Nucl - Phys.B v.765

Nuclear Physics B 765 (2007) 130
Inclusive-jet and dijet cross sections in deep

inelastic scattering at HERA
ZEUS Collaboration
S. Chekanov, M. Derrick, S. Magill, S. Miglioranzi 1 , B. Musgrave,
D. Nicholass 1 , J. Repond, R. Yoshida
Argonne National Laboratory, Argonne, IL 60439-4815, USA 45
M.C.K. Mattingly
Andrews University, Berrien Springs, MI 49104-0380, USA
N. Pavel , A.G. Yages Molina

Institut fr Physik der Humboldt-Universitt zu Berlin, Berlin, Germany
S. Antonelli, P. Antonioli, G. Bari, M. Basile, L. Bellagamba, M. Bindi,

D. Boscherini, A. Bruni, G. Bruni, L. Cifarelli, F. Cindolo, A. Contin,
M. Corradi 2 , S. De Pasquale, G. Iacobucci, A. Margotti, R. Nania,
A. Polini, L. Rinaldi, G. Sartorelli, A. Zichichi
University and INFN Bologna, Bologna, Italy 36
G. Aghuzumtsyan, D. Bartsch, I. Brock, S. Goers, H. Hartmann,

E. Hilger, H.-P. Jakob, M. Jngst, O.M. Kind, E. Paul 3 , J. Rautenberg 4 ,
R. Renner, U. Samson 5 , V. Schnberg, M. Wang, M. Wlasenko
Physikalisches Institut der Universitt Bonn, Bonn, Germany 33
N.H. Brook, G.P. Heath, J.D. Morris, T. Namsoo

H.H. Wills Physics Laboratory, University of Bristol, Bristol, United Kingdom 44
0550-3213/$ see front matter 2006 Published by Elsevier B.V.

doi:10.1016/j.nuclphysb.2006.09.018
RAPID COMMUNICATION
ZEUS Collaboration / Nuclear Physics B 765 (2007) 130
M. Capua, S. Fazio, A. Mastroberardino, M. Schioppa, G. Susinno,

E. Tassi
Calabria University, Physics Department and INFN, Cosenza, Italy 36
J.Y. Kim 6 , K.J. Ma 7

Chonnam National University, Kwangju, South Korea 38
Z.A. Ibrahim, B. Kamaluddin, W.A.T. Wan Abdullah

Jabatan Fizik, Universiti Malaya, 50603 Kuala Lumpur, Malaysia 49
Y. Ning, Z. Ren, F. Sciulli

Nevis Laboratories, Columbia University, Irvington on Hudson, NY 10027, USA 46
J. Chwastowski, A. Eskreys, J. Figiel, A. Galas, M. Gil, K. Olkiewicz,

P. Stopa, L. Zawiejski
The Henryk Niewodniczanski Institute of Nuclear Physics, Polish Academy of Sciences, Cracow, Poland 40
L. Adamczyk, T. Bod, I. Grabowska-Bod, D. Kisielewska, J. ukasik,

M. Przybycien, L. Suszycki
Faculty of Physics and Applied Computer Science, AGH-University of Science and Technology, Cracow, Poland 47
A. Kotanski 8 , W. Sominski
Department of Physics, Jagellonian University, Cracow, Poland
V. Adler, U. Behrens, I. Bloch, A. Bonato, K. Borras, N. Coppola,

J. Fourletova, A. Geiser, D. Gladkov, P. Gttlicher 9 , I. Gregor,
O. Gutsche, T. Haas, W. Hain, C. Horn, B. Kahle, U. Ktz, H. Kowalski,
H. Lim 10 , E. Lobodzinska, B. Lhr, R. Mankel, I.-A. Melzer-Pellmann,
A. Montanari, C.N. Nguyen, D. Notz, A.E. Nuncio-Quiroz,
R. Santamarta, U. Schneekloth, A. Spiridonov 11 , H. Stadie, U. Stsslein,
D. Szuba 12 , J. Szuba 13 , T. Theedt, G. Watt, G. Wolf, K. Wrona,
C. Youngman, W. Zeuner
Deutsches Elektronen-Synchrotron DESY, Hamburg, Germany
S. Schlenstedt
Deutsches Elektronen-Synchrotron DESY, Zeuthen, Germany
RAPID COMMUNICATION
G. Barbagli, E. Gallo , P.G. Pelfer

University and INFN, Florence, Italy 36
A. Bamberger, D. Dobur, F. Karstens, N.N. Vlasov 14

Fakultt fr Physik der Universitt Freiburg i.Br., Freiburg i.Br., Germany 33
P.J. Bussey, A.T. Doyle, W. Dunne, J. Ferrando, D.H. Saxon,

I.O. Skillicorn
Department of Physics and Astronomy, University of Glasgow, Glasgow, United Kingdom 44
I. Gialas 15
Department of Engineering in Management and Finance, University of Aegean, Greece
T. Gosau, U. Holm, R. Klanner, E. Lohrmann, H. Salehi, P. Schleper,

T. Schrner-Sadenius, J. Sztuk, K. Wichmann, K. Wick
Hamburg University, Institute of Experimental Physics, Hamburg, Germany 33
C. Foudas, C. Fry, K.R. Long, A.D. Tapper

Imperial College London, High Energy Nuclear Physics Group, London, United Kingdom 44
M. Kataoka 16 , T. Matsumoto, K. Nagano, K. Tokushuku 17 , S. Yamada,

Y. Yamazaki
Institute of Particle and Nuclear Studies, KEK, Tsukuba, Japan 37
A.N. Barakbaev, E.G. Boos, A. Dossanov, 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 38
J. de Favereau, K. Piotrzkowski
Institut de Physique Nuclaire, Universit Catholique de Louvain, Louvain-la-Neuve, Belgium 48
F. Barreiro, C. Glasman 18 , M. Jimenez, L. Labarga, J. del Peso, E. Ron,

J. Terrn, M. Zambrana
Departamento de Fsica Terica, Universidad Autnoma de Madrid, Madrid, Spain 43
RAPID COMMUNICATION
F. Corriveau, C. Liu, R. Walsh, C. Zhou

Department of Physics, McGill University, Montral, PQ, Canada, H3A 2T8 32
T. Tsurugai
Meiji Gakuin University, Faculty of General Education, Yokohama, Japan 37
A. Antonov, B.A. Dolgoshein, I. Rubinsky, V. Sosnovtsev, A. Stifutkin,

S. Suchkov
Moscow Engineering Physics Institute, Moscow, Russia 41
R.K. Dementiev, P.F. Ermolov, L.K. Gladilin, I.I. Katkov, L.A. Khein,
I.A. Korzhavina, V.A. Kuzmin, B.B. Levchenko 19 , O.Yu. Lukina,
A.S. Proskuryakov, L.M. Shcheglova, D.S. Zotkin, S.A. Zotkin
Moscow State University, Institute of Nuclear Physics, Moscow, Russia 42
I. Abt, C. Bttner, A. Caldwell, D. Kollar, W.B. Schmidke, J. Sutiak

Max-Planck-Institut fr Physik, Mnchen, Germany
G. Grigorescu, A. Keramidas, E. Koffeman, P. Kooijman, A. Pellegrino,

H. Tiecke, M. Vzquez 20 , L. Wiggers
NIKHEF and University of Amsterdam, Amsterdam, Netherlands 39
N. Brmmer, B. Bylsma, L.S. Durkin, A. Lee, T.Y. Ling

Physics Department, Ohio State University, Columbus, OH 43210, USA 45
P.D. Allfrey, M.A. Bell, A.M. Cooper-Sarkar, A. Cottrell,

R.C.E. Devenish, B. Foster, C. Gwenlan 21 , K. Korcsak-Gorzo,
S. Patel, V. Roberfroid 22 , A. Robertson, P.B. Straub, C. Uribe-Estrada,
R. Walczak
Department of Physics, University of Oxford, Oxford, United Kingdom 44
P. Bellan, A. Bertolin, R. Brugnera, R. Carlin, R. Ciesielski, F. Dal Corso,

S. Dusini, A. Garfagnini, S. Limentani, A. Longhin, L. Stanco,
M. Turcato
Dipartimento di Fisica dell Universit and INFN, Padova, Italy 36
RAPID COMMUNICATION
B.Y. Oh, A. Raval, J. Ukleja 23 , J.J. Whitmore

Department of Physics, Pennsylvania State University, University Park, PA 16802, USA 46
Y. Iga
Polytechnic University, Sagamihara, Japan 37
G. DAgostini, G. Marini, A. Nigro

Dipartimento di Fisica, Universit La Sapienza and INFN, Rome, Italy 36
J.E. Cole, J.C. Hart

Rutherford Appleton Laboratory, Chilton, Didcot, Oxon, United Kingdom 44
H. Abramowicz 24 , A. Gabareen, R. Ingbir, S. Kananov, A. Levy

Raymond and Beverly Sackler Faculty of Exact Sciences, School of Physics, Tel-Aviv University, Tel-Aviv, Israel 35
M. Kuze
Department of Physics, Tokyo Institute of Technology, Tokyo, Japan 37
R. Hori, S. Kagawa 25 , S. Shimizu, T. Tawara

Department of Physics, University of Tokyo, Tokyo, Japan 37
R. Hamatsu, H. Kaji, S. Kitamura 26 , O. Ota, Y.D. Ri

Tokyo Metropolitan University, Department of Physics, Tokyo, Japan 37
M.I. Ferrero, V. Monaco, R. Sacchi, A. Solano

Universit di Torino and INFN, Torino, Italy 36
M. Arneodo, M. Ruspa
Universit del Piemonte Orientale, Novara, and INFN, Torino, Italy 36
S. Fourletov, J.F. Martin

Department of Physics, University of Toronto, Toronto, ON, Canada, M5S 1A7 32
S.K. Boutle 15 , J.M. Butterworth, R. Hall-Wilton 20 , T.W. Jones,

J.H. Loizides, M.R. Sutton 27 , C. Targett-Adams, M. Wing
Physics and Astronomy Department, University College London, London, United Kingdom 44
RAPID COMMUNICATION
B. Brzozowska, J. Ciborowski 28 , G. Grzelak, P. Kulinski, P. uzniak 29 ,

J. Malka 29 , R.J. Nowak, J.M. Pawlak, T. Tymieniecka, A. Ukleja 30 ,
A.F. Zarnecki
Warsaw University, Institute of Experimental Physics, Warsaw, Poland
M. Adamus, P. Plucinski 31
Institute for Nuclear Studies, Warsaw, Poland
Y. Eisenberg, I. Giller, D. Hochman, U. Karshon, M. Rosin

Department of Particle Physics, Weizmann Institute, Rehovot, Israel 34
E. Brownson, T. Danielson, A. Everett, D. Kira, D.D. Reeder, P. Ryan,

A.A. Savin, W.H. Smith, H. Wolfe
Department of Physics, University of Wisconsin, Madison, WI 53706, USA 45
S. Bhadra, C.D. Catterall, Y. Cui, G. Hartner, S. Menary, U. Noor,

M. Soares, J. Standage, J. Whyte
Department of Physics, York University, ON, Canada, M3J 1P3 32
Received 21 August 2006; received in revised form 14 September 2006; accepted 18 September 2006
Available online 18 October 2006
* Corresponding author.
E-mail address: gallo@mail.desy.de (E. Gallo).

1 Also affiliated with University College London, UK.
2 Also at University of Hamburg, Germany, Alexander von Humboldt Fellow.
3 Retired.
4 Now at University of Wuppertal, Germany.
5 Formerly U. Meyer.
6 Supported by Chonnam National University in 2005.
7 Supported by a scholarship of the World Laboratory Bjrn Wiik Research Project.
8 Supported by the research grant No. 1 P03B 04529 (20052008).
9 Now at DESY group FEB, Hamburg, Germany.
10 Now at Argonne National Laboratory, Argonne, IL, USA.
11 Also at Institut of Theoretical and Experimental Physics, Moscow, Russia.
12 Also at INP, Cracow, Poland.
13 On leave of absence from FPACS, AGH-UST, Cracow, Poland.
14 Partly supported by Moscow State University, Russia.
15 Also affiliated with DESY.
16 Now at ICEPP, University of Tokyo, Japan.
17 Also at University of Tokyo, Japan.
18 Ramn y Cajal Fellow.
19 Partly supported by Russian Foundation for Basic Research grant No. 05-02-39028-NSFC-a.
20 Now at CERN, Geneva, Switzerland.
RAPID COMMUNICATION
Abstract
Inclusive-jet and dijet differential cross sections have been measured in neutral current deep inelastic ep
scattering for exchanged boson virtualities Q2 > 125 GeV2 with the ZEUS detector at HERA using an
integrated luminosity of 82 pb1 . Jets were identified in the Breit frame using the kT cluster algorithm.
Jet cross sections are presented as functions of several kinematic and jet variables. The results are also presented in different regions of Q2 . Next-to-leading-order QCD calculations describe the measurements well.
Regions of phase space where the theoretical uncertainties are small have been identified. Measurements in
these regions have the potential to constrain the gluon density in the proton when used as inputs to global
fits of the proton parton distribution functions.
2006 Published by Elsevier B.V.
1. Introduction
A well established testing ground for perturbative QCD (pQCD) is jet production in neutral
current (NC) deep inelastic ep scattering (DIS) at high Q2 , where Q2 is the negative of the square
of the virtuality of the exchanged boson V (V = , Z). The parton-model process (V q q) in
DIS gives rise to final states containing one jet of high transverse energy, the so-called current
and QCD-Comptonjet. At leading order (LO) in s , the bosongluon-fusion (V g q q)
scattering (V q qg) processes give rise to two hard jets.
jet
The hadronic final state in NC DIS may consist of jets with high transverse energy, ET ,
produced in the hard-scattering process, accompanied by the remnant (beam jet) of the incoming
proton. For the analysis of these types of process, the Breit frame [1] is advantageous, since it provides a maximal separation between the products of the beam fragmentation and the hard jets. In
this frame, the exchanged virtual boson is purely space-like, with 3-momentum q = (0, 0, Q),
21 PPARC Postdoctoral Research Fellow.
22 EU Marie Curie Fellow.
23 Partially supported by Warsaw University, Poland.
24 Also at Max Planck Institute, Munich, Germany, Alexander von Humboldt Research Award.
25 Now at KEK, Tsukuba, Japan.
26 Department of Radiological Science.
27 PPARC Advanced fellow.
28 Also at dz University, Poland.
29 dz University, Poland.
30 Supported by the Polish Ministry for Education and Science grant No. 1 P03B 12629.
31 Supported by the Polish Ministry for Education and Science grant No. 1 P03B 14129.
32 Supported by the Natural Sciences and Engineering Research Council of Canada (NSERC).
33 Supported by the German Federal Ministry for Education and Research (BMBF), under contract Nos. HZ1GUA 2,
HZ1GUB 0, HZ1PDA 5, HZ1VFA 5.
34 Supported in part by the MINERVA Gesellschaft fr Forschung GmbH, the Israel Science Foundation (grant No.
293/02-11.2) and the USIsrael Binational Science Foundation.
35 Supported by the GermanIsraeli Foundation and the Israel Science Foundation.
36 Supported by the Italian National Institute for Nuclear Physics (INFN).
37 Supported by the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT) and its grants
for Scientific Research.
38 Supported by the Korean Ministry of Education and Korea Science and Engineering Foundation.
39 Supported by the Netherlands Foundation for Research on Matter (FOM).
RAPID COMMUNICATION
and is collinear with the incoming parton. In the parton-model process, the virtual boson V is
absorbed by the struck quark, which is back-scattered with zero transverse momentum with respect to the V direction, whereas the beam jet follows the direction of the initial struck quark.
QCD-Compton and bosongluon-fusion events, on the other hand, are characterised by two partons in the final state with possibly non-vanishing transverse momentum with respect to the V
direction. Thus, while retaining hard QCD processes at leading order in s , the contribution from
the parton-model process can be suppressed by requiring the production of jets with high transjet
verse energy in the Breit frame. Therefore, measurements of inclusive-jet events at high ET or
dijet events in the Breit frame are directly sensitive to hard QCD processes, allowing direct tests
of the pQCD predictions and of the proton parton distribution functions (PDFs).
Jet cross sections in NC DIS have been measured previously at HERA. Inclusive-jet [24],
dijet [5] and multijet [6] production have been used to extract values of s . A previous dijet
analysis [7] has been used to test the gluon density extracted from global fits. Results from
inclusive-jet and dijet measurements [3] have been used to constrain the gluon density in the
proton. Recently, jet cross sections in NC DIS [2] and photoproduction [8] have been included
in a new NLO QCD analysis to extract the proton PDFs [9], resulting in a significant reduction
of the uncertainty on the gluon density at medium and high momentum fractions.
This paper presents new measurements of differential dijet cross sections as functions of
Q2 , of the mean jet transverse energy of the dijet system in the Breit frame, E T , of the dijet invariant mass, Mjj , of the half-difference of the jet pseudorapidities in the Breit frame,
jet1
jet2
= |B B |/2, and of the fraction of the proton momentum taken by the interacting parton,
= xBj (1 + Mjj2 /Q2 ). Here, xBj is the Bjorken scaling variable that defines, for the parton-model
process, the fraction of proton momentum carried by the struck parton. Measurements of the
dijet cross section as a function of are also shown for different regions of Q2 . Also presented
are measurements of the inclusive-jet cross section as a function of the jet transverse energy in
jet
the Breit frame, ET ,B , in different regions of Q2 . The analyses are based on data samples with
more than twice the statistics of the previous studies [2,7]. These measurements probe an extended kinematic regime with respect to previous analyses due to the increase in the proton beam
energy. The improvement in the experimental uncertainties obtained here will facilitate a more
precise determination of the gluon density in the proton at high .
40 Supported by the Polish State Committee for Scientific Research, grant Nos. 620/E-77/SPB/DESY/P-03/DZ
117/20032005 and 1P03B07427/20042006.
41 Partially supported by the German Federal Ministry for Education and Research (BMBF).
42 Supported by RF Presidential grant No. 1685.2003.2 for the leading scientific schools and by the Russian Ministry
of Education and Science through its grant for Scientific Research on High Energy Physics.
43 Supported by the Spanish Ministry of Education and Science through funds provided by CICYT.
44 Supported by the Particle Physics and Astronomy Research Council, UK.
45 Supported by the US Department of Energy.
46 Supported by the US National Science Foundation.
47 Supported by the Polish Ministry of Scientific Research and Information Technology, grant Nos. 112/E356/SPUB/DESY/P-03/DZ 116/20032005 and 1 P03B 065 27.
48 Supported by FNRS and its associated funds (IISN and FRIA) and by an Inter-University Attraction Poles Programme
subsidised by the Belgian Federal Science Policy Office.
49 Supported by the Malaysian Ministry of Science, Technology and Innovation/Akademi Sains Malaysia grant SAGA
66-02-03-0048.
Deceased.
RAPID COMMUNICATION
2. Experimental set-up
A detailed description of the ZEUS detector can be found elsewhere [10,11]. A brief outline
of the components that are most relevant for this analysis is given below.
Charged particles are tracked in the central tracking detector (CTD) [12], which operates
in a magnetic field of 1.43 T provided by a thin superconducting solenoid. The CTD consists
of 72 cylindrical drift-chamber layers, organised in nine superlayers covering the polar-angle50
region 15 < < 164 . The transverse-momentum resolution for full-length tracks can be parameterised as (pT )/pT = 0.0058pT 0.0065 0.0014/pT , with pT in GeV. The tracking
system was used to measure the interaction vertex with a typical resolution along (transverse to)
the beam direction of 0.4 (0.1) cm and to cross-check the energy scale of the calorimeter.
The high-resolution uranium-scintillator calorimeter (CAL) [13] covers 99.7% of the total
solid angle and consists of three parts: the forward (FCAL), the barrel (BCAL) and the rear
(RCAL) calorimeters. Each part is segmented into towers and subdivided in depth into one
electromagnetic section and either one (in RCAL) or two (in FCAL and BCAL) hadronic sections. The smallest subdivision of the calorimeter is called a cell. Under
test-beam conditions,
E for electrons and
the CAL single-particle
relative
energy
resolutions
were
(E)/E
=
0.18/
(E)/E = 0.35/ E for hadrons, with E in GeV.

The luminosity was measured from the rate of the bremsstrahlung process ep ep. The
resulting small-angle energetic photons were measured by the luminosity monitor [14], a leadscintillator calorimeter placed in the HERA tunnel at Z = 107 m.
3. Data selection and jet search
The data used in this analysis were collected during the period 19982000, when HERA
operated with protons of energy Ep = 920 GeV and electrons or positrons51 of energy Ee =
27.5 GeV, and correspond to an integrated luminosity of 81.7 1.8 pb1 .
A three-level trigger system was used to select events online [11,15]. At the third level, NC
DIS events were accepted on the basis of the identification of a scattered-electron candidate using
localised energy depositions in the CAL. An independent trigger selection which required at least
one (two) jet(s) with transverse energies above 10(6) GeV and pseudorapidities below 2.5 was
used to measure the fraction of scattered electrons that gave a trigger signal. The efficiency of
the trigger selection based on the scattered-electron candidate was found to be above 99%.
Events were selected offline using criteria similar to those reported previously [2]. The main
steps are briefly listed below. The inclusive-jet measurement is based on CAL cells for the reconstruction of the kinematic and jet variables, whereas the dijet analysis uses a combination
of track and CAL information [16]. The selected combinations of tracks and CAL clusters are
referred to as energy flow objects (EFOs).
The scattered-electron candidate was identified from the pattern of energy deposits in the CAL
[17]. The energy (Ee ) and polar angle (e ) of the electron candidate were determined from the
CAL measurements. The kinematic variables Q2 and xBj were reconstructed using the double
50 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. The polar angle, , and the azimuthal angle, , are defined with respect to the
proton beam direction. The pseudorapidity is defined as = ln(tan 2 ).
51 Here and in the following, the term electron denotes generically both the electron and the positron.
RAPID COMMUNICATION
10
angle (DA) method [18]. This method uses e and the angle h , which is equivalent to the angle
of the scattered quark in the quark-parton model. The angle h , defined as
cos h =
(1 y)xBj Ep yEe
,
(1 y)xBj Ep + yEe
was reconstructed using the hadronic final state [18], where y = Q2 /xBj s and s is the centre-ofmass energy.
The following requirements were imposed on the data sample for both the inclusive-jet and
dijet analyses:
an electron candidate of energy Ee > 10 GeV. This requirement ensured a high and well
understood electron-finding efficiency and suppressed background from photoproduction
events, in which the scattered electron escaped down the rear beampipe;
the total energy not associated with the electron candidate within a cone of radius 0.7 units
in the pseudorapidity-azimuth () plane around the electron direction should be less than
10% of the electron energy. This condition removed photoproduction and DIS events in
which part of a jet was falsely identified as the scattered electron;
the vertex position along the beam axis should be in the range |Z| < 34 cm. This condition
helped toselect events consistent with ep interactions;
PT ,miss / ET < 2.5 GeV1/2 , where PT ,miss is the missing transverse momentum as measured with the CAL and ET is the total transverse energy in the CAL. This cut removed
cosmic-ray events and beam-related background;
Q2 > 125 GeV2 ;
0.65 < cos h < 0.65. The lower limit avoided a region with limited acceptance due to
the requirement on the energy of the scattered electron, while the upper limit was chosen to
ensure good reconstruction of the jets in the Breit frame.
In addition, the following selection requirements were imposed for the inclusive-jet analysis:
ye < 0.95, where ye = 1 Ee (1 cos e )/(2Ee ). This condition removed events in which
fake electron candidates were found in the FCAL;
if the polar angle of the scattered electron was in the range 30 < e < 140 , it was required
that the fraction of the electron energy within a cone of radius 0.3 units in the plane
around the electron direction should be larger than 0.9; for e < 30 , the cut was raised to
0.98. This condition removed events in which jets were misidentified as electrons;
the event was not consistent with elastic Compton scattering (ep ep), namely no second
electromagnetic energy cluster above 10 GeV was allowed when the rest of the CAL energy,
besides the two electromagnetic energy clusters, was below 4 GeV;

38 < (E PZ ) < 65 GeV, where
E is the total energy, E = i Ei , and PZ is the Zcomponent of the vector P = i pi . The sums run over all final-state objects. This cut
removed events with large initial-state radiation and further reduced the background from
photoproduction.
For the dijet analysis, the following conditions had to be fulfilled:
Q2 < 5000 GeV2 ;
45 < (E PZ ) < 62 GeV.
RAPID COMMUNICATION
11
The kT cluster algorithm [19] was used in the longitudinally invariant inclusive mode [20] to
reconstruct jets in the hadronic final state both in data and in Monte Carlo (MC) simulated events
(see Section 4) assuming massless objects. In data, the algorithm was applied to the final-state
objects after excluding the scattered-electron candidate. The jet search was performed in the
plane of the Breit frame. The jet variables were defined according to the Snowmass convention
[21].
After reconstructing the jet variables in the Breit frame, the massless four-momenta were
jet
boosted into the laboratory frame, where the transverse energy (ET ,LAB ) and the pseudorapidity
jet
(LAB ) of each jet were calculated. Energy corrections [2,8,22,23] were then applied to the jets
jet
in the laboratory frame and propagated into ET ,B in order to compensate for energy losses in the
inactive material in front of the CAL. The following cuts were applied:
events were removed from the sample if the distance of any of the jets to the electron candidate in the plane of the laboratory frame was smaller than 1 unit, =

jet
jet
(LAB e )2 + (LAB e )2 < 1, where e and e are the azimuthal angle and pseudorapidity of the scattered electron, respectively. This requirement removed some background
from photoproduction and improved the purity of the sample;
events were removed from the sample if a jet was in the backward region of the detector
jet
(LAB < 2). This requirement removed events in which a radiated photon from the electron
was misidentified as a jet in the Breit frame;
jet
2 < B < 1.5. This cut restricted the jets to a region with large acceptance.
The inclusive-jet sample was then selected using the following conditions:
jet
ET ,LAB > 2.5 GeV. This cut removed a small number of jets for which the uncertainty on
the energy correction was large;
jet
ET ,B > 8 GeV.
For the dijet analysis, the following further requirements were imposed:
jet
ET ,LAB > 3 GeV;

jet
LAB < 2.5. This condition restricted the jets to the region in which the reconstruction of the
jet variables was optimal;
jet
jet
the two highest-ET ,B jets in an event, ordered according to ET ,B , were required to satisfy
jet1
jet2
ET ,B > 12 GeV and ET ,B > 8 GeV.

The selected sample of inclusive-jet (dijet) events consisted of 19908 (3868) events.
4. Monte Carlo simulation and acceptance corrections
Samples of MC events were generated to determine the response of the detector to jets of
hadrons and the correction factors necessary to obtain the hadron-level jet cross sections. The
hadron level is defined in terms of hadrons with lifetime 10 ps. The generated events were
RAPID COMMUNICATION
12
passed through the G EANT 3.13-based [24] ZEUS detector- and trigger-simulation programs
[11]. They were reconstructed and analysed by the same program chain as the data.
Neutral current DIS events including radiative effects were simulated using the H ERA CLES 4.6.1 [25] program with the D JANGOH 1.1 [26] interface to the hadronisation programs.
H ERACLES includes corrections for initial- and final-state radiation, vertex and propagator terms,
and two-boson exchange. The QCD cascade is simulated using the colour-dipole model (CDM)
[27] including the LO QCD diagrams as implemented in A RIADNE 4.08 [28] and, alternatively,
with the MEPS model of L EPTO 6.5 [29]. The CTEQ5D [30] proton PDFs were used for these
simulations. Fragmentation into hadrons is performed using the Lund string model [31] as implemented in J ETSET 7.4 [32].
The jet search was performed on the MC events using the energy measured in the CAL
cells or EFOs in the same way as for the data. The same jet algorithm was also applied to the
final-state particles (hadron level) and to the partons available after the parton shower (parton
level).
The data were corrected to the hadron level and for QED-radiative effects using bin-by-bin
correction factors obtained from the MC samples. For this approach to be valid, the uncorrected
distributions of the data must be well described by the MC simulations. This condition was in
general satisfied by both the A RIADNE and L EPTO MC. The A RIADNE model gave a slightly
better description of the inclusive-jet data and was thus used as the default model; L EPTO was
then used to estimate the systematic effect on the correction procedure due to the parton-shower
model (see Section 6). In contrast to this, the dijet distributions, shown in Fig. 1, were slightly
better described by the L EPTO model; here, A RIADNE was therefore used for systematic checks.
In all cases, the correction factors differed from unity by typically 10%. These correction factors
took into account the efficiency of the trigger, the selection criteria and the purity and efficiency
of the jet reconstruction.
5. NLO QCD calculations
The measurements were compared with NLO QCD (O(s2 )) calculations obtained using the
program D ISENT [33]. The calculations were performed in the MS renormalisation and factorisation schemes using a generalised version [33] of the subtraction method [34]. The number
of flavours was set to five and the factorisation scale was chosen to be F = Q. Calculations
with different choices of the renormalisation scale, R , were performed: the default choice was
jet
2R = (ET ,B )2 for the inclusive-jet analysis; cross checks were performed using 2R = Q2 and
2R = Q2 + E T2 ,B , where E T2 ,B is the mean transverse energy of the selected jets in an event. For
the dijet analysis 2R = Q2 + E T2 was used as default, since the resulting predictions describe the
data better than either scale alone; here the effects of using either Q2 or E T2 were investigated.
The effects on the NLO calculations of using different choices for R were found to be smaller
than the uncertainties described below. The strong coupling constant was calculated at two loops
(5)
with = 226 MeV, corresponding to s (MZ ) = 0.118. The calculations were performed usMS
ing the CTEQ6 [35] parameterisations of the proton PDFs. The kT cluster algorithm was also
applied to the partons in the events generated by D ISENT in order to obtain the jet cross section
predictions.
Since the measurements refer to jets of hadrons, whereas the NLO QCD calculations refer
to jets of partons, the predictions were corrected to the hadron level using the MC models. The
multiplicative correction factor (Chad ) was defined as the ratio of the cross section for jets of
RAPID COMMUNICATION
13
jet1
jet2
jet
Fig. 1. Uncorrected data distributions for dijet production with ET ,B > 12 GeV, ET ,B > 8 GeV and 2 < B < 1.5 in
the kinematic range given by | cos h | < 0.65 and 125 < Q2 < 5000 GeV2 (dots). For comparison, the predictions of the
ARIADNE (dashed histograms) and LEPTO (solid histograms) MC models are also included.
hadrons over that for jets of partons, estimated by using the MC programs described in Section 4. The ratios obtained with A RIADNE and L EPTO were averaged to obtain the value of Chad .
This value differs from unity by approximately 5 (510)% for the inclusive-jet (dijet) analysis.
The NLO QCD predictions for the inclusive-jet analysis were also corrected for the Z 0 exchange contribution by using MC simulated events with and without Z 0 exchange. This
correction is not required for the dijet analysis owing to the restricted Q2 range.
RAPID COMMUNICATION
14
jet
Fig. 2. Overview of theoretical uncertainties for the inclusive-jet analysis as functions of ET ,B in bins of Q2 for
jet
| cos h | < 0.65 and 2 < B < 1.5. Shown are the relative uncertainties induced by the variation of the renormalisation scale R (hatched area) and by the uncertainties on the proton PDFs (shaded area).
Several sources of uncertainty in the theoretical predictions were considered:

the uncertainty on the NLO QCD calculations due to terms beyond NLO, estimated by varyjet
ing R by a factor of two up and down, was below 7% at low Q2 and low ET ,B and
decreased to below 5% in the highest-Q2 region for the inclusive-jet analysis. For the dijet
measurement, the average uncertainty was about 10%, with the largest values approaching
20% at low and low Q2 values;
the uncertainty on the NLO QCD calculations due to that on s (MZ ) was estimated by
repeating the calculations using two additional sets of proton PDFs, CTEQ6A114 and
RAPID COMMUNICATION
15
Fig. 3. Overview of theoretical uncertainties for the dijet analysis as functions of log10 in bins of Q2 for | cos h | < 0.65,
jet1
jet2
jet
ET ,B > 12 GeV, ET ,B > 8 GeV and 2 < B < 1.5. Other details as in the caption to Fig. 2.
CTEQ6A122 [36], determined assuming s (MZ ) = 0.114 and 0.122, respectively. The difference between the calculations using these sets and CTEQ6 was scaled by a factor of 0.68
to reflect the uncertainty on s [37]. The resulting uncertainty in the cross sections was
typically below 4%;
the uncertainty on the modelling of the parton shower was estimated as half the difference between the multiplicative correction factors calculated from the L EPTO and
RAPID COMMUNICATION
16
jet
Fig. 4. NLO QCD predictions of the gluon-induced fraction of the inclusive-jet cross sections d/dET ,B as functions of
jet
jet
ET ,B in different regions of Q2 for | cos h | < 0.65 and 2 < B < 1.5 (shaded histograms). The CTEQ6 parameterisations of the proton PDFs were used.
A RIADNE models. The resulting uncertainty on the cross sections was typically less
than 3%;
the uncertainty on the NLO calculations due to the proton PDFs was estimated by
repeating the calculations using 40 additional sets from the CTEQ6 analysis, which
takes into account the statistical and correlated systematic experimental uncertainties
of each data set used in the determination of the proton PDFs. The resulting uncertainty in the cross sections was about 4% at low Q2 and decreased to around 2% at
high Q2 ;
the uncertainty of the calculations in the value of F was estimated by repeating the
calculations with F = Q/2 and 2Q. The variation of the calculations was negligible.
RAPID COMMUNICATION
17
Fig. 5. NLO QCD predictions of the gluon-induced fraction of the dijet cross sections d/d log10 as functions of
jet1
jet2
jet
log10 in different regions of Q2 for | cos h | < 0.65, ET ,B > 12 GeV, ET ,B > 8 GeV and 2 < B < 1.5. Other
details as in the caption to Fig. 4.
The total theoretical uncertainty was obtained by adding in quadrature the individual uncertainties listed above. Figs. 2 and 3 show an overview of various theoretical uncertainties for both
the inclusive-jet and the dijet analysis.
jet
The gluon-induced contribution to the cross section as a function of ET ,B for the inclusive-jet
analysis and of log10 for the dijet analysis in various regions of Q2 is shown in Figs. 4 and 5.
jet
The gluon fraction varies between 70% at low Q2 for both low and low ET ,B and less than 10%
jet
for the lowest ET ,B at high values of Q2 . For the dijet analysis, the gluon contribution is always at
least 30%. In addition, Fig. 6 shows, separately for the inclusive-jet and dijet analyses, the gluon-
RAPID COMMUNICATION
18
Fig. 6. NLO QCD predictions of the gluon-induced fraction of the dijet (inclusive-jet) cross section d/d log
jet
jet
(d/dET ,B ) as a function of Q2 for | cos h | < 0.65 and 2 < B < 1.5 and for log10 = 1.25
jet
(10 < ET ,B < 14 GeV); the uncertainty on the gluon-induced fraction due to those on the proton PDFs is represented by
the shaded (hatched) area. The CTEQ6 parameterisations of the proton PDFs were used.
induced contribution to the cross section as a function of Q2 together with the uncertainties as
taken from the CTEQ6 analysis.
6. Experimental uncertainties
The following sources of systematic uncertainty were considered for the measured cross sections:
jet
the uncertainty in the absolute energy scale of the jets was estimated to be 1% for ET ,LAB >
jet
10 GeV and 3% for lower ET ,LAB values [8,22,38]. The resulting uncertainty on the cross
sections was about 5% and increased to approximately 10% in certain regions of the dijet
phase space;
the uncertainty in the absolute energy scale of the electron candidate was estimated to be
1% [39]. The resulting uncertainty was below 1%;
the differences in the results obtained by using either A RIADNE or L EPTO to correct the data
for detector effects were typically below 8%;
RAPID COMMUNICATION
19
jet1
Fig. 7. The measured differential cross sections (a) d/dQ2 and (b) d/dxBj for dijet production with ET ,B > 12 GeV,
jet2
jet
ET ,B > 8 GeV and 2 < B < 1.5 (dots), in the kinematic range given by | cos h | < 0.65 and 125 < Q2 < 5000 GeV2 .
The inner error bars represent the statistical uncertainty. The outer error bars show the statistical and systematic uncertainties, not associated with the uncertainty in the absolute energy scale of the jets, added in quadrature. The shaded bands
display the uncertainties due to the absolute energy scale of the jets. The NLO QCD calculations with 2R = Q2 + E T2
(solid lines), 2R = Q2 (dashed lines) and 2R = E T2 (dash-dotted lines), corrected for hadronisation effects and using the CTEQ6 parameterisations of the proton PDFs, are also shown. The lower parts of the figures show the relative
differences with respect to the NLO QCD calculations with 2R = Q2 + E T2 : measurements (dots), NLO QCD calculations with 2R = Q2 (dashed lines) and with 2R = E T2 (dotted lines); the hatched bands display the total theoretical
uncertainty.
the analysis was repeated using an alternative technique [40] to select the scattered-electron
candidate. The resulting uncertainty was typically below 3%;
jet
the ET ,LAB cut was raised to 4(4.5) GeV for the inclusive-jet (dijet) analysis. The resulting
uncertainty was typically smaller than 1%;
jet
the cut in LAB used to suppress the contamination due to photons falsely identified as jets
in the Breit frame was changed to 3 and to 1.5. The resulting uncertainty was typically
below 1%;
the uncertainty due to the selection cuts was estimated by varying the values of the cuts
within the resolution of each variable. The effect on the cross sections was typically below
2%;
the uncertainty on the cross sections due to that in the simulation of the trigger was negligible.
The systematic uncertainties not associated with the absolute energy scale of the jets were
added in quadrature to the statistical uncertainties and are shown in Figs. 712 as error bars. The
uncertainty due to the absolute energy scale of the jets is shown separately as a shaded band in
RAPID COMMUNICATION
20
Fig. 8. The measured differential cross sections (a) d/d E T , (b) d/dMjj , (c) d/d and (d) d/d log10 for dijet1
jet2
jet
jet production with ET ,B > 12 GeV, ET ,B > 8 GeV and 2 < B < 1.5 (dots), in the kinematic range given by
| cos h | < 0.65 and 125 < Q2 < 5000 GeV2 . Other details as in the caption to Fig. 7.
each of these figures, due to the large bin-to-bin correlation. In addition, there was an overall
normalisation uncertainty of 2.2% from the luminosity determination, which is not included in
the figures.
RAPID COMMUNICATION
21
Table 1
Dijet cross section d/dQ2 for jets of hadrons in the Breit frame selected with the longitudinally invariant kT cluster
algorithm. The statistical, uncorrelated systematic and jet-energy-scale (ES) uncertainties are shown separately. The multiplicative corrections applied to the data to correct for QED radiative effects, CQED , and the corrections for hadronisation
effects to be applied to the parton-level NLO QCD calculations, Chad , are shown in the last two columns
Q2 bin (GeV2 )
d/dQ2 (pb/GeV2 )
stat
syst
125250
0.1183
0.0033
250500
0.0589
0.0018
5001000
0.02061
0.00074
10002000
0.00602
0.00028
20005000
0.001189
0.000075
+0.0041
0.0040
+0.0025
0.0023
+0.00094
0.00094
+0.00018
0.00018
+0.000083
0.000083
ES
+0.0081
0.0077
+0.0032
0.0030
+0.00096
0.00083
+0.00014
0.00013
+0.000025
0.000020
CQED
Chad
0.96
0.85
0.94
0.91
0.93
0.92
0.91
0.95
0.96
0.96
Table 2
Dijet cross section d/dxBj for jets of hadrons in the Breit frame selected with the longitudinally invariant kT cluster
algorithm. Other details as in the caption to Table 1
xBj bin
d/dxBj (pb)
stat
syst
+54
52
+37
37
+26
25
+4.0
4.0
+3.3
3.3
0.00010.01
1599
45
0.010.02
1849
48
0.020.035
647
24
0.0350.07
141.7
7.5
0.070.1
21.1
3.2
ES
+101
92
+98
95
+26
23
+2.9
2.6
+0.5
0.1
CQED
Chad
0.95
0.86
0.94
0.93
0.92
0.92
0.95
0.91
0.89
0.90
Table 3
Dijet cross section d/d E T for jets of hadrons in the Breit frame selected with the longitudinally invariant kT cluster
E T bin (GeV)
d/d E T (pb/GeV)
stat
1016
5.71
0.11
1622
1.935
0.064
2230
0.361
0.023
3060
0.0371
0.0043
syst
ES
+0.12
0.12
+0.086
0.084
+0.023
0.023
+0.0076
0.0076
+0.30
0.26
+0.081
0.091
+0.017
0.016
+0.0021
0.0018
CQED
Chad
0.94
0.90
0.94
0.90
0.94
0.91
0.97
0.91
7. Results
7.1. Dijet differential cross sections
Dijet cross sections were measured in the kinematic region 125 < Q2 < 5000 GeV2 and
jet
| cos h | < 0.65. These cross sections correspond to the two highest-ET ,B jets in each event with
jet1
jet2
jet
ET ,B > 12 GeV, ET ,B > 8 GeV and 2 < B < 1.5 and were corrected for detector and QED
radiative effects as described in Section 4.
The measured cross sections as functions of Q2 , xBj , E T , Mjj , and log10 are shown in
Figs. 7 and 8 and are listed in Tables 16. In these and subsequent figures, each data point is
plotted at the abscissa at which the differential cross section was equal to its bin-averaged value,
according to the NLO QCD calculation. The data distribution as a function of Q2 exhibits a
RAPID COMMUNICATION
22
Table 4
Dijet cross section d/dMjj for jets of hadrons in the Breit frame selected with the longitudinally invariant kT cluster
Mjj bin (GeV)
d/dMjj (pb/GeV)
stat
2032
2.382
0.051
3245
1.134
0.034
4565
0.222
0.012
65120
0.0180
0.0021
syst
ES
+0.055
0.054
+0.047
0.047
+0.006
0.006
+0.0017
0.0017
+0.123
0.109
+0.049
0.055
+0.009
0.009
+0.0013
0.0009
CQED
Chad
0.95
0.91
0.94
0.89
0.92
0.89
0.97
0.94
Table 5
Dijet cross section d/d for jets of hadrons in the Breit frame selected with the longitudinally invariant kT cluster
bin
d/d (pb)
stat
syst
ES
+4.0
3.9
+1.6
1.6
+1.3
1.1
+1.0
1.0
+0.54
0.52
+3.7
3.7
+3.6
3.7
+3.5
3.0
+2.5
2.2
+0.37
0.34
00.1
87.9
3.4
0.10.25
87.0
2.8
0.250.45
68.5
2.1
0.450.65
42.7
1.6
0.651.60
6.07
0.28
CQED
Chad
0.95
0.91
0.94
0.94
0.95
0.91
0.94
0.87
0.91
0.86
Table 6
Dijet cross section d/d log10 for jets of hadrons in the Breit frame selected with the longitudinally invariant kT cluster
log10 bin
d/d log10 (pb)
stat
syst
ES
+0.72
0.71
+2.8
2.8
+2.2
2.0
+2.4
2.4
+0.48
0.48
+1.34
1.17
+3.2
3.4
+3.4
3.2
+1.5
1.4
+0.32
0.21
21.5
22.22
0.78
1.51.35
74.4
2.6
1.351.1
73.9
1.9
1.10.85
31.0
1.3
0.850.5
3.93
0.36
CQED
Chad
0.95
0.88
0.94
0.92
0.94
0.92
0.94
0.87
0.93
0.79
fall-off of more than two orders of magnitude within the range studied. The Q2 -restricted range
implicitly limits the range of the measurements to 0.009 0.37, a region of phase space
where these measurements can significantly constrain the gluon density in the proton. The dijet
cross sections as functions of E T and Mjj are particularly suited to test the matrix elements in
the perturbative calculations. The measured distributions show a steep fall-off of more than two
orders of magnitude within the measured range.
The NLO QCD calculations with 2R = Q2 + E T2 are compared to the data in Figs. 7 and 8.
They give a good description of the data. The calculations with 2R = Q2 or 2R = E T2 are included in Figs. 7 and 8 as well. At low Q2 , low Mjj and low E T , the theoretical uncertainty is
dominated by the effect of the variation of the renormalisation scale. In the case of the dijet cross
sections as functions of E T and Mjj , the theoretical uncertainty is smaller than the experimental
at high values of E T and Mjj . The excellent agreement between data and theory for these distrib-
RAPID COMMUNICATION
23
jet1
jet2
Fig. 9. The measured differential cross section d/d log10 for dijet production with ET ,B > 12 GeV, ET ,B > 8 GeV
jet
and 2 < B < 1.5 in different regions of Q2 (dots). Other details as in the caption to Fig. 7.
utions demonstrates the validity of the description of the dynamics of dijet production by pQCD
at O(s2 ).
7.2. Dijet and inclusive-jet differential cross sections in different regions of Q2
Fig. 9 shows the measured dijet cross section as a function of log10 in different regions
of Q2 . The cross sections are also given in Table 7. The requirement that two jets with high
transverse energy be observed in the final state suppresses the cross section in the low- re-
RAPID COMMUNICATION
24
Table 7
Dijet cross sections d/d log10 for jets of hadrons in the Breit frame selected with the longitudinally invariant kT
cluster algorithm in different regions of Q2 . Other details as in the caption to Table 1
log10 bin
d/d log10 (pb)
stat
syst
ES
CQED
Chad
0.97
0.83
0.95
0.89
0.95
0.86
0.99
0.83
0.94
0.93
0.94
0.93
0.93
0.89
0.94
0.88
0.94
0.92
0.94
0.94
0.91
0.92
0.98
0.86
0.93
0.94
0.94
0.98
0.92
0.96
0.85
0.89
0.92
0.99
1.00
0.96
0.93
0.94
125 < Q2 < 250 GeV2

2.001.50
9.17
0.48
1.501.35
25.0
1.4
1.351.10
19.74
0.93
1.100.50
2.52
0.20
+0.44
0.44
+1.8
1.8
+0.43
0.41
+0.10
0.10
+0.70
0.62
+1.4
0.7
+1.17
1.65
+0.26
0.19
250 < Q2 < 500 GeV2

2.001.50
8.65
0.49
1.501.30
23.4
1.2
1.301.00
16.18
0.83
1.000.50
1.77
0.21
+0.59
0.57
+1.5
1.4
+0.55
0.44
+0.09
0.09
+0.42
0.38
+1.1
1.3
+1.02
0.78
+0.10
0.11
500 < Q2 < 1000 GeV2

1.901.50
4.32
0.40
1.501.20
17.50
0.88
1.200.90
9.45
0.63
0.900.60
1.48
0.26
+0.15
0.18
+1.00
1.00
+0.90
0.90
+0.09
0.09
+0.23
0.17
+0.72
0.75
+0.50
0.35
+0.06
0.06
1000 < Q2 < 2000 GeV2

1.701.40
2.72
0.37
1.401.20
10.98
0.86
1.201.00
10.37
0.84
1.000.60
2.17
0.25
+0.33
0.33
+1.23
1.23
+0.48
0.48
+0.15
0.15
+0.06
0.12
+0.22
0.01
+0.26
0.28
+0.06
0.08
2000 < Q2 < 5000 GeV2

1.601.20
2.14
0.26
1.201.00
8.28
0.80
1.000.60
2.48
0.30
+0.14
0.14
+0.54
0.54
+0.16
0.16
+0.01
0.02
+0.14
0.18
+0.08
0.04
gion, so that the measured cross section rises as log10 increases. At high values of log10 ,
the decrease of the cross section reflects the decrease of the gluon and quark densities at
high .
The NLO QCD predictions with 2R = Q2 + E T2 are compared to the measurements in Fig. 10
using the relative difference of the measured differential cross sections to the NLO calculations;
the uncertainty of the calculation is also shown in the figure. The predictions of NLO QCD give
a good description of the data. Also indicated are the predictions using 2R = Q2 or 2R = E T2 .
Inclusive-jet cross sections were measured in the kinematic region Q2 > 125 GeV2 and
jet
| cos h | < 0.65. These cross sections include every jet of hadrons in the event with ET ,B >
jet
8 GeV and 2 < B < 1.5 and were corrected for detector and QED radiative effects. The cross
jet
sections for different regions of Q2 as a function of ET ,B are presented in Fig. 11 and Table 8.
RAPID COMMUNICATION
25
Fig. 10. Relative differences between the measured differential cross sections d/d log10 presented in Fig. 9 and the
NLO QCD calculations with 2R = Q2 + E T2 (dots). The relative differences between the NLO QCD calculations with
2R = Q2 (2R = E T2 ) and those with 2R = Q2 + E T2 are also shown as dashed (dotted) lines. Other details as in the
caption to Fig. 7.
jet
The measured cross sections exhibit a steep fall-off within the ET ,B range considered. As Q2
jet
increases, the ET ,B dependence of the cross section becomes less steep.

jet
The NLO QCD predictions with 2R = (ET ,B )2 are compared to the measurements in Fig. 11.
They give a good description of the data. To study the scale dependence, calculations using 2R =
Q2 or 2R = Q2 + E T2 ,B are also compared to the data in Fig. 11; they provide an approximately
equally good description of the data.
RAPID COMMUNICATION
26
jet
jet
Fig. 11. The measured differential cross section d/dET ,B for inclusive-jet production with ET ,B > 8 GeV and
jet
jet
2 < B < 1.5 in different regions of Q2 (dots). The NLO QCD calculations with 2R = (ET ,B )2 (solid lines),
2R = Q2 (dashed lines) and 2R = Q2 + E T2 ,B (dash-dotted lines), corrected for hadronisation and Z 0 effects and
using the CTEQ6 parameterisations of the proton PDFs, are also shown. Other details as in the caption to Fig. 7.
Fig. 12 shows the relative difference of the measured differential cross sections to the NLO
jet
QCD calculations with 2R = (ET ,B )2 . The uncertainty of the NLO QCD calculations and the
results of the theory calculations using either 2R = Q2 or 2R = Q2 + E T2 ,B are also shown. The
data are well described by the predictions.
As indicated in Section 5, the theoretical uncertainties in both the inclusive-jet and dijet
regimes are dominated by the contribution from higher orders as estimated by varying the renor-
RAPID COMMUNICATION
27
Table 8
jet
Inclusive jet cross sections d/dET ,B for jets of hadrons in the Breit frame selected with the longitudinally invariant kT
cluster algorithm in different regions of Q2 . Other details as in the caption to Table 1
jet
ET ,B bin (GeV)
jet
d/dET ,B (pb/GeV)
stat
syst
ES
CQED
Chad
0.96
0.90
0.98
0.94
0.97
0.94
0.94
0.93
0.97
0.86
0.94
0.92
0.96
0.95
0.96
0.97
0.92
0.97
0.95
0.91
0.96
0.91
0.94
0.95
0.93
0.98
0.96
0.99
0.96
0.97
0.93
0.93
0.91
0.95
0.98
0.99
0.93
0.99
1.00
1.00
0.92
0.90
0.90
0.93
1.02
1.00
0.96
1.00
0.91
0.99
0.99
0.93
0.96
0.93
0.97
0.98
0.98
0.99
0.99
1.00
125 < Q2 < 250 GeV2

810
32.97
1014
13.00
1418
3.71
1825
0.835
25100
0.0160
810
18.40
1014
8.74
1418
3.30
1825
0.889
25100
0.0242
810
8.79
1014
4.69
1418
2.239
1825
0.701
+1.21
1.21
+0.19
0.22
0.19
+0.15
0.11
0.15
+0.013
0.037
0.012
+0.0027
0.0014
0.0027
250 < Q2 < 500 GeV2
+1.81
1.69
+0.79
0.75
+0.28
0.24
+0.051
0.056
+0.0010
0.0011
+0.74
0.74
+0.30
0.19
0.30
+0.15
0.11
0.15
+0.041
0.042
0.041
+0.0005
0.0020
0.0005
500 < Q2 < 1000 GeV2
+0.68
0.60
+0.33
0.35
+0.18
0.14
+0.052
0.057
+0.0012
0.0011
+0.34
0.34
+0.19
0.14
0.19
+0.137
0.093
0.137
+0.051
0.039
0.051
+0.0018
0.0027
0.0018
1000 < Q2 < 2000 GeV2
+0.26
0.15
+0.11
0.13
+0.091
0.074
+0.026
0.026
+0.0019
0.0019
+0.14
0.14
+0.077
0.091
0.077
+0.056
0.069
0.056
+0.039
0.034
0.039
+0.0043
0.0026
0.0043
2000 < Q2 < 5000 GeV2
+0.09
0.08
+0.017
0.037
+0.034
0.008
+0.009
0.018
+0.0020
0.0013
+0.120
0.120
+0.024
0.060
0.024
+0.070
0.052
0.070
+0.028
0.024
0.028
+0.0023
0.0021
0.0023
5000 < Q2 < 100000 GeV2
+0.033
0.022
+0.007
0.006
+0.023
0.017
+0.009
0.006
+0.0004
0.0005
+0.091
0.091
+0.022
0.023
+0.017
0.017
+0.019
0.019
+0.0042
0.0042
+0.011
0.006
+0.003
0.019
+0.003
0.005
+0.001
0.000
+0.0003
0.0001
0.49
0.38
0.26
25100
0.0335
810
3.30
1014
1.985
1418
1.115
1825
0.492
25100
0.0263
810
1.292
1014
0.858
1418
0.612
1825
0.242
25100
0.0185
810
0.225
0.037
1014
0.267
0.037
1418
0.122
0.024
0.16
0.095
1825
0.070
0.013
25100
0.0114
0.0022
RAPID COMMUNICATION
28
jet
Fig. 12. Relative differences between the measured differential cross sections d/dET ,B presented in Fig. 11 and the
jet
NLO QCD calculations with 2R = (ET ,B )2 (dots). The relative differences between the NLO QCD calculations with
jet
2R = Q2 (2R = Q2 + E T2 ,B ) and those with 2R = (ET ,B )2 are also shown as dashed (dash-dotted) lines. Other details
as in the caption to Fig. 7.
malisation scale. As was shown in Figs. 2 and 3, the scale uncertainty decreases as Q2 increases;
it is also lower for high values of than for low values of this variable. The contribution from the
PDF uncertainty is approximately constant in all variables and non-negligible. Especially at high
jet
values of (ET ,B ) for the dijet (inclusive-jet) analysis, the PDF uncertainty is even dominant.
In these regions, in which the gluon-induced contribution is still substantial (see Figs. 46), the
data will be able to constrain further the gluon density in the proton.
RAPID COMMUNICATION
29
8. Summary and conclusions

2
Inclusive-jet and dijet cross sections have been measured at high Q values at HERA for
s = 318 GeV. The jets were reconstructed using the kT cluster algorithm in the longitudinally
invariant inclusive mode in the Breit frame. The dijet cross sections were measured as functions
of Q2 , xBj , E T , Mjj , and . In addition, inclusive-jet and dijet measurements were performed
jet
in different regions of Q2 as functions of ET ,B and , respectively. The data are well described
by the NLO QCD predictions. The cross sections in different regions of Q2 are shown to be
sensitive to the gluon density of the proton. The precise measurements obtained here are therefore
of particular relevance for improving the determination of the gluon density in future QCD fits.
Acknowledgements
We thank the DESY Directorate for their strong support and encouragement. The remarkable
achievements of the HERA machine group were essential for the successful completion of this
work and are greatly appreciated. We are grateful for the support of the DESY computing and
network services. The design, construction and installation of the ZEUS detector have been made
possible owing to the ingenuity and effort of many people from DESY and home institutes who
are not listed as authors.
References
[1] R.P. Feynman, PhotonHadron Interactions, Benjamin, New York, 1972;
K.H. Streng, T.F. Walsh, P.M. Zerwas, Z. Phys. C 2 (1979) 237.
[2] ZEUS Collaboration, S. Chekanov, et al., Phys. Lett. B 547 (2002) 164.
[3] H1 Collaboration, C. Adloff, et al., Eur. Phys. J. C 19 (2001) 289.
[4] H1 Collaboration, C. Adloff, et al., Phys. Lett. B 542 (2002) 193.
[5] ZEUS Collaboration, J. Breitweg, et al., Phys. Lett. B 507 (2001) 70.
[6] ZEUS Collaboration, S. Chekanov, et al., Eur. Phys. J. C 44 (2005) 183.
[10] ZEUS Collaboration, M. Derrick, et al., Phys. Lett. B 293 (1992) 465.
[11] ZEUS Collaboration, U. Holm (Ed.), The ZEUS Detector, status report, unpublished, DESY, 1993, available on
http://www-zeus.desy.de/bluebook/bluebook.html.
[12] 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.
[13] 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.
[14] J. Andruszkw, et al., preprint DESY-92-066, DESY, 1992;
ZEUS Collaboration, M. Derrick, et al., Z. Phys. C 63 (1994) 391;
J. Andruszkw, et al., Acta Phys. Pol. B 32 (2001) 2025.
[15] W.H. Smith, K. Tokushuku, L.W. Wiggers, in: C. Verkerk, W. Wojcik (Eds.), Proceedings of Computing in HighEnergy Physics (CHEP), Annecy, France, September 1992, CERN, Geneva, Switzerland, 1992, p. 222. Also in
preprint DESY 92-150B.
[16] G.M. Briskin, Diffractive Dissociation in ep Deep Inelastic Scattering, Ph.D. Thesis, Tel Aviv University, DESYTHESIS-1998-036, 1998.
[17] H. Abramowicz, A. Caldwell, R. Sinkus, Nucl. Instrum. Methods A 365 (1995) 508;
R. Sinkus, T. Voss, Nucl. Instrum. Methods A 391 (1997) 360.
RAPID COMMUNICATION
30
[18] S. Bentvelsen, J. Engelen, P. Kooijman, in: W. Buchmller, G. Ingelman (Eds.), Proceedings of the Workshop on
Physics at HERA, vol. 1, DESY, Hamburg, Germany, 1992, p. 23;
K.C. Hger, in: W. Buchmller, G. Ingelman (Eds.), Proceedings of the Workshop on Physics at HERA, vol. 1,
DESY, Hamburg, Germany, 1992, p. 43.
[19] S. Catani, et al., Nucl. Phys. B 406 (1993) 187.
[20] S.D. Ellis, D.E. Soper, Phys. Rev. D 48 (1993) 3160.
[21] J.E. Huth, et al., in: E.L. Berger (Ed.), Research Directions for the Decade, Proceedings of Summer Study on High
Energy Physics, vol. 1990, World Scientific, 1992, p. 134. Also in preprint FERMILAB-CONF-90-249-E.
[24] R. Brun, et al., G EANT 3, Technical Report CERN-DD/EE/84-1, CERN, 1987.
[25] A. Kwiatkowski, H. Spiesberger, H.-J. Mhring, Comput. Phys. Commun. 69 (1992) 155;
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.
[26] K. Charcua, G.A. Schuler, H. Spiesberger, Comput. Phys. Commun. 81 (1994) 381;
H. Spiesberger, H ERACLES and D JANGOH: Event Generation for ep Interactions at HERA Including Radiative
Processes, 1998, available on http://www.desy.de/~hspiesb/djangoh.html.
[27] Y. Azimov, et al., Phys. Lett. B 165 (1985) 147;
G. Gustafson, Phys. Lett. B 175 (1986) 453;
G. Gustafson, U. Pettersson, Nucl. Phys. B 306 (1988) 746;
B. Andersson, et al., Z. Phys. C 43 (1989) 625.
[28] L. Lnnblad, Comput. Phys. Commun. 71 (1992) 15;
L. Lnnblad, Z. Phys. C 65 (1995) 285.
[29] G. Ingelman, A. Edin, J. Rathsman, Comput. Phys. Commun. 101 (1997) 108.
[30] H.L. Lai, et al., Eur. Phys. J. C 12 (2000) 375.
[31] B. Andersson, et al., Phys. Rep. 97 (1983) 31.
[32] T. Sjstrand, Comput. Phys. Commun. 82 (1994) 74;
T. Sjstrand, Comput. Phys. Commun. 39 (1986) 347;
T. Sjstrand, M. Bengtsson, Comput. Phys. Commun. 43 (1987) 367.
[33] S. Catani, M.H. Seymour, Nucl. Phys. B 485 (1997) 291;
S. Catani, M.H. Seymour, Nucl. Phys. B 510 (1998) 503, Erratum.
[34] R.K. Ellis, D.A. Ross, A.E. Terrano, Nucl. Phys. B 178 (1981) 421.
[35] J. Pumplin, et al., JHEP 0207 (2002) 012;
D. Stump, et al., JHEP 0310 (2003) 046.
[36] J. Pumplin, et al., JHEP 0602 (2006) 032.
[37] S. Bethke, J. Phys. G 26 (2000) R27, hep-ex/0407021.
[38] M. Wing (on behalf of the ZEUS Collaboration), in: R. Zhu (Ed.), Proceedings of the 10th International Conference
on Calorimetry in High Energy Physics, Pasadena, USA, 2002, p. 767, hep-ex/0206036.
[40] ZEUS Collaboration, J. Breitweg, et al., Eur. Phys. J. C 11 (1999) 427.
A new measurement of the Collins and Sivers

asymmetries on a transversely polarised deuteron target
E.S. Ageev y , V.Yu. Alexakhin h , Yu. Alexandrov s , G.D. Alexeev h ,
M. Alexeev ad , A. Amoroso ad , B. Badeek ae , F. Balestra ad , J. Ball z ,
J. Barth d , G. Baum a , M. Becker u , Y. Bedfer z , P. Berglund m,n , C. Bernet z ,
R. Bertini ad , M. Bettinelli t , R. Birsa ac , J. Bisplinghoff c , P. Bordalo p,1 ,
F. Bradamante ac , A. Bressan ac , G. Brona ae , E. Burtin z , M.P. Bussa ad ,
V.N. Bytchkov h , A. Chapiro ab , A. Cicuttin ab , M. Colantoni ad,2 ,
A.A. Colavita ab , S. Costa ad, , M.L. Crespo ab , N. dHose z ,
S. Dalla Torre ac , S. Das g , S.S. Dasgupta f , R. De Masi u , N. Dedek t ,
D. Demchenko q , O.Yu. Denisov ad,3 , L. Dhara g , V. Diaz ac,ab ,
A.M. Dinkelbach u , S.V. Donskov y , V.A. Dorofeev y , N. Doshita b,v ,
V. Duic ac , W. Dnnweber t , A. Efremov h , P.D. Eversheim c , W. Eyrich i ,
M. Faessler t , V. Falaleev k , P. Fauland a , A. Ferrero ad , L. Ferrero ad ,
M. Finger w , M. Finger Jr. h , H. Fischer j , J. Franz j , J.M. Friedrich u ,
V. Frolov ad,3 , U. Fuchs k , R. Garfagnini ad , F. Gautheron a ,
O.P. Gavrichtchouk h , S. Gerassimov s,u , R. Geyer t , M. Giorgi ac ,
B. Gobbo ac , S. Goertz b,d , A.M. Gorin y , O.A. Grajek ae , A. Grasso ad ,
B. Grube u , A. Guskov h , F. Haas u , J. Hannappel d,q , D. von Harrach q ,
T. Hasegawa r , S. Hedicke j , F.H. Heinsius j , R. Hermann q , C. He b ,
F. Hinterberger c , M. von Hodenberg j , N. Horikawa v,4 , S. Horikawa v ,
I. Horn c , C. Ilgner k,t , A.I. Ioukaev h , S. Ishimoto v , I. Ivanchin h ,
O. Ivanov h , T. Iwata v,5 , R. Jahn c , A. Janata h , R. Joosten c ,
N.I. Jouravlev h , E. Kabu q , V. Kalinnikov ac , D. Kang j , B. Ketzer u ,
G.V. Khaustov y , Yu.A. Khokhlov y , Yu. Kisselev a,b , F. Klein d ,
K. Klimaszewski ae , S. Koblitz q , J.H. Koivuniemi b,m,n , V.N. Kolosov y ,
E.V. Komissarov h , K. Kondo b,v , K. Knigsmann j , I. Konorov s,u ,
V.F. Konstantinov y , A.S. Korentchenko h , A. Korzenev q,3 ,
0550-3213/$ see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2006.10.027
RAPID COMMUNICATION
32
E.S. Ageev et al. / Nuclear Physics B 765 (2007) 3170
A.M. Kotzinian h,ad , N.A. Koutchinski h , O. Kouznetsov h , K. Kowalik ae ,

D. Kramer o , N.P. Kravchuk h , G.V. Krivokhizhin h , Z.V. Kroumchtein h ,
J. Kubart o , R. Kuhn u , V. Kukhtin h , F. Kunne z , K. Kurek ae ,
M.E. Ladygin y , M. Lamanna k,ac , J.M. Le Goff z , M. Leberig k,q ,
A.A. Lednev y , A. Lehmann i , J. Lichtenstadt aa , T. Liska x , I. Ludwig j ,
A. Maggiora ad , M. Maggiora ad , A. Magnon z , G.K. Mallot k, ,
C. Marchand z , J. Marroncle z , A. Martin ac , J. Marzec af , L. Masek o ,
F. Massmann c , T. Matsuda r , D. Matthi j , A.N. Maximov h , W. Meyer b ,
A. Mielech ac,ae , Yu.V. Mikhailov y , M.A. Moinester aa , T. Nagel u ,
O. Nhle c , J. Nassalski ae , S. Neliba x , D.P. Neyret z , V.I. Nikolaenko y ,
A.A. Nozdrin h , V.F. Obraztsov y , A.G. Olshevsky h , M. Ostrick d,q ,
A. Padee af , P. Pagano ac , S. Panebianco z , D. Panzieri ad,2 , S. Paul u ,
D.V. Peshekhonov h , V.D. Peshekhonov h , G. Piragino ad , S. Platchkov k,z ,
K. Platzer t , J. Pochodzalla q , J. Polak o , V.A. Polyakov y , G. Pontecorvo h ,
A.A. Popov h , J. Pretz d , S. Procureur z , C. Quintans p , S. Ramos p,1 ,
G. Reicherz b , A. Richter i , E. Rondio ae , A.M. Rozhdestvensky h ,
D. Ryabchikov y , V.D. Samoylenko y , A. Sandacz ae , H. Santos p ,
M.G. Sapozhnikov h , I.A. Savin h , P. Schiavon ac , C. Schill j , L. Schmitt u ,
P. Schnmeier i , W. Schroeder i , D. Seeharsch u , M. Seimetz z , D. Setter j ,
O.Yu. Shevchenko h , H.-W. Siebert l,q , L. Silva p , L. Sinha g ,
A.N. Sissakian h , M. Slunecka h , G.I. Smirnov h , F. Sozzi ac , A. Srnka e ,
F. Stinzing i , M. Stolarski ae , V.P. Sugonyaev y , M. Sulc o , R. Sulej af ,
V.V. Tchalishev h , S. Tessaro ac , F. Tessarotto ac , A. Teufel i ,
L.G. Tkatchev h , T. Toeda v , S. Trippel j , G. Venugopal c , M. Virius x ,
N.V. Vlassov h , M. Wagner i , R. Webb i , E. Weise c,j , Q. Weitzel u ,
R. Windmolders d , W. Wislicki ae , A.M. Zanetti ac , K. Zaremba af ,
M. Zavertyaev s , J. Zhao q,z , R. Ziegler c , A. Zvyagin t
a Universitt Bielefeld, Fakultt fr Physik, 33501 Bielefeld, Germany 6
b Universitt Bochum, Institut fr Experimentalphysik, 44780 Bochum, Germany 6
c Universitt Bonn, Helmholtz-Institut fr Strahlen- und Kernphysik, 53115 Bonn, Germany 6
d Universitt Bonn, Physikalisches Institut, 53115 Bonn, Germany 6
e Institute of Scientific Instruments, AS CR, 61264 Brno, Czech Republic 7
f Burdwan University, Burdwan 713104, India 9
g Matrivani Institute of Experimental Research & Education, Calcutta 700 030, India 10
h Joint Institute for Nuclear Research, 141980 Dubna, Moscow region, Russia
i Universitt ErlangenNrnberg, Physikalisches Institut, 91054 Erlangen, Germany 6
j Universitt Freiburg, Physikalisches Institut, 79104 Freiburg, Germany 6
k CERN, 1211 Geneva 23, Switzerland
l Universitt Heidelberg, Physikalisches Institut, 69120 Heidelberg, Germany 6
m Helsinki University of Technology, Low Temperature Laboratory, 02015 HUT, Finland
RAPID COMMUNICATION
33
n University of Helsinki, Helsinki Institute of Physics, 00014 Helsinki, Finland

o Technical University in Liberec, 46117 Liberec, Czech Republic 7
p LIP, 1000-149 Lisbon, Portugal 8
q Universitt Mainz, Institut fr Kernphysik, 55099 Mainz, Germany 6
r University of Miyazaki, Miyazaki 889-2192, Japan 11
s Lebedev Physical Institute, 119991 Moscow, Russia
t Ludwig-Maximilians-Universitt Mnchen, Department fr Physik, 80799 Munich, Germany 6
u Technische Universitt Mnchen, Physik Department, 85748 Garching, Germany 6
v Nagoya University, 464 Nagoya, Japan 11
w Charles University, Faculty of Mathematics and Physics, 18000 Prague, Czech Republic 7
x Czech Technical University in Prague, 16636 Prague, Czech Republic 7
y State Research Center of the Russian Federation, Institute for High Energy Physics, 142281 Protvino, Russia
z CEA DAPNIA/SPhN Saclay, 91191 Gif-sur-Yvette, France
aa Tel Aviv University, School of Physics and Astronomy, 69978 Tel Aviv, Israel 12
ab INFN Trieste and ICTPINFN MLab Laboratory, 34014 Trieste, Italy
ac INFN Trieste and University of Trieste, Department of Physics, 34127 Trieste, Italy 13
ad INFN Turin and University of Turin, Physics Department, 10125 Turin, Italy 13
ae Sotan Institute for Nuclear Studies and Warsaw University, 00-681 Warsaw, Poland 14
af Warsaw University of Technology, Institute of Radioelectronics, 00-665 Warsaw, Poland 15
Received 27 September 2006; accepted 30 October 2006

Available online 15 November 2006
Abstract
New high precision measurements of the Collins and Sivers asymmetries of charged hadrons produced
in deep-inelastic scattering of muons on a transversely polarised 6 LiD target are presented. The data were
taken in 2003 and 2004 with the COMPASS spectrometer using the muon beam of the CERN SPS at
160 GeV/c. Both the Collins and Sivers asymmetries turn out to be compatible with zero, within the present
statistical errors, which are more than a factor of 2 smaller than those of the published COMPASS results
from the 2002 data. The final results from the 2002, 2003 and 2004 runs are compared with naive expectations and with existing model calculations.
2006 Elsevier B.V. All rights reserved.
PACS: 13.60.-r; 13.88.+e; 14.20.Dh; 14.65.-q
Keywords: Transversity; Deuteron; Transverse single-spin asymmetry; Collins asymmetry; Sivers asymmetry;
COMPASS
E-mail address: gerhard.mallot@cern.ch (G.K. Mallot).

1 Also at IST, Universidade Tcnica de Lisboa, Lisbon, Portugal.
2 Also at University of East Piedmont, 15100 Alessandria, Italy.
3 On leave of absence from JINR Dubna.
4 Also at Chubu University, Kasugai, Aichi 487-8501, Japan.
5 Also at Yamagata University, Yamagata 992-8510, Japan.
6 Supported by the German Bundesministerium fr Bildung und Forschung.
7 Supported by Czech Republic MEYS grants ME492 and LA242.
8 Supported by the Portuguese FCTFundao para a Cincia e Tecnologia grants POCTI/FNU/49501/2002 and
POCTI/FNU/50192/2003.
RAPID COMMUNICATION
34
1. An introduction to transverse spin physics

1.1. Historical introduction
The importance of transverse spin effects at high energy in hadronic physics was first suggested by the discovery in 1976 that hyperons produced in pN interactions exhibited an
anomalously large transverse polarisation [1]. This effect could not be easily explained. For a
long time it was believed to be forbidden at leading twist in QCD [2], and very little theoretical
work was devoted to this field for more than a decade. Nevertheless some important theoretical progress for the understanding of single spin asymmetry (SSA) phenomena was done at that
time [3].
This situation changed in the nineties. After the first hints of large single transverse spin
asymmetries in inclusive 0 production in polarised pp scattering at CERN [4] and in IHEP [5],
remarkably large asymmetries were found at Fermilab both for neutral and charged pions [6]. The
discovery of the EMC Collaboration at CERN in the late eighties that the quark spin contributes
only a small fraction to the proton spin [7] caused a renewed interest in the origin of the nucleon
spin and proposals for new and versatile experiments. In parallel, intense theoretical activity was
taking place: the significance of the quark transversity distribution, already introduced in 1979 [8]
to describe a quark in a transversely polarised nucleon, was reappraised [9] in 1990, and its
measurability via the DrellYan process established. In 1991 a general scheme of all leading twist
and higher-twist parton distribution functions was worked out [10], and in 1993 a way to measure
transversity in lepton nucleon polarised deep-inelastic scattering (DIS) was suggested [11]. On
the experimental side, the RHIC-Spin Collaboration [12] and the HELP Collaboration [13] put
forward the first proposals to measure transversity. Today transversity is an important part of
the scientific programme of the HERMES experiment at DESY, of the RHIC experiments at
BNL, and of the COMPASS experiment at CERN, all presently taking data. An experiment to
measure transversity is being prepared at JLAB. First results on a transversely polarised proton
target have been published recently by the HERMES Collaboration [14], and on a transversely
polarised deuteron target by COMPASS [15].
The COMPASS published results refer to the data taken in 2002. Further data taking occurred
in 2003 and 2004, and in this paper results from these new data are presented, as well as a
reanalysis of the 2002 data and the final results for the 20022004 data.
1.2. The Collins mechanism
To fully specify the quark structure of the nucleon at the twist-two level, the transverse spin
distributions T q(x) must be added to the momentum distributions q(x) and the helicity distri9 Supported by DST-FIST II grants, Government of India.
10 Supported by the Shailabala Biswas Education Trust.
11 Supported by the Ministry of Education, Culture, Sports, Science and Technology, Japan; Daikou Foundation and
Yamada Foundation.
12 Supported by the Israel Science Foundation, founded by the Israel Academy of Sciences and Humanities.
13 Supported also by MIUR (Italy).
14 Supported by KBN grants Nos. 621/E-78/SPUB-M/CERN/P-03/DZ 298 2000, 621/E-78/SPB/CERN/P-03/DWM
576/20032006, and by MNII research funds for 20052007.
15 Supported by KBN grant No. 134/E-365/SPUB-M/CERN/P-03/DZ299/2000.
Deceased.
RAPID COMMUNICATION
35
Fig. 1. Definition of the Collins and Sivers angles. The vectors pTh , s and s are the hadron transverse momentum and
spin of the initial and struck quark, respectively.
butions q(x) [10], where x is the Bjorken variable. For a discussion on notation, see Ref. [16].
If the quarks are collinear with the parent nucleon (no intrinsic quark transverse momentum kT ),
or after integration over kT , these three distributions exhaust the information on the internal dynamics of the nucleon. More distributions are allowed admitting a finite kT , as we will see in
Sections 1.3 and 1.4, or at higher twist [1619].
The distributions T q are difficult to measure, since they are chirally odd and therefore absent
in inclusive DIS. They may instead be extracted from measurements of the single-spin asymmetries in cross-sections for semi-inclusive DIS (SIDIS) of leptons on transversely polarised
nucleons, in which a hadron is also detected in the final state. In these processes the measurable
asymmetry is due to the combined effect of T q and another chirally-odd function which describes the spin-dependent part of the hadronisation of a transversely polarised quark q into a
hadron h.
The existence of an azimuthal asymmetry in transversely polarised leptoproduction of spinless
hadrons at leading twist, which depends on a T -odd fragmentation function and arises from finalstate interaction effects, was predicted by Collins [11] and is now generally known as the Collins
effect. It is responsible for a leftright asymmetry in the fragmentation of transversely polarised
quarks. In some models [20], the Collins asymmetry is expected to be largest for the leading
hadron in the current jet, i.e. the hadron with the highest momentum.
Assuming the detected hadron to be spinless, and a collinear quark distribution in the nucleon,
at leading twist the SIDIS cross-section can be written as

2 s

d
2em
1
2
=
e
1 + (1 y)2 x q(x) Dqh z, pTh

q
h
4
2
Q
dx dy dz d 2 pT dS
q

+ (1 y)|S | sin(h s ) x T q(x) 0T Dqh z, pTh ,
(1)
where Bjorken x is Q2 /[2M(El El )], y is the fractional energy of the virtual photon, Q2 the
photon virtuality, M the target nucleon mass, and z = Eh /(El El ) the fraction of available
energy carried by the hadron. The energies Eh , El , and El are the energies of the hadron, the
incoming lepton, and the scattered lepton respectively in the target rest frame system. The hadron
transverse momentum pTh is evaluated with respect to the virtual photon direction. Referring
to Fig. 1, h and s are the azimuthal angles of the hadron and of the struck quark spin in
a coordinate system in which the z-axis is the virtual photon direction, and the xz plane is the
lepton scattering plane with positive x-direction along the scattered lepton transverse momentum.
S is the target spin normal to the virtual photon direction and S is its azimuthal angle with
respect to the lepton scattering plane.
RAPID COMMUNICATION
36
The pTh -dependent fragmentation function can be obtained by investigating the fragmentation
of a polarised quark q into a hadron h, and is expected to be of the form

DTh q z, pTh = Dqh z, pTh + 0T Dqh z, pTh sin(h s ),
(2)
where 0T Dqh (z, pTh ) is the T -odd part of the fragmentation function, responsible for the left
right asymmetry in the fragmentation of the transversely polarised quark.
The Collins angle C was originally defined in [11] as the angle between the transverse
momentum of the outgoing hadron and the transverse spin vector of the fragmenting quark, i.e.,
sin C =
(pTh q)s
|pTh q||s |
(3)
or
C = h s .
(4)
Since, as dictated by QED, the directions of the final and initial quark spins are related to each
other by s = s , Eq. (4) becomes
C = h + S .
(5)
By comparing the cross-sections on oppositely polarised target nucleons one obtains from
expression (1) the transverse single-spin asymmetry
AhT
d (S ) d (S )
= |S | DN N AColl sin C ,
d (S ) + d (S )
where the Collins asymmetry is

2
0 h
h
q eq T q(x) T Dq (z, pT )
,
AColl =
2
h
h
q eq q(x) Dq (z, pT )
(6)
(7)
and
1y
1 y + y 2 /2
is the transverse spin transfer coefficient from the initial to the struck quark [16].
DN N =
(8)
1.3. The Sivers mechanism

An entirely different mechanism was suggested by Sivers [21] as a possible cause of the
transverse spin effects observed in pp scattering. This mechanism could also be responsible for
a spin asymmetry in the cross-section of SIDIS of leptons on transversely polarised nucleons.
Sivers conjecture was the possible existence of a correlation between the transverse momentum
kT of an unpolarised quark in a transversely polarised nucleon and the nucleon polarisation
vector, i.e. that the quark distribution q(x) in expression (1) could be written as
qT (x, kT ) = q(x, kT ) + |S | T0 q(x, kT ) sin S ,
(9)
where the Sivers angle

S = q S
(10)
is the relative azimuthal angle between the quark transverse momentum kT and the target spin S .
RAPID COMMUNICATION
37
Under the assumption that the hadron produced in the fragmentation and the fragmenting
quark are collinear, i.e. that all the hadron transverse momentum originates from the intrinsic
transverse momentum of the quark in the nucleon (pTh = zkT ), the Sivers angle, shown in Fig. 1,
becomes
S = h S
and the SIDIS cross-section in leading order QCD is given by
2 s

2em
1
d
=
eq2 1 + (1 y)2 x
h
4
2
2
Q
dx dy dz d pT dS
q

q x, pTh /z + |S | sin(S )T0 q x, pTh /z Dqh (z).
(11)
(12)
Comparing the cross-sections on oppositely polarised target nucleons, the transverse spin
asymmetry of expression (6) becomes
AhT
d (S ) d (S )
= |S | ASiv sin S ,
d (S ) + d (S )
where the Sivers asymmetry

2
h
T
h
q eq 0 q(x, pT /z) Dq (z)
ASiv =
h
2
h
q eq q(x, pT /z) Dq (z)
(13)
(14)
could be revealed as a sin S modulation in the number of produced hadrons.

