209
B695 (2004) 3
B695 (2004) 3
B694 (2004) 443
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B691 (2004) 233
B695 (2004) 41
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B693 (2004) 195
B693 (2004) 223
B695 (2004) 3
B692 (2004) 3
B695 (2004) 3
B693 (2004) 51
B691 (2004) 3
B695 (2004) 3
B695 (2004) 3
Labarga, L.
Labes, H.
Lacagnina, G.
Lainesse, J.
Lammers, S.
Lang, C.B.
Lelas, D.
Levchenko, B.B.
Levy, A.
Li, L.
Lightwood, M.S.
Lim, H.
Lim, I.T.
Limentani, S.
Lin, Y.
Ling, T.Y.
Liu, X.
Loaiza-Brito, O.
Lhr, B.
Lohrmann, E.
Loizides, J.H.
Long, K.R.
B695 (2004) 3
B695 (2004) 3
B693 (2004) 36
B695 (2004) 3
B695 (2004) 3
B694 (2004) 170
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 169
B695 (2004) 3
B695 (2004) 3
B694 (2004) 405
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
332
Palmonari, F.
Papucci, M.
Parenti, A.
Park, I.H.
Patel, S.
Paul, E.
Pavel, N.
Pawlak, J.M.
Pelfer, P.G.
Pellegrino, A.
Pepe, M.
Perlt, H.
Pesci, A.
Phong, D.H.
Piclum, J.H.
Pilaftsis, A.
Piotrzkowski, K.
Plefka, J.
Plucinski, P.
Pokrovskiy, N.S.
Polini, A.
Posocco, M.
Pozzorini, S.
Proskuryakov, A.S.
Przybycien, M.
B695 (2004) 3
B695 (2004) 169
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 35
B693 (2004) 3
B695 (2004) 3
B695 (2004) 132
B695 (2004) 199
B692 (2004) 303
B695 (2004) 3
B692 (2004) 110
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B692 (2004) 135
B695 (2004) 3
B695 (2004) 3
QCDSF Collaboration
B693 (2004)
B695 (2004) 3
B695 (2004) 3
B691 (2004) 259
Rakow, P.E.L.
Randjbar-Daemi, S.
Rautenberg, J.
Raval, A.
Reeder, D.D.
Ren, Z.
Renner, R.
Repond, J.
Riederer, S.
Rinaldi, L.
Riser, R.
Riveline, M.
Rizos, J.
Robins, S.
Rosin, M.
Ross, G.G.
Ruan, J.
Rulik, K.
Runkel, I.
Rurua, L.
Ruspa, M.
Ryan, P.
Ryzhov, A.V.
B693 (2004) 3
B692 (2004) 346
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B693 (2004) 149
B695 (2004) 3
B691 (2004) 251
B695 (2004) 3
B695 (2004) 41
B695 (2004) 3
B695 (2004) 3
B692 (2004) 50
B692 (2004) 417
B694 (2004) 239
B694 (2004) 277
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B692 (2004) 3
B695 (2004)
B695 (2004)
Sacchi, R.
Sachrajda, C.T.
B695 (2004) 3
B693 (2004) 103
Longhin, A.
ukasik, J.
Lukina, O.Yu.
Ltken, C.A.
uzniak, P.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 73
B695 (2004) 3
Ma, K.J.
Maddox, E.
Magill, S.
Majumdar, P.
Mankel, R.
Margotti, A.
Marini, G.
Martin, J.F.
Martucci, L.
Mastroberardino, A.
Matsuzawa, K.
Mattingly, M.C.K.
McCubbin, N.A.
McInnes, B.
McLoughlin, T.
Melzer-Pellmann, I.-A.
Menary, S.
Metlica, F.
Meyer, U.
Miglioranzi, S.
Milite, M.
Mirea, A.
Moch, S.
Mohapatra, R.
Monaco, V.
Montanari, A.
Mueller, A.H.
Musgrave, B.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B692 (2004) 209
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B694 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B692 (2004) 270
B694 (2004) 115
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B691 (2004) 129
B695 (2004) 313
B695 (2004) 3
B695 (2004) 3
B692 (2004) 175
B695 (2004) 3
Nagano, K.
Nair, V.P.
Nakayama, R.
Namsoo, T.
Nania, R.
Nguyen, C.N.
Nigro, A.
Ning, Y.
Nooij, S.E.M.
Notari, A.
Notz, D.
Nowak, R.J.
Nuncio-Quiroz, A.E.
B695 (2004) 3
B691 (2004) 182
B693 (2004) 176
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 41
B695 (2004) 169
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
Oh, B.Y.
Olkiewicz, K.
Ossola, G.
Pac, M.Y.
Padhi, S.
3
3
Sadooghi, N.
Salehi, H.
Snchez, F.
Sartorelli, G.
Savin, A.A.
Saxon, D.H.
Schagen, S.
Schierholz, G.
Schiller, A.
Schioppa, M.
Schlenstedt, S.
Schleper, P.
Schmidke, W.B.
Schneekloth, U.
Schnitzer, H.J.
Schnitzer, H.J.
Schrner-Sadenius, T.
Schweigert, C.
Sciulli, F.
Seco, M.
Sezgin, E.
Shaynkman, O.V.
Shcheglova, L.M.
Shen, Z.
Shimono, Y.
Shoshi, A.I.
Sibiryakov, S.M.
Sirvanl,
B.B.
Skillicorn, I.O.
Sominski, W.
Smilga, A.V.
Smith, W.H.
Smolyakov, M.N.
Soares, M.
Soddu, A.
Sokal, A.D.
Solano, A.
Son, D.
Sosnovtsev, V.
Splittorff, K.
Stairs, D.G.
Stanco, L.
Standage, J.
Steinhauser, M.
Stifutkin, A.
Stonjek, S.
Stopa, P.
Stsslein, U.
Straub, P.B.
Strumia, A.
Suchkov, S.
Susinno, G.
Suszycki, L.
Sutiak, J.
B691 (2004) 111

