Nucl - Phys.B v.770

NUCLEAR PHYSICS B
Journal devoted to the experimental and theoretical study of the fundamental

constituents of matter and their interactions
Editorial Board
Supervisory Editors: G. ALTARELLI, Geneva, Switzerland; W. BARTEL, Hamburg, Germany; C. BECCHI, Genova,
Italy; R.H. DIJKGRAAF, Amsterdam, The Netherlands; D. KUTASOV, Chicago, IL, USA; L. MAIANI, Rome, Italy;
H. MURAYAMA, Berkeley, CA, USA; T. NAKADA, Geneva, Switzerland; H. OOGURI, CalTech, Pasadena, CA, USA;
H. SALEUR, USC, Los Angeles, CA, USA; A. SCHWIMMER, Rehovot, Israel
Associate Editors for Particle Physics: S.J. Brodsky, SLAC; M. Dine, UC, Santa Cruz; G. Dvali, New York; G.W.
Gibbons, Cambridge, UK; D. Gross, St. Barbara; S.S. Gubser, Princeton; L. Ibanez, Madrid; R.L. Jaffe, MIT; K. Kajantie,
Helsinki; M. Luescher, CERN; E. Rabinovici, Jerusalem; L. Randall, Princeton; A. Sen, Bombay; G. Steigman, Ohio
State; G. t Hooft, Utrecht; Y. Totsuka, KEK; P. van Baal, Leiden; S. Weinberg, Texas; J. Zinn-Justin, Saclay
Associate Editors for Field Theory and Statistical Systems: B. Duplantier, Saclay; A.W.W. Ludwig, Santa Barbara;
S. Lukyanov, Rutgers, Piscataway; T. Miwa, Kyoto; G. Mussardo, Trieste; M. Oshikawa, Tokyo; G. Parisi, Frascati;
A.M. Polyakov, Princeton; R. Sasaki, Kyoto; M. Zirnbauer, Cologne
Aims and Scope
Nuclear Physics B focuses on the domain of high energy physics and quantum field theory, and includes three main sections, i.e. particle physics, field theory and statistical systems and physical mathematics. The emphasis is on original
research papers. Conference Proceedings on the topics covered by Nuclear Physics B are published in the (separately
available) journal Nuclear Physics B-Proceedings Supplements.
Abstracted/Indexed in:
Chemical Abstracts/Current Contents: Physical, Chemical & Earth Sciences. Also covered in the abstract and citation
database SCOPUS . Full text available on ScienceDirect .
Subscription Information 2007
Volumes 760787 are scheduled for publication in 84 issues. Publication frequency: weekly. Subscription prices are
available upon request from the Publisher or from the Regional Sales Office nearest you or from this journals website
(http://www.elsevier.com/locate/nuclphysb). A combined subscription to Nuclear Physics A ISSN 0375-9474 (volumes
781797), Nuclear Physics B ISSN 0550-3213 (volumes 760787) and Nuclear Physics B Proceedings Supplements
ISSN 0920-5632 (volumes 164174) is available at a reduced rate. Subscriptions are accepted on a prepaid basis only
and are entered on a calendar year basis. Issues are sent by standard mail (surface within Europe, air delivery outside
Europe). Priority rates are available upon request. Claims for missing issues should be made within six months of the
date of dispatch.
Orders, claims, and journal enquiries: please contact the Customer Service Department at the Regional Sales Office
nearest you:
Orlando: Elsevier, Customer Service Department, 6277 Sea Harbor Drive, Orlando, FL 32887-4800, USA; phone: (+1)
(877) 8397126 [toll free number for US customers], or (+1) (407) 3454020 [customers outside US]; fax: (+1) (407)
3631354; e-mail: usjcs@elsevier.com
Amsterdam: Elsevier, Customer Service Department, PO Box 211, 1000 AE Amsterdam, The Netherlands; phone:
(+31) (20) 4853757; fax: (+31) (20) 4853432; e-mail: nlinfo-f@elsevier.com
Tokyo: Elsevier, Customer Service Department, 4F Higashi-Azabu, 1-Chome Bldg, 1-9-15 Higashi-Azabu, Minato-ku,
Tokyo 106-0044, Japan; phone: (+81) (3) 5561 5037; fax: (+81) (3) 5561 5047; e-mail: jp.info@elsevier.com
Singapore: Elsevier, Customer Service Department, 3 Killiney Road, #08-01 Winsland House I, Singapore 239519;
phone: (+65) 63490222; fax: (+65) 67331510; e-mail: asiainfo@elsevier.com
Advertising information
Europe, USA, Canada and ROW: Advertising orders and enquiries can be sent to: Miss Katrina Barton, phone: (+44)
(0) 20 7611 4117; fax: (+44) (0) 20 7611 4463; e-mail: commercialsales@elsevier.com.
South America: Advertising orders and enquiries can be sent to: Mr Tino DeCarlo, The Advertising Department,
Elsevier Inc., 360 Park Avenue South, New York, NY 10010-1710, USA; phone: (+1) (212) 633 3815; fax: (+1) (212)
633 3820; e-mail: t.decarlo@elsevier.com.
US mailing notice Nuclear Physics B (ISSN 0550-3213) is published weekly by Elsevier B.V., P.O. Box 211, 1000
AE Amsterdam, The Netherlands. The annual institutional subscription price in the USA is US$ 17015 (valid in North,
Central and South America), including air speed delivery. Periodical postage paid at Rahway NJ and additional mailing
offices. USA Postmaster: Send change of address: Nuclear Physics B, Elsevier, 6277 Sea Harbor Drive, Orlando, FL
32887-4800. Airfreight and mailing in the USA by Mercury International Limited, 365, Blair Road, Avenel, NJ 07001.
Nuclear Physics B 770 (2007) 146
Moduli stabilization in non-geometric backgrounds

Katrin Becker a , Melanie Becker a , Cumrun Vafa b,c , Johannes Walcher d,
a Texas A & M University, College Station, TX 77843, USA
b Harvard University, Cambridge, MA 02138, USA
c Center for Theoretical Physics, MIT, Cambridge, MA 02139, USA
d Institute for Advanced Study, Princeton, NJ 08540, USA
Received 24 November 2006; accepted 17 January 2007

Available online 15 February 2007
Abstract
Type II orientifolds based on LandauGinzburg models are used to describe moduli stabilization for
flux compactifications of type II theories from the world-sheet CFT point of view. We show that for certain
types of type IIB orientifolds which have no Khler moduli and are therefore intrinsically non-geometric, all
moduli can be explicitly stabilized in terms of fluxes. The resulting four-dimensional theories can describe
Minkowski as well as anti-de Sitter vacua. This construction provides the first string vacuum with all moduli
frozen and leading to a 4D Minkowski background.
2007 Elsevier B.V. All rights reserved.
1. Introduction
With the discovery of CalabiYau compactifications more than twenty years ago it became
evident that many aspects of the 4D theory can be traced back to the topology of the internal
manifold. It did not take long until backgrounds resembling the real world were constructed. At
the same time it became evident that string theory does not have a unique ground state because
the values of the moduli fields describing the deformations of the internal manifold could not be
determined. This has been an open problem for many years, not only for particle phenomenology
predictions coming from string theory, but also for string cosmology. This situation changed over
the past years, as it has been realized that flux compactifications of string theory can stabilize all
the moduli fields.
* Corresponding author.
E-mail address: walcher@ias.edu (J. Walcher).

0550-3213/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2007.01.034
K. Becker et al. / Nuclear Physics B 770 (2007) 146
Due to the incorporation of fluxes, the continuous choice of moduli parameters was restricted
to a large number of discrete choices. Thus this still left an extremely large number of string
vacua. These vacua are part of the string theory landscape, which at present is analyzed with
statistical methods [1] and techniques borrowed from number theory [3]. See [2] for a review.
All the more it is surprising that the number of explicit models known in the literature with
all geometric moduli stabilized is rather limited [4] and no models leading to four-dimensional
Minkowski space have been explicitly constructed. So far, moduli stabilization has been discussed in the literature in the supergravity approximation, a limit for which the radial modulus
is assumed to be large and the string coupling is small. In many cases the radial modulus is then
fixed in terms of non-perturbative corrections to the superpotential in a KKLT [5] like fashion,
leading to supersymmetric anti-de Sitter vacua. More recently, moduli stabilization in terms of
fluxes only (i.e. at the classical level) was achieved in [6] and in [7] in the context of type IIA
massive supergravity, where it was shown that fluxes stabilize all geometric moduli of a simple
T 6 /Z3 Z3 orientifold. In these models, the restrictions on the fluxes again result in a negative
cosmological constant.
One of the goals of this paper is to construct a set of simple models in which all moduli are explicitly stabilized by fluxes only and which have a vanishing 4D cosmological constant. We will
do so in the context of the type IIB theory, which allows the most freedom for dialing the fluxes.
It is known that in this theory fluxes generate a classical superpotential for the complex structure moduli. This is in contrast to the potential for Khler moduli, which is typically generated
through non-perturbative effects, which are less under control.
To avoid the complication of stabilizing Khler moduli, the basic idea underlying our work
is to start in the type IIB theory with a model which does not have any Khler moduli to begin
with, and stabilize the complex structure moduli and the dilaton by turning on appropriate RR
and NSNS fluxes. With ten-dimensional supergravity in mind, it appears quite hopeless to make
any progress with this idea. Indeed, in any geometric compactification with an ordinary manifold
M 6 as internal space, the overall size of that manifold will appear as a free parameter, a Khler
modulus. Thus, we will need M 6 to be non-geometric in one way or another. Thanks to string
theory, we know that such non-geometric models do exist. It might be expected that the flux superpotential stabilizes all moduli in such compactifications and that the resulting supersymmetric
vacua can be either Minkowski or AdS.
The fact that understanding string compactifications requires generalized notions of geometry
is well-appreciated. The best-known example is probably the correspondence between sigma
models on CalabiYau manifolds and an effective LandauGinzburg (LG) orbifold model as the
analytic continuation to small volume of the sigma model [810]. This correspondence also
plays a fundamental role in the understanding of mirror symmetry [11].
The existence of dualities has accentuated the relevance of string vacua without a tendimensional geometric interpretation. For instance, there are examples of CalabiYau manifolds
whose complex structure cannot be deformed. Mirror symmetry exchanges the complex structure with the Khler structure. Therefore, the mirror duals of such rigid manifolds would not
have Khler moduli and cannot correspond to a geometric manifold. Nevertheless, they have an
effective world-sheet description as LG models, in accord with the general ideas of [11].
When turning on fluxes, mirror symmetry as well as other dualities require an even broader
enlargement of the allowed class of compactification spaces. From the study of simple local
or toroidal models, it is well known that the mirror or T-duals of compactifications with generic
fluxes cannot be described by a conventional geometry, see e.g., [13]. More generally, by looking
at the panoply of RR and NSNS fluxes that are available in supergravity, and invoking (pertur-
bative and non-perturbative) dualities, one can argue that the most general flux compactification
will not allow a geometric description in any duality frame [14]. As mentioned above, the usual
ten-dimensional effective supergravity of string theory will not be useful for the study of such
vacua. Approaches which have been taken in the past include effective supergravity descriptions
in dimensions less than ten dimensions [1416], as well as exact world-sheet descriptions [14,
17,18].
In the present paper, we will use a combination of non-geometric world-sheet techniques
and 4D effective spacetime description to exhibit a simple class of models in which all moduli
can be stabilized by fluxes. Depending on the particular model, different values of the cosmological constant (Minkowski or AdS) are obtained. Charge conservation is accomplished by the
presence of orientifolds.
We will illustrate such a generic claim in a precise manner, by studying two explicit models
with Hodge numbers h11 (M) = 0 and h21 (M) = 84, 90 respectively. The underlying models
before turning on the fluxes are mirror duals of rigid CalabiYau manifolds and admit an effective
description as LG models [12]. At a particular point in moduli space, they are also equivalent to
some Gepner model [19].
Our models do not have a manifold interpretation, and therefore geometrical notions such as
cycles, differential forms, etc., do not have the conventional meaning. An appropriate description
of D-branes and supersymmetric cycles in LG models was developed in [20] using world-sheet
techniques. A great deal of information about D-branes and orientifolds in Gepner models and
their relation to LG and geometry is also available from Refs. [2123,25,26,59]. For convenience
we summarize the most important results in Section 2.
The first new aspect of our work is the description of fluxes in these models, which is done
in Section 3. The flux configurations we are interested in satisfy constraints coming from supersymmetry and the type IIB tadpole cancellation condition. Supersymmetric type IIB vacua
have fluxes belonging to the H 2,1 (M) H 0,3 (M) cohomology of the internal space M. We will
argue that these vacua are stable even non-perturbatively. This is due to the existence of a nonrenormalization theorem for the superpotential of [27]. This basically follows because there are
no relevant instantons to correct it. The existence of this non-renormalization theorem was crucial
for the works [28,29] where the classically generated superpotential (which is holographically
dual to the sum of the planar diagrams of the gauge theory) does not receive any corrections.
The fact that these results are in complete agreement with the exact non-perturbative dynamics
of supersymmetric theories is a powerful check on the non-renormalization of the superpotential. We will also discuss in Section 3 the other known consistency conditions of type IIB flux
compactifications, including flux quantization and the tadpole cancellation condition.
In Section 4, we will explicitly solve the supersymmetry equations which follow from the flux
superpotential and the tadpole cancellation condition for the simpler of our two models, related
to the so-called 19 Gepner model. We find supersymmetric Minkowski vacua for several types of
flux configurations.
It turns out that the spectrum of possibilities in the 19 model is quite constrained, and we have
not been able to find solutions leading to 4D AdS space in this model. One might ask whether
this has any significance or whether it is just an accident in this particular case. We address this
question in Section 5 by repeating the analysis for the so-called 26 Gepner model. We will see
that the range of possibilities is much larger, and that we can in particular find 4D AdS solutions.
We would like to point out that we have not been able to find solutions or sequences of
solutions in which the dilaton is stabilized at very small values, although we have no general
argument why this cannot be done. Because of the non-renormalization theorem for the super-
potential which we have mentioned above, having a string coupling of O(1) does not affect the
existence of the solutions. Nevertheless, it does mean that for other aspects of the solution, such
as the masses of the moduli, as well as introduction of supersymmetry breaking effects which
do receive quantum corrections, one should try to find a sequence of vacua which stabilize the
coupling constant at weaker values.
In Appendix A, we present an analysis of type IIB orientifolds of the quintic threefold from
the LG point of view. In particular, we discuss the computation of the O-plane charge in the LG
model, as well as the LG representation of the D3-brane sitting at a point of the quintic. The
orientifold actions which we study include exchanges of variables, and have not been treated
before in either the Gepner model or LG literature. This discussion, which is an application
of the general setup of [26], serves as background for our discussion of tadpole cancellation
condition in Section 3.4.
We discuss the implications of our results for studies of the string theory landscape in Section 6.
2. Branes and orientifolds in LandauGinzburg models
Due to their simplicity, LG models of string compactifications have been studied in great
detail over the years. They were among the first examples in which to access stringy geometry.
Moreover, as studied in [810], LG models can often be thought of as the analytical continuation
of CalabiYau sigma models to substringy volume.
The models for which this works most straightforwardly are CalabiYau hypersurfaces in
weighted projective space. Namely, they are given by the vanishing locus of a homogeneous
polynomial P (x1 , . . . , x5 ) = 0 in five variables of weight w1 , . . . , w5 , and of total degree H =

wi . Under the CY/LG correspondence, this sigma model corresponds to an N = (2, 2) LG
orbifold model with five chiral fields and world-sheet superpotential W = P . The orbifold group
is a ZH acting by phase rotations xi e2i/ hi xi , where hi = H /wi , which are assumed to be
integer. For a special choice of polynomial P , this LG model flows in the IR to a particular
CFT with a rational chiral algebra known as a Gepner model. These Gepner models are formally
defined as the tensor product of some number
r of N = 2 minimal models of level ki = hi 2,
such that the total central charge is c = (1 2/ hi ) = 3. Gepner models with a hypersurface
interpretation have r = 5, in which case the central charge condition is equivalent to the CY
condition on the degree of P .
The correspondence between LG models and CalabiYau sigma models can be extended to
the boundary sector, D-branes, and orientifolds. Boundary states in Gepner models were first constructed in [21], and their geometric interpretation was first addressed in [22]. In [20], A-branes
in LG models were shown to correspond to a particular type of non-compact cycle in the x-space
on which the superpotential W has a constant phase. More recently, B-branes in LG models
were studied in terms of matrix factorizations [30]. For the extension to orientifolds, we refer to
[23,25,26].
In the present paper, we will be concerned with a slightly different type of LG models which
do not have a manifold interpretation, as we have mentioned in the introduction. Nevertheless,
the techniques which are used to study LG models with such a connection can still be applied.
Let us now proceed to introduce this technology concretely in the relevant examples, referring to
the literature for the more general discussion.
2.1. The 19 /Z3 Gepner model

2.1.1. LandauGinzburg model
Our first example is a LG model M based on nine minimal models and world-sheet superpotential
W=
9
xi3 ,
i=1
divided by a Z3 generated by
g : xi xi ,
i = 1, 2, . . . , 9 and e2i/3 .
(2.1)
It is easy to compute the Hodge numbers of this model. As far as complex structure deformations
are concerned, they all come from deformations of W and a basis is given by the Z3 -invariant
monomials of the chiral ring C[x1 , . . . , x9 ]/(x12 , . . . , x92 ). In other words they are given by the
polynomials in the chiral fields
xi xj xk
with i = j = k = i,
(2.2)
and there are h2,1 (M) = 84 of them. To see that there are no Khler structure deformations,
we recall from [12] that ground states corresponding to the even cohomology always arise from
the twisted sectors in LG orbifolds. In a Z3 orbifold, there are only two non-trivial twisted sectors, and the first must contribute to h00 (M) = 1, while the second contributes in the conjugate
h33 (M) = 1. Hence h11 (M) = h22 (M) = 0. This reasoning also readily implies that there are no
contributions to h21 (M) from the twisted sectors. As a result the Hodge diamond is
1
0
0
1
0
0
84
0
0
84
0
0
1.
(2.3)
0
0
1
2.1.2. Geometric description
We can use the LG orbifold technology of [12] and the above-mentioned correspondence with
geometry to give alternative descriptions of the background. To illustrate this idea consider the
simpler example of the LG model based on the quotient of
W = x13 + x23 + x33 ,
(2.4)
by a Z3 action which sends (x1 , x2 , x3 ) (x1 , x2 , x3 ). This corresponds to a

torus model
at the Z3 symmetric point in both complex structure and Khler moduli space where
T2
and = .
(2.5)
This model has three Z3 symmetries that will be relevant for us. Two of them act geometrically
by phase rotations on the xi s, modulo the diagonal phase rotation which we have already divided
out. These Z3 s correspond to geometric symmetries of the T 2 . The third, somewhat less familiar,
Z3 is a so-called quantum symmetry [12], and is formally identified with the dual of the original
Z3 orbifold group defining our LG model,

quantum
= ZLG
Z3
= Z3 .
3
(2.6)
More concretely, the quantum symmetry multiplies a state in the lth twisted sector by l . Dividquantum
ing out by Z3
gives back the unorbifolded LG model described by the polynomial (2.4).
The model M of our interest is based on nine cubics instead of three, and it is natural to expect
a relation to geometry via (T 2 )3 = T 6 . The precise statement is that M is mirror to a certain rigid
CalabiYau Z which can be obtained as a quotient of T 6 by a Z3 Z3 action generated by

g 12 : (z1 , z2 , z3 ) z1 , 1 z2 , z3 ,

g 23 : (z1 , z2 , z3 ) z1 , z2 , 1 z3 .
(2.7)
Here z1 , z2 , z3 are the complex coordinates of T 6 = (T 2 )3 . This manifold has Hodge numbers
h11 (Z) = 84 and h21 (Z) = 0. Note that in order for g 12 , g 23 to be symmetries, we have to fix the
complex structure of the torus to be diagonal and equal to i = for i = 1, 2, 3. Being mirror
to Z, M can also be described as a toroidal orbifold, except that the orbifold group does not act
geometrically. To explain this in more detail, we state that M can be obtained by starting from
T 6 , where now we fix the Khler structure of T 6 at the Z3 orbifold point in each T 2 factor,
i = for i = 1, 2, 3, and divide out by a Z3 Z3 subgroup of the (Z3 )3 quantum symmetry
group. While preserving supersymmetry, this orbifold projects out all Khler moduli, so that we
end up with h11 (M) = 0. The complex structure moduli of the torus are projected, and new ones
appear in the twisted sector, for a total of h21 (M) = 84.
Alternatively, we can start with the 19 /Z3 Gepner model and turn it into a T 6 at the orbifold
point in Khler moduli space by modding out by a Z3 Z3 generated by
g1 : (x1 , x2 , x3 , x4 , x5 , x6 , x7 , x8 , x9 ) (x1 , x2 , x3 , x4 , x5 , x6 , x7 , x8 , x9 ),
g2 : (x1 , x2 , x3 , x4 , x5 , x6 , x7 , x8 , x9 ) (x1 , x2 , x3 , x4 , x5 , x6 , x7 , x8 , x9 ).
(2.8)
The quantum symmetry group (Z3 )3 of this W/(Z3 )3 LG orbifold model of (T 2 )3 is generated by
g1 , g2 and g3 , where g3 = gg11 g21 . The Z3 Z3 which turns T 6 back into M is then generated
by g1 (g2 )1 and g2 (g3 )1 , which are mirror duals of g 12 and g 23 in (2.7), respectively. Here, g
is the original orbifold generator in (2.1).
2.2. Orientifolds
2.2.1. Involutions
We intend to compactify type IIB string theory on an orientifold of M, which results in an
N = 1 theory in four dimensions. The orientifold is defined by dividing out B-type world-sheet
parity B dressed with a holomorphic involution such that the square of it is in the orbifold
group and such the superpotential is invariant up to a sign [23,26]
W ( x) = W (x).
(2.9)
This last condition comes from the fact that superpotential enters the world-sheet action as an
F-term,

d + d W (x),
(2.10)
Table 1
Number of invariant complex structure deformations for various orientifolds of M
Orientifold
h+
21
b3+
0
1
2
3
4
84
63
52
47
44
170
128
106
96
90
and B-type world-sheet parity exchanges the fermionic coordinates in superspace + . The
sign resulting from this parity transformation is compensated for by different types of involutions. The simplest involution that cancels the sign in (2.9) changes the sign of all nine bosonic
coordinates
0 : (x1 , x2 , . . . , x9 ) (x1 , x2 , . . . , x9 ).
(2.11)
Under this transformation the superpotential changes sign W (0 x) = W (x). Since we are already working with a Z3 orbifold by g in (2.1), we also have to divide out by the parity reversing
symmetries g0 and g 2 0 . The full orientifold group Z3 Z2
= Z6 is generated by g0 .
We can consider other orientifolds with a more non-trivial action on the xi s. In particular, we
can permute, respectively, 1, 2, 3, or 4 pairs of xi s:
1 : (x1 , x2 , x3 , x4 , x5 , x6 , x7 , x8 , x9 ) (x2 , x1 , x3 , x4 , x5 , x6 , x7 , x8 , x9 ),
4 : (x1 , x2 , x3 , x4 , x5 , x6 , x7 , x8 , x9 ) (x2 , x1 , x4 , x3 , x6 , x5 , x8 , x7 , x9 ).
(2.12)
One of the effects of the orientifold is to project the space of complex structure deformations
onto the subspace which is compatible with W ( x) = W (x). The number of invariant com+
+
plex structure deformations h+
21 , as well as invariant three-cycles, b3 = 2(h21 + 1) is tabulated in
Table 1. They can be obtained as follows. The 84 deformations of M correspond to the monomials xi xj xk with distinct i, j, k = 1, . . . , 9. A parity i acts on these monomials, leaving nfix (i ) of
them invariant up to the sign, and permuting the others pairwise. The number of (anti-)invariant
deformations is then given by
( ) 84 nfix (i )
i
h+
(2.13)
=
+ nfix (i ).
21 M
2
E.g., for 1 , invariant monomials are x1 x2 xj with j = 3, . . . , 9 as well as xi xj xk with distinct
i, j, k = 3, . . . , 9, so nfix (1 ) = 7 + 35 = 42 and h+
21 = 63.
2.2.2. Orientifold planes
One important piece of data of an orientifold in the geometric setting is the fixed point locus
of the involution that dresses world-sheet parity. The connected components of this fixed point
locus are referred to as orientifold planes (O-planes). O-planes carry RR charge and NSNS
tension. The crucial role of O-planes arises from the fact that both charge and tension can be
negative, while preserving spacetime supersymmetry. In fact, in compact models, O-planes are
necessary in order to cancel the tadpoles generated by D-branes and fluxes.
In our present setup, which is non-geometric, there is no straightforward notion of orientifold

plane as a geometric locus. Nevertheless, since the charge of O-planes can be detected on
the world-sheet by computing a crosscap diagram with RR field insertion, we can still ask in a
meaningful way for the RR charge sourced by the orientifold. We will present pertinent formulas
for these RR charges in Section 2.4.
In geometric type IIB orientifolds, the mutually supersymmetric O-planes that can occur
together are either O9- and O5-planes or O7- and O3-planes. In other words, the complex codimension of the fixed point locus of the dressing involution can be either 0 mod 2 or 1 mod 2. For
example, we can view the type I string as a type IIB orientifold with trivial involution dressing
world-sheet parity. This corresponds to the O9/O5-case, where O5-plane charge can be induced
from the curvature of the compactification manifold.
In our model, we also have a similar distinction between two types of orientifolds based on
the spacetime supersymmetry of the crosscap state resulting from the various involutions. For
instance, we can anticipate that the canonical action 0 corresponds to type I compactification on
M. Then, because they are of even codimension in the sense of having an even number of 1
eigenvalues, the orientifolds associated with 2 , and 4 are also of the O9/O5-type, and similarly
1 and 3 correspond to O7/O3-type. The latter is the class of orientifold models considered in
the supergravity regime in [33], and in which moduli can be stabilized by fluxes. In this section,
we will discuss all possible orientifolds of M. The fact that even in the non-geometric LG model
we study we find only two distinct types of O-planes is a strong indication that even in the nongeometric model, the intuition of the geometric case continues to hold.
2.3. D-branes and supersymmetric cycles
2.3.1. Supersymmetric cycles
Before we can describe the fluxes and the tadpole cancellation condition in our model, we need
to review some background material about the LG description of D-branes wrapping supersymmetric cycles. Generally speaking, there are two types of supersymmetric cycles in CalabiYau
type compactifications: A-cycles are middle-dimensional cycles represented by (special) Lagrangian submanifolds. They are relevant for flux compactifications as the cycles supporting
RR and NSNS three-form fluxes. B-cycles are even-dimensional cycles represented by holomorphic submanifolds carrying (stable) holomorphic vector bundles. They are the cycles that
can be wrapped by O-planes and D-branes of various types to, for instance, support standard
model gauge fields, cancel tadpoles from fluxes, etc.
In the non-geometric setting, the most useful way to talk about supersymmetric cycles is in
terms of D-branes. Namely, we can distinguish A- and B-branes by the boundary conditions on
the N = (2, 2) supercurrents on the world-sheet
A-type:
G
z) at z = z ,
L (z) = GR (
B-type:
G
z) at z = z .
L (z) = GR (
(2.14)
So let us review some of what is known about D-branes wrapping supersymmetric cycles in
LG models. There are generally two languages to describe the cycles. In [20], A-branes in LG
models were related to PicardLefshetz vanishing cycles of the singularity described by the LG
superpotential. In the simple homogeneous cases, these cycles can be pictured as wedges in the xplane, see e.g., Fig. 1. It is straightforward to implement orbifolding in this description. The other
language we will use is the more abstract language of matrix factorizations, originally introduced
Fig. 1. The piece of cake picture of Lagrangian (A-type) D-branes in LG model x 3 .
in [30]. These matrix factorizations describe B-cycles in LG models and their orbifolds and have
been studied extensively in the last few years.
Since A-model and B-model are mirror to each other, and the mirror of an LG model is an
orbifold of it, we can also combine the roles of these two descriptions. The wedge description is
easier to picture, while the matrix factorization approach is more general, in the sense that not all
matrix factorizations have a (known) mirror wedge description. Moreover, the O-plane charge is
in general only known how to compute in this language.
We will first describe these cycles in the parent LG model (including the orbifold, but before
orientifolding). Later, we will describe how the orientifold projects them and work out invariant
representatives.
2.3.2. A-branes
By studying the A-type world-sheet supersymmetry condition (2.14), one finds [20] that Atype D-branes in LG models correspond to middle-dimensional cycles. In the x-space D-branes
correspond to the pre-images of the positive real axis (or, by an R-rotation, some other straight
ray with constant slope) in the W -plane, i.e. to the pre-images of
Im(W ) = 0.
(2.15)
In the simplest case of LG models, based on the superpotential W = x k+2 (and its deformations),
this condition can be solved completely, and one can compare with RCFT results on boundary
states in N = 2 minimal models. Specifically, the condition (2.15) selects (k + 2) spokes through
the origin in the x-plane at an angle which is an integer multiple of e2i/(k+2) , and the branes
correspond to all possible wedges built from these spokes. A useful homology basis is provided
by the (k + 2) wedges V0 , . . . , Vk+1 of smallest angular size e2i/(k+2) .
For example, for the minimal model building block of the 19 Gepner model, each of the factors
comes with a set of three A-branes, see Fig. 1. We will call these three cycles V0 , V1 , and V2 .
The Vn s generate the homology of the minimal model, but satisfy the one relation
V0 + V1 + V2 = 0,
(2.16)
as can be seen from the figure.

On the other hand, the cohomology basis of the space of A-type D-brane charges in the minimal model is spanned by the two RR sector ground states, |l
, with l = 1, 2 [12]. These can be
10
equivalently represented by the chiral ring C[x]/x 2 = 1, x

. The correspondence is given by
|l
x l1 .
(2.17)
The RR charges of the Vn s can be computed as the overlaps [20] (disk one-point functions)

3
Vn |l
= x l1 ex dx = 1 l ln .
(2.18)
Vn
The normalization we have chosen here differs slightly from the ones of [20], but is more convenient for our purposes.
2.3.3. Intersection form of A-cycles
For later applications, it is useful to discuss the action of the Z3 symmetry, as well as the
intersection form on the A-cycles in the minimal model.
First of all, it is quite obvious that the Z3 symmetry which sends x x is represented on
the Vi s as

0 0 1
(V0 , V1 , V2 ) (V1 , V2 , V0 ) = (V0 , V1 , V2 )g with g = 1 0 0 .
(2.19)
0 1 0
Turning to the intersection form, this is best defined as the Witten index Tr(1)F in the Hilbert
space of open strings between two branes. Geometrically, there is clearly one RR ground state
localized at each geometric intersection point. In the more abstract setting such as the LG models
considered herein, we can still compute Tr(1)F as the cylinder amplitude for open strings
stretched between two branes with supersymmetric boundary conditions around the cylinder.
In the limit that the cylinder becomes very long, this amplitude factorizes on disk amplitudes
with closed string ground states inserted, thus providing an alternative (and often very simple)
derivation of D-brane charges from the Witten index. In the case at hand, the intersection product
between Vm and Vn , which is defined as TrHmn (1)F , follows most easily from the relation to
soliton counting [20]. One finds

1
0 1
F
TrHmn (1) = id g = 1 1
(2.20)
0 .
0 1 1
The index theorem which expresses this in terms of the RR overlap (2.18) is explicitly
1
1
TrHmn (1)F =
(2.21)
Vm |l
l|Vn
,
3
1 l
l=1,2
where l|Vn
= Vn |l
.
Clearly, (2.20) does not have full rank, which is a reflection of the relation in homology (2.16).
We can truncate the Z3 representation and intersection matrix by passing to a basis of A-cycles,
given for instance by (V0 , V1 ). The Z3 generator takes the form

0 1
A=
,
(2.22)
1 1
while the intersection matrix is

1 0
I=
.
1 1
(2.23)
11
We now tensor together nine such minimal models and orbifold by g acting diagonally as in
(2.1). The orbifolding projects RR ground states and chiral ring to those states
|l
= |l1 , . . . , l9
with li = 1, 2 and
9
li = 0 mod 3,
(2.24)
i=1
and identifies brane states by summing over orbits. We will denote these branes as

1
[n] =
(2.25)
Vni +
Vni +1 +
Vni +2
with n = (n1 , . . . , n9 ),
3 i
i
i
where the ni s are taken mod 3. To see that the factor 1/ 3 on the RHS is the correct normalization factor of the boundary states, we look at the open string spectrum. Let us look in particular
at the intersection index Tr(1)F between [m] and [n] . In the parent (unorbifolded) LG model,
each of them has three pre-images rotated by g. It is clear that the intersection of the two branes
in the orbifold is given by looking at the intersection points of all these 9 pre-images. Now if we
rotate the pre-images of both branes simultaneously, the intersection does not change, trivially
because g is a symmetry. In other words, the intersection points related by rotating both preimages simultaneously are gauge equivalent. Therefore, to obtain the intersection in the orbifold,
we fix any one pre-image of [m] and look for intersections with all pre-images of [n] . Thus,
the intersection form on the [n] is given by [22]

2

(1 g)9 1 + g 9 + g 9 ,
(2.26)
or, restricted to those [n] with representatives with all ni = 0, 1,

2

I = I 9 1 + A9 + A9 .
(2.27)
The same result can also be expressed in a form similar to (2.21)

1
1

(2.28)
1 li nlml ,
1 li
8
l
3
1i
l

where n l = ni li and the sum is over all l with
li = 0 mod 3, cf. (2.24). Since we are
9
summing over 170 intermediate states in (2.28), the 2 29 -dimensional matrix I has rank 170.
As it turns out, when restricting I to the first 170 (in alphabetical order) of the [n] with ni = 0, 1,
I is invertible and has determinant 1. We have thus described an algorithm for finding a minimal
integral basis of A-cycles in our 19 LG model.
Inm =
2.3.4. B-branes and matrix factorization

The analog of (2.15) for B-type boundary condition (2.14) would naively impose the holomorphic condition W =0 at the boundary. Both conditions arise from the fact that the F-term
world-sheet interaction d 2 W is supersymmetric only up to partial integration, and picks up
a contribution from the boundary if one is present. In the case of A-branes, this boundary term
is eliminated by imposing the boundary condition (2.15). But for B-branes, restricting to W = 0
would not allow for a great diversity of boundary conditions. In that case one introduces additional boundary degrees of freedom and boundary interactions whose susy variation will cancel
the boundary term.
12
As it was shown in [30], one way of encoding these boundary interactions is in terms of matrix
factorization. Briefly, a matrix factorization of the world-sheet superpotential W is a block offdiagonal matrix Q with polynomial entries in the LG variables and satisfying the equation
Q2 = W id.
(2.29)
Physically, the boundary interactions encoded by Q can be viewed as an open string tachyon
configuration between space filling branes and anti-branes: The blocks of Q correspond to the
ChanPaton spaces of the branes, resp., the anti-branes. The off-diagonal blocks correspond to
the open string tachyon. The diagonal blocks could carry a gauge field configuration, which
however can be gauge away in the standard LG models.
For example, for the minimal model x 3 , there is essentially only one non-trivial factorization
3
x = x x 2 , with associated matrix factorization

0 x
Q=
(2.30)
,
Q2 = x 3 id.
x2 0
The spectrum of massless open strings between two such branes is computed as the cohomology
of the matrix factorizations acting on matrices with polynomial entries. For the simple example
(2.30) for instance, it is easy to see that there are exactly two such states,

1 0
0 1
Q,
= 0,
Q,
= 0.
(2.31)
0 1
x 0
We must refer to the original literature [30], and citations thereof, for more details about the
matrix factorization description of B-branes in minimal models. We will review the essentials
here in view of their applications in our models.
Now let us discuss B-branes in orbifolds. The simplest, and also useful, example is the Z3
orbifold of a single minimal model x 3 divided by the action x x. As we have mentioned,
this orbifolding is nothing but an implementation of mirror symmetry, so we should compare the
result with the wedge picture of A-branes in the (unorbifolded) model we have described above.
Since we are dealing with space filling branes, we seek a representation of the orbifold group on
their ChanPaton spaces. It is easy to see that Z3 -generator is represented on the matrix Q of
(2.30) by

n 1 0
,
=
(2.32)
0
where n = 0, 1, 2. This generator satisfies
Q(x) 1 = Q(x).
(2.33)
The factorization Q equipped with these three representations of Z3 corresponds via mirror symmetry of the minimal model with its orbifold precisely to the three A-type wedges Vn discussed
in the previous subsection. One can also work out the projection on the open strings and thereby
recover (2.20). The expressions (2.18) and (2.21) are then special cases of the general formulas
in [31].

For our purposes, we are interested in the orbifold of W = 9i=1 xi3 by a diagonal Z3 . Before
orbifolding, we have just one factorization, given by tensoring nine copies of Q in (2.30). In the
orbifold, this yields three different branes: We tensor together nine copies of in (2.32), with
naively 39 choices of representation, but clearly only the sum of ns (mod 3) matters. Well now
call these three branes n , n = 0, 1, 2. A different way to think of the n s is as A-branes in the
13
mirror W/Z83 , where we are dividing by all symmetries that leave the product of xi s invariant.
This orbifold action allows to align the wedges in all nine factors, so n can be identified with
Vn in one of the factors, e.g., the first one.
2.3.5. Minimal integral basis for B-cycles
From either one representation, we find that the intersection matrix of the n s is given by

0
81 81
J = 81
(2.34)
0
81 .
81 81
0
For example, the evaluation of the formulas in [31] reads
9

9
1
1 km
TrHm ,n (1)F =
1 k
kn 1 k .
k
9
3
(1 )
(2.35)
k=1,2
From this, we could also read off the overlaps of the B-type boundary states with Ramond ground
states |k
in the twisted sectors k = 1, 2,
9

n |k
= Str g k = 1 k kn ,
(2.36)
where g = 9 , cf., (2.18).
In distinction to the A-brane situation, we see that the n s do not contain a minimal integral
basis of the B-type charge lattice. For the later discussion of tadpole cancellation in the flux
models, we will however need to know the precise quantization condition, so it is necessary to
digress a little further on the construction of such a minimal basis.
In the context of matrix factorizations, it is by now well known how to construct the minimal
basis. Namely, one has to use factorizations which are not obtained as tensor products of minimal
model factorizations. E.g., the potential x13 + x23 has the factorization1

0
x 1 + x2
(12)
Q
=
(2.37)
,
x12 x1 x2 + x22
0
which is not the tensor product of minimal model Cardy states. Such a tensor product would be a
4 4 matrix. It is thus no surprise that Q(12) is also smaller as far as RR charges are concerned.
For instance, the diagonal Z3 is represented on Q(12) by a single copy of in (2.32) instead of
(12)
two. Thus, if we denote by n the branes obtained by tensoring together the Q(12) in (2.37)
with 7 copies of the Cardy brane (2.30), their charges are
8

n(12) |k
= 1 k kn ,
(2.38)
and their mutual intersections are

0
F
TrH (12) (12) (1) = 27
m ,n
27

27 27
0
27 .
27
0
(2.39)
This is still not minimal, but its clear how to proceed. We denote e.g., by (12)(34) the tensor
product of Q(12) with Q(34) and 5 copies of Q, and then with obvious further notation, we find
1 These factorizations are also known as permutation branes [36]. The mirror A-model description of these branes is
not known in the LG formulation.
14
the overlaps
7

n(12)(34) |k
= 1 k kn ,
6

|k
= 1 k kn ,
(12)(34)(56)
n
5

(12)(34)(56)(78)
|k
= 1 k kn .
n
0
The intersection matrix of the last set is 1
(2.40)
1
1
0 1
1 1 0
, yielding a minimal basis. It is also not hard
to express the charges of these branes built from (2.37) in terms of the standard n s. Using
(1 )1 = (1 2 )/3, we find by comparing (2.36) with (2.38)
[n ] [n+2 ]
,
n(12) =
3
(12)(34)
[n+2 ]
n
=
,
3
(12)(34)(56)
[n+2 ] [n+1 ]
n
=
,
9
(12)(34)(56)(78) [n+1 ]
n
=
.
9

(2.41)
2.4. Charges of O-planes

The charges of the O-planes2 associated with the orientifold actions i described in (2.12)
can be computed using the formula (5.37) in [26]. This formula says that, in the same basis of
charges in which D-brane charges are given by the formulas in [31], such as we have, e.g., used
them in (2.36), (2.38), (2.40), the charge of the O-plane, namely the overlap of the crosscap state
|C
with the Ramond ground state |k
, is given by3
C|k
=
9

(1 + i ),
(2.42)
i=1
where i are the eigenvalues of the element of the orientifold group which squares to g k (where
g is the generator of the orbifold group).
For example, let us consider the trivial orientifold action, (2.11). The orientifold group has
three elements which reverse world-sheet parity, namely 0 , g0 and g 2 0 . To determine C|k
with k = 1, we notice that (g 2 0 )2 = g and the eigenvalues of g 2 0 are (2 , . . . , 2 ) for

i = 1, . . . , 9. Thus, C|1
= (12 )9 = (1)9 = 0 |1
. Similarly, C|2
= (12 )9 =
0 |2
.
Next, we consider the orientifold action involving a single permutation of variables. The
eigenvalues of g 2 1 are (2 , 2 , . . . , 2 ), and C|1
= (1 + 2 )(1 2 )8 = 2 (1 )8 .
Instead of going through the computations for the remaining cases, let us simply quote the result for the topological classes of the O-planes Oi associated with i . In terms of the basis n
2 As we have mentioned before, rather than implying that there is a geometric locus which can be identified as an
O-plane, we here simply mean the abstract world-sheet concept.

3 This formula gives the contribution to the charge from the internal theory only. The spacetime contribution is universal and multiplies the formulas in [26] by 4. See the next section for further discussion.
15
(n = 0, 1, 2), we find
[O0 ] = [0 ],
[1 ] [2 ]
[O1 ] =
,
3
[0 ]
,
[O2 ] =
3
[1 ] [2 ]
[O3 ] =
,
9
[0 ]
[O4 ] =
(2.43)
.
9
Notice that these charges coincide with the charges of particular permutation branes (2.41).
However, this is an accident of having level 1 minimal models. For example, a similar statement
is not true on the quintic. The charge of the permutation brane associated with the factorization
x15 + x25 = (x1 + x2 )( ), although it owes its existence to the permutation x1 x2 , does not
coincide with the charge of the orientifold associated with that permutation. We discuss this
explicitly in Appendix A.2.
3. Fluxes in LandauGinzburg models
Because our models do not have a radial modulus, it will now be shown that all moduli of the
internal space can be stabilized in terms of fluxes. Our solution is exact, i.e. there are no perturbative or non-perturbative corrections in the string coupling constant. The reason is that our analysis
is based on two ingredients, the supersymmetry constraints following from the spacetime superpotential and the tadpole cancellation condition. As argued in this section, both equations are
exact.
3.1. Flux superpotential
Let us first discuss the situation in the geometric case, and then explain how it continues to
hold in the non-geometric LG case of interest to us. In the type IIB theory there are three-form
fluxes in the RR and NSNS sector, HRR and HNS respectively, that can be combined into a
complex three-form
G = HRR HNS .
(3.1)
+ ie
is the axion-dilaton combination. In the type IIB theory the fluxes generate
Here = C0
a spacetime superpotential for the complex structure moduli

W = G .
(3.2)
M
This superpotential was derived in [27] and can be obtained with two different arguments. First,
supersymmetry constrains the form of the allowed flux components. These constraints were derived for M-theory/F-theory on four-fold compactifications in [34,32] and the superpotential is
such that it reproduces these constraints. Similarly, the type IIB superpotential can be derived
from the supersymmetry constraints imposed on the fluxes in type IIB. The second method involves general arguments which relate the tensions of domain walls to the superpotential [27,35].
16
As we will elaborate in the next subsection this latter derivation of the superpotential can be used
to show that the superpotential is exact and does not receive any corrections, perturbative or
non-perturbative, beyond the tree level.
Unbroken supersymmetry demands
Di W = i W + i KW = 0.
(3.3)
Here i runs over the complex structure moduli and . Further K describes the Khler potentials
for the complex structure moduli and the dilaton-axion.

K = K za + K( ),
(3.4)

a
K( ) = log i( ) ,
K z = log i .
(3.5)
M
Demanding
1
D W =
( )
= 0,
G

Da W =
G a = 0,
(3.6)
where a is a basis of harmonic (2, 1) forms, leads to the conclusion

G = HRR HNS H 2,1 (M) H 0,3 (M).
(3.7)
Since for the first case the superpotential vanishes it corresponds to a Minkowski solution, while
the second option corresponds to AdS. We will restrict to supersymmetric vacua, so that our
analysis can be based purely on a solution to the supersymmetry constraints. Notice that the
situation here is different than for geometric models, where a (0, 3) component breaks supersymmetry due to the presence of the radial modulus.
3.2. Non-renormalization theorem
It is important for our subsequent analysis that this superpotential does not receive any perturbative nor non-perturbative corrections, neither in nor in the string coupling constant gs .
Since we are dealing with type IIB model, corrections are already summed up in the LG
model. We will thus focus on the potential gs corrections. This will guarantee that our solutions
are valid to all orders in perturbation theory and even non-perturbatively. We will argue that W
does not receive perturbative nor any non-perturbative corrections.4 The arguments for the nonrenormalization of W in the geometric case are already known and used in the literature. Here
we will elaborate them in detail as it is important to argue for its non-renormalization even in the
non-geometric case which is the case of interest for us.
First the geometric case: Consider type IIB D5-branes and suppose we have some HRR flux
turned on in the internal CalabiYau manifold. If we wrap D5-branes over a three-cycle in the
CalabiYau, and let it be a domain wall in spacetime, then the flux value jumps from one side
of the domain wall to the other. The BPS computation for the tension of this domain wall is by
definition the change in the value of the superpotential W as we go from one side to the other.
4 There are other similar arguments for the perturbative non-renormalization. Despite the subtlety that the space
time superpotential depends explicitly on the dilaton, this can be shown perturbatively using the type IIB R-invariance,
PecceiQuinn symmetry as well as SL(2, Z) invariance [37].
On the other hand the BPS tension of the D5-brane wrapped on a 3-cycle C is given by5

T = .
17
(3.8)
Since we have
W = T =
(3.9)
and HRR has changed by one unit over each 3-cycle that intersects C in the positive sense, this
implies that

W = HRR .
(3.10)
M
Similarly if we also have HNS and adapt the above argument to the NS 5-brane we have a similar
story (as can also be deduced by the S-duality of type IIB) yielding

W = G .
(3.11)
M
Thus the question of whether there are quantum corrections to this formula translates to the
question of whether there are corrections to the BPS tension of the D5-brane domain walls. It is
known that this is not renormalized. To see this first of all note that by T-duality in spacetime part
(viewing the 4d on T 4 ) this is related to the quantum correction in the tension of D3-branes. But
it is well known that the BPS tension of electrically charged states is exact at the tree level (in the
N = 2 terminology. This follows from the fact that string coupling constant is a hypermultiplet
whereas the tension of the D3-brane is determined by vector multiplet data, which do not interact
with hypermultiplet terms in holomorphic terms). By S-duality this leads to the above formula
for the tension of the NS 5-brane domain walls as well and also to its non-renormalization.
Another way to argue for non-renormalization is to note that the only branes that could have
corrected the tension of D5-branes, should have been Euclidean instantons which break half
the supersymmetry and have three-dimensional world-volume6 : They would have wrapped the
corresponding internal three-cycles of the CalabiYau. But, the type IIB has no Euclidean twobrane which preserves supersymmetry. Therefore there is no candidate instanton.
The non-renormalization theorem is crucial for us, because we will fix the coupling constant
at a value of order 1. The non-renormalization theorem for the superpotential has passed some
highly non-trivial checks: In the context of N = 1 holography studied in [28,29] this statement
was equivalent with the statement that the exact non-perturbative superpotential fixing the glueball vevs of the gauge theory can be computed exactly by considering the planar diagrams only,
which in turn is equivalent to saying that the superpotential, determined by the fluxes is exact at
the tree level (which automatically sums up the planar diagrams). Note that the 1/N corrections
5 This point will be made more precise below.
6 One may have also worried about potential correction to the tension of the D5-brane domain walls from the D-
instantons. This could not have done the job by itself, without wrapped internal D-brane instantons, because we know
that for pure D(1) instantons, which would have therefore been present in the decompactified limit (as it does not
depend on the internal moduli) there is no contribution due to higher supersymmetry.
18
to the N = 1 superpotential would translate directly to gs corrections and if there were such
corrections it would have ruined the exact results of [28,29]. Needless to say, the fact that this
reproduces exact non-perturbative results for gauge theories is a strong check for the validity of
the non-renormalization theorem of the superpotential.
Now we turn to the non-geometric LG case, which is the case of main interest for us in this
paper. In such cases before even considering the superpotential we first have to argue that the
corresponding HNS , HRR degrees of freedom exist! Since the internal CFT is not geometric, we
cannot identify HRR and HNS with three-forms in the internal theory. However the notion of
degree of the form can be replaced by the notion of the internal U (1) charge of the underlying
(2, 2) SCFT. We would like to argue that for each chiral field with charge (1, 1) field, which
in our notation, corresponds to the H 2,1 (M) elements, there exists complex HNS and HRR fields
(complexification corresponding to the H 1,2 (M) elements). In fact we can use the world-sheet
construction to write down the corresponding vertex operators which is most naturally done in
the Berkovits hybrid formalism [57,58,24]:

j
j
ij qLi qL qRi qR + c.c. HNS ,
j
ij qLi qR + c.c. HRR ,
(3.12)
i
where qL,R
denote the left/right supersymmetry generators.7
We can now formulate the non-renormalization of W along the lines of the arguments discussed in the geometric case. Since we have identified all the relevant objects in terms of the
internal CFT theory, we can apply it to the LG case. In particular the superpotential can be
viewed as such an object: The internal D-branes of the geometric case, fixing the superpotential
in the geometric case, can now be translated to the non-geometric case, simply by formulating
the objects in the CFT language. For example the notion of a D3-brane wrapping an internal
cycles clearly has a CFT description. This directly leads to the CFT definition for a D5-brane
(simply by extending the D-brane in 2 of the spatial directions). Moreover the lack of instantons
to correct the D3 brane tension, still holds as in the geometric case (the N = 2 BPS charges are
not renormalized, as evidenced by the absence of relevant geometric objects; also the notion of
the brane tension certainly makes sense and the jump of flux across the corresponding domain
wall can also be formulated). Similarly the non-renormalization of the superpotential in terms of
the lack of availability of suitable branes follows in a similar form. We can thus formulate all the
relevant ingredients for non-renormalization of W in the non-geometric case.
There is however, one point to consider in more detail: Note that there is no similar nonrenormalization argument for the Khler potential K. Therefore one may worry about the renor-
7 In this language the turning on of the auxiliary fields in the N = 2 supersymmetry multiplet is what is responsible for
the generation of the superpotential. Moreover the non-renormalization of the superpotential would then follow directly
from the non-renormalization of the prepotential of the N = 2 theory. This was in particular the description used in [56].
Basically the point is that if denotes a vector multiplet N = 2 superfield and giving vev v to its 2 components yields

d 4 F0 () =
d 2 v F0 ,
which leads to the above formula for W when we include all contributions. This view of the non-renormalization theorem
is nice in that it follows directly from the 4d data, without assuming anything about whether the internal theory is
geometric or not. In particular the non-renormalization of the prepotential F0 which is crucial for the exact computations
in the context of the N = 2 theories directly leads to the non-renormalization theorem for the case with fluxes.
19
malization of the criticality of the potential, namely the condition

Di W = 0
(3.13)
also involves the Khler potential K. First of all note that this issue does not exist in the case of
Minkowski solutions that we will consider because in that case W = 0 and thus Di W = i W .
However for our AdS type solutions we would need to argue about the non-renormalization of
Di W . This may sound impossible if K gets renormalized. However we now argue that this is
indeed the case.
First of all note that W is not a holomorphic function but a holomorphic section of a line bundle. The fact that instead of i W we have the covariantized Di W reflects this fact. In particular
when we mentioned that the tension of the domain wall does not get renormalized and wrote the
tension as an integral of the holomorphic three-form , this reflects the fact that W is a section
of a line bundle. In this case the line bundle corresponds to the rescaling of
.
(3.14)
Thus the worry would have been if the covariantization of derivative could receive quantum
corrections. The solution we have found for the flux extremization states that the flux G should
lie in the H 0,3 H 2,1 . Since the rescaling of does not affect this statement, even if the section
receives quantum corrections, and may affect how W is expressed as a function, it would not
affect the form of the solution we have found which is gauge invariant, i.e. invariant under the
rescaling of . Another way to say this is as follows: Suppose we find our solution at tree level
at some fixed value of fluxes. We can choose our coordinates of moduli ti such that the Khler
potential will have an expansion
K = ti ti + aij ti tj f (t, t),
(3.15)
where ti = ti = 0 is the solution. Quantum corrections to i K evaluated at ti = ti = 0 will affect

the solution only by terms which are purely holomorphic, i.e., corrections of the form
K = f (t) + c.c.
(3.16)
But this can be reabsorbed to the definition of the holomorhic section of W , i.e. W
exp(f (t))W will get rid of it, without affecting our solution.
3.3. Dirac quantization condition
Throughout this paper, our basic strategy for finding vacua is to start from the effective fourdimensional superpotential induced by the fluxes and then find its critical points as described
in the previous subsection. On top of that, we impose all known string consistency conditions
which are not captured by the four-dimensional supergravity description. An example for such a
condition is the tadpole cancellation condition, which crucially puts a bound on the total amount
of flux that can be turned on. We will discuss this condition in the next subsection.
Another condition which cannot be seen purely within supergravity is the Dirac quantization
condition for the fluxes. This constraint arises from the requirement that the quantum mechanics
for various brane probes charged under these fluxes be consistent.8 Flux quantization is notoriously delicate to analyze in topologically non-trivial backgrounds, and this is even more true in
8 It has been argued [43] that other consistency conditions such as the tadpole cancellation can also be seen from the
brane probe point of view.
20
the presence of orientifolds. One potential subtlety is related to the so-called FreedWitten anomaly [3840], whose full consequences in orientifolds has, to our knowledge, not been rigorously
worked out even in the supergravity regime [45] (but see [44]). Conceivably, one could translate
these constraints to the worldsheet and check whether they are satisfied in our non-geometric
models.
Another subtlety of flux quantization in orientifolds was pointed out in [41,42].9 To discuss
this, let us consider a manifold X together with an involution , by which we wish to dress
world-sheet parity to construct an orientifold model. One usually calls X the covering space
of the orientifold X/ . It can then happen that the quotient space X/ has cycles that are not
inherited from X (see [41] for examples). Indeed, consider a p-cycle C X which is mapped
to itself by , but meets the fixed point locus of in a lower-dimensional cycle. Then C/ is a
p-cycle of X/ which is represented in X by C/2.
The FreyPolchinski puzzle arises [41] when turning on a p-form flux F in the orientifold.
Naively,
one would require that the periods over any
p-cycle in X/ be integral. In particular

F
Z.
This
means
that
in
the
covering
space,
C/
C F is an even integer. On the other hand,
orientifolding can be viewed simply as gauging world-sheet parity dressed by , and this point
of view only requires that F be integral on X and invariant under (or anti-invariant, depending
on the intrinsic parity of F ).
The conundrum was resolved in [41] in favor of the covering space point of view. Namely,
at least for a single cycle C, one can project any (even or odd) integral flux configuration. The
naive lack of integrality in the quotient space is repaired by noticing that the fixed loci of ,
the
Those discrete fluxes contribute to
O-planes, carry discrete versions of the p-form fluxes.
F
,
and
make
it
integral
independent
of
the
parity
of
C/
C F.
While this argument appears to work for a single cycle at a time, there are mutual consistency
conditions between different cycles. (Because the discrete fluxes at the O-planes require an overall choice.) We suspect that this condition must appear in the covering space as an obstruction to
choosing a gauge such that the gauge potential of F is invariant under , and not just the flux F
itself.
It would be important to understand this better, however we believe that this subtlety does
not affect our results: As far as discrete NSNS fluxes are concerned, we can argue from the
world-sheet perspective. Discrete NSNS fluxes correspond to discrete choices in the orientifold
action on the NSNS sector. But as is evident from the analysis in Section 2, our orientifolds do
not admit such discrete choices. Therefore, following [41], fluxes can have either parity in the
covering space, and there can be no consistency condition which exist when such choices are
possible. It is natural to believe the same holds for RR flux (which would also be natural from
the viewpoint of S-duality).
To summarize, we will simply impose the Dirac quantization condition on the fluxes in the
covering space of the orientifold. Namely, we require that the integral of RamondRamond and
NSNS flux through any three-cycle be integer,

HRR Z,
HNS Z,
(3.17)
9 It is possible that this condition is subsumed in the complete analysis of the FreedWitten anomaly for orientifolds.
We discuss it here as if it were independent.
21
for any H3 (M; Z). (We work with a normalization for the fluxes in which the periods are
directly integer.) The compatibility with the orientifold is simply the invariance condition

HRR =
HRR ,
HNS =
HNS .
(3.18)
( )
( )
3.4. Tadpole cancellation condition

Geometrically, the tadpole cancellation condition in type IIB reads10

HRR HNS + ND3 = Q3 (O-plane),
(3.19)
where the first term is the contribution of (integrally normalized) RR and NSNS three-form
fluxes, ND3 is the number of wandering D3-branes in the geometry, and Q3 (O-plane) is the
D3-brane charge of the orientifold plane(s). All charges are measured in the covering space, in
which a single D3-brane contributes one unit. So e.g., for the T-dual of type I compactification,
there are 64 O3-planes with total three-brane charge 32 in our units.
Our goal is to derive the CFT equivalent of the tadpole cancellation condition. Note that in
our non-geometric model M, we cannot readily evaluate equation (3.19). The reason is that while
we know explicitly the charge of the O-plane in terms of the LG basis of B-branes n , or the
overlaps with the closed string RR ground states, we do not know which one of these charges
should we identify with a D3-brane.
The LG analogue of (3.19) can be obtained by looking at models that are continuously connected with geometry. In the geometry, we can identify the charges appearing in (3.19) in the
large radius limit, and phrase them in CFT language. An important property of the tadpole cancellation condition is that it is a topological condition and hence does not depend on the moduli
we vary to reach the LG point. The tadpole cancellation condition will therefore take the same
form, no matter at what point in the moduli space it is phrased.
For some simple cases, the comparison between Gepner model orientifolds and geometry was
successfully done in [25]. We review this comparison and extend the check to orientifolds of the
quintic involving permutations in Appendix A.4. This will be a useful check on the methods used
in this section to obtain the tadpole cancellation condition of our non-geometric model in CFT
language.
3.4.1. Application to the non-geometric torus orbifold
As mentioned in Section 2, we can view the non-geometric LG/Gepner model 19 as a Z3 Z3
orbifold of T 6 = (T 2 )3 . The idea to identify the tadpole contribution due to fluxes in the nongeometric model is to first identify this contribution in T 6 . This we can do throughout the moduli
space because, being topological, it is locally constant over the moduli space and we know what
it is at large volume, where it can be expressed in supergravity. We then translate this knowledge
into a LG language. At this point, we forget that there ever was a geometric interpretation, and
10 A more rigorous description of the type IIB tadpole cancellation condition can be obtained in terms of twisted
K-theory. Such a description is addressed in [45].
22
simply track the flux contribution to the tadpole as we orbifold the LG model for T 6 to obtain
the non-geometric model of our interest.11
The flux contribution to the D3-brane tadpole on T 6 is simply that if we turn on one unit of
RR flux through cycle A and one unit of NSNS flux through cycle B, then the contribution to
the tadpole is precisely one unit of D3-brane charge for every intersection point. In other words
the RR charge generated by the fluxes is
[Flux] = (A B)[pt.],
(3.20)
where [pt.] is the class of a point on T 6 , where a D3-brane can sit. In this formula, we can give
a world-sheet interpretation to the intersection product A B, because it can be computed as
Tr(1)F in the open string sector between D-branes wrapped on the corresponding cycles.
To give a world-sheet (LG) interpretation to [pt.], we use the CalabiYau/LG correspondence
for branes, which we have reviewed in Appendix A. We start from the LG model for a single T 2 ,

W = x13 + x23 + x33 /Z3 .
(3.21)
Under the canonical CY/LG correspondence, the branes (0 , 1 , 2 ) (see Section 2) arise in
large volume by restricting to the elliptic curve the exceptional collection n (n) from the
ambient P2 (where is the cotangent bundle of P2 ). It is simple to compute the large volume
charges of these bundles. In terms of their Chern classes,

1 2 1
chi (n ) = Bin =
(3.22)
.
0 1 1
In words, 0 corresponds to a D2-brane wrapping the whole T 2 , 1 corresponds to a bound
state of 2 anti-D2-branes and 3 D0-branes, and 2 to a bound state of one D2-brane and 3 antiD0s. Here, the factor of 3 comes from the fact that the (hyper-)plane class H of P2 intersects the
elliptic curve {x13 + x23 + x33 = 0} P2 in three points. The LG monodromy n n+1 when
acting on the large volume charges looks as

2 3
A=
.
(3.23)
1
1
As is by now familiar, the n are not a minimal charge basis and do not contain a D0-brane.
Such a minimal basis can be constructed using permutation branes. Specifically, the D0-brane,
which has charges ch(D0) = (0, 1/3)T in our large-volume basis, arises in the charge orbit

0 1 1
(1 A)1 B = 1
(3.24)
.
1
23
3
3
Going back to the LG model, we note that the LG charges of the n in the 13 model are (
e2i/3 , k = 1, 2)
3

n |k
= 1 k kn ,
(3.25)
while the charges of the set containing the point on T 2 are
2

[pt.]T 2 |k
= 1 k .
(3.26)
11 We point out that for this procedure to be successful, it is crucial that the T 6 /Z Z orbifold has no B-type charges
3
3
from the twisted sectors.
23
(Again, these are represented by permutation branes.)

Let us now take three copies of this LG model for T 2 . We get three exceptional collections
(j )
n , where j = 1, 2, 3 labels the T 2 factor. The point on T 6 is of course the tensor product of
points on the T 2 s, and so has charges
3

(j ) 2
[pt.]T 6 k (1) k (2) k (3) =
1 k
.
(3.27)
j =1
Here, k (j ) = 1, 2 label the appropriate RR ground states in the T 2 s. Now we forget the geometric
interpretation and state that every intersection point (measured by Tr(1)F ) between the cycle
through which we put NSNS flux and the cycle through which we put RR flux contributes in

(j )
the class 3j =1 (1 k )2 . We can then orbifold the T 6 by Z3 Z3 as described in Section 2.
(j )
This has the effect of projecting the k (j ) , n so that a single set remains. This can be identified
as the set n from Section 2.3.4. In the orbifold then, the tadpole contribution will be

6
[Flux]|k
= (A B) 1 k ,
(3.28)
which can also be written as
[1 ] [2 ]
(3.29)
.
9
This is to be compared with the charges of the orientifold planes (2.43). Namely, we conclude
from this analysis that the tadpole cancellation condition between O-plane and fluxes in the nongeometric model 19 is (assuming no background D-branes are present)

12 for orientifold action 1 ,
HNS HRR =
(3.30)
4 for orientifold action 3 .
[Flux] = (A B)
Here, we have taken the O-plane charges for the orientifold action involving one or three permutations, from (2.43). The additional factor of four is the contribution from the uncompactified
spacetime directions. (The general formula is 2d/2 for a d-dimensional spacetime, and evaluates to 32 for d = 10, and 4 for d = 4.)
4. Solutions
We are now ready to present some explicit examples in which all moduli are stabilized by
fluxes along the lines we have sketched in the introduction. Since the foregoing two sections
have been quite detailed and technical, we will begin by rewriting explicitly the equations that
we are to solve.
4.1. Summary of the conditions
We have seen that unbroken supersymmetry requires that given integral three-form fluxes HRR
and HNS the complex structure of M and the dilaton must be adjusted so that
G = HRR HNS H 2,1 (M) H 0,3 (M).
(4.1)
Fluxes which have a non-trivial component along the (0, 3) direction lead to AdS spaces, while
fluxes with only (2, 1) components gives rise to Minkowski space solutions. Except for some
24
brief comments, we will mostly be interested in choosing the complex structure, and trying to
find a dilaton and an integral flux which is supersymmetric for those values of the moduli. In this
interpretation, Eq. (4.1) are simply linear equations in the flux quantum numbers, and at first they
are rather easy to solve.
The problem becomes more interesting when we also impose the tadpole cancellation condition,

1
= 12 ND3 ,
GG
HRR HNS =
(4.2)

where 12 is the contribution from the orientifold plane for the orientifold action, 1 , on which
we concentrate from now on (we have not been able to find any solutions for, 3 ). Here ND3
the number of D3-branes that we might want to allow. The LHS of Eq. (4.2) is quadratic and
positive definite in the flux quantum numbers. Moreover, the fluxes being quantized leads to
a quantization of the LHS of (4.2), and it is at priori not clear whether the smallest quantum
consistent with (4.1) will be sufficiently small.
As we will see, the simplest solutions of (4.1) do in fact not satisfy (4.2). We will nevertheless
present this simplest ansatz first and then improve on it, eventually exhibiting a supersymmetric
flux satisfying all requirements.
4.2. Ansatz
Recall that we have introduced an integral basis {n } of the lattice of A-cycles which are labeled by the first 170 non-negative integers in binary notation with 9 digits, n = (n1 , n2 , . . . , n9 ),
ni = 0, 1. We can then introduce a set of three-forms n which are Poincar dual to the n , i.e.,

n = m n = Imn ,
(4.3)
m
where Imn is the intersection form (2.27). For convenience, we also introduce a dual basis n
of three-forms, defined by the condition

n
n = m
.
(4.4)
m

Clearly, m = Imn n where Imn is the inverse of Imn , and m n = Imn .
Also recall the LG description of H 3 (M) according to which harmonic forms are represented
by RR sector ground states which are labeled by a set of nine integers
l = |l1 , . . . , l9
with li = 1, 2 and
9
li = 0 mod 3.
(4.5)
i=1
H 3 (M)
The Hodge decomposition of

at the Fermat point is displayed in Table 2. The pairing
between homology and cohomology is given by

l = Bl nl with n l =
ni li .
(4.6)
n
Here Bl is an l-dependent constant which will eventually drop out of our equations.
25
Table 2
LandauGinzburg representation of H 3 (M) at the Fermat point

Spans
li
H 0,3
H 1,2
H 2,1
H 3,0
18
15
12
9
There are now two ways to parameterize the general solution to (4.1) (which, again, we view
as an equation for the flux, fixing the complex structure at the Fermat point). One is to start from
the integral ansatz

N n n
M n n ,
G = HRR HNS =
(4.7)
and then impose

G l = 0 for all l with
li = 12, 18
(4.8)
as a constraint on the flux quantum numbers N n , M m . The alternative is to start from an ansatz

Al l ,
G=
(4.9)
li =12,18
and then adjust the coefficients Al in such a way that G has all integral periods

G = Nn Mn .
(4.10)
The two parameterizations are clearly related, by N n = Inm Nn and M m = Imn Mm (where all
N s and Ms are integer).
Even more explicitly, using (4.6) the conditions (4.8) reduce to

N n M n nl = 0 where
li = 12, 18.
(4.11)
n
Note that anyone of these equations implies that is of the form (this also follows alternatively
from (4.10))
a + b
(4.12)
,
c + d
where a, b, c, d are integers. As a result the value of is constrained. As we will see below,
solutions = (where is a third root of unity), which corresponds to one of the cusps in the
fundamental domain of the torus, can be explicitly constructed.12
Finally, we should also ensure that our flux is invariant under the orientifold action. To implement this, we study the action of i on our basis n . (We should restrict from now on to i = 1, 3,
since only in that case can fluxes be supersymmetric with respect to the orientifold plane at all.)
By using i (n ) = i (n) and our choice of basis explained in Section 2.3.2, it is not hard to find
=
12 Solutions in which = i, which would correspond to another fixed point of the fundamental domain, are not allowed
since they cannot be written in the form (4.12).
26
the expansion
i (n ) = Sin m m ,
(4.13)
where Si is a 170 170-dimensional matrix. The invariance condition on the flux quantum numbers in (4.7) can then be written as
N m Si n m = N m
M n (Si )n m = M m .
(4.14)
These equations are easy to solve over the integers and allow to rewrite equation (4.11) in terms
of 2h+
21 + 2 = 128,96 independent flux quantum numbers for 1 and 3 , respectively.
It is similarly simple to impose invariance under the orientifold on the ansatz (4.9). In either
way it turns out that the invariance condition on the fluxes is not a severe restriction on the
spectrum of possible solutions.
4.3. One flux component
A simple solution to the supersymmetry constraints (which, however, does not satisfy the tadpole cancellation condition) is provided by a flux proportional to the component corresponding
to the RR ground state with |l
= |222222222
. In this case

ni mod 3.
= A2|n| where |n| =
(4.15)
i
Here |n| takes three different values, and A is some constant. Taking into account that 1 + +
2 = 0 it turns out that the flux numbers are determined by four integers only, which we denote
by N0 , N1 , M0 and M1 , where the index on the flux numbers denotes the value of |n|. These
integers are constrained to satisfy the determining equations for the dilaton and the parameter A
=
N0 N1
M0 M1
and A = N1 M1 .
(4.16)
It is not hard to find that the contribution of this flux configuration to the tadpole is given by

HNS HRR = Mn Inm Nm = 27(N1 M0 N0 M1 ).
(4.17)
For integer values of M0 , N0 , M1 , N1 , this result (4.17) is clearly in excess of the tadpole contribution from the O-plane (4.2), but this will be improved upon shortly.
It is an instructive check to derive the result (4.17) in a different basis of three-cycles, called
the homogeneous basis. This basis is spanned by the cycles dual to the RR sector ground states
according to

l = ll .
(4.18)
Cl
Note that under complex conjugation the set of integers characterizing a RR sector ground state
transforms by interchanging li = 1 and li = 2. As a result we define the index li = 1 + li mod 2
and have

(4.19)
l = ll .
Cl
Indices are raised and lowered, again, with the help of the intersection matrix

Jll = l l = l,l ,
27
(4.20)
which turns out to be diagonal. Here = i27 3 is a normalization constant. This expression is
useful since it provides an alternative derivation of the intersection matrix Imn after transforming
back to the basis spanning the integral lattice using (4.6). We now apply the Riemann bilinear
identity and obtain

G G = Jll
(4.21)
G G.
Cl
C l
In case that one flux component in the (0, 3) direction is turned on, i.e. if G = A this yields

1
= 27(N1 M0 N0 M1 ),
GG
HNS HRR =
(4.22)

where we have used the result for and the quantization condition for A in Eq. (4.16). Here,
then, is an alternative derivation of (4.17). The homogeneous basis is practical since it results in
a diagonal intersection matrix. However, flux quantization becomes cumbersome in this basis.
Note that the same line of reasoning shows that a minimal non-trivial contribution to the
tadpole of 27 is present each time a single component in any of the (2, 1) directions is present.
Thus no flux along a single direction in H 2,1 (M) H 0,3 (M) provides a solution of the tadpole
cancellation condition. It remains to show that the combination of several flux components will
reduce the minimal non-trivial value of the tadpole. We do this in the next subsection.
4.4. The general supersymmetric flux
One may attempt to improve on the previous flux configuration, with smallest tadpole contribution of 27, by turning on some (small) number of supersymmetric flux components in the
homogeneous ansatz (4.9),
G=
N
(4.23)
Ai l(i) .
i=1
As we have noted, we need to make the flux invariant under the orientifold. In other words, we
+
+
have only b3+ /2 = h+
21 + 1 independent fluxes we can turn on, where b3 and h21 are tabulated
in Table 1. We denote the number of components with N . As a result the flux spans an hyperplane. We are interested in the sublattice created from the intersection of this hyperplane with
the integral lattice given by H 3 (M, Z), i.e.

2,1
H (M) H 0,3 (M) H 3 (M, Z).
(4.24)
Note that the contribution to the tadpole can be succinctly written in the form ( i37/2 )
1

=
GG
N

|Ai |2 ,

i=1
(4.25)
28
where the coefficients Ai have to be chosen so that the flux is integrally quantized. In order to
impose integrality we note that (4.8) and (4.23) implies
N

N n M n n ,
(4.26)
which becomes

G=
Aj mj = Nm Mm
(4.27)
Ai l(i) =
i=1
after integrating over the integral basis. Here m = (m1 , . . .) = (n l(1) , . . .), i.e., mj = 0, 1, 2.
Note that by squaring the two sides (4.26) one obtains
N

|Ai |2 = Nn Inm Mm .

(4.28)
i=1
As a result once (4.27) is imposed the contribution to the tadpole given by (4.25) will always
be integral. However, due to (4.11) not all the flux quanta are independent. Consequently even
though the entries of the intersection matrix Imn are 1 the flux numbers will appear on the righthand side of (4.28) with a certain multiplicity. This multiplicity is the origin of the factor 27 on
the right-hand side of (4.22).
Since the D3-brane charge originating from the orientifold plane is 12 the minimal non-trivial
contribution of the three-form fluxes can maximally be 12 in order to lead to a vanishing total
tadpole. Achieving this while at the same time satisfying the set of Diophantine equations (4.27)
is highly non-trivial and the existence of solutions is not a priori guaranteed. Below we show the
existence of solutions by presenting a set of explicit examples.
4.5. Some sample solutions
We have seen that the simplest supersymmetric flux, G makes a minimal contribution
to the tadpole of 27. It is not hard to see that by turning on 2 flux components (N = 2 in the
notation of the previous subsection), we can reduce this value to 18, which however is still too
large. Turning on more components makes the equations increasingly cumbersome and it is not
easy to find the general integral solution by working with the ansatz (4.9). One may instead
attempt to work with the integral ansatz (4.7), although this has the disadvantage that the tadpole
contribution is far less controlled.
In any event, after a tedious but in the end serendipitous search, we have found solutions satisfying all requirements, including the tadpole cancellation condition. Namely, we have found
supersymmetric flux configurations which are invariant under the orientifold action 1 and make
a tadpole contribution of 12 or 8. (We have not found any solutions consistent with the orientifold 3 .)
Let us write down explicitly three examples of solutions we have found, all corresponding to
an axio-dilaton combination
= ,
(4.29)
resulting in a string coupling constant, gs = 2/ 3. Let us write out these solutions in both the
integral basis and in the homogeneous basis. Namely, the flux
29
1
HRR
= 000010101 + 000010110 + 000011001 000011010 + 000100101
000100110 000101001 + 000101010 + 001000101 001000110

001001001 + 001001010 + 001100101 001100110 001101001
+ 001101010 + 001110101 001110110 001111001 + 001111010 ,
1
HNS
= +000000101 000000110 000001001 + 000001010 + 000010101

000010110 000011001 + 000011010 + 000110101 000110110
000111001 + 000111010 + 001010101 001010110 001011001
+ 001011010 001100101 + 001100110 + 001101001 001101010 ,

1
1
1
G1 = HRR
HNS
= (111122121 111122112 111121221 + 111121212 )
3
has tadpole

1
1
HNS
= 12,
HRR
(4.30)
(4.31)
the configuration
2
= 010010101 + 010010110 + 010011001 010011010 + 010100101
HRR
010100110 010101001 + 010101010 ,

2
HNS
= 000010101 + 000010110 + 000011001 000011010 + 000100101

000100110 000101001 + 000101010 ,
2
2
HNS
G = HRR
1
(111212121 + 111212112 + 111211221
=
3(1 )
111211212 + 111122121 111122112 111121221 + 111121212 )
2
contributes

2
2
HRR
HNS
= 8,
(4.32)
(4.33)
and finally, for

3
= +000000001 + 010000001 + 100000001 + 001000001 111000001
HRR
000010000 010010000 100010000 001010000 + 111010000 ,

3
HNS
= 010000001 100000001 110000001 001000001 011000001

101000001 + 010010000 + 100010000 + 110010000 + 001010000
+ 011010000 + 101010000 ,
3
HNS
1
=
(111222111 111221211 111221121
3(1 )
+ 111212112 + 111211212 + 111211121 111122211 111122121
3
HRR
111121221 + 111112212 + 111112122 + 111111222 )
(4.34)
30
one finds

3
3
HNS
= 12.
HRR
(4.35)
It is interesting that the point = that we have found for the axio-dilaton corresponds to
one of the cusps or fixed points in the fundamental domain of the torus.
As advertised before, these solutions correspond to Minkowski space, which follows from the
fact that the coefficient of 222222222 is zero in all solutions we have found.
5. The 26 Gepner model
We have seen in the previous section that it is rather hard to find supersymmetric flux configurations satisfying the tadpole cancellation condition in the 19 model, and in fact we have been
extremely lucky to find any solutions at all! From our perspective, the difficulty stems mainly
from the fact that the O-plane contribution to the tadpole is so small (12 or 4, depending on the
choice of involution). One naturally wonders why this is so. After all, our model is nothing but a
torus orbifold, and for the orientifold of T 6 in which world-sheet parity inverts all 6 torus directions, there are 64 O3-planes, with total charge 32. The reason we get something smaller in the
LG description can be traced back to the fact that the 13 Gepner model actually corresponds to a
T 2 with B-field B = 1/2. One can show that this forces some of the O3-planes to be exotic in
the sense that they have positive charge where regular O3-planes have negative charge. This reduces the contribution from the O-planes. However, this observation also indicates that it should
be easier to find solutions in a model related to T 6 with B = 0, for which tadpole canceling flux
configurations were for example discussed in [42]. Indeed, there is such a Gepner model, which
is the so-called 26 model. This is also a torus orbifold T 6 /Z4 Z4 (with zero B-field) with Hodge
numbers h11 = 0, h21 = 90. In this section, we will see that repeating the analysis for this model
has several payoffs.
5.1. The model
The so-called 26 model is best understood as emerging from the world-sheet superpotential
W=
6
xi4 + z2 ,
(5.1)
i=1
divided by a Z4 action
g : xi ixi ,
z z.
(5.2)
The extra z2 term in the superpotential might seem trivial and indeed it can be integrated out.
However, doing so, the orbifold action xi ixi has to be dressed by (1)F and this is somewhat
awkward to implement at the level of the branes. It is generally recommended13 to study LG
models with the number of fields congruent to c modulo 2.
Similarly to the 19 model, the 26 model is an orbifold T 6 /Z4 Z4 . So most of the previous
discussion carries over to the present case, however there are several subtleties associated with
the fact that the levels are now even. For example, we have new choices in the orientifold action.
13 Geometrically it is more natural to add three quadratic fields z2 .
i
31
Table 3
Number of invariant complex structure deformations for various orientifolds of the 26 model
Orientifold
h+
21
b3+
0
1
2
3
90
60
50
48
182
122
102
96
The canonical choice is

0 : (x1 , . . . , x6 ) (x1 , . . . , x6 ),
z iz,
(5.3)
where now = e2i/8 . The orientifold group is Z8 and fits into the sequence
Z4 Z8 Z2 .
(5.4)
The orientifold action can be dressed in various ways by symmetries. The restrictions are that
parity remain involutive up to the orbifold group and we count orientifold actions as equivalent
when they differ by conjugation by a symmetry. For example, we can have (suppressing z iz)
1 : (x1 , x2 , x3 , x4 , x5 , x6 ) (x1 , x2 , x3 , x4 , x5 , x6 ),
2 : (x1 , x2 , x3 , x4 , x5 , x6 ) (x1 , x2 , x3 , x4 , x5 , x6 ),
3 : (x1 , x2 , x3 , x4 , x5 , x6 ) (x1 , x2 , x3 , x4 , x5 , x6 ),
4 : (x1 , x2 , x3 , x4 , x5 , x6 ) (x1 , x2 , x3 , x4 , x5 , x6 ),
5 : (x1 , x2 , x3 , x4 , x5 , x6 ) (x1 , x2 , x3 , x4 , x5 , x6 ),
6 : (x1 , x2 , x3 , x4 , x5 , x6 ) (x1 , x2 , x3 , x4 , x5 , x6 ).
(5.5)
Note that 6 = g 2 1 , 5 = g 2 2 , 4 = g 2 3 , so these parities define the same orientifold. One

might also note that we did not have the similar option to dress parity action with a non-trivial
phase in the 19 model, where all levels are odd. In the present case, we can in addition to the
phase symmetries also consider permuting some of the variables, but these parities are always
equivalent by a change of variables to one of the i s, perhaps with a change of superpotential.
E.g., the action x1 x2 on x14 + x24 is equivalent to (x1 , x2 ) = (x1 x2 , x1 + x2 ) (x1 , x2 ),
with the superpotential x14 + 6x12 x22 + x24 . The projection of moduli is given in the Table 3 There
is one further option in the orientifold action which was not available for the 19 model, namely
the dressing by quantum symmetry. Recall that the quantum symmetry associated with the
Z4 orbifold is Z4
= Z4 and measures the twisted sector. Dressing parity by an element Z4
means that we multiply a state in the sector twisted by g Z4 by the phase (g). Any such
dressing is involutive, and those related modulo (Z4 )2 are equivalent. In upshot, we have one nontrivial dressing by quantum symmetry, and we will denote the corresponding orientifolds by i ,
i = 0, 1, 2, 3. The projection of complex structure moduli is unchanged, whereas the projection
of Khler parameters, if they were present, would be different. (See [25] for examples of such
situations.)
32
5.2. A-branes
In the x 4 minimal model, we have to divide the cake into 4 pieces, which we call V0 , V1 , V2 ,
V3 , satisfying the relation V0 + V1 + V2 + V3 = 0, and having intersection matrix
1
0
0 1
0
0
1 1
id g =
(5.6)
.
0 1 1
0
0
0 1 1
In the z2 factor, there are only two straight wedges, which only differ by orientation. Following
the same strategy as before in the 19 model, we obtain basic A-branes [n] in the 26 model,
where n = (n1 , . . . , n6 , nz ), ni ni mod 4, and nz nz mod 2. The orbifold equivalence is n
n + (1, 1, 1, 1, 1, 1, 1). The intersection matrix in the orbifold is

2
3

(1 g)6 (1 gz )6 1 + g 6 gz + g 6 gz + g 6 gz .
(5.7)
In practice, it is convenient to go to a truncated set by using the relations satisfied by the Vi s. In
the x 4 models, the symmetry generator and intersection matrix look, respectively,

0 0 1
1
0 0
A = 1 0 1 ,
I = 1 1 0 .
(5.8)
0 1 1
0 1 1
In the z2 model, we have, Iz = 1, Az = 1. Thus, the truncated intersection matrix of the 26
model is:

2
3

I = I 6 1 A6 + A6 A6 .
(5.9)
The formula in the closed string channel (cf., (2.21)) is

1
I[n],[m] = 5
1 i li i ni li mi li ,
4
(5.10)

where the sum is over l = (l1 , . . . , l6 ) with li = 1, 2, 3 with
li = 0 mod 4 (there are 182 of
them). The formula (5.10) can be understood, from the mirror symmetry construction using matrix factorizations, or from the wedge picture.
As in the 19 model, it turns out that the first (in alphabetical order) 182 [n] with ni = 0, 1, 2
form an integral basis of the charge lattice.
5.3. B-branes
The basic B-branes correspond to the tensor product of matrix factorizations x 4 = x x 3 , on
which the Z4 generator is represented by

1 0
=
.
(5.11)
0 i
In the z2 model, we only have the factorization z z with Z2 generator represented by

1 0
.
z =
0 1
(5.12)
33
We now tensor together and orbifold, which means choosing a representation of Z4 ,

g = i n 6 z .
(5.13)
We call the resulting branes n , n = 0, 1, 2, 3. There are two twisted Ramond ground states in
our model, from twisted sector k = 1 and k = 3. The brane charges are (cf., (2.36))
6

n |k
= Str g k = 2 1 i k .
(5.14)
One way to see that there is no Ramond ground state for k = 2 is that Str g 2 = 0 for any brane.
It is also easy to see that n is the anti-brane of n+2 , and consequentially 0 , 1 form a
(non-integral) basis, with intersection form

6

1
1
0 8
k 6 k(nm)
2 1 ik
2
1
i
i
=
.
Jnm =
(5.15)
8 0
4
2(1 i k )6
k=1,3
A minimal basis is provided by the maximal permutation branes with charges

3

|k
= 2 1 i k ,
(12)(34)(56)
n
(5.16)
which are related to the 0 , 1 basis by

(12)(34)(56)
[0 ] + [1 ]
,
4
(12)(34)(56)
[0 ] [1 ]
.
4
(5.17)
By following a similar analysis as for the 19 model, one can show that these charges (5.17) are
the charges corresponding to the point class in the non-geometric 26 model, and hence are the
normalization for the tadpole contribution from the fluxes.
5.4. O-plane charges
To get the charges of the O-planes associated with the various orientifold actions (5.5) as
well as their quantum symmetry twists, we turn again to Eq. (5.37) in [26]. There are now two
distinctions from the 19 model. First of all, we notice that for any given orbifold element g k
(k = 1, 3) with a ground state in the corresponding twisted sector, there are two parities which
square to it: If is one, then g 2 is the other. Thus, (5.37) becomes a sum of two terms.
To understand the possible dressing by quantum symmetry, we have to resolve the definition
of the phase c( ) from Eqs. (4.13), (4.14). Without quantum symmetry dressing, 1, the
phase is just an overall choice of sign of the O-plane. The non-trivial is defined by (g) = i,
and we get the values

c(g ) = ic( ),
c g 2 = c( ),
(5.18)
c g 3 = ic( ),
where c( ) = 1 , and = e2i/8 . (The sign is an overall choice, and well omit it.)
Now let us compute the O-plane charge associated with the canonical orientifold action 0 . As we said, for each orbifold element there are two parities which square to it.
E.g., for k = 1, the eigenvalues of 0 are (, , , , , , i), while those of g 2 0 are
(, , , , , , i). So we obtain

C|1
= (1 + i) (1 + )6 + (1 )6 = 56,

6
6
C|3
= (1 i) 1 + 1 + 1 1 = 56.
(5.19)
34
With dressing by quantum symmetry, we get

C|1
= i1 (1 + i) (1 + )6 (1 )6 = 40 40i,

6
6
C|3
= (i) i1 (1 i) 1 + 1 1 1 = 40 + 40i.
(5.20)
Continuing in this fashion, and expressing the charges in the 0 , 1 basis, we find the following
analog of (2.43)
7
[O0 ] = [1 ],
2
3
[O1 ] = [0 ],
2
[1 ]
[O2 ] =
,
2
[0 ]
[O3 ] =
,
2
[O 0 ] =

5
[0 ] + [1 ] ,
2
[O 1 ] = [0 ] + [1 ],
[O 2 ] =
[0 ] [1 ]
,
2
[O 3 ] = 0.
(5.21)
(5.22)
(5.23)
(5.24)
As before, these results should be multiplied by 4 to get the actual charge of the orientifold planes
in spacetime.
5.5. Simple ansatz
As we have done in the 19 model, it is a useful starting point to first study the tadpole con More
tribution of a flux with only a (0, 3) component turned on, i.e., G = HNS HRR .
precisely, we set G = Al0 where l0 = (3, 3, 3, 3, 3, 3) in the normalization in which the intersection form is given by (5.10), namely

l l = 45

1 i li = 213 i.
(5.25)
Imposing integrality on the fluxes means

G=A

1 i li i ni li = Nn Mn .
(5.26)
As in the 19 model, we can parameterize this solution in terms of just 4 integers, N0 , N1 , M0 ,

M1 . We find
=
N0 iN1
,
M0 iM1
A=
N1 M1
,
(1 + i)6
(5.27)
and as a result
1

(N1 M0 M1 N0 )2
|M0 iM1 |2
i213
2i(N0 M1 M0 N1 ) |(1 + i)6 |2 |M0 iM1 |2
= 64(N0 M1 M0 N1 ).
=
GG
(5.28)
35
5.6. Tadpole cancellation

When combining together (5.17), (5.21) and (5.28), we see that we ought to use the orientifolds which include twist by quantum symmetry in order to get O-plane charges in the direction
of a D3-brane, which we have identified with (5.17). Moreover, we should remember to multiply the results of (5.21) by 4 to take into account the spacetime contribution.
In this way, we obtain the following tadpole cancellation condition

HRR HNS = 40, 16, 8, 0,
(5.29)
respectively, for the four possible orientifolds. This Eq. (5.29) cannot be satisfied by the simple
ansatz and result (5.28) used above. But there are more complicated flux configurations which
do the job.
5.7. Sample solutions
We have made a search for supersymmetric flux configurations in the 26 model whose tadpole
contribution is within the bound imposed by the charge of at least one of the O-planes (5.29). As
anticipated, the spectrum of possibilities is wider than in the 19 model, due to the fact that the
O-plane contribution is larger. Nevertheless, most of the solutions are still quite complicated.
As an example, the configuration
1
HRR
= 000002 + 2000011 + 2000012 + 000020 + 2000021 000110 000112 000120
000122 000200 000210 + 000211 + 000221 001001 001002 001021

001022 002000 002001 + 002011 + 002012 ,
1
HNS
= 000001 + 000010 + 2000011 000012 000021 2000022 000100 000102

000110 000112 + 000201 + 000210 + 000211 + 000220 001000 001001
001020 001021 + 002001 + 002002 + 002010 + 002011 ,
1
1
G = HRR
HNS
1
121321 122311 + 123112 123211 + 131221 + 132112

2132211 + 211321 212311 + 213112 213211 + 221221 + 222112
2222211 + 231112 + 231121 231211 232111 + 311221 + 312112
2312211 + 321112 + 321121 321211 322111 ,
has a tadpole contribution of