The existence of the Sivers function requires final/initial state interaction, and an interference
between different helicity Fock states. In the absence of interactions the Sivers function would
vanish by time-reversal invariance of QCD (see e.g. Ref. [19]) and indeed it was believed for
several years that the Sivers function is zero. Recently it was shown however [2224] that these
interactions are represented naturally by the gauge link that is required for a gauge invariant definition of a transverse momentum dependent (TMD) parton distribution, thus the Sivers function
has become a very important piece in the most fundamental issues of QCD.
1.4. More general formalism
A combined treatment of the Collins and Sivers mechanisms requires a more general approach
and the introduction of TMD distributions and fragmentation functions [18,19,25]. Convolution
integrals appear in the SIDIS cross-section, whose expression becomes

1
2
eq
1 + (1 y)2 x q Dqh + |S | sin(S )T0 q Dqh
d
2
q

0 h

+ (1 y)|S | sin(C )xT q T Dq .
(15)
In the general expression of the SIDIS cross-section at leading order QCD other terms related
to different single and double spin azimuthal asymmetries appear. Here they are neglected since
they are beyond the scope of this paper. The symbol , which replaces the products in Eqs. (1)
and (12), indicates the convolution integral

DF F F = d 2 kT DF (x, kT ) F F z, pTh zkT ,
(16)
where DF and F F are generic TMD distributions and fragmentation functions, respectively.
RAPID COMMUNICATION
38
The transverse spin asymmetry is then given by

AhT
d (S ) d (S )
= |S | DN N AColl sin C + |S | ASiv sin S ,
d (S ) + d (S )
(17)
where the Collins and the Sivers asymmetries are still given by Eqs. (7) and (14) when replacing
the products of the distribution and fragmentation functions with the corresponding convolutions.
In the usual assumption of Gaussian distributions for the parton and the hadron transverse momenta in the DF and in the F F , the only effect of the convolution integral in Eqs. (7) and (14)
is the presence of a factor which depends on kT2
and pTh2
(see i.e. [26,27]).
Since the Collins and Sivers terms in the transverse spin asymmetry depend on the two independent angles C and S , measuring SIDIS on a transversely polarised target allows the
Collins and the Sivers effects to be disentangled and the two asymmetries can separately be
extracted from the data.
2. The COMPASS experiment
2.1. Physics objectives
The COMPASS experiment was proposed to CERN in 1996 to investigate hadron structure
and hadron spectroscopy by carrying on a number of key measurements using both hadron ( , K
and protons) and muon high-energy beams. Apart from a short pilot run in 2004 with 190 GeV/c
pions to measure the pion polarisability via Primakoff scattering on high-Z targets, the experiment in so far has focused on the investigation of the spin structure of the nucleon using a
160 GeV/c + beam and a polarised deuteron target. In the longitudinal target spin mode, the
main goal of the experiment is the measurement of G/G [28], the polarisation of the gluons
in a longitudinally polarised nucleon, but very precise Ad1 data are also collected [29]. In about
20% of the running time, the target polarisation was set along the vertical direction, orthogonal
to the beam axis, and transverse spin effects were measured, which are the subject of this paper.
2.2. The experimental set-up
The COMPASS spectrometer [30] has been set up in the CERN SPS North Area.
The layout of the spectrometer which was on the floor in 2003 is shown in Fig. 2. To combine large geometrical acceptance and broad dynamical range, the spectrometer comprises two
magnetic stages, and is made up of a variety of tracking detectors, a fast RICH, two hadronic
calorimeters, and provides muon identification via filtering through thick absorbers.
The first stage is centred around the spectrometer magnet SM1 (1 Tm bending power) and
the design acceptance is about 200 mrad in both planes, to fully contain the hadrons of the
current jet. The second stage uses the spectrometer magnet SM2 (operated at a bending power of
4.4 Tm), located 18 m downstream from the target, and the acceptance is 50 and 25 mrad in
the horizontal and vertical plane, respectively.
The design of detector components, electronics and data acquisition system allows to handle
beam rates up to 108 muons/s and about 5 107 hadrons/s. The triggering system and the
tracking system of COMPASS have been designed to stand the associated rate of secondaries,
and use state-of-the-art detectors. Also, fast front-end electronics, multi-buffering, and a large
and fast storage of events are essential.
RAPID COMMUNICATION
39
Fig. 2. Top view of the lay-out of the spectrometer for the COMPASS experiment in 2003. The labels and the arrows
refer to the major components of the trigger and PID systems. The thin vertical lines represent the tracking detectors.
The experiment has been run at a muon energy of 160 GeV. The muons originate from the
decay of and K mesons produced by the 400 GeV proton beam on a primary beryllium target.
The + beam is naturally polarised by the weak decay mechanism, and the beam polarisation
is about 80%. The beam polarisation contributes to the transverse spin-dependent part of the
cross-section only by higher twist effects, which are not considered in the leading-order analysis
of this paper. The + intensity is 2 108 per spill of 4.8 s with a cycle time of 16.8 s. The beam
profile presents a Gaussian core and a large non-Gaussian tail due to halo muons. The beam has
a nominal energy of 160 GeV and is focused at the target centre, with a spread of 7 mm (r.m.s.)
and a momentum spread of p /p = 0.05 for the Gaussian core. The momentum of each muon is
measured upstream of the experimental area in a beam momentum station consisting of several
planes of scintillator strips with 3 dipole magnets (30 Tm total bending power) in between. The
precision of the momentum determination is typically p/p 0.003.
Due to the major problems and delay in the construction of the new large-acceptance COMPASS polarised target (PT) magnet, the experiment has utilised in so far the magnet from the
SMC experiment, which has a similar design and identical magnetic properties, but a smaller
bore (26.5 cm diameter). The resulting angular acceptance is reduced, going from 170 mrad
at the downstream end to 69 mrad at the upstream end of the target. The polarised target system [31] consists of two oppositely polarised target cells (upstream u, downstream d), 60 cm
long each and 3 cm diameter, so that data are collected simultaneously for the two target spin
orientations. The PT magnet can provide both a solenoidal field (2.5 T) and a dipole field for
adiabatic spin rotation (up to 0.5 T) and for the transversity measurements (set at 0.42 T). Correspondingly, the target polarisation can then be oriented either longitudinally or transversely to
the beam direction. The target is cooled to temperatures below 100 mK by a 3 He4 He dilution
refrigerator. When operated in the frozen spin mode, temperatures of 50 mK are reached. The
polarisation was lost at rate of 0.41.0%/day in the 0.42 T dipole field and 0.050.1%/day in
2.5 T solenoid field. The use of two different target materials, NH3 as proton target and 6 LiD as
deuteron target, is foreseen. Polarisations of 90% [32] and 50% [30] have been reached, respectively. In so far only 6 LiD has been used as target: its favourable dilution factor of 0.4 is of the
utmost importance for the measurement of G. The dynamical nuclear polarisation system can
polarise the target only at 2.5 T, i.e. in the solenoidal field. To run in the transverse polarisation
RAPID COMMUNICATION
40
mode, the spins are frozen and rotated adiabatically by first lowering the longitudinal field to
0.5 T, then rotating the magnetic field to the vertical direction with the help of the dipole coils.
The polarisation values are obtained from corresponding measurements done at the beginning
and at the end of each data taking period, with the polarised target field set back at 2.5 T, and
take into account the relaxation time of the target polarisation.
To match the expected particle flux in the various locations along the spectrometer, COMPASS uses very different tracking detectors. The small area trackers consist of several stations
of scintillating fibres, silicon detectors, micromegas chambers [33] and gaseous chambers using
the GEM-technique [34]. Large area tracking devices are made from gaseous detectors (Drift
Chambers, Straw tubes [35], and MWPCs) placed around the two spectrometer magnets.
Muons are identified in large-area Iarocci-like tubes and drift tubes downstream of muon
absorbers. Hadrons are detected by two large iron-scintillator sampling calorimeters, installed in
front of the absorbers and shielded to avoid electromagnetic contamination. The charged particles
identification relies on the RICH technology [36]. In this paper we have not utilised the information of the RICH, and give results for non-identified hadron asymmetries only. The asymmetries
for RICH-identified hadrons (pions and kaons) will be the subject of a separate paper.
The trigger [37] is formed by a combination of signals indicating the presence of a scattered
muon at a given angle or in a given energy range. In most DIS events (Q2 > 1 (GeV/c)2 ), the
scattered muon is identified by coincidence signals in the trigger hodoscopes, that measure the
projection of the scattering angle in the non-bending plane and check its compatibility with the
target position. Several veto counters installed upstream of the target are used to avoid triggers
due to halo muons. In addition to this inclusive trigger mode, several semi-inclusive triggers
select events fulfilling requirements based on the muon energy loss and on the presence of a
hadron signal in the calorimeters. The acceptance is further extended toward high Q2 values by
the addition of a standalone calorimetric trigger in which no condition is set for the scattered
muon.
A complete description of the spectrometer can be found in Ref. [30].
2.3. Data taking and off-line system
The data acquisition system [38] was designed to read the 250 000 detector channels with an
event rate of up to 100 kHz and with essentially no dead-time. This required a full custom design
of the readout electronics (mounted directly on the detectors) and of the readout-driver modules
(named CATCH and GeSiCA), which perform local event building and trigger distribution to
the front-end boards (Fig. 3). The trigger control system (TCS) performs trigger distribution and
synchronisation of the time-to-digital converters to better than 50 ps. The data collected during
the 4.8 s spills is stored in 32 GByte buffers. The event building is performed on high performance
Linux PCs connected to the read out system via Gigabit Ethernet. The data are grouped in runs
corresponding to about 100 consecutive spills in 2003 and 200 in 2004. General information on
the run is extracted and stored in the meta-data tables of an Oracle database. The files are then
sent to the Central Data Recording facility in the CERN computer centre in parallel multiple
streams over a dedicated optical fibre network at an average speed of 70 MBytes/s.
Once at the computer centre, the files are registered in the name space of CASTOR (the
CERN hierarchical storage management system): from this moment onward, CASTOR controls
thoroughly the events data handling (copy to tape, managing of the disk space), while an Oracle
RAC database system is in charge of translating high-level requests of data into file requests.
The huge amount of data of about 350 TBytes/year is reconstructed at the CERN computer
RAPID COMMUNICATION
41
Fig. 3. General architecture of the DAQ system.
centre, requiring a computing power of about 200k SPECint2000. The event reconstruction is
carried on by CORAL, the COMPASS reconstruction and analysis framework, a fully objectoriented program with a modular architecture written in C++, which provides interfaces for the
event reconstruction algorithms and insulation layers to access the data and for external pluggable packages [39]. The reconstruction is carried out in parallel, by some 600 jobs running on
the CERN batch system. CORAL decodes the data, reconstructs tracks and vertexes, and performs the particle identification making use of the alignment and calibration data describing the
apparatus, which are stored as time/version-dependent information in a MySql database. The reconstructed data is output in a proprietary format to files called Data Summary Tapes (DSTs),
and in ROOT [40] format to mini-DSTs, which are selectively filtered out from the DSTs during production and turn out to have a size of about 1% of the original RAW data. DSTs and
miniDSTs are stored also centrally on tape, under CASTOR. The physics analysis is performed
on the mini-DSTs, replicated in the different institutes, by means of PHAST, the COMPASS
framework for the final data analysis [39]. Fig. 4 depicts the reconstruction and analysis system to which the data flow after they are stored centrally. For the simulations, the Monte Carlo
program COMGeant [39], based on GEANT 3 and LEPTO 6.5.1, has been written and used
extensively.
In the first three years of data taking, from 2002 to 2004, in which the experiment benefited
of about 12 105 spills, COMPASS collected 30 billion events, corresponding to a total data
sample of more than 1 PByte. Thanks to continuous work on alignment and improvement of the
reconstruction code, these data have been processed several times. About 20% of these data have
been taken in the transverse target spin mode.
RAPID COMMUNICATION
42
Fig. 4. Schematic view of the off-line system and reconstruction and analysis flow.
2.4. Principle of the measurement

As explained in Section 1.1, the Collins and the Sivers asymmetries can be estimated from
the SIDIS data by measuring, in a given kinematical bin defined by x, z, or pTh , the number of
events N + and N collected on oppositely polarised target nucleons, and fitting the measured
asymmetries (N + N )/(N + N ) either with a sin C or a sin S dependence. Since the
spectrometer acceptance is somewhat different for the two target halves, the cross-section asymmetries cannot be obtained from a direct comparison of the number of events collected in the
u- and d-cell. It is necessary to compare the number of events collected in each cell with the two
target spin orientations, and the two-cell system allows to reduce the systematic effects.
The orientation of the target polarisation cannot be flipped fast in the COMPASS target system. The holding field is 0.42 T, and transverse to the beam, thus it is out of question to flip the
target spin by inverting the magnetic field because the acceptance would change. Polarisation
reversal is done by exchanging the microwave frequencies for u- and d-cells, and to achieve
50% polarisation one needs two days. Therefore, once the target is polarised, data are taken
for several days (typically 5) before a polarisation reversal is done, and to evaluate an asymmetry
one compares the number of events collected almost a week apart. For this reason, the transverse
spin measurement usually was carried out at the end of the run, when the spectrometer was fully
operational and functioning smoothly, and great care was taken not to intervene on any detector
to avoid changes in efficiency.
The typical cycle (a data taking period) consisted therefore of 5 days of measurement in one
configuration (yielding e.g. Nu+ and Nd ), 2 days to change the target polarisation, and 5 more
days of data taking in the opposite configuration (yielding Nu and Nd+ ). At this point, one can
estimate the cross-section asymmetry. A straightforward procedure was used for the published
2002 data analysis [15], namely to derive one asymmetry from Nu+ and Nu , a second asymmetry
from Nd+ and Nd , check their consistency and then average the two. A different method, which
uses at once the four numbers and is less sensitive to differences in acceptance, has been used in
this analysis, and is described in Section 4.3.2.
RAPID COMMUNICATION
43
In this paper final results are presented for all the data collected with the deuteron target in
the three years 200220032004. Data were collected during two periods in 2002, one period in
2003, and during two periods in 2004. The focus is on the data collected in 2003 and 2004, and
most of the next section will be devoted to the analysis of these data, which is slightly different
from the analysis of the 2002 data (already published), thanks to various improvements in the
analysis code. The new chain of analysis has also been used to reanalyse the 2002 data: the results
are perfectly compatible, as will be shown in Section 6.1. Also, the new 20032004 data are in
excellent agreement with the 2002 data. At the end, overall results will be given for the whole
data set, 20022004.
3. Data analysis IEvent reconstruction and selection
3.1. Event reconstruction
In the event reconstruction, a track reconstructed before the target is assumed to be an incoming muon if it is reconstructed in the scintillating fibres and silicons upstream of the target, its
momentum is reconstructed in the BMS, and the track time is within 3 standard deviations of
BMS and trigger time. If several valid BMS-tracks are compatible with the track time, a backtracking algorithm is used to resolve ambiguities.
The scattered muon candidates are defined as the positively charged outgoing tracks with a
momentum larger than 1 GeV/c, going through SM1, and their extrapolation at the entrance and
at the exit of the target is within 5 cm from the target axis. In addition, for all triggers based on
the hodoscope information, the track must be compatible with the hodoscope hits as given in the
trigger matrix. In the case of a calorimetric trigger, a minimal number of hits is required in the
muon walls and the amount of material traversed in the spectrometer must be larger than 66 and
74 radiation lengths for tracks reconstructed in the first and in the second spectrometer stage,
respectively.
The muon interaction point (the so-called primary vertex) is defined by one beam particle
and one scattered muon. If in the event there are more than one beam particles and/or scattered
muons, several vertexes may be reconstructed. The distance of closest approach between the
scattered muon and beam track must be less than 10 (about 23 mm). The position of the vertex
along the spectrometer axis zvtx 16 is given by the average of the distance of closest approach of
all tracks in the vertex, while in the orthogonal plane the coordinates xvtx and yvtx are defined
from the beam track at zvtx . The tracks belonging to the vertex are selected using a Kalman fit.
The best primary vertex, used in the following steps of the analysis, is defined as the one with
the maximum number of tracks and, if the number of tracks is the same, the one with smaller
vertex 2 .
Only the events with at least one primary vertex reconstructed with at least one more outgoing
track, and Q2 > 1 (GeV/c)2 , which are about 1% of the initial raw event sample, are written on
miniDST and used in the physics analysis.
The event reconstruction was almost the same for the 2003 and 2004 data, and is very similar
to that used for the 2002 data. Each year an improvement in the ratio of reconstructed events
over the useful beam was achieved, due both to additional detectors in the spectrometer and to
the better tuned reconstruction and analysis software.
16 The used reference system is defined to have a right-handed frame with the z axis along the spectrometer axis (i.e. the
nominal beam direction), the y axis in the vertical direction pointing upwards, and the origin at the centre of the u-cell.
RAPID COMMUNICATION
44
3.2. Data quality checks

To monitor the performances of the apparatus before and after the target spin reversal (i.e. on
the two sub-periods) in each data taking period, the time stability of many distributions has
been checked dividing the whole sample into runs or clusters of neighbouring runs.
Using the histograms produced during the event reconstruction, the detector performance stabilities were scrutinised looking at the stability of the shape of the hit distributions in the about
360 detector planes. The time stability of the detector and reconstruction efficiencies was checked
looking at:
the number of clusters per plane and per event,

the mean number of tracks per event,
the mean number of track segments in the different spectrometer regions per event,
the mean number of primary vertexes per event,
the mean number of secondary vertexes per event.
Using the miniDST events, the stability was checked monitoring run per run
the number of reconstructed K 0 per primary vertex,

the reconstructed K 0 mass distribution,
the energy measured in the two hadronic calorimeters HCAL1 and HCAL2,
the xvtx and yvtx distributions in the two cells,
the vertex 2 distribution.
Also, the time stability of the distributions of several kinematical observables (like x, Q2 , y, the
azimuthal angles and the momenta of the scattered muon and of the hadrons) was investigated in
detail.
As an example, the mean invariant mass in the K 0 region from the data collected in the
first sub-period of the 2004 run is shown in Fig. 5 as a function of time.
Runs showing some instability were not used for the physics analysis. The runs rejected by
this criterion are 28 (over a total of 458) for the 2003 data, and 44 (over 462) for the 2004 data,
corresponding to about 5% and 4% of the initial raw event sample.
3.3. DIS events selection
To better define the DIS events, more refined cuts were applied. In particular:
(1) Primary vertex: if more primary vertexes were reconstructed in one event, the best primary
vertex was selected.
The vertex had then to be inside the target. A radial cut rvtx < 1.3 cm was applied on the
distance of the vertex from the target axis. The event was then accepted only if zvtx was
inside one of the two target cells (100 cm < zvtx < 40 cm or 30 cm < zvtx < 30 cm),
and assigned correspondingly to the u- or to the d-cell.
The distribution of the primary vertex z-coordinate for the final sample is shown in Fig. 6.
The increase in the number of events with zvtx is due to the increase in geometrical acceptance going from the upstream to the downstream end of the target, as explained in
Section 2.2, and is very well reproduced by the Monte Carlo simulation. The two target
RAPID COMMUNICATION
45
Fig. 5. invariant mass in the K 0 region as a function of time from August 13 to August 19, 2004.
Fig. 6. Distribution of the primary vertex z-coordinate for the final sample. The dashed histogram indicates the events
selected by the zvtx cut.
cells are clearly separated. One can also notice the continuum of events produced on the
helium bath, as well as a sample of events produced in the Al window, at z = 50 cm, which
seals the vessel of the PT magnet.
(2) Incoming muon: the cut pbeam < 200 GeV/c was applied on the momentum of the incoming
particle. A check was also performed on the quality of the beam track, and a safe cut on
2 / was applied.
To ensure an identical beam intensity in both target cells, i.e. a nearly identical luminosity, the
beam track projection at the entrance and at the exit of the target region (e.g. at z = 100 cm
RAPID COMMUNICATION
46
Fig. 7. Momentum distribution of the reconstructed incoming muons for the final sample of events.
and z = 30 cm) had to be inside the target region as defined for the primary vertex. The
momentum distribution for the final data sample is shown in Fig. 7.
(3) Scattered muon ( ): as well as for the incident muon, a suitable cut was applied on the
2 / of all reconstructed tracks from the vertex, including the scattered muon candidate.
Other outgoing tracks were flagged as muon candidate if they had hits in one of the two
Muon Walls. To achieve a clean muon identification, the amount of material traversed in the
spectrometer had to be larger than 30 radiation lengths. Only events with one and only one
muon candidate entered the following steps of the analysis.
In addition, standard DIS cuts Q2 > 1 (GeV/c)2 , mass of the final hadronic state W > 5 GeV/c2 ,
and 0.1 < y < 0.9 were applied.
All these cuts reduced the number of miniDST events by a further 45%.
3.4. Hadron identification
All the outgoing particles not flagged as candidates were assumed to be hadrons.
Only particles with the last measured coordinate after the first spectrometer magnet were
used, to reject tracks reconstructed in the fringe field on SM1 which have a poorer momentum
resolution.
The following requirements had to be satisfied in order to identify the particle with a hadron:
(1) The amount of material traversed in the spectrometer had to be smaller than 10 radiation
lengths.
(2) The particle must not give a signal in both the hadronic calorimeters HCAL1 and HCAL2.
(3) If the particle gave a signal in one calorimeter, the measured energy had to be EHCAL1 > 5
(2003 data) or 4 (2004 data) GeV, or EHCAL2 > 5 (2003 and 2004 data) GeV. The correlation
RAPID COMMUNICATION
47
Fig. 8. Correlation between the energy measured in HCAL1 (left) and HCAL2 (right) and the energy measured by the
spectrometer for the 2004 data.
between the energy measured in HCAL and that measured by the spectrometer is shown in
Fig. 8.
(4) The transverse momentum of the particle with respect to the virtual photon direction had to
be larger than 0.1 GeV/c.
The first requirement reduced the muon contamination, the second and the third the muon
and the electron contamination. The last cut was introduced to assure a good resolution in the
measured azimuthal angle.
Two different asymmetries have been measured. The all hadron asymmetries were evaluated
using all the particles identified as hadrons carrying a fraction of available energy z > 0.2. For
the leading hadron analysis, only the events with at least one hadron with z > 0.25 were used,
and the hadron with largest z was defined as leading. In the case the energy of the reconstructed
leading hadron was higher than the missing energy evaluated from all the charged reconstructed
hadrons, no further test was performed. If this was not the case, the event was accepted only if in
the hadron calorimeters no particle with energy larger (within the energy resolution) than that of
the leading hadron was detected.
After these cuts, the number of events with a leading hadron amounted to 22% of the number
of miniDST events.
4. Data analysis IIkinematical distributions and asymmetry evaluation
4.1. Final samples of events
In Table 1, the final statistics used for the asymmetry evaluation is given for all the periods
both for the positively and negatively charged leading hadron and the all hadron sample.
4.2. Kinematical distributions
The distributions of some kinematical quantities from the final 2004 leading hadron events are
shown in Figs. 9 to 11. The plots have been produced after all the cuts described in the previous
section, if not specified differently.
The Q2 vs x scatter-plot and its projections are shown in Fig. 9.
RAPID COMMUNICATION
48
Table 1
Final statistics used for the asymmetry evaluation
Year
Period
Leading hadron sample
All hadron sample
Positive hadrons
Negative hadrons
Positive hadrons
Negative hadrons
2002
2002
1
2
0.48 106
0.32 106
0.38 106
0.26 106
0.71 106
0.48 106
0.59 106
0.40 106
2003
1.68 106
1.33 106
2.46 106
2.03 106
2004
2004
1
2
1.44 106
1.87 106
1.13 106
1.47 106
2.12 106
2.75 106
1.74 106
2.26 106
Fig. 9. Scatter-plot of Q2 vs x, and the corresponding Q2 - and x-distributions for the 2004 positive plus negative leading
hadron sample.
The xz correlation (without the z cut), the xy correlation, and the zpTh correlation (without
the z and the pTh cuts) are shown in Fig. 10, while the y, pTh (without the pTh cut) and the z
(without the z cut) distributions are shown in Fig. 11.
4.3. Asymmetry evaluation
4.3.1. Binning
The Collins and Sivers asymmetries were evaluated as function of x, pTh , and z dividing
the corresponding kinematical range in bins (with variable width, in order to have a comparable
RAPID COMMUNICATION
49
Fig. 10. Scatter-plots of z vs x, y vs x, and z vs pTh for the 2004 all positive plus negative hadron samples.
Table 2
Final binning in x, z, and pTh
0.003 < x < 0.008
0.20 z < 0.25
0.10 < pTh 0.20 GeV/c
0.008 x < 0.013
0.25 z < 0.30
0.20 < pTh 0.30 GeV/c
0.013 x < 0.020
0.30 z < 0.35
0.30 < pTh 0.40 GeV/c
0.020 x < 0.032
0.35 z < 0.40
0.40 < pTh 0.50 GeV/c
0.032 x < 0.050
0.40 z < 0.50
0.50 < pTh 0.60 GeV/c
0.050 x < 0.080
0.50 z < 0.65
0.60 < pTh 0.75 GeV/c
0.080 x < 0.130
0.65 z < 0.80
0.75 < pTh 0.90 GeV/c
0.130 x < 0.210
0.80 z < 1.00
0.90 < pTh 1.30 GeV/c
0.210 x < 1.000
1.30 < pTh
statistics in each of them), and integrating over the other 2 variables. No attempt has been done to
extract the asymmetries in a 2- or 3-dimensional grid. In total, the asymmetries where evaluated
in the 9 x-bins, 9 pTh -bins, and 8 z-bins for the all hadron samples (7 for the leading hadron,
because of the z > 0.25 cut) given in Table 2.
Tables with the mean values of Q2 , z, pTh and y in all the x bins are available on HEPDATA [41]. As an example, Fig. 12 shows the mean values of Q2 (left), z, pTh and y (right) in
the different x bins for the all positive hadron sample from 2003 and 2004 data.
RAPID COMMUNICATION
50
Fig. 11. Upper plots: Distributions of y and pTh (without the pTh cut) for the 2004 all positive plus negative hadron
sample. Lower plots: z-distribution (without the z cut) for the same events (left) and for the 2004 leading positive plus
negative hadron sample (right).
Fig. 12. Mean values of Q2 (left), z, pTh and y (right) in the different x bins for all positive hadrons, 2003 and 2004 data.
4.3.2. Raw asymmetry evaluation

The Collins and Sivers asymmetries have been evaluated separately, in each kinematic bin,
for each data taking period, and for positive and negative hadrons.
RAPID COMMUNICATION
51
Using Eqs. (1) and (12), the distribution of the number of events for each cell and for each
polarisation state can be written as

Nj,k
(18)
(j ) = Fk nk aj,k
(j ) 1 j,k
sin j ,
where j = C, S, and F is the muon flux, n the number of target particles, the spin averaged
cross-section, and aj the product of angular acceptance and efficiency of the spectrometer. The
index k = u, d refers to the target cell. The signs indicate the two target spin orientations:
+ means spin up, spin down. Here the Collins and Sivers angles j are evaluated using the
expressions given in Sections 1.2 and 1.3, always assuming target spin up, at variance with what
was done in [15].
are
The asymmetries j,k
C,k
= cC,k
AColl ,
S,k
= cS,k
ASiv ,
where
cC,k
= f PT,k DN N ,
cS,k
= f PT,k .
(19)
(20)
The factor f is the polarised target dilution factor, PT,k the deuteron polarisation, and DN N
the transverse spin transfer coefficient from the initial to the struck quark given in Eq. (8). The
absolute values of the target polarisation PT in the two cells and for the two spin orientations in
the same cells differed up to about 7%.
from the measured number of events a method, which in the following

In order to estimate j,k
will be called ratio product method (RPM), has been used which minimises the effect on the
estimated asymmetries of possible acceptance variations.
In each data taking period, the ratio product quantities [15]
Aj (j ) =
+
+
Nj,u
(j ) Nj,d
(j )
Nj,u (j ) Nj,d (j )
j = C, S,
(21)
were evaluated from the number of events in the two cells. For small values of the quantities j,k
it is:
Aj (j ) = CF Ca,j
CF Ca,j
+
+
(1 + j,u
sin j ) (1 + j,d
sin j )
(1 j,u
sin j ) (1 j,d
sin j )

m
1 + Aj sin j ,
(22)
where
+
+
Am
j = j,u + j,d + j,u + j,d = 4 j
,
CF =
Fu+
Fu
Fd+
,
Fd
Ca,j =
+
+
aj,u
(j ) aj,d
(j )
.
aj,u (j ) aj,d (j )
(23)
(24)
m
The raw asymmetries Am
C and AS were evaluated by fitting respectively the quantities AC (C )
m
and AS (S ) with the functions p0 (1 + Am
C sin C ) and p0 (1 + AS sin S ), where p0 is a
free parameter.
In this method the only requirement to avoid systematic errors due to acceptance effects is
that the factor Ca,j does not depend on j . This is surely true, in particular, under the reason+
(j )/aj,d
(j ) before the polarisation reversal be equal to the
able assumption that the ratio aj,u
RAPID COMMUNICATION
52

Table 3
Target polarisation values for the 2003 and 2004 data taking
Year
Period/Sub-period
Polarisation
Upstream
Downstream
2003
1/1
2/2 (1st part)
2/2 (2nd part)
49.7
49.4
51.3
52.8
42.6
44.6
2004
1/1
1/2 (1st part)
1/2 (2nd part)
2/1 (1st part)
2/1 (2nd part)
2/2
50.7
44.8
38.6
46.1
46.4
49.9
43.5
46.0
40.4
47.4
47.4
42.8
+
corresponding ratio aj,u
(j )/aj,d
(j ) after the reversal in each j bin. In this case, Ca,j is
equal to 1 for all j values. Also, with the cuts applied on the incoming beam it is expected that
CF = 1 and indeed the fits to Am
j give p0 values always compatible with 1.
In addition to the fact that this estimator has soft requirements on the acceptance stability,
it is independent of the relative luminosity and combines all the data from the two target cells.
A further advantage of the use of the RPM, is that at first order (for small values of the involved
asymmetries) all spin-independent effect, e.g. Cahn asymmetry, are factored out.
Concerning the j binning, the interval (0, 2) was divided in 16 bins. Monte Carlo studies
indicated that the angular resolution (rms) is 0.07 rad, much smaller than the bin size.
4.3.3. Collins and Sivers asymmetries evaluation

To extract the Collins and Sivers asymmetries, the target polarisation values given in Table 3
have been used.
The dilution factor f has been taken constant and equal to 0.38 for all data taking periods.
Using the event kinematics, the mean values of the quantities cC,k
= f |PT,k | DN N and
= f |PT,k | have been evaluated for each bin and each period of data taking. These values
cS,k
have than been used to calculate the Collins and Sivers asymmetries from the corresponding raw
asymmetries Am
j .
5. Data analysis IIISystematic studies
Given the high statistics of our measurements, a number of systematic studies have been
performed in order to determine the size of possible systematic errors.
Extensive tests both to measure false asymmetries and to investigate the stability of the physics
results were done, for each measured asymmetry, and in each data taking period. They are briefly
described in the following.
5.1. False asymmetry studies
Two different kinds of asymmetries expected to be zero were measured using the final event
samples, for the Collins and Sivers angles, positive and negative hadrons, leading and all hadrons,
and in each x, z, pTh bin.
RAPID COMMUNICATION
53
The first one was built by splitting the two target cells in two parts and by combining the data
from the same cell. The mean values of all the resulting asymmetries were found to be compatible
with zero, as it should. The distributions of the pulls of these asymmetries with respect to zero,
i.e. the values of the resulting asymmetries divided by the corresponding standard deviation,
were well compatible with Standard Normal distributions, giving no indication of the presence
of systematic effects.
The second asymmetry was measured after scrambling the data collected in each period and
measuring the asymmetry in the standard way. The runs of the two sub-periods before and after
the polarisation reversal were mixed by labelling as spin up every 1st, 3rd, 5th, . . . run of the
two sub-periods and as spin down all the others. This selection was truncated in a way, that
each fake period had the same amount of runs from each real sub-period to ensure roughly the
same amount of events. As in the previous case, the asymmetries are expected to be zero, and the
measured asymmetries were compatible with zero. The refined statistical analysis of pulls with
respect to zero did not give any hint for the presence of systematic effects.
5.2. Stability of physics asymmetries vs acceptance and time
Possible acceptance effects on the physics results from each data taking period were tested
by checking the compatibility of the asymmetries evaluated after splitting the data according
to: the location inside the target; the region of the spectrometer in which the scattered muon
was measured; the trigger of each event. This work was done for both the Collins and Sivers
asymmetries of positive and negative hadrons, leading and all, and in each x, z, pTh bin.
The compatibility test consisted in comparing the distribution of the pulls
m
Am
i,k Ak
Pi,k =
A2 m A2 m
i,k
(25)
with the normal standard distribution. Here Am

k are the raw asymmetries evaluated from the
whole data sample of one specific period k. They are evaluated for positive and negative hadrons,
Collins and Sivers asymmetries in each x, pTh and z bin, for a total of 2 2 (9 + 9 + 7) =
100 values in each period. The corresponding variance is A2 m . The asymmetries Am
i,k are the
k
corresponding ones evaluated for all the sub-samples i in which the original data set was divided.
2
2
The variances of Am
i,k are Am , and the use of the difference with Am takes into account almost
i,k
m
completely all the correlations between Am
i,k and Ak . The compatibility with the normal standard
distribution was checked also for some specific group of asymmetries, like positive and negative
hadrons, and Collins and Sivers asymmetries.
The effect of the different acceptances for events with the primary vertex in the two target cells
was investigated, evaluating the physics asymmetries separately for the events with the primary
vertex in the inner half and in the outer half of the two target cells. Combining events from only
the outer halves (upstream part of cell u and downstream part of cell d) as well as from the inner
halves only (downstream part of cell u and upstream part of cell d) gave results compatible to
each other and to the ones using the full sample.
The asymmetries were also evaluated dividing the data samples accordingly to the azimuthal
angle of the scattered muon in the laboratory system, namely in the regions 0 < < /2,
/2 < < , < < 3/2, and 3/2 < < 2 .
RAPID COMMUNICATION
54
As in the previous case, the pull distributions turned out to be centred around zero, and in very
good agreement with the standard distribution. The RMS were statistically compatible with 1, as
expected in the absence of systematic effects.
The physics asymmetries have been evaluated splitting the data in different samples according
to the different triggers. The results turned out to be statistically compatible, giving once more
no evidence for systematic effect.
A further test was done to investigate the stability in time of the physics asymmetries. To do
this, each of the two sub-periods entering the standard asymmetry calculation was split in these 9
groups of subsequent runs (each corresponding to 10 hours of data taking). The asymmetries
were then calculated using the full data set of one sub-period and all the 9 groups of runs of
the other sub-period separately. In total 54 asymmetries were evaluated from the 2003 and 2004
data. This method is known to be quite powerful to single out significant time dependencies
within single periods. As for the previous test, the pulls relative to the mean asymmetry for
each time slot region were calculated, but no evidence for systematic errors was found in their
distributions.
A further test was done to check the possible effect of z and pTh acceptance. The asymmetries
were evaluated also in 4 complementary z and pTh bins: (z < 0.5, pTh < 0.5), (z < 0.5, pTh > 0.5),
(z > 0.5, pTh < 0.5), (z > 0.5, pTh > 0.5) and their weighted means were compared with the
standard results. The differences between the two sets of measurements are essentially invisible,
much smaller than the statistical errors, again as expected in absence of systematic effects.
5.3. Stability of the acceptance in the Collins and Sivers angles
A stringent test on the j (j = C, s) dependence of the acceptance ratio, already used for the
previously published data, consists in checking the j dependence of the ratios:
Rj (j ) =
Nj,u
(j ) Nj,d
(j )
+
Nj,u
(j ) Nj,d
(j )
(26)
At the first order, assuming the absolute value of the target polarisation to be the same in each
cell before and after reversal, it is
Rj (j )
Fu+ Fd aj,u (j ) aj,d (j )
+
Fu Fd+ aj,u
(j ) aj,d
(j )
(27)
+
(j )/aj,d
(j ) = aj,u
(j )/aj,d
(j ) it is
In the very likely case in which aj,u
Rj (j )
Fu+ Fd
Fu Fd+
+
aj,u
(j )
aj,u
(j )
2

Fu+ Fd
Fu Fd+
aj,d
(j )
+
aj,d
(j )
2
,
(28)
thus, the constancy in j of Rj (j ) implies for each cell the ratio of the acceptances before and
after the reversal to be constant in j . It must be noted that this constancy is not required to have
unbiased estimators in the case the asymmetries are evaluated with the RPM. Still, this test is
quite convincing on the stability of the apparatus.
The ratios Rj were calculated in each bin of x, z, pTh , for the Collins and Sivers angles, for
leading and all hadrons and for both charges. In each bin the j distribution was fitted with a
constant. Fig. 13 shows an example of the RC values for the Collins leading hadron sample in the
9 x-bins for the first period of data taking in 2004. The lines are the results of the fit. The quality
RAPID COMMUNICATION
55
Fig. 13. RC values vs C for the leading hadron sample in the nine x-bins for the first period of data taking in 2004. The
lines show the result of fits with to a constant.
Fig. 14. The 2 distribution of a constant fit on R() for the leading hadron sample compared to the normalised 2
distribution for ndf = 15.
of these fits are very good, as can be seen from Fig. 14 where the distribution of the 2 of all the
fits is compared with the expected 2 distribution for = 15 degrees of freedom normalised to
the number of entries. To conclude, this test gave an independent indication on the stability of
the acceptances.
RAPID COMMUNICATION
56
2
Fig. 15. 2 distribution of the two parameter fit used to extract the asymmetries Am
j compared to the normalised
distribution for ndf = 14.
Summarising, all the test we performed on the effects of acceptance and on fluctuation in
time of the measured asymmetries gave results statistically compatible with what expected in the
absence of systematic effects.
5.4. Further tests on the fit of the ratio product quantities
m
The quality of the fit of the double ratio quantities Am
j with the function p0 (1 + Aj sin j )
has been checked looking at the 2 distribution. Fig. 15 shows the 2 distribution for Collins and
Sivers asymmetries, in each bin of x, z, pTh for leading hadrons of both charges and for all the
5 data taking periods in 2002, 2003, and 2004. The curve is the expected 2 distribution for 14
degrees of freedom normalised to the number of entries. A perfect agreement can be observed.
The effect of using two other different fitting functions, namely (1 + Am,j sin j ) and
(p0 + Am,j sin j ), as well as changing the j bin size (8 bins instead of 16) have also been
investigated. In both cases the effects turned out to be negligible, of the order of a few percents
of the statistical error.
5.5. Different estimators

To estimate the size of possible systematic effects, the asymmetries have also been evaluated
using two other estimators which rely on different assumptions of the acceptance variations,
namely the difference method (DM, in the following) and the geometric mean method
(GMM), which are commonly used in evaluating asymmetries.
The DM was used to extract the final values of the asymmetries published in [15]. The evaluation of the asymmetries is performed separately for the two target cells and, after checking their
compatibility, they are combined by taking weighted averages. The asymmetries C,k and S,k ,
where k = u, d indicates the target cell, were evaluated from the number of events with the two
target spin orientations (+ spin up, and spin down) by fitting the quantities
m
Dj,k
(j ) =
Nj,k
(j ) rk Nj,k
(j )
+
Nj,k
(j ) + rk Nj,k
(j )
j = C, S
(29)
RAPID COMMUNICATION
57
with the functions C,k sin C and S,k sin S . The normalisation factor rk was taken equal to
the ratio of the total number of detected events in the two orientations of the target polarisation:
+
/Nh,tot,k
. This procedure is correct if the difference of the absolute value of the target
rk = Nh,tot,k
polarisation before and after the spin reversal is negligible. For what concerns the acceptances,
+
they cancel in Eq. (29) as long as the ratio aj,k

(j )/aj,k
(j ) is constant in j .
In the GMM, again the asymmetries are evaluated separately for the two target cells and then
combined by taking weighted averages. The method consists in building the measured quantities

+
+
Nj,k
(j ) Nj,k
(j + ) Nj,k
(j ) Nj,k
(j + )

Gm
(30)
(
)
=
.
j
j,k
+
+
Nj,k
(j ) Nj,k
(j + ) + Nj,k
(j ) Nj,k
(j + )
Assuming:
a negligible difference in the absolute value of the target polarisation before and after the
spin reversal (as for the DM), and
that the acceptances satisfy the relation
+
aj,k
(j )
+
aj,k
(j + )
(j )
aj,k
aj,k
(j + )
(31)
the asymmetries are evaluated by fitting the quantities Gm

j,k (j ) with the functions C,k
sin C and S,k sin S in the range 0 C,S < .
The advantage of this method is that the luminosity cancels. Still, the requirement on the
acceptance stability is more demanding than in the case of the ratio product method, which was
thus used to evaluate the final asymmetries.
The fitted asymmetry values from the three methods were very close. Given the advantages of
the RPM, the evaluation of the asymmetries with the GMM and DM has been considered only
a cross-check of the result, and the results of the experiment given in the following are those
obtained using the RPM method.
5.6. 2-D fits
In an entirely different approach, we have estimated the Collins and Sivers asymmetries using
the standard linear least square method (LSM) and fitted in each kinematics bin the measured
asymmetries
+
+
Nj,u
(h , S ) Nj,d
(h , S )

Aj (h , S ) =
Nj,u (h , S ) Nj,d (h , S )
(32)
with the function

H (h , S ) = a1 + a2 sin(h + S ) + a3 sin(3h S )
+ a4 sin(h S ) + a5 cos(h S ).
(33)
The first parameter should be one. The second, third and fourth terms arise from the transverse
component of the target nucleon spin when the lepton beam is unpolarised, while the last term
originates from the interaction between a longitudinally polarised lepton beam and a transversely
RAPID COMMUNICATION
58
polarised target (see Ref. [42]). The second term is the Collins term. The fourth term is the
Sivers term. The last term has also physics interest of its own, it is related to the g1T transverse
momentum distribution function.
In order to have enough statistics in each (h , S ) bin, we have plotted the data in 8 bins of h
and 8 bins of S . Each kinematic bin is thus split in 64 (h , S ) bins. We have performed both 5parameter fits and 4-parameter fits, in this second case fixing a1 to its expected value of 1. Also,
we have done two independent fits, one minimising a 2 with a linear LSM, the second using
MINUIT [43]. The results of the four fits are in excellent agreement, and both the fitted Collins
and Sivers asymmetries and their errors turn out to be essentially identical to the values given by
the one-dimensional fits, as expected from the orthogonality of the different terms. The Collins
and Sivers asymmetries as given by the 2-D fit turn out to be slightly correlated (the correlation
coefficient ranges from 0.25 to 0.25), a known effect due to the considerably non-uniform
population of the 64 (h , S ) bins.
A full discussion of the procedure and of the results will be the subject of a separate paper.
5.7. Systematic errors
As spelled out in the previous sections, all tests performed in the different data taking periods
did not give any evidence for the presence of systematic effects.
The conclusive test was to look at the compatibility of the physics results obtained separately
for all the data taking periods. As already mentioned, from 2002 to 2004 data were collected in
5 periods and the compatibility test is significant. It has to be stressed that the 2002 data from
which the results have already been published, have been reanalysed using the slightly different
event selection and analysis described here. The published data turned out to be very close to the
new ones, with differences of the order of half of the systematic errors only in the less populated
bins, essentially because of the different method used to evaluate the asymmetries (RPM instead
of the DM used previously).
The test was performed separately for all hadron and leading hadrons asymmetries, following
the procedure described in Section 5.2. The Collins and Sivers asymmetries in each x, pTh and
z bin, for positive and negative hadrons, obtained in the 5 periods were compared with their
weighted mean. The pulls are defined as
Pk,j
m
Am
k,j Aj
=
,
A2 m A2 m
k,j
Am
j
k = 1, 5,
(34)
are the weighted means of the asymmetries, and are expected to follow a normal
where
standard distribution. In total there are 500 Pk,j values for the leading hadron asymmetries and
520 for the all hadron asymmetries.
The distribution of the Pk,j values for all hadrons and for leading hadrons are shown in Fig. 16.
As expected, these pulls follow a normal distribution. A very good agreement with purely statistical fluctuations has been seen for all the various subsamples of the asymmetries; the RMS of
the different distributions are given for completeness in Table 4.
Since even in this last test we could not observe any indication for systematic effects, we
concluded that the systematic errors due to acceptance and efficiency effects are considerably
smaller than the statistical errors.
RAPID COMMUNICATION
59
Fig. 16. Distributions of Pk,j (see text) for the leading hadron (left) and all hadron (right) asymmetries.
Table 4
RMS of the pull distributions for the different samples of asymmetries
Asymmetries
RMS
Collins
0.882 0.056
Leading
Sivers
0.895 0.057
Positive hadrons
Collins
0.919 0.057
All
Sivers
0.974 0.060
Collins
1.043 0.066
Leading
Sivers
0.971 0.061
Negative hadrons
Collins
1.092 0.068
All
Sivers
0.991 0.061
Leading hadrons
0.950 0.030
0.996 0.031
All hadrons
Positive hadrons
0.919 0.029
1.026 0.032
Negative hadrons
The asymmetry scale uncertainty due to the uncertainties on PT is of 5%. The error on the
dilution factor f , which takes into account the uncertainty on the target composition, is of the
order of 6%. When combined in quadrature, these errors give a global scale uncertainty of 8%.
6. Results and comparison with models
6.1. Measured asymmetries
The results from the different data taking periods have been combined by making the standard
weighted mean.
Plots of the measured values of the Collins and Sivers asymmetries AColl and ASiv for the
20032004 leading hadron sample against the three kinematic variables x, z and pTh are given
in Fig. 17. Full points and open points refer to positive and negative hadrons, respectively. The
errors shown in the figure are only statistical. The same asymmetries are shown in Fig. 18 for the
20032004 all hadron sample. Again, the errors shown in the figure are only statistical.
The improvement in statistics with respect to the 2002 published result is clearly visible in
Fig. 19, where, as an example, the published Collins asymmetry data for all positive (left) and all
negative (right) hadrons are compared with the 20032004 results.
RAPID COMMUNICATION
60
Fig. 17. Collins asymmetry (top) and Sivers asymmetry (bottom) against x, z and pTh for positive (full circles) and
negative leading hadrons (open circles) from 20032004 data. Error bars are statistical only. In all the plots the open
circles are slightly shifted horizontally with respect to the measured value.
As already mentioned, the 2002 data have been reanalysed using the event selection and the
analysis described in this paper. The effect on the measured asymmetries is very small, and is
mainly due to use of the RPM in evaluating the raw asymmetries. As an example, Fig. 20 shows
the comparison between the published and the new results for the Collins asymmetry. The new
values from the 2002 data have been combined with the results from the 20032004 data to
evaluate the final asymmetries.
The overall results from 20022004 deuteron target for the leading hadron sample and for the
all hadron sample are given in Figs. 21 and 22, respectively. All these measured asymmetries are
available on HEPDATA [41].
6.2. Comments and comparison with models
As apparent from Figs. 21 and 22, all the measured asymmetries are small, if any, and compatible with zero. This trend already characterised the published data of the 2002 run, and is
confirmed by the new data with considerably improved precision. Small asymmetries is not a
surprise. From the very beginning it was predicted that transverse spin effects be small in the
deuteron due to the opposite sign which was expected for the u and d distributions, causing cancellations on the asymmetries of an isoscalar target, very much like in the helicity case. Still, it
was not obvious that they would have been so small.
An analysis of the results on the deuteron can be done only in conjunction with corresponding
proton data, which up to now have been measured only by the HERMES Collaboration [14].
RAPID COMMUNICATION
61
Fig. 18. Collins asymmetry (top) and Sivers asymmetry (bottom) against x, z and pTh for positive (full circles) and
negative hadrons (open circles) from 20032004 data. Error bars are statistical only. In all the plots the open circles are
slightly shifted horizontally with respect to the measured value.
Fig. 19. Collins asymmetry against x for all positive (left) and all negative (right) hadrons from 2002 data [15] (triangles)
and from 20032004 data (circles). The 2002 values are slightly shifted horizontally with respect to the measured value.
The measured non-zero Collins asymmetry on the proton has provided convincing evidence that
both the transversity distribution T u(x) and the Collins mechanism 0T Duh (z) are not zero.
Independent evidence that the Collins mechanism is a real measurable effect has come from the
recent analysis of the Belle Collaboration [44,45]. Furthermore, the HERMES data on a proton
target have provided convincing evidence that the Sivers mechanism is also at work.
RAPID COMMUNICATION
62
Fig. 20. Collins asymmetry against x for positive (left) and negative (right) hadrons from 2002 data. The triangles are the
published results [15] and the circles the results of the new analysis. The published values are slightly shifted horizontally
with respect to the measured value.
Fig. 21. Overall results for Collins asymmetry (top) and Sivers asymmetry (bottom) against x, z and pTh for positive (full
circles) and negative leading hadrons (open circles) from 2002, 2003, and 2004 data. Error bars are statistical only. In all
the plots the open circles are slightly shifted horizontally with respect to the measured value.
It is fair to say that the accuracy of the present HERMES data has allowed to extract only the
leading contribution to the proton transverse asymmetry, that is the u quark contribution. Also,
the interpretation of the HERMES data has led to surprising assumptions about the relative size
of the favoured and unfavoured spin-dependent fragmentation functions. A global analysis using
RAPID COMMUNICATION
63
Fig. 22. Overall results for Collins asymmetry (top) and Sivers asymmetry (bottom) against x, z and pTh for all positive
(full circles) and all negative hadrons (open circles) from 2002, 2003, and 2004 data. Error bars are statistical only. In all
the plots the open circles are slightly shifted horizontally with respect to the measured value.
all the available deuteron and proton data should allow now to provide first estimates of both the
u- and the d-quark contribution, and clearly constitutes the next step in this work.
In the following, first naive expectations, based on the simple formulas of Sections 1.2 and
1.3, are given for the deuteron asymmetries, then the new deuteron data are compared to a few
existing model calculations.
6.2.1. Collins asymmetry
Although the measured deuteron asymmetries refer to unidentified hadrons, in the following
it will be assumed that the hadrons be pions (actually more than 80% are pions), so that the
algebra considerably simplifies. Further simplification can be obtained by neglecting the sea
contribution (i.e. T q = T s = 0 and q = s = 0) and considering only the range 0.1 < x < 0.3.
This is justified by the fact that the PDFs are expected to be considerably different from zero in
the valence region, and the HERMES data show non-zero values in the range 0.05 < x < 0.3.
Assuming
+
Du = Dd = D1 ,
+
Dd = Du = D2 ,
0T Du = 0T Dd = 0T D1 ,
(35)
0T Dd = 0T Du = 0T D2 ,
(36)
and using Eq. (7), one gets for + on a proton target:

p, +
AColl
4T uv 0T D1 + T dv 0T D2

4uv D1 + dv D2
(37)
RAPID COMMUNICATION
64
and for :
p,
AColl
4T uv 0T D2 + T dv 0T D1

.
4uv D2 + dv D1
(38)
If, as suggested from the HERMES data, 0T D1 = 0T D2 , and taking D2 0.5D1 , dv 0.5uv ,
the previous expressions become
p, +
AColl
T uv 0T D1
uv
D1
(39)
and
p,
AColl
4 T uv 0T D1
,
2.5 uv
D1
(40)
respectively. As already stressed, the u-quark contribution is dominant.

For a deuteron target equations (37)(40) become respectively
+
T uv + T dv 40T D1 + 0T D2
,
uv + dv
4D1 + D2
T uv + T dv 0T D1 + 40T D2

,
uv + dv
D1 + 4D2
Ad,
Coll
Ad,
Coll
(41)
(42)
and
+
3 T uv + T dv 0T D1
,
7
uv
D1
3 T uv + T dv 0T D1

.
4.5
uv
D1
Ad,
Coll
Ad,
Coll
(43)
(44)
Both the + and the Collins asymmetries on the deuteron are proportional to T uv (x) +
T dv (x), therefore cancellation is expected to reduce considerably the effect which has been
measured on the proton. As a matter of fact, assuming T dv = 0 (no limit on the size of T dv is
provided by the HERMES data) one derives relations between the Collins asymmetry measured
by HERMES and COMPASS which are only marginally satisfied. Thus, the present precise data
on Ad,
Coll should allow to extract T dv .
This was not the case in so far. Three global analyses have been performed with the published
data, trying to derive bounds on the transversity distributions and the Collins fragmentation functions. In Ref. [46] the Soffer bound |T q| = (q + q)/2 was used, a fit of the HERMES data set
was performed, and the Collins functions were extracted. Two different scenarios for favoured
and unfavoured Collins fragmentation functions were considered, but the fits always favoured a
relation 0T D1 0T D2 . The comparison of the fit results with the COMPASS data shows a
fair agreement, as apparent from Fig. 23, although the data do not exhibit the trend with x which
is suggested by the model. The upper and lower curves in the figures correspond to the 1 errors
of the fitted parameters.
In Ref. [27] a chiral quarksoliton model was used for the transversity distributions, and
the Collins fragmentation function was derived from a fit to the HERMES data, which do not
constrain the T d distribution. A comparison with the present COMPASS results shows again a
fair agreement (Fig. 24). The upper and lower curves in the figures correspond to the uncertainty
in the Collins fragmentation functions as obtained from the fit. Independent extraction of the
RAPID COMMUNICATION
65
Fig. 23. Comparison between the present results for positive (left) and negative (right) hadrons with the calculations of
Ref. [46] (scenario 1).
Ref. [27].
Collins function was performed by fitting the Belle data. The result was found to be compatible
with the one obtained fitting the HERMES data.
Similar results were obtained in Ref. [47]. Two different scenarios were used for transversity,
either T q = q, or the Soffer bound, and the Collins fragmentation functions were extracted
from a fit to the HERMES data. The fits were very good in both cases. The extracted Collins
functions were then used to reproduce the published 2002 COMPASS data. The agreement is
acceptable for both scenarios, as apparent from Fig. 25, although also in this case the expected
increase of AColl with increasing x is not manifested by the COMPASS data. The upper and
lower curves in all the figures correspond to 1 deviations of the parameters. Also in this case
the Belle data are reproduced, and one can conclude that the Collins mechanism in SIDIS and in
e+ e hadrons are the same.
To summarise, the new data are compatible with the uncertainty bands given by the phenomenological calculations, but the trend of the data with x is not the one suggested by the central
value of the calculations: this is a clear indication that the COMPASS data will surely help in
constraining the parameters of the models.
RAPID COMMUNICATION
66
Ref. [47], where two scenarios were tested for transversity: T q = q was assumed in one scenario (upper figures),
while T q = (q + q)/2 was assumed for the calculations of the lower figures.
6.2.2. Sivers asymmetry

Also in this case it is useful to consider the expressions one obtains for ASiv in the hypothesis
that all hadrons are pions. Again, the simplified analysis is restricted to the valence region.
Neglecting the sea contribution (i.e. T0 q = T0 s = 0 and q = s = 0 at all x) and assuming
+
+
Du = Dd = D1 and Dd = Du = D2 , on a proton target, from Eq. (14) one gets for + :

p, +
ASiv
4T0 uv D1 + T0 dv D2
4uv D1 + dv D2
(45)
4T0 uv D2 + T0 dv D1
.
4uv D2 + dv D1
(46)
and for :
p,
ASiv
Assuming D2 0.5D1 , dv 0.5uv , the previous expressions become

p, +
ASiv
T0 uv
uv
(47)
2T0 uv + T0 dv
,
2.5uv
(48)
and
p,
ASiv
RAPID COMMUNICATION
67
Ref. [26].
respectively. Since the Sivers asymmetries for as measured by HERMES is about zero, in
this very simplified treatment it follows that
T0 dv 2T0 uv .
(49)
For a deuteron target the Sivers asymmetries can be written as

+
Ad,
Siv
T0 uv + T0 dv
uv + dv
(50)
T0 uv + T0 dv
,
uv + dv
(51)
and
Ad,
Siv
d,
which implies Ad,
Siv ASiv . The approximatively zero Sivers asymmetries for positive and
negative hadrons observed in COMPASS require
T0 dv T0 uv ,
(52)
a relation which is also obtained in some models.

From Eqs. (47)(49), a relation between the Sivers asymmetry measured by COMPASS and
those measured by HERMES can be derived
p, +
A
(53)
Siv ,
1.5
which is only marginally satisfied by the data.
Also the Sivers data have been looked upon independently by three different groups. In
Ref. [26] a fit of the HERMES data was performed, on the assumption that T0 d(x) = T0 u(x).
A good agreement with the HERMES data was obtained, and a zero asymmetry in case of a deuterium target was predicted, as shown in Fig. 26. The curves in the figure indicate the expected
size of the effect on ASiv of the 1/Nc -corrections. The COMPASS data fall well within the band
resulting from the model.
The authors of Ref. [48] could estimate the Sivers functions by fitting both the HERMES
and the published COMPASS data getting T0 dv T0 uv . Leading Order MRST01 sets of
+
Ad,
Siv
Ad,
Siv
RAPID COMMUNICATION
68
Ref. [48].
Ref. [46].
unpolarised distribution functions [49] were used, together with Kretzers set of fragmentation
functions [50]. A very good agreement with the experimental data was reached, as apparent from
Fig. 27, where the upper and lower curves correspond to 1 deviation of the parameters. In this
model the z and pTh dependence of the single-spin asymmetries are also well described.
In Ref. [46] it was assumed that the final hadron transverse momentum is the transverse momentum in the Sivers function, i.e. a collinear fragmentation was assumed. GRV98 leading order
distribution functions [51] were used, along with Kretzers fragmentation functions. The fit of
the HERMES data is very good, and gave as a result T0 dv 2T0 uv , but the prediction for
COMPASS are not in agreement with the new data, as shown in Fig. 28. The upper and lower
curves in the figures correspond to 1 errors in the fitted parameters. Here again, a new global fit
using the deuteron data should be able to extract the Sivers function for the d-quark and improve
the agreement with the data.
Very recently, the smallness of the Sivers asymmetry for positive and negative hadrons on the
deuteron has been interpreted as evidence for the absence of gluon orbital angular momentum in
the nucleon [52].
RAPID COMMUNICATION
69
7. Conclusions
After providing the very first SIDIS data on transverse spin asymmetries on a transversely
polarised deuteron target, COMPASS now publishes the overall results from the data collected
in 2002, 2003, and 2004, increasing the statistics as compared to the published 2002 data by a
factor of 7, so that even at large x the errors on the measured asymmetries are only a few
percent.
All the measured asymmetries are small, mostly compatible with zero within the measurement
errors. Presently, the most likely interpretation, taking into account the corresponding measurements of the HERMES Collaboration on a proton target, is that in the COMPASS isoscalar target
there is a cancellation between the proton and the neutron asymmetries. Also, the independent
evidence provided by the Belle data that the Collins effect is a real physical mechanism guarantees that the transversity distributions T q(x) can be extracted from the single-spin asymmetries
measured in SIDIS. A global analysis of all the presently available data is now mandatory, and
the inclusion of the new precise deuteron data from COMPASS will surely allow to provide
first estimates of the down quark transversity distribution T d and of the Collins fragmentation
functions 0T D. Already now the COMPASS data for the Sivers asymmetry provide convincing
evidence on the cancellation of the u- and d-quark Sivers distribution functions.
It has to be stressed, however, that the measured effects are rather small, of the order of a
few per cent at most, and that flavour separation requires much larger statistics than presently
available. Also, within the present errors, the physics interpretation is not straightforward and
has led to some surprises. The present data are of fundamental importance because they have
opened up the road to transverse momentum dependent distribution and fragmentation functions,
but they will by no means suffice to determine these new functions. New data are needed, and
particularly, new proton data. HERMES is still finalising the analysis of their 2005 proton run,
which will double the statistics of their analysed data. The Belle Collaboration is producing more
and more accurate data on the Collins fragmentation function. In the near future, COMPASS
plans proton runs, which should result in particularly reduced error bars at large x, thanks to
the much improved geometrical acceptance of the spectrometer which is obtained using the new
COMPASS polarised target magnet. A global analysis will then again be necessary, and from all
those measurements it should be able to provide the u- and d-quark transversity distributions.
Acknowledgements
We thank M. Anselmino, X. Artru, V. Barone, and A. Prokudin for many interesting discussions. We gratefully acknowledge the support of the CERN management and staff, as well as the
skills and efforts of the technicians of the collaborating institutes.
References
[1] G. Bunce, et al., Phys. Rev. Lett. 36 (1976) 1113.
[2] G.L. Kane, et al., Phys. Rev. Lett. 41 (1978) 1689.
[3] See, i.e. A. Efremov, O. Teryaev, Sov. J. Nucl. Phys. 39 (1984) 962, and the papers of the Lund team on polarisation.
[4] J. Antille, et al., Phys. Lett. B 94 (1980) 523.
[5] V.D. Apokin, et al., in: Proceedings of the 7th International Symposium on High Energy Spin Physics, vol. 2,
Protvino, USSR, 1986, Institute of High Energy Physics, Serpukhov, USSR, 1987, p. 83.
[6] Fermilab E704 Collaboration, A. Bravar, et al., Phys. Rev. Lett. 77 (1996) 2626.
RAPID COMMUNICATION
70
[7] European Muon Collaboration, J. Ashman, et al., Phys. Lett. B 206 (1988) 364;
European Muon Collaboration, J. Ashman, et al., Nucl. Phys. B 328 (1989) 1.
[8] J.P. Ralston, D.E. Soper, Nucl. Phys. B 152 (1979) 109.
[9] X. Artru, M. Mekhfi, Z. Phys. C 45 (1990) 669.
[10] R.L. Jaffe, X. Ji, Phys. Rev. Lett. 67 (1991) 552.
[11] J. Collins, Nucl. Phys. B 396 (1993) 161.
[12] G. Bunce, et al., Part. World 3 (1993) 1.
[13] HELP Collaboration, B. Vuaridel, et al., CERN reports LEPC 93-14, LEPC/P7, 1993.
[14] HERMES Collaboration, A. Airapetian, et al., Phys. Rev. Lett. 94 (2005) 012002.
[15] COMPASS Collaboration, V.Y. Alexakhin, et al., Phys. Rev. Lett. 94 (2005) 202002.
[16] V. Barone, A. Drago, P.G. Ratcliffe, Phys. Rep. 359 (2002) 1.
[17] R.L. Jaffe, X. Ji, Nucl. Phys. B 375 (1992) 527.
[18] A. Kotzinian, Nucl. Phys. B 441 (1995) 234.
[19] P.J. Mulders, R.D. Tangerman, Nucl. Phys. B 461 (1996) 197.
[20] X. Artru, The transverse spin, in: Proceedings of 10th Rhodanien Seminar, hep-ph/0207309.
[21] D. Sivers, Phys. Rev. D 41 (1990) 83.
[22] S.J. Brodsky, D.S. Hwang, I. Schmidt, Phys. Lett. B 553 (2003) 223.
[23] J.C. Collins, Phys. Lett. B 536 (2002) 43.
[24] A.V. Belitsky, X. Ji, F. Yuan, Nucl. Phys. B 656 (2003) 165.
[25] D. Boer, P.J. Mulders, Phys. Rev. D 57 (1998) 5780.
[26] J.C. Collins, et al., Phys. Rev. D 73 (2006) 014021.
[27] A. Efremov, et al., Phys. Rev. D 73 (2006) 094025.
[28] COMPASS Collaboration, E.S. Ageev, et al., Phys. Lett. B 633 (2006) 25.
[29] COMPASS Collaboration, E.S. Ageev, et al., Phys. Lett. B 612 (2005) 154;
COMPASS Collaboration, V.Y. Alexakhin, et al., The deuteron spin dependent structure function g1d and its first
moment, in preparation.
[30] COMPASS Collaboration, The COMPASS experiment at CERN, in preparation.
[31] J. Ball, et al., Nucl. Instrum. Methods A 498 (2003) 101.
[32] Spin Muon Collaboration, D. Adams, et al., Nucl. Instrum. Methods A 437 (1999) 23.
[33] Y. Giomataris, et al., Nucl. Instrum. Methods A 376 (1996) 29;
F. Kunne, Nucl. Phys. A 721 (2003) 1087c.
[34] F. Sauli, Nucl. Instrum. Methods A 386 (1997) 531;
B. Ketzer, et al., IEEE Trans. Nucl. Sci. 49 (2002) 2403.
[35] V.N. Bychkov, et al., Nucl. Instrum. Methods A 556 (2006) 66.
[36] E. Albrecht, et al., Nucl. Instrum. Methods A 504 (2003) 354.
[37] C. Bernet, et al., Nucl. Instrum. Methods A 550 (2005) 217.
[38] L. Schmitt, et al., IEEE Trans. Nucl. Sci. 51 (2004) 439.
[39] For all the COMPASS off-line packages see http://wwwcompass.cern.ch/compass/software/offline/welcome.
html#software.
[40] R. Brun, F. Rademakers, Nucl. Instrum. Methods A 389 (1997) 81;
See also http://root.cern.ch/.
[41] HEPDATA: The Durham HEP Database, http://durpdg.dur.ac.uk/HEPDATA/index.html.
[42] A.V. Efremov, Ann. Phys. 13 (2004) 651.
[43] F. James, MINUIT, CERN Program Library Long Writeup D506, http://wwwasdoc.web.cern.ch/wwwasdoc/
WWW/minuit/minmain/minmain.html.
[44] Belle Collaboration, K. Abe, et al., Phys. Rev. Lett. 96 (2006) 232002.
[45] A. Ogawa, et al., Spin dependent fragmentation function at Belle, hep-ex/0607014.
[46] W. Vogelsang, F. Yuan, Phys. Rev. D 72 (2005) 054028.
[47] M. Anselmino, et al., RICH + AGS annual Users Meeting June 6, 2006, http://www.bnl.gov/rhic_ags/
users_meetings/workshops/basp, and paper in preparation.
[48] M. Anselmino, et al., Phys. Rev. D 72 (2005) 094007.
[49] A.D. Martin, et al., Phys. Lett. B 531 (2002) 216.
[50] S. Kretzer, Fragmentation functions from flavour-inclusive and flavour-tagged e+ e annihilations, Phys. Rev. D 62
(2000) 054001.
[51] M. Gluck, E. Reya, A. Vogt, Z. Phys. C 67 (1995) 433.
[52] S.J. Brodsky, S. Gardner, Evidence for the absence of gluon orbital angular momentum in the nucleon, hepph/0608219.
Phenomenology of the triplet seesaw mechanism with

gauge and Yukawa mediation of SUSY breaking
Filipe R. Joaquim a,b , Anna Rossi a,
a Dipartimento di Fisica G. Galilei, Universit di Padova, I-35131 Padua, Italy
b Istituto Nazionale di Fisica Nucleare (INFN), Sezione di Padova, I-35131 Padua, Italy
Received 19 September 2006; received in revised form 16 November 2006; accepted 20 November 2006
Available online 15 December 2006
Abstract
We thoroughly discuss a new supersymmetric grand unified scenario of the triplet seesaw mechanism
where the exchange of heavy SU(2)W triplet states generates both neutrino masses and soft supersymmetry
breaking terms. This framework, recently proposed by us in a previous work, is highly predictive since it
contains only three free parameters connecting low-energy neutrino parameters, lepton and quark flavour
violation, sparticle and Higgs boson spectra and electroweak symmetry breakdown. These three parameters
are the triplet mass MT , the effective supersymmetry breaking scale BT and a coupling constant . We
perform a complete analysis of the parameter space taking into account the present experimental constraints
and considering different types of neutrino spectrum. A special emphasis is given to the particular features
of the sparticle and Higgs spectra and to the model independent predictions obtained for the processes
eX, e conversion in nuclei, eY and Y (X = , ee, Y = , ee, ). In the appendices,
we present some technical aspects relevant for our analysis.
1. Introduction
It has long been recognised that the realm of neutrino physics may offer some insights on the
search for physics beyond the Standard Model (SM). The evidence of non-vanishing neutrino
masses and of leptonic mixing angles, as provided by neutrino oscillations, calls for an extension of the SM particle content. Namely, the d = 5 operator Y (LH )2 /ML , describing neutrino
E-mail addresses: joaquim@pd.infn.it (F.R. Joaquim), arossi@pd.infn.it, anna.rossi@pd.infn.it (A. Rossi).

doi:10.1016/j.nuclphysb.2006.11.030
72
F.R. Joaquim, A. Rossi / Nuclear Physics B 765 (2007) 71117
masses at the effective level [1], can be generated by decoupling some heavy degrees of freedom
at the scale ML where lepton number (L) is broken. This is the essence of the well-celebrated
seesaw mechanism which can be realized at the tree-level by exchanging either singlet fermions
N [2], or SU(2)W triplets T with zero [3] or non-zero hypercharge [4], at ML . However, we
should notice that, from a theoretical perspective, the concept of non-zero neutrino masses by
itself does not tell us much about the new physics beyond ML . On the other hand, alternative
signals of lepton flavour violation (LFV) (besides neutrino oscillations) would be a clear and
dramatic manifestation of new physics, since they are strongly suppressed within the SM by the
smallness of neutrino masses. For this reason, and taking into account the increasing sensitivity of the present and future experiments, it is crucial to explore theoretical frameworks where
such LFV processes can be sizeable. A typical example is the minimal supersymmetric standard
model (MSSM) extended with renormalizable interactions that give rise to the d = 5 effective
neutrino mass operator. Hence, at least two attractive features of the supersymmetric version of
the above seesaw scenarios are worth to be recalled:
(1) supersymmetry (SUSY) alleviates the hierarchy problem of the SM, which would be exacerbated by the presence of one more high scale, ML (besides the Planck mass scale MPL ) [5];
(2) lepton flavour violating (LFV) processes (otherwise unobservable) can be enhanced
through one-loop exchange of the lepton superpartners if their masses are not too far from the
electroweak scale and do not conserve flavour [6].
Regarding the latter aspect, most of the literature has been focussing on the most conservative
scenario of universal sfermion masses at a scale larger than ML , which is realized in either minimal supergravity or gauge mediated supersymmetry breaking (GMSB) models with very large
SUSY breaking mediation scale. In such cases, flavour non-conservation in the sfermion masses
arises from renormalization group (RG) effects due to flavour-violating Yukawa couplings [79]
encoded, at the effective level, in Y . In this respect, it has been pointed out that, in the seesaw
realization with non-zero hypercharge triplets, the flavour structure of the slepton mass matrix
m2 can be univocally determined in terms of the low-energy neutrino parameters [9]. In contrast,
L
within the singlet seesaw, the determination of m2 from low-energy observables requires model
L
dependent assumptions [1012].
Recently we have proposed a new supersymmetric scenario of the triplet seesaw mechanism
in which the soft SUSY breaking (SSB) parameters of the MSSM are generated at the decoupling
of the heavy triplets. Moreover, the mass scale of such SSB terms is fixed only by the triplet SSB
bilinear term BT [13]. This scenario turns out to be highly predictive in the sense that it relates
neutrino masses, LFV in the sfermion sector, sparticle and Higgs boson spectra and electroweak
symmetry breaking (EWSB). In this work we aim to further elucidate and discuss a more general
version of this new framework including effects (previously neglected in Ref. [13]) which also
entail quark flavour violation (QFV).
The paper is organised as follows. In Section 2 we review the main features of the SUSY
triplet seesaw mechanism. Its embedding into the SU(5) gauge group is described in Section 3
where the role played by the heavy triplets as messengers of SUSY breaking is also discussed.
It will be shown that the exchange at the quantum level of the triplet states generate all the
MSSM SSB mass parameters. The results of the related analytical evaluations are presented and
analysed in Section 4. We proceed in Section 5 with a detailed description of the flavour structure exhibited by the SSB terms. Then, in Section 6, we bring forward the phenomenological
analysis of our framework. More specifically, we detail the experimental constraints imposed on
the parameter space (Section 6.1) and present our numerical results (Section 6.2). Section 7 is
devoted to the relevant phenomenological predictions such as the sparticle and Higgs boson mass
73
spectra (Section 7.1) and the expected size of lepton flavour violation (Section 7.2). Afterwards,
we discuss the predictions for several LFV processes ( e , , e , eee,
ee, e, , eee and e conversion in nuclei) and the peculiar correlations which arise in our scenario, together with a complete numerical analysis (Section 7.3).
Our conclusions and summary are drawn in Section 8. Several technical aspects are collected
in appendices: Appendix A is devoted to the generalization of the method based on the wavefunction renormalization to derive the SSB mass parameters; Appendix B presents our analytical
calculations of the MSSM coefficients for the LFV j i Z operators; finally, Appendix C regards
the computation of the box diagrams relevant for the ee and e amplitudes.
2. Recalling the triplet seesaw mechanism
Before starting the discussion of the main subject of our paper, let us briefly review the
key features of the supersymmetric triplet seesaw mechanism. The requirement of a holomorphic superpotential implies introducing the triplets as super-multiplets T = (T 0 , T + , T ++ ),
T = (T 0 , T , T ) in a vector-like SU(2)W U (1)Y representation, T (3, 1), T (3, 1)
[9,14]. The relevant superpotential terms are:
1
1
1 ij
YT Li T Lj + 1 H1 T H1 + 2 H2 T H2 + MT T T + H2 H1 ,
2
2
2
(1)
where i, j = e, , are family indices, Li are the SU(2)W lepton doublets and H1 (H2 ) is the
ij
Higgs doublet with hypercharge Y = 1/2(1/2). The matrix YT is a 3 3 symmetric matrix
and 1,2 are dimensionless unflavoured couplings. MT and denote the mass parameters of the
triplets and the Higgs doublets, respectively.
By decoupling the triplet states at the scale MT , one obtains the d = 5 effective operator
Y (LH2 )2 /2ML where Y /ML is identified as follows:
1 ij
2 ij
Y =
Y .
ML
MT T
(2)
At the electroweak scale the Majorana neutrino mass matrix emerges and is given by1 :
mij
=
v22 ij v22 2 ij
Y =
Y ,
ML
MT T
(3)
where v2 = v sin = H2 (v = 174 GeV). It is worth to emphasize that the flavour structure of
the matrix YT is the same as that of Y and hence of the neutrino mass m .
Without loss of generality, we choose to work in the basis where the charged-lepton Yukawa
matrix Ye is diagonal. Therefore, all the information about low-energy lepton flavour violation
is contained in Y or m :
m = U mD
U ,
mD
= diag(m1 , m2 , m3 ),
(4)
1 The appearance of neutrino masses can be also interpreted in terms of non-vanishing vacuum expectation value (vev)
v2
2 2 . In our scenario the values of T 0 will be smaller

induced by the EWSB on the scalar neutral state T 0 , i.e. T 0 = M
T
than 103 GeV, which is much below the upper bound of 2 GeV inferred from the global fits of the electroweak
data [15].
74
where m1 , m2 , m3 are the neutrino mass eigenvalues and the unitary lepton mixing matrix U can
be written as

U = V diag 1, ei1 , ei2 ,
c12 c13
s12 c13
s13 ei
V = s12 c23 c12 s23 s13 ei c12 c23 s12 s23 s13 ei
(5)
s23 c13 .
i
i
s12 s23 c12 c23 s13 e
c12 s23 s12 c23 s13 e
c23 c13
The mixing matrix V is responsible for LFV and, in particular, for neutrino oscillations. We have
adopted the notation sij sin ij , cij cos ij for the three mixing angles 12 , 23 and 13 , and
denoted the Dirac and Majorana CP-violating phases by and 1,2 , respectively.
The relations (3) and (4) clearly show that the high-energy structure of YT can be determined
by the low-energy neutrino parameters (taking also into account the RG effects on the d = 5
operator which, however, do not introduce unknown flavour structures). The implications of such
simple flavour structure become dramatic when one considers LFV induced by RG effects in the
mass matrix m2 of the left-handed sleptons [9]. Assuming, for instance, flavour-blind SSB terms
L
at the grand-unification scale MG (m2 = m20 1), the form of the LFV entries is
L

MG
, i
= j,
m2L ij m20 YT YT ij log
MT
which, in terms of the neutrino parameters, read

2

2
2

D 2
MG
MG
2 MT
2 MT
mL ij m0
m
V m V ij log
m
log
m
.
0
ij
MT
MT
2 v22
2 v22
(6)
(7)
We note that the L-conserving combination YT YT V(mD

)V depends only on the neutrino
oscillations parameters, since the Majorana phases have been absorbed. From here, one can
derive the relative size of LFV among the different leptonic family sectors [9]:
(m2 )
L
(m2 )e
L
2
[V(mD
) V ]
,
D
[V(m )2 V ]e
(m2 ) e
L
(m2 )e
L
2
[V(mD
) V ] e
.
D
2
[V(m ) V ]e
(8)
The above ratios depend only on the neutrino parameters which can be measured in low-energy
experiments. This observation renders the SUSY triplet seesaw mechanism much more predictive
when compared with the singlet one. Our scenario, therefore, constitutes a concrete and simple
realization of the so-called minimal lepton flavour violation hypothesis which has been recently
revived in Refs. [16,17]. In fact, relations like those shown in Eq. (8) cannot be obtained in the
latter case due to an ambiguity in the extraction of the high-energy neutrino parameters from
low-energy observables [10].
3. Gauge and Yukawa mediated SUSY breaking scenario
A brief comment on the SSB pattern is now in order to motivate and introduce the main
idea of this paper. We recall that the assumption of universality of the soft scalar masses at a
high scale below MPL may not be a justified one. As a matter of fact, flavour universality at
MPL , arising from some gravity-mediated SUSY breaking models [18], is likely to be spoiled
by Yukawa interactions in the energy range above MG [19,20]. Therefore, the common lore of
adopting the universality condition at MG should be regarded as a conservative approach to the
75
issue of flavour violation (FV). In this work, we discuss an alternative scenario which, in our
opinion, suggests a more motivated and predictive picture for the SSB pattern at high scales.
This is the consequence of the fact that the triplet-exchange at the quantum level gives rise to the
SSB terms2 of the MSSM, i.e., the triplets play the role of SUSY breaking messengers [13].
At first sight, the presence of extra SU(2)W triplet states at intermediate energies could spoil
the simple gauge coupling unification, which can be achieved within the MSSM. However, this
can be remedied by invoking a grand unified theory (GUT) where the triplets live in a complete
GUT representation, in such a way that gauge coupling unification can be preserved.3 To this
purpose, we arrange a SU(5) set-up where the T (T ) states fit into the 15 (15) representation,
15 = S + T + Z with S, T and Z transforming as S (6, 1, 23 ), T (1, 3, 1), Z (3, 2, 16 )
under SU(3) SU(2)W U (1)Y (the 15 decomposition is obvious).4
The SUSY breaking mechanism is parameterized by a gauge singlet chiral supermultiplet
X = SX + X + 2 FX , whose scalar SX and auxiliary FX components are assumed to acquire
a vev through some unspecified dynamics in the secluded sector (X is one of the goldstino
components). In order to prevent the tree-level generation of SSB terms in the observable sector,
the couplings of X to the MSSM fields must be forbidden. At the same time, X must couple
with the 15 and 15 fields in order to trigger SUSY breaking in the messenger sector. Both these
requirements can be achieved by, e.g., imposing that the SU(5) model conserves the combination
of baryon and lepton number B L = Q + 45 Y , where Y are the hypercharges and
1
Q10 = ,
5
4
Q15 = ,
5
3
Q5 = ,
5
Q5H (5 H ) =

2 2
,
5 5
6
Q15 = ,
5
QX = 2.
(9)
The SU(5) matter multiplets are understood as 5 = (d c , L), 10 = (uc , ec , Q) and the Higgs doublets fit with their coloured partners t and t, like 5H = (t, H2 ), 5 H = (t, H1 ). Given this, the
relevant superpotential terms [13], consistent with the SU(5) and B L symmetries, are
1
WSU(5) = (Y15 5 15 5 + 5H 15 5H ) + Y5 5 5 H 10 + Y10 10 10 5H
2
+ M5 5H 5 H + X 15 15.
(10)
The form of WSU(5) makes explicit the fact that, thanks to the coupling with X, the 15 and
15 states act as messengers of both BL and SUSY breaking to the MSSM observable sector.
Namely, while SX only breaks BL, FX breaks both SUSY and BL. These effects are
tracked by the superpotential mass term M15 15 15, where M15 = SX , and the bilinear SSB
term B15 M15 15 15, with B15 M15 = FX . Once SU(5) is broken to the SM group5 we find,
2 In [21] the authors discussed the finite radiative contributions (arising from the decoupling of the triplet states)
relevant to the SSB trilinear couplings in connection with the generation of the electric dipole moments.
3 This is not the only possibility to maintain gauge coupling unification in the presence of extra states at intermediate
scales [22].
4 We find it interesting that such supersymmetric SU(5) with a 15, 15 pair may be realized in contexts based on string
inspired constructions [23,24].
5 For the sake of brevity, we have omitted in the SU(5) invariant superpotential (10) other terms, as those involving the
adjoint 24 representation responsible for the SU(5) breaking.
76
below the GUT scale MG [9],

W = W0 + WT + WS,Z ,
W0 = Ye ec H1 L + Yd d c H1 Q + Yu uc QH2 + H2 H1 ,
1
WT = (YT LT L + H2 T H2 ) + MT T T ,
2
1
WS,Z = YS d c Sd c + YZ d c ZL + MZ Z Z + MS S S.
2
(11)
Here, W0 denotes the MSSM superpotential where the standard notation for the supermultiplets
is understood. The terms relevant for neutrino mass generation [cf. (1)] are contained in WT
while the couplings and masses of the coloured fragments S and Z are included in WS,Z . From
the comparison of Eqs. (11) and (1), we observe that the BL invariance forbids the coupling
1 H1 T H1 (which is not relevant for neutrino masses) leaving only the term proportional to
2 . Consequently, the number of independent real parameters in WT is reduced to eleven: MT ,
and the nine from the symmetric matrix YT .
Notice that we have relaxed the strict SU(5) symmetry relations for the Yukawa interactions and the mass term M5 by assuming SU(5) breaking effects (due to insertions of the
24-representation), which are necessary to correct the relation Ye = YTd [25] and to solve the
doublettriplet splitting problem [26]. Thus, beneath the scale MG , the coloured partners t and t
are considered to be decoupled in order to adequately suppress dangerous d = 5 baryon number
violating operators. The only d = 5 operator, generated by the exchange of the 15-fragments, is
the L violating neutrino mass operator (see Fig. 1). Instead, Fig. 2 shows that additional d = 6
operators arise from the exchange of T , S and Z [9]. However, these are BL conserving and,
being suppressed by MT2 , are irrelevant for low-energy phenomenology. Hence, the presence of
Fig. 1. The d = 5 supergraph from the 15, 15 exchange in the SU(5) symmetric phase (left). Once the coloured states
t 5H have been splitted from the Higgs doublet partners H2 and decoupled at the scale MG , there is only the supergraph
(right) with T , T exchange generating the L violating d = 5 effective superpotential operator (LH2 )2 .
Fig. 2. The d = 6 supergraph from the 15 exchange in the SU(5) symmetric phase (left). In the SU(5) broken phase
( MG ), the latter splits into the remaining three supergraphes (right) which generate the BL invariant d = 6 Khler
operators (LL)(L L ), (d c d c )(d c d c ) and (d c L)(d c L ) at low-energy.
77
S and Z does not introduce new sources of baryon-number violation which could speed up the
decay of the proton.
As mentioned above, the BL conserving superpotential of Eq. (10) contains the M5 -term
from which the -parameter emerges in W0 . This mass parameter is not predicted by our model
and will be determined through the requirement of EWSB (notice that the coupling XH1 H2 is
forbidden by BL).
Also the coupling in Eq. (10) could include 24-insertions, therefore allowing different
masses for the 15-components. For simplicity, we consider the minimal case which implies a
common mass MS = MZ = MT = M15 at the GUT scale. The Yukawa-unification condition
YS = YT = YZ = Y15 ,
(12)
can either hold or not at MG , depending on the type and size of the SU(5) breaking effects. In
this respect, we can discuss the following two extreme scenarios:
(A) Eq. (12) holds at the GUT scale. As a consequence, flavour violation is extended to all the
couplings YS , YT and YZ , originating FV effects both in the lepton and quark sector. In
this work we shall focus on this general case.
(B) The states S and Z have negligible Yukawa couplings (Eq. (12) does not hold because of
SU(5) breaking). In this case, FV is confined to the lepton sector since the couplings in YT
are the only ones which exhibit a non-trivial flavour structure. We have already studied this
scenario in Ref. [13].
Beneath the messenger scale MT , the particle content of our model is that of the MSSM and
so, the superpotential reduces to the sum of the MSSM terms in W0 and the effective neutrino
mass operator
W<MT = W0 +
YT
(LH2 )(LH2 ).
2MT
(13)
As mentioned above, the vev of FX induces the only tree-level SSB terms which, below the GUT
scale, read (BT = B15 at MG )
T
L>M
soft = BT MT (T T + Z Z + S S) + h.c.
(14)
Such terms remove the mass degeneracy between the scalar and fermionic messenger components. To avoid tachyonic scalar messengers we require that FX < MT2 (or BT < MT ). At
the tree-level, the ordinary MSSM supermultiplets are degenerate as they do not couple to the
superfield X. Nevertheless, in the presence of the bilinear terms given in Eq. (14), the mass splitting is radiatively generated at the scale MT through the gauge and Yukawa interactions between
the messenger states S, T and Z and the ordinary MSSM fields. Our scenario can be, therefore,
regarded as a gauge and Yukawa mediated SUSY breaking realization of the triplet seesaw mechanism. This will become clear in the next section where we present the complete MSSM SSB
Lagrangian.
4. The SSB mass parameters
We now discuss the SSB terms of the MSSM Lagrangian LMSSM
which are generated at the
soft
can be
decoupling scale of the heavy states S, T and Z. Below MT , the most general LMSSM
soft
78
Fig. 3. One-loop diagrams for the trilinear terms Ae , Ad and Au (first and second rows) and the bilinear Higgs term
and ( ) denote the scalar and fermionic components of
BH and gaugino masses Ma (third row). The fields ()
S),
(T , Z),
(S,
Z)
the T (T ), Z(Z), S(S) superfields in such a way that = (T , Z, S), (T , Z), (S, Z) and = (T , Z,
for a = 1, 2, 3, respectively (the assignments for and are obvious).
written as:
m2 Q
+ dc m2 c dc + uc m2 c uc + m2H H H1
= L m2L L + ec m2ec ec + Q
LMSSM
soft
1 1
u
Q
d

2 + 1 M a a a
+ m2H2 H2 H2 + H1 ec Ae L + H1 dc Ad Q + uc Au QH
2

+ BH H2 H1 + h.c. ,
(15)
where we have adopted the standard notation for the slepton, squark and Higgs soft masses, the
trilinear couplings Ae,d,u , the gaugino masses Ma and the Higgs bilinear term BH .
At the mass scale MT , one-loop finite contributions are generated6 for Ae,d,u , Ma and BH . In
Fig. 3 we draw the one-loop diagrams for Ae , Ad and Au (first and second rows), BH and Ma
(third row). The evaluation of these diagrams yields:

3BT
Ye YT YT + YZ YZ ,
2
16
3BT
Au =
||2 Yu ,
16 2
Ae =
6 Since the RGE-induced splitting on the masses of the 15-fragments is small, we decouple all the components at the
common threshold MT .
79

2BT
YZ YZ + 2YS YS Yd ,
2
16
7BT 2
Ma =
g ,
16 2 a
3BT
BH =
(16)
||2 .
16 2
In these expressions, ga (a = 1, 2, 3) indicates the GUT normalized gauge coupling constants
(g1 = g2 = g3 at the unification scale) and the factor N = 7 in the r.h.s. of the gaugino masses
Ma corresponds to twice the Dynkin index of the 15-representation. A noteworthy feature of
the above equations is that both the A-terms and the Higgs doublet bilinear parameter BH are
generated in our scenario,7 once the Yukawa couplings and the parameter are present in the superpotential, respectively.8 This is in contrast with the minimal GMSB realizations, where only
the gaugino masses emerge at one-loop [27,28]. We remark that the triplet states, being SU(3)
singlets, would not communicate SUSY breaking to the gluino. This is one more motivation, besides the aforementioned gauge coupling unification requirement, to consider the GUT extension
of the triplet seesaw with the 15 representation.
Regarding the SSB squared scalar masses, the leading O(FX2 /MT2 ) = O(BT2 ) contributions
vanish at the one-loop level due to a cancelation among the different terms [29]. For example, in the case of the soft masses m2 , the one-loop diagrams driven by the exchange of the
L
F -component of T cancel against those from the exchange of the F -component of L. Nonvanishing O(FX2 /MT2 ) contributions for the SSB squared scalar masses arise at the two-loop
level.9 They are finite and can be evaluated either by diagrammatic computations or by means
of a generalization of the wave function renormalization method proposed in Refs. [30,31]. For
the sake of illustration, we have depicted in Fig. 4 some representative two-loop diagrams which
4 . In
generate contributions to m2 . The first diagram gives flavour-blind terms proportional to g1,2
Ad =
2 Y Y and to
turn, the remaining two diagrams generate LFV contributions proportional to g1,2
T T
2
g1,2,3
YZ YZ . In Appendix A we revisit the wave function renormalization approach and provide
general formulas to extract all the SSB terms just by knowing the one-loop anomalous dimension
matrices of the different fields above and below the messenger scale MT . Our computation leads
to the following result:

|BT | 2 21 4 21 4
27 2
2
2
mL =
g + g2
g + 21g2 YT YT
10 1
2
5 1
16 2

2
2

21 2
2
2
g1 + 9g2 + 16g3 YZ YZ + 18 YT YT + 15 YZ YZ
15

+ 3 Tr YT YT YT YT + 12YZ YS YS YZ + 3 Tr YZ YZ YZ YZ

+ 9YT YTZ YZ YT + 9 YT YT YZ YZ + h.c. + 3YT YTe Ye YT + 6YZ Yd Yd YZ ,
7 The SSB terms M , B and the trilinear couplings require interactions which violate the U (1) symmetry. In this
a
H
R
context, the messenger bilinear terms (14) are those responsible for the U (1)R breaking.
8 Although the parameter is not predicted by our model, we nevertheless assume that whatever mechanism generates , it does not generate BH .
9 These O(F 2 /M 2 ) two-loop contributions dominate over the O(F 4 /M 6 ) = O(B 4 /M 2 ) one-loop ones for M >
T
X
T
X
T
T
T
(4 YT /g 2 )BT . In the following such a hierarchy is fulfilled.
80
Fig. 4. Examples of two-loop diagrams which contribute to m2 . The upper diagrams generate contributions proportional
L
4 (left) and to g 2 Y Y (right). Instead, the lower diagram leads to a term proportional g 2 Y Y .
to g1,2
a Z Z
1,2 T T

m2ec =
m2dc
|BT |
16 2

42 4
g1 6Ye YT YT + YZ YZ Ye ,
5

7 4 21 4 56 4
g1 + g2 + g3 2Yd YZ YZ + 2YS YS Yd 3||2 Yu Yu ,
30
2
3
2

56 4 56 4
|BT |
2
g
g
=
+
6||
Y
Y
u
u ,
15 1
3 3
16 2

|BT | 2 14 4 56 4
16 2
2
g + g3
g + 48g3 YS YS
=
15 1
3
5 1
16 2

2

14 2
32
g1 + 6g22 + g32 YZ YZ + 32 YS YS + 4 Tr YS YS YS YS
15
3
2

T
+ 8YS YZ YZ YS + 6YZ YT YT YZ + 8 YS YS YZ YZ + h.c. + 10 YZ YZ

+ 2 Tr YZ YZ YZ YZ + 8YS Yd YTd YS + 2YZ Ye Ye YZ ,

m2Q =
m2uc
|BT |
16 2

|BT | 2 21 4 21 4
g1 + g2 3 Tr YT YT + YZ YZ Ye Ye
2
10
2
16

6 Tr YZ YZ + 2YS YS Yd Yd ,

m2H1 =

m2H2 =
|BT |
16 2

21 4 21 4
27 2
g1 + g2
g1 + 21g22 ||2 + 9||2 Tr Yu Yu + 21||4 .
10
2
5
(17)
One can immediately recognize that the diagonal gauge-mediated contributions follow the exf
f
pected form 2ka Ca Nga4 (k1 = 3/5, k2 = k3 = 1), where Ca is the quadratic Casimir of
2
f
1
for the fundamental representation
the f -particle [C1 = Yf2 = (Qf Tf3 )2 and C = n 2n
of SU(n)].
81
The above expressions hold at the decoupling scale MT and are, therefore, meant as boundary
conditions for the SSB parameters which then undergo (MSSM) RG running down to the lowenergy scale s . All the soft masses have the same scaling property msoft BT /(16 2 ) which,
on the basis of the naturalness requirement msoft O(102 103 GeV), implies that BT 10 TeV.
In principle, the SSB parameters also receive gravity mediated contributions of order the gravitino mass m3/2 F /MPL (F 2 = |FX |2 + is the sum of F -terms in the secluded sector).
We assume such contributions to be negligible, which is the case if MT 1016 GeV FX /F .
This also implies that the gravitino is lighter than the MSSM sparticles.
It is apparent from Eqs. (16) and (17) (and Figs. 3 and 4) that the gauge interactions participate
in the mediation of SUSY breaking in the gaugino and soft scalar masses. In the particular case
of the sfermion mass matrices, the associated terms constitute the flavour blind contribution.
Exactly the same situation occurs in pure GMSB models where flavour violation is automatically suppressed if the SUSY-breaking mediation scale is lower than the flavour or GUT scale
[27,28,32]. In our case, SUSY breaking is also mediated by the Yukawa interactions YS , YT ,
YZ and , giving rise to the bilinear parameter BH , the trilinear couplings Ae,d,u and to additional contributions to the Higgs scalar masses and the sfermion mass matrices.10 In this way,
FV is transmitted to the SSB mass parameters. Since the origin of FV originally comes from
the couplings Y15 (felt by the 5 supermultiplets), one expects that, due to the condition (12),
flavour violation is inherited with comparable size by the L and dc SSB parameters Ae , Ad , m2
L
and m2c . We emphasize that the FV entries of e.g., (m2 )ij (i

= j ) show up as finite radiative
d
L
contributions induced by BT at MT , and they are not significantly modified by the running evolution to low-energy. This is different from a previous work [9] where a common SSB scalar
mass m0 O(100 GeV) was assumed at MG and the dominant LFV contributions to m2 were
L
generated by RG evolution from MG down to the decoupling scale MT (see Eq. (6)). In such a
case, finite contributions like those in Eqs. (16) and (17) also emerge at MT . Nevertheless, they
are subleading with respect to the RG corrections, since BT is of the same order as m0 . Instead,
in the present picture, there is a hierarchy between the SSB parameter BT and the remaining ones
(see Eqs. (16) and (17)), BT2 (BT g 2 /16 2 )2 m20 [13].
5. Flavour structure in m2 and m2c from neutrino parameters
L
In this section we discuss in detail the features of the flavour structure which emerges in the
mass matrices m2 and m2c , and in the trilinear couplings Ae and Ad . These parameters depend
L
d
on the Yukawa couplings YT , YS , YZ . The matrix YT is determined at MT according to the
matching expressed by Eq. (3) i.e.
YT =
MT
m ,
v22
(18)
where the effective neutrino mass matrix m has undergone the MSSM RG running from lowenergy to MT . The matrices YS and YZ are iteratively determined at MT under the unification
constraint of Eq. (12). For the determination of the mass matrix m at low-energy (cf. Eq. (4))
we consider three different types of neutrino spectra:
10 Other examples of Yukawa mediated SUSY breaking can be found e.g. in [29,31,33].
82
1. Normal hierarchy (NH): m1 m2 < m3 , with

m22 = m21 + m2S ,
m23 = m21 + m2S + m2A .
(19)
2. Inverted hierarchy (IH): m3 m1 < m2 , with

m21 = m23 m2S + m2A ,
m22 = m23 + m2A .
(20)
3. Quasi degenerate (QD): m1 m2 m3 , with

m22 = m21 + m2S ,
m23 = m21 + m2S + m2A ,
m21 m2A .
(21)
The neutrino mass squared differences m2A = |m23 m22 | and m2S = m22 m21 are responsible
for the atmospheric and solar neutrino oscillations, respectively, together with the corresponding mixing angles 23 and 12 of the mixing matrix V (see Eq. (5)). From global analysis of
the neutrino oscillation data, the following best fit values are available (with their allowed 3
range) [34]:
3
eV2 ,
m2A = 2.2+1.1
0.8 10
5
eV2 ,
m2S = 7.9+1.0
0.8 10
sin2 23 = 0.5+0.18
0.16 ,
sin2 12 = 0.3+0.1
0.6 .
(22)
For the lepton mixing angle 13 the following upper bound has been settled (at 3 ) [34]:
sin2 13 < 0.043.
(23)
It is worth noticing that, for a NH spectrum, the largest entries of m (and, hence, of YT ) are
those in the block, where the largest neutrino mass (m2A )1/2 dominates.11 Moreover, they
are of comparable size due to maximal atmospheric mixing. On the other hand, the remaining
matrix elements are smaller by one or two orders of magnitude, depending on the value of 13
(which introduces the dependence on (m2A )1/2 to these entries). In the IH case, while (m )ee
and the block are of the same order, (m )e and (m ) e are generically smaller. Finally,
when the neutrino mass spectrum is of the QD type, the leading mass m1 renders the matrix
pattern less hierarchical and enhances the overall magnitude of m (or, equivalently, of YT ).
Let us now turn to the flavour structure of the SSB parameters which are generated at the
messenger scale MT . For simplicity of our discussion, we temporarily assume that the unification
condition (12) remains approximately valid at MT , YT YS YZ . In addition, we disregard
the terms proportional to the Yukawa couplings Ye,d in the expressions of m2 and m2c given in
L
d
Eqs. (17). Under these assumptions, we can write:

2
2
2

|BT | 2
D 2
e 2
mL ij
V m V ik ca ga kj 72 V mD
V kj 6 Tr mD
kj ,
2
16

2
3BT
(Ae )ij
(Ye )ii V mD
V ij ,
2
8

2
|BT | 2
D 2
mdc ij
V m V ik
16 2
2
2

cad ga2 kj 72 V mD
V kj 6 Tr mD
kj ,
11 These considerations are valid for vanishing CP-phase, = 0. From this point forward we restrict ourselves to this
limit. The discussion about the impact of
= 0 on our results is postponed to Section 7.3.
(Ad )ij
3BT
D 2
V m V ij (Yd )jj ,
8 2

=
MT
v22
83
2
,
(24)
62
176
e
e
d
d
d
where cae , cad are: c1e = 102
15 , c2 = 30, c3 = 16, c1 = 15 , c2 = 6 and c3 = 3 (summation of repeated indices is implicit). The flavour indices are i, j = e, , (d, s, b) for the slepton (squark)
matrices m2 and Ae (m2c and Ad ). Wherever necessary, the correspondences e d, s,
L
d
b are understood.
From Eqs. (3)(5), and considering the three different kinds of neutrino mass spectra shown
in (19)(21), we have:
2
m1 ij + m2S Vi2 Vj 2 + (m2A + m2S )Vi3 Vj 3 (NH, QD),
D 2
V m V ij =
(25)
m23 ij + m2A Vi2 Vj 2 + (m2A m2S )Vi1 Vj 1 (IH),
which manifestly shows that the FV (i

= j ) entries are always independent of the lightest neutrino mass. The explicit expressions in terms of the neutrino parameters read:
2
2

2

D 2
2 2
2
,
c23 s23 c23 s23 s12
c12
s13 + c12 s12 s13 s23
c23
V m V = m2A c13
D 2

V m V e = m2A c13 s13 s23 c12 (c23 s12 + c12 s23 s13 ) ,
D 2
V m V e = m2A c13 c23 s13 c12 (s23 s12 c12 c23 s13 ) ,
(26)
where the upper (lower) sign applies for either the NH or QD (IH) neutrino spectrum, and
= m2S /m2A 0.04. Due to the smallness of the parameter , the quantities |(YT YT )ij |
2
|[V(mD
) V ]ij | are not sensitive to the type of neutrino mass spectrum, unless s13 c12 s12 c23 /
s23 (or c12 s12 s23 /c23 ) 0.02. For such values of s13 , the e and sd ( e and bd) matrix elements in Eqs. (24) would be strongly suppressed or even vanish for a IH (NH) mass pattern.
Instead, the entries (bs) do not change significantly with 13 , since the terms proportional to
2 2
s13 are suppressed by cos(223 ) 0. The quartic combinations [V(mD
) V ] can be approximately obtained from the corresponding quadratic ones of Eq. (26) by performing the following
replacements:
2

NH: m2A m2A (1 + 2), 2 (1 2),

2
2 2
IH: mA mA ,
2 1
,
2
QD:
m2A 2m21 m2A ,
(27)
It is then straightforward to obtain the explicit expressions for the r.h.s. of Eqs. (24), taking into
account Eqs. (26) and (27). However, it is more instructive to consider the results for certain
ranges of 13 . Let us focus on m2 , as the result for m2c will follow directly. Distinguishing the
L
d
three types of neutrino mass neutrino spectra, we find:
NH spectrum, for s13 2 c23 c12 s12 /s23 8 104 :

2

|BT | 2
D 2
e 2
mL
V m V ca ga 78m2A (1 + 2) ,
2
16

2

|BT | 2
D 2
mL e, e
V m V e, e cae ga2 6m2A (1 + 14) ,
2
16
(28)
84
while for s13 c23 c12 s12 /s23 0.02:

2

|BT | 2
D 2
e 2
mL ij
V m V ij ca ga 78m2A (1 + 2)
2
16
(29)
IH spectrum for s13 c23 c12 s12 /s23 0.02:

2
|BT | 2
D 2 e 2
V m V ca ga 84m2A ,
mL
2
16

2

|BT | 2
D 2
mL e, e
V m V e e cae ga2 156m2A ,
2
16
(30)
while for s13 2c23 c12 s12 /s23 0.04:

2

|BT | 2
D 2 e 2
mL ij
V m V ij ca ga 84m2A .
16 2
(31)
QD spectrum (for any value of s13 )

2
m2A
|BT | 2
D 2
e 2
2
c
1
+
.
V
m
V
g
162m
mL ij
1
ij a a
16 2
27m21
(32)
2
The above expressions exhibit the relevant flavour violating factors [V(mD
) V ]ij separated
2
from the piece composed by terms proportional to the gauge couplings ga and terms proportional to m2A (NH and IH) or m21 (QD). For a given effective SUSY breaking scale BT ,
the overall size of those entries is controlled by (MT /)2 . It is interesting to notice that,
for a certain value of MT /, the ga2 -terms are canceled and thus all the three off-diagonal
entries can be vanishing. If s13 0.02, this peculiar flavour-blind suppression happens when
MT / v22 (
cae ga2 1/2

)
78m2A
and MT / v22 (
cae ga2 1/2

) ,
84m2A
for the NH and IH neutrino spectrum,
respectively. Instead, for the QD case, the suppression occurs for MT / v22 (
cae ga2 1/2

) ,
162m21
irre-
spective of s13 . If s13 is tiny and the neutrino spectrum is either NH or IH, the entry ( ) is
suppressed for MT / smaller (larger) by 1/3 ( 1.4), with respect to the entry (e) [( e)]
for the NH (IH) spectrum. We can summarise this analysis saying that the off-diagonal entries of
m2 and m2c can undergo a strong reduction in two cases: (1) when s13 0.02 and so the quantiL
2
D 2
ties [V(mD
) V ] e and [V(m ) V ]e are vanishing for the NH and IH spectrum, respectively;
(2) when there is a cancelation between the quadratic and quartic contributions, which depends
on the parameters MT and and on the neutrino spectrum.
In the parameter range where the FV entries (m2 )ij are dominated by either the quadratic or
L
the quartic Yukawa terms, the relative size of LFV (QFV) in the (b s) and e (s d)
sectors, does not depend of the ratio MT /, and can be approximately predicted in terms of only
85
the low-energy observables12 V and mD

, as:
R23/12 =
R13/12 =
(m2 )
L
(m2 )e
L
(m2 ) e
L
(m2 )e
L
(m2c )bs
d
(m2c )sd
d
2
(m c )bd
d
(m2c )sd
d
2
[V(mD
) V ]
,
2
[V(mD
) V ]e
2
[V(mD
) V ] e
.
2
[V(mD
) V ]e
(33)
Plugging the results of Eqs. (26) into the above expressions, and taking the central values (22)
for the neutrino parameters, we have:
R23/12 |s13 =0 40(40),
R23/12 |s13 =0.2 3.2(3.8),
R13/12 |s13 =0 1(1),
R13/12 |s13 =0.2 0.8(1.2),
(34)
for the NH (IH) spectrum. The QD case gives the same results as the NH one.
6. Phenomenological viability
In the previous sections, we have set up a theoretical framework consisting of the superpotential (13) and the SSB Lagrangian (15), beneath the energy scale MT , whose relevant soft
mass parameters (16), (17) emerge at the messenger energy scale. Next, we describe the criteria
employed to constrain the parameter space spanned by MT , BT and . With the purpose of investigating the phenomenological viability of our scenario, a detailed numerical analysis is carried
out in Section 6.2. The results are then thoroughly discussed.
6.1. The phenomenological constraints
Our approach to relate the low-energy measured parameters with the high-energy quantities,
such as the Yukawa couplings, follows a bottomup perspective. In particular, this allows us to
determine the following quantities:
The matrix YT (as well as YS and YZ ) is determined at MT as already explained in the
previous section (see Eq. (18)).
The Yukawa matrices Ye , Yu , Yd are extracted from the corresponding charged fermion
masses, modulo tan .
The remaining low-energy input parameters (s ) (with its sign) and tan are determined
by the EWSB conditions:
2 =
tan2 m
2H2 + m
2H1
tan2 1
MZ2
,
2
sin 2 =
m
2H1
2BH
.
+m
2H2 + 22
(35)
We have included the one-loop tadpole corrections t1,2 through the redefinition of the Higgs
2H2 = m2H2 t2 . According to the standard
scalar masses, i.e. m
2H1 = m2H1 t1 , m
2v cos
2v sin
practice, the above minimisation conditions are imposed at the scale s = mt1 mt2 .
12 We recall that, for low/moderate values of tan , the off-diagonal entries (24) are not substantially modified by the RG
running from the high scale MT to low-energies. Therefore, in Eq. (33) we have disregarded the renormalization effects
on the neutrino mass (see Eq. (18)). However, this does not alter the form of m , since it amounts to an overall correction
factor, except possibly for the entries (m ) i which receive extra (overall) corrections in the regime of large tan .
86
Table 1
Present limits and future sensitivities for the branching ratios (BR) of several LFV processes. The bound on the e
conversion rate (CR) in Ti is also shown
BR
Present limits
Future sensitivity
e

e
e e + e
+
e + e
e e+ e
e +
CR( e; Ti)
1.2 1011 [35]