B695 (2004) 3
B695 (2004) 217
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B693 (2004) 3
B693 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 267
B695 (2004) 283
B695 (2004) 3
B694 (2004) 277
B695 (2004) 3
B692 (2004) 83
B692 (2004) 346
B692 (2004) 363
B695 (2004) 3
B692 (2004) 417
B693 (2004) 176
B692 (2004) 175
B691 (2004) 91
B692 (2004) 249
B695 (2004) 3
B695 (2004) 3
B694 (2004) 473
B695 (2004) 3
B695 (2004) 301
B695 (2004) 3
B692 (2004) 83
B691 (2004) 259
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 84
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 199
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 169
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
333
Sutton, M.R.
Swanson, I.
Sztuk, J.
Szuba, D.
Szuba, J.
B695 (2004) 3
B694 (2004) 115
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
Tandler, J.
Tapper, A.D.
Targett-Adams, C.
Tassi, E.
Tawara, T.
Taylor, T.R.
Terrn, J.
Tiecke, H.
Tokushuku, K.
Trigiante, M.
Trigiante, M.
Tseytlin, A.A.
Tsurugai, T.
Turan, G.
Turcato, M.
Tymieniecka, T.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 103
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B693 (2004) 261
B694 (2004) 239
B692 (2004) 3
B695 (2004) 3
B692 (2004) 249
B695 (2004) 3
B695 (2004) 3
Ukleja, A.
Ukleja, J.
Underwood, T.E.J.
B695 (2004) 3
B695 (2004) 3
B692 (2004) 303
Vasiliev, M.A.
Vzquez, M.
Velasco-Sevilla, L.
Veneziano, G.
Verbaarschot, J.J.M.
Vermaseren, J.A.M.
Vernizzi, G.
Vives, O.
Vlasov, N.N.
Vogt, A.
Voss, K.C.
B692 (2004) 363

B695 (2004) 3
B692 (2004) 50
B694 (2004) 206
B695 (2004) 84
B691 (2004) 129
B694 (2004) 59
B692 (2004) 50
B695 (2004) 3
B691 (2004) 129
B695 (2004) 3
Walczak, R.
Walsh, R.
Wang, M.
West, P.
Whitmore, J.J.
Wichmann, K.
Wick, K.
Wiese, U.-J.
Wiese, U.-J.
Wiggers, L.
Wing, M.
Wolf, G.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B693 (2004) 76
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B693 (2004) 149
B694 (2004) 35
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
Yamada, S.
Yamashita, T.
B695 (2004)
B695 (2004)
3
3
334
Yamazaki, Y.
Yang, D.s.
Yelnikov, A.
Ylmaz, N.T.
Ylmaz, U.O.
Yoshida, R.
Youngman, C.
B695 (2004) 3
B692 (2004) 232
B691 (2004) 182
B691 (2004) 223
B692 (2004) 249
B695 (2004) 3
B695 (2004) 3
Zambrana, M.
Zarnecki,
A.F.
B695 (2004)
B695 (2004)
3
3
Zawiejski, L.
Zeuner, W.
ZEUS Collaboration
Zhautykov, B.O.
Zhu, W.
Zichichi, A.
Ziegler, A.
Ziegler, Ar.
Zotkin, S.A.
Zwirner, F.
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B692 (2004) 417
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B695 (2004) 3
B691 (2004) 233