1
1
HNS
= 40,
HRR
(5.30)
(5.31)
exactly saturating the tadpole from [O 0 ]. Since G1 does not have a 333333 component turned
on, it corresponds to a Minkowski space solution.
As another example, let us look at
2
HRR
= 000101 + 000102 + 000110 + 2000111 + 000112 + 000120 + 000121 001001
001002 001010 2001011 001012 001020 001021 ,
36

2
HNS
= 000100 + 000101 + 000110 000112 000121 000122 001000 001001
001010 + 001012 + 001021 + 001022 ,

2
2
HRR
G = HNS
2
(1 + i)322111 + (1 i)321211 + 2i313111 2i311311

(1 i)232111 + (1 i)231211 + 2i223111 2i221311 + (1 + i)213211
(1 + i)212311 + 2i133111 2i131311 + (1 + i)123211 (1 + i)122311 .
(5.32)
This configuration has a smaller tadpole,

2
2
HNS
= 16.
HRR
(5.33)
As a result, if we use it in conjunction with the orientifold [O 0 ], the flux will not completely
cancel the charge of the O-plane. This gives the freedom to include additional D-branes into
the background. It is an interesting open question to determine whether there exist D-branes
with the correct charge but without continuous moduli solving the tadpole cancelation condition.
However, since the solutions presented herein are at gs = O(1) the description of the properties
of these D3 branes would be difficult to control.
Finally, we have searched for solutions which have a (0, 3) component turned on. There are
several possibilities for such solutions giving rise to 4d AdS space, one of which is of the form
G3 4i3,3,3,3,3,3 2,2,2,2,1,1 ,
with
(5.34)

3
3
HRR
HNS
= 40.
(5.35)
(This is the only solution we do not write out in the integral basis, as it would take several pages.)
6. Discussion and conclusions
In this paper, we have studied moduli stabilization by fluxes in LG compactifications of
type IIB string theory. We have given both a world-sheet and a 4D effective description of fluxes
in these theories. The particular models considered are non-geometric (as they do not have any
Khler moduli h11 = 0) and can be represented by orientifolds of LG models. It has been shown
that the complex structure moduli can be stabilized in terms of fluxes only, while the tadpole
cancellation condition is satisfied due to the presence of the orientifold charge. The value of the
string coupling constant for our solutions is of the order of unity, so that our solutions are at
strong coupling and describe points in moduli space of enhanced symmetry [3]. This type of
vacua are of interest for model building. So for example, low energy theories like the MSSM
have a discrete R-symmetry that helps to explain the stability of the proton.
Since our solutions are at strong coupling, our analysis heavily relies on supersymmetry and
non-renormalization theorems. The particular vacua that have been found have N = 1 supersymmetry, so that only a non-vanishing H 0,3 and H 2,1 component of the flux or a linear combination
thereof is allowed. It has been shown that the classical superpotential of [27] is exact, so that our
solutions persist even non-perturbatively.
Among our main results is a set of examples of totally explicit flux configurations which are
supersymmetric, invariant under the orientifold, and satisfy the tadpole cancellation condition.
37
Technically, these fluxes are solutions to a large number ( 100) of linear Diophantine equations,
and a single positive definite quadratic inequality. This type of problem was used in [46] to
argue that the landscape of string vacua might be so complex from the computational complexity
point of view as to preclude finding and studying individual vacua explicitly. From this point of
view, it can appear surprising that we have found a solution in such a high-dimensional case. In
fact, in all studies of flux stabilization so far (outside of the statistical approach), the number of
moduli has been of order 1. Nevertheless, our findings need not be viewed as being at odds with
the arguments of Ref. [46], which rely on statements about the generic problem in this class.
Moreover, our problem clearly has additional symmetry properties such as all periods being cube
roots of unity. Although we have not crucially used these structures to find the solution, it is likely
that one could.
Our models provide the first explicit examples of flux compactifications with all moduli stabilized by fluxes only and which have an external Minkowski spacetime. The reason for this
is the absence of Khler moduli. All models constructed in the literature before, lead to AdS
spacetimes, which is the generic case in all geometric models.
One can ask whether the existence of 4D solutions of string theory with all moduli stabilized
and exactly vanishing cosmological constant should have been expected. In particular, is this
consistent with the concept of landscape naturalness?14 A possible resolution of the cosmological constant problem is via a dense but discrete distribution of stable and meta-stable vacua in
the landscape. If zero is not a special value, is it natural to find it on the list of allowed values?
Clearly the solution we have found adds a new angle on this question and it would be interesting
to study in more detail the distribution of solutions of the type discussed in this paper. In the context of supersymmetric vacua, vanishing superpotential leads to an unbroken R-symmetry, which
might make such vacua look more natural. Some work on vacua with unbroken R-symmetry has
been done in [3] and [47].
One possible extension of our work would be to deform the LG model away from the Fermat
point in the complex structure moduli space. There is one particularly interesting limit in the
moduli space, namely the mirror of the large radius limit of the corresponding rigid CalabiYau
manifolds. Indeed, in this limit, our models should be related by mirror symmetry to certain
type IIA vacua studied in [7], which found infinite families of AdS solutions with all geometrical
moduli stabilized. It would be interesting to recover and generalize these solutions in the type IIB
setup.
It would also be interesting to gain a better understanding of the microscopic description of
fluxes in non-geometric LG models. If this can be achieved, one could also address a worldsheet derivation of the tadpole cancellation condition and ultimately the derivation of a dual CFT
theory description of the KKLT-like AdS vacua appearing in the string theory landscape.
Acknowledgements
We would like to thank Aaron Bergman, Nathan Berkovits, Jacques Distler, Mike Douglas,
Dan Freed, Simeon Hellerman, Shamit Kachru, Joe Polchinski, Jessie Shelton and Wati Taylor
for valuable discussions and communications. We are all grateful to the organizers of the 2006
Simons Workshop in Mathematics and Physics, where much of this work was done. J.W. would
also like to thank the KITP for hospitality during the program on string phenomenology.
14 We thank S. Hellerman for raising the question and S. Giddings for a discussion on this issue.
38
The work of K.B. was supported by NSF grants PHY-0612842 and PHY-0555575. The work
of M.B. was supported by NSF grants PHY-0505757 and PHY-0555575. The work of C.V. was
supported in part by NSF grants PHY-0244821 and DMS-0244464. The work of J.W. was supported in part by the Roger Dashen Membership at the Institute for Advanced Study and by the
NSF under grant number PHY-0503584. This research was supported in part by the National
Science Foundation under Grant No. PHY99-07949.
Appendix A. Analytic continuation
The purpose of this appendix is to provide some background checks on the connection between LG orientifolds and the large volume CalabiYau manifolds, when it exists. In particular,
we wish to review the canonical identification of the D0-brane in the LG/Gepner model. We also
provide a non-trivial check of the formulas of [26], which we have used to compute the O-plane
charges in the non-geometric LG. Namely, we verify that the O-plane charges of the exchange
orientifolds of the quintic agree between the LG and CY geometric description. This has not so
far been available in the literature, and could be useful for other purposes as well.
A.1. Geometric interpretation of Cardy states in Gepner models
A general Gepner model connected with a hypersurface in weighted projective space has an
LG description with five factors15
W=
5
xihi ,
(A.1)
i=1

with 1 2/ hi = 3, modded out by a ZH symmetry, where H l.c.m.(hi ). The corresponding
hypersurface is X = {W = 0} P4w1 ,...,w5 , where wi = H / hi .
The LG description yields H basic B-branes in these models, which we will call n ,
n = 0, 1, . . . , H 1. The corresponding matrix factorizations are based on factorizing xihi =
xi xihi 1 . After choosing a path in Khler moduli space which connects the LG model with
the large volume, we can ask for a geometric interpretation of the n s in terms of bundles on
the corresponding hypersurface. This was studied in great detail following the work [22], and
understood in generality in [48,49], using results of [20]. See also the recent work [50]. Namely,
following a particular path in Khler moduli space, the n reduce to the restriction to the hypersurface of a so-called exceptional collection of bundles on the ambient P4w1 ,...,w5 . Exceptional
collections are particularly nice bases of branes to work with, and have appeared previously e.g.,
in the description of mirror symmetry for Fano varieties [20].
A.2. The quintic
In the following we would like to consider the example of the quintic in P4 . In order to change
the basis from LG to large volume (LV) it is enough to determine how the charges of the branes
transform. The LG charges of these branes are 16
5

n |k
= 1 k kn ,
(A.2)
15 One of the h s could be equal to 2, which one would not see in the Gepner model.
i
16 In this section, of course, = e2 i/5 .
39
where n = 0, . . . , 4 and k = 1, . . . , 4.
At large volume BPS charges arising from D-branes wrapping cycles in the CalabiYau are
determined in terms of the topology of the embedded cycle and a choice of bundle E to be [54,
39,55]
X ).
Q = ch(E) A(T
(A.3)
Wrapping a p-brane on a cycle induces a Dp-brane charge given by the rank of E, for example,
while lower brane charges resulting from the expansion of the Chern character are also induced.
The Chern characters of the bundles in the exceptional collection n = n (n) (here n = 0, . . . , 4
and is the cotangent bundle of P4 ) corresponding to the fractional branes at the LG point are
ch(0 ) = 1,
H2 H3
+
,
2
6
H2 H3
+
,
ch(2 ) = 6 3H
2
2
H2 H3
ch(3 ) = 4 + 3H
,
2
2
H2 H3
ch(4 ) = 1 H +
,
2
6
ch(1 ) = 4 + H +
(A.4)
where H is the hyperplane class of P4 . In matrix notation
1 4 6
4
1
3
1
0 1 3
Bin = chi (n ) =
1
0 12 12 12
.
2
0
1
2
1
6
12
(A.5)
16
Combining (A.5) with (A.2), we can work out the change of basis between the LG and the LV
limit.
One can then derive the matrix A representing the LV counterpart of the LG monodromy g
(which sends n n+1 ), namely
j
chi (n+1 ) = Ai chj (n )
(A.6)
with
4 20
3
1
1
A=
1
1
2
5
0
1
1
6
1
2
5
0
.
0
1
The intersection matrix of the n s is

5

5
1
1
1 k km
1 k kn
k
5
5
(1 )
4
k=1
(A.7)
40
0
5
0
5
= 10
5
10 10
5 10
10 10
5
5
10 10
0
5
10 ,
5
0
5
10
5
0
(A.8)
and again, the n s are not a minimal integral basis of the charge lattice. We can improve on this,
as first pointed out in [51], by using permutation branes.
The permutation branes based on the exchange of x1 and x2 have LG charges
4
1

n(12) |k
= 1 k kn = 1 k
n |k
,
(A.9)
which in LV translates to the Chern characters
0
0

(12)
Bin
= chi n(12) =
0
1
5
1 3
0 1
0 12
1
5
1
30
1
1
1
,
2
3
2
0
7
15
(A.10)
1
30
where

(12)
Bin = (1 A)1 B in .
(A.11)
The first of those

(12)
H3
ch 0
= ch3 =
,
5
(A.12)
describes a point on the quintic. Remember that H is the hyperplane class of P4 and the quintic
(12)
is in the class 5H , so 0 intersects the quintic exactly once. Note that even though this set of
(12)
branes contains a D0-brane, the n are still not a minimal basis of the charge lattice.
(12)(34)
, which are based on the exchange of x1 and x2 ,
Continuing, the permutation branes n
and x3 and x4 , have LG charges
3
2

n |k
n(12)(34) |k
= 1 k kn = 1 k
(A.13)
and Chern characters
0
0
(12)(34)

= (1 A)2 B in =
chi n
15
1
5
0
0
15
1
0
15
15
2
1
3
10
7
30
1
1
3
.
10
(A.14)
7
30
These are now indeed a minimal basis of the charge lattice (but do not contain a D0-brane). Their
intersection form is
41
4
3

3
1
1
1 k km
1 k kn
k
5
5
(1 )
k=1
0
0 1 1
0
0
0 1 1
0
= 1
0
0
0 1 .
1 1
0
0
0
0 1 1
0
0
(A.15)
A.3. Identification of the D0-brane

(12)
It follows from the previous discussion that the set of fractional branes n containing the
D0-brane and the set of CardyRecknagelSchomerus branes n containing the D6-brane are
related by the formula
(12)
n = (1 g)nm m
.
(A.16)
Such an identification of the D0-brane in the LG model appears in fact to be canonical and holds
in particular for all hypersurfaces whose analytical continuation has been studied so far, see, e.g.,
[49,52].
(12)
More properly, the statement that one of the n is a D0-brane of course depends on the
analytical continuation that one has chosen to connect LG to LV. For example, encircling the LG
point leads to a cyclic permutation n n + 1. One of the consequences of this ambiguity in
the context of flux compactifications is that the statement the fluxes contribute to the D3-brane
tadpole is not invariant under all small volume monodromies: If the D0-brane on the CalabiYau
returns under monodromy as a general combination of even-dimensional cycles, this can only be
consistent with tadpole cancellation if after the monodromy, the fluxes become non-geometric
and contribute in other classes as well.
However, now we have to take into account that we are performing an orientifold projection.
In type IIB, this selects a real subspace of the Khler moduli space, which therefore eliminates
some of the possible monodromies. Moreover, as we will see in the next section for the particular
example of the quintic the orientifold projection fixes the ambiguity completely.
In the above discussion we have seen an interesting interplay between orientifolds, monodromies and tadpole contributions generated by fluxes. In the present context we have used
this interplay to identify the class of a point in the LG regime but we expect it to have implications beyond the present discussion and to play a pivotal role in the search for the LG theories
incorporating NSNS and RR fluxes.
A.4. Orientifolds of the quintic
We consider first the trivial involution
0 : (x1 , x2 , x3 , x4 , x5 ) (x1 , x2 , x3 , x4 , x5 ),
(A.17)
and
where, the full orientifold group consists of
0 for k = 0, . . . , 4. To compute C|k
,
following [26], we have to look for those elements of the orientifold group which square to the
element g k of the orbifold group, and then compute its eigenvalues. E.g., for k = 1 (g 3 0 )2 = g,
with eigenvalues (3 , . . . , 3 ) and so on:
5

C|k
= 1 3k
(A.18)
gk
gk
42
and using (1 )1 = 15 (4 + 3 + 22 + 3 ), we find for the class of the O-plane

[O] = 3[0 ] 5[2 ] 5[3 ],
(A.19)
which corresponds to large volume charges 7 + 5H 2 . After we recall that the formulas in [26]
are missing a factor of 4 from the extended directions, we see that this would correspond to a
rank 28 bundle, which cannot be correct for tadpole cancellation for a type I compactification on
the quintic, which we would have naively expected corresponds to this orientifold (and would
hence require a rank 32 bundle).
A solution to this was noted in [25]. Recall that the correspondence between n s and bundles
is in fact ambiguous by the path we choose to get to large volume. In the quintic case, the path is
fixed by the orientifold projection, except at the orbifold point. In fact,
g 2 [O] = 3[2 ] 5[4 ] 5[0 ]
(A.20)
corresponds to large volume charges 8 4H 4H 2 + 73 H 3 , and gives rank 32 after we multiply

by our factor of 4. Thus, if we modify our path by 2 LG monodromies (which is compatible with
the orientifold projection on the moduli space), we get agreement with large volume data.
To understand that this is in fact the path we must take, we recall that a global coordinate on
the Khler moduli space is the complex structure parameter of the mirror quintic
y15 + y25 + y35 + y45 + y55 5y1 y2 y3 y4 y5 .
(A.21)
More precisely, gives a five-fold cover of the moduli space, which is usually parametrized by
z = (5)5 . The LG monodromy corresponds to e2i/5 . Now, the orientifold acts on the
mirror quintic simply by complex conjugation yi yi , and hence restricts to be real. This is
a stronger condition than requiring z to be real. Navigating from positive real to negative real
in fact requires encircling the origin of the moduli space z = 0 twice in the positive direction
(or thrice in the negative direction).
That g 2 [O] does not seem to correspond to a real bundle in this case (odd Chern classes are
non-zero) is explained by the fact that we actually end up with a non-zero NSNS B-field (more
precisely B = H /2) under this analytical continuation. Namely
4eH /2 [O] = 32 20H 2
(A.22)
which is correct for anomaly cancellation in type I on the quintic.

To complete the story, we note that the naive result (A.19), 4(7 + 5H 2 ) differs from the
type I result with B = 0 simply by one unit of D6-brane charge, as well as a sign. Both can be
understood by noting that the path starting at large volume with B = 0 has to go through the
conifold singularity before reaching the LG point. At the conifold, the O-plane looses exactly
one unit of the vanishing cycle, which is the D6-brane, and also changes into an anti-orientifold
plane, see [53]. We thus see that we can understand completely the charge of the orientifold
plane under analytical continuation through the quintic moduli space, and that large volume and
Gepner/LG data agree beautifully.
A.4.1. Exchange orientifolds
Consider now the orientifold action
1 : (x1 , x2 , x3 , x4 , x5 ) (x2 , x1 , x3 , x4 , x5 ).
(A.23)
Its LG charges are

4

C|k
= 1 + 3k 1 3k ,
43
(A.24)
which gives at large volume

19 3
H ,
15
16
g 2 [O1 ] = 2H + H 2 + H 3 .
(A.25)
15
Again, we can check that this matches the geometric expectations. At large volume, the fixed
point locus of the involution consists of two components [23]: An O7-plane at a hyperplane
x1 = x2 , and an O3-plane at a point x1 = x2 , x3 = x4 = x5 = 0. The general formulas (see, e.g.,
[25]) give the O-plane charge of a fixed component Y X as
!
" 1
3
codim
(Y
)
" L( T Y )
R
2
# 4
[Y ]
(A.26)
,
L( 14 N Y )
A(X)
[O1 ] = 1 + 2H
where [Y ] is the Poincar dual of the fixed point locus, and the sign is the type of O-plane
2
(O+ or O ). For the quintic X in P4 , A(X)

= 1 + 10
12 H . The hyperplane has [Y ] = H , and
2
3
c(N Y ) = 1+H , so c(T Y ) = (1+10H 40H )/(1+H ) = 1H +11H 2 . We find L( 14 N Y ) =
1 + H 2 /48, L( 14 T Y ) = 1 21H 2 /48. For the point on the quintic, [pt] = H 3 /5, so the formula
evaluates altogether to
$

H3
(1 21H 2 /48)
31 3
H3
2H
(A.27)
2H
H
.
40
24
40
(1 + H 2 /48)(1 + 10H 2 /12)
Let us compare this with [O1 ] and g 2 [O1 ] we have computed above. First of all, we have to
add 1 to [O1 ] because the path to large volume crosses the conifold locus. Then we see that the
resulting O-plane is an O7 with an O3 of the same type (we cannot determine the overall type
from these considerations)
2H
31 3 H 3
19
H +
= 2H H 3 .
24
40
15
(A.28)
For g 2 [O1 ], we have to multiply it with eH /2 because of the B-field, and find that this is an O7
with an O3 of the opposite type

H3
16
31
eH /2 2H H 3
(A.29)
= 2H + H 2 H 3 .
24
40
15
It should be possible to understand geometrically why the B-field changes the type of the O3plane in this fashion.
Finally, we consider the orientifold action with two exchanges, which is in LG limit:
2 : (x1 , x2 , x3 , x4 , x5 ) (x2 , x1 , x4 , x3 , x5 ).
Its LG charges are
2
3

C|k
= 1 + 3k 1 3k
(A.30)
(A.31)
44
which gives at large volume

3
[O2 ] = 1 H 2 ,
5
2 2 1 3
2
g [O2 ] = H H .
(A.32)
5
5
In the geometric regime, the fixed point locus corresponds to an O5 at a degree 5 curve at x1 = x2 ,
x3 = x4 , 2x15 + 2x35 + x55 = 0 in cohomology class H 2 , plus an O5 at a rational curve x1 = x2 ,
x3 = x4 , x5 = 0 in class H 2 /5. The general formula evaluates to
1 H2 1 2
H .
2 5
2
Indeed, removing the 1 from [O2 ], this is
(A.33)
1
1
3
H 2 = H 2 H 2,
5
2
10
(A.34)
whereas for g 2 [O2 ], we get
1
2 2 1 3
1
H H = eH /2 H 2 H 2 .
5
5
2
10
(A.35)
Again, the type of one component of the O-plane changes as we navigate through the nongeometric phase, or as we change the B-field from 0 to 1/2.
References
[1] M.R. Douglas, The statistics of string/M theory vacua, JHEP 0305 (2003) 046, hep-th/0303194.
[2] M.R. Douglas, S. Kachru, Flux compactification, hep-th/0610102.
[3] O. DeWolfe, A. Giryavets, S. Kachru, W. Taylor, Enumerating flux vacua with enhanced symmetries, JHEP 0537
(2005) 0502, hep-th/0411061;
O. DeWolfe, Enhanced symmetries in multiparameter flux vacua, JHEP 0510 (2005) 066, hep-th/0506245.
[4] F. Denef, M.R. Douglas, B. Florea, Building a better racetrack, JHEP 0406 (2004) 034, hep-th/0404257;
F. Denef, M.R. Douglas, B. Florea, A. Grassi, S. Kachru, Fixing all moduli in a simple F-theory compactification,
hep-th/0503124;
P.S. Aspinwall, R. Kallosh, Fixing all moduli for M-theory on K3 K3, JHEP 0510 (2005) 001, hep-th/0506014;
B.S. Acharya, F. Benini, R. Valandro, Fixing moduli in exact type IIA flux vacua, hep-th/0607223;
D. Lust, S. Reffert, W. Schulgin, S. Stieberger, Moduli stabilization in type IIB orientifolds. I: Orbifold limits,
hep-th/0506090;
I. Antoniadis, A. Kumar, T. Maillard, Magnetic fluxes and moduli stabilization, hep-th/0610246.
[5] S. Kachru, R. Kallosh, A. Linde, S.P. Trivedi, De Sitter vacua in string theory, Phys. Rev. D 68 (2003) 046005,
hep-th/0301240.
[6] G. Villadoro, F. Zwirner, N = 1 effective potential from dual type-IIA D6/O6 orientifolds with general fluxes,
JHEP 0506 (2005) 047, hep-th/0503169.
[7] O. DeWolfe, A. Giryavets, S. Kachru, W. Taylor, Type IIA moduli stabilization, JHEP 0507 (2005) 066, hepth/0505160.
[8] B.R. Greene, C. Vafa, N.P. Warner, CalabiYau manifolds and renormalziation group flows, Nucl. Phys. B 324
(1989) 371.
[9] E. Witten, Phases of N = 2 theories in two dimensions, Nucl. Phys. B 403 (1993) 159, hep-th/9301042.
[10] E.J. Martinec, Algebraic geometry and effective Lagrangians, Phys. Lett. B 217 (1989) 431.
[11] K. Hori, C. Vafa, Mirror symmetry, hep-th/0002222.
[12] C. Vafa, String vacua and orbifoldized LG models, Mod. Phys. Lett. A 4 (1989) 1169.
[13] S. Kachru, M.B. Schulz, P.K. Tripathy, S.P. Trivedi, New supersymmetric string compactifications, JHEP 0303
(2003) 061, hep-th/0211182.
45
[14] S. Hellerman, J. McGreevy, B. Williams, Geometric constructions of nongeometric string theories, JHEP 0401
(2004) 024, hep-th/0208174.
[15] A. Dabholkar, C. Hull, Duality twists, orbifolds, and fluxes, JHEP 0309 (2003) 054, hep-th/0210209;
C.M. Hull, A geometry for non-geometric string backgrounds, JHEP 0510 (2005) 065, hep-th/0406102;
C.M. Hull, Doubled geometry and T-folds, hep-th/0605149.
[16] J. Shelton, W. Taylor, B. Wecht, Nongeometric flux compactifications, JHEP 0510 (2005) 085, hep-th/0508133;
J. Shelton, W. Taylor, B. Wecht, Generalized flux vacua, hep-th/0607015.
[17] A. Flournoy, B. Wecht, B. Williams, Constructing nongeometric vacua in string theory, Nucl. Phys. B 706 (2005)
127, hep-th/0404217.
[18] S. Hellerman, J. Walcher, Worldsheet CFTs for flat monodrofolds, hep-th/0604191.
[19] D. Gepner, Spacetime supersymmetry in compactified string theory and superconformal models, Nucl. Phys. B 296
(1988) 757.
[20] K. Hori, A. Iqbal, C. Vafa, D-branes and mirror symmetry, hep-th/0005247.
[21] A. Recknagel, V. Schomerus, D-branes in Gepner models, Nucl. Phys. B 531 (1998) 185, hep-th/9712186.
[22] I. Brunner, M.R. Douglas, A.E. Lawrence, C. Romelsberger, D-branes on the quintic, JHEP 0008 (2000) 015, hepth/9906200.
[23] I. Brunner, K. Hori, Orientifolds and mirror symmetry, JHEP 0411 (2004) 005, hep-th/0303135.
[24] A. Lawrence, J. McGreevy, Local string models of soft supersymmetry breaking, JHEP 0406 (2004) 007, hepth/0401034.
[25] I. Brunner, K. Hori, K. Hosomichi, J. Walcher, Orientifolds of Gepner models, hep-th/0401137.
[26] K. Hori, J. Walcher, D-brane categories for orientifolds: The LandauGinzburg case, hep-th/0606179.
[27] S. Gukov, C. Vafa, E. Witten, CFTs from CalabiYau four-folds, Nucl. Phys. B 584 (2000) 69, hep-th/9906070;
S. Gukov, C. Vafa, E. Witten, Nucl. Phys. B 608 (2001) 477, Erratum.
[28] F. Cachazo, K.A. Intriligator, C. Vafa, A large N duality via a geometric transition, Nucl. Phys. B 603 (2001) 3,
hep-th/0103067.
[29] R. Dijkgraaf, C. Vafa, A perturbative window into non-perturbative physics, hep-th/0208048.
[30] A. Kapustin, Y. Li, D-branes in LandauGinzburg models and algebraic geometry, JHEP 0312 (2003) 005, hepth/0210296;
A. Kapustin, Y. Li, Topological correlators in LandauGinzburg models with boundaries, hep-th/0305136;
I. Brunner, M. Herbst, W. Lerche, B. Scheuner, LandauGinzburg realization of open string TFT, hep-th/0305133.
[31] J. Walcher, Stability of LandauGinzburg branes, J. Math. Phys. 46 (2005) 082305, hep-th/0412274.
[32] K. Dasgupta, G. Rajesh, S. Sethi, M theory, orientifolds and G-flux, JHEP 9908 (1999) 023, hep-th/9908088.
[33] S.B. Giddings, S. Kachru, J. Polchinski, Hierarchies from fluxes in string compactifications, Phys. Rev. D 66 (2002)
106006, hep-th/0105097.
[34] K. Becker, M. Becker, M-Theory on eight-manifolds, Nucl. Phys. B 477 (1996) 155, hep-th/9605053.
[35] S. Gukov, Solitons, superpotentials and calibrations, Nucl. Phys. B 574 (2000) 169, hep-th/9911011.
[36] A. Recknagel, Permutation branes, JHEP 0304 (2003) 041, hep-th/0208119;
I. Brunner, M.R. Gaberdiel, Matrix factorisations and permutation branes, JHEP 0507 (2005) 012, hep-th/0503207;
H. Enger, A. Recknagel, D. Roggenkamp, Permutation branes and linear matrix factorisations, JHEP 0601 (2006)
087, hep-th/0508053.
[37] C.P. Burgess, C. Escoda, F. Quevedo, Nonrenormalization of flux superpotentials in string theory, JHEP 0606 (2006)
044, hep-th/0510213.
[38] E. Witten, On flux quantization in M-theory and the effective action, J. Geom. Phys. 22 (1997) 1, hep-th/9609122.
[39] D.S. Freed, E. Witten, Anomalies in string theory with D-branes, hep-th/9907189.
[40] D.E. Diaconescu, G.W. Moore, E. Witten, E(8) gauge theory, and a derivation of K-theory from M-theory, Adv.
Theor. Math. Phys. 6 (2003) 1031, hep-th/0005090.
[41] A.R. Frey, J. Polchinski, N = 3 warped compactifications, Phys. Rev. D 65 (2002) 126009, hep-th/0201029.
[42] S. Kachru, M.B. Schulz, S. Trivedi, Moduli stabilization from fluxes in a simple IIB orientifold, JHEP 0310 (2003)
007, hep-th/0201028.
[43] A.M. Uranga, D-brane probes, RR tadpole cancellation and K-theory charge, Nucl. Phys. B 598 (2001) 225, hepth/0011048;
A.M. Uranga, D-brane, fluxes and chirality, JHEP 0204 (2002) 016, hep-th/0201221.
[44] G. Villadoro, F. Zwirner, D terms from D-branes, gauge invariance and moduli stabilization in flux compactifications, JHEP 0603 (2006) 087, hep-th/0602120.
[45] J.Distler, D.S.Freed, G.W. Moore, in preparation.
[46] F. Denef, M.R. Douglas, Computational complexity of the landscape. I, hep-th/0602072.
46
[47] M. Dine, Z. Sun, R symmetries in the landscape, JHEP 0601 (2006) 129, hep-th/0506246.
[48] D.E. Diaconescu, M.R. Douglas, D-branes on stringy CalabiYau manifolds, hep-th/0006224.
[49] P. Mayr, Phases of supersymmetric D-branes on Kaehler manifolds and the McKay correspondence, JHEP 0101
(2001) 018, hep-th/0010223.
[50] M. Herbst, K. Hori, D. Page, Various Talks since Spring 2005, in preparation.
[51] S.K. Ashok, E. DellAquila, D.E. Diaconescu, Fractional branes in LandauGinzburg orbifolds, Adv. Theor. Math.
Phys. 8 (2004) 461, hep-th/0401135.
[52] E. Scheidegger, On D0-branes in Gepner models, JHEP 0208 (2002) 001, hep-th/0109013.
[53] K. Hori, K. Hosomichi, D.C. Page, R. Rabadan, J. Walcher, Non-perturbative orientifold transitions at the conifold,
JHEP 0510 (2005) 026, hep-th/0506234.
[54] R. Minasian, G.W. Moore, K-theory and RamondRamond charge, JHEP 9711 (1997) 002, hep-th/9710230.
[55] M.B. Green, J.A. Harvey, G.W. Moore, I-brane inflow and anomalous couplings on D-branes, Class. Quantum
Grav. 14 (1997) 47, hep-th/9605033.
[56] C. Vafa, Superstrings and topological strings at large N , J. Math. Phys. 42 (2001) 2798, hep-th/0008142.
[57] N. Berkovits, W. Siegel, Superspace effective actions for 4D compactifications of heterotic and type II superstrings,
Nucl. Phys. B 462 (1996) 213, hep-th/9510106.
[58] W.D. Linch, B.C. Vallilo, Hybrid formalism, supersymmetry reduction, and RamondRamond fluxes, hepth/0607122.
[59] A. Misra, Type IIA on a compact CalabiYau and D = 11 supergravity uplift of its orientifold, Fortschr. Phys. 52
(2004) 831, hep-th/0311186.
Flipped SU(5) from Z12I orbifold with Wilson line

Jihn E. Kim a, , Bumseok Kyae b
a Department of Physics and Astronomy, and Center for Theoretical Physics, Seoul National University,
Seoul 151-747, Republic of Korea

b School of Physics, Korea Institute for Advanced Study, 207-43 Cheongryangri-dong, Dongdaemun-gu,
Seoul 130-722, Republic of Korea

Received 8 September 2006; received in revised form 27 December 2006; accepted 1 February 2007
Abstract
We construct a three family flipped SU(5) model from the heterotic string theory compactified on the
Z12I orbifold with one Wilson line. The gauge group is SU(5) U (1)X U (1)3 [SU(2) SO(10)
U (1)2 ] . This model does not derive any non-Abelian group except SU(5) from E8 , which is possible
only for two cases in case of one shift V , one in Z12I and the other in Z12II . We present all possible
Yukawa couplings. We place the third quark family in the twisted sectors and two light quark families
in the untwisted sector. From the Yukawa couplings, the model provides the R-parity, the doublettriplet
splitting, and one pair of Higgs doublets. It is also shown that quark and lepton mixings are possible. So far
we have not encountered a serious phenomenological problem. There exist vector-like flavor SU(5) exotics
(including Qem = 16 color exotics and Qem = 12 electromagnetic exotics) and SU(5) vector-like singlet
0 = 3 at the full
exotics with Qem = 12 which can be removed near the GUT scale. In this model, sin2 W
8
unification scale.
PACS: 11.25.Mj; 12.10.Kt; 12.60.Jv
Keywords: Orbifold; Three families; Yukawa couplings; Flipped SU(5)
1. Introduction
At present, it is of utmost importance to connect the high energy string theory with the low
energy standard model, in particular with the minimal supersymmetric standard model (MSSM).
E-mail addresses: jekim@phyp.snu.ac.kr (J.E. Kim), bkyae@kias.re.kr (B. Kyae).

doi:10.1016/j.nuclphysb.2007.02.008
48
J.E. Kim, B. Kyae / Nuclear Physics B 770 (2007) 4782
The initial attempt of the CalabiYau space compactification, which is geometrical, has been
very attractive [1]. But the orbifold compactification, also being a geometrical device, got more
interest due to its simplicity in model buildings [2,3]. Initially, the standard-like models were
looked for [4], in an attempt to obtain minimal supersymmetric standard models (MSSMs), but
it became clear that the standard-like models have a serious problem on sin2 W to arrive at
MSSMs [5]. All ZN models without Wilson lines were tabulated a long time ago [6] and recently
all Z3 models with Wilson lines are tabulated in a book [7].
0 at the unification or
The sin2 W problem is that it is better for the bare value of sin2 W
string scale to be close to 38 [5] so that it reproduces the fact of the convergence of three gauge
couplings at one point near the unification scale [8]. The so-called flipped SU(5) does not fulfill
this requirement automatically due to the leakage of U (1)Y beyond SU(5).1 Thus, the sin2 W
problem directs toward grand unified theories (GUTs) from superstring without the electroweak
hypercharge Y leaking outside the GUT group. In this regard, one may consider simple GUT
groups SU(5) [12], SO(10) [13], E6 [14] and trinification SU(3)3 [15]. The simplest orbifold
without matter representations beyond the fundamentals require the KacMoody level K = 1.
With K = 1, one cannot obtain adjoint representations [7]; thus among the above GUT groups
the trinification group is the allowed one. Also, the PatiSalam SU(4) SU(2) SU(2) [16] can
be broken to the standard model without an adjoint matter representation; above the GUT scale
however three gauge couplings of the PatiSalam model diverge rather than evolving in unison.
Thus, trinification GUT seems to be the most attractive solution regarding the sin2 W problem.
The trinification is possible only in Z3 orbifolds [17].
Another interesting GUT group, though not giving sin2 W = 38 necessarily, is the flipped
SU(5) [9,10] where the exchanges d c uc and ec (neutral singlet c ) in the representations
of SU(5) are adopted. The matter representation of the flipped SU(5) is, under SU(5) U (1)X ,2
16flip 101 + 53 + 15 .
(1)
The electroweak hypercharge is given by

Y = 15 (X + Y5 )
(2)
where Y5 = diag( 13 13 31 12 12 ) and X = diag(x x x x x). Then, the electroweak hypercharges

of 15 and 53 are +1, 23 , and 12 , which are ec , uc , and electron doublet. There are some nice
features of flipped SU(5) [18].
From the string context, flipped SU(5) was considered before in the fermionic construction
scheme [11] and recently in orbifold construction also [19], CalabiYau compactification [20],
and intersecting D-brane models [21]. Let us call flipped SU(5) from string construction string
flipped SU(5). In string flipped SU(5), it does not necessarily predict sin2 W = 38 at the unification scale [5]. However, if we introduce more parameters intrinsic in flipped SU(5), we may fit
parameters so that the gauge couplings meet at one point at the string scale Ms , the unification
scale of SU(5) and U (1)X couplings. These parameters include the symmetry breaking scale
MGUT for SU(5) U (1)X SM breaking and intermediate scales of vector-like representations. Above MGUT the RG evolutions of SU(5) and U (1)X couplings are different, and we do
not expect a string scale around 0.7 1018 GeV [22] but can be determined by the unification
1 The terminology flipped SU(5) was used as an SU(5) U (1) subgroup of SO(10) [911]. In this paper, we still
X
use the same terminology if there appear 16s having the same quantum numbers as in the flipped SU(5).
2 The quantum number X of U (1) in the flipped SU(5) is highlighted.
X
49
ansatz at Ms and mass scales of vector-like representations [23]. In our string orbifold model,
sin2 W turns out to be 38 at the full unification scale due to the possibility of the electroweak hypercharge embedding in SO(10). Thus such vector-like fields should be removed near the GUT
scale.
String models in general include exotics. Electromagnetic exotics (E-exotics) are fractionally charged particles which are non-Abelian gauge group singlets. Color exotics (C-exotics) are
quarks with non-standard charges, i.e. color triplet quarks not having Qem = 23 or 13 and color
anti-triplet quarks not having Qem = 23 or 13 . Flipped SU(5) GUT exotics (G-exotics) are SU(5)
representations, not having the X charges of 101 , 101 , 53 , 53 , 52 , 52 , 15 , 10 . G-exotics contain C-exotics and fractionally charged leptons.
The fermionic construction of flipped SU(5) has shown the existence of E-exotics with
Qem = 12 and integer charged cryptons where cryptons are defined to be the composites
of the hidden sector confining group SU(4) [24]. Cosmological effect of cryptons was given
in [25]. Because of the possibility that fractionally charged particles exist in most string vacua,
discovery of fractionally charged particles may strongly hint the correctness of the idea of string
compactification.
In this paper, we present the orbifold compactification with Z12I twist. This contains a detailed account of Ref. [19]. In addition, we present another orbifold model having a hidden sector
SU(4) . We succeeded in constructing a phenomenologically desirable flipped SU(5) model
from Z12I .
We need three families of 16flip , where

16flip 101 + 53 + 15 = d c , q, c + uc , l + ec .
(3)
For spontaneous symmetry breaking, we need also the Higgs fields,
(101 + 101 ) + (52 + 52 ).
Sometimes, it is useful to represent the components in terms of
c
c

d
u
q
,
15 : e c ,
5
:
,
101 :
3
l
q c

D
D
52 :
52 :
,
,
hd
hu
(4)
(5)
where q and l are lepton and quark doublets, D is Qem = 13 quark, and hd,u are Higgs doublets
giving mass to d, u quarks. Spontaneous symmetry breaking of flipped SU(5) proceeds via VEVs
of 101 and 101 (components c , c ) and 52 and 52 (components in hd , hu ).
In this model, there appear two light families from the untwisted sector and the third heavy
family from twisted sectors. This is dictated from the Yukawa coupling structure. It also leads to
(i) the doublettriplet splitting, (ii) one pair of Higgs doublets, and (iii) the existence of R-parity.
Some standard models from fermionic construction are worthwhile to note since they may be
free from some problems of GUTs [26]. But these models use Z2 Z2 which may have a very
different phenomenology from the one we discuss here with Z12I .
In Section 2, we present a review on orbifold construction with order N = 12. Here we include
formulae with Wilson lines also. In Sections 3 and 4, the untwisted and twisted sector spectra are
calculated in detail. In Section 5, we collect all observable sector fields. In Section 6, we present
the Yukawa coupling structure and derive some phenomenological consequences. In Section 7
0 = 3 . Section 8 is a conclusion. In Appendix A, we provide another model
we show sin2 W
8

having SU(4) .
50
Fig. 1. Simplified showing of fixed points of the Z12I orbifold.
2. Orbifold method
An orbifold is constructed from a manifold by identifying points under a discrete symmetry
group. The six internal space is orbifolded by a twist vector s . With three complexified components, s has three components s1 , s2 and s3 . The twist of Z12I orbifold is [7]

5 4 1
1 7
s =
(6)
, ,
with s2 =
.
12 12 12
12 2
Torus corresponding to si is twisted by si . Hence the first and third tori have one fixed point
while the second torus being modded by Z3 has three fixed points as schematically shown in
Fig. 1. Multiplicity due to fixed points are three. These can be distinguished by Wilson lines.
In ten-dimensional (10D) heterotic string, left and right movers are treated as gauge group
degrees and N = 1 supersymmetry, respectively. The embedding possibility is in the gauge group
space of left movers, NS sector of left and right movers, and R sector of right movers. So we focus
on the embedding in the group space for left movers and in the R sector for right movers. In 10D,
the R sector embedding is given by s . The group space embedding is given by sixteen numbers,
1
V I (I = 1, 2, . . . , 16) {v(I = 1, . . . , 8), v (I = 9, . . . , 16)}. Factoring out 12
by defining sa =
1
1
1
,
v
=
w
,
v
=
w
,
we
must
satisfy
for
a
Z
orbifold
[27]
12I
12 a a
12 a a
12 a
3

a=1
a2
8

wa2
a=1
3a3 = 3a4 = 0,
8

wa 2 = 0 mod 24,
a=1
a1 = a2 = a5 = a6 = 0.
(7)
2.1. Dynkin diagram technique for finding gauge group

Just for finding out a gauge group structure, the Dynkin diagram technique is extremely useful [28]. In the Dynkin diagram, each simple root is endowed with a Coxeter label. A Dynkin
diagram technique of obtaining maximal subgroups is to strike out a simple root from the extended Dynkin diagram. In orbifold, this is generalized to strike out some roots where sum of the
eliminated Coxeter labels add up to order N of ZN . To have SU(5) only without a Wilson line,
3
there must remain four linearly connected simple
roots. So, for N 8 orbifold, it is impossible.
For Z12 , there is only possibility. Suppose i ci = N where ci is the Coxeter label of simple
root i . There is only one possibility which is c0 + c1 + c2 + c6 + c7 = 12. See Fig. 2. Thus,
from orbifold compactification, there are only two possibilities for constructing flipped SU(5)
3 By a two step process using Wilson lines, it is possible to obtain SU(5) in other Z orbifolds.
N
51
Fig. 2. The SU(5) subgroup of E8 . The Coxeter labels are shown inside circles.

models,
one in Z12I and another in Z12II . For N = i ci = N , the shift vector V is given
by V = i i where i are fundamental weights [27]. Thus, for Z12 orbifolds V is given by
V1 = 0 + 1 + 2 + 6 + 7 with 0 = 0. Thus, the shift vector for flipped SU(5) is

1 17 5 3 1 1 1 1 1
V1 =
( ).
12 2 2 2 2 2 2 2 2
5
, to obtain
Now we shift the origin by 24
1
(8)
(11 5 4 3 3 3 3 2)( ).
12
Both V1 and V2 give an unbroken SU(5). But the entries of V1 and V2 do not have five common
entries. So, we try to add an integer times Z12I shift s so that the resulting entries manifestly
show five common entries. In this way, of course the SU(5) non-Abelian group is kept. Usually,
if one adds s to three entries of E8 , some non-Abelian groups are broken. So our strategy is to
4
8
4
0 12
0 0 0 0 12
) to obtain
add a multiple of s such that an SU(5) survives. For this, we add ( 12
1
12 (15 5 12 3 3 3 3 6). Subtracting torus lattice and rearranging entries, we obtain
V2 =
1
(9)
(3 3 3 3 3 5 6 0)
12
which has five common entries. This form is perfectly simple enough in obtaining SU(5) weights
since there are five common entries. Otherwise, i.e. with V1 or V2 , it is cumbersome to work out
all the SU(5) weights as tried out in [7].
Since the five entries are common, the simple roots for SU(5) take the following standard
form,
V3 =
1 = (1 1 0 0 0; 0 0 0),
2 = (0 1 1 0 0; 0 0 0),
3 = (0 0 1 1 0; 0 0 0),
4 = (0 0 0 1 1; 0 0 0).
Then, the highest weights of some representations we use are

(1 0 0 0 0; 0 0 0),
5:
(+ 1 1 1 1 1 ; 0 0 0),
2 2 2 2 2
(1 0 0 0 0; 0 0 0),
5:
( 12 + 12 + 12 + 12 + 12 ; 0 0 0),

(1 1 0 0 0; 0 0 0),
10:
(+ 12 + 12 12 12 12 ; 0 0 0),
(10)
(11)
(12)
(13)
52

10:
(1 1 0 0 0; 0 0 0),
( 12 12 + 12 + 12 + 12 ; 0 0 0).
(14)
For the E8 , we find roots and weights in a similar way.

2.2. Choosing shift vector and Wilson lines
Embedding of the orbifold action can be found by satisfying the modular invariance conditions (7). If one tries to have a specific gauge group, the Dynkin diagram technique is very
helpful as we have discussed in Section 2.1. Or one can study the components of ( )( ) to
guess the gauge group, but this method of finding gauge group is completed only after one obtains all nonzero roots of the gauge multiplet. For example, if one tries V2 of (8) and V3 of (9)
then in both cases he will obtain SU(5). But guessing SU(5) from (8) is not straightforward. In
this sense, the Dynkin diagram technique is superior. On the other hand, a computer search of
gauge groups will cover all these cases. In the computer search, the identical shift vector is given
by several different forms as done in V2 and V3 . Usually, it is very difficult to identify all the
same shift vectors [7].
Choosing the shift vector (V ) and Wilson lines (a1 , . . . , a6 ) fixes the embedding of the orbifold
action in the group space. So gauge groups and representations are fixed by the shift vector and
Wilson lines, consistently with the modular invariance condition (7).
In summary, the compactification is specified by twisting represented by a shift form in the
six internal space s , shift V , and Wilson lines a1 , . . . , a6 ,
internal space:
group space:
s = (s1 , s2 , s3 ),
V , ai (i = 1, 2, . . . , 6).
(15)
Here, s0 determines the chirality and s encodes the orbifolding information of the R sector.
2.3. Massless modes
Finding out all the massless modes below the compactification scale is the key problem in the
compactification process. The left movers and right movers have different relations for the Einstein mass-shell condition, even though the form has a similarity. In the symmetric orbifold [2],
let us bosonize the Ramond sector of right movers, which is represented by four half integers
in s. The orbifold action is modding out the 6D torus, and s contains the orbifold information of
right movers under translation in the internal space. For left movers, momenta P corresponding
to translation in the group space have the orbifold information. All these satisfy the level matching condition, ML2 = MR2 . Thus, left moving and right moving states on torus have the following
vanishing vacuum energy for massless states,
left movers:
(P + kV )2 L
Nj j ck = 0,
+
2
j
right movers:
(s + ks
2
)2
NjR j ck = 0,
(16)
2,
3}.
Here i ki mod Z such that 0 < i 1, ki mod Z
where j runs over {1, 2, 3, 1,
i
such that 0 < i 1. (If ki is an integer, j = 1 [29,30].) For k = 0, conditions for massless
53
left and right movers are given by

P2 = 22
NjL j ,
j
s2 = 1 2
NjR j ,
(17)
where c0 = c0 + 12 .
The massless modes include graviton g , antisymmetric tensor field B , dilaton, gravitino,
gauge bosons, gauginos, and chiral matter. So the matter states (P , s) must satisfy

(P + kV )2 = 2ck 2
NjL j ,
j
(s + ks ) = 2ck 2
2
NjR j .
(18)
Among these massless modes, we are interested in a resulting N = 1 SUSY gauge theory.
The SUSY condition for orbifold compactification is given by right movers, which are given
by four component s, including three component s . When we compactify six dimensions, tendimensional supersymmetry generators can be decomposed into Q(10) = Q(4) Q(6) . The sixdimensional internal space part Q(6) transforms as 4 of SO(6) SU(4). Because the remaining
part Q(4) becomes the four-dimensional generator, the dimension 4 counts the number of supersymmetries. Its spinorial representation is given by |s = |s1 s2 s3 = | 12 , 12 , 12 with even
number of minus signs. Under point group, it transforms as
Q(6) exp(2is )Q(6) .
The invariant component corresponds to the unbroken supersymmetry generator. For N 1
supersymmetry, we need at least one solution, say if we choose s = (+ 12 , 12 , 12 ), the argument
of exponent vanishes for
1 2 3 = 0 mod Z.
The number of solutions, N , counts the number of unbroken supersymmetry generators from
5 4 1
, 12 , 12 ) the above condition is satisorbifold compactification. Note that for our choice s = ( 12
1
1
1
fied only for s = (+ 2 , 2 , 2 ). We can introduce another set, r , of half integers so that entries
of s + r become integers,
r = (r1 , r2 , r3 ).
(19)
Namely, the SUSY condition for orbifolded spectrum is s r = 0 [7].

2.3.1. Gauge multiplet
Gauge boson multiplets appear in the untwisted sector U , satisfying
gauge group:
P 2 = 2,
P V = 0,
P ai = 0
for all i,
(20)
where the first one is the masslessness condition and the second and the third ones are the orbifold
conditions. The corresponding right movers, satisfying the mass-shell and orbifold conditions
chooses two ss with s 2 = 1, which are always CPT conjugates of each other. In this way, we
obtain the gauge multiplet. As expected, the multiplicity (P) of gauge bosons is 1.
54
2.3.2. Matter multiplets

Other massless states can appear in U also, for
U1 :
P 2 = 2,
s + r = (1, 0, 0),
P V =
k
N
P ai = 0 for all i.
(21)
We combine s and r to distinguish untwisted matter so that they appear in three categories under
orbifolding, i.e. differing in sub-lattice shifts,4
untwisted matter:
U2 :
for k = 0 mod N,
s + r = (0, 1, 0),
s + r = (0, 0, 1).
U3 :
(22)
There are fixed points in field theory orbifolds. In string orbifolds also, we must consider
physics related to fixed points. Massless strings can sit at fixed points, which is found by the
mass-shell condition at fixed point. But noting that some linear combinations of strings sitting at
several fixed points may be taken to satisfy the orbifold condition, we consider twisted sectors.
For ZN orbifold, we consider k = 1, . . . , N 1 twisted sectors, Tk . The CPT conjugates of Tk
appear in TNk . Thus, in non-prime orbifolds Z4 , Z6 , Z8 , Z12 , the sector TN/2 contains CPT
conjugates also.
For untwisted matter, multiplicity (P) is given just by counting all possible states. Multiplicity
in the twisted sector is more involved. The method of linear combination can come with complex
numbers. This is taken into account in the (generalized) GSO projector, which projects out nonphysical states. It can be read off from the one loop partition function of string [7,31];
Pk =
N1
N1
1 k l 2il0
1 k l l
, e
, ,
N
N
l=0
(23)
l=0
where N is the order of ZN orbifolds, and

k
NjL NjR j V 2 s2 + (P + kV ) V (s + ks ) s + integer, (24)
0 =
2
j
where j denotes the coordinates of the 6-dimensional compactified space running over
2,
3}
in complexified coordinates, and i = si sgn( i ) where sgn( ) = sgn( i )
{1, 2, 3, 1,
i
[29]. It turns out that NjR = 0 generically for the massless right mover states in our Z12I orbifold compactification. The ( m , k ) in Eq. (23) denotes the degenerate factor tabulated in
Table 1 [7,32]. For the sectors wound by Wilson lines, the Wilson line modified shift vector Vf
is used instead of V .
Table 1
Degeneracy factor ( k , l ) in the Z12I orbifold
kl
0
1
2
3
4
1
3
3
3
3
3
2
3
3
3
3
3
4
1
1
4
1
3
4
27
3
3
3
27
3
3
3
3
3
5
6
16
1
1
4
1
5
3
3
1
3
3
1
6
3
3
4
3
3
16
7
3
3
1
3
3
1
8
3
3
1
27
3
1
4 For the same chirality, i.e. L, we use () instead of (+++). So, we take U = (1, 0, 0).
1
9
3
3
4
3
3
4
10
3
3
1
3
3
1
11
3
3
1
3
3
1
55
In the presence of a Wilson lines ( a I ), the GSO projector Eq. (23) needs to be modified
as [29]
e2il0
NW 1
1
e2ilf ,
NW
(25)
f =0
where NW = 3 for Wilson line of order three in case of Z12I , and in Eq. (24) should also be
modified as

k
f =
NjL NjR j Vf2 s2 + (P + kVf ) Vf (s + ks ) s + integer,
2
j
(26)
NW 1
where Vf (V + mf a3 ), and Eq. (23) can be rewritten as Pk = f =0 Pk (f ) where
Pk (f ) =
N1
1 k l 2ilf
, e
.
N NW
(27)
l=0
Note that in the Z12I model, f = {0, 1, 2} {f0 , f+ , f } and Pk (f0 ) = Pk (f+ ) = Pk (f ) for
k = 3, 6, 9.
3. Z12I model
We choose the following shift vector and Wilson lines

1 1 1 1 1 5 6
2 2
V=
0
000000 ,
4 4 4 4 4 12 12
12 12

2
1 1
a 3 = a 4 = 05 0
0 0 05 ,
3 3
3
a1 = a2 = a5 = a6 = 0,
(28)
(29)
which satisfies Eq. (7): it gives V 2 s2 = 12 modular 2 (integer)/12. Since the Wilson line is a
Z3 shift, it distinguishes three cases. These satisfy the following conditions [31,33]:

12 |V |2 |s |2 = 0 mod even integer,
12(V a3 ) = 0 mod integer,
12|a3 |2 = 0 mod even integer.
(30)
Thus, we consider the following effective shifts distinguishing twisted sector Tk to Tk0 , Tk+ , and
Tk (k = 0, 3, 6, 9) by
V0 V ,
1 1 1 1 1 5 2 4
4 4 4 4 4 12 12 12

2 2 8 5
0 ,
12 12 12

1 1 1 1 1 5 10 4
2 2 8 5
V V a3 =
0 .
4 4 4 4 4 12 12 12
12 12 12
V+ V + a 3 =
(31)
(32)
(33)
56
For future convenience, we list

1
V02 s2 = ,
2
5
V+2 s2 = ,
6
3
V2 s2 = .
2
(34)
3.1. Gauge group from untwisted sector

For the gauge multiplet, we search for roots satisfying P 2 = 2, P V = 0 and P a3 = 0, and
obtain the following unbroken gauge group

SU(5) U (1)X U (1)3 SU(2) SO(10) U (1)2
(35)
where we choose the U (1)X of flipped SU(5) as

QX = (2, 2, 2, 2, 2; 0, 0, 0) 08 .
(36)
For example, one can see that the following is the roots of SU(5):
(1 1 0 0 0; 0 0 0) nonzero roots among 24 of SU(5),
(37)
where the underlined entries allow permutations.

3.2. Matter from untwisted sector
3.2.1. Chirality
The chirality is determined by the 8 component SO(8) spinor of the Ramond sector of right
movers. It is labeled by s = (s0 , s ) = { 12 , 12 , 12 , 12 } with an even number of minus signs.
We define the 4D chirality as the one originating from the first entry of s denoted by or ,
i.e.

1 1 1
1
s = or , , ,
(38)
, s ,
2 2 2
2
= 2s0 = 2 or 2 .
(39)
Let us call = 1 (1) as right- (left-)handed, and s has three components in terms of 12 .
In the untwisted sector U , we have U1 , U2 and U3 as defined in Eq. (22). These Ui correspond
to the untwisted s = (), (++), (++), respectively [or to (+++), (+), (+),
respectively, since antiparticles can be used also].
The chirality in the twisted sector can be similarly defined by the 8 component SO(8) spinor
of the Ramond sector of right movers.
3.2.2. Spectrum
The massless matter fields are those with P V = 0 and satisfy the masslessness condition.
k
P must satisfy P 2 = 2, P a3 = integer and P V = 12
where k = 1, 2, . . . , 11. We will consider
k = 1, 2, . . . , 6 only, since the rest will provide their CT P conjugates. Here, it is sufficient to
look at 3 cases only, k = 1, 4, 5, to keep the GSO allowed states.
57
Table 2
Visible sector chiral fields from the U sector
P V
s , Ui
Visible states
SU(5) U (1)X
1
12
(++), U3
(+; +++)
5L
3
(+++; +)
10L
1
(+++++; +++)
1L
5
4
12
(++), U2
(1, 0, 0, 0, 0; 1, 0, 0)
5L
2
5
12
(+++), U1
(+++ + ; ++)
5R
3
(++; )
10R
1
(; ++)
1R
5
Table 3
Hidden sector chiral fields from the U sector
P V
4
12
s , Ui
Hidden states
[SU(2) SO(10)]
Label
(++), U2
(1, 1; 0, 0, 0, 0, 0, 0)
10
su
We have the convention that the highest weight of the complex conjugated representation is in
fact the lowest one so that all the weights of the complex conjugated representation are obtained
by adding simple roots.
1
Let denote the phase e2i/12 . For k = 1 or P V = 12
, the left movers obtain a phase . We
1
need an extra phase from the right movers, which is accomplished by e2iss where s =
1
12 (5 4 1) and s = ( ++). It is left-handed and allows U3 . For k = 4, s = ( ++) provides
the needed 4 ; thus it is left-handed and provides U2 . For k = 5, s = (+++) provides 5 ;
thus it right-handed and provides U1 . from the right movers is provided by s = (+)
which thus will couple to k = 11. It is right-handed. 4 from the right movers is provided by
s = (+) which will couple to k = 8. It is right-handed. 5 from the right movers is provided
by s = ( +) which will couple to k = 7. It is left-handed. These particles for k > 6 give the
antiparticle spectra. The chiralities and Ui s are shown in Tables 2 and 3.
1
4
7
Thus, the vectors for p V = 12
, 12
, and 12
give
(15 + 53 + 101 )L
U3 ,
(52 )L
U2 ,
(15 + 53 + 101 )L
U1 ,
(10 )L
U2
(40)
8
5
and their CT P conjugates appear in p V = 11
12 , 12 , and 12 . Here we listed only U (1)X charges
as boldfaced subscripts. Note that for the untwisted k = 6 sector there is no way to provide an
additional 6 at the massless level from the right movers.
4. Twisted sectors with Wilson line

1
(5, 4, 1) has three fixed points in the second torus for
The Z12I with the twist vector = 12
the prime order 1 and 5 twists. This is because it is the same as Z3 . For the first and the third
torus the origin is the only fixed point, viz. Fig. 1. For the other twists such as k = 4 and 6,
counting the number of massless states involves a more complicated nonvanishing projection
operator P k . But addition of Wilson lines can distinguish some fixed points. In fact, for Z12I
possible Wilson lines must satisfy 3a3 = 0, 3a4 = 0 and ai = 0 (i = 3, 4) so that any combination
58
of k(V + mf ai ) is another shift vector. Since Z12I allows 3a3 = 0 modular integer, the Wilson
line is a Z3 shift.
In the second torus, Wilson lines must be symmetric, a3 = a4 . Then, the kth twisted sector is
distinguished by kV , k(V + a3 ), and k(V a3 ), which is denoted as V ,
V = k(V + mf a3 ),
mf = 0, 1,
(41)
or
V = k{V0 , V+ , V }.
(42)
Since 3a3 = 6a3 = 0, 2V and 5V in the T2 and T5 sectors are equivalent to 2V a3 and
5V a3 , respectively. On the other hand, 4V in the T4 sector is equivalent to 4V a3 .
The masslessness condition is

(P + V )2 = 2ck 2
(43)
NjL j .
j
twist (k = 1, 2, . . . , 6), we have

210
192
144 , k = 4;
144 , k = 1;
216
210
2ck = 144 , k = 2;
144 , k = 5;
234
216
144 , k = 3;
144 , k = 6,
For the
for the left movers, and

11
24 , k = 1;
2ck = 12 , k = 2;
5
k = 3;
8,
1
3,
11
24 ,
1
2,
(44)
k = 4;
k = 5;
(45)
k = 6,
for the right movers.

For the sectors wound by Wilson lines, V are used instead of V0 . The untwisted sector k = 0
and twisted sectors for k = 3, 6, 9 are not affected by Wilson lines since the Wilson line condition,
3a3 = 0, makes it trivial. So, for k = 3, 6, 9, there is the additional condition, (P + kV ) ai = 0,
which is applicable to T6 only in our case. For k = 3, 6, 9, the multiplicity for each twisted sector
k(V + mf a3 ) is P = 13 Pk .
The formula for multiplicity, Eq. (23), is the GSO allowed number of states.5 For non-prime
orbifolds such as Z12 , the multiplicity (23) is nonvanishing even if were not 1. Only for those
with pure Z12 twists, i.e. k = 1, 2 and 5, the multiplicity is counted by those with the vanishing
phase.
The twisted sectors for k = 3, 6, 9 are not affected by the additional Wilson lines since
3a3 = 0. Note that the untwisted sector also is not distinguished by Wilson lines, but Wilson
lines give the modular invariance condition P ai = 0 in the untwisted sector. By the same token,
in the sectors where 3a3 = 0 (k = 0, 3, 6, 9, with 0 corresponding to the untwisted sector), the
modular invariance condition restricts Wilson lines [34],
(P + kV ) a3 = 0 mod Z,
k = 0, 3, 6, 9.
(46)
5 For the prime orbifold Z , the multiplicity is just 1 (1 + + 2 ) which can be either 1 for = 1 or 0 for =
3
3
e2 i/3 . So in Z3 it is sufficient to count those with the vanishing phase, i.e. (P + V ) V (s + ) = 0. It is so also
in the kth twisted sector of Z12I if (
k , l ) are the same for all l.
59
Table 4
Left-handed massless states satisfying (P + 6V ) a3 = 0 mod Z in T6
P + 6V
(53 ; 12 , 0, 0)(08 )
(101 ; 12 , 0, 0)(08 )
(15 ; 12 , 0, 0)(08 )
8
(53 ; 1
2 , 0, 0)(0 )
8
(101 ; 1
2 , 0, 0)(0 )
1
(15 ; 2 , 0, 0)(08 )
(10 ; 0, 12 , 12 )(08 )
(N L )j
P6
Labels
1
2
T 65
T 610
1
2
T 61I
T 65
T 610
T 61J
h1
h2
h3
1
6
1
3
1
6
13
0
0
13
11
11
1
8
(10 ; 0, 1
2 , 2 )(0 )
1
6
1
2
1
3
h4
h 1
13
1
2
1
3
11
h 2
h 3
1
6
h 4
13
11
But other twisted sectors are not affected, in particular the k = 1, 2, 4, 5 sectors. For these sectors,
the multiplicity for each of k(V + mf a3 ) is 13 P k where P k is given in Eq. (23).
Let us present twisted sectors for k = 6, 1, 2, 3, 4, 5 in order. The k = 6 twisted sector T6 contains the CT P conjugates in T6 again. The spectra in the k = 1, . . . , 5 twisted sectors accompany
their CT P conjugates in k = 11, . . . , 7.
4.1. Twisted sector T6

j = 3 , where P is the
The massless condition for the left mover is 12 |P + 6V |2 + j (N L )j ()
4

E8 E8 weight vectors. The left mover states satisfying the massless condition always appear
vector-like in the T6 sector. In general, however, they carry different phases from those of the
counterpart states with opposite quantum numbers. Thus, chiral matter spectrum is possible even
in T6 after imposing the GSO projection by Eq. (23). In view of (46), we additionally require
(P + 6V ) a3 = 0 mod Z.
(47)
The massless states with left-handed chirality satisfying this constraint, and their multiplicity
numbers determined by P6 are listed in Table 4.
For simplicity, here we employed the following abbreviations for the SO(10) spinors and
neutral singlets under SU(5) U (1)X ,
53 (+),
101 (+++),
15 (+++++),
(48)
53 (++++),
101 (++),
15 (),
(49)
10 (0, 0, 0, 0, 0).
(50)
60
From the T6 sector, thus we have the following massless states,

L L
L L
L
L
10L
1 + 3 101 , 101 + 2 53 , 53 + 2 15 , 15 + 22 neutral singlets
+ CT P conjugates.
(51)
L
L
While the multiplicity number for 10L
1 is 4, the multiplicity of 101 is 3. One of 101 s provides
L
c
c
one generation of the MSSM matter, {Q, d , }. The remaining vector-like pairs of {10L
1 , 101 }
could be utilized to break SU(5) U (1)X into the MSSM gauge group.
We present the calculation in detail for the first row. The 6 twist vectors are

1
1
= 6s =
, 0,
,
2
2

1 1 1 1 1 1
0 0 (0 0 0 0 0 0 0 0).
V = 6V
(52)
2 2 2 2 2 2
In the 6 twisted sector, consider P = (08 )(08 ). With some shifts, P can be ([1]5 ; 2, 3, 0)
3
(1, 1; 6, 05 ) . Then, from (43) and (44) we require (P + 6V )2 = 2c6 = 216
144 = 2 which is
certainly satisfied with N L = 0. Therefore, the state P = (08 )(08 ) is massless. The GSO projection is given when combined with the right movers. The masslessness condition for right movers
is (s + 6)2 = 12 ,6 from which we will determine the chirality.
Let us calculate the masslessness condition of the left movers first, the multiplicity and the
chirality.
The masslessness condition for left movers becomes

3
NjL j = .
(P + V )2 + 2
(53)
2
j
Thus, the vectors satisfying Eq. (53) with Eq. (52) constitute the representation 5 L . These weights
are
15 :
(0 0 0 0 0 0 0 0),
53 :
(1 0 0 0 0 1 0 0),
101 :
(1 1 1 0 0 1 0 0),
(54)
where we calculated U (1)X charges as (P + V ) QX . Their CT P conjugates are

(1 1 1 1 1 1 0 0),
(1 1 1 1 0 0 0 0),
(1 1 0 0 0 0 0 0).
Note that for P
(55)
= (08 )(08 ),
5
(P + 6V ) V = ,
6
1
s = .
4
(56)
To obtain the chirality , we look at s = ( 12 , s ) allowing nonvanishing multiplicities. Then the

s 1 k(V 2 s2 ). With k = 6, we have
phase of 6 is found as (P + V ) V (s + )
2
6 See Appendix D of [7].
61
Table 5
The multiplicity for s 2 = 1 and (P + 6V ) V = 16 . + and denote + 12 and 12 , respectively, and ( ) is R(L-)handed. Note that (P + V ) V 1 k(V 2 s2 ) = 1
2
1
2
s = (r + )
s s
phase
Multiplicity
+++
5
12
4 2
12
1
12
2
12
5
12
6
12
1
12
1
12
4
12
5
12
++
++
1
12
4
12
++
0 2
3 2
12
1
s2 ) = 32 12 , and the phase of is 2 times ( 12
s s ), and hence we obtain the
multiplicity listed in Table 5. Then, we read the chirality , or or , from the first entry of
s = ( 12 |s1 s2 s3 ) if that chirality is allowed by the right mover condition, which is shown in the
first column of Table 5. If the first column does not satisfy the right mover condition, we should
search for higher s0 , which will be discussed shortly. The components of the vector s are the last
three 12 s of s. The number of massless states are given by P 6 . For these, we use the Euler
numbers ( k , l ) given in Table 1 [7]. Therefore, we obtain
1
2
2 k(V
P6 = P 6 =

1
1 + 6 16 + 43 + (1 + ) 1 + 3 + 6 + 9
12
(57)
which becomes 4, 2, 2, and 3 for = 1, 1, 3 = 1, +1, respectively, and 0 for the other cases.
We know that s = s + (integer). Now consider the right mover condition. For s 2 = 1, the
2 = 1 leads to s = ( 1 , 1 , 1 ) where we used the shifted of
masslessness condition (s + )
2
2
2
2
Eq. (52). The relevant ones appear in the third and fifth rows of Table 5. Among these one set is
the CT P conjugates of the other. The U (1)X charge is (P + 6V ) QX = 5. Thus, we obtain two
singlets as shown in the third row of Table 4.
Consider P = (1 0 0 0 0 1 0 0)(08 ) and P = (1 1 1 1 0 0 0 0)(08 ). For P =
(1 0 0 0 0 1 0 0)(08 ),
1
(P + 6V ) V = ,
6
1
(58)
s = .
4
s 1 k(V 2 s2 ) = ( 7 s s ). We
For k = 6, the phase becomes (P + V ) V (s + )
2
12
8
add 12 2 to the fourth column entries of Table 5. The masslessness condition chooses s =
( 12 , 12 , 12 ), the third and fifth rows again, leading to the multiplicity 2. (1 0 0 0 0 1 0 0)
is 5 whose U (1)X charge is 3. For P = (1 1 1 1 0 0 0 0)(08 ),
1
s = .
(59)
4
s 1 k(V 2 s2 ) = ( 11 s s ) =
Now the phase becomes (P + V ) V (s + )
2
12
1
( 12 s s ). Thus we obtain the fourth column entries of Table 5. The masslessness condition chooses s = ( 12 , 12 , 12 ), the third and fifth rows again, leading to the multiplicity 2.
1
(P + 6V ) V = ,
6
62
(1 1 1 1 0 0 0 0) is 5 whose U (1)X charge is 3. The above four cases are shuffled to list
the CT P conjugates together,

5: (P + 6V ) = 12 12 12 12 21 12 0 0 , , QX = 3,
L
53
(60)
5: (P + 6V ) = 1 1 1 1 1 1 0 0, , QX = 3,
2 2 2 2 2
2
1 1 1 1 1 1

5: (P + 6V ) = 2 2 2 2 2 2 0 0 , , QX = 3,
5L
(61)

3
5: (P + 6V ) = 12 12 12 12 21 12 0 0 , , QX = 3.
Thus we obtain the first and fourth row entries of Table 4.
Next consider 10s and 10s. So consider P = (1 1 0 0 0 0 0 0)(08 ) and P =
(1 1 1 0 0 1 0 0)(08 ). For P = (1 1 0 0 0 0 0 0)(08 ), we have
1
(P + 6V ) V = ,
3
1
s = .
4
(62)
s 1 k(V 2 s2 ) = ( 5 s s ) = ( 7 s s ).
The phase becomes (P + V ) V (s + )
2
12
12
6
We add 12 2 to the fourth column entries of Table 5. The masslessness condition for right
movers chooses s = ( 12 , 12 , 12 ), the third and fifth rows again, leading to the multiplicity 3
and 4, respectively. For P = (1 1 1 0 0 1 0 0)(08 ), we have
1
(P + 6V ) V = ,
3
1
s = .
4
(63)
s 1 k(V 2 s2 ) = ( 1 s s ) = ( 11 s s ).
The phase becomes (P + V ) V (s + )
2
12
12
2
2 to the fourth column entries of Table 5. The masslessness condition chooses
We add 12
s = ( 12 , 12 , 12 ), the third and fifth rows again, leading to the multiplicity 4 and 3, respectively.
These four cases are shuffled to list the CT P conjugates together,

10: (P + 6V ) = 12 12 21 12 21 12 0 0 , , P6 = 3, QX = 1,
10L
(64)

1
10: (P + 6V ) = 12 12 12 21 21 12 0 0 , , P6 = 3, QX = 1,

10: (P + 6V ) = 12 12 12 12 21 12 0 0 , , P6 = 4, QX = 1,
L
101
(65)
10: (P + 6V ) = 1 1 1 1 1 1 0 0, , P6 = 4, QX = 1.
2
2 2 2 2 2
Thus we obtain the second and fifth row entries of Table 4.
Other singlets with nonvanishing oscillators are also allowed, which are shown in Table 4.
The massless condition for the right mover in the T1 sector is (s + s )2 = 11
24 . It allows only
1
one right-handed state s = ( ), which gives (s + s ) s = 8 .
Since k2 (V 2 2 ) = 14 for k = 1, the phase in Eq. (24) is given by 0 = (P + V ) V +

1
L
j (N )j ()j 8 .
The T1 sector is distinguished by Wilson lines: V0 = V , V+ = V + a1 , and V = V a1 .
These sectors are denoted as T10 , T1+ and T1 .
63
Note the degeneracy factor ( k , l ) for k = 1, 2, 5 in Table 1. They take the same value 3
along each horizontal line, as in the prime orbifolds such as Z3 and Z7 . Thus, only the states with
vanishing phase turn out to survive the projection by Eq. (23) in the T1 , T2 , T5 (and also in T11 ,
T10 , T7 ) sectors.
The masslessness condition for the left movers (43) gives
(P + kV )2 = 2
NjL j + 2ck .
(66)
For k = 1 we have 2c1 = 35

24 . The states satisfying the massless condition and (P + V0 ) V0 +

1
0
L
j (N )j ()j = 8 for T1 are listed in Table 6. The multiplicity 3 reduces to 1 due to the distinction by Wilson lines: V0 , V+ , and V .
j= 7
In the T1+ sector with V+ = V + a3 , only the states with (P + V+ ) V+ +(N L )j ()
24
survives the GSO projection by Eq. (23), which are listed in the middle part of Table 6.
j=5
In the T1 sector with V = V a3 , only the states with (P + V ) V + (N L )j ()
8
survives the GSO projection by Eq. (23), which are listed in the lower part of Table 6.
The CT P conjugates of T1 appear in T11 .
Table 6
1 4
3 1 4
Chiral matter fields satisfying 0, = 0 in the T10 and T1 sectors. Here, 51/2 ( 3
4 , [ 4 ] ), 51/2 ( 4 , [ 4 ] ), and
1
1
1
1
1
5
1

15/2 ( 4 , 4 , 4 , 4 , 4 ). There are two SU(2) doublets with ( 6 , 6 ; )
P +V
6
1 1 6
(51/2 ; 7
12 , 12 , 0)( 6 , 6 ; 0 )
5 , 6 , 0)( 1 , 1 ; 06 )
(51/2 ; 12
12
6 6
6 )( 1 , 1 ; 06 )
(51/2 ; 1
,
0,
12
12 6 6
6 , 0)( 1 , 1 ; 06 )
(15/2 ; 7
,
12 12
6 6
5 , 6 , 0)( 1 , 1 ; 06 )
(15/2 ; 12
12
6 6
6 )( 1 , 1 ; 06 )
(15/2 ; 11
,
0,
12
12 6 6
6 )( 1 , 1 ; 06 )
,
0,
(15/2 ; 1
12
12 6 6
(N L )j
P1 (f0 )
13
23
33
12 , 43
1+1
11 , 53 , {12 + 13 }
1+1+1
P + V+
(N L )j
P1 (f+ )
5 , 2 , 4 )( 5 , 1 ; 1 ; 05 )
(15/2 ; 12
12 12
6 6 3
5 , 2 , 4 )( 1 , 1 ; 2 ; 05 )
(15/2 ; 12
12 12 6 6 3
23
4 2 5 1 1 5
(15/2 ; 1
12 , 12 , 12 )( 6 , 6 ; 3 ; 0 )
13
4 2 1 1 2 5
(15/2 ; 1
12 , 12 , 12 )( 6 , 6 ; 3 ; 0 )
33
P + V
(N L )j
P1 (f )
4 2 1 1 2 5
(51/2 ; 1
12 , 12 , 12 )( 6 , 6 ; 3 ; 0 )
13
13
5 , 2 , 8 )( 1 , 1 ; 2 ; 05 )
(15/2 ; 12
12 12 6 6 3
2 4 1 1 2 5
(15/2 ; 7
,
12 12 , 12 )( 6 , 6 ; 3 ; 0 )
8 2 1 1 2 5
(15/2 ; 1
12 , 12 , 12 )( 6 , 6 ; 3 ; 0 )
64

The massless condition for the right mover in the T2 sector is (s + 2s )2 = 12 . It allows only
one right-handed state s = ( ), which gives (s + 2s ) s = 16 = 56 .
Since k2 (V 2 2 ) = 12 for k = 2, the phase of in Eq. (24) is given by 0 = (P + 2V )

j + 1 . The T2 sector is also distinguished by Wilson lines: V0 = V , V+ =
V + j (N L )j ()
3
V + a3 , and V = V a3 . These sectors are denoted as T20 , T2+ and T2 .
Since ( k , l ) with k = 2 in Eq. (23) takes the same value 3 along the horizontal line,
only the states with 0, = 0 survive the projection operator in the T2 sector. In the T20 sec
j = 1
tor, hence, the massless states satisfying the condition (P + 2V ) V + j (N L )j ()
3

+
j = 0 and
are selected. Similarly, in the T2 and T2 sectors, (P + 2V+ ) V+ + j (N L )j ()

j = 1 should be chosen. The massless condition for the left
(P + 2V ) V + j (N L )j ()
3

j = 3 . The allowed shifted E8 E weight vectors
mover is |P + 2V() |2 + 2 j (N L )j ()
8
2
(P + 2V )s in the T2 sector are shown in Table 7.
Consider the first row of Table 7. Since

2V =
1 1 1 1 1 5
; 10
2 2 2 2 2 6

1 1 6
0 ,
3 3
Table 7
Chiral matter fields satisfying 0 = 0, + = 0, and = 0 in the T20 , T2+ , and T2 sectors, respectively. Here
10 (0, 0, 0, 0, 0) and the hidden sector 16 and 10 are not bold-faced not to be confused with observable sector repre
sentations. There are four SU(2) doublets of D1,2
P + 2V
(N L )j
P2 (f0 )
Labels
2 1 1 6
(53 ; 1
6 , 0 )( 3 , 3 ; 0 )
T 25
T 21
21 , 23
1+1
C10 , C20
12 , {11 + 13 }
1+1
C30 , C40
11
C50
13
C60
C70
(N L )j
P2 (f+ )
Labels
T 2+
12 , {11 + 13 }
1+1
D1+ , D2+
21 , 23
1+1
C1+ , C2+
(N L )j
P2 (f )
Labels
T 2
O10
21 , 23
1+1
D1 , D2
12 , {11 + 13 }
1+1
C1 , C2
2 1 1 6
(15 ; 1
6 , 0 )( 3 , 3 ; 0 )
(10 ;
(10 ;
(10 ;
(10 ;
(10 ;
1 , 1 , 1 )( 1 , 1 ; 06 )
3 2 2
3 3
1 , 1 , 1 )( 1 , 1 ; 06 )
3 2 2 3 3
2 , 1 , 1 )( 1 , 1 ; 06 )
3 2 2 3 3
2 , 1 , 1 )( 1 , 1 ; 06 )
3
2
2
3 3
1 , 1 , 1 )( 2 , 2 ; 06 )
3 2 2
3
3
P + 2V+
(10 ;
(10 ;
(10 ;
1,
3
1,
3
1,
3
1 ,
6
1 ,
6
1 ,
6
1 )( 1 , 1 ; 1 ; 16)
6
6
6
6
1 )( 2 , 1 ; 1 ; 05 )
6
3 3 3
1 )( 1 , 1 ; 2 ; 05 )
6 3 3 3
P + 2V
(10 ;
(10 ;
(10 ;
(10 ;
1,
3
1,
3
1,
3
1,
3
1,
6
1,
6
1,
6
1,
6
1 )( 1 , 1 ; 1 ; 10)
6
3 3 3
1 )( 2 , 1 ; 1 ; 05 )
6
3 3 3
1 )( 1 , 1 ; 2 ; 05 )
6
3 3 3
1 )( 2 , 2 ; 2 ; 05 )
6
3
3 3
O16
C3
65
P = (1 1 1 1 0; 1 1 0)(08 ) satisfies the masslessness condition. The SU(5) representation is 5. Then, we have

1 1 1 1 1
1
1 1 6
P + 2V = + ; 0 0
(67)
0
2 2 2 2 2
6
3 3
which gives QX = 3. From the previous discussion on the right mover condition, we obtain the
chirality 2 .
For the T2 sectors, only neutral fields under SU(5) U (1)X arise, which are tabulated in
Table 7. Thus, from T2 we obtain the following SU(5) U (1)X representations
{53 , 15 } + 12 neutral singlets.
(68)
The CT P conjugates are provided from T10 .

There are four SU(2) doublets and one 16 and one 10 of SO(10) .
The shifted momenta in the T3 sector must satisfy (P + 3V ) a3 = 0 mod Z, viz. Eq. (46). It
turns out that there is no massless states satisfying this condition.
The massless condition for the right movers in the T4 sector is (s + 4s )2 = 13 . Taking a
shifted 4s as ( 23 31 13 ), only the right-handed state s = ( ) satisfies this condition. So it is
left-handed. Now, (s + 4s ) s = 0.
Since k2 (V 2 2 ) = (integer) for k = 4, it does not contribute to the phase. So, of Eq. (24)

j . The T4 sector is again distinguished by
is given by f = (P + 4V() ) V() + j (N L )j ()
Wilson lines: V0 = V , V+ = V + a3 and V = V a3 . These sectors are denoted as T40 , T4+
and T4 .
( 4 , k ) of the T4 sector are (27, 3, 3, 3)3 , hence the multiplicity is
P4 =

3
1 + 4 + 8 8 + 1 + + 2 + 3 .
12
(69)
So, = e2i/12 , e2i/6 give P4 = 0. P4 = 9, 6, 6 for = 1, 1, e2i/4 . Considering Wilson

lines, the nonvanishing multiplicities are 3, 2, 2 in each T4 . The massless fields of T40 are listed
in Table 8.
Consider first two rows of Table 8. They are left-handed. The massless condition for the left
movers is (P + 4V )2 = 43 . Since

5
2 2 6
4V = 1 1 1 1 1; 2 0
0 ,
3
3 3
the state

0, [1]4 ; 2, 2, 0 1, 1; 06
satisfies the masslessness condition. It is 5 and QX = 2 since P + 4V = (1 0 0 0 0 )( ) .
Since (P + 4V ) V = 0, we obtain = e2i0 from (23) and the multiplicity is 3. The state 5
66
Table 8
Chiral matter fields in the T40 , T4+ and T4 sectors. There are twelve SU(2) doublets of d1,2,3
P + 4V
(N L )j
P4 (f0 )
Labels
2 1 1 6
(52 ; 1
3 , 0 )( 3 , 3 ; 0 )
T 45
1
2
T 45
13
1
4
1
2
1
4
s10
s20
s30
s40
2+2
s50 , s60
2+2
s70 , s80
2 1 1 6
(52 ; 1
3 , 0 )( 3 , 3 ; 0 )
1 6
(10 ; 23 , 02 )( 1
3 , 3 ;0 )
12
11
(10 ; 23 , 02 )( 23 , 23 ; 06 )
1 1 6
(10 ; 1
3 , 0, 1)( 3 , 3 ; 0 )
1
2
1
4
1
4
P + 4V+
(N L )j
P4 (f+ )
Labels
1
2
d1+
s1+
2 1 2 1 1 5
(10 ; 1
3 , 3 , 3 )( 3 , 3 ; 3 ; 0 )
d2+
2 1 1 1 2 5
(10 ; 1
3 , 3 , 3 )( 3 , 3 ; 3 ; 0 )
1
4
1
4
s2+
1 2 2 1 1 5
(10 ; 1
3 , 3 , 3 )( 3 , 3 ; 3 ; 0 )
d3+
1 2 1 1 2 5
(10 ; 1
3 , 3 , 3 )( 3 , 3 ; 3 ; 0 )
1
4
1
4
s3+
P + 4V
(N L )j
P4 (f )
Labels
1
2
d1
s1
2 1 2 1 1 5
(10 ; 1
3 , 3 , 3 )( 3 , 3 ; 3 ; 0 )
d2
2 1 1 1 2 5
(10 ; 1
3 , 3 , 3 )( 3 , 3 ; 3 ; 0 )
1
4
1
4
s2
1 2 2 1 1 5
(10 ; 1
3 , 3 , 3 )( 3 , 3 ; 3 ; 0 )
d3
1 2 1 1 2 5
(10 ; 1
3 , 3 , 3 )( 3 , 3 ; 3 ; 0 )
s3
1 1 6
(10 ; 1
3 , 1, 0)( 3 , 3 ; 0 )
(10 ;
(10 ;
2,
3
2,
3
(10 ; 23 ,
(10 ; 23 ,
1 ,
3
1 ,
3
1,
3
1,
3
1 )( 2 , 1 ; 1 ; 05 )
3 3 3
3
1 )( 1 , 1 ; 2 ; 05 )
3
3
3 3
1 )( 2 , 1 ; 1 ; 05 )
3
3 3 3
1 )( 1 , 1 ; 2 ; 05 )
3
3
3
3
1
4
1
4
with QX = 2,

2, [1]4 ; 2, 2, 0 1, 1; 06
1
also satisfies the masslessness condition. Since (P + 4V ) V = 12 , we obtain = e2i( 2 )

from (23) and the multiplicity is 2.
In the T4 sectors, the phases in Eq. (26) are respectively given by

j + 1,
N L j ()
3
j

j,
= (P + 4V ) V +
N L j ()
+ = (P + 4V+ ) V+ +
where 13 in Eq. (70) comes from 42 (V+2 s2 ).
(70)
(71)
67
It turns out that from the 4(V a3 ) sectors, only neutral fields under SU(5) U (1)X arise,
which are listed in Table 8.
Thus, the massless states in T4 (+T8 ) sectors are
52 + 2{52 , 52 } + 30 neutral singlets + CT P conjugates.
(72)
There are twelve SU(2) doublets.

In the T5 (and T5 ) sector of the Z12I orbifold, only the right-handed chirality states appear
as massless states from the right mover condition.
The massless condition for the right mover in the T5 sector is (s + 5s )2 = 11
24 . Taking a
1
4 5
shifted 5s as ( 12
12
),
only
the
right
mover
s
=
(+)
satisfies
this
condition.
So it is
12
right-handed. Now, (s + 5s ) s = 18 .
Note that k2 (V 2 2 ) = 14 for k = 5. So, the phase in Eq. (24) is given by 0 = (P + 5V )

j 1 . The T5 sector is again distinguished by Wilson lines: V0 = V , V+ =
V + j (N L )j ()
8
V + a3 , and V = V a3 . These sectors are denoted as T50 , T5+ and T5 . In the T5 sectors, only
l)s are the
the states with 0, = 0 survive the GSO projection Eqs. (23), (25) since all (5,
same.
Only neutral and vector-like pairs of SU(5) singlets arise in T5 . In the T5+ sector, only the

j = 1 survive the GSO projection Eq. (23). In the
states with (P + 5V+ ) V+ + j (N L )j ()
24

j = 3 survive. These are shown
T5 sector, only the states with (P + 5V ) V + j (N L )j ()
8
in Table 9.
The CT P conjugates of T5 appear in T7 .
5. Summary of matter spectra
Collecting all the flipped SU(5) model fields, we obtain the following:
U:
(15 + 53 + 101 )L
U3 ,
(52 )L
U2 ,
(15 + 53 + 101 )L
U1 ,
(10 )L
U2 ,
from the untwisted sector, and

L
T6 : 10L
+ 22{10 },
1 + 2(15 + 15 + 53 + 53 ) + 3(101 + 101 )
(73)
(74)
T2 :
L
1L
5 + 53 + 12{10 },
(75)
T4 :
L
5L
2 + 2(52 + 52 ) + 30{10 },
(76)
from the twisted sectors. The chiral matter resulting from this spectra constitutes three families
of quarks and leptons.
In addition, we obtain G-exotics and E-exotics from T1 and T5 sectors,
T1 :
2(51/2 )L + 2(5+1/2 )L + 7(15/2 )L + 7(1+5/2 )L ,
T5 :
2(5+1/2 )R + 2(51/2 )R + 7(1+5/2 )R + 7(15/2 )R .

16
(77)
(78)
12
From (2), we note that G-exotics carry Qem =

quarks and Qem =
E-exotics and E5
1
exotics with X = 2 have Qem = 2 . All these exotics can be removed if all U (1)s except the
68
Table 9

Chiral matter fields satisfying 0, = 0 in the T50 and T5 sectors. There are two SU(2) doublets with ( 56 , 1
6 ; )
P + 5V
7 , 0, 6 )( 1 , 1 ; 06 )
(51/2 ; 12
12
6
6
5
1 , 1 ; 06 )
)(
(51/2 ; 12 , 0, 6
12
6
6
1 , 6 , 0)( 1 , 1 ; 06 )
(51/2 ; 12
12
6
6
7 , 0, 6 )( 1 , 1 ; 06 )
(15/2 ; 12
12
6
6
6 1 1 6
(15/2 ; 5
12 , 0, 12 )( 6 , 6 ; 0 )
6
1 1 6
(15/2 ; 11
12 , 12 , 0)( 6 , 6 ; 0 )
1 , 6 , 0)( 1 , 1 ; 06 )
(15/2 ; 12
12
6
6
P + 5V+
4 2 5 1 1 5
(15/2 ; 5
12 , 12 , 12 )( 6 , 6 ; 3 ; 0 )
4 2 1 1 2 5
(15/2 ; 5
12 , 12 , 12 )( 6 , 6 ; 3 ; 0 )
1
4 5 1 1 5
(15/2 ; 12 , 2
12 , 12 )( 6 , 6 ; 3 ; 0 )
1 , 2 , 4 )( 1 , 1 ; 2 ; 05 )
(15/2 ; 12
12 12
6
6
3
(N L )j
P5 (f0 )
11
21
31
12 , 41
1+1
13 , 51 , {11 + 12 }
1+1+1
(N L )j
P5 (f+ )
21
11
31
P + 5V
(N L )j
P5 (f )
1 , 2 , 4 )( 1 , 1 ; 2 ; 05 )
(51/2 ; 12
12 12
6
6 3
11
11
8 2 1 1 2 5
(15/2 ; 5
12 , 12 , 12 )( 6 , 6 ; 3 ; 0 )
7
4
1 1 2 5
(15/2 ; 12 , 12 , 2
12 )( 6 , 6 ; 3 ; 0 )
1 , 2 , 8 )( 1 , 1 ; 2 ; 05 )
(15/2 ; 12
12 12
6
6 3
U (1)Y are broken at the GUT scale and a sufficient number of neutral singlets develop VEVs,
and hence there are not left with dangerous half-integer charged fields below the GUT scale.
It will be explained in Section 6.1. The lightest of these half-integer charged fields (LHIC) is
absolutely stable since all the light observable SM fields (including color singlet composites) are
integer charged. If the mass of LHIC is much larger than the reheating temperature after inflation,
we expect that most of LHIC are diluted away by inflation.
For the hidden sector, there appear 20 SU(2) doublets and one 16 and one 10 of the hidden
SO(10) .
6. Yukawa couplings
Nonvanishing couplings of vertex operators are constructed by satisfying the Z12I symmetry [29,35,37]. It is summarized in [7]. In our notation for the shift vector, basically it amounts
for the operator OA OB OC to satisfy
Invariance under the group space shift V ,
Invariance under the internal space shift s .
and
(79)
(80)
For the shift V , it is easy to check the modular invariance: The relevant vertex operators have only
to satisfy the gauge invariance. The invariance under the shift s belongs to a generalized Lorentz
shift and the condition is sometimes called the H -momentum conservation. The (bosonic) H -
69
momentum is defined as

Ri = (s + ks + r )i NiL NiL ,
i = {1, 2, 3},
(81)
1 1
1 1 1
where r is ( 1
2 , 2 , 2 ) for left-handed states ( r ) and ( 2 , 2 , 2 ) for right-handed states
( r+ ). As discussed earlier, s should satisfy the mass-shell condition, |( 12 , s ) + ks |2 = 2ck .
Ri can be interpreted as a discrete R-charge. Thus, neglecting oscillator numbers, the H -momenta
for Z12I twist are
U1 :
T1 :
T4 :
(1 0 0),
7 4 1
12 12 12

1 1
3 3 3 ,
1
U2 :
,
(0 1 0),
U3 : (0 0 1),
1 4 1
3 1
T2 :
T3 :
6 6 6 ,
4 0 4 ,
1 4 7
1 1
T5 :
T6 :
12 12 12 ,
2 0 2 ,
(82)
which are used to check the generalized Lorentz invariance.