6.8 108 [37]
1.1 107 [39]
1.0 1012 [40]
1.9 107 [42]
1.9 107 [43]
2.0 107 [44]
3.3 107 [44]
1.7 1012 [45]
1014 [36]
109 [38]
109 [38]
1014 [41]
109 [38]
109 [38]
109 [38]
109 [38]
1018 [46]
Hence, we are left with three free parameters, MT , BT and the coupling , which can all be
taken to be real, without loss of generality. The parameter space spanned by these parameters is
constrained as follows.
We impose the constraint coming from the decay e , using the results in [11]. The
experimental upper bound for the branching ratio of this decay is shown in Table 1. The
relevant LFV entry is (m2 )e (see Eqs. (24) and (26)).
L
A conservative limit upon the lightest Higgs boson mass, mh > 110 GeV, is considered on
the basis of the negative LEP2 direct searches [47]. Our predictions on mh include the lowenergy radiative corrections which are obtained linking our code to FeynHiggs [48].
We also require that the sparticle masses (which pass the tachyonic test) respect the experimental lower bounds set by Tevatron and LEP direct searches [15].
The SUSY contribution to the anomalous magnetic moment of the muon a [11] has been
considered and subjected to the following constraint at 99% C.L. [49]:
13 1010 < a < 57 1010 .
(36)
However, we will see that this requirement does not imply an important constraint on the
parameter space.
We have checked that all the coupling constants remain in the perturbative regime up to
the scale MG . In particular, while the presence of a complete GUT representation below MG
2
does not alter the value of MG , the gauge coupling gG gets an additional contribution gG
=
N
N
8 ln(MT /MG ). The requirement of perturbativity implies 2 ln(MG /MT ) 24 which, for
N = 7, gives MT 107 GeV. The parameter space is further restricted by imposing the
perturbativity requirement on the coupling constants , YT , YS and YZ .
As it will be shown in the next section, the most stringent constraints come from the sparticle
spectrum, e decay, the Higgs boson mass and the requirement of radiative EWSB and
perturbativity.
6.2. Parameter space analysis
In Fig. 5 the constraints imposed on the parameter space (, MT ) are shown for BT = 20 TeV
and s13 = 0(0.2) in the upper (lower) panel, taking the NH neutrino spectrum (later we will com-
87
Fig. 5. The (, MT )-parameter space configuration for BT = 20 TeV and s13 = 0 (s13 = 0.2) in the upper (lower) panel.
Perturbativity and EWSB exclude the light-grey and red regions (see legend in the upper panel), respectively. The orange
region is excluded by the Higgs mass bound mh < 110 GeV for mt = 170.2 GeV, while the left-most (right-most)
dash-dotted line delimit the same region for mt = 172.5(174.8) GeV. Inside the yellow and green areas, BR( e ) is
above the present experimental upper bound and the lightest slepton mass is below 100 GeV, respectively. In the dotted
and hatched areas, the neutralino 10 is the lightest MSSM sparticle and 1 is tachyonic, respectively. The thick-solid
(dashed) lines correspond to the isocontours of tan ( in GeV) for mt = 172.5 GeV. The thin solid lines refer to the
d (defined in Eq. (49)). See text for more details. (For interpretation of the references to colour in this
FV parameter bs
figure legend, the reader is referred to the web version of this article.)
ment on the other cases). The light-grey regions are excluded by the perturbativity requirement.
For each value of MT , there is a minimum value of , scaling as 2 104 (MT /1011 GeV), below which the couplings YT and/or YS,Z reach the Landau pole between MT and MG . Similarly,
there is an upper bound for , above which itself blows up.
88
Fig. 6. The same as in Fig. 5 for BT = 50 TeV.
The EWSB constraint excludes the red region (which covers the range with 1 along
the whole interval of MT ) limited on the left by the least achievable value of tan 2.5. We
have studied the sensitivity of our results to the top mass, considering the 1 allowed range
mt = (172.5 2.3) GeV [50]. The most relevant effects of varying mt are the ones induced by the
top-Yukawa term in the RG running of the soft mass m2H2 . However, this variation does not affect
our results considerably and, therefore, we will mainly take the central value mt = 172.5 GeV.
89
EW (solid
Fig. 7. Left panel: Curves of BH (dashed lines) obtained at s from the running of BH (MT ), and of BH
lines) determined by the minimization conditions (35), as a function of tan for BT = 20 TeV, MT = 109 GeV and
= 104 , 0.75 (lower plot). In the upper plot, the corresponding values of the parameter are shown. Right panel: The
same as in the left one but for MT = 1013 GeV and = 0.25, 1.0.
The EWSB conditions select values of tan (thick solid-contours) up to moderate ones,13
(tan 25). Notice that the largest values of tan are achievable only for MT 109 GeV and
small values of 0.2. In fact, Eq. (35) shows that larger values of tan (or smaller sin 2) require suppressed BH (s ), which can be obtained by shortening the running energy interval (and
so, smaller MT are needed). We also display the iso-contours of the (s ) parameter (dashedlines) as obtained by the EWSB conditions. One can realize that this parameter slightly increases
with MT due to the enhancement of the RG factor which affects m2H2 (s ) in the minimisation
condition of Eq. (35). The whole allowed parameter space covers the range (450550) GeV.
To better understand the behavior of the and tan contours, it may be useful to consider
Fig. 5 together with Fig. 7. Indeed, the latter displays the solutions of Eq. (35) for (upper plots)
and BH (lower plots), when MT = 109 (1013 ) GeV in the left (right) panel. We have chosen four
distinct points in the (, MT ) parameter space and shown, for each case, the predicted parameter
BH (s ) = BH (MT ) + BH , where BH (MT ) is determined from Eq. (16) and BH englobes
the RG running between MT and s . The curves of BH (s ) have to be compared with those
EW in the plots). The crossing of these two
of BH as extracted from Eq. (35) (denoted by BH
curves (indicated by a black dot) signals the presence of a solution at the corresponding value
of tan . For instance, the left-panel of Fig. 7 shows that for BT = 20 TeV, MT = 109 GeV and
= 104 (0.75), EWSB occurs for tan 22(3) with 450(400) GeV. A similar example is
illustrated in the right-panel of Fig. 7 for BT = 20 TeV, MT = 1013 GeV and = 0.25 and 1.
Consider now the constraint imposed by the lower bound on mh . The orange region in Fig. 5
shows the portion of the parameter space forbidden by the condition mh > 110 GeV, taking the
lower limit of the 1 range for the top-mass (mt = 170.2 GeV). The dependence on mt comes
from the low-energy radiative corrections
m4t
m2W
ln(2s /m2t ). As mt increases, the region shrinks,
being delimited by the left-most (right-most) dashed-line when mt = 172.5(174.8) GeV. It is

13 As already mentioned above, varying m in its 1 interval does not lead to appreciable changes on the values of
t
and tan . Thus, we have only displayed the tan and iso-contours for the central value of mt .
90
also clear that, for each value of MT , the upper bound on is set by the Higgs mass constraint
e.g., < 0.450.55 when MT = 109 GeV. Since the dependence of the Higgs mass on tan
mostly comes from the tree-level contribution | cos 2|MZ , the iso-contours of mh closely follow
those of tan , setting in this way the least allowed value of tan 7.
In our framework, the lightest MSSM supersymmetric particle is tipically the lightest slepton
1 , except inside the dotted region where m 0 < m1 (which is almost entirely excluded by the
1
Higgs and e constraints). The mass of 1 turns out to be below the LEP2 lower bound
of about 100 GeV in the region of the parameter space filled in green, where the large values of
tan 22 reduce m1 through the leftright mixing at the electroweak-symmetry breaking (see
also Section 7.1). In the hatched region (lower-left corner), where tan 30, the lightest slepton
is tachyonic.
All the constraints discussed so far, being related to unflavoured observables, are not sensitive to the angle 13 , as it is shown by the comparison between the upper (sin 13 = 0) and lower
(sin 13 = 0.2) panels of Fig. 5. On the contrary, the size of LFV strongly depends on it (see
Eq. (26)) and hence, the region excluded by the bound on BR( e ) changes when different
values of 13 are considered. For MT 2 108 GeV, the e constraint provides the most
restrictive lower bound on . This stems from the fact that the size the LFV entry (m2 )e scales
L
as (MT /)2 (Eq. (24)). Consequently, the allowed range is wider for lower values of MT ,
closing up for MT 4 1013 GeV. By switching on s13 (lower panel), the size of e LFV is
enhanced as

(m2 )e |s13 =0.2
m2A 2 s13
L
(37)
1+
12
(m2 )e |s13 =0
m2S s12 c12
L
and, correspondingly, the e limit implies that the lower bound on increases by a factor
of (140)1/4 3.5 and so, the allowed parameter space shrinks. Taking into account all the
constraints considered above, one concludes that values of MT 2 1013 GeV are excluded for
BT = 20 TeV, independently of the value of .
An equivalent analysis is presented in Fig. 6 for BT = 50 TeV. The comparison with the
previous case indicates that the regions excluded by the perturbativity and EWSB requirements
are not significantly affected. On the other hand, the constraint on the Higgs mass implies a
slightly different upper bound on ( 0.6) and a decrease of the minimum allowed value of
tan (tan 4). Indeed, the sparticle spectrum is now heavier and thus, the radiative corrections
ln(
2s
)
m2t
to mh are larger. For this reason, smaller values of tan are tolerated in the tree-level
contribution MZ | cos 2|. Contrarily to the previous case, 1 is never tachyonic and its mass
lies above the LEP bound. Consequently, for MT = 107 108 GeV the allowed -range is much
more extended with respect to the case with BT = 20 TeV. Moreover, larger values of tan
are now possible (tan 40). The heavier spectrum also makes the e constraint weaker,
reducing the excluded yellow region. In conclusion, the allowed parameter space enlarges when
BT increases. Also for this case, the effect of non-zero 13 (lower panel) produces similar results
as for smaller BT , by raising the lower bound of .
Before concluding this section, a comment is in order about the influence of the type of neutrino spectrum on the allowed parameter space. In the IH case, the perturbativity constraint on
the Yukawa couplings YT ,S,Z would just imply a slightly larger minimum of , for each MT . All
other constraints would instead be unaffected. For a QD spectrum, the effect from the perturbativity requirement would be much stronger (depending on the magnitude of the overall neutrino
91
mass m1 ) since, as already mentioned, all the YT ,S,Z entries increase when compared to the NH
or IH cases. Therefore, the light-grey region would mostly cover the yellow area excluded by
the BR( e ) bound. In conclusion, either for the IH or QD spectrum, the resulting allowed
parameter space would not be much different from the NH case displayed in Figs. 5 and 6.
7. Phenomenological predictions
After describing the main phenomenological constraints imposed on the (, MT ) parameters space, we will now go through the specific features of the sparticle and Higgs spectra
(Section 7.1) of our scenario, and to the implications for several low-energy LFV processes
(Section 7.2).
7.1. Sparticle and Higgs spectroscopy
The spectrum of the superpartners is determined by the finite radiative contributions to the
SSB parameters at MT (see Section 4), acting as boundary conditions, and by the subsequent
MSSM RG running from MT to s . The physical scalar masses are obtained by taking into
account the latter effect at one-loop level [51] and the low-energy D- and F -term contributions.
To get some intuition on the main features of the physical spectrum, we present some qualitative
arguments in addition to the complete numerical results.
In the present framework, the boundary conditions for the gaugino and sfermion masses
are not universal at MT , even when MT is not far from MG . This is due to the different
gauge quantum numbers of the MSSM field representations (see Eq. (17)). At lowest order, the
gaugino masses M a at the messenger scale14 are in proportion to the gauge coupling squared,
M 1 : M 2 : M 3 = 1 : 2 : 3 (a = ga2 /4 ). This relation is maintained at low-energy, like in unified SUGRA scenarios. The low-energy gaugino masses are given by:
BT
(38)
Nga2 (s ), a = 1, 2, 3,
16 2
which leads to: M2 1.86M1 0.35M3 (taking s 700 GeV). The most interesting aspect
comes from the fact that the scalar and gaugino masses are mutually related at the messenger
scale. For the sake of our discussion, let us disregard for the moment the contributions proportional to the Yukawa couplings in the expressions of the scalar masses in Eq. (17), as well as the
Yukawa effects in the (1-loop) renormalization. In such a case, the low-energy sfermion masses
have the form:
Ma (s ) =
2ka Ca 2
Ma + Ra ,
N
f
2ka Ca (1 xa )
g 4 (s )
Ra = M a2
, xa = a 4 ,
ba
g a
m2f (s ) =
(39)
where b1 = 33/5, b2 = 1, b3 = 3. The first term in m2 (s ) corresponds to the high-energy

f
boundary contribution while the second one, Ra , accounts for the RG effects induced by
the gaugino masses. The squark masses mQ receive the main RG correction from the gluino
14 In this section, we denote by overbar any quantity evaluated at the scale M .
T
92
Fig. 8. The sparticle and Higgs spectrum for BT = 20 TeV as a function of MT and correspondingly taking (as
indicated in the upper horizontal axis) along the isocontour BR( e ) = 1.2 1011 . Left panel: Physical squark
masses, mu (black solid lines), md (red dashed) and the pole gluino mass mg (blue dash-dotted). Right panel: Charged
slepton masses m (solid black lines), sneutrino masses m (dotted blue), neutralino and chargino masses m 0 and m
(dashed blue). The dash-dotted green lines correspond to the Higgs boson masses mh , mH , mA and mH . In the small
inner plot (on the left panel) we show the behavior of tan (red line) and as obtained by the EWSB conditions. (For
interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
mass term, which amounts to a positive shift on m2 . The term R3 is larger than the boundQ
ary condition value (since |1 x3 | > |ba |/N = 3/7), by a factor of approximately 8(2) for
MT = 1014 (107 ) GeV. Notice that, in the minimal GMSB model (N = 1), the dominance of
R3 holds only for a messenger scale above 1012 GeV [52]. Therefore, we expect the lowenergy ratio mQ /M3 to be given as

1/2

q
2C3 1 x3
b3
,
1+
M3
b3
x3
N (1 x3 )
mQ
(40)
which lies within 0.90.8 for MT 1014 107 GeV. Hence, the gluino is the heaviest of the
coloured sparticles.
In Fig. 8 we display the sparticle and Higgs spectra for BT = 20 TeV. The values of (shown
in the upper horizontal axis) and MT , are the ones along the border which delimits on the right the
e excluded region15 (see upper panel of Fig. 5). The gluino pole mass includes the lowenergy finite corrections (see e.g. [53]), which amount to 2030% of the tree-level value M3 .
As anticipated the gluino is the heaviest superparticle. The first and second generation squarks
d (dashed curves) and u (solid) (mainly composed by left-handed squarks) are the heaviest.
Instead, the lightest squark t1 is mainly a right-handed tc whose mass is pushed down by a
negative shift, driven mainly by the top Yukawa-induced renormalization. This effect, together
with the leftright squark mixing, is enhanced for MT 1013 GeV where tan becomes smaller
(see the inner panel where we have drawn the behavior of tan and versus MT ). This analysis
has shown that for BT = 20 TeV in the allowed (, MT ) portion, where the e is close to
the present bound, the squark spectrum lies in the range 800950 GeV. This mass range will be
15 The sparticle and Higgs boson spectra do not depend much on and so, the results shown in Fig. 8 are quite representative of the whole parameter space at BT = 20 TeV.
93
soon explored by the large hadron collider (LHC) [54] with a luminosity L = 100 pb1 and a
center-of-mass energy s = 14 TeV.

The left-handed slepton masses can instead be compared with the SU(2)W gaugino mass M2 .
In this case the renormalization factor in Eq. (39) is comparable to the boundary contribution. At
low-energy, the ratio mL /M2 lies in the range 10.5 for MT = 1014 107 GeV. In the right panel
of Fig. 8 we show the physical spectrum of the electroweak states. The heaviest charged sleptons
4,5,6 (which are mainly left-handed sleptons) have masses around 250320 GeV in the allowed
range MT 108 1013 GeV. The sneutrino masses are splitted from the latter states mainly by
the SU(2)W D-term, m2 m2 = m2W | cos 2|.

The right-handed slepton masses receive contributions only from the U (1)Y gauge interactions and, therefore, are smaller than the other sfermion masses. In the range MT =
1014 107 GeV, the high energy contribution is larger than the renormalization term R1 by
a factor of 13. The low-energy ratio mec /M1 is approximately in the range 10.7 for
MT = 1014 107 GeV. To these estimates one has to add the negative shift from the Y -induced
renormalization and the effect from the leftright slepton mixing at the electroweak-symmetry
breaking. Both these contributions are important for large tan and lower the physical 1 mass
below m c . In Fig. 8 (right panel) we can see that for MT 108 GeV the 1 mass is pushed below
100 GeV and, in the range MT 108 1013 GeV, 1 is indeed the lightest MSSM sparticle. For
larger values of MT , corresponding to small tan , the neutralino 10 becomes the lightest MSSM
sparticle. Hence, either 1 or 10 would decay into the gravitino, which is in fact the lightest SUSY
particle in our framework. In conclusion, the slepton masses lie in the range 100320 GeV (for
BT = 20 TeV) and, therefore, are within the discovery potential of the LHC.
Concerning the physical charginos and neutralinos, the inner plot (right panel) shows that the
2 > M 2 hold for most of the parameter
parameter comes out to be larger than MZ and 2 M1,2
Z
space. These hierarchies imply that the lightest neutralino 10 is mainly a B-ino and has mass
m 0 M1 , while 20 and the lightest chargino 1 are almost degenerate and are mainly W 1
inos with mass M2 [30]. Therefore, both these masses do not exhibit a significant dependence
0 are mostly
on MT , as seen in Fig. 8 (right panel). The heaviest chargino 2 and neutralinos 3,4
higgsinos with mass set by the parameter, increasing therefore with MT .
Notice that, increasing the value of BT , all the sparticles become linearly heavier since they
scale as BT .
Finally, also the Higgs boson masses can be predicted in our scenario. As already mentioned in
Section 6.1, the mass of the lightest CP-even Higgs boson (mh ) has been computed by including
the low-energy radiative corrections. In the allowed parameter space of Fig. 5, mh turns out to
be in the range 110120 GeV, thus being testable in the near future at the LHC (mainly through
the Higgs decay into 2 photons [55]). The heaviest CP-even (H ), CP-odd (A) and charged Higgs
bosons are much heavier. At tree-level, mA = BH / sin 2 and for mA MZ all these states
are almost degenerate, mH mH mA . For BT = 20 TeV, Fig. 8 shows that mA , mH , mH
400470 GeV. For larger values of BT the masses of such non-standard Higgs bosons increase,
while mh increases by a few GeV due to the logarithmic sensitivity to BT . The Higgs sector is
therefore characterized by a decoupling regime with a light SM-like Higgs boson (h) and the
three heavy states (H, A, H ).
We conclude this section with a comment on the SUSY contribution to the muon anomalous
magnetic moment a . In Fig. 9 we show the a behavior as a function of MT in correspondence with different values of BR( e ) and for BT = 20 TeV. It is well known that a is
94
Fig. 9. Left panel: SUSY muon anomalous magnetic moment contribution a as a function of MT (taken with its
corresponding along the lines of constant BR( e ) = 1.2 1011 , 1014 ) for s13 = 0, 0.2 and BT = 20 TeV.
Right panel: curves of (lower plot) and tan (upper plot), for the same inputs considered in the left-panel.
induced by the dipole operator whose dominant contributions go as
g 2 m2 tan
16 2 m3soft
. Since the spar-
ticle spectrum does not change significantly with MT (Fig. 8), a directly reflects the demeanor
of tan (cf. left and right panels). The sign of a is the same as the one of and, in the allowed
range MT 2 108 GeV, a 4 109 respects the constraint (36). As a scales as 1/BT2 ,
it decreases for larger BT .
7.2. LFV: Model independent predictions
We have already described the structure and parameter dependence of the flavour violating
SSB mass parameters in Section 5. Here, we intend to further investigate the phenomenological implications by considering several LFV processes (besides e ) such as, eee,
, e , , eee, ee e and e conversion in 22
48 Ti.
The present experimental upper bounds and future sensitivities for the BRs of these decays are
collected in Table 1.
Let us briefly recall some points related to the computation of such processes. The radiative
decays j i + are induced by the effective dipole operator:

emj iDjLi i cj + iDjRi ci j + h.c. F ,
(41)
where e and F are the electric charge and the electromagnetic field strength, respectively. The
corresponding branching ratios are:
BR(j i + ) =
48 3
L 2 R 2
Dj i + Dj i BR(j i j i ),
G2F
(42)
where = e2 /(4), GF is the Fermi constant, BR( e e ) 1, BR( ) 17%

and BR( e e ) 18%. Since LFV resides in the parameters m2 and Ae , the coefficient D L
L
is the dominant one. In the mass-insertion approximation its parametric dependence is:
(m2 )

(m2 )j i
(Ae )j i
1
g2
L
L j i
L
Dj i
(43)
or
tan or
,
(Ye )jj msoft
16 2 m2soft m2soft
m2soft
95
where the coupling g can be either g2 or g = 3/5g1 . In the MSSM framework, the dipole coefficient D L receives contributions from three different diagram topologies which, under certain
features of the sparticle spectrum, could cancel against each other. In the latter case, these decays
would be strongly suppressed even if the sparticle spectrum were not too heavy [56]. This can be
realized if, for instance, the A-terms Ae are large, or the gaugino mass parameters M1 and M2
have opposite sign. In our scenario such peculiar situations do not occur, so the suppression of
the dipole operators can only arise from large mass parameters (i.e. by increasing BT ) or from
cancellations inherent to the m2 off-diagonal entries, as discussed in Section 5.
L
The strict predictions regarding the ratios (33) and (34) can be translated into predictions for
the ratios of the BRs [9] once Eqs. (42) and (43) are taken into account:
(m2 ) 2

300 [s13 = 0],
BR( )
BR( )
L
2
BR( e )
BR(
)
2(3)
[s13 = 0.2],
(m )e
e
L
(m2 ) 2

0.2 [s13 = 0],
BR( e e )
BR( e )
L e
(44)
BR( e )
0.1(0.3) [s13 = 0.2],
(m2 )e BR( e e )
L
where the results on the r.h.s. hold for all the three types of neutrino mass spectrum and the
numbers in parenthesis apply for IH case, whenever this is different from the others.
Regarding the 3-body decays j i k k and e conversion in Ti, these receive contributions from - (dipole D L,R and monopole C L,R operators), Z- (monopole AL,R operator) and
Higgs- (monopole L,R operator) exchange diagrams, as well as from box (B LL,RR , B LR,RL )
diagrams. The branching ratios are given by:

2 LR 2 RL 2

1
+
+ F + F + ak F RR 2
BR
=

ak FjLL
i
ji
ji
ji
j
i k k
2
8GF

LR + DjRi FjRL
RR
+ 4e2 Re DjLi ak FjLL
i + Fj i
i + ak F j i

m2j

R 2

4 L 2

+ 8e Dj i + Dj i
log 2 bk
BR
j i i j , (45)
mk
where ak = 1(2), bk = 3(11/4) for k
= i (k = i), and the coefficients F N P (N, P = L, R) are
combinations of AN , B N P , C N and N (see e.g. [56]). We have found some discrepancies
between our analytical computations concerning the coefficients AL,R with those published by
some authors [11]. In Appendix B we have reported and discussed our results for AL,R . Moreover, in Appendix C we present the results for the box coefficients B MN relevant for the decays
ee and e, calculated to all order in the electroweak-breaking effects. For the sake
of brevity, we refer to the existing literature (e.g. [11] and [56]) for the explicit formulas of all
the other coefficients, including the one relevant for the CR( e; Ti). Like for the dipole D L ,
the presence of LFV in the left-handed sector implies that only the coefficients AL , B LL , B LR ,
C L , L are important. We just recall their parametric dependence:
AL
ji,
BjLL(LR)
i
g2
, CjLi
2
g 2 (mL )j i
,
16 2 m4soft
L
ji
2
g 2 (mL )j i
.
16 2 m2soft
(46)
From what has been commented above about suppressing D L in our scenario, and by comparing Eq. (43) with (46), one can realize that, whenever the coefficients D L get suppressed, AL ,
2
B LL,LR and C L undergo the same fate. Regarding L , one has that L
j i 0 only if (m )j i 0
L
96
since these coefficients are insensitive to an overall mass scale increasing. Consequently, the D L
contributions to the 3-body BRs [CR( e)] are dominant with respect to those from the remaining operators, due to the tan and phase-space logarithmic-factor [tan ] enhancement.
Although the Higgs-exchange diagram contribution also benefits from a tan enhancement,
its numerical relevance requires the Higgs bosons A and H to be significantly lighter than
the sleptons, charginos and neutralinos [5659]. As already discussed in Section 7.1, in our
scenario this does not occur, thus the Higgs-mediated contributions come out to be subleading.
In the aforementioned dipole-dominance situation, the LFV processes under consideration
can be directly compared with the radiative decays:

m2 11
BR( eee)
ln 2
6 103 ,
BR( e )
3
4
me

m2 11
BR( eee)
ln 2
102 ,
BR( e )
3
4
me

m2
BR( e)
ln 2 3 2 103 ,
BR( e )
3
m

11
m2
BR( )
ln 2
2.2 103 ,
BR( )
3
4
m

m2
BR( ee)
ln 2 3 102 ,
BR( )
3
me

2
CR( e; Ti)
4
Z F (q) Ti 5 103 ,
16 4 Zeff
BR( e )
capt
(47)
Ti 2.6 106 s1 is the muon capture width in

where Z = 22, Zeff = 17.6, |F (q)|2 0.54, capt
5
1
Ti [11] and 4.5 10 s is the total muon decay width. In addition to these correlations,
we also find that the 3-body decays can be related to each other by using the ratios (34) as
follows:
2
(m2 ) 2 ln m2 11
4
m
BR( )
BR( )
2
2
m
BR( eee)
BR( e e )
(m )e
ln m2 11
L
4
e

100 [s13 = 0],
0.6(0.9) [s13 = 0.2],
(m2 ) 2 ln m2 3

550 [s13 = 0],
BR( )
BR( ee)
m2e
L
2
2
m
BR( eee)
3(4) [s13 = 0.2],
(m )e
ln 11 BR( e e )
L
m2e
(m2 ) 2 ln m2 11

0.3 [s13 = 0],
4
BR( eee)
BR( e e )
m2e
L e
2
2
m
BR( eee)
BR(
)
0.2(0.4) [s13 = 0.2],
(m )e
e
ln m2 11
L
4
e
97
Table 2
Expectations for the various LFV processes from Eqs. (44), (47) and (48), assuming BR( e ) = 1.2 1011 . The
results in parenthesis apply to the case of the IH neutrino spectrum, whenever these are different from those obtained for
NH and QD
Expectation
s13 = 0
s13 = 0.2
BR( )
BR( e )
3 109
2(3) 1011
1(3) 1012
6 1014
4(6) 1014
2(3) 1013
1(3) 1014
2(4) 1015
6 1014
2 1012
6 1014
7 1012
3 1011
2 1014
3 1015
6 1014
BR( e e+ e )
BR( + )
BR( e+ e )
BR( e e+ e )
BR( e + )
CR( e; Ti)
2
(m2 ) 2 ln m2 3
m
BR( e)
BR( e e )
L e
m2
BR( eee)
11 BR( e e )
(m2 )e
ln m2 4
L
e

0.06 [s13 = 0],
0.05(0.07) [s13 = 0.2],
(48)
where the parenthesis enclose the results for the IH spectrum (when different from the other
cases). In Table 2 we present a synoptic view of the correlation pattern predicted in our context,
assuming that the present bound on e (Table 1) is saturated, choosing s13 = 0(0.2) and
setting all the remaining neutrino parameters at their best fit points (22).
7.3. LFV processes: Numerical analysis
We are now in position to analyse our numerical results regarding the predictions on the
LFV decay branching ratios. At this point, we are interested in knowing how and to what extent
the LFV processes can probe the parameter space of our framework. In other words, will the
upcoming experimental sensitivities be enough to test the allowed parameter space of Figs. 5
and 6? Instead of plotting the contours relative to the various BRs in the allowed (, MT ) space,
we have fixed BT = 20 TeV, selected some values of MT and performed the analysis along one
direction, which can be either s13 (and the phase ) or . This is representative enough and allows
us to properly display the main features.
Consider the effective size of LFV and QFV, which can be parameterized by the following
dimensionless parameters [8,19]:
ijL
|(m2 )ij |
L
m2
L
(i, j = e, , ),
ijd
|(m2c )ij |
d
m2c
(i, j = d, s, b),
(49)
where m2 and m2c are the average L and dc masses, respectively. The parameters L,d are indeL
d
pendent of BT , while their overall size is determined by the ratio MT /. In Fig. 10 we have plotted ijL and ijd as a function of s13 for two selected points of the parameter space shown in Fig. 5:
(, MT ) = (4.8 105 , 109 GeV) (Fig. 10, upper panels) and (, MT ) = (2.4 105 , 109 GeV)
(Fig. 10, lower panel). The first point lies in the allowed portion of the parameter space studied in
Fig. 5 for s13 = 0 (very close to the region delimited by the present e bound), whereas it is
excluded for s13 = 0.2. The second point falls into the region excluded by the e bound, for
98
L (i, j = e, , ), d (i, j = d, s, b) and BR( ) as a function of s for the NH (left

Fig. 10. FV parameters ij
j
i
13
ij
upper-panel) and IH (right upper-panel and lower panel). The parameters MT , BT and are fixed as indicated on the
top of each panel. For each radiative decay, the present upper bound (horizontal solid-lines) and future experimental
sensitivity (horizontal dashed-lines) for the BRs are shown.
the NH (and QD) spectrum, irrespective of s13 (but it is allowed in the IH case for some values
of s13 ).
2
The behavior of L,d just reflects that of the relevant FV structure V(mD
) V in Eq. (26). For
2
example, notice that they scale as 1/ outside the cancelation dip. This figure clearly shows
d are insensitive either to s or to the type of neutrino spectrum (cf. the upper panthat L , bs
13
L ( d ) gets exchanged with L ( d ) when passing from the NH to the IH case.
els), while e
e bd
sd
This latter feature is due to the fact that 23 = /4 and so the flavours and (or b and s)
L , d / d and L / L , d / d just reproduce
are indistinguishable. The relative ratios L /e
e e bd sd
bs sd
the absolute values of R23/12 and R13/12 in Eq. (34), respectively. As we have already deduced
d and L , d are
from (26), this constant-ratio rule is violated for s13 0.02, where Le , bd
e
sd
strongly suppressed for the NH and IH spectrum, respectively. All the above peculiarities are
common to the related curves of BR(j i ), plotted on the right of each panel. In particular, the ratios (44) are remarkably reproduced, except for s13 0.02 where BR( e ) and
BR( e ) undergo a sharp suppression for the NH and IH spectrum, respectively. Hence, the
point with = 4.8 105 and s13 0.02 is excluded by the BR( e ) bound in the NH case,
whereas it is allowed in the IH with BR( ) 3 109 , BR( e ) O(1011 ). The
point with = 2.4 105 (lower panel) is phenomenologically viable if the neutrinos have IH
99
Fig. 11. Contours of the ratios BR( )/ BR( e ) (dashed lines) and BR( e )/ BR( e ) (solid lines)
in the (s13 , )-plane, for the NH (left panel) and IH (right panel) neutrino mass spectra.
masses and s13 0.02. In such a case, BR( e ) 1011 , BR( ) 5 108 and
BR( e ) O(1010 ). This is an example where BR( ) is close to the present bound
and e might be unobservable.
In the examples considered above, the size of the d s in each family sector is smaller by a
factor of 3.5 than the one of the corresponding L s. This results from different compensating
effects. Namely, the squark masses are about three times larger than the slepton ones (due to
the gluino-induced renormalization effect; see also Fig. 8), while the (m2c )ij are larger than
d
the (m2 )ij at the messenger scale MT (because of the major strong coupling contribution). In
L
d 1.5 103 (left panel). This is the maximal value attainable by d in

the first example, bs
bs
d ,
the allowed parameter space of Fig. 5. To perceive the phenomenological relevance of such bs
we need to confront it with the gluino and squarks masses. Recall that for BT = 20 TeV we got
d 103 is well below the experimental
mQ 900 GeV and M3 1.3 TeV (see Fig. 8). Then, bs
d 5 104 (obtained
bound posed by the measured BR(b s ) [60]. The predicted value sd
for s13 = 0.2) also lies below the bound inferred from K 0 K 0 mixing [60].
Fig. 11 is aimed to address the dependence of the ratios BR( )/ BR( e )
and BR( e )/ BR( e ) on s13 and the CP-violating phase for a NH (left panel)
and IH spectrum (right panel). The results do not depend on the ratio MT / as far as either the quadratic or the quartic terms dominate in the LFV entries m2 . Regarding the ratio
L
BR( )/ BR( e ), it slightly decreases (increases) with increasing s13 for < /2
( > /2). However, this ratio blows up at s13 0.02 and (0) for the NH (IH) spectrum. The BR( e )/ BR( e ) is also sensitive to both s13 and . In fact, for a given
value of s13 , the increase of this ratio with is quite modest in most of the s13 range. Still, for
s13 0.02, BR( e )/ BR( e ) goes to zero (infinity) when = 0() for the NH case.
Instead, the opposite occurs when the IH pattern is considered since there is an interchange of
the roles played by BR( e ) and BR( e ) (as already discussed). In short, barring the
range around s13 0.02, the predictions (44), (47) and (48) are not substantially altered when
the effects of
= 0 are included.
Finally, we come to a comparative analysis of radiative decays, 3-body decays and e
conversion in Ti considering the three types of neutrino spectrum. Fig. 12 shows the dependence
100
Fig. 12. Left panels: Branching ratios of the radiative decays j i and e conversion in Ti, as a function
of . Right panels: Branching ratios of the three-body decays j 3i and ee, e. All the results are shown
as a function of for MT = 109 GeV and BT = 20 TeV taking s13 = 0 (upper plots) and s13 = 0.2 (lower plots). The
left (right) vertical line indicates the lower bound on imposed by requiring perturbativity of the Yukawa couplings
YT ,S,Z when m1 = 0(0.3) eV (NH (QD) neutrino mass spectrum). The regions in green (grey) are excluded by the
m > 100 GeV constraint (perturbativity requirement when m1 = 0 eV). (For interpretation of the references to colour
1
in this figure legend, the reader is referred to the web version of this article.)
of the BRs on for MT = 109 GeV and s13 = 0(0.2) in the upper (lower) panels, assuming a NH
spectrum (m1 = 0 eV).
For s13 = 0 there is a cancelation dip only in BR( ) (left panel), BR( ee) and
BR( 3) (right panel) at 3.5 106 , which lies in the (green) region excluded by the
negative sparticle search, 6 106 (cf. also Fig. 5: upper panel). This dip originates from the
cancellations of the quadratic and quartic terms in m2 discussed in Section 5. In BR( e )
L
this would take place at a value of smaller by a factor of (6/78)1/2 0.3 (see Eq. (28)) and
so, would fall into the (grey) non-perturbative range. For s13 = 0.2 the dips are instead present
for all the BRs, as we expect on the basis of Eq. (29). Outside the cancelation regions, the relative
ratios of BR(j i ) are those announced in Eq. (44) and the BR(j i k k ) are correlated
with BR(j i ) according to Eq. (47). For comparison, we have also plotted all these BRs
for the case of the QD spectrum with m1 = 0.3 eV which can be obtained by continuously
rising the mass m1 in the NH case beyond (m2A )1/2 (cf. Eq. (19)). In the QD case the non-
101
perturbative range extends much above the one relative to the NH (grey) so that the perturbativity
lower bound on (indicated by the vertical solid line) is larger by about one order of magnitude,
3105 . Notice that a more restrictive lower bound is imposed by sparticle searches (green),
5 105 . The cancellations occur at approximately the same (inside the excluded regions)
for all the BRs. For 5 105 , the curves corresponding to the NH and QD are superimposed
and so the two scenarios are not distinguishable (cf. Eqs. (25) and (26)).
Finally, Fig. 12 reveals that, for MT = 109 GeV and assuming a NH spectrum with s13 0,
only BR( e ), BR( ), BR( 3e) and CR( e; Ti) are within the future experimental sensitivities for 5 105 7 105 , while if > 7 105 only e and
CR( e; Ti) could be accessible. All the other LFV processes would be undetectable in the
allowed range. Then, if for example the MEG experiment [36] detects BR( e ) at the
level of 8 1012 then BR( ), BR( 3e) and CR( e; Ti) are expected to be
2 109 , 5 1014 and 4 1014 , respectively. The case of the QD spectrum is similar:
In the range (67) 105 the e LFV decays are observable but BR( ) is predicted
to be 109 , and for larger only e and e conversion would be visible.
By switching on s13 , LFV is enhanced in the e and e sectors, but it is not essentially
altered in the sector. As a consequence, the lower bound on imposed by the present limit
on BR( e ) is more restrictive (cf. Fig. 5) and consequently, the BRs of the sector
are penalised. For instance, detecting e with a BR around 8 1012 would imply the
possibility to measure also BR( 3e) and CR( e; Ti) at the level of 5 1014 and 4
1014 , but BR( ) would be 1011 (well below the planned future capability). This
conclusion holds for both the NH and QD spectrum.
Fig. 13 presents a similar analysis for the IH spectrum with m3 = 0 eV. For comparison,
we also show the QD case which can be obtained by pushing the mass m3 to values larger
than (m2A )1/2 thus recovering a case very similar to the QD one reported in Fig. 12. When
s13 = 0 (upper panels) the main difference with the NH case is the fact that all BRs exhibit the
suppression dip in the perturbative range (in the sector this takes place for a smaller with
respect to the e one). However, due to the choice of BT these dips fall into the (green) range
excluded by sparticle searches. Outside these cancelation regions the values of all these BRs (and
the related correlations) are essentially the same as those obtained for the NH case leading to the
same phenomenological implications.
The IH scenario offers the possibility to reverse the dominance of the e LFV over the
LFV. Suppose that the effective SUSY breaking is increased up to BT 70 TeV so that all the
sparticle masses are enhanced by a factor of 3.5. Then the BR( e ) dip would lie in the
allowed range16 and BR( ) would be brought close to the present bound. As a result,
e , 3e and e conversion might be invisible, whereas BR( ) 5 108
could be measured (perhaps together with BR( ee) 5 1010 ).
By switching on s13 = 0.2 (lower panels) we find, as for the NH case, that all the LFV
processes undergo a dramatic suppression at the same which, however, occurs in the excluded
range. Outside this interval, the relative ratios of the BRs closely follow the predictions17 (44),
16 The effect of increasing B does not modify the perturbativity bounds, but for B 50 TeV the constraint from the
T
T
negative sparticle searches disappears, as the comparison between Figs. 5 and 6 has already demonstrated.
17 The fact that in Fig. 13 BR( 3) is slightly larger than BR( 3e), in contradiction with the predictions
reported for example in Table 2, is due the Y -driven RG running of the neutrino mass m which concerns mainly the
entries i (i = e, , ), and is more sizeable for s13
= 0 with IH spectrum.
102
Fig. 13. The same as in Fig. 12 for the case of an IH neutrino mass spectrum (m3 = 0 eV). For comparison, the results
with m3 = 0.3 eV are also shown.
(47) and (48). In such a case, the planned experimental sensitivities would allow to test only
e , 3e and e conversion.
Figs. 14 and 15 contain the results of a similar analysis performed by taking MT = 1012 GeV
for the NH and IH spectrum, respectively. The increasing of MT implies an overall decreasing
of all the BRs due to a heavier sparticle spectrum (cf. Fig. 8) with respect to the case with
MT = 109 GeV. We again find that for any (barring the cancelation ranges) the several BRs are
correlated according to the model-independent predictions contained in Eqs. (44), (47) and (48).
The features of the cancelation dips are those discussed in Section 5 and already encountered for
smaller MT . Consider the NH case (Fig. 14). At s13 = 0 and within (34) 102 , e ,
, 3e and e conversion are in the reach of the future experiments in both the
NH and QD cases. From 4 102 up to 1.5 101 only e and e conversion
could be detected. For s13 = 0.2 and 0.1 only e , 3e and e conversion
are accessible. We arrive at similar conclusions observing the case of IH spectrum displayed in
Fig. 15: The e sector is more favoured with respect to the sector, where only the
decay could be detected as far as s13 0. However, this conclusion could be contradicted for
larger BT and s13 = 0. Consider BT 50 TeV so that the sparticle spectrum increases by a factor
of 2.5. Then, in the suppression dips of the e LVF processes (see upper panels in Fig. 15),
103
Fig. 14. As in Fig. 12 but for MT = 1012 GeV.
BR( ) would be close to the present bound and BR( ee) 109 . Therefore, the
future detection of both and ee, together with a slepton and squark spectrum in
the range 400650 GeV and 1.82.5 TeV, respectively, would point to a IH neutrino masses and
s13 0.
The future perspectives to detect signals of new physics mostly rely on the observation of sparticles at the LHC or LFV decays at e.g. the B-factories [38], the incoming MEG experiment [36],
the super-flavour factory [61] or the PRISM/PRIME experiment at J-PARC [46]. It is, therefore,
extremely important to motivate and suggest theoretical scenarios which can be tested in more
than one direction. In this paper we have presented and discussed in detail a supersymmetric
SU(5) version of the triplet seesaw mechanism in which the triplet are messengers of both L and
SUSY breaking. The key-points of our model can be outlined as follows:
The tree-level exchange of the triplets T generates neutrino masses, so flavour violation
is induced also in the lepton sector of the SM, as required by the observation of neutrino
oscillations. All the LFV effects are parameterized by a single flavour structure YT ;
104
Fig. 15. As in Fig. 13 but for MT = 1012 GeV.
The quantum-level exchange of the 15-states T , S and Z generates all the SSB mass parameters of the MSSM via gauge and Yukawa interactions. Their mass scale is determined
by the effective SUSY breaking scale BT . Flavour violation is induced in the mass matrices
m2 and m2c , and in the trilinear terms Ae and Ad , by the Yukawa couplings YT (and, acL
d
cording to the SU(5) relation (12), also by YS and YZ ). Therefore, the flavour structure of
the SSB parameters is provided by YT or, according to Eq. (18), by the low-energy neutrino
parameters.
The number of free parameters is three: The triplet mass MT , the effective SUSY breaking
scale BT and the unflavoured coupling constant .
These aspects make the present scenario highly predictive since it relates neutrino masses and
mixing, sparticle and Higgs spectra, lepton and quark flavour violation in the sfermion masses
and electroweak symmetry breaking. We have performed a complete analysis of the parameter
space spanned by MT , BT and taking into account the present experimental constraints on the
above physical observables (see Figs. 5 and 6). This has demonstrated that there is a region in
the parameter space where our framework is compatible with experiment. In particular, despite
the very constrained structure of our SSB mass parameters, EWSB can be radiatively realized
105
analogously to the conventional MSSM case. Our predictions allow us to further test the allowed
parameter space as follows:
Regarding the MSSM sparticle spectrum, we predict that the gluino is the heaviest sparticle
while, in most of the parameter space, 1 is the lightest. However, the gravitino is lighter
than the MSSM sparticles. For instance, if BT = 20 TeV, the squark and slepton masses
lie in the ranges 800950 GeV and 100300 GeV, respectively. The chargino masses are
m 320 GeV and m 450550 GeV. Moreover, m 0 190 GeV, m 0 m and
1
2
1
1
2
m 0 m . These mass ranges are within the discovery reach of the LHC. Increasing the
2
3,4
parameter BT implies a linearly heavier spectrum. The measurement of only a few sparticle
masses will provide a hint on the value of the effective SUSY breaking scale BT and a test
of the correlation pattern shown in Fig. 8.
The mass range of the electroweak sparticles implies that the SUSY contribution to the muon
anomalous magnetic moment a never exceeds the maximum value of the interval (36).
Moreover, since a > 0, the discrepancy between the experimental determination and SM
prediction is alleviated in our model.
The Higgs sector is characterized by a decoupling regime with a light SM-like Higgs boson (h) and the three heavy states (H, A and H ) with mass mH,A,H 450550 GeV
(again, for BT = 20 TeV). These masses increase almost linearly with BT .
We have considered several LFV processes: eX, e conversion in nuclei, eY
and Y (X = , ee, Y = , ee, ). Our framework is characterized by peculiar LFV
and QFV patterns (intimately related to each other; see Eq. (33)), which are mostly determined by low-energy neutrino masses and mixing. The size of QFV, when confronted with
the coloured sparticle spectrum, is well below the present phenomenological bounds extracted from b s and K 0 K 0 mixing, etc. Therefore, QFV processes do not constrain
our parameter space.
Concerning LFV, we stress that strict predictions have been obtained for the relative branching
ratios of the radiative j i (see Eq. (44)) and 3-body j i k k decays (see Eq. (48)). The
latter, as well as e conversion in Ti, are also correlated with the radiative decays as shown
in Eq. (47). All these results are model-independent in the sense that they do not depend on MT ,
BT or (they are given only in terms of the low-energy neutrino observables). If the present
bound on BR( e ) is saturated, the branching ratios of the remaining LFV processes are
predicted as shown in Table 2, where the three types of neutrino spectrum have been considered.
The experimental signatures of our scenario crucially depend on the value of the lepton mixing
angle 13 :
Tiny 13 : The analysis has shown that, in the allowed parameter space, the future experimental sensitivity will allow to measure at most BR( e ), BR( 3e), BR( )
and CR( e Ti) according to the relations (44) and (47). In particular, being BR(
)/ BR( e ) 300, BR( ) is expected not to exceed 3 109 , irrespective of
the type of neutrino spectrum. All the decays i k k would have BR < O(1011 ).
Sizeable 13 : If s13 is close to the upper bound (23), the sector is hardly accessible and only
the decays e , 3e and e conversion in Ti can be observed in the future. This
conclusion holds for the NH, IH and QD neutrino spectra. For instance, if the MEG experiment
measures BR( e ) 8 1012 , then BR( 3e) and CR( e Ti) are expected to be
5 1014 and 4 1014 . The latter is in the reach of the PRISM/PRIME sensitivity. Values
106
of s13 O(101 ) will be explored soon by several neutrino experiments like MINOS, OPERA
and Double Chooz [62].
Deviations to this very specific model-independent pattern occur when (m2 )ij cancel (see
L
discussion in Section 5). This can be the case if s13 0.02(0.02) and the neutrino spectrum is
IH (NH). Then, all the e LFV processes might be invisible and detected with a BR
close to the present bound (taking BT = 20 TeV). Moreover, ee would be O(109 1010 )
and the remaining decays below 1010 . Notice that, increasing BT the BRs are suppressed since
they scale as 1/BT4 . This specific value of s13 is in the sensitivity range of neutrino factories [63].
Alternatively, (m2 )e could be vanishing because of a cancelation between the quadratic and
L
quartic Yukawa contributions (see Eq. (30)). For the IH case, if BT 50 TeV and s13 is very
small, all the e LFV processes would be strongly suppressed whereas the sector would be
favoured with BR( ) 5 108 and BR( ee) 109 . This scenario is correlated
with a slepton and squark spectrum in the ranges 400650 GeV (in the limit of the LHC detection
capabilities) and above 2 TeV (within the LHC reach), respectively.
Given the increasing interest on the problem of finding a possible relation between leptogenesis [64] and low-energy neutrino physics [65], we would like to comment on this issue in the
framework of our work. Within the present version of the supersymmetric triplet seesaw mechanism, leptogenesis can be realized by considering that the soft bilinear term BT produces a mass
splitting between T and T , leading to resonant leptogenesis [66]. For this to work, BT must be
around the electroweak scale. In our case, since BT 20 TeV, the BAU turns out to be too small.
Still, leptogenesis could be made effective either by adding an additional pair of triplets [14] or
by including heavy singlet neutrinos [67]. The former case would imply the appearance of one
more flavour source YT which, depending on its size, could have some impact in the predictions
of our framework. Instead, we would like to comment on the second possibility which requires
heavy neutrino singlets (with mass MN ) coupled to the lepton doublets through the Yukawa
couplings YN . To maintain the predictions made along this work we must require that the singlet contribution to neutrino masses is much smaller then the one generated by the triplet i.e.,
Y2N YT MN /MT . Moreover, YN YT is also required to suppress LFV arising in the SSB
parameters from the singlet exchange at the quantum level18 (notice that the neutrino singlets
could couple to the spurion field X). For the purpose of leptogenesis, two different situations
can be envisaged: MN > MT or MN < MT . In the former case, the CP-asymmetry generated
through the decay of the triplets into two leptons is directly proportional to Y2N /MN [67], so it
turns out to be very tiny (Y2N YT MN /MT ). On the contrary, if MN < MT the CP-asymmetry
is weakly sensitive to YN [67], therefore a viable value for the BAU can be achieved.
We conclude our discussion by remarking that our scenario is not only extremely predictive
but it can also be tested in view of the present and future programmes of LFV and neutrino
oscillation experiments.
Acknowledgements
We thank A. Brignole for valuable comments and suggestions and T. Hambye for useful discussions. The work of F.R.J. is supported by Fundao para a Cincia e a Tecnologia (FCT,
Portugal) under the grant SFRH/BPD/14473/2003, INFN and PRIN Fisica Astroparticellare
(MIUR). The work of A.R. is partially supported by the project EU MRTN-CT-2004-503369.
18 For a discussion related to this in the context of EWSB see [68].
107
Appendix A. Extracting the SSB terms from wave function renormalization

In this appendix we derive the general expressions for the soft supersymmetry breaking scalar
masses, the bilinear and trilinear couplings at the messenger scale M. We employ a generalization
of the method suggested in Ref. [30] and subsequently presented in Ref. [31]. Consider the case
in which the scales of SUSY breaking and its mediation to the observable sector are determined
by the vev of the auxiliary and scalar components of a chiral singlet superfield X, X = M + 2 F
(in the following M is taken to be real and |F | < M 2 , for consistency of the method). The leading
contributions (at lowest order in F ) to the SSB terms arise from X-dependent wave-function
renormalizations of the chiral superfields ZQ . The effective Lagrangian reads19

4
a b
2
a b
a b c
d ab Q Q + fabc Q Q Q + h.c. ,
L = d Qa ZQ X, X b Q +
(A.1)
where ab and fabc are superpotential mass parameters and dimensionless coupling constants,
respectively.
The wave-function renormalization ZQ is a Hermitian matrix which depends on
|X| = XX . Its -expansion at X = X is given by

1 ZQ
1 2 ZQ 2 2 F F
2F
2F

ZQ |X| X=X = ZQ (M) +

(A.2)

.
+
+
2 ln M
M
M
4 ln M 2
M2
After expressing Eq. (A.1) in terms of the canonically normalized superfields Q as

Z1 ZQ 2
1/2
Q= 1+ Q
F ZQ Q ,
2 M
(A.3)
we can extract the SSB masses for the scalar component of Q from the quartic terms 2 2 in
the first integral and the SSB bilinear and trilinear couplings from the quadratic terms 2 in the
second and third integral, respectively

1 1/2 2 ZQ
ZQ 1 ZQ
1/2 F F
2
,
Z
Z
m = ZQ
Q
Q
Q
4
ln M
ln M 2 ln M
M2

a

b
1
1/2 ZQ
1/2
1/2 ZQ
1/2
Aabc = fa bc ZQ
+ fab c ZQ
ZQ
ZQ
2
ln M
ln M
a
b

c
F
1/2 ZQ
1/2
+ fabc ZQ
Z
,
ln M Q
M
c

a

b
1
1/2 ZQ
1/2
1/2 ZQ
1/2
Bab = a b ZQ
(A.4)
+ ab ZQ
ZQ
ZQ
.
2
ln M
ln M
a
b
In order to find the explicit expressions for the SSB parameters at an energy scale we recall
that the -dependence of ZQ is expressed by the RG equation:

1/2
ZQ
1/2
dZQ
dt
= Q,
(A.5)
19 Here the index a labels either the superfield and its associated charges or only the charges; the context should
make clear the case.
108

1/2
1/2
where t = ln and Q is the matrix of anomalous dimension. By defining ZQ = 1 + ZQ ,

1/2
where ZQ encodes the quantum corrections, Eq. (A.5) reads
1/2
dZQ
dt

1/2
= Q + ZQ , Q ,
2
4
(A.6)
at the lowest order. By following the procedure outlined in Ref. [31], we have to formally in1/2
tegrate the above equation to obtain ZQ in terms of the anomalous dimension. Afterwards,
the solution can be plugged into the expressions (A.4) to extract the SSB terms. For the sake of
brevity, we do not report all the intermediate steps (which can be easily performed) and, instead,
give the final expressions at = M:

1 d Q (M)
FF
d
2
<
m =
,

Q (M)
Q
4
d ln M
d ln M
M2
1

F
,
Aabc = fa bc Q (M)aa + fab c Q (M)bb + fabc Q (M)cc
2
M
1

F
Bab = a b Q (M)aa + ab Q (M)bb
(A.7)
,
2
M
<
where >
Q (M) [ Q (M)] is the anomalous dimension above (below) the mass scale M,
d
>
<
Q (M) Q (M) <
Q (M) and ( d ln M ) Q (M) means considering the difference of the
<
beta-functions of the couplings contained in Q (M) above and below M. Notice that our result (A.7) for m2 differs from the one obtained in Ref. [31] because an extra term proportional
Q
<
to the commutator [ >
Q , Q ] appears in that work, which is manifestly inconsistent for a Hermitian quantity such as m2 . We believe that the appearance of such a commutator is due to the
Q
improper definition of the RG equation for ZQ given in Eq. (3.5) of Ref. [31].
In the following, we provide the explicit expressions for the relevant anomalous dimensions
needed to extract the SSB parameters from Eq. (A.7) in our specific framework. The anomalous
dimensions below the scale M are:

3 2 3 2
2 <
16 L = 2Ye Ye 4
g + g
,
20 1 4 2

12 2
16 2 <
g ,
ec = 4Ye Ye
5 1

1 2 3 2 4 2
2 <
g + g + g
,
16 Q = 2Yd Yd + 2Yu Yu 4
60 1 4 2 3 3

1 2 4 2
16 2 <
g + g
,
d c = 4Yd Yd 4
15 1 3 3

4 2 4 2
4Y
Y
4
+
g
g
,
16 2 <
c
u
u
u
15 1 3 3

3 2 3 2
2
Tr
Y
Y
+
3Y
Y
+
4
16 2 <
g
g
,
e e
H1
d d
20 1 4 2

3 2 3 2
4
=
6
Tr
Y
Y
+
g
g
.
16 2 <
(A.8)
u u
H2
20 1 4 2
109
The differences Q (M) are instead the following:

16 2 L = 6 YT YT + YZ YZ ,

16 2 d c = 4 YZ YZ + 2YS YS ,
16 2 H2 = 6||2 ,
16 2 F = 0,
F = ec , Q, uc , H1 .
(A.9)
Regarding the expressions for the RG equations at one loop in the MSSM framework with the
(15 + 15) SU(5) representation we refer to Ref. [9]. Finally, we obtain the explicit formulas given
in Eqs. (16) and (17).
Appendix B. Coefficients of the j i Z operators
In this appendix we compute the coefficients AL,R of the monopole operators

L(R)
L(R),c
L(R),n
c c
j + AR
gZ AL
= Aj i
+ Aj i
j i i
j i i j + h.c. Z , Aj i
(B.1)
where gZ = g2 /cW (cW = cos W , W is the weak mixing angle) and AL(R),c [AL(R),n ] stand
for the contributions from the chargino/sneutrino [neutralino/charged-slepton] loop diagrams.
The two-component spinor notation is used such that, for example, i (ci ) is the left-handed
(right-handed) component of the lepton i field (i = e, , ). Different one-loop results for such
coefficients have been presented in the literature. For instance, the authors of Ref. [11] provided an all-order calculation in the electroweak breaking effects, while the authors of Ref. [56]
performed a lowest-order calculation. We found dramatic numerical discrepancies between the
two aforementioned results, which cannot be ascribed to the approximation used in Ref. [56].
Moreover, the authors of Ref. [59] have recently re-evaluated the contributions to AL,R and
claimed to have found additional contributions disregarded in Ref. [11]. We have independently
performed the all-order computation to compare with the previous results and to clarify this issue. The notation of Ref. [11] has been adopted to define the mass eigenstates of the charged
sleptons X (X = 1, . . . , 6), sneutrinos X (X = 1, . . . , 3), charginos A (A = 1, 2) and neu ,
tralinos A0 (A = 1, . . . , 4). The corresponding (unitary) mixing matrices are denoted by UXi
UXi , (OL )A , (OR )A and (ON )A (where () is the current-basis index for charged or neutral
gauginos/higgsinos). The relevant interactions of sleptons/leptons with charginos and neutralinos
are:
R
c L
c L c 0
R 0
X +
L = i CiAX
A + ei CiAX X A + i NiAX X A + ei NiAX X A + h.c.,
(B.2)
where
mi
R
= g2 (OR )A1 UXi

,
CiAX
= g2
,
(OL )A2 UXi
CiAX
2mW cos

g2
mi
R

NiAX
= (ON )A2 + (ON )A1 tan W UXi
+
(ON )A3 UX(i+3)
,
mW cos
2

g2
mi
L

= 2(ON )A1 tan W UX(i+3)
+
(ON )A3 UX(i+3)
.
NiAX
mW cos
2
The interactions of charginos and neutralinos with the Z boson are the following:

L = gZ + A B+ RAB A B LAB + 0 A B0 NAB Z ,
(B.3)
(B.4)
110
where

1
2
2
sW OR A2 (OR )B2 ,
RAB = cW OR A1 (OR )B1 +
2

1

(ON )B4 .
ON A3 (ON )B3 ON
NAB =
A4
2

,
LAB = RAB
RL
(B.5)
The chargino contributions are:
AL,c
ji
R C R
CiBX
j AX
16 2

AB
RAB
F (xA X , xB X ) + ln
2
2
MA2
MA MB
H (x ) + ln 2
A X
1
MA
ZL
,
2
2
m2 X
LAB G(xA X , xB X )
L,c
AR,c
j i = Aj i LR ,
(B.6)
(B.7)

= T3L(R) Q s2W (T3L = 12 , T3R = 0 and Q = 1).
here A, B = 1, 2, X = 1, 2, 3 and ZL(R)
(A summation over repeated indices is understood.) In the above equations the first and second
terms come from the diagrams in which the Z boson line is attached to the chargino line, the
third one where it is attached to the sneutrino and the fourth term comes from the wave function
renormalization. For completeness, we have also displayed the terms proportional to ln(2 /MA2 )
in Eq. (B.6), coming from the divergent diagrams, where is the renormalization scale. Obviously, such terms cancel out. As for the argument of the loop functions we have adopted the
convention xab = m2a /m2b , then e.g. xA X = M2A /m2 X . The loop functions are defined as follows:
2

1
1
x ln x y 2 ln y
F (x, y) = + ln x +
,
2
x y 1x
1y

x ln x
y ln y
1
,
G(x, y) =
x y 1x 1y

3
1
(1 2x) ln x
H (x) = +
(B.8)
.
2
1x
(1 x)2
Using the relation H (x) = F (x, x) 2xG(x, x), one can easily verify the validity of the Ward
Takahashi (WT) identity in the SU(2)W U (1)Y unbroken phase, which entails the vanishing of
the coefficient, AL(R),c |v1 =v2 =0 = 0. By exploiting this identity in Eq. (B.6), the above expressions (B.6) simplify to:

F (xA X , xB X )
(OR )A2 OR B2
2
4
16

MA MB
G(xA X , xB X ) ,
OL A2 (OL )B2
m2 X
L,c
AR,c
j i = Aj i LR .
AL,c
ji =
R C R
CiBX
j AX
(B.9)
(B.10)
111
The result obtained for AL in (B.9) coincides with that of Ref. [11] and is consistent with
the one in Ref. [56]. The formulas (B.6), (B.7) are also in agreement with those reported20 in
Ref. [59], which, however, have not been reduced to the form (B.9), (B.10). We observe that
AR,c AL,c because of the Yukawa coupling suppression (this coefficient has been set directly
to zero in [11]).
The coefficients AL(R),n from the neutralino-exchange contributions are given by:
AL,n
ji
R N R
NiBX
j AY
XY NAB
16 2
+ NAB
MA MB
m2 X

UY k
UXk
AR,n
ji =

AB 2
2
G(xA X , xB X )
sW XY H (xA X ) + ln 2
2
MA
16 2

XY NAB
MA MB
m2 X

UY k
UXk
2
MA2

+ AB XY ZL
2

2
MA2
I (xA Y , xX Y ) + ln
L N L
NiBX
j AX
+ NAB
H (xA X ) + ln
2
MA2

,
2
MA2

AB 2
2
G(xA X , xB X ) +
sW XY H (xA X ) + ln 2
2
MA
I (xA Y , xX Y ) + ln
2
2
MA2

AB XY ZR
H (xA X ) + ln
2
2
MA2

, (B.11)
where A, B = 1, . . . , 4, X = 1, . . . , 6, k = e, , and the loop function I (x, y) = 1 + F (x, y).

The first and second terms derive from the contributions with the Z attached to the neutralino
line, the third and fourth from those with the Z attached to the slepton line, and the fifth one
from the wave-function renormalization diagram. By using again the WT the above expressions
simplify as21 :
AL,n
ji =
R N R
NiBX
j AY
16 2

+ AB

F (xA X , xB X )
MA MB
G(x
,
x
)
+ NAB
XY NAB
A X B X
2
m2
X

I (xA Y , xX Y )
H (xA X )

UY k
XY
,
UXk
4
4
20 In fact, the agreement between Eqs. (B.6), (B.7) and the corresponding formulas in Ref. [59] does not regard the constant terms in the loop-functions F (x, y) and H (x). Nevertheless, such terms do not contribute because of the unitarity
relations.
21 Using the simplified formulas (B.10), (B.10) and (B.12) is more convenient also because cancellations are already
accounted for. Needless to say that the constant numerical addenda appearing in the loop functions (B.8) do not contribute
to the final amplitudes because of unitarity of the mixing matrices U and U . Still, they are essential to prove the WT
identities and then to yield the simplified formulas (B.9), (B.10) and (B.12).
112
AR,n
ji
L N L
NiBX
j AY
16 2

UY k
AB UXk

XY NAB
F (xA X , xB X )
I (xA Y , xX Y )
4
2

M A M B
+ NAB
m2
X

G(xA X , xB X )
(B.12)
Our results (B.11) are compatible with those of Ref. [59] (see, however, the comment in footnote 20), and are also compatible with Ref. [56]. Instead, our final expressions (B.12) differ from
those of Ref. [11], because of the third and fourth (third) terms in AL,n (AR,n ).
Appendix C. Box coefficients for the j i k k (i = k) amplitudes
We consider the four-fermion operators which are relevant for the amplitude of the LFV decays j i k k :
LL
c c RL

c
c
k + B RR c c + h.c.
i j Bj i;k k k + BjLR
k
i;k k k + i j Bj i;k k
j i;k k
(C.1)
Each coefficient B MN (M, N = L, R) receive contributions from box diagrams with either
charginos/sneutrinos or neutralinos/charged-sleptons exchange (see Fig. 16):
MN(c)
BjMN
i;k = Bj i;k
MN(n)
+ Bj i;k
(C.2)
We have obtained the following results valid for i

= k:

R

1 2
2
2
2
R
R
R
R
R
R
R
BjLL(c)
i;k = 4 J4 M A , M B , m X , m Y CiAY Cj AX CkBX CkBY + Cj AX CiBX CkAY CkBY ,
R
1
LR(c)
L
L
CjR
Bj i;k = J4 M2A , M2B , m2 X , m2 Y CiAY
AX CkBX CkBY
4

R
1
L
L
MA MB I4 M2A , M2B , m2 X , m2 Y CiBX
CjR
AX CkAY CkBY ,
2

R

1
LL(n)
R
R
R
R
R
NjR
NkAY
NkBY
+ NiAY
NjR
NkBX
NkBY
Bj i;k = J4 M2A , M2B , m2 , m2 NiBX
AX
AX
X
Y
4
2

R
1
R
R
NjR
+ MA MB I4 MA , M2B , m2 , m2 NiBX
AX NkBY NkAY
X
Y
2

R
R
R
+ NiBY
NjR
AX NkBX NkAY ,

R
1 2
2
2
2
R
L
L
R
R
L
L
BjLR(n)
i;k = 4 J4 M A , M B , mX , mY NiAY Nj AX NkBY NkBX NiBX Nj AX NkAY NkBY

R
L
L
+ NiBY
NjR
AX NkAY NkBX

R
1
L
L
NjR
MA MB I4 M2A , M2B , m2 , m2 NiBX
AX NkBY NkAY ,
X
Y
4
RR(c)
LL(c)
RL(c)
LR(c)
Bj i;k = Bj i;k LR ,
RR(n)
LL(n)
RL(n)
LR(n)
(C.3)
113
RR
Fig. 16. Box diagrams relevant for the computation of the coefficients BjLL
i;k (first and second rows), Bj i;k (third and
RL
fourth rows), BjLR
i;k (fifth and sixth rows) and Bj i;k (seventh and eighth rows) (see Eqs. (C.1)(C.3)).
L(R)
L(R)
where the coefficients NiAX , CiAX have been defined in Eq. (B.2) and the loop integrals are
given as:
114

i
I4 m21 , . . . , m24
16 4

i
J4 m21 , . . . , m24
16 4

d4 k
(k 2 m21 ) (k 2
k 2 d4 k
m24 )
(k 2 m21 ) (k 2 m24 )
,
.
(C.4)
The box coefficients for the decays ee and e correspond to the replacements
(j, i, k) (, , e) and (j, i, k) (, e, ), respectively. Our results for these coefficients are
in numerical agreement with those of Ref. [56].
References
[1] S. Weinberg, Phys. Rev. Lett. 43 (1979) 1566.
[2] P. Minkowski, Phys. Lett. B 67 (1977) 421;
M. Gell-Mann, P. Ramond, R. Slansky, in: P. Van Nieuwenhuizen, D. Freedman (Eds.), Supergravity, North-Holland,
Amsterdam, 1979, p. 315;
T. Yanagida, in: O. Sawada, A. Sugamoto (Eds.), Proceedings of the Workshop on the Unified Theory and the
Baryon Number in the Universe, KEK, Tsukuba, 1979, p. 95;
S.L. Glashow, in: M. Lvy, et al. (Eds.), Quarks and Leptons, Plenum, New York, 1980, p. 707;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
[3] R. Foot, H. Lew, X.G. He, G.C. Joshi, Z. Phys. C 44 (1989) 441;
E. Ma, Phys. Rev. Lett. 81 (1998) 1171;
B. Brahmachari, E. Ma, U. Sarkar, Phys. Lett. B 520 (2001) 152.
[4] R. Barbieri, D.V. Nanopoulos, G. Morchio, F. Strocchi, Phys. Lett. B 90 (1980) 91;
R.E. Marshak, R.N. Mohapatra, VPI-HEP-80/02 Invited talk given at Orbis Scientiae, Coral Gables, FL, 1417
January 1980;
T.P. Cheng, L.F. Li, Phys. Rev. D 22 (1980) 2860;
M. Magg, Ch. Wetterich, Phys. Lett. B 94 (1980) 61;
J. Schechter, J.W.F. Valle, Phys. Rev. D 22 (1980) 2227;
G. Lazarides, Q. Shafi, C. Wetterich, Nucl. Phys. B 181 (1981) 287;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. D 23 (1981) 165;
E. Ma, U. Sarkar, Phys. Rev. Lett. 80 (1998) 5716.
[5] J.A. Casas, V. Di Clemente, A. Ibarra, M. Quiros, Phys. Rev. D 62 (2000) 053005.
[6] J.R. Ellis, D.V. Nanopoulos, Phys. Lett. B 110 (1982) 44;
R. Barbieri, R. Gatto, Phys. Lett. B 110 (1982) 211;
M.J. Duncan, Nucl. Phys. B 221 (1983) 285;
J.F. Donoghue, H.P. Nilles, D. Wyler, Phys. Lett. B 128 (1983) 55.
[7] F. Borzumati, A. Masiero, Phys. Rev. Lett. 57 (1986) 961.
[8] F. Gabbiani, A. Masiero, Nucl. Phys. B 322 (1989) 235.
[9] A. Rossi, Phys. Rev. D 66 (2002) 075003.
[10] See A. Casas, A. Ibarra, in Ref. [12];
S. Davidson, A. Ibarra, JHEP 0109 (2001) 013.
[11] J. Hisano, T. Moroi, K. Tobe, M. Yamaguchi, Phys. Rev. D 53 (1996) 2442.
[12] J. Hisano, D. Nomura, T. Yanagida, Phys. Lett. B 437 (1998) 351;
J. Hisano, D. Nomura, Phys. Rev. D 59 (1999) 116005;
J.R. Ellis, M.E. Gomez, G.K. Leontaris, S. Lola, D.V. Nanopoulos, Eur. Phys. J. C 14 (2000) 319;
J.A. Casas, A. Ibarra, Nucl. Phys. B 618 (2001) 171;
D.F. Carvalho, J.R. Ellis, M.E. Gomez, S. Lola, Phys. Lett. B 515 (2001) 323;
S. Lavignac, I. Masina, C.A. Savoy, Phys. Lett. B 520 (2001) 269;
A. Kageyama, S. Kaneko, N. Shimoyama, M. Tanimoto, Phys. Rev. D 65 (2002) 096010;
A. Masiero, S.K. Vempati, O. Vives, Nucl. Phys. B 649 (2003) 189;
K.S. Babu, B. Dutta, R.N. Mohapatra, Phys. Rev. D 67 (2003) 076006;
T. Blazek, S.F. King, Nucl. Phys. B 662 (2003) 359;
S.T. Petcov, S. Profumo, Y. Takanishi, C.E. Yaguna, Nucl. Phys. B 676 (2004) 453;
M. Hirsch, J.C. Romao, S. Skadhauge, J.W.F. Valle, A. Villanova del Moral, Phys. Rev. D 69 (2004) 093006;
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
E. Arganda, M.J. Herrero, Phys. Rev. D 73 (2006) 055003;

L. Calibbi, A. Faccia, A. Masiero, S.K. Vempati, hep-ph/0605139;
P. Hosteins, S. Lavignac, C.A. Savoy, hep-ph/0606078;
S. Antusch, E. Arganda, M.J. Herrero, A.M. Teixeira, hep-ph/0607263.
F.R. Joaquim, A. Rossi, Phys. Rev. Lett. 97 (2006) 181801.
T. Hambye, E. Ma, U. Sarkar, Nucl. Phys. B 602 (2001) 23.
S. Eidelman, et al., Particle Data Group, Phys. Lett. B 592 (2004) 1.
L.J. Hall, L. Randall, Phys. Rev. Lett. 65 (1990) 2939.
V. Cirigliano, B. Grinstein, G. Isidori, M.B. Wise, Nucl. Phys. B 728 (2005) 121;
B. Grinstein, V. Cirigliano, G. Isidori, M.B. Wise, hep-ph/0608123;
S. Davidson, F. Palorini, hep-ph/0607329;
G.C. Branco, A.J. Buras, S. Jager, S. Uhlig, A. Weiler, hep-ph/0609067.
V.S. Kaplunovsky, J. Louis, Phys. Lett. B 306 (1993) 269;
A. Brignole, L.E. Ibanez, C. Munoz, Nucl. Phys. B 422 (1994) 125;
A. Brignole, L.E. Ibanez, C. Munoz, Nucl. Phys. B 436 (1995) 747, Erratum.
L.J. Hall, V.A. Kostelecky, S. Raby, Nucl. Phys. B 267 (1986) 415.
R. Barbieri, L.J. Hall, Phys. Lett. B 338 (1994) 212.
E.J. Chun, A. Masiero, A. Rossi, S.K. Vempati, Phys. Lett. B 622 (2005) 112.
C. Bachas, C. Fabre, T. Yanagida, Phys. Lett. B 370 (1996) 49;
S. Chaudhuri, G. Hockney, J.D. Lykken, Nucl. Phys. B 469 (1996) 357;
T. Han, T. Yanagida, R.J. Zhang, Phys. Rev. D 58 (1998) 095011;
B. Brahmachari, Phys. Rev. D 65 (2002) 067502;
I. Dorsner, P.F. Perez, Nucl. Phys. B 723 (2005) 53;
I. Dorsner, P.F. Perez, R. Gonzalez Felipe, Nucl. Phys. B 747 (2006) 312.
P. Langacker, B.D. Nelson, Phys. Rev. D 72 (2005) 053013, hep-ph/0507063.
M. Cvetic, I. Papadimitriou, G. Shiu, Nucl. Phys. B 659 (2003) 193;
M. Cvetic, I. Papadimitriou, G. Shiu, Nucl. Phys. B 696 (2004) 298, Erratum;
M. Cvetic, P. Langacker, hep-th/0607238.
H. Georgi, C. Jarlskog, Phys. Lett. B 86 (1979) 297;
J.A. Harvey, P. Ramond, D.B. Reiss, Phys. Lett. B 92 (1980) 309;
For more recent works see for example: K.S. Babu, S.M. Barr, Phys. Rev. D 56 (1997) 2614;
Z. Berezhiani, Phys. Lett. B 417 (1998) 287;
R. Barbieri, L.J. Hall, S. Raby, A. Romanino, Nucl. Phys. B 493 (1997) 3;
Z. Berezhiani, A. Rossi, JHEP 9903 (1999) 002;
Z. Berezhiani, A. Rossi, Nucl. Phys. B 594 (2001) 113;
G. Altarelli, F. Feruglio, I. Masina, JHEP 0011 (2000) 040.
S. Dimopoulos, H. Georgi, Nucl. Phys. B 193 (1981) 150;
N. Sakai, Z. Phys. C 11 (1981) 153;
E. Witten, Nucl. Phys. B 188 (1981) 513.
M. Dine, W. Fischler, M. Srednicki, Nucl. Phys. B 189 (1981) 575;
S. Dimopoulos, S. Raby, Nucl. Phys. B 192 (1981) 353;
L. Alvarez-Gaume, M. Claudson, M.B. Wise, Nucl. Phys. B 207 (1982) 96;
C.R. Nappi, B.A. Ovrut, Phys. Lett. B 113 (1982) 175;
S. Dimopoulos, S. Raby, Nucl. Phys. B 219 (1983) 479;
M. Dine, A.E. Nelson, Phys. Rev. D 48 (1993) 1277;
M. Dine, A.E. Nelson, Y. Shirman, Phys. Rev. D 51 (1995) 1362.
For a recent review see G.F. Giudice, R. Rattazzi, Phys. Rep. 322 (1999) 419.
M. Dine, Y. Nir, Y. Shirman, Phys. Rev. D 55 (1997) 1501, hep-ph/9607397.
L.V. Avdeev, D.I. Kazakov, I.N. Kondrashuk, Nucl. Phys. B 510 (1998) 289;
G.F. Giudice, R. Rattazzi, Nucl. Phys. B 511 (1998) 25;
N. Arkani-Hamed, G.F. Giudice, M.A. Luty, R. Rattazzi, Phys. Rev. D 58 (1998) 115005;
C.E.M. Wagner, Nucl. Phys. B 528 (1998) 3.
Z. Chacko, E. Ponton, Phys. Rev. D 66 (2002) 095004.
For a string-inspired realization of GMSB see e.g. E. Floratos, C. Kokorelis, hep-th/0607217.
G.R. Dvali, M.A. Shifman, Phys. Lett. B 399 (1997) 60;
Z. Chacko, E. Katz, E. Perazzi, Phys. Rev. D 66 (2002) 095012.
115
116
[34] See e.g., M.C. Gonzalez-Garcia, Phys. Scr. T 121 (2005) 72;
G.L. Fogli, E. Lisi, A. Marrone, A. Palazzo, Prog. Part. Nucl. Phys. 57 (2006) 742;
A. Strumia, F. Vissani, Nucl. Phys. B 726 (2005) 294;
R.N. Mohapatra, A.Y. Smirnov, hep-ph/0603118;
J.W.F. Valle, hep-ph/0603223.
[35] M.L. Brooks, et al., MEGA Collaboration, Phys. Rev. Lett. 83 (1999) 1521.
[36] M. Grassi, MEG Collaboration, Nucl. Phys. B (Proc. Suppl.) 149 (2005) 369.
[37] B. Aubert, et al., BaBar Collaboration, Phys. Rev. Lett. 95 (2005) 041802.
[38] A.G. Akeroyd, et al., SuperKEKB Physics Working Group, hep-ex/0406071.
[39] B. Aubert, et al., BaBar Collaboration, Phys. Rev. Lett. 96 (2006) 041801.
[40] U. Bellgardt, et al., SINDRUM Collaboration, Nucl. Phys. B 299 (1988) 1.
[41] J. Aysto, et al., hep-ph/0109217;
B.L. Roberts, M. Grassi, A. Sato, Nucl. Phys. B (Proc. Suppl.) 155 (2006) 123.
[42] B. Aubert, et al., BaBar Collaboration, hep-ex/0312027.
[43] Y. Yusa, H. Hayashii, T. Nagamine, A. Yamaguchi, Belle Collaboration, eConf C0209101 (2002) TU13, Nucl. Phys.
B (Proc. Suppl.) 123 (2003) 95, hep-ex/0211017.
[44] M. Hodgkinson, BaBar Collaboration, Nucl. Phys. B (Proc. Suppl.) 144 (2005) 167.
[45] J. Kaulard, et al., SINDRUM II Collaboration, Phys. Lett. B 422 (1998) 334.
[46] Y. Mori, et al., PRISM/PRIME Working group, LOI at J-PARC 50-GeV PS, LOI-25, http://psux1.kek.jp/~jhfnp/LOIlist/LOIlist.html.
[47] R. Barate, et al., LEP Working Group for Higgs Boson Searches, Phys. Lett. B 565 (2003) 61.
[48] S. Heinemeyer, W. Hollik, G. Weiglein, Comput. Phys. Commun. 124 (2000) 76.
[49] See for a review see e.g., M. Passera, Nucl. Phys. B (Proc. Suppl.) 155 (2006) 365.
[50] Tevatron Electroweak Working Group, hep-ex/0603039.
[51] See e.g., N.K. Falck, Z. Phys. C 30 (1986) 247;
S.P. Martin, M.T. Vaughn, Phys. Rev. D 50 (1994) 2282.
[52] For studies on the sparticle spectrum in GMSB models see e.g., S. Dimopoulos, S.D. Thomas, J.D. Wells, Nucl.
Phys. B 488 (1997) 39;
A. Strumia, Phys. Lett. B 409 (1997) 213.
[53] D.M. Pierce, J.A. Bagger, K.T. Matchev, R.j. Zhang, Nucl. Phys. B 491 (1997) 3.
[54] For a review see N.V. Krasnikov, V.A. Matveev, Phys. Usp. 47 (2004) 643, Usp. Fiz. Nauk 174 (2004) 697,
http://CMSinfo.cern.ch/Welcome.html/CMSdocuments/CMSplots.
[55] M. Carena, H.E. Haber, Prog. Part. Nucl. Phys. 50 (2003) 63, and references therein.
[56] A. Brignole, A. Rossi, Nucl. Phys. B 701 (2004) 3.
[57] K.S. Babu, C. Kolda, Phys. Rev. Lett. 89 (2002) 241802;
A. Dedes, J.R. Ellis, M. Raidal, Phys. Lett. B 549 (2002) 159.
[58] A. Brignole, A. Rossi, Phys. Lett. B 566 (2003) 217.
[59] See E. Arganda, M.J. Herrero in Ref. [12].
[60] F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321;
J. Foster, K.I. Okumura, L. Roszkowski, JHEP 0603 (2006) 044.
[61] M.A. Giorgi, et al., SuperB group, INFN Roadmap Report, March 2006.
[62] M. Komatsu, P. Migliozzi, F. Terranova, J. Phys. G 29 (2003) 443;
P. Migliozzi, F. Terranova, Phys. Lett. B 563 (2003) 73;
P. Huber, J. Kopp, M. Lindner, M. Rolinec, W. Winter, JHEP 0605 (2006) 072.
[63] For a recent review, see A. Blondel, A. Cervera-Villanueva, A. Donini, P. Huber, M. Mezzetto, P. Strolin, hepph/0606111.
[64] M. Fukugita, T. Yanagida, Phys. Lett. B 174 (1986) 45.
[65] G.C. Branco, R. Gonzalez Felipe, F.R. Joaquim, M.N. Rebelo, Nucl. Phys. B 640 (2002) 202;
W. Buchmuller, P. Di Bari, M. Plumacher, Nucl. Phys. B 643 (2002) 367;
G.C. Branco, R. Gonzalez Felipe, F.R. Joaquim, I. Masina, M.N. Rebelo, C.A. Savoy, Phys. Rev. D 67 (2003)
073025;
S. Pascoli, S.T. Petcov, W. Rodejohann, Phys. Rev. D 68 (2003) 093007;
E.K. Akhmedov, M. Frigerio, A.Y. Smirnov, JHEP 0309 (2003) 021;
G.F. Giudice, A. Notari, M. Raidal, A. Riotto, A. Strumia, Nucl. Phys. B 685 (2004) 89;
T. Hambye, Y. Lin, A. Notari, M. Papucci, A. Strumia, Nucl. Phys. B 695 (2004) 169;
R. Gonzalez Felipe, F.R. Joaquim, B.M. Nobre, Phys. Rev. D 70 (2004) 085009;
117
W. Buchmuller, P. Di Bari, M. Plumacher, Ann. Phys. 315 (2005) 305;

G.C. Branco, R. Gonzalez Felipe, F.R. Joaquim, B.M. Nobre, Phys. Lett. B 633 (2006) 336;
A. Abada, S. Davidson, F.X. Josse-Michaux, M. Losada, A. Riotto, JCAP 0604 (2006) 004;
E. Nardi, Y. Nir, E. Roulet, J. Racker, JHEP 0601 (2006) 164;
A. Abada, S. Davidson, A. Ibarra, F.X. Josse-Michaux, M. Losada, A. Riotto, hep-ph/0605281;
V. Cirigliano, G. Isidori, V. Porretti, hep-ph/0607068;
K. Bhattacharya, N. Sahu, U. Sarkar, S.K. Singh, hep-ph/0607272.
[66] G. DAmbrosio, T. Hambye, A. Hektor, M. Raidal, A. Rossi, Phys. Lett. B 604 (2004) 199;
For previous works on resonant leptogenesis in different contexts see e.g. M. Flanz, E.A. Paschos, U. Sarkar, Phys.
Lett. B 345 (1995) 248;
M. Flanz, E.A. Paschos, U. Sarkar, Phys. Lett. B 382 (1996) 447, Erratum;
L. Covi, E. Roulet, F. Vissani, Phys. Lett. B 384 (1996) 169;
M. Flanz, E.A. Paschos, U. Sarkar, J. Weiss, Phys. Lett. B 389 (1996) 693;
G. DAmbrosio, G.F. Giudice, M. Raidal, Phys. Lett. B 575 (2003) 75;
Y. Grossman, T. Kashti, Y. Nir, E. Roulet, Phys. Rev. Lett. 91 (2003) 251801;
Y. Grossman, T. Kashti, Y. Nir, E. Roulet, Nucl. Phys. B 692 (2004) 303.
[67] T. Hambye, G. Senjanovic, Phys. Lett. B 582 (2004) 73;
S. Antusch, S.F. King, Phys. Lett. B 597 (2004) 199;
T. Hambye, hep-ph/0412053;
T. Hambye, M. Raidal, A. Strumia, Phys. Lett. B 632 (2006) 667.
[68] Y. Farzan, JHEP 0502 (2005) 025.
Heterotic string dynamics

in the solvable Lie algebra gauge
Nejat T. Ylmaz
a Department of Mathematics and Computer Science, ankaya University, gretmenler Cad. No. 14,
06530 Balgat, Ankara, Turkey

Received 11 July 2006; received in revised form 18 October 2006; accepted 17 November 2006
Abstract
A revision of the toroidal KaluzaKlein compactification of the massless sector of the E8 E8 heterotic
string is given. Under the solvable Lie algebra gauge the dynamics of the O(p, q)/(O(p) O(q)) symmetric space sigma model which is coupled to a dilaton, N Abelian gauge fields and the ChernSimons
type field strength is studied in a general formalism. The results are used to derive the bosonic matter field
equations of the massless sector of the D-dimensional compactified E8 E8 heterotic string.
1. Introduction
The supergravity theory which has the highest spacetime dimension is the D = 11, N = 1
supergravity [1]. There are three types of supergravity theories in ten dimensions namely the IIA
[24], the IIB [57] and as the third supergravity, the ten-dimensional N = 1 type I supergravity
theory which is coupled to the YangMills theory [8,9]. One can obtain the D = 10, IIA supergravity theory by the KaluzaKlein dimensional reduction of the D = 11 supergravity on the
circle, S 1 . The supergravity theories in D < 10 dimensions can be obtained from the D = 11 and
the D = 10 supergravities by the dimensional reduction and the truncation of fields. A general
treatment of the supergravity theories can be found in [1013].
The ten-dimensional IIA supergravity and the IIB supergravity theories are the massless sectors or the low energy effective limits of the type IIA and the type IIB superstring theories
respectively [14]. The type I supergravity theory in ten dimensions on the other hand is the low
E-mail address: ntyilmaz@cankaya.edu.tr (N.T. Ylmaz).
doi:10.1016/j.nuclphysb.2006.11.031
N.T. Ylmaz / Nuclear Physics B 765 (2007) 118141
119
energy effective limit of the type I superstring theory and the heterotic superstring theory [14].
The eleven-dimensional supergravity is conjectured to be the low energy effective limit of the
eleven-dimensional M-theory.
The symmetries of the supergravity theories are important to understand the symmetries and
the duality transformations of the string theories. Especially the global symmetries of the supergravities give the non-perturbative U-duality symmetries of the string theories and the M-theory
[15,16]. An appropriate restriction of the global symmetry group G of the supergravity theory to
the integers Z, namely G(Z), is conjectured to be the U-duality symmetry of the relative string
theory which unifies the T-duality and the S-duality [14,15].
In supergravity theories which possess scalar fields, the global (rigid) symmetries of the scalar
sector can be extended to the other fields as well, thus the global symmetry of the scalars will be
the global symmetry of the entire bosonic sector of the theory. For the majority of the supergravities, the scalar manifolds are homogeneous symmetric spaces which are in the form of cosets
G/K [17], and the scalar Lagrangians can be formulated as the non-linear coset sigma models,
in particular the symmetric space sigma models. The dimension of the coset space G/K is equal
to the number of the coset scalars of the theory. In general G is a real form of a non-compact
semi-simple Lie group and K is its maximal compact subgroup. Thus the coset G/K can locally
be parameterized through the exponential map by using the solvable Lie subalgebra of G [18].
This parametrization is an effective tool for the dynamics of the supergravity theories and it is
called the solvable Lie algebra gauge [17].
When we apply the KaluzaKlein dimensional reduction to the bosonic sector of the D = 11
supergravity [1] over the tori T n , where n = 11 D, we obtain the D-dimensional maximal
supergravity theories [1921]. The global (rigid) symmetry groups of the bosonic sectors of the
reduced Lagrangians are in split real form (maximally non-compact). In other words for the
scalar coset manifolds, G/K of the maximal supergravities, G which is the global symmetry
group, is a semi-simple split real form and K is its maximal compact subgroup. Thus the coset
spaces G/K can be parameterized by the Borel gauge which is a special case of the solvable Lie
algebra gauge. Therefore the scalar sectors of the maximal supergravities can be formulated as
symmetric space sigma models. However we should remark that in certain dimensions in order
to formulate the scalar sectors as symmetric space sigma models one has to make use of the
dualisation methods to replace the higher-order fields with the newly defined scalars. On the
other hand when one considers the KaluzaKlein compactification of the bosonic sector of the
ten-dimensional N = 1 simple supergravity that is coupled to N Abelian gauge multiplets on
the tori T 10D [8,9], one can show that when a single scalar is decoupled from the others, the
rest of the scalars of the lower, D-dimensional theories can be formulated as G/K symmetric
space sigma models [22]. One can even enlarge the coset formulations of the scalars by using
partial dualisations. Unlike the maximal supergravities for this class of supergravities, the global
symmetry group G is not necessarily a split real form but it is in general a semi-simple, noncompact real form and again K is a maximal compact subgroup of G. Therefore in this case one
makes use of the more general solvable Lie algebra gauge [17] to parameterize the scalar coset
manifolds for the reduced theories.
In this work considering the achievements of [22] we focus on the dynamics of the toroidally
compactified low energy effective E8 E8 heterotic string in the solvable Lie algebra gauge. In
accordance with the formulation of [22] we will first show how one can obtain a lower dimensional bosonic Lagrangian starting from the ten-dimensional N = 1 type I supergravity theory
which is coupled to the Abelian YangMills theory [8,9]. We will formulate the D-dimensional
bosonic Lagrangian in a compact form in which the terms governing the scalars and the coupling
120
Abelian gauge fields are constructed in the solvable Lie algebra gauge. Then in a general formalism we will work out in detail the bosonic field equations of the symmetric space sigma model
which is coupled to a dilaton, an arbitrary number of Abelian gauge fields and a two-form field
whose kinetic term is expressed in terms of its ChernSimons field strength. Our formulation
will be a generalized one which uses arbitrary coupling constants. Also it is performed for a general, O(p, q) global symmetry group of the scalar coset. In our generalized formulation we will
make use of the construction of [23] to derive the field equations of the coset scalars. To denote
the relevance of our construction with the supergravity theories we will give three examples of
MaxwellEinstein supergravities on which the derivation we perform is applicable and whose
bosonic field equations correspond to the ones we derive in algebraic terms for the solvable Lie
algebra gauge of the scalar coset manifold. Finally we will apply our general results to the case of
the D-dimensional compactified E8 E8 heterotic string to derive its bosonic matter field equations which show a degree of complexity due to the coupling between the coset scalars and the
Abelian gauge fields. We will also discuss the relation between different coset parametrizations.
Besides we will reckon the correspondence of the two equivalent formulations of the symmetric
space sigma model with couplings which are based on different scalar coset representations.
In Section 2 we give a brief review of the KaluzaKlein reduction of the ten-dimensional
N = 1 supergravity that is coupled to the Abelian YangMills theory [8,9] which is studied
in detail in [22]. The main objective of Section 2 will be the construction of the D-dimensional
bosonic Lagrangian in a compact form by using the solvable Lie algebra gauge. In Section 3 again
under the solvable Lie algebra gauge we will derive the field equations of the symmetric space
sigma model which is generally coupled to a dilaton, Abelian gauge fields and a ChernSimons
type field strength. In addition we will mention about three examples on which the results can be
applied. In Section 4 we will write down the bosonic matter field equations of the D-dimensional
toroidally compactified heterotic string by using the results obtained in Section 3. We will also
discuss the possible field transformations between the two equivalent formulations of the bosonic
Lagrangian of the compactified heterotic string.
2. Toroidally compactified heterotic string
In this section we will focus on the KaluzaKlein reduction on the Euclidean torus T 10D
of the bosonic sector of the ten-dimensional simple N = 1 supergravity [8,9] which is coupled
to N Abelian gauge multiplets. We will not cover the details of the reduction steps. A detailed
study of the reduction can be referred in [22] in which the structure of the scalar cosets and the
matter couplings which appear in each dimension upon the dimensional reduction of the basic
fields of the ten-dimensional theory are given in detail. When as a special case, the number of
the U (1) gauge fields is chosen to be 16, the ten-dimensional supergravity which is coupled
to 16 Abelian U (1) gauge multiplets becomes the low energy effective limit of the E8 E8
ten-dimensional heterotic string theory which forms the massless background coupling [14].
Thus when N = 16 our formulation corresponds to the dimensional reduction of the low energy
effective bosonic Lagrangian of the ten-dimensional E8 E8 heterotic superstring theory. For
N = 16, the D = 10 YangMills supergravity [8,9] has the E8 E8 YangMills gauge symmetry,
however the general Higgs vacuum structure causes a spontaneous symmetry breakdown so that
the full symmetry E8 E8 can be broken down to its maximal torus subgroup U (1)16 , whose Lie
algebra is the Cartan subalgebra of E8 E8 . Thus the ten-dimensional YangMills supergravity
reduces to its maximal torus subtheory which is an Abelian (MaxwellEinstein) supergravity
theory. The bosonic sector of this Abelian YangMills supergravity corresponds to the low energy
121
effective theory of the bosonic sector of the fully higgsed ten-dimensional E8 E8 heterotic
string theory [14,22]. Therefore our main concern will be the Abelian YangMills supergravity
in ten dimensions, however for the purpose of generality we will consider the coupling of N
U (1) gauge field multiplets to the graviton multiplet.
As we have discussed before the scalar sectors of the lower dimensional theories can be formulated as non-linear sigma models, more specifically as the symmetric space sigma models.
One needs to apply the method of dualisation for some of the fields in certain dimensions to
construct the scalar cosets as symmetric space sigma models. The scalar cosets G/K are based
on the global internal symmetry groups G which are in general non-compact real forms of semisimple Lie groups. In supergravity theories the global symmetry of the scalar Lagrangian can be
extended to the entire bosonic sector of the theory. Under certain conditions the global internal
symmetry groups may be maximally non-compact (split) real forms but in general they are elements of a bigger class of Lie groups which contains the global internal symmetry groups of
the maximal supergravities namely the split real forms as a special subset. The main difference
between the scalar cosets based on the non-compact and the maximally non-compact numerator
groups is the parametrization one can choose for the coset representatives. For the general noncompact real forms one can make use of the solvable Lie algebra gauge [17] to parameterize the
scalar coset. The solvable Lie algebra is a subalgebra of the Borel algebra of g (the Lie algebra
of G). It is simply composed of certain Cartan and positive root generators of g [18].
The bosonic Lagrangian of the D = 10, N = 1 Abelian YangMills supergravity which is
coupled to N U (1) gauge multiplets is [8,9,22]
1
1
1 1 I
G(2) GI(2) ,
d1 d1 e1 F(3) F(3) e 2 1
2
2
2
N
L10 = R 1
(2.1)
I =1
I are the N U (1) gauge field strengths. In (2.1) is a scalar field and the
where GI(2) = dB(1)
1
subscripts for the rest of the fields and the field strengths denote the degree of the fields. We also
define the field strength of the field A(2) as
1 I
I
dB(1)
.
F(3) = dA(2) + B(1)
(2.2)
2
The pure supergravity sector of (2.1) can be truncated from the IIA supergravity [24] by choosing the extra RR fields to be zero in the IIA Lagrangian. The KaluzaKlein ansatz for the
ten-dimensional spacetime metric upon T n -reduction where n = 10 D is [24]
2
ds10
=e
s
2
dsD
11D
2
e2i hi .
(2.3)
i=2
Here we have
s = (s2 , s3 , . . . , s(11D) ),

fi = 0, . . . , 0, (10 i)si , si+1 , . . . , s(11D) ,
where in the second line there are i 2 zeros and