Nucl - Phys.B v.695

Transféré par

Informations du document

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Nucl - Phys.B v.695

Transféré par

Droits d'auteur :

Formats disponibles

Nuclear Physics B 695 (2004) 337

Exclusive electroproduction of J / mesons

P. Antonioli, G. Bari, M. Basile, L. Bellagamba, D. Boscherini,

G. Aghuzumtsyan, D. Bartsch, I. Brock, S. Goers, H. Hartmann,

D.S. Bailey 4 , N.H. Brook, J.E. Cole, G.P. Heath, T. Namsoo,

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

M. Capua, A. Mastroberardino, M. Schioppa, G. Susinno

J.Y. Kim, I.T. Lim, K.J. Ma, M.Y. Pac 5

M. Helbich, Y. Ning, Z. Ren, W.B. Schmidke, F. Sciulli

J. Chwastowski, A. Eskreys, J. Figiel, A. Galas, K. Olkiewicz,

L. Adamczyk, T. Bod, I. Grabowska-Bod 6 , D. Kisielewska,

V. Adler, U. Behrens, I. Bloch, K. Borras, V. Chiochia,

G. Barbagli, E. Gallo, C. Genta, P.G. Pelfer

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

A. Bamberger, A. Benen, F. Karstens, D. Dobur, N.N. Vlasov

M. Bell, P.J. Bussey, A.T. Doyle, J. Ferrando, J. Hamilton,

T. Carli, T. Gosau, U. Holm, N. Krumnack, E. Lohrmann, M. Milite,

C. Collins-Tooth, C. Foudas, R. Gonalo 12 , K.R. Long,

M. Kataoka 13 , K. Nagano, K. Tokushuku 14 , S. Yamada,

A.N. Barakbaev, E.G. Boos, N.S. Pokrovskiy, B.O. Zhautykov

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

F. Barreiro, C. Glasman 15 , O. Gonzlez, L. Labarga, J. del Peso,

M. Barbi, F. Corriveau, S. Gliga, J. Lainesse, S. Padhi, D.G. Stairs,

A. Antonov, P. Danilov, B.A. Dolgoshein, D. Gladkov,

R.K. Dementiev, P.F. Ermolov, I.I. Katkov, L.A. Khein,

I. Abt, C. Bttner, A. Caldwell, X. Liu, J. Sutiak

N. Coppola, S. Grijpink, E. Koffeman, P. Kooijman, E. Maddox,

N. Brmmer, B. Bylsma, L.S. Durkin, T.Y. Ling

A.M. Cooper-Sarkar, A. Cottrell, R.C.E. Devenish, B. Foster,

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

A. Bertolin, R. Brugnera, R. Carlin, F. Dal Corso, S. Dusini,

E.A. Heaphy, F. Metlica, B.Y. Oh, J.J. Whitmore 17

G. DAgostini, G. Marini, A. Nigro

C. Cormack 18 , J.C. Hart, N.A. McCubbin

H. Abramowicz, A. Gabareen, S. Kananov, A. Kreisel, A. Levy

T. Fusayasu, S. Kagawa, T. Tawara, T. Yamashita

R. Hamatsu, T. Hirose 2 , M. Inuzuka, H. Kaji, S. Kitamura 19 ,

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

M. Costa, M.I. Ferrero, V. Monaco, R. Sacchi, A. Solano

T. Koop, J.F. Martin, A. Mirea

J.M. Butterworth 20 , R. Hall-Wilton, T.W. Jones, M.S. Lightwood,

J. Ciborowski 21 , R. Ciesielski 22 , P. uzniak 23 , R.J. Nowak,

Y. Eisenberg, D. Hochman, U. Karshon, M. Riveline

A. Everett, L.K. Gladilin 26 , D. Kira, S. Lammers, L. Li,

S. Bhadra, C.D. Catterall, S. Fourletov, G. Hartner, S. Menary,

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

Received 2 June 2004; accepted 11 June 2004

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

0.35/ E in the hadronic sections, with E in GeV.

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

3. Kinematics and cross sections

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

The virtual photonproton cross section, pJ /p T + L , can be used to

4. Reconstruction and selection of the events

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

5. Monte Carlo simulation

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

6. Extraction of the J / signal

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337

6.1. Non-resonant background

ZEUS Collaboration / Nuclear Physics B 695 (2004) 337