1
As an example, consider the T6 H -momentum, ( 1
2 0 2 ). It is derived in the following way.
The right mover mass-shell condition is
1
1 2
MR = (s + 6s )2 = 2c =
2
2
(83)
24 6
where 6s = ( 30
12 12 12 ). There are two solutions of Eq. (83),

5 3 1
5 5 1
s+ = ;
,
,
and s = ;
.
2 2 2
2 2 2
For the left-handed states appearing in Table 4, let us focus on s whose corresponding sum-toodd-integer solution is (3 2 0). So bosonization of s is s + r . Thus, H -momentum, in
1
analogy with P + kV , is (s + r ) + ks = ( 1
2 0 2 ) appearing in Eq. (82).
5 4 1
The H -momenta conservation conditions with s = ( 12
, 12 , 12 ) can be reformulated only in
terms of the bosonic H -momenta as follows:

R1 (z) = 1 mod 12,
R2 (z) = 1 mod 3,
R3 (z) = 1 mod 12,
(84)
z
where z ( A, B, C, . . .) denotes the index of states participating in a vertex operator. In addition,

m
space group selection rules requires a vertex operator with z-states in Tk f sector (k = 0 for the
untwisted sector) should satisfy

k(z) = 0 mod 12,
(85)
z
[kmf ](z) = 0 mod 3.
(86)
One can easily check the following cubic couplings are allowed in Z12I orbifold models fulfilling Eqs. (84) and (85), if NiL = NiL = 0 [35,37]:
U 1 U2 U3 ,
T 6 T 6 U2 ,
T4 T4 T4 ,
T2 T4 T6 ,
T1 T4 T7 .
(87)
Note that in considering the superpotential couplings, one should consider only the same chirality, and in our model there is no massless states from the T3 and T9 sectors.
70
Table 10
H -momenta for some combinations of neutral singlets under SU(5) U (1)X appearing in our model. All the combinations are neutral under all gauge symmetries in this model, and fulfill the space group selection rules
Comb. of singlets
h1 h 1
H -momenta
Comb. of singlets
H -momenta
(1 0 1)
h1 h 2
h1 h 4
(1 0 1)
(0 0 2)
H -momenta
h2 h 2
h3 h 3
(1 0 3)
h4 h 4
(3 0 1)
h2 h 3
h2 h 4
Comb. of doublets
H -momenta
Comb. of doublets
D1+ d2+ h 1
D + d + h 2
1 2
Comb. of singlets
C1+ s2+ h 1
C + s + h 2
1 2
C1+ s2+ h 3
C1+ s2+ h 4
C2+ s2+ h 1
C2+ s2+ h 2
C2+ s2+ h 3
C2+ s2+ h 4
(1 0 1)
(2 0 0)
(2 0 2)
(1 3 2)
D1+ d2+ h 3
D + d + h 4
H -momenta
Comb. of singlets
H -momenta
(1 1 0)
C1 s2 h1
(1 3 0)
(1 3 0)
(1 1 2)
(2 1 1)
(0 1 1)
(1 1 2)
(1 1 0)
(0 1 1)
(2 1 1)
1 2
C1 s2 h2
C1 s2 h3
C1 s2 h4
C2 s2 h1
C2 s2 h2
C2 s2 h3
C2 s2 h4
(0 3 1)
(2 3 1)
(1 3 2)
(0 3 1)
(2 3 1)
(0 1 1)
(0 1 1)
(1 1 0)
(1 1 0)
For future convenience, we display the H -momenta for combinations of some singlets appearing in our model in Table 10.
In Ref. [30], it has been shown that one can always find a vacuum where only neutral singlets
under a symmetry of interest develop VEVs of string scale preserving the F and D flatness conditions. This is possible because (a) once a superpotential term W is allowed by the selection rules,
then W N+1 is also allowed in the ZN orbifold model, and (b) there exists in general a transformation rescaling the VEVs leaving intact the above F flatness conditions but making accordingly
D flatness conditions satisfied by adjusting VEVs. In our model, there are enough superpotential
terms constructed with only neutral singlets, e.g. W = WS + WS13 + WS25 + , where

WS = C1+ s2+ h 1 C1 s2 h4 + C1+ s2+ h 4 C1 s2 h1

+ C1 s2 h2 C2 s2 h1 + C1 s2 h1 C2 s2 h2 +
where the string scale is set Mstring = 1. We assume a vacuum where only neutral singlets develop
large string scale VEVs.
6.1. Flipped SU(5) spectrum
There exist (101 , 101 ) whose VEV (in the SU(2) singlet direction c ) breaks the flipped
SU(5) down to the standard model. Also, there exist the needed electroweak Higgs fields
(52 , 52 ).
In T1 and T5 , there appear G-exotics and E-exotics. These are removed by T1 T4 T7 couplings
6 1 1
6
of Eq. (87) via 10 (T4 ) and other singlets VEVs. For instance, (15/2 ; 1
12 , 0, 12 )( 6 6 ; 0 ) with
4 2
1 1 2 5
L
NjL = 11 in T10 , and (15/2 ; 7
12 , 12 , 12 )( 6 6 ; 3 ; 0 ) with Nj = 11 in T7 get a mass from
71
2 4 1 1 2 5
L
the coupling with s1+ in T4+ . Similarly (15/2 ; 7
12 , 12 , 12 )( 6 6 ; 3 ; 0 ) with Nj = 13 in T1 ,
+
6
1 1 6
0
L
and (15/2 ; 1
12 , 12 , 0)( 6 6 ; 0 ) with Nj = 13 in T7 achieve a mass also from s1 . It turns out
that the contributions of oscillator numbers carried by the other states to the H -momenta can
be always cancelled by (multi-) products of C1+ s3+ h3 , C2+ s2+ h 3 , C1 s2 h3 , C1 s2 h 1 , etc. When
NjL = {12 + 13 } is involved in a T1 T7 T4 vertex operator, for instance, one could additionally
multiply C1+ s2+ h 4 to the vertex operator in order to cancel the oscillator number contribution.
When NjL = 43 is involved, e.g. (C1 s2 h3 )4 needs to be multiplied.
In T6 and T4 , there appear vector-like representations (101 + 101 )s, (15 + 15 )s, and
(52 + 52 )s. These are removed by the survival hypothesis that all vector-like representations
are removed at the GUT scale [36]. Indeed, it is so in our model by allowing large VEVs to
L
neutral singlets. Thus, we obtain just only one 5L
2 from T4 and 52 from the untwisted sector.
One can anticipate that this may happen if the twisted sectors provide large Yukawa couplings.
Indeed, such couplings are present.
The fields 2(52 + 52 )L of T4 can be removed by T4 T4 T4 couplings where one T4 is s40
in Table 8. The vector-like representations in T6 are also removed by the couplings, e.g. with
(C1+ s2+ h 1 )(C1 s2 h1 ) and (C2+ s2+ h 3 )(C1 s2 h3 ), etc. Thus, {2(15 + 15 + 53 + 53 ) + 3(101 +
101 )}L and 2(52 + 52 )L are expected to be removed at the GUT scale. In all these, several
singlets with QX = 0 are expected to develop GUT scale VEVs. Then, there result the following
light fields,
U:
R
R
(15 + 53 + 101 )R
U1 + (15 + 53 + 101 )U3 + (52 )U2 ,
(88)
T6 :
10R
1,
R
15 + 5R
3 ,
5R
2.
(89)
T20 :
T4 :
(90)
(91)
These constitute the three families and one pair of Higgs quintets. Relabeling R to L, we can
obtain the standard form of left-handed W interactions. In the remainder, however, we will
keep R to compare with the entries of tables.
It is interesting to note that the pair of Higgs quintets, (52 + 52 ), survives the above analysis.
Certainly, it is not allowed to write MGUT 52 52 since there is no coupling of the form U2 T4 .
In fact, it is not easy to construct a 52 52 consistent with the H -momentum conservation. Thus
Higgs doublet mass can be far below the GUT scale, which is a good thing. But the colored
scalars in (52 + 52 ) must be removed at high energy scale toward MSSM. It is achieved by the
doublettriplet splitting mechanism we discuss below.
6.2. GUT breaking
The GUT breaking in the flipped SU(5) model proceeds by 101 = 101 = MGUT , which
is a D-flat direction. This vector-like representation, 101 + 101 , is present in the T6 sector. Note
that the H -momentum of (101 101 )L is (1, 0, 1).
The T6 sector also contains twenty two Qem = 0 singlets. Some combinations of them, e.g.
h1 h 2 in Table 10 also give the H -momentum of (1, 0, 1). h3 h 4 , h2 h 1 , h4 h 3 also provide the
same H -momentum. The GUT scale VEV of them could induce 101 = 101 = MGUT along
an F-flat direction, for instance, through a non-renormalizable superpotential,

H
W = D1+ d2+ h 1 C1+ s2+ h 1 10H

(92)
1 101 h1 h2 h3 h4 +
72
with D1+ or d2+ vanishing. Note that there are 20 SU(2) doublets. Thus, the gauge coupling
of SU(2) would not blow up at lower energies. The neutrino direction of 101 , 101 allows
the symmetry breaking
SU(5) U (1)X SU(3)c SU(2) U (1)Y .
(93)
6.3. Doublettriplet splitting

H
Let us call the three vector-like ten and tenbars of Eq. (74) as Higgs fields, 10H
1 and 101 .
Among the three two are purely vector-like and removed at the GUT scale. One remaining vectorlike pair joins the Higgs mechanism. Let us consider this remaining pair.
H
c c
c c
The Higgs 10H
1 and 101 contain {q, d , }+{q, d , } in terms of the standard model quanc
c
tum numbers. { , } obtain GUT scale masses from Eq. (92) when SU(5) U (1)X is broken.
H
{q, q} contained in 10H
1 and 101 are absorbed by the heavy gauge sector. On the other hand,
c
c
{d , d } still remain light. In order to make the standard model vacuum stabilized, somehow they
should get superheavy masses.
H
H
H
Let us consider W = 10H
1 101 5h + 101 101 5h , where 5h and 5h indicate the five-plets
H
inducing electroweak symmetry breaking [38,39]. When 10H
1 and 101 develop VEVs in the
H
H
c
c
c
c
included in 5h , 5h pair
= = MGUT direction, {d , d } in 101 , 101 and triplets (D, D)
up to be superheavy [40]. This achieves the doublettriplet splitting in flipped SU(5).
H
L
This mechanism can be realized also in our model. The [10H
1 101 5h ] term arises from
H
L
T6 T6 U2 , which satisfies the H -momentum conservation. The [10H
1 101 5h ] is still also possible
+ u
H
H
L
from highly non-renormalizable interactions, e.g. [101 101 5h ] (s1 s1 s )(C2+ s2+ h 3 )(C2 s2 h3 ),
which satisfies all selection rules discussed above. Thus we should assume cutoff scale VEVs for
the neutral singlets.
Thus, we obtain the so-called MSSM spectra with one pair of Higgs doublets at cubic level.
However, the Higgs doublets would obtain mass since there exist a lot of singlets. Indeed, there
exist many couplings. The simplest two terms for are

u 0 2
8
1
1 1 6
(52 ; 1, 0, 0) 0 U
s s2 + f s20 s40 [T 0 ]2
, 0, 0
,
;0
52 ;
2
4
3
3 3
T40
(94)
0
where f is a relative strength. Before, s4 was needed for charged lepton masses. On the other
hand, s20 was needed for mixing of charged leptons. Let us set s20 a free parameter. Assuming that
52 and 52 obtain VEVs, the F-flat direction chooses s20 = f s40 /2s u so that the term is of
order f2 s402 /4s u . This can be linked to the charged lepton masses. For example, if we choose
s40 105 from electron mass and s u O(1), we require f = O(102 ) to obtain a TeV scale
term. This fine tuning is a huge improvement over a fine tuning of 1015 . Since we neglected
many higher order terms, this is just an illustration of smoothing the fine tuning problem. The s20
mimics the axion multiplet of the solution [41].
6.4. Fermion masses

There is one cubic coupling relevant for u-type quarks and Dirac neutrinos, T6 T2 T4 , which
is interpreted as the top quark and (Dirac) tau neutrino Yukawa coupling,
t quark + (Dirac) neutrino coupling:
[101 53 52 ]R .
(95)
73
Non-renormalizable terms allow the other u-type quark (Dirac neutrino) Yukawa couplings. Most
of all, through the couplings with C50 , C60 in T20 and h1 , h 3 in T6 , [101 (T6 )53 (U1 )52 (T40 )]L ,
[101 (T6 )53 (U3 )52 (T40 )]L , [101 (U3 )53 (T20 )52 (T40 )]L , and [101 (U1 )53 (T20 )52 (T40 )]L are
allowed. With h1 C50 , h 3 C60 , h 3 C50 , and h1 C60 , the masses and mixing for the first two families of
u-type quarks and Dirac neutrinos are possible.
The bottom quark mass arises in terms of U2 Higgs doublet. Indeed, such a coupling is
present from T6 T6 U2 . If the coupling strength of T6 T2 T4 and T6 T6 U2 are comparable, a large
tan is needed to obtain mt /mb 35. But the couplings depend on the location of the respective
fields and one may treat the ratio as a free parameter of order 1 [42].
In fact [101 (U1 )101 (U3 )52 (U2 )]R exists. For smallness of it we could assume a proper volume of the 6-dimensional compact space. As an alternative way, one could consider (1, 1) and
(2, 2) components of the d-quark mass matrix. They can be induced with h1 h1 and h 3 h 3 . By
h1 and h 3 , the mixing between the first two and the third families of d-type quarks are also
permitted.
For the mass of charged lepton in the twisted sector T20 , we need a coupling containing
[15 (T20 )53 (T20 )52 (U2 )]L . This coupling is possible by being supplemented with (s20 )2 . On the
other hand, cubic couplings [15 (U1 )53 (U3 )52 (U2 )]R [15 (U3 )53 (U1 )52 (U2 )]R are possible.
It may be that is placed in the untwisted sector. In this sense, the leptonic sector does not go
parallel to the quark sector. The b unification is not achieved in this model. (1, 1) and (2, 2)
components of the charged lepton mass matrix are generated via, e.g. h 1 h 1 and h3 h3 . The
mixing terms between the first two and the third families of charged leptons should be mediated
by h3 s20 and h 1 s20 .
In flipped SU(5), Majorana neutrino masses arise from [101 101 101 101 ]L . The H momentum of this operator is (2, 0, 2). Thus this operator can be induced when supplemented,
e.g. by the coupling with (C1+ s2+ h 3 )(C2+ s2+ h 1 )[(C1 s2 h1 )(C2 s2 h3 )]2 . The other components of
the Majorana neutrino mass matrix should require additional singlet VEVs such as h1 , h 3 , (h1 )2 ,
and (h 3 )2 . If heavy Majorana neutrino masses are around 1014 GeV, thus we obtain the mass
of order 0.1 eV,
m2top
0.1 eV
(1014 GeV)
for mtop = mDirac neutrino , which is valid in flipped SU(5).
m
(96)
6.5. R-parity
If we consider cubic couplings Eq. (87), we can define an R-parity in the standard way,
R = 1 for matter fermions and R = +1 for Higgs bosons.
Firstly, consider the coupling 101 (U1 )101 (U3 )52 (U2 ). A nontrivial parity can be defined as
R = 1 for 101 (U1 ), 101 (U3 ), 53 (U1 ), 53 (U3 ), 15 (U1 ), 15 (U3 ),
(97)
R = +1 for 52 (U2 ).
(98)
As discussed above, mixing between the first two families of matter in the untwisted sector
and the third family of matter in T5 and T20 are always possible if VEVs of some neutral singlets
are supposed. Such neutral singlets should also be inert under all symmetries relevant at low
energies. Otherwise, symmetries must be broken at low energies by their VEVs. Even with the
mixing terms between untwisted and twisted matter fields, the R-parity relevant in low energies
74
can still be defined by assigning R = 1 for the neutral singlets developing VEVs and
L
L
R = 1 for 10L
1 (T6 ), 53 (T2 ), 15 (T2 ).
(99)
Then, the allowed Yukawa coupling T6 T2 T4 determines

R = +1 for 52 (T4 ).
(100)
Thus, R-parity can survive down to low energies and hence R-parity conservation for proton
longevity is fulfilled in the present model.
7. Electroweak hypercharge in flipped SU(5)
7.1. GUT value of weak mixing angle
Flipped SU(5) was originally considered as a subgroup of SO(10) [9], which can be called
SO(10)-flipped SU(5). In this case, the GUT value for sin2 W should be 38 . In this case, symmetry breaking must proceed via adjoint Higgs field. The shift vector must take a form

(0 0 0 0 0 x y z)( ) , or

V = 1 1 1 1 1
(101)

2 2 2 2 2 x y z ( )
so that an SO(10) group is obtained.
String flipped SU(5), for example Eq. (28) with five 14 s, is basically different from SO(10)
flipped SU(5) even though it includes SO(10) flipped SU(5) if the shift vector takes the form
Eq. (101). If the electroweak SU(2) U (1)Y is embedded in a simple group in a field theory
GUT, the GUT value of sin2 W is 38 as calculated from
sin2 W =
Tr T32
.
Tr Q2em
(102)
Many possibilities of sin2 W appear because of many possibilities of embedding the electroweak
hypercharge Y in an Abelian group. In the E8 E8 heterotic string, the embedding of Y in U (1)s
is basically looked from the untwisted sector spectrum, which sets the embedding in E8 . Thus, in
string compactification, sin2 W depends on the spectrum in the untwisted sector [7]. Therefore,
if the untwisted sector spectrum includes 16flip , then sin2 W is the same as that of SO(10). On
the other hand, if 16flip cannot be obtained from the fields in the U sector, sin2 W = 38 is not
warranted. In the present model, a complete multiplet 16flip appears in the U sector, we obtain
sin2 W = 38 . This is explicitly shown below.
The electroweak hypercharge Y is a combination of SU(5) and U (1)X generators, Y =
1
5 (X + Y5 ). As will be shown in the next section, the U (1)Y gauge coupling is
gY2 =
g52
g 2
+ X 2,
15
25u
(103)
where u denotes an employed unit of U (1)X charges. So far we tacitly supposed u = 1, but
its absolute value should be determined by the string theory with g5 = gX at the compactification scale. Along the standard model direction, Eq. (103) is still valid above the flipped SU(5)
breaking scale. Thus at the compactification scale, where SU(2) gauge coupling g2 is identified
75
with g5 , the Weinberg angle is given by

0
sin2 W
=
1
1+
g52
gY2
1
1
1 + ( 15
1
)
25u2
(104)
Including the unit factor u, the U (1)X charge operator, QA

X 2(1, 1, 1, 1, 1, 0, 0, 0)
u can be expressed in terms of a proper orthonormal basis vector q5 [43],
(08 )
1
QA
X = c 5 q5 ,
2
(105)
where
c5 2 10u

1
and q5 (1, 1, 1, 1, 1, 0, 0, 0) 08 .
5
(106)
2 can be calculated in the string theory framework [7],

With this expression, g52 /gX
g52
2
gX
= c52 = 40u2 .
(107)
1
from
Hence, the string theory determines the absolute values of U (1)X charges with u2 = 40
g5 = gX at the compactification scale. With this unit, the bare value of the Weinberg angle is also
determined from Eq. (104):
3
0
sin2 W
= .
8
(108)
7.2. Embedding of Y in flipped SU(5)

A covariant derivative in flipped SU(5) includes the term

3
Y5 g5 A5 + XugX Ax ,
5
(109)
1
where Y5 = diag( 13 , 13 , 13 , 1
2 , 2 ) and X [= x diag(1, 1, 1, 1, 1)] denotes the U (1)X charge
operator employed in this paper, u a unit of U (1)X charge. So far u is tacitly assumed to be
unity, but generically it is not necessarily 1.
3
5 Y5
is one of SU(5) generator normalized with
Tr TSU(5) TSU(5) = A5 and Ax stand for the SU(5) and U (1)X gauge fields, respectively. After
symmetry breaking, one linear combination of A5 and Ax becomes the U (1)Y gauge field B of
the standard model. We introduce a mixing angle ,

cos sin
B
A5
=
,
(110)
Ax
C
sin
cos
1
2.
Eq. (109) is recast in the following form,

5 gX
1
1
tan u 15 cos g5 B + [C terms],
Y5 + X
5
5
3 g5
(111)
76
c , c . From
where C achieves a superheavy mass from the 10-dimensional Higgs VEVs H
H
this expression, one can read off the U (1)Y charges and coupling,

5 gX
1
1
tan u,
Y = Y5 + X
(112)
5
5
3 g5
gY
(113)
= 15 cos .
g5
Since Y ( c ) = 0 is the SM direction, Eq. (112) should be
1
1
Y = Y5 + X, or
5
5

5
g5
=
tan u.
gX
3
With Eqs. (113) and (115), the relation between g5 , gX , and gY is derived:
gY2 =
g52
g 2
+ X 2.
15
25u
(114)
(115)
(116)
8. Conclusion
We constructed a supersymmetric flipped SU(5) model from a Z12I orbifold compactification. The notable features of the model are:
From E8 , the only non-Abelian group is the needed SU(5). This is possible only for one shift
vector in Z12I . In this sense, it is the unique Z12I flipped SU(5) model.
Three families are obtained. The third family is located in twisted sectors while the first two
families are in the untwisted sector. Separating the third family from the first two families
enables a mass hierarchy of fermions.
There exists a doublettriplet splitting from a kind of the missing partner mechanism.
There results only one pair of Higgs doublets.
The R-parity is present.
H
Allowed Yukawa couplings can generate a GUT scale VEVs of 10H
1 and 101 for the GUT
breaking down to the standard model.
There exist Qem = 12 particles. But these form vector-like representations and most are
removed at the GUT scale. Thus, sin2 W is similar to that of SO(10) GUT.
In this paper, all the relevant Yukawa couplings are derived from string construction. So far,
we have not encountered any serious phenomenological problem. Successful Yukawa couplings
may be generated by appropriate GUT scale VEVs of singlet fields which are treated here as
free parameters. Finally, it is expected that a standard model can be derived from Z12 without
going through the intermediate stage of the flipped SU(5) with features discovered in the present
model [44]. In a future communication, we will tabulate all computer-searched Z12 orbifold
models.
Acknowledgements
We thank Ji-hun Kim for numerical checking of the spectra, and thank K.-S. Choi and
I.-W. Kim for useful discussions. One of us (J.E.K.) also thank the Kavli Institute for Theo-
77
retical Physics for the hospitality during this work was finished. This research was supported
in part by the National Science Foundation under Grant No. PHY99-07949. J.E.K. is also supported in part by the KRF grants, Nos. R14-2003-012-01001-0, R02-2004-000-10149-0, and
KRF-2005-084-C00001.
Appendix A. Flipped SU(5) with SU(4)
One can eliminate some exotic particles carrying QX = 12 , 52 observed in the model discussed in the main text (Model I) by employing more complicated shift vector and Wilson line,

2 2
1 1 1 1 1 5 6
, , , , ; , ,0
, ; 1, 0; 04 ,
V=
(A.1)
4 4 4 4 4 12 12
12 12

1 1
2 2 1 1
,
a3 = 0, 0, 0, 0, 0; 0,
(A.2)
,
0, 0; 0, 0; , ,
,
3 3
3 3 3 3
which satisfy the modular invariance conditions. Model II, constructed with these shift vector
and Wilson line, eliminates in particular all 51/2 and 51/2 from the massless spectrum, leaving
intact the MSSM fields obtained in Model I. Here, the gauge group is further broken down to

SU(5) U (1)X U (1)3 SU(4) SU(2)1 SU(2)2 SU(2)3 U (1)2 . (A.3)
Model II gives the same spectrum as Model I for the T6 , T20 , and T40 sectors and visible sector of
U in Model I. As in Model I, there is no massless states satisfying (P + 3V ) a3 = 0 mod Z in
the T3 sector. The spectrum of Model II is summarized as follows:
Fields of flipped SU(5): 3 16flip + 1 {52 , 52 , } + 1 {101 , 101 }, where 16flip
{101 , 33 , 15 }.
(Regularly charged) vector-like fields: 2 {16flip , 16flip } + 2 10flip , where 10flip
{52 , 52 }.
Table 11
Chiral fields from the U sector. + () denotes 12 ( 1
2 )
P V
s , Ui
Visible states
SU(5) U (1)X
1
12
(+; +++)
5L
3
(+++; +)
10L
1
(+++++; +++)
1L
5
(++), U3
4
12
(++), U2
(1, 0, 0, 0, 0; 1, 0, 0)
5L
2
5
12
(+++), U1
(++++; ++)
5R
3
(++; )
10R
1
(; ++)
1R
5
[SU(4) SU(2)3 ]
P V
s , Ui
Hidden states
4
12
(++), U2
(; ++; ) , (; ; )
(; ++; ) , (; ; )
(; ++; ) , (; ; )
4
12
(++), U2
(1, 1; 0, 0; 0, 0, 0, 0)
(6, 2, 1, 1)L
singlet
78
Table 12
Massless states satisfying (P + 6V ) a3 = 0 mod Z in T6 . The definitions of 53 , 53 , 101 , 101 , and 15,0 are found
in the main text
P + 6V
(53 ; 12 , 0, 0)(08 )
(101 ; 12 , 0, 0)(08 )
(15 ; 12 , 0, 0)(08 )
8
(53 ; 1
2 , 0, 0)(0 )
1
8
(101 ; 2 , 0, 0)(0 )
8
(15 ; 1
2 , 0, 0)(0 )
(10 ; 0, 12 , 12 )(08 )
(N L )j
1
2
1
2
3
4
L
L
1
6
1
6
1
3
1
6
13
0
0
1
6
1
2
1
3
13
11
11
1
8
(10 ; 0, 1
2 , 2 )(0 )
P6
1
2
1
3
13
13
11
11
Table 13
Chiral matter fields satisfying 0,+, = 0 in the T20,+, sectors
(N L )j
P2 (f0 )
1 1 6
(10 ; 13 , 12 , 1
2 )( 3 , 3 ; 0 )
21 , 23
1+1
(10 ;
12 , {11 + 13 }
1+1
11
P + 2V
2 1 1 6
(53 ; 1
6 , 0 )( 3 , 3 ; 0 )
1
(15 ; 6 , 02 )( 13 , 13 ; 06 )
(10 ;
(10 ;
(10 ;
1 , 1 , 1 )( 1 , 1 ; 06 )
3 2 2 3 3
2 , 1 , 1 )( 1 , 1 ; 06 )
3 2 2 3 3
2 , 1 , 1 )( 1 , 1 ; 06 )
3
2
2
3 3
1 , 1 , 1 )( 2 , 2 ; 06 )
3 2 2
3
3
13
P + 2V+
(N L )j
P2 (f+ )
1

(10 ; 13 , 1
6 , 6 )(4, 1, 1, 2)2+
1

(10 ; 13 , 1
6 , 6 )(1, 1, 2, 2)2+
21 , 23
1+1
21 , 23
1+1
(N L )j
P2 (f )
(10 ; 13 ,
(10 ; 13 ,
1 ,
6
1 ,
6
1 )(1, 2, 1, 1)
2+
6
1 )(1, 1, 1, 1)
2+
6
P + 2V
(10 ;
(10 ;
(10 ;
(10 ;
(10 ;
1,
3
1,
3
1,
3
1,
3
1,
3
1,
6
1,
6
1,
6
1,
6
1,
6
1 )(6, 1, 1, 1)
2
6
1 )(4, 1, 2, 1)
2
6
1 )(1, 1, 1, 1)
2
6
1 )(1, 2, 1, 1)
2
6
1 )(1, 1, 1, 1)
2
6
12 , {11 + 13 }
1+1
12 , {11 + 13 }
1+1
79
Exotic particles: 16 {15/2 , 15/2 }.

A lot of neutral singlets under SU(5) U (1)X .
The full massless spectrum of Model II is presented in Tables 1115.
In Table 13, the abbreviated symbols denote

1
2 1 1 1
(4, 1, 1, 2)2+ 2
3 , 3 ; 0, 0; 3 , 3 , 3 , 3 ,

1 1 1 1 1 1
(1, 1, 2, 2)2+ 56 , 1
6 ; 2, 2 ; 6 , 6 , 6 , 6 ,

1
1
1 1 1 1 1
(1, 2, 1, 1)2+ 1
6 , 6 ;2,2; 6 , 6 , 6 , 6 ,

(1, 1, 1, 1)2+ 13 , 13 ; 0, 0; 13 , 13 , 13 , 13 ,
Table 14
Chiral matter fields in the T40 , T4+ and T4 sectors. In the T4+ and T4 sectors, (1, 2, 1, 1)4+ ( 16 , 16 ; ; 16 , 16 , 16 , 16 ) ,
1
1 1 1 1
1 1
1 1 1 1

(1, 1, 1, 1)4+ ( 1
3 , 3 ; 0, 0; 3 , 3 , 3 , 3 ) , (1, 2, 1, 1)4 = ( 6 , 6 ; ; 6 , 6 , 6 , 6 ) and (1, 1, 1, 1)4 =
1
1 1 1 1
( 1
3 , 3 ; 0, 0; 3 , 3 , 3 , 3 )
P + 4V
(N L )j
P4 (f0 )
2 1 1 6
(52 ; 1
3 , 0 )( 3 , 3 ; 0 )
1
2
2+2
2+2
2 1 1 6
(52 ; 1
3 , 0 )( 3 , 3 ; 0 )
1 6
(10 ; 23 , 02 )( 1
3 , 3 ;0 )
11
12
13
(10 ;
(10 ;
(10 ;
2 , 02 )( 2 , 2 ; 06 )
3
3 3
1 , 1, 0)( 1 , 1 ; 06 )
3
3
3
1 , 0, 1)( 1 , 1 ; 06 )
3
3
3
0
0
0
1
4
1
2
1
4
1
2
1
4
1
4
P + 4V+
(N L )j
P4 (f+ )
1

(10 ; 23 , 1
3 , 3 )(1, 2, 1, 1)4+
1

(10 ; 23 , 1
3 , 3 )(1, 1, 1, 1)4+
1
4
1
4
(10 ; 1
3 ,
1
(10 ; 3 ,
1 ,
3
1 ,
3
(10 ; 1
3 ,
1
(10 ; 3 ,
2,
3
2,
3
2 )(1, 2, 1, 1)
4+
3
2 )(1, 1, 1, 1)
4+
3
1
4
1
4
(N L )j
P4 (f )
1 2

(10 ; 1
3 , 3 , 3 )(1, 2, 1, 1)4
1
4
1
4
1 2

(10 ; 1
3 , 3 , 3 )(1, 1, 1, 1)4
2 1

(10 ; 1
3 , 3 , 3 )(1, 2, 1, 1)4
2 1

(10 ; 1
3 , 3 , 3 )(1, 1, 1, 1)4
1 )(1, 2, 1, 1)
4+
3
1 )(1, 1, 1, 1)
4+
3
P + 4V
(10 ; 23 ,
(10 ; 23 ,
1,
3
1,
3
1 )(1, 2, 1, 1)
4
3
1 )(1, 1, 1, 1)
4
3
1
4
1
4
80
Table 15
Chiral matter fields satisfying 0,+ = 0 in the T10 , T1+ , T50 , and T5+ sectors. There are no massless states in
1 1
T1 and T5 . Here, 15/2 ( 14 , 14 , 14 , 14 , 14 ). In T10 and T1+ , (4, 1, 1, 1)1+ ( 16 , 16 ; 0, 0; 23 , 1
3 , 3 , 3 ),
+
1
1
1
1
1
1
5
1
(1, 1, 2, 1)1+ ( 3 , 3 ; ; 6 , 6 , 6 , 6 ) . In T50 and T5 , (1, 1, 1, 2)5 = ( 6 , 6 ; 0, 0; 0, 0, 0, 0) , (4, 1, 1, 1)5+ =
1
2 1 1 1
1 1
1 1 1 1

( 1
6 , 6 ; 0, 0; 3 , 3 , 3 , 3 ) , and (1, 1, 2, 1)5+ = ( 3 , 3 ; ; 6 , 6 , 6 , 6 )
(N L )j
P1 (f0 )
13
(N L )j
P1 (f+ )
13
13
P + 5V
(N L )j
P5 (f0 )
1 , 6 , 0)(1, 1, 1, 2)
(15/2 ; 12
5
12
11
6

(15/2 ; 5
12 , 0, 12 )(1, 1, 1, 2)5
P + 5V+
(N L )j
P5 (f+ )
11
11
P +V
5 , 6 , 0)(1, 1, 1, 2)
(15/2 ; 12
1
12
1

)(1,
1,
1,
2)
(15/2 ; 12 , 0, 6
1
12
P + V+
5 , 2 , 4 )(4, 1, 1, 1)
(15/2 ; 12
1+
12 12
5 , 2 , 4 )(1, 1, 2, 1)
(15/2 ; 12
1+
12 12
4 , 2 )(4, 1, 1, 1)
(15/2 ; 1
,
1+
12 12 12
4 , 2 )(1, 1, 2, 1)
(15/2 ; 1
,
1+
12 12 12
1 , 2 , 4 )(4, 1, 1, 1)
(15/2 ; 12
5+
12 12
1 , 2 , 4 )(1, 1, 2, 1)
(15/2 ; 12
5+
12 12
4 , 2 )(4, 1, 1, 1)
(15/2 ; 5
,
5+
12 12 12
4 2

(15/2 ; 5
12 , 12 , 12 )(1, 1, 2, 1)5+

1 1
1
, ; 0, 0; 23 , 23 , 1
3 , 3 ,
3 3

1 1
, 3 ; 0, 0; 1
, 1
, 23 , 23 ,
(6, 1, 1, 1)2
3
3
3
1 , 1 ; 0, 0; 2 , 1 , 2 , 1 ,
3 3
(4, 1, 2, 1)2
1
6

1
5 1 1 1
6 ; ; 6 , 6 , 6 , 6 ,

1 1 1 1
, 2
3 ; 0, 0; 3 , 3 , 3 , 3 ,

1
1 1 1 1
(1, 2, 1, 1)2 1
6 , 6 ; ; 6 , 6 , 6 , 6 ,

1 1 1
(1, 1, 1, 1)2 13 , 13 ; 0, 0; 1
3 , 3 , 3 , 3 .
(1, 1, 1, 1)2
2
3
References
[1] P. Candelas, G.T. Horowitz, A. Strominger, E. Witten, Nucl. Phys. B 258 (1985) 46.
[2] L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 261 (1985) 678;
L.J. Dixon, J.A. Harvey, C. Vafa, E. Witten, Nucl. Phys. B 274 (1986) 285.
[3] L.E. Ibanez, H.P. Nilles, F. Quevedo, Phys. Lett. B 187 (1987) 25.
[4] L. Ibaez, J.E. Kim, H.P. Nilles, F. Quevedo, Phys. Lett. B 191 (1987) 282;
See, also J.A. Casas, C. Munoz, Phys. Lett. B 214 (1988) 63.
[5] J.E. Kim, Phys. Lett. B 564 (2003) 35.
[6] Y. Katsuki, Y. Kawamura, T. Kobayashi, N. Otsubo, Y. Ono, K. Tanioka, Kanazawa Univ. preprint DPKU-8904,
1989.
81
[7] K.-S. Choi, J.E. Kim, Quarks and Leptons from Orbifolded Superstring, Springer-Verlag, Heidelberg, Germany,
2006.
[8] P. Langacker, M. Luo, Phys. Rev. D 44 (1991) 817;
C. Giunti, C.W. Kim, U. Lee, Mod. Phys. Lett. A 6 (1991) 1745;
J.R. Ellis, S. Kelly, D.V. Nanopoulos, Phys. Lett. B 260 (1991) 131;
U. Amaldi, W. de Boer, H. Frstenau, Phys. Lett. B 260 (1991) 447.
[9] S.M. Barr, Phys. Lett. B 112 (1982) 219.
[10] J.-P. Derendinger, J.E. Kim, D.V. Nanopoulos, Phys. Lett. B 139 (1984) 170.
[11] I. Antoniadis, J.R. Ellis, J.S. Hagelin, D.V. Nanopoulos, Phys. Lett. B 231 (1989) 65;
I. Antoniadis, J.R. Ellis, J.S. Hagelin, D.V. Nanopoulos, Phys. Lett. B 208 (1988) 209;
I. Antoniadis, J.R. Ellis, J.S. Hagelin, D.V. Nanopoulos, Phys. Lett. B 213 (1988) 562, Addendum;
K. Dienes, A. Faraggi, Nucl. Phys. B 457 (1995) 409;
A.E. Faraggi, R.S. Garavuso, J.M. Isidro, Nucl. Phys. B 641 (2002) 111.
[12] H. Georgi, S.L. Glashow, Phys. Rev. Lett. 32 (1974) 438.
[13] H. Georgi, Talk presented at AIP Conference, William and Mary, 1974, AIP Conf. Proc. 23 (1975) 575;
H. Fritsch, P. Minkowski, Ann. Phys. 93 (1975) 193;
F. Wilczek, A. Zee, Phys. Rev. D 25 (1982) 553.
[14] F. Grsey, P. Ramond, P. Sikivie, Phys. Lett. B 60 (1977) 177.
[15] S.L. Glashow, Trinification of all elementary particle forces, in: K. Kang, et al. (Eds.), Proc. of IV Workshop on
Grand Unification, World Scientific, Singapore, 1985, p. 88;
A. De Rujula, H. Georgi, S.L. Glashow, unpublished, 1984.
[16] J. Pati, A. Salam, Phys. Rev. D 8 (1973) 1240.
[17] K.-S. Choi, K.-Y. Choi, K. Hwang, J.E. Kim, Phys. Lett. B 579 (2004) 165;
J.E. Kim, Phys. Lett. B 591 (2004) 119.
[18] Q. Shafi, Z. Tavartkiladze, hep-ph/0606188, and references therein.
[19] J.E. Kim, B. Kyae, hep-th/0608085.
[20] R. Blumenhagen, S. Moster, T. Weigand, Nucl. Phys. B 751 (2006) 186.
[21] C.M. Chen, G.V. Kraniotis, V.E. Mayes, D.V. Nanopoulos, J.W. Walker, Phys. Lett. B 611 (2005) 156;
C.M. Chen, V.E. Mayes, D.V. Nanopoulos, Phys. Lett. B 633 (2006) 618;
C.M. Chen, T. Li, D.V. Nanopoulos, Nucl. Phys. B 751 (2006) 260;
M. Cvetic, P. Langacker, hep-th/0607238.
[22] V. Kaplunovsky, Phys. Rev. Lett. 55 (1985) 1036.
[23] I.W. Kim, J.E. Kim, in preparation.
[24] J. Ellis, L. Lopez, D.V. Nanopoulos, Phys. Lett. B 247 (1990) 257.
[25] J. Ellis, V.E. Meyers, D.V. Nanopoulos, Phys. Rev. D 70 (2004) 075015;
C. Coriano, A.E. Faraggi, M. Plmacher, Nucl. Phys. B 614 (2001) 233.
[26] A.E. Faraggi, D.V. Nanopoulos, K. Yuan, Nucl. Phys. B 335 (1990) 347;
A. Faraggi, Phys. Lett. B 278 (1992) 131;
A. Faraggi, Phys. Lett. B 455 (1999) 135;
G.B. Cleaver, A.E. Faraggi, D.V. Nanopoulos, Phys. Lett. B 455 (1999) 135.
[27] K.-S. Choi, J.E. Kim, Quarks and Leptons from Orbifolded Superstring, Springer-Verlag, Heidelberg, Germany,
2006, Appendix D.
[28] V.G. Kac, D.H. Peterson, in: Anomalies, Geometry and Topology, Proceedings of Argonne-Chicago Conference,
1985, p. 276;
T.I. Hollowood, R.G. Myhill, Int. J. Mod. Phys. A 3 (1988) 899;
K.-S. Choi, K. Hwang, J.E. Kim, Nucl. Phys. B 662 (2003) 476.
[29] T. Kobayashi, S. Raby, R.J. Zhang, Nucl. Phys. B 704 (2005) 3.
[30] W. Buchmuller, K. Hamaguchi, O. Lebedev, M. Ratz, Nucl. Phys. B 712 (2005) 139;
W. Buchmuller, K. Hamaguchi, O. Lebedev, M. Ratz, Phys. Rev. Lett. 96 (2006) 121602, hep-th/0606187.
[31] L.E. Ibanez, J. Mas, H.P. Nilles, F. Quevedo, Nucl. Phys. B 301 (1988) 157.
[32] Y. Katsuki, Y. Kawamura, T. Kobayashi, N. Otsubo, Y. Ono, K. Tanioka, Nucl. Phys. B 341 (1990) 611.
[33] S. Forste, H.P. Nilles, P.K.S. Vaudrevange, A. Wingerter, Phys. Rev. D 70 (2004) 106008;
S. Forste, H.P. Nilles, A. Wingerter, Phys. Rev. D 73 (2006) 066011.
[34] T. Kobayashi, S. Raby, R.J. Zhang, Phys. Lett. B 593 (2004) 262.
[35] S. Hamidi, C. Vafa, Nucl. Phys. B 279 (1987) 465;
L.J. Dixon, D. Friedan, E.J. Martinec, S.H. Shenker, Nucl. Phys. B 282 (1987) 13;
D. Friedan, E.J. Martinec, S.H. Shenker, Nucl. Phys. B 271 (1986) 93.
82
[36] H. Georgi, Nucl. Phys. B 156 (1979) 126.

[37] T. Kobayashi, N. Ohtsubo, Phys. Lett. B 245 (1990) 441.
[38] K. Hwang, J.E. Kim, Phys. Lett. B 540 (2002) 289;
B. Kyae, Q. Shafi, Phys. Lett. B 635 (2006) 247.
[39] C.S. Huang, T. Li, C. Liu, J.P. Shock, F. Wu, Y.L. Wu, hep-ph/0606087.
[40] I. Antoniadis, J.R. Ellis, J.S. Hagelin, D.V. Nanopoulos, Phys. Lett. B 194 (1987) 231.
[41] J.E. Kim, H.P. Nilles, Phys. Lett. B 138 (1984) 150.
[42] H.D. Kim, J.E. Kim, H.M. Lee, Europhys. J. C 24 (2002) 159.
[43] P. Ginsparg, Phys. Lett. B 197 (1987) 139.
[44] J.E. Kim, J.-H. Kim, B. Kyae, hep-ph/0702278.
On massive spin 2 interactions

Yu.M. Zinoviev
Institute for High Energy Physics, Protvino 142280, Moscow Region, Russia
Received 19 January 2007; accepted 1 February 2007
Abstract
In this paper we use a constructive approach based on gauge invariant description of massive high spin
particles for investigation of possible interactions of massive spin 2 particle. We work with general case of
massive spin 2 particle living in constant curvature (A)dSd background, which allows us carefully consider
all flat space, massless or partially massless limits. In the linear approximation (cubic terms with no more
than two derivatives in the Lagrangians and linear terms with no more than one derivative in gauge transformations) we investigate possible self-interaction, interaction with matter (i.e. spin 0, 1 and 1/2 particles)
and interaction with gravity.
0. Introduction
In all investigations of massless particles interactions gauge invariance plays a crucial role.
Not only it determines a kinematic structure of free theory and guarantees a right number of
physical degrees of freedom, but also to a large extent fixes possible interactions of such particles.
This leads, in particular, to formulation of so-called constructive approach for investigation of
massless particles models [15]. In such approach, starting with appropriate collection of free
massless particles and requiring conservation of (modified) gauge invariance in the process of
switching on an interaction, one can consistently reproduce such physically important theories
as YangMills, gravity and supergravity.
Usual description of massive particles does not include gauge invariance and it is hard to
formulate one simple principle for constructing consistent theories with such particles. A lot of
different requirements, such as conservation of right number of physical degrees of freedom,
smooth massless limit, tree level unitarity and causality, was used in the past [613].
E-mail address: yurii.zinoviev@ihep.ru.
doi:10.1016/j.nuclphysb.2007.02.005
84
Yu.M. Zinoviev / Nuclear Physics B 770 (2007) 83106
There exist two classes of consistent models for massive high spin particles, namely, for massive non-Abelian spin 1 particles and for massive spin 3/2 ones. In both cases masses of gauge
fields appear as a result of spontaneous gauge symmetry breaking. One of the main ingredients of this mechanism is the appearance of Goldstone particles with non-homogeneous gauge
transformations. This, in turn, leads to the gauge invariant description of such massive spin 1 and
spin 3/2 particles. But such gauge invariant description of massive particles could be constructed
for higher spins as well, e.g. [1423]. This allows one to extend the above mentioned constructive
approach to the case of massive high spin particles.
In Section 1 of our paper we give two simple examples of such constructive approach. One of
them deals with triplet of massive spin 1 particles with gauge group SU(2) O(3). We show that
constructive approach allows one to reproduce two well-known possibilities: one based on the
non-linear -model [24], and, with the help of introduction of additional scalar field, usual model
of spontaneous symmetry breaking with the doublet of Higgs fields. Another example devoted to
electromagnetic interaction for massive spin 3/2 particle [9,10]. In the first linear approximation
(i.e. cubic terms in the Lagrangian and linear terms in the gauge transformations) we construct
the most general gauge invariant Lagrangian. Our results unambiguously show that any model
with minimal gauge interaction of massive spin 3/2 particle in flat space must be a part of some
(spontaneously broken) supergravity theory.
Main part of our paper devoted to massive spin 2 particle interactions. It is an old but still
unsolved problem and the main difficulty is connected with the so-called sixth degree of freedom [25]. Indeed, for the free particle it is easy to choose a structure of mass terms so that there
are exactly five degrees of freedom in the model. But in the interacting theory this dangerous
(because it is a ghost) sixth degree of freedom reappear. Note that it is, in principle, possible
that in the full non-linear theory this DoF become physical [26,27], but in what follows we will
not consider such possibility. One more difficulty that appears in such theories is the absence
of smooth massless limit [28,29]. Moreover, when one considers such theory in a background
space with non-zero cosmological term (de Sitter or anti-de Sitter spaces) one faces an ambiguity
between flat space and massless limits [3032].
Here we use constructive approach based on the gauge invariant description of massive particles for the investigation of possible massive spin 2 interactions. It turns out that some results
depends on the dimensionality of spacetime as well as on the presence (or absence) of cosmological term. So we start in Section 2 with the gauge invariant formulation of free massive spin 2
particle in the spacetime of arbitrary d > 2 dimension with non-zero cosmological term, that
could be positive or negative. This includes, in particular, the so-called partially massless spin 2
particle, which could exist in the de Sitter space only [16,3336].
Then, in Section 3 we consider possible self-interaction of such particles in linear approximation. We will not make any suggestion about gauge algebra which could stand behind such
theory nor we will not insist on any possible geometrical interpretation. Instead, we use a brute
force method for our construction. Namely, we write the most general Lorentz invariant cubic
vertexes in the Lagrangian as well as the most general ansatz for gauge transformations linear
in fields and require that the Lagrangian will be gauge invariant. The only restriction (besides
Lorentz invariance) we impose is the restriction on the number of derivatives. In this paper we
consider only interactions with no more than two derivatives (and gauge transformations with no
more than one derivative) leaving investigation of possible higher derivatives interactions for the
future.
Clearly, the existence of linear approximation does not guarantee that the construction of
full non-linear theory is possible because obstruction can appear in the next approximations.
85
Nevertheless, the investigation of linear approximation is a very important step, because structure
of this approximation (and its whole existence) does not depend on the presence of any other field
in the theory. Only in the next quadratic approximation one faces the problem that the closure
of gauge algebra requires the presence of right (finite or infinite) collection of fields. Thus, the
results obtained in our paper are essentially model independent (up to restriction on the number
of derivatives).
For the generic values of mass and cosmological constant the linear approximation for selfinteraction exists in any spacetime dimension d 3, but for partially massless spin 2 particle
it exists in d = 4 dimensions only. This clearly related with the fact that only in d = 4 partially
massless theories are conformally invariant [37]. One of the non-trivial results is that the structure
of gauge transformations for Goldstone field A turns out to be non-canonical, so if one tries to
interpret (as usual in gravity) the gauge transformations as general coordinate ones, then this
field should give some non-linear realization of this transformations. One more non-trivial result
is that two gauge transformationsvector and scalar ones, do not commute when the mass
m = 0 forming an algebra similar to Weyl one.
In the next section we consider possible cubic couplings of massive spin two particles with
matter. Namely, we consider interaction with (massive or massless) spin 0, spin 1 and spin 1/2
particles. In all cases (except massless spin 1 in d = 4) our results show an ambiguity between
flat and massless limits, which shows itself through non-trivial dependence of coupling constant
for scalar Goldstone field and matter fields interactions on mass and cosmological constant.
As a result, partially massless spin 2 particle can interact with matter having traceless energy
momentum tensor only, the most important example being the coupling of partially massless
spin 2 with massless spin 1.
Our last section deals with possible interactions for massive spin 2 particles with gravity. As
is well known, it is impossible to construct non-trivial consistent interacting theory for collection
of massless spin 2 particles [3842], but it still leaves the possibility to construct non-trivial
interaction of gravity with (one or many) massive spin 2 particles. Indeed, such possibility attend
much interest recently in very different contexts. First of all, we show that it is impossible to
construct cubic vertex with two massless and one massive spin 2 particles while vertex with two
massive and one massless ones does exist. Moreover, it is easy to construct general covariant
vertex with two massive fields and arbitrary number of massless gravitons. As for the interaction
of partially massless spin 2 particles with gravity, no restrictions on the spacetime dimension
arise in this approximation.
1. Toy models
In this section we present two simple examples of constructive approach to massive high spin
particles interactions. First one deals with the triplet of massive spin 1 particles, while second
one devoted to possible gauge interactions for massive spin 3/2 particle. Results of this section
where already known before, but it is instructive to see how they can be reproduced using gauge
invariant formulation alone.
1.1. Spin 1
In this subsection we illustrate our approach by the simplest non-Abelian SU(2) O(3)
gauge theory with triplet of vector fields A a , a = 1, 2, 3. It is well known that with the help
of triplet scalar Goldstone fields a one can easily construct gauge invariant description of free
86
massive particles. In this, the Lagrangian and gauge transformations look like:
2 m2 a 2
2
1
1
mA a a + a ,
A a +
A
4
2
2
0 A a = a ,
0 a = m a .
L0 =
(1)
Now, if we switch on usual non-Abelian gauge interaction:

A a = A a g abc A b A c ( ),
0 A a = a g abc A b c
(2)
then the Lagrangian will not be gauge invariant any more:

0 L0 = gm abc A a b c
(3)
but at the linear approximation the gauge invariance could be easily restored with additional
vertex in the Lagrangian and appropriate corrections to gauge transformations:
L1 = a1 abc A a b c ,
1 a = g1 abc b c
(4)
provided a1 = g/2, g1 = g/2!. Note, that non-canonical value g/2 in the gauge transformations of scalar fields a shows that these fields do not transform under linear triplet representation
of gauge group. As we already mentioned in the introduction, this first step does not depend on
the presence of any other fields in the model. Now we can proceed in at least two different ways.
At first, we can evade introduction of any additional fields and go on by introducing into the
Lagrangian terms of the type A 2 , 2 ( )2 and so on. It would lead us to the essentially
non-linear theory with the Lagrangian [24]:
2 m2 a 2
1
1
mA a E ak () k + g kl () k l
A a +
A
4
2
2
and gauge transformations for scalar fields:

ka
k = m E 1 a
L=
(5)
(6)
where g kl = E ak E al . This Lagrangian will be gauge invariant provided the triad E ak satisfy
the equation
E ak E al
= g abc E bk E cl .
l
k
(7)
It is easy to check that non-canonical value g/2 in the linear approximation for the a gauge
transformations is a direct consequence of this equation.
But there is another way. We can introduce one more scalar field and try to stop iterations at
some order. And indeed, if we add to the Lagrangian of linear approximation all possible cubic
and quartic terms with new field:
2

1
L2 = ( )2 + a2 (A)a a + a3 A a a + a4 m A a
2

2

2
+ a5 A a 2 + a6 2 A a
(8)
as well as appropriate corrections to the gauge transformations:

= g2 ( ),
2 a = g3 a
(9)
87
we can obtain full gauge invariant theory, provided:

g2 = a2 =
g
,
2
a3 = g,
g
g3 = a 4 = ,
2
a5 = a6 =
g2
.
8
Moreover, if we introduce shifted variable = 2m

g , then total Lagrangian could be rewritten
in a familiar form:
2 1
2 1
g
1
L = A a + a + ( )2 abc A a b c
4
2
2
2
g 2 a 2 2

g a
a
a
+ 2 .
+ A +
(10)
A
2
8
Up to the arbitrary potential depending on the invariant combination 2 + 2 only this is just the
well-known model for spontaneous gauge symmetry breaking through the Higgs mechanism. In
this, four scalar fields a and are transformed under the linear doublet representation of gauge
group.
Note, that one more possibility arises when one consider systems with infinite number of
gauge fields such as so-called KaluzaKlein models e.g. [4348]. Indeed, let us consider simplest
examplefive-dimensional YangMills theory with gauge fields A a , where M = 0, 1, 2, 3, 5,
and some non-Abelian gauge algebra [t a , t b ] = f abc t c . Suppose now that one dimension is a
compact one, being a circle of radius R. From a four-dimensional point of view, such fields
represent combinations of vector and scalar ones: AM (x , x5 ) A (x, x5 ), (x, x5 ), which
equivalent to infinite number of four-dimensional fields:
A a (x , x5 ) =
a (x , x5 ) =
n=
A(n) a (x) exp(iMnx5 ),
n=
n=
(n) a (x) exp(iMnx5 )
(11)
n=
where M 1/R, A(n) = A(n) and similarly for . Performing integration on x5 , one obtains
four-dimensional gauge theory with infinite dimensional gauge algebra:
a b
tn , tm = f abc tn+m c .
In this, gauge transformations of all fields have the form:
A(n) a = n a + f abc k b A(nk) c ,
n a = Mnn a + f abc k b (nk) c .
(12)
From the last line we see, that all the scalar fields, except 0 , have non-homogeneous transformation laws and play the role of Goldstone fields. As a result, such model describes finite number
of massless vector A0 a and scalar 0 a fields and infinite tower of massive n = 0 vector fields.
Introducing appropriate covariant derivatives for these fields:
D n a = n a f abc A(k) b (nk) c MnA(n) a
the total Lagrangian could be written in a compact form:

n=
1
1
a
a
a
a
A(n) A(n) + D n D n .
L=
4
2
n=
88
More complicated examples, corresponding to partial breaking of initial gauge algebra, could be
found in [47].
1.2. Spin 3/2
Our second example devoted to the possible electromagnetic interactions for massive spin 3/2
particles [9,10]. As is well known using two fieldsvector-spinor and spinor one can
construct gauge invariant formulation of massive spin 3/2 particle. So we start with the sum of
free massive spin 3/2 and massless spin 1 Lagrangians:
i
i
1
L0 = 5 + A 2
2
4
2
3
m
+ i
(13)
m( ) m
2
2
where gauge transformations leaving this Lagrangian invariant look like:

3
im
0 = +
(14)
0 =
,
m,
A = .
2
2
Here is a spinor parameter, while is a scalar. We prefer to work with Majorana spinors, so in
what follows all spinor objects , and will be considered as doublets of (anticommuting)
Majorana spinors. In this, minimal gauge interactions corresponds to the replacement of ordinary
partial derivatives by covariant ones:

0 e
D = + qA , q =
(15)
, q 2 = e2 I.
e 0
As a result of such replacement, the Lagrangian lost its gauge invariance (just because covariant
derivatives do not commute):
1
A = A .
(16)
2
So we try to restore gauge invariance by adding to the Lagrangian non-minimal terms linear in
gauge field-strength A :
0 L0 = i q A 5 ,

1
mL1 = a1 A + a2 5 A + a3 g ( A) + a4 A + A q
2

a7
+ i a5 A + a6 5 A q + q( A)
2
as well as all possible linear terms for -transformations for all three fields:
m1 = iq(1 A + 2 5 A ) ,
m1 A = 4 ( q) + i5 ( q).
(17)
m1 = q3 ( A),
(18)
Straightforward calculations show that the gauge invariance could be restored (up to terms
quadratic in A ) provided all the unknown coefficients are expressed in terms of two parameters, say 1 and 3 ,
2 = 1 ,
a2 = 21 ,
4 = 21 ,
a3 = a4 = 0,
5 = 23 ,
a1 = 21 ,
a5 = a6 = 23 ,
a7 = 2
2
3
3
while this two parameters satisfy the relation:

3
3 + 1 = 0.
21 4
2
89
(19)
So we have one parameter family of Lagrangians. Now, if we calculate the commutator of two
transformations on the vector field A , we obtain:

[1 , 2 ]A = 4ie2 1 2 + 23 2 2 1 A .
(20)
Due to relation on the parameters 1 and 3 it is impossible to set both parameters simultaneously
equal to zero. This means that commutator of two transformations always give a translation
and to proceed further on we must introduce graviton as well. So any model with minimal gauge
interaction for massive spin 3/2 particle in flat space must be a part of some (spontaneously
broken) supergravity theory. In this case all coupling constants are related with gravitational
one k 1/mpl and we obtain usual for supergravity models relation between spin 3/2 mass,
Planck mass and gauge coupling constant m empl ! Note in pass that one more example of
constructive approach, namely electromagnetic interactions for massive spin 2 particles, could
be found in [15].
2. Free massive spin 2
Our main starting point is the gauge invariant description of free massive spin 2 particle. As
we will see part of the results obtained crucially depend on the spacetime dimension d and/or
the presence or absence of cosmological term. So in what follows we will work in spacetime
with arbitrary d 3 dimension and with non-zero (positive or negative) cosmological term. We
denote metric of such space as g and appropriate covariant derivatives as D . Till the last
section, where g will become dynamical, it is just fixed background metric. Our convention
for the commutator of two covariant derivatives is:

[D , D ]v = g g v
(21)
2
where = (d1)(d2)
, cosmological term.
One of the nice features of gauge invariant description is that it allows one to consider general case of non-zero mass and cosmological term including all possible massless and partially
massless limits [16]. To describe the massive spin 2 particle in d-dimensional constant curvature
space we will use the following Lagrangian:
1
1
1
1
L0 = D h D h D h D h (Dh) (Dh) + (Dh) D h D hD h
2
2
2
2

1
2(d 1)
(D A D A )2 +
(D )2 + 2m h D A h(DA)
2
d 2

m2 + (d 2)
4(d 1)c0
A D
h h h2
d 2
2
2(d 1)mc0
2d(d 1)m2 2
(22)
+ 2(d 1)A 2
h +
d 2
(d 2)2

where c0 = m2 + (d 2). Here symmetric tensor h is the main gauge field, while vector
90
A and scalar are Goldstone fields necessary for gauge invariance. To simplify the calculations
we have chosen non-canonical normalization for kinetic terms of these fields. Also note that
there is an ambiguity in the structure of kinetic terms for the h field. Indeed, in flat space
(where derivatives commute) the second and third terms in the first line differ by total divergence
only. But in spaces with non-zero cosmological term they lead to slightly different structure of
mass-like terms and the choice we made gives the most simple and convenient structure of these
terms. It is not hard to check that this Lagrangian is invariant under the following local gauge
transformations:
0 h = D + D +
0 A = D + m ,
2m
g ,
d 2
0 = c0 .
(23)
Let us note, that these and all subsequent results depend crucially on use of canonical description of massless spin 2 particles. As is known [49], there exists alternative possibility based
on traceless symmetric tensor (or on additional Weyl symmetry). Canonical analysis shows that
right number of physical degrees of freedom in this case is achieved due to combination of first
and second class constraints. So the relation between alternative and canonical formulations is
similar to that between, say, non-gauge invariant and gauge invariant formulations of massive
spin 1 particle. Indeed, additional scalar degree of freedom (trace h in this case) promotes second class constraints to the first class ones. As we have already note in the introduction, it is hard
to formulate constructive approach for theories with second class constraints. Moreover, as it was
shown in [49], any attempt to give mass to graviton in such alternative formulation unavoidably
leads to the appearance of ghosts.
Note also, that there exists a disagreement on the definition of mass in (A)dS space. Indeed,
even for the massless particles gauge invariance requires some mass-like terms in the Lagrangians to be present and one often tries to interpret these terms as real mass. Gauge invariant
description of massive particles gives, above all, simple and unambiguous definition of massless
limit, namely it is a limit where Goldstone fields decouple from the main gauge field.
Let us stress the important difference between massless and massive cases for spin 2 particle.
In the massless case we have one gauge symmetry with vector parameter only, while for
non-zero mass we have two gauge symmetries with vector and scalar parameters and only
these two symmetries together could guarantee the right number of physical degrees of freedom
even after switching on an interaction. Thus, models based on the invariance only like those
based on OSP(4) symmetry [50] or models exploring transformations only [51] always require
additional checks for consistency. One more example is a BRST construction of [52], where for
massive spin 2 particles authors promote gauge transformations with vector parameter only (and
introduce vector Goldstone field u only). As a result, such theory describes additional scalar
degrees of freedom and all previous results point that it must be ghost. At the same time, BRST
construction for pure massive spin 2 particle does exist [18], but requires additional scalar gauge
transformations and scalar Goldstone field.
Note also that if one introduces gauge invariant under the transformations derivatives:
h = D h
2m
A ,
d 2
= D c0 A
then the Lagrangian could be rewritten in more simple form:
(24)
1
1
1
L0 = h h h h (h) (h) + (h) h
2
2
2
1
1
2(d 1)
2
( )2
h h (D A D A ) +
2
2
d 2
2(d 1)mc0
2d(d 1)m2 2
m2 + (d 2)
h h h2
h +
.
2
d 2
(d 2)2
91
(25)
As could be easily seen from the formulas given above in the massless limit m 0 the
Lagrangian breaks into two independent parts. One of them (for symmetric tensor h ) gives
usual description of massless spin 2 particle in (A)dSd background, while the other one (with
vector A and scalar fields) gives gauge invariant description of massive spin 1 particle (or
sum of massless spin 1 and spin 0 particles in a flat case). But in de Sitter space < 0 there exist
one more special limit c0 0. In this, the scalar field completely decouples, while pair h ,
A corresponds to the so-called partially massless spin 2 particle. Note that one can use
gauge transformation in order to set vector A = 0. In this, the resulting simple Lagrangian for
h will still be invariant under the transformations, provided we supplement it with restoring
gauge A = 0 transformation with = m1 D :
1
2m
(D D + D D ) +
g .
(26)
m
d 2
We have explicitly checked that such gauge fixed Lagrangian is indeed invariant under this transformation.
h =
3. Self-interaction
In this section we consider self-interaction of such massive spin 2 particles in linear approximation. As we have already mentioned in the introduction, we will not make any suggestion
on the structure of gauge algebra which could stand behind this theory nor we will not insist
on any geometrical interpretation. Instead, we will construct the most general Lorentz invariant cubic terms for the Lagrangian as well as the most general ansatz for gauge transformations
and require that the Lagrangian will be gauge invariant. As a result this section will be the most
technical part of the paper, but it is important to make sure that our results are completely general.
Even for the massless particles in (A)dSd gauge invariance require introduction of masslike terms into the Lagrangian and appropriate corrections to gauge transformation laws so that
the structure of the theory resembles that of massive theory in flat space. Thus, working with
general case of massive particles in (A)dSd , it is convenient to organize the calculations just
by the number of derivatives. So we start here with the Lagrangian terms with two derivatives
and gauge transformations with one derivative and analyze all possible vertexes compatible with
gauge invariance up to the lower derivatives terms.
hhh-vertex. In this case the most general ansatz for gauge transformations linear in h-field
and containing one derivative looks as follows:

h = c1 D h + c2 (Dh) + (Dh) + c3 g (Dh)
+ c4 g D h + c5 (D h + D h ) + c6 (D h + D h )

+ c7 h D + h D + c8 h(D + D ) + c9 h (D )

+ c10 h D + h D + c11 g h D + c12 g h(D ).
92
But due to gauge invariance of free Lagrangian the structure of this transformations could be defined only up to arbitrary field dependent gauge transformations (or in other words up to possible
redefinitions of parameter) which in this case have the form:

h = d1 D h + D h + d2 D (h ) + D (h ) .
Moreover, as in all theories where interacting Lagrangian contains the same number of derivatives as the free one, there always exists a family of physically equivalent Lagrangians related
by trivial field redefinitions. In the case at hands, we have a possibility to make such redefinition
with four arbitrary parameters:
h h + s1 h h + s2 hh + s3 g h 2 + s4 g h2 .
We use all these freedom to bring the gauge transformations to the following more simple form:

h = c1 D h + c2 (Dh) + (Dh) + c3 g (Dh)

+ c4 g D h + c7 h D + h D + c9 h (D ).
At last, we require that the algebra of these transformations be closed, i.e.

(1 ), (2 ) = (3 ).
Simple calculations immediately give c2 = c3 = c4 = c9 = 0 and c7 = c1 . In the massless case the
only remaining parameter c1 corresponds to gravitational coupling constant k 1/mpl . In what
follows, we set c1 = 2. Thus, we obtain a very simple final form for these gauge transformations:

h = 2 D h + D h + D h .
(27)
In this,
3 = 2 D 1 1 D 2 .
Certainly, these transformations look exactly like general coordinate transformations for covariant second rank symmetric tensor, but let us stress that in our case these are just gauge
transformations for spin 2 field living in fixed (A)dSd background. Recall also, that concrete
form of these transformations (i.e. that of covariant tensor and not that of contravariant one or
tensor density) is just a matter of our choice. This and all our results that follows are always
defined up to possible field redefinitions of the type shown above.
Now we construct the most general Lorentz invariant cubic terms with two derivatives:

Lhhh = h a1 D h (Dh) + a2 D h D h + a3 D h D h + a4 D hD h
+ a5 D h D h + a6 D h(Dh) + a7 D h (Dh) + a8 D h D h

+ a9 D h D h + a10 D h D h + a11 (Dh) (Dh)

+ h a12 D h D h + a13 D hD h + a14 (Dh) (Dh)

+ a15 (Dh) D h + a16 D h D h .
(28)
As in the free case, we face an ambiguity because in a flat space not all these terms are independent, namely up to total divergence we have (schematically):
(a1 ) = (a5 ) + (a9 ) (a11 ),
(a2 ) = (a6 ) + (a14 ) (a16 )
93
so there exists a family of equivalent Lagrangians with two arbitrary parameters. In a constant
curvature space different choices lead to slightly different structure of mass-like h3 terms and
again we use this freedom to bring these terms to the most simple form. Then the requirement
that the Lagrangian will be gauge invariant (up to lower derivative terms given below) gives:
3
3
a1 = a11 = ,
a4 = a15 = 1,
a2 = ,
a3 = 1,
2
2
5
5
a8 = 2,
a6 = ,
a7 = a10 = 2,
a5 = ,
2
2
1
1
a13 = a14 = a16 = .
a9 = a12 = ,
2
2
In this, in the non-flat space background we still have non-compensated variations to be taken
into account in the lower derivatives orders:

0 Lhhh + 1 L0 = (3d 8)h D h 2(2d 5)h D h + 2(d 3)hD h

2(d 3)h D h + (3d 8)h (Dh) 2(d 3)h(Dh) .
(29)
Till now all calculations are the same for massive as well as massless spin 2 particles. Let us
make here a comment on the so-called gravity reconstruction, i.e. on possibility to reconstruct
full gravity theory starting from free massless spin 2 particle in flat or constant curvature space
background [53]. As usual, in all calculations of the type given above, all the variations which
are total divergence are dropped out and of course none of such terms can be reconstructed
in such iterations. But nothing prevent of to add to the Lagrangian any such terms to bring the
result in a convenient form. As for the ambiguity which arise in this iterative process and which is
often related with ambiguity in the definition of energymomentum tensor for gravity or matter
fields, we have seen that this ambiguity clearly related with trivial field redefinitions and up to
this freedom the results are essentially unique.
hAA-vertex. Here we start with the most general ansatz for transformations:
A = c1 D A + c2 D A + c3 (DA) + c4 A D + c5 A D + c6 A (D ).
As in the previous case there exists a freedom connected with the possible redefinitions of field
A and parameter :
A = d1 D (A ),
A A + s1 h A + s2 hA
so without lost of generality we can restrict ourselves to

A = c1 D A + c2 D A + c3 (DA) .
Usually, then one considers an interaction of gravity with vector fields (Abelian or non-Abelian)
one assumes that gravitational field h is inert under the gauge transformations of these fields.
But now vector field A is just a component of massive spin 2 field, so we have to consider all
the possible gauge transformations with parameter also. Most general form looks like:
h = c4 D( A) + c5 g (DA) + c6 A( D) + c7 g A D ,
A = c8 D h + c9 (Dh) + c10 hD + c11 h D .
Once again using the freedom that exists here:
h = d2 D( (A) ),
A = d3 D (h),
h h + s3 A A + s4 g A 2
94
we can reduce these transformations to the form:

h = c5 g (DA),
A = c9 (Dh) + c10 hD + c11 h D .
Now we proceed by constructing the most general Lorentz invariant cubic terms with two derivatives:

LhAA = h a1 D A (DA) + a2 D A D A + a3 D A D A + a4 D A D A

+ h a5 D A D A + a6 D A D A + a7 (DA)(DA)
+ a8 D h D A A + a9 (Dh) D A A + a10 (Dh) D A A
+ a11 (Dh) (DA)A + a12 D hD A A + a13 D h(DA)A .
Not all these terms are independent here. Indeed, it is easy to check that up to terms without
derivatives:
D h D A A = h D A D A h D A (DA) + (Dh) D A A ,
D hD A A = h(DA)(DA) hD A D A + D h(DA)A ,
D h D A A = (Dh) (DA)A + (Dh) D A A D h D A A .
Using this freedom and requiring that the Lagrangian will be gauge invariant we finally get:
c2
hA 2
4
while the only non-trivial transformation is:
LhAA = c2 h A A
(30)
A = c2 A .
(31)
The result obtained is of course familiar and could seems trivial, but for what follows it is very
important that linear approximation does not fix the value of c2 coupling constant. In the normal massless gravity one expects c2 = 2 but as we will see later on massive theory is possible
provided c2 = 1 so that Goldstone field A must have non-canonical transformations.
hh-vertex. Now the most general ansatz for transformations has the form:

h = c1 (D + D ) + c2 g D + c3 (D + D ) + c4 g (D ),
= c5 (Dh) + c6 D h + c7 h D + c8 h(D )
and using one more time the freedom to make redefinitions:

h = d1 D ( ) + D ( ) ,
h h + s1 h + s2 g h,
+ s3 h h + s4 h2
one can leave c2 , c5 and c6 as the only non-zero parameters. Then writing most general cubic
terms:

Lhh = a1 D h D h + a2 D h D h + a3 (Dh) (Dh) + a4 (Dh) D h

+ a5 D hD h + D a6 D h h + a7 D h h + a8 (Dh) h

+ a9 D hh + a10 D hh + a11 (Dh) h
95
and using the fact that up to lower derivative terms we have:

(a2 ) = (a3 ) (a7 ) + (a8 )
one can check that there is no non-trivial solution for such vertex.
AA-vertex. In this case one has to consider transformations only, so we write:
A = c1 D + c2 D ,
= c3 (DA) + c4 A D .
Possible field and parameter redefinitions

A = d1 D (),
A A + s1 A ,
+ s2 A 2
allows one to leave c3 as the only non-zero parameter. Then considering the most general cubic
terms:

LAA = a1 D A D A + a2 D A D A + a3 (DA)(DA)

+ D a4 D A A + a5 (DA)A
and taking into account that
D A D A = (DA)(DA) + D (DA)A D D A A
we obtain the following simple Lagrangian
LAA = a0 A 2
(32)
which is trivially gauge invariant.

h-vertex. Here the only possible terms for transformations are:
= c3 D + c4 (D ).
Moreover, using field redefinition
+ s1 h
we can leave c3 as the only non-zero parameter. Then the requirement that the Lagrangian
Lh = a1 h D D + a2 hD D + a3 (Dh) D + a4 D hD
will be gauge invariant gives:
2c3 (d 1)
c3 (d 1)
,
a2 =
,
a3 = a4 .
d 2
d 2
In this, the arbitrary parameter a3 is related with one more field redefinition (which does not
change the structure of gauge transformations)
a1 =
h h + s2 g 2 .
So without lost of generality we can set a3 = 0 and obtain:

2(d 1)
c3
c3 h D D + h(D)2 ,
Lh =
d 2
2
= c3 D .
(33)
Again this result could seems trivial but it is important that coupling constant c3 is not fixed yet.
96
-vertex. In this last and simplest case we have only one possible term in the Lagrangian,
no non-trivial gauge transformations and one possible field redefinition:
L 3 = a1 D D ,
+ s1 2
showing that this vertex is trivial one.

Let us collect together all the pieces obtained so far. Total Lagrangian with two derivatives:
c2
L0 = Lhhh + c2 h A A hA 2 + a0 A 2
4

2(d 1)
c3
+
c3 h D D + h(D)2
d 2
2
(34)
while the only non-trivial gauge transformations look like:

h = 2 D h + D h + D h ,
A = c2 A ,
= c3 D .
(35)
Let us stress once again that parameters c2 and c3 are not fixed yet.
Now we proceed to the next order. Namely, we construct the most general cubic terms with
one derivative for the Lagrangian as well as the most general linear terms without derivatives for
gauge transformations laws. At this order there are no ambiguities related with field redefinitions
so all calculations are completely straightforward. The resulting part of the Lagrangian has the
form:

L1 = m 4h (Dh) A h(Dh) A + 2h D h A

3h D hA 3h D h A + hD hA
+ 2c0 b0 h A D c0 b0 h(AD) + b0 mA D
while additional terms to the gauge transformations are:

4
6d
h = m A + A
g (A ) +
h ,
d 2
d 2
mc3
= c0 c3 (A )
.
A = mh + 2a0 m ,
2
(36)
(37)
(d1)
, a0 = 2cm(d4)
, c2 = 1 and c3 = 1 2cm 2(d4)
. As could easily be seen from
Here b0 = 2c3d2
0 (d2)
0 (d2)
these formulas, in particular from the expressions for the a0 and c3 parameters, in arbitrary space
time dimension d general solution exists for c0 = 0 only. But in d = 4 dimensions there exists
another solution with c0 = 0 then scalar field completely decouples which corresponds to the
self-interaction for partially massless gravity. This fact is clearly connected with the conformal
invariance of partially massless theories namely in d = 4 dimensions [37]. Now we proceed by
considering general solution with c0 = 0 in arbitrary spacetime dimension but we will comment
on partially massless gravity at the end of this section.
The last but not least part of our calculations deals with cubic terms without derivatives in the
Lagrangian. The most general form for these terms looks like:
L0 = b1 h h h + b2 hh h + b3 h3 + b4 h h + b5 h2
+ b6 2 h + b7 3 + b8 h A A + b9 hA 2 + b10 A 2 .
(38)
97
There are no any new terms for the gauge transformations here, so the gauge invariance must be
achieved with the gauge transformations obtained at the previous orders. Indeed, it turns out to
be possible and gives:
(7d 16)
5m2 (3d 7)
,
b2 =
,
6
4
2
2c0 m(d 1) (d 4)m3
b4 = 2b5 =
,
d 2
2c0 (d 2)
(d 2 + 5d 6)m2 (3d 2)(d 4)m4
,
b6 =
2(d 2)2
4c0 2 (d 2)2

d(d + 2)(d 4)m5
b7 = 2dc0 2 d 2 11d + 10 +
,
6c0 3 (d 2)3
m2
b9 =
+ (d 1),
b8 = m2 2(d 1),
2
2c0 (d 1)m (d 4)m3
+
.
b10 =
d 2
c0 (d 2)
b1 = m2 +
b3 =
m2 (2d 5)
+
,
4
6
The dependence of
these coefficients on the spacetime dimension d and cosmological term
(recall that c0 = m2 + (d 2)) appears to be rather complicated so let us give explicit expression for non-derivative terms in d = 4 (any way this dimension is special for us)

2

5
1 3
L0 = m + 2 h h h hh h + h
4
4
3c0 m 2 15m2 2
3m3 3
h +
h
2
4
c0

m2 + 6
+ m2 6 h A A +
hA 2 3c0 mA 2 .
2
A few comments are in order.
+ 3c0 mh h
(39)
One of the most important results of our investigation is that possibility to switch on selfinteraction in linear approximation crucially depends on the non-canonical form of gauge
transformations for vector Goldstone field A (recall, that results of linear approximation are
model independent up to restriction on the number of derivatives). Gauge transformations we
obtained:

h = + + 2 h + D h + D h ,
A = m + A + mh
resembles very much the situation with massive spin 1 particles which we discussed in Section 1:
g
A a = a g abc A b c ,
a = m a abc b c .
2
So by analogy with spin 1 case one can suggest that there are two possible ways to proceed
beyond linear approximation. At first, if one evade an introduction of any additional fields
in the model, then this results in highly non-linear theory with higher and higher derivatives.
98
Indeed, we have explicitly checked that one cannot construct quadratic approximation without introduction of such higher derivatives terms. On the other hand, one can try to introduce
additional fields (analog of Higgs field) in order to restrict the number of derivatives.
Note that exactly as in the free case all the terms containing vector fields (except kinetic
ones) could be hidden if one introduce -covariant derivatives:
2m
6d
A g
mA h ,
d 2
d 2
mc3
A .
= D c0 A +
(40)
2
One can always use the and local gauge transformations to choose the gauge where
A = 0 and = 0. In this, the main difference of massive and massless theories are the
cubic completion of famous FierzPauli quadratic mass terms:

5
1
m2 h h h hh h + h3 .
4
4
h = D h
It is instructive to compare this result with that of [54] (see also [55]) obtained in a very
different context. The investigation of [54] shows that allowed cubic terms must be combination of h h h hh h and h h h 32 hh h + 12 h3 . It is easy to see that
our terms correspond to combination with coefficients 1/2, so at this point our results agree
with that of [54].
In the discussions of possible mass terms for gravity it is very often assumed that the full
theory could be the sum of usual massless gravity plus some invariant mass terms, e.g.
[17].
If we denote effective metric as g and assume that in the lowest approximation g
g + h and g g 2h then we easily obtain:

5
1
m2
g g g (h h h h ) 2m2 h h h hh h + h3
2
4
4
(41)
so the structure of cubic terms does not contradict (up to a factor 2) such assumption. Nevertheless, let us stress that in the massive case the gauge transformations of tensor h is
changed so it is hard (if at all possible) to represent massive theory as a sum of two parts
which are separately gauge invariant.
In the massless limit massive spin 2 particle in (A)dSd decompose into massless spin 2
particle and massive spin 1 (or massless spin 1 and spin 0 in flat space). In this, two gauge
transformations and are completely independent and transformation is just usual
gauge transformation of Abelian vector field. But then mass m = 0 the commutator of
and transformations gives:

(), ( ) h = m D ( ) + D ( ) ,

(), ( ) A = m2
(42)
so if we denote a generator of transformation as P and that of one as D, we get
commutation relation for Weyl group [P , D] mP !
As we have already mentioned, apart form general solution which exists in arbitrary dimension
d, in dS4 and only in this dimension there exist another solution with c0 =
m2 + 2 = 0 which corresponds to self interaction of partially massless spin 2 particle

in de Sitter background. In this, scalar field completely decouples, leaving us with h
99
and A fields (note the absence of h3 terms):

1
Lint = Lhhh + h A A hA 2
4

+ m 4h (Dh) A h(Dh) A + 2h D h A

3h D hA 3h D h A + hD hA .
(43)
Using local gauge transformation one can always choose the gauge A = 0. As in the
free case, the resulting Lagrangian will still be invariant under transformation provided we
supplement it with the appropriate restoring gauge A = 0 transformation. Indeed, let us
look at the form of these transformations when A = 0:

h = D + D + mg + 2 D h + D h + D h + mh ,
A = D + m + mh .
(44)
Then it is easy to see that compensating gauge transformation has to be:

1
D + h D .
m
We have explicitly checked that the Lagrangian for the h field is indeed invariant under
such combined and transformations. In the A = 0 gauge the Lagrangian looks as the
linearization of usual gravity in dS4 background but this holds in the linear approximation
only. We have checked that it is impossible to proceed with quadratic approximation without
introduction of higher derivative terms and/or some other fields.