2
.
si =
((10 i)(9 i))
(2.4)
(2.5)
122
Also we define
1
1
i = s fi ,
2
2
(2.6)
and hi is
hi = dzi + Ai(1) + Ai(0)j dzj ,
(2.7)
where Ai(1) are the D-dimensional Maxwell gauge fields and Ai(0)j are the D-dimensional
scalars. The coordinates {zi } are the coordinates on the n-torus T n . We further define the vector

= 2 , 3 , . . . , (11D) ,
(2.8)
whose components are the dilatons of the KaluzaKlein reduction.
We should also mention how to built up an ansatz to reduce a general D-dimensional (n 1)1
form potential field AD
(n1) on S , with the assumption that the D-dimensional spacetime is
composed of the Cartesian product of a (D 1)-dimensional subspacetime and S 1 . The ansatz
can be chosen as [21]
AD
(n1) (x, z) = A(n1) (x) + A(n2) (x) dz,
(2.9)
where the coordinates x are on the (D 1)-dimensional spacetime and z is the coordinate on S 1 .
When we take the exterior derivative of (2.9) we get
D
(x, z) = dAD
F(n)
(n1) (x, z) = dA(n1) (x) + dA(n2) (x) dz.
(2.10)
While choosing the (D 1)-dimensional field strength of A(n2) as F(n1) = dA(n2) , one does
not simply choose the (D 1)-dimensional field strength of A(n1) as F(n) = dA(n1) but for
the purpose of obtaining a nice looking lower dimensional Lagrangian one defines the lower
dimensional field strengths through the transgression relations
F(n) (x) = dA(n1) (x) dA(n2) (x) A(x),
F(n1) (x) = dA(n2) (x).
(2.11)
In terms of these (D 1)-dimensional field strengths the D-dimensional field strength can be
given as

D
(x, z) = F(n) (x) + F(n1) (x) dz + A(x) .
F(n)
(2.12)
D in terms of the (D 1)-dimensional field
We can express the D-dimensional kinetic term of F(n)
strengths we have defined as
1
D
D
LD
F = F(n) F(n)
2

1 2(n1) (D1)
1 2(Dn1) (D1)
= e
F(n) F(n) e
F(n1) F(n1) dz
2
2
(D1)
= LF
dz.
(2.13)
If we perform the S 1 -reduction step by step on the ten-dimensional metric by using the ansatz
(2.3) and on the other fields in (2.1) as well as on the by-product lower dimensional fields by
123
using the ansatz (2.9) then the ten-dimensional Lagrangian (2.1) can be written as
L10 = LD dz1 dzn ,
(2.14)
where the S 1 -reduction is performed n times. The D-dimensional Lagrangian in (2.14) can be
calculated in terms of the D-dimensional fields as
1
1

LD = R 1 d d ea1 F(3) F(3)
2
2
1 a1i
1 a1ij
e
F(2)i F(2)i
e
F(1)ij F(1)ij
2
2
i
1
2
i<j
ec GI(2) GI(2)
1
2
eci GI(1)i GI(1)i
i,I
1 bi
1 bij
i
i
i
i
e
F(2)
F(2)
e
F(1)j
F(1)j
,
2
2
i
(2.15)
i<j
where i, j = 2, . . . , 11 D and = (1 , 2 , . . . , (11D) ) in which except 1 , the rest of the

scalars are the KaluzaKlein scalars which originate from the ten-dimensional spacetime metric.
The transgression relations which define the field strengths in (2.15) in terms of the potentials
are as follows

1 I
1 I
1 I
j
I
I
I
F(3) = dA(2) + B(1) dB(1) dA(1)i + B(0)i dB(1) + B(1) dB(0)i ji A(1)
2
2
2

1
j
I
I
i k
m
+ dA(0)ij B(0)[i
dB(0)j
] k A(1) m A(1) ,
2

j l

1 I
1 I
k
I
I
I
I
F(2)i = i dA(1)k + B(0)k dB(1) + B(1) dB(0)k + dA(0)kj B(0)[k dB(0)j ] l A(1) ,
2
2

j
l m
I
I
i
F(2)
= ji d k Ak(1) ,
F(1)ij = i j dA(0)lm B(0)[l dB(0)m] ,
j
I
I
GI(2) = dB(1)
dB(0)i
ji A(1) ,
I
GI(1)i = i dB(0)j
,
i
F(1)j
= jk dAi(0)k ,
(2.16)
where we have omitted the wedge product. The dilatons , the KaluzaKleinMaxwell poj
tentials A(1) as well as the axions Ai(0)k in each dimension are the descendants of the tendimensional spacetime metric. The potentials A(0)lm , A(1)k and A(2) are the KaluzaKlein
I
I are the
and B(1)
descendants of the two-form potential in D = 10. The potentials B(0)j
I as a result of the
D-dimensional remainders of the ten-dimensional YangMills potentials B(1)
i
ansatz (2.9) applied in each S 1 -reduction step. We define the matrix j as
i

ji = (1 + A(0) )1 j ,
(2.17)
and ji is the inverse of it. The dilaton vectors which couple to the scalars in various field
strength terms in (2.15) are defined as [22]
a1 = (1, 2s ),
bi = (0, fi ),

1
, s ,
c =
2
a1i = (1, fi 2s ),

bij = (0, fi + fj ),

1
ci =
, fi s .
2
a1ij = (1, fi + fj 2s ),
(2.18)
124
In [22] the scalar sector and the Abelian matter coupling in each dimension are reformulated
through a series of field and field strength redefinitions. The dualisation of certain fields is also
performed in various dimensions when necessary. In this way the global symmetry groups are
identified. Also the Lagrangian (2.15) is written in a more compact form whose scalar sector
is formulated as a G/K non-linear sigma model. We will not mention about the details of this
calculation here, we will only present the result which will be a reference point for our further
analysis in the following sections. In [22] it is shown that when a single dilaton is decoupled
from the rest of the scalars then the G/K coset representatives and the internal metric
M = T ,
(2.19)
generated by the remaining scalars are elements of O (10 D + N, 10 D)1 which is composed
of the (20 2D + N )-dimensional real matrices A which satisfy
AT A = ,
(2.20)
where the (20 2D + N ) (20 2D + N ) matrix is

0
0
1(10D)
,
=
0
1(N )
0
0
0
1(10D)
(2.21)
in which 1(n) is the n n unit matrix. The scalar Lagrangian of the D-dimensional compactified
theories can be described in the form

1
Lscalar = tr dM1 dM ,
(2.22)
4
with an additional decoupled dilatonic kinetic term after certain field redefinitions. The coset
representatives are parameterized as
1 i
H
= e 2
Ai(0)j Ei
ij
e 2 A(0)ij V eB(0)i UI .
(2.23)
j
{Hi , Ei , V ij , UIi }
are (20 2D + N )-dimensional matrices and following the notation of

Here
[22] we have modified the ranges of the indices i, j as i, j = 1, . . . , 10 D. We also assume that
j
i < j . The matrices {Hi , Ei , V ij , UIi } are [22,25]

ej i 0 0
i eii 0
0
ic
j

,
Ei =
Hi =
0
0
0
0 0 ,
0
0
0 i ci eii
0
0 eij

0 0 eij ej i
0 eiI 0
V ij = 0 0
(2.24)
,
UIi = 0 0 eI i ,
0
0 0
0
0 0
0
where the matrices are partitioned with appropriate dimensions which can be read from the nonzero entries and eab is a matrix with appropriate dimensions and which has zero entries, except
a {+1} entry at the ath row and the bth column. Notice that the indices i, j = 1, . . . , 10 D
and I = 1, . . . , N determine the dimensions of the matrices eab . As it is mentioned in [22] the
j
matrices {Hi , Ei , V ij , UIi } form up an algebra and they can be embedded in a fundamental rep
resentation of o (10 D + N, 10 D). As a matter of fact they generate the solvable Lie algebra
1 The reason why we use a prime will be clear later. Here we slightly change the notation used in [22].
125
of o (10 D + N, 10 D) in each dimension thus the parametrization of the coset in (2.23) is

nothing but an example of the solvable Lie algebra gauge. This result justifies the prediction that
the global internal symmetry group of the scalar Lagrangian is O (10 D + N, 10 D) and the
scalar manifold for the D-dimensional compactified theory with N gauge multiplet couplings
becomes
O (10 D + N, 10 D)
R.
O(10 D + N ) O(10 D)
(2.25)
The extra R factor arises since there is an additional dilaton which is decoupled from the rest of
the scalars in the scalar Lagrangian. The group O (10 D + N, 10 D) R is the global internal
symmetry of not only the scalar Lagrangian but the entire D-dimensional bosonic Lagrangian as
well. The Lagrangian (2.15) upon the above mentioned restoration, in terms of the newly defined
fields can be written as

1
1
d d + tr dM1 dM
2
4
1 8/(D2)
1
T
e
F(3) F(3) e 2/(D2) H(2)
MH(2) ,
2
2
LD = R 1
(2.26)
where
H(2) = dC(1) ,
(2.27)
in which we define
A
(1)i
I
C(1) = B(1)
,
(2.28)
Ai(1)
which is a column vector of dimension (20 2D + N ). The field strength F(3) in (2.26) can be
given in a compact form for any dimension as
1 T
F(3) = dA(2) + C(1)
(2.29)
dC(1) .
2
The definitions of the potentials introduced in (2.26) in terms of the original potentials which
come from the KaluzaKlein reduction in (2.15) can be referred in [22]. In this work, to derive
the field equations in the following sections we will consider the potentials used to construct
(2.26) as our starting point. In addition to the general D-dimensional Lagrangian (2.26) furthermore the D = 4 and the D = 3 cases can be studied separately, since they have global symmetry
enhancements over the general scheme of O (10 D + N, 10 D) R [22]. One may define additional scalars in these dimensions by dualizing certain fields by applying the Lagrange
multiplier methods. When the two-form potential A(2) is dualized with an additional axion in
D = 4, an axion-dilaton, SL(2, R) system is decoupled from the rest of the scalars in the scalar
Lagrangian and the enlarged D = 4 scalar manifold becomes
O (6 + N, 6)
SL(2, R)
.
O(6) O(6 + N )
O(2)
(2.30)
I ) can be dualized to give 7 + 7 +

On the other hand in D = 3, the bosonic fields (Ai(1) , A(1)i , B(1)
N additional axions. In this case the entire bosonic sector comes out to be composed of only the
126
scalars. The D = 3 enlarged scalar manifold then becomes

O (8 + N, 8)
.
O(8) O(8 + N )
(2.31)
We see that all of the global internal symmetry groups in (2.25), (2.30) and (2.31) apart from the
contributions of the decoupled scalars namely the groups O (10 D + N, 10 D), O (6 + N, 6),
O (8 + N, 8) are non-compact real forms of semi-simple Lie groups and they enable solvable
Lie algebra parametrizations of the cosets that are generated by the denominator groups O(10
D + N ) O(10 D), O(6) O(6 + N ), O(8) O(8 + N ), respectively.
3. The general formalism for the SSSM with couplings
In this section we will construct the general formulation of the Abelian gauge matter, dilaton
and ChernSimons couplings of the generic symmetric space sigma model (SSSM) [19,2629]
under the solvable Lie algebra parametrization. We will derive the field equations for a general
theory and then we will give a number of examples on which we can apply the results. Also in
the next section we will use the results to write down the field equations of the D-dimensional
low energy effective theory of the E8 E8 heterotic string which is studied in the previous
section. The general formalism and the field equations are already derived for the symmetric
space sigma model and the Abelian matter coupled symmetric space sigma model in [28,29] and
[23] respectively under the solvable Lie algebra gauge. Therefore in this section we extend the
construction of [23] further by including a dilaton and a ChernSimons coupling.
We will first construct the scalar Lagrangian by using the solvable Lie algebra parametrization
[17] for the coset representatives. In Section 2 we have seen that the toroidally compactified heterotic string gives the MaxwellEinstein supergravities in D-dimensions. The scalars decoupled
from a single dilaton are governed by symmetric space sigma models. The scalar fields parameterize the scalar coset manifold O (10 D + N, 10 D)/O(10 D + N ) O(10 D) where
O (10 D + N, 10 D) is in general a non-compact real form of a semi-simple Lie group and
O(10 D + N) O(10 D) is its maximal compact subgroup. For this reason the scalar manifold is a Riemannian globally symmetric space for all the O (10 D + N, 10 D)-invariant
Riemannian structures on it [18].
In this section we will use a slightly different notation. We will take the representation of the
global symmetry group as O(10 D, 10 D + N ) whose (20 2D + N )-dimensional real
matrix elements A satisfy the defining relation
AT A = ,
(3.1)
where
= diag(, , . . . , , +, +, . . . , +),
(3.2)
is the indefinite signature metric with 10 D minus signs and 10 D + N plus signs. Although
the set of matrices defined in (2.20) and (3.1) may differ the groups they form are isomorphic to
each other [30]. Thus in this section we will base our formulation on the scalar coset which is the
conventional one used in the construction of the MaxwellEinstein supergravities in [3133].
To construct the symmetric space sigma model Lagrangian for the scalar coset manifold,
O(10 D, 10 D + N )
,
O(10 D) O(10 D + N )
(3.3)
127
one may make use of the solvable Lie algebra parametrization [17] for the parametrization of the
coset representatives. The solvable Lie algebra gauge or the parametrization is a consequence of
the Iwasawa decomposition [18]
o(10 D, 10 D + N ) = k0 s0 = k0 hk nk ,
(3.4)
where k0 is the Lie algebra of O(10 D) O(10 D + N ) which is a maximal compact subgroup of O(10 D, 10 D + N ) and s0 is a solvable Lie subalgebra of o(10 D, 10 D + N ).
In (3.4) hk is a subalgebra of the Cartan subalgebra h0 of o(10 D, 10 D + N ) which generates
the maximal R-split torus in O(10 D, 10 D + N ) [18,27,29]. The nilpotent Lie subalgebra
nk of o(10 D, 10 D + N ) is generated by a subset {Em } of the positive root generators of
+
o(10 D, 10 D + N ) where m +
nc . The roots in nc are the non-compact roots with respect
to the Cartan involution induced by the Cartan decomposition [18,23,29]
o(10 D, 10 D + N ) = k0 u0 ,
(3.5)
where u0 is a vector subspace of o(10 D, 10 D + N ). We should remark that when the

Lie group O(10 D, 10 D + N ) is in split real form (maximally non-compact) [18] then
for a Cartan decomposition and a resulting Iwasawa decomposition the solvable Lie algebra s0
in (3.4) coincides with the Borel subalgebra which is generated by all the simple root Cartan
generators {H } (or any other basis) of h0 and the positive root generators {E | + } [19].
The split real form global symmetry groups are of the form O(n, n), O(n + 1, n), O(n, n + 1)
[21].
Since we have [21]
dim s0 = (10 D) (10 D + N ),
(3.6)
by coupling (10 D) (10 D + N ) scalars to the generators of the solvable Lie algebra s0
we can parameterize the representatives of the scalar coset manifold in the solvable Lie algebra
gauge as [18]
1 i
H
= e2
mE
(3.7)
where {Hi } for i = 1, . . . , dim(hk ) r are the generators of hk and {Em } for m +
nc are the
positive root generators which generate nk . The scalars { i } for i = 1, . . . , r are called the dilatons and { m } for m +
nc are called the axions. The coset representatives satisfy the defining
relation of O(10 D, 10 D + N ) namely
T = .
(3.8)
If we assume that we choose the (20 2D + N )-dimensional fundamental representation for the
algebra o(10 D, 10 D + N ) and if we define the internal metric
M = T ,
(3.9)
then the scalar Lagrangian which governs the (10 D) (10 D + N ) scalar fields of the theory
can be constructed as

1
Lscalar = tr dM1 dM .
(3.10)
4
Since in this work our main perspective is the derivation of the field equations of the toroidally
compactified heterotic string we take the scalar coset as O(10 D, 10 D + N )/O(10 D)
128
O(10 D + N ) however it is obvious that our analysis is valid for any scalar coset of the form
O(p, q)/O(p) O(q). As we have assumed a (20 2D + N )-dimensional representation for
the Lie algebra o(10 D, 10 D + N ) we can couple (20 2D + N ) Abelian gauge fields AI
and a dilaton to the scalars as [23,2629]
Lm = c2 e1 F MF = c2 e1 MI J F I F J ,
(3.11)
where the field strengths of the gauge fields are defined to be F I = dAI . Furthermore one can
couple a two-form field B to the other field content by introducing the ChernSimons three-form
G = dB + c4 I J AI F J ,
(3.12)
which becomes the field strength of B indeed. We are ready to write down the general Lagrangian
of the O(p, q)/O(p) O(q) scalar coset which has the Abelian gauge, ChernSimons and the
dilaton couplings in D-dimensional spacetime,
L = c1 d d + c2 e1 F MF +

1
tr dM1 dM
4
+ c3 e2 G G.
(3.13)
Our field content becomes {, AI , B, j , m } in which we have (10 D) (10 D + N ) scalars
of the coset, (20 2D + N ) Abelian gauge fields AI , a two-form B and a single dilaton which
we take to be decoupled from the rest of the scalars that parameterize the scalar coset manifold.
We will first vary the Lagrangian (3.13) with respect to the dilaton . The EulerLagrange
equation [34] which yields the vanishing of the variation of (3.13) with respect to gives
(1)(D1) d(2c1 d ) = c3 2 e2 G G + c2 1 e1 MI J F I F J .
(3.14)
The left-hand side comes from the variation of the kinetic term of the dilaton in (3.13) and the
terms in the right-hand side arise due to the coupling of the dilaton to the other fields. Since the
Lagrangian (3.13) does not explicitly depend on B the variation of it with respect to B will yield
the EulerLagrange equation which defines a closed form. It is

(1)(D+1) d c3 e2 2 dB + 2c4 I J AI F J
(3.15)
= 0.
By using the definition of the ChernSimons three-form given in (3.12) we can write (3.15) as

d e2 G = 0.
(3.16)
We will now vary the Lagrangian (3.13) with respect to the Abelian gauge fields AI . The Euler
Lagrange equations of motion read

0 = (1)3(D3) 2c3 c4 e2 I K F K dB + c4 LM AL F M

+ d c2 e1 MI J dAJ + MJ I dAJ

+ 2c3 c4 e2 dB + c4 LM AL F M J I AJ .
(3.17)
Bearing in mind that from its definition in (3.9) the internal metric M is symmetric so that
MI J = MJ I ,
(3.18)
also by using the definition of the ChernSimons form in (3.12) the EulerLagrange equations
(3.17) for the gauge fields AI can be written as
(1)3(D3) c3 c4 e2 I K F K G

= d c2 e1 MI J dAJ d c3 c4 e2 G J I AJ .
Now by acting the exterior derivative the last term in (3.19) can be written as

d c3 c4 e2 G J I AJ

= c3 c4 d e2 G J I AJ + (1)(D3) e2 G J I dAJ .
129
(3.19)
(3.20)
If we use the field equation (3.16) of the two-form field B above and insert the result back in
(3.19) we obtain
(1)3(D3) c3 c4 e2 I K F K G

= d c2 e1 MI J dAJ (1)(D3) c3 c4 e2 G J I dAJ .
(3.21)
Noting that is symmetric,

I J = J I ,
(3.22)
we can finally write the field equations of AI as

d c2 e1 MI J F J = (1)D 2c3 c4 e2 I J F J G,
(3.23)
which we will compactly express in matrix form as

d c2 e1 M F = (1)D 2c3 c4 e2 F G.
(3.24)
One can take the internal metric out of the exterior derivative on the left-hand side above to obtain

d c2 e1 F = c2 e1 M1 dM F + (1)D 2c3 c4 e2 M1 F G.
(3.25)
We may also express the field equations (3.23) in terms of the CartanMaurer form
G = d 1 .
(3.26)
Before going in that direction we should remark on certain identities which the general coset
representatives (3.7) satisfy. First of all since
1 = 1 = 1,
(3.27)
taking the exterior derivative of both sides yields the identities

d 1 = d 1 ,
d 1 = 1 d.
(3.28)
Also as is a symmetric matrix and

2 = 1,
(3.29)
starting from (3.8) one can derive the following identities

1 = T ,
T = ,
T = 1 ,

1
1
1
1 T
= ,
= T
,
T
= .
(3.30)
Since is a constant matrix one can take the exterior derivative of the above identities to obtain
1
d 1 = d T ,
d = d T
.
d T = d 1 ,
(3.31)
130
Due to the definition of the coset elements in (3.7) we also have

T 1 1 T
=
.
(3.32)
In order to write the field equations of AI in terms of the CartanMaurer form G starting from
(3.24) we can proceed as follows

d c2 e1 M F

= c2 e1 d T F + T c2 e1 d F + d c2 e1 F .
(3.33)
Inserting this result back in (3.24) we obtain

T

d c2 e1 F = c2 e1 1 d F c2 e1 1 d 1 F
1
+ (1)D 2c3 c4 e2 1 T
F G.
(3.34)
Now multiplying the above equation with on both sides and then using the identities (3.28)
(3.32) effectively we obtain
1

1

F + c2 e1 T d 1 T
F
d c2 e1 F = c2 e1 d T T
+ (1)D 2c3 c4 e2 MF G.
(3.35)
By using
1
1
= T GT T
,
d T T
(3.36)
finally we can express (3.24) in terms of the CartanMaurer form (3.26) as

d c2 e1 F
1

F + (1)D 2c3 c4 e2 MF G.
= c2 e1 T G T + G T
(3.37)
For the generic non-split scalar cosets the CartanMaurer form G is calculated explicitly in terms
of the coset scalars namely the dilatons { i } and the axions { m } in [29]. It reads

1
1
1
i
G = d i Hi + e 2 i U E = d i Hi + E d,
2
2
(3.38)
where {Hi } for i = 1, . . . , r are the generators of hk and {E } for +

nc are the generators
of nk as we have already defined before. We have introduced the column vector
U = d ,
(3.39)
and the row vector

( E ) = e 2 i E ,
i
where the components of the roots
(3.40)
+
nc
[Hi , E ] = i E .
are defined as
(3.41)
Also is a dim nk dim nk matrix

=

m=0

m
= e I 1 .
(m + 1)!
(3.42)
131
The components of the matrix are
= K .
(3.43)
The structure constants K are defined as
[E , E ] = K E .
(3.44)
In other words since

[E , E ] = N, E+ ,
(3.45)
we have
K
= N,
if + = ,
K
=0
if + = ,
= 0.
(3.46)
The reason why we have written the field equations of the gauge fields AI (3.24) in terms of
the CartanMaurer form G in (3.37) becomes clear now. After choosing a (20 2D + N )dimensional representation for the Lie algebra o(10 D, 10 D + N ) (thus for the solvable
Lie subalgebra s0 ) and after expressing the coset representatives (3.7) and the internal metric
(3.9) as matrices one can insert (3.38) in (3.37) to write the later explicitly in terms of the coset
scalars i and m .
The terms in the Lagrangian (3.13) which contain the coset scalars i and m are
1

L i , m = tr dM1 dM + c2 e1 F MF.
4
By using the results of [26,28,29] we can write (3.47) as

1
L i , m = tr G G T + G G + c2 e1 F MF.
2
Furthermore by using (3.38) we have
(3.47)
(3.48)
r

1
1 i i
d i d i
e
U U
L i , m =
2
2
+
i=1
+ c2 e
nc
F MF.
(3.49)
U
as independent fields and then use the Lagrange

Following the outline of [26,27] which treat
multiplier formalism the field equations of i and m arising from the variation of (3.49) have
been derived in [23]. They can be given as

1 i

1
1
1
i
i 1
i
d e 2 i U = j e 2 i d j U +
e 2 i e 2 i N, U U ,
2
=

1
1
1
j
j
d d i =
i e 2 j U e 2 j U
2
+
nc

N A
M
J
+ (1)D c2 e1 (Hi )A
N M J F F ,
(3.50)
A
, , +
nc and i = 1, . . . , dim(hk ) r. The matrices {(Hi )N } are the ones corresponding to
the generators {Hi } in the (20 2D + N )-dimensional representation chosen. By also referring
132
to [28] the field equations in (3.50) can more compactly be written as

i
i
d ei U =
N, U ei U ,
=
1

1
1
j
j
i e 2 j U e 2 j U + (1)D c2 e1 F T Hi F.
d d i =
2
+
(3.51)
nc
It is worth stating that the last term in the second equation above arises from the variation of
the matter Lagrangian (3.11) with respect to the dilaton i . A similar term is missing for the
axions m owing to the Lagrange multiplier method used in the derivation of the corresponding
field equations [26,28]. Therefore we have
(1)D c2 e1 F T Hi F = (1)D
Lm
.
i
(3.52)
From (3.11)
Lm (c2 e1 MI J F I F J ) (c2 e1 IA JA F I F J )
=
=
l
l
l

1
A A
A
A
I
J
l I J + I l J F F .
= c2 e
(3.53)
Since in the parametrization (3.7) the dilatons i are coupled to the Cartan generators Hi which
commute among themselves within the Cartan subalgebra structure of o(10 D, 10 D + N )
one can easily show that
1
l = Hl .
l
2
Thus we have

A
1
1
N
=
= (Hl )A
H
l IA = (l )A
l
I
N I .
2
2
I
Inserting this result back in (3.53) we obtain

1 A
Lm
1 1
A N A
A N
=
c
e
)
+
(H
)
(H
FI FJ.
2
l
l
N I J
N J
l
2
2 I
(3.54)
(3.55)
(3.56)
Now since both F I and F J are two-forms by using the identity

N A
I
J
A N A
J
I
(Hl )A
N I J F F = (Hl )N I J F F ,
(3.57)
and by arranging the indices above we conclude that

Lm
N
I
J
1
(3.58)
= c2 e1 IA (Hl )A
F T Hl F,
N J F F = c2 e
l
which justifies its existence in (3.51). In this section we have derived the set of Eqs. (3.14), (3.16),
(3.24) and (3.51) which are the field equations of the O(p, q)/O(p) O(q) symmetric space
sigma model coupled generally to a dilaton, a certain number of Abelian gauge fields and a twoform which is coupled through its ChernSimons field strength. We have shown that Eq. (3.24)
can be expressed more explicitly in terms of the coset scalars through the calculation of the
CartanMaurer form. Thus we have also derived a simplified set of equations namely Eq. (3.37)
as alternatives of (3.24). Eq. (3.50) are also written more compactly in (3.51). Before we link our
results to the D-dimensional toroidally compactified heterotic string we will give three examples
on which the results we have obtained are applicable.
133
3.1. The D = 7 case

The bosonic Lagrangian of the N = 2, D = 7 MaxwellEinstein supergravity is constructed
in [31] as

1
5
1
1
L = R 1 d d e2 G G + tr dM1 dM
2
8
2
4
1
e F M F,
2
(3.59)
where the coupling between the field strengths F I = dAI for I = 1, . . . , N + 3 and the scalars
which parameterize the coset SO(N, 3)/SO(N ) SO(3) can be explicitly written as
1
1
e F M F = e MI J F I F J .
(3.60)
2
2
In writing (3.59) we have assumed the (N + 3)-dimensional matrix representation of so(N, 3).
The ChernSimons form G is defined as [31]
1
G = dB I J AI F J ,
(3.61)
2
where = diag(, , , +, +, . . . , +) is the invariant tensor of SO(N, 3). If we compare (3.59)
with (3.13) we see that
5
c1 = ,
8
1 = 1,
1
c2 = c 3 = ,
2
2 = 2.
(3.62)
Also comparing (3.61) with (3.12) yields

1
c4 = .
(3.63)
2
Thus with the above identification of the coefficients (3.14), (3.16), (3.24) and (3.51) give us
the bosonic field equations of the N = 2, D = 7 MaxwellEinstein supergravity in a general
solvable Lie algebra parametrization of the coset SO(N, 3)/SO(N ) SO(3). We should state
that although this coset is different than the one we have assumed in (3.3) the formal, abstract
construction of the Lagrangian in the solvable Lie algebra gauge we have performed for (3.3) is
also valid for SO(N, 3)/SO(N ) SO(3).
3.2. The D = 8 case
The Lagrangian of the bosonic sector of the N = 1, D = 8 MaxwellEinstein supergravity is
given as [32]

1
3
1
1
L = R 1 + d d e2 G G + tr dM1 dM
4
8
2
4
1
e F M F.
2
The ChernSimons three-form G is
G = dB + I J F I AJ ,
(3.64)
(3.65)
134
with I, J = 1, . . . , N + 2. Apart from the single dilaton the rest of the scalars parameterize the coset manifold SO(N, 2)/SO(N ) SO(2) and the SO(N, 2) invariant tensor is =
diag(, , +, +, . . . , +). We again assume that we choose an (N + 2)-dimensional matrix representation of so(N, 2). Again if we compare (3.64) with (3.13) in this case we have
3
c1 = ,
8
1 = 1,
1
c2 = c 3 = ,
2
2 = 2.
(3.66)
Also from (3.65) and (3.12) we have

c4 = 1.
(3.67)
If the coefficients are chosen as above then (3.14), (3.16), (3.24) and (3.51) correspond to the
bosonic field equations of the N = 1, D = 8 MaxwellEinstein supergravity in a general solvable
Lie algebra parametrization of the coset SO(N, 2)/SO(N ) SO(2).
3.3. The D = 9 case
As a final example we will consider the N = 1, D = 9 MaxwellEinstein supergravity [33].
The bosonic Lagrangian of the N = 1, D = 9 MaxwellEinstein supergravity can be given
as [33]

1
1
7
1
L = R 1 + d d + e4 G G + tr dM1 dM
4
4
2
4
1 2
e
F M F.
2
(3.68)
In this case the scalars of the coupling Abelian vector multiplets parameterize the scalar coset
SO(N, 1)/SO(N ). The ChernSimons three-form is taken as
G = dB + I J AI F J ,
(3.69)
with I, J = 1, . . . , N + 1. We assume an (N + 1)-dimensional matrix representation of so(N, 1).

Similar to the D = 7 and D = 8 cases we have = diag(, +, +, . . . , +) which is the invariant
tensor of SO(N, 1). The comparison of (3.68) and (3.69) with (3.13) and (3.12) respectively
specifies the coefficients as
7
c1 = ,
4
1 = 2,
1
c2 = ,
2
2 = 4,
1
c3 = ,
2
c4 = 1.
(3.70)
Therefore again with the above identification of the coefficients (3.14), (3.16), (3.24) and (3.51)
become the bosonic field equations of the N = 1, D = 9 MaxwellEinstein supergravity for
a general solvable Lie algebra parametrization of the coset SO(N, 1)/SO(N ). We should remark
that the discussion we have made for the replacement of (3.3) with the coset SO(N, 3)/SO(N )
SO(3) of the D = 7 case is also valid for the D = 8 coset SO(N, 2)/SO(N ) SO(2) and the
D = 9 coset SO(N, 1)/SO(N ).
135
4. Field equations of the D-dimensional heterotic string

In Section 2 we have given the bosonic Lagrangian of the toroidally compactified E8 E8
low energy effective heterotic string. The KaluzaKlein reduction produces the scalar manifolds
which are in the form of (2.25). The scalar Lagrangian is written as (2.22) by using the explicit
representation (2.24) of the generators of the solvable Lie algebra of o (10 D + N, 10 D)
in the coset parametrization (2.23). As we have already mentioned before the importance of
this formulation is that it is based on the KaluzaKlein descendant fields. In Section 3 however
we have taken our scalar coset manifold explicitly as O(10 D, 10 D + N )/O(10 D)
O(10 D + N) whose definition originates from the indefinite signature metric (3.2). We have
to remark two points now. Firstly in the definition of the generalized orthogonal groups O(p, q)
the indefinite signature metric can be replaced by any symmetric matrix which has p positive and
q negative eigenvalues. The resulting group of matrices may differ from each other from their set
content point of view however their group structures are all isomorphic to each other [30]. Also
since Spin(p, q) is a double cover of O(p, q) and since it is isomorphic to Spin(q, p) [35] we
have the isomorphisms
O (10 D + N, 10 D) O (10 D, 10 D + N )
O(10 D + N, 10 D) O(10 D, 10 D + N ). (4.1)
Thus we observe that from the algebraic point of view the choice of the global symmetry group
representation among (4.1) which defines the scalar coset manifold is unimportant. However the
choice of O (10 D + N, 10 D) is distinct since it arises naturally during the KaluzaKlein
reduction in [22]. On the other hand the choice of the global symmetry group representation
effects the structure of the field equations via the indefinite metric which explicitly appears in
the equations. The second point we should emphasize on is that the properties of which we
have made use of to derive the bosonic field equations of the scalar coset with couplings in the
previous section are all shared by . First of all as it can be deduced directly from (2.21)
2 = 1.
(4.2)
Also since is a symmetric matrix and the coset representatives (2.23) of O (10 D + N, 10
D)/O(10 D + N) O(10 D) satisfy (2.20) the identities (3.27)(3.32) are shared by
and the coset representatives (2.23). Therefore we may conclude that we may replace by
in the formulation performed in the last section, thus our results can be extended to the
KaluzaKlein scalar coset representation of Section 2. In other words all the formulas of Section 3 are legitimately correct when one replaces by taking the scalar coset manifold as
O (10 D + N, 10 D)/O(10 D + N ) O(10 D). One final remark we should do before writing down the field equations of the heterotic string is about the coset parametrization.
Comparing the coset parametrizations (2.23) and (3.7) which are both based on the solvable Lie
algebra of the global symmetry group one sees that one needs a transformation to relate the fields
I } to the ones { i , m }. Thus one should consider the equality of the two
{ i , Ai(0)j , A(0)ij , B(0)i
coset parametrizations
1 i
H
e2
Ai(0)j Ei
ij
1 i
H
e 2 A(0)ij V eB(0)i UI = e 2
mE
(4.3)
bearing in mind that while the solvable Lie algebra generators on the left-hand side are fixed
by the representation chosen and by their embedding in o (10 D + N, 10 D) which is discussed in [22], the solvable Lie algebra generators on the right-hand side are completely arbitrary.
136
The transformation can be calculated in two ways; either after choosing the basis on both sides of
(4.3) the same, one considers local scalar fields whose ranges in the exponential generate solvable
Lie algebra elements in a sufficiently small neighborhood of the identity element so that (4.3) is
valid in any representation chosen for the algebra, thus giving the transformations explicitly as
a matrix equation2 or after determining another solvable Lie algebra s0 of o (10D +N, 10D)
through a selection of an Iwasawa decomposition on the right-hand side of (4.3), one may identify
the matrix representatives of any basis of this solvable Lie algebra {Hi , Em } in the representation generated by (2.24) then one can find the field transformations from (4.3) explicitly again.
Despite the locality of the first method by following the second method one may find global field
transformations which depend on the particular representation of Section 2. Therefore keeping
in mind that one can always find such transformations between the scalar field definitions of
the two different coset parametrizations as discussed above, in writing the field equations of the
D-dimensional compactified heterotic string we will assume a coset parametrization in the form
of (3.7) for the scalar coset manifold O (10 D + N, 10 D)/O(10 D + N ) O(10 D)
which is the one used in the previous section. Comparing the Lagrangians (2.26) and (3.13) we
firstly observe that
.
(4.4)
Then

1
c 1 = c2 = c3 = ,
2
2
1 =
,
(D 2)
2 =
8
.
(D 2)
(4.5)
I
F I H(2)
.
(4.6)
Also
,
G F(3) ,
I
C(1)
AI ,
B A(2) ,
Furthermore
1 T
1
I
J
F(3) = dA(2) + C(1)
dC(1) = dA(2) + I J C(1)
dC(1)
,
2
2
which is in the form of (3.12) with the identification of the coupling constant as
(4.7)
1
c4 = .
(4.8)
2
We may conclude that in the light of these identifications the formalism which appears in Section 2 is in parallel with the general one constructed in Section 3 differing only in the scalar coset
representation. Finally upon the discussion we have made about the coset parametrizations, if
we consider the field redefinitions and the substitutions mentioned above in the formalism of
the previous section a similar line of derivation would give us the bosonic field equations of the
D-dimensional toroidally compactified E8 E8 low energy effective heterotic string as

(1)D d(d) =
8
1
8/(D 2)e (D2) F(3) F(3)

2

2
1
(D2)
I
J
2/(D 2)e
+
MI J H(2)
H(2)
,
2
2 A similar discussion leads to the differential form of such scalar field transformations in [29].
137
8

d e (D2) F(3) = 0,

2

8
d e (D2) M H(2) = (1)D e (D2) H(2) F(3) ,

i
i
d ei U =
N, U ei U ,
=
1

1
1
j
j
i e 2 j U e 2 j U
d d i =
2
+
nc
(1)D e
2
where the coset representatives
1 i
H
= e2
mE
2
(D2)
H(2) T Hi H(2) ,
(4.9)
(4.10)
satisfy the defining relation of O (10 D + N, 10 D) namely

T = ,
(4.11)
and they parameterize the coset

O (10 D + N, 10 D)/O(10 D + N ) O(10 D),
(4.12)
as we have discussed in detail above. Before concluding we will discuss one more point. One
may investigate the relation between the two formulations considered in this section also in Section 2 and the one introduced in Section 3 which are constructed over different scalar coset
representations although they are two equivalent formulations of the same theory. This is a matter of interest since when we take a look at (4.7) and (4.9) we see that as is a non-diagonal
matrix there is a degree of mixing of the Abelian gauge fields and their field strengths in the
ChernSimons three-form (4.7). This mixing brings a certain complexity of coupling in the field
equations (4.9). However in the formulation of Section 3 this mixing is avoided since is a diagonal matrix. Thus the field equations obtained by using the scalar coset representation based
on involve less coupling which may be a simplification and an advantage in seeking solutions. One may find solutions in the -formulation and then pass to the KaluzaKlein field
content by using the field transformations upon the assumption that the Lagrangians of these
two distinct formulations are equal. This approach is different than the one used above which
derives the field equations of the D-dimensional heterotic string solely in the -formulation.
More specifically one would consider the equality of the two Lagrangians which are based on
the scalar coset manifolds of O (10 D + N, 10 D)/O(10 D + N ) O(10 D) and
O(10 D, 10 D + N )/O(10 D) O(10 D + N ) respectively.3 In this case one has to
calculate the field transformations between the constructions based on two different scalar coset
representations. We may calculate the transformations of the field contents by equating the two
Lagrangians (2.26) and (3.13) which are based on different scalar coset matrix structures. If we
do so we see that all the previous coupling constant and field equivalences mentioned in this
3 Note that although we have used the first coset structure for the D-dimensional heterotic string in Sections 2 and 3 and
the second one in our general formalism in Section 3 we have made use of the results of Section 3 for the D-dimensional
heterotic string by simply substituting instead of owing to the similar properties of these two metrics which take
part in the derivation of the field equations. However we have not mentioned relating these two formulations before.
138
section are valid except the ones for AI and the coset scalars since in this case we have
1
I
J
dC(1)
= c4 I J AI F J ,
I J C(1)
2
(4.13)
and
1
T
MH(2) = c2 e1 F MF.
e 2/(D2) H(2)
(4.14)
2
On the left-hand side of the above equation the coset parametrization is (2.23) and on the rightI and AI of the
hand side it is (3.7). If one assumes a transformation between the gauge fields C(1)
two formulations in the form
C(1) = T A,
(4.15)
where T is a (20 2D + N ) (20 2D + N ) matrix then by choosing c4 = 1/2 again one finds
that from (4.13) T must satisfy
T T T = .
(4.16)
Inserting (4.15) in (4.14) one also finds that

= T .
(4.17)
Since is a symmetric matrix and since any symmetric matrix can be diagonalized in the form
(4.16) with T O(20 2D + N ) [3638] one can find the transformation matrix T to relate the
two different set of gauge fields. After determining the solvable Lie algebra s0 of O(10 D, 10
D + N) through a selection of an Iwasawa decomposition (3.4) one may choose a (20 2D + N )dimensional matrix representation to express the basis {Hi , Em } and then from (4.17) one may
also find the transformation rules of the coset scalars between the two scalar coset formulations.
In this way one may solve the field equations in the -formulation which are less coupled and
then pass to the field content obtained by the direct KaluzaKlein reduction.
Although this method may look tempting we should also point out that the transformations
obtained in this way cannot cover the entire solution space thus they may be highly restrictive.
The reason for this lies in the process of equating the two Lagrangians (2.26) and (3.13) and obtaining (4.17). We have stated before that although the groups are isomorphic the matrix contents
of the sets O (10 D + N, 10 D) and O(10 D, 10 D + N ) may be quite different for
example
O (10 D + N, 10 D),
O(10 D, 10 D + N ),
/ O(10 D, 10 D + N ),
/ O (10 D + N, 10 D).
(4.18)
Thus the transformations (4.17) can exist only for the localized coset scalars which would lead to
the coset representatives that lie in O (10 D + N, 10 D) when the representatives in O(10
D, 10 D + N) are translated by T from the right. On the other hand due to the generalized
formulation of Section 3 one has the degree of freedom of defining { i , m } from a rich choice
of alternatives of the Iwasawa decomposition, the solvable Lie algebra, the basis {Hi , Em } and
the matrix representations of the Lie algebra o(10 D, 10 D + N ). Finally we may state
that to justify the transformations (4.17) one should focus on matching the coset representative
images of the solvable Lie algebra parametrizations of the two distinct scalar coset structures of
Sections 2 and 3 via the right-translation by T .
139
5. Conclusion
A review of the KaluzaKlein dimensional reduction [22] of the bosonic sector of the tendimensional N = 1 simple supergravity that is coupled to N Abelian gauge multiplets [8,9] on
the tori T 10D which describes the massless sector of the E8 E8 heterotic string theory [14]
is given in Section 2. Then we have derived the field equations of the general symmetric space
sigma model with dilaton, Abelian matter and ChernSimons couplings in Section 3. In Section 4
we have used the results of Section 3 to express the bosonic field equations of the D-dimensional
heterotic string whose bosonic Lagrangian is in parallel with the general Lagrangian we have
considered in Section 3. Besides as we have discussed by the end of Section 3 our general
formulation corresponds to the one that occurs when the MaxwellEinstein supergravities are
constructed by the use of the Noethers method [3133]. We have also mentioned about the possible transformations between the two different scalar coset parametrizations of Sections 2 and 3
which are both in the solvable Lie algebra gauge. Finally in Section 4 we have discussed an alternative way of deriving the field equations of the compactified heterotic string whose field content
directly arises from the D = 10 heterotic string via the KaluzaKlein ansatz [22]. By this method
one can express the field equations of the D-dimensional heterotic string exactly as in Section 3
whose scalar coset matrix representations are different from the ones used in Section 2. One can
then search for the field transformations which originate from the equality of the two distinct
Lagrangians of Sections 2 and 3. The limitations associated with this method are also pointed
out in Section 4.
In [23] the dynamics of the symmetric space sigma model with only Abelian gauge field couplings is studied under the solvable Lie algebra gauge. Therefore apart from deriving the bosonic
matter field equations of the D-dimensional E8 E8 heterotic string this work generalizes the
formulation of [23] to include further dilaton and ChernSimons couplings. Our formulation in
Section 3 is based on a generic global symmetry group O(p, q) and we derive the field equations
for a general solvable Lie algebra parametrization of the scalar coset without specifying neither
the basis used nor the representation chosen as in [23]. From this point of view the field equations we have obtained are extensive in their coverage and they are purely in algebraic terms.
On the other hand the major achievement of this work is the explicit derivation of the bosonic
matter field equations of the D-dimensional toroidally compactified E8 E8 heterotic string
for a generic and unspecified solvable Lie algebra parametrization of the scalar coset. Although
the bosonic Lagrangian constructed in [22] under the solvable Lie algebra gauge assumes a certain representation of the coset generators we have obtained the bosonic matter field equations
of the D-dimensional heterotic string in a representation free formalism. Therefore for the field
equations obtained in Section 4 we have a degree of freedom of choosing an appropriate representation of the solvable Lie algebra. This would provide a useful machinery for seeking solutions
of the theory. As a matter of fact one can inspect various representations to simplify the field
equations bearing in mind that choosing a basis for the solvable Lie algebra also corresponds
to defining the coset scalar fields. Appropriate representations would decrease the complexity
of the field equations. When one finds a suitable representation step by step one can construct
the explicit equations of motion which would involve less couplings. If the field equations in
Sections 3 and 4 are inspected carefully one can see that the most essential coupling is between
the coset scalars and the gauge fields beside the non-linearity of the structure of the coset scalars
themselves. Thus focusing on these couplings we have also derived various versions of the corresponding field equations in Section 3 which may provide a set of tools when certain solutions
of these equations are studied.
140
In Section 4 we have discussed that the formulation of Section 3 which is based on the
invariant metric is advantageous over the formulation descending from the KaluzaKlein compactification so that is a diagonal matrix and it prevents cross couplings of the gauge fields in the
field equations. Since our aim in this work is to formulate the field equations rather than studying
the solutions we have not given specific examples for which this simplicity may be exploited.
However we may state that the -formulation brings simplicity in deriving the first-order field
equations of the MaxwellEinstein supergravities and in constructing the non-linear realizations
of these theories [3941]. For the -formulation from [3941] we observe that the non-mixing
of the gauge fields in the field equations is reflected in the coset algebra of the non-linear sigma
model construction of the theory such that the operators which correspond to different gauge
fields do not mix under the algebra product (they anti-commute with each other).
Although we have mentioned about the relation between the two different solvable Lie algebra
parametrizations of the scalar coset in Section 4 we have not shown any attempt to construct
the transformations explicitly. As we have discussed one can search for the local or the global
scalar field transformations which would make use of the group theoretical structure of the global
symmetry group. One can also inspect the field transformations mentioned in the last part of
Section 4 which relate the field contents of the two equivalent formulations of the theory based
on two distinct scalar coset representations. As we have stated before this would remove some
complexity of the field equations. Therefore one can workout the solutions in the formulation of
Section 3 and then transform the fields to the original KaluzaKlein field content of Section 2.
The formalism of Section 3 can be extended for theories which posses different global symmetry groups than O(p, q) but which still can pertain the solvable Lie algebra gauge. In this case
the scalar coset G/K must be such that again the global symmetry group G must be a real form
of a non-compact semi-simple Lie group and K must be its maximal compact subgroup to be
able to apply the solvable Lie algebra gauge to parameterize the scalar coset manifold [18]. However when defining the internal metric one must use a generalized transpose which is induced by
the Cartan involution of the Lie algebra g [18,23,2629] instead of the ordinary matrix transpose.
Since the supergravity theories give way to first-order formulations [19,20,42] the first-order formulation of the field equations can be performed for both the general formalism of Section 3 and
the D-dimensional compactified heterotic string. Finally we should state that although owing
to the non-linear structure of the coset scalars also the complexity of the coupling between the
scalars and the gauge fields we have focussed on the bosonic matter sector in this work one can
extend the solvable Lie algebra gauge formalism developed here to derive the field equations in
the presence of the graviton and the fermionic sectors too.
References
[1] E. Cremmer, B. Julia, J. Scherk, Supergravity theory in eleven-dimensions, Phys. Lett. B 76 (1978) 409.
[2] I.C.G. Campbell, P.C. West, N = 2, d = 10 non-chiral supergravity and its spontaneous compactification, Nucl.
Phys. B 243 (1984) 112.
[3] M. Huq, M.A. Namazie, KaluzaKlein supergravity in ten-dimensions, Class. Quantum Grav. 2 (1985) 293.
[4] F. Giani, M. Pernici, N = 2 supergravity in ten-dimensions, Phys. Rev. D 30 (1984) 325.
[5] J.H. Schwarz, P.C. West, Symmetries and transformations of chiral N = 2, d = 10 supergravity, Phys. Lett. B 126
(1983) 301.
[6] P.S. Howe, P.C. West, The complete N = 2, d = 10 supergravity, Nucl. Phys. B 238 (1984) 181.
[7] J.H. Schwarz, Covariant field equations of chiral N = 2, d = 10 supergravity, Nucl. Phys. B 226 (1983) 269.
[8] E. Bergshoeff, M. de Roo, B. de Wit, P. van Nieuwenhuizen, Ten-dimensional MaxwellEinstein supergravity, its
currents, and the issue of its auxiliary fields, Nucl. Phys. B 195 (1982) 97.
141
[9] G.F. Chapline, N.S. Manton, Unification of YangMills theory and supergravity in ten-dimensions, Phys. Lett. B 120
(1983) 105.
[10] P.C. West, Introduction to Supersymmetry and Supergravity, second ed., World Scientific, 1990.
[11] A. Salam, E. Sezgin (Eds.), Supergravities in Diverse Dimensions, World Scientific, 1989.
[12] P.C. West, Supergravity, brane dynamics and string duality, hep-th/9811101.
[13] Y. Tanii, Introduction to supergravities in diverse dimensions, hep-th/9802138.
[14] E. Kiritsis, Introduction to superstring theory, hep-th/9709062.
[15] C.M. Hull, P.K. Townsend, Unity of superstring dualities, Nucl. Phys. B 438 (1995) 109, hep-th/9410167.
[16] E. Witten, String theory dynamics in various dimensions, Nucl. Phys. B 443 (1995) 85, hep-th/9503124.
[17] L. Andrianopoli, R. DAuria, S. Ferrara, P. Fre, M. Trigiante, RR scalars, U-duality and solvable Lie algebras, Nucl.
Phys. B 496 (1997) 617, hep-th/9611014.
[18] S. Helgason, Differential Geometry, Lie Groups and Symmetric Spaces, Graduate Studies in Mathematics, vol. 34,
American Mathematical Society, Providence, 2001.
[19] E. Cremmer, B. Julia, H. L, C.N. Pope, Dualisation of dualities. I., Nucl. Phys. B 523 (1998) 73, hep-th/9710119.
[20] E. Cremmer, B. Julia, H. L, C.N. Pope, Dualisation of dualities II: Twisted self-duality of doubled fields and
superdualities, Nucl. Phys. B 535 (1998) 242, hep-th/9806106.
[21] C.N. Pope, Lecture notes on KaluzaKlein theory, unpublished.
[22] H. L, C.N. Pope, K.S. Stelle, M-theory/heterotic duality: A KaluzaKlein perspective, Nucl. Phys. B 548 (1999)
87, hep-th/9810159.
[23] T. Dereli, N.T. Ylmaz, Dualisation of the symmetric space sigma model with couplings, Nucl. Phys. B 705 (2005)
60, hep-th/0507007.
[24] H. L, C.N. Pope, p-brane solitons in maximal supergravities, Nucl. Phys. B 465 (1996) 127, hep-th/9512012.
[25] E. Bergshoeff, D.C. Jong, E. Sezgin, Noncompact gaugings, chiral reduction and dual sigma models in supergravity,
Class. Quantum Grav. 23 (2006) 2803, hep-th/0509203.
[26] A. Keurentjes, The group theory of oxidation, Nucl. Phys. B 658 (2003) 303, hep-th/0210178.
[27] A. Keurentjes, The group theory of oxidation II: Cosets of non-split groups, Nucl. Phys. B 658 (2003) 348, hepth/0212024.
[28] N.T. Ylmaz, Dualisation of the general scalar coset in supergravity theories, Nucl. Phys. B 664 (2003) 357, hepth/0301236.
[29] N.T. Ylmaz, The non-split scalar coset in supergravity theories, Nucl. Phys. B 675 (2003) 122, hep-th/0407006.
[30] A.W. Knapp, Lie Groups Beyond an Introduction, Progress in Mathematics, vol. 140, second ed., Birkhauser,
Boston, 2002.
[31] E. Bergshoeff, I.G. Koh, E. Sezgin, YangMills/Einstein supergravity in seven dimensions, Phys. Rev. D 32 (1985)
1353.
[32] A. Salam, E. Sezgin, D = 8 supergravity: Matter couplings, gauging and Minkowski compactification, Phys. Lett.
B 154 (1985) 37.
[33] S.J. Gates, H. Nishino, E. Sezgin, Supergravity in d = 9 and its coupling to non-compact model, Class. Quantum
Grav. 3 (1986) 21.
[34] W.E. Thirring, A Course in Mathematical Physics, I and II: Classical Dynamical Systems and Classical Field Theory, Springer-Verlag, Berlin, 1991.
[35] P. Lounesto, Clifford Algebras and Spinors, London Mathematical Society Lecture Note Series, vol. 286, second ed.,
Cambridge Univ. Press, 2001.
[36] K.M. Hoffman, R.A. Kunze, Linear Algebra, second ed., Prentice Hall, 1971.
[37] B. Schutz, Geometrical Methods of Mathematical Physics, Cambridge Univ. Press, 1980.
[38] J.F. Cornwell, Group Theory in Physics: An Introduction, Academic Press, 1997.
[39] N.T. Ylmaz, Dualisation of the d = 7 heterotic string, JHEP 0409 (2004) 003, hep-th/0507008.
[40] T. Dereli, N.T. Ylmaz, Dualisation of the SalamSezgin d = 8 supergravity, Nucl. Phys. B 691 (2004) 223, hepth/0407004.
[41] N.T. Ylmaz, Dualisation of the d = 9 matter coupled supergravity, JHEP 0506 (2005) 031, hep-th/0507011.
[42] B. Julia, Superdualities: Below and beyond U-duality, hep-th/0002035.
Type-II seesaw mass models and baryon asymmetry

Amal Kr. Sarma a,b, , H. Zeen Devi a , N. Nimai Singh a
a Department of Physics, Gauhati University, Guwahati 781014, India
b Department of Physics, D.R. College, Golaghat 785621, India
Received 13 September 2006; received in revised form 24 October 2006; accepted 4 December 2006
Abstract
We compute and also compare the contributions of canonical and non-canonical mass terms towards
baryon asymmetry by considering type-II seesaw mass models of neutrinos: degenerate (3 varieties), normal
hierarchical and inverted hierarchical (2 varieties). We have shown that for particular choices of parameter
(the so-called discriminator) for different neutrino mass models, the baryon asymmetry is largely dominated by canonical term. Within such type-II seesaw scenario, we find normal hierarchical (NHT3) mass
model as the most favourable choice of nature.
PACS: 14.60.Pq; 11.30.Er; 11.30.Fs; 13.35.Hb
Keywords: Seesaw neutrino mass; Yukawa coupling; Lepton and baryon asymmetry
1. Introduction
The search for a suitable mass model consistent with the observations: masses, mixings and
baryon asymmetry, is a standing agenda of present day neutrino physics. With this motivation different neutrino mass models [1,2] were constructed on the basis of celebrated seesaw mechanism
[3]. All the models were checked for masses, mixings and stability under radiative corrections.
In the electroweak baryogenesis scenario [4,5], the heavy right-handed Majorana neutrinos play
a decisive role. In this model of baryogenesis, lepton asymmetry is produced by the decay of
lightest of heavy Majorana neutrinos, and non-perturbative sphaleron [6] processes convert this
lepton asymmetry to baryon asymmetry. Again the heavy Majorana neutrinos are coupled to light
E-mail address: amalk_s@yahoo.co.in (A.Kr. Sarma).

doi:10.1016/j.nuclphysb.2006.12.002
A.Kr. Sarma et al. / Nuclear Physics B 765 (2007) 142153
143
left-handed neutrinos via seesaw mechanism. In that sense any mass model constructed on the
basis of seesaw mechanism should have the capacity to explain the observed baryon asymmetry.
Sakharovs three conditions [7] for baryogenesis: (1) baryon number violation, (2) C and CP
violation and (3) out of equilibrium decay can be realised by the decay of heavy right-handed
Majorana neutrinos. Recently a non-thermal scenario of baryogenesis for these models was considered [8], where our estimation [9] of baryon asymmetry for different mass models within the
thermal leptogenesis scenario was extensively considered. One point missed in Ref. [9] is that
all the calculations to estimate the baryon asymmetry were confined to type-I seesaw term only,
the contributions of non-canonical term were not considered. For completeness we compute the
contribution of type-II term also.
We address in this paper, the estimation of baryon asymmetry of the universe using the parameters which are already fixed at the seesaw stage within a common framework. This may
possibly discriminate the correct pattern of neutrino mass model in question.
In Section 2 we review the discrimination of neutrino mass models in type-II seesaw framework. In Section 3, with a brief discussion on the expressions for lepton and baryon asymmetry,
we present our numerical calculations and results. Section 4 concludes with a summary and
discussion. The zeroth-order mass matrices for various models are collected in Appendix A. Important expressions related to mILL and MRR for three neutrino mass models are relegated to
Appendix B.
2. Type-II seesaw mechanism and discrimination of neutrino mass models
The type-I seesaw formula relates the light left-handed Majorana neutrino mass matrix mILL
and heavy right-handed Majorana mass matrix MRR in a simple way:
1 T
mILL = mLR MRR
mLR .
(1)
Here, mLR is the Dirac neutrino mass matrix. In some LR symmetric theories such as
SO(10) GUT, the left-handed Higgs triplet L acquires vacuum expectation value. Taking
this contribution, seesaw formula can be modified as [10]: mLL = mILL + mII
LL where, the
first term is the usual canonical seesaw term (1) and the second term can be expressed as
2
mII
LL = (MW /vR ) MRR . Such non-canonical seesaw formula (generally known as type-II seesaw formula) can be written as:
1 T
mLL = mLR MRR
mLR + (MW /vR )2 MRR .
(2)
In the light of above type-II seesaw formula, the neutrino mass matrices mLL are realised for
I
II
I
II
I
three different situations: (1) mII
LL mLL , (2) mLL mLL and (3) mLL mLL . It has been
reported that the inequality (1) can naturally provide a connection [11] between the large atmospheric mixing and b unification in the context of the minimal supersymmetric SO(10)
theory [12]. But we are interested to study the relative strength of these two contributions in all
three cases. The second term denoted by mII
LL is heavily constrained by the definition of vR ,
and vR can be extracted from MRR appeared in mILL , thus indicating the dependence of mII
LL on
mILL . This point is addressed in the present work.
The solar and atmospheric neutrino oscillation experiments usually measure only the masssquare differences and in general we have three possible patterns of neutrino masses [13]:
(a) degenerate m1 m2 m3 , (b) inverted hierarchical m1 m2 m3 and (c) normal hierarchical m1 m2 m3 . Depending upon the relative CP-phase of (m1 , m2 ) pair, the degenerate and
inverted hierarchical models are further classified as: degenerate type-1A (DegT1A), degenerate
144
type-1B (DegT1B), degenerate type-1C (DegT1C), inverted hierarchical type-2A (InvT2A) and
inverted hierarchical type-2B (InvT2B), and these zeroth-order left-handed Majorana mass matrices [2,13] are collected in Appendix A for ready reference. The present forms [1,2] of various
mass models given in Appendix B, are derived from seesaw mechanism (1).
The vacuum expectation values vL and vR of left- and right-handed Higgs triplets are con2 where the free parameter depends on various
nected [14] to W -boson mass via vL vR = MW
couplings. Without fine tuning, we generally have 1, but we are searching the input values of
where the contribution of second term arising from the presence of left-handed Higgs triplets
is just sufficient to restore the good predictions of mILL already acquired [2] in type-I seesaw
framework. If the second term is dominant, the relevant good predictions of mILL are spoiled.
This can be avoided with the proper choice of values. We call this parameter as discriminator of the models. A neutrino mass model is said to be favourable and hence stable when its
canonical term dominates over the non-canonical (perturbative) term, and this condition is used
here as a criterion for discriminating neutrino mass models [2].
The Dirac neutrino mass matrix mLR appeared in seesaw formula can have any arbitrary
structure which either is diagonal or non-diagonal, and it plays a significant role [1] in the
construction of mILL . In the seesaw mechanism, for a specific structure of mILL , we can have
three possible combinations of mLR and MRR : (a) both mLR and MRR are non-diagonal, (b)
mLR diagonal and MRR non-diagonal, (c) mLR non-diagonal and MRR diagonal. These three
combinations can be realised in different physical situations. For example, when one calculate
lepton asymmetry, one needs to consider the diagonal basis of heavy right-handed neutrinos,
and combination (c) becomes relevant. In the present calculations, the diagonal form of mLR is
chosen for different neutrino mass matrices. To see this let us consider [2] the seesaw relation
mILL = mLR MR1 mTLR , where both mLR and MR are non-diagonal. Using some left- and rightdiag
handed rotations, the Dirac neutrino mass matrix can be diagonalised [15] as mLR = UL mLR UR .
In terms of diagonal basis of mLR , the seesaw relation reduces to
1
mLR UL ,
mILL = UL mLR MRR
diag
diag
(3)
1
= UR MR1 URT . It is assumed that eigenvalues of mLR are hierarchical (similar to
where MRR
quarks or charged leptons). In absence of Dirac left-handed rotations [15], we can set UL 1.
For slight deviation from unity we can assume UL UCKM , where UCKM is the quark mixing
matrix. Again this can be set to unity as quarks mixings are very small. This type of approximations do not produce significant change in numerical calculations. For UL 1 the Eq. (3)
diag 1 diag
mLR where MRR is in the diagonal basis of mLR . We follow this
reduces to mILL = mLR MRR
representation in the present calculation.
In some grand unified theory such as SO(10) GUT, the possible structure [16] of mLR =
diag(m , n , 1)v, where v is the overall scale factor representing electroweak vacuum expectation values. In the present calculation we take = 0.3 and v = 174 GeV. We consider two
choices of (m, n) pair: case (i) (m, n) (6, 2) for charged lepton and (ii) (m, n) (8, 4) for
up-quark mass matrices representing the Dirac neutrino mass matrix.
As a representative example, we consider normal hierarchical mass model and the corresponding heavy right-handed neutrino mass matrix and light left-handed neutrino mass matrices
are [1,2] collected from Appendix B:
diag
Normal hierarchical type-3 (NHT3):

2m1
m+n1 m1
MRR = m+n1 m+n2
v0 ,
0
0
1
m1

mILL =
3
1
1
1 1 3
145
(4)

(5)
m0 ,
where we take the input values = 0.3, m0 = 0.03 eV and v0 = 1.01 1015 GeV.
The numerical calculations are carried out for two choices of diagonal structure of Dirac
neutrino mass matrix: case (i) Dirac neutrino mass matrix taken as charged lepton mass matrix
and case (ii) Dirac neutrino mass matrix taken as up-quark mass matrix. Within the type-I seesaw
framework, the left-handed neutrino mass matrix mILL leads to correct neutrino mass parameters
and mixing angles consistent with recent data [1,2]:
m221 = 9.04 105 eV2 ,
tan2 12 = 0.55,
m223 = 3.01 103 eV2 ,
sin2 223 = 0.98,
sin 13 = 0.074.
In type-II seesaw scenario, one has to take care of the non-canonical term mII
LL . In order to
maintain the existing good predictions of mILL , one has to use the freedom in choosing . We
make a search programme [2] for finding the value of the minimum departure of from the
canonical value of one, i.e., < 1.0, in which the good prediction of neutrino masses and mixing
parameters can again be restored in mLL . In other words, such least value of is just enough
to suppress the perturbative effect due to type-II seesaw formula. The value of = 0.007 is
extracted for NHT3 mass model. The values of in case (i) and case (ii) for other neutrino
mass models given in Appendix B, are presented in Tables 2, 3, along with the corresponding
predictions of neutrino oscillation parameters. Higher the values of , more the stability of the
neutrino mass model in question. Following the results of calculation in Tables 13. It has been
observed that InvT2A model with = 0.1 gives most favourable choice of nature, followed
by NHT3 and InvT2B with 102 . However, InvT2A model appears to be unstable under
radiative corrections in MSSM. Now we are left with NHT3 and InvT2B as favourable candidates
which are generally stable under radiative corrections in MSSM.
For further discrimination of the neutrino mass models, particularly between NHT3 and
InvT2B, we now consider baryon asymmetry estimation within the type-II seesaw formalism
as an extra criteria and this will be carried out in the following section as the main thrust of the
present investigation.
3. Type-II seesaw mechanism and the estimation of baryon asymmetry
For our calculation of lepton asymmetry we consider the model [5] where the asymmetric
decay of the lightest of the heavy right-handed Majorana neutrinos, is assumed. In this model the
physical Majorana neutrino NR can decay into two modes:
NR lL + lL + ,
where lL is the lepton and lL is the antilepton. For CP-violating decay through the one-loop
radiative correction by Higgs particle, the branching ratio for these two decay modes is likely to
146
Table 1
Predicted values of the solar and atmospheric neutrino mass-squared differences and three mixing parameters calculated
from mILL derived from type-I seesaw formula in Appendix B
Type
m221 [105 eV2 ]
m223 [103 eV2 ]
tan2 12
sin2 223
sin 13
DegT1A
DegT1B
DegT1C
InvT2A
InvT2B
NHT3
8.80
7.91
7.91
8.36
9.30
9.04
2.83
2.50
2.50
2.50
2.50
3.01
0.98
0.27
0.27
0.44
0.98
0.55
1.0
1.0
1.0
1.0
1.0
0.98
0.0
0.0
0.0
0.0
0.0
0.074
Table 2
Predicted values of the solar and atmospheric neutrino mass-squared differences and three mixing parameters along with
the values of , extracted from mLL in type-II seesaw formula for case (i)
Type
m221 [105 eV2 ]
m223 [103 eV2 ]
tan2 12
sin2 223
sin 13
DegT1A
DegT1B
DegT1CC
InvT2A
InvT2B
NHT3
105
8.45
7.97
7.93
8.20
9.40
9.41
2.73
2.30
2.47
2.50
2.40
2.98
0.98
0.28
0.27
0.49
0.98
0.54
1.00
1.00
1.00
1.00
1.00
0.98
0.0
0.0
0.0
0.0
0.01
0.09
104
105
0.1
0.009
0.007
Table 3
Predicted values of solar and atmospheric neutrino mass-squared differences, and three mixing parameters along with
the values of , extracted from mLL in type-II seesaw formula for case (ii)
Type
m221 [105 eV2 ]
m223 [103 eV2 ]
tan2 12
sin2 223
sin 13
DegT1A
DegT1B
DegT1CC
InvT2A
InvT2B
NHT3
105
8.56
7.69
7.69
8.3
9.40
9.18
2.74
2.30
2.54
2.5
2.40
2.80
0.98
0.27
0.29
0.47
0.98
0.55
1.00
1.00
1.00
1.00
1.00
0.98
0.0
0.0
0.0
0.0
0.0
0.07
104
105
0.1
0.02
0.007
be different. The CP-asymmetry which is caused by the interference of tree level with one-loop
corrections for the decays of lightest of heavy right-handed Majorana neutrino N1 is defined as
[5,17]
=

,
+
and = (N1 lL ) are the decay rates. Considering hierarchical

where = (N1 lL )
structure of heavy Majorana neutrinos a perturbative calculation from the interference between
tree level and vertex plus self-energy diagrams, gives [18] the lepton asymmetry I in terms of
light neutrino mass matrix mILL as
I =
3M1 Im[(h mILL h )11 ]

.
(hh )11
16v 2
(6)
147
This accounts for the type-I contribution. Again due to the presence of Higgs triplet as a virtual
particle in the decay of N1 , we have the additional contribution [18] (type-II contribution) to
lepton asymmetry II as
II =
3M1 Im[(h mII

LL h )11 ]
.
16v 2
(hh )11
(7)
For quasi-degenerate spectrum i.e., for M1 M2 < M3 the asymmetry is largely affected by a
resonance factor [19,20] R = |M1 |/2(|M2 | |M1 |). In such situation, the lepton asymmetry is
modified [19,20] to
=
M2 Im[(h mLL h )11 ]

R.
(hh )11
8v 2
(8)
The total asymmetry 1 is the sum of I and II .

The sphaleron process convert [4,21] this lepton asymmetry to baryon asymmetry. The baryon
asymmetry YBSM is expressed in terms of the dilution factor 1 and lepton asymmetry 1 as [22]

YBSM
nB
n
SM
1.08 102 1 1 .
(9)
For our calculation of baryon asymmetry we use the above expression.

How much the produced asymmetry is washed out is described by Boltzmann equation and
the solution can be described by a parameter 1 known as dilution factor [14,23]
K(ln0.3
if 10 K 106 ,
K)0.6
1
(10)
12
if 0 K 10,
2
K +9
where K = m
1 /m with m
1 = (hhM)111 v is the effective neutrino mass, v the electroweak
scale, M1 the mass of N1 , h the matrix for the neutrino Yukawa couplings. And m =
5
2
16
3 5
g2 Mv pl 1.08 103 eV is known as equilibrium neutrino mass [24] with g = 106.75,
1 < m , the heavy right-handed neutrino will satisfy the

Mpl = 1.2 1019 GeV. Thus, for m
out-of-equilibrium condition. For our calculation we consider this condition also.
3.1. Numerical calculation and results
diag
For our calculation of lepton asymmetry, we choose a basis UR where MRR = URT MRR UR =
diag(M1 , M2 , M3 ) with real and positive eigenvalues [15,25]. We transform mLR = diag(m ,
n , 1)v to the UR basis by mLR m
LR UR . In the prime basis the Yukawa coupling becomes
h = (mLR UR )/v. For demonstration as an example, we consider here the normal hierarchical
mass model (NHT3) described in Section 2. The corresponding heavy right-handed neutrino
mass matrix and light left-handed neutrino mass matrices are given in Eqs. (4) and (5).
For case (i): charged lepton mass matrix, mLR = mE = diag(6 , 2 , 1)v, (m, n) = (6, 2), v0 =
1.01 1015 GeV, m0 = 0.03 eV, = 0.007. For the above input values the three heavy masses
diag
are MRR = diag(4.28 109 , 1.16 1010 , 3.84 1013 ) in GeV. From the construction of h,
we have (hh )11 = 5.31 107 , Im(h mILL h )11 = 2.45 1018 , Im(h mII
LL h )11 = 2.52
24
I
10 , K = 0.23 and 1 = 0.17. These leads to lepton asymmetry = 5.92 107 and II =
148
6.09 1013 resulting 1 = I + II = 5.92 107 . The corresponding baryon asymmetry is

YB = 1.08 102 1 1 = 1.08 109 .
For all other models including the normal hierarchy case (ii) we follow the same procedure
to calculate the lepton and baryon asymmetry. For case (ii) the Dirac neutrino mass matrix is
taken as the up-quark mass matrix i.e., mLR = diag(8 , 4 , 1)v. As quasi-degenerate pattern of
heavy masses appears in inverted hierarchy type-2B (InvT2B), so for this model we consider the
resonance enhancement factor R, and a different expression [19] YB = 2.16 102 1 1 is used
for the estimation of baryon asymmetry. The estimated values of lepton and baryon asymmetry
for all the models in Appendix B are collected in Tables 68. As emphasised earlier, we use
the only parameters fixed at seesaw stage in Section 2 for estimating baryon asymmetry as an
additional criteria for discriminating neutrino mass models.
4. Summary and discussion
To summarise, we have considered various neutrino mass models: degenerate models
(DegT1A, DegT1B, DegT1C), inverted hierarchical (InvT2A, InvT2B) and normal hierarchical
( NHT3). We have estimated the contributions of type-I (canonical) and type-II (non-canonical)
seesaw mass term towards baryogenesis. Following the predictions of SO(10) grand unified theory, we estimate the baryon asymmetry for two choices of Dirac neutrino mass matrix: case (i)
Dirac neutrino mass matrix taken as charged lepton mass matrix and case (ii) Dirac neutrino
mass matrix taken as up quark mass matrix. For specific values of the discriminator extracted
from type-II seesaw formula [2], we observe that the baryon asymmetry is dominated by type-I
term only.
All the eigenvalues of MRR are collected in Table 4, and the mass parameter m
1 , dilution
factor 1 are presented in Table 5. From Table 5 it is clear that only inverted hierarchical model
type-2B (InvT2B) and normal hierarchical model type-3 (NHT3) satisfy the out-of-equilibrium
decay condition. In these two models, the effective neutrino mass is less than the equilibrium
neutrino mass m ( 1.08 103 ) i.e., m
1 < m .
In Tables 68 we present the estimations of lepton and baryon asymmetry. We present the
numerical predictions of the neutrino mass-squared differences and three mixing angles in type-I
and type-II seesaw framework [2] in Tables 13. Only two models i.e., inverted hierarchical
model type-2A (InvT2A) and normal hierarchical model type-3 (NRT3) satisfy all the neutrino
oscillation parameters (Tables 13). However, the inverted hierarchical type-2A (InvT2A) model
is not stable under radiative corrections in MSSM, whereas normal hierarchical type-3 (NHT3)
is stable.
Table 4
The three right-handed Majorana neutrino masses in GeV for both case (i) and (ii). The expressions of MRR are collected
from Appendix B
Type
Case (i) |Mj |
Case (ii) |Mj |
DegT1A
DegT1B
DegT1C
InvT2A
InvT2B
NHT3
4.28 109 , 1.16 1010 , 3.84 1013

4.05 107 , 6.16 1011 , 7.6013
4.05 107 , 6.69 1012 , 6.69 1012
3.28 108 , 9.70 1012 , 6.79 1016
5.6527 1010 , 5.6532 1010 , 5.38 1016
6.51 1010 , 7.97 1011 , 1.01 1015
3.47 107 , 9.42 107 , 3.81 1013

3.28 105 , 4.98 109 , 7.60 1013
3.28 105 , 4.85 1011 , 7.81 1011
2.64 106 , 7.92 1010 , 6.70 1016
4.5971 108 , 4.5974 108 , 5.34 1016
5.27 108 , 6.45 109 , 1.01 1015
149
Table 5
Estimation of the values of effective mass parameter m
1 in eV, and dilution factor 1
Type
For case (i)
For case (ii)
m
1
m
1
DegT1A
DegT1B
DegT1C
InvT2A
InvT2B
NHT3
3.76 103
0.40
0.40
0.05
2.85 104
2.47 104
3.48
370
370
46.3
0.26
0.23
0.11
2.79 104
2.79 104
2.9 103
0.17
0.17
3.74 103
0.40
0.40
0.05
2.83 104
2.47 104
3.46
370
370
46.3
0.26
0.23
0.11
2.79 104
2.79 104
2.9 103
0.17
0.17
Table 6
Calculation of lepton asymmetry I and II for case(i) for respective neutrino mass models given in Appendix B
Type
I
II
1 = I + II
DegT1A
DegT1B
DegT1C
InvT2A
InvT2B
NRT3
2.10 106
1.19 1015
2.10 106
2.661018
1.741018
1.591014
1.47 102
5.92 107
2.66 1018
1.74 1018
1.59 1014
1.47 102
5.92 107
3.88 1027
2.24 1025
5.30 1026
2.52 1013
6.09 1013
Table 7
Calculation of lepton asymmetry I and II for case (ii) for respective neutrino mass models given in Appendix B
Type
I
II
1 = I + II
DegT1A
DegT1B
DegT1C
InvT2A
InvT2B
NRT3
1.71 108
2.16 1020
1.69 1020
1.27 1016
1.62 104
4.78 109
7.98 1020
2.54 1031
1.43 1026
2.46 1028
5.08 1017
4.90 1017
1.17 108
2.161020
1.691020
1.271016
1.62 104
4.78 109
Table 8
Calculation of baryon asymmetry YB for respective neutrino mass models given in Appendix B
Type
YBSM case (i)
YBSM case (ii)
DegT1A
DegT1B
DegT1C
InvT2A
InvT2B
NHT3
2.49 109
8.001024
5.201024
4.971019
5.40 105
1.08 109
2.031011
6.501026
5.101026
3.981021
5.94 107
8.801012
10 , a competitive naIf we consider the observed baryon asymmetry [26] YB = (6.10.3

0.2 ) 10
ture appears for both degenerate type-1 model (DegT1A) and normal hierarchical type-3 (NHT3).
150
But considering the mixing parameters and stability condition, one can easily rule out the degenerate type-1A (DegT1A) model. For NHT3 the mass of M1 is also within the IbarraDavidson
bound [27]. Considering the above results we have found that in type-II scenario normal hierarchical type-3 (NHT3) model is the most favourable choice of nature.
The present calculation may also be useful to discriminate the choices of Dirac neutrino mass
matrix. The charged lepton, up-quark and down-quark approximation of Dirac neutrino mass
matrix may not be the right choice of nature. Once a neutrino mass model is fixed by experiment,
then one can search for the possible structure of Dirac neutrino mass matrix by constraining
observed baryon asymmetry to such model [28].
Acknowledgements
One of us (A.K.S.) would like to thank UGC (NER), India, for awarding a fellowship under
FIP programme, Xth plan.
Appendix A
We list in Table 9 zeroth-order left-handed Majorana neutrino mass matrices [2,13] with texture zeros, mILL , corresponding to three models of neutrinos, viz., degenerate (DegT1A, DegT1B,
DegT1C), inverted hierarchical (InvT2A, InvT2B) and normal hierarchical (NHT3).
Table 9
Type
[DegT1A]
[DegT1C]
[InvT2A]
[NHT3]
1 0 0
0 1
0 0
1 0
0 0
0 1
1 0
0 1
[InvT2B]
1
2
1
2
12
1
2
1
2
1
2
12
[DegT1B]
diag
mLL
0
1
2
m0
0 m0
1
0
1 m0
0
0
12
Diag(1, 1, 1)m0
Diag(1, 1, 1)m0
Diag(1, 1, 1)m0
1
2 m0
1
2
0 1 1
1 0 0 m0
1 0 0
0
0
0
0
1
12
m0
2
0
mLL
1
2
Diag(1, 1, 0)m0
Diag(1, 1, 0)m0
Diag(0, 0, 1)m0
151
Appendix B
Here we collect [1,2] the various right-handed Majorana mass matrices MRR and the light
left-handed neutrino mass matrix mILL along with the input values of the parameters.
Degenerate type-1A (DegT1A):
2 2m
( 1 + )m+n
( 1 + )m
2
m+n
1
MRR = ( 2 + 1 )
( 1 + 1 )m
2
(2 + 2 )
1
2
( 1 )
I
1
mLL =
2
( 1
2
1 )
(1/2 + 1 2 )2n
(1/2 + 1 2 )n
( 1 1 )
2
(1/2 + 2 )
(1/2 + 2 )
(1/2 + 1 2 )n
v0 ,
(1/2 + 1 2 )
( 1 1 )
2
(1/2 + 2 )
m0 .
(1/2 + 2 )
Here 1 = 0.0061875, 2 = 0.0031625, = 0.3, m0 = 0.4 eV and v0 = 7.57 1013 GeV, =

105 .
Degenerate type-1B (DegT1B):

(1 + 21 + 22 )2m
1 m+n
1 m
m+n
2n
n
MRR =
v0 ,
1
(1 + 2
2
1 m
2 n
(1 + 2 )

mILL
(1 21 22 )
1
1

1
1
m0 .
(1 2 )
2
2
(1 2 )
Here 1 = 7.2 105 , 2 = 3.9 103 , = 0.3, m0 = 0.4 eV and v0 = 7.57 1013 GeV,
= 104 .
Degenerate type-1C (DegT1C):

(1 + 21 + 22 )2m
1 m+n
1 m
m+n
2n
n
MRR =
1
2
(1 + 2 ) v0 ,
1 m
(1 + 2 )n
2

mILL
(1 21 22 )
1
1

1
1
2
(1 2 ) m0 .
(1 2 )
2
Here 1 = 7.2 105 , 2 = 3.9 103 , = 0.3, m0 = 0.4 eV and v0 = 7.57 1013 GeV,
= 105 .
Inverted hierarchical type-2A (InvT2A):

(1 + 2)2m
m+n
m
v0
MRR =
,
m+n
1/22n
(1/2 )n
m
(1/2 )n
1/2

mILL =

(1 2)

1/2
(1/2 ) m0 .

(1/2 )
1/2
Here = 0.0045, = 0.0055, = 0.3, m0 = 0.05 eV and v0 = 6.06 1013 GeV, = 0.1.
152
Inverted hierarchical type-2B (InvT2B):

2m+7
m+n+4 m+4
MRR = m+n+4
2n
n v0 ,
m+4
n
1
0
1
1
mILL = 1 (3 4 )/2 (3 + 4 )/2 m0 .
1 (3 + 4 )/2 (3 4 )/2
Here = 0.3, m0 = 0.035 eV and v0 = 5.34 1016 GeV. For case (i) = 0.009 and for case
(ii) = 0.02.
Normal hierarchical type-3 (NHT3):
2m1
m+n1 m1
m+n1
m+n2
v0 ,
MRR =
0
0
1
m1

mILL
3
1
1
1 1 3

m0 .
Here = 0.3, m0 = 0.03 eV and v0 = 1.01 1015 GeV, = 0.007.

References
[1] N. Nimai Singh, M. Patgiri, Int. J. Mod. Phys. A 17 (2002) 3629;
M. Patgiri, N. Nimai Singh, Int. J. Mod. Phys. A 18 (2003) 743.
[2] N. Nimai Singh, M. Patgiri, M.K. Das, Pramana J. Phys. 66 (2) (2006) 361.
[3] M. Gell-Mann, P. Ramond, R. Slansky, in: P. Van Nieumenhuizen, D. Freedman (Eds.), Supergravity, Proceeding
of the Workshop, Stony Brook, NY 1979, North-Holland, Amsterdam, 1979;
T. Yanagida, KEK Lectures, 1979, unpublished;
R.N. Mahapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
[4] V.A. Kuzmin, V.A. Rubakov, M.E. Shaposhnikov, Phys. Lett. B 155 (1985) 36.
[5] M. Fukugita, T. Yanagida, Phys. Lett. B 174 (1986) 45;
M.A. Luty, Phys. Rev. D 45 (1992);
U. Sarkar, hep-ph/9810247.
[6] N.S. Manton, Phys. Rev. D 28 (1983) 2019;
F.R. Klinkhamer, N.S. Manton, Phys. Rev. D 30 (1984) 2212.
[7] A.D. Sakharov, JETP Lett. 5 (1967) 24.
[8] G. Panotopoulos, hep-ph/0606127.
[9] A.K. Sarma, H.Z. Devi, N. Nimai Singh, hep-ph/0604040.
[10] R.N. Mohapatra, G. Senjanovic, Phys. Rev. D 23 (1981) 165;
G. Lazarides, Q. Shafi, C. Wetterich, Nucl. Phys. B 181 (1981) 287;
C. Wetterich, Nucl. Phys. B 187 (1981) 343.
[11] B. Bajc, G. Senjanovic, F. Vissani, Phys. Rev. Lett. 90 (2003) 051802, hep-ph/0110310;
H.S. Goh, R.N. Mahapatra, S.P. Ng, hep-ph/0303055.
[12] C.S. Aulakh, B. Bajc, A. Melfo, G. Senjanovic, F. Vissani, hep-ph/0306242.
[13] G. Altarelli, F. Feruglio, Phys. Rep. 320 (1999) 295, hep-ph/9905536.
[14] A.S. Joshipura, E.A. Paschos, W. Rodejohann, JHEP 0108 (2001) 029;
A.S. Joshipura, E.A. Paschos, W. Rodejohann, Nucl. Phys. B 611 (2001) 227.
[15] E.K. Akhmedov, M. Frigerio, A.Y. Smirnov, JHEP 0309 (2003) 021, hep-ph/0305322.
[16] P.M. Fishbane, P. Kaus, J. Phys. G: Nucl. Part. Phys. 25 (1999) 16291640.
[17] E.W. Kolb, M.S. Turner, The Early Universe, AddisonWesley, New York, 1990.
[18] N. Sahu, S.U. Sankar, Phys. Rev. D 71 (2005) 013006, hep-ph/0406065;

T. Hambye, G. Senjanovic, Phys. Lett. B 582 (2004) 73, hep-ph/0307237;
S. Antush, S.F. King, Phys. Lett. B 597 (2004) 199, hep-ph/0405093.
[19] C.H. Albright, S.M. Barr, Phys. Rev. D 70 (2004) 033013, hep-ph/0404095.
[20] T. Hambye, ICTP Summer School Lecture Notes, 2004.
[21] J.A. Harvey, M.S. Turner, Phys. Rev. D 42 (1990) 3344;
U. Sarkar, hep-ph/9810247.
[22] L. Covi, E. Roulet, F. Vissani, Phys. Lett. B 384 (1996) 169, hep-ph/9605319.
[23] P. Di Bari, hep-ph/0406115;
P. Di Bari, hep-ph/0211175.
[24] W. Buchmuller, hep-ph/0306047.
[25] K.S. Babu, A. Bachri, H. Aissaoui, Nucl. Phys. B 738 (2006) 7692, hep-ph/0509091.
[26] D.N. Spergel, et al., Astophys J. Suppl. 148 (2003) 175, astro-ph/0302209.
[27] S. Davidson, A. Ibarra, Phys. Lett. B 535 (2002) 25.
[28] A.Kr. Sarma, Eur. Phys. J. C (2006), in press.
153
The effective electroweak mixing angle sin2 eff

with two-loop bosonic contributions
W. Hollik a , U. Meier a , S. Uccirati b,
a Max-Planck-Institut fr Physik (Werner-Heisenberg-Institut), D-80805 Mnchen, Germany
b Dipartimento di Fisica Teorica, Universit di Torino, INFN Sezione di Torino, 10125 Torino, Italy
Received 15 November 2006; accepted 4 December 2006

Abstract
We present the results for the full electroweak two-loop bosonic contributions to the effective leptonic
mixing angle of the Z boson, sin2 eff , in the Standard Model. A method applied to extract collinear divergences from two-loop vertex-functions is described. Comparisons of our results with those from a recent
previous calculation show complete agreement.
1. Introduction
Measurements of the various asymmetries of the Z resonance have determined the effective leptonic mixing angle sin2 eff with high accuracy. The current experimental value is
0.23153 0.00016 [1]; a linear electronpositron collider with GigaZ capabilities could even
reach an accuracy of 1.3 105 [2,3]. Comparison with the theoretical prediction of the Standard Model yields stringent bounds on the Higgs-boson mass MH . The importance and precision
of sin2 eff requires sufficient control on the theoretical accuracy of the prediction by providing
adequate higher-order calculations.
sin2 eff is determined from the ratio of the dressed vector and axial vector couplings gV ,A of
the Z boson to leptons [4]
E-mail address: uccirati@to.infn.it (S. Uccirati).

doi:10.1016/j.nuclphysb.2006.12.001
W. Hollik et al. / Nuclear Physics B 765 (2007) 154165

1
gV
sin eff =
1 Re
.
4
gA
2
155
(1)
In the on-shell renormalization scheme, it is related to the vector-boson mass ratio or, equiva2 = 1 M 2 /M 2 via
lently, to the on-shell quantity sW
W
Z
2
sin2 eff = sW
,
= 1 + ,
(2)
involving the factor, which is unity at the tree level and accommodates the higher-order contributions in . MW can be related to the precisely known Fermi constant G via the relation

M2
2
(1 + r),
1 W2 =
MW
(3)
MZ
2GF
where r summarizes the higher-order contributions. Beyond the one-loop order, the universal QCD corrections [5,6], the complete electroweak fermionic [7] and bosonic [8] two-loop
corrections, as well as leading higher-order contributions via the -parameter [9,10] and the Sparameter [11] are known for the Standard Model.
The universal higher-order terms to the self-energies from QCD and through the - and Sparameter enter also the quantity in (2). The fermionic electroweak two-loop corrections
(involving at least one closed fermion loop) have been established [12,13]; the MH -dependence
of the bosonic corrections to was given in [14], and only recently a complete calculation of
the bosonic contributions was reported [15].
In this paper we complete our result for the bosonic two-loop contributions by accomplishing
the MH -independent part of , which was missing in [14]. Together with our result from [14],
this yields an independent result for the complete bosonic two-loop corrections for sin2 eff , based
on different methods.
2. Calculational strategy
Expanding the dressed couplings in (1) according to gV ,A = gV(0),A (1 + gV(1),A + gV(2),A + ) in
powers of , the O( 2 ) contribution to sin2 eff in the loop expansion

(0)
(1)
(2)
sin2 eff = sin2 eff + sin2 eff + sin2 eff + O 3
(4)
is obtained,
(0)
(2)
sin2 eff =
gV
(0)
4gA
(2)
(2)
(1) (1)
(1)
Re gV gA + gA gA gV .
(5)
Hence, besides two-loop diagrams, also products of one-loop contributions have to be taken into
account. They play an important role in the cancellation of IR-divergences (see Section 3). But
the major task consists in the calculation of irreducible two-loop Z-vertex diagrams ( = e,
to be precise). The Z-boson couplings in (1) appear in the renormalized Z vertex for on-shell
Z bosons; for the two-loop contributions entering the expression (5) we need the renormalized
two-loop vertex

(2)
Z(2) MZ2 = gV(2) gA
(6)
5 .
As done in [13,14] we split the renormalized vertex into two UV-finite pieces,
156

Z(2) MZ2 = Z(2) MZ2 + CT

= Z(2) (0) + CT + Z(2) MZ2 Z(2) (0) .
(7)
Z(2)
(MZ2 ) denotes the corresponding unrenormalized Z vertex for on-shell leptons and
momentum transfer P 2 = MZ2 , and CT is the two-loop counter term. Details on the renormalization are given in [13], where the expressions are general and comprise also the bosonic
two-loop contributions.
The first term in (7) can be computed as in the fermionic case, which means generating Feynman diagrams with the help of FeynArts [16] and applying TwoCalc [17] to reduce the amplitudes
to standard integrals. The resulting vacuum integrals are calculated using analytic results [18,19],
whereas the two-loop self-energies with non-vanishing external momenta, which appear in the
counter terms, are calculated with the help of one-dimensional integral representations [20] or
the methods described in [21].
The IR-finite contributions to the second term in (7) can be calculated in analogy to the MH dependent part of the bosonic corrections [14]. This means that vertex corrections are computed
applying the methods from [21] together with further improvements in order to increase the
numerical stability. Non-planar diagrams are calculated using the method described in [14], and
diagrams with self-energy insertions are computed using dispersion relations as described in [13].
3. Treatment of IR- and collinear divergences
As a new feature compared to the MH -dependent subset of the bosonic corrections, diagrams
with internal photons appear that can be IR-divergent. These divergences, however, cancel in the
complete result rendering sin2 eff as a IR-finite quantity.
The IR-divergent diagrams are shown in Fig. 1. The grey circles represent one-loop insertions.
In order to verify the cancellation of these divergences, we have used the methods described
in [22]. These methods allow to extract all appearing IR-divergences in terms of IR-divergent
one-loop integrals. After this extraction the cancellation of these integrals can be checked analytically. The cancellation occurs between the various two-loop diagrams within the set of
Fig. 1(a)(c), and between the two-loop diagrams of Fig. 1(d) and the product of one-loop diagrams occurring in (5). This is shown schematically in Fig. 2.
(a)
(b)
(c)
(d)
Fig. 1. Diagrams with IR-divergences.
157
Fig. 2. Compensation of IR-divergences.
(a)
(b)
Fig. 3. Examples for collinearly divergent diagrams. All momenta are incoming, and P = p1 + p2 .
Moreover, collinear divergences appear. In order to regularize them we have kept the electron
mass me finite where necessary. In addition we have expanded the resulting expressions in me ,
so that the divergent behavior shows up as terms proportional to ln2 (m2e ) and ln(m2e ). Afterwards
it was checked analytically that the terms proportional to ln2 (m2e ) cancel, whereas the cancellation of the ln(m2e )-terms was checked numerically. Our methods to extract these logarithms are
explained in the following Section 4 by means of the two non-planar diagrams shown in Fig. 3.
4. Extraction of collinear divergences
Collinear divergences can show up in massive diagrams which do not contain any soft divergences, or they can overlap with soft divergences. In both cases we have managed to extract
explicitly the coefficients of the logarithms in m2e (which is the natural regulator for these divergences) and the constant term. For overlapping divergences we have used the results of [22],
while for the other ones we have used a subtraction method in parametric space. In order to explain it we consider the two collinear configurations of the non-planar diagrams in Fig. 3, which
arise in the computation.
In momentum space the generic diagram contains the following scalar and tensor integrals,

(1, q1 , q1 q1 , q1 q1 q1 )
1
(1,,,)
4
= 4 d q1 d 4 q2
,
V222
(8)
[1][2][3][4][5][6]
[1] = q12 m21 ,
[2] = (q1 p2 )2 m22 ,
[4] = (q1 q2 p2 )2 m24 ,
[5] = q22 m25 ,
[3] = (q1 q2 + p1 )2 m23 ,

[6] = (q2 p1 )2 m26 .
Following the discussion given in [21], we first combine the propagators [1] and [2] with a
Feynman-parameter z1 , the propagators [3] and [4] with a Feynman-parameter z2 , and the propagators [5] and [6] with a Feynman-parameter z3 . Then we combine the q1 and q1 q2 propagators
158
with a parameter x. After the q1 -integration we combine the residual propagators with a parame(1)
ter y and carry out the q2 integration. For the scalar integral V222 , the resulting expression reads
(we have changed x 1 x with respect to [21]):
(1)
V222
1
=
dx dy dz1 dz2 dz3 xy(1 x)(1 y)U 2 ,
(9)

U = xy(1 x)(1 y) + y(1 x) z1 ; p22 ; m1 , m2 + xy z2 ; P 2 ; m4 , m3

+ x(1 x)(1 y) z3 ; p12 ; m5 , m6 ,

= P 2 (z2 z3 )(1 z1 z2 ) + p12 (z2 z3 )(1 z1 z3 )

+ p22 (1 z1 z3 )(1 z1 z2 ) ,

z; p2 ; m, M = p 2 z(1 z) + m2 (1 z) + M 2 z.
Here and in the following we use the short-hand notation
1
1
dx dy dz3 =
1
dy
dx
0
1
dz3 .
0
4.1. Non-planar diagram: Two internal fermions

In Fig. 3(a) the following mass configuration of the V222 -family appears,
m1 = m5 = 0,
P
= MZ2 ,
m2 = me ,
p12
= p22
m3 = m4 = m6 = MW ,
= m2e .
(10)
The divergence comes entirely from the fermionphoton interaction, so it is possible to set p12 =
(1)
0. Inserting these values into (9) the resulting expression for the scalar integral V222 reads
(1)
V222
1
=
1
dy dz1 dz2 dz3 y(1 y)I,
dx
0
where a, b and c(x) depend on MW ,

particular, b is linear in z1 and z3 ,
b = z1 z3 + z1 + z3 + ,
with I =
P 2,
x(1 x)
,
[x(ax + b) + m2e c(x)]2
(11)
and on the Feynman-parameters y, z1 , z2 , and z3 . In
with
= MZ2 y(1 y),

= MZ2 y(1 y)z2 ,

2
MZ2 y(1 z2 ) ,
= (1 y) MW
2

= y MW
MZ2 yz2 (1 z2 ) .
Moreover, the following relations hold,
2

a + b = y MW
MZ2 z2 (1 z2 ) .
c(0) = yz12 ,
(12)
(13)
159
From (11) one can see that, for me = 0, I is divergent at x = 0. The divergence can be extracted
as ln m2e in the following way:

x(1 x)
x
x
+
.
[x(ax + b) + m2e c(x)]2 [xb + m2e c(0)]2
[xb + m2e c(0)]2
0
(14)
Ia is not divergent and we are free to set me = 0,
1
I = I a + Ib =
1
Ia =
dx
0
dx

1
1
1x
1
a+b
+
1
.
=
ln
x (ax + b)2 b2
b
b2
(15)
In Ib , the divergence can be extracted by integrating in x,

1
Ib = 2 ln m2e + ln c(0) ln b + 1 .
(16)
b
Thanks to the linearity of b with respect to z1 and z3 , we can now perform the z1 , z3 integrations
explicitly. Taking into account (12) and (13), the following types of integrals remain:
1
1
dz1 dz3 2 =
b
1
0
1
0

dz1 dz3
1
d
= ln 1 +
,
[][ ]
(z1 z3 + z1 + z3 + )2 d

ln z1
1
[]
dz1 dz3 2 =
Li2
Li2
,
d
[ ]
1
dz1 dz3
0

ln b 1 1
d
d
d
d
=
Li
Li
Li
+
Li
2
2
2
2
2
d
[ ]
[]
[ ]
b

d
d
+ ln[] ln 1
ln[ ] ln 1
[ ]
[]

d
d
+ ln[ ] ln 1
(17)
ln ln 1
,
[ ]
where we have introduced d = and [1 n ] = 1 + + n . It is important to note

that the only denominator appearing in the result (d) is now multiplied by regulator functions
which vanish when d goes to 0. The result is therefore smooth enough to be directly inserted into
(11), and the remaining z2 , y integrations are done numerically.
(,,)
The computation of the tensor integrals V222
is processed in the same way. The new
feature is the presence of z3 to some power m > 0 in the integrand. The previous formulae are
now replaced by the following ones, where an integration by parts in z3 is performed:
1
dz1 dz3
z3m
=
b2
1
dz3
0
ln z1
dz1 dz3 z3m 2
b
1

mz3m1
(1 z3 )d
ln 1 +
,
d
[ ]([ ]z3 + [])
1
=
dz3
0

mz3m1
[]
z3 +
Li2
Li2
,
d
[ ]
z3 +
160
1
dz1 dz3 z3m
(ln b 1)
b2

mz3m1
d
d
Li2
Li2
d
[ ]
([ ]z3 + [])
0

d
d
d
+ Li2
ln[ ] ln 1
Li2
[ ]
( z3 + )
[ ]

d
d
+ ln [ ]z3 + [] ln 1
+ ln[ ] ln 1
([ ]z3 + [])
[ ]

d
ln( z3 + ) ln 1
.
( z3 + )
1
=
dz3
(18)
4.2. Non-planar diagram: Four internal fermions

Fig. 3(b) corresponds to the configuration
m1 = 0,
P
m2 = m3 = m4 = m6 = me ,
= MZ2 ,
p12
= p22
m5 = MZ ,
= m2e .
(19)
This is an example for a diagram with a double collinear divergence. In order to regularize these
divergences the masses of all internal electrons are kept and p12 = m2e = p22 is used. In the follow(1)
ing the scalar integral V222 is considered; the tensor integrals can be calculated as in Section 4.1.
We apply the same parametrization as in Section 4.1 and change parameters according to
z2,3 1 z2,3 , z1 z3 . The following expression is obtained,
(1)
V222
1
=
1
dy dz3 y(1 y)J,
with J =
dx dz1 dz2
0
x(1 x)
,
( + m2e i)2
(20)

= x yz2 (1 z2 ) (1 x)(1 y) z1 + y(z1 z2 )(z2 z3 ) MZ2 ,
= xy(1 x)(1 y)(z1 z3 )2 + (1 x)yz32 + xy + x(1 x)(1 y)(1 z1 )2 .
Since is of the type = x(z1 A + z2 B + z1 z2 C), one can easily see from Eq. (20) that, for
(1)
me = 0, V222 diverges at x = 0 and at z1 = 0 = z2 . J can be calculated in the following way:
J = J a + Jb + Jc + Jd
= [J Jx Jz + Jxz ] + [Jx Jxz ] + [Jz Jxz ] + Jxz ,
1
Jx =
dx dz1 dz2
0
1
Jz =
0

x(1 x)
,
( + m2e )2 x 2 =xm2e =0

x(1 x)
dx dz1 dz2
,
2
2
( + m2e )2 z1 =z22 =z1 z22 =0
z1 me =z2 me =0
(21)
1
Jxz =
0

x(1 x)
dx dz1 dz2
( + m2e )2
x 2 =z12 =z22 =z1 z2 =0

xm2e =z1 m2e =z2 m2e =0
161
Ja is finite for me = 0, while Jb and Jc involve a single divergence (at x = 0 and z1 = 0 = z2 ,

respectively), giving raise to a simple logarithm in m2e . Finally, Jd contains the double divergence
at x = 0 and z1 = 0 = z2 , and we extract also a ln2 m2e term accordingly. Now we define:
a + b = yz2 (1 z2 )MZ2 i,

b = (1 y) z1 + y(z1 z2 )(z2 z3 ) yz2 (1 z2 ) MZ2 i,
a0 + b0 = yz2 MZ2 i,

b0 = (1 y) z1 yz3 (z1 z2 ) yz2 MZ2 i,
(22)

B = y 1 (1 x)(1 y)z3 MZ2 i,

B0 B|x=0 = y 1 (1 y)z3 MZ2 i,
A0 A|x=0 = (1 y)(1 yz3 )MZ2 i,

0 = |z1 =z2 =0 = 1 x(1 y) x + (1 x)yz32 ,
A = (1 x)(1 y)(1 yz3 )MZ2 i,
00 = |x=0 = |x=z1 =z2 =0 = yz32 .

With these notations we obtain for Ja , setting me = 0,
1
Ja =
0

1
1
1x
1
1x
dx dz1 dz2
+
x (ax + b)2 b2 (a0 x + b0 )2 b02
1
=
dz1 dz2
0

ln b ln(a + b) 1 ln b0 ln(a0 + b0 ) 1
.
b2
b02
(23)
The resulting expressions are of the form (17) and can be integrated in z1 and z3 accordingly.
Since in this case both , and d of (17) factorize a z2 , we get a factor 1/z2 , that is not multiplied
by regulator functions. However, even if the integrals containing 1/b2 and 1/b02 are separately
divergent, the combination of the two is finite, because they have the same behaviour around
(1)
z2 = 0. Therefore, after applying (17), the contribution from Ja to V222 can be safely integrated
numerically.
For Jb we proceed in an analogous way integrating in x, and obtain
1
Jb =
dx dz1 dz2
0
1
=
dz1 dz2
0
x
x
2
2
(xb + me 00 )
(xb0 + m2e 00 )2

2 1
1
ln b ln 00 1 ln b0 ln 00 1
ln me
.
b2
b2 b02
b02
(24)
Again, these terms are of the same form as in the expression for Ja and can be integrated analytically in z1 and z3 and then numerically in y and z2 .
162
To extract the logarithmic behaviour for Jc we have to integrate in z1 and z2 , yielding

1
Jc =
dx dz1 dz2
0
x
x(1 x)
2
2
[x(Az1 + Bz2 ) + me 0 ]
[x(A0 z1 + B0 z2 ) + m2e 00 ]2
1 (1 x)[ln(A + B) ln A ln B ln x + ln 0 ]
= dx
x
AB
0

2 1 x
1
ln(A0 + B0 ) ln A0 ln B0 ln x + ln 00
+ ln me
A0 B0
AB
A0 B0
1
(25)
The factors y, (1 y), and (1 x) in the denominators of this expression are always cancelled
by those present in the numerator. The overall factor 1/x is again regularized by the difference
between the terms with A, B, 0 and those with A0 , B0 , 00 . The remaining y and z3 integrations
can be done numerically.
Finally, for Jd we integrate in x, z1 and z2 , yielding
1
x
[x(A0 z1 + B0 z2 ) + m2e 00 ]2
0

1
A0 + B0
A0
B0
=
Li2 2
Li2 2
Li2 2
.
A0 B0
me 00
me 00
me 00
Jd =
dx dz1 dz2
(26)
The dilogarithms can be expanded according to

2

c
2 1
ln c ln m2e + O m2e ,
Li2 2 =
6
2
me
(27)
giving rise to a term proportional to ln2 m2e . The resulting expression is suitable for numerical
integration.
5. Results
In Table 1 the set of input parameters for our numerical evaluation is listed. MW and MZ are the
experimental values of the W - and Z-boson masses, which are the on-shell masses. They have to be conTable 1
Input parameters. MW and MZ are the experimental values for the W - and Z-boson mass, whereas M W and M Z are
calculated quantities in the pole mass scheme
Parameter
Value
MW
MZ
Z
mt
2)
(MZ
80.404 GeV
91.1876 GeV
2.4952 GeV
172.5 GeV
0.05907
2)
s (MZ
G
M W
M Z
0.119
1.16637 105
80.3766 GeV
91.1535 GeV
163
Table 2
Bosonic two-loop corrections to in comparison with the fermionic ones, and the results from [15]
( 2 )
( 2 )
( 2 )
( 2 )
MH [GeV]
ferm 104
bos 104
ferm 104 [15]
bos 104 [15]
100
200
600
1000
1.07
0.33
2.89
2.62
0.74
0.47
0.18
1.11
1.07
0.32
2.89
2.61
0.74
0.47
0.17
1.11
Table 3
Fermionic and bosonic electroweak two-loop corrections to sin2 eff . MH is given in GeV; the values for sin2 eff have
to be multiplied by 105
( 2 )
( 2 )
( 2 )
( 2 )
MH
sin2 eff (MW,ferm )
sin2 eff (kferm )
sin2 eff (MW,bos )
sin2 eff (kbos )
100
200
600
1000
93.89
98.51
106.89
105.36
2.38
0.73
6.43
5.83
1.93
0.97
0.19
1.16
1.65
1.05
0.40
2.47
verted to the values in the pole mass scheme [7], labeled as M W and M Z , which are used internally for
2 /(2M
the calculation. These quantities are related via MW,Z = M W,Z + W,Z
W,Z ). For Z the experi
3 /(2 2 )(1 +
mental value (Table 1) and for W the theoretical value has been used, i.e. W = 3G MW
2 )/(3 )) with sufficient accuracy.
2s (MW
Our results in terms of for the bosonic contributions are shown in Table 2, in comparison with the
fermionic ones. The last two columns of Table 2 contain the corresponding results from [15], which agree
with ours.
One can see from Table 2 that the bosonic corrections are basically of the same order of magnitude as the
fermionic ones, as long as the W -mass is taken as a fixed input parameter. The situation changes, however,
when instead of the W -mass the Fermi constant is taken as an input parameter and MW is derived via (3).
In order to identify the various sources of the two-loop contributions to sin2 eff , we expand both quantities
2
MW and according to {MW , } = {MW , }(tree+) + {MW , }( ) + O( 3 ), yielding

2
(tree+)
( 2 )
sin2 eff = sin2 eff
+ sin2 eff ( ) + O 3
+ sin2 eff MW

(tree+) 2
2
(MW
)
(tree+) 2 mW M ( 2 ) + 1 mW ( 2 ) + O 3 .
= 1
(28)
W
2
MZ
m2Z
m2Z
The first term represents the one-loop result, while the other two terms correspond to the electroweak
( 2 )
two-loop contributions to the shift in the W mass, MW , and to the shift in , ( ) . As can be seen
in Table 3, the fermionic corrections from MW dominate, while the contributions from MW and in the
bosonic sector cancel to a large extent.
In Table 4, the results for MW , as taken from [15], and sin2 eff are shown with only the fermionic
ferm and sin2 ferm ) and with both fermionic and bosonic two-loop corrections
two-loop corrections (MW
eff
ferm+bos
ferm+bos
and sin2 eff
) for various values of MH . Besides the electroweak one- and two-loop
(MW
2
corrections, the results for sin eff also contain the QCD-corrections of order O(s ) and O(s2 ) [6], as
well as the leading three-loop corrections of order O( 2 s m4t ) and O( 3 m6t ) [9].
The shifts in sin2 eff originating from the full set of bosonic two-loop contributions are displayed in the
last column of Table 4. According to the cancellations mentioned above, they are much smaller than the
corresponding shifts for a fixed value of the W mass as an input parameter, which would result from Eq. (2)
with the values of given in Table 2.
164
Table 4
bos is the purely bosonic
MW and sin2 eff without (ferm) and with (ferm + bos) bosonic two-loop corrections. sin2 eff
2
contribution to sin eff
MH [GeV]
100
200
600
1000
ferm [GeV]
MW
ferm+bos
MW
[GeV]
ferm
sin2 eff
80.3694
80.3276
80.2491
80.2134
80.3684
80.3270
80.2490
80.2141
0.231459
0.231792
0.232346
0.232587
ferm+bos
sin2 eff
0.231461
0.231792
0.232352
0.232599
bos [104 ]
sin2 eff
0.02
0
0.06
0.12
6. Conclusion
In conclusion, we have completed the calculation of the electroweak bosonic two-loop corrections to the effective leptonic mixing angle sin2 eff . A comparison of our results with the ones
in [15] was performed and full agreement was found. The bosonic corrections are basically of
the same order of magnitude as the fermionic ones when the W -mass is taken as a fixed input
parameter. If instead the Fermi constant is taken as input, with MW as a derived quantity, the
bosonic corrections to sin2 eff are quite small.
In addition, the use of the general techniques of [22] to treat IR-divergences provides a good
test of their applicability to physical processes. In particular we have shown that in this approach
the cancellation of the infrared poles can be verified analytically. Moreover, our complete calculation of the two-loop electroweak corrections of sin2 eff is the first physical application of the
numerical methods of [21] showing that these methods are really applicable in practice for the
calculation of physical quantities. In particular within this approach massive two-loop vertices
with many mass scales are computed directly, so not using mass expansions, which could represent a good way to deal also with calculations beyond the Standard Model, where often many
(unknown) massive particles come into the game.
Acknowledgement
W.H. is grateful to the Institute of Theoretical Physics, University of Vienna, where this paper
was finalized while he was Erwin Schrdinger visiting professor.
References
[1] LEP Collaborations, LEP Electroweak Working Group, SLD Electroweak, Heavy Flavour Groups, Phys. Rep. 427
(2006) 257, hep-ex/0509008;
D. Wood, XXIII International Conference on High Energy Physics, Moscow, 2006.
[2] J.A. Aguilar-Saavedra, et al., TESLA technical design report part III: Physics at an e+ e linear collider, hep-ph/
0106315;
T. Abe, et al., American Linear Collider Working Group Collaboration, in: R. Davidson, C. Quigg (Eds.), Proceedings of the APS/DPF/DPB Summer Study on the Future of Particle Physics (Snowmass 2001), SLAC-R-570,
Resource book for Snowmass 2001, hep-ex/0106055, hep-ex/0106056, hep-ex/0106057, hep-ex/0106058;
K. Abe, et al., ACFA Linear Collider Working Group Collaboration, ACFA Linear Collider Working Group report,
hep-ph/0109166.
[3] U. Baur, R. Clare, J. Erler, S. Heinemeyer, D. Wackeroth, G. Weiglein, D.R. Wood, in: N. Graf (Ed.), Proceeding
of the APS/DPF/DPB Summer Study on the Future of Particle Physics (Snowmass 2001) eConf C010630 (2001)
P122, hep-ph/0111314.
[4] D.Y. Bardin, et al., hep-ph/9709229, in: D. Bardin, W. Hollik, G. Passarino (Eds.), Precision Calculations for the Z
Resonance, CERN 95-03.
165
[5] A. Djouadi, C. Verzegnassi, Phys. Lett. B 195 (1987) 265;

A. Djouadi, Nuovo Cimento A 100 (1988) 357;
B.A. Kniehl, Nucl. Phys. B 347 (1990) 86;
F. Halzen, B.A. Kniehl, Nucl. Phys. B 353 (1991) 567;
B.A. Kniehl, A. Sirlin, Nucl. Phys. B 371 (1992) 141;
A. Djouadi, P. Gambino, Phys. Rev. D 49 (1994) 3499, hep-ph/9309298;
A. Djouadi, P. Gambino, Phys. Rev. D 53 (1996) 4111, Erratum.
[6] K.G. Chetyrkin, J.H. Khn, M. Steinhauser, Phys. Rev. Lett. 75 (1995) 3394, hep-ph/9504413;
K.G. Chetyrkin, J.H. Khn, M. Steinhauser, Nucl. Phys. B 482 (1996) 213, hep-ph/9606230.
[7] A. Freitas, W. Hollik, W. Walter, G. Weiglein, Phys. Lett. B 495 (2000) 338, hep-ph/0007091;
A. Freitas, W. Hollik, W. Walter, G. Weiglein, Phys. Lett. B 570 (2003) 260, hep-ph/0007091, Erratum;
A. Freitas, W. Hollik, W. Walter, G. Weiglein, Nucl. Phys. B 632 (2002) 189, hep-ph/0202131;
A. Freitas, W. Hollik, W. Walter, G. Weiglein, Nucl. Phys. B 666 (2003) 305, Erratum;
M. Awramik, M. Czakon, Phys. Lett. B 568 (2003) 48, hep-ph/0305248.
[8] M. Awramik, M. Czakon, Phys. Rev. Lett. 89 (2002) 241801, hep-ph/0208113;
A. Onishchenko, O. Veretin, Phys. Lett. B 551 (2003) 111, hep-ph/0209010;
M. Awramik, M. Czakon, A. Onishchenko, O. Veretin, Phys. Rev. D 68 (2003) 053004, hep-ph/0209084.
[9] J. van der Bij, K. Chetyrkin, M. Faisst, G. Jikia, T. Seidensticker, Phys. Lett. B 498 (2001) 156, hep-ph/0011373;
M. Faisst, J.H. Khn, T. Seidensticker, O. Veretin, Nucl. Phys. B 665 (2003) 649, hep-ph/0302275.
[10] R. Boughezal, J.B. Tausk, J.J. van der Bij, Nucl. Phys. B 713 (2005) 278, hep-ph/0410216;
Y. Schroder, M. Steinhauser, Phys. Lett. B 622 (2005) 124, hep-ph/0504055;
K.G. Chetyrkin, M. Faisst, J.H. Kuhn, P. Maierhofer, C. Sturm, hep-ph/0605201;
R. Boughezal, M. Czakon, hep-ph/0606232.
[11] R. Boughezal, J.B. Tausk, J.J. van der Bij, Nucl. Phys. B 725 (2005) 3, hep-ph/0504092.
[12] M. Awramik, M. Czakon, A. Freitas, G. Weiglein, Phys. Rev. Lett. 93 (2004) 201805, hep-ph/0407317;
M. Awramik, M. Czakon, A. Freitas, G. Weiglein, Nucl. Phys. B (Proc. Suppl.) 135 (2004) 119, hep-ph/0408207.
[13] W. Hollik, U. Meier, S. Uccirati, Nucl. Phys. B 731 (2005) 213, hep-ph/0507158.
[14] W. Hollik, U. Meier, S. Uccirati, Phys. Lett. B 632 (2006) 680, hep-ph/0509302.
[15] M. Awramik, M. Czakon, A. Freitas, hep-ph/0605339, hep-ph/0608099.
[16] J. Kblbeck, M. Bhm, A. Denner, Comput. Phys. Commun. 60 (1990) 165;
T. Hahn, Nucl. Phys. Proc. Suppl. 89 (2000) 231, hep-ph/0005029;
T. Hahn, Comput. Phys. Commun. 140 (2001) 418, hep-ph/0012260;
FeynArts Users Guide, available at http://www.feynarts.de.
[17] G. Weiglein, R. Scharf, M. Bhm, Nucl. Phys. B 416 (1994) 606, hep-ph/9310358;
G. Weiglein, R. Mertig, R. Scharf, M. Bhm, PRINT-95-128 Prepared for 2nd International Workshop on Software Engineering, Artificial Intelligence and Expert Systems for High-energy and Nuclear Physics, La Londe Les
Maures, France, 1318 January 1992.
[18] G. t Hooft, M.J.G. Veltman, Nucl. Phys. B 153 (1979) 365.
[19] A.I. Davydychev, J.B. Tausk, Nucl. Phys. B 397 (1993) 123.
[20] S. Bauberger, M. Bhm, G. Weiglein, F.A. Berends, M. Buza, Nucl. Phys. B (Proc. Suppl.) 37 (1994) 95, hepph/9406404;
S. Bauberger, F.A. Berends, M. Bhm, M. Buza, Nucl. Phys. B 434 (1995) 383, hep-ph/9409388;
S. Bauberger, M. Bhm, Nucl. Phys. B 445 (1995) 25, hep-ph/9501201.
[21] A. Ferroglia, M. Passera, G. Passarino, S. Uccirati, Nucl. Phys. B 680 (2004) 199, hep-ph/0311186;
A. Ferroglia, M. Passera, G. Passarino, S. Uccirati, Nucl. Phys. B 650 (2003) 162, hep-ph/0209219;
S. Actis, A. Ferroglia, G. Passarino, M. Passera, S. Uccirati, Nucl. Phys. B 703 (2004) 3, hep-ph/0402132;
G. Passarino, Nucl. Phys. B 619 (2001) 257, hep-ph/0108252;
G. Passarino, S. Uccirati, Nucl. Phys. B 629 (2002) 97, hep-ph/0112004;
S. Uccirati, Acta Phys. Pol. B 35 (2004) 2573, hep-ph/0410332.
[22] G. Passarino, S. Uccirati, Nucl. Phys. B 747 (2006) 113, hep-ph/0603121.
Superstring scattering from O-planes

Mohammad R. Garousi
Department of Physics, Ferdowsi University, P.O. Box 1436, Mashhad, Iran
Institute for Studies in Theoretical Physics and Mathematics IPM, P.O. Box 19395-5531, Tehran, Iran
Received 25 November 2006; received in revised form 9 December 2006; accepted 13 December 2006
Abstract
We write the vertex operators of massless NSNS and RR states of Type II superstring theory in the
presence of orientifold p-planes. They include the usual vertex operators of Type II theory and their images.
We then calculate the two-point functions of these vertex operators at the projective plane PR2 level. We
show that the result can be written in the Veneziano-type formulae, with the same kinematic factor that
appears in the Dp -branes amplitudes. While the scattering amplitudes with the usual vertex operators are
not gauge invariant, the above amplitudes are invariant. From the amplitude describing scattering of two
NSNS states off the O-plane, we find the low energy effective action of O-planes. The result shows a
relative factor 2p6 between couplings to O-planes and to D-branes at ( )2 order.
1. Introduction
String theory appears to contain various extended objects, other than just strings. Dp -branes,
for instance, are dynamical extended objects that have remarkably simple word sheet description.
World-sheet with boundary on which spacetime coordinate fields satisfy Dirichlet and Neumann
boundary conditions. Using this simple description, it was shown in [1] that D-branes carry
RamondRamond (RR) charges in Type II superstring theory. Another set of extended objects
which have simple world sheet description are orientifold p-planes. These are non-dynamical
planes at the fixed points of spacetime. The world-sheet in this case has crosscap instead of
boundary. Having world sheet description, one can study these O-planes using well understood
tools of two-dimensional quantum field theory. Interaction between O-plane and D-brane, for
instance, has been found. From this amplitude one finds that O-planes have negative tension, and
E-mail address: garousi@ipm.ir (M.R. Garousi).
doi:10.1016/j.nuclphysb.2006.12.007
M.R. Garousi / Nuclear Physics B 765 (2007) 166184
167
carry RR charge in the case of Type II superstring theory (see e.g., [2]). We are interested, in this
paper, in studying closed string scattering with O-planes. Closed string scattering off O-planes
have been studied in [3,4] using boundary state formalism. However, we would like to study
them here using projective plane calculation.
In string theory, the description for scattering of closed string off Dp -brane is as follows: The
string background is taken to be simply flat empty space, however, interactions of closed strings
with a Dp -brane are described by world-sheets with boundary. The boundary of world sheet
must be fixed to the surface at the position of the Dp -brane. The latter is accomplished by imposing Neumann boundary conditions on the fields along Dp -brane world volume, and Dirichlet
boundary conditions on the fields transverse to the Dp -brane [2,5]. The scattering amplitudes
describing the interactions of closed strings with D-branes has been found in [69], i.e.,
(t/2)(2s)
(1 t/2 2s)
where K(1, 2) is kinematic factor that depends on momentum and polarization of external states.
It satisfies various Ward identities. In this equation, t is the closed string channel and s is open
string channel.
The description for scattering of closed string off Op -plane is as follows: The string background is again taken to be flat empty space, however, interactions of closed strings with these
Op -planes are described by world-sheets with a crosscap instead of boundary. The crosscap must
be fixed to the surface at the position of the Op -plane. The latter is accomplished by choosing
appropriate boundary conditions [10]. The scattering amplitude of closed string off O-plane has
been studied in [3,4] using the boundary states formalism. In these studies, the standard vertex
operator of closed string states, i.e., the same vertex operator that one uses in scattering with
D-brane, have been used. In [4], it has been argued that the scattering of NSNS closed string off
O-plane is not gauge invariant because the standard vertex operators have been used in calculating the amplitude. We note here that in the presence of O-plane, the vertex operators must satisfy
the constraint that when they are on the world volume of O-plane they should describe unoriented
states. Using this constraint, we will find the vertex operators in the presence of O-plane and will
find that the scattering amplitudes describing the interactions of closed strings with O-planes are
given by
AD2 K(1, 2)
(t/2)(u/2)
(1)
(1 t/2 u/2)
where K(1, 2) is exactly the same kinematic factor that appears in the D-brane case. Hence, the
amplitude is gauge invariant. In this amplitude, both channels are closed string channel. Here
t = (p1 + p2 )2 and u = (p1 + p2 D)2 where pi is momentum of external state and pi D
is momentum of its image.
The paper is organized as follows: In the following section, after writing the NSNS vertex
operator in the presence of O-plane, we describe the calculation of the scattering of two such
states from an O-plane using conformal field theory techniques. The result is fully covariant,
satisfy the Ward identity and can be written as (1). In Section 3, we repeat the same calculation for
scattering amplitude of two RR states from O-plane. In this case also we show that the result can
be written as (1). We then examine the NSNS amplitude at low energy in Section 4. We show in
this section that the massless poles of the amplitude are in exact agreement with analogous field
theory calculations, the contact terms at ( )0 order are consistent with negative cosmological
constant of O-plane, and the effective action at ( )2 order is the same as the effective action of
D-brane with the relative factor 2p6 . We conclude with a discussion of our results in Section 5.
APR2 K(1, 2)
168
2. NSNS scattering amplitudes

T-duality of unoriented Type I theory gives Type II theory with O-planes at the fixed points
of the T-dual space. In the original Type I theory, states are invariant under world-sheet parity
. In the T-dual theory, this operator has two effects, world sheet parity and spacetime reflection
on the T-dualized coordinates x i . If one considers only the center of mass dependence of the
string wave function, then the projection onto = +1 determines the wave function at x i to
be the same as that at x i , up to a sign.1 Various components of massless states in NSNS sector
satisfy [11]

Bab x a , x i = Bab x a , x i ,
Gab x a , x i = Gab x a , x i ,

Gai x a , x i = Gai x a , x i ,
Bai x a , x i = Bai x a , x i ,

Gij x a , x i = Gij x a , x i ,
(2)
Bij x a , x i = Bij x a , x i .
The orientifold fixed plane is at x i = 0. The wave functions that satisfy the above conditions are

a
a
Bab = iab eipa x sin pi x i ,
Gab = ab eipa x cos pi x i ,

a
a
Bai = ai eipa x cos pi x i ,
Gai = iai eipa x sin pi x i ,

a
a
Bij = iij eipa x sin pi x i ,
Gij = ij eipa x cos pi x i ,
(3)
where is the polarization tensor. The momentum and polarization tensor satisfy p 2 = 0, and
p = 0 = p . One can rewrite the above wave functions as

1
eipx + (D D) eipDx ,
2

1
B = eipx (D D) eipDx ,
(4)
2
where matrix D is ab for world volume directions and is ij for transverse directions. The
vertex operator corresponding to the above wave functions is
G =
V PR2 (, p, z, z ) =
1
:V (p, z)::V (p, z ):
2

+ D T D :V (p D, z)::V (p D, z ): .
(5)
The holomorphic components V (p, ) in 0 and 1 picture are given by

V1 (p, z) = e(z) (z)eipX(z) ,

V0 (p, z) = X (z) + ip (z) (z) eipX(z) .
(6)
The antiholomorphic components take the same form as in Eq. (6) but with the left-moving fields
z), (z) (
z), and (z)
replaced by their right-moving counterpartsi.e., X(z) X(
1 Our index conventions are that lowercase Greek indices take values in the entire ten-dimensional spacetime, e.g.,
= 0, 1, . . . , 9; early Latin indices take values in the world volume, e.g., a, b, c = 0, 1, . . . , p; and middle Latin indices
take values in the transverse space, e.g., i, j = p + 1, . . . , 9. Thus, for example, G denotes the entire spacetime
metric, while Gab and Gij denote metric components for directions parallel and orthogonal to the O-planes/D-branes,
respectively.
169
z). The vertex operator (5) can also be written as

(
V PR2 (, p, z, z ) =
1
:V (p, z)::V (p, z ):
2

+ :V (p, z)::V (p, z ): (z)D(
z), (z)(
z), X(z)DX(
z) .
(7)
For the special case that D = i.e., no T-duality on Type I theory, one should recover the
NSNS vertex operator of unoriented Type I theory. Interchanging the holomorphic part with the
antiholomorphic part in the second line above, one finds that the two lines are the same except
the fact that the transpose of polarization appears in the second line. Hence, the KalbRamond
antisymmetric tensor is projected out. The surviving states are graviton and dilaton which are the
only NSNS states in unoriented Type I theory. Note also that on the Op -plane world volume
X i = 0 = i . Hence, only matrix D ab = ab appears in the vertex operator. As a result, as
expected, the vertex operator represents unoriented states on the world volume of O-plane.
For D-brane and in the absence of O-planes, there is no constraint on the closed string states,
hence, the wave function for graviton/antisymmetric tensor is
eipx
(8)
and the corresponding vertex operator is twice the first term in (5).
The amplitudes describing the scattering of two massless NSNS states from an O-plane/Dbrane are calculated as two closed string vertex operator insertions on a PR2 /D2 with appropriate
crosscap/boundary conditions [2,5]. The amplitude is given as

A d 2 z1 d 2 z2 V1 (1 , p1 , z1 , z 1 )V2 (2 , p2 , z2 , z 2 )
(9)
both integrals run over the unit disk. The choices for the picture of vertex operators are arbitrary,
however, the sum of superghost charge must be 2. The amplitude can be separated as
APR2 = A1 + A2 + A3 + A4 ,
AD2 = A1 ,
(10)
where

1
A1 1 2 d 2 z1 d 2 z2 :V1 (p1 , z1 )::V0 (p1 , z 1 )::V1

(p2 , z2 )::V0 (p2 , z 2 ): ,
4

1
A2 1 D 2T D
4

d 2 z1 d 2 z2 :V1 (p1 , z1 )::V0 (p1 , z 1 )::V1

(p2 D, z2 )::V0 (p2 D, z 2 ): ,
A3
A4

1
D 1T D 2
4

d 2 z1 d 2 z2 :V1 (p1 D, z1 )::V0 (p1 D, z 1 )::V1

(p2 , z2 )::V0 (p2 , z 2 ): ,

1
D 1T D D 2T D
4

(p2 D, z2 )::V0 (p2 D, z 2 ): .

d 2 z1 d 2 z2 :V1 (p1 D, z1 )::V0 (p1 D, z 1 )::V1
(11)
170
To calculate the above correlators, one needs propagators for various world-sheet fields. Separately, the left- and right-moving fields have standard propagators on sphere world sheet, e.g.,

X (z) X (w) = log(z w),

,
(z) (w) =
zw

(z) (w) = log(z w),
(12)
with analogous expressions for the right-movers (see [12]). However, the projective plane/disk
can be obtained as a Z2 identification of the sphere. This identifies points z and z with z = n/z
where n = 1 for projective plane and n = 1 for disk. The method of images can then be used
to obtain the propagators between left- and right-moving fields [13]. Alternatively, the boundary
state of O-plane is obtained by inserting (1)m in every terms of the exponent of the D-brane
boundary state [10]. This means every zi z j combination of D-brane gets replaced by zi z j [3].
In the propagator between left- and right-moving fields each zw should be replaced by zw.
Hence,

X (z)X (w)
= D log(1 + nzw),

nD
=
,
(z) (w)
1 + nzw

nD
(w)
(z) =
,
1 + nzw

w)
(z)(
= log(1 + nzw).
(13)
For Dp -brane case, the fields X a for a = 0, 1, . . . , p should satisfy Neumann boundary conditions, and X i for i = p + 1, . . . , 9 should satisfy the Dirichlet boundary conditions [2,5]. These
facts are incorporated by the matrix D . In the scattering amplitude, this matrix has also the
effect of having conservation of momentum along the world volume of Dp -brane. To obtain PR2
propagators, we have replaced zw by zw.
In the above propagators, we have fixed the coefficient of the propagator of world sheet fermion by requiring that the coefficient of unphysical
vertex operator in the OPE of V0 (p1 , z1 ) and V0 (p2 , z 2 ) must be a total derivative [14], i.e.,
:V0 (p1 , z1 )::V0 (p2 , z 2 ): n
p1 D p2 1
eip1 X+ip2 X + .
(1 + nz1 z 2 )p1 Dp2 2
Given the above propagators, it is now straightforward to evaluate the correlation functions
appearing in the amplitude (11). After evaluating these correlators, one finds

1
d 2 z1 d 2 z2 |z1 z2 |2p1 p2 |1 + nz1 z 2 |2p1 Dp2
A1 1 2
4
(1 + nz1 z 1 )p1 Dp1 (1 + nz2 z 2 )p2 Dp2
p+1
1
2
3
4
I
+ I
+ I
+ I
(p1 + p1 D + p2 + p2 D),
(14)
where delta function gives conservation of momentum along world volume of O-plane/D-brane.

X C C
for PR2 case and i(2)p+1 CDX2 CD2 CD2 for disk
There is a factor of i(2)p+1 CPR
2 PR2 PR2
X , for example, is the functional determinant for the string field X a [13]. We are not
where CPR
2
trying to calculate these constants here. Hence, we omit them here and at the end normalize the
171
amplitude by comparing with low energy effective action. In the above equation,

p1 D nz1
p2 D nz2
p2
1
+
+
I =
(1 + nz1 z 1 ) (1 + nz2 z 1 ) (z2 z 1 )
(z1 z2 )2 (z1 z 2 )2

p1 D nz1
p2 D nz2
p1

+
,
(1 + nz1 z 2 ) (1 + nz2 z 2 ) (z1 z 2 )

p1 D nz1
p2 D nz2
p1
n
2
I =
+
(z1 z2 )
(1 + nz1 z 2 ) (1 + nz2 z 2 ) (z1 z 2 )

p1 D D p1 D D
,
(1 + nz1 z 1 )(1 + nz2 z 1 )

n
p1 D nz1
p2 D nz2
p2
3
I =
+
(z1 z2 )
(1 + nz1 z 1 ) (1 + nz2 z 1 ) (z2 z 1 )

p2 D D p2 D D
,
(1 + nz1 z 2 )(1 + nz2 z 2 )

p1 D ( p2 D p2 D ) D (p2 D p1 p1 p2 D )
n
4
I
=
(z1 z2 )
(1 + nz1 z 1 )(z1 z 2 )(1 + nz2 z 2 )
p2 D ( p1 D + p1 D ) + D (p1 D p2 p1 p2 D )
+
(1 + nz1 z 2 )(z1 z 2 )(1 + nz2 z 1 )
(p1 p2 p1 p2 )
n
(15)
.
(z1 z2 )(z1 z 2 )2
Now the integrand should be invariant under the CKG of the projective plane/disk which is
SL(2, C) group of sphere that respects the identification z = n/z. Transformation of z and z
under this group are
d z nc
az + b
,
z n
,
(16)
cz + d
bz na
where a, b, c, d are four complex parameters that satisfy ad cb = 1. Moreover, comparing the
second transformation above with the complex conjugate of first transformation gives b = nc
and d = a. Under above transformation, one finds
z
z1 z2
,
(cz1 + d)(cz2 + d)
z 1 z 2
,
z 1 z 2
(bz1 na)(bz2 na)
1 + nz1 z 2
,
1 + nz1 z 2
(cz1 + d)(bz2 na)
dzd z
.
dz d z
(cz + d)2 (bz na)2
z1 z2
(17)
One can check that the integrand in the amplitude (14) is invariant under the above transformations. In checking this, one should replace one of the momentum in each bracket [ ] in (15) in
terms of other two momentum using conservation of momentum and the physical condition on
the polarization of external states, i.e., pi i = 0. To gauge fix this symmetry, we first fix one
vertex operator at z1 = z 1 = 0. The residual symmetry is then z aa z which is a rotation in the
z-plane. To fix the residual symmetry, we fix the polar coordinate of the second vertex operator,
172
i.e., |z2 | = |z2 | = r. Ignoring the volume of the symmetry group, i.e.,
d 2 z1 d 2 z2 dr 2
(18)
one is left with a single real integral of the form

A
D2
iTp
=
2
1
0
2
APR
1
iTp

p p
p Dp2 a1
a2
a3
dr 2 r 2 1 2 1 r 2 2
+
+
,
r2 1 r2 r4
1

2 p1 p2

a2
a3
2 p2 Dp2 a1
1+r
dr r
+
+
,
r2 1 + r2 r4
2
8
0
(19)
where a1 , a2 and a3 are three kinematic factors depending only on the spacetime momenta and
polarization tensors. We have also normalized the amplitudes at this point by the introduction
of factors of , the closed string coupling, and Tp (Tp ), the O-plane (D-brane) tension, respectively.2 Note that the constants that we have omitted in (14) and in (18) have the same sign for
both PR2 and D2 . Hence, the constants Tp and Tp must have the same sign. The kinematic factors
are

a3 = Tr 1 2T (1 + t/2),

a1 = n Tr(1 D) p1 2 p1 p1 2 D 1 p2 p1 2 1T D p1

1
p2 1T 2 p1 + p1 2 1T p2
2

t
Tr(1 D) Tr(2 D) Tr(1 D 2 D) + {1 2} ,
4
a2 = Tr(1 D)(p1 2 D p2 p2 D 2 D p1 )
p1 2T 1 D p1
+ p1 D 1 D 2 D p2 p2 D 2 1T D p1
+ Tr(1 D) Tr(2 D) (t/4) + {1 2},
(20)
where t = (p1 + p2 )2 = 2p1 p2 is the momentum transfer to the O-plane/D-brane, and

s = p1a p1b ab = 12 p1 D p1 is the momentum flowing parallel to the world-volume of the
O-plane/D-brane. Our notation is such that e.g., p1 2 1T D p1 = p1 2 1 D p1 .

Using the following integrals:
1
0
1
dx x (1 + x) =
F (, 1 ; 2 ; 1)
,
1
dx x (1 x) = B(1 , 1 ).
(21)
2 Here and in the subsequent amplitudes, we omit the Dirac delta-function which imposes momentum conservation in
the world-volume directions. We have introduced a phase i though which corresponds to that of the analogous field
theory amplitudes calculated in Minkowski spacesee Section 4.
173
One finds
AD 2 =
2
APR
1

iTp
a3 B(1 t/2, 1 2s) + a1 B(t/2, 1 2s) + a2 B(1 t/2, 2s) ,
2

F (2s, 1 t/2; t/2; 1)
=
a3
8
1 t/2
F (2s, t/2; 1 t/2; 1)

F (2s + 1, 1 t/2; 2 t/2; 1)
+ a2
.
+ a1
t/2
1 t/2
(22)
iTp
(23)
As is evident the final amplitude is symmetric under the interchange of the two string states, i.e.,
1 2, despite the asymmetric appearance of the initial integrand in Eq. (9). From the gamma
function factors appearing D2 amplitude, we see that the amplitudes contain two infinite series
poles3 corresponding to closed string states in the t -channel with m2 = 4n, and to open string
states in the s-channel with m2 = n, with n = 0, 1, 2, . . . . However, the PR2 amplitude has
poles only in the t-channel with m2 = 4n. Having no open string pole is consistent with the fact
that the Op -planes, unlike the Dp -branes, are not dynamical objects. The amplitude AD2 is the
one that has been found in [9] using upper-half z-plane calculation. While this amplitude satisfies
the Ward identities associated with the gauge invariance of external states, i.e., the amplitude
2
vanishes upon substituting i pi qi or qi pi , where qi pi = 0 [9], the amplitude APR
1
does not satisfy the Ward identities. Using the package [15] for expanding the hypergeometric
2
functions, one finds that APR
has the following expansion at low energy4 :
1

iTp

1
PR2
(t 4s) Tr 1 2T + 2a1
A1 =
8
t

+ ln(2) 4s Tr 1 2T + a2

2

2
2s(t + 4s) Tr 1 2T 4sa1 + ta2 ln(2) 4s 2 Tr 1 2T + sa2
24
+ .
(24)
+
The massless pole is reproduced in field theory assuming linear coupling of gravity to O-plane.
However, the ( )0 contact term is not consistent with the cosmological constant of the O-plane
effective action. Moreover, the and ( )2 contact terms above are not consistent with what is
expected for the effective action of O-plane. This part of amplitude has been also found in [4]
using the boundary state formalism. However, they have used another picture for the graviton
vertex operator, i.e., they have used V(0,0) whereas we use V(1,0) . Their result is the same as
2
above up to some contact terms. It has been argued in [4] that the ln(2) terms should be
APR
1
absent in the final gauge invariant amplitude since their presence would be in conflict with the
fact that such terms are not present in the effective action of Type I string theory. Using this,
2
an expansion for the amplitude APR
has been found in [4] that is consistent with expected low
1
energy effective action. We will show shortly, however, that when we add the contributions from
A2 , A3 and A4 the final result satisfies the Ward identities, it does not contain the ln(2) terms,
3 We explicitly restore here. Otherwise our conventions set = 2.
4 I would like to thank H. Ghorbani for finding this expansion.
174
it is independent of the choice for the picture of vertex operators, and it produces expected low
energy effective action.
Before calculating A2 , A3 , A4 , we would like to compare the t -channel poles in the amplitudes
(22) and (23). The Feynman rule for calculating the t-channel poles in field theory is as follows:
It is a summation of infinite terms. Each is a product of a vertex of three states where two of
them are the on-shell external states, the propagator of the off-shell state, and the coupling of
the off-shell state to O-plane/D-brane. Using the fact that the external states are the same in both
amplitudes, the t -channel analysis gives information about the linear coupling of off-shell closed
string states to O-plane and to D-brane. To compare them, consider the following definition of
the beta function (see e.g., [16]):
B(, ) =

(1)n ( 1)( 2) ( n)
+n
n!
n=0
(25)
This expansion can be rewritten as
B(, ) =
(n + 1 )(n + )
1
(1 )
(n + + 1)
n=0
1
= F (1 , ; 1 + , 1).
(26)
From this, one finds the following expansion for the hypergeometric function appearing in (23):
1 ( 1)( 2) ( n)
1
F (1 , ; 1 + , 1) =
.
n!
+n
(27)
n=0
Using (25) and (27), one finds the t -channel expansion for the amplitudes in (22) and (23). Noting
that a1D2 = a1PR2 , one finds that the two amplitudes have the same pole structure, however, the
overall sign of each pole in two amplitudes is different. The signs are such that the massless
poles have opposite sign, the first massive poles have the same sign, the second massive poles
have again opposite sign, and so on. This observation can also be made from the interaction of
D-brane with O-plane [2].
Now to calculate A2 , A3 , A4 , one may try to find them from A1 by replacing appropriate
momenta and polarization tensors, e.g., if (p1 , 1 , p2 , 2 ) (p1 , 1 , p2 D, D 2T D), then
A1 A2 . In this way, using the identity (D D) = , one can easily show that
A1 = A4 ,
A2 = A3 .
(28)
APR2 ,
would be in terms of some hypergeometric funcFinding A2 in this way, the final result,
tions which we have found it difficult to simplify the final result. Instead, we calculate A2 using
another arrangement for the picture of the vertex operators in (11). We write it as
A2 1 (D 2 D)

d 2 z1 d 2 z2 :V1 (p1 , z1 )::V0 (p1 , z 1 )::V0 (p2 D, z2 )::V1

(p2 D, z 2 ): .
After evaluating the correlators and fixing the CKG symmetry as before, one will find the following result:
2
APR
2
iTp
1
8
0

p Dp2
p Dp2 a1
a2
+
a
1 + r2 2
dr 2 r 2 1
+
,
3
r 2 r 2 (1 + r 2 )
(29)
175
where a1 , a2 , and a3 are the kinematic factors (20). In terms of the hypergeometric function it is

iTp
F (2s, 1 u/2; 2 u/2; 1)
A2 =
a3
8
1 u/2
F (2s, u/2; 1 u/2; 1)

F (2s + 1, u/2; 1 u/2; 1)
+ a1
+ a2
,
u/2
u/2
where u = (p1 + p2 D)2 = 2p1 D p2 . This amplitude has an infinite tower of closed string
2
poles in u-channel, and has no open string s-channel. This amplitude like APR
in (22) does not
1
satisfy the Ward identities. However, we will see shortly that the combination of the two parts
does. Considering the following identity [17]:
F (a, b; b + 1; 1) =
b
(a b)(b + 1)
F (a, a b; a b + 1; 1) +
ba
(a)
(30)
and the on-shell condition 4s + t + u = 0, one can write the final result as Veneziano-type amplitude with the interplay between the t and u channels,
APR2 = 2(A1 + A2 )
iTp

a3 B(1 t/2, 1 u/2) + a1 B(t/2, u/2) + a2 B(1 t/2, u/2) .
=
4
(31)
To show that the amplitude satisfies the Ward identities, we compare APR2 with AD2 . Both
amplitudes can be rewritten as

iTp
t
(t/2)(2s)
us Tr 1 2T 2sa1 a2
,
AD2 =
2
2
(1 t/2 2s)

iTp

t
(t/2)(u/2)
PR2
T
=
us Tr 1 2 + 2sa1 a2
.
A
(32)
4
2
(1 t/2 u/2)
Using the fact that a1PR2 = a1D2 and a2PR2 = a2D2 , one realizes that the kinematic factors are
exactly the same. On the other hand, it has been shown in [9] that kinematic factor of disk
amplitude satisfies the Ward identities. Hence the projective plane amplitude satisfies the Ward
identities too. Moreover, if one uses the graviton vertex operators in any other picture, the final
result would be the same as above. We will use the above amplitude in Section 4 to study the
low energy effective action of O-planes. We note here that the amplitude APR2 is invariant under
replacements (p1 , 1 , p2 , 2 ) (p1 , 1 , p2 D, D 2T D), (p1 , 1 , p2 , 2 ) (p1 D, D 1T
D, p2 , 2 ), and (p1 , 1 , p2 , 2 ) (p1 D, D 1T D, p2 D, D 2T D).
3. RR boson amplitude
The RR vertex operator in Type II superstring theory in the absence of O-plane is [11]
V D2 (, p, z, z ) = (P i(n) )AB :V1/2 A (p, z)::V1/2 B (p, z ):.
(33)
The holomorphic components above are given by

V1/2 A (p, z) = e(z)/2 SA (z)eipX(z)
(34)
176
and the antiholomorphic components have the same form, but with the left-moving fields replaced by their right-moving counterparts. We refer readers to [9] for our conventions on P and
i(n) . Using the same discussion as for NSNS vertex operator (7), in the presence of O-plane,
the RR vertex operator is given by

1
V PR2 (, p, z, z ) = (P i(n) )AB :V1/2 A (p, z)::V1/2 B (p, z ):
2

+ :V1/2 A (p, z)::V1/2 B (p, z ): (z)D(
z),(z)(
z),X(z)DX(
z) .
(35)
z), one finds the following replacement for the spin field
Now under replacement (z) D (
[13]:
SA (z) (M)A B SB (z).
(36)
The RR vertex operator (35) becomes
V PR2 (, p, z, z )

1
= (P i(n) )AB :e(z)/2 SA (z) eipX(z) ::e(z)/2 SB (z)eipX(z) :

2

+ :e(z)/2 M 1 A D SD (z)eipDX(z) ::e(z)/2 (M)B C SC (z)eipDX(z) : .
(37)
Now one can work with the above vertex operator and the propagators (13) to calculate any
scattering amplitude. Alternatively, one may use the replacement [9]
z) D X(z),
X(
z) D (z),
(
to simplify the world sheet propagators. The matrix

= log(1 + nzw),
X (z) X (w)

n
(z) (w)
=
,
1 + nzw

n
(z) =
,
(w)
1 + nzw

(z) (w)
= log(1 + nzw).
z) (z)
(
D
in (13) is replaced with
(38)
,
i.e.,
(39)
In the RR vertex operator, one should first note that the replacement (38) makes the following
replacement for the spin field [9]:
SA (z) MA B SB (z),
(40)
where matrix M is the same matrix that appears in (36). Hence, the RR vertex operator (37)
becomes5
5 The replacement (38) makes the NSNS vertex operator (7) to the form

1
(41)
( D) :V (p, z)::V (p D, z ): + :V (p, z )::V (p D, z): .
2
The first term above is the vertex operator in the presence of D-brane. In general, the vertex operator in the presence of
O-plane can be written as
V PR2 (, p, z, z ) =
V PR2 (, p, z, z ) =

1 D2
V (, p, z, z ) + V D2 (, p, z , z) .
2
(42)
177

1
V PR2 (, p, z, z ) = (P i(n) M)AB :e(z)/2 SA (z)eipX(z) ::e(z)/2 SB (z)eipDX(z) :
2

+ :e(z)/2 SA (z)eipX(z) ::e(z)/2 SB (z) eipDX(z) : .
(43)
For the special case that D = , one should recover the RR vertex operator of Type I theory
which results from unorienting Type IIB theory. In this case, the matrix M = 1. Interchanging
the holomorphic part with the antiholomorphic part in the second line above, one finds an extra
minus sign which results from fermionic nature of R vertex operator. Now using the relation
((n) )AB = (1)n(n+1)/2 ((n) )BA (see Appendix of [9]), one finds zero result for n = 1, 5, and
non-zero result for n = 3 which is the only RR vertex operator surviving under unorienting
Type IIB.
In the presence of D-brane and absence of O-plane, the term in the first line of (43) represents
the RR vertex operator.
The amplitude describing two RR states scattering from O-plane/D-brane can be written as

A d 2 z1 d 2 z2 V1 (p1 , 1 , z1 , z 1 )V2 (p2 , 2 , z2 , z 2 ) ,
where the RR vertex operator appear in (43). The amplitude can be separated as
APR2 = A1 + A2 + A3 + A4 ,
where
A1
A2
A3
A4
AD2 = A1 ,

1
AB
CD
d 2 z1 d 2 z2
(P 1(n) M) (P 2(m) M)
4

:V1/2 A (p1 , z1 )::V1/2 B (p1 D, z 1 )::V1/2 C (p2 , z2 )::V1/2 D (p2 D, z 2 ): ,

1
(P 1(n) M)AB (P 2(m) M)CD d 2 z1 d 2 z2
4

:V1/2 A (p1 , z1 )::V1/2 B (p1 D, z 1 )::V1/2 C (p2 , z 2 )::V1/2 D (p2 D, z2 ): ,

1
AB
CD
d 2 z1 d 2 z2
(P 1(n) M) (P 2(m) M)
4

:V1/2 A (p1 , z 1 )::V1/2 B (p1 D, z1 )::V1/2C (p2 , z2 )::V1/2 D (p2 D, z 2 ): ,

1
(P 1(n) M)AB (P 2(m) M)CD d 2 z1 d 2 z2
4

:V1/2 A (p1 , z 1 )::V1/2 B (p1 D, z1 )::V1/2 C (p2 , z 2 )::V1/2 D (p2 D, z2 ): .
The correlation function for superghost in A1 is

(z )/2 (z )/2 (z )/2 (z )/2
1
1
2
2
:e
::e
::e
::e
:

1/4
2
2
= (1 + nz1 z 1 )|z1 z2 | |1 + nz1 z 2 | (1 + nz2 z 2 )
.
Now to find the correlation function between spin fields, we use the following result [18]:

:SA (z1 )::SB (z2 )::SC (z3 )::SD (z4 ):
z14 z23 ( )AB ( )CD z12 z34 ( )AD ( )BC
,
=
2(z12 z13 z14 z23 z24 z34 )3/4
where zij = zi zj . In the above relation the propagator of world-sheet fermion is
(zi ) (zj ) = /(zi zj ). In our case, the propagator is the same when both fields
178
are holomorphic or antiholomorphic. However, when one field is holomorphic and the other is
antiholomorphic, the propagator is given by (39). Then, for the correlation function of spin fields
in A1 , one expects the following result:
n
|1 + nz1 z 2 |2 ( )AB ( )CD + (1 + nz1 z 1 )(1 + nz2 z 2 )( )AD ( )BC

.
2((1 + nz1 z 1 )|z1 z2 |2 |1 + nz1 z 2 |2 (1 + nz2 z 2 ))3/4
(44)
With this correlation function, one can easily verifies, after performing the X-correlators, that the
integrand in A1 is invariant under CKG of projective plane/disk. Fixing this symmetry as before,
one finds

D2
AB
CD 1
AB ( )CD B(t/2, 2s)
A (P 1(n) M) (P 2(m) M)
2
1
+ AD ( )BC B(t/2, 1 2s) ,
2
1
PR2
A1 (P 1(n) M)AB (P 2(m) M)CD
4
F (1 + 2s, t/2; 1 t/2; 1)
1
AB ( )CD
2
t/2
F (2s, t/2; 1 t/2; 1)

1
.
+ AD ( )BC
(45)
2
t/2
The disk amplitude is the one that has been found in [7,9]. Both amplitudes have the same
massless pole, however, the overall sign is different. This is consistent with the observation that
the massless closed string couplings to D-brane and to O-plane have different sign.
Similar calculation for A2 , A3 , and A4 gives the following results:
1
2
(P 1(n) M)AB (P 2(m) M)CD
APR
2
4
F (1 + 2s, 1 u/2; 2 u/2; 1)
1
AB ( )CD
2
1 u/2
F (2s, u/2; 1 u/2; 1)

1
,
+ AD ( )BC
2
u/2
2
2
= APR
APR
3
2 ,
2
2
= APR
APR
4
1 .
Now summing all contributions and using the identity (30), one finds

1
PR2
AB
CD 1
(P 1(n) M) (P 2(m) M)
AB ( )CD B(t/2, 1 u/2)
A
2
2
1
+ AD ( )BC B(t/2, u/2) .
2
(46)
(47)
In this case, like (32), the PR2 and D2 amplitudes have the same kinematic factor, i.e.,
(t/2)(2s)
,
(1 t/2 2s)
1
(t/2)(u/2)
K(1, 2)
,
2
(1 t/2 u/2)
AD2 K(1, 2)
APR2
(48)
179
where the kinematic factor is

u
K(1, 2) = Tr P 1(n) M Tr(P 2(m) M ) + s Tr P 1(n) M 2(m) M . (49)
4
To write the kinematic factor in terms of only momenta and polarization tensors, one must perform the traces over gamma matrices which for general n, m, p is rather complicatedsee [7].
4. Low energy effective action
Now using the following expansion for gamma function:
2
1
( a)( b)
= 2
+ O( )

(1 a b)
6
ab
(50)
one finds the low energy expansion for the scattering amplitudes (63). In the case of NSNS
amplitude, one finds

1

1 a2

iTp
AD2 =
Tr 1 2T 4s Tr 1 2T + 2a1
2
t
s 2
2

+
2s(t + 4s) Tr 1 2T + 4sa1 + ta2 + ,
12

iTp 1

2

4s Tr 1 2T 2a1 (a1 + a2 )
APR2 =
4
t
u

2
+
(51)
2s(t + 4s) Tr 1 2T 4sa1 + ta2 + .
12
For D2 case, it has been shown that the massless poles are reproduced by field theory which
includes supergravity action for the bulk and the DBI action,

S0D2 = Tp d p+1 x e det(gab + Bab )
for D-brane action [9]. In particular, the t-channel in field theory for gravitongraviton amplitude
is
i
gg
At (p1 , 1 , p2 , 2 ) = Tp 2p1 2 1 p2 + 2p2 1 2 D p2
t
+ 2p1 2 1 D p1 + 2p1 2 D 1 p2
p1 D p1 Tr(1 2 )
Tr(1 D)p1 2 p1 Tr(2 D)p2 1 p2

iTp Tr(1 2 ) + Tr(D 1 2 ) ,

(52)
which is the same as the massless t -channel in AD2 up to some ( )0 contact terms. Moreover,
the ( )0 contact terms are exactly the ones that are reproduced by the above DBI action. This
conforms that the constant Tp in the D-brane amplitude (19) is tension of D-brane. The ( )2
contact terms, on the other hand, are shown to be reproduced by the following action [19]:
S2D2 =

2 2
Tp e det gab Rabcd R abcd 2Rab R ab Rij ab R ij ab + 2Rij R ij . (53)
48
180
For PR2 case, the ( )0 order of O-plane action should be [13]

S0PR2 = Tp d p+1 x e det gab ,
(54)
where we take tension of O-plane to be negative, i.e., Tp is absolute value of O-plane tension.
To check that the massless poles are reproduced by the effective actions, we recall that the wave
function for gravitons are given by (4). So the t -channel of scattering of two gravitons is
2
APR
t (g, g) =
Tp
gg
At (p1 , 1 /2, p2 , 2 /2)
Tp

gg
+ At (p1 D, D 1 D/2, p2 D, D 2 D/2)

Tp
1
gg
=
At (p1 , 1 , p2 , 2 ).
2
Tp
One can check that the above massless pole and the massless t -channel in (51) are the same
up to some ( )0 contact terms. The u-channel can be read from (52) by the replacement
(p1 , 1 , p2 , 2 ) (p1 , 1 /2, p2 D, D 2 D/2) and (p1 , 1 , p2 , 2 ) (p1 D, D 1
D/2, p2 , 2 /2), i.e.,
2
APR
u (g, g) =
Tp
Tp
gg
At (p1 , 1 /2, p2 D, D 2 D/2)

gg
+ At (p1 D, D 2 D/2, p2 , 2 /2)

Tp
1
gg
=
At (p1 , 1 , p2 D, D 2 D).
2
Tp
(55)
One can again check that the above massless pole and the massless u-channel pole in (51) are
exactly the same up to some ( )0 contact terms. To find these contact terms, one should subtract
2
2
the field theory amplitudes APR
+ APR
from string theory amplitude (51), i.e.,
u
t
PR2
2
APR2 (g, g) APR
t (g, g) Au (g, g)

iTp 1
Tr(1 D) Tr(2 D) Tr(1 2 )
=
2
2
Tr(1 D 2 D) 2 Tr(1 D 2 )

2
2s(t + 4s) Tr(1 2 ) 4sa1 + ta2 + .
+
24
(56)
The above amplitude is symmetric under replacements (p1 , 1 , p2 , 2 ) (p1 , 1 , p2 D, D

2 D), (p1 , 1 , p2 , 2 ) (p1 D, D 1 D, p2 , 2 ), and (p1 , 1 , p2 , 2 ) (p1 D, D 1
D, p2 D, D 2 D). So instead of considering wave function (4), one can consider (8) to find
the effective action consistent with the above couplings. The two graviton couplings from (54) is
exactly the ( )0 contact terms in the two lines above. This confirms that the tension of O-plane
is negative, i.e., Tp .
It was shown in [19] that the two gravitons couplings from action (53) with the wave function
(8) are exactly equal to the ( )2 contact terms of AD2 , e.g., the terms in the second line of (51).
On the other hand the ( )2 contact terms of APR2 are the same as the ( )2 contact terms of AD2
up to overall factor. Therefore, the action which is consistent with ( )2 contact terms of APR2 ,
e.g., the terms in the third line of (56) is

Tp D2
1
PR2
S2 (gravity) =
S (gravity).
2
Tp 2
181
(57)
The same factor of Tp /2Tp = 2p6 , where we have used the relation between the O-plane and
D-brane in Type II superstring theory [11], appears in the R 2 couplings in WessZumino part of
O-plane effective action [20]. The above action has been also found in [4].
Now we would like to show that the same factor appears in the KalbRamond couplings at
( )2 order. To see this, consider the scattering of two antisymmetric KalbRamond tensors from
O-plane. In field theory, the scattering amplitude with wave function (8) is given by
ABB
t (p1 , 1 , p2 , 2 ) =
i
T 2p1 2 1 p2 + p1 D p1 Tr(1 2 )
t p
+ 2(p1 D 2 1 p2 + p1 2 D 1 p2

+ p2 D 1 2 p1 )
iTp Tr(D 1 2 ).
(58)
The t-channel amplitude in field theory is

BB
2
APR
t (B, B) = At (p1 , 1 /2, p2 , 2 /2)
+ ABB
t (p1 D, D 1 D/2, p2 D, D 2 D/2)
1
(p1 , 1 , p2 , 2 ).
= ABB
2 t
The u-channel amplitude in field theory is
(59)
BB
2
APR
u (B, B) = At (p1 , 1 /2, p2 D, D 2 D/2)
+ ABB
t (p1 D, D 1 D/2, p2 , 2 /2)
1 BB
= At (p1 , 1 , p2 D, D 2 D).
2
(60)
These poles are exactly the ones appear in APR2 . Subtracting them, one finds
PR2
2
APR2 (B, B) APR
t (B, B) Au (B, B)

iTp 2

2su Tr(1 2 ) 4sa1 + ta2 + .
=
2
24
(61)
The above amplitude is invariant under replacement (p1 , 1 , p2 , 2 ) (p1 , 1 , p2 D, D 2

D), (p1 , 1 , p2 , 2 ) (p1 D, D 1 D, p2 , 2 ), and (p1 , 1 , p2 , 2 ) (p1 D, D 1
D, p2 D, D 2 D). Note that there is no ()0 contact terms in the above amplitude which is
consistent with the ( )0 order action of O-plane (54) which has no B-field. On other hand, the
( )2 contact terms are the same as the ( )2 contact terms of AD2 (B, B) in (51) up to the overall
factor. Hence, the ( )2 action of O-plane is proportional to the action of D-brane, i.e.,

Tp D2
1
S2PR2 (B, B) =
(62)
S (B, B).
2
Tp 2
An ( )2 correction to the D-brane action, S2D2 (B, B), has been proposed in [4]. One expects the
above correspondence between couplings to O-plane and to D-brane should be hold for other
182
massless fields. However, at higher order, the contact terms of APR2 and AD2 are not proportional to each other. So their effective actions are not proportional to each other either.
5. Discussion
In this paper, we have presented detailed calculations of two-point amplitudes describing
massless NSNS and RR closed string scattering from an orientifold p-plane in ten dimensions.
Using the result for NSNS amplitude, we have derived the effective action of O-plane up to
order ( )2 . While the action at ( )0 order is consistent with the negative cosmological constant term of O-plane, the couplings at ( )2 are the same as the corresponding coupling in the
effective action of D-brane with relative factor 1/2(Tp /Tp ) = 2p6 .
We have seen that the RR amplitude (48) and NSNS amplitude (32) can be written as
iTp
(t/2)(2s)
K(1, 2)
,
2
(1 t/2 2s)
iTp
(t/2)(u/2)
=
K(1, 2)
,
4
(1 t/2 u/2)
AD2 =
APR2
(63)
where the kinematic factor K(1, 2) is exactly the same in both amplitudes. This form of amplitude in general is consistent with the fact that the massless t -channel pole in O-plane amplitude
is proportional to minus massless t -channel pole in D-brane amplitude. Hence, we speculate
that the final result for any other massless amplitude should also be written as in (63) with the
appropriate kinematic factor that has been found in [9] in the case of scattering from D-brane.
We have seen that our calculation cannot fix the value of constants Tp and Tp in amplitude
(19). However, it indicates that the coupling of massless closed string to O-plane is given by
Tp , and to D-brane by Tp . On the other hand, as we have clarified in the paragraph under (19)
the two constants Tp and Tp have the same sign. Using the fact that the coupling of graviton to an
extended object is given by the tension of the object, this indicates that the PR2 calculations fix
the tension of O-plane to be negative. This is unlike the derivation of O-plane tension from interaction of O-plane with D-brane which gives both positive and negative result for the tension of
O-plane, i.e., Tp = 2p5 Tp , depending on the ChanPaton factor of the open string stretching
between D-brane and its image [2].
We have seen that the massless closed string vertex operators in the presence of O-plane must
be in the form of (7) in order to have gauge invariant scattering amplitude. This vertex includes
the usual vertex operator of Type II theory and its image. Using this vertex, one can calculate
the coupling of this state with a massless NS open string on the world volume of a D-brane [21].
The result is zero. Alternatively, one can calculate the scattering amplitude of NSNS massless
closed string from a D-brane. It is given by (22) and another terms which can be read from (22)
by replacements p2 p2 D and 2 D 2T D. One can easily find the massless open string
pole is cancelled in the final amplitude. All these are consistent with the fact that a D-brane
at the fixed plane is unoriented. However, in the scattering of external states from D-brane at
disk level, one cannot fix the position of D-brane. Hence, in the above amplitude the D-brane
is not necessary to be coincident with O-plane. One should not conclude that this D-brane is
also unoriented. If a D-plane is not coincident on O-plane then there is another D-brane which
is image of the first one. In the scattering of closed string from these branes at disk level, the
off-shell open strings see both D-branes to be coincident. Hence the open strings see only two
unoriented D-branes at the position of O-plane.
183
We have discussed the massless poles of the string scattering amplitudes in previous section.
As well as these poles, the string amplitudes contain an infinite number of massive poles as well.
Each of the higher poles represents a massive closed string state coupling to the Op -plane. From
this point of view, the Op -plane provides a -function source for each of these massive states,
just as it does for the massless states. The same set of -functions in the transverse coordinates
appears in the construction of a boundary state description of a O-plane [10,22]. This would then
lend itself to an interpretation of O-planes as objects of zero thickness. However, since O-plane is

not only a source of the massless fields but also massive fields with m2 = 4n/
, the conventional
(low energy) spacetime picture will breakdown at distances of the order of . It is within this
range that the full closed string spectrum makes
its presence felt. Hence from this perspective,
one would ascribe a thickness of the order of to O-planes.
An exception to the above form of the amplitudes is the special case of p = 8. In this case,
the O8 -plane forms a domain wall dividing the ten-dimensional spacetime into two halves. The
transverse space is only one-dimensional running along x 9 , and so it is only the ninth component
of the momentum vectors which is not conserved. Hence,
2
2
s = p19 = p29 .
(64)
Since there is a single component of momentum appearing here, one has either p19 = p29 or
p19 = p29 . In the first case, one finds t = 0, and in the second case one finds u = 0. This result
allows one to rewrite the amplitude (32) in a drastically simple form

iTp 1

2

T
4s Tr 1 2 2a1 (a1 + a2 ) .
APR2 =
(65)
4
t
u
As we have seen in the previous section, this is exactly the amplitude that one finds in the low
energy limit. Hence, all massive poles disappear in O8 -plane amplitudes. Similar disappearance
of massive states happens in the D1 -brane amplitude [6,7,9].
While there are no massive poles in the above amplitude, one should not conclude that no massive string states couple to O8 -plane. The coupling of massive states to external massless states
are such that in the O8 -plane amplitude all massive poles disappear. One expects then the massive poles make their appearance in the scattering amplitudes involving massive external strings.
Similar discussion can be used for D1 -brane case [9]. To clarify this point, one may consider
bosonic string theory and work with massless and tachyon states. The scattering amplitude of
two tachyons with O-plane can be found to be
APR2
(1 u/2)(1 t/2)

.
(2 u/2 t/2)
For p = 24, this simplifies to

APR2
1
1
+
,
u/2 t/2
where only massless poles survives. Now consider the scattering amplitude of one tachyon and
one graviton with Op -plane. The result is
(1 u/2)(1 t/2)
(u/2)(t/2)
+ p2 D 2 D p2
(u/2 t/2)
(u/2 t/2)
(u/2)(1 t/2)
+ 2p2 D 2 D p1
(1 u/2 t/2)
APR2 Tr(2 D)
184
+ p1 D 2 D p1
(1 u/2)(1 t/2)

.
(2 u/2 t/2)
(66)
For the special case of p = 24, the amplitude for u = 0 case simplifies to

1
1
p1 2 p1
PR2
A
Tr(2 D)
+
+
t/2 u/2
1 t/2
which has both massless and tachyon poles. If one consider other external massive states, one
would find other massive poles. Hence one concludes that all massive states couple to the domain
wall O-planes like any other O-plane. This should also be evident given the complex form of the
boundary states describing a O8 -plane, and the complex form of the potential energy between
an
O8 -plane and a D8 -brane [2]. Hence, domain wall O-planes also have thickness of order .
Acknowledgements
I gratefully acknowledge useful conversations with A. Sen, M.M. Sheikh-Jabbari and
A. Ghodsi. I would also like to thank H. Ghorbani for double checking the result in equation
(20).
References
[1] J. Polchinski, Dirichlet-branes and RamondRamond charges, hep-th/9510017.
[2] J. Polchinski, S. Chaudhuri, C.V. Johnson, Notes on D-branes, hep-th/9602052.
[3] B. Craps, F. Roose, Phys. Lett. B 445 (1998) 150, hep-th/9808074;
B. Stefanski Jr., Nucl. Phys. B 548 (1999) 275, hep-th/9812088;
J.A. Scrucca, M. Serone, Nucl. Phys. B 552 (1999) 291, hep-th/9812071.
[4] H.J. Schnitzer, N. Wyllard, JHEP 0208 (2002) 012, hep-th/0206071.
[5] J. Dai, R.G. Leigh, J. Polchinski, Mod. Phys. Lett. A 4 (1989) 2073;
J. Polchinski, Phys. Rev. D 50 (1994) 6041.
[6] I.R. Klebanov, L. Thorlacius, The size of p-brane, hep-th/9510200.
[7] S.S. Gubser, A. Hashimoto, I.R. Klebanov, J.M. Maldacena, Nucl. Phys. B 472 (1996) 231, hep-th/9601057.
[8] J.L.F. Barbon, Phys. Lett. B 382 (1996) 60, hep-th/960198.
[9] M.R. Garousi, R.C. Myers, Nucl. Phys. B 475 (1996) 193, hep-th/9603194.
[10] C. Callan, C. Lovelace, C. Nappi, S. Yost, Nucl. Phys. B 293 (1987) 83.
[11] J. Polchinski, TASI lectures on D-branes, hep-th/9611050.
[12] D. Friedan, Notes on string theory and two-dimensional conformal field theory, in: M.B. Green, D.J. Gross (Eds.),
Unified String Theories, World Scientific, Singapore, 1986.
[13] J. Polchinski, String Theory, vols. I and II, Cambridge Univ. Press, Cambridge, 1998.
[14] M.B. Green, N. Seiberg, Nucl. Phys. B 299 (1988) 559.
[15] T. Huber, D. Maitre, Comput. Phys. Commun. 175 (2006) 122, hep-ph/0507094.
[16] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, Cambridge Univ. Press, Cambridge, 1987.
[17] The wolfram function site, http://functions.wolfram.com/.
[18] D. Friedan, S. Shenker, E. Martinec, Phys. Lett. 160B (1985) 55;
D. Friedan, S. Shenker, E. Martinec, Nucl. Phys. B 271 (1986) 93.
[19] C.P. Pachas, P. Bain, M.B. Green, JHEP 9905 (1999) 011, hep-th/9903210.
[20] K. Dasgupta, D.P. Jatkar, S. Mukhi, Nucl. Phys. B 523 (1998) 465, hep-th/9707224.
[21] M.R. Garousi, R.C. Myers, Nucl. Phys. B 542 (1999) 73, hep-th/9809100.
[22] C.G. Callan, I.R. Klebanov, Nucl. Phys. B 465 (1996) 473, hep-th/9511173.
Non-relativistic non-BPS Dp-brane

J. Kluson 1
Dipartimento di Fisica & Sezione INFN, Universit di Roma Tor Vergata, Via della Ricerca Scientifica,
I-00133 Roma, Italy
Received 10 October 2006; accepted 18 December 2006
Abstract
We construct non-relativistic non-BPS Dp-brane action. Then we will study the properties of the tachyon
kink solution on its world-volume. We will argue that this tachyon kink describes non-relativistic D(p 1)brane.
Keywords: D-branes; Tachyon condensation
1. Introduction and summary

Non-relativistic string theory was determined some years ago in the context of the study of
the string theories in the background with the electric field [16]. Then it was soon recognised in
[79] that non-relativistic string theory is consistent sector of the string theory with the manifest
world-sheet conformal field theory description that has appropriate Galilean symmetry. For the
case of strings, this can be accomplished if we consider wound strings in the presence of a
background B-field and tuning the B-field so that the energy coming from the B-field cancels
the tension of the string. Then it was shown in [13] that the similar procedure can be performed
in case of Dp-branes.2 Since Dp-branes are charged with respect to RamondRamond form
Cp+1 it is possible to find the limit where the tension of the wound Dp-brane is cancelled by the
coupling to the Cp+1 field. As a result we obtain a world-volume kappa symmetric action of a
non-relativistic Dp-brane [13].
E-mail addresses: josef.kluson@roma2.infn.it, klu@physics.muni.cz (J. Kluson).
1 On leave from Masaryk University, Brno.
2 For related works, see [1012,14,15].
doi:10.1016/j.nuclphysb.2006.12.010
186
J. Kluson / Nuclear Physics B 765 (2007) 185199
We can also ask the question whether some symmetries of the relativistic string theories remain symmetries of the non-relativistic string theories as well. It was shown in [15] that in the
non-relativistic string theories one can find T-duality relation between effective actions for nonrelativistic Dp-branes and also one can argue that non-relativistic D1- and D2-brane actions arise
in the dimensional reduction from non-relativistic M2-brane action.
On the other hand it is well known that all Dp-branes can be interpreted as the result of
the tachyon condensation on the world-volume of unstable D-branes or on the world-volume of
D-braneanti-D-brane pair [16,17].3 One can then ask the question whether non-relativistic BPS
Dp-branes can be considered as the result of the tachyon condensation on the world-volume of
non-relativistic non-BPS Dp-brane. The goal of this paper is to study this problem.
Our approach closely follows [13] with some mild differences. First of all we would like to
find non-relativistic non-BPS Dp-brane action where the tachyon contribution has the schematic
form V (T )F (ij i T j T ) with F (u) u1/2 for large u. The reason for this requirement is that
for such a form of the non-BPS Dp-brane action one can find the tachyon kink solution that represents codimension one D(p 1)-brane [34]. This condition however implies that in the scaling
limit that defines non-relativistic Dp-brane from the relativistic one the tachyon has to scale in
the same way as the world-volume scalar modes that define the embedding of Dp-brane along
the directions spanned by the background RamondRamond Cp form. On the other hand if we
try to apply procedure given in [13] where these modes scale with the factor that goes to infinity
we obtain that the argument of the tachyon potential goes to infinity that however implies that
the tachyon potential goes to zero. As a result the non-BPS Dp-brane disappears in this definition of the non-relativistic limit. To resolve this paradox we use the fact that the non-relativistic
limit can be interpreted as the limit where the world-volume scalar modes that parametrise directions transverse to the RamondRamond background are small. More precisely, since the correct
definition of the non-relativistic limit of BPS Dp-brane needs background RamondRamond
form Cp+1 it is natural to split the world-volume scalar modes to two sets, one that corresponds
to the modes that parametrise directions along the RamondRamond form Cp+1 and the second one that parametrises directions transverse to it. In the case of the ordinary BPS Dp-brane
its world-volume can span the directions specified by the background RamondRamond Cp+1
form. However in case of a non-BPS Dp-brane the situation is different since there does not exist
such a Cp+1 form. On the other hand it is well known that the non-BPS Dp-brane couples to
the RamondRamond forms through the WessZumino (WZ) term that contains gradient of the
tachyon and the tachyon potential [2933]. We will argue that thanks to the existence of this WZ
term it is still possible to define the correct non-relativistic non-BPS Dp-brane as well.
As the further support of our result we will study the tachyon kink on the world-volume of
a non-relativistic non-BPS Dp-brane. Using the approach given in [34] we will construct the
singular tachyon kink on the world-volume of the non-relativistic non-BPS Dp-brane that can be
interpreted as a lower dimensional non-relativistic D(p 1)-brane. This result then suggests that
we can define non-relativistic non-BPS Dp-brane that has the same properties as its relativistic
version.
This paper is organised as follows. In Section 2 we introduce the bosonic part of the non-BPS
Dp-brane action in type IIA or type IIB theories. Then in Section 3 we will consider its nonrelativistic limit. In Section 4 we will study the tachyon kink on its world-volume. In Section 5
3 For review, see [18,19].
187
we review the basic properties of the supersymmetric form of the non-BPS Dp-brane in type IIA
theory. Finally in Section 6 we consider its non-relativistic limit.
2. Review of non-BPS Dp-brane
In this section we review the basic facts considering the bosonic part of the non-BPS Dp-brane
action [2023].4 The bosonic part of a non-BPS Dp-brane action takes the form5
S = SDBI + SWZ ,
SDBI = p d p+1 e V (T ) det A,

SWZ = p V (T ) dT C eF +B ,
(2.1)
where
Aij = i X M j X N GMN + i X M j X N BMN + Fij + i T j T ,
Fij = i Aj j Ai ,
(2.2)
where Ai , i, j = 0, . . . , p, and X M,N , M, N = 0, . . . , 9, are gauge and the transverse scalar

fields on the world-volume of the non-BPS Dp-brane and T is the tachyon field. V (T ) is the
tachyon potential that is symmetric under T T , has maximum at T = 0 equal to V (0) = 1
and has its minimum at T = where it vanishes. p is a tension
of the non-BPS Dp-brane
that is related to the tension of the BPS Dp-brane Tp as p = 2Tp . Finally GMN , BMN and
are background metric, B-field and dilaton respectively. We restrict to the background with
flat Minkowski metric GMN = MN = diag(1, 1, . . . , 1), with vanishing B-field and constant
dilaton 0 . In what follows we also include e0 into the definition of p .
Note also that SWZ given in (2.1) is WZ term for a non-BPS Dp-brane that expresses the
coupling of the non-BPS Dp-brane to the RamondRamond forms. In (2.1) denotes the worldvolume of the non-BPS Dp-brane and C collects all RR n-form gauge potentials (pulled back to
the world-volume) as

C=
(2.3)
C(n) .
n
The form of the WZ term was determined from the requirement that the RamondRamond charge
of the tachyon kink is equal to the charge of D(p 1)-brane [2933]. In fact it was argued in
[29,34] that the consistency requires:

p
V (T ) dT = Tp1 ,
where Tp1 is a tension of a BPS D(p 1)-brane.

4 General properties of the tachyon effective actions were also discussed in [2428].
5 We work in units (2 ) = 1. Note also that p is odd in type IIA theory while p is even in type IIB theory.
(2.4)
188
3. Non-relativistic limit of the non-BPS Dp-brane

In this section we define non-relativistic limit of a non-BPS Dp-brane. As was argued in [13]
the correct definition of the non-relativistic limit of BPS Dp-brane is based on the existence of
the constant background RamondRamond form Cp+1 and the BPS Dp-brane is extended in
the directions determined by the orientation of Cp+1 in order to cancel Dp-brane tension by the
coupling to the Cp+1 field. For non-BPS Dp-brane there is not suitable Cp+1 form with the rank
equal to the dimension of the world-volume of a non-BPS Dp-brane. Moreover the form of the
WZ term given in (2.1) contains tachyon so that we cannot proceed exactly in the same way as
in [13]. As we argued in the introduction we define the non-relativistic limit as the limit where
the world-volume scalar modes that parametrise directions transverse to the directions spanned
by the background RamondRamond form Cp are small. Then we also demand that the tachyon
and tachyon potential does not scale in the non-relativistic limit. More precisely let us presume
that the RamondRamond field Cp takes following form
C0 ...p1 = (1)p
0 ...p1 ,
, = 0, . . . , p 1.
(3.1)
Then (3.1) suggests following scaling of the world-volume variables

X = x ,
p = 2 NRp ,
Ai = Wi ,
X a = X a ,
Fij = fij ,
T = T,
(3.2)
where a, b = p, . . . , 9. Note that we have also presumed that the world-sheet gauge field Ai
scales in the same way as the transverse modes. The non-relativistic limit is defined as limit
0.
Now for (3.2) the matrix A takes the form
Aij = gij + fij + i T j T + 2 x a x b gab Bij + Cij + 2 Dij ,
Bij = gij + i T j T ,
Cij = fij ,
gij = i x j x ,
Dij = X a X b gab .
(3.3)
Inserting (3.3) into the DBI part of the tachyon effective action and considering the terms up to
orders 2 we obtain the non-relativistic Dp-brane in the form
NRp
d p+1 V (T ) det B
SNR = 2
NRp
d p+1 V (T ) det B
2

2
1 ij
1 1 ik 1 j l
a
b
j X i X gab B
fkj B
fli + O .
B
(3.4)
2
Before we turn to the non-relativistic limit in the WZ term let us study in more details the matrix
B. Its form suggests that it is natural to temporarily introduce the notation

Y I x , T , Y p = T , I, J = 0, . . . , p
(3.5)
so that we can rewrite B into the suggestive form
Bij = i Y I j Y J gI J EjI EjJ I J ,
pp = 1.
(3.6)
189
Using this notation it is easy to see that the divergent contribution in (3.4) can be written as

NRp
1
Sdiv = 2
(3.7)
V (T )E I0 E Ip ,
(p + 1)!
where E I EjI d j .
Let us study the non-relativistic limit in the WZ term. We insert the pull-back of Ramond
Ramond background form (3.1) into it and we obtain

NRp
1
SWZ = 2
V (T ) dT (1)p
0 ...p1 dX 0 dX p1
p!

NRp
1
V (T )
0 ...p1 dX 0 dX p1 dT
= 2
p!

NRp
V (T )
I0 ...Ip E I0 E Ip .
= 2
(3.8)
(p + 1)!
If we add (3.8) to (3.7) we obtain that all divergent contributions in the full non-relativistic nonBPS Dp-brane cancel. As the result we obtain finite non-relativistic non-BPS Dp-brane action

ij

NRp
fin
d p+1 V (T ) det B B1 j X a i X b gab
SNRp
=
2

1 1 ik 1 j l
B
fkj B
fli .
(3.9)
2
In the next section we will study the tachyon kink on the world-volume of the action (3.9).
4. Tachyon kink on the world-volume of non-relativistic non-BPS Dp-brane
In this section we will study the tachyon kink solution on the world-volume of the nonrelativistic non-BPS Dp-brane following [34].6
As in [34] we start with the following ansatz

T = f a p t ,

Wp = 0,
W = w ,

x = x , = 0, . . . , p 1,

X a = x a , a = p, . . . , 9,
(4.1)
where , , = 0, 1, . . . , p 1, are coordinates tangential to the world-volume of the kink. The
function f introduced in (4.1) satisfies
f (u) = f (u),
f (u) > 0 u,
f () =
(4.2)
but is otherwise an arbitrary function of the argument u. a is a constant that we shall take to
in the end. In this limit we have T = for p > 0 and T = for p < 0.
The first goal of our analysis is to show that the action (3.9) evaluated on the ansatz (4.1)
reproduces in the limit a the action for non-relativistic D(p 1)-brane

TNR(p1)
1
p
a
b
d det g g x x gab g f g f , (4.3)

SNR(p1) =
2
2
6 For generalisation of this approach to the curved background see [35,36].
190
where
g = x x ,
f = w w ,
, = 0, . . . , p 1,
, = 0, . . . , p 1,
(4.4)
and x a , a, b = p, . . . , 9, parametrise positions of non-relativistic Dp-brane transverse to its

world-volume. For letter purposes we also determine the equations of motion that arise from
(4.3)

1
1
x g det g g x a x b gab g f g f
2
2

a
b
det g g x g x x gab

+ det g g f g f g x
=0
(4.5)
that is the equation of motion for x . On the other hand the equation of motion for x a takes much
simpler form

det g g x b gba = 0.
(4.6)
Finally, we determine the equation of motion for w

det g g f g = 0.
Let us now insert the ansatz (4.1) into the matrix B defined in (3.3)

g + a 2 f 2 t t a 2 f 2 t
Bij =
a2f 2
a 2 f 2 t
and evaluate its determinant

B By B1 Bp
pp
det B =
Bp

0
= a 2 f 2 det g
Bpp
and its inverse matrix B1

1 pp
1 p
B
= g t t,
B
= t g ,
1 p
1
B
= g t,
B
= g .
(4.7)
(4.8)
(4.9)
(4.10)
Note that the form of the matrix B1 is exact for all a. If we now insert (4.1), (4.9) and (4.10)
into (3.9) and consider the limit of large a we obtain

TNR(p1)
1
fin
p
a
b
d det g g x x gab g f g f ,
SNRp =
(4.11)
2
2
where we have used the fact that

NRp

TNR(p1)
NRp
d p V f a p af p =
dm V (m) =
.
2
2
2
(4.12)
Comparing (4.3) with (4.11) we obtain the result that the tachyon kink reproduces the nonrelativistic D(p 1)-brane action. We also see that t ( ) does not appear in (4.11) and hence its
interpretation is unclear at present.
191
Note however that this result does not prove that the dynamics of the kink is governed by the
action (4.3). To do this we have to show, following [34], that any solution of the equations of
motion derived from (4.3) will produce a solution of the equations of motions derived from (3.9)
under the identification (4.1).
In order to establish this correspondence we firstly determine the equations of motions that
follow from (3.9). By varying the action (3.9) with respect to T we obtain

1 ij
1
1 1 ik 1 j l
a
b
j X i X gab B
fkj B
fli
V (T ) det B B
2
2

1 j i
1 ij
1 1 ik 1 j l
1
a
b
det B B
j X i X gab B
fkj B
fli
i j T B
2
2

il
kj

l V (T ) det B B1 k T B1 i X a j X b

im
nk
j l

+ m V (T ) det B B1 n T B1 fkj B1 fli
= 0.
(4.13)
In the same way we determine the equations of motion for x

j i
ij

1
i V (T ) j x B1
det B B1 j X a i X b gab
2

ik
j l
1
B1 fkj B1 fli
2

im

kj

m V (T ) B1 k x B1 j X a i X b gab

im

nk
j l

+ m V (T ) det B B1 n x B1 fkj B1 fli = 0
(4.14)
and the equations of motion for X a

ij

i V (T ) det B B1 j X b gba = 0.
(4.15)
Finally, we determine the equation of motion for Wi

j k
li

= 0.
j V (T ) det A B1 fkl B1
(4.16)
Now we will solve the equations of motions (4.13), (4.15), (4.16) with the ansatz (4.1). We start
our discussion with Eq. (4.15). Inserting (4.9), (4.10) and (4.1) to (4.15) we obtain

x

p V (T ) det B B1 X b gba + V (T ) det B B1 X b gba
t g x b gba + a [Vf ] det g g x b gba

= ap [Vf ] det g

+ af V (T ) det g g x b gba

= af V (T ) det g g x b gba ,
(4.17)
where we have used the fact that

V (f )af = tp V (f )af .
(4.18)
Now the final form of Eq. (4.17) implies that the equation of motion (4.15) is obeyed for the
ansatz (4.1) on condition that x a obeys (4.6). In the same way we proceed with (4.16). For i = p
192
we obtain

j k
lp

j V (T ) det B B1 fkl B1

= p [V af ] det g
t g f g t af V (T ) det g g f g t

= af V (T ) det g g f g t af V (T ) det g g f g t

= af V (T ) det g g f g t,
(4.19)
where in the first step we have used the fact that the expression proportional to p [V af ] vanishes
thanks to fact that g is symmetric while f is antisymmetric. In the second step we have again
used the antisymmetry of f so that
g f g t = 0.
(4.20)
Then the final form of Eq. (4.19) implies that the ansatz (4.1) solves the equation of motion (4.16)
on condition that the modes w , x a , x obey (4.7).
We can proceed in the same way in case when i in (4.16) is equal to and we get

j k
l

j V (T ) det B B1 fkl B1
(4.21)
= af V det g g f g .
This result again implies that the ansatz (4.1) obeys the equation of motion (4.16) on condition
that w , x , x a solve (4.7).
Finally we will analyse the tachyon equation of motion (4.13). To simplify the calculation we
use the fact that

j i
1 ij
Bjp = pi ,
Bpj B1 = pi
B
(4.22)
and then using (4.8) we obtain

B1
ip
i

B1 t =
1
i .
2
a f 2 p
(4.23)
Let us start with the expression on the first and the second line in (4.13)

1 ij
1
1 1 ik 1 j l
a
b
j X i X gab B
fkj B
fli
V (T ) det B B
2
2

1 j i
1 ij
1
1 1 ik 1 j l
a
b
i j T B
det B B
j X i X gab B
fkj B
fli
2
2

ij
ik
j l

1
1
= V (T ) det B B1 j X a i X b gab B1 fkj B1 fli
2
2

ij
ik
j l

1
1
1
i V (f ) 2 2 pi det B B1 j X a i X j gij B1 fkj B1 fli
2
2
a f
= 0,
(4.24)
where we have used (4.23). Moreover, using (4.23) in remaining terms in (4.13) we get that
all these terms vanish. We see that the ansatz (4.1) solves the equation of motion (4.13) for
arbitrary t ( ). In fact this is satisfactory result since we work with the action where the worldvolume diffeomorphism invariance is not fixed and hence p is equivalent to p + t ( ). This
fact also implies that it is natural to consider t ( ) as a parameter of the gauge diffeomorphism
transformations and hence it has no physical significance.
193
Let us outline the results derived above. We have shown that the dynamics of the massless
modes given in (4.1) is governed by the action (4.3). This result supports the interpretation of the
tachyon kink as the D(p 1)-brane.
5. Supersymmetric relativistic non-BPS Dp-brane
In this section we define supersymmetric form of a non-relativistic non-BPS Dp-brane. We
firstly review the basic facts about non-BPS D-branes in type IIA theory, following [20].7 Let
i , i, j = 0, . . . , p, are world-volume coordinates on a D-brane. On a non-BPS Dp-brane worldvolume we have a 32 component anti-commuting field that transforms as a Majorana spinor
of the 10-dimensional Lorentz group. We also denote M the ten-dimensional gamma-matrices
that can be chosen real by taking charge conjugation matrix C = 0 . These matrices obey the
anti-commutation relation
M N
= 2MN
,
(5.1)
with MN = (1, 1, . . . , 1). We also introduce 11 = 0 9 , (11 )2 = 1.
The supersymmetric generalisation of the DBI part of the effective action for a non-BPS Dpbrane has a form [2023]:

p+1
V (T ) det(Gij + Fij + i T j T ) = LDBI ,
SDBI = p d
(5.2)
where
iM = i X M + i M i ,
Gij = MN iM jN
(5.3)
and
Fij = Fij bij ,
Fij = i Aj j Ai .
There also exists the supersymmetric generalisation of the WZ term

SWZ = p V (T ) dT p = LWZ .
(5.4)
(5.5)
The form p expresses the coupling of the non-BPS Dp-brane to the RamondRamond
fields [21]. In particular, its bosonic part denotes the pull back of the RamondRamond fields
and the combination of the field strength F . This part vanishes for vanishing RamondRamond
background but there is a part of p that survives even in the absence of any RamondRamond
background [3740]. Since the explicit form of p is complicated we introduce, following [13]
(p + 1) form hp+1 such that
hp+1 = dp .
(5.6)
For type IIA non-BPS Dp-branes (p-odd) the forms b, hp+1 are equal to

i
b = i 11 M d M M d ,
2
7 For non-BPS D-brane in type IIB theory the situation is basically the same with difference in chirality of the
MajoranaWeyl fermions.
194
p1
(5.7)
,
2
where Tp1 is p 1 form. To define it, we introduce the formal sum of differential forms

Tp = eF CA ,
TA =
(5.8)
hp+1 = (1)n i d Tp1 d,
n=
p =even
where
CA = 11 +
1 2 1
1
+ 11 4 + 6 +
2!
4!
6!
(5.9)
and
= M M .
(5.10)
One can shown that the non-BPS Dp-brane action is invariant under rigid supersymmetry variations but it is not invariant under local symmetry transformations [20]. The absence of this
local symmetry implies that we cannot gauge away one half of the fermions and hence the worldvolume theory contains the correct number of massless degrees of freedom [20].
We can generalise the analysis presented in the previous section and switch on one more
coupling in the world-volume of non-BPS Dp-brane consistent with all symmetries of the action
[13]. From the spacetime point of view, this is equivalent to turning on a closed Ramond
Ramond p-form Cp which does not modify the supergravity equations of motion. Then the total
non-BPS Dp-brane action takes the form

S = d p+1 LDBI + LWZ + LCp ,
(5.11)
where
LCp = p V (T ) dT f Cp ,
(5.12)
where f is the pullback of Cp on the world-volume of non-BPS Dp-brane.