4. Interaction with matter

In this section we investigate possible interactions of massive spin 2 particles with matter
fields of lower spins. The strategy will be the same as in the previous sectionwe construct the
most general cubic terms with no more than two derivatives in the Lagrangians and corresponding linear terms with no more than one derivative for gauge transformation laws with the only
requirement that the Lagrangian be gauge invariant.
4.1. Spin 0
We start with the simplest case of massive spin 0 particles with the usual free Lagrangian:
1
m0 2 2
L0 = ( )2
(45)
.
2
2
All calculations here are very simple and one can easily check that up to possible field redefinitions an interacting Lagrangian has a form:

c2 m0 2
m
c2
1
m
2
2
h h( ) + ( )
h 4 2
L1 =
(46)
2
2
c0
4
c0
while gauge transformations for field look like:
= c2 .
100
We see that vector field A does not couple at all (because field is inert under transformations), while coupling of scalar field clearly shows an
ambiguity between flat space and
massless limits. Indeed, this coupling depends on m/c0 = m/ m2 + (d 2) and we have two
different limits. On one hand, we can take a massless limit m 0 while keeping cosmological
term small but non-zero. In this, scalar field completely decouples so that massless limit
of massive theory agrees with purely massless theory results. On the other hand, if we take flat
space limit 0 while keeping non-zero mass m, then the coupling constant for scalar field
tends to 1 and does not depend on mass any more. As a result, in the massless limit scalar field
does not decouple from matter field .
4.2. Spin 1
Our next exampleinteraction with massive spin 1 particles. Due to our usage of gauge invariant description this includes massless limit as well. Thus, we introduce two fieldsvector B
and Goldstone scalar and start with the free Lagrangian which has its own gauge invariance:
1
1
m1 2 2
B ,
L0 = B 2 + (D )2 m1 B D +
4
2
2
B = ,
= m1 .
(47)
An investigation of possible hBB and h vertexes with two derivatives goes exactly the same
way as that of hAA and h vertexes in the previous section, so we will not repeat these calculations here and write the corresponding part of Lagrangian:
L1 =
c1
c1
c2
c2
h B B hB 2 h D D + h(D )2
2
8
2
4
+ a0 (D )2 + a1 B 2 + a2 A B
(48)
which is invariant under the following transformations:

B = c1 B ,
= c2 D .
(49)
Now we proceed by adding all possible lower derivatives terms to the Lagrangian and corresponding terms to gauge transformations. This procedure results in:

m
1
L2 = m1 c1 h B D h(BD) + (BD)
2
c0

2
m1 c1
1
m
2
2
h B B hB + B
(50)
2
2
c0
with the only new term in gauge transformations:
= c1 m1 B .
In this, gauge invariance requires c1 = c2 , a0 = cm0 , a1 = c4c1 m(d4)
, a2 = 0. Once again, our results
0 (d2)
show an ambiguity between flat space and massless limits. Only for massless vector field in d = 4
dimensions this ambiguity is absent.
101
4.3. Spin 1/2

Our last exampleinteraction with massive spin 1/2 particles (recall that we prefer to work
with Majorana spinors). As is well known, to describe spinor fields living in curved background
one has to use first order formalism in terms of tetrad e a and Lorentz connection so that
= ea a and
1
[D , D ] = R ab ab = .
(51)
4
2
In this, to describe an interaction of spinor fields with our massive spin 2 particles it is also
convenient to use first order formulation of such particles [56] in terms of three pairs of fields:
(ea , ab ), (A , F ab ) and (, a ). We will not reproduce Lagrangian for such formulation
here (it could easily be found in [56]) because all we need here is the structure of gauge transformations:
m
ha = D a + a + ea ,
2

c0 2 a b
ab
ab
= D
e e b a ,
2
= c0
A = D + m ,
(52)
where apart from a and gauge transformations we have now one more transformation with
parameter ab = ba . Again we start with the free Lagrangian for massive spinor :
m1/2
i
L0 =
(53)
D

2
2
and construct the most general cubic terms (this time with no more than one derivative) compatible with all gauge symmetries. We obtain for interaction Lagrangian:

m1/2
i
i
i
m
ab

h
L1 = h D + h D ab
2
2
8
2
2c0
(54)
while for gauge transformations we get:
1
3m
= D ( )
(55)
.
4
4
In this case an ambiguity between flat space and massless limits exists for massive m1/2 = 0
spinor field only, while the results for massless spinor agree with that of purely massless gravity.
5. Interaction with gravity
There exist well-known difficulties one faces in any attempt to construct interacting theory
for a collection of massless spin 2 particles [3842]. For example, for the case of just two massless spin 2 particles there are only two possibilities. One of them corresponds to two copies of
usual gravities completely independent of each other, while the other possibility which does have
interaction requires that one of these particles has wrong sign of kinetic terms and be a ghost.
Moreover, there are examples of theories of such kind, where one of spin 2 particles is massive,
coming from higher derivative gravity models e.g. [57]. Also, as is well known, there are examples of consistent theories with infinite number of massive spin 2 particles in KaluzaKlein
models, but it could be shown [45,48] that it does not contradict with general results of [38,39].
102
But all this still leaves a possibility to construct a consistent theory where one massless spin 2
particle interacts with one or several massive spin 2 ones. In this section we investigate the case
of one massless and one massive particles. In this, there are two possible cubic vertexes (besides self-interaction). We have checked (though we will not present details of these calculations
here) that it is impossible to construct linear approximation with two massless and one massive
particles (once again up to the restriction on the number of derivatives). As for the other case,
i.e. interaction of massive spin 2 particles with usual gravity, it is not hard to constrict general
covariant vertex with arbitrary number of massless fields but still bi-linear in massive fields, as
we are going to demonstrate.
First of all, let us note that in this section metric g is not just a fixed background any more,
but it is a dynamical field with its own equation of motion:
1
(d 1)(d 2)
g = 0.
R g R +
(56)
2
2
As usual in gravity, we will assume that connection is metric compatible D g = 0 and we have
usual identities:
D R, = D R D R ,
1
D R = D R.
2
As in the case of massive particle living in constant curvature background, it is convenient to
organize the calculations by the number of derivatives. So we start with the sum of kinetic terms
of our three fields h , A and :
1
1
L2 = D h D h D h D h + (Dh) D h D hD h
2
2
1
2(d 1)
2
(D A D A ) +
D D
2
(d 2)
and corresponding gauge transformations with one derivative:
1 h = D + D ,
1 A = D .
(57)
(58)
Covariant derivatives do not commute and, as a result, this Lagrangian is not invariant under these
gauge transformations. But gauge invariance could be restored (up to lower derivative terms we
will take into account later) if one adds to the Lagrangian:
1
1
L = 2R h h + R h h + Rh h Rh2
2
4
and requires that metric g has non-trivial transformation
1 g = 2(D h D h D h ) .
(59)
(60)
Now we proceed with lower derivative terms. The part of the Lagrangian with one derivatives
turns out to be:

4(d 1)c0
L1 = 2m h D A h(DA)
(61)
A D
(d 2)
while corrections to gauge transformation look like:
0 h =
2m
g ,
(d 2)
0 A = m ,
0 = c0 .
(62)
103
Also this requires additional modification of g gauge transformations:

0 g

= 2m A + A

2
2m(d 4)
g (A )
h .
(d 2)
(d 2)
(63)
At last, we must add possible terms without derivatives, the most general form being:
L0 =
b1
b2
h h + h2 + b3 h + b4 2 + b5 A 2 .
2
2
(64)
There are no any additional corrections to gauge transformations at this level, so the gauge invariance must be achieved with the ones we already have. And indeed it turns out to be possible
giving:
(d 1)(d 2),
2
2m2 d(d 1)
,
b5 = 2(d 1).
b4 =
(d 2)2
b1 = m2 + (d 1)(d 2),
b3 =
2(d 1)mc0
,
(d 2)
b2 = m2 +
This time also there is a possibility greatly simplify all the expressions by introduction of covariant derivatives:
h = D h
2m
A g ,
(d 2)
= D c0 A .
(65)
In this, total Lagrangian could be rewritten in the form:

1
1
L = h h h h + (h) h h h
2
2
2(d 1)
1

(D A D A )2 +
2
(d 2)
1
1
2R h h + R h h + Rh h Rh2
2
4
b1
b2 2
h h + h + b3 h + b4 2
2
2
(66)
while gauge transformations for the metric field g look like:

g = 2( h h h ) +

2m(d 4)
g (A ) h .
(d 2)
(67)
Thus, in this approximation an interaction of massive spin 2 particles with gravity exists in any
spacetime dimension d 3 (though from the last equation one can see that d = 4 case is also
special here). In particular, nothing prevent us to consider the c0 0 limit, i.e. interaction of
partially massless spin 2 particle with gravity. But as in the case of self-interaction, as we have
explicitly checked, if one tries to proceed with terms quartic in massive fields than one will find
that higher derivatives interactions and/or some additional fields are necessary. It is instructive
to compare our results here with the investigations of massive spin 2 particle in gravitational
background [1113]. In general, results are similar, but the structure of Rhh terms is slightly
different.
104
6. Conclusion
We hope that the main lesson from our paper is that constructive approach based on gauge
invariant description of massive high spin particles does allow one efficiently investigate possible
interactions of such particles. It is important that due to peculiarity of linear approximation the
results obtained for any particle are completely model independent and do not depend on the
presence of any other fields in the theory. In particular, an impossibility to construct an interaction
in linear approximation means that such interaction does not exist at all. One of the examples of
such no-go results is the absence of self-interaction for partially massless spin 2 particles in
d = 4 dimensions.
One of the striking and unexpected results is that the existence of self-interaction for massive
spin 2 particles crucially depends on non-canonical form of gauge transformations for Goldstone
A field. By analogy with massive spin 1 case, one can assume that in full interacting theory (if it
exists at all) one must deal with non-linear realization of symmetry with higher and higher
derivatives. We are not aware on any works on non-linear realization of spacetime symmetries
where something similar appears.
In the investigation of the massive spin 2 particle interacting with matter (i.e. spin 0, 1, and 1/2
particles) out results clearly show the ambiguity between flat space and massless limits which
reveals itself through the dependence of scalar Goldstone field coupling constant on mass and
cosmological constant. Let us stress once again that this results are also model independent.
Throughout the paper we restrict ourselves with interaction terms with no more than two
derivatives in the Lagrangians (and correspondingly no more than one derivative in gauge transformations). But many of our results clearly show that to construct full interacting theory beyond
linear approximations one unavoidably will have to introduce higher derivatives interactions. But
such higher derivatives interactions could, in principle, change the results obtained here for linear
approximation. This question deserves further study.
References
[1] V.I. Ogievetsky, I.V. Polubarinov, Interacting field of spin 2 and the Einstein equations, Ann. Phys. 35 (1965) 167.
[2] J. Fang, C. Fronsdal, Deformations of gauge groups. Gravitation, J. Math. Phys. 20 (1979) 2264.
[3] K.A. Milton, L.F. Urrutia, R.J. Finkelstein, Constructive approach to supergravity, Gen. Relativ. Gravit. 12 (1980)
67.
[4] R.M. Wald, Spin-two fields and general covariance, Phys. Rev. D 33 (1986) 3613.
[5] I.L. Buchbinder, A. Fotopoulos, A.C. Petkou, M. Tsulaia, Constructing the cubic interaction vertex of higher spin
gauge fields, hep-th/0609082.
[6] C.R. Hagen, L.P.S. Singh, Search for consistent interactions of the RaritaSchwinger field, Phys. Rev. D 26 (1982)
393.
[7] S. Ferrara, M. Porrati, V.L. Telegdi, g = 2 as the natural value of the tree-level gyromagnetic ratio of elementary
particles, Phys. Rev. D 46 (1992) 3529.
[8] A. Cucchieri, S. Deser, M. Porrati, Tree-level unitarity constraints on the gravitational couplings of higher-spin
massive fields, Phys. Rev. D 51 (1995) 4543, hep-th/9408073.
[9] S. Deser, V. Pascalutsa, A. Waldron, Massive spin 3/2 electrodynamics, Phys. Rev. D 62 (2000) 105031, hepth/0003011.
[10] S. Deser, A. Waldron, Inconsistencies of massive charged gravitating higher spins, Nucl. Phys. B 631 (2002) 369,
hep-th/0112182.
[11] I.L. Buchbinder, V.A. Krykhtin, V.D. Pershin, On consistent equations for massive spin-2 field coupled to gravity
in string theory, Phys. Lett. B 466 (1999) 216, hep-th/9908028.
[12] I.L. Buchbinder, D.M. Gitman, V.A. Krykhtin, V.D. Pershin, Equations of motion for massive spin 2 field coupled
to gravity, Nucl. Phys. B 584 (2000) 615, hep-th/9910188.
105
[13] I.L. Buchbinder, D.M. Gitman, V.D. Pershin, Causality of massive spin 2 field in external gravity, Phys. Lett. B 492
(2000) 161, hep-th/0006144.
[14] Yu.M. Zinoviev, Gauge invariant description of massive high spin particles, preprint IHEP 83-91, Protvino, 1983.
[15] S.M. Klishevich, Yu.M. Zinoviev, On electromagnetic interaction of massive spin-2 particle, Phys. At. Nucl. 61
(1998) 1527, hep-th/9708150.
[16] Yu.M. Zinoviev, On massive high spin particles in (A)dS, hep-th/0108192.
[17] N. Arkani-Hamed, H. Georgi, M.D. Schwartz, Effective field theory for massive gravitons and gravity in theory
space, Ann. Phys. 305 (2003) 96, hep-th/0210184.
[18] S. Hamamoto, Possible nonlinear completion of massive gravity, Prog. Theor. Phys. 114 (2006) 1261, hepth/0505194.
[19] M. Bianchi, P.J. Heslop, F. Riccioni, More on la grande bouffe, JHEP 0508 (2005) 088, hep-th/0504156.
[20] K. Hallowell, A. Waldron, Constant curvature algebras and higher spin action generating functions, Nucl. Phys.
B 724 (2005) 453, hep-th/0505255.
[21] I.L. Buchbinder, V.A. Krykhtin, Gauge invariant Lagrangian construction for massive bosonic higher spin fields in
D dimensions, Nucl. Phys. B 727 (2005) 537, hep-th/0505092.
[22] I.L. Buchbinder, V.A. Krykhtin, P.M. Lavrov, Gauge invariant Lagrangian formulation of higher spin massive
bosonic field theory in AdS space, hep-th/0608005.
[23] R.R. Metsaev, Gauge invariant formulation of massive totally symmetric fermionic fields in (A)dS space, hepth/0609029.
[24] L.D. Faddeev, A.A. Slavnov, Invariant perturbation theory for non-linear chiral Lagrangians, Teor. Mat. Fiz. 8
(1971) 297.
[25] D.G. Boulware, S. Deser, Can gravitation have a finite range?, Phys. Rev. D 6 (1972) 3368.
[26] S.V. Babak, L.P. Grishchuk, Finite-range gravity and its role in gravitational waves, black holes and cosmology, Int.
J. Mod. Phys. D 12 (2003) 1905, gr-qc/0209006.
[27] A.N. Petrov, The field theoretical formulation of general relativity and gravity with non-zero masses of gravitons,
gr-qc/0505058.
[28] V.I. Zakharov, Linearized gravitation theory and graviton mass, JETP Lett. 13 (1970) 312.
[29] H. van Dam, M. Veltman, On the mass of graviton, Gen. Relativ. Gravit. 3 (1972) 215.
[30] I.I. Kogan, S. Mouslopoulos, A. Papazoglou, The m 0 limit for massive graviton in dS4 and AdS4 . How to
circumvent the van DamVeltmanZakharov discontinuity, Phys. Lett. B 503 (2001) 173, hep-th/0011138.
[31] M. Porrati, No van DamVeltmanZakharov discontinuity in AdS space, Phys. Lett. B 498 (2001) 92, hep-th/
0011152.
[32] C. Deffayet, G. Dvali, G. Gabadadze, A.I. Vainstein, Nonperturbative continuity in graviton mass versus perturbative
discontinuity, Phys. Rev. D 65 (2002) 044026, hep-th/0106001.
[33] S. Deser, A. Waldron, Gauge invariance and phases of massive higher spins in (A)dS, Phys. Rev. Lett. 87 (2001)
031601, hep-th/0102166.
[34] S. Deser, A. Waldron, Partial masslessness of higher spins in (A)dS, Nucl. Phys. B 607 (2001) 577, hep-th/0103198.
[35] S. Deser, A. Waldron, Null propagation of partially massless higher spins in (A)dS and cosmological constant
speculations, Phys. Lett. B 513 (2001) 137, hep-th/0105181.
[36] E.D. Skvortsov, M.A. Vasiliev, Geometric formulation for partially massless fields, hep-th/0601095.
[37] S. Deser, A. Waldron, Conformal invariance of partially massless higher spins, hep-th/0408155.
[38] R.M. Wald, A new type of gauge invariance for a collection of massless spin-2 fields: I. Existence and uniqueness,
Class. Quantum Grav. 4 (1987) 1267.
[39] R.M. Wald, A new type of gauge invariance for a collection of massless spin-2 fields: II. Geometrical interpretation,
Class. Quantum Grav. 4 (1987) 1279.
[40] N. Boulanger, T. Damour, L. Gualtieri, M. Henneaux, Inconsistency of interacting, multi-graviton theories, Nucl.
Phys. B 597 (2001) 127, hep-th/0007220.
[41] N. Boulanger, Multi-graviton theories: Yes-go and no-go results, Fortschr. Phys. 50 (2002) 858, hep-th/0111216.
[42] S.C. Anco, On multi-graviton and multi-gravitino gauge theories, Class. Quantum Grav. 19 (2003) 6445, gr-qc/
0303033.
[43] L. Dolan, M.J. Duff, KacMoody symmetries of KaluzaKlein theories, Phys. Rev. Lett. 52 (1984) 14.
[44] C.S. Aulakh, D. Sahdev, The infinite-dimensional gauge structure of KaluzaKlein theories I. D = 1 + 4, Phys.
Lett. B 164 (1985) 293.
[45] M. Reuter, Consistent interaction for infinitely many massless spin-two fields by dimensional reduction, Phys. Lett.
B 205 (1988) 511.
[46] Y.M. Cho, S.W. Zoh, Explicit construction of massive spin-two fields in KaluzaKlein theory, Phys. Rev. D 46
(1992) 2290.
106
[47] V.A. Tsokur, Yu.M. Zinoviev, N = 2 supergravity models with gauge KacMoody groups, hep-th/9607034.
[48] O. Hohm, On the infinite-dimensional spin-2 symmetries in KaluzaKlein theories, Phys. Rev. D 73 (2006) 044003,
hep-th/0511165.
[49] E. Alvarez, D. Blas, J. Garriga, E. Verdaguer, Transverse FierzPauli symmetry, Nucl. Phys. B 756 (2006) 148,
hep-th/0606019.
[50] A.H. Chamseddine, Spontaneous symmetry breaking for massive spin-2 interacting with gravity, Phys. Lett. B 557
(2003) 247, hep-th/0301014.
[51] S. Deser, A. Waldron, Partially massless spin 2 electrodynamics, hep-th/0609113.
[52] D.R. Grigore, G. Scharf, Massive gravity as a quantum gauge theory, Gen. Relativ. Gravit. 37 (2005) 1075, hep-th/
0404157.
[53] T. Padmanabhan, From gravitons to gravity: Myths and reality, gr-qc/0409089.
[54] P. Creminelli, A. Nicolis, M. Papucci, E. Trincherini, Ghosts in massive gravity, JHEP 0509 (2005) 003, hep-th/
0505147.
[55] C. Deffayet, J.-W. Rombouts, Ghosts, strong coupling and accidental symmetries in massive gravity, Phys. Rev.
D 72 (2005) 044003, gr-qc/0505134.
[56] Yu.M. Zinoviev, First order formalism for massive mixed symmetry tensor fields in Minkowski and (A)dS spaces,
hep-th/0306292.
[57] G. Magnano, L.M. Sokolowski, Nonlinear massive spin-two field generated by higher derivative gravity, Ann.
Phys. 306 (2003) 1, gr-qc/0209022.
Worldline Green functions for arbitrary

Feynman diagrams
Peng Dai , Warren Siegel
C.N. Yang Institute for Theoretical Physics, Stony Brook University, Stony Brook, NY 11790-3840, USA
Received 25 October 2006; received in revised form 1 February 2007; accepted 6 February 2007
Abstract
We propose a general method to obtain the scalar worldline Green function on an arbitrary 1D topological
space, with which the first-quantized method of evaluating 1-loop Feynman diagrams can be generalized
to calculate arbitrary ones. The electric analog of the worldline Green function problem is found and a
compact expression for the worldline Green function is given, which has similar structure to the 2D bosonic
Green function of the closed bosonic string.
2007 Published by Elsevier B.V.
1. Introduction
The first-quantized method for the calculation of particle scattering amplitudes was suggested
and used by Feynman [1], Nambu [2] and Schwinger [3] in as early as the 1950s. However, it was
not developed much until string theory was invented and used first-quantization as a key method
to calculate the string scattering amplitude [4]. Then, how to apply this method back to particles
was studied in detail. From a certain limit of string theory, Bern and Kosower obtained a set of
first-quantized rules for calculating the scattering amplitudes in YangMills theory at the 1-loop
level [5]. This new method was shown to be equivalent to the ordinary second-quantized method
and much more efficient when calculating 1-loop gluongluon scattering [6]. Later Strassler gave
an alternative derivation of the same rules directly from the first-quantized formalism of the field
theory [7].
E-mail addresses: pdai@grad.physics.sunysb.edu (P. Dai), siegel@insti.physics.sunysb.edu (W. Siegel).

0550-3213/$ see front matter 2007 Published by Elsevier B.V.
doi:10.1016/j.nuclphysb.2007.02.004
108
P. Dai, W. Siegel / Nuclear Physics B 770 (2007) 107122
The generalization to the rules for scalar theory at multi-loop level was then studied by
Schmidt and Schubert [8] and later by Roland and Sato [9]. Green functions on multi-loop vacuum diagrams were obtained by considering these diagrams as one-loop diagrams with insertions
of propagators. The Green function on any vacuum diagram containing a Hamiltonian circuit
could be found by this method.
A natural hope of further generalization is to find the worldline Green function on an arbitrary
one-dimensional topology, without the limitation that this topology must be one-loop with insertions. In this paper, we give such a general method to obtain the Green function for scalar field
theory at arbitrary multi-loop level. We show that the electric circuit can be an analog in solving
this problem.
On the other hand, Mathews [10] and Bjorken [11], from the second-quantized method (usual
field theory), gave a method with the electric circuit analogy to evaluate Feynman diagrams at
arbitrary loop level. Their result of the Feynman parameter integral representation of scattering
amplitudes was, in principle, the most general result, but because of the limitation of secondquantization, diagrams with the same topology except for different number or placement of the
external lines were treated separately, and the analogy between the kinetic quantities on the
Feynman diagram and the electric quantities on the circuit was not completely clear.
Fairlie and Nielsen generalized Mathews and Bjorkens circuit analog method to strings, and
obtained the Veneziano amplitude and 1-loop string amplitude [12]. Although they did not use
the term Green function, they in fact obtained the expression of the Green function on the disk
and annulus worldsheets of the bosonic string with the help of the 2D electric circuit analogy.
These attempts indicate that the problem of solving for the Green function on a certain topological space and the problem of solving a circuit may be related. Indeed, we show that there is
an exact analogy between the two kinds of problems in the 1D case.
In this paper, we first give an introduction to the first-quantized formalism in Section 2. In
Section 3, we find that the differential equation the Green function should satisfy can be solved
by an analogous method for solving the electric circuit. We give a complete analogy among
the problems of finding the Green function, the static electric field and the electric circuit. In
Section 4, we derive a general method to solve the electric circuit and give a compact expression
of the Green function,
1
1
G(,
) = s + vT 1 v,
2
2
where the scalar s, vector v, and matrix are quantities depending only on the topological and
geometrical properties of the 1D space and the position of the external sources and . This
expression is similar to that of the bosonic string [13]. In Section 5, a calculation of the vacuum
bubble amplitude is given to complete the discussion. The method is summarized in Section 6.
Examples of Green functions on topologies at the tree, 1-loop and 2-loop levels are given in
Section 7. Section 8 contains the conclusions.
2. First-quantization
The first-quantized path integral on the worldline of any given topology (graph) for a scalar
particle interacting with a background potential can be written as

1 1
DeDX
2
exp d
e X X + em + eV (X) ,
A(M) =
(1)
Vrep
2
M
109
where M means the particular topology of the worldline being considered and the potential V (X)
is specialized to a background consisting of a set of plane waves for our purpose to calculate the
scattering amplitude
V (X) =
eiki X(i ) ,
i=1
Vrep in (1) is the volume of the reparametrization group. The reparametrization symmetry of the
worldline must be fixed to avoid overcounting.
The path integral (1) gives the sum of all graphs of a given topology with an arbitrary number
of external lines. This can be seen by expanding the potential exponential. If we consider only
graphs with a certain number of external lines, the amplitude can be written as
N

1
DeDX
exp
d e1 X X + em2
di e eiki X(i ) , (2)
A(M, N) =
Vrep
2
M
i=1 M
where N is the number of external lines.

To fix the reparametrization symmetry, we can simply set the vielbein e to 1. However, by doing this, we have left some part of the symmetry unfixed (the Killing group, like the conformal
Killing group in string theory), as well as over-fixed some non-symmetric transformation (the
moduli space, also like in string theory). To repair these mismatches by hand, we add integrals
over the moduli space and take off some of the integrals over the topology. The general form
of the amplitude (2) after fixing the reparametrization symmetry is (up to a constant factor from
possible fixing of the discrete symmetry which arises due to the indistinguishable internal lines)
A(M, N)

1
dc
dTa DX exp
d X X + m2
eiki Xi (i ) ,
=
2
i
a=1F
c
/C
M
M
a
(3)
where C denotes the external lines whose position has to be fixed to fix the residual symmetry,
and in the particle case the moduli Ta are just the proper times represented by the lengths of the
edges in the graph of topology M. Further evaluation by the usual method of 1D field theory gives
the following expression (with the delta function produced by the zero mode integral suppressed
since it trivially enforces the total momentum conservation in the calculation of the scattering
amplitude),

1
dc exp
dTa VM (Ta )
ki kj GM (i , j ) ,
A(M, N) =
(4)
2
i,j
a=1F
c
/C
M
a
where VM (Ta ) is the amplitude of the vacuum bubble diagram,

1
2
VM (Ta ) = DX exp
d X X + m
2
(5)
and GM is the Green function which satisfies the following differential equation on the worldline
of topology M:
M (, ) = ( ) + ,
G
(6)
110
where is a constant, of which the integral over the whole 1D space gives 1, i.e., the inverse of
the total volume (length) of the 1D space
1
.
M d
=
This constant is required by the compactness of the worldline space.

The scattering amplitude (4) is a general expression for a worldline of any topology with an
arbitrary number of external lines. The form of the vacuum bubble amplitude and Green function
depends on the topology. For example, the amplitude of the one-loop 1PI graph with N external
lines (Fig. 1(A)) is

A(, N) =
dT V
N1

dc
c=1

1
ki kj G (i , j ) ,
exp
2
i,j
where

1
V (T ) = exp T m2 T D/2 ,
2
and

1
( )2

.
G (, ) = | |
2
T
Another example is the amplitude of the two-loop 1PI graph with N external lines (Fig. 1(B)),
3
A(, N) =
a=1 0
dTa V (T1 , T2 , T3 )

N

c=1

dc

1
ki kj G (i , j ) ,
exp
2
i,j
where V and G will be determined in the following sections.
Fig. 1. (A) One-loop diagram with N external lines. The topology of a circle has one modulus, the circumference T
of the loop. Also there is one residual symmetry, which has to be fixed by taking off one of the proper-time integrals.
(B) Two-loop diagrams with N external lines. The topology of this kind has three moduli, the lengths T1 , T2 and T3 of
the three edges. And there is no unfixed residual symmetry.
111
3. Green function
We note that the differential Eq. (6) is just the Poisson equation the electric potential should
satisfy when there is a unit positive charge at and a constant negative charge density of magnitude over the whole space. This suggests to us to consider the corresponding static electric
problem where the Green function is just the electric potential at of the above setup.
To demonstrate the general solution of the Poisson equation, we solve for the Green function on the two-loop worldline as an example. Consider the 1D topological space as shown in
Fig. 2(A) with a unit positive charge at and a unit negative charge uniformly distributed over
the whole space. According to the above argument, the Green function G(, ) is the electric
potential at . Now we can use Gauss law and single-valuedness of the potential to write down
equations and solve for the expression of the potential. The potential indeed gives a right form
for the Green function, but it contains many terms that will be canceled out when calculating the
scattering amplitude using Eq. (4), and hence can be further simplified.
Note that the setup of the static electric problem in Fig. 2(A) can be regarded as the superposition of the 2 setups in Fig. 2(B), (C). Let G(, ) denote the potential at of the setup shown
in Fig. 2(A), and G(,

) denote the potential at in Fig. 2(B). The potential at in Fig. 2(C) is
then G(, ). Thus
G(, ) = G(,
) + G(, ).
We further define G(,

) as the symmetric part of G(,
), i.e.,

1
G(,
) G(,
)
) + G(,
2

1
1
1
1
= G(, ) + G( , ) G(, ) G( , ).
2
2
2
2
(7)
Fig. 2. (A) Two-loop topological space with a unit positive charge at and a unit negative charge uniformly distributed
over the whole space. (B) Two-loop topological space with a unit positive charge at and a unit negative charge at .
(C) Two-loop topological space with a unit positive charge at and a unit negative charge uniformly distributed over the
whole space.
112
If we use G(,
) instead of G(, ) in Eq. (4), the sum can be rewritten as
1
i , j )
ki kj G(
2
i,j

1
1
=
ki k j
G(i , j ) + G(j , i ) G(i , i ) G(j , j )
2
2
i,j

1
1
=
ki k j
G(i , j ) + G(j , i )
2
2
i,j
1
2
ki kj G(i , j ).
i,j

We have used conservation of momentum ki = 0 in the second step and rearranged the sum
mands in the third step. This shows that using G(,
) in Eq. (4) is equivalent to using G(, ).
Note that same procedure is usually applied to construct the Green function for bosonic strings
[13],
w) = G(z, w) 1 G(z, z) 1 G(w, w),
G(z,
2
2
which is similar to Eq. (7).
) is the electric
Now we have to look for an electric field analog of G(,
). Since G(,
potential at of the setup shown in Fig. 3(A), it can be written as the potential at plus the po , ) is the potential at of the setup shown in Fig. 3(B).
tential difference from to . And G(

The sum of G(, ) and G( , ) just gives the potential difference from to of Fig. 3(A)
because the potential at of Fig. 3(A) cancels the potential at of Fig. 3(B). G(,
) is half

that potential difference, and therefore is just the potential difference from to of Fig. 3(C).
Fig. 3. (A) Two-loop topological space with a unit positive charge at and a unit negative charge at . (B) Two-loop
topological space with a unit negative charge at and a unit positive charge at . (C) Two-loop topological space with a
half unit positive charge at and a half unit negative charge at .
113
It is now clear that, to write down the expression of the scattering amplitude, we only have to
and use the following formula:
know the symmetric Green function G,

i , j ) .
dTa VM
dc exp
ki kj G(
A(M, N) =
(8)
a=1 F
cC
/ M
i<j
This simplifies the expression of the Green function.
Now that G(,

) is the potential difference from to in Fig. 3(C), we can apply Gauss
law (of 1D space) and single-valuedness of the potential to write down the equations the Green
function (electric potential ) and its first derivative (electric field E) should satisfy. We assume
the direction and value of E as shown in Fig. 4(A) and and are respectively on T1 and T3 .
According to Gauss law, we have the following equations:
1
a c + d = 0,
b + c + e = 0,
a+b= ,
2
The single-valuedness of the potential requires
cT2 + b(T1 ) = a ,
1
d e = .
2
cT2 + d = e(T3 ).
If we regard the electric field E as the current I , the electric potential as the voltage V , and the
length on the 1D space l as the resistance r, these equations are just the Kirchhoff equations of
a circuit of the same shape as the worldline and with half unit current going into the circuit at
and half unit current coming out at , as shown in Fig. 4(B). It is easy to see that this equivalence
between 1D static electric field and circuit is valid for an arbitrary 1D topological space. We have
the following relations for 1D (note that there is no cross-sectional area in the 1D case):
E = I,
l = r,
= V,
where and (= 1/ ) are respectively the 1D conductivity and resistivity, and the relation
El = is equivalent to Ohms law I r = V . In addition, Gauss law is equivalent to Kirchhoffs
Fig. 4. (A) Two-loop topological space with a half unit positive charge at and a half unit negative charge at . The
lengths of the three arcs are T1 , T2 , T3 . and are respectively on T1 and T3 and denote the lengths from the origin. The
magnitudes of the electric field on each part of the space are denoted by a e and the directions are chosen arbitrarily.
(B) Two-loop circuit with a half unit current input at and withdrawn at . The resistances of the three arcs are T1 , T2 ,
T3 . and are respectively on T1 and T3 and denote the resistance from the origin. The currents on the parts of the
circuit are denoted by a e and the directions are chosen arbitrarily.
114
Table 1
Analogy among quantities in three kinds of problems
Particle
Static electric field
Electric circuit
Green function G
Position x
Momentum p
Proper time
Action S
External force F
Potential difference
Electric potential
Electric field E
Length l
Energy U
Electric charge Q
Voltage difference V
Voltage V
Current I
Resistance r
Power P
emf E
current law, and the single-valuedness of the potential is equivalent to Kirchhoffs voltage law.
Thus the two problems are indeed equivalent. The Green function G(,
) on a particular 1D
topological space M can be understood as the voltage difference from to when a half unit
current is input into the circuit (1D space) at and withdrawn at .
Above we have shown that in the 1D case the problem of solving for the first-quantized particle Green function, the potential difference in a static electric field and the voltage difference in a
circuit are analogous. Before we solve the circuit problem for the Green function, it is interesting
to give a complete analogy among the quantities in these three kinds of problems (Table 1).
4. Solving the circuit
The next thing we have to do is to find a general expression for the Green function, i.e.,
a method to obtain the voltage difference described above. A formula has already been given
in graph theory (see, e.g., [14]). Here we summarize this result and develop it into a formula
which is similar to the known form of the Green function on the 2D worldsheet of bosonic string
theory [13].
The voltage difference from to can be obtained by the following steps:
(1) Connect and by a wire with zero resistance.
(2) The resulting graph has vertices V = v1 , . . . , vn and edges E = e1 , . . . , em1 , em . Denote by
em the edge (wire) we have just added in. Set the directions of all the edges arbitrarily.
(3) Assign voltage, current and resistance on each edge; they can be written in the form of
vectors:
U = (U1 , U2 , . . . , Um )T ,
I = (I1 , I2 , . . . , Im )T ,
r = (r1 , r2 , . . . , rm )T .
Also define the external-force-driven voltage u, which has, in our case, only one non-zero
component (the last one). Assume it has magnitude 1:
u = (u1 , u2 , . . . , um )T = (0, 0, . . . , 1)T .
(9)
(4) Find all the independent loops in the graph: There should be (m n + 1). These loops
can be identified by the following method: Choose an arbitrary spanning tree (a connected
subgraph that contains all the vertices and is a tree) of the graph. There are always (mn+1)
chords (the edges not belonging to the spanning tree). Adding a chord to the spanning tree
will generate a one-loop graph. Thus each chord gives a loop in the graph, and all the loops
obtained by this way are independent of each other. Therefore there are (m n + 1) loops.
Assign an arbitrary direction to each loop.
115
(5) Define matrices B, C and R as follows:

1
if vi is the initial vertex of ej ,
Bij = 1 if vi is the terminal vertex of ej ,
0
otherwise,

1
if ei is in the same direction of the loop lj ,
Cij = 1 if ei is in the opposite direction of the loop lj ,
0
otherwise,

r i = j,
Rij = i
0 i = j.
(6) With the above definitions, we can write down Kirchhoffs current law, Kirchhoffs voltage
law and Ohms law in compact forms as follows:
BI = 0,
C T U = 0,
U = RI + u.
The solution to the current on each edge is

1
I = C C T RC C T u.
(7) The total resistance between and (excluding the added [last] edge) is then minus the
voltage on the last edge divided by the current on that edge, i.e.,
Um
um
1
=
= .
Im
Im
Im
The Green function is then minus this resistance times the current, one half, since it is the
voltage difference from to ,
R(, ) =
1
1
1
,
G(,
) = R(, ) =
(10)
= T
T
2
2Im
2u C(C RC)1 C T u
where the last step comes from extracting the last component of I by using the vector u
defined in Eq. (9).
We can further simplify Eq. (10), by considering the physical meaning of each part of the
denominator:
(1) C T u and uT C: Since u = (0, . . . , 0, 1)T , C T u is an (m n + 1)-component column vector
whose ith component is the direction of the last (mth) edge on the ith loop (1 for same,
1 for opposite, and 0 for not on the loop). If we appropriately choose our independent
loops, we can achieve that the mth edge is only on the (m n + 1)th (last) loop and has the
same direction as this loop. This is always achievable in steps: (a) Choose the spanning tree in
such a way that the mth edge does not belong to the spanning tree, i.e., is a chord. (b) Define
the loop generated by adding the mth edge to the spanning tree to be the (m n + 1)th loop.
(c) Define the direction of the (m n + 1)th loop to be the same as the mth edge. By doing
so, we find that C T u is just a (m n + 1) component column vector with the last component
non-vanishing and of value 1.
C T u = (0, . . . , 0, 1)T .
And uT C is the transpose of C T u. Define for convenience
P C T u.
116
Note that PT MP gives the [(m n + 1), (m n + 1)] component of any matrix M with
dimension (m n + 1) (m n + 1).
(2) C T RC: C T RC is an (m n + 1) (m n + 1) matrix. The components can be interpreted
as the sum of the signed resistances,

T
C RC ij =
f (k, i, j )rk ,
k all edges
where rk is the resistance on kth edge and f is the sign:

1
nth edge is on both loops i and j with the same orientation,
f (n, i, j ) = 1 nth edge is on both loops i and j with opposite orientations, (11)
0
nth edge is not on both loops i and j.
We define for convenience

v
M
C T RC,
vT s
(12)
where is an (m n) (m n) matrix, v is an (m n)-component vector, and s is a

scalar.
Now the formula for the Green function (10) becomes
=
G
1
2PT M1 P
Since PT M1 P is just the [(m n + 1), (m n + 1)] component of M1 , we have

PT M1 P =
det
.
det M
So, we have
= det M .
G
2 det
Next we evaluate det M. By the usual matrix algebra (e.g., defining the determinant by a
Gaussian integral and doing the m n integrals first),

v
= (det ) s vT 1 v .
det M = det T
s
v
So we have the following expression for the Green function:
= det M = 1 s + 1 vT 1 v.
G
2 det
2
2
(13)
5. Vacuum bubble
To complete this general method of writing down the scattering amplitude, we need to give the
expression for the vacuum bubble amplitude with a given topology VM (Ta ) defined in Eq. (5).
This can always be achieved by evaluating the bubble diagram by the second-quantized method,
but here we give a derivation by direct calculation in first-quantization. Note that the path integral
(5) is the sum over all possible momentum configurations of the product of the expectation values
117
of the free evolution operator between two states at the ends of each edge:

1

T (p2 +m2 )/2

2
2
a

pa =
Ta pa + m
pa e
,
VM (Ta ) =
exp
2
{p}
{p}
a=1
a=1
where pa is the momentum of the particle traveling on the ath edge. The sum over all the configurations of pa can be written as the integration over all the possible values of pa , but they are
not independent from each other. Each of the pa s can be written as a linear combination of L
ki s where L is the number of loops of the graph and ki is the loop momentum on the ith loop.
So the amplitude VM (Ta ) can be written as

L

L
2

1
1
2
Ta m
dki exp
Ta
gai ki
,
VM (Ta ) = exp
2
2
i
a=1
a=1
(14)
i=1
where
1
edge a has the same direction as loop i,

1 edge a has the opposite direction to loop i,
0
edge a is not on loop i.
gai =
It is easy to see the following points: (a) If edge a is on loop i, there is a Ta ki2 term in the
underlined sum in Eq. (14), and vice versa. (b) If edge a is on both loop i and loop j , there
is a term 2Ta ki kj in the sum, and vice versa. Further, if on edge a both loops have the same
(opposite) direction, there is a positive (negative) sign before the term, and vice versa. Thus if
we use the factor f (a, i, j ) defined in Eq. (11), we can write the underlined part in Eq. (14) in
a compact form, and further in terms of the period matrix according to the definition (12), or
definition (17) below:

Ta
a=1
2
gai ki
i=1

L

f (a, i, j )Ta ki kj =
a=1 i,j =1

L

i,j =1

f (a, i, j )Ta ki kj
a=1
ij ki kj .
i,j =1
The amplitude VM (Ta ) can then be calculated easily:

L

L
1
1
2
VM (Ta ) = exp
Ta m
dki exp
ij ki kj
2
2
i
a=1
i,j =1

1
Ta m2 (det )D/2 ,
= exp
2
a=1
where D is the dimension of the spacetime.
(15)
118
6. Summary
We now summarize our results for general amplitudes. To find the Green function:
(1) Consider the circuit (topology M with two more vertices at and respectively) as a
graph. Assign a number to each edge. Define arbitrarily the direction of each edge. Assign a
number to each loop. Define arbitrarily the direction of each loop.
(2) Find the period matrix by the following definition:

f (k, i, j )rk ,
ij =
(16)
k all edges
where
1
nth edge is on both loops i and j with the same orientation,

1 nth edge is on both loops i and j with opposite orientations,
0
nth edge is not on both loop i and j.
f (n, i, j ) =
Each rk may be , , a Ta of the topological space M without external lines, Ta , or

Ta . It is obvious that does not depend on and , and only depends on the properties
of M: To see this, note that the graph of the circuit has just two more vertices (at positions
and ) than the graph of M and each of them may separate an edge in the graph of M
into two sub-edges. But these changes on the graph do not affect the period matrix: They
neither give new loops nor eliminate loops and the sub-edges will always be on or off a loop
simultaneously. Thus the period matrix of the graph of the circuit is just the period matrix of
the graph of M. And one only has to calculate it once for one certain topology,
ij =
(17)
f (a, i, j )Ta .
a=1
(3) Find a path (call it reference path) connecting and . Choose its direction arbitrarily.
Define the scalar s as the total resistance on the reference path.
(4) Find the vector v defined as follows:

f (k, i, 0)rk ,
vi =
k all edges
where 0 means the reference path.

(5) The Green function is given by
= 1 s + 1 vT 1 v.
G
2
2
The amplitude is then given by

i , j )
A(M, N ) =
dTa VM
dc exp
ki kj G(
a=1F
cC
/ M
i<j
with the Green function as above, and the vacuum bubble amplitude is

1
Ta m2 (det )D/2 .
VM (Ta ) = exp
2
a=1
119
Although we have defined and derived all these quantities s, v and in terms of the concepts
in the circuit problem, it is easy to give the worldline geometric interpretation by noting that the
resistance is an analog of the proper time, i.e., the length of the worldline (Table 1). s is just the
total proper time of the reference path, and the components of v are the sums of the signed
proper time on the common edges of the reference path and each loop. Entries of are sums
of the signed proper time on the common edges of each pair of loops. (All the signs are given
by f .) The components of v and entries of can also be expressed as integrals of the Abelian
differentials on the loops of the worldline,

vi =

i ,
ij =
j ,
i
where i is the line element on loop i and the second integral is around loop i along its direction. Then our expression for the particle amplitude has a similar structure to the bosonic string
amplitude, with the Green function on the 2D worldsheet [13]:

G(w,
z) = 2 lnE(w, z) + 2 Im
z
w
(Im )1 Im
z
,
w
where E is the prime form, the vector contains the basis of the Abelian differentials and the
matrix is the period matrix.
7. Examples
Here we give some examples of obtaining the Green functions on different topologies. For the
finite line of length T in Fig. 5, there is no loop, so there is no period matrix nor vector v. s is
just the total resistance between and . So the Green function is
(, ) = 1 s = 1 | |
G
2
2
and

1
2
V (T ) = exp T m .
2
The amplitude is then given by Eq. (8). One just has to note that there is one modulus T for this
case and to fix the residual symmetry: Two of the external lines should be fixed at one end of the
line and another two should be fixed at the other end.
For the circle, there is one loop as shown in Fig. 6(A). So the period matrix is 1 1. The only
entry of is then
= T.
Fig. 5. The topology of a line with length T . There is no loop and hence no period matrix nor vector v. The only path
between and is the edge connecting the two vertices e1 , so we choose it as the reference path.
120
And according to the definition, s and v are

s = | |,
v = | |.
Thus the Green function on the circle is

2
(, ) = 1 s + 1 vT 1 v = 1 | | + ( )
G
2
2
2
2T
and the vacuum bubble amplitude is

1
V (T ) = exp T m2 T D/2 .
2
For the Green function on the 2-loop graph shown in Fig. 6(B), the period matrix is 2 2 and
v is a two-component vector. , s and v are

T2
T1 + T2
v = ( , )T .
,
s = + ,
=
T2
T 2 + T3
So the Green function is, by plugging all the above into Eq. (13),

2

(, ) = T1 (T2 + T3 ) + T1 (T2 + T3 ) T3 + T2 (T3 )( + )
G
2(T1 T2 + T2 T3 + T3 T1 )
and the vacuum bubble amplitude is

1
2
V (T1 , T2 , T3 ) = exp (T1 + T2 + T3 )m (T1 T2 + T2 T3 + T3 T1 )D/2 .
2
Fig. 6. (A) The one-loop topology. T is the circumference of the circle. There are two edges (e1 , e2 ) and two vertices
(v1 , v2 ). and are the lengths from v1 and v2 to the origin though a counterclockwise path. The directions of the edges
are chosen arbitrarily and marked in the figure. There is only one loop and it is marked by dotted lines. The reference
path is marked by a dashed line. (B) A two-loop topology. T1 , T2 and T3 are the lengths of the three arcs. and are
respectively the length on the 3rd and 1st arc from the origin. (If the Green function with and on different arcs is
needed, simply repeat the steps for this special case.) There are 5 edges (e1 to e5 ) and 4 vertices (v1 to v4 ). The directions
of the edges are chosen arbitrarily and marked in the figure. There are two independent loops and they are marked by
dotted lines. The reference path is marked by a dashed line.
121
8. Conclusions
Based on the first-quantized formalism, the discussion in this paper gives a new method to
evaluate the scattering amplitude of scalar field theory at arbitrary loop level. (The procedure
was given in Section 6.) The form of the Green function is shown to be similar to that on the
worldsheet in bosonic string theory. The method applies not only to 1PI graphs, but arbitrary
graphs that might appear in S-matrices.
The amplitude obtained by the first-quantized method in this paper easily can be shown to
be equivalent to the amplitude from second-quantization. Further, the singularities of general
diagrams can be discussed and the Landau conditions can be obtained. The occurrence of a singularity on the physical boundary was shown to be related to a sequence of interactions connected
by real particle paths by Coleman and Norton [15]. Here this same picture emerges more naturally: The Feynman diagram is interpreted as particle path (through background fields) and the
integral over the proper time is introduced from the beginning.
The extension to diagrams with spinning particles will be the next thing to understand. It is
expected to be parallel to Strasslers discussion on various theories [7], with the Green function
on a circle replaced by those on other topologies. However, although the 2-loop Green function
has been known for a long time, no full-fledged extension to even 2-loop YangMills theory has
yet been found, because of the complexity and difficulty entering at multi-loop level. Previous
attempts include generalization from Bern and Kosowers string approach [16,17] and Strasslers
particle approach [18,19].
It is now clear that scalar field theory can be seen as the limit of the bosonic string theory, by
suppressing the length of the string. Can the bosonic string worldsheet be seen as the sum of all
the Feynman diagrams of the scalar field theory that are random lattices of the worldsheet? This
is another interesting question that may be related to the result in this paper.
Acknowledgement
This work was supported in part by National Science Foundation Grant No. PHY-0354776.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
R.P. Feynman, Phys. Rev. 80 (1950) 440.

Y. Nambu, Prog. Theor. Phys. 5 (1950) 82.
J.S. Schwinger, Phys. Rev. 82 (1951) 664.
Y. Nambu, in: Proceedings of International Conference on Symmetries and Quark Model, Detroit, USA, June 1969;
L. Susskind, Phys. Rev. D 1 (1970) 1182;
L. Susskind, Nuovo Cimento A 69 (1970) 457;
H.B. Nielsen, in: 15th International Conference on High Energy Physics, Kiev, Ukraine, September 1970.
Z. Bern, D.A. Kosower, Nucl. Phys. B 379 (1992) 451.
Z. Bern, D.C. Dunbar, Nucl. Phys. B 379 (1992) 562.
M.J. Strassler, Nucl. Phys. B 385 (1992) 145, hep-ph/9205205.
M.G. Schmidt, C. Schubert, Phys. Lett. B 331 (1994) 69, hep-th/9403158.
K. Roland, H.T. Sato, Nucl. Phys. B 480 (1996) 99, hep-th/9604152.
J. Mathews, Phys. Rev. 113 (1959) 381.
J.D. Bjorken, PhD Thesis, Stanford University, 1959;
J.D. Bjorken, S.D. Drell, Relativistic Quantum Field Theory, McGrawHill, 1965, pp. 220226.
D.B. Fairlie, H.B. Nielsen, Nucl. Phys. B 20 (1970) 637.
E.P. Verlinde, H.L. Verlinde, Nucl. Phys. B 288 (1987) 357;
E. DHoker, D.H. Phong, Rev. Mod. Phys. 60 (1988) 917.
122
[14]
[15]
[16]
[17]
[18]
[19]
B. Bollobas, Graph Theory, Springer-Verlag, 1979, p. 40.

S. Coleman, R.E. Norton, Nuovo Cimento 38 (1965) 438.
P. Di Vecchia, A. Lerda, L. Magnea, R. Marotta, Phys. Lett. B 351 (1995) 445, hep-th/9502156.
P. Di Vecchia, L. Magnea, A. Lerda, R. Russo, R. Marotta, Nucl. Phys. B 469 (1996) 235, hep-th/9601143.
H.T. Sato, M.G. Schmidt, Nucl. Phys. B 560 (1999) 551, hep-th/9812229.
H.T. Sato, M.G. Schmidt, C. Zahlten, Nucl. Phys. B 579 (2000) 492, hep-th/0003070.
T-duality with H -flux:

Non-commutativity, T-folds and G G structure
Pascal Grange a,b , Sakura Schfer-Nameki a,b,c,
a II. Institut fr Theoretische Physik der Universitt Hamburg, Luruper Chaussee 149, 22761 Hamburg, Germany
b Zentrum fr Mathematische Physik, Universitt Hamburg, Bundesstrasse 55, 20146 Hamburg, Germany
c California Institute of Technology, 1200 E California Blvd., Pasadena, CA 91125, USA

Abstract
Various approaches to T-duality with NSNS three-form flux are reconciled. Non-commutative torus fibrations are shown to be the open-string version of T-folds. The non-geometric T-dual of a three-torus
with uniform flux is embedded into a generalized complex six-torus, and the non-geometry is probed by
D0-branes regarded as generalized complex submanifolds. The non-commutativity scale, which is present
in these compactifications, is given by a holomorphic Poisson bivector that also encodes the variation
of the dimension of the world-volume of D-branes under monodromy. This bivector is shown to exist
in SU(3) SU(3) structure compactifications, which have been proposed as mirrors to NSNS-flux backgrounds. The two SU(3)-invariant spinors are generically not parallel, thereby giving rise to a non-trivial
Poisson bivector. Furthermore we show that for non-geometric T-duals, the Poisson bivector may not be
decomposable into the tensor product of vectors.
1. Introduction
Compactifications with H -flux are known to give rise to topology changes and even to nongeometric situations when T-duality is performed along directions which have non-trivial support
of the NSNS H -flux [18]. Non-geometry occurs for example in the very simple situation of a
three-torus endowed with an H -flux proportional to its volume form. Consider namely the threetorus as a trivial T 2 -fibration over a circle. Upon T-duality along the fibre, the metric picks up
* Corresponding author at: California Institute of Technology, 1200 E California Blvd., Pasadena, CA 91125, USA.
E-mail addresses: pascal.grange@desy.de (P. Grange), ss299@theory.caltech.edu (S. Schfer-Nameki).

doi:10.1016/j.nuclphysb.2007.02.003
124
P. Grange, S. Schfer-Nameki / Nuclear Physics B 770 (2007) 123144
a factor that makes it shrink under monodromy around the base circle. The monodromy around
the base is a non-trivial element of the O(2, 2; Z) group acting on the two-torus. This prevents
a three-dimensional global Riemannian description from existing. Further T-dualizing along the
base leads to more pathological situations, where points do not exist even in a local coordinate
patch, and the fibres are conjectured to become non-associative [9,10]. We will restrict ourselves
to the case of two T-dualities, and assume that local coordinate patches do exist. Progress in the
description of non-associative T-duals was achieved in the recent paper [11], which also contains
observations on the open-string metric and non-commutativity for two T-dualities that have some
overlap with ours.
Essentially three conjectures have been put forward for the description of the T-dual of a torus
with H -flux:
(I) Field of non-commutative tori: Mathai and Rosenberg proposed that T-dualizing along a
two-torus with non-zero H -flux yields a fibration by (or more precisely: field of) noncommutative tori. In particular, this fibration is encoded in a closed one-form, which is
obtained by integrating the NSNS-flux along the fibre directions [7,12,13].
(II) T-folds: these are spaces where T-dualities can act as transition functions between local
patches [8]. The T-dualized directions are doubled, and T-duality transformations may patch
the doubled fibres together. A sigma model with a T-fold as its target space was proposed,
and its boundary conditions were studied in [1418].
(III) G G structure compactifications: SU(3) SU(3) structure manifolds are characterized in terms of a pair of pure spinors, constructed as bilinear combinations of a pair
SU(3)-invariant spinors of Cliff(6). In case the SU(3)-invariant spinors are not parallel
to each other, their linear independence is encoded by a non-vanishing one-form, and
the discrepancy between left- and right-moving complex structures is a potential source
of non-geometry and/or non-commutativity. Moreover, [19,59] suggest the relevance of
SU(3) SU(3) structures for mirrors of NSNS-flux compactifications.
These directions of research have developed somewhat independently from each other, and it
is natural to ask if they are compatible. It is also natural to expect that techniques from generalized
complex geometry la Hitchin and Gualtieri [20,21] should bring some insights into the problem
for at least two reasons:
Firstly, generalized complex (GC) spaces have been related to non-commutativity in two instances: a non-commutativity scale is induced by the (0, 2) component of a B-field [22], and the
master equation of the generalized B-model [23] admits deformations by holomorphic Poisson
bivectors into a Poisson sigma model, which is known to induce star-products in the algebra of
observables [24];
Secondly, the doubling of the torus fibres in T-folds reminds one of the sum of tangent
and cotangent spaces considered in generalized complex geometry. But GC spaces have more
structure than T-folds, indeed, in [8,17] T-folds were pointed out to be a real version of GC
spaces. Moreover, elements of O(2, 2; Z) called B-transforms and -transforms act on maximally isotropic subspaces as symmetries of the inner product.
We shall therefore use as a main technical tool the geometry of pure spinors, that are in oneto-one correspondence with generalized complex branes, and building blocks for SU(3) SU(3)
structure compactifications.
Our conjectures, which we will justify in the case of tori with H -flux, are:
125
(I) vs. (II): The proposal (I) by Mathai and Rosenberg claims that the T-dual to a T 3 compactification with H -flux along two of the T-dualized directions yields a non-commutative
torus fibration. This is reconciled with Hulls T-fold proposal by showing that the metric seen
by the open strings on a T-fold is precisely the one on the non-commutative torus fibration.
Thus, the proposal (I) is the open-string version of (II). This connection is discussed from
various independent angles in Sections 2, 3 and 4.
(II) vs. (III): when both approaches are applicable as for the T 6 with H -flux, they yield the
same T-dual or mirror geometry.
(III) vs. (I): We show that for a generic SU(3) SU(3) structure compactification, where the
two SU(3)-invariant spinors are not aligned, there exists a Poisson bivector which parametrizes non-commutative deformations. The non-commutativity is however again only relevant for the open-string sector. This relation is discussed in Section 5. As for the mirror of a
six-torus with H -flux, we observe that the Poisson bivector can in fact not be decomposed in
terms of vectors, which seems to indicate that not all the possible non-commutativity scales
are inherited from SU(3) SU(3) structures.
2. T-folds and non-commutative tori
In this section we shall mainly be concerned with the connection between non-commutativity
and T-folds. We shall study this in the case of the simplest non-trivial example, which already
illustrates the main point: the MathaiRosenberg non-commutative torus-fibrations are the openstring version of T-folds. This observation will then be discussed from the generalized geometry
point of view in the next section.
The simplest example that exhibits all the key features is the T 3 -compactification with k units
of NSNS three-form flux H H 3 (T 3 , Z). We shall generally refer to NSNS-flux supported on
a torus bundle E with base B and fibre F of the type H H n (F ) H 3n (B) as an n-legged
H -flux. Thus, the one-legged case is known to have a purely geometric T-dual. Our main focus
is on the two-legged case, which will be shown to have a non-geometric T-dual.
In order to understand the T-dual along two fibre directions, we consider the three-torus as a
T 2 -bundle over S 1 (parametrized by x) and dualize along the fibre directions parametrized by y
and z. The metric and B-field can be chosen as
ds 2 = dx 2 + dy 2 + dz2 ,
B = kx dy dz.
(2.1)
Due to the B-field the monodromy Mk around the S 1 is non-trivial and reads
1
0
Mk =
0
k
0
1
k
0
0
0
1
0
0
0
,
0
1
(2.2)
in a basis adapted to the coordinates (y, z, y,

z ), where y and z are T-dual to y and z. Naively
applying the standard Buscher rules along the fibres yields the T-dual background
ds 2 = dx 2 +
2

1
dy + dz2 ,
2
2
1+k x
B=
kx
dy dz,
1 + k2x 2
(2.3)
126
and the monodromy obtained after action of the T-duality matrix
0 0 1 0
0 0 0 1
gyz =
1 0 0 0
0 1 0 0
along the fibres is
1 0 0
0 1 k
1
Wk = gyz Mk gyz =
0 0 1
0 0 0
k
0
.
0
1
(2.4)
As this is a non-trivial element (which is not merely a B-field shift or an element in the geometrically acting SL2 (Z) SL2 (Z)) of the T-duality group O(2, 2; Z), the resulting space is an
example of a T-fold as defined by Hull in [8].
The alternative proposal by Mathai and Rosenberg [7,12,13] claims that the T-dual is a field
C of non-commutative tori,1 A C S 1 , where the non-commutativity scale depends on
the base-coordinate x as
= kx.
(2.5)
This proposal arose from a K-theoretical point of view by showing that the H -twisted K-theory
of T 3 , KH (T 3 ), is the same as the algebraic K-theory of the algebra associated to the field of
non-commutative tori

KH T 3 = K(C).
(2.6)
It is furthermore supported by the fact that it consistently generalizes the case of geometric fluxes
and the T-duality action defined in this fashion is, thanks to Morita equivalence, of order two. In
this approach, the action of T-duality is realized in terms of taking the crossed-product algebra
[12].
We propose that both pictures are in fact valid, and are describing different aspects of the same
T-dual compactification. More precisely, we shall argue that the proposal (I) is the open-string
version of the T-fold proposal (II). Starting from the T-fold compactification (2.3), there is an
associated open-string metric G and theta-tensor introduced and studied in [2528], which
are related to the closed-string metric g and B-field B by (setting 2 = 1)
Gij = (g + B)1
(ij ) ,
ij = (g + B)1
[ij ] .
(2.7)
These are the metric and spacetime non-commutativity parameter, which the open-strings see.
For the background in (2.3) we obtain
ds 2 = dx 2 + d y 2 + d z 2 ,
= kx y z .
(2.8)
This is precisely the non-commutative torus fibration which was proposed as the T-dual spacetime in [7]. Similar backgrounds with a varying, meaning space-dependent, non-commutativity
parameter have been discussed before in [29,30].
1 The precise definition is in terms of the direct integral of non-commutative torus algebras C =
S 1 A d , with
non-commutativity parameter varying along the base S 1 .
127
How do we interpret this connection? The key point is to realize that the K-theory analysis depends on the open-string data (or open-string algebra). As advocated by Witten in [31],
the K-theory for H -flux backgrounds has a formulation in terms of the algebraic K-theory of a
(non-)commutative algebra [32], which on the other hand can be interpreted as the open-string
algebra [31,33]. This algebra is non-commutative when H = 0. Thus in order to prove the conjectured correspondence, it remains to show that the algebra C is precisely the algebra of open-string
field theory in this background.
On more general grounds one is then led to propose the following relation: consider a principal
T 2 -bundle E M with H -flux such that H2 = 0, where H2 H 2 (T 2 ) H 1 (M) (two-legged
case). Then the T-dual along the fibre-directions is given by a T-fold. The associated open-string
metric and -tensor can be computed from (2.7) and the resulting space will generically be noncommutative, with an associated non-commutative algebra, A. The conjecture is then, that A is
precisely the algebra proposed by Mathai and Rosenberg as the T-dual, i.e., it is obtained as a
crossed product algebra A = C(E, H ) R2 , where C(E, H ) is the C -algebra of the T 2 -bundle
E with H -flux and the crossed product is taken with respect to the R2 -action, which is induced
from the T 2 -action on the bundle, with the K-theory of the two algebras agreeing.
3. Probing non-geometry by generalized complex branes
In this section the same conclusion is reached as in the last section by embedding the discussion into the setup of generalized complex geometry. It is shown that the T-dual of the background
with H -flux is given by a -transformed background. Again, this is observed in the open-string
sector, and we show this by probing the T-fold geometry with generalized complex D-branes.
3.1. Generalized complex structures, B-transforms and -transforms
Let us recall a few definitions from generalized complex (GC) geometry [21]. Given an ndimensional manifold M, a generalized almost complex structure on M is defined as an almost
complex structure on the sum of tangent and cotangent bundles T M T M. For example, such
a structure can be induced by an ordinary complex structure J on M

J
0
JJ =
(3.1)
,
0 J
in which case it will sometimes be termed a diagonal GC structure, or by a symplectic form
on M

0 1
J =
(3.2)
,
0
where the matrices are written in a base adapted to the direct sum. Hybrid examples, other than
these two extreme ones, are classified by a generalized Darboux theorem [21], saying that any GC
space is locally the sum of a complex space and a symplectic space. For the existence of hybrid
GC structures with no underlying complex or symplectic structure, and their relevance for N = 1
supersymmetric compactifications in string theory see [34,35]. For the present discussion where
the (non-)geometry is probed by D0-branes, we shall restrict ourselves to GC structures of the
form JJ , thus generalizing the B-model.
Here we would like to relax the requirement that the space on which the GC structure acts
be globally of the form T M T M, and we only assume that it is made of patches that look
128
like the sum of local tangent and cotangent spaces. The definitions are therefore to be understood
in the neighborhood of some point p (which we assume to be still well-defined), that is on
Tp M Tp M.
The sum Tp M Tp M is naturally endowed with an inner product of signature (n, n),
1
X + , Y + = (X + Y ),
2
whose matrix in the same basis as above reads

0 1
G=
.
1 0
(3.3)
(3.4)
The inner product is conserved by an action of the group O(n, n) whose generic element
decomposes into a block-diagonal part (encoding an orthogonal transformation of the tangent
space and the induced orthogonal transformation of the cotangent space), and off-diagonal blocks
that can be exponentiated into B-transforms

1 0
exp(B) =
,
(3.5)
B 1
B : X + X + + X B,
(3.6)
and -transforms

1
exp() =
,
0 1
: X + X + + ,
(3.7)
(3.8)
.
where B and are antisymmetric blocks identified with a two-form B and a bivector
A B-transform acts by conjugation on generalized almost complex structures, thus mapping
the two generalized almost complex structures JJ and J to the structures

J
0
JJ (B) =
(3.9)
BJ + J t B J t
and

J (B) =
1 B
+ B1 B
1
B1
(3.10)
which we will encounter in Section 5.

3.2. D-branes as generalized complex submanifolds
Let H be a closed three-form. A generalized submanifold is defined in [21] as a submanifold
N endowed with a two-form B such that H |N = dB. The generalized tangent bundle NB of this
generalized submanifold is defined as the B-transform of the sum of the tangent bundle T N and
conormal bundle (or annihilator) Ann T N , namely:

NB = X + T N T M|N , |N = X B ,
(3.11)
so that N0 = T N Ann T N . A generalized tangent bundle is a maximally isotropic subspace
(i.e., it is isotropic with respect to G and it has the maximal possible dimension for an isotropic
space in ambient signature (n, n), namely n). Moreover, all the maximally isotropic subspaces
129
are of this form, for some submanifold N and two-form B. This is the origin of the one-toone correspondence between generalized submanifolds and pure spinors, which will be used in
Section 3.4.
Given a GC structure J , a generalized complex brane was defined in [21] to be a generalized
submanifold whose generalized tangent bundle is stable under the action of J . In the case of
J = JJ , the compatibility condition gives rise to the B-branes, as expected due to the localization
properties of the B-model on complex parameters [36]. The submanifold N namely has to be a
complex submanifold, and F has to be of type (1, 1) with respect to J
J (T N) T N,
J (X F ) + J X F = 0.
(3.12)
In the other extreme case of J = J , it yields all possible types of A-branes, including the
non-Lagrangian ones [37,38]. These are two tests of the idea that D-branes in generalized geometries are generalized submanifolds. This idea has passed further tests: calibrating forms and pure
spinors encoding stability conditions [39,40] for topological branes are correctly exchanged by
mirror symmetry [4146], and the study of morphisms between generalized tangent bundles [47]
generalizes the K-theoretic description of D-branes by taking winding numbers into account in
the resolution of vortex equations of the YangMillsHiggs model [36,48,49]. Although all the
generalized tangent bundles are n-dimensional, a generalized submanifold associated to a pdimensional submanifold N will be sometimes referred to as a generalized Dp-brane, and p will
be called the ordinary dimension of the brane.
It is important for the description of D-branes in generalized geometries to note that the projection of a subspace on the tangent space is unchanged under a B-transform. A B-transform just
switches on an Abelian field strength with magnitude B along the brane. However, a -transform
shifts the dimension of the projection of the brane on the tangent space (the ordinary dimension
of the brane) by the rank of . Let us review the linear-algebraic case where the ambient space is
V V for some vector space V . A -transform of a maximally isotropic subspace Ann F F ,
where F is a subspace of V , reads as a graph over F , in the notations of [21]

L(F, ) = X + V F, X|F = .
(3.13)
The intersection of this space and V is just the annihilator of F , because it is trivially embedded
in V F as

L(F, ) V = X + 0 V F, X|F = 0 = Ann F = L(F, 0) .
(3.14)
The vector part of any element of L(F, ) therefore decomposes into an element of Ann F and
an element of the image of : V V , and the decomposition is unique because the graph
condition X|F = implies that the intersection between Ann F and the image of is zerodimensional. Let V : V V V denote the projection onto V . We have therefore argued
that

V L(F, ) = Im L(F, ) V ,
(3.15)
and therefore

dim V L(F, ) = dim L(F, ) V + rk = dim Ann F + rk

= dim V L(F, 0) + rk .
(3.16)
130
3.3. T-duality maps B-transforms to -transforms

As we have just motivated the idea that Abelian D-branes may be identified with GC submanifolds, and since -transforms can change the ordinary dimension of such submanifolds,
it is natural to look for the connection between -transforms and monodromies on T-folds, in
the picture (II) of non-geometry. D-branes wrapped on T-folds can come back to themselves
with a different dimension after monodromy. We are going to describe how T-dualities map Btransforms to -transforms, together with the corresponding effects on D-branes.
3.3.1. Geometric three-torus with H -flux and B-transforms
Consider again the flat three-torus with uniform H -flux, with the same coordinates as above.
Consider two D2-branes wrapping fibres over two points of the base, one at x = 0 and one at
generic x. Going from the first to the second involves a B-transform by the two-form
B(x) = kx dy dz.
(3.17)
Going from x = 0 to generic x namely switches a two-form along the brane. The boundary conditions for open strings ending on a D2-brane wrapping a torus over the point x (with embedding
coordinates X, Y, Z(, ) and the obvious notation) read
Y + kx Z = 0,
Z kx Y = 0.
(3.18)
The matrix of the B-transform in a basis adapted to the coordinates (y, z) and the dual coordinates (y,
z ) reads
1
0
0 0
1
0 0
0
g=
(3.19)
.
0 kx 1 0
kx
0
0 1
3.3.2. Geometric T-dual with a connection
It is instructive to perform first the T-duality along the y direction. The D2-branes wrapping
the two fibres in question become D1-branes, and parametrizing the base by an angle with
kx = tan , we observe that the D1-branes are rotated with respect to each other within the fibre.
This reflects the fact that they now live on a torus with a connection
( sin Z + cos Y ) = 0,
(cos Z + sin Y ) = 0.
In the same basis as before, T-duality is encoded by the matrix
0 0 1 0
0 1 0 0
gy =
,
1 0 0 0
0 0 0 1
and the B-transform is therefore replaced by one with matrix
1 kx
0
0
0
0
0 1
g = gy1 ggy =
.
0 0
1
0
0 0 kx 1
(3.20)
(3.21)
(3.22)
131
Let us describe these D1-branes in terms of maximally isotropic2 subspaces. Start at x = 0

with a D1-brane wrapping the y circle. The corresponding pure spinor is the sum of the tangent
and conormal bundles of the y circle, with coordinates

S 1 Ann S 1 = y, z = 0, 1 = 0, 2 .
(3.23)
Acting on it with g yields the coordinates (y, kxy, kx 2 , 2 ), which means that there are
Dirichlet conditions along the one-dimensional subspace of the two-torus at x = l with equation:
tan Y Z = 0.
(3.24)
This is consistent with the fact that there is now a connection on the torus, and taking x to be
equal to 1 (the period of the coordinate along the base) and requiring the D1-brane to come back
to itself does indeed give rise to the identification of the twisted torus
(x, y, z) (x + 1, y, z + ky),
(3.25)
as it should [3].
3.3.3. Non-geometric T-dual space and -transforms
Let us perform one more T-duality, along the z direction, and get to the non-geometric space.
The matrix acting on the T 2 -fibre, in going from x = 0 to generic x, in a basis adapted to the real
coordinates (y, z, y,
z ) is obtained from g through conjugation by the T-duality matrix
0 0 1 0
0 0 0 1
gyz =
.
1 0 0 0
0 1 0 0
It therefore reads
1 0 0
0
1 kx
1
g = gyz
ggyz =
0 0 1
0 0 0
kx
0
,
0
1
(3.26)
which we recognize as a -transform by the bivector field

(x) = kxy z .
(3.27)
Since -transforms affect the vector part of maximally isotropic subspaces, there is no way of
twisting the torus to bring the D0-brane back to itself after a monodromy around the base. Moreover, -transforms are also associated to open paths on the base, showing that attaching an open
string to two D0-branes sitting over different points of the base is impossible, unless T-dualities
are allowed to patch the coordinate charts together. As open strings can wind around the base
before attaching themselves to the second brane, they are sensitive to the global effect of nongeometry, even if the two points on the base can be put in one single coordinate patch for the
2 Isotropic is understood with respect to the inner product on the sum of the two-torus and the dual two-torus; we do
not specify the embedding into T 6 yet; the coordinate on base only plays the role of a parameter as it is not acted on by
the T-dualities we consider.
132
purposes of local differential geometry. It is crucial for such a global effect that the base be
non-simply-connected.
To sum up, T-dualities therefore relate D-branes located in different fibres. Hence they are
needed as changes of charts, as predicted by the proposal (II). Moreover, the transformations of
the corresponding pure spinors are dictated by a bivector field (x) = kxy z depending on
the coordinate along the base in the same way as the tensor of the proposal (I).
3.4. Generalized D0-branes on the non-geometric T-dual
As points might be disturbed by global effects in non-geometric spaces, we would like to
probe non-geometry by generalized D0-branes. Of course, in order to be able to use techniques
from generalized geometry for describing T-duals of the three-torus with H -flux, we first have to
embed the three-torus into a six-torus.
Let us consider a generalized B-model, and pick a complex structure of the form JJ , with J
an ordinary complex structure on the six-torus. We still have a choice for the complex structure J :
we can either consider the T 2 -fibre as an elliptic curve in this complex structure (which would
make B a tensor of type (1, 1) and a valid field strength for a D2-brane of type B wrapping the
elliptic curve), or pick a complex structure in which y and z are components of different complex
coordinates. This way B would have a non-zero component of type (0, 2) and the dual torus with
coordinates y and z could not support a D2-brane of the B-model. Let us choose the second
option in order to single out the role of the (0, 2) components and their possible influence on
non-commutativity.
The way we embed the three-torus into a six-torus is therefore the following: the T 2 -fibre
coordinates y and z are real parts of complex coordinates y + iy and z + iz , where y and z are
coordinates along additional circles, and the base is combined with a third additional circle with
coordinate x into an elliptic curve. In the sequel we shall denote the local complex coordinates
we have just described by
z1 = x + ix ,
z2 = y + iy ,
z3 = z + iz .
(3.28)
This way B is not of type (1, 1) and will therefore contribute non-commutative deformations
as argued in [22]. Moreover, the x-dependence means that Morita equivalence cannot be used
to gauge non-commutativity away, since the B-field will assume non-rational values. But for
the time being, we are interested in the effect of the (0, 2) and (2, 0) components of the Bfield in terms of T-duality transformations, as an illustration of (II). The connection with noncommutativity using the language of (I) and (III) will be made in Sections 4 and 5.
A few comments about the choice of GC structure are in order: we restrict ourselves to diagonal GC structures, thus generalizing the B-model. We shall see in Section 4 that deformations of
the generalized B-model are indeed sufficient to explain the connection between non-geometry
and non-commutativity, but the reason why it is a priori sufficient to consider a diagonal GC
structure is that only such structures allow generalized D0-branes, which are the point-like objects with which one would like to probe non-geometry. Of course it is well known that D-branes
corresponding to GC structures of the form J do not include D0-branes, moreover the generalized Darboux theorem implies that hybrid GC structures locally have some A-type boundary
conditions that forbid D0-branes.
Let us consider generalized D0-branes for the GC structure we have just described, and the
way they transform under monodromy. They are not affected by B-transforms because the graph
133
condition of definition (3.11) is empty. On the other hand, their ordinary dimension is raised
upon a -transform by an amount equal to the rank of , as was explained above.
In order to work out the effect of the monodromy on D0-branes, we are going to use the description of generalized tangent bundles by pure spinors. The mapping between isotropic spaces
and spinors is made manifest by the action of sections of T M T M on M, which carries a
representation of Clifford(n, n):
(X + ) . = X + ,
(3.29)

(X + ) . (X + ) . =
X + , X + .
(3.30)
V .
Maximally isotropic subspaces

Given a spinor, one can associate to it its null space in V
are therefore in one-to-one correspondence with pure spinors.
In the case of a generalized D0-brane, the pure spinor to be considered is the holomorphic
three-form
:= dz1 dz2 dz3 ,
(3.31)
z1 , z2 , z3
in a local patch where

are complex coordinates associated to the complex structure we
have described on the six-torus. The annihilator is locally of the form

(0,1)
(1,0)
TM
(3.32)
T M
= Vect
,
,
Vect dz1 , dz2 , dz3 .
z 1 z 3 z 3
Let us write the components of in a way adapted to the local complex coordinates, so that
is the (2, 0) part of , and and do not appear in the -transform because they
act on components of the annihilator that are zero (and stay so, because the one-form part is not
transformed by ). The transformation rules are therefore

+ dz + + dz ,
(3.33)
where

=
+ ,

= .
The vector space spanned by the vectors ( )

the projection of the annihilator of on T M (1,0)
(3.34)
T M (0,1) ,
is still the whole subspace

whereas
is made two-dimensional by the monodromy,
since rk = rk(dz2 dz3 ) = 2.

As expected, the dimension of the projection on the tangent space is shifted by the rank of
the (0, 2) part of the bivector field . This establishes that there is no zero-dimensional global
section of the vector part. This phenomenon was observed in the context of non-commutative
deformations in [22], where it was called the uncertainty principle for topological D-branes
(commutators of equations of complex submanifolds cannot vanish, and this prevents D-branes
from wrapping maximal-codimension submanifolds). In the present case, the change of type3 of
a pure spinor under monodromy is equivalent to the lack of a global splitting between momenta
and winding numbers. The space obtained by T-duality from the generalized complex T 6 with
H -flux can therefore not be globally of the form L L , with L a maximally isotropic subspace.
This is the absence of global polarization that appeared in the real case for T-folds, and it is
encoded by the (0, 2) part of the bivector field .
3 A pure spinor can be written in a unique way as 1 n eF , where 1 , . . . , n are complex one-forms and F
is a complex two-form; the integer n is called the type of the pure spinor.
134
4. Non-commutativity from Lagrangian deformations in the BV procedure

In the previous section our consideration of the T-dual of a complex six-torus with H -flux
has shown the necessity of T-folds for the description of the global topology, when no global
polarization exists, which reproduces the criterion of [8] for the objects (II). What about proposal
(I) and non-commutativity along the dual T 2 -fibres? The connection comes from the construction
of topological string theory on GC spaces in [23], where it was explained that a tensor of type
(0, 2) can deform the B-model with generalized complex target space, inducing star-products
on the fibre. In order to make contact with [12] we are going to show how the -transform
induces this very deformation of the generalized B-model on the T-dual of the GC six-torus. For
a categorical viewpoint on the FourierMukai equivalence between deformations of complex tori
(either in a non-commutative direction parametrized by a holomorphic Poisson structure or in a
B-field direction), see [50].
4.1. -transforms and the generalized B-model
The BatalinVilkovisky (BV) formalism requires a nilpotent operator Q and an odd differential operator of second order . They act on the graded space of fields and induce an odd
symplectic structure, for which the master action S is a Hamiltonian function. The odd Laplacian induces an antibracket via the formula
(F, G) = (1)|F | (F G) F G (1)|F | F G.
(4.1)
The condition Q2 = 0 then induces the master equation

(S, S) = 0.
(4.2)
From now on, as is required by the BV procedure, we shall give fermionic statistics to the
vector and form coordinates, or in other words reverse the parity on the fibres of the tangent and
cotangent bundles. When computing the action of a sigma model, one has to pull back vector and
form fields on the world-sheet , which induces a change of statistics on the tangent bundle of
the world-sheet, which is now denoted . As far as the B-model is concerned, the graded space
of fields is (in a local coordinate patch) the space of observables of the B-model. The operators
Q and are the antiholomorphic and holomorphic differentials,
Q = = d
= = dz

=
,

(4.3)
where in re-expressing as a second-order differential operator, use has been made of the observation that one-forms may act on the de Rham complex as derivatives with respect to vector
coordinates [51]. This way the coordinates and z are canonically conjugate to each other, and
one has to add antifields to be paired with z and (because is degenerate on the subspace
and . As shown in [23], the master action for the generalized B-model
they span), called z
then reads

S=
(4.4)
dz + z
.
The allowed deformations of the generalized B-model involve holomorphic bivector fields.
A Lagrangian submanifold L of the space of fields has indeed to be chosen to compute the
gauge-fixed partition-function

ZL := DXeS[X] ,
135
(4.5)
and the invariance of this path integral under change of the gauge-fixing condition is equivalent to
its invariance under the Lagrangian deformations of L. A variation in the gauge-fixing condition
amounts to a Lagrangian deformation of the Lagrangian submanifold L, namely one where the
momenta are derived from a density
.
(4.6)
x i
Starting with a Lagrangian submanifold with equation given by the vanishing of all momenta
p i = 0, invariance of ZL under Lagrangian deformations is expressed by the following chain of
equalities

S(xi )
S(xi )
L Z = p i i eS(xi ) =
(4.7)
e
= 0,
e
=
i
i
p
x p
pi x i
p i =
which implies the quantum master equation

eS[X] = 0.
(4.8)
Expanding in powers of a deformation of the master action gives rise to the MaurerCartan
equation. In the case of the generalized B-model, splitting into tensors of different types shows
[23] that the deformation of the generalized B-model by a holomorphic bivector field is allowed
(moreover, the sum of tangent and cotangent spaces is one of the geometries recently addressed
by Ikeda in the deformation theory of BV structures [52]).
In the present context, the lack of global polarization induces deformations of the BV structure
when going from one patch of coordinates to another. It is instructive to see how derivatives
are affected by the monodromies described in (3.33). Let us work out the deformation of the
antibracket adapted to the isotropic subspace T M (0,1) T M (1,0) we started with in the previous
section

GF
G.
(F, G) = F
(4.9)

The change of coordinates induced by a -transform
= + ,
= + ,
(4.10)
induces the following changes in derivatives on the cotangent space

=
+
= ,

=
+
= + ,

(4.11)
and the antibracket now includes pairs of derivatives with respect to the vector coordinates, so
that the monodromy shifts the antibracket by a Poisson bracket:
(F, G) = (F, G) + 2F
G.

(4.12)
136
We have therefore shown that the data of the BV structure do change from patch to patch in the
non-geometric T-dual of the GC six-torus. We are going to work directly on the master action,
since the -transform is of the Lagrangian type, so that the T-duality that has been seen to bind
together the coordinate patches, is also deforming the master action.
4.2. From the generalized B-model to the Poisson sigma model
We have seen in the previous section that in a special case an obstruction to the existence of a
global generalized complex form L L for the T-duals is encoded by a holomorphic bivector
field . The link with [7] is provided by the choice of an isotropic submanifold involved in
the BV gauge-fixing procedure. A global such choice is impossible as soon as the -transform
is non-trivial, and this leads to a deformation of the generalized B-model by the holomorphic
bivector field . The resulting model is precisely the Poisson sigma model that appears as the
-deformation of the topological J -model constructed by Pestun [23]. Star-products emerge
from the Poisson structure by deformation quantization [53,54]. Of course this is no accident.
Relevance of the Kontsevich formula in non-commutative gauge theory along D-branes appeared
for example in [27,55,56].
Consider the master action that is obtained from the BV procedure for the B-model with a
generalized complex manifold as a target space [23], i.e., a target space endowed with a GC
structure of the diagonal form JJ . We therefore start with the master action on the patch with
complex coordinates (z , z )

dz + z
S=
(4.13)
.
As for the , they span the bundle T M (1,0) that is not modified by the monodromy, and the
are conjugate to the antiholomorphic base coordinates that are untouched by the monodromy.
z
The second term is decoupled in the initial patch, and will stay so under monodromy.
But the first term, as it is endowed with holomorphic indices, is affected by the monodromy.
This is due to the fact that the circle base cannot be covered by a single patch. Let us choose a
construction of the spinor bundle where the differential forms act by differentiation with respect
to the dual coordinates. This corresponds to choosing the pure spinor as the vacuum, and
vector fields as creation operators, as explained by Witten in [51]. In a local patch we therefore
identify dz with / , so that the relevant term in the master action transforms as follows

=
+ ,
(4.14)
and the result, in a representation where differential forms act by multiplication, as

S =
dz + ,
(4.15)
which of the form S + S, with S induced by the holomorphic bivector . Of course we

can rewrite the expression for S in terms of coordinates, vectors and forms with conventional
statistics, both on the world-sheet and the target space, by taking multiplications to be wedge
products.
137
Since the non-zero components of the bivector field are along the directions y and z, and
they depend only of the coordinate x, the parameter is a Poisson bivector field
+ + = 0.
(4.16)
The resulting model with action S is the Poisson sigma model studied by Cattaneo and Felder
in [24]. We are therefore left with T-dual fibres forming a field of non-commutative tori over a
circle. A subspace of the T-dual T 6 is therefore non-commutative, with the non-commutativity
scale predicted by (I).
For a given topology of the world-sheet, we find a deformation of the product of observables
into a star-product as in [24]:

f (a1 ) g(a) =
(4.17)
DX f X(0) g X(1) ei(S+ S)[X] ,
X =a
where 0, 1, are the coordinates of the points of insertion of observables on the boundary of the
world-sheet, and a1 is the component of the coordinates of the point a along the direction x. The
continuous dependence on a1 comes from the definition (3.27) for the bivector . The fact that
non-commutativity shows up in the boundary correlators is the sign that the open-string sector is
crucial for the equivalence between (I) and (II).
5. SU(3) SU(3) structure and non-commutativity
It is argued in [19] that the mirror of a CalabiYau compactification with magnetic H -flux [2]
possesses an SU(3) SU(3) structure. SU(3) SU(3) structure compactifications are described
in terms of pure spinors, made from bilinears of SU(3)-invariant spinors of Cliff(6). In case
those invariant spinors are not parallel and the type of the associated pure spinors is not globally
defined, the resulting compactification is conjectured to be non-geometric. In case the two pure
spinors still have a globally constant type, the compactification has global SU(2) structure and
is still geometric, an example of which is T 2 K3. The proposal in [19] should in particular be
consistent with the non-commutative T-dual conjecture, when both setups are applicable.
In this section we show that precisely in the case when the two SU(3)-invariant spinors are
not parallel, there is a non-trivial Poisson bivector, which yields a non-commutative deformation
of the open-string background. If the pure spinors have a uniform expression, the Poisson bivector is constant and thus one can dispose of the non-commutativity by Morita equivalence. This
of course is not possible in the case when the Poisson bivector is dependent on the remaining
coordinates.
5.1. SU(3) SU(3) structure manifolds
Consider type II compactifications on six-manifolds with SU (3) SU (3) structure [5,20,
21,5759] (for a detailed introduction and references see [60]). As such they are characterized
by a pair of no-where vanishing SU(3)-invariant spinors 1,2 , which arise in the decomposition
of the two SO(9, 1) spinors 1,2 of type II under SO(3, 1) SO(6). If 1 = 2 the structure
group is reduced to SU(3), which in particular incorporates the case of standard CalabiYau
compactifications. If on the other hand the spinors are not parallel to each other throughout the
manifold, one speaks of an SU(2) structure. The latter is characterized by a non-vanishing vector
138
field. Defining
2
1
1
= c+
+ (v + iw)
,
+
c C,
(5.1)
the vector field in question is
1
2
m := +
m
= vm iwm .
(5.2)
The spinors 1,2 can also be combined to construct a pair of SU(3, 3) bi-spinors
1
2

,
= +
(5.3)
which are pure (i.e., they are annihilated by half of the -matrices). Moreover, via the standard
Clifford map, they are in one-to-one correspondence with (formal sums of) differential forms.
On an SU(3) structure manifold the pure spinors correspond to the (3, 0) form and the
(exponential of the) (1, 1) form J . Generically however, there are two independent two- and
three-forms
J = j v w,
= (v iw).
(5.4)
Here j and parametrize the local SU(2) structure in the transverse directions to v and w. For
the present purposes it is instructive to note that in the case of non-geometric spaces the notion
of transversality holds only locally, since it can be spoiled by a -transform.
The associated two pure spinors are

1 ij
1
(5.5)
i eivw ,
= eij + ic (v + iw).
ce
8
8
Raising an index on the two-forms J we obtain two complex structures, I , which on the other
hand define a generalized complex structure I as will be discussed in the next section. Thus, one
important point to notice is that SU(3) SU(3) structure implies generically that there are two
independent complex structures, which arise from the two SU(3)-invariant spinors 1,2 .
+ =
5.2. Non-commutative deformations

To begin with, let us review some known facts about non-commutativity: In [22] Kapustin
presents a criterion when a compactification allows for non-commutative deformations. Consider
first the case of closed B. Then for a CalabiYau manifold with metric G and B-field and two
(not necessarily equal) complex structures I compatible with the Levi-Civita connection, one
may define the generalized complex structure

1
P
2 (I+ + I ) + P B
I=
(5.6)
,
J + BP B + 12 B(I+ + I ) + 12 (I+ + I )t B 12 (I+ + I )t BP
where the bivector part is defined as

1 1
J J1 ,
2 +
with J = GI . Furthermore J = 1/2(J+ J ). The complex structure is
P =
(5.7)
1
I = (I+ + I ) + P B,
2
(5.8)
139
whereas the Poisson bivector, which parametrizes the non-commutative deformations, is given
by
1
= I P .
(5.9)
2
In particular, the non-commutativity is non-trivial only if the two complex structures are unequal I+ = I , since otherwise P = 0. In [40] this was generalized to the case of H = dB = 0.
The only difference from the above equations is that the complex structures I now have to be
covariantly constant with respect to the connection with torsion Hijk . Note however that as in
the T-fold case, non-commutativity arises only at the level of D-branes, i.e., the closed-string
background remains commutative, albeit not necessarily geometric, as the left- and right-moving
modes on the world-sheet differ. In particular, a generic SU(3) SU(3) structure compactification yields a pair of distinct complex structures, and thus two distinct realizations of the N = 2
super-conformal algebra for left- and right-movers, respectively. In this sense, the world-sheet
theory is very much alike the situation for asymmetric orbifolds.
In particular, we can then determine P in the case of SU(3) SU(3) structure compactifications

1
P = (GI+ )1 (GI )1 ,
(5.10)
2
which has again non-vanishing Poisson bivector = 1/2I P if the two complex structures
differ.
In case of the two spinors being never parallel, which corresponds to v + iw = 0, i.e., we
have an SU(2) structure at least locally, the corresponding is non-zero. So this is indeed the
case, when there are non-trivial non-commutative deformations. In fact we can write the Poisson
bivector entirely in terms of the one-form that characterizes the SU(2) structure
1
,
(5.11)
2i
which means that the non-commutativity is governed only by the vectors in (5.2).
If the SU(2) structure is global, such as for K3-compactifications, the resulting noncommutative deformations are constant and thus of minor interest to the present discussion. The
interesting cases arise, when the above description is only local. Then the Poisson bivector is
not constant and one cannot get rid of it by Morita equivalence. Thus, the lack of global definition for the type of the pure spinors (because of the change of dimension between two different
base-points that was illustrated above) makes it unlikely that the linear independence between
two SU(3) spinors can be described by a globally-defined vector field. We may obtain a Poisson
bivector P , but it need not be of the form v w. We shall present an explicit example in the
next section.
P = w v =
5.3. Torus with H -flux and mirror symmetry

An illustrative example is T 6 in the complex coordinates (3.28) with H -flux H H 3 (T 6 , Z),
which allows to use the language of SU(3) SU(3) structures. We consider a triple T-duality
along x , y, z, which in this context corresponds to
considering the mirror. The three-form is
= dz1 dz2 dz3 and the (1, 1)-form is J = i dzi d z i . In particular, the T 3 that the
T-duality acts upon is a special Lagrangian cycle. Note that this is different from the T-duality
transformations encountered in the previous sections for T 3 with H -flux, however, as in that
140
case, we consider only two-legged H -flux, i.e., T-duality acts in only two directions supporting
the H -flux.
The setup for the square torus is simple enough, and the generalized complex structure I is
diagonal. However switching on the H -flux yields

I
0
I(B) =
(5.12)
.
BI + I t B I t
The T-dual complex structure was determined in [22] to be
t

I BI + I t B
I (B) =
,
0
I
(5.13)
so that the Poisson bivector is read off to be P = BI + I t B, which is not necessarily vanishing,
as expected. Moreover, if we insist that the -transformed D0-brane be a generalized D0-brane
with respect to I after monodromy, we obtain the constraint = (0,2) , in terms of the decomposition with respect to I . The deformation parameter is once more seen to be a (0, 2) tensor.
Let us connect this non-block-diagonal GC structure to the language of maximally isotropic
subspaces we used to probe the non-geometry by D0-branes. Consider again a graph of some
bivector field over some subspace F of the cotangent space

L(F, ) = X + V F, X|F = ,
(5.14)
and require stability of this graph under the action of I . We are led to the following equation,
that must hold for every element in F

I t + (P ) = I .
(5.15)
Eliminating the coordinates we observe that the (0, 2) part of must equal the Poisson
bivector field
I + I t = P .
(5.16)
We therefore see that the non-diagonal block of the GC structure in the T-dual picture is precisely
the parameter of the -transform that is undergone by any D0-brane. Whenever P is non-zero,
the dimension of the projection of a D-brane onto the tangent space is non-zero. Therefore P
induces non-geometry in the sense that point-like D-branes cannot be put on a GC space with
non-zero bivector block.
As stated in theorem (5.4) of [61] for mirrors of complex tori, the two generalized complex
structures JJ (B) and J (B) shown in formulae (3.9) and (3.10) are exchanged by mirror symmetry. Indeed, if gx yz is the element of O(6, 6) encoding T-duality in the x , y, z directions, the
mirror exchange

JJ (B) = gx yz I (B)gx1
(5.17)
yz
holds.
As we already argued, the image of JJ by T-duality is not block-diagonal anymore, and we
may read off the Poisson bivector P as
1
P = kx(y z + y z ),
(5.18)
2
which is not decomposable as the tensor product of two vectors. Hence this is a case of generalized type of T-fold. In this example of a non-geometric T-dual, the non-commutativity scale
141
depends on the base coordinate x in a way that prevents to gauge it away by Morita equivalence,
and furthermore it does not come from a globally defined vector field that encodes the linear
independence between two SU(3)-invariant spinors. The lack of a global polarization in T-folds
(II) can therefore be traced in the formalism of (III) as the non-decomposability of the Poisson
bivector field P .
So far our consideration of the T-dual of a complex torus with H -flux has shown the necessity
of T-folds for the description of the global topology, when no global choice of type for a pure
spinor exists, which reproduces the criterion of [8]. What about non-commutativity along the
dual T 2 -fibres? The connection comes from the construction of topological string theory on GC
spaces and sigma models with bistructures in [23,6264]. It was explained that a tensor of type
(0, 2) can deform the B-model with generalized complex target space, yielding star-products
induced by the Poisson bivector . In order to make contact with [12] it would be interesting to
see how the -transform induces this very same deformation of the generalized B-model.
The results in this section should also be derivable from the action of T-duality on spinors
as advocated by Hassan [6567]. In the geometric case, in particular for the LuninMaldacena
background [68], the analysis was performed in [69] and it is easily observed from their results
that P in this case is of the form v w, and non-commutativity does not occur between coordinates but as relative phases between ordered products of fields, corresponding to global U (1)
symmetries.
6. Conclusion
T-duality in the presence of NSNS-fluxes provides the first stepping stone to understanding
generalized versions of mirror symmetry la StromingerYauZaslow (SYZ) [70]. The present
paper discusses this issue, thereby merging various existing proposals for the T-dual. If the NSNS
H -flux is supported only on one T-dualized direction the dual is again geometric and consensus
has been reached on the T-dual geometry throughout the literature. Controversy starts when two
T-dualized directions are spanned by the H -flux. Our present investigations concern the case
of two directions, applied to tori and torus fibrations. We have shown that the proposal (I) of
Mathai and Rosenberg, claiming the dual to be a field of non-commutative tori, can be viewed
as the open-string version of Hulls T-fold proposal (II). Secondly, we have shown that generalized geometries provide an alternative setup for studying the T-dual or mirror, and can be
reconciled with the non-commutativity proposal by explicit construction of a Poisson bivector,
which depends crucially on the background H -flux. As argued in [22,71], this bivector parametrizes non-commutative deformations of the open strings. We found that in the case of two-legged
H -flux, the bivector field is in fact not uniform, but varies along the base of the torus-fibration.
On more general grounds it would be interesting to understand the precise conditions for the
dual space to be non-geometric. The key ingredient for the deformation by the bivector field is
of course the multiple-connectedness of the base of the fibration. The present discussion could
be extended to complicated fibrations over a multiply-connected base with H -flux. On the other
hand, T-dualizing along a two-torus carrying a non-zero B-field and fibered over a contractible
space, as in the sequence of -deformations in [68], can lead to a geometric T-dual (orbifolds
with torsion in that instance). This is consistent with the fact that any loop on the base can be
shrunk and included in a local coordinate patch (thus removing the -transformed term in the
formula (4.15) in the simply-connected cases), and also with the fact that non-commutativity of
the dual two-torus in the proposal of [12] is measured by classes in the first integral cohomology
142
group of the base (thus setting the star-product to the ordinary product in the simply-connected
cases).
Clearly it would be very interesting to extend the present discussions to more general setups,
in particular, a generalized SYZ construction would be the most natural next problem to be
addressed. The argument of SYZ for the existence of T 3 -fibrations of CalabiYau manifolds
rested on the fact that a D0-brane had only its position as a modulus. In the case of T-folds, the
modulus is a modulus of Lagrangian deformations and prevents D0-branes from existing, just
as non-commutativity does.
Acknowledgements
We thank Mariana Graa, Simeon Hellerman, Jan Louis, Varghese Mathai, Ruben Minasian,
Bastiaan Spanjaard and Dan Waldram for discussions and comments on various occasions.
P.G. thanks the Institute for Advanced Study for hospitality, as well as the organizers of the
38th International Symposium Ahrenshoop for a stimulating conference, and is funded by the
GermanIsraeli Foundation for Scientific Research and Development. S.S.N. thanks the Erwin
Schrdinger Institute, Wien, for hospitality and is funded by a Caltech John A. McCone Postdoctoral Fellowship in Theoretical Physics. This work was supported in part by the DFG and the
European RTN Program MRTN-CT-2004-503369.
References
[1] E. lvarez, L. lvarez-Gaum, J.L.F. Barbn, Y. Lozano, Some global aspects of duality in string theory, Nucl.
Phys. B 415 (1994) 71100, hep-th/930903.
[2] S. Gurrieri, J. Louis, A. Micu, D. Waldram, Mirror symmetry in generalized CalabiYau compactifications, Nucl.
Phys. B 654 (2003) 61113, hep-th/0211102.
[3] S. Kachru, M.B. Schulz, P.K. Tripathy, S.P. Trivedi, New supersymmetric string compactifications, JHEP 0303
(2003) 061, hep-th/0211182.
[4] S. Hellerman, J. McGreevy, B. Williams, Geometric constructions of nongeometric string theories, JHEP 0401
(2004) 024, hep-th/0208174.
[5] S. Fidanza, R. Minasian, A. Tomasiello, Mirror symmetric SU(3)-structure manifolds with NS fluxes, Commun.
Math. Phys. 254 (2005) 401423, hep-th/0311122.
[6] P. Bouwknegt, J. Evslin, V. Mathai, T-duality: Topology change from H -flux, Commun. Math. Phys. 249 (2004)
383415, hep-th/0306062.
[7] V. Mathai, J.M. Rosenberg, On mysteriously missing T-duals, H -flux and the T-duality group, hep-th/0409073.
[8] C.M. Hull, A geometry for non-geometric string backgrounds, JHEP 0510 (2005) 065, hep-th/0406102.
[9] P. Bouwknegt, K. Hannabuss, V. Mathai, Nonassociative tori and applications to T-duality, Commun. Math.
Phys. 264 (2006) 4169, hep-th/0412092.
[10] J. Shelton, W. Taylor, B. Wecht, Nongeometric flux compactifications, JHEP 0510 (2005) 085, hep-th/0508133.
[11] I. Ellwood, A. Hashimoto, Effective descriptions of branes on non-geometric tori, hep-th/0607135.
[12] V. Mathai, J.M. Rosenberg, T-duality for torus bundles via noncommutative topology, Commun. Math. Phys. 253
(2004) 705721, hep-th/0401168.
[13] V. Mathai, J. Rosenberg, T-duality for torus bundles with H -fluxes via noncommutative topology. II: The highdimensional case and the T-duality group, Adv. Theor. Math. Phys. 10 (2006) 123158, hep-th/0508084.
[14] A. Dabholkar, C. Hull, Generalised T-duality and non-geometric backgrounds, JHEP 0605 (2006) 009, hepth/0512005.
[15] C.M. Hull, Doubled geometry and T-folds, hep-th/0605149.
[16] C.M. Hull, Global aspects of T-duality, gauged sigma models and T-folds, hep-th/0604178.
[17] A. Lawrence, M.B. Schulz, B. Wecht, D-branes in nongeometric backgrounds, hep-th/0602025.
[18] S. Hellerman, J. Walcher, Worldsheet CFTs for flat monodrofolds, hep-th/0604191.
[19] M. Graa, J. Louis, D. Waldram, in preparation.
[20] N. Hitchin, Generalized CalabiYau manifolds, Quart. J. Math. Oxford Ser. 54 (2003) 281308, math.DG/0209099.
143
[21] M. Gualtieri, Generalized complex geometry, math.DG/0401221.