6. Non-relativistic non-BPS Dp-brane with presence of fermions
We derive the action for non-relativistic non-BPS Dp-brane from the supersymmetric form of
the relativistic non-BPS Dp-brane action. We again closely follow [13].
The non-relativistic limit is obtained by decoupling some charged light degrees of freedom
that obey non-relativistic dispersion relation from the full relativistic theory. This decoupling is
achieved in the similar way as in Section 2. In other words we demand that the scalar modes
that parametrise directions transverse to the background RamondRamond p-form Cp are small.
Considering fermionic degrees of freedom we will scale them in the way that is equivalent to the
scaling introduced in [11,13]
X = x ,
Fij = fij ,
= + + ,
X a = X a ,
Ai = Wi ,
p = 2 NRp ,
T = T,
C0 ...p1 =
0 ...p1 ,
(6.1)
where X M has been split in X , = 0, . . . , p 1, and X a , a = p, . . . , 9. The X are coordinates

of target space parallel to the orientation of Cp and X a are transverse to it. Finally is scaling
parameter that is sent to zero in the end.
195
The scaling of fermions depends on the splitting of the fermions under the matrix :
= .
(6.2)
The matrix is defined as

n+1
,
= (1)n+1 0...p1 11
2n = p 1
(6.3)
and it obeys following relations

= (1)
p+1
2
= (1)
p+1
2
11 = 11 ,
a = (1)
p1
2
a .
(6.4)
Note also that 2 = I. Now we insert (6.1) to the supertranslation 1-form given in (5.3) and we
obtain
= e + i2 + d+ ,
a = ua ,
e = e + i d ,
ua = dx a + 2i + a d ,
e = dx ,
x a = X a + i a + .
(6.5)
In the same way we determine the scaling of the form F

F = F (1) + 3 F (3) ,

i
i
(1)
F = f + (+ 11 d + i 11 d ) e d
2
2

i
i
+ i a 11 d ua a d+ + a d+ ,
2
2
1
F (3) = (+ 11 d + 11 d )+ d+
2

i
i
a
a
a
+ i + 11 a d+ u d+ + d+ .
2
2
(6.6)
Following [13] we will keep small but finite in the intermediate computations and only send
to zero in the end. For that reason we keep explicitly terms in the action that scale as 1 (that are
divergent) and terms that are independent on (that are finite). We also drop terms that scale as
since they cannot contribute when we take the limit 0 at the end of the analysis.
More precisely, if we insert (6.5) into G we obtain

(0)
(2)
Gij = iM jN MN = Gij + 2 Gij + O 3 ,
G(0)
ij = ei ej ,
Gij = 2i e + d+ + uai ubj ab ,

(2)
(6.7)
where we have restricted to the terms up to order 2 . Then if we insert (6.6) and (6.7) into the DBI
part of the tachyon effective action (5.2) we obtain supersymmetric form of the non-relativistic
non-BPS Dp-brane action in type IIA theory
NRp
SNR,DBI = 2
d p+1 V (T ) det B

ij (2) 1
ij (1)
kl (1)
NRp
d p+1 V (T ) det B B1 Gj i B1 Fj k B1 Fli

2
2
+ O(),
(6.8)
196
where
Bij = G(0)
ij + i T j T = ei ej + i T j T .
(6.9)
In order to carefully treat with the divergent term in (6.8) we perform the same simplification as
in Section 2 and temporarily write
Bij = eiI ejJ I J ,
eiI = ei ,
I, J = 0, . . . , p,
for = I = 0, . . . , p 1,
ei = i T .
With the help of (6.10) we can rewrite the divergent term in (6.8) as

NRp
NRp
V (T )
I0 ...Ip eI0 eIp
d p+1 V (T ) det B = 2
2
(p + 1)!

1
2 d p+1 Ldiv
DBI .
Now we insert (6.1) into the WZ term. As in [13] we easily obtain

NRp (2)
(0)
p hp+1 = 2 hp+1 + NRp hp+1 + O 2 .
The superficially divergent term can be determined as in [13] and takes the form
1
e1 ep1
1 ...p1 d
(p 1)!
1
=
1 ...p1
d e e1 ep1 .
(p 1)!
(6.10)
(6.11)
(6.12)
hp+1 = i d
(2)
Now we consider following expression

div
d d p+1 Ldiv = d d p+1 Ldiv
DBI + d V (T ) dT p

1
= NRp V (T ) d
I0 ...Ip eI0 eIp + dT hp+1

(p + 1)!

1
= NRp V (T )
J I1 ...Ip d eJ eI1 eIp + dT hp+1

p!
=0
(6.13)
(6.14)
using the fact that

dV ep =
dV p
e ep = 0,
dT
d(dT ) = 0
(6.15)
and also
1
d e e1 ep1
(p 1)!
1
= (1)(p+1)
1 ...p1
d e e1 ep1 dT
(p 1)!
1
=
J I1 ...Ip d eJ eI1 eIp ,
p!
dT
1 ...p1
(6.16)
where we have used the definition of the exterior derivative d and also the fact that for type IIA
non-BPS Dp-brane p is odd.
197
As the last term in (6.14) involves only the terms with fermions [13], this cancellation removes
div
the terms with fermions in Ldiv
DBI . There remains the pure bosonic term in LDBI that is
d p+1 Ldiv
DBI,bos = NRp V (T )
I ...I eI0 eIp .

(p + 1)! 0 p
(6.17)
This term can be cancelled by turning on a closed RamondRamond form Cp that gives an
contribution to the WZ term

1
div
LCp = NRp V (T ) dT
0 ...p1 dx 0 dx p1
p!

1
= NRp V (T )
(6.18)
I ...I eI0 eIp .

(p + 1)! 0 p
Then we obtain the finite part of the supersymmetric generalisation of the non-relativistic nonBPS Dp-brane action in the form

ij

SNR = NRp d p+1 V (T ) det B i B1 + ei j +

ij
ij
1 kl (1)
1
1
B
+ B1 uai ubj gab B1 Fj(1)
F
k
li
2
4

+ NRp V (T ) dT p(0) ,
(6.19)
(0)
where p is the non-relativistic WZ term that has the same form as in case of non-relativistic
BPS D(p 1)-brane.
Our goal is to shown that the world-volume action for non-relativistic D(p 1)-brane in type
IIA theory anises from the action (6.19) as the tachyon kink. Recall that the action for nonrelativistic D(p 1)-brane in type IIA theory has the form [13]

1
BPS
SNR
= TNR(p1) d p det g i g + e + + g u a u b gab
2

1
(1)
(1)
g F g F + TNR(p1) p(0) ,
(6.20)
4
where
g = e
e ,
e = e + i d ,
u a = d x a + 2i + a d ,
e = d x ,
x a = X a + i a + ,
(6.21)
and where F (1) has the same form as in (6.6) with the gauge field w .
In order to show that the tachyon kink on the world-volume of the supersymmetric nonrelativistic non-BPS Dp-brane describes the non-relativistic BPS D(p 1)-brane we should
perform the same analysis as in Section 2. As the first step we consider the ansatz

T = f a p t ,

Wp = 0,
W p , = w ,

p

X a p , = x a ,
x , = x ,

p , =
+ p , = + ,
(6.22)
198
and insert it to the matrix B so that we obtain

g + a 2 f 2 t t a 2 f 2 t
.
Bij =
a2f 2
a 2 f 2 t
(6.23)
Then we also get

det B = a 2 f 2 det g
(6.24)
pp
1 p
= g t t,
B
= t g ,
B1
1
1 p
= g t,
B
= g .
B
(6.25)
and

Then it is easy to see that when we insert (6.22), (6.23), (6.24) and (6.25) into (6.19), perform
the integration over p and use the relation (4.12) we obtain the non-relativistic D(p 1)-brane
action (6.20).
To show the complete equivalence we should check that the ansatz (6.22) solves the equations
of motion that arise from (6.19). However from the form of the ansatz (6.22) and corresponding
matrix B it is clear that the ansatz solves these equations on conditions that the massless modes
given there solve the equations of motion that follow from (6.20).
Acknowledgements
This work was supported in part by the Czech Ministry of Education under Contract No.
MSM 0021622409, by INFN, by the MIUR-COFIN contract 2003-023852, by the EU contracts
MRTN-CT-2004-503369 and MRTN-CT-2004-512194, by the INTAS contract 03-516346 and
by the NATO grant PST.CLG.978785.
References
[1] N. Seiberg, L. Susskind, N. Toumbas, Strings in background electric field, space/time non-commutativity and a new
non-critical string theory, JHEP 0006 (2000) 021, hep-th/0005040.
[2] R. Gopakumar, J.M. Maldacena, S. Minwalla, A. Strominger, S-duality and noncommutative gauge theory,
JHEP 0006 (2000) 036, hep-th/0005048.
[3] N. Seiberg, L. Susskind, N. Toumbas, Space/time non-commutativity and causality, JHEP 0006 (2000) 044, hepth/0005015.
[4] R. Gopakumar, S. Minwalla, N. Seiberg, A. Strominger, OM theory in diverse dimensions, JHEP 0008 (2000) 008,
hep-th/0006062.
[5] E. Bergshoeff, D.S. Berman, J.P. van der Schaar, P. Sundell, Critical fields on the M5-brane and noncommutative
open strings, Phys. Lett. B 492 (2000) 193, hep-th/0006112.
[6] T. Harmark, Open branes in spacetime non-commutative little string theory, Nucl. Phys. B 593 (2001) 76, hep-th/
0007147.
[7] J. Gomis, H. Ooguri, Non-relativistic closed string theory, J. Math. Phys. 42 (2001) 3127, hep-th/0009181.
[8] U.H. Danielsson, A. Guijosa, M. Kruczenski, IIA/B, wound and wrapped, JHEP 0010 (2000) 020, hep-th/0009182.
[9] U.H. Danielsson, A. Guijosa, M. Kruczenski, Newtonian gravitons and D-brane collective coordinates in wound
string theory, JHEP 0103 (2001) 041, hep-th/0012183.
[10] J. Brugues, T. Curtright, J. Gomis, L. Mezincescu, Non-relativistic strings and branes as non-linear realizations of
Galilei groups, Phys. Lett. B 594 (2004) 227, hep-th/0404175.
[11] J. Gomis, K. Kamimura, P.K. Townsend, Non-relativistic superbranes, JHEP 0411 (2004) 051, hep-th/0409219.
[12] J. Gomis, J. Gomis, K. Kamimura, Non-relativistic superstrings: A new soluble sector of AdS(5) S5 , JHEP 0512
(2005) 024, hep-th/0507036.
199
[13] J. Gomis, F. Passerini, T. Ramirez, A. Van Proeyen, Non-relativistic Dp-branes, JHEP 0510 (2005) 007, hep-th/
0507135.
[14] M. Sakaguchi, K. Yoshida, Non-relativistic AdS branes and NewtonHooke superalgebra, hep-th/0605124.
[15] K. Kamimura, T. Ramirez, Brane dualities in non-relativistic limit, JHEP 0603 (2006) 058, hep-th/0512146.
[16] E. Witten, D-branes and K-theory, JHEP 9812 (1998) 019, hep-th/9810188.
[17] P. Horava, Type IIA D-branes, K-theory, and matrix theory, Adv. Theor. Math. Phys. 2 (1999) 1373, hep-th/9812135.
[18] A. Sen, Non-BPS states and branes in string theory, hep-th/9904207.
[19] A. Sen, Tachyon dynamics in open string theory, Int. J. Mod. Phys. A 20 (2005) 5513, hep-th/0410103.
[20] A. Sen, Supersymmetric world-volume action for non-BPS D-branes, JHEP 9910 (1999) 008, hep-th/9909062.
[21] E.A. Bergshoeff, M. de Roo, T.C. de Wit, E. Eyras, S. Panda, T-duality and actions for non-BPS D-branes,
JHEP 0005 (2000) 009, hep-th/0003221.
[22] M.R. Garousi, Tachyon couplings on non-BPS D-branes and DiracBornInfeld action, Nucl. Phys. B 584 (2000)
284, hep-th/0003122.
[23] J. Kluson, Proposal for non-BPS D-brane action, Phys. Rev. D 62 (2000) 126003, hep-th/0004106.
[24] A. Sen, Time and tachyon, Int. J. Mod. Phys. A 18 (2003) 4869, hep-th/0209122.
[25] A. Fotopoulos, A.A. Tseytlin, On open superstring partition function in inhomogeneous rolling tachyon background,
JHEP 0312 (2003) 025, hep-th/0310253.
[26] A. Sen, Open and closed strings from unstable D-branes, Phys. Rev. D 68 (2003) 106003, hep-th/0305011.
[27] D. Kutasov, V. Niarchos, Tachyon effective actions in open string theory, Nucl. Phys. B 666 (2003) 56, hepth/0304045.
[28] V. Niarchos, Notes on tachyon effective actions and Veneziano amplitudes, Phys. Rev. D 69 (2004) 106009, hepth/0401066.
[29] K. Okuyama, WessZumino term in tachyon effective action, JHEP 0305 (2003) 005, hep-th/0304108.
[30] P. Kraus, F. Larsen, Boundary string field theory of the DD-bar system, Phys. Rev. D 63 (2001) 106004, hepth/0012198.
[31] C. Kennedy, A. Wilkins, RamondRamond couplings on braneantibrane systems, Phys. Lett. B 464 (1999) 206,
hep-th/9905195.
[32] M. Billo, B. Craps, F. Roose, RamondRamond couplings of non-BPS D-branes, JHEP 9906 (1999) 033, hepth/9905157.
[33] T. Takayanagi, S. Terashima, T. Uesugi, Braneantibrane action from boundary string field theory, JHEP 0103
(2001) 019, hep-th/0012210.
[34] A. Sen, DiracBornInfeld action on the tachyon kink and vortex, Phys. Rev. D 68 (2003) 066008, hep-th/0303057.
[35] J. Kluson, Tachyon kink on non-BPS Dp-brane in the general background, JHEP 0510 (2005) 076, hep-th/0508239.
[36] J. Kluson, Note about tachyon kink in nontrivial background, JHEP 0508 (2005) 032, hep-th/0506250.
[37] M. Cederwall, A. von Gussich, B.E.W. Nilsson, A. Westerberg, The Dirichlet super-three-brane in ten-dimensional
type IIB supergravity, Nucl. Phys. B 490 (1997) 163, hep-th/9610148.
[38] M. Aganagic, C. Popescu, J.H. Schwarz, D-brane actions with local kappa symmetry, Phys. Lett. B 393 (1997) 311,
hep-th/9610249.
[39] M. Cederwall, A. von Gussich, B.E.W. Nilsson, P. Sundell, A. Westerberg, The Dirichlet super-p-branes in tendimensional type IIA and IIB supergravity, Nucl. Phys. B 490 (1997) 179, hep-th/9611159.
[40] E. Bergshoeff, P.K. Townsend, Super-D-branes, Nucl. Phys. B 490 (1997) 145, hep-th/9611173.
Erratum
Erratum to: Scalar mode analysis of the warped

SalamSezgin model
[Nucl. Phys. B 747 (2006) 294]
Hyun Min Lee a , Antonios Papazoglou b,
a Department of Physics, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburg, PA 15213, USA
b Institut dAstrophysique de Paris, 98bis boulevard Arago, F-75014 Paris, France
Received 20 October 2006

Available online 10 November 2006
We correct a typo in [1] which changes the conclusions regarding the stability of the general
axisymmetric warped SalamSezgin vacuum.
There is a sign error in definition (63) of the mass eigenstates in [1], as has been pointed
out in Appendix B of [2]. The correct definition of the mass eigenstates is

1
= d + 1 + d 2 1 2 .
(63)
2
Therefore the coefficients K read
B + C(d + 1 + d 2 )
K =
(66)
,
1 + d2
and are always positive definite. Thus, the warped solution is always stable for any value of
4g/q. This corrects our comments regarding the warped vacuum stability in the abstract, the
introduction, the end of Section 4 and the conclusions.
References
[1] H.M. Lee, A. Papazoglou, Nucl. Phys. B 747 (2006) 294, hep-th/0602208.
[2] C.P. Burgess, C. de Rham, D. Hoover, D. Mason, A.J. Tolley, hep-th/0610078.
DOI of original article: 10.1016/j.nuclphysb.2006.05.001.

E-mail addresses: hmlee@andrew.cmu.edu (H.M. Lee), papazogl@iap.fr (A. Papazoglou).

doi:10.1016/j.nuclphysb.2006.10.025
Nuclear Physics B 765 [PM] (2007) 201239
The extended algebra of the SU(2)

WessZuminoWitten models
Pierre Mathieu, David Ridout
Dpartement de Physique, de Gnie Physique et dOptique, Universit Laval, Qubec, Canada G1K 7P4
Received 29 September 2006; accepted 17 November 2006
Abstract
The WessZuminoWitten model defined on the group SU(2) has a unique (non-trivial) simple current of
conformal dimension k/4 for each level k. The extended algebra defined by this simple current is carefully
constructed in terms of generalised commutation relations, and the corresponding representation theory is
investigated. This extended algebra approach is proven to realise a faithful (free-field-type) representation
of the SU(2) model. Subtleties in the formulation of the extended theory are illustrated throughout by the
k = 1, 2 and 4 models. For the first two cases, bases for the modules of the extended theory are given and
rigorously justified.
1. Introduction
1.1. Motivation and outline
This work initiates a series of articles devoted to the study of conformal field theories reformulated in terms of an extended algebra, in which the algebra is augmented by a simple current.
The extension studied here is that of the diagonal WessZuminoWitten models defined on
SU(2) [1,2]. These models are unitary, rational conformal field theories parametrised by a nonnegative integer k called the level. The irreducible (unitarisable) representations from which the
(chiral) conformal field theory is constructed are the level-k integrable highest weight modules
This work is supported by NSERC.
E-mail addresses: pmathieu@phy.ulaval.ca (P. Mathieu), darid@phy.ulaval.ca (D. Ridout).

doi:10.1016/j.nuclphysb.2006.11.019
202
P. Mathieu, D. Ridout / Nuclear Physics B 765 [PM] (2007) 201239
of
sl(2) [3,4], of which there are k + 1. We denote the corresponding highest weight states by
| , for = 0, 1, . . . , k, so labels the corresponding sl(2)-weight. Among the corresponding
primary fields (z), there is a unique (non-trivial) simple current k (when k = 0), whose
fusion rules take the form [5]
= k .
(1.1)
This simple current therefore acts as a permutation on the irreducible modules of the chiral
theory. In particular, = 0 , the identity.
The construction considered here consists of augmenting the
sl(2) symmetry algebra of the
theory by including the modes of the simple current. It is actually convenient to consider the corresponding multiplet of zero-grade descendant fields, rather than just the affine primary, whose
corresponding states form an irreducible representation of sl(2). We will therefore enlarge our algebra to include the modes of each zero-grade field (n) (z), n = 0, 1, . . . , k. Because these fields
each have conformal dimension k/4 (see Eq. (1.9)), this augmentation will in general lead to an
extended algebra defined by generalised commutation relations. Our aim is to precisely define
the structure of this extended algebra.
We note that such an investigation of this extended symmetry algebra has already been carried
out for k = 1 [6,7], under the guise of spinon bases. For higher values of k, our description differs
from the spinon formalism presented in [8,9], in that our algebra is generated by the simple
current = k rather than 1 , whose fusion rules with the different primary fields are more
complicated than a simple permutation. We present here a formalism which applies to the SU(2)
models for all k, providing a detailed model for similar treatments of more general extended
symmetries.
Reformulations of conformal field theories of the type considered here are of particular interest in that they embody a quasi-particle description of the space of states, that is, the states are
generated by the repeated action of the simple current modes. This action is subject to restriction
rules that can be interpreted as a generalised exclusion principle. Such quasi-particle reformulations have a number of immediate implications, the most obvious being that they shed new light
on the representation theory of the class of models under consideration. In the particular case of
the WessZuminoWitten models, this translates into a novel point of view on the representation
theory of affine Lie algebras.
The key point underlying this quasi-particle reformulation is the absence of singular vectors,
meaning that the restriction rules are built into the algebraic structure (the operator product expansions) of the extended theory. Establishing this very point in the present context is our main
result. It follows that whenever the generating function for the quasi-particles can be found in a
closed form, it will take the form of a positive-definite multiple sum, yielding a so-called fermionic character formula. We will derive such character formulae in several simple cases. This
contrasts with the usual formulae deduced via the subtraction of singular vectors, leading to the
non-positive-definite sums usually referred to as bosonic character formulae. Equating the fermionic and bosonic characters therefore results in non-trivial q-series identities that are of much
combinatorial interest.
These two motivations are clearly rather mathematical in nature. On a more physical note, we
stress that quasi-particle formulations share the underlying spirit of the condensed-matter/particle
formalism of conformal field theory, rather than that of the algebraic formalism focussing largely
on the representation theory of the chiral algebra. The quasi-particle interpretation is therefore
bound to be better adapted to the study of off-critical perturbations of the theory. In that vein, a
precise objective is to formulate the problem of computing correlation functions in a way that
203
is suited to the faithful (free of singular vectors) description we construct here. Moreover, we
expect that it could provide a starting point for the evaluation of off-critical correlation functions.
Along integrable lines, these should be governed by integrable differential equations. In brief,
we expect this formulation to allow us to cross the free-fermion barrier in the description of
correlation functions as solutions of solitonic equations, and hope to report on this elsewhere.
The extended algebra approach that we outline here was inspired by [10,11], which present
a description of the M(3, p) minimal models from the point of view of the extended algebra
generated by 2,1 . Its extension to all minimal models will be reported in [12].
1.2. Organisation of the article
The construction of the extended algebra of the SU(2) WessZuminoWitten models is presented in Section 2. As mentioned above, this algebra combines the usual affine currents with
the k + 1 zero-grade components of the simple current. The first issue we address is the question
of the mutual locality of these two types of currents. Overlooking this point quickly leads to
subtle inconsistencies in the extended theory. It turns out that these considerations, and in particular imposing a natural definition for the adjoint of the simple current modes, force (somewhat
surprisingly) the components of this simple current to anticommute with certain of the affine
generators. This is shown to be compatible with the Jacobi identities and forms the conclusion
of Section 2.1.
The extended algebra is essentially defined by the operator product expansions (m) (z) (n) (w)
of the simple current component fields, as the other generators, the affine currents, will be recovered (roughly speaking) as bilinears in these fields. The first step towards computing the different
fields appearing in these expansions is to consider the corresponding states. These are computed
for the first few levels, and are expressed directly in terms of the modes of the affine generators
in Section 2.2. We then consider the constraints imposed by associativity in Section 2.3. The key
point here is to show that the fields of our operator product expansions can be constructed from
these states in the usual way (the state-field correspondence).
More precisely, we have to address the following technical issue. In the analysis of the
M(3, p) models in [10], it was observed that the operator product expansion corresponding
2,1 =
to the fusion rule 2,1 2,1 = I must necessarily involve an operator S satisfying S
The question of the potential necessity of a similar operator in the present context
(1)p 2,1 S.
is addressed here, but is answered in the negative. Nevertheless, we find it convenient to introduce
trivial S-type
operators, denoted by Sm,n into each of our operator product expansions, whose
eigenvalues keep track of various constants that appear in the extended theory. The trivial nature
of these operators reflects the fact that they commute with the modes of the current generators
and those of the simple current components. With these technicalities settled, the path to the formulation of the generalised commutation relations defining the extended algebra is completely
paved. These relations are displayed in Section 2.4.
The representation theory of this extended algebra is developed in Section 3. Previously, in
order to write down the generalised commutation relations, we expanded the simple current components into modes acting on the vacuum module. To formulate this expansion in full generality,
we introduce the monodromy charge (Section 3.1). With this tool in hand, the notion of highest weight states for the extended algebra is defined (Section 3.2), and the Verma modules for
the extended algebra are constructed from the action of the simple current modes on such a
state. Within each extended Verma module, we readily identify two affine highest weight states.
204
Finally, we evaluate the eigenvalues (necessary for computation) of the Sm,n on these Verma
modules in Section 3.3.
We then illustrate the general theory we have developed with two examples, k = 2 and k = 4,
worked out in detail (Section 4). When k = 2, the dimension of the three components of the
simple current is 1/2, which suggests a fermionic interpretation. Linear combinations of these
three components are readily seen to correspond to three independent free fermions. It is actually
well-known that the affine
sl(2)2 algebra can be represented in terms of three free fermions
[13] (see also [5, Section 15.5.5 and Ex. 15.16]). We, however, show something more precise:
The k = 2 extended algebra is isomorphic to the tensor product of three copies of the extended
symmetry algebra of the Ising model (defined by its own simple current), which is the symmetry
algebra of a free fermion. As the spectra (the highest-weight states) of the two theories also
match, this suggests an equivalence of conformal field theories.
Another special case of interest is k = 4, for which the extended algebra is shown to be isomorphic to the (affine) symmetry algebra of the SU(3) WessZuminoWitten model at level 1,
sharpening the correspondence
sl(2)4
sl(3)1 [14] (see also [5, Section 17.5.2]). However, the
comparison of the spectra is more delicate in this case, because we must take the non-diagonal

sl(2)4 model in this correspondence. We show that the monodromy charge automatically accounts for this subtlety in our formalism.
The central result of this paper is presented in Section 5. It states that the representations
of the extended algebras we have constructed are faithful. In other words, the Verma modules
have no singular vectors, hence are irreducible. This result is established in various steps. First,
we demonstrate that not only are there two affine highest-weight states in each extended Verma
module, but that there are no more than two. For k = 1 and k = 2, we then show explicitly that
there are no singular vectors in the (extended) Verma modules by demonstrating that the primitive singular vectors of the affine algebra vanish identically. These calculations then motivate the
general argument establishing the absence of singular vectors as presented in Section 5.3. This
allows us to strengthen our conclusions concerning the cases k = 2, 4 and their connection to the
theory of three fermions and the
sl(3)1 model respectively. The isomorphisms, previously established for the symmetry algebras, are now extended to isomorphisms of irreducible modules.
Finally, a basis of states is rigorously derived for the two simplest models, k = 1 and 2. The
generating functions for these bases are constructed explicitly and shown to agree with known
fermionic expressions for the corresponding characters. We defer the derivation of the analogous
basis of states for general k to a sequel [15].
1.3. Basics and notation
The SU(2) models are unitary, rational conformal field theories parametrised by a nonnegative integer k called the level. The chiral algebra of these models may be described as
sl(2) associated to the complexifollows. We first define [16] the (untwisted) affine Lie algebra
fication sl(2) of su(2). This is the (graded) complex Lie algebra

sl(2) = sl(2) C t, t 1 spanC {K} J
(1.2)
spanC {L0 }
(
J denotes a semidirect sum) whose non-trivial commutation relations are

L0 , J t m = mJ t m and
1

J t m , J 2 t n = J 1 , J 2 t m+n + m J 1 , J 2 m+n,0 K.
(1.3)
205
Note that K is central. Throughout we will follow the standard practice of abbreviating J t n
as Jn . Here, (J 1 , J 2 ) denotes the Killing form of sl(2), which we normalise so that

1
J 1 , J 2 = tr ad J 1 ad J 2 ,
(1.4)
4
where ad() is the usual adjoint action of sl(2) on itself. The chiral algebra is obtained from the
universal enveloping algebra of
sl(2) by dropping L0 (it will shortly be recovered through the
Sugawara construction) and restricting K to have eigenvalue k. Formally, we take the universal
enveloping algebra of the derived subalgebra of
sl(2) and form the quotient by the ideal generated
by K k id. We denote this chiral algebra by1 U(
sl(2)k ).
We denote the elements of this basis by {E, H, F }, so that
[H, E] = 2E,
[E, F ] = H
and [F, H ] = 2F.
(1.5)
The normalisation of the Killing form then gives

(E, E) = (E, H ) = (H, F ) = (F, F ) = 0,
(E, F ) = 1
and (H, H ) = 2.
(1.6)
The corresponding affine modes are therefore denoted by En , Hn and Fn .

For a given level k, the Virasoro symmetry is realised as a subalgebra Virc of the chiral
algebra through the Sugawara construction (see for example [5, Eq. (15.70)]):
1
1
Ln =
(1.7)
: Hm Hnm + Em Fnm + Fm Enm :.
2(k + 2)
2
mZ
Through this construction, the L0 of Eq. (1.3) is recovered, and is associated to the Virasoro
energy operator. The central charge of our conformal field theory is then found to be
3k
(1.8)
.
k+2
As previously indicated, the highest weight states are denoted by | , for = 0, 1, . . . , k, where
labels the sl(2)-weight. The conformal dimensions of these states are given by
c=
h =
( + 2)
.
4(k + 2)
(1.9)
|0 is the vacuum of the theory, and will be denoted in what follows by |0.
We choose a basis of these zero-grade descendant states carefully in order to simplify later
calculations. First, we normalise the highest weight states

| (0)
(1.10)
1 We remark that it is far more commonplace in the physics literature to denote this chiral algebra by su(2)
k . Whilst
the name we ascribe an object is largely arbitrary, there is much to be said for maintaining consistent notation within a
field. That said however, we prefer to use the more precise notation given above as in the mathematics literature. The
reason why we explicitly indicate the universal enveloping algebra is that our chief object of study is a certain extension
of this chiral algebra, and this extension is not a Lie algebra. We want to treat these extended algebras on an equal footing
with more familiar chiral algebras, and so we emphasise chiral algebras as associative algebras, not Lie algebras. The
reason for writing sl rather than su is really a matter of tastesu(2) does not, strictly speaking, contain the root vectors
or coroot that form the most convenient basis for our study.
206
to have length 1, and then define

(n+1)
1/2 (n)

= (n + 1)( n)
F0 ,
(1.11)
(n)
for all n = 0, 1, . . . , 1. The sl(2)-weight of | is therefore 2n:

(n)
(n)
H0 = ( 2n) .
(1.12)
A simple induction argument now shows that

1/2 (n+1)

F0 (n) = (n + 1)( n)
(n+1)
1/2 (n)

.
= (n + 1)( n)
E0
(1.13)
It follows from these definitions that the zero-grade descendant states are all normalised:
(n+1) (n+1)
1/2
(n+1) (n)
(n) (n)

F0
(1.14)
= (n + 1)( n)
= ,
since F0 = E0 .
2. The extended algebra
Before starting the analysis of the extended algebra per se, we first need to clear up a little
technical issue, namely the analysis of the mutual locality of the two families of fields that we
wish to combine into an extended chiral algebra.2
2.1. Preliminaries
Recall [5] that the zero-grade descendant fields may be characterised through their affine
operator product expansions,
J (z) (m) (w) =
(J0 (m) )(w)

+ ,
zw
(2.1)
where J (z) is an (arbitrary) affine current and (J0 (m) )(w) denotes the (zero-grade) descendant
field corresponding to J0 | (m) . We would therefore expect to obtain commutation relations of
the form

Jr , s(m) = J0 (m) r+s .
(2.2)
However, it is not clear whether this expectation is correct, because we have not justified the
assumption that J (z) and (m) (w) commute (are mutually bosonic) in operator product expansions.
Let us suppose therefore that
J (z) (m) (w) = J,m (m) (w)J (z),
(2.3)
2 Chiral algebras are usually defined in relation with modular invariants and pertain to algebras spanned by fields with
integer dimension. Here we understand chiral algebra in a generalised sense, without reference to a modular invariant
and thus without this dimensional restriction.
for some constants J,m . The corresponding mode relations are then
y
q

Jr , s(m) Jr s(m) J,m s(m) Jr = J0 (m) r+s .
207
(2.4)
We set Js , Jr K = 1
J,m JJr , s K, and consider the constraints imposed by the corresponding
Jacobi identity:
q a q b (m) yy
q
yy
q (m)
q (m)
y
Jr , Js , t
+ J a ,m Jsb , t , Jra + J a ,m J b ,m t , Jra , Jsb = 0. (2.5)
(m)
(m)
This becomes
J a ,m J b ,m a b

a b (m)
J0b J0a (m) r+s+t =
f J J J c J0c (m) r+s+t ,
J0 J 0
r+s+t
J c ,m
c
upon expanding the affine commutator in terms of the structure constants of sl(2):
a b J aJ b
a b
c
Jr , J s =
f
J c Jr+s + r J , J r+s,0 k.
(2.6)
(2.7)
The constraints are therefore that

J a ,m J b ,m = J c ,m
if f J
aJ b
Jc
= 0.
(2.8)
Up to permutation, the non-vanishing structure constants are f EF H = 1 and f H E E =

F = 2 (Eq. (1.5)), so the constraints give
f FH
H,m = 1 and E,m = 1

F,m .
(2.9)
In other words, the Jacobi identity requires H (z) to be mutually bosonic with each (m) (w), but
does not determine the analogous behaviour of E(z) and F (z). The most natural solution of these
constraints is to (arbitrarily) choose E,m = F,m = 1, in agreement with our original expectation. However, as we will shortly see, this choice must be consistent with another consideration,
and the natural choice in that case leads to a contradiction with the choice made above. We will
therefore proceed in full generality so as to completely understand this intriguing situation.
A chiral algebra comes equipped with an involutive antilinear anti-automorphism defining
the adjoint on representations. In other words, chiral algebras are not just (graded) associative
algebras, but (graded) -algebras. We therefore need to be able to extend the adjoint of
sl(2) to
the modes r(n) in a manner consistent with the algebraic structure of the extended theory. The
full structure of the extended theory will not be derived until Section 2.4, but the part we are
currently investigating places strong constraints on this extended adjoint nonetheless.
Consideration of the conformal dimension and sl(2)-weight force
(m)
(km)
= m s ,
s
(2.10)
where m is a constant (which we take to be independent of s). The involutive nature of the
adjoint then requires that
km = 1.
m
(2.11)
The natural choice in this case is therefore to take m = 1 for all m. However, this leads to the
promised contradiction with E,m = F,m = 1, which is detailed in the following computation:
208
Er , s(m)

1/2 (m1)
1/2 (k+1m)
r+s
= m(k + 1 m)
= m(k + 1 m)
rs

(km)
= s(m) , Er .
= Fr , s
(2.12)
This demonstrates that m = 1 does not define an anti-automorphism of the extended theory
when E,m = F,m = 1.
By repeating this computation (and others) for general m and E,m , F,m , we find that the
extended adjoint thus defined avoids this contradiction precisely when
m1 = E,km m
and m+1 = F,km m .
(2.13)
Of course, this does not itself imply that we have constructed an anti-automorphism of the extended algebra (we will not even finish the derivation of this algebra until Section 2.4), merely
that the above constraints are necessary to obtain an anti-automorphism. We note that these constraints imply that E E,m and F F,m are independent of m:
1
= 1
E,km = m1 m
F,k+1m = E,k+1m ,
(2.14)
by Eq. (2.9). It follows that m = (F )m 0 , hence that

(E )k = |0 |2 > 0,
(2.15)
by Eq. (2.11).
Setting E,m = F,m = 1 is therefore consistent with m = (1)m 0 if k is even, but not when
k is odd, and never with m = 1. Likewise, the choice m = 1 is consistent with E,m = F,m =
1 for all k, but not with E,m = F,m = 1. In other words, the most convenient choices we can
make to describe the algebra of -modes are the following:
(m)
(km)
(1) For all k, define (s ) = s

so that the (m) (w) are mutually bosonic with respect to
H (z) but must be mutually fermionic with respect to E(z) and F (z). The defining relations
(m)
between the Hr and the s are therefore commutation relations whereas those between the
(m)
Er or Fr and the s are anticommutation relations.
(m)
(km)
(2) Define (s ) = (1)km s
so that the (m) (w) can be mutually bosonic with respect
to all affine fields when k is even, but mutually fermionic with respect to E(z) and F (z) when
(m)
k is odd. The defining relations between the Hr and the s are therefore commutation rela(m)
tions whereas those between the Er or Fr and the s are commutation or anticommutation
relations according to the parity of k.
We prefer to work with choice (1) in which the adjoint and defining relations are independent
of k. However, in specific examples with k even, it is often preferable to work with choice (2)
because the formulation involving bosonic operator product expansions facilitates comparison
with other theories (as we will do in Section 4 for example).
When k is even, the equivalence of these two choices is easy to establish with the help3 of an
auxiliary operator F satisfying
F Er = Er F,
F Hr = Hr F,
F Fr = Fr F
and
F 2 = id .
3 We could also establish this with Klein factors, but not as elegantly as in the present development.
(2.16)
209
Distinguishing the -modes under choice (2) by tildes, this equivalence may be explicitly described by
s(m) = s(m) F s(m) = s(m) F.
(2.17)
Then (for instance),

1/2 (m1)

r+s F
Er s(m) + s(m) Er F = m(k + 1 m)

1/2 (m1)
(m)
(m)
Er s s Er = m(k + 1 m)
r+s .
(2.18)
In this way, we transform anticommutation relations into commutation relations.

(m)
We also need to settle the manner in which F commutes with the s . Supposing that
(m)
(m)
Fs = m s F , we have

1/2 (m1)

Fr+s
F Er s(m) + s(m) Er = m(k + 1 m)

1/2 (m1)
(m)
(m)
m Er s s Er = m1 m(k + 1 m)
r+s .
(2.19)
Therefore m = m1 . Considering the corresponding relations for the other affine modes
yields no further constraints, hence we are free to choose m = (1)m . In other words, F and
(m)
s may be taken to commute when m is even, but anticommute when m is odd.
Finally, we may take F to be self-adjoint. This follows from
(km) (km)
= (1)m s(m) = (1)m s(m) F = F s(m) ,
F s(m) = s F = s
(2.20)
remembering that k is assumed to be even.

To summarise then, we have an extension of the chiral algebra U(
sl(2)k ) by the modes s(m)
of the simple current. We choose to extend the adjoint anti-automorphically as in choice (1),

s(m)
(km)
= s
,
(2.21)
which determines the commutativity (mutual locality) of the operator product expansions to be
E(z) (m) (w) = (m) (w)E(z),
H (z) (m) (w) = (m) (w)H (z)
and
F (z) (m) (w) = (m) (w)F (z).
(2.22)
The corresponding mode relations are therefore

1/2 (m1)

r+s ,
Er , s(m) = m(k + 1 m)

1/2 (m+1)

(m)
Hr , s(m) = (k 2m)r+s and Fr , s(m) = (m + 1)(k m)
r+s ,
(2.23)
where {, } denotes an anticommutator. The equivalence of this extension with that which would
be obtained through choice (2) (when k is even) is effected by introducing a self-adjoint operator
(m)
F , as in Eq. (2.17), which squares to id, commutes with the Hr and the s with m even, and
(m)
anticommutes with the Er , Fr , and the s with m odd.
To complete the description of this extended chiral algebra, we have to determine the algebraic
(m)
(n)
relations between the modes r and s . These relations are derived from the corresponding
operator product expansions whose computation we now turn to.
210
Fig. 1. The first few grades of a generic vacuum

sl(2)k -module (this triangular pattern continues until we arrive at the
affine singular vectors at grade k + 1). The states are written in the form (, h) where denotes the sl(2)-weight, h
denotes the conformal dimension, and the subscript denotes the multiplicity.
2.2. Operator product expansions

Let us consider the operator product expansions
(m)
(z)
(n)
(w) =
A(m,n;j ) (w)(z w)j k/2 ,
(2.24)
j =0
for m, n = 0, 1, . . . , k, the unknown fields A(m,n;j ) (w) have conformal dimension j . Since
= 0 , the fields A(m,n;j ) (w) may be expressed in terms of fields descended from the identity
field, and we will compute the form of these expressions by considering the corresponding states:

(m,n;j )
dz
(m)
(n)
A
= lim (m) (z) (n) (w)(z w)k/2j 1
|0 = k/4j k/4 |0.
(2.25)
w0
2i
w
To do this, we require some structural knowledge of the vacuum

sl(2)k -module. We indicate this
structure in Fig. 1 for the first few grades. The three states at grade 1 will be denoted by |E, |H
and |F , as they are the states corresponding to the three affine current fields.
Taking j = 0 in Eq. (2.25), we see that |A(m,n;0) must be a state in the vacuum
sl(2)k -module
of conformal dimension 0 and sl(2)-weight 2(k mn). It is therefore proportional to m+n,k |0,
and the proportionality constant may be evaluated by comparing

(m,n;0)

(m) (n)
0 A
(2.26)
= 0|k/4 k/4 |0 = (km) (n) = m+n,k
and 0|0 = 1. It follows that |A(m,n;0) = m+n,k |0.
If m + n = k 1, then |A(m,n;1) is a state of conformal dimension 1 and sl(2)-weight 2,
hence is proportional to |E. Again, we evaluate the constant of proportionality by comparing
E|E = 0|F1 E1 |0 = 0| H0 + k|0 = k,
where we have used Eq. (1.3), and
(m,n;1)

1/2 (m+1) (n)
(m)
E A
k/4
= 0|F1 k/41 (n) = 0| (m + 1)(k m)

1/2
km1,n ,
= (m + 1)(n + 1)
where we have used Eqs. (1.13) and (2.23). It follows that
(m,n;1) 1
1/2
A
= (m + 1)(n + 1)
|E if m + n = k 1.
k
(2.27)
(2.28)
(2.29)
211
Similar calculations give

(m,n;1)
1
A
= (n m)|H
2k
if m + n = k,
(2.30)
if m + n = k + 1.
(2.31)
and
(m,n;1) 1
A
= [mn]1/2 |F
k
Clearly |A(m,n;1) vanishes otherwise.

We can repeat these computations to determine |A(m,n;2) . However, we note that the relevant
states in Fig. 1 have non-trivial multiplicities, and some of them will be singular when k = 1. It
is therefore appropriate to choose bases of the corresponding state spaces accordingly. We will
use the following bases for the spaces of weights 2, 0 and 2:

L2 |0, H2 |0, 02 |0 and F2 |0,
E2 |0, +
(2.32)
2 |0 ,
2 |0 ,
where
+
2 = H1 E1 + E1 H1 ,
02
2 = H1 F1 + F1 H1
= H1 H1 E1 F1 F1 E1 .
and
(2.33)
It is not hard to check that each of these bases consists of mutually orthogonal states, and that
they have the nice property that the singular vector is a basis element (the 2 |0) when k = 1,
hence can be easily removed.
as before, omitting the details. The non-zero results are as
We therefore compute the |A(m,n)
2
follows:
m + n = k 2:
m + n = k 1:
m + n = k:
m + n = k + 1:
m + n = k + 2:
(m,n;2)
A
=

1/2
1
E1 E1 |0,
(m + 1)(m + 2)(n + 1)(n + 2)
2k(k 1)

(m,n;2) 1
1/2 1
nm +
A
= (m + 1)(n + 1)
E2 +
2 |0,
k
2
4(k 1)

(m,n;2)
k
+
2
n
m
1
mn
A
=
L2 +
H2 +
02 |0,
6
4k
12 2k(k 1)

(m,n;2) 1
n
m
1
1/2
A
= [mn]
F2 +
|0,
k
2
4(k 1) 2
(m,n;2)

1/2
1
A
=
F1 F1 |0.
(2.34)
m(m 1)n(n 1)
2k(k 1)
When k = 1, these results are modified by simply dropping the states whose coefficients diverge
(these correspond to singular vectors).
This can obviously be extended to |A(m,n;j ) for j > 2, though with significant increase in
effort as j increases. We would like to point out one important (and useful) feature of this
extension: For arbitrary m and n, the first |A(m,n;j ) which does not identically vanish is that
with j = |k m n|. This may be seen by noting that the vacuum descendant of sl(2)-weight
2(k m n) and minimal conformal dimension is obtained by repeatedly acting on the vacuum
j
j
with either E1 or F1 . This first state is therefore proportional to either E1 |0 or F1 |0 (depending on whether the weight is positive or negative). This proportionality constant is easily
212
determined (as above), giving

(m + n)! k n k m 1/2 kmn
E1
|0
(m,n;|kmn|)
k!
m
n
A
=

1/2
m
n
(2k m n)!
m+nk
F1
|0
k!
kn km
if m + n k,
if m + n k.
(2.35)
Note that these constants are invariant under (m, n) (n, m) and (m, n) (k m, k n).
It is tempting to conclude from the above results that (for example)
A(m,n;2) (w) =

1/2
1
:F (w)F (w):
m(m 1)n(n 1)
2k(k 1)
if m + n = k + 2.
(2.36)
However, such conclusions are not yet justified, because we have no a priori guarantee that these
conclusions will preserve the associativity of the operator product expansions. Associativity is
a fundamental requirement of a consistent conformal field theory, but it need not be automatically satisfied. As mentioned in Section 1.2 (see also [12]), it is sometimes necessary (and often
convenient) to introduce an operator S into the operator product expansion of two fields in order
to ensure the associativity of the operator product algebra. This will not be in conflict with the
state-field correspondence if this S-operator acts as (a multiple of) the identity on the vacuum
module. We therefore turn to a careful (and rigorous) study of associativity.
2.3. Associativity
We begin with the commutativity of the radially-ordered product. As noted above, the first
contributing term in the operator product expansion (2.24) corresponds to j = |k m n|. It
follows that the most general form of commutativity may be expressed as
(z w)k/2|kmn| (m) (z) (n) (w) = m,n (n) (w) (m) (z)(w z)k/2|kmn| ,
(2.37)
where m,n = 1
n,m is a phase (compare Eq. (2.22) for an example exhibiting non-trivial such
phases). We do not cancel the powers of z w and w z above as the resulting power of 1
will not be integral when k is odd.
Substituting Eq. (2.24) into Eq. (2.37), we derive the relation
A(m,n;j ) (w) = m,n (1)kmn
j

(1)j i
i=0
i!
i A(n,m;j i) (w).
(2.38)
Putting j = |k m n|, only one term (i = 0) contributes on the right-hand side, and the corresponding relation between states becomes

(m,n;|kmn|)

A
= m,n A(n,m;|kmn|) .
(2.39)
It now follows from Eq. (2.35) that the two states appearing in this equation are identical and
non-vanishing, hence m,n = 1 for all m and n. We therefore drop these phases in what follows.
We now consider the consistency of the operator product algebra. In particular, we investigate
its associativity by analysing what happens when we use Eq. (2.37) twice to obtain

(m) (x) (n) (z) (p) (w)(x z)k/2 (x w)k/2|kmp| (z w)k/2|knp|
z
213
dx
2i
dx
,
2i
z
(2.40)
where Z is arbitrary. Inserting our operator product expansion (2.24) and evaluating the
contour integral gives
=
(p) (w) (m) (x) (n) (z)(x z)k/2 (w x)k/2|kmp| (w z)k/2|knp|

1

k/2 |k m p| (m,n;j )
(z) (p) (w)(z w)k|kmp||knp|+j +1
A
j 1
j =0

(1) j 1 (p) (w)A(m,n;j ) (z)(w z)k|kmp||knp|+j +1 = 0.
(2.41)
Note that the common exponent of z w and w z is an integer, so we may factor out this power
of z w, leaving the second term with a (well-defined) relative sign: (1)k|kmp||knp| =
(1)kmn . Note also that this equation now has the form of a matrix equation M . v = 0, where
M is lower-triangular (its entries are the binomial coefficients appearing above) with ones on the
diagonal, hence invertible. We therefore conclude that associativity requires that
A(m,n;j ) (z) (p) (w) = (1)kmn (p) (w)A(m,n;j ) (z).
(2.42)
The field A(m,n;j ) (z) has sl(2)-weight 2(k m n), so it must be decomposable into terms
containing an even or an odd number of E(z) and F (z) (and their derivatives), according as to
whether k m n is even or odd (respectively). Comparing Eq. (2.42) with Eq. (2.22) now,
we see that this is exactly the commutativity that we would expect, hence it is not necessary
to introduce S-operators to preserve associativity. Nevertheless, it turns out to be convenient
to introduce operators Sm,n into our simple current operator product expansions (we will see
why shortly). The lack of necessity for introducing these operators translates into the fact that
they must commute with every mode of the extended symmetry algebra (to be defined next),
hence will act as multiples of the identity on the Verma modules of the extended algebra (to be
introduced in Section 3.2).
2.4. Generalised commutation relations
Having settled the question of associativity, it is now convenient to redefine the operator product expansion (2.24) to have the form
(m) (z) (n) (w) = Sm,n

j =0
A(m,n;j ) (w)(z w)j k/2
Sm,n
m,n (z, w).
(z w)k/2
(2.43)
In other words, we explicitly factor out the Sm,n from the definition of the fields A(m,n;j ) (w).
The eigenvalues of the Sm,n on the vacuum are given by Eq. (2.35) as

(m + n)! k n k m 1/2
|0
if m + n k,
k!
m
n
Sm,n |0 =
(2.44)

1/2
m
n
(2k m n)!
|0 if m + n k,
k!
kn km
214
and the m,n (z, w) which do not vanish to second order are (compare Section 2.2):
m + n = k 2:
m + n = k 1:
:E(w)E(w):(z w)2 + ,

nm
+ (w) (z w)2 + ,
E(w)(z w) + 12 E(w) + 4(k1)
m + n = k:
1+
nm
H (w)(z w)
1
2k
k+2
+ 6 T (w) + nm
4k H (w) + 12
0

mn
(w) (z w)2 + ,
2k(k1)

nm
(w) (z w)2 + ,
F (w)(z w) + 12 F (w) +
4(k 1)
(2.45)
:F (w)F (w):(z w)2 + .
m + n = k + 1:
m + n = k + 2:
When k = 1, these series are modified by simply ignoring all fields whose coefficients contain the
divergent factor (k 1)1 . We note that Eq. (2.35) implies that the leading term of m,n (z, w) is
kmn

:E(w) E(w):(z w)kmn
m+nk
or

:F (w) F (w):(z w)m+nk ,
(2.46)
according as to whether m + n k or m + n k (respectively).

The generalised commutation relations are obtained from these operator product expansions
by evaluating

dz dw
(m,n)
Rr,s ( ) =
(2.47)
(m) (z) (n) (w)zrk/4+ 1 w s+k/41 (z w)k/2
,
2i 2i
0 w
where Z is arbitrary, in two distinct fashions: Expanding as a difference of two contours,

dz dw
dz dw
dz dw
,
(2.48)
2i 2i
2i 2i
2i 2i
0 w
0 0
|z|>|w|
0 0
|z|<|w|
and using Eq. (2.37) (with m,n = 1), or expanding directly using Eq. (2.43). This procedure
yields the set of generalised commutation relations (parametrised by , m, n, r and s):

k/2 + 1 (m) (n)
(n)
(m)
r s+ (1)kmn+ s+k/2 rk/2+ +

=0
= Sm,n

1
j =0

r k/4 + 1 (m,n ;j )
Ar+s .
1j
(m,n;j )
(2.49)
refer to the fields appearing in Eq. (2.43) (with the Sm,n explicitly
Note that the modes Ar+s
factored out), not those of Eq. (2.24).
We observe that if k is even, we may choose = k/2, obtaining a genuine commutation or
anticommutation relation (according as to the parity of m n). For k odd however, every
generalised commutation relation will have infinitely many terms on the left-hand side. We also
observe that for 0, the right-hand side of the generalised commutation relations clearly
vanish. When positive, is a measure of the depth at which the operator product expansion
(2.43) is probed. Indeed, represents the (maximal) number of terms (singular or regular) of the
operator product expansion which contribute to the generalised commutation relation. As these
relations will be frequently used in what follows, we will find it convenient to employ a shorthand
215
notation = to indicate that the generalised commutation relation of order has been used to
obtain an equality.
In any case, these generalised commutation relations define the algebraic structure of the
extended chiral algebra. To be precise, the extended chiral algebra is the -algebra Ak generated
(m)
by the modes r , subject to these generalised commutation relations, and equipped with the
adjoint chosen in Section 2.1. The Sm,n with m + n = k are generated by the = 1 generalised
commutation relations

1 k/2 (m) (km)
(km)
(m)
Sm,km =
(2.50)
r r + k/2r1 rk/2++1 ,

=0
and the other Sm,n are multiples of these (as we shall see in Section 3.3). The affine modes are
then generated by the = 2 generalised commutation relations, for example

2 k/2 + 1 (0) (k1)
(k1)
(0)
r s+ + s+k/22 rk/2++2 ,
S0,k1 Er+s =
(2.51)

=0
and the affine modes commute or anticommute with the r(m) as in Eq. (2.23). It follows that
sl(2)k ) through
U(
sl(2)k ) is a subalgebra of Ak , just as U(Virc ) is identified as a subalgebra of U(
the Sugawara construction (Eq. (1.7)).
We conclude by checking that the adjoint defined in Section 2.1 is indeed an involutive antiautomorphism of Ak , completing the check of the self-consistency of the extended chiral theory.
(m,n)
( ) defining the generalised commutation relations is
Indeed, we find that the function Rr,s
(kn,km)
sent to Rs,r
( ) by the adjoint (this is easiest to see with the left-hand side of Eq. (2.49)).
Hence, the adjoint preserves the set of generalised commutation relations and therefore is consistently defined on Ak . We remark that this implies that Sm,n = Skn,km , which is obvious from
Eq. (2.44).
3. Representation theory
3.1. Monodromy charge
We now turn to a study of the expansion of the fields (m) (z) into modes. As is usual
with non-bosonic fields, this expansion depends upon which state the field acts. On each zero(n)
(m,n)
grade descendant state | , we therefore define a set of k + 1 u(1)-valued charges
(m,n)
(m = 0, 1, . . . , k) as follows:
is the negative of the leading (smallest) power of z w in
(n)
(m) (z) (w) (whose coefficient is non-zero):
(n)
(m) (z) (w) =
(w)
(m,n)
(z w)
+ .
(3.1)
From the fusion rules, it might appear that this power is
hk + h hk = ,
(3.2)
2
by Eq. (1.9), but this need not be the case in general, as evidenced by the case = k which we
studied in Section 2.2. Indeed, the sl(2)-weight is conserved in operator product expansions, so
the leading term in the operator product expansion (3.1) should involve a zero-grade descendant
216
Fig. 2. A schematic diagram of the level k irreducible

sl(2)k -module of highest weight k . The states are represented
in the form (, h) where is the sl(2)-weight and h is the conformal dimension (and we have ignored any multiplicity
label).
field of k (w), whose weight is k 2m + 2n. However, such a field may necessarily
vanish.
To analyse this question of existence, we consider the structure of the irreducible
sl(2)k module of highest weight k . This is illustrated schematically in Fig. 2. We see that there will
be a zero-grade descendant field of weight k + 2(m + n) if and only if m + n k. In all
other cases, the leading term in the expansion (3.1) will involve a descendant field at non-zero
grade. However, as 0 m + n k + , these cases are easily analysed (in particular, it is not
necessary to consider the affine singular vectors), and we find that
+ m + n if 0 m + n ,
(m,n)
=
if m + n k,
+ k m n if k m + n k + ,
(3.3)
(m,n)
where = /2. Note that k

= k/2 |k m n| in agreement with the general operator
(m,n)
product expansions given in Section 2.4, especially Eq. (2.46). Of course,
= (mod 1) for
all m and n (this is a direct consequence of being a simple current), so the corresponding U(1)valued charges are independent of m (and n). We shall refer to this common charge modulo 1
as the monodromy charge, because its negative represents the common monodromy of the z w
(n)
factors in (m) (z) (w). We note that the monodromy charge of a zero-grade descendant state
is always in Z/2.
The importance of assigning a monodromy charge to states is that it tells us how to expand
the fields (m) (z) into modes. To be more precise, when expanding
(n)
(n)
(m) (z) = lim (m) (z) (w)|0
w0
(3.4)
217
in powers of z, each term must have monodromy (mod 1), so we are forced into an expansion
of the form

(n)
(n)
(m) (z) =
(3.5)
r(m) zrk/4 .
rZ+ k/4
In this way, the mode expansion of the simple current fields depends upon the state on which the
modes act, as claimed.
The importance of the u(1)-valued charges (m,n) is that they provide the additional informa(m)
tion required to determine the largest value of r in the above mode expansion such that r does
(n)
not annihilate | . In other words, these charges determine the first non-trivial descendant
(n)
states of | with respect to the algebra Ak . It is easily verified that these descendants are of
the form
(n)
(m)

(m,n)
(3.6)
(first descendants),
k/4
(m,0)
=0
a result that will prove invaluable to us in what follows. In particular, the vacuum has 0
(m)
(m)
for all m (n must be 0), so its first Ak -descendants are the zero-grade states | = k/4 |0
(as we should expect).
It remains to determine the monodromy charge of descendant states. Suppose then that |
is a state of monodromy charge Z/2 (this is true for zero-grade descendants, so we may
suppose that it is true as part of an inductive procedure). Then, r(m) | is defined if and only if
r Z + k/4, so by Eq. (2.23),
(m)
r(m) Hs | = Hs r(m) | (k 2m)r+s |
(3.7)
must be defined for r Z + k/4. It follows that Hs | must also have monodromy charge .
The same is true for Es | and Fs |, so we conclude that acting with affine modes does not
change the monodromy charge (modulo 1).
To determine the effect of acting with a -mode on the monodromy charge, we apply the
generalised commutation relation (2.49) to the state |. For this to make sense, s + and r
k/2 + + must belong to Z + k/4. In other words, r Z + + k/4 and s Z + k/4
(hence r + s Z as required on the right-hand side of the generalised commutation relation). It
now follows from the term r(m) s(n) | that the monodromy charge of s(n) | must be + k/2,
hence we conclude that acting with -modes changes the monodromy charge (modulo 1) by k/2
(and by induction that the monodromy charge always takes values in Z/2).
We also mention that these results for descendant states can be derived directly from the
definition in terms of the corresponding operator product expansion. We will usually only be
interested in the charge (modulo 1) of descendant states, so we will omit this demonstration.
3.2. Highest weight modules
We define an Ak -highest weight state to be an affine highest weight state | |(0) such
that a -mode acting upon it cannot increase its sl(2)-weight or lower its conformal dimension,
just as the affine modes cannot in the affine case. This requires
(0)
r(m) = 0 for r > 0, or r = 0 and m < k/2.
(3.8)
218
We now define an Ak -Verma module in the usual way: It is the module generated from an Ak highest weight state by the action of all elements of Ak , modulo the algebra relations.
It remains to see whether the condition (3.8) imposes constraints on . Being an affine highest
weight state, the Ak -highest weight state will obviously have a well-defined conformal dimension and sl(2)-weight with 0 k. In addition, we may also associate to it the u(1)-charges
(0)
computed in Eq. (3.3). Since each Ak -highest weight state must be of the form | , its first
non-vanishing descendants must be, by (3.6), of the form (m = 0, 1, . . . , k)

(0)
m /2 if m ,
(m)
(m,0)

, where
=
(m,0)
(3.9)
k/4
/2
if m .
We observe that for = k, k(m,0) k/4 = m 3k/4, which is positive for m = k. In other words,
(0)
we see that the affine highest weight state |k has a non-vanishing descendant of lower
conformal dimension, hence we conclude that |k(0) cannot be an Ak -highest weight state.
(m,0)
: m = 0, 1, . . . , k} = /2, a sufficient
This observation is easily generalised. Since max{
(0)
condition for | to have a descendant of lower conformal dimension (hence not be an Ak highest weight state) is that > k/2. If = k/2, we find that the first non-vanishing descendants
(0)
(indexed by m as above) have higher conformal dimensions than |k/2 for m < k/2, and the
conformal dimension is unchanged for m k/2. It is easy to check that in this latter case, the
(0)
sl(2)-weight of the descendants is less than or equal to that of |k/2 , so there is no contradiction
(0)
to |k/2
being an Ak -highest weight state.
(0)
To summarise, we have shown that an affine highest weight state | is an Ak -highest

weight state if and only if k/2. The remaining affine highest weight states must therefore
appear as descendants of the genuine Ak -highest weight states. The sl(2)-weights and conformal
dimensions of the affine highest weight states suggest that the exact relationship is given by
(0)
(0)
()

(3.10)
k = /2k/4 ,
which is verified by directly showing that the right-hand side is an affine highest weight state:
(0)
(0)
()
()
Er /2k/4 = Er , /2k/4
(0)

1/2 (1)
= (k + 1 )
(3.11)
/2+rk/4 = 0,
(1,0)
since
= /2 1 < /2 + r, for r 0 (and similar calculations with Hr and Fr , but
r > 0).
Thus we see that a generic Ak -Verma module contains at least4 two affine highest weight
(0)
(0)
states: The Ak -highest weight state, | , and the Ak -descendant |k . It is necessary to
consider separately the exceptional case where k is even and = k/2. Then these two affine
(0)
(0)
highest weight states, |k/2 and | k/2 say, with
(0)
(k/2) (0)

k/2 ,
(3.12)
k/2 = 0
coincide (hence both would qualify as Ak -highest weight states according to the discussion
(k/2)
above), and are related by the self-adjoint mode 0 . It is really a matter of convenience as
4 In fact, there are no more, as we shall prove in Proposition 5.2.
219
to whether we should identify these two states or not. We note that the simultaneous normalisation of all zero-grade states gives
(0) (0)
(0) (k/2) 2 (0) (k/2) 2 (0) (0)

= ,
1 = k/2 k/2 = k/2 0
(3.13)
k/2 0
(k/2)
gives at most 2 independent states.

so repeated application of 0
If we choose to identify these two highest weight states, we have to conclude that this excep(k/2)
tional Ak -Verma module has but one affine highest weight state, and that 0
leaves this state
invariant. However, we will find it convenient (and not just for regularity of exposition) to not
(0)
(0)
identify these states, and declare that |k/2 is an Ak -highest weight state, whilst | k/2 is not.
With this understanding, this exceptional module may be treated just like any other.
We point out that in the preceding discussion, we have only verified that Eq. (3.10) holds up
to a multiplicative constant. This reflects the fact that highest weight states are only unique up
to constant multipliers. Recall from Section 1.3 that we have chosen to normalise all the affine
highest weight states (and their zero-grade descendants). Identifying the affine highest weight
states as in Eq. (3.10) is consistent with this, because we may absorb any factors arising from
this identification into the eigenvalues of the Sm,n . We now turn to an elaboration of this claim.
3.3. Sm,n -eigenvalues
Let us consider the computation of the eigenvalues of the Sm,n on each Ak -Verma module.
These eigenvalues can be calculated directly from the general commutation relations, but it can
be rather tiresome to compute each of them in this way (there are 14 (k + 2)2 independent
eigenvalues on each of the 12 (k +1) modules). To illustrate this, and the method of computation,
the simplest case is exhibited in Section 4.1.
However, recall that the purpose of the Sm,n is to absorb some unpleasant factors in the operator product expansions, which derive from our insistence on simultaneously normalising the
zero-grade states of a given module. This is equivalent to independently choosing the innerproduct on each affine module, and we have chosen it so that every affine highest weight state
has unit norm. In the extended theory, affine highest weight states are paired in the Ak -Verma
modules, so the algebra could potentially interfere with our ability to independently choose the
inner-products. As mentioned above, however, the Sm,n are able to absorb the factor that the
algebra dictates should relate the two inner-products, assuming Eq. (3.10), thereby restoring our
ability to choose to normalise every zero-grade state. This factor will clearly depend upon the
module, but not upon m and n.
The upshot of this is that in the formalism we have developed, the eigenvalue of Sm,n on
some Ak -Verma module is the product of its eigenvalue on |0 (given by Eq. (2.44)) and a factor
independent of m and n. For each module, this latter factor may therefore be computed from
(0)
the eigenvalue of a single Sm,n . A convenient eigenvalue on | | ( k/2) for this
computation would be that of Sk, , because
1 = k |k
+1
(k)
()
= |k/4/2 /2k/4 | =

/2
(k,;j )
|Sk, A0
| .
j
(3.14)
j =0
(k,;j )
are sl(2)-weightless, dimensionless, normally-ordered combinations of affine

The A0
modes, hence their action on an affine highest weight state is by scalar multiplication. Evalu-
220
ating this action then allows us to determine the eigenvalue of Sk, , hence the eigenvalues of
all of the Sm,n .
We illustrate this with the first few cases. If = 0 (| = |0), then Sk,0 is easily computed
to have eigenvalue 1. The eigenvalues of a general Sm,n on the vacuum are therefore given by
Eq. (2.44) (as they must be). The first non-trivial case is = 1 which gives

1
1 k2
2
1 = 1 |Sk1,1
(3.15)
H0 |1 = 1 |Sk1,1 |1 ,
2
2k
k
hence eigenvalue k. The eigenvalues of the Sm,n on |1 are therefore k times the values given
by Eq. (2.44). For = 2 we get

k+2
4k
1
k2
4k
3
1 = 2 |Sk2,2
H0 +
L0
H0 +
00 |2
2k
6
4k
12 k(k 1)
2
=
(3.16)
2 |Sk2,2 |2 ,
k(k 1)
where we note that

00 |2 =
:Hr Hr Er Fr Fr Er :|2 = H02 H0 |2 = 2|2
(3.17)

and (H )0 = H0 . The Sm,n -eigenvalues are therefore k2 times the eigenvalues given in
Eq. (2.44).

It appears likely from these computations that this ratio will be k in general, but it is difficult to prove this using the above method. The problem is essentially that we have to use the
(k,;j )
generalised commutation relation of order = + 1, thereby needing to compute the A0
for j up to , a rather non-trivial task when is large. This non-triviality derives from the large
number of terms contributing to the right-hand side of the generalised commutation relation with
m + n = k when = + 1 is large. Such general computations would be far easier if we were
only required to use a generalised commutation relation with a single (known) term on the righthand side. Remarkably, this can indeed be arranged.
Proposition 3.1. The eigenvalue of Sm,n on | is
k
times its eigenvalue on |0.
Proof. By iterating Eq. (1.13), we may write

1 = | =
1
(0) ()
E0 .
!
(3.18)
()
Since is a zero-grade state, we have

E0 () =

:Er1 Er :() ,
(3.19)
r1 ,...,r Z
r1 ++r =0
where the normal-ordering is nested to the right as usual. But, this is (up to an Sm,n factor)
precisely the right-hand side of the generalised commutation relation with = + 1, m + n =
k and r + s = 0. We choose n = 0 and s = /2 k/4 (in accord with the monodromy charge
221
()
of | ):

k/2 + (k)
(0)
(0)
(k)
k/4/2 /2k/4+ + k/4/21 /2k/4+1+

=0

+1
= Sk,0
:Er1 Er :.
(3.20)
r1 ,...,rZ
r1 ++r =0
()
(m,)
Applying this to | , almost all of the terms vanish because

= /2 when m k (by
Eq. (3.3)). We therefore conclude that

() +1 1
(0) 1
()
1
(0)
(k)
(0)
1=
:Er1 Er : =
Sk,0 k/4/2 /2k/4

!
!
r1 ,...,r Z
r1 ++r =0
()
1
(0) 1
(0)
,
k Sk,0 /2k/4
(3.21)
!
by (the adjoint of) Eq. (3.10).
(0)
()
Consider now the combination /2k/4 | . Eq. (1.13) allows us to exchange the sl(2)weight between the mode and the state in the following manner:
=
()
(1)
1 (0)
(0)
/2k/4 = 1/2 /2k/4 F0
1/2
(1)
(1)
k
1
(1)
(0)
=
1/2 F0 /2k/4
.
/2k/4
(3.22)
(0)
The F0 at the left of the last term will annihilate k |, so iterating this exchange gives

(0) 1 k 1/2
(0) 1 (0)
1 k 1/2
(0) 1
()
1=
(3.23)
k Sk,0 /2k/4 =
k Sk,0 k .
!
!
The eigenvalues of Sk,0 on |k and on the vacuum (see Eq. (2.44)) are therefore given by

(k )! k 1/2
1 k 1/2
and
(3.24)
!
k!
respectively, whence the result.
4. Examples
4.1. The
sl(2)2 extended algebra
We consider the case k = 2, so there are three affine primaries: The identity field, a field
1 (z) of conformal dimension 3/16, and a simple current (z) of conformal dimension 1/2.
(0)
(1)
1 (z) gives rise to two zero-grade descendant fields, 1 (z) 1 (z) and 1 (z) (of respective
weights 1 and 1), whereas the simple current gives rise to three: (z) (0) (z), (1) (z) and
(2) (z) (of respective weights 2, 0, and 2). The operator product expansions of the latter take
the form:
(0) (z) (0) (w) = S0,0 :E(w)E(w):(z w) + ,
222
1
(z) (w) = S0,1 E(w) +
E(w) +
2

1
E(w)
(1) (z) (0) (w) = S1,0 E(w) +
2
(0)
(1)

1 +
(w) (z w) + ,
4

1 +
(w) (z w) + ,
4
(0) (z) (2) (w)

1
2
1
1 0
1
= S0,2
+ H (w) +
T (w) + H (w) + (w) (z w) + ,
zw 2
3
4
12

1 0
2
1
(1)
(1)
+
T (w) (w) (z w) + ,
(z) (w) = S1,1
zw
3
6
(2) (z) (0) (w)

1
1
1
2
1
= S2,0
H (w) +
T (w) H (w) + 0 (w) (z w) + ,
zw 2
3
4
12

1
1
F (w) + (w) (z w) + ,
(1) (z) (2) (w) = S1,2 F (w) +
2
4

1
1
F (w) (w) (z w) + ,
(2) (z) (1) (w) = S2,1 F (w) +
2
4
(2) (z) (2) (w) = S2,2 :F (w)F (w):(z w) + .
(m) (z) (n) (w)
We have explicitly listed both

and
mutativity given by Eq. (2.37) (with m,n = 1):
(n) (z) (m) (w)
(m) (z) (n) (w) = (1)m+n+1 (n) (w) (m) (z).
(4.1)
in order to illustrate the com(4.2)
We also note that the generalised commutation relations with = 1 are the super-commutation
relations
1
r(m) s(n) + (1)m+n s(n) r(m) = Sm,n m+n,2 r+s,0 .
(4.3)
From Eq. (2.44), the eigenvalues of the Sm,n on the vacuum A2 -module are given (in matrix
form with rows and columns indexed by 0, 1 and 2) by
1
1/2 1/ 2
S|0 = 1/ 2
(4.4)
1
1/ 2 |0.
1
1/ 2 1/2
We determine the eigenvalues on |1 , to compare with (and illustrate) the discussion of Section 3.3. The easiest determination is
1 1
(1) (1)
1 = 1 | 1 = 1 |0 0 |1 = 1 |S1,1 |1 ,
(4.5)
2
hence S1,1 |1 = 2|1 .
The computation of the other eigenvalues provides a convenient illustration of the use of
generalised commutation relations. First, note that = 1 gives
1
(0) (1)
(1) (0)
(1) (1)
1 |0 0 F0 |1 = 1 |0 0 F0 |1 = 21 |0 0 |1 = 2,
(4.6)
(0)
since 0 annihilates |1 , by Eq. (3.9). Alternatively,

(0) (1)
1 |0 0 F0 |1 = 1 |S0,1 E0 F0 |1 = 1 |S0,1 |1 ,
(4.7)
223
and
(0) (2)
(0)
(1)
21 |0 0 |1 1 |0 F0 0 |1
1
(1) (1)
= 21 |S0,2 |1 21 |0 0 |1

= 2 1 |S0,2 |1 1 ,
(4.8)
from which we get S0,1 |1 = 2|1 and S0,2 |1 = 2|1 . Finally, the eigenvalue of S0,0
may be obtained by comparing
(0) (1)
1 |0 0 F0 |1 =
(0) (0)
(1) (1)
1 |F1 1 1 F1 |1 = 21 |0 0 |1 = 2,
(4.9)
and

3 1
(0) (0)
1 |F1 1 1 F1 |1 = 1 |S0,0 F1 E02 + 2E1 E1 F1 |1 = 21 |S0,0 |1 , (4.10)
2
whence S0,0 |1 = |1 , and
1
2 2
S|1 = 2 2
(4.11)
2 |1 .
2 1
2
The eigenvalues of the Sm,n on |1 are therefore precisely double the corresponding eigenvalues
on |0, explicitly confirming the discussion in Section 3.3.
We conclude this example by showing how A2 decomposes into three copies of the extended
symmetry algebra of the Ising model M(3, 4). By this, we mean the Ising model extended by its
own simple current 2,1 , which effectively describes a free fermion. We define two fermions by

1
+ (z) = (0) (z) + (2) (z)
2
and

i
(z) = (0) (z) (2) (z) .
2
(4.12)
Then, the operator product expansions give:

+ (z) + (w)

1
2
1
1
= S0,2
+
T (w) + 0 (w) + :EE + F F :(w) (z w) + ,
zw
3
12
4

i
1
+ (z) (w) = S0,2 H (w) + H (w) :EE F F :(w) (z w) + ,
2
2
(z) (w)

1
2
1 0
1
= S0,2
+
T (w) + (w) :EE + F F :(w) (z w) + ,
zw
3
12
4
(4.13)
where we have used the fact that S0,0 = 12 S0,2 .

The third fermion field should be (1) (z), but we note that this field commutes with the other
fermions, by Eq. (4.2), instead of anticommuting with them. The remedy is to introduce an operator F into the fermion definitions so as to restore anticommutativity. Recall from Section 2.1
that when k is even, we may consistently choose a different adjoint in the extended theory which
results in the affine modes E and F commuting with the -modes, rather than anticommuting
(this is referred to as choice (2) in Section 2.1). The equivalence of the theory defined by this
224
adjoint and the adjoint we have used throughout was given by such an F in Eq. (2.17), whose
properties were summarised at the end of Section 2.1.
Redefining the first two fermions as + (z) + (z)F and (z) (z)F has no effect
on their operator product expansions. We now define the third fermion to be 0 (z) = i (1) (z)F ,
so that the three fermions are all mutually anticommutative. The remaining operator product
expansions then become
+ (z) 0 (w)

i
1
1
= S0,2
(E + F )(w) +
(E + F )(w) + (+ )(w) (z w) + ,
2
2
4

1
2
1
0
0
0
+
T (w) (w) (z w) + ,
(z) (w) = S0,2
zw
3
6
(z) 0 (w)

1
1 +
1
= S0,2
(E F )(w) +
(E F )(w) + ( + )(w) (z w) + ,
2
2
4
(4.14)
using S1,1 = S0,2 and S0,1 = 1 S0,2 . Note that the operator product expansions of distinct fermi2
ons are regular.
These operator product expansions should be compared with the operator product expansion
of the simple current5 (z) of the Ising model [5,12]:

1
(z) (w) = S
(4.15)
+ 2T (w) + .
zw
Here, S is a scalar operator taking (when the Virasoro highest weight states are simultaneously
normalised) eigenvalue 1 on the Ising model vacuum and 2 on the highest weight state of conformal dimension 1/16 (so we identify it with S0,2 = S1,1 ), and T (w) denotes the Ising model
Virasoro field. It follows that the A2 -Virasoro field decomposes into three Ising model Virasoro
fields, T (z) = T + (z) + T 0 (z) + T (z), where
1
1
1
1
T + (z) = T (z) + 0 (z) + :EE + F F :(z) = :H H :(z) +
3
24
8
16
1
1
1
1
0
0
:H H :(z) + :EF + F E:(z),
T (z) = T (z) (z) =
3
12
16
8
1
1
1
1
0
T (z) = T (z) + (z) :EE + F F :(z) = :H H :(z)
3
24
8
16
1
:EE + F F :(z),
8
and
1
:EE + F F :(z).
8
(4.16)
One may explicitly check (we used OPE DEFS [17] in M ATHEMATICA) that these are indeed
Virasoro fields of central charge 1/2, and that the operator product expansions between distinct
Virasoro fields are regular.6
5 Quantities with bars such as these will always be reserved for those relating to the extended Ising model (free fermion)
theory. They do not refer to antiholomorphic sectors.
6 Actually, OPE DEFS reports that these operator product expansions all have the form
1 :EH E:(w) :F H F :(w)

T i (z)T j (w) =
+
8
zw

i, j {+, 0, }, i = j .
However, it is easy to check that the two composite fields appearing above are actually null.
(4.17)
225
we have therefore explicitly demonDenoting the extended algebra of the Ising model by A,
strated the isomorphism
A
A.
A2
=A
(4.18)
We note that this isomorphism preserves the adjoints of the theories (the fermions constructed
above are self-adjoint), hence that we have constructed an isomorphism of L0 -graded -algebras
(the grading by sl(2)-weight is lost on the fermions). Moreover, it is not hard to verify that the
A2 -highest weight states |0 and |1 decompose under this isomorphism as

|0
|0
and |1 1 1 1 ,
|0 |0
(4.19)
16
16
16
denotes a A-highest
respectively, where |h
weight state of conformal dimension h (the affine
highest weight state | does not even correspond to an eigenstate of the Virasoro modes L
0
corresponding to the fields T (z)). This isomorphism of algebras suggests that there are corresponding isomorphisms of modules, hence character identities. We will study this in Sections 5
and 6.
4.2. The
sl(2)4 extended algebra
When k = 4, there are three A4 -highest weight states, |0, |1 and |2 , and the simple
current has five component fields (m) (z), m = 0, . . . , 4, of conformal dimension 1. The Sm,n
are given by Eq. (2.44) as
6/12 1/2
1
1/24
1/12
1/12
1/4
6/4
1
1/2
S|0 = 6/12
(4.20)
6/4
1
6/4
6/12 |0,
1/2
6/4
1/4
1/12
1
6/12 1/12
1/24
1
1/2
and Section 3.3 gives the corresponding eigenvalues on |1 and |2 as 4 and 6 times these,
respectively. The simple current component fields are supercommutative in the sense that
(m) (z) (n) (w) = (1)m+n (n) (w) (m) (z),
and the non-zero operator product expansions (to second order) are

1
(1) (z) (1) (w) = S2,2 :EE:(w) + ,
4

6
S2,2 :EE:(w) + ,
(0) (z) (2) (w) =
12

6
S2,2 :F F :(w) + ,
(2) (z) (4) (w) =
12

1
(3)
(3)
(z) (w) = S2,2 :F F :(w) + ,
4

1
1
E(w) 1
(0)
(3)
(z) (w) = S2,2
+ E(w) + + (w) + ,
2
zw 2
4

6
1 +
E(w) 1
(1)
(2)
(z) (w) =
S2,2
+ E(w) + (w) + ,
4
zw 2
12
(4.21)
226

6
1
F (w) 1
S2,2
+ F (w) + (w) + ,
4
zw 2
12

1
1
F (w) 1
+ F (w) + (w) + ,
(1) (z) (4) (w) = S2,2
2
zw 2
4

H (w)
1
1
+
+ T (w) + H (w) +
(0) (z) (4) (w) = S2,2
4
(z w)2 2(z w)

H (w)
1
1
+
+ T (w) + H (w)
(1) (z) (3) (w) = S2,2
8
(z w)2 4(z w)

1 0
1
+
T
(w)
(w)
+
(2) (z) (2) (w) = S2,2

12
(z w)2
(2) (z) (3) (w) =

1 0
(w) + ,
12

1 0
(w) + ,
24
(4.22)
the rest being obtained by these from Eq. (4.21).

sl(3)1 , meaning that the corresponding coset
There is a conformal embedding of
sl(2)4 into

model sl(3)1 /sl(2)4 has vanishing central charge. This does not directly imply the equivalence of
the two conformal field theories. Indeed, the spectrum of the
sl(3)1 -WessZuminoWitten theory indicates that the correct
sl(2)4 -WessZuminoWitten theory is the non-diagonal one [14].
Nevertheless, the chiral algebras of the two theories need to correspond in some way. We will
sl(3)1 ) are isomorphic.
detail this correspondence by proving that the chiral algebras A4 and U(
Roughly speaking, the three sl(2) generators are augmented by the five components of the
simple current (whose conformal dimension is 1) to fill out the eight sl(3) generators:
E
(0)
(1)
(4)
(3)
f1
e1
F
H, (2)
f .
h1 , h2
e2
(4.23)
f2
We choose the sl(3) generators in the following way: For the two simple roots, we let h1 and
h2 be the corresponding coroots, and let e1 , f1 and e2 , f2 be the corresponding root vectors,
normalised by [ei , fi ] = hi . The Killing form7 therefore satisfies
(e
i , fj ) = (f
i , ej ) = ij ,

2 1
, the Cartan matrix ,
A=
(h
i , hj ) = Aij
1 2
(4.24)
and vanishes for all other combinations of ei , fi and hi . We define e = [e1 , e2 ] and f =
, f ) =
e = [f2 , f1 ]. Invariance of the Killing form and the Jacobi identity then give (e
(h
1 , h2 ) = 1, hence
[e2 , f ] = f1 ,
[f , e1 ] = f2 ,
[e , f2 ] = e1 ,
[f1 , e ] = e2
(4.25)
(invariance), and finally, [e , f ] = h1 + h2 (Jacobi identity).

sl(3)1 ) may be derived by comparing the operator prodThe isomorphism between A4 and U(
uct expansions of each theory.8 If we suppose that H = ah1 + bh2 and that (0) is proportional
7 We denote the Killing form of sl(3) by (,
) to distinguish it from that of sl(2).
8 Alternatively, we can directly compare the
sl(3)1 commutation relations with the supercommutation relations
(m,n)

(m) (n)
(n) (m) 2
s (1)m+n s r = Sm,n Jr+s + rm+n,4 r+s,0 ,
(4.26)
where J (m,n) denotes E, 18 (n m)H or F according as to whether m + n is 3, 4 or 5 (respectively), and vanishes
otherwise.
227
to e , then by comparing
(a + b)e (w)
+ ,
zw
(4.27)

we derive that a + b = 4. Similarly, comparing H (z)H (w) with its sl(3)1 counterpart gives
a 2 ab + b2 = 4, hence a = b = 2. If we now put (2) = h1 + h2 , then considering its
operator product expansions with H and with itself, we conclude that = and 2 = S2,2 /6.
At this point we pause to note that the
sl(3) currents are all mutually bosonic, whereas our
simple current fields are not (Eq. (4.21)). Again, we must remedy this with an operator F , and
Eq. (2.17) advises that we must put
!
S2,2
(2)
(h1 h2 ).
F=
(4.28)
6
The remaining calculations are routine and offer no new subtleties. The final result is as follows:
H (z) (0) (w) =
4 (0) (w)
+
zw
and (ah1 + bh2 )(z)e (w) =
"
(0) F = S2,2 e
E=
(1)
F=
1 (0)
F,
e = "
S2,2
1
1
(1) F,
e1 = E + "
2S2,2
2 2
2(e1 + e2 )
S2,2
(e1 e2 )
2
H = 2(h1 + h2 )
!
(2)
F=
F=
S2,2
(h1 + h2 )
6
2(f1 + f2 )
(3)
F =
S2,2
(f1 f2 )
2
"
(4) F = S2,2 f
1
1
(1) F,
e2 = E "
2S2,2
2 2
#
3
1
h1 = H +
(2) F,
4
2S2,2
#
3
1
(2) F,
h2 = H
4
2S2,2
1
1
(3) F,
f1 = F "
2S2,2
2 2
1
1
f2 = F + "
(3) F,
2S2,2
2 2
1 (4)
f = "
F.
S2,2
(4.29)
sl(3) are also preserved, hence we have another isomorphism

We note that the adjoints of A4 and
of -algebras. We mention that if we had tried to continue these computations without such an
F , we would have quickly reached a contradiction in the algebra.
It is not difficult to verify that this isomorphism identifies the Virasoro fields of our two theories. Of more interest is the fact that H = 2(h1 + h2 ) limits the allowed A4 -highest weight states
to |0 and |2 , as the eigenvalues of H are constrained to be even (this is consistent with the
formulae given for h1 and h2 if we remember that S2,2 has eigenvalue 6 on |2 ). Actually, this
is just the constraint imposed by monodromy charge: Linear combinations such as those appearing on the right of (4.29) only make sense at the level of modes when acting upon states whose
charge is integral, so that the mode subscripts are integers.
228
(2)
Assuming that 0 interchanges the highest weight states |2 and | 2 , the precise correspondence is then

(1, 0) ,
2

|

+
|
2
|0 (0, 0) and
(4.30)

(0, 1) ,
|2 | 2
sl(3)1 -highest weight states whose hi -eigenvalue is i . This corwhere |(1 , 2 ) denotes the
respondence between highest weight states can be lifted to the irreducible modules and their
characters. To see this, we now turn to a study of the singular vectors of the Ak -Verma modules.
5. Singular vectors
5.1. Preliminary remarks
Recall from Section 3.2 that we have identified two affine highest weight states in each Ak Verma module: The Ak -highest weight state itself, and its Ak -descendant given in Eq. (3.10).
That there are only two (non-singular) affine highest weight states is to be expected as the simple
current has order two in the fusion ring. Nevertheless, we can give a direct proof of this fact using
the following lemma, which states that the action of any bilinear in the -modes within a highest
weight module can be replaced by a combination of affine modes.
(m) (n)
Lemma 5.1. Any state of the form r s |, where | is an arbitrary state in a highest
weight module, may be written as a linear combination of affine descendants of |.
Proof. The generalised commutation relations given in Eq. (2.49), for different choices of ,
all express an infinite linear combination of quadratics in the -modes, in terms of pure affine
modes (recall that the Sm,n act as constants within highest weight modules). We therefore seek
to invert this relation. Write the generalised commutation relations in the form

k/2 + 1 (m) (n)
(n)
(m)
r s+ (1)kmn+ s+k/2 rk/2+ + |

=0
(m,n)
= Rr,s
( )|,
(5.1)
(m,n)
where Rr,s ( ) denotes the right-hand side of Eq. (2.49). We determine an integer N 0
(n)
such that s+N + annihilates | for all 0, and then choose sufficiently large so that
(m)
rN k/2+ + also annihilates | for all 0. The generalised commutation relations now
become (with r r i and s s + i)
N1
k/2 + 1 (m)
(n)
(m,n)
ri s+i+ | = Rri,s+i ( )| (for i = 0, . . . , N 1).

=0
(5.2)
(m)
(n)
This is a set of linear equations for the ri s+i+ |, and the corresponding matrix is
triangular with diagonal entries equal to one, hence invertible. It therefore follows that each
(m)
(n)
(m,n)
ri s+i+ | can be expressed as a linear combination of the Rrj,s+j ( )|. But these are
expressed as affine modes acting on |, as required. 2
We are now in a position to prove the announced result, namely:
229
Proposition 5.2. There are only two non-singular affine highest weight states in each Ak -Verma
module (though they may coincide).
()
k | ,
24
Proof. We have already identified two such highest weight states, | and |k =
in each Ak -Verma module. We therefore need only show that there are no others.
)
1)
r(m
| . If = 0, then we are
Consider therefore an arbitrary state of the form r(m

1
considering the Ak -highest weight state. If = 1, then we are considering r(m) | , and Eq. (3.9)
gives r (m,0) k/4, where (m,0) is /2 or m /2 according as to whether m or not.
Acting on these states with E0 and F1 suffices to show that we have already found the only affine
highest weight state with = 1. It follows that every other state with = 1 must be an affine
descendant of this highest weight state, as an affine descendant of | must be expressed in
terms of at least two -modes (as follows from the generalised commutation relations).
But if > 1, Lemma 5.1 allows us to replace the first two -modes with (a linear combination
(m )
)
)
1)
r(m
| is an affine descendant of r3 3 r(m
| . Inducof) affine modes. Thus, r(m

1
tively, it follows that states with > 1 are affine descendants of | or |k , according as
to whether is even or odd (respectively). Any such descendants which are also affine highest
weight states are singular, hence the proposition is proved. 2
(k)
(k) denote the corLet M denote the Ak -Verma module of highest weight , and let M
responding
sl(2)k -Verma module. Then, Proposition 5.2 implies that the inclusion of U(
sl(2)k )
into Ak induces a surjection of the form
(k) M(k) .
(k) M
M
(5.3)
Furthermore, our simultaneous normalisation of all zero-grade states (in particular of the affine
highest weight states) implies that this surjection is an isometry. It follows that the singular
vectors of the affine modules (affine singular vectors for short) must therefore be mapped to
Ak -singular vectors, and all Ak -singular vectors arise in this way.
We can now investigate the Ak -singular vectors by determining the kernel of our surjection
(5.3). What we will find is that this kernel coincides with the set of affine singular vectors. In
other words, each affine singular vector is mapped to zero in the Ak -Verma module, so the Ak Verma module is actually irreducible, and decomposes as the direct sum of two irreducible affine
modules (this is the content of Corollary 5.4). Before demonstrating this result, we verify it in
the following section within the context of the two simplest models, k = 1 and 2.
5.2. Some examples
The easiest case to analyse is of course k = 1. The operator product expansions defining the
extended algebra are as follows:
1
(0) (z) (0) (w) = E(w)(z w)1/2 + E(w)(z w)3/2 + ,
2
(0) (z) (1) (w)

1
1
1
1
1/2
+ H (w)(z w) + T (w) + H (w) (z w)3/2 + ,
=
2
4
(z w)1/2 2
1
(1) (z) (1) (w) = F (w)(z w)1/2 + F (w)(z w)3/2 + .
2
(5.4)
230
The = 2 generalised commutation relation is therefore

2 (m,n)
+ 12 (m) (n)
(n)
(m)
r s+ + (1)m+n s3/2 r+3/2+ = Jr+s ,

(5.5)
=0
where J (m,n) stands for E, 12 (n m)H , or F , according as whether m + n is 0, 1, or 2, respectively (note also that Sm,n = id for all m and n).
The singular vectors of the affine modules are generated by F0 |0, F0 | (1) , E1 | (0) , and
2
E1 |0. But, if we take m = n = 1 and s = 3/4, we find that

1
2 + 2 (1)
(1)
(1)
(1)
F0 |0 =
(5.6)
|0 = 0
+ 3/4
3/4+
3/4 3/4+

=0
(1)
r
as
annihilates the vacuum for all r > 1/4. Similarly, we find that F0 | (1) vanishes when
(1)
(1,1)
1/4 = 3/4
we take m = n = 1 and s = 3/4, because r annihilates | (1) when r > 1
(0)
(Section 3.1). Putting m = n = 0 and s = 1/4, we learn that E1 | also vanishes identi2 |0. The demonstration of vanishing is less
cally (we use 1(0,0) = 1/2), so we are left with E1
immediate in this case, and may be discerned by noting that

2
(0)
(0)
(0)
2
|0 = E1 3/4 1/4 |0 = 3/4 E1 (0) = 0,
E1
(5.7)
since we have already established that E1 | (0) vanishes.
When k = 2, we have three affine modules, hence six primitive singular vectors to consider.
The = 1 generalised commutation relation was given in Eq. (4.3), and = 2 gives

(m) (n)
(n)
(m) 2
(m,n)
r s+ (1)m+n s1 r+1+ = Sm,n Jr+s ,
(5.8)
=0
where J (m,n) now stands for E, 14 (n m)H , or F , according as to whether m + n is 1, 2, or 3,

respectively (and vanishes if m + n is 0 or 4).
(1)
As with the k = 1 case, the singular vectors F0 |0, F0 |1 , and F0 | (2) can be easily
checked to vanish identically when the = 2 generalised commutation relation is applied with
n = 2 and s = 1/2, 0, or 1/2, respectively. Similarly, E1 | (0) vanishes identically upon taking
s = 1/2 and n = 0. This then gives
2
(1) (0)
(0) (0)
3
2
= 2E1 3/2
= 0,
|0 = 2E1
1/2
E1
(5.9)
where we anticommute the E1 to the right, where they annihilate | (0) .
Verifying the vanishing of the last singular vector requires a little more delicacy. One straightforward demonstration is as follows:

(0) 2 1 (0) (0) (1) (1) (0)

1 (0)
(0) (1) (0)
2 (0) 2 1
E1
1 = E1 1 0 1 = 1 E1 1 =
1
2 1 1 0 0
2
2
(0) 1 (0) (0) (0) 1
1 1 (0) (0)
(5.10)
=
= 0.
1 1 S1,1 1 =

4
2 1 1 1
A more elegant demonstration involves using the = 3 generalised commutation relation,

3
(0) (0)
(0)
(0)
( + 1) r s+ + s2 r+2+ = S0,0
:Et Er+st :,
=0
tZ
(5.11)
and noting that
2
t :Et E2t :| = E1 |
231
when | is an affine highest weight state:

2 (0) 3 (0) (0) (0) 1
1 = 1 1 1 = 0.
E1
(5.12)
This belies the fact that the vanishing may be demonstrated using generalised commutation relations with < 3. Nevertheless, it does suggest a strategy by which this vanishing may be
demonstrated for general k.
5.3. Vanishing in general
Remarkably, it is possible to demonstrate explicitly that all affine singular vectors vanish
identically when the algebra is extended, hence that these extended algebras define faithful (or
free-field-like) representations of the SU(2) WessZuminoWitten models. This may sound surprising, but recall from Section 2.2 that we have used the structure of the irreducible affine
vacuum module to derive the extended algebra, and in Section 3.1 we used the structures of the
(m,n)
. A knowledge of the affine
other irreducible modules to derive the u(1)-valued charges
singular vectors was therefore built in to the extension, in a sense, so it should not be surprising
to find that they do not appear in the extended theory. This suggests that the extended algebra
approach we have constructed above should be interpreted as a natural free-field reformulation
of the SU(2) WessZuminoWitten models.
k+1
| , for =
Theorem 5.3. The affine singular vectors generated by F0+1 | and E1
0, . . . , k, span the kernels of the surjections of Eq. (5.3).
Proof. It is enough to show that the primitive singular vectors vanish in the corresponding Ak modules. We consider then F0+1 | . As in the proof of Proposition 3.1, we note that

j
(5.13)
:Ft1 Ftj :| = F0 |
t1 ,...,tj Z
t1 ++tj =0
when | is a highest weight state, and that this is just (a multiple of) the right-hand side of the
generalised commutation relation with = j + 1, m + n = j + k and r + s = 0, applied to |.
We therefore take = + 2, m + n = k + + 1, and choose r = s = /2 k/4 + 1. This
gives
+2 1
F0+1 | = Sm,n

k/2 + + 1
=0
(m)

(n)
(n)
(m)
k/4/21 /2+1k/4+ + k/4/21 /2+1k/4+ |
= 0,
(5.14)
because (m,0) /2 for all m. This proves that this singular vector vanishes in all cases when
m and n can be chosen to sum to k + + 1. This covers all cases except = k, for which we
offer the simple argument that
(0)
F0k+1 | = F0k+1 k/4 |0 = 0,
(5.15)
232
as F0 has already been shown to annihilate |0, so anticommuting the F0 to the right eventu(k)
ally gives something proportional to F0 k/4 |0, which obviously vanishes upon one further
anticommutation.
k+1
The demonstration for E1
| is similar if more intricate. The fact that | is an affine
highest weight state again leads to

k+1
E1
(5.16)
| =
:Et1 Etk+1 :| ,
t1 ,...,tk+1 Z
t1 ++tk+1 =k+1
which is (a multiple of) the right-hand side of the generalised commutation relation with =
k + 2 and m + n = 1, applied to | . The conformal dimension determines r + s to be
k + 1.
When is odd, we choose s = 1/2 k/4, so the modes acting immediately on | are
(n)
(m)
(m,0)
(n,0)
= m/2 and
= n/2,
1/2k/4+ and 1/2k/4+ . Since m+n = 1, we have
hence these modes will annihilate | if m and n are both less than ( + 1)/2. We therefore
choose m = n = ( 1)/2, demonstrating that this singular vector vanishes identically. When
is even, we cannot choose s = 1/2 k/4 due to the monodromy charge of | . Instead, we take
(n)
(m)
s = k/4, so the modes acting on | are k/4+ and 1k/4+ . Complete annihilation will
then occur if n < /2 and m < /2 + 1. Taking m = /2 and n = /2 1 then proves that the
singular vector also vanishes identically in this case.
Finally, we again note that m + n = 1 cannot be satisfied when = 0, but we can quickly
verify that
(k1) (0)
(k1)
k+1
1
1
k
k
E1
|0 = Sk1,0
E1
k/41
k/4 |0 = Sk1,0
E1
k/41
| = 0,
2
by anticommuting the E1 to the right where they annihilate |.
(5.17)
Corollary 5.4. The Ak -Verma module is irreducible, and is isomorphic to the direct sum of two
affine irreducible highest weight modules (irreducible modules are denoted by L):
(k)
(k) (k)
L (k)
Lk = M = L .
(5.18)
5.4. Consequences when k = 2 or 4

To complete our study of singular vectors, let us consider what the vanishing of the singular
vectors implies for the examples studied in Section 4. In Section 4.1, we showed that A2 was
of the Ising model
isomorphic to the tensor product of three copies of the extended algebra, A,
(that is, three free fermions + , 0 and ). The A2 -vacuum thereby decomposed into the tensor
product of three fermionic vacuum states, and the A2 -highest weight state |1 was identified
with the product of three copies of the fermionic highest weight state of conformal dimension
1/16. The vanishing of the singular vectors, both in the A2 -vacuum module and the fermionic
vacuum modules, then leads to the conclusion that we have an isomorphism of modules (graded
by conformal dimension only):
(2)
(2)
(2)
L 0 L 2
= L0
= L 0 L 0 L 0 ,
(5.19)
where L h denotes the fermionic module whose highest weight state has conformal dimension
h. We shall explicitly confirm this fact in Section 6.2, by showing that the character of the A2 (2)
vacuum module L0 is the cube of the character of the fermionic vacuum module L 0 .
233
(2)
Navely, we might suppose that the character of L1 is also the perfect cube of the character
of L 1/16 . However, this is not correct (see Section 6.2), though it is nearly so. The obstruction is
subtle: The chiral algebras are -isomorphic and there are no singular vectors to consider, in L(2)
1
by Theorem 5.3 and in L 1/16 as is well-known. But, the natural definition of a highest weight
module is slightly different for the two algebras, and it is this difference which is not preserved
by our isomorphism. Explicitly, |1 is an affine highest weight state, so it is annihilated by
1
is only a Virasoro highest weight state (by
E0 , whereas the fermionic highest weight state | 16
definition), hence it is not annihilated by any mode of zero-grade. In other words, this slight
incompatibility in the respective definitions leads to
2
0 = E0 |1 = 0(1) 0(0) |1
1 1 1
1
1
i 1
0 16
0 16
16 + i 16 0 16
0 16
= 0,
2
(5.20)
using the explicit form of the isomorphism given in Section 4.1.

It should be clear that this difference is only relevant for zero-grade modes, and that it was not
relevant in our consideration of the vacuum module. Explicitly, we have 8 independent states at
(2)
1
zero-grade in L 3
1/16 (each | 16 may have a 0 acting on it or not), but only 4 such states in L1 :
(0)
(0)
(1) (0)

|1 = 1 ,
= 0 1 ,
1
(1)
(1)
(0)
(0)
1
1

= F0 1
and 1 = F0 1 .
1
2
2
(5.21)
The isomorphism between the chiral algebras and the absence of singular vectors now allow us
to conclude that L 3
1/16 has exactly twice as many independent states at each conformal grade as
(2)
(2)
L1 . In other words, the character of L1 is precisely half the cube of the character of L 1/16 .
We will use this fact in Section 6.2.
The isomorphism A4
sl(3)1 ) constructed in Section 4.2 also extends to isomorphisms
= U(
between modules. Because this is a -isomorphism, norms are preserved, hence Theorem 5.3
implies that any singular vectors in the
sl(3)1 -modules correspond to 0 (this can of course be
checked explicitly). The definitions of highest weight state are consistent in this case (we did
after all model the Ak -definition on the affine one). This is obvious for positively-graded modes,
and for the zero-grade modes we note that
E0 | = 0(1) | = 0(0) | = 0
(e1 )0 | = (e2 )0 | = (e )0 | = 0,
(5.22)
(2)
using the explicit correspondence (4.29) (and similarly for H0 , 0 , (h1 )0 and (h2 )0 ).
(1)
Denoting the irreducible
sl(3)1 -module whose sl(3)-highest weight is (a, b) by V (a,b)
, we
have therefore proved the following module isomorphisms:
(4) 2
(4)
(4)
(4)
(1)
(4)
(1)
(1)
L 0 L 4
(5.23)
= L0
= V (0,0) and L 2
= L2
= V (1,0) V (0,1) ,
and thus the corresponding character identities. As noted in Section 4.2, monodromy charge
(4)
considerations prevent us from constructing a similar isomorphism for L1 .
We remark that these are isomorphisms of L0 -graded modules, or more precisely, of H0 graded modules where the H0 -grading on the
sl(3)1 -modules is determined by (4.29). This
grading on the
sl(3)1 -modules is strictly coarser than the usual one by sl(3)-weight, but the
234

(2)
usual grading may be recovered by keeping track of the 0 -eigenvalues on the A4 -modules.
The isomorphisms (5.23) of L0 -graded modules thus imply identities of the specialised characters (tr q L0 ) only. To obtain an identity involving the full
sl(3)1 -characters (distinguishing the
(2)
conjugate modules), we would have to incorporate the 0 -eigenvalues in the A4 -grading. We
will not pursue this subtlety here.
6. Module bases and characters
6.1. The A1 character
When k = 1, we saw in Section 5.2 that the generalised commutation relation with = 2
was sufficient to demonstrate that the singular vectors vanish. We claim that to determine the
character of the (single) irreducible A1 -module, we only need to consider the = 1 identity:

12
=0
1
(m) (n)
(n)
(m)
r s+ (1)m+n s1/2 r+1/2+ = m+n,1 r+s,0
(6.1)
(note Sm,n = id for all m, n).

(m) (n)
Consider then a given quadratic term r s , for which we define the distance-1 difference
to be r s. Eq. (6.1) then allows us to replace this quadratic term, whenever it appears, by a linear
combination of terms whose distance-1 differences are r s 2 (for 1) and s r 1 2
(for 0), as well as a possible constant term. This replacement results in a strict decrease in
the distance-1 difference when r s > 1/2, hence we find that the vector space of all such
quadratic terms (and constants) is spanned by those which satisfy r s 1/2 (and constants).
We can sharpen this conclusion by considering Eq. (6.1) at the critical difference r s =
1/2:

12
=0
(m)
(n)
(n)
(m) 1
s1/2 s+ (1)m+n s1/2 s+ = m+n,1 s,1/4 .
(6.2)
When m = n, this is vacuous (both sides vanish identically), but for m = n, say m = 0 and n = 1,
(0)
(1)
(1)
(0)
it allows us to express s1/2 s as a linear combination of s1/2 s , other quadratic terms of
strictly lower distance-1 difference, and a constant. Thus, our spanning set for the vector space
of quadratic terms may be culled slightly by imposing the constraint that if r s = 1/2, then
we require that m n.
Consider now a state of the form
1 ) (m ) |0.
r(m
r
1
(6.3)
The monodromy charge of the vacuum is zero, so we must have r Z 1/4. As acting with
a -mode changes the monodromy charge by k/2 = 1/2, ri Z 1/4 when i is even and
(m ,0)
ri Z + 1/4 when i is odd. Similarly, the u(1)-charge 0 of the vacuum also vanishes,
hence r 1/4.
(m) (n)
We have just seen that any quadratic term r s with r s > 1/2, or r s = 1/2 and
m < n may be expressed via the generalised commutation relation with = 1 as a linear combination of those desired terms with r s < 1/2 or r s = 1/2 and m n (and constants). We
can extend this result to states of the vacuum module considered above, so that these constraints
235
apply (simultaneously) to every pair of consecutive modes occurring therein.9 We have therefore
the following result.
(1)
Proposition 6.1. The irreducible A1 -highest weight module L0 is spanned by states of the form

1
1
1 ) (m ) |0

0,
m
{0,
1},
r
Z
+
,
r
,
r(m
(6.4)
i
2i
2i1
r
1
4
4
where r 1/4, ri ri+1 1/2, and ri ri+1 = 1/2 implies mi mi+1 .
We claim that the spanning set given in Proposition 6.1 is actually a basis. To prove this directly, we would have to verify that this set is linearly independent, a difficult task. However, this
can be rigorously confirmed by computing the module character corresponding to the spanning
set of Proposition 6.1, and comparing it to the sum of the characters of the affine modules L (0)
0
(0)
and L 1 (as in Corollary 5.4).
To this end then, let us work out the combinatorics corresponding to this proposed basis. To
(m )
(m )
each basis state r1 1 r |0, we associate a partition10 by
3 1
(6.5)
( i).
4 2
It is easy to check that is indeed a partition, and that if we colour its parts by the mi {0, 1},
then the only condition on the coloured partitions is that if i = i+1 , then mi mi+1 . That is,
if any parts of the partition coincide, those that are coloured 1 must occur to the left of those that
are coloured 0.
Given such a coloured partition , we can split it into two monochromatic subpartitions 0
and 1 (whose parts are all coloured 0 or 1, respectively). Conversely, given any two partitions,
we can put the parts together to make a composite partition, reordering the component parts
to satisfy the partition condition. If we colour the parts so as to distinguish from which of the
original two partitions the part derived, then the only ambiguity in constructing the composite
partition is in deciding the order when a part from one of the constituent partitions coincides with
a part from the other. This ambiguity is resolved by requiring that when the parts coincide, the
parts of one colour (call it 1) are ordered to the left of those of the other colour (0 say).
We therefore have a bijective correspondence between the coloured partitions corresponding
to our proposed basis and pairs of ordinary (monochromatic) partitions. Specifically, we have
proved that

%
%

%
partitions
partitions
coloured partitions &
.
(6.6)
=
of length 0
of length 1
of length
0 +1 =
$
As the conformal dimensions of our basis states are related to the weights || = i i of the
associated partitions by
i = ri +
1
h = || + ( 4),
4
(6.7)
9 This extension is not entirely trivial, but for brevity we will omit it. We will prove this result in much greater generality
in [15], and note that the analogous result for the minimal models will also be presented in [12].
10 We intend to follow the standard practice [18] of denoting partitions by , despite our previous usage of this symbol
for weights. For the remainder of this section we shall use for weights, trusting that our use of will not cause any
confusion.
236
and the sl(2)-weights are related to the colourings by

mi = 0 1 ,
w=2
(6.8)
the character of the A1 -vacuum module is

(1)
0 (q; z) =
q h zw =

q |
0 |+|1 |+ 1 ( + )2
1
0
1
4 0
z0 1
0 ,1 =0 0 ,1
basis
2
2

q (0 +1 ) /4 0 1
q /4
=
z
=
z2m ,
(q)0 (q)1
(q)m (q)m
0 ,1 =0
(6.9)
=0 m=0
'
j
i
where (q)m = m
j =1 (1 q ) as usual, and the primed summation indicates that the partition
has length i . Here, we have used the well known partition identity [18, Eq. (2.2.5)]

q || z =
=0
(
j =1
q
1
=
z .
j
1 zq
(q)
(6.10)
=0
The affine characters are obtained by restricting to be even or odd, by Lemma 5.1. They then
take the form
(1)
0 (q; z) =
2n

n=0 m=0
(1)
1 (q; z) = zq 1/4
qn
z2(nm)
(q)m (q)2nm
2n+1

n=0 m=0
and
q n +n
z2(nm) ,
(q)m (q)2n+1m
2
(6.11)
where the subscript denotes the highest weight.

It is shown in [19] that these formulae (ignoring the z-dependence), agree with the usual
bosonic character formulae, completing our proof that the spanning set given in Proposition 6.1
is indeed a basis as claimed.
(1)
Theorem 6.2. A basis for the (irreducible) A1 -vacuum module L0 is given by the states:

1
1
1 ) (m ) |0

0,
m
{0,
1},
r
Z
+
,
r
,
r(m
(6.12)
i
2i
2i1
r
1
4
4
where r 1/4, ri ri+1 1/2, and ri ri+1 = 1/2 implies mi mi+1 . The character of
this module is
(1)
0 (q; z) =

q (0 +1 ) /4 0 1
z
.
(q)0 (q)1
(6.13)
0 ,1 =0
We mention that the character formula we have derived agrees with that obtained in [7] from a
proposed spinon basis, even though our basis is quite different. Our Theorem 6.2 may therefore
be viewed as an elementary proof of their results (for k = 1).
237
6.2. The A2 characters

The defining operator product expansions when k = 2 have already been given in Section 4.1,
where it was noted (Eq. (4.3)) that the generalised commutation relations with = 1 are supercommutation relations. We claim that considering these supercommutation relations is sufficient
for computing the characters of the A2 -modules. We present a proof of this claim below which
is entirely analogous to that given for k = 1.
(m) (n)
As in Section 6.1, the distance-1 difference r s of a quadratic term r s may be strictly
reduced by using the supercommutation relation if r > s (we swap the modes). The vector space
of quadratic terms is thus spanned by those with distance-1 difference at most 0. When r = s,
we may swap a quadratic term with m < n for one with m > n, and those with m = n are seen to
be equivalent to constants. The analogue of Proposition 6.1 is therefore:
(2)
Proposition 6.3. The irreducible A2 -highest weight module L ( {0, 1}) is spanned by states
of the form

1 ) (m ) |
0, mi {0, 1, 2}, ri Z + ( 1)/2 ,
r(m
(6.14)
r
1
(m ,0)
where r
1/2, ri ri+1 , and ri = ri+1 implies mi > mi+1 .

(m1 )
Let us therefore associate a partition to each basis state r1
(m )
r
| by
i = ri + ( + 1)/2.
(6.15)
(m ,0)
Since /2, it is easy to check that is a genuine partition which, when coloured by the
mi , satisfies the ordering condition that if i = i+1 , then mi > mi+1 .
We therefore decompose such a partition into its three monochromatic subpartitions 0 , 1
and 2 , noting that the ordering condition implies that each m is a partition into distinct parts.
Conversely, given three partitions into distinct parts, we can uniquely merge them into a coloured
partition satisfying our ordering condition. This bijection of partitions leads to an expression for
the character of the vacuum module ( = 0):
(2)
0 (q; z) =

0 ,1 ,2 =0
0 ,1 ,2
q |
0 |+|1 |+|2 | 1 ( + + )
1
2
2 0
z2(0 2 ) .
(6.16)
The prime on the summation indicates that we are summing over partitions m of length m into
distinct parts, and we have noted that the conformal dimension of our state and its sl(2)-weight
are given by

mi = 2(0 2 ),
h + || ( + 1)/2 and 2
(6.17)
i
respectively. The generating function for partitions into distinct parts [18, Eq. (2.2.6)]

=0
q || z =
(

j =1

q (+1)/2
z ,
1 + zq j =
(q)
=0
(6.18)
238
then gives
0(2) (q; z) =
q (0 +1 +2 )/2 2(0 2 )

z
.
(q)0 (q)1 (q)2
2
0 ,1 ,2 =0
(6.19)
Dropping the z-dependence, this is exactly the character of the fermionic vacuum module cubed,
(2)
which we noted in Section 5.4 is the character of L0 . This proves that our spanning set is indeed
a basis (for the vacuum module), because it gives the correct character.
(m ,0)
To compute the other character ( = 1), there is one further constraint, because = /2
in general (but equality always holds when = 0). We therefore need to impose the condition
32 (m ,0) . This bound is 2 or 1, according as to whether m is 0 or not. It follows that the
monochromatic subpartition 0 must be further constrained not to have a part equal to 1. Such
subpartitions are in bijective correspondence with partitions into distinct parts (subtract 1 from
each part of the former), so this additional condition is easily accommodated. The final result for
the character is then
1(2) (q; z) = zq 3/16
q (0 +1 +2 )/2+(0 1 2 )/2 2(0 2 )

z
.
(q)0 (q)1 (q)2
2
0 ,1 ,2 =0
(6.20)
If we substitute z = q 1 into Eq. (6.18), we derive that

q (1)/2
=0
(q)
(

j =0
(

q (+1)/2
.
1 + qj = 2
1 + qj = 2
(q)
j =1
(6.21)
=0
It now follows that

(2)
1 (q; 1) =
)
*3
1 1/16 q (1)/2
,
q
2
(q)
(6.22)
=0
that is, half the cube of the fermionic module whose highest weight state has conformal dimen(2)
sion 1/16. As this is indeed the correct character of L1 (Section 5.4), we may again conclude
that our spanning set is in fact a basis for this module. We have therefore completed the proof of
the following theorem:
(2)
Theorem 6.4. A basis for the (irreducible) A1 -vacuum module L0 is given by the states:

1
(m1 )
(m )
0, mi {0, 1, 2}, ri Z
,
r1 r |0
(6.23)
2
where r 1/2, ri ri+1 , and ri = ri+1 implies mi > mi+1 . The corresponding character is
(2)
0 (q; z) =

0 ,1 ,2 =0
q (0 +1 +2 )/2 2(0 2 )

z
.
(q)0 (q)1 (q)2
2
(6.24)
(2)
A basis for the (irreducible) A2 -module L1 is given by the states

1 ) (m ) |
0, mi {0, 1, 2}, ri Z ,
r(m
1
r
1
(6.25)
239
where r 0, r = 0 implies m > 0, ri ri+1 , and ri = ri+1 implies mi > mi+1 . The corresponding character is
(2)
1 (q; z) = zq 3/16

0 ,1 ,2 =0
q (0 +1 +2 )/2+(0 1 2 )/2 2(0 2 )

z
.
(q)0 (q)1 (q)2
2
(6.26)
References
[1] V. Knizhnik, A. Zamolodchikov, Current algebra and WessZumino model in two dimensions, Nucl. Phys. B 247
(1984) 83103.
[2] E. Witten, Non-Abelian bosonization in two dimensions, Commun. Math. Phys. 92 (1984) 455472.
[3] D. Gepner, E. Witten, String theory on group manifolds, Nucl. Phys. B 278 (1986) 493549.
[4] G. Felder, K. Gawedzki, A. Kupiainen, The spectrum of WessZuminoWitten models, Nucl. Phys. B 299 (1988)
355366.
[5] P. Di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Graduate Texts in Contemporary Physics,
Springer-Verlag, New York, 1997.
[6] D. Bernard, V. Pasquier, D. Serban, Spinons in conformal field theory, Nucl. Phys. B 428 (1994) 612628, hepth/9404050.
[7] P. Bouwknegt, A. Ludwig, K. Schoutens, Spinon bases, Yangian symmetry and fermionic representations of Virasoro characters in conformal field theory, Phys. Lett. B 338 (1994) 448456, hep-th/9406020.
[8] P. Bouwknegt, A. Ludwig, K. Schoutens, Spinon basis for higher level SU(2) WZW models, Phys. Lett. B 359
(1995) 304312, hep-th/9412108.
[9] B. Feigin, M. Jimbo, T. Miwa, E. Mukhin, Y. Takeyama, A monomial basis for the Virasoro minimal series
M(p, p ): The case 1 < p /p < 2, Commun. Math. Phys. 257 (2005) 395423, math.QA/0405468.
[10] P. Jacob, P. Mathieu, A quasi-particle description of the M(3, p) models, Nucl. Phys. B 733 (2006) 205232, hepth/0506074.
[11] P. Jacob, P. Mathieu, Embedding of bases: From the M(2, 2 + 1) to the M(3, 4 + 2 ) models, Phys. Lett. B 635
(2006) 350354, hep-th/0511040.
[12] P. Mathieu, D. Ridout, The extended algebra of the minimal models, in preparation.
[13] P. Goddard, D. Olive, KacMoody algebras, conformal symmetry and critical exponents, Nucl. Phys. B 257 (1985)
226252.
[14] F. Bais, A. Taormina, Accidental degeneracies in string compactification, Phys. Lett. B 181 (1986) 8790.
[15] P. Mathieu, D. Ridout, Characters of the extended algebra of the SU(2) WessZuminoWitten models, in preparation.
[16] V. Kac, Infinite-Dimensional Lie Algebras, Cambridge Univ. Press, Cambridge, 1990.
[17] K. Thielemans, A M ATHEMATICA package for computing operator product expansions, Int. J. Mod. Phys. C 2
(1991) 787798, Package available at http://www.hammersmithimanet.com/~kris/string/.
[18] G. Andrews, The Theory of Partitions, in: Encyclopedia of Mathematics and its Applications, vol. 2, Addison
Wesley, Reading, 1976.
[19] R. Kedem, B. McCoy, E. Melzer, The sums of Rogers, Schur and Ramanujan and the BoseFermi correspondence in
(1 + 1)-dimensional quantum field theory, in: P. Bouwknegt, et al. (Eds.), Recent Progress in Statistical Mechanics
and Quantum Field Theory, World Scientific, New Jersey, 1995, pp. 195219, hep-th/9304056.
Tachyon condensation on the elliptic curve

Suresh Govindarajan 1 , Hans Jockers , Wolfgang Lerche,
Nicholas P. Warner 2
Department of Physics, Theory Division, CERN, Geneva, Switzerland
Received 11 May 2006; accepted 11 December 2006
Abstract
We use the framework of matrix factorizations to study topological B-type D-branes on the cubic curve.
Specifically, we elucidate how the brane RR charges are encoded in the matrix factors, by analyzing their
structure in terms of sections of vector bundles in conjunction with equivariant R-symmetry. One particular
advantage of matrix factorizations is that explicit moduli dependence is built in, thus giving us full control
over the open-string moduli space. It allows one to study phenomena like discontinuous jumps of the cohomology over the moduli space, as well as formation of bound states at threshold. One interesting aspect
is that certain gauge symmetries inherent to the matrix formulation lead to a non-trivial global structure
of the moduli space. We also investigate topological tachyon condensation, which enables us to construct,
in a systematic fashion, higher-dimensional matrix factorizations out of smaller ones; this amounts to obtaining branes with higher RR charges as composites of ones with minimal charges. As an application, we
explicitly construct all rank two matrix factorizations.
E-mail address: hans.jockers@cern.ch (H. Jockers).

1 On leave from the Department of Physics, Indian Institute of Technology Madras, Chennai 600036, India.
2 On leave from the Department of Physics and Astronomy, University of Southern California, Los Angeles, CA
90089-0484, USA.
doi:10.1016/j.nuclphysb.2006.12.009
S. Govindarajan et al. / Nuclear Physics B 765 [PM] (2007) 240286
241
1. Introduction
1.1. Motivation and overview
D-branes play a very important rle in our understanding non-perturbative properties of string
and field theories, as well as in building semi-realistic models. However, the naive geometrical
notion of a D-brane, in which it is thought of as wrapping some p-dimensional cycle of a Calabi
Yau manifold, is a classical concept that is valid only in certain limiting situations, such as the
large radius limit. When distances are small or curvatures large, quantum corrections tend to
blur notions of classical geometry, such as the dimension of a wrapped submanifold. Moreover,
branes can become unstable and decay in ways that are not visible classically.
Therefore one needs to adopt a more suitable language for describing general D-brane configurations. For topological B-type D-branes, the proper mathematical framework is a certain
enhanced, bounded derived category of coherent sheaves [13] (and via homological mirror
symmetry [4,5], this maps to the Fukaya category of A-type branes wrapping special Lagrangian
cycles). This framework retains more data than the more familiar characterization just in terms of
K-theory (i.e., RR charges) and thus provides a much sharper description of D-branes. That is,
the category also contains the information about the brane locations, and other possible (bundle
or sheaf) moduli. For instance, a configuration consisting of an anti-D0-brane located at some
point 1 of the compactification manifold, plus a D0-brane located at some other point 2 , is trivial from the K-theory point of view, but is a non-trivial object in the categorical description as
long as 1 = 2 . Obviously, this extra information is crucial for understanding questions such as
whether, in a given D-configuration, deformations are obstructed or not (i.e., what is the effective
superpotential and the moduli space of its flat directions). Moreover, the language of categories
is tailor-made for addressing questions about stability and bound state formation, which can be
described more physically by tachyon condensation. Excellent reviews of these matters may be
found in Refs. [6,7].
Often physicists associate derived categories with just an abstract collection of objects (the
D-branes) and maps (open strings) between them, and wonder what concrete physical benefit
such a picture might provide. Indeed, by merely tracing arrows around a quiver diagram, all
one obtains is a list of possible terms in the effective superpotential and these terms are merely
added up with unit coefficients. However, there is more to these maps than just being pointers
between objects: in general they depend on the various parameters like brane-location moduli,
and thus encode valuable extra information beyond mere combinatorics. Thus, superpotential
terms derived from quiver diagrams will, in general, have pre-factors depending on the various
moduli of the geometry, a fact that is often neglected in the physics literature.
The abstract notions of objects and morphisms (maps) can, in fact, be easily translated into a
language more familiar to physicists via the following two logical steps. First, as has been proven
recently [8] in quite some generality (see also [9]), the relevant category of topological B-type
D-branes is isomorphic to a certain category of matrix factorizations [10,11], which encodes
the specific D-geometry in question. Second, such matrix factorizations have a direct interpretation [12,13] in terms of two-dimensional topological (twisted N = 2 supersymmetric) boundary
LandauGinzburg theory [14,15]. Specifically, the maps alluded to above feature (partly) as
boundary superpotentials. Thus, via this chain of arguments, boundary LandauGinzburg theory
provides a very explicit realization of the topological field theory of B-type D-branes. A sample
computation was presented in Ref. [16] demonstrating how it can be used, in conjunction with
242
mirror symmetry, to explicitly determine moduli-dependent, instanton-corrected contributions to

superpotentials on intersecting branes.
The purpose of the present paper is to use the language of matrix factorizations to develop,
from a physicists point of view, a better understanding of tachyon condensation and the process
of composite formation of B-type D-branes. Specifically, we will analyze, in some detail, Dbranes on the cubic curve, , which is the simplest situation with both bulk and boundary (brane)
moduli and can be studied fairly explicitly.
Mathematically, the classification of bundles on the elliptic curve is a completely solved problem [17]: An indecomposable bundle is uniquely determined by rank and first Chern class of the
bundle E, plus a continuous parameter : (r(E), c1 (E), ) (N2 , N0 , ). In physical terms, this
corresponds to the number of D2 and D0-branes plus, essentially, the location of the D0-brane
on . The mirror map to the Fukaya category of A-type branes is understood as well [18]. Matrix factorizations describing bundles on have been described in Ref. [19], but only for fixed
moduli and not via tachyon condensation.3 We will make use of these mathematical results to
construct and analyze matrix factorizations explicitly depending on moduli, extending prior work
[16,21] in a systematic fashion. Specifically we will show how the bundle data (r(E), c1 (E), )
are explicitly encoded by certain properties of the matrices. This leads to an algorithm that allows
one to recover the brane data encoded in a given matrix factorization.
As an application of this we study (topological) tachyon condensation [22] of pairs of branes.
This problem has two parts: first, determining the open-string spectrum by solving the relevant
cohomology problem, and second, identifying the bundle data (N2 , N0 , ) of the matrix factorization that results from perturbing the direct product of matrix factorizations with an open string
cohomology element. While we will encounter a few minor subtleties (such as discontinuous
jumps of the cohomology upon varying moduli, and the formation of composites at threshold),
the physical results are entirely as expected: Brane composites can be formed according to the
vector addition of brane charges, provided one properly chooses the perturbing tachyonic operators and appropriately tunes the moduli.
We will show explicitly how all configurations with rank r = N2 2 can be built out of a
minimal generating set of two-dimensional factorizations, which correspond to D-branes whose
RR charges generate the full charge lattice. It is pretty clear that by iteratively applying the same
logic, brane composites corresponding to arbitrary points (N2 , N0 ) on the charge lattice can be
generated. Of course, this does not come as a surprise, but this is not the point of the present
paperthe point is to understand how rather abstract mathematical concepts can be realized in
a concrete physical framework, namely boundary LandauGinzburg theory and how to understand condensation within that framework. We expect that the insight we gain will be useful for
attacking more complicated geometries, like branes on threefolds.
The plan of the paper is as follows: In the remainder of this section, we will review the description of B-branes by matrix factorizations in very simple terms; this is aimed at non-experts.
Moreover, we will outline the main points of tachyon condensation in such models and present
a few examples. Section 2 is then devoted to a general discussion of how the bundle data of a
given brane configuration are encoded in the corresponding matrix factorization. An important
rle is played by the holomorphic sections of bundles on the elliptic curve, which are given by
3 While writing up this paper, we received a paper [20] that also deals with matrix factorizations pertaining to the
elliptic curve, and which has some overlap with our work.
243
Riemann theta functions and Appell functions, for rank one line bundles and rank two vector
bundles, respectively.
In Section 3 we will reconsider the known 2 2 and 3 3 dimensional factorizations, and
discuss in detail their structure in terms of transition functions of the relevant bundles. Section 4
deals with the open-string moduli space of the 3 3 factorization, which has a non-trivial global
structure due to gauge symmetries inherent to the factorization; also, we will find how its moduli
space can be compactified by adding an exceptional 4 4 factorization at the boundary, which
appears to describe a pure, rigid anti-D2-brane.
In Section 5 we address how to properly formulate tachyon condensation in term of equivariant R-symmetry; this is necessary for disentangling the various different branes that are described
by a given matrix factorization, and for subsequently identifying the various tachyon channels between pairs of them. Section 6 provides some more tools for obtaining new matrix factorizations
from known ones; in particular we introduce certain bound states at threshold, which resolve a
certain singularity in the multi-brane moduli space. Finally, in Section 7 we apply the techniques
developed in the preceding sections, and show how the factorizations describing certain rank two
vector bundles can be systematically generated by condensation of lower rank branes. We have
also relegated some more technical material to the appendices.
1.2. Recapitulation: B-type branes on the elliptic curve
We will consider B-type branes on the cubic curve, , defined by the following hypersurface
in P2 :

W (x) x13 + x23 + x33 3ax1 x2 x3 = 0.
(1.1)
Here, a is the complex structure modulus that is related to the standard modulus of the torus via:

3a(a 3 + 8) 3
(1.2)
= j ( ),
a3 1
where j ( ) = q 1 + 744 + is the familiar modular invariant function in terms of q = e2i ,
and is the flat coordinate of the complex structure moduli space, which coincides with the
The coordinates on the cubic can be uniformized in
Khler modulus of the mirror curve, .
terms of theta functions by writing x = ( ), where

1
(1 ) 12
(1)
3
( ) ( | ) =
(1.3)
3, 3 ,
12
with e2i/3 . For further details we refer the reader to Appendix A. Here, is an arbitrary
point on the Jacobian of , which coincides with itself. One should also note that ( ) is
not a function on , but is actually a section of the line bundle, L3 , with first Chern number
c1 = 3. This will lead to a natural ambiguity in determining the bundle data associated with a
given matrix factorization.
As has been discussed by now in many papers [12,13,2336], topological B-type D-branes
can be described by a two-dimensional, twisted N = 2 supersymmetric boundary Landau
Ginzburg model based on matrix factorizations of the form
J (x)E(x) = E(x)J (x) = W (x)1nn .
(1.4)
244
Here J (x) and E(x) are n n polynomial matrices4 with values in C[x1 , x2 , x3 ], whose precise
form depends on the D-brane configuration in question. It will be important later to make use of
the fact that (1.4) is invariant under gauge transformations of the form:
J (x) UL (x)J (x)UR (x),
E(x) UR1 (x)E(x)UL1 (x),
(1.5)
for polynomial matrices UL,R (x) that are invertible over C[x1 , x2 , x3 ]. In particular, this means
we can do arbitrary row and column reduction operations on J (respectively, E) so long as
one does the corresponding inverse operations on E (respectively, J ). For further details of this
procedure we refer the reader to Appendix B.
Mathematically, matrix factorizations form a category [10,11] with objects P of the form
P1
EP ,
P
(1.6)
JP
P0
where P0 and P1 are certain projective modules over C[x1 , . . . , xm ]. This composite form of
objects, P , has a simple physical interpretation. Recall that the usual bulk LandauGinzburg action is not conformally invariant, but represents an action that will flow, in the infra-red, to the
conformal field theory of interest, and the superpotential will remain unrenormalized along the
flow. By the same token, in the boundary theory, the constituents P0,1 correspond to D-branes
and anti-D-branes in C[x1 , . . . , xm ] and the maps J and E correspond to tachyon profiles that
trigger the condensing of P0,1 into P [12,25]. Thus we construct a particular D-brane, P , by setting up an action that will generate it via an infra-red boundary flow (i.e., tachyon condensation).
In this formulation, J has the interpretation as a boundary superpotential while E appears in a
modified chirality condition of fermionic boundary superfields [37,38].
Anti-D-branes are associated with objects commonly denoted by P [1], and look like P
in (1.6) except that JP EP and EP JP ; we will often use the notation P for them.
The trivial object in the category, denoted by V , is described by the simplest factorization,
P11 , for which J = 111 and E = W 111 , or vice versa. It corresponds to the situation where
P0 and P1 have completely annihilated to leave no net branes at all. Factorizations of different
dimension are thus considered to be equivalent if they differ by appending or removing such
trivial matrix blocks.5
On the cubic curve, , the simplest non-trivial factorizations are two- and three-dimensional [16,21]. The first ones are given by:

1 Q2
,
J22 = Q
L2 L1
P22 :
(1.7)

E22 = L1 Q2 ,
L2 Q1
4 We do not require n to be the dimension of a Clifford algebra, which would be needed if we introduced boundary fermions [14]. The implementation of more general boundary couplings not involving fermions, was discussed in
Refs. [2529].
5 In the derived category of matrix factorizations these objects are called perfect complexes and are divided out [8],
because they do not have any non-trivial morphisms with any other object in the category. This means that there are no
open-string states associated to these perfect complexes, and hence such configurations are isomorphic to the open-string
vacuum [7].
245
where the linear entries read

L1 = 3 x1 2 x3 ,
L2 = 3 x2 + 1 x3 ,
(1.8)
and the quadratic entries are6

1
1 2 x12 + 22 x1 x2 12 x22 1 3 x32 ,
1 2 3

1 2 2
Q2 =
x 12 x1 x2 1 2 x22 + 32 x1 x3 .
1 2 3 2 1
Q1 =
(1.9)
There is also the three-dimensional factorization given by:

x x x
1 1
2 3
3 2
J33 = 3 xx3 1 xx2 2xx1 ,
2 2
3 1
1 3
3
1
2
1 2
1 2
1
2
P33 :
1 x1 2 3 x2 x3 3 x3 1 2 x1 x2 2 x2 1 3 x1 x3
E33 = 1 x32 2 x1 x2 1 x22 1 x1 x3 1 x12 3 x2 x3 .
2
1 3
1
2 3
3
1 2
3
2
1
1 2
1 2
1 2
3 x2 1 2 x1 x3
2 x1 1 3 x2 x3
(1.10)
1 x3 2 3 x1 x2
Both (1.7) and (1.10) represent valid matrix factorizations satisfying (1.4) precisely when the
parameters, just like the coordinates x , satisfy the cubic equation:
3

1 + 23 + 33 3a1 2 3 = 0.
(1.11)
One may therefore uniformize these parameters by taking, once again:
= ( ),
(1.12)
for some parameter, . The physical interpretation is that the factorization parameters, , are
moduli of the D-branes, and is the associated flat coordinate that labels a point on the curve
corresponding to the location of the D0-brane component of the (D2, D0) brane configuration.
As mentioned above, (indecomposable) B-type D-branes are labeled by their RR charges
and location: (r(E), c1 (E), ) (N2 , N0 , ). In discussing the corresponding holomorphic vector
bundles we will adopt a common notation, E(r, c1 ), that suppresses the parameter, . We will also
use the notation, Ln , to denote the nth power of the degree-one line bundle, L.
The question naturally arises as to the precise map between these bundle data and the structure
of the matrices J and E that define a given factorization, P . This will be discussed in detail in
Sections 2 and 3 below. We recall here that the two factorizations under discussion have been
shown to correspond each to a triplet of branes with the following charges:

S P22 : (r, c1 )LG (Sa ) = (1, 0), (0, 1), (1, 1) ,

L P33 : (r, c1 )LG (La ) = (2, 1), (1, 1), (1, 2) .
(1.13)
6 This particular matrix factorization is valid for = 0 because then L and L are linear independent. For the
3
1
2
singular limit 3 = 0 one needs to go to a different coordinate patch [16]. For 1 = 0 the matrix E also becomes
singular. However, this can be fixed by apply a gauge transformation of the form Q1 Q1 +
Q2 22 1 2 32 x1 L1 . Similarly one proceeds for 2 = 0.
22
1 2 32
x1 L2 , Q2
246
Fig. 1. The 2 2 and 3 3 factorizations correspond, via mirror symmetry, to D1-branes that stretch along the short (S)
and long (L) diagonals on the elliptic curve, respectively. Here we show these D1 branes (suppressing the anti-branes
with opposite charges, which correspond to factorizations where J and E are exchanged). This figure also provides a
useful and simple graphical representation of the RR charge lattice for the B-branes.
As indicated above, we will denote the two factorizations by S and L, each comprising three
branes denoted by Sa and La .7 These charge assignments, at least for the 2 2 factorization,
were originally computed rather indirectly [16,26]: The intersection matrices were computed
using the matrix formulation and then the results were compared with the intersection matrices
computed from the conformal field theory and from the geometry. Part of our purpose here is to
give a far more direct algorithm for computing these geometric data from the matrices.
Under mirror symmetry, Khler and complex structure moduli exchange and the B-type
branes map into A-type D1-branes, which are labeled by the winding numbers p and q:
(r, c1 )B = (p, q)A . Moreover, the brane modulus, , maps into a complex modulus comprising both position and Wilson line moduli of the D1-brane. Thus, the lattice of brane charges can
be drawn on the covering space of , as shown in Fig. 1.
We have listed in (1.13) the charges corresponding to the Gepner-point in the Khler moduli
space, which is natural from the LandauGinzburg perspective. However, recall that RR charges
are ambiguous due to monodromy, or flow of gradings [1,41], in the Khler moduli. Matrix
factorizations, which pertain to the topological B-model and thus depend only on the complex
structure moduli, are insensitive to variations of the Khler moduli and therefore cannot distinguish bundles differing by such monodromies. For example, looping around the large radius limit
induces a monodromy that amounts to tensoring with powers of the line bundle L3 , which means
that the first Chern number of a rank-r bundle will jump by 3r.
7 In BCFT language, the L are the RecknagelSchomerus branes [39,40] with smallest charges, which correspond
a
to holomorphic vector bundles inherited from the ambient P2 and as such do not generate the full RR charge lattice. The
Sa correspond to the recently-discovered permutation branes [26,33,35] that form an integral basis of the charge lattice
on the curve.
247
Fig. 2. This shows schematically where boundary preserving () and boundary changing (, ) operators are located
on intersecting D-branes and on the boundary of the world-sheet, Z.
On the other hand, performing a partial monodromy by moving from the Gepner point to
the large radius limit, the charges (1.13) flow according to tensoring with the line bundle8 L2 ,
i.e., (r, c1 ) (r, c1 2r), and this results in the following list of charges [41,43]:

2 2: (r, c1 )LR (Sa ) = (1, 2), (0, 1), (1, 1) ,

3 3: (r, c1 )LR (La ) = (2, 3), (1, 3), (1, 0) .
(1.14)
In the following, we will adopt this labeling convention because it refers to the large radius limit,
which is semi-classical from the point of view of the sigma-model and coincides with the labeling
in the mathematics literature.
In order to discuss topological tachyon condensation, we need to determine the relevant part
of the open-string spectrum. Since, in the twisted theory, the tachyons become fermionic operators [44] (coupling to bosonic deformation parameters), we will consider only fermionic
operators here. There are two classes of such operators. First, there are boundary preserving
operators, represented
by2nP 2nP dimensional, block off-diagonal matrices of the form:

P (P ,P ) = E0P J0P , which are tied to a single brane and describe moduli corresponding
to infinitesimal deformations of the brane (such as position shifts). Most of the physics literature on open-string TFT deals with this class only. Mathematically these operators correspond to
endomorphisms of the object P .
The other class consists of boundary changing operators, (P ,Q) , which correspond to open
strings stretching between pairs of branes P and Q and thus are localized at their intersection.9
See Fig. 2. These are the topological version of tachyons, and indeed they typically have Rcharges q < 1, which means that they are relevant operators inducing a non-trivial boundary RG
flow. They can be written
in terms of 2nP 2nQ dimensional, block off-diagonal matrices of the
form: (P ,Q) =
0 0
1 0
. The rle of 0 and 1 as maps between the composite objects P and
8 This can be understood from a linear sigma model point of view [9]. Note that this is also closely related to Seiberg
dualities [42].
9 Non-intersecting branes have a trivial topological open-string spectrum between them because the open strings are
massive and so are not part of the cohomology.
248
Fig. 3. The quiver diagram displaying the fermionic open string states related to the short- and long-diagonal branes, Sa
and La . The arrows depict the multiplicities. For simplicity, we do not show the anti-branes Sa , L a .
Q can be visualized by the following diagram:

P1
JP
EQ
JQ
EP
P0
Q1
1
0
(1.15)
Q0
There is a similar structure for the bosonic operators, .

The physical operators then correspond to thenon-trivial
cohomology elements of the BRST

0 J
operator, which can be written in the form: Q = E 0 . That is, we require them to be closed:
0 = E Q 0 + 1 JP ,
0 = J Q 1 + 0 E P ,
(1.16)
modulo exactness
0ex = JQ 0 1 JP ,
1ex = EQ 1 0 EP ,
(1.17)
for any choice of matrices 0 and 1 .

The open-string spectrum pertaining to the 2 2 and 3 3 matrix factorizations can be represented by the quiver diagram shown in Fig. 3. The number of arrows indicates the number
of inequivalent fermionic cohomology elements, and coincides with the number of intersection
points of the mirror A D1-branes on the curve. For each arrow there is implicitly another one run-
249
ning in the opposite direction, which corresponds to the Serre dual, bosonic operator and which
we do not show.10 Moreover, closed loops denote the boundary preserving deformations, P .
1.3. Examples of tachyon condensation
To illustrate one of the basic techniques we use in this paper, we sketch the two simplest
possible examples of tachyon condensation; a more extensive analysis will be presented in the
subsequent sections. To recapitulate the basic point: One wants to find, and turn on, a suitable
boundary changing operator (P ,Q) , whose effect is to condense two sets of branes, P and Q,
to form some composite, R. This process can be visualized by collapsing the two-brane system
shown in Eq. (1.15) into a single object as follows:
Q1
P1
R1
1
EQ JR
JQ
ER R,
JP
EP
(1.18)
0
R0
Q0
P0
where the composite maps, JR and ER , can be thought of as (nP + nQ ) (nP + nQ ) block
matrices of the form:

JQ 0
E Q 1
JR =
(1.19)
,
ER =
.
0 JP
0 EP
If is a non-trivial cohomology element, these matrices satisfy the factorization condition (1.4)
and represent a new B-type of D-brane.11 In the following, we will often use the following
shorthand notation to denote the process of tachyon condensation:
P Q R.
(1.20)
Note that this construction is well-known in the mathematical literature and goes by the name of
the cone construction. That is, one writes a sequence of maps in the form of a distinguished
triangle:
P [1]
[1]
P,
(1.21)
where the composite R coincides with what is called the mapping cone (for details see, for
example, Ref. [7], and for subtleties concerning off-shell versus on-shell physics, see Refs. [2,
6]).
In practice, one would like to find a simple way to determine exactly what this new D-brane
is. From the point of view of matrix factorization, the obvious, but rather impractical method is
to try to reduce the matrices to some standard set of canonical forms. The most efficient way of
achieving this end is, in fact, to determine the underlying bundle data for the new brane, R. Of
course, as far as the K-theoretic data are concerned, ranks and first Chern classes of bundles are
additive under condensation and thus can be trivially determined. It is, however, not so obvious
how to determine data beyond RR charges, that is, the extent to which R is decomposable, and
10 Fermionic operators
(P ,Q) correspond to Ext(P , Q) while bosonic operators (Q,P ) correspond to Hom(Q, P ).
Serre duality implies Hom(Q, P ) Ext(P , Q).
11 Note that in principle the maps need not be upper triangular. Factorization then becomes a highly non-trivial condition, which in general is satisfied only on a sub-locus of the combined open/closed string moduli space; for examples,
see Ref. [30].
250
Fig. 4. In addition to the two simple condensation processes discussed in this section, we show here how the rank two
respectively, by switching on suitable
composites can be generated by condensing S and L branes or anti-branes, S and L,
tachyons. The anti-branes are denoted by dashed arrows. Details will be discussed in Section 7.
how the parameters of P and Q combine into the parameter(s) of R. Here we illustrate this issue
with two examples.
The simplest possible example is combining a pair of 2 2-matrix factorizations. There are
two ways to achieve this, namely either combining a pair of branes, or a brane with an anti-brane.
We start with the second possibility and take the first D-brane to be given by the 2 2-matrix
factorization S() = (J (), E()) in Eq. (1.17), while the anti-D-brane is represented by the
2 2-matrix factorization S()

= (E(), J ()). The outcome of the condensation depends
on which precise members, Sa and Sb , of the two factorizations S and S we choose to condense,
and this is tied to which specific tachyonic operator in the cohomology between the factorizations
we choose to switch on.
From the vector addition of RR charges shown in Fig. 4 we expect that there should be two
types of tachyonic perturbations, which either lead to complete annihilation, or to a composite
The simplest possibility is of course the complete
corresponding to a 3 3 factorization, L or L.
annihilation of the brane/anti-brane pair, which we will discuss momentarily; the other, more
involved situation will be analyzed in Section 5.
For generic values of the moduli, i and i , of the branes S1 () and S1 (), the relevant cohomology of fermionic open-string operators (determined by the physical state condition (1.16)
modulo (1.17) turns out to be empty. This reflects the fact that, in the A-model mirror picture,
the anti-parallel D1-brane/anti-brane pair does not intersect, so that there is no operator that can
be used to form a tachyon condensate. However, upon tuning i = i it is easy to check that the
cohomology jumps and now contains the fermionic operator (S1 ,S1 ) (0 , 1 ) given by
0 = 122 ,
1 = 122 .
(1.22)
This tuning of the open-string moduli corresponds to the situation where the anti-parallel brane
and anti-brane move on top of each other and where the Wilson line moduli of the branes are
tuned to match. In fact the cohomology then also contains an extra bosonic operator, so that the
251
intersection index S1 ,S1 = TrS1 ,S1 (1)F = dim(Hom(S1 , S1 )) dim(Ext(S1 , S1 )) = 0 does not
change.12
To see that the result of the condensation induced by (1.22) is indeed the expected trivial
ground state, one can make use of the gauge transformations (1.5), and most particularly, of the
constant entries of the 0,1 , to make elementary row and column reductions of JR and ER to
show that they are gauge equivalent to the trivial factorization (given by matrices with either 1 or
W on the diagonal). See Appendix B for more discussion of gauge transformations and row and
column reduction.
The next-to-simplest possibility is to condense two of the same type of matrix factorization,
but with different values of the moduli: S() condensed with S(). We find that there are two
distinct choices for the tachyon, and the choice of the tachyon again relates to which particular
branes, Sa () and Sb (), in the families described by the factorizations, S() and S(), participate in the condensation. In Section 4 we will analyze the interrelation between the tachyonic
spectrum and the choice of brane pairs in complete detail, using equivariant R-symmetry. For
the present we will choose the tachyon that leads to the condensation of S1 and S2 to form a
composite anti-brane, S3 (cf. Fig. 4).
This tachyon, represented by the boundary changing fermionic operator (S1 ,S2 ) (, ), is
found to be13

H1 (, ) H2 (, )
G2 (, ) H2 (, )
0 (, ) =
,
1 (, ) =
.
G1 (, ) G2 (, )
G1 (, ) H1 (, )
(1.23)
The constant entries are given by14
G1 (, ) = 1 32 3 1 3 ,
G2 (, ) = 2 32 + 3 2 3 ,
while the linear entries are:

1 2
1 1
3 3
H1 (, ) = 2
x1 1
x2 3
x3 ,
1
2 2
2 2
2

1 2 2 3 12 3 22
2 1
+
H2 (, ) =
x1 + 1
x2 .
2
1
2 3
1 3
2
(1.24)
(1.25)
The fermionic operator (1.23) can be used to construct a condensate R via the cone construction
mentioned above, and one obtains the following 4 4-matrix factorization:

J () 0 (, )
E() 1 (, )
,
ER (, ) =
.
JR (, ) =
(1.26)
0
J ()
0
E()
Since the tachyon operator, (S1 ,S2 ) , contains the constant entries (1.24), the matrix-factorization
can again be simplified by the process of row and column elimination. After a few straightforward
12 The appearance of such pairs is not a special feature of the 2 2 factorizations, but instead this operator, as well as
its bosonic partner, appear for any anti-parallel brane/anti-brane pair as long as the open-string moduli of both branes
coincide.
13 These expressions have also been obtained by J. Walcher who participated in early stages of this project.
14 These vanish for = . However, at = there exists a cohomology element of different form, for which the
i
i
i
i
following arguments hold analogously.
252
steps of algebra, the matrix factorization (1.26) can be cast into its gauge-equivalent form:
0 L1 ( ) 0 Q2 ( )
0
0 1
0
Q2 ( )
0
1
0
0
Q1 ( ) 0 0
JR ( ) =
, ER ( ) =
,
0
W 0
0
W
0
0
0
L2 ( ) 0 0 L1 ( )
0
L2 ( ) 0 Q1 ( )
(1.27)
where
1 = 22 1 2 1 3 32 ,
2 = 32 2 3 1 2 12 ,
3 = 12 1 3 2 3 22 .
(1.28)
The parameters i also satisfy the cubic torus equation, just like the parameters i in (1.11).
Indeed, the i become much simpler when written in terms of the uniformizing variables and, as
we will discuss in later sections, (1.28) merely represents the addition formula for theta functions.
If we now drop the two trivial D-branes in the matrix factorization (1.27) (each corresponding
to J = 111 and E = W 111 , or vice versa), we finally obtain for the condensate R the following
2 2 factorization:

L1 ( ) Q2 ( )
Q1 ( ) Q2 ( )
JR ( ) =
(1.29)
,
ER ( ) =
.
L2 ( ) Q1 ( )
L2 ( ) L1 ( )
) = (E( ), J ( )), which corresponds to an anti-brane (cf.
We can readily identify it with S(
(1.17)) depending on the open-string modulus, .
2. Holomorphic vector bundles and matrix factorizations
As we discussed earlier, the classification of B-type branes is given by that of holomorphic
vector bundles15 and this is equivalent to finding matrix factorizations. Since the primary focus
of this paper is upon matrix factorization, and how to generate new matrix factorizations from
old ones via tachyon condensation, we will discuss in some detail how one can extract the bundle
data from the matrices. The converse construction is also possible [19], but we will not address
it directly here.
More concretely, we will see in the following how one can associate a bundle with each of the
matrix factors, J and E. We will denote these bundles by EJ and EE , respectively. The charges of
the branes will then be determined by EJ and the charges of the anti-branes by EE . The impatient
reader, who would like to skip the technical details, is invited to move on to Section 3.4, where
we summarize our findings with regard to the brane charges.
Before starting with the discussion, one should also note that the bundle data that we seek
has a natural ambiguity coming from the fact that the coordinates, xj , are themselves sections
of the line bundle, L3 , and since we allow ourselves to multiply and divide vectors by the xj ,
any bundle data that we get will be ambiguous up to tensoring by powers of this line bundle.
Therefore, E(r, d), will be indistinguishable from E(r, d 3r); note that this is compatible with
the ambiguity induced by the large radius monodromy in the Khler moduli space.
15 More generally, B-type branes are described in terms of coherent sheaves, which is a notion more general than vector
bundles. As we will explain later and as pointed out in [26], in order to describe sheaves that are not vector bundles, the
construction detailed below needs to be refined by using equivariant R-symmetry.
253
There is a closely related issue with the definitions of J (x) and E(x): The fact that the xj are
sections of a line bundle means that the matrices will generically have non-trivial transformations
between patches on . That is, suppose that between two patches one has xj gxj for some
transition function, g. Then one has W (x) g 3 W (x) and
J (gx) = g k G1 (g)J (x)G2 (g)1 ,
E(gx) = g 3k G2 (g)E(x)G1 (g)1 ,
(2.1)
where k Z is arbitrary and G1 and G2 are, in fact, matrices that describe the action of the Rsymmetry. We will discuss R-symmetry in more detail in Section 5, but here we note that if the
matrix elements of J and E each have a well-defined (but possibly different) degree, then G1
and G2 will be diagonal with integer powers of g.
2.1. The direct approach
We start by taking the most naive approach to the problem and seeing how far we can get.
Indeed, the most elementary way to exhibit the holomorphic vector bundles associated with a
matrix factorization is to look at the kernels of the matrix factors. Given an n n-matrix factorization of the form (1.4), it follow that
det(J ) = W nr ,
det(E) = W r ,
(2.2)
for some r Z+ . On the surface, , defined by W = 0 in a complex projective space, the kernel
of E is thus a rank-r vector bundle, EJ , on and it is spanned by the columns of J . Conversely,
the columns of E span the rank-(n r) vector bundle, EE , that is the kernel of J . In terms of the
objects, P , in (1.6), the idea is that we are extracting the data about the condensation that leads to
P by passing to the kernels of the maps. The two bundles extracted in this way correspond to the
constituent D-brane and its anti-D-brane, P0 and P1 in (1.6). In more physical terms, the same,
naive argument [45] that leads from the bulk LandauGinzburg model to the surface W = 0 in
projective space, when extended to the boundary suggests that one should look at zeroes of J
and E, for branes and anti-branes respectively, on the boundary. A proper justification of this
argument may, however, require the boundary linear sigma-model [9].
There is a technical problem with the naive construction above: The matrices may have nontrivial transition properties (2.1) and so taking the linear span of columns may not be well-defined
upon . The simplest remedy is to consider the kernels of:

1

1
J(x) G1 xp1 J (x)G2 xp1 ,

(2.3)
E(x)
G2 xp1 E(x)G1 xp1 ,
for some p = 1, 2, 3 and where G1 and G2 are defined in (2.1). This is similar to the gauge
transformation in the sense of (1.5) but it is not invertible in the polynomial ring. It does, however,
make the entries of J and E into rational functions of the xj . However, J and E have well1 =G
2 = 1) and thus E and E are well defined
defined kernels and images on (because G
J
E
in . This is what we really mean by multiplying and dividing by powers of xj , and this is the
reason for the ambiguity, E(r, d + 3j r) for j Z, in the bundle data. We will proceed with the
mild abuse in terminology by thinking of EJ and EE as vector bundles, with the understanding
that they can be turned into vector bundles using the foregoing construction.
Since the matrices are polynomials in the xj , the spanning sets (the matrix columns) of these
vector bundles are holomorphic. These vector bundles will clearly have transition functions induced from those of the xj and it is tempting to assume that these will be sufficient to define
the vector bundles EJ and EE . However, it is not that simple: The columns of J (or E) are not
254
linearly independent and therefore do not constitute a basis. The obvious remedy is to choose
a linearly independent subset of columns and use this as a basis, but the problem is that such a
choice of basis cannot be done globally on : One needs to introduce further patches on and
use different sets of columns in each of these new patches. There will also be non-trivial transition functions between such patches. Mathematically, this amounts to constructing an explicit
local trivialization of the vector bundle. These transition functions can be written as rational functions of the matrix elements and the patches can be arranged so that these transition functions
are holomorphic on the intersections. Thus one can easily see that EJ and EE are holomorphic
vector bundles, and one can, in principle, compute their properties in this manner.
For example, consider the 3 3 factorization given by (1.10). Every matrix element of J
has degree one and every matrix element of E has degree two and so G1 = G2 = 1. Since
det(J ) = W , EE has rank 1, and so the columns of E must all be multiples of one another, which
can easily be verified (mod W = 0). Thus EE must be a line bundle on , and the matrices
G1 and G2 are trivial. Now observe that the j th column of E vanishes identically if xj = 1 ,
xj +1 = 3 and xj +2 = 2 , where the subscripts are taken mod 3. One therefore needs to use at
least two of the columns in two different patches, U1 and U2 , if one is to define the line bundle
globally. The transition functions between these two patches will be a ratio of quadratic functions
in the xj , and thus meromorphic on . Indeed, the patches can be chosen so that the transition
function on U1 U2 is biholomorphic.
From this simple example, we see that to characterize the line bundle we not only need G1 ,
G2 and the properties of the xj , but also the non-trivial transition functions between patches in
which different sets of columns are linearly independent. Thus the naive construction of the holomorphic vector bundles EJ and EE works nicely, but it may not result in the simplest, canonical
description of the bundle.
To simplify and generalize this discussion, we start with a brief review of holomorphic vector
bundles on a torus.
2.2. Holomorphic vector bundles on an elliptic curve
To define the elliptic curve, q , corresponding to the surface with complex structure modulus, q, it is most convenient to think of it as an annulus in the complex plane with the interior
and exterior edges identified. That is, one considers C C\{0} with the identification z qz
and where q C is a parameter. It is also useful to use the additive parametrization where
q = e2i , z = e2i and + 1 + . There is a natural projection map, : C q ,
and given a holomorphic vector bundle on q , one can pull it back to C . One can then show
(see, for example, [18,46]) that the pull-back must be a trivial bundle on C , and so the bundle
on q is determined entirely by the gluing matrix under z qz. The trivial, rank-r bundle on
C is simply C Cr , and to obtain a rank r bundle on q one must specify an invertible r r
matrix, A(z), of holomorphic functions on C and then one has:

Vr (A) (z, v) C Cr : (z, v) qz, A(z)v .
(2.4)
Conversely, the set of non-trivial bundles is determined by the choices of A(z) up to gauge
equivalence: A(z) B(qz)A(z)B 1 (z) for some invertible r r holomorphic matrix, B(z).
Thus we have a complete characterization of the vector bundles on q .
There is also a well-known result of Atiyah [17] that states that every indecomposable holomorphic vector bundle on q is characterized by three parameters: (i) the rank, r, (ii) the first
Chern number, c1 , and (iii) a point, y on q . Indeed, one can always write Vr (A) as L Vr (eN )
255
where L is an appropriately chosen line bundle and N is a constant, indecomposable, nilpotent

r r matrix [18]. We will discuss and illustrate this result extensively in subsequent sections,
but here we will merely note that the classical theta function:
1
q 2 n(n1) (yz)n ,
(z, y)
(2.5)
nZ
is a global holomorphic section of the line bundle V1 (y 1 z1 ) on q , where Ay (z) = y 1 z1 .

Note that this transition function involves a factor of z1 , which has winding number 1, and
thus c1 = 1. Also observe that the transition matrix for (z, qy) becomes Aqy (z) = q 1 Ay (z),
y) and (z, qy) are global
which is gauge equivalent to Ay (z) with B(z) = z1 . Hence both (z,
sections of the same line bundle V1 (y 1 z1 ) and therefore the parameter y may be thought of as
living in q .
Returning to the line bundle spanned by the columns of the 3 3 matrix, E, of (1.10), an
elementary theta-function identity reveals that the matrix elements of E may be rewritten in
terms of theta functions:
Eij =
2 ( )
i ( )j ( + ),
1 2 3
(2.6)
where x = ( ), = ( ) and ( ) is the Dedekind -function. Thus the columns of E are

all holomorphic multiples of the single (nowhere-vanishing) basis vector:

1 ( )
3i( )
v1 e
(2.7)
2 ( ) .
3 ( )
The phase pre-factor has been included so as to render v1 periodic under + 1. Under
+ one has v1 y 3 z3 v1 , where z = e2i and y = e2i , and therefore the holomorphic
transition function A(z) reads
A(z) = y 3 z3 .
(2.8)
(y 3 z3 )
(y 1 z1 ))3 ,
Thus the kernel of J leads to the line bundle, V1

= (V1
or E(1, 3), with
c1 = 3 and a parameter, y. This bundle is, modulo the ambiguity outlined at the beginning of
the section, equivalent to E(1, 3k), k Z and both E(1, 0) and E(1, 3) are the anti-bundles
of two of the bundles whose charges were listed in (1.14). We will discuss the relation with all
the charges listed in (1.14) in more detail after we have identified all the matrix bundles EJ , EE ,
associated with the 2 2 and 3 3 factorizations.
2.3. MCM modules
We started this section by specializing to the surface, W = 0, and then identifying the holomorphic vector bundles associated with the matrix factorization. There is another, somewhat
more abstract, approach that realizes these vector bundles in terms of Maximal CohenMacaulay
(MCM) modules. Since this description is the standard approach to matrix factorizations in the
mathematics literature, it is useful to relate it to our discussion here. Moreover, the formulation
in terms of MCM modules can be used to show that holomorphic vector bundles are sufficient to
reconstruct the matrix factorization (see, for example, Ref. [19]).
The first step is to introduce a local ring, R, defined by the superpotential, W . That is, let
P = C[x1 , . . . , xm ] be the ring of complex polynomials in the xj , and let [W ] be the ideal of P
256
generated by the superpotential, W . One then defines a new ring, R, by:

R C[x1 , . . . , xm ]/[W ],
(2.9)
which is simply the ring of polynomials taken modulo W . Once again one considers the kernels
and images of the matrices E and J , but this time one thinks of these kernels and images as
modules over the ring R. Specifically, J and E are now thought of as maps from Rn to Rn ,
and the columns of the matrices, of course, generate the images of these maps. For instance, the
columns of J generate a module, M, however, this module is not necessarily freely generated.
That is, elements of M can generically be written only as a non-trivial linear combination (over
the ring R) of all the columns, and yet the columns are not linearly independent. There are thus
relations between the columns of J , and the set of such relations are called the first syzygy,
1 (M), of M, which is also an R-module. Indeed, the columns of the matrix E exactly generates this set of relations. Again, the module, 1 (M), is generically not freely generated and so
there are relations between the relations and this defines the second syzygy, 2 (M), of M. In a
matrix factorization the set of such relations between the columns of E is of course generated by
the columns of J again. In other words, we have 2 (M) = M. This identity, in fact, defines an
MCM module.
One can summarize the foregoing by stating that the following is an exact sequence:
Rn
Rn
Rn
Rn
Rn
0,
(2.10)
with a similar sequence with E and J interchanged.

The ideas and techniques of MCM modules are very useful if one is going to generate matrix
factorizations from bundle data [19], but since this is not our focus here, we will not need to
discuss these ideas any further.
3. Some examples of matrix factorizations and their vector bundles
Here we will describe, in detail, the vector bundles associated with the 2 2 and 3 3 factorizations. Before we proceed to the examples, we need to summarize some of the properties of the
classical theta and Appell functions that arise in the description of holomorphic vector bundles
on a torus. More details may be found in Appendix A.
3.1. Classical elliptic functions
In the present paper we mainly deal with line bundles and rank two vector bundles. We therefore briefly discuss their canonical sections, which are given by Riemann theta functions and
Appell functions, respectively. The standard definitions are:
1 2
( ) = ( | )
(3.1)
q 2 n zn ,
nZ
(, ) = (, | )
q 12 n2 zn
nZ
qn y
(3.2)
where y = q m , m Z and
q e2i ,
z e2i ,
y e2i .
(3.3)
257
These functions satisfy the following periodicity relations:

1
( + ) = q 2 z1 ( ),
( + 1) = ( ),
(, + 1) = ( + 1, ) = (, ),

1
( + , ) = q 2 z y(, ) + ( ) .
(3.4)
(, + ) = y(, ) + ( ),
(3.5)
Consider these functions on the torus whose fundamental cell in C is defined by + 1

+ . It is a fairly familiar fact that the theta functions provide global holomorphic sections of
line bundles on this torus: They are periodic under + 1 and have a transition function of
1
q 2 e2i under + . The fact that this function has winding number 1 on the circle
1
defined by [0, 1] means ( ) is a holomorphic section of the line bundle, V1 (q 2 z1 ), with
c1 = +1 on the torus.
The Appell function provides a global holomorphic section of a non-trivial rank two vector
bundle, F . In particular, the vector:

(, )
(3.6)
( )
is globally holomorphic, is periodic under + 1, but has the non-trivial transition matrix
under + :

y
1
.
A(z) =
(3.7)
1
0 q 2 z1
1
This vector bundle, F , is a non-trivial extension of V1 (q 2 z1 ) by V1 (y):

0
V1 (y)
1

V1 q 2 z1
0.
(3.8)
On the cubic curve, , we need the theta functions appearing in (1.3) (see Appendix A for more
details). The functions, , have the following periodicity properties:
( + 1) = ( ),
( + ) = q 2 z3 ( ),
but it is also useful to note that:

+ 13 = (1) ( ),

1
13 = q 6 z+1 ( ),
(3.9)
(3.10)
where the subscript on +1 is taken mod 3. Holomorphy and the periodicity properties on the
torus defined by + 13 + uniquely define the functions ( ). Note that (3.9) implies
that ( ) is not a function on , but is a section of the line bundle, L3 . To be more precise,
e3i ( ) is periodic under + 1, and is a section of L3 = V1 (z3 ), where L V1 (z1 ).
Since the functions ( ) are global holomorphic sections of L3 , each of the ( ) has three
zeroes. In particular, one has:

2 (0) = 3 (0) = i( ).
1 (0) = 1 13 = 1 23 = 0,
(3.11)
The values of the at other third-periods can then be deduced from this using (3.10).
One can define Appell functions, , associated with the (see Appendix A for details).
These Appell functions are defined by their periods:

(, + ) = y 3 (, ) + ( ).
, + 13 = (1) (, ),
(3.12)
258
3.2. The 3 3 factorization revisited

Consider, once again, the matrix factorization defined by (1.10). We have seen that EE is a
line bundle and that a global, non-vanishing holomorphic section can be taken to be16 :

1 ( )
v1 2 ( ) .
(3.13)
3 ( )
The bundle EJ has rank two, and may be defined, via the kernel of E, as the set of vectors
orthogonal to:

1 ( + ), 2 ( + ), 3 ( + ) .
(3.14)
Given that the columns of J are expressed as sections of a line bundle, one might, at first, expect
that the rank two, holomorphic vector bundle, EJ , is itself a trivial sum of line bundles. For
example, one might try taking the last two columns of J as a basis. However, for = + n3 one
has:
x = n(1) ,
(3.15)
and thus the last two columns of J are multiples of one another, and so they do not represent
a good global basis for the vector bundle. In the neighborhood of such points one must use a
different pair of columns as a basis, and the transition function is:

3 x2
1 ( + ) 1 x1
2 ( + ) 2 x3
(3.16)
2 x1 =
3 x3
1 x2 .
3 ( + ) x
3 ( + ) x
x
1 3
2 2
3 1
This change of basis is singular at the three zeroes of 3 ( + ) (i.e. at = + 23 + n3 ). One

therefore has to break the torus into patches if one is to use the columns of J as a basis.
On the other hand, one can use Appell functions to obtain a basis of holomorphic sections for
the two-dimensional vector bundle, EJ . Consider the vectors:

1 x 1
1 1 (, )
v1 3 x3 ,
(3.17)
v2 3 3 (, ) .
2 x2
2 2 (, )
One can show that
1 1 ( + )1 (, ) + 3 2 ( + )3 (, ) + 2 3 ( + )2 (, ) = 0,
(3.18)
provided that:
= 12 .
(3.19)
This means that v2 is in the kernel of E. One can prove this identity by considering the periodicity
properties of the function, F (, ), defined to be the left-hand side of (3.18). One first notes that
the anomalous shift term in the Appell functions amounts to shifting v2 by v1 and since v1 is
in the kernel of E, the shift term does not contribute to F (, ) under + . This means
that, considered either as a function of or as a function of , F (, ) represents a global section
16 For simplicity, we have dropped the pre-factor in (2.7).
259
of a line bundle. One then uses holomorphy and standard theta-function methods to write it in
terms of theta functions or, in this instance, conclude that it is zero provided that one chooses the
Appell function parameter according to (3.19).
Using the same kind of argument one can establish the following identities:

3 x 2
3

6i
98 (3 ) 3i( + )
2 2 , v2 = iq
e
2 ( )v1 e
(3.20)
2 x1 ,
( )
1 x3

2 x 3
3

6i
98 (3 ) 3i( + )
3 ( )v1 e
(3.21)
3 2 , v2 = iq
e
1 x2 .
( )
x
3 1
In other words, the set of columns of J are given by holomorphic linear combinations of the
holomorphic vectors, v1 and v2 .
Finally, observe that the vectors uj e3i vj are periodic under + 1 and under
+ they have the transition matrix:

3 6i
1
q e
.
A(z) =
(3.22)
0
z3
Summarizing, the bundle, EJ , is the non-trivial, non-split bundle with (r(EJ ), c1 (EJ )) = (2, 3).
Up to the ambiguity of tensoring with L3n , this is equivalent to E(2, 3), which indeed belongs
to the charges listed in (1.14).
3.3. Vector bundles of the 2 2 factorization
Consider the 2 2 factorization given by (1.17). The determinants of E and J are both W and
so the kernels of both are one-dimensional. Because the matrix elements have different degrees,
the matrices in (2.1) are now non-trivial, that is, taking g = q 3/2 e6i one has:

J x( + ) = g k G1 (g)J x( ) G2 (g)1 ,

E x( + ) = g 3k G2 (g)E x( ) G1 (g)1 ,
(3.23)
where
G1 =
g2
0

0
,
g
G2 = 122 .
(3.24)
Since G2 = 1 there are no technical issues in defining EE . The columns of E must be proportional
to each other, and by looking at the common zeroes of the Lj and Qj one can argue that the
kernel should be spanned by theta functions of characteristic 2. However there is a simpler way
to obtain the kernels of the matrices using the functions .
By taking constant (i.e. independent of ) linear combinations of the rows of J in (1.10), one
can show that

L 1 1 ( ) + L 2 2 ( ) + 3 ( ) = 0,
(3.25)
where

L 1 12 x1 + 22 x2 + 32 x3 ,
L 2 (2 3 x1 + 1 3 x2 + 1 2 x3 ).
(3.26)
One can similarly derive relations between 1 and 2 + k 3 . Taking the latter relationships for
k = 1, 2 one can multiply by each by factors that are linear in the xj so as to obtain a relationship
260
of the form:

2 1 ( ) = 0,
Q 1 2 ( ) + 3 ( ) Q
(3.27)
j are quadratic in the x . Thus we get a 2 2 matrix with a null vector of the form:
where the Q

s1 ( )
2 ( ) + 3 ( )
(3.28)
.
s2 ( )
1 ( )
This matrix is gauge equivalent to J22 in (1.17), except one must replace j in (1.7) according
to:

1 2 33 13 ,
(3.29)
2 1 23 33 ,
3 3 13 23 .
Using (2.6) at = , along with (3.10) and (3.11), one can see that the replacement (3.29) is
equivalent to
2 + 13 .
(3.30)
To understand the underlying vector bundle, one should note that this derivation of the matrix
factorization implicitly relies on the fact that s1 ( ) and s2 ( ) have a common zero at =
and so (3.28) vanishes identically at one point. To cover this vanishing point we could find
another section and the transition function, but it is simpler to note that there is a global, nowherevanishing section of this bundle given by:

e3i
s1 ( )
2 ( ) + 3 ( )
(3.31)

.

s2 ( )
1 ( )
( ) 1 (1 + )
2
The sj are thus theta functions of characteristic 2 and satisfy:

sj ( + 1) = sj ( ),
sj ( + ) = z2 sj ( ),
(3.32)
and so the bundle EE is simply E(1, 2).

Conversely, given the sj , the ratio s2 /s1 is an elliptic function with two poles and two zeroes
on the torus, and so there is a unique way to write it in the form:
s2 a 1 x 1 + a 2 x 2 + a 3 x 3
=
,
s1 b1 x1 + b2 x2 + b3 x3
(3.33)
and this defines L 1 and L 2 . Similarly, one can write s2 /s1 as a ratio of quadratics in the xj .
j , has six parameters, one of which is a scale and the other five can be used to
Each quadratic, Q
j are
adjust the locations of the six zeroes on the torus.17 The locations of two zeroes in each Q
2 , but are otherwise
fixed by the sj , while the other zeroes of Q 1 must coincide with those of Q
free. Thus there is a three-parameter family of ways of writing s2 /s1 as ratio of quadratics. On
the other hand, one can trivially generate such a quadratic ratio by multiplying the numerator
and denominator of (3.33) by the same arbitrary linear function of the xj . This choice of linear
function has three parameters, and so the quadratic ratio is unique up to gauge transformations.
Thus one can reconstruct the 2 2-matrix factorization uniquely from (3.31).
17 The sum of the zeroes of Q must be zero modulo the lattice + 1, + .
j
261
By keeping track of the locations of the zeroes in the foregoing argument, it is easy to check
that one can write the columns of E as:

2
s1 ( )
s1 ( )
L 1
Q
(3.34)
(
)
(
)
=
,
,
1
2
1
L 2
s2 ( )
s2 ( )
Q
for some global holomorphic sections, 1 ( ) and 2 ( ), of E(1, 1) and E(1, 4) respectively. From
this one can write the columns of J as:

2
1
2 ( )
2 ( )
Q
Q
(3.35)
= s2 ( )
= s1 ( )
,
,
1 ( )
1 ( )
L 2
L 1
and hence the columns of EJ correspond to the bundle E(1, 1) (or E(1, 4)).
3.4. Summary
In this section, we considered both the 2 2 and 3 3 factorizations and determined the data
of the bundles EJ and EE , associated with the matrix factors, J and E. We now like to match
these data to the RR charges of the B-branes. Specifically, as mentioned before, the charges of
the branes will be associated with EJ , while the anti-branes will be associated with EE .
To proceed, recall that for extracting the bundle data from the matrix factorization, we had
to impose W 0, and therefore we implicitly characterized the branes in the large radius limit.
Thus we should refer to the RR charges listed in (1.14), which apply to the large radius limit.
Indeed we find a perfect match between the data of EJ , and the charges of one of the members
of {Sa } or {La }, respectively, provided we tensor uniformly with the line bundle L3 (recall
that brane charges are defined only up to such shifts, reflecting the monodromy around the large
radius limit). There is also an analogous match between the data of EE , and the charge of one of
the anti-branes in {Sa } or {L a }, respectively. We have summarized these findings in Table 1.
Note that if we view the charge lattice in Fig. 1 as the SU(3) root lattice, the specific charges of
the vector bundles at large radius lie within what one may call the positive cone, or fundamental
region, spanned by the simple roots S1 and S2 ; this is indeed natural from the point of view of
quiver representations (see the appendix of Ref. [41]).
What then about the other charges in the lists (1.14), some of which are related to more
general objects than vector bundles (like D0-branes described by point-like sheaves)? The point
is, of course, that in order to obtain the TFT on the elliptic curve from the LandauGinzburg
model, one must implement a Z3 orbifold projection, and it is only then that the full (quantum)
Z3 orbits of the branes will emerge. This is what we will discuss in Section 5, however before
doing so, we will first complete the discussion of P33 by analyzing its open-string moduli space.
Table 1
Here we summarize some properties of the short- and long-diagonal branes, associated to the 2 2 and the 3 3-matrix
factorizations. In particular the relationship between the bundle data of the matrix factors EJ and EE , and the large radius
RR charges is shown
2 2 factorization
Bundle
Matrix bundle E
E L3 (r, c1 )LR
Brane & anti-brane in
the positive cone
EJ
E(1, 1)
(1, 2)LR
S1
3 3 factorization
EE
E(1, 2)
(1, 1)LR
S3
EJ
E(2, 3)
(2, 3)LR
L1
EE
E(1, 3)
(1, 0)LR
L 3
262
4. Compactifying the moduli space of the 3 3 factorization

We now return to the 3 3-matrix factorization, P33 , in order to examine in greater detail
its open-string moduli space. This will eventually lead us to an exceptional 4 4 matrix factorization, which appears only at a certain point in that moduli space, and which compactifies the
moduli space of the 3 3 factorization P33 . It is this 4 4 matrix factorization which naturally
appears in the boundary fermion construction.
First of all one observes that the 3 3 matrices, J = J33 and E = E33 , given in (1.10) are
quasi-(double-)periodic for + 1 and + due to the quasi-periodicity (3.9) of the
parameters in terms of the uniformizing open-string modulus of Eq. (1.12). In other words,
such a shift in results in a multiplication of both J and E by an overall factor. However, the
factors of J and E are inverse to each other, and hence they are easily compensated by a gauge
transformation (1.5). Therefore, modulo gauge transformations the 3 3-matrix factorization is
indeed periodic for + 1 and + .
However, it is conceivable that there are further identifications in the open-string moduli space
of due to additional gauge equivalences: According to Eq. (3.10) the uniformizing functions
( ) of have also nice periodicity properties for one-third of a period. In particular, for the
3 3 matrices J and E, we readily find:
2

1 x1 2 x3
3 x2

1
2
J + 3 = C1
(4.1)
3 x3
1 x2 2 x1 ,
2 x2
3 x1
2 1 x3
and
J +

= C2
3 x 1
2 x3
1 x2
1 x3
3 x2
2 x1
2 x2
1 x1
3 x3

(4.2)
with overall factors C1 and C2 . Hence, apart from these factors, for + 13 the entries of
J are multiplied by cube roots of unity, whereas for + 3 the coefficients of x in J are
permuted. However, both transformations can be compensated by suitable gauge transformations,
i.e. J ( ) = UL,1 J ( + 13 )UR,1 and J ( ) = UL,2 J ( + 3 )UR,2 with18

UL,1 = Diag 1, 2 , ,
UR,1 = C11 Diag , 2 , 1 ,
(4.3)
and

UL,2 =
0
0
1
1
0
0

0
1 ,
0

UR,2 =
1
C2
0
0
1
1
0
0

0
1 .
0
(4.4)
We conclude that, modulo gauge transformations, the 3 3 matrix factorization is doubleperiodic in the open-string modulus with + 13 and + 3 , and thus obtain for the
3 3-matrix factorization the following toroidal open-string moduli space:

M33 = C | + 13 , + 3 .
(4.5)
One might suspect that there are other identifications in the open-string moduli space (4.5) of
the 3 3-matrix factorization and that the space M33 is yet again just the covering space of
18 One also has E( ) = U 1 E( + 1 )U 1 and E( ) = U 1 E( + )U 1 .
R,1
R,2
3 L,1
3 L,2
263
the true moduli space. However, the construction of the vector bundles encoded in the columns
of the matrix E and in the columns of the matrix J shows that the corresponding vector bundle
transition matrices (2.8) and (3.22) contain both the factor e2i3 , which independently confirms
the stated periodicity of J and E. Thus M33 is indeed the open-string moduli space for the
3 3-matrix factorization.
The reduction of the periodicity by one third will be an important ingredient in the construction of the 3 3-matrix factorization via tachyon condensation from 2 2 matrix building blocks
as discussed in Section 5.
Our next task is to examine the moduli-space (4.5) of the factorization P33 in greater detail.
For 0 the uniformizing function, 1 , approaches zero, cf. Eq. (3.11), for which the 3 3
matrix factorization becomes singular.19 Note that, in contrast to the singularity in the matrix
factorization P22 , this singularity is not a mere gauge artifact, because it cannot be removed
by a gauge transformation (1.5). This observation, however, is at first puzzling since there is no
obvious physical reason for a singularity in the toroidal open-string moduli space.
However, we should keep in mind that the topological B-type D-branes really are objects in
a category with certain equivalence relations, and a given matrix-factorization is just a particular
realization of a topological B-type D-brane. In other words, singularities of matrix factorizations
can also be an artifact of using the wrong representative for a D-brane in a given patch in the openstring moduli space. As has been discussed in Ref. [21], the apparent singularity encountered in
the 3 3-matrix factorization is indeed of this type, and this is what we want to make manifest
in the following.
Following Ref. [21], we replace as a first step the 3 3 matrix factorization by a 4 4-matrix
factorization that is obtained by adding a trivial braneanti-brane pair to (1.10), i.e.:
J44 =
W
1

J ()
E44 =
1
E()
(4.6)
The next task is to perform an appropriate gauge transformation, (1.5), so that the singularity
disappears for 0. This is achieved by first rewriting the open-string modulus, , by =
.20 Using Eq. (2.6) with x , replaced by and respectively, we may re-express
by21
1 = 22 1 3 1 3 22 ,
2 = 12 1 2 2 3 32 ,
3 = 32 2 3 1 2 12 ,
(4.7)
where and are uniformized by = () and = (). Now the limit 1 0 becomes
, which in terms of the uniformizing parameters translates into . In terms of these
19 At first sight there seem to be nine singularities, namely = 0 for the three choices of , which, according to (3.10)

and (3.11), have each three zeroes. However, Eqs. (4.1) and (4.2) show that all nine choices are gauge-equivalent.
20 This step is a little ad hoc but it is motivated by viewing the brane L( ) as a composite of the two branes L()
and
L().
21 Strictly speaking the parameters, , must be rescaled by a homogeneous factor, i.e. 2 ( ) ( + ) .