[22] A. Kapustin, Topological strings on noncommutative manifolds, Int. J. Geom. Methods Mod. Phys. 1 (2004) 4981,
hep-th/0310057.
[23] V. Pestun, Topological strings in generalized complex space, hep-th/0603145.
[24] A.S. Cattaneo, G. Felder, A path integral approach to the Kontsevich quantization formula, Commun. Math.
Phys. 212 (2000) 591611, math.QA/9902090.
[25] M.R. Douglas, C.M. Hull, D-branes and the noncommutative torus, JHEP 9802 (1998) 008, hep-th/9711165.
[26] C.S. Chu, P.M. Ho, Noncommutative open string and D-brane, Nucl. Phys. B 550 (1999) 151, hep-th/9812219.
[27] V. Schomerus, D-branes and deformation quantization, JHEP 9906 (1999) 030, hep-th/9903205.
[28] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032, hep-th/9908142.
[29] M. Anazawa, D0-branes in a H -field background and noncommutative geometry, Nucl. Phys. B 569 (2000) 680
692, hep-th/9905055.
[30] D.A. Lowe, H. Nastase, S. Ramgoolam, Massive IIA string theory and matrix theory compactification, Nucl. Phys.
B 667 (2003) 5589, hep-th/0303173.
[31] E. Witten, Overview of K-theory applied to strings, Int. J. Mod. Phys. A 16 (2001) 693706, hep-th/0007175.
[32] P. Bouwknegt, V. Mathai, D-branes, B-fields and twisted K-theory, JHEP 0003 (2000) 007, hep-th/0002023.
[33] E. Witten, Noncommutative geometry and string field theory, Nucl. Phys. B 268 (1986) 253.
[34] G. Cavalcanti, M. Gualtieri, Generalized complex structures in nilmanifolds, J. Symplectic Geom. 2 (2004) 393
410, math.DG/0404451.
[35] M. Graa, R. Minasian, M. Petrini, A. Tomasiello, A scan for new N = 1 vacua on twisted tori, hep-th/0609124.
[36] H. Ooguri, Y. Oz, Z. Yin, D-branes on CalabiYau spaces and their mirrors, Nucl. Phys. B 477 (1996) 407430,
hep-th/9606112.
[37] A. Kapustin, D. Orlov, Remarks on A-branes, mirror symmetry, and the Fukaya category, J. Geom. Phys. 48 (2003),
hep-th/0109098.
[38] S. Chiantese, Isotropic A-branes and the stability condition, JHEP 0502 (2005) 003, hep-th/0412181.
[39] M. Mario, R. Minasian, G.W. Moore, A. Strominger, Nonlinear instantons from supersymmetric p-branes,
JHEP 0001 (2000) 005, hep-th/9911206.
[40] A. Kapustin, Y. Li, Topological sigma-models with H -flux and twisted generalized complex manifolds, hepth/0407249.
[41] P. Grange, R. Minasian, Modified pure spinors and mirror symmetry, Nucl. Phys. B 732 (2006) 366378, hepth/0412086.
[42] P. Koerber, Stable D-branes, calibrations and generalized CalabiYau geometry, JHEP 0508 (2005) 099, hepth/0506154.
[43] O. Ben-Bassat, M. Boyarchenko, Submanifolds of generalized complex manifolds, J. Symplectic Geom. 2 (3)
(2004) 309355.
[44] O. Ben-Bassat, Mirror symmetry and generalized complex manifolds. I. The transform on vector bundles, spinors,
and branes, J. Geom. Phys. 56 (4) (2006) 533558.
[45] L. Martucci, P. Smyth, Supersymmetric D-branes and calibrations on general N = 1 backgrounds, JHEP 0511
(2005) 048, hep-th/0507099.
[46] L. Martucci, D-branes on general N = 1 backgrounds: Superpotentials and D-terms, JHEP 0606 (2006) 033, hepth/0602129.
[47] P. Grange, R. Minasian, Tachyon condensation and D-branes in generalized geometries, Nucl. Phys. B 741 (2006)
199214, hep-th/0512185.
[48] R. Minasian, G.W. Moore, K-theory and RamondRamond charge, JHEP 9711 (1997) 002, hep-th/9710230.
[49] E. Witten, D-branes and K-theory, JHEP 9812 (1998) 019, hep-th/9810188.
[50] O. Ben-Bassat, J. Block, T. Pantev, Non-commutative tori and FourierMukai duality, math.AG/0509161.
[51] E. Witten, A note on the antibracket formalism, Mod. Phys. Lett. A 5 (1990) 487.
[52] N. Ikeda, Deformation of BatalinVilkovisky structures, math.SG/0604157.
[53] P. Schaller, T. Strobl, Poisson structure induced (topological) field theories, Mod. Phys. Lett. A 9 (1994) 31293136,
hep-th/9405110.
[54] M. Kontsevich, Deformation quantization of Poisson manifolds. I, Lett. Math. Phys. 66 (2003) 157216, hepth/9709040.
[55] L. Cornalba, D-brane physics and noncommutative YangMills theory, Adv. Theor. Math. Phys. 4 (2000) 271281,
hep-th/9909081.
[56] B. Jurco, P. Schupp, Noncommutative YangMills from equivalence of star products, Eur. Phys. J. C 14 (2000)
367370, hep-th/0001032.
144
[57] M. Graa, R. Minasian, M. Petrini, A. Tomasiello, Generalized structures of N = 1 vacua, JHEP 0511 (2005) 020,
hep-th/0505212.
[58] M. Graa, J. Louis, D. Waldram, Hitchin functionals in N = 2 supergravity, JHEP 0601 (2006) 008, hep-th/0505264.
[59] I. Benmachiche, T.W. Grimm, Generalized N = 1 orientifold compactifications and the Hitchin functionals, Nucl.
Phys. B 748 (2006) 200252, hep-th/0602241.
[60] M. Graa, Flux compactifications in string theory: A comprehensive review, Phys. Rep. 423 (2006) 91158, hepth/0509003.
[61] A. Kapustin, D. Orlov, Vertex algebras, mirror symmetry, and D-branes: The case of complex tori, Commun. Math.
Phys. 233 (2003) 79136, hep-th/0010293.
[62] R. Zucchini, A sigma model field theoretic realization of Hitchins generalized complex geometry, JHEP 0411
(2004) 045, hep-th/0409181.
[63] R. Zucchini, Generalized complex geometry, generalized branes and the Hitchin sigma model, JHEP 0503 (2005)
022, hep-th/0501062.
[64] P.S. Howe, U. Lindstrm, V. Stojevic, Special holonomy sigma models with boundaries, JHEP 0601 (2006) 159,
hep-th/0507035.
[65] S.F. Hassan, O(d, d; R) deformations of complex structures and extended world sheet supersymmetry, Nucl. Phys.
B 454 (1995) 86102, hep-th/9408060.
[66] S.F. Hassan, T-duality, spacetime spinors and RR fields in curved backgrounds, Nucl. Phys. B 568 (2000) 145161,
hep-th/9907152.
[67] S.F. Hassan, SO(d, d) transformations of RamondRamond fields and spacetime spinors, Nucl. Phys. B 583 (2000)
431453, hep-th/9912236.
[68] O. Lunin, J.M. Maldacena, Deforming field theories with U (1) U (1) global symmetry and their gravity duals,
JHEP 0505 (2005) 033, hep-th/0502086.
[69] R. Minasian, M. Petrini, A. Zaffaroni, Gravity duals to deformed SYM theories and generalized complex geometry,
hep-th/0606257.
[70] A. Strominger, S.-T. Yau, E. Zaslow, Mirror symmetry is T-duality, Nucl. Phys. B 479 (1996) 243259, hepth/9606040.
[71] P. Grange, Branes as stable holomorphic line bundles on the non-commutative torus, JHEP 0410 (2004) 002, hepth/0403126.
Partial breaking of N = 2 supersymmetry and

decoupling limit of NambuGoldstone fermion
in U (N ) gauge model
K. Fujiwara
Department of Mathematics and Physics, Graduate School of Science Osaka City University 3-3-138,
Sugimoto, Sumiyoshi-ku, Osaka 558-8585, Japan
Received 20 September 2006; received in revised form 31 January 2007; accepted 7 February 2007
Abstract
We study the N = 1 U (N ) gauge model obtained by spontaneous breaking of N = 2 supersymmetry.
The FayetIliopoulos term included in the N = 2 action does not appear in the resulting N = 1 action and
the superpotential is modified to break discrete R symmetry. We take a limit in which the Khler metric
becomes flat and the superpotential preserves non-trivial form. The NambuGoldstone fermion is decoupled
from other fields but the resulting action is still N = 1 supersymmetric. It shows the origin of the fermionic
shift symmetry in N = 1 U (N ) gauge theory.
2007 Published by Elsevier B.V.
1. Introduction
It was conjectured in [1] that non-perturbative quantities in a low energy effective gauge
theory can be computed by a matrix model. This conjecture was confirmed by [2] for the case of
an N = 1 U (N ) gauge theory with a chiral superfield in the adjoint representation of U (N ).
The N = 1 action is obtained from soft breaking of N = 2 supersymmetry by adding the
tree-level superpotential

d 2 Tr W ().
E-mail address: fujiwara@sci.osaka-cu.ac.jp.

0550-3213/$ see front matter 2007 Published by Elsevier B.V.
doi:10.1016/j.nuclphysb.2007.02.007
(1.1)
146
K. Fujiwara / Nuclear Physics B 770 (2007) 145153
The group SU(N ) is confined and there is a symmetry of shifting the U (1) gaugino by an anticommuting c-number W W 4 . It is called fermionic shift symmetry. Thanks to
this symmetry, effective superpotential is written as

Weff = d 2 F,
(1.2)
for some function F . The fermionic shift symmetry is due to a free fermion and should be related
to a second, spontaneously broken supersymmetry.
AntoniadisPartoucheTaylor (APT) constructed the U (1) gauge model which breaks N = 2
supersymmetry to N = 1 spontaneously by electric and magnetic FayetIliopoulos (FI) terms
[3]. (See also [4].) The U (N ) generalization of the APT model was given in [5] which is
described by N = 1 chiral superfields and N = 1 vector superfields. The NambuGoldstone
fermion appears in the overall U (1) part of U (N ) gauge group and couples with the SU(N )
sector because of the fact that the 3rd derivatives of the prepotential are non-vanishing [6]. A
manifestly N = 2 formulation of U (N ) gauge model [5,6] with/without N = 2 hypermultiplets has been realized in [7]. It overcomes the difficulty in coupling hypermultiplets to the APT
model. Partial breaking of local N = 2 supersymmetry was discussed in a lot of papers [8,9].
This paper is organized as follows. In Section 2, we derive the resulting N = 1 U (N ) action
from the N = 2 U (N ) gauge model [5]. In Section 3, we take a limit in which the Khler metric
becomes flat, while the superpotential preserves its non-trivial form. After taking this limit the
NambuGoldstone fermion is decoupled from other fields, but partial breaking of N = 2 supersymmetry is realized as before. We get a general N = 1 action discussed in [1,2]. It shows that
the fermionic shift symmetry is due to the decoupling limit of the NambuGoldstone fermion. In
Appendix A, we derive the resulting N = 1 supercharge algebra.1
2. Spontaneous partial breaking of N = 2 supersymmetry and resulting N = 1 action
The on-shell action of the N = 2 U (N ) gauge model [5] takes the following form:
L N =2 = Lkin + Lpot + LPauli + Lmass + LFermi4 ,
(2.1)
1
1
a
a
b
Lkin = gab Dm Aa Dm Ab gab vmn
v bmn Re(Fab ) mnpq vmn
vpq
4
8
1
1
1
Fab a m Dm b Fab
Dm a m b Fab a m Dm b
2
2
2
1
a m b
Fab Dm ,
2

1 ab 1
1
0
0
Lpot = g
Da + 2 a
Db + 2 b g ab a W b W ,
2
2
2
2
2 a m n c b
c m n a b
LPauli = i
Fabc vmn + i
F vmn ,
8
8 abc
(2.2)
on-shell
with
1 We follow the notation of [10].
(2.3)
(2.4)

i
1
i
cd
Lmass = Fabc g d W a b W a b Fabc g cd d W a b
4
2
4

1
1
cd
0
c
+ Fabc g Dd + 2 2 d + gac kb a b + c.c.,
4 2
2
i
i
1
LFermi4 = Fabcd c d a b + Fabcd
c d a b + gab F a F b + gab D a D b
8
8
2
i
i
1
+ Fabc F c a b + Fabc F c a b + Fabc D c a b
4
4
2 2
i c a b i c a b
1
c a b
Fabc F Fabc F + Fabc
D
4
4
2 2
147
where D a
(2.5)
(2.6)

d c ), F a i g ab (F c d Fbcd c d ) and W =
+ Fbcd
bcd
4

L
eA0 + mF0 . Let us examine the case with F = nk=0 tr gk!k k . The vacuum condition Apot
a =0
reduces to
F00 =
2 ab
d c
4 g (Fbcd
e i
,
m
(2.7)
where . . . denotes . . . evaluated at Ar = 0 (indices r represent non-Cartan generators). For the
0. It is revealed in [6] that the

sake of simplicity, we choose + sign in (2.7) and this means m
NambuGoldstone fermion exists in the overall U (1) part of U (N ) gauge group,
0

0
N =2
= 2im(1 + 2 ),
0

+ 0
N =2
= 0.
(2.8)
We use . . . for vacuum expectation values which satisfy (2.7). 1 (0 0 ) is the Nambu
2
Goldstone fermion and it will be included in the overall U (1) part of the resulting N = 1
U (N ) vector superfield. The vacuum expectation value of the scalar potential V Lpot is
V = 2m . As is pointed out in [5], the second term in the RHS of the local version of N = 2
supersymmetry algebra enables us to add a constant 2m to the action (2.1) in order to set
V = 0. In the formalism of harmonic superspace, this freedom to add a constant number
comes from arbitrariness to choose the imaginary part of the magnetic FI term in [7].2
To obtain the resulting N = 1 action for the case that U (N ) gauge symmetry is not broken at
vacua, we shift the scalar fields Aa by its vacuum expectation value and mix the spinor fields a
and a . We define

A a Aa A0 0a ,

1
a a a ,
2

1
+a a + a .
2
(2.9)
Substitute these into (2.1), we get the resulting N = 1 U (N ) gauge action after spontaneous
breaking of N = 2 supersymmetry,
L N =1 = L kin + L pot + L Pauli + L mass + L Fermi4 ,
on-shell
2 In [3], such freedom comes from the electric FI term.
(2.10)
148
with
1
1
a
a
b
L kin = g ab Dm A a Dm A b g ab vmn
v bmn Re(F ab ) mnpq vmn
vpq
4
8
1
1
1
F ab a m Dm b F ab
Dm a m b F ab +a m Dm +b
2
2
2
1
+a m +b
Fab Dm ,
2
1
aD
b g ab a W b W ,
L pot = g ab D
8
2
2 a m n +c b
b
Fabc +c m n a vmn
Fabc vmn ,
+i
L Pauli = i
8
8

i
i
1
L mass = F abc g cd d W a b W +a +b F abc g cd d W a b
4
2
4

1
1
d + g ac k c +a b + c.c.,
+ Fabc g cd D
b
4 2
2
i
i
L Fermi4 = F abcd +c +d a b + F abcd
+c +d a b + g ab F a F b
8
8
1
i
i
+ g ab D a D b + F abc F c +a +b + F abc F c a b
2
4
4
1
i
i c a b
c
+a
b
c
+a
+ Fabc D F abc F +b F abc
F
4
4
2 2
1
c +a +b
+ F abc
D ,
2 2
(2.11)
(2.12)
(2.13)
(2.14)
(2.15)
where
a b
a b c
A)
F + Fa A a + Fab A A + Fabc A A A + ,
F(
2!
3!
2 F
A)
F ab F ab
F(
,
F a
,
F ab
, . . . , g ab
2i
A a
A a A b

a i g ab f b A c A d ,
D
ka b i g bc
Da ,
cd
A c
i
i
c d
g ab F bcd +c +d ,
F a g ab F bcd
4
4
2
2 ab +c d
a
ab
+c
d
D
g F bcd
g Fbcd ,
W (e i )A 0 + mF 0 ,
4
4
W
2 W
a W
,
a b W
.
A a
A a A b
Here we have used
1
ia Db + ib Da g cd Fabc Dd = 0,
g ab Da b0 = 0,
2
c
Fabc +a n m +b vmn
= 0,
Fabcd +a +b +c +d = 0.
Take notice that we have added the constant 2m to Lpot as mentioned above.
(2.16)
(2.17)
149
As a result, the action (2.10) agrees with the action (2.1) except for the superpotential term
and FI term. There is no FI term in (2.10), and the superpotential W = eA0 + mF0 get shifted to
W = (e i )A 0 + mF 0 (we neglected a constant term). Because the coefficient (e i ) in W is
a complex number, (2.10) is not invariant under the discrete R transformation.3
We can write the off-shell N = 1 action by introducing auxiliary fields F and D,

1
1
a
a
b
v bmn Re(F ab ) mnpq vmn
vpq
L N =1 = g ab Dm A a Dm A b g ab vmn
4
8
off-shell
1
1
1
F ab a m Dm b F ab
Dm a m b F ab +a m Dm +b
2
2
2
1
1
Fab Dm +a m +b + g ab F a F b + F a a W + F a a W + g ab D a D b
2
2

1 a
i
1
i
+ D Da +
Fabc F c a b W +a +b + F abc F c a b
2
4
2
4

1
1
i
1
c
+ g ac kb c + F abc D c +a b + F abc
F a b W +a +b
2
4
2
2

i c a b
1
1 c +a b
F abc
F + g ca kb c + F abc
D
4
2
2

2
a m n +c b
i
Fabc +c n m a F abc
vmn
8
i
i
F abcd +c +d a b + F abcd
(2.18)
+c +d a b .
8
8
Component fields (A a , +a , F a ) form massive N = 1 chiral multiplets a . Other component
a , a , D
a ) form massless N = 1 vector multiplets V a . The NambuGoldstone fermion
fields (vm
0
is contained in the overall U (1) part of V a .
3. Reparametrization and scaling limit
We consider a limit in which the NambuGoldstone fermion 0 is decoupled from other
fields with N = 2 supersymmetry breaking to N = 1. If the prepotential F is a second-order
polynomial, there are no Yukawa couplings in (2.18) and 0 will be a free fermion. However,
derivatives of the superpotential become zero, a b W = mF 0ab = 0 and a W = (e i )a0 +
mF 0a = (e i )a0 + mF0a = 0. This means that the superpotential does not contribute to
(2.18) and it preserves N = 2 supersymmetry. This problem can be solved by a large limit of the
parameters (e, m, ), i.e. large limit of electric and magnetic FI terms.
We reparametrize gk =
F=
n

k=0
3 R:
gk
(k
3) and (e, m, ) = (e , m , ). The prepotential F is

n
gk k
g2 2
1 gk k
tr = tr g0 1 + g1 + +
tr ,
k!
2
k!
a
+a
.
+a
k=3
(3.1)
150
and we see the dependence of the following terms:

1
A c A d
e + i

F ab = Fab +
A c + Fabcd

Fabc
+ =
ab + O 1 ,
2!
m

1
1

F abc = O ,
F abcd = O ,
g ab = ab + O 1 ,
m

i
a = i g ab f b A c A d = ab f b A c A d + O 1 ,
D
(3.2)
cd
cd
m

g
where F = tr(g0 1 + g1 + g22 2 ) + nk=3 tr k!k k . Note that the scaling parameter is cancelled out in the superpotential term:

1

A b + F0abc
A b A c + ,
a W = (e i )a0 + mF 0a = m F0ab
(3.3)
2!

1

+ F0abc
A c A d + .
a b W = m F0ab
(3.4)
A c + F0abcd
2!
Take a limit , and the action (2.18) is converted into

L = ab Dm A a Dm A b i+a m Dm +b
m

2 b c +a d
2 b c +a d
i b a c d
a b
+ F F fcd D A A +
f A +
f A
2
2 dc
2 dc

1 a b
1 a bmn 1 e mnpq a b
a m
b
+
vmn vpq i Dm + D D

+ ab vmn v
m
4
8
2
1
1
+ F a a W + F a a W a b W +a +b a b W +a +b .
2
2
(3.5)
The matrix form of the superpotential W is given as4

1
1

A a A b + F0abc
A a A b A c +
F0ab
W m
2!
3!

1
1
a b
a b c
F0ab A A + F0abc A A A +

=m
2!
3!

n
m gk
A0 k1
1
mF0 mF0a A a
=
tr A +
(k
1)!
2N k=1
2N
=m
n2

hk
tr A k+1 ,
k+1
(3.6)
k=1
where we define hk
binomial coefficient.
(k+1)
2N
n2k
=0
A0
gk++2
.
(k++1)! (k++1) C
2N
Here the symbol (k++1) C is a
4 We normalize the standard u(N) Cartan generators t as tr(t t ) = 1 , which implies that the overall u(1) generator
i
i j
2 ij
is t0 = 1 1N N .
2N
151
We can rewrite the action (3.5) in superfield formalism as

e + i
+
W
+ c.c. ,
L = Im
d 2 W ()
2 d 4 tr + eV + d 2 tr W
m
(3.7)
where W is the field strength of V . The factor 2 in the first term comes from the normalization
of the standard u(N) Cartan generators. Note that the NambuGoldstone fermion 0 , which is
contained in the overall U (1) part of N = 1 U (N ) vector superfields V , is decoupled from other
fields in (3.7). However N = 2 supersymmetry is broken to N = 1 because of existence of the
superpotential. We get a general N = 1 action (3.7), which is known as a soft broken N = 1
action, from spontaneously broken N = 2 supersymmetry. We conclude that the fermionic shift
symmetry in [2] is related to the decoupling limit of the NambuGoldstone fermion.
Acknowledgements
The author would like to thank Hiroshi Itoyama, Makoto Sakaguchi, Kazunobu Maruyoshi
and Hironobu Kihara for very useful discussions.
Appendix A. Supercharge algebra
The N = 2 transformation rule are given by a combination of following transformation rules,5
1 Aa = 21 a ,
1 a = i 2 m 1 Dm Aa + 21 (F a g ab b W ),
1 ab
a + i (D
a
0
1 a = 12 m n 1 vmn
1 2 g (Db + 2 2 b )),
a
m
a
a
m
1 vm = i1 i 1 ,
2 Aa = 22 a ,
1 ab
0
a i (D
a
2 a = 12 m n 2 vmn
2 + 2 g (Db 2 2 b )),
a
m
a
a
ab
2 Dm A 22 (F g b W ),
2 a = i 2
2 vm = i2 m a i a m 2 ,
where spinors k (k = 1, 2) are transformation parameters. The N = 2 supersymmetric transm from the
formation rules are N =2 a = 1 a + 2 a . We can find the 1st supercurrent S1
action (2.1):

m c
1
a

m a Da + i 2 ec0 + mF0c
S1m = igab np m b vpn
2
2 m 0 2gab n m a Dn Ab + ,
(A.1)
m is given by the
where the dots denote terms involving three fermions. The 2nd supercurrent S2
m
discrete R transformation of S1 with a flip of the sign of the FI parameter ,

m c
1
a
S2m = igab np m b vpn

m a Da i 2 ec0 + mF0c
2
+ 2 m 0 + 2gab n m a Dn Ab + .
(A.2)
5 It is easy to give proof that L = 0 (up to total derivative) with the use of L = 0 and RL = L|
2
1
. (See [5].)
As in [11], the FI term does not break N = 2 supersymmetry.
152
Supercharge algebra is derived by

0
A , S 0 = i QA , S 0 + i Q , S 0 ,
= i A QA + A Q
A SB
A
B
B
A
A B
(A.3)
where A, B = 1 or 2. It may be irrelevant to denote supercharges as Q1 , Q2 because N = 2 supersymmetry is broken to N = 1 spontaneously and the supercharge corresponding to the broken
supersymmetry is ill-defined. We ignore this point here and write the divergent part explicitly.
We obtain the central charge

{Q1 , Q2 } = 2i dx 3 i Ab Re Fab 2ia K 0ij k vjak + 2gab Ab v a0i

+ 8 d 3 xi A0 0i .
(A.4)
Here K = 2i (Aa Fa Aa Fa ) is the Khler potential. To get the resulting N = 1 supercharge
algebra, we define Q 1 (Q1 Q2 ) and Q+ 1 (Q1 +Q2 ). Anti-commutators of Q (Q+ )
2
2
(Q + ) are given as
and Q

i
3
c
d q r
Q , Q = i d x g ab gac vnp
n p + iDa 0 gbd vqr
+ iDb
4

a
b n 0 p
ab
,
2igab Dp A Dn A 2ig a W b W +

+
Q+
, Q = i

i
c
d q r
d 3 x g ab gac vnp
n p + iDa 0 gbd vqr
+ iDb
4

2igab Dp Aa Dn Ab n 0 p 2ig ab a W b W 0 +

8m 0
d 3 x,
(A.5)
where the dots indicate terms involving fermion fields. This result agree with the supersymmetry
algebra in [12]. Finally, we conclude that Q is the unbroken generator and Q+ is the broken
one.
References
[1] R. Dijkgraaf, C. Vafa, Matrix models, topological strings, and supersymmetric gauge theories, Nucl. Phys. B 644
(2002) 3, hep-th/0206255;
R. Dijkgraaf, C. Vafa, On geometry and matrix models, Nucl. Phys. B 644 (2002) 21, hep-th/0207106;
R. Dijkgraaf, C. Vafa, A perturbative window into non-perturbative physics, hep-th/0208048.
[2] F. Cachazo, M.R. Douglas, N. Seiberg, E. Witten, Chiral rings and anomalies in supersymmetric gauge theory,
JHEP 0212 (2002) 071, hep-th/0211170.
[3] I. Antoniadis, H. Partouche, T.R. Taylor, Spontaneous breaking of N = 2 global supersymmetry, Phys. Lett. B 372
(1996) 83, hep-th/9512006.
[4] H. Partouche, B. Pioline, Partial spontaneous breaking of global supersymmetry, Nucl. Phys. B (Proc. Suppl.) 56
(1997) 322, hep-th/9702115.
[5] K. Fujiwara, H. Itoyama, M. Sakaguchi, Supersymmetric U (N ) gauge model and partial breaking of N = 2 supersymmetry, Prog. Theor. Phys. 113 (2005) 429, hep-th/0409060;
K. Fujiwara, H. Itoyama, M. Sakaguchi, U (N ) gauge model and partial breaking of N = 2 supersymmetry, hepth/0410132.
153
[6] K. Fujiwara, H. Itoyama, M. Sakaguchi, Partial breaking of N = 2 supersymmetry and of gauge symmetry in the
U (N ) gauge model, Nucl. Phys. B 723 (2005) 33, hep-th/0503113.
[7] K. Fujiwara, H. Itoyama, M. Sakaguchi, Partial supersymmetry breaking and N = 2 U (N (c)) gauge model with
hypermultiplets in harmonic superspace, Nucl. Phys. B 740 (2006) 58, hep-th/0510255;
K. Fujiwara, H. Itoyama, M. Sakaguchi, Supersymmetric U (N ) gauge model and partial breaking of N = 2 supersymmetry, hep-th/0602267.
[8] S. Ferrara, L. Girardello, M. Porrati, Minimal Higgs branch for the breaking of half of the supersymmetries in N = 2
supergravity, Phys. Lett. B 366 (1996) 155, hep-th/9510074;
S. Ferrara, L. Girardello, M. Porrati, Spontaneous breaking of N = 2 to N = 1 in rigid and local supersymmetric
theories, Phys. Lett. B 376 (1996) 275, hep-th/9512180;
P. Fre, L. Girardello, I. Pesando, M. Trigiante, Spontaneous N = 2 N = 1 local supersymmetry breaking with
surviving local gauge group, Nucl. Phys. B 493 (1997) 231, hep-th/9607032;
M. Porrati, Spontaneous breaking of extended supersymmetry in global and local theories, Nucl. Phys. B (Proc.
Suppl.) 55 (1997) 240, hep-th/9609073;
J.R. David, E. Gava, K.S. Narain, Partial N = 2 N = 1 supersymmetry breaking and gravity deformed chiral
rings, JHEP 0406 (2004) 041, hep-th/0311086;
L. Andrianopoli, R. DAuria, S. Ferrara, M.A. Lledo, N = 2 super-Higgs, N = 1 Poincar vacua and quaternionic
geometry, JHEP 0301 (2003) 045, hep-th/0212236.
[9] H. Itoyama, K. Maruyoshi, U (N ) gauged N = 2 supergravity and partial breaking of local N = 2 supersymmetry,
hep-th/0603180.
[10] J. Wess, J. Bagger, Supersymmetry and Supergravity, second ed., Princeton Univ. Press, 1992.
[11] P. Fayet, FermiBose hypersymmetry, Nucl. Phys. B 113 (1976) 135.
[12] J. Hughes, J. Polchinski, Partially broken global supersymmetry and the superstring, Nucl. Phys. B 278 (1986) 147.
Nuclear Physics B 770 [FS] (2007) 155178
The spectrum of open string field theory at the stable

tachyonic vacuum
C. Imbimbo
Dipartimento di Fisica, Universit di Genova, Istituto Nazionale di Fisica Nucleare,
Sezione di Genova, via Dodecaneso 33, I-16146 Genova, Italy
Received 30 November 2006; accepted 15 January 2007
Abstract
We present a level (10, 30) numerical computation of the spectrum of quadratic fluctuations of open
string field theory around the tachyonic vacuum, both in the scalar and in the vector sector. Our results
are consistent with Sens conjecture about gauge-triviality of the small excitations. The computation is
sufficiently accurate to provide robust evidence for the absence of the photon from the open string spectrum.
We also observe that ghost string field propagators develop double poles. We show that this requires nonempty BRST cohomologies at non-standard ghost numbers. We comment about the relations of our results
with recent work on the same subject.
1. Introduction, summary and discussion

In this paper we extend and improve the analysis of bosonic open string field theory (OSFT) at
the stable vacuum that we started in a previous work [1]. OSFT possesses a classical, translational
invariant solution whose energy density exactly cancels the brane tension and which is thought
to represent the closed string vacuum with no open strings [2,3]. The existence of such a solution
has been persuasively demonstrated first [68] within the level truncation (LT) expansion [5],
and, more recently, analytically [9]. The closed string interpretation requires that the spectrum
of quadratic fluctuations around this classical solution be not only tachyon-free but also gaugetrivial. This expected property of the tachyonic vacuum goes under the name of Sens OSFT third
conjecture.
E-mail address: camillo.imbimbo@ge.infn.it.
doi:10.1016/j.nuclphysb.2007.01.035
156
C. Imbimbo / Nuclear Physics B 770 [FS] (2007) 155178
In [1] we explained what Sens third conjecture implies for the gauge-fixed quadratic OSFT
action expanded around the stable vacuum: each pole of the open string field propagator should
cancel with appropriate poles of the (second quantized) ghost string fields. To state it more
(n)
precisely, let us denote by L 0 (p) (where n = 0, 1, . . . is (minus) the second quantized ghost
number1 ) the gauge-fixed kinetic operators for both matter and ghost string fields acting on states
(n)
of momentum p and ghost number n. L 0 (p) is the restriction to states of ghost number n and
momentum p of the operator
b0 }
L 0 = {Q,
(1.1)
is the BRS operator associated with the classical stable OSFT solution and b0 is the zero
where Q
(n)
mode of the 2d CFT antighost field that implements the Siegel gauge condition. If det L 0 (p)
2
2
has a zero of order dn for p = m , the number of physicali.e. gauge-invariantdegrees of
freedom of mass m is given by the FadeevPopov index:
IFP (m) = d0 2d1 + 2d2 + =
(1)n dn .
(1.2)
n=
The spectrum of gauge-invariant quadratic fluctuations is empty if, and only if, the above index
vanishes for all p 2 .
Both [1] and the present paper study the spectrum of quadratic fluctuations of OSFT around
the stable vacuum within the framework of the LT expansion. The key drawback of LT expansion
is that it breaks (second quantized) BRS invariance of the gauge-fixed OSFT action around the
stable vacuum. Consequently, poles of propagators of matter and ghost string fields which are
degenerate in the exact theory correspond, in the level truncated theory, to multiplets of poles that
are only approximately degenerate. The FadeevPopov index (1.2) should therefore be defined,
(n)
in the level truncated theory, by including zeros of gauge-fixed kinetic operator L 0 (p) which
belong to the same approximately degenerate multiplet. In order for this definition to make sense
the level has to be large enough that the splitting among zeros belonging to the same multiplet is
significantly smaller than the separation between multiplets. It is expectedand explicit numerical computations confirm thisthat matter and ghost propagators poles begin clustering together
into well-defined approximately degenerate multiplets for levels which are increasingly large as
m2 = p 2 . In practice, therefore, the level truncated numerical analysis can probe reliably
only a limited range of values of m2 .
The analysis of [1] was limited to the Lorentz scalar sector of the theory. Numerical computations were performed using the approximation which, in terminology of [6], was of type (L, 3L)
for levels L up to 6 and of type (L, 2L) for levels L = 7, 8, 9. For p 2 > 6.0 propagators poles
were found only in the twist-odd sector: it was observed that they form an approximately degenerate multiplet with vanishing FadeevPopov index around p 2 = m2scalar, 2.1. In the
twist-even scalar sector, propagators have no poles up to p 2 6: at the level reached by the
computation, poles with p 2 < 6.0 do not show yet any clear and stable multiplet structure.
Although these findings are well consistent with gauge-triviality of the spectrum of quadratic
excitations, an unexpected result was also obtained in [1]. The non-vanishing FadeevPopov
1 We are adopting the convention in which the SL(2, R) invariant vacuum |0 has ghost number 1. In the natural
(n)
= (L (n) ) . Thus ghost and antighost string fields form canonically conjugate pairs (n , n )
hermitian product, L 0
with n > 0.
157
degrees dn of the approximately degenerate multiplet of zeros of det L 0 (p) at p 2 = m2scalar,

were found to be
(n)
d0 = 2,
d1 = 2,
d2 = 1.
(1.3)
It was moreover observed that while the (approximate) double zero of the determinant of the
(0)
matter kinetic operator is associated to two distinct vanishing eigenvalues of L 0 (p), the double
zero of det L (1)
(p) is due to a single eigenvector of the kinetic operator of ghost numbers 1
0
whose eigenvalue
2

p 2 p 2 + m2scalar, ,
(1.4)
has an (approximate) double zero at p2 = m2scalar, . In other words there exists a ghost
antighost string field pair with ghost number 1 whose propagator develops a double pole for
p 2 = m2scalar, .
Physical states are elements of the Q-cohomology

with zero ghost number. Let us denote
the cohomology of Q
on states of ghost number n. Because of (1.1), the zeros of
by H(n) (Q)
acting on H(n) (Q)
with n different
the gauge-fixed kinetic operators also encode properties of Q
than zero. In [1] it was argued that the double pole of the propagator of the ghostantighost string
requires as well
field pair at p 2 = m2scalar, , although consistent with the vanishing of H(0) (Q),
that
= dim H(2) (Q)
=1
dim H(1) (Q)
(1.5)
p2 .
for the same value of

In the present paper we confirm and refine this analysis in various ways. First, we compute
(n)
the gauge-fixed kinetic operators L 0 (p) for both the scalar and the vector Lorentz sector. We
also improve our LT computation, by performing the numerical evaluations in the (L, 3L) approximation2 up to level L = 10. We find, on top of the multiplet of poles of the twist-odd scalar
propagators already discovered in [1], multiplets of propagator poles, which are approximately
degenerate and have vanishing FadeevPopov indices, both in the twist-even and in the twist-odd
vector sector. The increased level reached by the computation allows for a simple linear extrapolation of the locations of the poles of the approximately degenerate FadeevPopov multiplets.
The poles of propagators of different ghost numbers when extrapolated at level L = become
indeed nearly coincident: See Table 3 and Figs. 35 of Section 4. The extrapolated values of the
masses of the degenerate multiplets are
m2scalar, = p 2 = 2.0,
m2vector,+ = p 2 = 4.0,
m2vector, = p 2 = 5.9,
(1.6)
where the indices refers to the twist even/odd sectors. The values (1.6) of the poles are intriguingly consistent with integer even values of m2 . The observed structure of the three multiplets is
the same. FadeevPopov degrees are as in Eq. (1.3); in all the three cases listed in (1.6), there exists a single ghostantighost string field pair with ghost numbers 1 whose propagator develops
an (approximate) double pole.
2 Our numerical findings confirm the conclusion, shared by other authors, that the approximation of type (L, 2L) is
not satisfactory for this problem.
158
These numerical results confirm Sens conjecture for both scalars and vectors in the region
p 2 6. In particular our approximation should be quite accurate for p 2 = 0: hence, our
computation provides (the first) robust direct evidence for the absence of the photon in the nonperturbative stable vacuum.
At the same time, the arguments developed in [1] imply that non-empty cohomologies at nonstandard ghost numbers must exist in the vector sectors as well as in the scalar sector. Since
this conclusion is somewhat unexpected and appears to contradict other works [10,11,14], we
reconsider and strengthen in the present paper the analysis of [1], which relates the double pole
of the ghostantighost string field pair to non-vanishing cohomologies at non-standard ghost
numbers.
The main mathematical tool used in this analysis is the long infinite cohomology sequence
H(n) (Q)
h (n1) (Q)
h (n+1) (Q)

h (n) (Q)
(1.7)
in terms of relative tilde h (n) (Q)

and check h (n) (Q)
that computes the cohomologies H(n) (Q)
Q-cohomologies. Both tilde and check relative cohomologies are defined on fields which satisfy
on fields
the Siegel gauge condition. The tilde relative h (n) cohomology is the cohomology of Q
n that are in the kernel of both L0 and b0

b0 n = L 0 n = 0.
(1.8)
b0 }. In [1] we also introduced the

This is a consistent cohomological problem since L 0 = {Q,
check relative Q-cohomology

h (n) on the space of fields that satisfy both (1.8) and
n = L 0 n+1
Z
(1.9)
in the c0 and b0 algebra

where the operators Z and L 0 are defined by the decomposition of Q
Q = c0 L 0 + b0 D + M + c0 b0 Z.
(1.10)
Since
We will review in Section 3 why this also is a well posed cohomological problem for Q.
Z = [c0 , L0 ], the BRS operator associated with the unstable perturbative vacuum represented
by the bosonic 25-brane has Z = 0. Consequently, the tilde and the check cohomologies of the
perturbative BRS operator coincide. Our numerical computations indicate that Z does not
vanish for the non-perturbative Q.

In [1] the cohomology sequence (1.7) was derived by making certain technical assumptions
that were not otherwise proven. It was assumed that the space of gauge-fixed states decomposes
as the direct sum of the kernel and the image of L 0 . In this paper we relax this hypothesis and
establish the validity of (1.7) beyond reasonable doubt. We emphasize that this is an exact result,
independent of any numerical computation.
We also show, without invoking any of the mathematical structures provided by the technical
assumption made in [1], that the experimentally observed propagator double poles of the ghost
string field pairs with ghost number 1 are compatible with the exact long sequence only if
Eq. (1.5) for non-standard cohomologies holds for the corresponding values of p 2 listed in (1.6).
Furthermore we refine this result by making use of the following observation.3 It is known [18]
that L 0 commutes with an SU(1, 1) symmetry whose J3 generator is half of the ghost number.
3 In [1] it was established that a certain number of different assignments for the relative cohomologies, all of which
implied (1.5), were consistent with the long exact sequence. The analysis could not provide a unique possibility for the
explicit representatives of the absolute cohomologies. This is achieved in the present paper.
159
the J+ generator does. ExploitAlthough the full SU(1, 1) symmetry does not commute with Q,
ing this symmetry we show that the null vectors of the kinetic operators for ghost number 1
and 2 provide explicit representatives of the non-vanishing cohomologies:
L (2)
(p)v2 = 0,
0
[v2 ] H(2) (Q),
(1)
L 0 (p)v1 = 0,
[v1 ] H(1) (Q).
(1.11)
We provide the detailed proof of (1.11) in Section 5. However it is useful to summarize here
acts on the space W (p)
the basic reasons for this result. The important observation is that Q
spanned by the vectors which are in the kernel of both b0 and L0 at a given p 2 . These are the null
vectors responsible for the poles of the string field propagators. The approximately degenerate
poles that we found numerically correspond in the exact theory to a null space W (p) with the
same dimension six for all the three sectors and values of p 2 listed in Eq. (1.6). For these values
acting on W (p) reduces to the operator M that appears in the decomposition (1.10).
of p 2 , Q
Therefore M 2 = 0 on the six-dimensional space W (p). The (Witten) index of this supersymmetry
is 4 2 = 2 and equals the index of the relative tilde and check cohomologies

index h =
(1.12)
(1)n dim h (n) = index h =
(1)n dim h (n) = 2.
n
This implies that the relative cohomologies cannot all vanish.

W (p) decomposes into a singlet, a doublet and a triplet representation of the SU(1, 1) symmetry. Clearly, there exists only a finite number of inequivalent representations of the supersymmetry operator M acting on the finite-dimensional space W (p). Among these representations
There are four such inequivalent
one should focus on those for which J+ commutes with M.
representations, as shown in Section 5. Since W (p) is finite dimensional, the infinite long exact
= 0 causes the exact sequence
sequence becomes a finite exact sequence. The fact that H(0) (Q)
to split into four shorter sequences, putting further constrains on the possible representations of
In the end, it becomes a matter of simple linear algebra to show that only one representation
M.
of M is compatible with the exact sequences:
2 = 0,
Mv
1 = 0,
Mv
t = 0,
Mv
0
s = v1 ,
Mv
0
(1.13)
{v0t , v2 }
where is the SU(1, 1) singlet, {v1 } the doublet, and

the triplet.
We went into these details to clarify that the result (1.11) relies exclusively on the assumption
that the multiplets of approximately degenerate propagators poles that we found numerically
for the values of p2 listed in (1.6) do really correspond in the exact theory to multiplets of
exactly degenerate poles. This assumption cannot of course rigorously be proven by numerical
methods alone. A priori one can imagine that, as the level increases, either some of the poles we
found disappear or some new pole shows up. This, although possible in principle, seems however
unlikely for the following reasons.
To start with, the zeros pn2 (L) of the determinants det L (n)
0 (p) as the level is increased from
L = 4 to L = 10 are nicely interpolated by linear relations
v0s
pn2 (L) = pn2 +
qn
L
(1.14)
with intercepts pn2 (corresponding to L = ) which are independent of the ghost numbers n =
0, 1, 2 with remarkably good approximation: See Table 3 and Figs. 35 of Section 4. The zeros
with n = 0, 2 correspond to eigenvalues which vanish linearly in p 2 : on topological grounds,
160
they are stable and therefore unlikely to disappear altogether. As we have mentioned above, the
zeros with n = 1 correspond instead to eigenvalues which vanish quadratically in p 2 and thus
they are not protected by topological reasons. Therefore one could think thatcontrary to what
our numerical computations seem to indicatethe pair of zeros with n = 1 do not correspond,
in the exact theory, to a single eigenvalue with a double zero: rather, one might fear that, as the
level is increased, this pair of zeros would eventually be lifted. However, if this were the case, the
FadeevPopov index would become positive and physical degrees of freedom would appear. In
other words, if we assume Sens conjecture, it becomes very difficult to provide an interpretation
of our numerical findings different than what we have proposed.
that
Two more independent tests support the conclusion about the exotic cohomologies of Q
we inferred from the numerical computation. The first argument was already developed in [1]. If
= 0, then the
Q(p)
has non-vanishing cohomologies for discrete values of p2 only and H(0) (Q)
following equation must hold in the exact theory
dim H(2) (Q)
+ .
0 = dim H(1) (Q)
(1.15)
Our result (1.11) does satisfy this constraint and this seems to be a non-trivial check of its correctness.
One more reasoning that strengthen our belief in the correctness of our conclusion (1.11) is
suggested by the numerical extrapolations (1.6) obtained in the present paper. Poles of propagators that correspond to gauge-trivial excitations are, in general, gauge-dependent. According
to our extrapolations, the poles of the string fields seem to correspond, in the exact theory, to
integer even values of m2 . It becomes difficult to understand why this is so unless such poles
have a gauge-invariant meaning, like the one provided by (1.5).
After we completed our numerical computations, the paper [14] appeared where, following
previous work [11], an analytical proof of the absence of physical states of OSFT around the
stable vacuum is presented. The idea of this proof is to show that the identity state I is Q-exact:
I = QA.
(1.16)
The authors of [14] provide an explicit expression for the trivializing state A
1
B0 I
(1.17)
L0
where the operators L0 and B0 are obtained from the usual perturbative operators L0 and b0 by
means of a certain coordinate transformation. The exactness of the identity implies not only the
but also that of H(n) (Q)
for n generic: this of course contradicts our result
emptiness of H(0) (Q)
4
in (1.11).
We do not have yet a definite understanding of this conflict. The discussion we presented
above makes it clear that the only reasonable way to avoid our conclusion regarding cohomologies at non-standard ghost numbers is that level truncation is simply not appropriate to study the
spectrum of OSFT at p 2 < 0, regardless of the level. In other words one has to admit that the
approximately degenerate multiplets of poles that we detected are a finite level artifact that have
no correspondence in the exact theory. We just reviewed the reasons why this, although possible
in principle, seems difficult to understand.
A=
4 The conflict between our result about cohomologies at non-zero ghost number and the triviality of the identity
prompted us to both improve our numerical results and to relax the technical hypothesis that were assumed in [1] to
derive the exact long sequence (1.7).
161
On the other hand, it should be remarked that the proof presented in [14] has a somewhat
formal character. The reason is that the identity state is not a completely good state of OSFT
star algebra, since it has several anomalous properties, discussed for example in [1517]. The
argument of [14] is that any state which is Q-closed

is also Q-exact
since Eqs. (1.16) and
(1.17) imply
).
= Q(A
(1.18)
The question we are raising therefore is if the state A is well-defined or, more precisely, if the
star product of A with any string field is well-defined.
We think this question deserves further investigation. Here, we limit ourselves to observe that
the assumption of the existence of the identity leads to consequences that look quite dramatic
from the point of view of the gauge-fixed second quantized theory. We explained that the absence
of physical states is a statement regarding the null vectors of the gauge-fixed kinetic operators
(n)
L 0 (p) for any p 2 , any ghost number n and any Lorentz quantum number. Now, assuming that
the identity state does exist, one could consider, in analogy with (1.17), the following state
1
b0 I.
A =
L0
(1.19)
A = I P I
Q
V1
(1.20)
Then
where PV1 is the projector on the subspace V1 of states that are in the kernel of L 0 , have
momentum p = 0, ghost number 1 and are twist-parity even Lorentz scalars. Therefore if
V1 is empty, A trivializes the identity. In other words, the existence of the identity implies
that a property of the propagators at p 2 = 0 in some definite Lorentz and ghost sectorthe
vanishing of V1 would determine the behaviour of the propagators for all p 2 and all quantum
numbers. This seems a very (and maybe too) strong statement and, we feel, suggests caution
when manipulating the identity.
Note that the vanishing of V1 is a question that can reliably addressed in the LT expansion,
since it involves a sector with vanishing momentum, definite twist parity, Lorentz and ghost
quantum numbers. To test the emptiness of V1 it is enough to compute the determinants of the
twist-even scalar kinetic operators at zero momentum and ghost number 1 and 2. In fact,
this is a very particular case of the computation we performed in this paper (and in [1], for that
matter). For p = 0 our level 10 approximation should by all means be reliable: the same sector
(but with ghost number 0) is the one where the same approximation turned out to capture quite
accurately the properties of the stable classical vacuum solution. Since we have not detected any
zeros of the determinants of the twist-even scalar kinetic operators (for both ghost numbers 1
and 2) at p 2 = 0, we can confidently assert that V1 is empty.
In this sense therefore the numerical computations presented both in this paper and in [1]
are coherent with what proven in [14]. If the expression in Eq. (1.19) (analogous to Eq. (1.17))
defined a good state, our numerical computations, when restricted to p = 0, ghost number
1, and to the twist-even scalar sector, could be interpreted as a reliable numerical proof of
the triviality of the identity. The extension of the same analysis to non-vanishing p 2 appears to
contradict this conclusion, however: in summary, this might signals either a failure of LT when
extended to p 2 < 0 or the formal character of expressions like (1.19) and (1.17).
162
In order to elucidate this question it should be helpful to extend our computations to gauges
different than the Siegel gauge. This would allow testing the gauge invariant meaning of the
propagator double poles. We leave this to future work.
We add a final comment. In [19] the spectrum of OSFT around so-called universal solutions
was studied analytically. It was found there that for such background while H(0) vanishes, BRS
cohomologies with ghost number 1 and 2 are not empty. Although this result is intriguingly
reminiscent of ours, it also differs from it in various respects. Cohomologies at non-standard
ghost numbers around the universal solutions are isomorphic to perturbative, ghost number zero,
cohomologies. In particular they exist for p 2 = m2 = 2, 0, 2, . . . . We do not find cohomologies for m2 = 2, 0. Moreover the cohomologies that we do find at m2 = 2, 4, 6 do not have the
same quantum numbers as the perturbative ones. More work is needed to understand the relation,
if there is one, between cohomologies around the stable classical solution and around universal
solutions.
The rest of this paper is organized as follows. In Section 2 we briefly review, for selfcontainedness, the gauge-fixing procedure of OSFT in the classical stable vacuum. In Section 3
we derive, relaxing the additional technical assumptions of [1], the long exact cohomology sequence (1.7). In Section 4 we report the result of our numerical level (10, 30) computation. In
Section 5 we work out the unique action (1.13) of M on the null vectors of the gauge-fixed kinetic operators, which should correspond, in the exact theory, to the approximately degenerate
propagators poles found in Section 4.
2. Gauge-fixed open string field action
The open string field theory (OSFT) action around the tachyonic background writes [4]
1
) + 1 (, ),
[ ] = (, Q
(2.1)
2
3
is the classical open string field, a state in the open string Fock space of ghost number zero.
(A, B) is the bilinear form between states A and B of ghost numbers gA and gB , respectively.
(A, B) vanishes unless gA + gB = 1. is Wittens associative and non-commutative open string
is the BRS operator around the non-perturbative vacuum
product. Q
Q + [ , ]
Q
(2.2)
where
[A,B] A B ()(gA +1)(gB +1) B A
(2.3)
and Q is the perturbative BRS operator, which is (anti)symmetric with respect to the bilinear
inner product (, ) based on BPZ conjugation. is the solution of the classical equation of
motion
Q + = 0
(2.4)
that represents the tachyonic vacuum. The flatness Eq. (2.4), together with the associativity of
Q is (anti)symmetric with respect to the product (, )
the -product, ensures the nilpotency of Q.
thanks to the property
(A, B) = (A , B).
(2.5)
163
The action (2.1) is thus invariant under the following gauge transformations
+ [ , C]
= QC
(2.6)
where C is a ghost number 1 gauge parameter.

CFT ghost number g provides a grading for string fields: matter string field have g = 0. It
is useful to introduce another grading, the second quantized string field ghost number, that we
will denote by nsft . Matter fields have nsft = 0, by definition. Fields with second quantized ghost
(n)
number nsft = n and CFT ghost number g will be denoted with g .
The gauge invariance (2.6) of the classical OSFT action translates into the second quantized
BRS symmetry

(0)
(1) + (0) , (1)
BRS 0 = Q
(2.7)
1
0
1
(1)
where 1 is the ghost string field of first generation.

We will gauge-fix the invariance (2.7) by going to Siegel gauge:
b0 0(0) = 0
(2.8)
Gauge-fixing the OSFT action requires an infinite number of ghost field generations [12]. We
will adopt the Siegel gauge for all higher-generation ghost string fields:
(n)
b0 n = 0.
(2.9)
(n)
For any field m one can write the decomposition

(n)
(n)
+ c0 m1
m(n) = m
(n)
(2.10)
(n)
where m and m1 are fields that do not contain c0 :

(n)
b0 m
=0
m, n.
(2.11)
The corresponding, completely gauge-fixed, quadratic action is

(2)
g.f. =

1 (0)
(0)
(n)
n(n) , c0 L 0 n
0 , c0 L 0 0 +
2
(2.12)
n=1
where
b0 }.
L 0 {Q,
(2.13)
(n)
n
Thus the gauge-fixed OSFT action depends on fields

n which are b0 -invariant states
of the first quantized Fock space with CFT ghost number n and second quantized ghost number
n. We will denote this state space with n .
It is convenient to define the following non-degenerate bilinear form
, on n n
, (, c0 ).
From the definition (2.13) of L 0 and from the Jacobi identity one obtains:

b0 }, c0 = b0 , {Q,
c0 } = [b0 , D]
[L 0 , c0 ] = {Q,
(2.14)
(2.15)
164
c0 }. This ensures that L 0 is an operator on n which is symmetric with respect

where D {Q,
the bilinear form
, :
n , L 0 n =
L 0 n , n .
(2.16)
In conclusion the quadratic part of the gauge-fixed OSFT action at the tachyonic background
writes as

1
(2)
g.f.
=
0 , L 0 0 +
n , L 0 n .
2
(2.17)
n=1
3. Relative and absolute cohomologies

the Q
Let Fn be the space of states of CFT ghost number n. Let us denote by H(n) (Q)
(n)
cohomologies on Fn . We will refer to H (Q) as the absolute BRS state cohomologies. As we

recalled in the Introduction, the number of physical states of open string theory is given by the
cohomology.
dimension of the H(0) (Q)
is based on the preliminary computation of a different kind of
One way to compute H(n) (Q)
Q-cohomologiesthe
relative cohomologies. Let W n be the subspace of Fn of states n of ghost
number n which are both b0 and L 0 invariant:
def
n W n b0 n = L 0 n = 0.
(3.1)
The relative Q-cohomology

of ghost number n is given by the Q-closed
states n W n
n=0
Q
(3.2)
image of W n1
modulo the states which are in the Q
n1
n n = n + Q
(3.3)
where n1 W n1 . Such a definition is consistent since

b0 } = L 0 .
{Q,
(3.4)
We will denote the relative cohomologies of Q by

in the b0 , c0 algebra:
Let us decompose Q
h (n) .
Q = c0 L 0 + b0 D + M + c0 b0 Z
(3.5)
M and Z are independent of c0 and b0 . The crucial difference between the decomwhere L 0 , D,
and its perturbative analogue is the term proportional to
position (3.5) of the non-perturbative Q
c0 b0 , which is absent in the perturbative case. Note that
b0 } = L 0 + b0 Z,
L 0 {Q,
c0 } = D c0 Z
D {Q,
(3.6)
= Z,
in agreement with the Jacobi identity (2.15). The first
and therefore [L 0 , c0 ] = [b0 , D]
equation of (3.6) implies that W n is the kernel of L 0 on n (i.e. the space of b0 -invariant states
of ghost number n):
(n)
W n = ker L 0 .
(3.7)
are equivalent to the following equations

The nilpotency of Q
M 2 + D L 0 = 0,
Z}
+ Z 2 = [D,
L 0 ],
{M,
L 0 = 0,
L 0 M (M + Z)
M + Z)
= 0.
M D D(
165
(3.8)
These equations show that the b0 -relative cohomology h (n) is the cohomology of the operator M
on W n :
W n ).
h (n) = H(n) (M,
(3.9)
Indeed, the first of Eq. (3.8) says that M 2 = 0 on W n and the third of Eq. (3.8) guarantees that
M : W n W n+1 .
Let us denote by Vn the kernel of L 0 on Fn :
(n)
Vn = ker L 0 .
(3.10)
on Fn is identical to the cohomology of Q on Vn .

Thanks to Eq. (3.4), the cohomology of Q
on Vn in terms
Now we come to the main point. We want to describe the cohomology of Q
of cohomologies defined on the Wn s, the gauge-fixed (b0 -invariant) spaces. There exists two
natural maps between these spaces: the immersion map
: W n Vn ,
(n ) = n
(3.11)
and the projection :

: Vn W n1 ,
(n + c0 n1 ) = n1 .
(3.12)
The problem is that, although is injective, the projection is not in general surjectiveif Z
is not vanishing. For this reason we introduce the image of Vn by the map and denote it by
W n1 :
n1 W n1 n1 W n1 ,
n1 = L 0 n ,
Z
def
n n
(3.13)
W n is in general a subspace of W n which reduces to the latter when Z vanishes.

Therefore, by construction, the following is an exact short sequence
0 W n Vn W n1 0.
(3.14)
= M and M
= Q.
Also, it is easily verified that
Moreover Q
M : W n W n+1
(3.15)
by virtue of the nilpotency relations (3.8).

In conclusion, the following diagram is (anti)-commutative
W n
W n+1
Vn
W n1
Vn+1
(3.16)
W n
From this diagram, one obtains (see, for example, [13]), along the usual lines, the following
exact long sequence of Q-cohomologies

D
h (n1) h (n+1) .
h (n) H(n) (Q)
(3.17)
166
In this exact sequence a new kind of relative cohomology appears, h (n) , to which we will refer
as the check relative cohomology. This is defined as the cohomology of M on W n :
W n ).
h (n) = H(n) (M,
(3.18)
The map D is known in homology theory as the connecting map and it is defined as follows.
Let vn1 be an element of W n1 . Thus, there exists n Fn such that
n1 = L 0 n .
Zv
(3.19)
Therefore n + c0 vn1 Vn and

(n + c0 vn1 ) = vn1 .
(3.20)
We define
n1 ) Dv
n1 + M
n.
D(v
(3.21)
The commutativity of the diagram (3.16) and the nilpotency relations (3.8) ensure that D descends to a cohomology map.
Then
Indeed, suppose vn1 W n1 is in the kernel of M.
n1 ) = M Dv
n1 + M 2 n = D(
M + Z)v
n1 D L 0 n
M D(v
M + Z)v
n1 D Zv
n1 = 0
= D(
and
n1 ) = L 0 Dv
Zv
n1 = D L 0 vn1 Z Mv
n1 = 0.
n1 + (M + Z)
L 0 D(v
Therefore D maps the kernel of M on W n1 to kernel of M on W n+1 .
Suppose now that vn1 is trivial in check cohomology:
n2 ,
vn1 = Mv
L 0 vn2 = 0,
n2 = L 0 n1 .
Zv
(3.22)
Hence
n1 ) = D Mv
n2 + M
n = M Dv
n2 D Zv
n2 + M
n
D(v
n2 D L 0 n1 + M
n = M(
Dv
n2 + M
n1 + n ).
= M Dv
Moreover
n2 + M
n1 + n ) = L 0 Dv
Zv
n2 + Zv
n1
n2 + (M + Z)
L 0 (Dv
Zv
n2 + Zv
n1
n2 + (M + Z)
= L 0 Dv
n2 + Z Mv
n2 = 0.
= D L 0 vn2 Z Mv
Thus D maps trivial states to trivial states. Hence, Eq. (3.21) defines a map between h (n1) and
h (n+1) .
We have observed that in the perturbative case Z = 0 and therefore h (n) = h (n) . In this case,
therefore, the sequence (3.17) allows determining the absolute cohomologies by means of the
relative ones. On the other hand, if Z = 0, the knowledge of both h (n) and h (n) is needed, in
general, for the computation of the absolute cohomologies by means of the exact sequence. One
can however derive few general relations connecting tilde and check relative cohomologies.
One such relations is the following: define the tilde relative index,

index h =
(1)n dim h (n)
167
(3.23)
and the check relative index

index h =
(1)n dim h (n) .
(3.24)
Then, the duality between absolute cohomologies,

H(1n) (Q)
H(n) (Q)
(3.25)
together with the sequence (3.17) leads to the identity of the relative cohomology indices:
index h = index h.
(3.26)
Further relations between tilde and check cohomologies are somewhat obvious and yet useful
(n)
inequalities which rest on mathematical properties of the operators L 0 that one can assume on
physical grounds. Let us list such properties:
(n)
(a) For the Siegel gauge to be a good gauge, the kernels of L 0 i.e. W n must vanish for
2
p generic. This is equivalent to the requirement that propagators be well-defined after gaugefixing.
(b) It is also physically reasonable to assume that the dimensions of the kernels W n at a
given discrete value of p 2 for which they are not emptyremain finite. This amounts to say that
we expect a finite number of fields of a given mass.
(c) Last, we should assume that at a given value of p2 there is only a finite number of W n
with different ghost number n that are non-empty: in other words, for a given p 2 , there exists
a maximal ghost number g > 0 such that W n = 0 for |n| > g. This assumption is essential to
give a mathematical precise meaning to the FadeevPopov index (4.7) that counts the number
of physical states. More generally, this assumption gives mathematical sense to the BRS gaugefixing construction for OSFT which, as we have seen, involves an infinite number of ghost fields
generations.
All these three conditions are obviously verified in the level truncated theory, for fixed level
L. The validity of LT as a computational scheme of OSFT is based on the assumption that these
properties are stable as L . This means that for a given interval of p2 there should exist
a level L such that for levels L > L the dimensions of the W n do not jump even if the values of
p 2 at which non-trivial W n s appear move a bit. To state it a little more precisely: given p 2 , if
then for any L > L there should exist a p 2 for which
dim W n (pL ) = 0 for pL2 > p 2 and L > L,
L
dim W n (pL ) = dim W n (pL ) and |pL2 pL2 | 0 as both L and L go to infinity.
Let us remark that we are not assuming uniform convergence on the p 2 axis: L may well
depend on p 2 and, indeed, our numerical computations suggest that it grows linearly as p 2
.
We have no formal proof of this stability property of level truncation, although our numerical computation are consistent with it. On the other hand, if level truncation did not enjoy
this property its use in OSFT would have in general no justification, putting aside the specific
problem we are considering.
Let us now come back to the inequalities between dimensions of relative cohomologies that
one can prove assuming (a)(c) in the exact theory. Let g > 0 be the maximal ghost number, such
168
that W n = 0 for |n| > g, as specified in (c). The image of M in W g vanishes, since W g1 = 0.
Therefore h (g) reduces to the kernel of M on W g while h (g) is the kernel of M restricted to
W g W g . Therefore
dim h (g) dim h (g) .
(3.27)
An analogous inequality is derived as follows. The kernel of M at ghost number g consists of the
whole W g , since W g+1 = 0. Given any vector vg in W g we can decompose it as follows
g = L 0 3 + vg+1
Zv
(3.28)
where 3 3 and vg is an element of the cokernel coker L 0
(g+1)
of L 0 on 3
(g+1)
(g+1)
vg coker L 0
3 / img L 0
.
By hypothesis W g+1 = ker L 0
(g+1)
= 0: therefore
(g+1)
(g+1)
= dim ker L 0
= 0.
dim coker L 0
In other words,
vg+1
(3.29)
(3.30)
= 0 in Eq. (3.28) above and
W g = W g = ker M (g) .
(3.31)
On the other hand, W g1 W g1 and thus the image of W g1 via M is contained in the image
of W g1 . We conclude that
dim h (g) dim h (g) .
(3.32)
4. The numerical computation

The field spaces n can be decomposed as direct sum of spaces with fixed spacetime momentum p , = 0, 1, . . . , 25:

n =
(4.1)
n (p).
p
Because of translation invariance the kinetic operator L 0 is diagonal with respect to this decom(n)
(n)
position. For each space n (p) choose a basis {ein (p)}. Let us denote by L 0 (p) the matrix
representing in this basis the operator L 0 acting on n (p). Let G(n) (p) be the square matrix
whose elements are given by
(n)
(n)

(n)
G (p) i j = ein (p), ejn (p) .
(4.2)
n n
For n > 0 the symmetric square matrix that specifies the kinetic operator for the fields (n , n )
is

(n)
0
G(n) (p)L 0 (p)
1
(n)
C
(4.3)
(p)
.
2 G(n) (p)L (n) (p)
0
0
For the matter string field 0 the kinetic quadratic form is instead
C (0) G(0) (p)L (0)
0 (p).
(4.4)
The determinants of the kinetic operators

(n)
(n) p 2 det L 0 (p)
169
(4.5)
are functions of p 2 . The zeros of such determinants encode the information about physical states
of OSFT. Suppose that

d

(n) p 2 = (n) p 2 = an p 2 + m2 n 1 + O p2 + m2
(4.6)
where the first equality is a consequence of the symmetry property (2.16) of L 0 . Then, the number
of physical states of mass m is given by the index:
IFP (m) = d0 2d1 + 2d2 + =
(1)n dn .
(4.7)
n=
This is so since the ghost and antighost pairs (n , n ) are complex fields of Grassmanian parity
(1)n . The numbers dn are in general gauge-dependentin our case they capture properties of
the b0 -invariant spaces n . The index IFP (m) is gauge-invariant and coincides with the dimen of Q
on the total space of (non-b0 -invariant) states of ghost
sion of the cohomology H(0) (Q)
number 0. In a physically sensible theory IFP (m) must be non-negative. Sens conjecture is that
IFP vanishes for all m.
Typically, in the exact (not level truncated) theory, (n) (p 2 )s with different ghost numbers n
L 0 ] = 0; so, if
vanish at the same value of p2 , as a consequence of BRS invariance. Indeed [Q,
(n)
2
2
2
(p ) vanishes for some p = m , then there exists a n such that
n ).
L 0 n = 0 = b0 (Q
(4.8)
Therefore
L 0 (n+1 ) = 0 = b0 n+1 ,
(4.9)
n . If n+1 does not vanish,

m2
where n+1 = Q
(n)
2
are associated to a multiplet of determinants (p ) with different ns that vanish simultaneously at p 2 = m2 .
Since level truncation breaks BRS invariance we expect that the zeros of the determinants in
the same multiplet, when evaluated at finite L, would be only approximately coincident. Thus
using the index formula (4.7) to compute the number of physical states is meaningful when the
splitting between approximately coincident determinant zeros is significantly smaller than the
distance between the masses of different multiplets.
(n)
In the theory truncated at level L, the operators L 0 (p) reduce to finite dimensional matrices;
(n)
moreover for a given L, the L 0 (p) vanish identically for n greater than a certain nL which
scalar (p) and vector (p),
depends on the level.5 We evaluated the LT matrices L (n)
n
0 (p) on both n
the subspaces of n (p) containing the states which are either scalars or vectors with respect to
spacetime Lorentz symmetry.
commutes with the twist
The computation is simplified by noting that the non-perturbative Q
N
parity operator (1) . Therefore the kinetic operators decompose as follows
(n+1) (m2 ) = 0. Thus physical states of mass
(n)
(n,+)
(n,)
L 0 (p) = L 0 (p) L 0 (p)
5 n is the greatest integer which satisfies the inequality n (n + 1)/2 L.
L
L L
(4.10)
170
where L 0 (p) are the kinetic operators acting on the subspaces n (p) of n (p) with twist
parity .
Another symmetry of L 0 is the SU(1, 1) symmetry generated by:
(n,)
()
J+ = {Q, c0 } =
ncn cn ,
n=1
J =

1
n=1
bn bn ,
J3 =
1
(cn bn bn cn ),
2
(4.11)
n=1
J+ and J3 are derivatives of the -product [18]. They obviously commute both with b0 and the
perturbative L0 and hence they are a symmetry of the OSFT equations of motion in the Siegel
gauge:
L0 + b0 ( ) = 0.
(4.12)
The tachyon solution turns out to be a singlet of the SU(1, 1) algebra: it follows that J+ and J3
commute with L 0 since

L 0 = L0 + b0 , [ , ] .
(4.13)
Thus the multiplets of determinants (n) (p 2 ) that vanish at a given p 2 = m2 organize themselves into representations of SU(1, 1). The symmetry (4.11) is not broken by LT since its
generators commute with the level: therefore the SU(1, 1) symmetry of the multiplets of vanishing determinants (n) (m2 ) is exact even at finite L. Because of the SU(1, 1) symmetry, the
FadeevPopov formula for the number of physical states of mass m rewrites in Siegel gauge as
follows

IFP (m) =
(4.14)
(1)2J (2J + 1)dJ ,
J
where the sum is over the SU(1, 1) spin J of the representations formed by the zeros of the
determinants of the kinetic operators at p 2 = m2 and dJ are their associated exponents.
(n)
We computed numerically the matrices L 0 (p) as functions of p in the theory truncated at
various levels L, from L = 4 up to L = 10.6
For L 10 the subspaces nscalar (p) (nvector (p)) are non-empty for |n| 4 (|n| 3). The
dimensions of the matrices L (n,+)
(p) (L (n,)
(p)) for scalars and vectors at even (odd) levels are
0
0
listed in Tables 1 and 2.
We looked for zeros of the determinants
(n)
(n,)
p 2 det L 0 (p)
(4.15)
in the scalar and vector sector. The zeros of the determinants for p 2 > 10 are plotted in Fig. 1.
We found that the determinants have zeros only on the negative p 2 axis, corresponding to
physical (positive) values of m2 = p 2 . The first zeros (on the negative p 2 axis, closest to the
origin) of the scalar even determinants are located around p 2 = 6 (graph (a) of Fig. 1). However
up to level 10 they are not stable yet. Their number keeps jumping as one increases the level:
6 We adopted the approximation that, in the terminology of [6], is of type (L, 3L). We verified that the approximation
of type (L, 2L) is not satisfactory for this problem.
171
Table 1
Number of b0 -invariant scalar states at up to level 10
Level
ghost # 0
ghost # 1
ghost # 2
ghost # 3
ghost # 4
3 (odd)
4 (even)
5 (odd)
6 (even)
7 (odd)
8 (even)
9 (odd)
10 (even)
9
24
45
99
183
363
655
1216
6
13
30
61
125
240
458
841
1
2
7
14
35
68
145
272
0
0
0
1
2
7
15
36
0
0
0
0
0
0
0
1
Table 2
Number of b0 -invariant vector states up to level 10
Level
ghost # 0
ghost # 1
ghost # 2
ghost # 3
3 (odd)
4 (even)
5 (odd)
6 (even)
7 (odd)
8 (even)
9 (odd)
10 (even)
7
16
40
85
184
367
730
1385
3
9
22
52
113
238
478
936
0
1
3
10
24
59
127
272
0
0
0
0
1
3
10
25
no multiple structure is detectable up to this level. We cannot therefore draw any conclusions
about the fate of the Sens conjecture in the even scalar sector.
For scalars in the odd sector and vectors in both odd and even sectors there exists a first group
of zeros on the negative p 2 axis which are closest to p 2 = 0 and well separated from other zeros
located at more negatives values of p 2 (graphs (b), (c), (d) of Fig. 1). These groups of almost
degenerate zeros become stable starting with level L = 4 or L = 5. As the level increases these
zeros move on the p 2 axis but their number does not jump. The almost degenerate zeros of the
scalar odd determinant are located around p 2 = 2; those of the vector even determinant around
p 2 = 4; and those of vector odd determinant around p 2 = 6. In all these three cases, the
almost degenerate zeros form a reducible representation of the SU(1, 1) which is the sum of a
scalar with J = 0, two doublets with J = 1/2 and a vector with J = 1. The associated Fadeev
Popov index vanishes

(1)2J (2J + 1)dJ = 0.
(4.16)
J =0,1/2,1
The observation which is important for our analysis is the following: the two zeros with J = 1/2
do correspond to a single eigenvalue of the kinetic operator with a zero of order two. This seems
to be unequivocal looking at the graphs of the vanishing eigenvalue that is reported, for level 10
or 9 in Fig. 2. The conclusion is that the two doublets with J = 1/2 should correspond in the
exact theory to a single zero with dJ = 2.
The almost degenerate zeros of the determinants for the scalar odd, vector even and vector
odd sectors, are plotted, with the corresponding levels, in Figs. 35. In the same plots we also
show the linear fits of the zeros locations as function of the inverse of the level, 1/L, for the
different SU(1, 1) spins J .
172
(n)
(n)
(n)
(n)
Fig. 1. Zeros of scalar + (p2 ) (a), scalar (p2 ) (b), vector + (p2 ) (c),vector (p2 ) (d) at levels L = 4, . . . , 9
up to p 2 = 10.
Table 3
Determinant zeros extrapolated at L =
Sector
J =0
J = 1/2
J =1
Scalar odd
Vector even
Vector odd
1.99172
3.98938
5.97751
2.03279; 1.97541
3.99494; 3.99087
5.96576; 6.00275
2.04905
3.98803
5.78701
The linearly extrapolated values of the zeros of the determinants for spins J = 0, 1/2, 1 in the
various Lorentz/twist-parity sectors are listed in Table 3.7
Extrapolated zeros with different J agree with remarkable accuracy. It is very tempting to
conjecture from these data that the exact values for the degenerate zeros in the corresponding
7 The linear fits in Figs. 35 have been performed by excluding the values of the zeros with lowest level, for which one
can expect the corrections to the linear dependence in 1/L are largest. Including these zeros in the fits does not change
the extrapolated values in Table 3 significantly: it worsen slightly the convergence between zeros with different ghost
number.
173
(b)
(a)
(c)
Fig. 2. The vanishing eigenvalue of the kinetic operator in Eq. (4.3) for ghost number 1, at level 10 (or 9) in the scalar
odd sector (a), vector even (b) and vector odd (c).
(n)
Fig. 3. The first group of zeros of (p2 ) in the scalar odd sector at p2 2.0 for levels L = 3, 5, 7, 9. J = 0
dot-dashedred, J = 1/2 solidgreen, J = 1 dashedblue. (For interpretation of the references to colour in this
figure legend, the reader is referred to the web version of this article.)
sectors are
m2scalar, = 2.0,
m2vector,+ = 4.0,
m2vector, = 6.0.
on the zeros of L
0
5. The action of Q
(n)
We found, numerically, that the kernels W n of L 0 corresponding to multiplets of approximately degenerate propagators poles, form a singlet, a doublet and a triplet of the SU(1, 1)
174
(n)
Fig. 4. The first group of zeros of + (p2 ) in the vector even sector at p 2 4.0 for levels L = 4, 6, 8, 10. J = 0
dot-dashedred, J = 1/2 solidgreen, J = 1 dashedblue. (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)
(n)
Fig. 5. The first group of zeros of (p2 ) in the vector odd sector at p 2 6.0 for levels L = 5, 7, 9. J = 0
dot-dashedred, J = 1/2 solidgreen, J = 1 dashedblue. (For interpretation of the references to colour in this
figure legend, the reader is referred to the web version of this article.)
symmetry, in all sectors listed in Table 3. Let us denote by v2 and v1 the vectors that generate, respectively, the kernels W 2 and W 1 . Let v0s and v0t be the vectors in W 0 that belong to,
respectively, the singlet and the triplet of SU(1, 1). One has
J+ v0s = 0,
v0t = J+ v2 ,
v1 = J+ v1 ,
v2 = J+ v0t ,
J+ v2 = 0,
J+ v1 = 0.
(5.1)
It follows that it commutes with M,

L 0 , D,
Z and Z.
J+ commutes both with L 0 and with Q.
Therefore J+ maps not only Wn into Wn+2 but also Wn into Wn+2 .
The goal of this section is to evaluate the dimensions of the tilde and check relative cohomologies
n (n) dim h (n) ,
n (n) dim h (n) .
(5.2)
As explained above, our numerical computations indicate that the dimensions of the kernels W n
of L (n)
0 in the exact theory are
dim W 0 = 2,
dim W 1 = 1,
dim W 2 = 1,
dim W n = 0 for n 3.
(5.3)
p2 ,
the relative indices are

Therefore, for the same values of

(1)n dim W n = 2
index h = index h =
n
(5.4)
175
while the FadeevPopov index vanishes and

= H(1) (Q)
=0
H(0) (Q)
(5.5)
= H(1) (Q)
= 0 the long sequence (3.17)
in agreement with Sens conjecture. When H(0) (Q)
breaks into the short exact sequence
D
0 h (1) h (1) 0
(5.6)
and into the two semi-infinite exact sequences
D

h (2) h (0) 0
h (3) h (1) H(1) (Q)
D

0 h (0) h (2) H(1) (Q)
h (1) h (3) .
(5.7)
From (5.6) one obtains

h (1) = h (1)
(5.8)
and this should hold for any p 2 if Sens conjecture is true.

Eq. (5.3) implies that the semi-infinite exact sequences (5.7) break up into finite sequences
D
h (2) h (0) 0,
0 h (1) H(1) (Q)
h (1) 0,
0 h (0) h (2) H(2) (Q)
0,
0 h (2) H(2) (Q)
D

0 H(3) (Q)
h (2) 0.
(5.9)
Hence, we obtain
= H(3) (Q)
= h (2) = h (2)
H(2) (Q)
(5.10)
from the last two sequences above, while the first two give
= n (1) + n (2) n (0) = dim H(2) (Q)
= n(0) + n (2) + n (1) .
dim H(1) (Q)
(5.11)
Eqs. (5.8), (5.10), and (5.11) establish the following relations between the dimensions of the
relative tilde and check cohomologies:
n (1) = n (1) ,
n (0) = 2 + n (1) n (2) + n (1) n (2) ,
n (2) = n (2) .
(5.12)
We now want to look for solutions of Eqs. (5.4)(5.11) with

n (0) , n (0) = 0, 1, 2,
n (1) , n (1) , n (2) , n (2) = 0, 1.
(5.13)
Moreover, relations (3.27) and (3.32) require that

n (2) n (2)
and
n (2) n (2) .
The most general action of M on the kernel of L 0 takes the form

2 = 0,
2 = v1 ,
Mv
Mv
1 = v s + v t ,
1 = v2 ,
Mv
Mv
0
(5.14)
176
s = v1 ,
Mv
0
t = MJ
+ v2 = v1 ,
Mv
0
(5.15)
where , , and are numbers. (Without loss of generality we can assume these numbers to
be real, since their phases can be reabsorbed into the normalizations of the vectors vn .) M 2 = 0
is equivalent to the relations
= = 0,
= = 0.
(5.16)
The possible solutions of (5.16) are

(a)
= = = = 0,
(b)
= 0,
= = 0,
(c)
= 0,
= 0,
(d)
= = = 0,
(, ) = (0, 0),
= 0.
Only (d), however, is compatible with the constraints that come from the sequence (5.9) and
Let us show why.
the fact that J+ commutes with Q.
Solution (a) leads to
n (2) = 1,
Since
n (0)
n (1)
n (1) = 1
n (0) = 2,
n (1) = 1,
n (2) = 1.
(5.17)
= 1 it follows from (5.11) that
n (2) = 1,
= 0.
dim H(1) (Q)
(5.18)
= 0 the first two sequences in (5.9) split farther and give

If H(1) (Q)
n (1) = n (1) = 0
(5.19)
in conflict with Eq. (5.17) above.

Solution (b) leads to the following values for the tilde cohomologies
n (2) = 0,
n (1) = 0,
n (0) = 1,
n (1) = 0,
n (2) = 1.
(5.20)
Again, n (0) n (1) = 1 and therefore, as in (5.18) n (2) = 1. But this is inconsistent with the
inequalities (5.14) which require n (2) n (2) = 0.
Two different sets of values for tilde cohomologies are a priori possible in case of solution (c):
n (2) = 1,
n (1) = 0,
n (0) = 1,
n (1) = n,
n (2) = n
(5.21)
n (0) n (1)
with n = 1 if = 0 and n = 0 if = 0. In both cases,

= 1, and therefore Eqs. (5.18)
and (5.19) hold as well. Therefore the values of the check cohomologies are
n (2) = 1,
n (1) = n,
n (0) = n,
n (1) = 0,
n (2) = 1.
(5.22)
However, n = 1 = n (1) requires = = 0, otherwise M would have no kernel on W 1 W 1 .

This reduces again to solution (a), which we have already ruled out. Therefore n = 0. Since
n (2) = 1, v2 W 2 . Thus v0t W 0 and v2 W 2 . n (0) = 0 dictates that v0t be trivial in check
cohomology. Therefore, = 0 and
1 = v t
Mv
0
(5.23)
177
where = 0 and v1 W 1 . As J+ sends W 1 to W 1 we conclude that

v1 = J+ v1 W 1 .
(5.24)
1 means that n (2) = 0, in conflict

We reached a contradiction: v2 W 2 , v1 W 1 and v2 = Mv
with (5.22).
We are left therefore with solution (d), for which
n (2) = 1,
n (1) = 1,
n (0) = 1,
n (1) = 0,
n (2) = 1.
(5.25)
The first sequence in (5.9) implies that n (2) does not vanishsince n (0) = 1. Thus n (2) = 1.
t = 0 and Mv
1 =, we conclude that
Therefore v2 W 2 and v0t = J+ v2 W 0 . Since Mv
0
(0)
n = 1. This, together with the general relations (5.12), determines all of the check cohomologies:
n (2) = 1,
n (1) = 0,
n (0) = 1,
n (1) = 1,
n (2) = 1.
(5.26)
To sum up, the action of the operators M on the kernel W n and the subspaces W n which are
compatible with our sequence are
2 = 0,
Mv
W 2 = {v2 },
1 = 0,
Mv
W 1 = 0,
t = 0,
Mv
0
W 1 = {v1 },
s = v1 ,
Mv
0

W 0 = v0t .
(5.27)
and H(2) (Q)

are v1 and v2 , respectively. The dual
Non-trivial representatives of H(1) (Q)
(2)
(3)
non-empty cohomologies H (Q) and H (Q) have representatives

2 = c0 v1 + 2
(5.28)
3 = c0 v2 + 3
(5.29)
and
respectively, where 2 and 3 are given by

1 = L 0 2 ,
Zv
2 = L 0 3 .
Zv
(5.30)
Note that 2 and 3 are defined by these equations up to elements in the kernels of L 0 on W 2
and W 3 . The latter is empty and the former is spanned by v2 : the exactness of the sequence (5.9)
cohomology.
ensures that v2 is trivial in the Q
Acknowledgements
I thank Sofia Mosci for collaborating on setting up the low-level numerical computations.
I thank Stefano Giusto and Martin Schnabl for simulating conversations. I am indebted to M. Beccaria for useful suggestions regarding computational algorithms. I gratefully acknowledge the
hospitality of the Theory Group of CERN, where part of this work was done. This work is supported in part by Ministero dellUniversit e della Ricerca Scientifica e Tecnologica.
178
References
[1] S. Giusto, C. Imbimbo, Physical states at the tachyonic vacuum of open string field theory, Nucl. Phys. B 677 (2004)
52, hep-th/0309164.
[2] A. Sen, Descent relations among bosonic D-branes, Int. J. Mod. Phys. A 14 (1999) 4061, hep-th/9902105.
[3] A. Sen, Universality of the tachyon potential, JHEP 9912 (1999) 027, hep-th/9911116.
[4] E. Witten, Noncommutative geometry and string field theory, Nucl. Phys. B 268 (1986) 253.
[5] V.A. Kostelecky, S. Samuel, On a nonperturbative vacuum for the open bosonic string, Nucl. Phys. B 336 (1990)
263.
[6] A. Sen, B. Zwiebach, Tachyon condensation in string field theory, JHEP 0003 (2000) 002, hep-th/9912249.
[7] N. Moeller, W. Taylor, Level truncation and the tachyon in open bosonic string field theory, Nucl. Phys. B 583
(2000) 105, hep-th/0002237.
[8] D. Gaiotto, L. Rastelli, Experimental string field theory, hep-th/0211012.
[9] M. Schnabl, Analytic solution for tachyon condensation in open string field theory, hep-th/0511286.
[10] L. Rastelli, A. Sen, B. Zwiebach, String field theory around the tachyon vacuum, Adv. Theor. Math. Phys. 5 (2002)
353, hep-th/0012251.
[11] I. Ellwood, B. Feng, Y.H. He, N. Moeller, The identity string field and the tachyon vacuum, JHEP 0107 (2001) 016,
hep-th/0105024.
[12] M. Bochicchio, Gauge fixing for the field theory of the bosonic string, Phys. Lett. B 193 (1987) 31;
C.B. Thorn, Perturbation theory for quantized string fields, Nucl. Phys. B 287 (1987) 61.
[13] R. Bott, L.W. Tu, Differential forms in Algebraic Topology, Springer-Verlag, New York, 1997.
[14] I. Ellwood, M. Schnabl, Proof of vanishing cohomology at the tachyon vacuum, hep-th/0606142.
[15] M. Schnabl, Wedge states in string field theory, JHEP 0301 (2003) 004.
[16] L. Rastelli, B. Zwiebach, Tachyon potentials, star products and universality, JHEP 0109 (2001) 038.
[17] I. Kishimoto, K. Ohmori, CFT description of identity string field: Toward derivation of the VSFT action, JHEP 0205
(2002) 036.
[18] B. Zwiebach, Trimming the tachyon string field with SU(1, 1), hep-th/0010190.
[19] I. Kishimoto, T. Takahashi, Open string field theory around universal solutions, Prog. Theor. Phys. 108 (2002) 591,
hep-th/0205275.
Gauging N = 4 supersymmetric mechanics II:

(1, 4, 3) models from the (4, 4, 0) ones
F. Delduc a, , E. Ivanov b
a ENS Lyon, Laboratoire de Physique, 46, alle dItalie, 69364 Lyon cedex 07, France
b Bogoliubov Laboratory of Theoretical Physics, JINR, 141980 Dubna, Moscow Region, Russia

Abstract
Exploiting the gauging procedure developed by us in hep-th/0605211, we study the relationships between
the models of N = 4 mechanics based on the off-shell multiplets (4, 4, 0) and (1, 4, 3). We make use of the
off-shell N = 4, d = 1 harmonic superspace approach as most adequate for treating this circle of problems.
We show that the most general sigma-model type superfield action of the multiplet (1, 4, 3) can be obtained
in a few non-equivalent ways from the (4, 4, 0) actions invariant under certain three-parameter symmetries,
through gauging the latter by the appropriate non-propagating gauge multiplets. We discuss in detail the
gauging of both the PauliGrsey SU(2) symmetry and the Abelian three-generator shift symmetry. We
reveal the (4, 4, 0) origin of the known mechanisms of generating potential terms for the multiplet (1, 4, 3),
as well as of its superconformal properties. A new description of this multiplet in terms of unconstrained
harmonic analytic gauge superfield is proposed. It suggests, in particular, a novel mechanism of generating
the (1, 4, 3) potential terms via coupling to the fermionic off-shell N = 4 multiplet (0, 4, 4).
PACS: 11.30.Pb; 11.15.-q; 11.10.Kk; 03.65.-w
Keywords: Supersymmetry; Gauging; Isometry; Superfield
1. Introduction
An extended d = 1 supersymmetry possesses some notable features which are not shared by
its higher-dimensional counterparts. In view of the distinguished role of supersymmetric quan* Corresponding author.
E-mail addresses: francois.delduc@ens-lyon.fr (F. Delduc), eivanov@theor.jinr.ru (E. Ivanov).

doi:10.1016/j.nuclphysb.2007.02.001
180
F. Delduc, E. Ivanov / Nuclear Physics B 770 [FS] (2007) 179205
tum mechanics, both as the appropriate simplified laboratory for studying various aspects of
supersymmetric quantum field theories and as a theory providing superextensions of some intrinsically one-dimensional systems, it is of importance to fully understand these specific features
of the d = 1 supersymmetry and their dynamical manifestations in the corresponding models
of supersymmetric mechanics. One of these peculiarities is the so-called 1D automorphic duality [13] which relates off-shell d = 1 supermultiplets with the same number of physical
fermions and different divisions of the set of bosonic fields into physical and auxiliary components (see also [46]). The procedure generating the multiplets with a greater number of
auxiliary fields from those with the lesser number and the procedure inverse to it can be referred to as the reduction and oxidation, respectively [7].1 The non-linear versions of this
duality were considered in [810]. Using these dualities, the relations between different supersymmetric mechanics models can be studied. In particular, it was argued in [10] that various
models of N = 4 supersymmetric mechanics based on the off-shell multiplets with 4 physical
fermions can be obtained, through the reduction procedure, from the models based on the root
[12] off-shell multiplet with the field content (4, 4, 0) [1,8,9,11,12] (hereafter, the abbreviating
(n1 , n2 = n1 + n3 , n3 ) stands for the off-shell multiplet with n1 physical bosons, n2 physical
fermions and n3 auxiliary bosonic fields).
In most of the above studies, the reduction procedure was accomplished at the component
level and by hands: basically by treating the time derivative of some initial physical bosonic
field as a new auxiliary field. Recently, we proposed a superfield version of this procedure ensuring the manifest off-shell supersymmetry at all steps [7]. The process of reduction was shown to
amount to gauging some isometries of the superfield actions of the multiplet with the maximal
number of the physical bosonic fields ((1, 1, 0), (2, 2, 0) and (4, 4, 0) in the N = 1, N = 2 and
N = 4 cases) by a topological gauge multiplet. The characteristic property of the latter (specific just for the d = 1 case) is that its only surviving component in the WessZumino gauge is
the bosonic gauge field. The residual gauge freedom is always realized in such a way that it
can be used to kill one or few original bosonic fields. After fully fixing the gauge freedom in this
way, one is left with the new off-shell multiplet in which the place of killed bosonic fields is
occupied by the former gauge fields which possess no kinetic terms and so are auxiliary. Thus the
supersymmetric mechanics of d = 1 supermultiplets with one or another numbers of auxiliary
fields naturally arises upon fixing a gauge in a coupled system of the extreme multiplet having
no auxiliary fields at all and a topological supermultiplet which gauges one or another isometry
realized on the extreme multiplet. Besides choosing the WessZumino gauge (which basically
corresponds to the component consideration of Refs. [16,10]) one is free to choose another,
manifestly supersymmetric gauge in which the whole reduction procedure can be performed in
terms of superfields. This possibility to accomplish the reduction in a manifestly supersymmetric superfield fashion is one of the merits of the gauging approach. It can be regarded as an
efficient tool of deducing off-shell superfield descriptions of d = 1 supersymmetric mechanics
systems, starting from the system associated with the basic (root) multiplet. In most cases, the
potential terms of the resulting multiplet are generated by FayetIliopoulos terms of the gauge
superfield and/or the gauge-covariantized WessZumino-like terms of the root multiplet. The
inverse oxidation procedure amounts to constraining the gauge multiplet to be pure gauge
and so eliminating it altogether.
1 In an obvious analogy with the nomenclature used in the context of the spacetime dimensional reduction.
181
In [7] we concentrated on the N = 4 mechanics and the superfield reduction from the N = 4
root multiplet (4, 4, 0) to the multiplet (3, 4, 1) (both its linear [1315] and non-linear [8,9]
versions). As a by-product we constructed a new non-linear superfield version of the multiplet
(4, 4, 0) in N = 4, d = 1 harmonic superspace (see also [11,1618]). Some further examples of
our gauging procedure leading to the N = 4 multiplets (0, 4, 4) and (1, 4, 3) were also considered. The last case corresponds to gauging non-Abelian SU(2) symmetry realized on the
multiplet (4, 4, 0).
In [7] we limited our study to a particular case of such gauging, with the free action of the multiplet (4, 4, 0) as the input. In the present paper we consider the most general situation. We start
from the most general SU(2) invariant (4, 4, 0) action in the harmonic N = 4, d = 1 superspace
and show that its SU(2) gauging generates the generic sigma-model type superfield action of
the multiplet (1, 4, 3) [19,20]. Also, we show that the latter can be equally reproduced by gauging some other three-parameter isometries admitting a realization on the multiplet (4, 4, 0). As
such one can choose three commuting shift isometries. We discuss the (4, 4, 0) origin of various
mechanisms of generating potential terms of the multiplet (1, 4, 3) and show how the superconformally invariant actions of the latter can be reproduced within the gauging procedure from the
superconformal (4, 4, 0) actions. As a by-product we find out a new description of the multiplet
(1, 4, 3) in terms of unconstrained harmonic analytic gauge prepotential. This description suggests a new mechanism of generating potential terms of the multiplet (1, 4, 3) via coupling it
to the off-shell fermionic N = 4 multiplet (0, 4, 4). Also, using this formulation, off-shell couplings of the multiplet (1, 4, 3) to the (3, 4, 1) one and some extra (4, 4, 0) multiplets can be
easily constructed. Finally, we discuss the description of the mirror (1, 4, 3) multiplet (with a
different SU(2) assignment of the component fields) in the N = 4, d = 1 harmonic superspace
and present a simple coupling of the mirror multiplet to the initial (1, 4, 3) multiplet.
2. N = 4, d = 1 harmonic superspace
2.1. Basics
Because the N = 4, d = 1 harmonic superspace (HSS) plays the central role in our construction, we start by recollecting the basics of this approach following Refs. [21,22] and [7,8].
The N = 4, d = 1 superspace is defined as the following coordinate set

z = t, i , i , i = (i ).
(2.1)
The covariant spinor derivatives are defined as
Di =
+ i i t ,
i

i
D , D j = 2iji t ,

D i = i + ii t = D i ,

i j
D , D = {D i , D j } = 0.
(2.2)
The N = 4, d = 1 harmonic superspace (HSS) in the central basis is the following extension of
(2.1)

(z, u) = t, i , i , u
(2.3)
i .
Here u
i SU(2)A /U (1) are the SU(2)A harmonic variables:

+
+i ,
u
u+i u
u+
i = u
i =1
i uk uk ui = ik .
(2.4)
182
The coordinates of N = 4, d = 1 HSS in the analytic basis are

tA = t i + + + , = i u , = i u , u .
(2.5)
The analytic subspace of HSS is defined as

tA , + , + , u
i (, u).
(2.6)
It is closed under the N = 4 supersymmetry

i +
+
tA = 2i i u
,
+ = i u+
i ui
i ,
+ = i u+
i ,
u
i = 0, (2.7)
and is real with respect to the generalized conjugation [21]

= ,

t
A = tA ,
= ,

i
u
i =u ,

i = u .
u
i
(2.8)
In the central basis (z, ui ), the harmonic derivatives and the harmonic projections of spinor
derivatives are defined by
D = = u
i
,
u
i
i
D = u
i D ,
i
D = u
i D .
(2.9)
In the analytic basis, the same spinor and harmonic derivatives read
,
D + = ,
D = + + 2i tA ,

D+ =
D =
+ 2i tA ,
+
+ + ,

D = 2i tA + + + + .
(2.10)

They satisfy the following non-zero (anti)commutation relations

+

= D , D + = 2itA ,
D , D = D ,
D , D
D , D = D,
0
++
= D0,
= 2D ,
D ,D
D ,D
(2.11)
D 0 = u+
(2.12)
u
+ + + + + + .
i
i

u+
u
D ++ = ++ 2i + + tA + +
D+ ,
D +
The derivatives
are short in the analytic basis, whence it follows that one can define
analytic N = 4 superfields (q) (, u)
D + (q) = D + (q) = 0
(q) = (q) (, u),
(2.13)
where q is the external harmonic U (1) charge. This Grassmann harmonic analyticity is preserved
by the harmonic derivative D ++ : when applied to (q) (, u), this derivative yields an analytic
N = 4, d = 1 superfield of charge (q + 2).
The measures of integration over the full HSS and its analytic subspace are given, respectively,
by

(2)
H = du dt d 4 = du dtA D D D + D + = A D + D + ,

(2)
A = du d (2) = du dtA d + d + = du dtA D D .
(2.14)
The integration over the harmonic two-sphere S 2 = {u+

i , uk } is normalized so that

du 1 = 1,
183
(2.15)
and the integral of any other irreducible monomial of the harmonics is vanishing [21,22].
2.2. Multiplet q +a and its symmetries
In this paper we shall deal with the multiplet (4, 4, 0) which is described by a doublet analytic
superfield q +a (, u) of charge 1 satisfying the harmonic constraint2
D ++ q +a = 0
+ a
+ a
q +a (, u) = f ia (t)u+
i + (t) + (t)
+ 2i + + t f ia (t)u .
i
It satisfies the pseudoreality condition (see (2.8))

+a = q +a
f ia = ab ik f kb ,
q

a = a .
(2.16)
(2.17)
The Grassmann analyticity conditions together with the harmonic constraint (2.16) imply that
in the central basis
q +a = q ia (t, , )u+
i ,
D (i q k)a = D (i q k)a = 0.
We may write a general off-shell action for the (4, 4, 0) multiplet as

q a D q +a .
Sq = du dt d 4 L q +a , q b , u ,
(2.18)
(2.19)
After solving the constraint (2.16) in the central basis of HSS, the superfield q a may be written
ia
(i k)a = D
(i q k)a = 0. We then use the notation
in this basis as q a = u
i q (t, , ), D q

+a b

ia
Sq = dt d 4 L q ia .
L q = du L q , q , u ,
(2.20)
The free action is given by

i

1
du dt d 4 q +a qa =
du d (2) q +a t qa+ .
Sqfree =
4
2
(2.21)
The action (2.19) produces a sigma-model type action in components, with two time derivatives on the bosonic fields and one derivative on the fermions. One can also construct an invariant
which in components yields a WessZumino type action, with one time derivative on the bosonic
fields (plus Yukawa-type fermionic terms). It is given by the following general integral over the
analytic subspace

WZ
Sq = du d (2) L+2 q +a , u .
(2.22)
The free action (2.21) and constraint (2.16) exhibit a number of symmetries. Some of them
can be extended to the interaction case, leading to certain restrictions on the form of the general
action (2.19). In terms of the component fields, these symmetries become isometries of the target
2 For brevity, in what follows we frequently omit the index A of t .
A
184
bosonic metric. We shall list here all symmetries of this sort, having in mind to gauge some of
them in the next sections. We will be interested only in those symmetries which commute with
N = 4 supersymmetry and so can be gauged without passing to the local supersymmetry [7].
Rotational isometries
The free action (2.21) and constraint (2.16) are manifestly invariant under the so-called Pauli
Grsey SU(2) group

su(2) q +a = a b q +b ,
(2.23)
a b = a b .
a a = 0,
An arbitrary one-parameter subgroup of this group is singled out as
1 q +a = 1 ca b q +b 1 I1 (c)q +a ,
I1 (c) = ca b q +b
,
q +a
(2.24)
where c(ab) is an isotriplet of constants

cab = cab .
ca a = 0,
(2.25)
One more isometry of this type is the target space dilatations
(2.26)
.
q +a
The constraint (2.16) is obviously covariant under these rescalings, while the action (2.21) is
not. The simplest invariant action is of the sigma model type [7]. The transformations (2.26)
and (2.23) commute with each other.
2 q +a = 2 q +a 2 I2 q +a ,
I2 = q +a
Shift isometries
The action (2.21) (up to boundary terms) and constraint (2.16) are also invariant under the
Abelian shifts
(a) 3 q +a = 3 u+a 3 I3 q +a , I3 = u+a +a ;

q
(b) 4 q +a = (a b) u+b .
(2.27)
An arbitrary one-parameter subgroup of (2.27b) is singled out as
5 q +a = 5 b(a d) u+d 5 I5 (b)q +a ,
I5 (b) = b(a d) u+d
,
q +a
(2.28)
where b(ab) is some real constant isotriplet (in general, it is different from c(ab) ). In what follows,
without loss of generality, we normalize all these isotriplets as in [7]
c2 = cab cab = 2,
b2 = bab bab = 2.
(2.29)
q +a ,
The rotational and shift isometries, being realized on the same

do not commute with
each other. Their closure is an extension of SU(2)PG by some solvable subgroups. Below we list
some subgroups with two and three generators from this closure.
Commuting subsets

GI = I1 (c), I2 ,
GIII = I5 (c), I5 (b) ,
GII = I3 , I5 (b) ,

I1 (c), I2 = I3 , I5 (b) = I5 (c), I5 (b) = 0.
(2.30)
(2.31)
185
Non-commuting subsets

GV = I2 , I5 (b) ,
GIV = {I2 , I3 },

I2 , I5 (b) = I5 (b),

GVI = I1 (c), I3 , I5 (c) ,

I1 (c), I5 (c) = I3 ,
[I3 , I2 ] = I3 ,
(2.32)
I1 (c), I3 = I5 (c),
(2.33)

GVII = I1 (c), I5 (b), I5 (d) ,
(c b) = (c d) = (b d) = 0,

ab
(a
f b)
I1 (c), I5 (b) = I5 (d),
d c fb ,

I1 (c), I5 (d) = I5 (b),

GVIII = I2 , I3 , I5 (c) ,
I2 , I5 (c) = I5 (c), (2.35)
[I2 , I3 ] = I3 ,

GIX = I2 , I5 (c), I5 (b) ,

I2 , I5 (b) = I5 (b).
(c b) = 0,
(2.34)
I2 , I5 (c) = I5 (c),
(2.36)
The subclasses of the general q-actions (2.19) and (2.22) revealing invariance under various
subgroups listed above will be defined in the next sections, as far as necessary. In fact, some of
the above subgroups coincide up to a redefinition of q +a by some constant matrix (preserving
the constraint (2.16) and the free action (2.21)). The full list of such isomorphisms is
GII GIII ,
GIV GV ,
GVI GVII ,
GVIII GIX .
(2.37)
For further use, we also present the full structure of the closure of the rotational and shift
isometries. Denoting, in the proper basis, the generators of SU(2)PG (2.23) by TM (M = 1, 2, 3),
the generator of target dilatations (2.26) by T and the generators of the singlet and triplet shifts
in (2.27) by F and FM , we have
[TM , TN ] = iMNK TK ,
[FM , FN ] = [FM , F ] = 0,
i
[TM , FN ] = MNK FK iMN F,
2
[T , TM ] = 0,
[T , FM ] = FM ,
i
[TM , F ] = FM ,
4
[T , F ] = F. (2.38)
This algebra generates a subgroup of the Weyl group in 4-dimensional target Euclidean space,
with F and FM forming the 4-translation operator and SU(2)PG being one of two SU(2) factors
of the rotation group SO(4) SU(2) SU(2).3
2.3. Topological gauge N = 4 superfield
The N = 4, d = 1 gauge multiplet is described by a charge 2 unconstrained analytic superfield V ++ (, u) the gauge transformation of which in the Abelian case reads
V ++ = D ++ ,
(2.39)
3 The second factor is the automorphism SU(2) acting on the doublet indices of harmonics ui , central basis Grassmann variables and the left index of the superfield q ia in (2.18). Since it does not commute with N = 4 supersymmetry,
it cannot be gauged without turning on the whole world-line N = 4 supergravity [7].
186
with (, u) being a charge zero unconstrained analytic superfield parameter. Using this gauge
freedom, one can choose the WessZumino gauge, in which the gauge superfield becomes

V ++ (, u) = 2i + + A(t),
A(t) = t 0 (t),
0 = (, u)|=0 .
(2.40)
We observe that the gauge N = 4, d = 1 multiplet locally carries (0 + 0) degrees of freedom

and so it is topological. Globally the field A(t) can differ from a pure gauge, and this feature
allows for its treatment as an auxiliary field in the unitary gauges.
As in the N = 2, d = 4 HSS [21,22], V ++ gauge-covariantizes the analyticity-preserving
harmonic derivative D ++ . Assume that the analytic superfield (q) is transformed under some
Abelian gauge isometry as
(q) = I (q) ,
(2.41)
where I is the corresponding generator. Then the harmonic derivative D ++ is covariantized as

D ++ (q)

D++ (q) = D ++ V ++ I (q) .
(2.42)
One can also define the second, non-analytic harmonic connection V

D = D V I,
V = D .
(2.43)
From the requirement of preserving the algebra of harmonic derivatives (2.11),

++
D ,D
= D0,
D 0 , D = 2D ,
(2.44)
the well-known harmonic zero-curvature equation follows

D ++ V D V ++ = 0.
(2.45)
It specifies V in terms of V ++ . One can also define the covariant spinor derivatives

D = D , D + = D + D + V I,

D = D , D + = D + D + V I,
(2.46)
as well as the covariant time derivative Dt :

D + , D = 2iDt ,
Dt = t
i + +
I.
D D V
2
(2.47)
The vector gauge connection

V D + D + V ,
V = 2itA ,
(2.48)
is an analytic superfield, D + V = D + V = 0, so Dt preserves the analyticity. In the WZ gauge

(2.40)
V
2iA(t).
We will exploit these relations and their non-Abelian generalization in next sections.
(2.49)
187
3. Multiplet (1, 4, 3) from gauging SU(2)PG

In the full set of rotational and shift symmetries realized on q +a there are six different subgroups with three generators: SU(2)PG defined in (2.23), the Abelian 3-parameter translation set
(2.27b) and four solvable subgroups GVI GIX defined by Eqs. (2.33)(2.36) (modulo isomorphisms (2.37)). In this paper we shall limit our study to the first two options4 and start with
gauging SU(2)PG . The simplest version of this gauging, with the free q +a action as the point of
departure, was already considered in [7].
3.1. From (4, 4, 0) to (1, 4, 3)
Let us gauge the SU(2)PG symmetry (2.23) by substituting a b a b (, u),
q +a = a b q +b .
(3.1)
The constraint (2.16) is covariantized to

++ q +a D ++ q +a V ++a b q +b = 0,
where the traceless analytic gauge connection V ++
(3.2)
a
is transformed as
V ++a b = D ++ a b + a c V ++c b V ++a c c b .
(3.3)
The analytic superspace form of the free action (2.21) is covariantized by replacing
t q +a
i
t q +a = t q +a V a b q +b ,
2
(3.4)
where
V a b = D + D + V a b ,
D
++
a
b
=D
(3.5)
++a
++a
b V
a
c
b + cV
b
b +V
a
c
V
c b.
cV
cV
++c
= 0,
(3.6)
(3.7)
The equivalent form of the action (2.21) in the central basis is covariantized by replacing
q a = D q +a
q a q +a = D q +a V a b q +b .
(3.8)
It is straightforward to check the identity of both forms of the covariantized free action.
Using the constraint (3.2) and the harmonic zero curvature condition (3.6), it is easy to check
that
++
,
(3.9)
= D0,
++ q a = q +a ,
q a = 0.
The subclass of general q + actions (2.19) enjoying gauge SU(2)PG symmetry is defined as
follows

Sgauge = dt d 4 du L q +a qa , u .
(3.10)
4 See the concluding Section 5 for some comments on the remaining options.
188
Taking into account the relations (3.2), (3.9), the SU(2)PG invariant
J q +a qa
(3.11)
is the only gauge invariant quantity which one can construct. Also, it is easy to show that
D ++ J = 0,
(3.12)
whence it follows that J does not depend on harmonics in the central basis. Therefore, without
loss of generality, one can neglect the harmonic integral in (3.10) together with the dependence
on the explicit harmonics in L(J, u ). Thus, the most general gauge invariant action is obtained,
via the replacement (3.8), from the most general globally SU(2)PG invariant action

+a
1 ia
4
4
SPG = dt d L q qa = dt d L q qia .
(3.13)
2
As was already shown in [7], the SU(2)PG gauging of the multiplet q +a gives rise to one sort of
the N = 4, d = 1 supermultiplet (1, 4, 3), in such a way that three physical bosonic components
of q +a become purely gauge while V ++(ab) supplies three auxiliary degrees of freedom. In [7]
this was demonstrated in the WZ gauge
Vba = 2i Aa b (t),
V ++a b = 2i + + Aa b (t),
V a b = D + D + V a b = 2iAa b (t),
(3.14)
r Aa b = t (0) a b + (0) a c Ac b Aa c (0) c b ,
(3.15)
q +a ,
f ia (t)
f ia u+
i
r f ia = (0) a c f ic ,
= q +a |=0 .
is the first component of

In this gauge, the solution of the
where
covariantized constraint (3.2) is obtained from the solution (2.16) just by the replacement
t f ia
Splitting
f ia
t f ia = t f ia + Aa b f ib .
(3.16)
as
1
f ia = ia f + f (ia) ,
(3.17)
2
and assuming that f has a non-vanishing constant vacuum part, f = f + , f = 0, one
observes that the symmetric part in (3.17) can be fully gauged away by the residual SU(2) gauge
freedom
1
f ia ia f.
(3.18)
2
So one ends up with the fields f (t), ia (t), A(ab) , which is just the off-shell field content of the
multiplet (1, 4, 3). Note that in this gauge the only manifest SU(2) symmetry is the diagonal one
in the product SU(2)A SU(2)PG . It plays the role of automorphism SU(2) group. As usual in
WZ gauge, N = 4 supersymmetry is not manifest, it should be accompanied by a field-dependent
gauge transformation to preserve the WZ gauge and the additional gauge (3.18).
Here we show how to arrive at the multiplet (1, 4, 3) while preserving manifest N = 4 supersymmetry.
To this end, we project all doublet SU(2)PG indices on the harmonics ui using the completeness relation (2.4)
q +a = u+a l ++ ua ,
++
= q +a u
= q +a u+
a, l
a,
(3.19)
189
V (ab) = u+a u+b V () + ua ub V (++) 2u+(a ub) V (+) ,
V () = V (ab) u
a ub ,
V () = V (ab) u
a ub ,
V (+) = V (ab) u+
a ub .
(3.20)
In terms of these projections, the transformation laws (3.1), (3.3) and (3.7) read
= + + l ++ ,
V
++(++)
++(+)
++()
(++)
(+)
()
=D
++
=D
++
=D
++
=D
=D
=D
++
++
l ++ = + l ++ ++ ,
++
++
++(+)
2
2
++
++
+ 2
++()
++
++()
(+)
()
()
+ 2
+ 2
+ 2
++(++)
(3.21)
++(++)
++(+)
(++)
(++)
(+)
(3.22)
,
(3.23)
while the constraint (3.2) takes the form

(a)
D ++ l ++ V ++(+) l ++ + V ++(++) = 0,
(b)
D ++ l ++ V ++() l ++ + V ++(+) = 0.
(3.24)
Assuming that = 0, we observe from the transformation law (3.21) that one can choose
the following manifestly N = 4 supersymmetric gauge
l ++ = 0
= 1,
q +a = u+a .
(3.25)
In this gauge, the constraints (3.24) imply

V ++(++) = V ++(+) = 0,
V ++() V = 0.
(3.26)
The residual gauge freedom is given by

++ = 0,
r V = D
++
+ = 0,
= 0,
(3.27)
(3.28)
r V (++) = 0,
r V (+) = + V (++) ,
r V () = D + 2 V (+) .
(3.29)
Here = (, u) is the only unconstrained residual analytic gauge parameter. The harmonic zero-curvature condition (3.6) is rewritten as
D ++ V (++) = 0,
D
++
++
(3.30)
(+)
(1 + V)V
()
(++)
+ V = 0,
V 2(1 + V)V
(+)
(3.31)
= 0.
(3.32)
These equations determine V (++) , V (+) and V () as functions of the analytic gauge
potential V.
Thus in the supersymmetric gauge (3.25) we are left with the analytic gauge superfield
V(, u), r V = D ++ , as the basic object encompassing the whole field content of the system consisting of the (4, 4, 0) multiplet and gauge SU(2)PG superfield. The general action (3.10)
190
takes the simple form

Sgauge = dt d 4 L(J ),
J = q +a qa+ = 1 V (++) ,
(3.33)
D ++ J = 0,
(3.34)
where we took into account Eq. (3.30). Using the harmonic independence of J in the central
basis, it is easy to find from (3.31) the expression of J in terms of V:

1
k +
u
, i u+
, W t, i , k du V t 2i i k u+
J=
(3.35)
i , uk , ul .
(i
k)
1+W
To reveal the field content carried by V, we should fully exploit the residual infinitedimensional gauge freedom (3.28). The full WZ form of V is easily found to be
+ + (ik)
+ i
(tA )u
V(, u) = v0 (tA ) + + i (tA )u
i + (tA )ui + 3 A
i uk ,
(3.36)
= 0. Thus we end up with the off-shell

without any further residual gauge freedom,
r
N = 4 supermultiplet (1, 4, 3) in the new formulation in terms of the analytic gauge prepotential
V(, u). The off-shell transformation properties of the component fields in (3.36) can be found
from the transformation law

V = i Qi + i Q i V + D ++
(3.37)
comp ,
where the first part is induced by the N = 4 supertranslations (2.7) (Qi = u+

i + + ,
i = u+i + + ), while the second part is the compensating gauge transformation needed
Q

to preserve the WZ gauge (3.36):

+ i
1 (i k)
+ i (kl) u u u .
(3.38)
+ (i k) u
i uk + A
i k l
2
The meaning of the N = 4 superfield W defined in (3.35) can be also easily understood. First
of all, by construction it is invariant under the gauge transformations (3.28), so one can always
choose WZ form (3.36) for V in (3.35), i.e. the field content of W is just (1, 4, 3). Secondly,
using the analyticity of V, D + V = D + V = 0, and the completeness relation (2.4), it is easy to
show that

D i D i W = 2 du D D + V = 0,
D i Di W = 2 du D D + V = 0,

i

D , D i W = 2 du D D + + D D + V = 0.
(3.39)
comp =
These are just the constraints which define the (1, 4, 3) multiplet in the ordinary N = 4 superspace [19,20].
Thus we have shown that the most general sigma-model type action of the N = 4 multiplet
(1, 4, 3) can be reproduced from the most general SU(2)PG invariant action of the multiplet
(4, 4, 0) by gauging the SU(2)PG symmetry using a topological gauge N = 4 supermultiplet.
For further use, let us note that, before imposing any gauge-fixing condition, one can use the
constraints (3.24) to covariantly express the gauge superfields V ++(+) and V ++(++) in terms
of V ++() , and l ++
V ++(+) =

1 ++
1 + V ++() D ++ ,
l
191

1 ++ 2
l
1 + V ++() D ++ l ++ .
(3.40)
2
Taking into account that the analytic superfields ( 1) and l ++ , in view of their inhomogeneous
transformation laws (3.21), can be treated as Goldstone superfields related to the spontaneous
breaking of local SU(2)PG symmetry down to its Abelian subgroup with the analytic parameter , Eqs. (3.40) supplies a nice example of the inverse Higgs phenomenon [23]. This
phenomenon, in particular, provides a possibility to covariantly express gauge fields associated
with the coset generators of the given non-linearly realized symmetry in terms of the Goldstone
fields and gauge fields belonging to the linear stability subgroup. Using the transformation laws
(3.21) and (3.22), it is easy to check that the above composite gauge superfields V ++(++)
and V ++(+) possess the correct gauge transformation properties. In the unitary gauge = 1,
l ++ = 0 these superfields vanish as it should be. Also, in accord with the general reasoning of
Ref. [23], one can construct a new gauge connection
V ++(++) =
1
V ++() = 2 1 2 + V ++()
which has the would-be Abelian SU(2)PG gauge transformation law

1
.
V ++() = D ++
2
(3.41)
(3.42)
We have chosen V ++() in such a way that it is equal to V ++() in the unitary gauge. In what
follows, this gauge connection will be used to construct an invariant FayetIliopoulos (FI) term
for the multiplet (1, 4, 3).
3.2. Superconformal properties
Now we turn to discussing the superconformal properties of the new description of the multiplet (1, 4, 3) and some immediate applications thereof. We shall need to know the realization of
the most general N = 4, d = 1 superconformal group D(2, 1; ) in the analytic basis of N = 4
superspace [8]. Since the whole set of the D(2, 1; ) transformations is contained in the closure
of Poincar supersymmetry (2.7) and conformal supersymmetry, it is sufficient to explicitly give
only the transformations of the latter:

+ = + tA + 2i(1 + ) + + ,
+ = + tA + 2i(1 + ) + + ,
(3.43)

+ +
++
+ +
u+
D ++ sc ,
u
++
sc = 2i
i = sc ui ,
i = 0,

++ 2

sc = 2i + + ,
D
sc = 0,
(3.44)
= tA + 2i (1 + ) + + + + 2i + 2i(1 + )+ , (3.45)

.
=
tA = 2itA + + ,
(3.46)
i
i i
Here, = i u
i , = ui and , are the corresponding Grassmann parameters and the
involution is defined in (2.8). We also need the transformation properties of the harmonic
derivatives, the (4, 4, 0) superfield q +a , gauge potentials V and the measures of integration
over the full and analytic harmonic superspaces (2.14)

0
D = D ++
,
D 0 = 0,
D ++ = ++
(3.47)
sc D ,
sc D
192

q +a = sc q +a ,
V ++ = 0,
V = D ++
,
sc V

(2)
(2)
+
+
A = 0,
A = A tA + ++
sc + +

++
+
+
H = A tA + sc + + H

= 2i (1 ) + + (1 + ) + + H .
(3.48)
(3.49)
(3.50)
The integration measures are evidently invariant under the N = 4 Poincar supersymmetry (2.7).
Note that for the special values of the parameter ,
(a)
= 1,
(b)
= 0,
(3.51)
the supergroup D(2, 1; ) is reduced to two different PSU(1, 1|2) supergroups, such that one
of the two commuting R-symmetry subgroups SU(2) D(2, 1; ) is identified with SU(2)
PSU(1, 1|2), while the other decouples and acts as outer automorphisms of this PSU(1, 1|2). In
particular, as follows from (3.44), the PSU(1, 1|2) supergroup corresponding to the choice = 0
does not affect harmonic variables at all.
It is easy to check superconformal covariance of the constraints (2.16), (3.2) and the harmonic
zero-curvature conditions (2.45), (3.6) at any . The quantity J defined in (3.34) is transformed
as

D ++ 0 = 0.
J = 2sc D ++
(3.52)
sc J 0 J,
The superconformal properties of the basic quantities in the gauge (3.25) can be easily found
from the condition of preserving this gauge (which fixes the relevant compensating gauge transformations) and the transformation property of the harmonic measure du in the central basis
du = du D ++
sc .
(3.53)
We obtain
V = 2sc (1 + V),
V (++) = 0 1 V (++) ,
W = 0 (1 + W).
(3.54)
= 1 V (++) ,
Taking into account that in this gauge J

Also, it is convenient to define U = 1 + W, so that
we see that (3.54) agrees with (3.52).
1
U = 0 U.
,
J = 0 J,
(3.55)
U
The object U satisfies the same constraints (3.39) as W, but has simpler transformation properties.
Let us recall the form of superconformally invariant actions of the multiplet (1, 4, 3). Within
the above gauging procedure, for all except the special value = 0, they are uniquely defined
by the corresponding actions of the multiplet (4, 4, 0). The latter are formulated in terms of
q +a qa q ia qia [8]. To obtain the invariant action of the multiplet (1, 4, 3), one just should make
the replacement q +a qa J = q +a qa in the corresponding (4, 4, 0) action. For = 0, 1 such
subclass of the general sigma-model action (2.20) is given by

= dt d 4 J 1/ = dt d 4 U 1/ ,
S(sc)
(3.56)
J=
where is a normalization constant. In particular, the free q + action (2.21) is invariant under
the supergroup D(2, 1; = 1) OSp(4 |2), and the associated superconformal (1, 4, 3) action
193
is just

=1
S(sc)

dt d J =
4
dt d 4 U 1 .
(3.57)
This action contains a non-trivial self-interaction, despite the fact that it was obtained by gauging
the free q +a action. It is the only one which admits a local representation as an integral over the
analytic superspace. This is just the example we have considered in [7]. On the other hand, the
choice of = 1/2 yields the free (1, 4, 3) action

=1/2
S(sc)
(3.58)
= dt d 4 U 2 ,
though it is obtained by gauging a non-trivial sigma-model q + action, with the superfield Lagrangian L (q ia qia )2 in (2.20).5
The cases of = 1 and = 0 at which D(2, 1; ) degenerates into the PSU(1, 1|2) supergroups (times outer SU(2) automorphisms) require a separate consideration.
At = 1 the action (3.56) is still invariant, but now the Lagrangian is just U = 1 + W and
the action identically vanishes as a consequence of the constraints (3.39). The meaningful action
in this case reads [15,19]

=1
S(sc)
(3.59)
dt d 4 U log U,
which is invariant under PSU(1, 1|2) (and the extra SU(2) automorphisms), up to a total derivative in the integrand.
In the case of = 0 the situation is even more subtle. As seen from (3.44), in this case the
harmonic variables are inert under the corresponding P SU(1, 1|2) 6 and the superfields V, U are
transformed with zero conformal weight. On the other hand, the integration measure H is still
not invariant,

H = 2i i i k k H .
(3.60)
The only way to construct the invariant action in this case is to modify the transformation law of
the analytic prepotential V under the conformal supersymmetry:

mod
(3.61)
V = 4i + +
(actually, the coefficient in the r.h.s. can be an arbitrary non-zero real number; it was chosen
in this way just for further convenience, using the freedom of rescaling V). Respectively, the
superfield U is now transformed as

mod
(3.62)
U = 2i i i k k .
The extra terms in some other PSU(1, 1|2) transformations can be found by taking the Lie brackets of (3.61), (3.62) with Poincar supersymmetry (in fact, when acting on V, the symmetry
PSU(1, 1|2) closes modulo some particular gauge transformation). The invariant action for the
5 In Ref. [10] there is an erroneous statement that all superconformally invariant actions of the reduced multiplets are
generated from the free q +a action.
6 They are still transformed under the extra automorphisms SU(2) acting on the indices i, k.
194
case = 0 is then as follows

=0
dt d 4 eU .
S(sc)
(3.63)
The modified = 0 superconformal transformation (3.61) cannot be obtained from any modification of the = 0 transformations of q +a and/or V ++(ab) before fixing the unitary gauge
with respect to the local SU(2)PG . This becomes possible within the alternative gauging of the
(4, 4, 0) system considered in the next section.
Let us now discuss the issue of superconformally invariant potentials of the multiplet (1, 4, 3)
in the description through the analytic prepotential V. It is easy to see that there exist no SU(2)PG
invariant WZ-type q +a Lagrangians among those in (2.22). Thus it seems impossible to generate
the potential terms for the multiplet (1, 4, 3) by SU(2)PG gauging of any q +a action. Nevertheless, such terms can be constructed with the help of the gauge superfield V ++(ab) . Prior to
imposing any SU(2)PG gauge, one can define the gauge invariant FI term

(2) +2 ++()
c V
= du d (2) c+2 V,
SFI = du d
+
1
c+2 = cik u+
i uk , [c] = cm ,
(3.64)
where V ++() was defined in (3.41), (3.42). This action is invariant under (3.42) and (3.28)
thanks to the condition D ++ c++ = 0, and in (3.64) we made use of the property that V ++() =
V in the unitary gauge = 1, l ++ = 0. In the WZ gauge (2.40) the component Lagrangian
following from (3.64) is cik Aik . When (3.64) is added to some non-trivial action (3.33), eliminating the auxiliary field Aik gives rise to a non-trivial potential of the physical bosonic field v0
(plus the appropriate Yukawa-type fermionic terms) [19,20].
Inspecting the superconformal properties of (3.64), one can check that it is superconformally
(PSU(1, 1|2)) invariant only at = 0 (the shift (3.61) is linear in the analytic coordinates + , +
and so does not affect (3.64)). Thus one possibility to construct the superconformally invariant
action of the multiplet (1, 4, 3) with a non-trivial scalar potential is to sum up (3.64) with (3.63),

=0
S(sc)
(3.65)
= dt d 4 eU + du d (2) c+2 V.
After elimination of Aik there comes out the conformal potential e(1+v0 ) with the strength
c2 = cik cik (see below). The extra automorphisms SU(2) acting on the indices i, k is obviously
broken down to some U (1) due to the presence of the constant triplet cik in (3.64), (3.65).
The only alternative mechanism of generating superconformally invariant potential term for
the multiplet (1, 4, 3) which was known to date [19] requires = 1. Though (3.64) is not
superconformally invariant in this case and so cannot be used, one can modify the constraints
(3.39) by inserting two arbitrary constants (one complex and one real) in their r.h.s. They form
a constant isotriplet with respect to the second R-symmetry SU(2) subgroup (the one which
provides outer automorphisms of PSU(1, 1|2) corresponding to the choice = 1). Exploiting
this broken SU(2) symmetry, one can choose the frame where
i
2 U = 0,
(D)2 U = (D)
(3.66)
D , D i U = f, f = f .
With U transforming as in (3.55), at = 1 the expressions in the l.h.s. of these constraints
can be checked to be scalars of zero conformal weight, so one can equate them to some non-zero
constants without contradiction with the superconformal PSU(1, 1|2) symmetry. The substitution
195
of U into (3.59) for U once again yields the conformal potential for v0 in the component action
(with strength f 2 ).
It is interesting to see how the modified constraints emerge within the analytic
prepotential de
scription of the multiplet (1, 4, 3). The superfield U is related to U = 1 + du V in the following
way

1
1
U = U + f i i = U + f + + .
(3.67)
2
2
This superfield has precisely the same transformation properties with respect to PSU(1, 1|2)
(=1)
U , provided that the analytic
corresponding to = 1 as the superfield U , i.e. U = 0
prepotential V possesses the following modified transformation rules with respect to the Poincar
and conformal supersymmetries

(=1)
= 2sc
V
(3.68)
(1 + V) + f + tA + + + tA + .
It is easy to check that the closure of the modified transformations coincides with the original one
(i.e. for f = 0) modulo some special gauge transformation of the form (3.28). The latter does not
make contribution to W = du V owing to the u-integral. Note that the second term in (3.67)
cannot be re-absorbed into V because it contains non-analytic Grassmann coordinates.
It is curious that the prepotential realization of the (1, 4, 3) multiplet suggests one more mechanism of generating conformal potential for the bosonic field v0 . Again, it only works for the
PSU(1, 1|2) case with = 1. It is non-minimal, because it uses a superconformal coupling to
the extra off-shell multiplet (0, 4, 4). The latter contains no physical bosons at all and comprises
4 fermionic fields and 4 bosonic auxiliary fields. It is described by the fermionic analog of q +a ,
+m ) = + , subjected to the constraint [8]

the superfield +m , (
m
D ++ +m = 0
+ m
+ m + 2i + + t im u .
+m = im u+
i + a + a
i
(3.69)
With respect to the doublet index m (m = 1, 2), it is transformed by some extra SU(2) which
commutes with N = 4 and so is an analog of SU(2)PG (it does not necessarily coincide with the
latter). The requirement of superconformal covariance of the constraint (3.69) uniquely fixes the
superconformal D(2, 1; ) transformation rule of +m , for any , as follows
+m = sc +m .
(3.70)
+m ,
The free action of

Sfree = du d (2) +m m+ ,
(3.71)
is obviously not invariant under D(2, 1; ). However, recalling the transformation law (3.54), we
observe that this action can be easily promoted to a superconformal invariant by coupling +m
to the (1, 4, 3) multiplet

S(sc) = du d (2) (1 + V) +m m+ .
(3.72)
This action is superconformal at any , and it also respects the gauge invariance (3.28) as a
consequence of the constraint (3.69). However, a simple analysis shows that in components it
yields only a bilinear term in the auxiliary fields a m , a m and therefore cannot produce a nontrivial potential of the field v0 . To get such a potential, one needs to add an FI-type term to (3.72)

SFI = du d (2) + m +m + + m m+ ,
(3.73)
196
where m , m is a constant doublet which breaks the extra SU(2) acting on the indices m. Using
the transformation properties (3.43), (3.49) and (3.70), as well as the constraint (3.69), it is easy
to show that (3.73) at = 1 is invariant under both Poincar and conformal supersymmetries,
up to a total derivative in the integrand. After passing to components, this action produces terms
which are linear in the auxiliary fields a m , a m . Eliminating these fields in the total D(2, 1; =
1) invariant action
=1
+ S(sc) + SFI ,
Stot = S(sc)
(3.74)
one again reproduces the conformal potential for v0 . The price for this is the enlargement of
the physical fermionic sector of the model from 4 to 8 fields, still with the presence of only
one physical bosonic field. Taking into account that the N = 4 multiplets (1, 4, 3) and (0, 4, 4)
can be joined into the off-shell N = 8 multiplet (1, 8, 7) [24], one can expect a hidden N = 8
supersymmetry in this combined system.
Note that the two mechanisms of obtaining superconformally invariant potential terms at
= 1 described above cannot coexist since the action (3.72) is not invariant under the modified
superconformal transformations (3.68).
3.3. Examples of component actions
Let us give a few examples of bosonic component actions. We shall need the form of the
bosonic sector of the superfield J and U = 1 + W defined in (3.34) and (3.35). Passing to the
central basis in the prepotential V(, u), we find from (3.35)

U = 1 + W = (1 + v0 ) + A++ + + + A+ + + + A
+ + + t2 v0 ,
(3.75)
where all component fields are functions of t and we still used the harmonic projections of the
Grassmann coordinates (in fact, the harmonic dependence in (3.75) is fake, which immediately
follows from the easily checkable relation ++ U = 0). The corresponding expression for the
superfield J = U 1 is

1
1 ++ +
J=
A + + A+ + + + A
1
1 + v0
1 + v0

1
1
+ + t2 v0
Aik Aik .
(3.76)
1 + v0
1 + v0
Using these explicit expressions, we find, in particular,

1 6 ik
=1
=1
2
Sbos dt (t ) A Aik , = (1 + v0 )1/2 ,
S(sc)
(3.77)
8

1
=1
=1
Sbos
dt (t )2 2 Aik Aik , = (1 + v0 ),
S(sc)
(3.78)
8

1
1
=0
=0
Sbos
dt (t )2 2 Aik Aik , = 2e 2 (1+v0 ) ,
S(sc)
(3.79)
8
where the superfield actions were defined in (3.57), (3.59) and (3.63). For U = U + 12 f ( +
+ ) there appears the additional (conformal) potential term in (3.78)

f2
1
dt 2 .
16
The FI term (3.64) yields

bos
SFI SFI
= i dt cik Aik .
197
(3.80)
After eliminating Aik from the total action (3.65) (with = 1 for simplicity), one obtains

2c2
=0
=0
S(sc)
(3.81)
Sbos
= dt (t )2 2 .
The sum of actions (3.72) and (3.73) gives rise to the following bosonic contribution

dt 2(1 + v0 )a m a m + m a m m am ,
S(sc) + SFI
(3.82)
which, upon eliminating the auxiliary fields a m , a m , again adds the conformal potential to the
action (3.78) in the total action (3.74):

m m
1
=1
=1
dt 2 .
Sbos
(3.83)
Sbos
3.4. Mirror (1, 4, 3) multiplet

To close this section, we make a few comments on the description of the mirror (1, 4, 3)
multiplet in the considered setting. The basic difference between this multiplet [20] and the one
discussed above is that its three auxiliary fields form a triplet with respect to the second (hidden) SU(2) automorphism group of N = 4, d = 1 Poincar superalgebra. They are singlets with
respect to the manifest automorphism SU(2) acting on the doublet indices i of harmonics and
Grassmann coordinates. One could consider an alternative N = 4, d = 1 harmonic superspace,
with just this second SU(2) being harmonized. In this superspace the mirror (1, 4, 3) multiplet
is described in the same way as the multiplet we dealt with here, the only difference being in the
D(2, 1; ) superconformal properties, such that the special cases = 0 and = 1 switch with
respect to each other (the formal coincidence with the description in the N = 4, d = 1 harmonic
superspace considered here can be restored by passing to the parameter = ( + 1)). On the
other hand, in the framework of the harmonic superspace considered here this alternative (1, 4, 3)
multiplet is described by a general superfield subjected to the following constraints [7]:
(a)
D + D + = 0,
(b)
D ++ = 0.
(3.84)
In the analytic basis, these constraints imply

= (, u) + i + (, u) + + (, u) ,

(a) D ++ + = D ++ + = 0,
(b) D ++ + i + + + + + = 0.
(3.85)
(3.86)
The general solution of (3.86) is

+
+ +
i
+
+ = i u+
i + s + r + 2i t ui ,
+ = i u+ + r + + s 2i + + t i u ,
(3.87)
+ i
= i i u
i + i ui ,
(3.88)
i
+
where
Re r = t .
(3.89)
198
The independent fields (t), i (t), s(t), Im r(t) constitute the alternative off-shell (1, 4, 3) multiplet.
The superconformal properties of the superfields , , can be easily defined and the
relevant actions can be constructed analogously to those presented above. Once again, the superconformally invariant potential terms can be constructed only for = 0 and = 1. At
= 0, the expression in the l.h.s. of (3.84a) has the conformal weight zero, so one can consider
the more general condition
D + D + = c++ ,
+
c++ = cik u+
i uk ,
D ++ c++ = 0.
(3.90)
The constants cik have the dimension of mass and, via the constraint (3.84b), properly modify
(3.85)(3.88). After substitution of the modified superfields into the sigma-model type superconformal action of , one obtains the conformal potential for 0 , in the same way as for v0 in the
case = 1. In the latter case, the mechanism of activating the superconformal potential for the
mirror multiplet resembles the = 0 one for the multiplet (1, 4, 3) of the first kind: it consists
in adding the proper FI-type term to the corresponding invariant sigma-model action

SFI = du d (2) g1 + + + g2 + + + c.c. .
(3.91)
This action can be shown to be D(2, 1; ) invariant only for = 1. The non-minimal mechanism of obtaining the superconformally invariant potentials exists in this case too, now for the
choice = 0. Also, it is easy to couple the two kinds of (1, 4, 3) multiplets to each other through
an interaction similar to (3.72)

SIII du d (2) (1 + V) + + ,
(3.92)
where , satisfy the constraints (3.86a) corresponding to the choice (3.84) (i.e. for cik = 0 in
(3.90)). In the future we hope to come back to a more detailed analysis of these models and their
possible implications in such long-standing problems as constructing N = 4 extensions of the
Calogero and CalogeroMoser integrable systems [25].
4. Gauging shift isometries
The multiplet (1, 4, 3) can equally be reproduced by gauging three mutually commuting shift
isometries (2.27b).
After promoting a b in (2.27b) to a b (, u),
q +a = a b u+b ,
a a = 0,
(4.1)
the constraint (2.16) should be covariantized by introducing three Abelian analytic gauge connections V ++(ab)
D ++ q +a + V ++(ab) u+
b = 0,
V ++(ab) = D ++ (ab) .
(4.2)
Like in the SU(2)PG case, one can introduce non-analytic gauge connection V (ab) ,
D ++ V (ab) D V ++(ab) = 0,
V (ab) = D (ab) ,
(4.3)
and define the non-analytic superfield q a as

q a = q +a = D q +a + V (ab) u+
b,
q a = ab ub ;
(4.4)

+a
++ q a = D ++ q a + V ++(ab) u
b =q ,
199
q a = D q a + V (ab) u
b = 0.
(4.5)
Note that q a = 0, and the last relation in (4.5) follows from the equation ++ q a =
D ++ q a = 0, which can be easily proved using the constraint in (4.2) and the harmonic
flatness condition in (4.3).
Passing to the harmonic projections , l ++ and V ++() , V ++(+) by the same relations as
in the previous section, we observe that
= + ,
l ++ = ++ ,
(4.6)
where the involved analytic parameters are the proper projections of (ab) . Eqs. (4.6) suggest
the choice of the manifestly supersymmetric unitary gauge in this case as
= l ++ = 0.
(4.7)
In this gauge, as follows from (4.2),

V ++(++) = V ++(+) = 0
(4.8)
and the only residual gauge symmetry is

V ++() = D ++ ,
= (, u).
(4.9)
The harmonic flatness condition is reduced to the set

D ++ V (++) = 0,
D
++
++
(+)
()
(4.10)
(++)
+ V = 0,
V 2V
(+)
V V
= 0,
++()
(4.11)
(4.12)
which is just the Abelian version of (3.30)(3.32). We see that, like in the case of SU(2)PG
gauging, the basic object encoding the irreducible field content in the unitary gauge is the analytic
superfield V = V ++() with the gauge transformation law V = D ++ which allows one
to choose the WZ gauge (3.36) with the (1, 4, 3) off-shell field content. From Eqs. (4.10), (4.11)
we deduce the expression for V (++) in terms of V which coincides with the one given in
Eq. (3.35)

V (++) W = du V.
(4.13)
The remaining projections V (+) and V () can be also expressed through V from
(3.30)(3.32); like in the SU(2)PG case, they seem to be of no need for constructing the corresponding invariant actions. Indeed, the superfield V (++) = W in Eq. (4.13) by construction
satisfies the constraints (3.39) defining the off-shell multiplet (1, 4, 3) in the ordinary N = 4
superspace. The most general sigma-model type off-shell action of this multiplet is given by an
expression similar to (3.33)

S = dt d 4 L(W).
(4.14)
It clearly has the same degree of generality as (3.33) in view of the relation J = 1/(1 + W).
However, these actions are obtained by gauging different subclasses of the general action of the
superfield (4, 4, 0). While in the previous case one should proceed from the SU(2)PG invariant
200
subclass, which corresponds to the restriction to the Lagrangians (3.13) depending on the only
SU(2)PG invariant structure J = q +a qa , in the case considered here we need to start from the
subclass possessing an invariance under the shifts (2.27b). It is easy to construct the appropriate
unique invariant combination of the superfields q +a and q a = D q +a :
+a
I0 = q a u+
a q ua ,
D ++ I0 = 0.
The sigma-model q +a actions invariant under (2.27b) are then constructed as

Sshift = H L(I0 , u) = dt d 4 L (I0 ),
(4.15)
(4.16)
where the second form of the action is achievable due to the property that I0 does not depend
on harmonics. Passing to the gauge invariant actions is then accomplished by covariantizing the
constraint (2.16) as in (4.2) and making the substitution q a q a in (4.15) and (4.16):

+a
++
I0 I = q +a u+
(4.17)
I = 0,
a q ua , D

loc
Sshift Sshift
(4.18)
= dt d 4 L(I ).
The unitary gauge (4.7) implies q +a = 0, so the invariant I is reduced just to W
I = V (++) = W.
(4.19)
Like in the previous case, there exist no WZ-type q +a actions (2.40) invariant under (2.27b),
so the only way of generating potential terms of the eventual (1, 4, 3) multiplet from some gauge
invariant actions of the system of superfields q +a and V ++(ab) is the FI term of the gauge
superfield. Due to the Abelian structure of the gauge group, such term is given, before any gaugefixing, by

shift
c(ab) = c(ab) .
= i du d (2) c(ab) V ++(ab) ,
SFI
(4.20)
Using the constraint (4.2), one can replace V ++(++) and V ++(+) by their inverse Higgs expressions
V ++(++) = D ++ l ++ ,
V ++(+) = l ++ D ++ .
Then, up to a total harmonic derivative, (4.20) can be rewritten as

shift
= i d (2) c++ (V 2).
SFI
(4.21)
(4.22)
It is still gauge invariant up to a total derivative in the Lagrangian. In the unitary gauge (4.7) it
coincides with (3.64) of the SU(2)PG case. The object
V = V 2,
V = D ++
(4.23)
is the Abelian analog of the modified gauge connection (3.41).

Despite the formal coincidence of the final outputs in the manifestly supersymmetric unitary
gauge in both cases, there is one important difference related to the superconformal invariance.
In the SU(2)PG case, the gauge covariantization preserves the superconformal D(2, 1; ) covariance of the original constraint (2.16). Also, the invariant J has nice superconformal properties
both before and after performing the SU(2)PG gauging. As a result, for any = 0 there is a oneto-one correspondence between the superconformally invariant sigma-model type actions of q +a
201
and those of the multiplet (1, 4, 3) emerging as a particular gauge of the original q +a , V ++(ab)
system. In the case of the gauging of three shift isometries, the gauge-covariantized constraint
(4.2) breaks the original superconformal invariance for any except = 0. Also, the superconformal transformations, at any = 0, do not take the invariant I into itself, as opposed to the
SU(2)PG invariant J . On the other hand, staying in the unitary gauge with I as the only object
accommodating the irreducible (1, 4, 3) field content, one can forget about the precise (4, 4, 0)
origin of this multiplet and construct from I any actions of the multiplet (1, 4, 3), including the
superconformally invariant ones described in the previous section. The property that the same
(1, 4, 3) actions can be obtained by gauging two non-equivalent global symmetries realized on
the multiplet (4, 4, 0) is in fact one more manifestation of the non-uniqueness of the oxidizing
procedure which is inverse to gauging. Indeed, given a sigma-model type (1, 4, 3) action, it can
be oxidized either to the (4, 4, 0) action (3.13) or to (4.16). Only the first oxidation inherits the
superconformal invariance (at = 0): starting from a superconformally invariant (1, 4, 3) action
one arrives at the (4, 4, 0) action which also respects the same superconformal invariance. The
second version of the oxidizing procedure generically lacks superconformal covariance.
As an example, let us discuss the covariantization of the free q +a action (2.21) within the
alternative gauging under consideration. Like in the previous case, we shall deal with the full
superspace form of this action. While in the SU(2)PG case the gauging is accomplished just by
the replacement D q +a q +a , it is not so in the shift case, just because even in the rigid
case the action (2.21) is invariant under (2.27b) up to a total derivative in the Lagrangian. The
gauge-invariant (once again, up to a total harmonic derivative) superfield Lagrangian in this case
proves to be as follows
+
+a +
(ab) ++c) +
Lfree
qa 2V (ab) u+
V(b u(a uc) .
gauge = q D
a qb + 2V
(4.24)
In the unitary gauge q +a = 0, it is simplified to

++c) +
u(a uc) .
(ab)
V(b
Lfree
gauge = 2V
(4.25)
It is curious that the Lagrangian (4.24) coincides, modulo a total harmonic derivative, with the
square of the gauge invariant quantity I defined in (4.17)
(++)
2
2
,
Lfree
(4.26)
gauge = I = V
where the second equality is valid in the unitary gauge (recall (4.19)). To prove the equivalence
of (4.24) and (4.26), it is sufficient to compare their gauge-fixed forms. The r.h.s. in (4.25) can
be rewritten as
V (++) V ++() V () V ++(++) = V (++) V ++() ,
(4.27)
where we used the property (4.8) which is valid in the unitary gauge. Then we represent one
of two V (++) in (4.26) as V (++) = V (ab) D ++ u+

(a ub) , integrate by parts with respect to
D ++ , use the relations (4.3), (4.10) and once again (4.8) to reduce [V (++) ]2 just to the form
(4.27).
We see that in the shift case the gauging of the free (4, 4, 0) action yields the free action of the
multiplet (1, 4, 3). This should be contrasted with the SU(2)PG gauging which produces from the
free q +a action the (1, 4, 3) action (3.57) involving a non-trivial self-interaction [7]. This simple
example illustrates the non-compatibility of the shift gauging with superconformal invariance
at = 0: the superconformal symmetry leaving invariant the free q +a action is D(2, 1; = 1),
202
while the free action of the multiplet (1, 4, 3) is invariant under D(2, 1; = 1/2). No such an
inconsistency takes place in the case of the SU(2)PG gauging.
It is interesting that the only N = 4, d = 1 superconformal symmetry which is consistent with
the gauging considered here corresponds to the exceptional case = 0 in which D(2, 1; ) degenerates into PSU(1, 1|2) and an extra SU(2) automorphisms group. Indeed, the covariantized
constraint (4.2) is manifestly invariant under the = 0 version of the transformations (3.43)
(3.48). What is even more essential is that the constraint (4.2) is also invariant under the following
modified transformation of q +a

mod
q +a = 2i a + a + .
(4.28)
Here, is a constant which can be fixed at any non-zero value by simultaneously rescaling
q +a , V and the gauge parameters (ab) . We will choose = 1. Note that the possibility of
such a modification (missed in [8]) exists already at the rigid level, since the original constraint
(2.16) is invariant under such an additional shift. In the unitary gauge, with = 1 and taking into
account the appropriate compensating gauge transformations, the analytic prepotential V and the
superfield W transform as

mod
(4.29)
mod
V = 4i + + ,
W = 2i i i i i ,
which coincides with the = 0 transformation laws (3.61) and (3.62) of the previous section.
In the SU(2)PG case these transformations cannot be derived from the first principles, i.e.
prior to imposing any gauge-fixing condition, because the constraint (3.2) is not covariant under
(4.28). On the other hand, in the alternative approach where Abelian shift symmetries are gauged,
this becomes possible since (4.2) is covariant under (4.28). Thus, as regards the superconformal
properties, the two different ways of deducing the multiplet (1, 4, 3) by gauging three-parameter
rigid isometries of the (4, 4, 0) multiplet are complementary to each other: the SU(2)PG gauging
is compatible with the D(2, 1; ) symmetries for all = 0, while the second gauging suits for
treating the exceptional = 0 case. Note that the gauge invariant quantity I defined in (4.15),
(4.17) has the following = 0 transformation properties

mod
(4.30)
I = 2i i i i i
both in the rigid and local cases, so the superconformally invariant Lagrangian (3.63) of the
multiplet (1, 4, 3) is obtained via the Abelian gauging, q a q a = q +a , of the following
particular case of the Lagrangians in (4.16)
I0
q
L=0
(sc) = e = e
a u+
a
eq
+a u
a
= e(q
ia )
ai
(4.31)
The corresponding q + action is invariant under the = 0 superconformal group PSU(1, 1|2).
5. Concluding remarks
In this paper, we continued the study of implications of the gauging procedure of Ref. [7]
in the models of N = 4 supersymmetric mechanics. We have shown that the general models
associated with the off-shell multiplet (1, 4, 3) can be recovered, in a manifestly supersymmetric
superfield form, by gauging certain three-parameter symmetries appearing in special subclasses
of the superfield actions of the multiplet (4, 4, 0), thereby confirming the role of the latter as the
basic (or root) multiplet for constructing various models of N = 4 mechanics. We have found a
new description of the multiplet (1, 4, 3) in terms of the unconstrained harmonic analytic gauge
203
superfield V(, u), V = D ++ (, u). Since the multiplets (4, 4, 0), (3, 4, 1) and (0, 4, 4)
also admit a natural description as N = 4 harmonic analytic superfields [8], we conclude that the
N = 4, d = 1 harmonic analytic superspace plays a key role in N = 4 mechanics. Actually, the
chiral N = 4, d = 1 multiplets (2, 4, 2), both linear [19,20,26] and non-linear [9], also admit an
alternative description in terms of N = 4 analytic superfields [27]. The new off-shell formulation
of the multiplet (1, 4, 3) allowed us to find a new mechanism of generating potential terms for
this multiplet and to write simple off-shell couplings of this multiplet to the mirror (1, 4, 3)
multiplet (which can also be formulated in the N = 4, d = 1 harmonic superspace). Also note
that it is easy to couple the (1, 4, 3) multiplet to the off-shell multiplet (3, 4, 1) in the description
via the analytic superfield W (, u), D ++ W ++ = 0 [8]. The superconformally invariant form of
this coupling is given by the following unique analytic superspace integral7

SV W du d (2) (1 + V)W ++ .
(5.1)
It is gauge invariant because of the constraint D ++ W ++ = 0. In the bosonic sector it yields direct
couplings of the physical fields of one multiplet to the auxiliary fields of the other one and can
also be used to generate non-trivial scalar potentials in the coupled system of two multiplets after
eliminating the auxiliary fields. One can also couple the multiplet (1, 4, 3) to some extra (4, 4, 0)
++ Q+a c Q+b
multiplet Q+a , D ++ Q+a = 0, via the substitutions W ++ Q+a u+
ab
a or W
in (5.1) (only the second one preserves the superconformal invariance [8]).
We hope that these findings and new tools will help us to gain further insights into the problem
of constructing N = 4 extensions of some important bosonic systems, such as the integrable
many-component Calogero-type models [25].
In the process of our study we exhibited (in Section 2.2) the full set of symmetries inherent to
the free superfield action of the multiplet (4, 4, 0). Some of them admit an extension to more general (4, 4, 0) actions (like SU(2)PG (2.23) or its Abelian shift analog (2.27b)) while some others
do not. In particular, it seems impossible to construct, out of q +a , q a = D q +a and harmonics u
i , any tensorial invariant of the symmetries associated with the solvable three-generator
algebras (2.33)(2.36). However, even in this case we can get a (1, 4, 3) action with a non-trivial
interaction as the result of the appropriate gauging of the free q +a action. Let us end up with an
example of such gauging.
For definiteness we choose the symmetry associated with (2.33). Its local version is spanned
by the following set of gauge transformations
1 q +a = 1 ca b q +b ,
2 q +a = 2 u+a ,
3 q +a = 3 ca b u+b .
(5.2)
The gauge covariantization of the constraint (2.16) and of q a = D q +a can be easily constructed
D ++ q +a V1++ ca b q +b V2++ u+a V3++ ca b u+b = 0,
(5.3)
V1 ca b q +b
(5.4)
+a
=D
+a
V2 u+a
V3 ca b u+b .
Here the gauge potentials are transformed as

V1 = D 1 ,
V2 = D 2 1 V3 + 3 V1 ,
7 The superconformal invariance can be broken by adding, to the Lagrangian in (5.1), the term W ++ with an
arbitrary coupling constant.
204
V3 = D 3 + 1 V2 2 V1 ,
(5.5)
and satisfy the following zero-curvature conditions

D ++ V1 D V1++ = 0,
D ++ V2 D V2++ + V1++ V3 V3++ V1 = 0,

D ++ V3 D V3++ V1++ V2 + V2++ V1 = 0.
(5.6)
The correct gauge covariantization of the free q +a Lagrangian q +a D qa+ in the present
case is given by

q +a D qa+ q +a D qa+ V1 q +a cab q +b 2V2 q +a u+
a

2V3 q +a cab u+b + 2 V3++ V2 V2++ V3 u+a cab ub .
(5.7)
Under (5.2) this expression transforms into a total harmonic derivative and so is not a tensor. The
corresponding action can of course be rewritten as an integral over the analytic superspace. One
can also add an FI term

du d (2) V1++ .
(5.8)
After passing to the WZ gauge in (5.3), (5.7) and (5.8) (VB = AB ), descending to
components, properly fixing the residual 3-parameter gauge freedom and eliminating the auxiliary fields AB , one is left with a non-trivial action of a self-interacting (1, 4, 3) multiplet.
Acknowledgements
The work of E.I. was supported in part by the NATO grant PST.GLG.980302, the RFBR
grant 06-02-16684, grant INTAS 05-7928 and a grant of HeisenbergLandau program. He thanks
Laboratoire de Physique, ENS Lyon, for the kind hospitality extended to him during the course
of this study.
References
[1] S.J. Gates Jr., L. Rana, On Extended Supersymmetric Quantum Mechanics, Maryland Univ. Preprint #UMDPP
93-24, October 1994.
[2] S.J. Gates Jr., L. Rana, Phys. Lett. B 345 (1995) 233, hep-th/9411091.
[3] S.J. Gates Jr., L. Rana, Phys. Lett. B 342 (1995) 132, hep-th/9410150.
[4] A. Pashnev, F. Toppan, J. Math. Phys. 42 (2001) 5257, hep-th/0010135.
[5] Z. Kuznetsova, M. Rojas, F. Toppan, JHEP 0603 (2006) 098, hep-th/0511274.
[6] S. Bellucci, S.J. Gates Jr., E. Orazi, A journey through garden algebras, in: Lectures Given at Winter School on
Modern Trends in Supersymmetric Mechanics (SSM05), Frascati, Italy, 712 March 2005, hep-th/0602259.
[7] F. Delduc, E. Ivanov, Nucl. Phys. B 753 (2006) 211, hep-th/0605211.
[8] E. Ivanov, O. Lechtenfeld, JHEP 0309 (2003) 073, hep-th/0307111.
[9] E. Ivanov, S. Krivonos, O. Lechtenfeld, Class. Quantum Grav. 21 (2004) 1031, hep-th/0310299.
[10] S. Bellucci, S. Krivonos, A. Marrani, E. Orazi, Phys. Rev. D 73 (2006) 025011, hep-th/0511249.
[11] S. Hellerman, J. Polchinski, Supersymmetric quantum mechanics from light cone quantization, in: M.A. Shifman
(Ed.), The Many Faces of the Superworld, hep-th/9908202.
[12] S.J. Gates Jr., W.D. Linch III, J. Phillips, When superspace is not enough, hep-th/0211034.
[13] E. Ivanov, A. Smilga, Phys. Lett. B 257 (1991) 79.
[14] V. Berezovoj, A. Pashnev, Class. Quantum Grav. 8 (1991) 2141.
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
205
E. Ivanov, S. Krivonos, O. Lechtenfeld, JHEP 0303 (2003) 014, hep-th/0212303.

F. Delduc, S. Krivonos, unpublished.
C. Burdik, S. Krivonos, A. Shcherbakov, Czech. J. Phys. 55 (2005) 1357, hep-th/0508165.
S. Krivonos, A. Shcherbakov, Phys. Lett. B 637 (2006) 119, hep-th/0602113.
E.A. Ivanov, S.O. Krivonos, V.M. Leviant, J. Phys. A 22 (1989) 4201.
E.A. Ivanov, S.O. Krivonos, A.I. Pashnev, Class. Quantum Grav. 8 (1991) 19.
A. Galperin, E. Ivanov, V. Ogievetsky, E. Sokatchev, Pisma Zh. Eksp. Teor. Fiz. 40 (1984) 155, JETP Lett. 40
(1984) 912;
A.S. Galperin, E.A. Ivanov, S. Kalitzin, V.I. Ogievetsky, E.S. Sokatchev, Class. Quantum Grav. 1 (1984) 469.
A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, E.S. Sokatchev, Harmonic Superspace, Cambridge Univ. Press, 2001,
306 pp.
E.A. Ivanov, V.I. Ogievetsky, Teor. Mat. Fiz. 25 (1975) 164.
S. Bellucci, E. Ivanov, S. Krivonos, O. Lechtenfeld, Nucl. Phys. B 684 (2004) 321, hep-th/0312322.
G.W. Gibbons, P.K. Townsend, Phys. Lett. B 454 (1999) 187, hep-th/9812034.
S. Fubini, E. Rabinovici, Nucl. Phys. B 245 (1984) 17.
F. Delduc, E. Ivanov, in preparation.
Functional integral for ultracold fermionic atoms

S. Diehl , C. Wetterich
Institut fr Theoretische Physik, Philosophenweg 16, 69120 Heidelberg, Germany
Received 6 February 2006; received in revised form 7 February 2007; accepted 20 February 2007
Available online 12 March 2007
Abstract
We develop a functional integral formalism for ultracold gases of fermionic atoms. It describes the BEC
BCS crossover and involves both atom and molecule fields. Beyond mean field theory we include the
fluctuations of the molecule field by the solution of gap equations. In the BEC limit, we find that the low
temperature behavior is described by a Bogoliubov theory for bosons. For a narrow Feshbach resonance
these bosons can be associated with microscopic molecules. In contrast, for a broad resonance the interaction between the atoms is approximately pointlike and microscopic molecules are irrelevant. The bosons
represent now correlated atom pairs or composite dressed molecules. The low temperature results agree
with quantum Monte Carlo simulations. Our formalism can treat with general inhomogeneous situations
in a trap. For not too strong inhomogeneities the detailed properties of the trap are not needed for the
computation of the fluctuation effectsthey enter only in the solutions of the field equations.
PACS: 03.75.Ss; 05.30.Fk
1. Introduction
Ultracold fermionic atoms can exhibit both the phenomena of a BoseEinstein condensate (BEC) [1,2] of molecules and the condensation of correlated atom pairs similar to BCSsuperconductivity [3,4]. Recent experimental progress [58] in the crossover region reveals the
universality of the condensation phenomenon, as anticipated theoretically [914].
E-mail addresses: s.diehl@thphys.uni-heidelberg.de (S. Diehl), c.wetterich@thphys.uni-heidelberg.de

(C. Wetterich).
doi:10.1016/j.nuclphysb.2007.02.026
S. Diehl, C. Wetterich / Nuclear Physics B 770 [FS] (2007) 206272
207
In this paper we develop a systematic functional integral formulation for the treatment of the
equilibrium state1 of ultracold fermionic atoms. We discuss in detail how to arrive at a formulation that treats the fermionic fluctuations of unbound atoms and the bosonic fluctuations of the
molecule or di-atom field on equal footing. An approach based on a HubbardStratonovich transformation is an ideal starting point for a unified inclusion of fluctuations of molecules or Cooper
pairs. We show how this formalism can be implemented in practice in a self-consistent approximation scheme. We carefully discuss the renormalization procedure that is needed to absorb the
ultraviolet divergences in this non-relativistic quantum field theory. The result is an effective low
energy formulation which is insensitive to the microphysical cutoff scale .
We concentrate on dimensionless quantities by measuring all quantities in units of the
Fermi momentum kF or the Fermi energy F = kF2 /2M. The inverse Fermi momentum is the
most important length scale in the problem, related to the total number density of atoms by
kF = (3 2 n)1/3 . It measures the typical interparticle spacing. We show that the atom density
does not appear as an independent parameter in the computations which can be performed in
terms of dimensionless ratios. This renders our formalism highly universal, since the results of
experiments with different atoms and densities can be related by simple scaling laws.
For this purpose, we introduce three dimensionless parameters that characterize the crossover
problem efficiently: First, the concentration c describes the ratio between the in-medium scattering length and the average distance between two unbound atoms or molecules. Its inverse,
c1 , smoothly connects the weakly coupling BCS regime (c < 0, |c| 1) with the BEC regime
(c > 0, |c| 1). The perhaps most interesting region is the crossover regime in between,
|c1 | 1. Second, a Yukawa or Feshbach coupling h characterizes the interaction between
atoms and molecules. The third parameter is the temperature in units of the Fermi energy,
T = T /F . No further details of the microphysics are needed for the macroscopic quantities. In this sense the description becomes universal. In the language of quantum field theory or
critical phenomena the parameters c and h describe relevant couplings for the long distance
physics.
Our formalism is well suited to describe all regimes of coupling and temperature, including
the superfluid phase with broken symmetry. As a first application we compute the phase diagram
for the crossover problem as described by the dependence of the critical temperature on c. The
function T (c1 ) only depends on the value of h and shows universal limits for large |c1 |, i.e.
in the BCS or BEC regime, respectively. In the crossover region the dependence on the Yukawa
coupling h is strongest. Nevertheless, we find that for the two limits h 0 (narrow resonance)
and h (broad resonance) the crossover becomes independent of h . For broad Feshbach
resonances the Yukawa coupling becomes an irrelevant parameter.
In the narrow resonance limit for h 0 the bosonic degrees of freedom can be associated with microscopic molecules. In this limit the molecule fluctuations can be neglected and
mean field theory becomes valid. In contrast, for h the microscopic molecule degrees
of freedom play no role and the model is equivalent to a purely fermionic model with pointlike interactions. Nevertheless, effective molecular bound states (dressed molecules) become a
crucial feature of the crossover physics.
For an actual comparison with experimental results one needs to relate the universal parameters to experimental observables, in particular the strength of the magnetic field. This is done
in [15], where we connect the scattering properties of atoms with the values of our universal
1 We do not touch in this work on the highly interesting non-equilibrium physics of the atom gas.
208
parameters. The present paper may therefore be viewed as the theoretical basis for the more detailed comparison with experiment in [15]. Here we concentrate on the conceptual and formal
developments.
A further aspect of universality concerns the geometry of the trap. We will present a systematic
derivative expansion which computes the effects of fluctuations and interactions independently
of the shape of the trap. The details of the trap only enter at the end through the solution of effective field equations in presence of a trap potential. The systematic character of the derivative
expansion allows for a quantitative estimate of the reliability of simple approximations, like the
local density resp. ThomasFermi approximation or the local condensate approximation
frequently used in Refs. [1620]. The effective field equation for the condensate has the same
status as the (time independent) GrossPitaevskii equation [2123] for a BoseEinstein condensate.
Our approach basically relies on two ingredients: The functional integral and the Hubbard
Stratonovich transformation or partial bosonization. The partial bosonization permits us to formulate the problem microscopically as a Yukawa theory, thereby allowing to deal with nonlocal
interactions. This route is also taken in [16,17,2433]. The power of the functional integral techniques, however, is so far only marginally used. A first attempt in this direction was made by
Randeria [11,34]. Later, other approaches employed this concept more as an argumentative tool
than as a method for concrete calculations [25,30,31,35].2
Beyond the systematic functional integral formulation and the emphasis on the universal aspects of the phase transition our work extends previous results by the systematic inclusion of the
molecule fluctuations. These fluctuations are important for the quantitative understanding of the
phase transition for a broad Feshbach resonance with a large dimensionless Yukawa coupling
h , as relevant for the present experiments in 6 Li and 40 K. For zero temperature our calculations
agree well with quantum Monte Carlo simulations [37] at the resonance.
This paper is organized as follows:
In Section 2 we investigate the Feshbach resonance and introduce the important molecule de
grees of freedom in terms of a di-atom or molecule field (x).
This allows us to cover the whole
range of temperature and the crossover. The BoseEinstein condensate or the superfluid order
We establish the equivalence
parameter corresponds to a non-vanishing expectation value .
of our formulation with a purely fermionic formulation for which the effective interaction between the atoms contains a nonlocal piece reflecting the molecule exchange. Only in the broad
resonance limit h this interaction becomes pointlike.
Our functional integral formulation in terms of an independent field (x)

is particularly well
adapted to the crossover from a BEC to a BCS condensate: Molecules and Cooper pairs are
described by the same field. Of course, the dynamical properties depend strongly on the BEC or
BCS regime. In particular, we compute in Section 3 the gradient coefficient A which determines
the gradient contribution to the free energy for a spatially varying molecule field. For the BEC
regime, A is dominated by the classical value, corresponding to the dominance of the tightly
bound molecules. In contrast, for the BCS regime the fluctuation effects dominate A . In this
case (x)
can be associated to a collective degree of freedom (Cooper pair). One could omit
the classical contribution to A such that becomes on the microscopic level an auxiliary field.
Indeed the presence of a molecular bound state is no longer crucial in the BCS regime. One may
2 The present paper covers part of a longer first version of [15]. A publication using functional integral computations
has appeared more recently [36].
209
work in a purely fermionic setting with a local interaction which depends on a magnetic field B.
For a broad Feshbach resonance this feature holds for the entire crossover region.
We discuss the derivative expansion of the effective action in Section 3. In particular, Section 3.1 addresses the issue of additive renormalization of the detuning and Section 3.2 computes the wave function renormalization Z which distinguishes the renormalized field for the
dressed molecules and the field for the microscopic of bare molecules. We turn in Section 4 to
the discussion of the relevant parameters that describe the universal aspects of ultracold atoms.
In terms of the dimensionless concentration c and Yukawa coupling h the system becomes
independent of the detailed short distance properties.
Section 5 discusses the fraction of atoms bound in molecules. It is crucial to distinguish between the bare or microscopic molecules and the dressed molecules [38,39]. Their numbers are
related by the multiplicative wave function renormalization Z . In order to achieve a complete
symmetry between the fermionic fluctuations of unbound atoms and the bosonic fluctuations
of molecules we adapt our functional integral setting in Section 6. In Section 7 we turn to the
BEC limit. For a broad Feshbach resonance (large h ) one finds very large Z , such that the
condensate fraction (condensed dressed molecules) exceeds by far the number of microscopic
molecules. The latter becomes completely negligible for h . Nevertheless, we find a Bogoliubov theory for bosons in the low temperature BEC regime for all values of h . Finally, we
include the molecule fluctuations (or collective fluctuations of di-atom states) in Section 8 in the
form of new bosonic gap equations. We draw conclusions in Section 9.
While the main part of this paper deals with a homogeneous situation our formalism can be
extended to cover the inhomogeneous situation in a trap of atoms if the inhomogeneity is not too
large. Since the main part is independent of the discussion of inhomogeneities we display the
formalism for inhomogeneous situations in Appendix A.
We introduce a general formalism for a functional integral which applies to arbitrary fermionic systems and is easily generalized to systems with bosons, far beyond the particular case of a
Feshbach resonance (where di-atom states play a role). In addition to the (Grassmann) field vari
ables (x)
for the fermionic atoms we employ a bosonic field variable (x). It corresponds to a
varying effective chemical potential which is associated to the density field n(x). This procedure
allows computations for the inhomogeneous setting of atoms in a trap beyond the small density
approximation or beyond the ThomasFermi approximation. The bosonic field variable is introduced by partial bosonization. We formulate the effective action [ ] and establish the exact
formal relations between (x), n(x), the chemical potential and the local trap potential Vl (x).
2. Dilute gas of ultracold atoms
The ultracold gas of fermionic atoms in the vicinity of a Feshbach resonance can be treated
in the idealization of two stable atomic states denoted by a two component spinor . (For the
example of 6 Li these states may be associated with the two lowest hyperfine states |1, |2.) The
molecular state responsible for the Feshbach resonance can be treated as a bosonic particle. In
our idealization it is stable for negative binding energy and can decay into a pair of fermionic
atoms for positive binding energy.
For a realistic description our formalism has to be capable to describe the following phenomena: (i) Condensates of atom pairs may form at low temperature, similar to the BCS description
of superconductivity. (ii) Molecules of two atoms can be exchanged between the single atoms,
thus contributing to the interaction. Also these molecules may form a BoseEinstein condensate
210
at low temperature. In our formalism both effects find a unified description as will be discussed
in detail in this paper.
We work with a microscopic action which explicitly includes a bosonic field with atom
number two [25],

1

SB = dx
+
+ 2
2M
4M

1 T

.
h
(1)
2
Here we employ the Matsubara formalism where the Euclidean time is wrapped on a torus with
circumference = 1/T with conventions

x = (, x),
dx =

d
d 3 x.
(2)
(Our units are h = c = kB = 1.) We will shortly see how this microscopic model relates to a
purely fermionic setting by the means of a partial bosonization or HubbardStratonovich transformation.
The complex two-component spinors (x)

are anticommuting Grassmann variables. We assume an equal mixture of the two atomic states. In this case the chemical potential associated to
the difference in the number of atoms in the up and down states precisely cancels the energy
difference between the two states such that both can be omitted. (For unequal mixtures the action contains an additional term 3 .) The bosonic molecules are described by a complex
The propagator for these bare molecules is obtained from simple symmetry
bosonic field .
considerations, i.e. we assume a mass 2M, leading to a non-relativistic kinetic energy q 2 /4M.
The quadratic term involves the bare binding energy or detuning ( ) which typically
depends on the magnetic field. In order to make contact to physical observables, has to be
additively renormalized, which is implemented in Section 3.1.
We use here two different chemical potentials and for the fermionic atoms and bare
molecules. This is useful if we want to obtain separately the densities of fermionic atoms or bare
molecules by differentiation of the free energy with respect to or . Since only the total number
of atoms is conserved, one has to set = at the end of the computations. The distinction
between and is appropriate if one wants to understand explicitly the role of the microscopic
(or bare) molecules. In Section 6 we will drop this distinction in favor of a more unified approach.
There we will set = from the outset such that will be conjugate to the total density of atoms
irrespective of a microscopic distinction between unbound atoms and molecules. In the main part
of this paper we will treat and as constant classical source terms. However, we stress that
this source term can be straightforwardly promoted to a fluctuating field, (x). This issue,
and its use for the description of inhomogeneities beyond the usual local density approximation,
will be discussed in Appendix A.
The Yukawa or Feshbach coupling h describes the coupling between the single atoms and
molecules, = , 12 = 1. For h 0 the molecular states decouple. However, in an appropriately performed narrow resonance limit h 0 which keeps the scattering length fixed,
an exact solution of the many-body problem becomes feasible above the critical temperature.
A detailed analysis of this limit is given in [15]. Broad Feshbach resonances correspond to large
h , and we will see that the limit h describes a purely fermionic theory with pointlike
interactions where microscopic molecules can be neglected.
211
The thermodynamic equilibrium situation is described by the partition function. The basic
ingredient for our formalism is the representation of this object in terms of a functional integral
with weight factor eSB , with SB the Euclidean action (1)

]
+ dx (x) (x) + (x)
exp SB [,
Z[, j ] = DD

.
+ j (x) (x) + j (x)(x)
(3)
This formulates the full quantum theory in terms of the sources and j for the fermion and the
boson fields. All one particle irreducible n-point functions, including the order parameter and the
correlation functions, can be directly extracted from the effective action , which obtains by a
Legendre transform,

= ln Z + dx (x) (x) + (x)(x) + j (x) (x) + j (x)(x)
[, ]
(4)
.
The effective action is a functional of the classical fields (or field expectation values) = ,
= in the presence of sources. They are defined as
(x)
= (x)
=
ln Z
j (x)
(5)
and analogous for the fermion fields. Of course, due to Paulis principle, the fermion field cannot
acquire a non-vanishing expectation value for = = 0 such that the physical value is simply
= 0. It is often convenient to write as an implicit functional integral over fluctuations ,

around background fields ,

= ln DD exp S[ + , + ] +
[, ]
j + + h.c. , (6)
This form is particularly useful for the construction of the equation of

with j (x) = / (x).
state, i.e. the explicit expression for the total atom number density.
The total number density of atoms n includes those from unbound or open channel atoms
and the ones arising from the bare molecules or closed channel atoms. It obeys

n(x) = n F (x) + n B (x) = (x)(x)

(7)
+ 2 (x)(x)
.
Indeed, the action is invariant under U (1) phase transformations of the fermions and bosons,
ei ,
e2i
(8)
and the corresponding Noether charge is the total atom number N = d 3 x n(x). We emphasize that Eq. (7) is no ad hoc assumptionit is an exact expression for the particle number
and
and directly follows from the microscopic formulation. More technically speaking,
2 represent the full two-point correlation functions of the bare fields which appear in the
microscopic action (1) and are quantized by means of the functional integral. In a homogeneous
situation, the conserved particle number can be replaced by a fixed constant particle density,
n = N/V .
The bosonic part n B counts the total number of atoms contained in the microscopic or bare
molecules. This number receives a contribution from free molecules and from the condensate
as discussed in more detail in Section 5. In the language often used for a Feshbach resonance,
212
n B measures the closed channel microscopic atoms. Using the formalism of the present paper
we have computed n B as a function of the magnetic field B in [15]. We find very good agreement
with observation [8] over several orders of magnitude in n B .
Since the action (1) contains only terms quadratic or linear in it is straightforward to express
the expectation value 0 in terms of a local fermion-bilinear. It obeys (for vanishing source for )

2

h T

+ 0 =
.
4M
2
(9)
In particular, for constant 0 we find

0 =
T
h
.
2 4
(10)
This demonstrates directly that our formalism makes no difference between a condensate of
are simply related by a multimolecules and a condensate of atom pairs T they

plicative constant.
We finally show the equivalence of our formalism with a model containing only fermionic
atoms and no microscopic molecules. The interaction in this fermionic description is, in general,
not local. It becomes local, however, in the limit of a broad Feshbach resonance for h .
For this purpose we use again the quadratic form of the bosonic part of the microscopic action
(1), which allows us to integrate out the field. Expressed only in terms of fermions our model
contains now a momentum dependent four-fermion interaction (Q4 = Q1 + Q2 Q3 ; Q =
(n , q
) with discrete bosonic Matsubara frequencies n = 2nT at finite temperature)

1
2 ) (Q4 )(Q
Sint =
(Q1 )(Q
3)
2
Q1 ,Q2 ,Q3
h 2
2 + 2i(n1 n4 )T + (
q1 q
4 )2 /4M

.
(11)
We emphasize that there is no difference between the Yukawa type model described by the action (1) and a purely fermionic model with interaction (11). All physical observables can be
computed in either one or the other of the two formulations. However, Eq. (11) reveals that our
model describes a setting beyond pointlike interactions via the classical frequency and momentum dependence of the four-fermion interaction. The momentum structure of (11) is compatible
with interactions in the -channel. The action (1) hence models a nonlocal coupling between
the fermionic constituents. Reversing the logic, Eq. (1) could also be obtained by starting from
Eq. (11) and performing a HubbardStratonovich transform or partial bosonization [40,41].
Finally, we note that we could also choose classical gradient coefficients A (cl)
(see below)
different from 1/(4M) in order to model an experimentally determined effective range.
In the pointlike limit the momentum dependence and the dependence on can be neglected
and the coupling term in Eq. (11) is replaced by the local interaction approximation
=
h 2

(12)
This limit obtains formally for h 2 , , while keeping fixed. It is relevant for broad
Feshbach resonances, as discussed in detail in [15].
213
3. Derivative expansion for the effective action

The condensation of atoms pairs or molecules is signalled by a non-vanishing expectation
= 0 . The associated symmetry breaking of the global continuous symmetry of phase
value
rotations of and (related to the conservation of the number of atoms) induces a massless
Goldstone boson. This is the origin of superfluidity. In this section, we show how to describe
this phenomenon in the effective action formalism. For this conceptual issue, it is sufficient to
work in the mean field approximation or a simple extension thereof (extended MFT). The more
sophisticated approximation schemes beyond mean field are presented in Sections 6, 8.
In this work we will treat the effective action in a derivative expansion, i.e. we write

+ ,
= dx U ()
+ Z + A
[]
(13)
and compute the wave function renormalization Z , the gradient coefficient A and the ef In the present paper we do not compute the corrections to the part of
fective potential U ().
the effective action involving fermionsfor this part we have simply taken the classical action.
(See [15] for the renormalization of the Yukawa coupling in presence of a background fourfermion interaction.) We therefore omit the fermionic part of the effective action from now on.
For the concrete calculation we work in momentum space. We emphasize, however, that the
above expression can be used for the investigation of weak inhomogeneities as encountered in
atom traps. The fluctuation problem, i.e. the computation of Z , A , U in the above truncation,
can then be solved in momentum space, while the effects of weak inhomogeneities can be investigated by solving the classical field equations derived from (13). This reaches substantially
beyond the usual local density approximation, which ignores the kinetic terms in Eq. (13). We
discuss the implementation of an external trapping potential in our functional integral formalism
in Appendix A. Here, however, we focus on the homogeneous situation.
In its most general form, the effective action depends on the parameters T , and and on
The dependence on the chemical potentials and can already be inferred
the classical field .
from the partition functionthey are only spectators w.r.t. the Legendre transform. Following
the thermodynamic construction, the total particle density in a homogeneous setting is obtained
as

U
U
.
n=
(14)

This prescription precisely reproduces Eq. (7). Here U has to be taken at its minimum.
In the absence of sources the field equation for (x)

reads
!
= 0.
(x)
(15)
For a homogeneous situation the stable solution corresponds to a minimum of the effective potential
U
= U = 0,
(16)

with
U =
U
,

= .
(17)
214
Here we have introduced the U (1) invariant = the

effective potential only depends on
this combination. This simple equation can be used to classify the thermodynamic phases of the
system,
Symmetric phase (SYM):
0 = 0,
U (0) > 0,
Symmetry broken phase (SSB):
0 > 0,
U (0 ) = 0,
Phase transition (PT):
0 = 0,
U (0) = 0,
(18)
where 0 = 0 0 denotes the minimum of U (). For high temperatures, the minimum of the
effective potential occurs at = 0 and we deal with a normal gas phase. For low enough T ,
to occur for 0 = 0. The spontaneous
on the other hand, we expect the minimum of U (, )
breaking of the U (1) symmetry is signalled by a nonzero field expectation value and indicates
the condensation phenomenon. This has an important aspect of universality: one and the same
criterion can be used for the whole parameter space, both for the BCS-type condensation of
Cooper pairs and for the BEC of microscopic molecules.
In the remainder of this section, we will evaluate the effective action in the mean field approximation (MFT). This scheme is defined by only considering the effects generated by fermion
fluctuations. We will include bosonic fluctuations in later chapters. Beyond MFT, we first include the contribution to the density from dressed molecules. This is connected to the effective
bosonic two-point function (connected part of the bosonic particle density). This effect is included in different current approaches to the crossover problem in the limit of broad Feshbach
resonances [19,29,38]. We will refer to it as extended mean field theory. Furthermore, we include
in Section 8 the modifications of the effective potential due to bosonic fluctuations, using suitable
SchwingerDyson equations.
In a realistic physical situation the validity of our model is restricted to momenta smaller
than some microphysical ultraviolet cutoff . In turn, is typically given by the range of
the van der Waals interactions. One may use aB /100 with aB the Bohr radius. For practical
computations it is often convenient to consider the limit such that no explicit information
about the cutoff physics is needed. This requires to express the couplings of the theory in terms
of suitably defined renormalized couplings that stay finite for . In the next subsection
we will discuss the additive renormalization of the detuning. The functions Z , A , in contrast,
are UV finite. We will also subtract field-independent pieces linear in and that obtain in
a naive MFT computation. This additive density renormalization will be traced back to the
relation between the functional integral and the operator formalism.
3.1. Effective potential and additive renormalization
In the mean field approximation, the effective potential reads, after carrying out the Matsubara
summation and omitting an irrelevant infinite constant,
= ( 2) + U (F ) ,
U (, )
1
U1(F )
where
1
=
2T

= 2T
d 3q
ln cosh
(2)3
q2
2M
1/2
2
+r
1/2

= 2 + 2
,
(19)
215
q2
2M
r 1/2
(20)
,
=
,
r = h 2 .
2T
2T
Here we have added an index in order to remind that this form still depends on an ultraviolet
cutoff . We regularize the effective potential by limiting the integration over spacelike momenta
by an upper bound , q 2 < 2 .
Let us now discuss the additive renormalization needed to properly describe the physics encoded in this object. It is needed in two instances: The first one concerns a zero-point shift of
the two-point function and is related to the quantization via the functional integral. Its removal
does not involve a physical scale and can thus be seen as a normalization of a certain observable,
the particle number. The second one is related to a true ultraviolet divergence which needs to be
cured by an appropriate counterterm. For the ultraviolet renormalization we can restrict to the
simpler situation where there is no spontaneous symmetry breaking, since this effect occurs only
in the low energy sector and cannot affect the ultraviolet physics. For the physical value of the
field expectation value, this implies = 0 and = .
The fermionic part of the particle number is naively obtained from the thermodynamic relation
n = U / in a homogeneous setting. This yields the explicit expression
=

n F, =
d 3q
tanh
(2)3
(21)
and we observe that this number may get negative for large negative /T . In order to clarify the
precise relation between n and the particle density, we first consider the simpler situation of a
can be related to
single fermionic degree of freedom. In this case the expectation value
the expectation values of products of the usual annihilation and creation operators a, a , which
obey the anticommutation relation a a + aa = 1,
1

1
1
1
(22)
= a a aa = a a = n .
2
2
2
2
Here the second equality holds since this combination of operators is covariant with respect to
permutations of the ordering. For a lattice model (as, for example, the Hubbard model) with
f degrees of freedom per site the fermion number per site therefore reads n = n + f/2. For
electrons in a solid (f = 2) one can associate n with the difference of the electron density from
half filling (where n = 1), i.e. the average number of electrons minus holes per site as compared
to the half-filling density. For relativistic charged fermions measures the difference between
particle and antiparticle density and the additive constant drops out. For non-relativistic atoms,
however, the relation between the atom density n and n becomes3

n F, (x) = (x)(x)

=
i (x)j (y) j (y)i (x) ij (x y)
2
y i,j

ai (x)aj (y) aj (y)ai (x) ij (x y)

=
2
=

i,j
3 Note that the constant shift n diverges if the ultraviolet cutoff for the momentum integration (q 2 < 2 ) goes to
infinity, n = 3 /(6 2 ).
216
1
=
2

2ai (x)aj (y) ij (x
y) ij (x y)
i,j
f

f
= a (x)a(x) (0) = nF (x)
2
2
= nF (x) n.
d 3q
(2)3
(23)
The volume factor in momentum space, (0), diverges in the limit of infinite momentum cutoff. The physical fermionic particle density nF (x) and the relative particle density n F, (x) =
(x)(x)
are therefore related by an additive shift that depends on the momentum cutoff.
In consequence, one finds now a manifestly positive total fermionic particle number

1
d 3q
3
NF = d x (n F, + n)
(24)
exp(2 ) + 1 .
=V
3
(2)
The momentum integral is now ultraviolet finite for . It becomes exponentially insensitive to the ultraviolet cutoff , such that we dropped the index. We recover the familiar Fermi
distribution. An analogous argument holds for the connected part of the bosonic two-point function which will be implemented below. Formally, this additive renormalization appears in the
form of field independent terms linear in and in the classical potential.
Let us now proceed to the second instance where UV renormalization is needed. The microscopic action (1) depends explicitly on two parameters , h . A third parameter is introduced
implicitly by the ultraviolet cutoff for the momentum integration in the fluctuation effects.
(Besides this, the results will depend on the thermodynamic variables T and , .) Contact to
experiment is established by relating the microscopic parameters to observables of the concrete
atomic system. Here we choose these parameters to be the magnetic field dependent physical detuning (B) and the Feshbach coupling h .4 The parameters ( , h ) can, however, be replaced
by an equivalent set (a 1 , h ) (a the scattering length) in a second step as pointed out below.
Once the parameters (B) and h are fixed by the properties of the molecules or atom scattering
in empty space we can proceed to compute the properties of the atom gas at nonzero temperature
and density without further free parameters.
In the vicinity of the Feshbach resonance at B = B0 we may approximate (B) by a linear
behavior (linear Zeeman effect)

(25)
= B .
B
Here B = + + M reflects the difference between the sum of the magnetic moments of
the two atomic states (+ + ) and the molecule magnetic moment M . We relate the physical
detuning to by an additive, B-independent shift
h 2 M

(26)
=
.
B
B
This is motivated by a consideration of the fermionic contribution to the boson mass term,
=
2 2
m
2 = U (0 = 0).
(27)
4 For vanishing background or open channel coupling, h is a free parameter. Else, a further UV renormalization of
the Feshbach coupling is necessary [42].
217
Indeed, the fluctuation contribution diverges for

m
2
1
2
h 2 d 3 q
q 2 /2M
q
=
tanh
3
2
2T
(2) 2M
1
2

h 2 d 3 q
q 2 /2M
2M
q
tanh
.
=
2
2M
2T
(2)3
q2
(28)
In the second equation the linear dependence of the fluctuation correction on the cutoff is
absorbed into the definition of the physical detuning . The remaining integral in the second line
of (28) is very insensitive with respect to the precise value of , and we can formally send to
infinity.
More generally, we choose such that the zero of coincides with the Feshbach resonance at B0
= B (B B0 ).
(29)
In vacuum (n = 0, T = 0) the Feshbach resonance corresponds to a vanishing binding energy

2 = 0. There is also no condensate in the vacuum, i.e.
M = 0. This is realized [15] for = 0, m
0 = 0. We note that there are no boson fluctuations contributing to the renormalization of the
mass term in the physical vacuum n = T = 0 as can be seen from a diagrammatic argument [42].
The scattering length a can be defined in terms of the scattering amplitude at zero momentum
and zero energy [15]. At the present stage we may consider a resonant scattering length aR
related to by
M
M
= 2 = 2 +
.
4aR
2 2
h
h
(30)
It accounts for the contribution of the molecule exchange to the atom scattering. For B = B0
and low enough momenta and energies (low enough temperature) the molecule exchange can
be described by an effective pointlike interaction. We therefore also define the renormalized
resonant four-fermion vertex R ,
R =
4aR
.
M
(31)
Comparison with Eq. (12) yields
M
1
M
1
= 2 = 2 +
=
+
.
2
R
h
h
2
2 2
(32)
This demonstrates that the renormalization of (26) corresponds to the renormalization of the
effective atom interaction strength .