3
264
auxiliary variables we perform the gauge transformation
1
1 1 1 1 x1 1 2 2 x2 1 3 3 x3
1 x1
0
1 1
1
0
0 ,
UL =
UR = 2
0
1 2 x2
0
1
0
0
0
0
1
3 x
1 3 3
0
0
,
0
0 0
1 0
which generates the following 4 4-matrix factorization:
1 1
2 2
1 2
2 2
0
1 x1 2 3 x2 x3
2 x2 1 3 x1 x3
1 1 x x 1 x 2
1 x1
2 x3
2 3
1 1
J44 = 2223
2
2
x1 x3 x2
3 x3
1 x2
1 3
2
3 3
1 2 x1 x2
1
1 x
1
E 44 = 21
x2
2
3
3 x3
33 x32
2 x2
3 x1
1
2
1 x1
2 x2
3 3
1
1 2
2 3 x2 x3
3 x3 + 2 1 2 x1 x2
3 3
1 2
2 1 3 x1 x3
2 x3 + 3 2 1 x1 x2
2 2
1 1
1 2
1 2
3 x2 + 2 3 1 x1 x3
2 x1 + 3 3 2 x2 x3
3 2
3 x3
(4.8)
3 3
1 2 x1 x2
3 x2
2 x1
1 x3
3
3 x3
2 2
1 2
2 x2 + 3 1 3 x1 x3
,
1 1
1 2
3 x1 + 2 2 3 x2 x3
1
2 3 x1 x2
(4.9)
which we will denote by P44 . We want to emphasize again that P44 describes the same physical D-brane as the original factorization P33 . As we will see in a moment, this factorization is
indeed most canonical from the LandauGinzburg point of view: While P33 is odd-dimensional
so that it cannot be given a standard representation in terms of boundary fermions, it is the equivalent factorization P44 that can be given a concise expression in terms of LandauGinzburg
fields.
In addition, P44 has, by construction, the virtue that it is also well-defined for = 0, and
thus we may proceed and take the limit 1 0, . Upon doing so, we prefer to perform
yet another gauge transformation:
12
2

2 (13 33 ) 1 3 2 + a x1 2ax2
2 3 3a x3
0
1
0
0
,
UL =
0
0
1
0
0
0
0
1
1
UR =
(4.10)
ULT ,
2 (13 33 )
in order to obtain the following form of P44 at = 0:
0
x12 ax2 x3 x22 ax1 x3
2
x + ax2 x3
0
x3
1
0
= 2
J44
x3
0
x2 + ax1 x3
x 2 + ax1 x2
3
0
x1
x1
0
0
E 44
=
2
x2 x3 ax1 x2
x3
x22 + ax1 x3
x32 ax1 x2
x2
,
x1
x1
x2
x2
x32 + ax1 x2
0
x12 ax2 x3
x3
x22 ax1 x3
.
x12 + ax2 x3
0
(4.11)
265
0 . In terms of the boundary BRST operator, Q, we can rewrite

We will denote this limit by P44
this factorization in a very simple form
0
P44
:
Q=
3

x + 13 x W (x, a) ,
(4.12)
=1
where W (x, a) is the LandauGinzburg potential in (1.1) and , , = 1, 2, 3, are boundary

fermions obeying {i , j } = { i , j } = 0, {i , j } = ij .22 Indeed, for an appropriately chosen
basis of the Clifford algebra, for which the chirality gamma matrix, 5 , takes the form 5 =
Diag 1, . . . , 1, 1, . . . , 1, the boundary BRST operator (4.12) is precisely given by the matrix
factorization (1.17)four, i.e.,

0
0
J44
.
Q=
(4.13)
0
E 0
44
For non-vanishing open-string modulus, , (4.12) gets deformed in leading order to Q Q +

Q, where

Q 1 2 3 + a 3 1 x1 x2 x3 1 2 3 + .
(4.14)
The ellipsis indicates terms that are linear and quadratic in x and are multiplied by products
of boundary fermions of the type i j k and i j k . Note that the deformation, Q, contains
a constant, x-independent term, which generates the constant entry, 1 , of the matrix E 44 in
Eq. (4.9).
Let us summarize and analyze the foregoing results. Starting from the matrix factorization,
P33 , we encountered a singularity. This singularity was removed by taking an equivalent description in terms of the 4 4-matrix factorization P44 in (4.9), which, however, is reducible
due to an extra constant entry in the matrix, E. This constant entry disappears as 0, which
0
implies that the corresponding factorization P44
becomes irreducible. This irreducible factorization is rigid in the sense that it exists only at one point in the open-string moduli space, and
specifically is not part of a parametric family of irreducible matrix factorizations.23 For a simple
depiction of the situation, see Fig. 5.
In order to give this mathematical description a physical interpretation, we observe that the
long-diagonal brane, L 3 , is associated to the line bundle, E(1, 0), and located in the positive
cone of the charge lattice in Fig. 1. Furthermore, according to (1.14), the brane L 3 has the (large
radius) RR charges of a pure D2-brane. This, however, does not necessarily imply that the brane
L 3 corresponds to a pure D2-brane, rather more generally it describes a D2-brane with a D0
D0-brane pair resolved on its world-volume. We will now argue that the rigid exceptional matrix
factorization, P 044 , describes a pure D2-brane, which also on physical grounds does not depend
on an open-string modulus, whereas the matrix factorization, P33 , captures the situation of a
D2-brane with an additional resolved D0D0-brane pair.24 This D-brane configuration is indeed
expected to depend on a (relative) open-string modulus.
22 That this represents a valid boundary BRST operator that obeys Q2 = W (x, a)1, follows from the fact that,
due to homogeneity of the LandauGinzburg potential, W (x, a), we can write W (x, a)1 = 13 x x W (x, a)1
1
i,j xi i j W (x, a) j .
3
23 There does, however, exist another family of irreducible 4 4-matrix factorizations (denoted by P
44 , without the
tilde) which describes D-branes with different charges (cf. Section 7).
24 Note again that such D-brane configurations cannot be distinguished at the level of K-theory.
266
Fig. 5. This shows the open-string moduli space of the factorization P44 . It is compactified at = 0 by an exceptional
indecomposable object, which is rigid in the sense that any deformation of it produces a decomposable object. It is
associated with the most canonical factorization, (4.12), of the LandauGinzburg superpotential. In physical terms, it can
be interpreted as a single rigid anti-D2 brane, and the deformation away from it corresponds to adding and pulling apart
an extra D0D0-brane pair.
This line of arguments can be substantiated by using the relationship between line bundles and
divisors. Each line bundle in the class, E(1, 0), can be specified by a divisor O(1 2 ), where 1
and 2 denote points on the torus, . The distance between them corresponds to the open-string
modulus that appears in the transition function, A(z), of the line bundle E(1, 0). Physically,
the point 1 describes the position of a D0-brane, whereas the point 2 refers to the position of a
D0-brane. If, on the other hand, 1 and 2 coincide, or in other words, if the D0-brane is on top
of the D0-brane, we arrive at the divisor of a trivial line bundle, which precisely corresponds to
the exceptional 4 4-matrix factorization of a pure, rigid D2-brane.
The appearance of distinguished matrix factorizations at special points in the moduli space has
been observed before in Ref. [19]. According to the classification of holomorphic vector bundles
on the torus [17], each vector bundle of rank r and degree zero is classified by a degree zero
line bundle on , namely by the determinant line bundle of the vector bundle in question. Moreover, each line bundle of degree zero has vanishing first Chern class and according to Eq. (2.4) is
given by the transition function A(z) = y 1 on C [18,46]. Note that for y = 1, or for = 0, the
transition function becomes the identity and hence the determinant bundle becomes the trivial
line bundle.25 In Ref. [19] it is shown that such exceptional vector bundles in E(r, 0) with trivial determinant bundle, correspond to exceptional MCM modules Mr ,26 and furthermore give
rise to exceptional matrix-factorizations, which are isolated in the sense that there exist no (open
string) deformations that would stay within the class of irreducible factorizations.
To conclude this section, we consider how the bundle data encoded in J33 and E33 evolves
for which (up to an unimover the moduli space, M33 . Consider first the anti-D-brane, L,
portant sign) the rles of J33 and E33 are exchanged, i.e. the vector bundle data is encoded
in the matrix E33 given in (1.10). At a generic point in the open-string moduli space M33 ,
25 Those exceptional bundles of rank r and degree zero are also distinguished by the fact that they have a global
non-trivial section.
26 The MCM modules M are self-dual R-modules, i.e. M Hom(M , R) [19].
r
r
r
267
the vector bundle spanned by the columns of E33 is associated to a line bundle in E(1, 0) (cf.
Eq. (3.13)). Therefore one might naively expect that the exceptional 4 4-matrix factorization at
= 0, corresponds to the exceptional rank one MCM module M1 (it is actually the anti-version,
This, however, is not
P 044 , that we talk about here, as we started out with the anti-brane, L).
true, because we should keep in mind that this 4 4 matrix factorization has been constructed
by adding a trivial braneanti-brane pair. In particular the new column of E44 in (4.6) has increased the number of linearly independent columns of E33 by one, and thus the 4 4-matrix
factorization is correctly identified with the rank two exceptional module, M2 .27 Comparison
with Ref. [19] does indeed confirm that the exceptional 4 4-matrix factorization is associated
to the exceptional MCM module M2 .
On the other hand, if we now consider the D-brane, L, the bundle data are encoded in the
matrix J of (1.10). As explained in Section 3.2, the D-brane L gives rise to the rank two vector
bundle E(2, 3). As before, in order to analyze the limit 0, we must not neglect the added
trivial braneanti-brane pair. However, the matrix J is only enhanced by the block-diagonal
entry, W , (cf. Eq. (4.6)), which does not increase the number of linear independent columns of
J because W 0 on the torus, . Therefore the exceptional 4 4-matrix factorization assigned
to the brane, L, at = 0 must still be associated to a vector bundle in E(2, 3). Once again
comparing with Ref. [19] reveals that the 4 4-matrix factorization (4.11) is really mapped to
the exceptional 4 4-matrix factorization in E(2, 3), which, furthermore, is associated to the first
syzygy module, 1 (M2 ), of M2 . Note that this is precisely in accord with the physical picture
because passing over to the first syzygy module, 1 (M2 ), of M2 corresponds to switching from
to the D-brane, L.
the anti-D-brane, L,
5. Tachyon condensation
The aim of this section is to generate new matrix factorizations via tachyon condensation.
In particular, we will show that any matrix factorization on the elliptic curve may, in principle,
be obtained by iteratively condensing building blocks of 2 2-matrix factorizations. So as to
generate new matrices in a controlled fashion, we introduce the notion of equivariant matrix
factorizations. This enables us to explain how different choices of tachyons for a given pair of
matrix factorizations will lead to different condensates. Conversely, if we seek a certain condensate, equivariance provides a very useful set of constraints upon the form of the requisite
tachyon.
In the present and the next two sections we will use these ideas to generate explicitly all rank
two matrix factorizations.
5.1. Equivariant matrix factorizations
In order to systematically generate, via tachyon condensation, a particular matrix factorization for a given set of RR charges, it is necessary to have control over the RR charges of the
constituents of the condensate. For instance, we need to distinguish among the three D-branes
comprised in the 2 2-matrix factorization, S, and in the 3 3-matrix factorization, L, in order
to realize the different composites illustrated in Fig. 4. This is achieved by refining the descrip27 The self-duality of M is reflected in the fact that the matrices J
2
44 and E44 in (4.11) are antisymmetric.
268
tion of B-type D-branes in terms of equivariant matrix factorizations along the lines of [26] (see
also [32]).
Since our analysis takes place at the Gepner point of the Khler moduli space, the appropriate
LandauGinzburg model is specified in terms of the superpotential (1.1) together with the natural
Z3 orbifold action (k), k Z3
:
(k)x k x
with = e
2i
3
(5.1)
Obviously, this orbifold action must also be taken into account in the characterization of B-type
D-branes, for which the equivariant formulation of matrix factorizations becomes the appropriate
framework. In practice this means that a D-brane, P , represented by a n n-matrix factorization
(1.6) is supplemented by two Z3 representations R0 and R1 of dimension n:
(P1 , R1P )
EP ,
P JP
(5.2)
(P0 , R0P )
such that, in addition to the factorization condition (1.4), one also requires the equivariance condition [26]:

JP (x) = R1P k 1 JP (k)x R0P (k),

EP (x) = R0P k 1 EP (k)x R1P (k).
(5.3)
Note that these relations also require an adjustment of the notion of gauge transformations.
Namely, it is easy to infer that a gauge transformation (1.5) induces also a conjugation transformation acting on the representations R0P and R1P by
R0P (k) UR1 R0P (k)UR ,
R1P (k) UL R1P (k)UL1 .
(5.4)
We demonstrate this idea by going through the matrix factorizations that we have encountered
so far. The easiest example is the trivial D-brane configuration, V , given by the 1 1-matrix
factorization, JV = 1, EV = W , which corresponds to the vacuum. For this configuration the
equivariance conditions (5.3) are fulfilled as long as R0V (k) = R1V (k) = ak for any a = 1, 2, 3.
The triviality of the D-brane configuration manifests itself in the fact that the two representations
R0V (k) and R1V (k) must be the same.
For the 2 2-matrix factorization, S, the equivariance yields three possible representations
R0S and R1S , which read

R1S (k) = Diag ak , (a+1)k .
R0S (k) = Diag (1a)k , (1a)k ,
(5.5)
Taking into account the orbifold Z3 action therefore results in three different 2 2-matrix factorizations, which are distinguished by a = 1, 2, 3. Thus in the orbifolded LandauGinzburg model
the 2 2-matrix factorization, Sa , gains an additional label, a, which specifies the choice of
representations R0S and R1S .
Analogously, we find for the 3 3-matrix factorization three possible representations
R0L (k) = (1a)k 133 ,
R1L (k) = (a+1)k 133 ,
which give rise to the three distinct branes, La , labeled by a = 1, 2, 3.
(5.6)
269
Note that the equivariant labels allow us to unambiguously distinguish among all the shortand long-diagonal branes illustrated in Fig. 1. This use of equivariance means that we can go
beyond the vector bundles discussed in Ref. [19] and enables us to deal with objects that are
more general than vector bundles in the large radius limit. In particular, recall that one of the
short branes, S2 , is the D0-brane and this is not associated to a vector bundle but rather to a
point-like sheaf.
Tachyon condensation provides the opportunity to make extensive direct tests of this equivariant formulation of the Z3 orbits of branes. We will now examine this in detail but first we need
the appropriate equivariant modification of the open-string spectrum [26]. Just as in (1.15), a
boundary changing operator (P ,Q) of an open string stretching between the brane P and Q can
be pictured by the diagram:
(P1 , R1P )
JP
JQ
EP
(P0 , R0P )
(Q1 , R1 )
EQ
(5.7)
(Q0 , R0 )
In the equivariant context, the physical state conditions, (1.16), of the maps, 0 and 1 , of (the
fermionic) operator, (P ,Q) , are now supplemented by the equivariance condition28

Q
0 (x) = R1 k 1 0 (k)x R0P (k),

Q
1 (x) = R0 k 1 1 (k)x R1P (k).
(5.8)
This condition amounts to a selection rule upon the original (non-equivariant) tachyon spectrum.
The selection rule tells us which, if any, tachyon appears between different members of the Z3
families. It also tells us the Z3 label of the condensed state. We now illustrate this with some
examples.
5.2. A reprise of tachyon condensations of P22
In calculating tachyon condensations, our modus operandi consists of two basic steps: First
we analyze the boundary changing operators of the constituents in terms of equivariant matrix
factorizations. Then we build up the composites via the cone construction, (1.18), and simplify
the result (if necessary) using gauge transformations, (1.5) and (5.4). We begin by returning to
the examples of tachyon condensation already introduced in Section 1.2.
The simplest example is brane/anti-brane annihilation between Sa () and Sb (). Recall that
the fermionic identity operator, (1.22), appears in the open-string spectrum of the boundary
changing sector provided that = . Moreover, equivariance imposes an additional constraint on
the existence of this operator, i.e. due to (5.8) the representation R0S and R1S acting on 0 = 122
and the representations R1S and R0S acting on 1 = 122 must pairwise coincide. In other words
the fermionic identity operator, (1.22), can only form a condensate of Sa () with Sa ().29
28 A similar equivariance condition applies for the bosonic operators.
29 In equivariant matrix factorizations going from the brane P to the anti-brane, P , does not only exchange the matrices
(JP , EP ) with (EP , JP ) but is also accompanied by a flip of the representations (R0P , R1P ) to (R1P , R0P ).
270
Applying the cone construction to the constituents Sa () and Sa () using the boundary changing operator, (1.22), one obtains the following composite:

J22 ()
E22 ()
122
122
JV =
(5.9)
,
EV =
.
0
E22 ()
0
J22 ()
Note that the constant entries in the matrices, JV and EV , mean that one may make elementary
simplifications by row and column elimination. Here we simply observe that because there
are two independent constant entries in JV and in EV , the method of row and column elimination shows that the composite, (5.9), is gauge equivalent to four trivial braneanti-brane pairs.
Thus, within the D-brane category of B-type D-branes, the tachyon condensation of Sa and Sa
describes the annihilation to the vacuum:
Sa ( ) Sa ( ) V ,
(5.10)
where is given by = ( ).
The other example considered in Section 1.2 was the condensation of Sa () with Sb (). The
relevant boundary-changing operator, (Sa ,Sb ) (, ), is given in Eq. (3.23) and the condensation
process has already been described in Section 1.3. Here, we want to re-examine the discussion
so as to elucidate the selection rules arising from equivariance. First, the form of the tachyon,
(Sa ,Sb ) (, ), constraints the equivariance labels, a and b. That is to say, that evaluating condition (5.8) together with the representations (5.5) of Sa and Sb yields the relation b = a + 1.
Hence, the boundary changing operator (Sa ,Sb ) (, ) only appears in the open-string cohomology for the equivariant pair Sa () and Sa+1 (), and only for such a pair can this fermionic
boundary operator be used to construct the composite, Sc ( ), determined in Eq. (1.29).
Furthermore, we can also determine the equivariance label, c, of the resulting composite,
Sc ( ). The representations, (5.5), of the constituents, Sa () and Sa+1 (), yield, for the untransformed composite, (1.26) the representations

R0 (k) = Diag ak , ak , (1a)k , (1a)k ,

R1 (k) = Diag (2a)k , (1a)k , ak , (a+1)k .
(5.11)
As discussed in Section 1.3 by acting with a gauge transformation, UL , UR , the composite,
(1.26), can be cast into the 4 4-matrix factorization (1.27). However, this gauge transformation
also induces a corresponding conjugation action on the representations, (5.11), according to (5.8),
and hence, after dropping all the trivial braneanti-brane pairs, we readily read off the resulting
equivariant 2 2-matrix factorization, which turns out to be Sa+2 ( ).
Before we move on to the next example, we can gain some further insight into the openstring parameters, , given in (1.28), by rewriting them in terms of uniformizing parameter
along the lines of Eq. (1.12). Namely, with = ( ) and = () the parameter can
be uniformized by = ( ), where we used Eq. (2.6) after substituting x by . Thus
schematically the tachyon condensation process is summarized by
Sa ( ) (Sa ,Sa+1 ) Sa+1 () Sa+2 ( ).
(5.12)
The next task is to construct the long-diagonal D-branes, L, by tachyon condensation of two
short-diagonal D-branes S. From Fig. 4 we readily infer that one is required to condense the
brane Sa () with Sa+2 () in order to reach L as a condensate. Thus we seek a fermionic boundary changing operator, (Sa ,Sa+2 ) (, ), in the spectrum of open-strings stretching from Sa to
271
Sa+2 .30 By once again taking advantage of the equivariance condition, (5.8), we readily deduce
the degrees of the entries of the fermionic boundary changing operators:

l1 (x) l2 (x)
l5 (x) q(x)
0 =
(5.13)
,
1 =
.
l3 (x) l4 (x)
c
l6 (x)
Here c has degree zero, l1 , . . . , l6 are linear and q is quadratic in x . A detailed analysis of the
corresponding fermionic cohomology element along the lines of Eqs. (1.16) and (1.17) eventually
reveals:

G1 (, ) G2 (, )
G4 (, ) H (, )
0 =
(5.14)
,
1 =
,
G3 (, ) G1 (, )
C(, ) G4 (, )
with
C(, ) = 3 3 (3 2 2 3 ),
(5.15)
and
G1 (, ) = (3 1 + 1 3 )x2 +
G2 (, ) = (2 3 3 2 )x2 ,
G3 (, ) = (3 2 2 3 )x1 +
G4 (, ) =
3 (23 + 33 )
x3 ,
1 3

3 22 22 3
x3 ,
3
3
2 3 (23 + 33 )
3 2 (3 2 2 3 )
3 (2 1 + 1 2 )
x1 +
x2
x3 ,
1 3
2
1 2 3
(5.16)
and
H (, ) =
2 2 (3 2 2 3 ) 2 (2 3 12 12 2 3 ) 2
x1 +
x2
1 3 1 3
2 3 2 3
( 3 3 23 33 )
(23 12 3 12 3 23 )
3 2 2 3
x1 x2 + 3 2
x1 x3 +
x2 x3 .
1 2 3 1 2 3
1 2 3 1 2 3
2 2
(5.17)
Next we use this fermionic boundary-changing operator, (Sa ,Sa+2 ) (, ), to form the composite

E22 () 0 (, )
J22 () 1 (, )
,
EL (, ) =
.
JL (, ) =
0
J22 ()
0
E22 ()
(5.18)
Note that the matrix E L contains a constant entry, which allows us to perform row and column
eliminations so as to obtain a 33-matrix factorization. Moreover, from the degrees of the entries
of E L , it is apparent that the row and column operations precisely remove the linear entries in
E L . Therefore the matrix E L becomes, after removing the trivial braneanti-brane pair, precisely
+
30 The precise argument is actually a bit more involved: From the A-model mirror picture we know that S intersects
a
with Sa+2 once, and hence we expect (at least) either one fermionic operator or one bosonic operator in the spectrum.
The orientation of the intersection tells us whether the operator is bosonic or fermionic. Since the intersection of Sa and
Sa+1 has the same orientation as Sa with Sa+2 the fermionic operator, (Sa ,Sa+1 ) , implies the presence of a fermionic
operator (S ,S ) .
a
a+2
272
the matrix E33 with only quadratic entries. Conversely, after row and column elimination, JL
contains only linear entries and takes the form the matrix J33 .
The final task is to relate the result of this row and column elimination to the standard form
of the 3 3-matrix factorization stated in (1.10), and thereby determine the precise form of the
brane, L. This is achieved by an appropriate gauge transformation that allows us to determine the
open-string parameter of the composite in terms of and . The result of this, straightforward
but tedious, analysis yields (in terms of the parameters and of the uniformized functions
= ( ), = ()):

Sa ( ) (Sa ,S ) Sa+2 () La 13 ( ) .
(5.19)
a+2
Before we conclude this section, a few comments are in order: First, note that the factor 13 in
the uniformizing function correctly reproduces the periodicity of the 3 3-matrix factorization observed in Eq. (4.5). Furthermore, the enhancement of the 3 3-matrix factorization to the
4 4-matrix factorization observed in Section 3.4 also becomes manifest. Namely, the indecomposable 4 4-matrix factorization is obtained for , or . In this limit, the constant
entry (5.15) vanishes and the composite, (5.14), becomes gauge equivalent to the exceptional
0 in (4.11).
factorization P44
6. Constructing more general matrix factorizations
So far we have analyzed some examples of tachyon condensation among the known factorizations P22 and P33 , and shown how the latter can be obtained as a condensate of the 2 2
factorization. We have also seen how the 4 4 factorization P44 , which is more naturally associated with boundary fermions, can be produced by condensation and how it connects to the
moduli space of the 3 3 factorization.
We now wish to go beyond this and obtain new factorizations. There are several techniques
that we can use to do this, and in this section we will describe them and determine the tachyons
that we will need to create a number of new factorizations. In the subsequent sections, we will
present a list of the resulting factorizations and discuss some of their properties.
6.1. New matrix factorizations and tachyons from transpositions
Given a matrix factorization P = (J, E), one can obtain another matrix factorization by transposing the matrices J and E. We denote the resulting factorization by

P T = J T, ET .
(6.1)
According to the equivariance condition (5.3), transposition also acts upon the equivariant repreP , by:
sentations, R0,1

T
T
T
T
R1P (k) = R0P k 1 .
R0P (k) = R1P k 1 ,
(6.2)
Applying the transposition operation to the branes Sa and La we find (modulo simple gauge
transformations) the corresponding transposed matrix factorizations to be:
SaT ( ) Sa1 ( ),
LTa ( ) La ( ).
(6.3)
In general the transposed matrix factorization can give rise to new matrix factorizations, which
are not related to the original factorization by changing the open-string modulus or by passing
273
over to the anti-brane. We will encounter such an example in with the two 5 5 factorizations,
T , that describe two distinct classes of D-branes.
P55 and P55
In the language of MCM modules, the transposed matrix factorization is associated to the dual
MCM module. In particular, Eq. (6.3) illustrates that the brane, L, at = 0 is invariant under
transposition, which reflects the self-duality of the corresponding exceptional MCM module at
= 0 in the open-string moduli space as discussed in Section 4.
Suppose we have a known tachyon, (P ,Q) , satisfying the physical state condition (1.16).
Then the transposed physical state condition reads
T
+ JPT 1T ,
0 = 0T EQ
T
+ EPT 0T ,
0 = 1T JQ
(6.4)
which we readily identify with the physical state conditions for a fermionic boundary changing
operator, (QT ,P T ) (0 T , 1 T ), of an open string stretching between QT and P T .
To illustrate this procedure we choose as an example the fermionic boundary changing operator, (Sa (),La+1 ( )) of an open string stretching between Sa () and La+1 ( ). This immediately
gives rise to the boundary changing operator, (LT ( ),S T ()) , of an open string stretching bea+1
tween LTa+1 ( ) and SaT (), which we can finally convert according to (6.3) to the fermionic
boundary changing operator
(La1 ( ),Sa1 ()) (LT
T
a+1 ( ),Sa ())
(6.5)
where those two boundary changing operators are related by the same gauge transformation that
gives rise to the corresponding identification in Eq. (6.3).
6.2. Creating bound states at threshold
In obtaining matrix factorizations via condensation, there is a significant difference between
the situation where the D-brane charges, (r, c1 ), of the end product are co-prime or have common
factors. In particular, when they have a common factor, it turns out that there is a simple canonical
procedure for obtaining the matrix factorization as a bound state at threshold.
To illustrate the basic physical idea it is easiest to use the A-brane mirror picture where the
charges (r, c1 ) correspond to winding numbers (p, q). The tension m of a BPS-saturated D1brane is then simply proportional to its length on the covering space of , as shown in Fig. 1.
That is,
m = |q + p|,
(6.6)
where is the complex structure parameter of the mirror torus. It coincides with the Khler
parameter of the original B-model, which decouples in the topologically twisted theory so that
we have no control over it.31 However, things are simple on the torus; in particular, there are no
lines of marginal stability on the Khler moduli space, so all what matters is that the tension is
linear in the charges. This implies that if we only know the charges, (p, q), we cannot distinguish
a single string with winding numbers (np, nq) from n strings, each with winding number (p, q).
31 In the LandauGinzburg model, which corresponds a Z orbifold of the torus CFT, is fixed to be a third root of
3
unity, but in the orbifolded LandauGinzburg model, the Khler modulus becomes a free parameter.
274
Such single-string configurations are commonly referred to as bound states at threshold [47].32
In general, it is often unclear whether a combination of states with equal charges will actually
form a true bound state at threshold, or whether the state is simply a multi-particle superposition
of the constituents.
In matrix factorizations, where we have additional information in form of the open-string
moduli, we can, in fact, easily see the difference between a bound state at threshold and a mere
direct sum of its constituents. In the following, we consider only the condensation of a pair of
identical branes but more general configurations can be treated in a similar way.
Consider first a pair of identical branes, P , but with distinct position moduli, and . It
is clear that as long as = , the branes do not intersect and thus there isnt a tachyon that
could possibly lead to bound state formation. The combined system is characterized by a blockdiagonal matrix factor

JP ()
0
JP P (, ) =
(6.7)
,
0
JP ()
and similarly for EP P . The vector bundles corresponding to JP P and EP P are thus trivial direct
sums. What is the moduli space of such a configuration?33 The naive answer is that it is simply
given by the direct product of the individual moduli spaces. However, there exists a gauge transformation that switches the rles of the block matrices. That is, JP P (, ) is gauge equivalent to
J (, )P P , and since gauge equivalences are equivalences in the full category, it follows that the
open-string moduli space is given by the symmetrized product of the individual moduli spaces,
i.e.,
MP P = Sym2 (),
(6.8)
which is entirely to be expected because the D0-branes are indistinguishable.

The identifying gauge transformations have a fixed point when the branes coincide and this
leads to an orbifold singularity in the moduli space when = . This is reflected in the physics in
that, when the branes move on top of one another, the cohomology jumps and a tachyon appears
(in a similar manner to the results discussed in Section 1).34 Since the physical state condition
at = is identical to the equation that determines the boundary preserving endomorphism,
P , the new operator (P ,P ) (P ,P ) has the exactly the same form as P , however, due to
the implicit ChanPaton labels it is really a boundary changing operator because it acts between
different D-branes.
(J )
(E)
Because the tachyon (P ,P ) = ((P
,P ) , (P ,P ) ) has charge one, it is a marginal operator
of the theory that couples to a dimensionless modulus, :

(J )
JP () P P ()
JP P (, ) =
(6.9)
.
0
JP ()
Switching on condenses the two-brane system into one indecomposable object, which is a
genuine bound state at threshold. Indeed, the off-diagonal terms in (6.9) generically transform
32 Of course, when (p, q) are co-prime, all multiple string configurations with the same net winding number cannot
satisfy the BPS bound because the triangle inequality implies that their total length will not be minimal.
33 We thank Robert Helling for asking this question, which then lead to the following discussion.
34 Again, both a bosonic and a fermionic tachyon appear at = , so that there is no net contribution to the intersection
index.
275
Fig. 6. The upper part of this diagram shows the moduli space, MP P Sym2 (), of two identical branes but with
independent moduli and . A new branch emerges at the singularity at = . Switching on moves the factorization
onto this new branch and the reducible two-brane system condenses into an indecomposable bound state at threshold.
non-trivially between coordinate patches on the torus and thereby make the vector bundle defined by (6.9) into a non-split extension of the original vector bundles. This clearly shows the
usefulness of the extra moduli information carried by the factorization, as the existence of such
an object cannot be inferred from the K-theory charges alone. Mathematically, the interpretation
of this extra degree of freedom is that it resolves the singularity of MP P . We have depicted the
situation in Fig. 6.
The tachyon, (P ,P ) , responsible for creating bound states at threshold, can be explicitly
obtained by following an observation in Ref. [16]. Let P = (JP (), EP ()) be a matrix factorization, where are the usual functions of the brane moduli: = ( ). One has
EP () JP () = W 1.
(6.10)
Taking the derivative of the above equation with respect to , one obtains
EP () JP () + EP () JP () = 0,
(6.11)
which becomes the physical state condition, (1.16), for the following tachyon between identical
branes:
(J )
(P ,P ) = JP (),
(E)
(P ,P ) = EP ().
(6.12)
That this operator is not exact follows from the fact that its boundary preserving version describes
the non-trivial marginal operator associated with the open-string modulus, .
For our present purposes, it is simpler to work with derivatives with respect to the , rather
than with respect to the flat coordinate, . However, the are not independent since they must
lie on the cubic curve. This is easily taken into account by taking a particular linear combinations
of derivatives. We find that the
D()

,

(6.13)
276
where (2 ), does the job in lieu of the -derivative in (6.12).35 To summarize, we can
write the following concise cohomology representative for the tachyonic modulus:
(P ,P ) () = D()QP ().
(6.14)
6.3. Completing the links in the quiver diagram: Tachyons between S and L
The spectrum of open strings connecting long branes to short branes/anti-branes, falls into
four distinct classes (ignoring equivariance labels):
(S,L) ,
(L,S) ,
(L,S)
and
(6.15)
(S,L)
.
The tachyons for the last two classes may be obtained from the first two using transposition as
outlined in Section 6.1. Thus, we shall focus here on the first two classes alone. The prediction
based on the quiver diagram shown in Fig. 3 is that there should be one tachyon of type (Sa ,La+1 ) ,
one tachyon of type (Sa ,La ) and two tachyons of type (La ,Sa+1 ) . The degrees of the terms in the
relevant tachyon operators are obtained by using the equivariance condition (5.8). We find that
the number of solutions to (1.16) is indeed consistent with the quiver diagram.
We now present the explicit expressions that we find for the tachyons. We denote the moduli
associated with the L and S branes by ( ) and (), respectively, and let
( ). We then find the following rectangular matrices for the first kind of tachyons:

1 3
3 2
0 = 2 2 1 1 ,
3 1 2 3
x1 2 1 + x3 1 1 x2 1 2
(Sa ,La+1 ) :
(6.16)
3 3
1 1
3 2
2 2
1 = 3 1 x112 + x332 x233 ,
3 1
2 2
1 2
x2 2 1 x1 3 3 x3 2 3
3 2
3 2
2 1
1 2
where a = 1, 2, 3. This tachyon gives rise to the composite as given below:

Sa () (Sa ,La+1 ) La+1 ( ) Sa+1 (3 + ).
(6.17)
In arriving at (6.17), we used the constant entries in the tachyon to carry out row and column
reductions and discarded three trivial P11 and P11 vacuum pieces.
The tachyon (Sa ,La ) can be shown to be gauge equivalent to x1 times the tachyon (Sa ,La+1 )
in (6.16):
(Sa ,La ) = x1 (Sa ,La+1 ) .
(6.18)
It is easy to see that any function linear in the xi multiplying (Sa ,La+1 ) will satisfy the physical
state condition. However, it turns out that there is only one non-trivial cohomology element that
can be obtained in this way. This is the tachyon that leads to a new 5 5 factorization that will
be discussed later.
35 The
satisfy the following two identities, enabling us to show that D / :

2 = 0 and
1 2 3 + 2 3 1 + 3 1 2 = 0.
277
The last kind of tachyons comes with a multiplicity of two, and we distinguish them by adding
a superscript:
(1)
(L
:
a ,Sa+1 )
x x x x

2 3 1 3 1
2 2 1 + 3 2 x3 1 + x2 3 2 + x1 1 2
1
3
2
1
3
3
2
3
1
2
1
2
2
1
2
3
1
3
,
0 =
0
2
1
x1 3 1 x2 1 x3 2 x1 3 x2 2 x1 2 2 x3 1
1 2 1 + 3 3 + 2 1 3 3 1 3 3 1 3 1 1
x2 (32 3 1 22 1 2 )
1 =
x2 3
x2 3
x1 1
x1 2
1 2 3 1 3
3 3 2 3
(6.19)
3 3 2 1
and
(2)
(La ,Sa+1 ) :
x x x x x

3 2 2 2 3 1 3 3
1 2 2 2 1 x2 1 2 x3 3
2 2 3
1 2
,
0 = 1 2 1 1 2 2 1 3 1 1 3 1 3
2
0
3
x2 3 x1 1 x3 2 x1 2 3 x1 2 2 x2 2 x3 3
3 3 2 3 3 1 2 1 3 1 3 1 2 3 1 1
.
1 = x x x x
x2 t (2 32 3 1 12 2 )
1 3
1 2
2 1
2 2
1 3 2 3
1 1 2 3
(6.20)
1 2 1 2 3
We can form the same composite with either one of the tachyons, using as before the constant
entries in 0 to carry out row and column reductions and thereby discarding one trivial, P11 ,
vacuum piece. The result is the 4 4-matrix factorization, P44 , as shown in Fig. 4 and discussed
in Section 7.
7. All rank two factorizations obtained via tachyon condensation
We will now use the tachyons found in the previous to write down explicitly all rank-two
matrix factorizations. Note that these do not describe all rank-two vector bundles, but only those
for which the charge vector, (r, c1 ), lies in the positive cone. By equivariance, a general rank-two
vector bundle for which (r, c1 ) is outside of the positive cone, can be mapped to a bundle within
the positive cone but with r > 2. These correspond to higher rank matrix factorizations.
From Fig. 1 and Fig. 4 one can see that there are six different factorizations to be considered.
These are distinguished by their Chern numbers, c1 , and we will take 2 < c1 +3. These
factorizations are shown on the right edge of the diagram in Fig. 4.
There is one minor subtlety which arises from exceptional matrix factorizations. Recall from
Section 4 that the moduli space of some bundles may have singularities and that their smooth
resolution can involve an extension of the matrix factorization and thus, to a corresponding
extension of the underlying bundle to one of higher rank. We encountered this with the exceptional bundle that appears in the special matrix factorization, P 1044 . Such factorizations are
rigid and do not exist at generic moduli. Thus our statement about there being only six different
factorizations is meant to be true only for generic moduli.
We have already seen that the vector bundle E(2, 3) is obtained from P33 , and thus it remains to construct the factorizations with c1 = 0, 1, 2. Below we will explain how to do this
explicitly, and obtain a new 4 4-matrix factorization, denoted by P44 and P44 , which describes the vector bundles E(2, 2) and E(2, 2) respectively; a 6 6-matrix factorization, P66 ,
and corresponding to E(2, 0); and a 5 5-matrix factorization, P55 , which gives E(2, 1). The
T .
bundle E(2, 1) is associated with the transpose, P55
278
7.1. Chern number 2

Since we already know that the 2 2 factorizations correspond to line bundles with c1 = 1,
the construction of bound states at threshold will lead us to rank two bundles with c1 = 2
respectively. Using (6.14), we consider the 4 4-matrix factorization given by

J = J22 () 0
0
J22 () ,

P44 :
(7.1)
E = E22 () 1
0
E () .
22
The explicit form of the tachyon that we obtain is given below:

0 = D()J22 ()
!

x32 2 " ! 2x2 1
3 x12
2x22 1
2 2 x12
x22 3 "
2x3 3
2x1 x2 2
x
1
1 3
2 3
2 3
1 2
1 3
32
22
32
=
,
x1 3 x3 2
x3 1 x2 3
! 2 2 x12 2x3 3 2x2 1

x22 3 "
x1 3 x3 2
1 3 + 1 2 2 3 x1 + 2
3
1 = D()E22 () =
! 3 x12 2x2 2 x1 2x22 1 x32 2 " ,
2 + 1 3 2 3 + 2
x2 3 x3 1
3
(7.2)
where denotes (2 ). The matrix bundles, EJ and EE , for the 4 4 factorization are
associated with the vector bundles E(2, 2) and E(2, 2), respectively. Because of the behavior
T is gauge equivalent to P
of P22 under transposition, it follows that P44
44 .
7.2. Chern number 0
The 3 3 factorization gives us two matrix bundles, EJ and EE , one having rank one and
degree zero and the other having rank two and degree 3. Therefore, the composite factorization
will consist of one matrix bundle of rank two and degree zero, plus a bundle of rank four and
degree 6, of the following form:

J = E33 ()
,
0
E33 ()

P66 :
(7.3)
1
J33 ()
.
E =
0
J33 ()
The explicit form of the tachyon is then given by (with l denoting l (2 ))
x 2
x 2
x 2
1 21 2x22x33 1 3 23 2x11x22 3 2 22 2x11x33 2
1
3
2
2 x 2 2x x
1 x22
3 x12
2x1 x3 1
2x2 x3 3
3
1 2 2
0 = D()E33 () = 2 1 3
2 2 3
2 1 2 ,
1
3
2 2
1 x32
3 x2
2 x12
2x1 x3 3
2x2 x3 2
2 1 2
2 1 3
2 2x12x23 1
3
1

2
x1 1 x3 2 x2 3
1 = D()J33 () = x3 3 x2 1 x1 2 .
(7.4)
x2 2 x1 3 x3 1
Recalling the results for P33 from Table 1, we see that the matrix bundles, EJ and EE , are
T ( ) is gauge
associated with the vector bundles E(2, 0) and E(4, 6) respectively. Note that P66
equivalent to P66 ( ), which follows from the known behavior (6.3) of P33 under transposition.
279
7.3. Chern number 1

These two composites arise as 5 5-matrix factorizations that are transposes of each other.
The composite obtained from the tachyon (Sa ,L a ) is given by

J = E33 () 0 (,) ,
E22 ()

0
P55 :
(7.5)
E = J33 () 1 (,) ,
0
J ()
22
where the tachyon is explicitly given by:
2 1 x12 1 1 x1 x3 1 2 x1 x2
x
+
3 2
2 2
3 3 1 1 12
1 2 x1
3 2 x1 x3 3 3 x1 x2
0 = 3 1 x1 + ,
3 1
2 2
1 2
2
(Sa ,L a ) :
2 3 x1 x3
2 1 x1 x2 3 3 x1
+
+
3 2 1
3 2
2 1
1 2
x x
1
3
1
3
2
1
1 = 2 2 x1 1 1 x1 ,
(7.6)
3 1 x1 2 3 x1
where denotes ( ). For P55 , the matrix bundle EJ is associated with the vector
bundle E(2, 5) E(2, 1). The factorization as given in (7.5) seemingly depends on two openstring moduli, and , of the 3 3 and 2 2 matrix factorizations that form the composite.
However, one can carry out a gauge transformation such that the factorization depends on only
the combination (3 + ). This combination is to be identified with the physical open-string
modulus of the composite. The form of the factorization after such a gauge transformation has
been given in Appendix C.
As we already mentioned in Section 6, remaining 5 5 matrix factorization, corresponding
T . This then
to E(2, 1), can be obtained by transposition and we will therefore denote it by P55
completes the list of all matrix factorizations related to rank two bundles.
Our purpose has been to show how matrix factorizations, physically realized in terms of
boundary LandauGinzburg models, can be used as an effective computational tool for dealing
with topological B-type branes. Once again we stress that to precisely understand these B-branes,
one must understand the derived category and not merely the K-theory of the branes. That is,
it is not sufficient to classify the branes in terms of their charges, but one must also understand
the equivariant maps between the branes (as well as the dependence of all these quantities on the
complex structure moduli). In more physical terms, this means classifying the tachyons between
branes and understanding the condensates they produce. We have seen that matrix factorizations
provide a very explicit way of realizing the branes and of performing such computations. Moreover the description of B-type branes via matrix factorization appears to be complete in that the
results of Refs. [8,9] show that the derived category of topological B-type D-branes is isomorphic
to the category of matrix factorizations [10,11] (at least for a large class of geometries). Thus we
have an explicit and relatively simple way of representing the category of topological B-branes,
as well as performing computations that capture the dependence on moduli, in a manner that is
easily accessible for physicists.
To flesh-out this picture, we have shown how such computations work in practice, and while
we have focused on the cubic torus, the ideas involved appear universal. We showed how the bun-
280
dle data for the branes and anti-branes can be extracted directly from the matrices. These bundle
data are ambiguous in a way that is familiar from the bulk linear sigma model: Generically, there
is a monodromy around the large-radius limit in the Khler moduli space that makes the bundle data ambiguous up to an overall tensoring with the line bundle LN . We have seen that this
ambiguity is mirrored in the B-model by a corresponding ambiguity in extracting the bundle data
from the matrix factors. Moreover, the ZN monodromy at the Gepner point in the Khler moduli
space is matched by the ZN orbit of brane charges that one can associate with a given matrix factorization.36 We have seen that these charges can be resolved by refining the matrix factorization
in terms of equivariant R-symmetry. In this way, the LandauGinzburg model is able to describe
general sheaves and not merely vector bundles. A more complete treatment, for instance based
on the forthcoming description of the boundary linear sigma model [9], may be needed in order
to precisely track states between LandauGinzburg and large radius phases, and determine how
the bundle data evolve over the Khler moduli space.
Having explicitly shown how the matrix factorizations encode the details of the underlying vector bundles, we showed, in some detail, how tachyon condensation works within this
framework. In particular, we demonstrated how repeated tachyon condensation of the 2 2 factorization can be used to construct, at least in principle, the matrix factorizations corresponding
to general vector bundles on the torus. Indeed, we showed how to obtain all previously known
factorizations, including their dependence on moduli, from the 2 2 factorization (prior to this
paper, the connection with the 3 3 factorization was only known implicitly, based upon the
RR charges). We went on to construct all the matrix factorizations corresponding to rank two
bundles on the torus.
One of the crucial ingredients in understanding and organizing the process of tachyon condensation was the application of equivariant R-symmetry [26,32]. The resulting orbits of branes
are labeled by representations of ZN , and when translated to the language of matrix factorization, these representations impose selection rules upon tachyons, and so effectively organize the
tachyon spectrum between any given pair of branes. This enabled us to determine the correct
tachyon to generate each and every possible condensate of the members of different ZN multiplets.
One of the other interesting features we uncovered in this study of tachyon condensation, and
the bundles associated with matrix factorization, is how the bundle structure can jump, apparently
discontinuously, on the open-string moduli space of the branes. We saw how such behavior is
easily, and smoothly, described by matrix factorizations. For example, the 3 3 factorization is
apparently singular at one point in its moduli space, however, we showed (following Ref. [31])
that this singularity could be resolved smoothly by passing to 4 4 factorization that, at generic
points in the moduli space, is equivalent to the 3 3 factorization. We found that the 4 4
factorization corresponds to a pair of rank two vector bundles, and, at generic points in the moduli
space, one of these rank two bundles is split, i.e., it is a trivial sum of line bundles, but at the
special point, where the 3 3 factorization is singular, both the rank two vector bundles are
indecomposable. We interpreted this configuration physically in terms of a pure anti-D2-brane,
which is rigid; deforming away from the special point corresponds to adding and pulling apart
36 Note that these two monodromies are very different: tensoring with a line bundle merely shifts the Chern number
c1 , while the ZN monodromy at the Gepner point changes all the Chern numbers, including the rank. Indeed, these
two monodromies taken together generate a (sub-)group of the full duality group of the theory, and thus a given matrix
factorization will describe a whole orbit under this group.
281
an extra D0-D0-brane pair, which amounts to a reducible bundle configuration depending on an

open-string modulus.
A general class of examples that exhibits similar behavior was encountered in studying bound
states at threshold. In this instance one combines two parallel branes and, if their moduli do not
match, then there is no tachyon and the overall vector bundle is simply the sum of the component
bundle for each brane. However when the moduli coincide then there is a new tachyon, formally
given by a boundary preserving operator, and this condenses the branes into a true bound state at
threshold that corresponds to a non-split extension of the component vector bundles.
Another, related feature that we found interesting is that the gauge symmetry, inherent to
any matrix factorization, sometimes leads to identifications in the open-string moduli space; this
can impose a non-trivial global structure on it, like an unexpected shortening of periodicities, or
orbifold points that connect to new branches.
We believe that many of the ideas and techniques we have used here will be applicable to
more general CalabiYau manifolds. Once one finds a set of matrix factorizations and suitable
tachyonic, boundary changing cohomology elements, one should be able to build more by condensation.37 However, the obvious approach of trying to find matrix factorizations pertaining to
something like the quintic threefold does not seem to be straightforward. One can easily convince
oneself by elementary counting arguments that matrix factorization of a generic quintic may be
difficult: One simply totals up the degrees of freedom in the matrix elements, taking into account
the gauge invariances, and subtracts the constraints imposed by (1.4). The result is generically
negative, and thus one should expect interesting new matrix factorizations of the complete quintic
only for special branes, and/or for special points in its complex moduli space.
For example, at the Fermat point one has all the matrix factorizations arising from tensoring
the matrices corresponding to RecknagelSchomerus states of the underlying minimal models;
the task would be to try to extend these away from the Fermat point. However, while for the torus
the sections relevant for the simplest branes are explicitly known to be given by theta and Appell
functions, much less is known about sections on threefolds. One the other hand, deformations of
branes on threefolds will often be obstructed, so that there are no open string moduli to begin
with; this may facilitate the construction of the corresponding matrix factorizations.
Moreover, it might also be rewarding to focus on a kind of hybrid description of B-type
branes. The same counting arguments that show that matrix factorizations for the complete
quintic are rare, also show that if there are only three variables of any degree, then there will
generically be matrix factorizations with possibly several moduli. This suggests that it might be
productive to look at B-type branes defined via vector bundles on sub-manifolds. For example,
by adding two further equations to the function(s) that define the CalabiYau manifold, one can
recast the CalabiYau, at least locally, as a Riemann surface fibered over a base, which may be
amenable to a treatment similar to the one described in this paper. Specifically, if one focuses
on elliptically fibered CalabiYau manifolds, one may make direct use of many of the results
presented here; this is presently under investigation.
37 Of course, a major new ingredient and complication is the problem of stability [1], but since this is tied to the Khler
parameters it is not clear to what extent this can be addressed within the topological B-model; for some ideas in this
direction, see Ref. [32]. It is also known that some of the permutation branes cross lines of marginal stability, i.e., they
can decay, when going to large radius. It is thus natural to carry out Seiberg dualities even in the topological B-model,
where they appear as a change of basis.
282
Acknowledgements
We would like to thank Robert Helling and Manfred Herbst for helpful comments, and especially Ilka Brunner, Dennis Nemeschansky and Johannes Walcher for participating in various
stages of this work. The work of N.W. is supported in part by the DOE grant DE-FG03-84ER40168.
Appendix A. Theta functions and Appell functions
The theta functions and Appell functions are defined by:
1 2
q 2 n zn ,
( ) = ( | )
(A.1)
nZ
(, ) = (, | )
q 12 n2 zn
nZ
qn y
(A.2)
where y = q m , m Z and
q e2i ,
z e2i ,
y e2i .
(A.3)
These functions also satisfy the periodicity relations:

1
( + ) = q 2 z1 ( ),
( + 1) = ( ),
(, + 1) = ( + 1, ) = (, ),

1
( + , ) = q 2 z y(, ) + ( ) .
(A.4)
(, + ) = y(, ) + ( ),
(A.5)
In addition, the Appell functions satisfy an interchange identity:

1

12 (1 + ) (, ) = q 2 zy 1 ( ) + 12 (1 + ), 12 (1 + ) .
(A.6)
By taking the limit as 0 on both sides of this identity one can then show that:
, 12 (1 + )
q 8 3 ( )

,
12 (1 + )
(A.7)
where
( ) q
1
24
#
1 qn
(A.8)
n=1
is the Dedekind -function. To uniformize the curve (1.1), we need the theta functions with
characteristics:

c1
2
q (m+c1 ) /2 e2i( +c2 )(m+c1 ) ,
,
(A.9)

c2
mZ
and then define:

%

( ) ( | ) =
3, 3 ,

3 1 1
3
2
= i(1)(1) q 2 ( 2 + 3 (1)) z( 2 +(1)) 3 32 + ( 1) 12 3 ,
$
(1)
283
1
1
3 (1 ) 2
12
(A.10)
where e2i/3 and = 1, 2, 3. One can then show that (1.1) is uniformized by taking x =
( ) with the modulus, a, related to via:

3a(a 3 + 8) 3
= j ( ).
(A.11)
a3 1
The associated Appell functions are then defined by:
(, ) i(1)
(1)
2
31 1
2 2 + 3 (1)
31
2
3
2 +(1)
3

n
1 2
3
(q 3 ) 2 n q ( 2 +(1)) z3
3
nZ
(q 3 )n y 3 q ( 2 +(1))
= i(1)(1) q 2 2 + 3 (1) z 2 +(1)

3 + 32 + ( 1) , 3 32 + ( 1) 12 |3 .
(A.12)
We now sketch some of the manipulations involving the l that we used in the main text of the
paper. A change in the sign of the argument of l is the permutation (up to an overall sign):
1 ( ) = 1 ( ),
2 ( ) = 3 ( ),
3 ( ) = 2 ( ).
(A.13)
In the following, let us denote ( ), () and ( ) by , and , respectively. The

addition formulae for the ( ) may be deduced from (2.6). For example, by setting = ,
one obtains formulae such as the one given below with the argument of the doubled:
1 1 (2 ) =
1 (23 33 )
.
i( )3
(A.14)
Similar manipulations also lead to (3.30) and (4.7). The following identities also follow from
(2.6):
1 1 1 + 2 2 2 + 3 3 3 = 0,
1 2 3 + 2 3 1 + 3 1 2 = 0,
1 3 2 + 2 1 3 + 3 2 1 = 0.
(A.15)
These identities are useful in solving the physical state condition and in identifying the openstring modulus of condensates.
Appendix B. Row and column elimination of matrix factorizations
Specifying a matrix factorization, given by J and E satisfying (1.4), corresponds to choosing
a representation for the C[x1 , x2 , x3 ]-module homomorphism J and E. In general, however,
there are different representatives for the same module homomorphisms. In physical terms, the
choice of representative means that we can act upon a given matrix factorization with a gauge
transformation to obtain another matrix factorization that satisfies the defining condition (1.4)
284
and thus describes the same D-brane configuration. From the point of view of the category, such
gauge transformations act trivially on the objects and thus lead to an equivalent description.
For a n n-matrix factorization, the gauge transformations are given by Eq. (1.5) with the
transformation matrices UL and UR as elements of GL(n, C[x1 , x2 , x3 ]). That is to say UL and
UR are invertible matrices in the polynomial ring C[x1 , x2 , x3 ].38
In the following we often make use of the gauge freedom (1.5) in order to rewrite a given
D-brane configuration in a convenient gauge. There are three basic gauge transformations which
allow us to simplify matrix factorizations step by step. The idea is to simplify in each step one
of the two matrices of a matrix factorization, e.g., J . A gauge transformation acting on J via UL
and UR then also defines according to (1.5) the corresponding gauge transformation acting on E.
The first basic gauge transformations are simply given by multiplying a matrix row r and a
matrix column t of J by non-zero constants a and b respectively. For a n n-matrix factorization
this gives rise to the diagonal transformations matrices
UL (a, b) = Diag(1, . . . , 1, a, 1, . . . , 1),
UR (a, b) = Diag 1, (. . . , 1, b, 1, . . . , 1),

UL (a, b)1 = Diag 1, . . . , 1, a1 , 1, . . . , 1 ,
UR (a, b)1 = Diag 1, . . . , 1, b1 , 1, . . . , 1 .
(B.1)
The second kind of gauge transformations are given by either adding the row r of J , multiplied
by a polynomial, pr (x), to the row s of J , or analogously by adding the column t of J , multiplied by a polynomial, pc (x), to the column u of J . The appropriate transformation matrices are
respectively given by

UL r, s, pr (x) = 1nn + pr (x)s,r ,
UR r, s, pr (x) = 1nn ,

1
1
UL r, s, pr (x)
(B.2)
= 1nn pr (x)s,r ,
UL r, s, pr (x)
= 1nn ,
and

|
UL t, u, pc (x) = 1nn ,
1
|
UL t, u, pc (x)
= 1nn ,
with

|
UR t, u, pc (x) = 1nn + pc (x)t,u ,
1
|
UL t, u, pc (x)
= 1nn pc (x)t,u ,
&
(B.3)
1 for r = k, s = l,
(B.4)
0 else.
With these simple transformation rules we can already deduce an important property of a
given matrix factorizations P : If either the matrix J or the matrix E contains a constant, that is
to say a non-vanishing entry of degree 0, the matrix factorization P simplifies as follows: For
concreteness let us assume that the matrix J of an n n-matrix factorization has a non-vanishing
constant in the top left corner. First, we act on the matrix factorization with a transformation of
the type (B.1) in order to normalize this constant to 1. Then we apply (n 1) times the gauge
transformations (B.2) with r = 1, s = 2, . . . , n and appropriate polynomials pr (x) such that the
first row of JP becomes (1, 0, . . . , 0). In a third step we apply the gauge transformation (B.3)
n 1 times with t = 1, u = 2, . . . , n and with suitable polynomials pc (x) so as to also reduce
the first column to (1, 0, . . . , 0). After this chain of gauge transformations we obtain a gauge
equivalent matrix factorization. As a consequence of (1.4) both the first row and the first column
(r,s )kl =
38 A matrix is invertible in C[x , x , x ] if and only if its determinant is a non-vanishing constant.

1 2 3
285
of E have automatically been transformed into (W, 0, . . . , 0)! Hence the original n n-matrix
factorization is really a (n 1) (n 1)-matrix factorization with a trivial irrelevant summand,
P11 . This technique, which reduces the dimension of a matrix factorization by applying gauge
transformations, is in the main text referred to row and column elimination.
Appendix C. An explicit form of the 5 5 matrix factorization
We present a form of the 5 5 matrix factorization with the open-string modulus is given
by l . This is gauge equivalent to the form given in (7.5) when l l (3 + ).
x2
x32 3ax1 x2
x22
x1 x2
1
x1 x1 2 32
3
2
1
3
0
x3 1
x2 2
x1 2
0
x
0
x1 3
0
x2 3
J55 = 3 1
(C.1)
,
3
3
2
2
2
2
x x +x
x2 1 2 x1 (2 +3 )
x2 2 x1 3
0
2 1 1 2 3 3
2
1
0
x1 3
x
2 2
E55 = x3 1
0
x1 x2 2
1
Q
x2
x2
x3 1 x2 3
22
x1 x3
3 )
x1 (x2 3 x3 1 )
2 3
( 12
x2 (x2 3 x3 1 )
2 3
x1 (x3 2 x1 3 )
32
x2 (x2 12 x1 22 +x3 32 )
1 2 3
x1 (x1 (23 +33 )x2 12 2 )
12 32
x1 x3 2
x3 1 (x1 3 x3 2 )
2 32
x3 1 (x2 3 x3 1 )
2 32
Q 3
Q 2
x1 x2 22
1 32
x12
3
x2 (x1 3 x3 2 )
32
x2 (x2 3 x3 1 )
32
x1 3 x3 2
Q 4
(C.2)
where
2 x12 3ax2 3 x1 + x32 3
,
1 3
x2
x2
2 = 2 + 3ax1 x2 3 ,
Q
3
1
1
1 2 3 x22 x3 12 2 x2 + x1 x3 (23 + 33 )
3 =
,
Q
1 2 32
1 =
Q
4 =
Q
2 3 x12 + x3 22 x1 x2 x3 12 x32 32
2 32
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
M.R. Douglas, D-branes, categories and N = 1 supersymmetry, J. Math. Phys. 42 (2001) 2818, hep-th/0011017.
C.I. Lazaroiu, Generalized complexes and string field theory, JHEP 0106 (2001) 052, hep-th/0102122.
P.S. Aspinwall, A.E. Lawrence, Derived categories and zero-brane stability, JHEP 0108 (2001) 004, hep-th/0104147.
M. Kontsevich, Homological algebra of mirror symmetry, alg-geom/9411018.
K. Hori, S. Katz, A. Klemm, R. Pandharipande, R. Thomas, C. Vafa, R. Vakil, E. Zaslow, Mirror Symmetry, American Mathematical Society, 2003.
C.I. Lazaroiu, D-brane categories, Int. J. Mod. Phys. A 18 (2003) 5299, hep-th/0305095.
P.S. Aspinwall, D-branes on CalabiYau manifolds, hep-th/0403166.
D. Orlov, Derived categories of coherent sheaves and triangulated categories of singularities, math.AG/0503632.
M. Herbst, K. Hori, D. Page, in preparation.
286
[10] M. Kontsevich, unpublished.

[11] D. Orlov, Triangulated categories of singularities and D-branes in LandauGinzburg models, math.AG/0302304.
[12] A. Kapustin, Y. Li, D-branes in LandauGinzburg models and algebraic geometry, JHEP 0312 (2003) 005, hepth/0210296.
[13] I. Brunner, M. Herbst, W. Lerche, B. Scheuner, LandauGinzburg realization of open string TFT, hep-th/0305133.
[14] N.P. Warner, Supersymmetry in boundary integrable models, Nucl. Phys. B 450 (1995) 663, hep-th/9506064.
[15] S. Govindarajan, T. Jayaraman, T. Sarkar, Worldsheet approaches to D-branes on supersymmetric cycles, Nucl.
Phys. B 580 (2000) 519, hep-th/9907131.
[16] I. Brunner, M. Herbst, W. Lerche, J. Walcher, Matrix factorizations and mirror symmetry: The cubic curve, JHEP
0611 (2006) 006, hep-th/0408243.
[17] M.F. Atiyah, Vector bundles over an elliptic curve, Proc. London Math. Soc. V II (3) (1957) 414452.
[18] A. Polishchuk, E. Zaslow, Categorical mirror symmetry: The elliptic curve, Adv. Theor. Math. Phys. 2 (1998) 443,
math.AG/9801119.
[19] R. Laza, G. Pfister, D. Popescu, Maximal CohenMacaulay modules over the cone of an elliptic curve, J. Algebra 253 (2) (2002) 209236.
[20] C. Baciu, Maximal CohenMacaulay modules over the Affine cone of the simple node, math.AC/0511405.
[21] K. Hori, J. Walcher, F-term equations near Gepner points, JHEP 0501 (2005) 008, hep-th/0404196.
[22] A. Sen, Tachyon condensation on the brane antibrane system, JHEP 9808 (1998) 012, hep-th/9805170.
[23] A. Kapustin, Y. Li, Topological correlators in LandauGinzburg models with boundaries, Adv. Theor. Math. Phys. 7
(2004) 727, hep-th/0305136.
[24] A. Kapustin, Y. Li, D-branes in topological minimal models: The LandauGinzburg approach, JHEP 0407 (2004)
045, hep-th/0306001.
[25] C.I. Lazaroiu, On the boundary coupling of topological LandauGinzburg models, JHEP 0505 (2005) 037, hepth/0312286.
[26] S.K. Ashok, E. DellAquila, D.E. Diaconescu, Fractional branes in LandauGinzburg orbifolds, Adv. Theor. Math.
Phys. 8 (2004) 461, hep-th/0401135.
[27] K. Hori, Boundary RG flows of N = 2 minimal models, hep-th/0401139.
[28] S.K. Ashok, E. DellAquila, D.E. Diaconescu, B. Florea, Obstructed D-branes in LandauGinzburg orbifolds, Adv.
Theor. Math. Phys. 427 (2004) 8, hep-th/0404167.
[29] M. Herbst, C.I. Lazaroiu, Localization and traces in open-closed topological LandauGinzburg models, JHEP 0505
(2005) 044, hep-th/0404184.
[30] M. Herbst, C.I. Lazaroiu, W. Lerche, D-brane effective action and tachyon condensation in topological minimal
models, JHEP 0503 (2005) 078, hep-th/0405138.
[31] K. Hori, J. Walcher, D-branes from matrix factorizations, C. R. Physique 5 (2004) 1061, hep-th/0409204.
[32] J. Walcher, Stability of LandauGinzburg branes, J. Math. Phys. 46 (2005) 082305, hep-th/0412274.
[33] I. Brunner, M.R. Gaberdiel, Matrix factorisations and permutation branes, JHEP 0507 (2005) 012, hep-th/0503207.
[34] I. Brunner, M.R. Gaberdiel, The matrix factorisations of the D-model, J. Phys. A 38 (2005) 7901, hep-th/0506208.
[35] H. Enger, A. Recknagel, D. Roggenkamp, Permutation branes and linear matrix factorisations, hep-th/0508053.
[36] E. DellAquila, D-branes in toroidal orbifolds and mirror symmetry, hep-th/0512051.
[37] S. Govindarajan, T. Jayaraman, Boundary fermions, coherent sheaves and D-branes on CalabiYau manifolds, Nucl.
Phys. B 618 (2001) 50, hep-th/0104126.
[38] S. Hellerman, S. Kachru, A.E. Lawrence, J. McGreevy, Linear sigma models for open strings, JHEP 0207 (2002)
002, hep-th/0109069.
[39] A. Recknagel, V. Schomerus, D-branes in Gepner models, Nucl. Phys. B 531 (1998) 185, hep-th/9712186.
[40] I. Brunner, M.R. Douglas, A.E. Lawrence, C. Rmelsberger, D-branes on the quintic, JHEP 0008 (2000) 015, hepth/9906200.
[41] M.R. Douglas, B. Fiol, C. Rmelsberger, The spectrum of BPS branes on a noncompact CalabiYau, JHEP 0509
(2005) 057, hep-th/0003263.
[42] D. Berenstein, M.R. Douglas, Seiberg duality for quiver gauge theories, hep-th/0207027.
[43] B. Ezhuthachan, S. Govindarajan, T. Jayaraman, A quantum McKay correspondence for fractional 2p-branes on LG
orbifolds, JHEP 0508 (2005) 050, hep-th/0504164.
[44] C. Vafa, Brane/anti-brane systems and U(N|M) supergroup, hep-th/0101218.
[45] B.R. Greene, C. Vafa, N.P. Warner, CalabiYau manifolds and renormalization group flows, Nucl. Phys. B 324
(1989) 371.
[46] A. Polishchuk, M.P. Appells function and vector bundles of rank 2 on elliptic curve, math.AG/9810084.
[47] A. Sen, Dyon-monopole bound states, selfdual harmonic forms on the multi-monopole moduli space, and SL(2, Z)
invariance in string theory, Phys. Lett. B 329 (1994) 217, hep-th/9402032.
Intrinsic moment of inertia of membranes as bounds

for the mass gap of YangMills theories
M.P. Garcia del Moral a,b,c, , L. Navarro d , A.J. Prez d , A. Restuccia d
a Perimeter Institute for Theoretical Physics, Waterloo, ON, N@J 2Y5, Canada
b Department of Physics and Astronomy, MacMaster University,
1280 Main Street West, Hamilton, ON, L8S 4M1, Canada

c DAMTP, Centre for Mathematical Sciences, University of Cambridge CB3 0WA, UK
d Departamento de Matemticas, Universidad Simn Bolvar, Apartado 89000, Caracas 1080-A, Venezuela
Received 28 November 2006; accepted 13 December 2006

Abstract
We obtain the precise condition on the potentials of YangMills theories in 0 + 1 dimensions and D0brane quantum mechanics ensuring the discreteness of the spectrum. It is given in terms of a moment of
inertia of the membrane. From it we obtain a bound for the mass gap of any D + 1 YangMills theory in the
slow-mode regime. In particular we analyze the physical case D = 3. The quantum mechanical behavior of
the theories, concerning its spectrum, is determined by harmonic oscillators with frequencies given by the
inertial tensor of the membrane. We find a class of quantum mechanic potential polynomials of any degree,
with classical instabilities that at quantum level have purely discrete spectrum.
1. Introduction
The characterization of the discreteness of bosonic potentials in quantum mechanics represents a very challenging problem by itself and in relation to matrix model regularizations of field
theories and to dimensional reduction of YangMills to 0 + 1 dimensions. It would provide information about the spectrum of quantum non-Abelian gauge theories, as well as theories containing
gravity (bosonic membrane of M-theory).
E-mail addresses: mmoral@perimeterinstitute.ca, m.g.d.moral@damtp.cam.ac.uk, garcia@to.infn.it (M.P. Garcia

del Moral), lnavarro@ma.usb.ve (L. Navarro), ajperez@ma.usb.ve (A.J. Prez), arestu@usb.ve (A. Restuccia).
doi:10.1016/j.nuclphysb.2006.12.006
288
M.P. Garcia del Moral et al. / Nuclear Physics B 765 [PM] (2007) 287298
The potential of the 2-brane theories is quartic on the fields and exactly the same, after
a SU(N ) regularization, to the dimensional reduction of YangMills theories to 1 dimension
(time). The potential in terms of the membrane fields has valleys extending to infinity along the
directions of zero potential. There are configurations, which have been called string-like spikes
[1,2] in the context of D = 11 supermembrane theory, for which the energy is zero. They can
be attached to the membrane and even connect two membranes without any change in the energy of the system. Those configurations correspond from the YangMills point of view to gauge
fields valued on Abelian subalgebras of SU(N ), in particular on the Cartan subalgebra. A toy
model for these class of potentials was first studied in [3] and then extended to a toy model of
the D = 11 supersymmetric membrane in [1]. The potentials are classically unstable, quantum
mechanically stable and its supersymmetric extension quantum unstable. Although the bosonic
potential presents flat directions extending to infinity the walls of the valleys become narrower
as we move to infinity in a way that the wave function cannot escape to infinity. A proof of discreteness of the spectrum of the SU(N ) regularized Hamiltonian of the bosonic membrane was
presented in [4], where a bound
, H ,
(1)
in terms of a function (x), with (x) as x was obtained. It is a sufficient condition for the discreteness of the spectrum. However in order to understand the exact property
of the membrane, and, as a consequence of YangMills, ensuring the discreteness of the spectrum one should look for a necessary and sufficient condition. That condition was discovered by
Molchanov, [5] and recently extended by Mazya and Shubin [6], and makes use of the mean
value, in the sense of Molchanov, of the potential on an n-dimensional cube. If the mean value
goes to infinity at large distances in configuration space the spectrum is discrete, otherwise is
continuous. We consider the Molchanov condition for the membrane potential. By performing
the explicit calculation we obtain that at large distances in the direction of the valleys, the mean
value of the potential behaves as an harmonic oscillator with frequencies determined by a moment of inertia of the membrane. There is a inertial tensor associated to the potential of the
membrane, or equivalently of the D0 mechanics, which is strictly positive. One may interpret the
mean value of the potential as a rotational energy of D0 branes.
This is the precise property ensuring the discreteness of the membrane and of YangMills in
0 + 1 dimensions. This approximation for YangMills describe the spectrum of the theory when
only the modes with zero momenta but nonzero energy are kept. This regime has been thought
to contain relevant information of the IR phase of the theory in which the gluons are expected to
be confined forming bound states called glueballs.
We then obtain bounds for the YangMills Hamiltonian in terms of the moment of inertia
of the membrane. The Hamiltonian bound implies that the resolvent of the Schrdinger operator is compact. It also provide a bound for the mass gap. The same results are proven for the
YangMills theories in the slow mode regime and give an strong indication that the complete
YangMills theories in D-dimensions have a mass gap.
The interesting properties of the membrane and YangMills potentials are also present in more
general potentials, of higher degree than four. We discuss a class of these potentials in Section 3,
in particular they are related to the D = 11 5-brane potential. We prove discreteness of these
class of potentials.
The paper is organized as follows: In Section 2 some preliminary results are presented. In
Section 3 we obtain a class of potentials for which the spectrum of the Schrdinger operator
is discrete, extending previous results. In Section 4 we introduce the moment of inertia of the
289
membrane In Section 5 we obtain a lower bound to the Hamiltonian in terms of it. In Section 6.
the discreteness of the YangMills Hamiltonian in the slow mode regime is proved. In Section 7
we present the conclusions.
2. Preliminary results
In 1953 Molchanov (see [5]) proved that in dimension n = 1, the discreteness of the spectrum
of the Schrdinger operator + V in L2 (Rn ) with a locally integrable and semi-bounded
below potential V is equivalent to

For every fixed d > 0
V (x) dx as Qd
Qd
where Qd is an open cube with the edges parallel to coordinate axes and of edge length d and
Qd means that the distance from Qd to 0 goes to infinity. In the same paper Molchanov
showed that this condition is not sufficient for dimension n 2, his counter-example is based on
Theorem 2.1 of his paper:
The spectrum of the Schrdinger operator + V is not discrete if and only if there exist
a constant C > 0 and a sequence of disjoint cubes {Qk }
k=1 with the same edge length such
that for all k = 1, 2, . . .

(x)2 + V (x)(x)2 dx < C
(Qk ) = inf
Qk
where
the infimum is taken over all the functions such that |Qk = 0 and

2 dx = 1.
|(x)|
Qk
After, Molchanov stated and proved a modified necessary an sufficient condition for the discreteness of the spectrum of Schrdinger operators for n 2, which involve the class of negligible
compact subsets of the cube. We will only state the V. Mazya and M. Shubins generalization
(see [6]). Before we state this theorem, let us remember the following definition of capacity. For
the details and the full-general version of the theorem see [6].
Definition 1. Let n 3, F Rn be compact, and Lipc (Rn ) the set of all real-value functions with
compact support satisfying a uniform Lipschitz condition in Rn . Then the Wieners capacity of F
is defined by

u(x)2 dx u Lipc Rn , u|F = 1
cap(F ) = capRn (F ) = inf
Rn
In physical terms the capacity of the set F Rn is defined as the electrostatic energy over Rn
when the electrostatic potential is set to 1 on F .
Definition 2. Let Gd Rn be an open, bounded and star-shaped set of diameter d, let (0, 1).
The negligibility class N (Gd ; Rn ) consists of all compact sets F Gd satisfying cap(F )
cap(Gd ).
290
Balls and cubes in Rn are useful examples of such Gd . In what follows we denote the ball of
diameter d and center x by Bd (x) and the n-dimensional Lebesgue measure by Vol().
Theorem 1 (Mazya and Shubin). Let V L1loc (Rn ), V 0.
Necessity: If the spectrum of + V in L2 (Rn ) is discrete then for every function
: (0, +) (0, 1) and every d > 0

inf
(2)
V (x) dx + as Gd .
F N (Gd ;Rn )
Gd \F
Sufficiency: Let a function d (d) (0, 1) be defined for d > 0 in a neighborhood of 0 and
satisfying
lim sup d 2 (d) = +.
d0
Assume that there exists d0 > 0 such that (2) holds for every d (0, d0 ). Then the spectrum of
+ V in L2 (Rn ) is discrete.
Remark 2. It follows from the previous theorem that a necessary condition for the discreteness
of spectrum of + V is

V (x) dx as Gd .
(3)
Gd
Let us recall that in 1934, K. Friedrichs (see [6] for further references) proved that the spectrum of the Schrdinger operator + V in L2 (Rn ) with a locally integrable potential V is
discrete provided V (x) as |x| .
Remark 3. The presence of negligibility sets F in the formulation of the Mazya and Shubins
theorem is related to the first term in the expression of (Qk ). In fact this term describes also an
electrostatic energy, when = 1 on F Q the energy is minimized by the harmonic solution to
the corresponding Dirichlet problem. The energy of the harmonic solution defines the capacity
of the capacitor [F, Q] while the capacity of F is the capacity of the capacitor [F, ].
The following lemma is very useful tool in the next sections.
Lemma 4. For each given Gd = Gd (x0 ),
cd :=
inf
F N (Gd ;Rn )
Vol(Gd \ F ) > 0.
Proof. Let V (x) = |x|. Then by Friedrichs theorem the spectrum of + V is discrete, so by
Theorem 1 we have

V (x) dx as |x0 | .
inf
F N (Gd ;Rn )
Gd \F
291

Now Gd \F V (x) dx (|x0 | + d) Vol(Gd \ F ) implies that

inf
V (x) dx |x0 | + d
inf
Vol(Gd \ F ),
F N (Gd ;Rn )
F N (Gd ;Rn )
Gd \F
from which follows that infF N (Gd ;Rn ) Vol(Gd \ F ) > 0, as we claimed.
3. Some potentials for which the spectrum of its Schrdinger operator is discrete
In this section we use the sufficiency conditions for the discreteness of spectrum provided by
Theorem 1.
Theorem 5. Let V 0, V L1loc (Rn ) and cd as in Lemma 4. Suppose that there is some d0 > 0
such that for all d (0, d0 ), there exists an open neighborhood d of V 1 (0) with the following
properties:
(i) In Rn \ d , V (x) as |x| .
(ii) Vol(Bd d ) < c2d for every ball Bd (x0 ) with |x0 | d0 .
Then the spectrum of the Schrdinger operator + V in L2 (Rn ) is discrete.
Proof. Let Gd = Bd (x0 ) with d (0, d0 ) fixed. Let F N (Gd ; Rn ) and E = (Gd \ F ) \
(Gd d ). Notice that E Rn \ d . By condition (i) for each M > 0, there exists R > 0 such
that
If
x Rn \ Gd
and |x| > R,
then V (x) > M.
In this way, if |x0 | d > R then |x| > R for all x Gd , and as V 0 we have

cd
V (x) dx V (x) dx M Vol(E) M Vol(Gd \ F )
2
Gd \F
and by Lemma 4

V (x) dx M
inf
F N (Gd ;Rn )
Gd \F
cd
2
if |x0 | > R + d.
This together with Theorem 1 implies that the spectrum of + V is discrete.
Now, let us apply the previous theorem.

Proposition 6. Let V (x) = nk=1 |xk |k , where k > 0 for all k = 1, 2, . . . , n. Then the spectrum
of the Schrdinger operator + V in L2 (Rn ) is discrete.
n
Proof. Let d0 = n2
Conn , where n denotes the area of the unit ball, and let d (0, d0 ) fixed.
n
n
sider Uk = {x R | |xk | < d }, where 0 < d < 1 will be determined below. Let d = k=1 Uk .
Then V 1 (0) d .
292
(a) In Rn \ d , V (x) as |x| : First, x Rn \ d implies |xk | d for all k =

1, 2, . . . , n.
Let |xj | = maxk=1,...,n |xk |, M = maxk=1,...,n k and m = mink=1,...,n k . Then, denoting
|| = 1 + + n we have
j

n
j

||j
||m |x|
|xk |k d
d
.
max |xk |
V (x) =
k=1,...,n
n
k=1
Thus, for |x| > 1 we have

||
m
d
V (x)
|x|m ,
( n)M
from which follows that V (x) as |x| .

(b) Vol(Bd d ) < c2d for every ball Bd (x0 ) with |x0 | d0 : For |x0 | large enough, if Bd (x0 )
d = then Bd (x0 ) Uk = for only one k {1, 2, . . . , n}, namely k = j . In consequence,

Bd (x0 ) d x Rn | |xj | < d , |xk x0j | < d, k = 1, . . . , j 1, j + 1, . . . , n ,
from which follows that Vol(Bd (x0 ) d ) 2n d d n1 and this, in turn, implies what we
wanted to prove if we choose
d <
cd
.
n
2 d n1
(4)
Thus, by the previous theorem we conclude that the spectrum of + V is discrete.

n
Remark 7. Given that cd nnd then inequality (4) implies that d <
n
as d 0 and also d < 2nnnd < 2nndn0 = 2nnn n2
n = 1.
n d
2n n ,
which tends to zero
4. Moment of inertia and discreteness of the potential for membrane theories

The matrix regularization of the bosonic membrane in the light cone gauge was done long
time ago by [2,710]. It consists in expanding the bosonic fields X m in an orthonormal basis
ZA ( ), defined on the Riemann surface ,
A
Xm = Xm
( )ZA ( ).
(5)
C , are those of the infinite

The structure constants arising from the symplectic bracket on , gAB
group of area preserving diffeomorphisms.
The idea of matrix regularization is to integrate the dependence on the spatial coordinates
and to regularize this model by an SU(N ) one, in such a way that we end up with a theory of
C converge in the large N limit to
quantum mechanics of matrices, whose structure constants fAB
C
the original ones gAB , recovering the original symmetries in the large N limit. This limit is in
general not rigorously known, and a discussion on it will be presented in a separate paper [12].
In this paper we are only concerned with the theory at finite N . The potential associated to the
membrane is described by

C 2
V = YmA YnB fAB
(6)
,
293
where Ym , m = 1, . . . , d, denote the transverse direction to the light cone coordinates Y + Y .

A, B, C denote SU(N ) indices and the structure constants,

AB
C
C
fAB = 2iN sin
(7)
,
A+B
N
A = (a1 , a2 ) a1 , a2 = 1, . . . , N 1 modulo N and with |m| + |n| = 0. The structure constants
in the large N limit are proportional to the structure constants of the area preserving diffeomorphisms, the residual gauge symmetry of the membranes in the light cone gauge. The large N
limit is defined by taking indices A, B on a fixed set and taking N . We are interested in
the evaluation of

inf
V,
F N (Gd ;Rn )
Gd \F
where Gd is the star-shaped set and F the negligible set satisfying cap F cap Gd defined in
Section 1, as the distance of Gd , to a fixed point in Rn goes to infinity. V satisfy the assumptions
of Theorem 1. The integration is performed with the Lebesgue measure dY . We introduce for
a given F N (Gd ; Rn ), the center of mass of the volume Gd \ F by

A
YmA dY = Xm
Vol(G/F ).
(8)
Gd F
We decompose,
A
A
YmA = Xm
+ xm
,
with
(9)

=
A
xm
dY
Gd \F
A
xm
dx = 0
(10)
[Gd \F ]0
where [Gd \ F ]0 denote the set with its center of mass at the origin of coordinates. We then have

A + x A Dm XB + x B ,
V (Y ) =
(11)
Xm
m
m
m
AB
m
where

E E
XnC + xnC XnD + xnD fBC
fAD
DAB
n=m
= 4N

XnC
n=m
and
V (Y ) =

+ xnC

B C
AD
A+C
D
D
Xn + xn sin
,
sin
B+D
N
N

A m
B
X m
DAB Xm
+ O(Xm ) .
(12)
(13)
m
m)
Here limXm O(X
Xm = 0 assuming the other coordinates of the center of mass bounded. We
will consider first the case when only one Xm tends to infinity, namely Xm0 , while all the
294
others Xm , m = m0 remain bounded. We get

V (Y ) dY
Gd \F

V (X + x) dx
=
[Gd \F ]0

A
X m

m
B
DAB
dx Xm
+ O(Xm )
[Gd F ]0
A m
X m
IAB + 4N 2
XnC X nD sin
n=m
AC
B D
A+C
B
sin
B+D Vol(Gd F ) Xm
N
N
+ O(Xm ),
where
Im
AB
(14)

4N
n=m
xnC xnD sin
2
[Gd \F ]0
AC
B D B+C
sin
A+D dx
N
N
(15)
is an intrinsic inertial tensor of the membrane associated to the distribution [Gd \ F ]0 . Im

AB is
a self-adjoint positive matrix. The diagonal terms are

2
C C
2 AC
=
4N
x
x
sin
dx.
Im
(16)
n n
AA
N
[Gd \F ]0 n=m
Its trace is
tr Im =
4
Im
AA = 2N
xnC xnC dx.
(17)
[Gd \F ]0 n=m
This is the moment of inertia of the mass distribution Gd \ F with respect to the subspace genA . The second term in Eq. (14) is quartic on the center of mass coordinates. Its
erated by Xm
contribution to the integral of the potential is the value of the potential at the center of mass times
the volume of the Gd \ F ,
A C E 2
Xm Xn fAC Vol(Gd \ F ).
(18)
This term becomes zero when we take the center of mass at the bottom of the valleys, in the
directions where V is zero. We then conclude that along these directions where the potential is
zero we obtain,

lim
V dx
inf
Xm0 F N (Gd ;Rn )
=
=

lim
Gd \F
inf
Xm0 F N (Gd ;Rn )

lim
Xm0
Xm0 2
A m0 B
X m
I X
0 AB m0

inf
F N (Gd ;Rn )
A
Xm0
Xm0
IAB0
B
Xm
0
Xm0
(19)
295
and we notice that

A
B
Xm0 m 0 X m
0
inf
F ,
IAB

inf
Xm0
F N (Gd ;Rn ) Xm0
F N (Gd ;Rn )
m
where F is the minimum eigenvalue of IAB0 . It may be shown by explicit calculations and
using Lemma 4 that the infimum of the right-hand side of last inequality is greater than 0. One
may also obtain that conclusion in an indirect way from Theorem 1. In fact we know from an
independent argument (see next section or [4]), that the spectrum of the Schrdinger operator we
are considering is discrete with finite multiplicity. According to Theorem 1, the left hand member
of Eq. (19) must then go to infinity when Xm0 , hence it is necessary that
inf
F N (Gd ;Rn )
F > 0.
If the center of mass goes to infinity in more than one direction Xm , m = 1, . . . , d, (19) remains valid. The inertial tensor has now an additional index Imn
AB where m, n = 1, . . . , d. And it
is a strictly positive self adjoint matrix, the diagonal terms have the same expression (16). The
corresponding equality (19) holds when the center of mass of Gd \ F is located on the directions
where the potential is zero. In other directions where the potential does not vanish the left-hand
side of (19) is greater or equal to the quadratic right-hand side of (19).
We remark that this relation is the most precise one can obtain for the quantum Schrdinger
operator, since the condition of the theorem in Section 1 is a necessary and sufficient condition.
The mean value, in the sense of Molchanov, of the membrane potential is bounded by the mean
value of an harmonic oscillator at large distances in the configuration space and it is equal to it
at the directions of zero potential. This bound implies, using Mazya and Shubins theorem, that
the corresponding Schrdinger operator is discrete with finite multiplicity.
5. Bounds for the Hamiltonian
We present in this section a bound for the Hamiltonian of the regularized bosonic M2-brane
in terms of the integrand of the moment of inertia we discussed in the previous section. It is an
independent argument showing the discreteness of the spectrum and providing bounds to all the
eigenvalues of the membrane. It represents a generalization of the proof given in [3].
A instead of Y A for the configuration variables. We rearrange
Hereafter we use the notation xm
m
the potential of the M2-brane in the form
V (x) =
d

A B m
xm
xm MAB ,
(20)
C E C D
fAE
xn fBD xn
(21)
where
m
MAB
=

n=m
is a positive hermitian matrix. Then, the Hamiltonian can be written as

1
1
H = + ( + 2V ),
2
2

2

1/2
A B m
+
2x
x
M
2 tr M m
,
+ 2V =
m m AB
A
xm
m
(22)
296
hence

1
m 1/2
tr M
+ 2
.
H
2
m
(23)
The computation for tr M m gives

xnE xnE .
tr M m = 2N 4
(24)
n=m
We finally get the operatorial bound,

1/2
1
xnE xnE
H
+ 2N 2
H .
2
m
(25)
m=n

The potential V 2N 2 m ( m=n xnE xnE )1/2 , satisfies the criteria: V when x =

( n xnE xnE )1/2 implying that the resolvent of H is compact, H has a discrete spectrum
with finite multiplicity. The formula also provides a bound for the mass gap of H given by the
first eigenvalue of the operator H .
6. Spectrum of YangMills theories in the slow-mode regime
In this section we will consider the explicit relation between dimensional reduced Yang
Mills theories to 0 + 1 dimensions and the membrane moment of inertia condition ensuring the
discreteness of the spectrum.
The Hamiltonian of YangMills in D + 1 spacetime dimensions has the following Hamiltonian
2
g i 2
1
+ 2 Fija Fija Di ai Aai
Hym =
(26)
0 ,
2
4g
2
where ai represents the conjugate momenta to Aai . The indices of color are a = 1, . . . ,
dim SU(N ) = N 2 1, and i = 1, . . . , D labels the spatial components of gauge fields. The last
piece of the Hamiltonian represents the Gauss law:
Di ai = 0.
(27)
We assume we are in the slow mode regime and all of the fast modes has been integrated out
in such a way that we deal with an effective Hamiltonian containing gauge fields Ai (t) without
spatial dependence. This approximation has been considered to keep relevant information of the
theory in the IR limit where the gluons are supposed to be confined forming glueballs bound
states. The Hamiltonian has then the expression
2

g Vm i 2 Vm abc b c 2
red
i a ia
Hym = tr
(28)
+ 2 f Ai Aj Ai , A0 (t).
2
4g
In order to make comparison with the Hamiltonian of the preceding sections we redefine the
fields in terms of dimensionless ones. Following [11] we take
Ai
g 1/3
1/3
Vm
Ai ,
Pi
1
2/3
g 2/3 Vi
Pi .
(29)
The Hamiltonian can be re-written as,

g 2/3
g 2/3
1 i 2 1 abc b c 2
red
i a ia
+ f Ai Aj Ai , A0 (t) = 1/3 H.
Hym = 1/3 tr
2
4
Vm
Vm
297
(30)
The mass operator of the regularized bosonic membrane also in terms of dimensionless variables
can be written as,

M2
1 i 2 1 abc b c 2
2/3
i a ia
+ f Ai Aj Ai , A0 (t) = T 2/3 H.
(31)
= T tr
2
2
4
This means that the two theories are equivalent and the discreteness of YangMills in the slow
mode regime is related to the existence of a tensor of moment of inertia of the membrane. We
interpret it as an strong evidence that the configuration space for YangMills theories in the
large N limit is compact. The discreteness condition then is realized as a condition of rotational
energy of D0-branes in the configuration space of the bosonic membrane.
6.1. Mass gap and analytic eigenvalues distribution
An interesting problem is to estimate the mass gap of the theory for a particular value of N .
We have from the previous section

1
2
E E 1/2
H
(32)
H .
+ 2N
xn xn
2
m
m=n
Let us now consider, as an example, the case of SU(3) in D = 3 + 1 since it is the gauge
group realized in the nature. The number of degrees of freedom once the gauge fixing condition
has been taken into account is d = 2, A = 1, . . . , 8, and N = 3. H is bounded from below by the
following Hamiltonian

1/2
H1 = + 2N 2 X I X I
(33)
, I, J = 1, . . . , 16.
The eigenfunctions of H1 are expressed in terms of Bessel functions. The eigenvalues of H1 are
lower bounds, one by one, of the eigenvalues of H . That is, SU(3) YangMills in the slow-mode
regime in 3 + 1 dimensions has a discrete spectrum and its eigenvalues and mass gap are bounded
2/3
by those of g 1/3 H1 .
Vm
7. Conclusions
We showed that the most precise condition ensuring the discreteness of the spectrum of the
membrane theories and YangMills in the slow mode regime, is given in terms of an intrinsic
moment of inertia of the membrane. It may be interpreted as if the membrane, or equivalently
the D0 branes describing it, have a rotational energy. It is a quantum mechanical effect. The condition is obtained from the Molchanov, Mazya and Shubin necessary and sufficient condition on
the potential of a Schrdinger operator to have a discrete spectrum. The criteria is expressed not
in terms of the behavior of the potentials at each point, but by a mean value, on the configuration
space. The mean value in the sense of Molchanov considers the integral of the potential on a finite
region of configuration space. It can be naturally associated to a discretization of configuration
space in the quantum theory. We found that the mean value in the direction of the valleys where
298
the potential is zero, at large distances in the configuration space, is the same as a harmonic oscillator with frequencies given by the tensor of inertia of the membrane. The interesting feature is
that all previously known bounds for the membrane and YangMills potential were linear on the
configuration variables, while our bound is quadratic on the configuration variables. The bounds
we obtained should also be relevant on the analysis of the spectrum of wrapped supermembrane
with nontrivial central charges [1316].
We also obtained bounds, based on the moment of inertia for the Hamiltonian of YangMills
theories in the slow mode regime which allows to obtain interpretation about the mass gap of the
theory. In particular a bound for SU(3) in 3 + 1 dimensions is given by a Hamiltonian whose
spectrum and eigenvalues are known, and its eigenfunctions are expressed in terms of Bessel
functions.
In Section 3 we obtained a class of potentials, of any degree in the configuration coordinates,
presenting the same valley properties of the quartic potentials discussed in Sections 46. In particular, one of them has similarities with the potential for the 5-brane in D = 11. We proved that
the associated Schrdinger operator has discrete spectrum, with finite multiplicity.
Acknowledgements
We are very grateful to L. Boulton, F. Cachazo, H. Nicolai, Sh. Matsuura, V. Strauss for
helpful discussions and comments; to C. Burgess, F. Cachazo, R. Myers, F. Quevedo, for their
great support and encouragement. M.P.G.M. and A.R. would like to thank Perimeter Institute for
kind hospitality while part of this work was done. M.P.G.M. acknowledges support from NSERC
of Canada and MEDT of Ontario, and NSERC Discovery grant.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
B. de Wit, M. Lscher, H. Nicolai, The supermembrane is unstable, Nucl. Phys. B 320 (1989) 135.
H. Nicolai, R. Helling, Supermembranes and M(atrix) theory, hep-th/9809103.
B. Simon, Ann. Phys. 146 (1983) 209.
M. Lscher, Some analytic results concerning the mass spectrum of YangMills gauge theories on a torus, Nucl.
Phys. B 219 (1983) 233261.
A.M. Molchanov, On the discreteness of the spectrum conditions for selfadjoint differential equations of the second
order, Proc. Moscow Math. Soc. 2 (1953) 169199 (in Russian).
V. Mazya, M. Shubin, Discreteness of spectrum and positivity criteria for Schrdinger operators, Ann. Math. 162
(2005) 919942.
B. de Wit, J. Hoppe, H. Nicolai, On the quantum mechanics of supermembranes, Nucl. Phys. B 305 (1988) 545.
M. Claudson, M.B. Halpern, Supersymmetric ground state wave functions, Nucl. Phys. B 250 (1985) 689.
B. de Wit, A. Marquard, H. Nicolai, Area preserving diffeomorphisms and supermembrane Lorentz invariance,
Commun. Math. Phys. 128 (1990) 3962.
J. Hoppe, MIT PhD thesis, 1982.
G. Gabadadze, Modeling the glueball spectrum by a closed bosonic membrane, Phys. Rev. D 58 (1998) 094015,
hep-ph/9710402.
L. Boulton, M.P. Garcia del Moral, A. Restuccia, in preparation.
J. Bellorin, A. Restuccia, D = 11 supermembrane wrapped on calibrated submanifolds, Nucl. Phys. B 737 (2006)
190208, hep-th/0510259.
L. Boulton, M.P. Garcia del Moral, A. Restuccia, Discreteness of the spectrum of the compactified D = 11 supermembrane with non-trivial winding, Nucl. Phys. B 671 (2003) 343358, hep-th/0211047.
L.S. Boulton, M.P. Garcia del Moral, I. Martin, A. Restuccia, On the spectrum of a matrix model for the D = 11
supermembrane compactified on a torus with non-trivial winding, Class. Quantum Grav. 19 (2002) 2951, hepth/0109153.
M.P. Garcia del Moral, A. Restuccia, On the spectrum of a noncommutative formulation of the D = 11 supermembrane with winding, Phys. Rev. D 66 (2002) 045023, hep-th/0103261.

Nucl - Phys.B v.765

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nucl - Phys.B v.765

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 765 (2007) 130

Inclusive-jet and dijet cross sections in deep

N. Pavel , A.G. Yages Molina

S. Antonelli, P. Antonioli, G. Bari, M. Basile, L. Bellagamba, M. Bindi,

G. Aghuzumtsyan, D. Bartsch, I. Brock, S. Goers, H. Hartmann,

N.H. Brook, G.P. Heath, J.D. Morris, T. Namsoo

0550-3213/$ see front matter 2006 Published by Elsevier B.V.

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

M. Capua, S. Fazio, A. Mastroberardino, M. Schioppa, G. Susinno,

J.Y. Kim 6 , K.J. Ma 7

Z.A. Ibrahim, B. Kamaluddin, W.A.T. Wan Abdullah

Y. Ning, Z. Ren, F. Sciulli

J. Chwastowski, A. Eskreys, J. Figiel, A. Galas, M. Gil, K. Olkiewicz,

L. Adamczyk, T. Bod, I. Grabowska-Bod, D. Kisielewska, J. ukasik,

V. Adler, U. Behrens, I. Bloch, A. Bonato, K. Borras, N. Coppola,

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

G. Barbagli, E. Gallo , P.G. Pelfer

A. Bamberger, D. Dobur, F. Karstens, N.N. Vlasov 14

P.J. Bussey, A.T. Doyle, W. Dunne, J. Ferrando, D.H. Saxon,

T. Gosau, U. Holm, R. Klanner, E. Lohrmann, H. Salehi, P. Schleper,

C. Foudas, C. Fry, K.R. Long, A.D. Tapper

M. Kataoka 16 , T. Matsumoto, K. Nagano, K. Tokushuku 17 , S. Yamada,

A.N. Barakbaev, E.G. Boos, A. Dossanov, N.S. Pokrovskiy,

F. Barreiro, C. Glasman 18 , M. Jimenez, L. Labarga, J. del Peso, E. Ron,

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

F. Corriveau, C. Liu, R. Walsh, C. Zhou

A. Antonov, B.A. Dolgoshein, I. Rubinsky, V. Sosnovtsev, A. Stifutkin,

I. Abt, C. Bttner, A. Caldwell, D. Kollar, W.B. Schmidke, J. Sutiak

G. Grigorescu, A. Keramidas, E. Koffeman, P. Kooijman, A. Pellegrino,

N. Brmmer, B. Bylsma, L.S. Durkin, A. Lee, T.Y. Ling

P.D. Allfrey, M.A. Bell, A.M. Cooper-Sarkar, A. Cottrell,

P. Bellan, A. Bertolin, R. Brugnera, R. Carlin, R. Ciesielski, F. Dal Corso,

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

B.Y. Oh, A. Raval, J. Ukleja 23 , J.J. Whitmore

G. DAgostini, G. Marini, A. Nigro

J.E. Cole, J.C. Hart

H. Abramowicz 24 , A. Gabareen, R. Ingbir, S. Kananov, A. Levy

R. Hori, S. Kagawa 25 , S. Shimizu, T. Tawara

R. Hamatsu, H. Kaji, S. Kitamura 26 , O. Ota, Y.D. Ri

M.I. Ferrero, V. Monaco, R. Sacchi, A. Solano

S. Fourletov, J.F. Martin

S.K. Boutle 15 , J.M. Butterworth, R. Hall-Wilton 20 , T.W. Jones,

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

B. Brzozowska, J. Ciborowski 28 , G. Grzelak, P. Kulinski, P. uzniak 29 ,

Y. Eisenberg, I. Giller, D. Hochman, U. Karshon, M. Rosin

E. Brownson, T. Danielson, A. Everett, D. Kira, D.D. Reeder, P. Ryan,

S. Bhadra, C.D. Catterall, Y. Cui, G. Hartner, S. Menary, U. Noor,

E-mail address: gallo@mail.desy.de (E. Gallo).

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

(E)/E = 0.35/ E for hadrons, with E in GeV.

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

ET ,LAB > 3 GeV;

ET ,B > 12 GeV and ET ,B > 8 GeV.

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

Several sources of uncertainty in the theoretical predictions were considered:

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130

ZEUS Collaboration / Nuclear Physics B 765 (2007) 130