It is instructive to consider in the mean field framework the explicit equation which determines
the order parameter 0 in the superfluid phase. A nonzero 0 obeys
2
1
=
4T
h 2

d 3 q 1
tanh 4MT /q 2 .
3
(2)
(33)
218
In an alternative purely fermionic formulation we may compute 0 in the local interaction approximation (12) by solving the SchwingerDyson equation [43,44]. Eq. (33) would correspond
precisely to the BCS gap equation (lowest order SchwingerDyson equation) for the purely
fermionic formulation with local interaction [45,46], provided we choose
1
2
=

h 2
or
1
2
.
=
h 2
(34)
This suggests the definition of a or density dependent effective coupling ()

(cf. Eq. (11)) for
the atom-molecule model. In the many-body context and in thermodynamic equilibrium, where
fermions and bosons share a common chemical potential = , the latter is determined by the
density. In the physical vacuum, obtained by sending the density to zero, describes the binding
energy of a molecule, M = 2 = 2 and does not vanish on the BEC side of the resonance
[15]. We emphasize that the resonant scattering length R in Eq. (31) describes the scattering
of fermions throughout the crossover and is directly related to the observed scattering length for
two atom scattering. On the other hand, ()

is a universal combination characteristic for the
ground state of the system [42]. For broad Feshbach resonances (h ) the two quantities
coincide.
Expressing the effective potential U (19) in terms of the momentum integral becomes ultraviolet finite,
= ( 2) + U (F ) (, ),
U (, )
1

h 2 M

d 3q
(F )

= 2T
U1 (, )
+
e
ln
e
.
(2)3
2T q 2
(35)
The remaining cutoff dependence is O(1 ), the precise value of therefore being unimportant.
For definiteness, the cutoff can be taken = (3 2 ngs )1/3 with ngs the density in the liquid or
solid ground state at T = 0. This choice will be motivated in Appendix B.
At this point we have reached a simple but nevertheless remarkable result. When expressed
in terms of measurable quantities, namely scattering length aR (or R ) and the open channel
atom density n F the dependence of the effective potential becomes very insensitive to the
microscopic physics, i.e. the value of the cutoff . Without much loss of accuracy we can take
the limit for the computation of U . The effective chemical potential depends on n F
and via U (, )/
= n F .
3.2. Wave function renormalization
The wave function renormalization Z is another important ingredient for the description of
the crossover physics in the Yukawa model. This can be seen from the fact that rescaling all
couplings in the effective action with the appropriate power of Z , we end up with an effective
bosonic Bogoliubov theory in the BEC regime, as will be discussed in detail in Section 7. Here
we focus on the explicit computation of the wave function renormalization.
As can be read off from Eq. (13), the wave function renormalization Z is related to the inverse
molecule propagator P . In Appendix C we compute P in the mean field approximation,
q2
P (Q) = 2inT +
h 2
4M

Q
219
PF (Q )
[PF (Q ) ][PF (Q ) ] + h 2
PF (Q + Q)
[PF (Q Q) ][PF (Q + Q) ] + h 2
= P (Q)
(36)
where the kinetic part of the inverse fermion propagator reads

PF (Q) = iB +
q2
2M
(37)
with Q = (B , q
), and the frequency variable B represents the discrete fermionic Matsubara
frequencies at finite temperature, B = (2n + 1)T .
Here we interpret the wave function renormalization Z as a renormalization of the term in
the effective action with a timelike derivative. We may then evaluate the propagator correction
(36) for analytically continued frequencies B B + i and set B = 0. Now P (, q
)
becomes a continuous function of . Defining

P
Z = 1
(38)
=0
one finds
Z = 1 +
h 2
16T 2
d 3 q
3 tanh cosh2 .
(2)3
(39)
Here we have rescaled the integration variable and the Yukawa coupling as q = q/kF , h 2 =
4M 2 h 2 /kF . In the symmetric phase the simplification = applies. We note that Z is closely
related to the spectral function for the molecules. If we only consider fermionic diagrams, we can
give an equivalent definition of the wave function renormalization using the mean field effective
potential,
Z = 1
1 3U
.
2
(40)
The property Z = Z holds since the integral in Eq. (36) depends on the combination 2
only.
3.3. Gradient coefficient
In order to compute the gradient coefficient, we proceed in complete analogy to the wave
function renormalization: For the spacelike momenta, we define
P (0, q) P (0, 0)
A (q) =
,
q2
A = lim A (q).
q 2 0
More explicit formulae are given in Appendix C in Eqs. (C.13), (C.14).
(41)
(42)
220
4. Relevant parameters, momentum and energy scales

4.1. Concentration
For a mean field computation with nonzero density the effective renormalized four-fermion
coupling is related by an equation similar to Eq. (12) to the resonant scattering length in vacuum.
We define5
1
2
M
+ 2.
=
( ) 4aR
h
(43)
Therefore the effective scattering length in an atom gas differs from the (vacuum) scattering
length which is measured by the scattering of individual atoms. This density effect is reflected
by the dependence on the effective chemical potential and we define
1
8
1
+ 2 .
=
a
aR h M
(44)
On the BCS side is positive and aR is negativethe size of |a|

is therefore larger than |aR |.
A similar enhancement occurs in the BEC regime where will turn out to be negative and aR
positive. Roughly speaking, on the BCS side the presence of a non-vanishing atom density favors
the formation of (virtual) molecules by reducing the cost of forming molecules with positive
(or positive binding energy) to 2 . In the BEC regime the presence of molecules reduces the
absolute size of the effective binding energy. It should, however, be pointed out that the effect of
a density dependence is small in case of broad resonances h , as visible from Eq. (43).
The atom density n defines a characteristic momentum scale by the Fermi momentum kF , i.e.
kF3
.
(45)
3 2
The inverse of the Fermi momentum defines the most important characteristic length scale in
our problem. Roughly speaking, it corresponds to the average distance d between two unbound
atoms or molecules. We emphasize that our definition of kF involves the total density n and
therefore includes unbound atoms, molecules and condensed atom pairs. In terms of kF we can
form a characteristic dimensionless concentration
n=
c = ak
F.
(46)
The concentration is a measure for the ratio between the in-medium scattering length a and
average distance, c a/d.
As mentioned in the introduction, the concentration is the crucial

parameter for the description of the crossover between the BEC and BCS regimes. For a small
concentration |c| the gas is dilute in the sense that scattering can be treated as a perturbation. In
this range mean field theory is expected to work reasonably well. On the other hand, for large
|c| the scattering length exceeds the average distance between two atoms and fluctuation effects
beyond mean field might play a crucial role.
An alternative definition of the concentration could use the measured (vacuum) scattering
length, c = a(B)kF . This definition has the advantage of a simple relation to the magnetic field.
5 For = this combination appears in the term quadratic in in Eq. (35).
221
Fig. 1. Crossover phase diagram Tc (c1 ). For large h we compare a calculation with a gap equation modified by boson
fluctuations (dashed) to the result obtained with the standard BCS gap equation (solid). The universal narrow resonance
limit h 0 is indicated by the dashed-dotted line. We further plot the result of the standard BCS approach (BCS gap
equation, only fermionic contributions in the density equation; long-dashed rising line) and the result for noninteracting
bosons (long dashed horizontal line). The solid line at resonance Tc (c1 = 0) = 0.292 coincides with the result obtained
in [47] which omits boson fluctuations. The dashed line Tc (c1 = 0) = 0.259 is in good agreement with the value
obtained in [48] (Tc = 0.266), who work in the broad resonance limit from the outset.
The choice between the two definitions is a matter of convenience. We adopt here the definition (46) since this reflects universality in an optimal way. For broad Feshbach resonances the
two definitions coincide since a = a(B). In presence of an additional background scattering
length, a(B) = abg + aR (B) we will include abg in the definition of c.
The concentration c is the most important parameter for the description of the crossover (besides T and n). The inverse concentration c1 corresponds to a relevant parameter which
vanishes at the location of the resonance. Once described in terms of c1 the ultracold fermionic
gases show a large universality. We demonstrate this universality on the crossover phase diagram in Fig. 1 which plots the critical temperature Tc = Tc /F for the transition to superfluidity
as a function of c1 . The narrow resonance limit h 0 is exact, while our results for broad
resonances (h ) still have substantial uncertainties, as demonstrated by two approximations that will be explained later. For large h the actual value of h becomes irrelevant and all
curves coincide with the broad resonance limit. Intermediate h interpolate between the broad
and narrow resonance limits.
We have argued in Section 2 that in the limit of a pointlike approximation for the effective
fermionic interaction all results should only depend on the effective scattering length. This only
involves the ratio /h 2 such that for fixed c the separate value of h should not matter. On the
BCS side for small |c| we therefore expect a universal behavior independent of the value of the
Yukawa coupling h . This is clearly seen in Fig. 1 where the critical line approaches the BCS
result independently of h . Furthermore, all results become independent of the value of h in the
broad resonance limit (h ). The concentration remains then as the only parameter (besides
T and n) for the description of the crossover. These new universal aspects, adding to those that
will be presented in Section 4.3, are discussed in [15]. In particular, the broad resonance limit
h corresponds to a pointlike microscopic interaction.
For the broad resonance limit the first approximation (solid line) corresponds to extended
MFT and neglects the modifications of the effective potential induced by the molecule fluctuations. The second approximation (dashed line) includes these fluctuation effects via the solution
222
Fig. 2. Gradient coefficient A = 2M A and wave function renormalization Z in dependence on c1 . We divide by h 2

in order to get numbers O(1) and use h 2 = 3.72 105 as appropriate for 6 Li [15]. The ratio A = A /Z is displayed
in Fig. 9.
of a gap equation for the molecule propagator (Section 8). The fast approach to the BEC value
for c1 0+ does not reflect the expected behavior Tc = TcBEC + c with a dimensionless constant . This shortcoming of our treatment is remedied by a functional renormalization group
study [49]. The solid line at resonance Tc (c1 = 0) = 0.292 coincides with the result obtained in
[47] which omits boson fluctuations. The dashed line Tc (c1 = 0) = 0.259 is in good agreement
with the value obtained in [48] (Tc = 0.266), who work in the broad resonance limit from the
outset.
In Fig. 2 we show the wave function renormalization Z and the dimensionless gradient coefficient A = 2M A as a function of c1 for T = 0. We use the large value for the Feshbach
coupling in 6 Li, h 2 = 3.72 105 for kF = 1 eV (cf. Eq. (47)). We note that for large h , as
appropriate for the Feshbach resonances in 6 Li or 40 K, the wave function renormalization Z is
large, the ratio Z /h 2 being an O(1) quantity. We observe a strong increase of A when evolving to the BCS side of the resonance. This strongly suppresses the propagation of the effective
bosonic degrees of freedom, leading to a situation where they are completely irrelevant. This is
an aspect of universality, where the system looses memory of the bosonic degrees of freedom
and a purely fermionic BCS-type description becomes appropriate. These curves are essentially
identical in the limit h 2 .
Indeed, large values of h 2 exhibit an enhanced universality [15]. In this case, all microscopic quantities depend only on the concentration c. In the limit h at fixed scattering
length there is a loss of memory concerning the details of the bosonic sector in the microscopic
action (1). These aspects are worked out in more detail in [15]. In a renormalization group treatment this universality will be reflected in strongly attractive partial infrared fixed points, similar
to the quark meson model in strong interactions [50]. This universality property will be valid
beyond mean field theory.
4.2. Dimensionless parameters
The characteristic energy scales for T , and the gap = h are set by the Fermi energy
F = kF2 /(2M). It is appropriate to define dimensionless quantities
2M
,
kF2
T
2MT
,
T = 2 =
F
kF
1/2
h = 2MkF h ,
A = 2M A .
q =
q
,
kF
2M h

=
,
F
kF2
223
r =
(47)
Once all quantities are expressed in this way in units of the Fermi momentum or the Fermi energy
the atom mass M will no longer be present in our problem. Indeed, in dimensionless units one
has for the expressions (20)
=

1 2
q ,
2T
2
1/2
1 2
q + r
2T
(48)
such that the atom mass M drops out in the computation (19), (35) of the appropriately rescaled
effective potential U . All dimensionless quantities in are typically of the order one. For practical computations we may therefore choose units kF = 1 eV.
The rescaled potential
u = kF3
T
U = 2MkF5 U
T
(49)
)
is composed of a classical contribution6 and a contribution from the fermion fluctuations (u (F
1 )
u =
r
(F )
+ u 1 ,
8 c
1
1

.
=
2
8 c 8c h MkF
(50)
Then the equation determining (r , T ) becomes

1 n F
u
= 2
(51)

3 n
and is indeed independent of M. It depends, however, on the ratio n F /n since we have defined
in Eq. (45) kF as a function of n rather than n F . In the mean field approximation the effective
chemical potential obeys, using the dimensionless shorthands (48),

d q q

2 n F
tanh 1 =
.
3 n
(52)
In particular, for r = 0 this determines as a function of T and n F /n.

For given , T and c one can compute the order parameter r in the low temperature phase
according to
u
= 0,
r
(F )
u 1
r
1
.
8 c
(53)
The critical temperature for the phase transition corresponds to the value Tc where r vanishes.
This part of the mean field computation is independent of the Yukawa coupling h . In Section 5
we will determine n F /n as a function of , T and c such that Eq. (51) can be used to determine
as a function of c and T . As an alternative, we will modify in Section 6 the definition of U
such that a modified equation u/
= 1/3 2 becomes independent of n F /n. Again, can
6 In the formulation of Section 6 c will be replaced by c since and will be identified.
224
now be fixed as a function of c and T . We will see that the relation (c, T ) depends on the choice
of the Yukawa coupling h . The results will therefore depend on the additional dimensionless
parameter h (47). Away from the narrow and broad resonance limits, the model is therefore
characterized by two dimensionless quantities, c and h .
4.3. Universality
All observables can be expressed in terms of c, T and h . For example, using the definition
=
2M
,
kF2
(54)
one finds the relation

2 =
h 2
8c
(55)
In particular, the phase diagram Tc (c) depends only on h as shown in Fig. 1 for the case of
narrow and broad resonances. For T = Tc we find that the relation (c) depends only mildly on
the value of h as shown in Fig. 3 where we compare narrow and broad resonance limits again.
Furthermore, for small |c| the universal curves Tc (c) and (c, T ) become independent of h :
Both in the BEC and BCS regime, an enhanced universality sets in, making h irrelevant for all
its possible values! These issues are investigated more systematically in [15].
In summary we have now an effective low energy formulation where neither M nor enter
anymore. Everything is expressed in terms of kF and three dimensionless parameters, namely
c, h and T . This scaling property is an important aspect of universality. In [15] we discuss the
relation of the parameters c and h with the physical observables for an experimental setting,
namely the magnetic field B and the binding energy in vacuum. In the present paper we treat h
as a free parameter. The values of h for 6 Li and 40 K turn out to be large such that the broad
resonance limit applies to these systems.
Fig. 3. Crossover at the critical temperature in the broad resonance limit: Effective dimensionless chemical potential
at the critical temperature as a function of the inverse concentration c1 . We compare the results for two versions of
the gap equation as in Fig. 1 (solid and dashed line). Additionally, the result for the narrow resonance limit is indicated
(dashed-dotted).
225
Table 1
(3.7290 103 aB )1 )
Typical values for the dimensionless scale ratios for 6 Li and 40 K (kF = 1 eV =
M/kF
/kF
F /kF
6 Li
5.65 109
1.6 103
40 K
40.0 109
1.2 103
8.9 1011
1.2 1011
We display in Table 1 the values of the dimensionless scale rations M/kF , /kF and F /kF .
The first two give an idea how well the detailed microphysics decouples for realistic ultracold
fermionic gases. We use a density n = 4.4 1012 cm3 , kF = 1 eV =
(3.7290 103 aB )1 .
5. Molecule and condensate fraction
At this point the functional integral setting for the ultracold fermionic atom gases is fully specified. The parameters aR and h 2 can be extracted from a computation of two-atom scattering in
the vacuum, taking the limit T 0, n 0 [15]. Rescaling with appropriate powers of kF yields
the parameters c and h 2 for the many body system. The relation between and n is determined
by Eq. (14) and we identify = at the end of the computation. An approximate solution of
the functional integral for the effective action gives access to thermodynamic quantities and
correlation functions.
The relation (14) between and n seems to be rather formal at this stage. In this section we
will develop the physical interpretation of this formula. In this context the important distinction
between microscopic and dressed molecules will appear. In a quantum mechanical computation
for the physics of a Feshbach resonance the concept of dressed molecules arises from mixing
effects between the open and closed channels. This is an important ingredient for the understanding of the crossover. Our functional integral formalism has to reproduce this channel mixing
in vacuum and to extend it to the many body situation. In the functional integral formulation
the quantum mechanical mixing effects are closely related to the wave function renormalization Z . The concept of dressed molecules and their contribution to the density is directly related
to the interpretation of Eq. (14). We stress however, that the functional integral evaluation of n
for given = can be done completely independently of this interpretation.
5.1. Exact expression for the bare molecule density
The total density of atoms is composed of three components according to Eq. (7),
n = n F + 2n M + n C .
(56)
Here we have split up the contribution n B from (7) in a contribution from uncondensed bare
molecules (connected two-point function)

n B
n M =

= c n B
(57)
and one from the condensate which only occurs below the critical temperature Tc

=
n C = 2
kF4
r
r = 2kF3 2 .
2
2
2h M
h
(58)
226
The ratio of atoms in the condensate arising from bare molecules is defined as
n C
6 2
C =
= 2 r
n
h
(59)
1/2
and involves the Yukawa coupling h = 2M h /kF .
The density of uncondensed bare molecules n M (57) can be written in terms of the bare
molecule propagator or connected two-point function

(x, ; x, ) n B .
n M (x) = T d G
(60)
This formula is exact.7 It involves an additive renormalization n B of a similar origin as for the
fermionic density, cf. Section 3.1, but with opposite sign due to Bose-statistics. In the homogeneous limit we can write Eq. (60) as a sum over integer Matsubara frequencies 2nT
d 3q
(q, n) n B .
n M = T
(61)
G
(2)3
n
The ratio of bare molecules to open channel atoms, n M /n F , is important for the understanding
of the atom gas. Technically, it enters the field equation for the effective chemical potential (51)
which involves n F /n. In the regions in parameter space where n M /n is not very small a reliable
computation of n M is crucial for a quantitative understanding of the phase diagram. Such situations are e.g. realized in the BEC regime for narrow and intermediate resonances. However, for
the crossover region for a broad Feshbach resonance (as for 6 Li and 40 K) it turns out that n M /n
is small such that an approximation n F = n yields already a reasonable result.
Nevertheless, the mean field approximation (52) does not remain valid in all regions of the
phase diagram. This is due to an additional -dependence of u arising from the boson fluctuations which are neglected in MFT. We will argue below that this contribution from the bosonic
fluctuations can be interpreted as the density of dressed molecules. Thus, again an estimate of the
molecule density will be mandatory for the understanding of the crossover physics. A simple estimate of the dressed molecule density is the minimal ingredient beyond mean field theory needed
for a qualitatively correct description. We will refer to this as extended mean field theory.
As a first step the evaluation of the molecule density we may evaluate the classical approximation where the Yukawa interaction between open and closed channel atoms is neglected. This
(cl)
corresponds to the limit h 0 (for fixed a).
Then G (q, n) is the free propagator8

1
q2
(cl)
G (q, n) = 2inT +
.
(62)
+ 2
4M

= 1/2 d 3 q/(2)3 one obtains the faPerforming the Matsubara sum and inserting n B = n/2
miliar expression involving the occupation numbers for bosons
1
(cl)

2
P (q)
d 3q
(cl) (q) = q + 2.
n M =
(63)
,
P
1
exp
T
4M
(2)3
7 Eq. (60) is valid for the normal phase T T . For the superfluid phase G
becomes a 2 2 matrix and the correc
sponding generalization of Eq. (60) is discussed in Section 5.8.

8 Note that we use instead of this includes already the dominant fluctuation effects as motivated by the following
paragraphs. Strictly speaking, the classical propagator features .
227
We note the role of n B for the removal of pieces that do not vanish for .
5.2. Fluctuation effects
(q, n). Details of the computation of G
=
The fluctuation effects will change the form of G
1
P are presented in Appendix C. In the symmetric phase we may account for the fluctuations
by using

P (Q) = 2inT Z (, T ) + P (q) ab ,

2 (, T ).
P (q) = A , T , q 2 q 2 + m
(64)
We include here the fluctuations of the fermions (open channel atoms) for the computation of A
and Z . The mass term m
2 will be evaluated by a gap equation which also includes the bosonic
molecule fluctuations. In the superfluid phase ( = 0) the diagonalization of the propagator has
to be performed carefully as discussed below in Sections 5.7 and 5.8.
The gradient coefficient A (, T , q 2 ) is defined by Eq. (41) and depends on the Yukawa
coupling h . For large q 2 it comes close to the classical value 1/4M. We plot the renormalized
gradient coefficient A (q 2 ) = 2M A (q 2 )/Z for different c corresponding to the BCS, BEC
and crossover regimes in Fig. 4. It is obvious that this gradient coefficient plays a major role.
Large A leads to an additional suppression of the occupation number for modes with high q 2 ,
as anticipated in Section 3.3.
In the symmetric phase ( 0 = 0) the mass term is given by

2 U
2
m
=
(65)
.
=0
We use here a language familiar in quantum field theory since the molecule wave behaves
like a massive field for m
2 > 0. (The propagator G has a gap.) For m
2 > 0 the symmetric
solution of the field equation, 0 = 0, is stable whereas it becomes unstable for negative m
2 .
For a second order phase transition the critical temperature therefore corresponds precisely to
Fig. 4. Momentum dependence of the gradient coefficient in the broad resonance regime h 2 1. We plot A = 2M A /
Z for T = Tc in the different regimes, BCS (c1 = 1.5, Tc = 0.057, topmost), crossover (c1 = 0, Tc = 0.29) and
BEC (c1 = 1.5, Tc = 0.218). The dashed line is the value for elementary pointlike bosons, A = 1/2.
228
a vanishing mass term, m

2 (T = Tc ) = 0. The mass term reads in the mean field approximation
(F )2
m

(F )
2 U1
2 UMF T
=
= 2 +
.
=0
=0
(66)
(F )2
For T = 0, = 0 the definition of (26) implies m

= 2. We note that the fermion
2
fluctuations lower m
as compared to the microscopic term m
2, = 2, cf. Eq. (19). This
effect enhances the occupation number for moleculesusing m
2, instead of m
2 would yield a
much too small density of molecules! At the critical temperature the mass term m
2 vanishes (see
below). For T = Tc the fluctuation effects therefore concern the size and shape of A (q).
As we have seen in our mean field computation in Section 3 the quantities Z , A and m
2
depend strongly on . We may imagine to integrate first the fermion fluctuations in the functional integral (6) (with = = 0). The result is an intermediate mean field action for the
remaining functional integral over bosonic fluctuations. The quadratic part (Q)(Q) in this
action will be of the type of Eq. (64). Performing now the boson fluctuations will induce an additional contribution to U/ = n F . This will be interpreted below as the density of dressed
molecules.
5.3. Dressed molecules
Dressed molecules are quasi-particles with atom number two. They are described by renormalized scalar fields with a standard non-relativistic -derivative in the effective action. This
allows for a standard association of the number density of quasi-particles with the correlation
function for renormalized fields. With the effective action (13) the relation between the fields for
dressed and bare molecules reads
1/2
R = Z .
(67)
Correspondingly, the dressed molecule density nM becomes

2nM
(68)
n
and we find for large Z a very substantial enhancement as compared to the bare molecule
density n M .
We may define a renormalized gradient coefficient A and mass term m2 by
nM = Z n M ,
A
,
A ,R =
Z
M =
m
2,R
m
2
Z
(69)
or, in dimensionless units,

A =
A
,
Z
m2 =
m
2
Z
2M 2
m
.
kF2 Z
(70)
Then the quadratic part in the effective action for the bosons can be written in terms of and
m2 , A , without explicit reference to Z . (Similar rescalings can be made for any quantity entering our calculations. For a complete list of dimensionful, dimensionless and dimensionless
renormalized quantities, cf. Appendix F.) Using Eq. (64) in Eq. (61) the wave function Z can
229
be factored out in P such that

1

P,R (q)
d 3q
nM = Z n M =
(71)
.
1
exp
T
(2)3
This is a standard bosonic particle number without the appearance of Z , now expressed in terms
of the effective renormalized inverse boson propagator (dimensionful and dimensionless version
displayed)
P (q)
= A ,R q 2 + m
2,R ,
P,R (q) =
Z
2M P (q)
P (q) = 2
= A q 2 + m2 .
kF Z
(72)
The dressed molecules [31,38] include both bare and fluctuation induced, effective molecules
(cf. Eq. (38)). We will see how the dressed molecule density nM emerges naturally in the equation
of state for the particle density below.
We plot in Fig. 5 the mode occupation number for the dressed molecules (q = |
q |/kF )

1

1
NM (q)
(73)
q | /T 1
= exp P,R |
= exp P (q)/
T 1
(weighted by a volume factor q 3 ). There we compare the classical case P = q 2 /4M with the
result including the fluctuation corrections. The normalization in the figure reflects directly the
relative contribution to nM

kF3
d(ln q)
q 3 NM (q).
nM =
(74)
2 2
We observe a large fluctuation effect in the BCS regime (beyond the renormalization of m
2 ). In
this regime, however, the overall role of the molecules is subdominant. On the other hand, in the
BEC regime the molecule distribution is rather insensitive to the details of the treatment of the
fluctuations. The most important uncertainty from the molecule fluctuations therefore concerns
the crossover regime. In the BEC and crossover regime the replacement of A (q 2 ) by A (q 2 = 0)
results only in a moderate error. For simplicity we neglect the momentum dependence of A (q 2 )
for the numerical results in this work.
5.4. Contribution of molecule fluctuations to the effective potential
The computation of nM evaluates a one loop integral which involves the molecule fluctuations. (Graphically, Eqs. (61), (63) correspond to a closed loop for the molecule fluctuations
with an insertion of a -derivative.) A self-consistent approximation should therefore also include the effects of the molecule fluctuations in the computation of U and therefore 0 or Tc .
Our functional integral approach makes this necessity particularly apparent: The computation
of the partition function Z involves the fluctuations of open channel atoms and molecules in a
completely symmetric way (the variables and in Eq. (1)). There is no reason why the fluctuations of the fermionic atoms should be included and not the ones for the bosonic molecules.
In particular, the critical region very close to Tc will be dominated by the boson fluctuations.
For the effective potential the incorporation of the molecule fluctuations is achieved by adding
(B)
to U the one loop contribution U1 from the fluctuations of
(B)
U = UMF T + U1
(F )
(B)
= ( 2) + U1 + U1 .
(75)
230
(a)
(b)
(c)
T =T
Fig. 5. Weighted Bose distribution NM c (q)
= (exp(A q 2 /T ) 1)1 as a function of dimensionless momentum
q = q/kF , for T = Tc . We show the three regimes: (a) BEC (c1 = 1.5, Tc = 0.218), (b) Crossover (c1 = 0,
Tc = 0.292) and (c) BCS (c1 = 1.5, Tc = 0.057). We compare our best estimate using A (q)
= A (q)/Z
(solid
curve) with the approximation A = A (q = 0) (dashed curve) that we employ for the numerical estimates in this paper.
Also indicated is the result for the classical gradient coefficient A = 1/2 (dashed-dotted) that strongly overestimates the
molecule number in the crossover and BCS regimes.
231
We can construct the bosonic contribution as the leading order correction to the mean field result which omits boson fluctuations completely. For this purpose, we note that the fermion field
appears only at quadratic order in the classical action (1). We can therefore integrate them out,
which turns Eq. (6) into a purely bosonic functional integral,

[ = 0, ] = ln D exp SMF T [ + ] +
(76)
j + h.c.
with an intermediate action SMF T depending on the field = + . This is given by the exact
expression
= S (cl) []
1 ln det S () []
= S (cl) []
1 Tr ln S () []
SMF T []
2
2
(77)
where S () denotes the second variation w.r.t. the fermion fields. In the classical approxima = SMF T [],
while the one loop approximation corresponds to an expansion
tion one has []
of SMF T [ + ] to second order in .

The mean field effective potential (19) is obtained from Eq. (76) in the classical approximation. The next order contribution takes the Gaussian approximation for the fluctuations of
the molecule field into account. In principle, this requires the evaluation of highly nonlocal
objectsthe one-loop fermion fluctuations encoded in the Tr ln-term in Eq. (77) feature a complex frequency and momentum dependence. However, since we are interested in the observable
low energy properties of the system, we may apply a derivative expansion which only keeps the
leading order terms in the frequency and momentum dependence. This precisely generates the
wave function renormalization Z (38), (39) and the gradient coefficient (41), (C.13). In this approximation, we find (up to an irrelevant infinite constant and evaluated in the symmetric phase)
the one loop result

d 3q
(B)
U1 = T
(78)
ln1 eP,R /T ,
3
(2)
where the spacelike part of the boson propagator is precisely given by Eq. (72) in the symmetric phase (for the result in the symmetry broken phase, cf. Eq. (104)). This one loop formula
has shortcomings and we will improve on it by solving appropriate gap equations in Section 8.
Nevertheless, it already contains the essential information for the different contributions to the
density.
Let us compare the effect of the - and -derivatives on the bosonic part of the effective
potential, Eq. (78). With the classical inverse boson propagator P(cl) = q 2 /4M + 2 one
has P / = 2 and we note the simple relation
U1(B)
= 2n M .
(79)
(B)
On the other hand, the fermion loop corrections induce a -dependence (at fixed ) of U1 ,
which contributes to n F . This contribution can be interpreted as the number of open channel
atoms that are bound in the dressed molecules

U1(B)
2nF M =
(80)
.

232
The total number of dressed molecules is then given by

nM = n M + nF M = Z n M .
(81)
This could be taken as a possible alternative definition of Z

(B)
Z 1 =
U1 /
nF M
=
.
(B)
n M
U1 /
(82)
In the limit where the -dependence of Z and A can be neglected and P / = 2 one finds
from
2
P = A q 2 + m
(83)
(cf. Eq. (64)) the relation

Z 1 =

3
2

P /
1 m
= 1 U .
=

2
2
P /
(84)
We recover the MFT result in Eq. (40). The combined effect of the derivatives with respect to
and yields directly the number of dressed molecules
(B)
(B)
U1
1 U1
+
nM = Z n M =
(85)
.
2

The quantitative results shown in the figures are obtained by including the - and -dependence
of P . They will be discussed in more detail in the next sections.
5.5. Open and closed channel atoms
Some characteristic properties of the ultracold gas involve the fractions of open channel atoms,
uncondensed bare molecules and condensed bare molecules
n F
2n M
n C
F =
,
M =
,
C =
,
n
n
n
F + M + C = 1.
(86)
For example, the sum B = M + C measures the fraction of closed channel atoms, as observed in [8]. The formal use of two different effective chemical potentials in the action SB (1),
i.e. for the fermions and for the bosons, allows the simple association

U
U
2
2
3 F =
(87)
,
3 B =
,

(B)
and we recall that F also receives a contribution from U1 . We have computed B in [15] and
the results agree well with [8] over several orders of magnitude.
In the symmetric phase for T Tc one has F + M = 1. For small h 1 the BEC
BCS crossover is the crossover from small to large F . In Fig. 6(b) we plot F and M
as a function of the inverse concentration c1 for T = Tc . The modifications of Z , A and
m
2 in the inverse propagator P (64) depend on the Yukawa coupling h . However, this influences the precise shape of the crossover between the BEC and BCS regime only for moderate h 1. For smaller h , the density fractions are insensitive to the fluctuation modifications.
233
(a)
(b)
Fig. 6. Crossover at the critical temperature: Contributions to the total particle number, showing the crossover from
fermion to boson dominated physics. (a) Fractions of dressed densities in the large h limit. We compare the results
for two versions of the gap equation as in Fig. 1. (b) Fractions of bare densities in the exact narrow resonance limit
h 0. Though the pictures are similar, the physical interpretation of the two plots differs as described in the text.
For the large values of h encountered for the broad Feshbach resonances in 6 Li and 40 K
the contributions from the closed channel molecules M , C become very small (cf. Fig. 7(b)).
The dressed molecules differ substantially from the bare molecules (large Z ) and the crossover
physics is better described in terms of dressed molecules. We display in Fig. 6(a) the fraction
of dressed unbound atoms F and dressed molecules M for large values of h and T = Tc .
The fractions are not sensitive to the precise value of h in the broad resonance limit h ,
similar to the behavior found for small h .
5.6. Condensate fraction
The total number of atoms in the condensate depends on the expectation value of the renor1/2
malized field R = R = Z .
We will mainly use the dimensionless field
3/2
= kF
3/2
R = kF
1/2
Z .
(88)
234
(a)
(b)
Fig. 7. (a) Contributions to the total particle density for T = 0 in the large h limit: The fraction of dressed molecules
M (dashed line) is largest in the crossover regime. The condensate fraction C (solid line) grows to one in the BEC
regime. The solid line corresponds to C = Z C whereas the dashed-dotted line uses Z (Eq. (96)) instead of Z .
(b) Fractions of the bare or closed channel molecules. In contrast to the dressed molecules, they are O(h 2 ). The
dominant contribution arises from the condensed bare molecules C (solid line). The contribution from non-condensed
bare molecules M at T = 0 remains very small.
With = , 0 = 0 0 the dressed condensate fraction can be defined as

C =
2 R R 2kF3 0 0
=
= 6 2 0 = Z C .
n
n
(89)
In Fig. 7(a) we plot the condensate fraction C as a function of c1 for T = 0. We also show
the fraction of closed channel atoms9 in the condensate, C , in Fig. 7(b). Both correspond to the
broad resonance limit and we choose the Yukawa coupling appropriate for 6 Li, h = 610.
For large Z one has C C indeed, the probability Z1 that a condensed di-atom state
contains a bare molecule is small. In this case the major part of the condensate is due to open
9 The figure for
C includes renormalization effects for h discussed in [15].
channel atoms, i.e. nC is dominated by a contribution from n F

C
F
nF C = nC n C = (Z 1)n C =
1
n F .
C + M
F
235
(90)
Here the last term defines implicitly F by the requirement

F + M + C = 1.
(91)
Z1
1 one concludes F 1, and for low enough T

(From nC + nM n, C + M
one further expects M < C .) In the BCS limit this result is not surprising since we could
have chosen a formulation without explicit molecule fields such that all atoms are described
by n F . (If the chemical potential multiplies only the total condensate fraction must be
C = 1 F .)10
The total number of open channel atoms can therefore be found in three channels, n F =
nF + nF C + 2nF M . Here nF denotes the unbound dressed fermionic atoms, 2nF M the open
channel atoms contained in dressed molecules and nF C the ones in the condensate. As long as
the fermionic and bosonic contributions to U can be separated we have the identities
(B)
(F )
(B)
U
U1
1 U1
nF,0 = nF + nF C = 1 ,
(92)
+
nM =
,
2

and
n = nF,0 + 2nM + n C = nF + 2nM + nC = n F + 2n M + n C .
(93)
The definition of a condensate fraction in terms of the superfluid order parameter 0 is rather
simple and appealing. Nevertheless, this may not correspond precisely to the condensate fraction
as defined by a given experimental setup. The ambiguity is even larger when we come to the
concepts of uncondensed fermionic atoms and molecules. The distinction becomes somewhat
arbitrary if we include higher loops for the computation of it, where bosonic and fermionic
fluctuations are mixed. As an example for the ambiguities in the definition one may try to extract
the number of open channel atoms in the condensate directly from the -dependence of n F . We
can decompose the fermion contribution into a part for vanishing condensate and a condensate
contribution due to = 0
(F )
U1
= U1 ( = 0) + U1 .
(F )
(F )
(94)
The association
(F )
U1
yields an alternative definition of Z ,

nF C =
Z 1 =
(95)
(F )
nF C
1 U1
=
n C
2
(96)
10 The difference between and 1 can be traced back to the appearance of the chemical potential in the
C
F
effective four-fermion interaction (11).
236
With this definition the number of unbound atoms nF becomes

U1 ( = 0)
(F )
nF = n F nF C nF M =
(97)
which amounts to the standard number density in a Fermi gas. We note that Z (96) coincides
(F )
with Z (84) in the limit where 3 U/ is dominated by the term quadratic in in U1 .
From Fig. 7(a) we see that this is the case in the BEC and BCS regimes.
5.7. Excitations in the superfluid phase
The bosonic excitations in the superfluid phase are analogous to a purely bosonic theory as
for superfluidity in 4 He [51]. Due to the non-vanishing order parameter the matrix for the inverse
renormalized propagator P contains now off-diagonal entries form terms 0 or 0
.
It is convenient to use a basis of real fields 1 , 2
where 0 = kF3 Z
1
= (1 + i2 ),
2

1 2
1 + 22 .
2
(98)
In this basis P and G = P1 are 2 2 matrices and the exact Eq. (61) for the bare molecule
density is replaced by

T
(x, ; x, ) n B .
n M (x) = tr d G
(99)
2
We expand in the superfluid phase around the minimum of the potential at = 0
u =
( 0 )2 +
2
such that the mass matrix

2
2 u
m ab =
a b =0
(100)
(101)
becomes

m2 =
2 0
0

0
.
0
(102)
Without loss of generality we have taken here the order parameter in the 1 direction, 1,0 =
20 , 2,0 = 0, and we recognize the flat direction in the potential (vanishing eigenvalue of m2 )
in the Goldstone direction 2 . In contrast, the radial mode 1 has a non-vanishing mass term
2 0 .
In the basis (1 , 2 ) the term containing the -derivative is off-diagonal (neglecting total
derivatives)

d = i d 1 2 = i d 2 1 .
(103)
237
In momentum space one therefore finds for the renormalized inverse propagator P =
2M/(Z kF2 )P 11

A q 2 + 2 0
P =
(104)
A q 2
where we use Eq. (70) and the renormalized order parameter and quartic coupling
0 = Z 0 ,

Z2
(105)
(For a list of the relations between dimensionless and dimensionless renormalized parameters
cf. Appendix F.) This has an important consequence: The propagating excitations correspond to
frequencies which obtain from the Matsubara frequencies B by analytic continuation B
i. This corresponds to the analytic continuation from Euclidean time to real or Minkowski
time t = i . Now G has a pole or P a zero eigenvalue. The eigenvalues of P therefore
obey

A q 2 + 2 0 A q 2 2 = 0.
(106)
Vanishing eigenvalues therefore lead to the dispersion relation

2 = A q 2 2 0 + A q 2 .
(107)
For small q 2 this yields the linear dispersion relation characteristic for superfluidity

= 2A 0 q 2 ,
(108)
from which we can read off the speed of sound

vs =
kF
2A 0 .
2M
(109)
5.8. Molecule density in the superfluid phase

The density of dressed molecules in the superfluid phase obeys

T
d 3 q 1
nM = Z n M = kF3 tr
P (q,
n ) n B
2
(2)3
n
d 3 q
A q 2 + 0
n B ,
= kF3 T
(2)3 n2 + A q 2 (A q 2 + 2 0 )
n
(110)
where the Matsubara summation can be performed analytically

A q 2 (A q 2 + 2 0 )
kF3
A q 2 + 0
d 3 q

nM =
1 . (111)
coth
2
(2)3
2T
A q 2 (A q 2 + 2 0 )
Due to the subtraction of n B (the term 1 in the curled bracket) the momentum integral is UV
finite.

11 Note the structure 1/2

T
2
q (Q)P (Q)(Q).
238
We have used dimensionless units and may introduce

(A q 2 + m2 )/(2T ) symmetric phase,
=
superfluid phase,
A q 2 /(2T )

0
symmetric phase,
=
superfluid phase,
0 /(2T )
where (for SSB)

= A q 2 (A q 2 + 2 0 )/(2T ) = 2 + 2
(112)
(113)
(114)
is the bosonic analog to Eq. (20). This allows us to write the dressed molecule density in both
phases as

d 3q
(SYM)
nM
(115)
=
(exp 2 1)1 ,
(2)3

1
d 3q +
(SSB)
nM =
(116)
coth
1
.
(2)3
At the phase boundary the two definitions coincide since = (m2 = 0, 0 = 0 and = 0).
Eq. (115) for the density of dressed molecules involve the renormalized coupling . This
describes the full vertex and therefore is a momentum dependent function. This also holds
for A . We will neglect the momentum dependence of A , as motivated by Fig. 5 for T = Tc .
For this issue is more involved since (q) vanishes for q 0 (T Tc ) due to the molecule
fluctuations, as shown in Appendix E. We observe that for small T the momentum integration
in Eq. (111) is dominated by the range q 2 0 /A . The infrared suppression of (q 0)
is therefore not effective. For the density equation we will simply omit the contribution of the
(F )
(F )
molecule fluctuations to and approximate = with evaluated at q 2 = 0.
The inclusion of the molecule density nM is important for quantitative accuracy even at T = 0.
This is demonstrated in Fig. 8 where we show the crossover for the effective chemical potential
as a function of c1 . The agreement with quantum Monte Carlo simulations
and the gap
[37] is substantially improved as compared to mean field theory, and also compares reasonably
= 0.53, = 0.445).
well with other analytical approaches (Strinati et al. [52],
6. Effective field for atom density
Before proceeding in Section 8 to a description of our computation of the effects from the
molecule fluctuations on the effective potential we present in this section an improvement of
our formalism which treats the fermionic and bosonic fluctuations in an even more symmetric
way. Indeed, in the thermodynamic equilibrium the molecule propagator should involve the same
effective chemical potential as the propagator of the unbound atoms. (So far it involves
instead of .)
If one is interested in the separate contributions from open and closed channel atoms one may
formally consider different chemical potentials and multiplying and in the action.
Then n F and n B can be associated to the variation of the effective action with respect to and .
At the end of the computations one has to identify = . For many purposes, however, one only
needs the total number of atoms n. It seems then advantageous to modify our formulation such
that the field is associated to n instead of n F . In this section, we treat as a classical field such
239
(a)
(b)
Fig. 8.
Solutions of the coupled gap and density equations at T = 0, (a) effective chemical potential, (b) gap parameter
= r . We also show (dashed) the mean field result (obtained by setting the bosonic contribution nM = 0 in the density

= 0.54.
equation). Our result can be compared to QMC calculations [37] performed at c1 = 0 which find = 0.44,
= 0.53, and improves as compared to the MFT result = 0.63,
= 0.65.
Our solution yields = 0.50,
that its role is reduced to a source term = J . In Appendix A we will consider the more general
case where is treated as a fluctuating field.
Identifying = in the bare action (1), the -derivative of the effective potential generates
the total particle number density,
= n = n F + 2n M + 2 .
(117)
In the presence of interactions it is hard to evaluate the full bare correlation functions in the above
form explicitlyit premises the solution of the full quantum field theory. As we have discussed
in Sections 5.4, 5.5 it is more practicable to decompose n according to
(B)
n=
(UMF T + U1 )
= n C + nF,0 + 2nM
(118)
240
which is a mixed representation involving both bare and renormalized quantities. In the next
section we will see that this equation reduces to an equation of state for fundamental bosons in
the BEC regime, which involves dressed quantities only (Eq. (124)).
(B)
Beyond the classical bosonic propagator we have implemented12 the -dependence of U1
by the -dependence of m
2 . This yields
1 m
1 3U
=
2
2
2
Z =
(119)
where we note that the contribution from the bare molecules is already included in the classical
-dependence of m
2 , leading to the replacement Z Z 1 in Eq. (84). In practice, we use
the approximation
(F )
1 3 U1
Z = 1
.
2
(120)
This has a simple interpretation: due to the fermionic fluctuations the classical cubic coupling
We can compute nM directly from
2 is replaced by a renormalized coupling 2Z .
Eqs. (74), (73), (71), (64).
7. BEC limit
In Section 5.7 we have defined a multiplicative renormalization scheme for the bosons by
rescaling all couplings and fields with an appropriate power of Z such that the timelike derivative in the effective action has a unit coefficient. This leads to the notion of dressed molecules in
our formalism. In this section we discuss the implications of this prescription in the BEC regime.
Here we will see that the dressed molecules behave just as elementary bosons.
The propagator for the dressed molecules has still a nontrivial renormalization factor for its
dependence on the spacelike momenta (64). The ratio A = A /Z is shown as a function of c
in Fig. 9. It is instructive to evaluate A in the BEC limit where the propagator should describe
the propagation of dressed molecules. The integrals in A and Z can be evaluated analytically
(cf. Appendix D, Eqs. (D.3), (D.4)) irrespective to the thermodynamic phase of the system. The
universal result does not depend on h ,
1
A .
2
(121)
This makes it particularly clear that for both the limit of small h 0 and the limit of large h
we recover composite, but pointlike bosons with wave function A = 1/2, corresponding to a
kinetic energy p 2 /4M. However, in the first case of small h we deal with microscopic closed
channel molecules, while in the second case of large h the emerging bosons are open channel
pairs, however behaving just as if they were pointlike bosons in the BEC limit. For large h we
could equally well drop the classical piece in A and Z . This corresponds to a purely pointlike
interaction for the fermions without any explicit reference to molecules at all, cf. Eq. (11), and the
discussion in [15]. Nevertheless, bound atom pairs emerge that behave just as pointlike particles.
12 A more rigorous derivation of the equation of state via the Noether construction for the conserved charge associated
to the global U (1)-symmetry reveals that this choice is uniquely fixed by requiring consistency within our truncation
(linear frequency dependence) [42].
241
Fig. 9. Gradient coefficient A = A /Z for T = 0, as a function of the inverse concentration. In the BEC limit, A
takes the classical value 1/2 as appropriate for elementary bosons of mass 2M.
A similar aspect of universality is found for the four-boson coupling = /Z2 . In the
approximation where we neglect bosonic contributions to the four-boson coupling and to the gap
(F )
equation for the mass (BCS gap equation), we find = = 8/ such that the bosonic
scattering length becomes aM = 2aR . This is the Born approximation for the scattering of the
composite bosons. However, more accurate treatments on this issue have shown aM /aR = 0.6.
This result is obtained from the solution of the four-body Schdinger equation [53] and form
a numerically demanding diagrammatic approach [54], and has also been confirmed in QMC
simulations [55]. A resummation of the effective boson interaction vertices in vacuum as done
in [56] yields aM /aR = 0.75(4). Our present SchwingerDyson approach does not correctly
account for the real situation since the contribution from molecule fluctuations in the vacuum
is missing. In the frame of functional renormalization group equations, this deficiency has been
remedied. Simple truncations yield aM /aR = 0.710.92 [49].
It is instructive to investigate Eq. (115) in the limit c1 (BEC limit) where we have
A 1/2, aM = 2aR . The density in the superfluid phase nM (115) then coincides precisely
with the Bogoliubov formula for bosons of mass 2M and the above scattering length,

|uq |2 + |vq |2
d 3q
1
2
|
+
|v
nM =
(122)
q
2
exp 2 1
(2)3
where we used the relations connecting the Bogoliubov transformation coefficients and our expressions (112), (114),

1 +
+
1 ,
|uq |2 + |vq |2 =
.
|vq |2 =
(123)
2
We emphasize again that we can perform first the limit h where we recover a purely
fermionic model with pointlike interaction and no explicit molecule degrees of freedom (cf.
[15]). Subsequently we may consider large c1 where the approximations leading to Eq. (122)
become valid. This shows that the Bogoliubov formula for weakly interacting fundamental
bosons can be recovered from a purely fermionic model! In our approach, this result emerges in
the simultaneous limit c1 (BEC regime), h (broad resonance regime). A similar
result has been established by Strinati et al. [1820,56], who work in a purely fermionic setting,
or, in our language, in the broad resonance limit h from the outset.
242
This observation is strengthened by an investigation of the dimensionless density equation (118) at T = 0 in the BEC limit:

r
r
1 = 3 2
(124)
+ 2nM = 3 2 2Z 2 + 2nM = C + M .
16
h
Here we have dropped the term C = O(h 2

) and we may similarly neglect M . The first
term of Eq. (124) is the explicit result for F,0 . The second term uses the explicit result for
Z (D.4) and we recover the definition of C (89). This is precisely the density equation one
obtains when assuming fundamental bosons of mass 2M. This is an important result: Our Z renormalization procedure generates precisely the macrophysics we would have obtained when
starting microscopically with a purely bosonic action. In our approach however, the bosons
emerge dynamically.
8. Gap equation for the molecule propagator
The formulation of the problem as an Euclidean functional integral for fermionic and bosonic
fields permits the use of many of the highly developed methods of quantum field theory and
statistical physics. It is an ideal starting point for systematic improvements beyond MFT. We
have mentioned in the previous sections that one possible alternative to the standard one loop
approximation could be to first integrate out the fermions and then perform the remaining integral in one-loop order. This procedure has, however, some problems.
Let us consider the dimensionless renormalized inverse bosonic propagator (64) (B = 0) in
the approximation
= A q 2 ab + m2 .
P (q)
,ab
(125)
The exact mass matrix

m2,ab

2 u
=
a b =0
(126)
is diagonal but in general non-degenerate in the 1 , 2 basis. It vanishes at the critical temperature
of a second order phase transition. Already after the first step of the -integration
P differs from
the classical inverse molecule propagator. The molecule propagator will appear in the molecule
fluctuations as described by the partition function
SMF T []
Z = De
(127)
,
where SMF T is the intermediate action resulting from the Gaussian integration of the fermion
fields. It is composed of the classical boson piece and a loop contribution from the fermion
fields,

A []
+ 2 + U (F ) [ ]
+ .
SMF T = d 4 x Z []
(128)
1
At this stage the contribution from the fermion loop still depends on the fluctuating boson field.
For example, the formula for the dressed molecule density in the symmetric phase

(B)
1
d 3 q P (q
)/T
1 U1
e
1
=
nM =
(129)
2
(2)3
243
Fig. 10. Lowest order SchwingerDyson equation for the inverse molecule propagator (double dashed line) in the symmetric phase. The first two terms on the r.h.s. denote the mean field inverse propagator after integrating out the
fermionic fluctuations with a dashed line for the classical inverse molecule propagator and a solid line for the fermion
(F )
propagator. The third term on the r.h.s. accounts for the molecule fluctuations. Here is the molecule self-interaction
induced by the fermion fluctuations.
involves P (125). It generalizes13 Eq. (63) with replaced by . We see that the inverse prop(B)
agator P appears in the bosonic fluctuation contribution U1 to the effective potential and
influences the field equation for .
For a simple one loop evaluation of the functional integral (127) one would have to use a
(F )2
propagator with m instead of m2 in Eq. (125). This has an important shortcoming. Due to
(F )2
the difference between UMF T and U the masslike term m vanishes at some temperature different from Tc . As a consequence of this mismatch one observes a first order phase transition for
sufficiently large values of h . Clearly such a first order phase transition may be suspected to
arise from an insufficient approximation. (A similar fake first order transition has been observed
for relativistic scalar theories. There it is well understood [57,58] that an appropriate resummation (e.g. by renormalization group methods) cures the disease and the true phase transition
is second order. For the crossover problem the second order nature of the phase transition has
recently been established by functional renormalization group methods for the whole range of
concentrations, cf. [49].)
In order to improve this situation we use for the bosonic fluctuations SchwingerDyson type
equations where the inverse propagator P involves the second derivative of the full effective
potential U (rather than UMF T ). It is obvious that this is needed for a reliable estimate of nM
since the exact expression (60) and therefore also (129) involves the full molecule propagator
including the contribution of the molecule fluctuations. Since our functional integral treats the
molecule fluctuations exactly on the same footing as the fermionic atom fluctuations we can
derive the lowest order SchwingerDyson or gap equation for m2 in the standard way. For the
symmetric phase it is graphically represented in Fig. 10 and involves the full propagator and
therefore m2,ab .
8.1. Symmetric phase
We will consider a Taylor expansion of u = 2MkF5 U in terms of the invariant
13 Eq. (129) approximates the exact Eq. (61) in the limit where corrections to the dependence of P on the Matsubara
frequencies can be neglected, see Appendix C. The second part can be viewed as a SchwingerDyson equation for the
-dependence of U .
244
= kF3 Z .
(130)
Correspondingly we use a renormalized Yukawa coupling and four-boson vertex,

h =
2M h
,
(kF Z )1/2

Z2
(131)
For the symmetric phase we expand

1
u = m2 + 2 + ,
2
1 (F )
u MF T = m2 + 2 + .
2
(132)
The gap equation for m2 takes the form

)2
+ m(B)2
m2 = m(F
m(B)2
=2

n
d 3 q (F )
(q,
n )P1 (q,
n ).
(2)3
(133)
Here is the effective vertex involving four bosonic fields ( )2 , as induced by the
fermion fluctuations. It depends on q 2 and n . Neglecting the n -dependence, i.e. replacing
(F )
(F )
(F )
n ) (q)
(q,
n = 0), one can perform the Matsubara sum
(q,
(F )
(B)2

=

A q 2 + m2
d 3 q (F )
.
(
q)
coth
(2)3
2T
(134)
(F )
We have not yet computed the momentum dependence of (q)

but a simple qualitative consideration of the relevant diagram shows that for large q 2 one has a fast decay (q)
q 4 .
This makes the momentum integral (134) ultraviolet finite. It will be dominated by small values
(F )
(F )

of q 2 . For our purpose we consider a crude approximation where we replace (q)
(F )
(F )
(q = 0). Of course, we have now to restrict the momentum integration to low momenta. This
can be done efficiently by subtracting the leading UV divergence similar as in the computation
of nM , i.e. replacing coth x by coth x 1. This procedure yields

d 3 q
(B)2
(F )
m = 2
(135)
(exp 2 1)1 .
(2)3
We recognize on the r.h.s. of Eq. (135) the expression for the number density of dressed
molecules and obtain the gap equation
(F )2
m2 = m
)
(F
M
3 2
(136)
where we recall that M depends on m2 . Our gap equation has a simple interpretation: The
bosonic contribution to m2 vanishes in the limit where only very few dressed molecules play a
role (M 0) or for vanishing coupling. We are aware that our treatment of the suppression
of the high momentum contributions is somewhat crude. It accounts, however, for the relevant
(F )
).
physics and a more reliable treatment would require a quite involved computation of (q,
This complication is an inherent problem of gap equations which often require the knowledge
of effective couplings over a large momentum range. As an alternative method one may employ
245
functional renormalization [59] for the crossover problem [49], where only the knowledge of
couplings in a narrow momentum interval is required at every renormalization step.
(F )2
The fermionic contribution to the mass term m (cf. Eq. (66)) reads explicitly
(F )2
m
(F )
u
2
= (Z F )
=
+ 1
Z

h2
2
d 3 q
2T
1
=
tanh 2 .
Z
(2)3
q
4T
1
(F )2
m
(137)
The expression in the last line has to be evaluated with = in the symmetric phase. The r.h.s.
)
of the gap equation (136) involves the coupling (F
for the moleculemolecule interactions
(F )
(F )
2MkF (F ) 2 u 1
2
Z2
h4 d 3 q 3

tanh 2 cosh2 .
=
3
3
(2)
32T
(138)
(F )
Again, in the symmetric phase has to be evaluated at = 0, i.e. = .

The molecule fluctuations give a positive contribution to m2 , opposite to the fermionic fluctuations. This has a simple interpretation. The fermion fluctuations induce a self-interaction between
)
the molecules (F
. In turn, the fluctuations of the molecules behave similarly to interacting
fundamental bosons and modify the two point function for the molecules. The quantum corrections to the fermionic and bosonic fluctuations to the inverse molecule propagator are represented
graphically in Fig. 10. We emphasize that an additional microscopic molecule interaction could
(F )
(cl)
now easily be incorporated by adding to the mean field value for a classical part . In
this case the renormalization of discussed in Section 3.1 would be modified. For a constant
(cl)
the UV-divergent part would contribute to m2 (T = 0, n = 0). One would again end up with
(F )
(cl)
(F )
a contribution of the form (135), now with replaced by + . Our approximation (135)
therefore treats the interactions of dressed molecules similar to fundamental interacting bosons.
Actually, the SchwingerDyson equation for the molecule propagator (Fig. 10) also describes
the contribution of molecule fluctuations to the momentum dependent part encoded in A . In the
(F )
limit of a momentum independent the contribution of the boson loop to A vanishes in the
symmetric phase.
8.2. Superfluid phase
In the superfluid phase we choose an expansion of u around the minimum at
0 =
kF3 r
r
=
,
h 2
h 2
0 =
r
,
h2
(139)
namely
1
u = ( 0 )2 +
2
(140)
246
and define correspondingly

u MF T
(F )2
,
m
=
0
(B)
Our truncation of u 1
(F )

2 u MF T
.
2 0
(141)
approximates
1 (B)
( 0 ) + ( 0 )2 + .
2
The location of the minimum 0 is determined by the condition
(B)
u
(F )2
(B)2
(B)2
= 0, m
= 1
.
m
+m
=0
(B)
u 1
(B)2
=m
(142)
(143)
This defines the gap equation for 0 , which is the equivalent of Eq. (136) for the superfluid phase.
(B)2
(B)2
The computation of the bosonic contribution m
encounters the same problems as for m
(F )
(F )
in the symmetric phase. Again we replace (q) by a constant evaluated for q = 0 and
subtract the leading UV divergence of the momentum integral. This results in

A q 2 + 0 /2
d 3 q
(B)2
(F )

m
= 2
(2)3 A q 2 (A q 2 + 2 )
0

1

.
exp A q 2 A q 2 + 2 0 /T 1
(144)
In terms of the shorthands , , (112), (113), (114) we arrive at the gap equation for 0

d 3 q + /2
(F )2
(F )
m + 2
(exp 2 1)1 = 0.
(145)
(2)3
)
The quantity contains a mass term 2 0 which involves the full vertex = (F
+
(B)
. We have computed the SchwingerDyson equation for in Appendix E. For zero momentum q 0 we find that vanishes in the superfluid phase. This is, however, only part
In order to
of the story since the gap equation involves a momentum dependent vertex (q).
we present in Fig. 11
demonstrate the uncertainty arising from our lack of knowledge of (q)
our results for different choices of in the gap Eq. (145).
In detail, we show in Fig. 11 four approximation scenarios: (i) the standard BCS gap equa(F )
tion (long dashed) neglects the molecule fluctuations, i.e. = = 0 in both the gap and
the density equation. This yields a second order transition but disagrees with QMC results for
(F )
T 0. (ii) Bogoliubov density (short dashed) with in the density equation, while the molecule fluctuations in the gap equation are neglected. This improves the behavior for T 0, but
induces a fake first order phase transition for T Tc . (iii) Neglection of molecule fluctuations
(F )
in the effective coupling (dashed-dotted), i.e. we use = in the density and gap equation. (iv) Our best estimate (solid line) includes also corrections from molecule fluctuations for
. As described in this section we use in the propagator of the diagram in the gap equation,
(F )
(F )
whereas the coefficient multiplying the diagram is given by . (We use in the density
equation.) The first order nature is weaker than in (iii), but still present. In a recent functional
renormalization group treatment we have established the second order nature of the phase transition throughout the crossover, indicating that we now control the universal long range physics
governing the phase transition [49].
247
= r at the resonance. The role of molecule fluctuations and the unFig. 11. Temperature dependence of the gap
certainties in their treatment for T Tc are demonstrated by four choices of in the gap and density equation. The
critical temperatures are indicated by vertical dashed lines, with values Tc = 0.259, 0.292.
8.3. Phase transition

For T = Tc the gap equations (136) and (145) match since 0 = 0, = , = 0. Also the
expression for the molecule density becomes particularly simple
nM = kF3

(3/2) (3/2) T 3/2
.
A
4 2
(146)
9. Conclusions
Our functional integral investigation of ultracold fermionic atoms strengthens the picture of a
smooth crossover between BoseEinstein condensation of molecules and BCS type superfluidity.
One and the same field can both describe molecules and collective atom excitations of the type
of Cooper pairs. The two different pictures correspond to two regions in the space of microscopic
couplings and external parameters (T , n). In dependence on a concentration parameter c, which is
related to the magnetic field B, one can continuously change from one to the other region. Away
from the critical temperature the macroscopic observables are typically analytic in c1 despite
the divergence of the scattering length for two-atom scattering at c1 = 0. Only for T Tc one
expects to encounter the non-analytic critical behavior which is characteristic for a second order
phase transition in the universality class of O(2)-Heisenberg magnets.
For small and moderate Yukawa couplings h the physical picture of the crossover for T > Tc
is rather simple. Far on the BEC side the low temperature physics is dominated by tightly bound
molecules. As the concentration increases, the molecule waves start to mix with collective diatom states. More precisely, the molecule state with momentum p
mixes with pairs of atoms with
momenta q
1 , q
2 that are correlated in momentum space such that q
1 + q
2 = p
(Cooper pairs).
On the BCS side the relevant state becomes dominantly a Cooper pair. Our formalism uses only
one field for both the molecule and Cooper pair states. Nevertheless, the relative importance
of the microscopic molecule versus the collective Cooper pair is reflected in the propagator of ,
i.e. the wave function renormalization Z and the gradient coefficient A .
248
On the BEC side the propagation of corresponds to free molecules and A takes the classical
(cl)
(cl)
value A = 1/4M, and similarly for the wave function renormalization, Z = 1. The increas(cl)
(cl)
ing mixing in the crossover region results in A and Z growing larger than A , Z . Indeed,
the diagram responsible for the increase of A , Z precisely corresponds to a molecule changing
(cl)
to a virtual atom pair and back. Finally, far in the BCS regime one has A A (cl)
, Z Z
such that the classical contribution is negligible. The existence of a microscopic molecule is not
crucial anymore. At long distances the physics looses memory of the microscopic detailsthe
approximation of a pointlike effective interaction between atoms becomes valid.
For a broad Feshbach resonance the situation is, in principle, similar. However, the region
(cl)
where Z and A /A are large covers now both sides of the Feshbach resonance. The difference between the bare (microscopic) and dressed molecules is a crucial feature. For a broad
Feshbach resonance the microscopic molecules are irrelevant for the whole crossover region. For
this limit we have demonstrated in Section 7 that our formalism is equivalent to a purely fermionic model with a pointlike interaction. This demonstrates that in the broad resonance limit, the
use of the two channel model is rather a question of computational ease than a physical issueit
describes the same physics as a single channel model.
On the other hand, the two channel model is more general and also covers the case of narrow
resonances. This issue is discussed in more detail in [15]. It remains to be seen if it will be
possible to investigate the narrow resonances also experimentally in the future.
The universality of the low temperature physics and the crossover can be traced back to the
renormalizable couplings of an effective non-relativistic quantum field theory for long distances. More precisely, the long distance physics will only depend on the relevant (or marginal)
couplings in the infrared renormalization flow. These are precisely the dimensionless parameters
c and h . Improved approximations will influence the relation between microscopic atomic and
molecular physics and (c, h ). At this point also subleading interactions like the -exchange
or other interactions contributing to the background scattering length abg will play a role.
However, universality predicts that the relations between long-distance (macroscopic) quantities should become computable only in terms of the renormalizable couplings m2 (or c) and
h . In consequence, we find a universal phase diagram in terms of these parameters, where the
memory of the microscopic physics only concerns the values of c and h , providing for a universal phase diagram in terms of these parameters. In addition, the universal critical exponents
and amplitude ratios for T Tc will be independent of c and h . Furthermore, the BCS limit
is independent of h since only one parameter (c) characterizes the effective pointlike atom interaction. Also the BEC limit does not depend on h since the fluctuations of unbound atoms
become irrelevant. On the other hand, the behavior in the crossover region can depend on h
as an important universal parameter. This demonstrates that a pointlike approximation for the
effective atom interaction is not always applicable for small |c1 |.
1/2
The couplings h and h are related by the multiplicative factor Z . In the broad resonance
limit h one finds that the renormalized coupling h is given by an infrared fixed point. It
therefore ceases to be an independent coupling.
For a more general form of the microscopic action we expect that deviations from universality
become most visible far in the BCS regime, where further channels may play a role for the
effective interaction, as well as in the far BEC regime, where more details of the microscopic
dispersion relation for the molecules and their microscopic interactions may become relevant.
Going beyond our particular ansatz for the microscopic action universality should actually work
249
best in the crossover region of small |c1 |. On the other hand, for a given microscopic action
the results are most accurate in the BEC and BCS regimes where fluctuation effects are most
easily controlled. It is precisely the presence of strong fluctuation or renormalization effects in
the crossover regime that is responsible for the loss of microscopic memory and therefore for
universality!
The formulation as a Yukawa model solves the problems with a large effective scattering
length for the atoms in the crossover region in a simple way. The large scattering length is simply
due to the small mass term m
2 of the molecule field. It does not require a large atommolecule
interaction h . As long as the dimensionless Yukawa or Feshbach coupling h remains small,
the region of small and vanishing |c1 | (large or diverging effective atom interaction) poses
no particular problem. In the symmetric phase all quantities can be computed in a perturbative
expansion in h . For h 0 a nontrivial exact limiting solution of the functional integral has
been found [15]. Nevertheless, for large h a new strong interaction appears and the bosonic
fluctuations become important. Indeed, the crossover for a broad Feshbach resonance amounts
to a theoretical challenge. One has to deal with a genuinely non-perturbative setting.
In the present paper we have included the molecule fluctuations by the solution for
SchwingerDyson or gap equations. For T = 0 the results are quite satisfactory as shown by
the comparison with quantum Monte Carlo simulations [37] in Fig. 8. However, as T increases
towards the critical temperature Tc our method becomes less reliable since the details of the
treatment of the molecule fluctuations play an increasing role, cf. Fig. 11. The main problem
results from the neglected momentum dependence of the four-boson-vertex . This is related to
a general difficulty for SchwingerDyson equations: the momentum integrals involved require
the knowledge of effective couplings in a large momentum range. A promising alternative may
be the use of functional renormalization [59]. At every renormalization step only a narrow momentum range plays a role and the momentum dependence of suitably chosen effective vertices
is much less important. First results of a functional renormalization group treatment can be found
in [49].
In this light the present paper should be viewed as a starting point for more accurate investigations with further functional integral techniques. It is well suited for systematic extensions,
among which we would like to stress a more appropriate treatment of the momentum dependence
of the couplings, an extended inclusion of the effect of boson fluctuations, and the modification
of the fermion propagator by the renormalization effects originating from mixed fermionboson
contributions.
It remains to be seen if theoretical improvements, together with a reduction of systematic
experimental uncertainties, will finally lead to an understanding of ultracold fermionic atoms
with high quantitative precision.
Acknowledgement
We would like to thank T. Gasenzer, H. Stoof and M. Zwierlein for useful discussions.
Appendix A. Partial bosonization and particle density
In this appendix we extend the formulation for the functional integral in such a way that we
are able to discuss situations which are (i) inhomogeneous and (ii) beyond the small density approximation (SDA). The treatment of inhomogeneities is particularly desirable in the context of
250
ultracold gases which are prepared in traps of various shapes. In particular, we show how the local density approximation emerges from our formalism. Further we consider situations beyond
small densities. The price to pay is the inclusion of a further functional integration over a now
fluctuating field . In the SDA, treating the chemical potential as a source term is appropriate,
and we will quantify this statement here.
We will start the discussion for a single fermion field. This describes a situation far off a
Feshbach resonance, where molecules of effective collective states are unimportant. Actually,
the discussion in this appendix covers a very large class of fermionic systems with effective
pointlike interactions. Besides ultracold fermionic atoms, it may be applied to neutrons (e.g. in
neutron stars), dilute gases with short range interactions (e.g. dipole interactions) and also covers certain aspects of electron gases (where the Coulomb interaction is replaced by a pointlike
repulsion). We then extend the discussion to the case of fermions and bosons in order to account
for strong interactions via a Feshbach resonance. The concept presented here will also be technically useful for the analysis of strongly interacting ultracold atoms in the frame of the functional
renormalization group.
A.1. Functional integral
we define the partition funcFor an arbitrary fermionic theory with classical action SF []
tion as a functional of the local source J (x)

ZF [J ] =

D exp SF [] + dx J (x) (x)(x) .
(A.1)
With WF [J ] = ln ZF [J ] the (relative) particle density becomes

WF
=
.
n (x) = (x)(x)
J (x)
(A.2)
The physical particle density is related to n (x) by a constant shift that we have discussed in
Section 3. The source may be composed14 of a bare chemical potential and a local potential
Vl (x),
J (x) = Vl (x).
(A.3)
For ultracold atoms the local potential Vl (x) represents the trapping potential.
In this section we develop a general functional integral approach for the computation of the
density n(x). For this purpose we introduce a field (x) which is conjugate to J (x). It will play
the role of an effective chemical potential which differs from Eq. (A.3) unless the density is small.
Our approach can be used for arbitrary n and is not restricted to a small density approximation.
The effects of density fluctuations can be incorporated by functional integration over via partial
bosonization.
14 The split between a constant part in V and
is arbitrary. For interacting atoms in a homogeneous situation (Vl
l
0), is related to the true chemical potential by a constant shift (see below). For electronic systems Vl typically
corresponds to an electrostatic potential.
251
A.2. Partial bosonization

Partial bosonization is achieved by inserting a Gaussian integral which is equivalent to one up
to an irrelevant multiplicative constant (HubbardStratonovich transformation [40,41])

ZF [J ] = DD exp SF [] + dx J (x) (x)(x)

2

1
1
.
m2 (x) J (x) 2 (x)(x)
2
m
(A.4)
The partition function ZF [J ] can now be computed from an equivalent functional integral which
also involves bosonic fields (x)

1 2
2
ZF [J ] = DD exp SB [ , ] + m
(A.5)
dx J (x) (x) J (x)
2
with
+
SB = SF []

dx

1 2 2
1 2

m +
.
2
2m2
(A.6)
Here the expectation value (x) = (x) is related to the density n (x) by

WF
= m2 (x) J (x) ,
J (x)

1
+ J (x).
(x) = (x) = 2 (x)(x)
m
n (x) =
(A.7)
After partial bosonization the new action SB contains a mass term for , a Yukawa interaction
2 /(2m2 ). The resulting explicit
and an additional four-fermion interaction ( )
four-fermion vertex in SB therefore becomes
1
= + 2 .
m
(A.8)
We can choose m2 such that vanishes by requiring

1
= .
m2
(A.9)
The limit of non-interacting fermions obtains then for m2 . We note that such a cancellation
is possible only for < 0. It is not compulsory for our formalism an we will not always assume
the relation (A.9) in the following.
The partition function ZF (A.5), (A.6) is closely related to the standard formulation of a
scalar-fermion-model with

ZB = DD exp SB [ , ] + dx j (x) (x)

(A.10)
by (WB = ln ZB )
j (x) = m J (x),
2
m2
W B = WF +
2

dx J 2 (x).
(A.11)
252
In this formulation we may define the (one particle irreducible) effective action by the usual
Legendre transformation

WB
= WB + dx j (x) (x), (x) =
(A.12)
.
j (x)
For a given source j (x) the expectation value (x) obeys the field equation
= j (x).
(x)
(A.13)
The effective action may be decomposed into a classical part cl [ ] = SB [ = , = 0] and a

quantum part q

m2
dx 2 (x) + q [ ].
= cl + q =
(A.14)
2
We next turn to the relative particle density n (x). It can be computed from by decomposing the field equation
q
= m2 (x) +
= m2 J (x)
(x)
(x)
(x) = J (x) + 1 (x),
(A.15)
as
n (x) = m2 1 (x) =

q
J (x) + 1 (x) .
(x)
(A.16)
This establishes the exact general relation between n (x) and the -functional derivative of the
quantum part of the effective action.
The limit |1 | |J | corresponds (for fixed J (x)) to small densities or small interactions
(m2 ) according to

n (x) m2 J (x).
(A.17)
In this limit we may expand

q
J (x) + 1 (x) = n0 (x) + b(x)1 (x) +
(x)
with
n0 (x) =

q
,
(x) J (x)
b(x) =

2 q
.
2 (x) J (x)
(A.18)
(A.19)
This yields
n (x) =
n0 (x)
.
1 + b(x)/m2
(A.20)
We emphasize, however, that Eq. (A.16) remains valid for arbitrary densities.
The explicit computation of n (x) needs an evaluation of q . In mean field theory the
fluctuation part q is estimated by including only the fermionic fluctuations in the functional
integral (A.10), while keeping (x) = (x) as a fixed background. In this scheme, the generalization of Eq. (21) for the relative particle density reads, according to Eq. (A.16)

q
1
PF
3
3

= Tr tanh
,
d x n (x) = d x
(A.21)
(x)
2
2T
253
r is over three dimensional phase space and spinor indices in

where the remaining trace T
d3q

momentum space it reads Tr =
V3 (2)
3
, with V3 the volume of (three-dimensional) space.
At this point the construction from Section 3.1 can be performed, replacing the relative (n ) by
the physical particle number density n + n.
This yields

1

P
r exp F
=T
.
N = d 3 x (n + n)
(A.22)
+1
T
r over an operator A can be evaluated in a
This formula has a simple interpretation. The trace T
complete orthonormal system of complex functions fm (x),

3
m (x) = Am m ,
d x fm (x)fm (x) = m m ,
d 3 x fm (x)Af

rA =
Amm .
T
(A.23)
m
With (x) = V (x) (in analogy to Eq. (A.3)) we can define the Hamilton operator H =
PF + V (with P = /(2M) for non-relativistic atoms). Choosing the fm to be eigenfunctions
of the Hamiltonian with eigenvalue Em Eq. (A.22) becomes
1
Em
N=
(A.24)
+1
exp
T
m
and we recognize the well known fermionic occupation number. In the limit of non-interacting
atoms (m2 ) we can cut the Taylor expansion in Eq. (A.18) at the lowest term, n = n0 .
In consequence, q / is evaluated with the local potential Vl in Eq. (A.3). In this limit one
has = Vl and V therefore equals the trap potential Vl . As it should be the Hamiltonian
H reduces to the quantum Hamiltonian of a single atom in a potential and the density becomes
independent of m2 .
A.3. Universal field equation for the density
The crucial advantage of our formalism is the possibility to compute the effective action [ ]
without any specification of the local potential (trapping potential) Vl (x). The detailed geometry
of the trap only enters at a second step when one solves the field equation (A.13). This offers
the great advantage that the fluctuation problem can be solved in a homogeneous setting, i.e. one
encounters standard momentum integrals in the loop expressions rather than summations over
spherical harmonics or other function systems adapted to the geometry of the trap. The results
are universal and apply to all geometries, provided the inhomogeneity is sufficiently weak.
For (
x ) mildly varying in space and independent of one may use a derivative expansion

1
[ ] = dx U ( ) + A ( ) + .
(A.25)
2
We emphasize that the derivative expansion always applies for situations close enough to homogeneity. The following results are therefore valid independently of the way how [ ] is
computedin particular, they do not rely on MFT. However, for definiteness, let us give the
result for A in the latter scheme (U ( ) can be read off from Eq. (19) for = ). For this
purpose, we proceed along the lines in Section 3.3 and extract from q,MF T the term quadratic
254
in Q
V3
(2)
q,MF T = Q
P (Q)Q ,
(A.26)
T

1
1
P (Q) =
+
,
(PF (Q ) )(PF (Q + Q) ) (PF (Q ) )(PF (Q Q) )

Q
(A.27)
d3q

with Q = n T (2)3 . Inserting the background (X) = +exp(iQX)Q +exp(iQX)
into Eq. (A.25) yields
Q
P (0, q)
q 2 0 q 2

1
tanh
q 2 2 cosh2 3
d 3q
=
.
9MT
4MT 2
(2)3 cosh2
cosh4
A ( ) = lim
(A.28)
We find that A ( ) is ultraviolet finite. The zero temperature limit of this expression reads15 (for
> 0)

2M
1
A =
(A.29)
,
12 2
and diverges for 0.
A.3.1. Density field equation
We are interested in time-independent situations and therefore consider (x) = (
x ) independent of . Variation with respect to yields the stationary field equation (with U = U / ,
etc.)
1
= m2 ( Vl )
U ( ) A ( ) A ( )
(A.30)
2
x ) is found one can compute the particle density from
where and Vl depend on x
. Once (
n = U + A + A
2

n
(A.31)
= m2 + Vl + 2 ,
m
1
U = U m2 2 n.
(A.32)
2
Here we have chosen the definition of U such that the physical particle density from Eq. (23) is
reproduced in Eq. (A.31).
In the second line of Eq. (A.31) appears a shifted or additively renormalized chemical potential,
=
n
.
m2
15 The prefactor has been determined numerically with an error of less than 0.3%.
(A.33)
255
This combination is the true physical chemical potential as can be seen along the lines of
Eqs. (22), (23). Indeed, the interaction term in the operator language (a (x)a(x))2 is translated
to the interaction term in the functional integral (A.1) for the microscopic action by the use of
2
2
1
1
a (x)a(x)
+ n .
(x)(x)
2
2
2m
2m
(A.34)
precisely shifts to in the functional integral (A.1). A detailed

The term (n/m
2 ) (x)(x)
account for a similar chemical potential shift for bosonic atoms can be found in [60].
Hence the second part of Eq. (A.31) yields a linear relation between the physical particle
density n and ,
= Vl +
n
.
m2
(A.35)
This will be our central equation relating the density n, the effective chemical potential , the
chemical potential and the local trap potential Vl . We observe that a great part of the ultraviolet
divergencies (for ) present in the functional integral description is related to and n.

These divergencies do not appear if the physical quantities and n(x) are used.
Eq. (A.35) can now be used to eliminate in favor of the physical particle density. Insertion
into the first Eq. (A.30) yields the central field equation for the density in an inhomogeneous
situation,

n
n
1
1
1
A
+
A
+
n
n
V
n
+
V
V
l
l
l
m2
m2
m2
m2
2m2

n
n
1
n

= U Vl + 2 A Vl + 2 Vl + A Vl + 2 V
l Vl .
2
m
m
m
(A.36)
Once the functions U ( ) and A ( ) have been computed this equation permits the determination of n(
x ) as a function of . For its solution the boundary conditions of n(
x ) have to be
specified appropriatelyfor trapped atoms n(
x ) has to vanish away from the trap. In practice,

one often does not know in a given experimental situation, but rather has information about
the total particle number N = d 3 x n(x). In this case one may compute N () by integrating the
solution of Eq. (A.36). Our setting can be reformulated in a more intuitive form in terms of a
density functional [n] to which we proceed in the next section.
A.3.2. Density functional
We can write the effective action in a more intuitive form as a functional of the particle density.
It is convenient to add the source explicitly such that the field equation becomes

= 0.
n(x)
Using Eq. (A.35) one has
(A.37)
256
[n] = [ ] + dx m2 (Vl ) n

n
n2
= dx U
+
V
+
l
m2
2m2

2
1
n
n
+ A
+
V
+
,
(A.38)
l
l
2
m2
m2
where again a density independent term has been dropped. The explicit form of Eq. (A.37) is
given by Eq. (A.36).
In particular, a homogeneous situation is governed by the effective potential
n2
(A.39)
+ U (n).
2m2
In this case we can use n as the independent variable, replacing . In turn, the chemical potential
(n) follows from the minimum condition Un /n = 0. In practice, one first solves for (n) by
inverting
Un (n) =
U ( ) = n.
(A.40)
In MFT this step does not depend on m2 . The interaction strength m2 only enters the determination of through Eq. (A.35). Similarly, in the local density approximation we can trade for
the density at a given reference location x0 , n0 = n(x0 ). Using again Eq. (A.40), the computation
of the density profile employs
n(x) n0
(A.41)
.
m2
For a given central density n0 (e.g. x0 = 0) and given trap potential Vl (x) Vl (0) we can now
evaluate U as a function of n and determine n(x) from U ( (n)) = n.
On the other hand, if n(x) is known we may use a functional of the variable for fixed n(x).
Using
(x) = (n0 ) + Vl (x0 ) Vl (x) +
= m2 n + n
we may define

m2 2

= + dx (n n)

,
= 0.
2
Here n(x) is considered as a fixed function. The corresponding potential
(A.42)
(A.43)
m2 2
= U + n
+ (n n)
U = U
(A.44)
2
is particularly useful for a homogeneous situation where n is a fixed constant. The solution for
(n) corresponds to the minimum of U .
A.3.3. Small and local density approximations
In the small density approximation (SDA) is given by
= Vl (x).
(A.45)
We can now specify the validity of the small density limit more precisely and replace the condition (A.17) by
n m2 | Vl |.
257
(A.46)
In lowest order in the small density approximation one obtains the density of non-interacting
fermions in a local potential as
1
l V
l.
n = U ( Vl ) A ( Vl )Vl + A ( Vl )V
(A.47)
2
We emphasize that Eq. (A.36) and its low density limit (A.47) are rather universal formulae
which determine the density of fermions in a local potential. Their possible applications reach
far beyond the particular problem of ultracold trapped atoms.
Another useful approximation, the local density approximation (LDA), obtains by setting
A = 0 in Eq. (A.36),

n

n = U l + 2 ,
(A.48)
m
where we define the local chemical potential
l (x) = Vl (x).
(A.49)
Its validity does not require a small density but rather that the derivative terms in Eq. (A.36) (or
(A.25)) are small as compared to U . For a given size of A this always applies for a sufficiently
homogeneous trap. Indeed, the local character of Eq. (A.48) guarantees that weak changes in
l (x) result in weak changes of n(x). The error of the LDA can now easily be estimated by an
iterative procedure. Inserting the solution of Eq. (A.48) into Eq. (A.36) allows for a simple direct
computation of the subsequent terms. Obviously, the error depends on the size of A which is
often a rather small quantity (see below).
If both the density is sufficiently small and the trap sufficiently homogeneous one can work
with the small local density approximation (SLDA). This results in the simple formulae
U
n=
(A.50)
and
= l .
(A.51)
In this approximation we can regard (x) as a fixed external parameter and compute n(x) by
Eq. (A.50). We will find that for realistic ultracold atom systems like 6 Li the LDA is valid
whereas Eqs. (A.50), (A.51) are oversimplifications (except for the BEC regime far away from
the Feshbach resonance).
We emphasize again that our method is valid quite generally and not bound to the case of small
density. If necessary, mean field theory can be improved by including the bosonic fluctuations in
the computation of U and A by performing the functional integral over . Our method can also
be applied in the presence of additional condensate fields. As a simple application we compute in
Appendix B the density at low T in the mean field approximation. In particular, this shows that
the dilute gas of ultracold atoms is a metastable state, the ground state being a liquid or solid.
A.4. Fermions and bosons
We can proceed in complete analogy to Appendices A.1, A.2 with ZF replaced by

ZF M [J ] = DD exp SF M [, ] + dx J (x) (x)(x) + 2 (x)(x) ,

(A.52)
258
and
ln ZF M
= n(x),
J (x)
J (x) = + (1 )
n
Vl (x).
m2
The classical action reads now

2

1 1
SF M = dx
2
2M
2 m

2
n
+
+ + 2 (1 2 + ) + VM (x) 2Vl (x)
4M
m

h T
2 2 2

2 2

2
m
m
(A.53)
(A.54)
and the partition function describes the coupled system of atoms in a local potential Vl and molecules in a molecule potential VM . (The term 2Vl in SF M cancels the corresponding
The atoms are coupled to the molecules by the Yukawa coupling h . Again,
term from 2J .)
the bare parameters , h and m2 have to be fixed by appropriate observable renormalized
parameters. We have also included a possible local self-interaction of the molecules 2/m2
2 . We concentrate on VM = 2Vl and = = 1.
and between free atoms and molecules 2/m
This simply counts the molecules as two atoms as far as the local interactions are concerned, e.g.
the local self-interaction is (n F + 2n M )2 and the energy in the trap potential is n F + 2n M .
We note that in this particular case n drops out. (The corrections to are computed similar to
Eqs. (A.33), (A.34).)
Replacing in the HubbardStratonovich transformation (A.4) + 2 all steps
in Appendix A.3 can be performed in complete analogy, with the only difference that ZB in
and SB reads
(A.10) involves now also an integration over and an appropriate source for ,
(for m2 = )

SB = dx
+
+ 2
2M
4M

2
h
m 2
+
(A.55)
T
.
2
2
This is a simple model for fermions with Yukawa coupling to scalar fields and .
A.4.1. Effective action
obtains again by a Legendre transform similar to Eq. (A.12) and
The effective action [, ]
obeys now
= m2 l = m2 n.
(A.56)
In particular, a homogeneous situation is characterized by an extremum of U = U (m2 /2) 2 +

n ,
U = n
h 2 M
4a
2 + U1 = U + n.
(A.57)
259
Similar to Section 5.4 we may proceed beyond the Gaussian functional integration for (MFT)
by adding to U1 a piece from the one loop contribution from the -fluctuations
(F )
U 1 = U1
(B)
+ U1 .
(A.58)
The contributions from dressed uncondensed molecules is now given by

(B)
U1
= 2nM .
(A.59)
Indeed, Eq. (A.59) follows from Eq. (85) if is replaced in the classical bosonic propagator by
the effective chemical potential .
A.4.2. Field equations
Collecting the different pieces the extremum of U (Eq. (A.57)) occurs for
U
= 0 = n 2 nF,0 2nM .
(A.60)
(F )
This is precisely the relation (118) as it should be. (Recall nF,0 = U1 / .) In other words,
the density obeys
U
.
(A.61)
With J = we may derive the relation between the effective chemical potential and the chemical potential
n
= 2
(A.62)
m
directly from the identity

1
2
DD D
J = 0.
ZB
(A.63)
exp SB + m

n=
The evaluation of U (, ) is now sufficient for the computation of the total atom density n. For
the homogeneous setting one can therefore determine by
1
u
= 2.

3
(A.64)
A.4.3. Effective chemical potential

In summary, our problem is now formulated as a functional integral for a Yukawa model. In
the small density approximation we can treat (x) = + Vl (x) as an external parameter. The
partition function becomes

;
] + source terms
exp SB [,
Z = DD
(A.65)
where SB is given by Eq. (A.55) and the density obtains from Eq. (A.64). Beyond the small
density approximation is treated as a field and the partition function involves an additional
integration over . In the limit where the -fluctuations can be neglected we may consider the
effective chemical potential instead of as a free external parameter. In particular, this
offers for the homogeneous case the advantage that we are not bound to the validity of the small
260
density approximation (which may not be accurate for realistic systems as 6 Li). It is sufficient to
compute the value of (T ) for a given density or given kF .
For the homogeneous case we can both deal with situations at fixed effective chemical potential or at fixed density n. For a fixed we may choose an arbitrary fiducial kF (not determined
by n) and do all the rescalings with kF (instead of kF ). One can then work with a fixed value of
and compute n or kF by the relation
u
1 k3
= 2 F3 .

3 kF
(A.66)
Many experimental settings can be idealized, however, by a fixed value of n. Then the choice
kF = kF and the determination of via Eq. (A.64) provides directly all results for fixed n. For
an inhomogeneous setting it is sensible to use a suitable fiducial kF .
Appendix B. Metastability
At low T the thermal equilibrium state of atoms is a liquid or solid, with density essentially
determined by the size of the atoms (typically set by the van der Waals length). When we
deal with a dilute gas of atoms at ultracold temperature we obviously do not consider the stable
thermal equilibrium state which minimizes the free energy. Indeed, we will see in this section that
the metastability can be captured within our formalism if the weak cutoff dependence O(1 ) is
not neglected. For this purpose, we consider a homogeneous system and neglect the effects from
molecules.
Under the above circumstances the field equation (A.36) reduces to a simple self-consistency
relation for the density,

1
n
d 3q
n = U + 2 = 2
(B.1)
.
m
(2)3 e(q 2 /(2M)n/m2 )/T + 1
For the gross features we first consider the T 0 limit where the last equation reduces to

3/2
n
1
.
2M + 2
n=
(B.2)
3 2
m
The resulting cubic equation for y = n/(m2 )
(y + 1)3 y 2 = 0,
9 4 m4
8M 3
(B.3)
may have several solutions, depending on . We concentrate on > 0 and consider first 1.
The solution with small y 1/2 ,
n1 =
1
(2M)3/2
3 2
(B.4)
describes a dilute gas of atoms. In lowest order the relation between n and is indeed independent of m2 and . For large the second solution with positive y occurs for y
n2 =
9 4 m6
8M 3
(B.5)
261
while the third solution has negative y and should be discarded. As decreases (i.e. increases)
the two solutions approach each other and melt for c = 27/4, yc = 2 or
nc = 2m2 =
4 m6
,
3M 3
(B.6)
which delimits the metastable gas phase. No solution with y > 0 exists for < c . The stability
of the solution depends on the second derivative of Un with respect to n, i.e. for T = 0

1
3 1 + y 1/2
2 Un
=
1
.
2
n2
m2
(B.7)
The solution n1 (with small y) turns out to be stable ( 2 Un /n2 > 0) whereas the solution n2
(with y ) is unstable. For c one has 2 Un /n2 = 0.
What happens for n > nc ? In order to study this question we cannot neglect the ultraviolet
cutoff anymore. In the evaluation of Eq. (B.1) the upper bound of the momentum space integral
is actually given by qmax = instead of infinity. For < (2M( + n/m2 ))1/2 this multiplies
the r.h.s. of Eq. (B.2) by an additional factor 3 /(2M( + n/m2 ))3/2 such that
n=
3
3 2
for
n
2
+
>
.
2M
m2
(B.8)
This solution is again stable with 2 Un /n2 = 1/m2 . We associate it with the liquid or solid
thermal equilibrium state. Indeed, the free energy for ngs = 3 /(3 2 ) is much lower than for the
solution n1 . Also the density is determined by the effective size of the atoms 1 . Inversely,
we can define our cutoff by the density ngs of the liquid or solid ground state for T = 0, =
(3 2 ngs )1/3 . Of course, in this region our approximations are no longer valid, but the detailed
properties of the liquid or solid state are not the purpose of this paper.
Beyond the limit T 0, a computation of the phase boundary for the existence of a
metastable gas for arbitrary temperature requires the solution of the condition for the melting
of the stable and unstable extrema,
!
0=
2 Un
1
= 4 M2
2
n
m
(B.9)
with
M2 = m2 + U = m2
1
2T
d 3q
cosh2 .
2 3
(B.10)
One of the dependent variables, or n, has to be eliminated via the equation of state (B.1). For
the range of T considered in this paper the influence of the temperature is negligible and the
qualitative result (B.5) is not affected.
Appendix C. Boson propagator in MFT
In mean field theory the gradient terms for the fields and obtain by evaluating the fermion
loop in an inhomogeneous background field. We write the inverse fermion propagator as P + F
We decompose the fields in a position independent part
with P the part independent of and .
and a harmonic (single momentum mode) small local perturbation,
262

F(X, Y ) = Fh + Finh =
h (X)
(X)

(X)
(X Y ),
h (X)
(C.1)
where is the 2-dimensional totally antisymmetric symbol and
(X)
= + exp(iKX),
(X) = + exp(iKX) + exp(iKX).
(C.2)
The one-loop fermionic fluctuations can now be expanded around the homogeneous fields,
1
q(MF T ) = Tr ln(P + Fh + Fin )
2

1
1
= Tr ln(P + Fh ) Tr ln 1 + Fin (P + Fh )1
2
2

2
4
1
= V UMF T + Tr Fin (P + Fh )1 + O Fin
.
4
(C.3)
The terms with odd powers in Finh vanish due to translation invariance.
For a practical computation, we switch to Fourier space where

P(Q1 , Q2 ) =
0
PF (Q1 )

PF (Q1 )
(Q1 Q2 ),
0
(C.4)
and
(P + Fh )1 (Q1 , Q2 ) = R(Q1 )(Q1 Q2 ),

(PF (Q1 ) )
h
R(Q1 ) =
h
(PF (Q1 ) )
1
.
[PF (Q1 ) ][PF (Q1 ) ] + h 2
(C.5)
We concentrate first on the gradient term for the molecules (A ) and set = 0. Then the inhomogeneous part reads in momentum space
Fin (Q1 , Q2 ) = h
Q1 ,Q2 +K
0
Q1 ,Q2 K

(C.6)
(R,S = (R S)). This yields
2 1
1
Tr Fin (P + Fh )1 = tr
4
4

Fin (Q1 , Q2 )R(Q2 )Fin (Q2 , Q1 )R(Q1 )
Q1 ,Q2
where tr is the trace over the internal 4 4 matrix. One obtains
(C.7)
qMF T =
V3 2
h
T
[PF (K

Q
263
PF (Q )
[PF (Q ) ][PF (Q ) ] + h 2
PF (Q K)
.

Q ) ][PF (Q K) ] + h 2
(C.8)
Insertion of the ansatz (C.2) into the effective action16

=
V3
P (K)
T
(C.9)
yields Eq. (36). We note the simpler form for = 0 (symmetric phase) where

h 2
q2
.
P (K) = 2inT +
4M
[PF (Q) ][PF (K Q) ]
(C.10)
The quantum corrections differ from P (Eq. (A.26)) by the overall factor h 2 /2 and the different momentum structure in the denominator.
We note that in the superfluid phase the symmetries would also be consistent with a gradi
(Q).
The coefficient of this term cannot be computed
ent term of the form Q (Q)
= + + exp(iKX) +
with the ansatz (C.2)it would require a more general ansatz (X)
exp(iKX). We will omit this term in the present paper.

C.1. Momentum dependence
For spacelike momenta the loop integral depends on the square of the external momentum
only and we define A (q) and A by Eqs. (41), (42). The fluctuation correction to P at n = 0
is given by
d 3q
f (q )
P = h 2 T
3
(2) [((2m + 1)T )2 + f 2 (q ) + r]
m
f (q q)
[((2m + 1)T )2 + f 2 (q q) + r]

h 2 d 3 q f (q )f (q q)
b
a
1
1
=
tanh
tanh

3
2
2T
2T
(2) f (q ) f (q q)
a
b
(C.11)
with f (q) = q 2 /2M , a = f 2 (q ) + r, b = f 2 (q q) + r. The momentum dependent gradient coefficient reads

P (q 2 ) P (0)
1
(cl)
= A + A q 2 .
+
A q 2 =
2
4M
q
(C.12)
As argued in Sections 5 and 6, the physically relevant quantity is A /Z . We have plotted the
momentum dependence of A (q)/Z in the broad resonance limit h 2 1 in Fig. 4. At large
momenta, the gradient coefficient slowly tends to zero. This has no impact on observables, since
16 Note the slight abuse of conventions. Here, P stands for the 11-entry of the inverse propagator matrix in the ,
basis, instead of the full matrix.
264
the thermal distribution functions are suppressed at lower momenta already (see below, also
Fig. 5). It might be an artefact of the neglected momentum dependence of Z .
In the present paper we neglect for the numerical results the momentum dependence of A and
approximate A (q) = A A (q = 0). Let us discuss here the validity of this approximation in
the broad resonance regime at the critical temperature. The impact of the momentum dependence
of A (q)/Z is most clearly seen for the Bose distribution in Fig. 5. There we explicitly compare
the results with a momentum dependent A (q)/Z and an approximation of constant A /Z .
In the BEC regime the error is very small with A (q)/Z close to the classical value 1/2 for
all q. The main difference from the classical propagator concerns the renormalization of the
mass term m
2 . In the crossover regime the approximation of constant A /Z underestimates the
number of molecules with large q 2 , but the error remains acceptable. In contrast, the deviation
from the result with the classical molecule propagator is already substantial. In the BCS regime
the underestimate of NM (q) for the dominant range in q is quite substantial. Though the overall
effect of the boson fluctuations is small in the BEC regime, this may affect the quantitative results
for the number density of molecules and the condensate. In view of Fig. 5(c) the estimates of M
and C and M , C in the present paper are most likely too small.
We next discuss A = A (q = 0) more explicitly. With A = 1/4M + A we can compute
A as the term linear in q 2 in the Taylor expansion of P (C.11). Using A = 2M A we find

the result
h 2 d 3 q 2 7 4

1
q 3 5 5 2 2 + 24
A = +
3
2 288T 3
(2)

tanh cosh2 + 2 2 2 2

cosh4 6 tanh 2 tanh2 ,

(C.13)
simplifying in the symmetric phase = to
h 2 d 3 q 2 3
tanh cosh2 .
q
A = +
(C.14)
2 48T 3
(2)3
The loop correction to the gradient coefficient is strictly positive and monotonically growing for
increasing from negative values (BEC side) to its saturation value = 1 on the BCS side. For
the BEC regime it vanishes in the limit . However, the physical quantity A = A /Z
approaches the finite value 1/2 in the BEC regimethis is an indicator of the emergence of an
effective bosonic theory. For 1 instead, A /Z is much larger than the value for elementary
pointlike bosons, 1/2. Indeed the integral is dominated by modes peaked around , explaining
the strong increase as 1.
In the limiting BEC and BCS cases, A can be approximated by much simpler formulae. The
BEC result is given in Appendix D, Eq. (D.3). In the BCS regime, the critical temperature is very
small (T 0.1) and 0.4 1. We then find an approximate behavior for the r -dependent
gradient coefficient
2 3/2
1 7 (3) h
.
A = +
2 12 4 4T 2 + r
(C.15)
(cl)
In the temperature range of interest the classical contribution to A is small (A = 1/2). Neglecting it and restricting to the symmetric phase ( = 0), this is consistent with the symmetric
BCS result in [11], and we see how the condensate regulates the divergence for T 0.
265
C.2. Frequency dependence

Similarly the loop correction can be evaluated as a function of external Matsubara frequency
m (Q = (m , q
)) for vanishing external spacelike momentum q. This amounts to the renormalization correction to the operator . In momentum space this corresponds to the part in
in the inverse molecule propagator. Hence we only need to consider the imaginary part of the
loop integral. The denominator in the integrand in Eq. (C.8) is real and the imaginary part of the
numerator becomes 2inTf (q ). This yields ( = 2nT )17
Im P (, q = 0)
d 3q

2
1
2
= h T
f (q ) (2m + 1)T + f 2 (q ) + r
3
(2)
m

2
1
(2m + 1)T + + f 2 (q ) + r
.
(C.16)
Obviously Im P vanishes for n = 0 ( = 0). For n = 0 we can perform the Matsubara sum
(we suppress the argument of f ),

h 2 d 3 q tanh( f 2 + r/2T )
.
Im P =
(C.17)
4
(2)3 (/2)2 + f 2 + r
We may define the coefficient of the Matsubara frequencies as (1 = 2T )
Im(P (, 0))
,

h 2 d 3 q tanh( f 2 + r/2T )
P (1 , 0)
.
=1+
Z, (1 ) = Im
1
4
(2)3 (T )2 + f 2 + r
Z, () =
(C.18)
We can study the bosonic propagator as a function of m. The loop corrected imaginary part of
the inverse boson propagator can be brought into the form

im 1 + c q 2 , T , , m2 , m = 2m.
(C.19)
Each Matsubara mode is renormalized by an m-dependent quantity. In the present paper we
neglect these corrections, i.e. we take c(q 2 , T , , m2 ) = 0.
Appendix D. Analytical results in the BEC regime
We exploit the fact that in the BEC regime /2T , which means that we can replace the
functions tanh 1, cosh2 0. In the superfluid phase, we use additionally r /| | 0.
The loop integrals can then be evaluated analytically. They are temperature independent. Furthermore, their values coincide in the symmetric and superfluid phase, if terms of O(r / ) or higher
order in 1 are ignored. We find
h 2
(F )2
,
m
=
(D.1)
8
17 As in Eq. (C.9), P stands for the 11-entry of the inverse propagator matrix in the , basis.
266
h 4
3,
128
h 2
,
A =
64
h 2
Z =
.
32
(F )
=
(D.2)
(D.3)
(D.4)
For the fermionic particle density contribution we find a term O(r / ),

r
.
16
The BCS gap equation is solved by using (D.1),

c1 = ,
nF,0 = kF3
(D.5)
(D.6)
independently of the value of r . Hence in the BEC limit, the relation between binding energy and
scattering length [15] is independent of the density scale kF . Indeed, many body effects should be
unimportant in this regime. Approaching the resonance, the impact of the pairing gap r becomes
important and = M /F .
In the limit of large Yukawa couplings, we can then evaluate the gradient coefficient of the
effective Bose distribution (cf. Eqs. (115), (116)):
A =
A
1
= .
Z
2
(D.7)
Similarly, the fermionic contribution to the four-boson coupling evaluates to

(F )
(F )

Z2
8
=
.
(D.8)
Using the relation p = 4ap /Mp between coupling strength and scattering length, the molecular scattering length in the BEC limit is given by
aM = 2aR
(D.9)
where we have changed back to dimensionful quantities. This reproduces the Born approximation
for the scattering of particles of mass Mp = 2M. In approaching the resonance for the fermionic
scattering length (crossover regime) c1 = 0, the bosonic scattering length however remains
(F )
finite. Note that both A and are effectively independent of h in the broad resonance limit.
Appendix E. SchwingerDyson equations for the molecule couplings
In this appendix we provide details of our computation of the effective moleculemolecule
interaction . We work in dimensionless renormalized units. In the symmetric phase we expand
the effective potential u and the mean field effective potential u MF T in powers of ,
1
u = m2 + 2 + ,
2
1 (F )
(F )2
u MF T = m + 2 + .
2
(E.1)
267
The contribution from the boson loop reads

(B)
u 1
1 (B)
+ 2 +
2

1
)2
) 2
= m2 m(F
+ (F
+
2
(B)2
= m
(E.2)
(B)
(see below). In the symmetric phase we evaluate u 1 in an approximation where we truncate in

quadratic order in (E.2).

2 U
(F )2
(B)2
)
(B)
2
m = m + m ,
(E.3)
=
= (F
+ .
2 =0
We determine the coupling from the SchwingerDyson equation
(B)
2 U1
(F )
= +
2 =0
)

3(F
d 3 q 1

(F )
=
(exp 2 1)1 + 2 sinh2
3
(2)
2T
(F )
= + I ,
(E.4)
which has the solution

=
)
(F
1 I
(E.5)
For m2 0 the last term in Eq. (E.4) becomes infrared divergent. Divergences of this type of
quantum corrections to quartic couplings are familiar from quantum field theory and statistical
physics of critical phenomena. Indeed, the point m2 = 0 corresponds to the critical line (or hypersurface) for the phase transition to superfluidityfor negative m2 the symmetric phase becomes
unstable. The remedy to this infrared problem has been well understood from the solution of
functional renormalization group equations: the strong fluctuation effects drive to zero at the
critical line [57,58,61].
Our gap equations recover this important feature in a direct way. As m2 approaches zero the
negative last term in Eq. (E.4) becomes more and more important as compared to on the
left-hand side. The solution to Eq. (E.4) implies
lim (m ) 0.
m 0
(E.6)
For small values of m2 in the vicinity of the phase transition we can expand the integral in
Eq. (E.4) as

2
d 3 q
(F )
I = 15T
(E.7)
A q 2 + m2
.
(2)3
One infers m according to
=
8 3/2
A m .
15T
(E.8)
268
In the superfluid phase, we expand the effective potential around 0
( 0 )2 .
2
We choose again a basis of real renormalized fields 1 , 2 according to
u = m2 ( 0 ) +
1
= (1 + i2 ),
2

1 2
1 + 22 .
2
(E.9)
(E.10)
2 = 2 ,
Without loss of generality we may consider a background of real , i.e. 1,0
0 2,0 = 0.
With 1 = 1 1,0 the potential (E.9) becomes

1
1
u = m2 22 + m2 + 2 0 1 2 + 0 /21 22
2
2
4 2 2 4
+
+
+
+ .
8 2
4 2 1
8 1
(E.11)
We can associate m2 and with the terms quadratic and quartic in 2 . The dots denote cubic and
quintic terms 1 3 , 24 1 , 22 1 3 that will not contribute in our approximation and we neglect
terms of O( 6 ). We use the SchwingerDyson equations for the 22 and 24 vertices which result
in Eq. (145) and
)

3(F

d 3 q 3
( )2 (exp 2 1)1
(2)3
2T
+ 2( + /2)2 sinh2 .
(F )
(E.12)
Again, we observe in Eq. (E.12) the appearance of infrared divergences in the contribution
(F )
from the Goldstone fluctuations. If we would define () by the 24 -vertex evaluated at some value > 0 they would be regulated. Taking the limit 0 we obtain similar to
Eq. (E.6) ( 0 ) 0. The Goldstone boson fluctuations renormalize to zero, as found
in [57,58,61]. In our approximation vanishes for all T < Tc in the superfluid phase. In consequence, the mass term 2 0 of the radial mode vanishes for all T < Tc . However, as we have
discussed in the main text, this vanishing of concerns only the effective vertex at zero external
momentum, whereas the loop integrals often involve vertices at nonzero momentum.
Furthermore, the contribution from the fluctuations of the radial mode in Eqs. (E.4) and (E.12)
are not treated very accurately. First, the 1 2 22 vertex contains in principle a contribution 0
(F )
(F )
(F )
( the coefficient of the contribution ( 0 )3 ) which shifts + 2 0 and is
neglected here. Second, the structure of the inverse propagator of the radial mode is actually not
simply A q 2 with constant A . Indeed, the effective quartic coupling only vanishes for zero
external momentum, with a typical momentum dependence |q|. For a definition of a mass
term at q = 0 this is consistent with 0 0. However, this effect will then become visible
as an infrared divergence of the gradient coefficient for the radial mode, A ,r 0 |q|1 (which
differs from A for the Goldstone mode). In consequence, one has P (q 0) |q| and the
radial mode contribution to the SchwingerDyson equation is not infrared divergent. We note,
however, that the radial mode contribution is subleading as compared to the Goldstone mode
contribution such that our approximation catches the dominant features for the behavior of .
Finally, in the superfluid phase the potential (E.11) also contains a cubic term 1 22
1/2
(B)2
0 . This term contributes to the SchwingerDyson equation for m . The coefficient of this
269
(F )
contribution 0 vanishes for = 0. Nevertheless, for a momentum dependent one

has to take it into account, as well as similar corrections to the SchwingerDyson equation for
which involve quintic couplings.
Appendix F. Numerical procedures
In this section we give a short summary which equations are actually used for numerical
solutions. All quantities are given in dimensionless renormalized units. We first give a complete
list relating dimensionful, dimensionless and dimensionless renormalized parameters, couplings
and fields.
(i) Relations between dimensionful and dimensionless parameters (the Fermi energy is given by
F = kF2 /2M)
q =
q
,
k
T
,
T =
F
=
m
2
A = 2M A ,
m
2 =
3/2
= kF ,
3/2
= kF ,
F
,
F
c = ak
F,
,
F
2M h
h = 1/2 ,
kF
= 2MkF ,
r
r = h 2 = 2 .
F
= = kF3 ,
(F.1)
(ii) Relations between dimensionless and dimensionless renormalized parameters

A
,
Z

= 2 ,
Z
A =
m2 =
h =
m
2
Z
h
1/2
,
Z
= Z .
(F.2)
All other parameters, couplings and fields are invariant under a rescaling with Z . In particular,
note the invariance of the concentration c and the dimensionless superfluid gap parameter r .
In our approximation, we neglect details of the renormalization coefficients for the Matsubara
frequencies and use
Z,
= 1.
Z
(F.3)
The input parameters are the detuning , the temperature T and the Yukawa coupling h ,
with c following from Eq. (55). Alternatively, we could choose c1 , h , T . We have to solve the
field equations for the effective chemical potential (density equation) and the field expectation
value . The latter has the general form
u
(0 ) 0 = 0.
(F.4)
In the superfluid phase it determines the expectation value 0 . In the normal or symmetric phase
the field expectation value vanishes, 0 = 0. The field equation for is now replaced by the gap
equation for the boson mass term m2 = u(0)/. The two equations provide a closed system,
whose solution for and 0 or m2 determines all observables of the crossover problem.
270
In the superfluid or symmetry-broken phase we have 0 = 0 such that u(0 )/ = 0 must

Howhold. The density equation and the latter condition are then solved for and r = h 2 .
ever, the full four-boson coupling enters both the density equation and the gap equation.
Therefore, an additional gap equation for is needed, cf. Section 8 and Appendix E. Hence the
solution is given by , r , . In the zero momentum approximation we find = 0.
All other quantities of interest can then be reconstructed from the solution. For example, we
can obtain the bare noncondensed molecule fraction and the condensate fraction by
M = 6 2 kF3 n M ,
C = 6 2
r
h2
(F.5)
where n M is given by Eqs. (115). The dressed non-condensed molecule fraction and condensate
fraction are the given by
M = Z M ,
C = Z C .
(F.6)
Let us now give the explicit formulae used in the different regions of the phase diagram.
(i) Normal phase.
The full boson mass is determined by the condition
(F )2
m2 = m
1 (F )
M .
3 2
(F.7)
(F )2
Here M depends on the full boson mass term m2 , and m

be evaluated at r = 0.
The field equation for is equivalent to the condition
1 = F + M
which uses the dressed density fractions. Here

1
d 3 q 2
F = 6 2
e +1 ,
3
(2)

1
d 3 q (A q 2 +m2 )/T
e
1
M = 6 2
(2)3

2
3 (3/2) T 3/2
=
Li3/2 em /T .
2
A
(F )
(Eqs. (137), (138)) have to
(F.8)
(F.9)
(F.10)
(ii) Superfluid phase.

Now r and are determined by the equations
1 (F )
M = 0,
3 2
F,0 + M + C = 1.
)2
m(F
+
(F.11)
(F.12)
These equations are still coupled to a gap equation for (cf. Appendix E for details). Here M
is given by Eq. (115) (M = 6 2 kF3 nM ), C by Eq. (F.5) and

d 3 q
tanh
1
.
F,0 = 3 2
(F.13)
(2)3
271
(iii) Phase transition.

At the critical line T = Tc we have m2 = 0 = 0 and we solve
1 (F )
M = 0,
3 2
F + M = 1
(F )2
(F.14)
(F.15)
at r = 0 for and Tc . Here M is evaluated at m2 = 0 and reads

M

3 (3/2) (3/2) T 3/2
=
.
2
A
(F.16)
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
A. Einstein, Sitzungsber. K. Preuss. Akad. Wiss. 1924 (1924) 261.

A. Einstein, Sitzungsber. K. Preuss. Akad. Wiss. 1925 (1925) 3.
L. Cooper, Phys. Rev. 104 (1956) 1189.
J. Bardeen, L. Cooper, J. Schrieffer, Phys. Rev. 108 (1957) 1175.
C.A. Regal, M. Greiner, D.S. Jin, Phys. Rev. Lett. 92 (2004) 040403.
M. Zwierlein, A. Stan, C.H. Schunck, S.M.F. Raupach, A.J. Kerman, W. Ketterle, Phys. Rev. Lett. 92 (2004) 120403.
C. Chin, M. Bartenstein, A. Altmeyer, S. Riedl, S. Jochim, J. Hecker Denschlag, R. Grimm, Science 305 (2004)
1128.
G.B. Partridge, K.E. Strecker, R.I. Kamar, M.W. Jack, R.G. Hulet, Phys. Rev. Lett. 95 (2005) 020404.
A.J. Leggett, in: A. Pekalski, R. Przystawa (Eds.), Modern Trends in the Theory of Condensed Matter, SpringerVerlag, Berlin, 1980.
P. Nozires, S. Schmitt-Rink, J. Low Temp. Phys. 59 (1985) 195.
C.A.R. Sa de Melo, M. Randeria, J.R. Engelbrecht, Phys. Rev. Lett. 71 (1993) 3202.
H.T.C. Stoof, M. Houbiers, C.A. Sackett, R.G. Hulet, Phys. Rev. Lett. 76 (1996) 10.
R. Combescot, Phys. Rev. Lett. 83 (1999) 3766.
H. Heiselberg, C.J. Pethick, H. Smith, L. Viverit, Phys. Rev. Lett. 85 (2000) 2418.
S. Diehl, C. Wetterich, Phys. Rev. A 73 (2006) 033615, cond-mat/0502534.
M.L. Chiofalo, S.J. Kokkelmans, J.N. Milstein, M.J. Holland, Phys. Rev. Lett. 88 (2002) 090402.
A. Griffin, Y. Ohashi, Phys. Rev. A 67 (2003) 033603.
P. Pieri, G.C. Strinati, Phys. Rev. Lett. 91 (2003) 030401.
A. Perali, P. Pieri, G.C. Strinati, Phys. Rev. A 68 (2003) 031601.
A. Perali, P. Pieri, L. Pisani, G.C. Strinati, Phys. Rev. Lett. 92 (2004) 220404.
L.P. Pitaevskii, Zh. Eksp. Teor. Fiz. 40 (1961) 646.
E.P. Gross, Nuovo Cimento 20 (1961) 454.
E.P. Gross, J. Math. Phys. 4 (1963) 195.
S.J. Kokkelmans, M.L. Chiofalo, R. Walser, M.J. Holland, Phys. Rev. Lett. 87 (2001) 120406.
M.L. Chiofalo, S.J. Kokkelmans, J.N. Milstein, M.J. Holland, Phys. Rev. A 66 (2002) 043604.
A. Griffin, Y. Ohashi, Phys. Rev. Lett. 89 (2002) 130402.
E. Timmermans, K. Furuya, P.W. Milonni, A. Kerman, Phys. Lett. A 285 (2001) 228.
Q. Chen, K. Levin, J. Stajic, S. Tan, Phys. Rev. A 69 (2004) 063610.
Q. Chen, M.L. Chiofalo, M. Holland, K. Levin, J.N. Milstein, J. Stajic, Phys. Rep. 412 (2005) 1.
R.A. Duine, H.T.C. Stoof, J. Opt. B: Quantum Semiclass. Opt. 5 (2003) S212.
R.A. Duine, H.T.C. Stoof, Phys. Rep. 396 (2004) 115.
G.M. Falco, H.T.C. Stoof, Phys. Rev. Lett. 92 (2004) 140402.
G.M. Falco, H.T.C. Stoof, Phys. Rev. A 71 (2005) 063614.
C.A.R. Sa de Melo, M. Randeria, J.R. Engelbrecht, Phys. Rev. B 55 (1997) 15153.
F. Pistolesi, G.C. Strinati, Phys. Rev. B 53 (1996) 15168.
H. Vivas, cond-mat/0504600.
J. Carlson, S.-Y. Chang, V. Pandharipande, K. Schmidt, Phys. Rev. Lett. 91 (2003) 050401.
M.W.J. Romans, H.T.C. Stoof, cond-mat/0506282.
Q. Chen, K. Levin, cond-mat/0505689.
272
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
J. Hubbard, Phys. Rev. Lett. 3 (1959) 77.

R. Stratonovich, Dokl. Akad. Nauk. SSSR 115 (1957) 1097.
S. Diehl, Ph.D. Thesis, Heidelberg, 2006.
F.J. Dyson, Phys. Rev. 75 (1949) 1736.
J.S. Schwinger, Proc. Natl. Acad. Sci. 37 (1951) 452.
T. Baier, E. Bick, C. Wetterich, Phys. Rev. B 10 (2000) 15471.
J. Jaeckel, C. Wetterich, Phys. Rev. D 68 (2003) 025020.
Q. Chen, J. Stajic, K. Levin, Phys. Rev. Lett. 95 (2005) 260405.
P. Pieri, G.C. Strinati, Phys. Rev. B 68 (2003) 144507.
S. Diehl, H. Gies, J.M. Pawlowski, C. Wetterich, cond-mat/0701198.
D.-U. Jungnickel, C. Wetterich, Phys. Rev. D 53 (1996) 5142.
T. Gollisch, C. Wetterich, Phys. Rev. B 65 (2002) 134506.
D. Petrov, C. Salomon, G. Shlyapnikov, Phys. Rev. Lett. 93 (2004) 090404.
I. Brodsky, A. Klaptsov, Y.M. Kagan, R. Combescot, X. Leyronas, cond-mat/0507240.
G. Astrakharchik, J. Boronat, J. Casulleras, S. Giorgini, Phys. Rev. Lett. 93 (2004) 200404.
N. Tetradis, C. Wetterich, Nucl. Phys. B 398 (1993) 659.
J. Berges, N. Tetradis, C. Wetterich, Phys. Rev. Lett. 77 (1996) 873.
J. Berges, N. Tetradis, C. Wetterich, Phys. Rep. 363 (2000) 223.
T. Gollisch, C. Wetterich, Phys. Rev. Lett. 86 (2001) 1.
N. Tetradis, C. Wetterich, Nucl. Phys. B 422 (1994) 541.
Non-rational 2D quantum gravity:

I. World sheet CFT
I.K. Kostov a , V.B. Petkova b,
a Service de Physique Thorique, CNRS-URA 2306, CEA-Saclay, F-91191 Gif-Sur-Yvette, France
b Institute for Nuclear Research and Nuclear Energy, 72 Tsarigradsko Chausse, 1784 Sofia, Bulgaria
Received 1 February 2006; received in revised form 11 February 2007; accepted 23 February 2007
Abstract
We address the problem of computing the tachyon correlation functions in Liouville gravity with generic
(non-rational) matter central charge c < 1. We consider two variants of the theory. The first is the conventional one in which the effective matter interaction is given by the two matter screening charges. In the
second variant the interaction is defined by the Liouville dressings of the non-trivial vertex operator of zero
dimension. This particular deformation, referred to as diagonal, is motivated by the comparison with the
discrete approach, which is the subject of a subsequent paper. In both theories we determine the ground
ring of ghost zero physical operators by computing its OPE action on the tachyons and derive recurrence
relations for the tachyon bulk correlation functions. We find 3- and 4-point solutions to these functional
equations for various matter spectra. In particular, we find a closed expression for the 4-point function of
order operators in the diagonal theory.
1. Introduction
The exact results in Liouville theory obtained in the last decade [18] allowed to improve
some old techniques developed in c 1 string theories (reviewed in [913]) and find new links
between the world-sheet and matrix model descriptions. In particular, the fundamental OPE identities, used in [3,58] to evaluate various Liouville structure constants, are similar in nature to
the ground ring relations in string theories [1422]. Recently the ground ring structure was re* Corresponding author.
E-mail address: petkova@inrne.bas.bg (V.B. Petkova).

doi:10.1016/j.nuclphysb.2007.02.014
274
I.K. Kostov, V.B. Petkova / Nuclear Physics B 770 [FS] (2007) 273331
considered in [2325], where it was applied to study the solitons, or D-branes, in c 1 string
theories.
In this work we generalize and exploit this approach to derive and solve finite difference
equations for the tachyon correlators on the sphere. We recall that the ground ring is the ring with
respect to the operator product expansions (OPE), modulo QBRST -exact terms, of the physical
operators of zero ghost number. Physical operators of fixed ghost number, like the tachyons,
represent modules under the action of the ring [14].
So far the technique has been tested in the simplest model of 2D string theory that can be
considered as a marginal deformation by Liouville interaction of a two-component Gaussian
field action with background charges. One can perturb in a similar way the matter component of
the Gaussian field by the matter screening charges. As a result one obtains a gravitational analog
of the DotsenkoFateev Coulomb gas construction. The correlation functions depend on two
coupling constants: the Liouville coupling (cosmological constant) L and its matter counterpart
M , associated with the matter screening charge. Unlike most of the previous studies, which
deal with the minimal string theories, we shall consider non-rational values of the matter central
charge characterized by a real parameter b,

2
1
c=16
(1.1)
b .
b
One of the motivations for this work was to compare the correlation functions in the continuous (world sheet), and the discrete (target space) approaches to the 2D quantum gravity. This
is an old problem and there are few matrix model results on the correlation functions with a
non-trivial matter. Such results are known only for the simplest examples of the rational, minimal theories, as the Ising model [26], recently reconsidered in [27]. Moreover, there is no matrix
model to match the non-rational case, except the O(n) matrix model [28], whose poor field content is too restricted. In a subsequent paper [29] we construct such a model, in which the matter
degrees of freedom are parametrized by a semi-infinite discrete set, generalizing the ADE string
theories [30]. In this matrix chain modelwhose target space is the A Dynkin graph, one develops a finite diagram technique for the explicit computation of the n-loop amplitudes; see [31]
for an early application of this technique in the rational case. Shrinking the loops, one extracts
the n-point local correlation functions. However, this procedure is not unique on a fluctuating
lattice. Moreover, it happens that the most natural definition of the local fields leads to a different
interpretation of the matter screening than that in the conventional theory. Namely, the charge
conservation condition involves only multiples of the matter background charge e0 = 1/b b,
and the Liouville dressings of the order parameter fields on the diagonal of the infinite Kac table close under fusion. This has led us to introduce and study another variant of the c < 1 gravity
in the continuum. Instead of the matter screening charges, i.e., the tachyons of matter charge b
and 1/b, the interaction terms for the matter field are now generated by the two Liouville dressings of the vertex of charge e0 = 1/b b. The effective action of this diagonal string theory is
no more a sum of Liouville and matter parts. Nevertheless it is possible to extend to this theory
the ground ring technique and to find solutions of the corresponding functional equations. A class
of these solutions reproduces the 4-point tachyon correlators found in the discrete approach.
The paper is organized as follows.
After some preliminaries collected in Section 2, we determine in Section 3 the generic tachyon
3-point function as a product of the c > 25 Liouville [1,2] and a c < 1 matter OPE structure
constants. The latter constant is given by an expression derived as in [3], which in particular
reproduces the Coulomb gas OPE constant of [32]; see also [33].
275
Fig. 1. The general structure of the 4-point function.
The ground ring is discussed in Section 4. Since its elements are built of vertex operators,
such that both the matter and Liouville parts are labelled by degenerate Virasoro representations,
one can compute their OPE with the tachyons using the free field Coulomb gas in the presence
of integer number of screening charges. We determine the action of the two generators of the
ground ring, a and a+ , on a tachyon of arbitrary momentum, taking into account both Liouville
and matter interactions. This amounts in the computation of the general 3-point function of two
tachyons and one ring generator, the details are collected in Appendices A.1, A.2. The result
confirms the expectation, see [24] for the rational case, that the deformed ring is isomorphic to
sl(2) sl(2) type fusion ring. In the diagonal matter string theory the deformed ground ring
element a+ a generates a diagonal sl(2) projection, preserving in particular the order parameter
tachyons.
We use this action in Section 5 to derive recurrence functional relations for the bulk tachyon
4-point correlators. The equations are written for the correlators satisfying the chirality rule
in the terminology of [9] and they extend the ones previously obtained for the case of Gaussian
matter [17,18,25]. The contact 3-point terms in these relations require the computation of a set
of free field 4-point functions containing a ring generator and an integrated tachyon, see Appendices A.3, A.4. Appendix A.5 contains a computation of some chiral OPE constants relevant
for the boundary ground ring, in particular the OPE coefficients of the boundary ground ring
generators and a tachyon of generic momentum.
In the remaining four sections we look for solutions of the equations for the 4-point tachyon
correlation functions. In Section 6 we reproduce the correlators found in [34] for the case of
Gaussian matter field. The set of fixed chirality solutions are shown to serve as a local basis
for another, fully symmetric in the momenta correlator, also described in [34]. We interpret this
symmetry as locality requirement and discuss the relation between the two types of correlators
and the respective equations they satisfy. In Section 7 we solve the functional equations for a
class of correlators such that the total matter (or total Liouville) charge can be compensated by
integer number of screening charges. In Section 8 we find correlators in the diagonal theory
in which one or all four tachyons is degenerate, i.e., its momentum labels a degenerate c < 1
Virasoro representation on the diagonal of the infinite Kac table. The case of degenerate fields in
the conventional, non-diagonal theory, is considered in Section 9. Here we find 4-point functions
with one matter (or Liouville) degenerate and three generic momenta. A formula for the 4-point
correlators with four degenerate fields is conjectured by analogy with the diagonal case.
The results for the 4-point tachyon correlators as functions of the momenta P1 , P2 , P3 , P4 are
summarized by a partial wave expansion formula sketched in Fig. 1.
The 4-point function is a sum of 1-particle irreducible (1pi) piece and a sum of the contributions of the three channels:

1pi

GP()
(1.2)
GP1 ,P2 ,P3 ,P4
(NP1 ,P2 ,P P NP
,P3 ,P4 + permutations),
1 ,P2 ,P3 ,P4
P
where is a sign determined by the chiralities. The form of the 1pi term G 1pi and the interpretation of the 3-point fusion multiplicities in (1.2) depend on the considered spectrum of
276
momenta. The case of Gaussian matter in [34] corresponds to a single contribution in each of
the three channels in (1.2). The formula (1.2) is symmetrised with the intermediate momentum
P replaced by |P |. This universal choice solves the locality requirement preserving the fusion rules, which in the conventional theory typically match those determined by the underlying
c < 1 (or c > 25) local correlators. It is further supported by a recursive procedure extending
the initial identities to equations for the local correlators. In the diagonal theory the term G 1pi
is proportional to the corresponding 4-point fusion multiplicity NP1 ,P2 ,P3 ,P4 , also expressed in
terms of the 3-point vertices. The diagonal theory correlators are not a special case of those in
the conventional string theory.
We conclude in Section 10 with a summary of the results and a discussion on the open problems of this investigation.
This paper is a detailed and extended presentation of the results announced in the short letter [35] and reported at conferences in Dubna, Bonn, Varna and Santiago de Compostela. In the
mean time, two papers appeared, which partially overlap with our results. Ref. [36] deals with
the boundary ground ring relations in the minimal c < 1 gravity. In Ref. [37] a different method
for evaluating the bulk tachyon correlation functions in 2D gravity is developed. The class of
4-point correlators computed in [37], namely those containing one matter degenerate field, are of
the type discussed in our Section 9; see the text for a comparison. We have been also informed by
V. Fateev about an unpublished recent work of him on the direct computation of some particular
examples of such correlators.
2. Preliminaries: Effective action, observables
2.1. Effective action
Consider Liouville gravity on the Riemann sphere. The effective action in the conformal gauge
and locally flat reference metric g ab = ab is a perturbation of the Gaussian action

1

1
z c ].
Afree =
d 2 x (a )2 + (a )2 + (Q + ie0 )R g +
d 2 x [bz c + b
4
(2.1)
Here is the Liouville field, is the matter field, and {b, c} is a pair of reparametrization ghosts
of scaling dimensions {2, 1}. The reference scalar curvature R is localized at the infinite point.
The Liouville and matter fields background charges Q and e0 are parametrized by a real constant b,
Q=
1
+ b,
b
e0 =
1
b.
b
(2.2)
We will consider the generic situation when b2 is not a rational number. With the choice (2.2)
the full central charge is

1
1
2
2
ctot cM + cL + cghosts = 13 6 b + 2
+ 13 + 6 b + 2
26 = 0. (2.3)
b
b
We consider a marginal deformation of the Gaussian action (2.1) by the Liouville interaction and
its matter counterpart, which is one of the two screening charges in the c < 1 CFT,

Aint = d 2 x L e2b + M e2ib ,
(2.4)
A int =
277

d 2 x L e2/b + M e2i/b .
(2.5)
In fact the interaction depends on the type of the correlators to be computed. In some cases we
shall take into account one or the two of the terms in the dual action, but it will be reduced
to a source of integer number of screening charges. We shall refer to the above theory, in which
the Liouville and matter parts of the action factorize, as the conventional c < 1 string theory.
In Section 4.1 we will also introduce another interaction which does not have this factorization
property.
2.2. Vertex operators for the closed string tachyons
We shall consider the BRST invariant fields that correspond to the vertex operators e2ie of
the matter CFT. In general, the matter vertex operators should be dressed by Liouville vertex
operators, e2 ,

Ve, = 1 2 + e2 e2ie e2 ,
(2.6)
so that the dressed operator has conformal weight (1, 1) [38,39]:
M (e) + L () = e(e e0 ) + (Q ) = 1.
(2.7)
To have simpler expressions we normalized by the leg factors as in [34], where we used the
standard notation (x) = (x)/ (1 x).
The condition (2.7) is the on-mass-shell condition for the closed string tachyons propagating
in the euclidean 2D target space. The simplest examples satisfying it are the four operators in
which only one of the two vertex operators in (2.6) appears, i.e., the sources of the Liouville
and matter screening charges in (2.4), (2.5). In general the solutions can be parametrized by the
tachyon target space momentum P and the chirality = 1.1 The matter and Liouville charges
are expressed in terms of P and as
1
1
= (Q P ) = e + b , = 1.
e = (e0 P ),
(2.8)
2
2
We denote by V := Vb , the vertex operators (2.6) with and e related by (2.8). To com()
pare with the microscopic theory it is convenient to introduce also the alternative notation VP ,

()
VP V = b1 P ei(e0 P )+(QP ) ,
(2.9)
where we used the relation
1 2 + e2 = b (Q 2) = b P .
(2.10)
We shall not restrict in general to the physical Seiberg bound P = Q 2 > 0

The BRST invariant operators can be represented in two pictures: either as (1, 1)-forms integrated over the world-sheet, or as QBRST -closed 0-forms:
2
d x
()
()
TP T =
(2.11)
or WP W = ccV .
V

[40].2
1 For simplicity we shall assume, unless stated otherwise, that the momenta are real, P R.
2 We recall that in the rational minimal gravity [41] the operators corresponding to the solutions of (2.7) with e and
e0 e are identified, and one is left with the two Liouville dressings, distinguished by the sign of Q/2 . We shall not
assume in general such identification.
278
In the n-point tachyon correlators, n 3 vertex operators are integrated over the worldsheet,
and three are placed, as usual, at arbitrary points, say 0, 1 and , and the ghost zero mode
contribution c1 c0 c1
is normalized to 1. The correlation function should not depend on the
choice of the three operators.
2.3. Normalization of the couplings and duality transformations
It is also convenient to redefine the couplings in the effective action (2.4) according to the
normalizations (2.9) of the vertex operators,

(+)
+
2b
(+)
= L Tb = L Te0 ,
M e2ib = M T0+ = M TQ ,
L e

2i
2
()
e b = M T0 = M TQ ,
= L Te()
,
L e b = L T1/b
(2.12)
M
0
where the new coupling constants are related to the old ones by

L = b2 L ,
M = (b2 )M ,

1
L ,
b2

1
= 2 M .
b
L =
(2.13)
All correlation functions in Liouville theory are invariant (for fixed charges) w.r.t. the substitution [5]
1
b ,
b
1/b2
L L = L
(2.14)
As we will see in the next section, for a consistent description of the matter correlation functions
one should introduce the dual matter coupling constant so that the functions in the c < 1 theory
obey the symmetry
1
b ,
b
M M = (M )1/b .
2
(2.15)
The duality transformation (2.14) (or (2.15), or their composition) relates the tachyon correlators
to those of a conjugated theory, obtained by flipping the sign of e0 (or Q, or both) respectively;
thus effectively we can restrict the real parameter b to the region (0, 1). On the other hand,
the composition of (2.14) with and M M preserves the free action (2.1) and
interchanges the two interaction actions (2.4), (2.5). The same effect yields the matter duality
transformation (2.15), accompanied with and L L . In parameter space these duality transformations are formulated as

1
{b, L , M , Pi , i }
, L , M , Pi , i ,
b

1
{b, L , M , Pi , i } , L , M , Pi , i .
b
(2.16)
(2.17)
279
3. The tachyon 3-point function as a product of Liouville and matter 3-point functions
In this simplest case the correlation function factorizes to a product of the matter and Liouville
three-point OPE constants
2 3
G31
C Liou (1 , 2 , 3 )C Matt (e1 , e2 , e3 )

(1 , 2 , 3 ) = W11 W22 W33 =
.
3
2
2
j =1 (j ej )
(3.1)
The case of the 3-point function is unique in the sense that for n > 3 the factorization holds
only before the integration over the n 3 moduli while for n < 3 there is a residual conformal
symmetry which does not allow the direct evaluation.
3.1. The case
e i = e0

First we assume that
ei = e0 , in which case the matter 3-point OPE constant is equal to
one. For the Liouville 3-point OPE constant we take the DOZZ formula [1,2]
Q1 2 3
1/b
(b) (21 ) (22 ) (23 )
C Liou (1 , 2 , 3 ) = L b2e0
(3.2)
1 ) ( 2 ) ( 3 )
(123 Q) (23
12
13
3 = + ,
with notation 12
1
2
3
123 = 1 + 2 + 3 , etc. Here i and ei are solutions (2.8)
of the mass-shell condition (2.7). Imposing the constraint

ei = e 0
i i bi = e0
(3.3)
i
and using the basic property of the function = b = 1/b ,

(x + b )
= b(12b x) xb , = 1
(x)
one checks that the constant (3.2) reduces to
1
Liou
Lb
(1 , 2 , 3 ) = 3
i=1 b
(3.4)
(Q1 2 3 )
i [bi (Q 2 )]
i
for

i i bi = e0 .
(3.5)
Thus one finds for the 3-point function (3.1), using (2.10),
1

G31 2 3 (1 , 2 , 3 ) =
Lb
(Q1 2 3 )
b1 +2 +3
(3.6)
Comment:
To compare with the expressions in [34] and the perturbative Coulomb
gas computation let us

choose, say, the action (2.4). The constraint (3.3) combined with 3i=1 i Q = sb for a positive integer s implies 23 = 1/b sb or 23 = (s 1)b for the choice of chiralities (+ + )
or ( +) respectively. (For the other two choices ( ) the above two conditions are inconsistent.) The Coulomb gas computation of the unnormalized by the leg factors 3-point functions
gives for generic values of 1 , 2 a finite expression for the case (+ + ) and zero for the case
( +).3 On the other hand in both cases the normalized by leg factors correlators are rendered
3 As a consequence, the derivatives with respect to , i.e., the (unnormalized) n-point tachyon correlators
L
G+++
(1 , 2 , 3 , b, . . . , b) are all vanishing.
n
280
finite by the overall zero mode divergence factor [34] and this corresponds to our expression (3.6).
For the dual interaction (2.5) the roles of (+ + ) and ( +) are interchanged.
3.2. A general formula for the matter 3-point function
Now consider the case of arbitrary matter charges e1 , e2 , e3 , when the matter OPE constant is
no more equal to one. The general c < 1 matter 3-point OPE constant satisfies the identities
1 + b))
1 (b(2e1 b)) (b2e1 ) (b(e23
C Matt (e1 b, e2 , e3 )
,
=
3 )) (b(e2 ))
C Matt (e1 , e2 , e3 )
b4 M (b(e123 e0 )) (b(e12
13
1 1 ))
C Matt (e1 + b1 , e2 , e3 )
b4 ( b1 (2e1 + b1 )) ( b1 2e1 ) ( b1 (e23
b
=
.
3 )) ( 1 (e2 ))
C Matt (e1 , e2 , e3 )
M ( b1 (e123 e0 )) ( b1 (e12
b 13
(3.7)
The change of variables b 1/b, M M interchanges the two relations (3.7). These functional relations come from the locality requirement on the 4-point matter functions with one of
1
inserted. Their derivation is analogous to the
the two simplest degenerate fields e = b2 , e = 2b
one for the Liouville case in [3], where the DOZZ formula (3.2) was reproduced as the unique
solution of the cL > 25 functional relations for positive, irrational b2 . Identifying the dual coupling constant as in (2.15), the solution of (3.7) is expressed in terms of the Liouville constant
C Liou in (3.2), with i = i ei + bi , i = 1, 2, 3 [35]4

1 (e0
Matt
(e1 , e2 , e3 ) = 3
M b
i=1 b
i ei )
Lb
i (bi (Q 2 ))
i
(Q
i )
C Liou (1 , 2 , 3 )
The relation holds for any choice of the three signs i . The overall constant is fixed by

ei = e 0 ,
C Matt (e1 , e2 , e3 ) = 1 for
(3.8)
(3.9)
which is checked using (3.5).

The formula (3.8) is obtained alternatively as analytic continuation of the particular (thermal)
DotsenkoFateev constant computed with
one of the matter screening charges, in full analogy
with the derivation of (3.2) in [1,2]. For i ei e0 = mb nb , n, m non-negative integers, the
expression (3.8) for the matter structure constant is finite for generic b2 and reproduces the 3point DotsenkoFateev constant in (B.10) of [32], times the powers (M )m ( M )n . In other
words, in the Coulomb gas range of the three parameters ei (3.8) can be looked as a compact
representation of the DF constant. Introducing the function
b (x) :=
1
1
=
= b (e0 x) = 1 (x)
b
b (x + b) b (x + b1 )
(3.10)
we can rewrite (3.8) in a form analogous to the DOZZ formula (3.2)5

e +e +e e
1/b
C Matt (e1 , e2 , e3 ) = M b2Q 1 2 3 0
(0) (2e1 ) (2e2 ) (2e3 )

.
1 ) (e2 ) (e3 )
(e123 e0 ) (e23
12
13
(3.11)
4 The derivation of the matter constant C Matt (e , e , e ) with a different choice of the normalization, has been carried
1 2 3
out independently by Al. Zamolodchikov, published in [33].

5 This expression and the function (3.10) have been earlier considered, see e.g. [42], without discussion of the relation
to the c < 1 DF constant.
281
The functional relations (3.4) are replaced by

(x + b1 )
1
= x b12x/b .
b
(x)
(x b)
= (bx)b12bx ,
(x)
(3.12)
The logarithm of the function (x) admits an integral representation as the one for the logarithm
of , with Q replaced by e0 (so that it is invariant under the change b 1/b),
log b (x) =
dt
t
e0
x
2
2
e
sinh2 ( e20 x) 2t
t
sinh b 2t sinh 2b
= log 1/b (x),
(3.13)
which converges (for b > 0) in the strip b < Re x < b1 .

Examples:

C Matt
b
b
, e, e0 e +
2
2

= M
((2e e0 )b)
,
(b2 ) (2eb)

((e0 2e) b1 )
1
1
C Matt , e, e0 e
= M
.
2b
2b
( 12 ) (2e b1 )
(3.14)
The Liouville three-point OPE constant satisfies the reflection property [2]
C Liou (1 , 2 , 3 ) =
1 ( 1 (21 Q))
1 Q2
b
b
C Liou (Q 1 , 2 , 3 ).
L
(b(Q 21 ))
b2
(3.15)
This identity has been used to write the r.h.s. of (3.8) in various equivalent ways corresponding
to the different choices (2.8) of the relation between ei and i . Analogously,
2e1 e0
C Matt (e1 , e2 , e3 ) = b2 M b
(b(2e1 e0 ))
( b1 (2e1 e0 ))
C Matt (e0 e1 , e2 , e3 ).
(3.16)
In particular for e1 = e2 = e, e3 = 0 (3.16) implies

2ee0
C Matt (e, e, 0) = b2 M b
(b(2e e0 ))
( b1 (2e e0 ))
(3.17)
The last formula reproduces the 2-point constant found in [43]. As pointed out in [43], the choice
e3 = 0 does not force e1 = e2 , or e1 = e0 e2 ; see also [42] for a different solution in the
case c = 1 and a different interpretation of the identity operator which avoids this problem,
not essential for our purposes.6 The construction of a consistent non-rational matter theory for
arbitrary momenta is beyond the scope of this paper.
6 We recall that the DF constants [32,44] for rational b2 are non-vanishing for some values in the minimal spectrum,
but beyond the restrictions of the fusion rules. In that sense these constants by themselves do not determine the fusion.
However, due to certain identities satisfied by the constants and the fusing matrix elements, there occur cancellations in
the block expansion of the local 4-point functions, so that each channel is consistent with the fusion rules; see e.g. [45].
282
3.3. The tachyon 3-point function for generic momenta

Having evaluated the matter three-point function (3.8), we insert it together with (3.2) in (3.1)
and obtain a simple expression for the tachyon 3-point function. If we factorize the dependence
on the coupling constants L , M and the powers of b, denoting the remaining momentumdependent factor by NP1 ,P2 ,P3 , we get
1
1 2 3
(P1 , P2 , P3 ) =
L2b
= b
i i Pi Q)
1
2b
M
(e0
Pi )
NP1 ,P2 ,P3
b1 +2 +3
i i
Lb
(Q
i ) b1 (e0
M
i ei )
(3.18)
or NP1 ,P2 ,P3 = 1. This reduces to (3.6) for M = 1. The simple 3-point function (3.18) satisfies
reflection properties inherited from those of its Liouville (3.15) and matter (3.16) parts. That is,
in the expectation value (3.1) given by (3.18) the following identities hold:
P /b
WP(+) = b2 L WP() = b2 M
P /b
()
(+)
WP
= (L /M )P /b WP
.
(3.19)
Notice however that the relations (3.19) do not necessarily hold within the 4-point functions, as
will be discussed below.
Given the expression (3.18) for the 3-point constant, the 2-point tachyon correlators are
conventionally defined [1] as integrals of 3-point ones over some of the interaction constants.
(b, , ) =
For example, the 2-point tachyon function for e1 + e2 = e0 is determined from G++
3
L G+
(,
),
=
e
+
b
=
Q/2
P
/2,
i.e.
1
2
P /b
L
+
(,
)
=
G
(P
,
P
)
=
G+
2
2
P
1
= (Q 2)b2 Lb
(Q2)
(Q2)
b
= L
(Q 2)
+
0|c(z)zc(z)z W+ (z, z )WQ
(z , z )|0
.
(3.20)
The composition of matter times Liouville reflections reproduces up to a sign the correlator
++
(0, , Q ) = M G+
G+
2 (e, e) determined analogously from G3
2 (, Q ).
The same convention leads to the partition function Z and its dual Z defined as
3L Z(L , M , b)
= Wb+ Wb+ Wb+
Z(L , M , b) =
b Qb eb 0
,
Qe0 L M
L , M , b)
3 Z(
L
e
Q
1
1
0
bQ be

= W
Lb M b ,
L M 0 =
1 W 1 W 1
Z(L , M , b) =
bQe0
bQe0
b
b
b

1
1
L , M , b) = Z(L , M , b).
= Z(
Z L , M ,
b
b2
(3.21)
We can then introduce normalized functions

2 3

i ei
(1 , 2 , 3 ; L , M , b)
i i
= b i i 1 e0 QL b M b ,
Z(L , M , b)

1 2 3

i ei
(1 , 2 , 3 ; L , M , b)
G3
ib i
i i +1
=b
e0 QL
M b .
L , M , b)
Z(
G31
(3.22)
283
These correlators are interchanged by the duality transformations (2.16), (2.17), which become
equivalent since (3.22) is expressed only through the variables7

2

Pi
1
b , L , M , bPi , i 2 , L , M , , i .
(3.23)
b
b
Furthermore each of the simple constants (3.22) is invariant under analytic continuation b
ib, transforming the charges and the coupling constants as
{b, L , M , ei , i } {ib, M , L , ii , iei }
(3.24)
or, in terms of target space momenta, Pi ii Pi , i i . This matter-Liouville duality

transformation reflects the invariance of the actions (2.1), (2.4) and (2.5) under the respective
interchange of the matter and Liouville fields {, } {, }.
The final simple expression for the 3-point functions (3.18) satisfies the identities

b
b
++
++
M G3
1 , 2 , 3 = G3
1 , 2 , 3
2
2

b
++
= L G3
(3.25)
1 , 2 + , 3 ,
2
and hence,
L ++
G
(1 , 2 , 3 ).
(3.26)
M 3
The last identity is also a direct consequence of the matter functional relations (3.7) and the
corresponding Liouville ones [3], and can be used itself as a relation determining the 3-point
tachyon correlator.
However the r.h.s. of the matter (3.7) or the corresponding Liouville functional relations may
become singular, so both the functional relation (3.26) and the simple solution (3.18) are valid
for generic momenta. For momenta such that some of the factors in (3.1) becomes singular, there
is a 0 indeterminacy. This ambiguity leads us to reconsider the problem of determining the
tachyon 3-point function and not rely on factorization. Then the arbitrary multiplicity factors
NP1 ,P2 ,P3 in the first line in (3.18) must satisfy a pair of difference equations, to be derived below
as part of the set of functional identities for the n-point tachyon correlators. These equations are
weaker than (3.25), (3.26):
(1 , 2 b, 3 ) =
G++
3
NP1 +b ,P2 ,P3 + NP1 b ,P2 ,P3 = NP1 ,P2 +b ,P3 + NP1 ,P2 b ,P3 ,
= 1.
(3.27)
The expression (3.18), i.e. NP1 ,P2 ,P3 = 1, is only the simplest of their solutions. We shall deal
with basically two deviations from this generic solution. One is the case when the factor NP1 ,P2 ,P3
has the meaning of a fusion multiplicity and can take values 1 or 0. In the second, this factor
will be rather a distribution. Thus in the simplest example of a Gaussian matter NP1 ,P2 ,P3 =
(P1 + P2 + P3 e0 ) replaces the normalization condition (3.9). The duality relations for the
3-point correlators are preserved if NP1 ,P2 ,P3 (b) = NP1 ,P2 ,P3 ( b1 ) = NP1 ,P2 ,P3 ( b1 ).
7 The first duality relation is satisfied by the constants (3.18), but the second holds up to a sign, i.e.
G {i } ({Pi }; b, L , M ) = G {i } ({Pi }; b1 , L , M ) = G {i } ({Pi }; b1 , L , M ). Nevertheless to simplify notation
we shall work with the unnormalized correlators or with other normalizations. Another possible though less intuitive
definition is to relate the partition function to the correlator Wb Wb+ Wb+
; according to (3.19) this removes a factor
e /b
b2 M 0
from Z.
284
4. The ground ring

The ground ring operators are BRST invariant fields obtained by applying raising operators of
level rs 1 to the product of two degenerate matter and Liouville fields with Kac labels r, s. The
resulting operators have conformal weights (0, 0), see [46] for an explicit construction of some
of the corresponding states. The ring is generated by the lowest two operators a = a (z)a (z)
[14],

b((z)i(z))
1
:
a (z) = : b(z)c(z) z (z) + i(z) e
b

1
a+ (z) = : b(z)c(z) bz (z) i(z) e b ((z)+i(z)) :
(4.1)
The derivatives z a and z a are QBRST -exact, and therefore any amplitude that involves a
and other BRST invariant operators does not depend on the position of a . This property allows
to write recurrence equations for the correlation functions from the OPE of a and the tachyons
W [1618,25], which will be generalized below.
4.1. The action of the ring generators on the tachyons
The recurrence equations were initially derived for the free fields with no interaction, or at
most accounting for the perturbative first order contribution of the Liouville interaction. The
momenta were therefore assumed to satisfy the charge conservation, or neutrality condition

1
(e0 Pi )
ei = e 0 .
2
i
(4.2)
More generally, treating the Liouville and matter screening charges in (2.4) and (2.5) as perturbations amounts to modifying the original ring generators as

a a 1 L Tb+ + = a ,

a+ a+ 1 L T1/b
(4.3)
+ = a + .
Summarizing, for generic momenta, i.e., taking any complex values excluding the lattice
1
L := Zb + Z ,
b
(4.4)
one finds that the action of the ring generators on the tachyons W = ccV of given chirality
contains two terms, up to QBRST commutators:
a W+ = L W +
+ b2
a W
M W + b ,
2
= L M W b
+ 2
a + W = L W
1
+ 2b
W b ,
(4.5)
M W
W+
a + W+ = L M W +
1
+ 2b
or, in the alternative notation,
1
2b
1
2b
(4.6)
(+)
= L WP b M WP +b ,
()
= WP b L M WP +b ,
()
= L W
(+)
= W
a WP
a WP
a + WP
a + WP
(+)
(+)
()
()
()
P + b1
(+)
P + b1
285
M W
(4.7)
()
,
P b1
(+)
L M W 1 .
(4.8)
Pb
The duality transformations interchange the two ring generators and the two pairs of module
relations.
The OPE coefficients in (4.5), (4.6) are found either by direct evaluation of the 3-point function of the ring generator and two tachyons in the presence of a number of screening charges,
or by exploiting the factorization to the related known c < 1 and c > 25 Coulomb gas 3-point
(,)
constants, see Appendix A for more details. The coefficients, say in (4.5), C b , are expressed
as products of the corresponding matter and Liouville constants
1+ 1
(b (Q 2)) (Q 2)2
b2
= M2 L2 =
C (,)
b
2
(b (Q 2 + b))
b2

b
b
b
b
C Matt , e, e0 e + C Liou , , Q + ,
2
2
2
2
= 1.
(4.9)
Similar formula holds for (4.6). The matter OPE constants in (4.9) are either
1,
as
in
(3.9),
or

given by the first example in (3.14). The constants C Liou (1 , 2 , 3 ) with i i Q = ( 1) b2
are the analogous c > 25 Coulomb gas expressions, which are alternatively obtained as residue
of the Liouville constant C Liou in (3.2).
The case of momenta in the lattice L
For some momenta P L on the lattice (4.4), the free field 3-point function determining
the OPE coefficients is non-vanishing for more than two values, leading to additional terms
in (4.5) and (4.6). This typically requires an integer power of one of the screening charges in
the dual interaction action, while the generic OPE (4.5) and (4.6) correspond to deformations
with the respective actions (2.4), (2.5); see Appendix A.2 for details. The additional OPE terms
correspond to reflections with respect to the matter or Liouville, or both, charges of the terms in
(4.5) and (4.6). In the first two cases the chirality is inverted.
Let us restrict the consideration to momenta labelled by degenerate matter representations

n
L
(4.10)
:=
P
2e
=
L.
mb
0
M
b
m,nN
For one of the signs in (4.10), P = Pm,n = n/b mb, the r.h.s. of (4.5) and (4.6) contains two
more terms,8 while for the other sign, P = mb n/b, the generic formulae (4.5), (4.6) hold. For
example,
1
(2ebe0 )
b
M
W+ b
2
W+ b
2
b
+ b 2 M
b2
b2
b2
1
W b ,
(2ebe0 )
W+ b ,
2
(1 1)b
n
+ mb ,
2
b
(1 1)b mb
n
2 Q =
+ .
2
2
2b
2 Q =
8 This happens as well for the border lines m = 0, n = 0 outside of (4.10).
(4.11)
(4.12)
286
In (4.11) and (4.12) appear the combinations invariant under matter reflection,
1
1 2b1 (2ee0 )
2b (e0 2e)
W+ + M
W
W + : = bM
b
1

2b (e0 2e)
= bM
b(Q 2) e2ie + C Matt (e, e, 0)e2i(e0 e) e2
(4.13)
or, in terms of momenta,

1
1
() = b1 2b Pm,n W () + 1 2b Pm,n W () , Pm,n = n mb.
W
Pm,n
Pm,n
Pm,n
M
b
b1 M
The relative constant for the unnormalized vertex operators in this linear combination is given by
the 2-point function (3.17).
A similar argument can be carried out for the momenta labelled by degenerate Liouville representations

n
L
(4.14)
:=
P
2
=
mb
+
L.
L
b m,nN
In this case Liouville reflected terms appear in the OPEs which correspond to the plus sign in
(4.14).
In fact the appearance of the Liouville or matter reflected points is universal and is a consequence of the properties (3.15) and (3.16) of the 3-point functions (3.2) and (3.11); the only
peculiarity of the degenerate cases discussed here is that both OPE coefficients are given by a
Coulomb gas 3-point correlator, while in general the reflection images correspond to functions
satisfying a relation obtained by a reflection from the Coulomb gas charge conservation condition. Taking this into account in particular removes the above asymmetry of the OPEs for the
tachyons of momenta Pm,n and Pm,n , related by a matter charge reflection. To make sense of
these relations one should be able to identify in the correlators the matter reflected tachyons in
(4.13) (or the Liouville reflected ones in the case of degenerate Liouville case (4.14)). At this
stage we shall merely assume that the action of the ring generators is given again by the generic
formulae (4.5) and (4.6). This is analogous to what is done in field theory, where only one of
the two charges of the same dimension is included in the block decomposition of the 4-point
functions.
4.2. The ground ring at non-rational b2
Assuming that the tachyons at the border lines of the degenerate set (4.10) vanish (at least
in the averages), one gets a semi-infinite set, in one to one correspondence with the irreps of
sl(2) sl(2). The modules of given chirality are generated from the corresponding tachyon of
momentum P = e0 serving as an identity. After absorbing the constant L in the normalization
of the vertex operators, the relations (4.5) and (4.6) are equivalent to the multiplication rule of
the characters of sl(2) irreps of dimensions respectively m and n, with the character of the fundamental representation of dimension 2. It allows to represent any character as a polynomial of the
fundamental onethe above rule is the functional identity defining the Chebyshev polynomials
Um1 of second kind. Analogously (4.5) and (4.6) imply (setting M = 1)

1
1
1
(Pmn e0 )
()
WPmn = L2b
(4.15)
Um1 O21 Un1 O12 We()
,
0
2
2
1
287
with O21 = L 2 a , O12 = L 2 a + . The polynomial acting on the tachyon We()

0 represents
the ground ring element Omn . The formula (4.15) derived from (4.5) and (4.6) confirms the
Ansatz in [24] used in the context of the minimal string theory, see also [37]. The equalities
(4.5), (4.6), and hence (4.15), all hold true up to Q-exact terms, which in general disappear only
in the 3-point tachyon functions.
4.3. Further OPE channels

The two-term relations (4.5) and (4.6) describe the OPEs of the ring generators a perturbed
by the screening charges in the action. In presence of one or more tachyons, given by an integrated vertex operators, as happens in any n-point function with n > 3, the OPE will contain more
terms. Indeed now any integrated tachyon serves as a screening charge. Instead of computing
explicitly in an operator form the Q-exact terms, appearing in the product of free exponential
fields, and then moving them to the right or left, one can compute all possible OPE relations

which send a tachyon W to a tachyon W

a W T11 Ttt W .
(4.16)
Thus the effect of the skipped Q-exact terms (if any) is already accounted for in the added new
OPE channels, which in turn are valid again up to such terms.
The mass-shell condition implies a relation on the possible combinations of chiralities and
momenta in (4.16), see Appendix A.4 for a summary of the consequences of these constraints.
The coefficients of these OPEs are computed from free field correlators, see Appendix A.4, here
we summarize these results.
The simplest example consistent with the mass-shell condition is given by the OPE relations
[17,18]
a W+ T+1 = W +
+1 b2
a+ W T1 = W
1
+1 2b
(4.17)
The relations (4.17) have been already used for particular values of 1 and the chiralities in the
derivation of the linear terms in the first lines of (4.5), (4.6). They are generalized for generic
values of to a whole series, with k = 0, 1, 2, . . . ,9
k
(T0 T1/b
)
= W+
,
+1 b2 + bk
k!2
(T + T + )k
.
a+ W T1 0 b2 = W
1
+kb
+1 2b
(k!)
a W+ T+1
(4.18)
(4.19)
Taken for different k the relations demonstrate the effect of the Q-exact terms,

k

k
+ + k
a+ W T1 T0+ Tb+ = (0 + )T1 T0+ Tb+ = W
1 + T0 Tb
+1 2b

2
+ + kp
k
= p!2
W
+
T0 Tb
1
+pb
+
1 2b
p
= k!2 W
1
+kb
+1 2b
+ .
9 The first non-trivial example k = 1 of these OPE coefficients has been computed by P. Furlan.
(4.20)
288
For the product of interacting fields we obtain combining with (4.5), (4.6),
(++)
(+++)
a W+ T+1 =
C b W+ T+1 +
C b W+

2 1

= L W +
+ b2
M W +
b2
T+1 +
k
L M W +
P +P1 2k+1
b
k=0
()
()
a + W T1 =
C 1 W T1 +
C 1
W
2b
2b 1

()
() +
= L W 1 M W 1 T1 +
(L M )k WP+P1 +(2k+1)b .
P+b
Pb
k=0
(4.21)
The 4-point OPE coefficients in (4.21) are expressed in terms of products of matter and Liouville
Coulomb gas 3-point constants, see formula (A.30) below. The relations hold for values for
which each of those constants is well defined. In particular, (4.18) extends to degenerate values
of (with shifted compared with (4.10), (4.14) notation),
n+1
(m + 1)b, m, n Z0 ,
b
for any k n. Similarly (4.19) extends for k m if
P=
(4.22)
n+1
(m + 1)b, m, n Z0 .
(4.23)
b
For these values the infinite sums in (4.21) truncate to the first n + 1 (respectively m + 1) terms,
see Appendix A.4. We can interpret the set {W + 1
2k , k = 0, 1, 2, . . .} as the states of a
P =
P1 b +P
sl(2) Verma module of h.w. bP = n + 1. The state k = n + 1 of weight bP corresponds to the

singular vector and it is set to zero for the (n + 1)-dimensional irrep.
The kinematical mass-shell constraints imply that for generic momenta P , P1
/ L and
P + P1
/ L the identities (4.21) exhaust all OPEs in presence of one integrated tachyon.
Furthermore under the same type of restrictions there is no contribution of two or more such
integrated tachyons to the OPE of the interacting fields. Therefore for a product of p tachyons
Ti there are p terms in the OPE of the type in (4.21).
However for values in L there are more solutions already for the case of one integrated
tachyon. In particular its momentum P1 can be given any c < 1 or c > 25 degenerate value. The
OPE coefficients in this case are computed for generic values of and any integers n, m 0:
()
a W T+1
C b W T+1
=
=

n
2 1

(+)
C b W
2 1
s
ns
M L W
s=0
a + W+ T1
=

+ 2sn
+ (m+1)b
2
2b

C (++)
W+ T1
1
2b

(++) +
C 1
W
2b 1
m+1
M
m+1
L
if P1 =
if P1 =
n+1
b
n+1
b
(m + 1)b,
+ (m + 1)b,
m
kM mk
L W
+ (m2k)b
+ n+1
2
2b
k=0
n+1
M
n+1
L
if P1 = n+1
b (m + 1)b,
if P1 = n+1
b (m + 1)b.
289
(4.24)
More generally, one can have a product of any number of arbitrary tachyons with partial sums
of momenta in L, depending on the chiralities. The simplest computable examples with two
integrated tachyons are given by
a W T+1 T+2 = W b ,
+ 2
1 + 2 = b,
1
1 + 2 = ,
(4.25)
b
or P1 + P2 = 2/b, 2b, respectively. These identities were used for 1 = 0 in the derivation of the
last terms in (4.5) and (4.6); for this value they reduce to (4.24). There are also cases in which
the chirality of the tachyon in the OPE is inverted. In our consideration below we shall restrict to
combinations of momenta which allow at most the basic series in (4.21), (4.24).
a+ W+ T1 T2 = W +
1
+ 2b
4.4. Diagonal ground ring

As we mentioned in the introduction, it is possible to construct a discrete model of nonrational 2D quantum gravity in which the order operators, i.e. the degenerate fields labelled by
the diagonal (m = n) of the infinite Kac table, have a simple realization as observables.
It happens that in this theory the 4-point function of order fields contains only order fields
in the intermediate channels. Therefore in the corresponding CFT the order field tachyons must
form a closed algebra under OPE. This is not possible in the matter CFT on a rigid surface, where
the OPE of the diagonal fields generates the whole spectrum of degenerate fields. The question
arises, is it possible, after switching on the Liouville field, that the order fields form a closed
algebra? We shall argue that such a theory exists.
First we notice that the ground ring element O2,2 obtained by combining (4.5) and (4.6), has
four term OPE with the tachyons of given chirality. They involve shifts of the momenta with
Q and e0 . To preserve the diagonal m = n of (4.10) we need rather a projection to the two
terms with shifts by e0 . Indeed, such a projection exists but it requires a different deformation
of the free field ring elements. This new theory is defined by an interaction which contains the
two Liouville screening charges, as well as the two possible Liouville dressings of the non-trivial
(+)
()
+
= Te0 and Tb = Te0 :
vertex operator with zero dimension: T1/b

2
int
2b
2/b
2ie0 2
2b
2
2/b
b L e
+ L e
2 M M e
+ b L e
A =
L e
e0
= L T + + L T + + L T + L T , L = L M M .
(4.26)
b
1/b
1/b
The matter charges in the correlators computed with this action can be screened only by
multiples of e0 = b1 b, whence the name diagonal, by which we refer to it.10 The dual10 In the same way one can consider an interaction theory described by the two matter screening charges and their
Liouville reflected counterparts TQ

dg
Aint =

2i
2 L L 2Q 2
M 2i
2ib +
b
M e2ib + M e b
e
e
e
b
.
M
Q2
b2
(4.27)
290
ity transformations (2.14), or (2.15), with a simultaneous change of sign of one of the two
fields, as discussed above, exchange the Liouville screening charges as well as the two new
terms, so that the action is invariant. On the other hand, the matter-Liouville transformations
{b, L , M , , } {ib, M , L , , } map it to the action (4.27).
The deformation (4.26) leads to an operator a a+ A with the following OPEs
AW+ = L W +
AW = L W
e0
2
e0
2
+ L M M W +
+ L M M W
e0
2
e0
2
,
.
(4.28)
These relations are obtained combining the free field formulae used in the derivation of (4.5),
(4.6) and (4.17); see Appendix A.1. Comparing with the composition of (4.5) and (4.6), the
+
T0 T0 of two matter and one Liouville screening charges, which leads to the shift
product T1/b
(+)
+
+ e0 /2, is now traded for the tachyon T1/b
= Te
. This explains the expression =
0
L M M for the coupling constant in (4.26), (4.28).
Note that we now need all the four terms in the interaction action (4.26) in order to determine
the OPEs of the ring generator with tachyons of both chiralities, in contrast with any of the
relations (4.5), or (4.6). Let us stress that at this stage we will consider the diagonal action (4.26)
as a formal tool, which provides us in a systematic way with certain rules. In particular, we will
not discuss its possible semiclassical limits.
It appears that in this theory the mass-shell condition applied to the potential OPE channels
is much more restrictive. Thus there are no additional terms in the OPE as far as we consider
either generic momenta, or the set of diagonal momenta P = ke0 . Similarly for the momenta
of interest (generic, or diagonal degenerate) there are no more OPE terms in the presence of
integrated tachyons besides (4.17). The operator a+ a , perturbed by the diagonal action (4.26),
generates a sl(2) type ring, as does each of the operators a perturbed by the actions (2.4) and
(2.5). Applying (4.28) to the set of order parameter fields we get a formula analogous to (4.15),
representing the diagonal ring elements as Chebyshev polynomial of the generator A.
5. Functional relations for the closed string tachyon amplitudes

5.1. 3-point solutions of the ring identities
In this section we shall apply the free field computed OPEs of the ring generators assuming
that they hold in a general tachyon correlator.
The general 4-point function with one of the ring generators and arbitrary three tachyons
Wi can be computed in two ways exploiting the operator product expansions (4.5), (4.6). This
leads to the finite difference identities (3.27) for the tachyon 3-point correlators. Similarly in the
diagonal theory (4.28) implies the relation
NP1 +e0 ,P2 ,P3 + NP1 e0 ,P2 ,P3 = NP1 ,P2 +e0 ,P3 + NP1 ,P2 e0 ,P3 .
(5.1)
The simplest solution of (5.1), as that for the identities (3.27), is NP1 ,P2 ,P3 = 1. To fix here the
overall normalization constant we assume that the correlators with zero overall matter charge are
the same in both theories, i.e., they are given by the normalized with the leg factors Liouville
3-point constant (3.5).
Non-trivial solutions exist whenever the momenta take values corresponding to degenerate
Virasoro representations. Let us first consider the correlators in the diagonal theory for tachyons
291
with momenta P = Ze0 . We require that the 3-point function vanishes whenever one of the
momenta is zero, that is, outside of the set of degenerate values. We choose for definiteness the
sign in (4.10), taken for m = n, to coincide with the chirality i . Then the diagonal ring relation
(5.1) for the 3-point function
e0
Q
mi ,
(5.2)
2
2
turns into the standard recurrence relation for the tensor-product decomposition multiplicities of
the irreps of sl(2) of finite dimensions mk :

1 if |m1 m2 | + 1 m3 m1 + m2 1 and m1 + m2 + m3 = odd,
Nm1 ,m2 ,m3 =
0 otherwise.
(5.3)
Any of the two sides in (5.1) is equal to the 4-point multiplicity Nm1 ,m2 ,m3 ,2 , where,

Nm1 ,m2 ,m Nm,m3 ,m4 = Nm1 ,m2 ,m3 ,m4
NP1 ,P2 ,P3 ,P4 =
NP1 ,P2 ,P3 = Nm1 ,m2 ,m3 ,
Pi = i mi e0 i =
m=1

1
(5.4)
min(m1 + m2 , m3 + m4 ) max |m1 m2 |, |m3 m4 | .
2
Similarly for general degenerate momenta (4.10) the identities (3.27) are solved by the product
=
NP1 ,P2 ,P3 = Nm1 ,m2 ,m3 Nn1 ,n2 ,n3 ,
Ps = (ns /b ms b),
(5.5)
assuming the vanishing of the tachyons on the border lines m = 0, or n = 0 of (4.10); see also
[24] for the rational case. The solution is symmetric with respect to matter charge reflections
Ps Ps and thus can be identified as a correlator of the invariant combinations (4.13). These
sl(2)sl(2) decomposition multiplicities are the fusion multiplicities in the quasi-rational matter
theory at generic values of b2 , described by the infinite set of fields of momenta Pm,n . The
same solution of (3.27) is found if the tachyon momenta take the Liouville degenerate values
s Ps = (ns /b + ms b) as in (4.14).
The fusion multiplicities (5.5) coincide, when restricted to the diagonal ms = ns , with those
obtained in the diagonal theory, (5.2) and (5.3). However the 4-point fusion multiplicities

NP1 ,P2 ,P3 ,P4 =
(5.6)
NP1 ,P2 ,Pm,n NPm,n ,P3 ,P4 = Nm1 ,m2 ,m3 ,m4 Nn1 ,n2 ,n3 ,n4
m,n=1
taken for such values differ from their counterparts (5.4) in the diagonal theory.
For other 3-point solutions see also Appendix B.
5.2. Recurrence relations
We next apply the OPE relations inserting a ring generator in the 4-point function

G41 2 3 4 (1 , 2 , 3 , 4 ) = W11 (0)W22 (1)T33 W44 () .
(5.7)
The relation one gets for a approaching the first or the second tachyons, reads
()
()
()
()
C b G4 (, 2 , 3 , 4 ) +
C b
G3 (, 2 , 4 )
2 1
C ()b
2 2
G()
4 (1 , , 3 , 4 ) +
2 1 3
C ()b
2 2 3
G()
3 (1 , , 4 ),
(5.8)
292
where we have omitted for simplicity the explicit dependence of the tachyon correlators G()
and the OPE coefficients C () on each of the chiralities. The OPE relations (4.5) determine the
first terms in both sides of (5.8). On the other hand (4.21), and for special values of the momenta, (4.24), give the explicit expressions for the 4-point OPE coefficients, which determine
()
the inhomogeneous, 3-point contact, terms G3 in (5.8). For generic momenta the ring
relation (5.8) generalizes straightforwardly to a correlator with an arbitrary number p 3 of
integrated tachyons T+i with summations over (p 1)-point contact terms. If however some
partial sums of momenta degenerate, i.e., lie on the lattice L, there are other possible solutions
of the mass-shell conditions, as explained in Appendix A.3, and hence potentially new m-point
contact terms, 3 m p 1. The OPE coefficients would require the computation of higher
p m + 3-point free field functions matrix elements, generalizing the p = 4 = m + 1 case of
Appendix A.4.
The inhomogeneous associativity identities (5.8) can be interpreted as string analogs of the
duality equations for the local 4-point correlators of the c < 1 or c > 25 Virasoro theory. Given
the OPE coefficients C () and a choice of the 3-point terms, the set of these relations determines
the 4-point tachyon correlators. What has also to be added to this set of recursive difference
equations is a choice of some boundary conditions, i.e. particular known values of the 4-point
tachyon correlators.
We shall now specialize the contact terms in (5.8) for two basic classes of 4-point tachyon
correlators. For the 3-point tachyon correlators in (5.8) we shall take the generic solution (3.18).
Let us choose 2 = 3 = 1 = 1 . Then the first series in (4.21) contributes to the r.h.s. of (5.8)
(with W+ T+1 now denoted W+2 T+3 ). As we have mentioned above, if the tachyon W+2 is labelled
by the degenerate momentum (4.22) of matter or Liouville type, the r.h.s. of (4.21) terminates.
Hence only the first n + 1 contact terms have to be taken into account in the r.h.s. of (5.8). Using
the relation (3.25) satisfied by the generic solution (3.18) one obtains (n + 1) times a power of
/ L, and P2 + P3
/ L these are the
the coupling constants L , M . For generic values P1 , P3
only contact terms in (5.8) and we obtain the equation
(1 , 2 , 3 , 4 ) + L M G+++
(1 + b, 2 , 3 , 4 )
G+++
4
4

b
b
b
b
+++
+++
L G4
1 + , 2 + , 3 , 4 M G4
1 + , 2 , 3 , 4
2
2
2
2

n

b
b k
( L M )k G++
=
1 + , 2 + 3 + , 4
3
2
2 b
k=0

b
b
++
= (n + 1)G3
(5.9)
1 + , 2 + 3 , 4 .
2
2
In the last line we have used the relation (3.25) satisfied by the generic solution (3.18); in general
one should keep the r.h.s. of the first equality.
There is another class of correlators in which the OPE relations (4.21) produce finite number
of contact terms in (5.8). These are the correlators with all generic momenta, but restricted by an
overall charge conservation condition. It can be a relation involving the two matter charges
4

i=1

n
2n
Pi = 2e0 2mb + ,
b
b
4
ei e0 = mb
i=1
m, n Z0 ,
(5.10)
293
or it can be c > 25 charge conservation condition

n
2n
i Pi = 2Q + 2mb + ,
b
b
4
i Q = mb
m, n Z0 .
(5.11)
i=1
), k = 0, 1, . . . , n, can be assigned
For a fixed n in (5.10) (or (5.11)) k of the charges T0 (or T1/b
to the OPE in (4.21). The general identity (5.8) takes again the form of (5.9).
Eq. (5.9) is recursive with m, e.g., in the case of the matter type charge conservation condition
(5.10), the M -independent and M -dependent terms in the l.h.s. correspond to m and m 1,
respectively. In the case of degenerate P2 (4.22), terms with the three values m, m 1 appear.
If on the other hand the degenerate field appears as an integrated tachyon, i.e. in our notation
P3 is degenerate as in the r.h.s. of (4.24), then (changing the notation n n3 in (4.24)), this
OPE relation leads to new (n3 + 1) contact terms. This implies, using once again the simple
solution (3.18), that the coefficient in the r.h.s. of (5.9) is modified to (n n3 ). For n3 = n the
contact term in the r.h.s. disappears and the relation (5.9) becomes homogeneous.
In particular in the class of correlators with n = 0 = n3 the r.h.s. of (5.9) simplifies to one or
zero contact terms, which we write as

b
b
+++
G
+
,
G+++
1
2 3 4
L M 4
1
2 3 4
4
2
2

b
b
+++
+++
L G4
1 , 2 + , 3 , 4 M G4
1 , 2 , 3 , 4
2
2

b
1 , 2 + 3 , 4
= G++
3
2

m +1

(m3 + 1)b
m +1
M3 ,b+ m3 b + L 3 , m3 b G++
+
,
.
+
1
2
4
3
3
3
2
2
2
m3 =0
(5.12)
The relation dual to (5.12) reads

1
1
+
1 , 2 , 3 , 4 + L M G4
2 + , 2 , 3 , 4
2b
2b

1
1
+
+
,
G
,
L G+
1 2
3 4
M 4
1 2
3 4
4
2b
2b

1
,
= G+
1
2
3
4
3
2b

n +1
+
(n3 + 1)
n3 +1
3
n
n
, 2 , 4 .
+
1 +
M 3 , 1 + 3 + L 3 , 3 G3
2b
b 2b
2b
G+
4
(5.13)
n3 =0
In these simplified equations only one of the matter and one of the Liouville charges is effectively
contributing, namely the pairs in the action (2.4), or (2.5) respectively. Eqs. (5.12), (5.13) become
homogeneous if, in particular, T3 coincides with one of the four screening charges. For example,
if 3 = 0 (e3 = b), the simplest solution of the homogeneous relation (5.12) is
294
1
(+++)
G4
(1 , 2 , 0, 4 ) = 2
b
ei e0 + b Lb
(Q
i ) b1 (e0
M
i ei )
=
G3 (1 , 2 , 4 ),
M
while for a3 = b (e3 = 0) it is
(5.14)
1

1
1
b (Q i i )
M b (e0 i ei )
i
L
2
b
G(+++) (1 , 2 , b, 4 ) =
=
G(1 , 2 , 4 ).
L
(5.15)
Setting M = 0 in (5.13) (or M = 0 in (5.12)) reduces furthermore (5.12), (5.13), so that

they apply to correlators with momenta satisfying trivial matter condition m = 0 = n in (5.10).
Note that besides the main contact terms, each of these equations with two terms in the l.h.s. still
contains an accidental contact term, e.g., m3 = 0 in (5.12), missed in the old considerations.
Taking into account such terms in the (4 + m)-point generalisations of the reduced two term
equations, derived with only the Liouville interaction included, leads to an alternative derivation
of the four term identities (5.12), (5.13).
In a similar way, one derives generically homogeneous relations fusing the ring generators
with two tachyons of the same chirality. E.g., for s = 1, s = 2, 3, 4

b
b
L G4 1 , 2 , 3 + , 4 + M G4 1 , 2 , 3 , 4
2
2

b
b
= L G4 1 , 2 + , 3 , 4 + M G4 1 , 2 , 3 , 4 ,
(5.16)
2
2
cancelling the difference of the two contact terms. We stress that this and the above discussed
cancellations occur when the simplest constant 3-point solution (3.18) is used; in general we
should keep the full linear combinations of 3-point contact terms.
The functional equations take a more compact form after rescaling
G()
n b
i i
Lb
(Q)
(ee0 )
b
M
()
G
n ,
n

i=1
i , e =
n
ei .
(5.17)
i=1
n do not depend on the constants L , M as is standardly checked

The normalized correlators G
by shifting log2bL , + log2biM . The rescaling by the power of b in (5.17) is equivalent to a change of the leg factor normalization
V V = b V
(5.18)
which removes the chirality-dependent power of b in the 3-point function (3.18). This normalisation does not change the OPE ring identities (4.5), (4.6), but changes the coefficients in front
of the contact terms. With a slight abuse of notation, in what follows we shall write Gn for the
n
corresponding correlators with just the powers of b removed, i.e., the ones differing from G
only by the powers of L and M . For further reference we write the ring relations (5.12) and
(5.13) also in terms of the target space momenta. For the rescaled functions
()
(1 , 2 , 3 , 4 ),
G4 (P1 |P2 , P3 , P4 ) G
i Pi = Q 2i ,
295
the equations take the form

(+)
2(cosh bP1 cosh bP2 )G4 (P1 |P2 , P3 , P4 ) = b(n + 1)NP1 ,P2 +P3 b1 ,P4 ,

1
1
1
()
2 cosh P1 cosh P2 G4 (P1 |P2 , P3 , P4 ) = (m + 1)NP1 ,P2 +P3 +b,P4 .
b
b
b
(5.19)
In the diagonal theory the ring relations read

e0
e
e
L G,+,, 1 + 0 , 2 , 3 , 4 + L b G,+,, 1 0 , 2 , 3 , 4
M
4
4
2
2

e0
e0
e0
b ,+,,
,+,,
L G4
1 , 2 , 3 , 4 L M G4
1 , 2 + , 3 , 4
2
2

G,+,
(1 , 2 + 3 Q
3
2 , 4 ), if = +,
=
(5.20)
,+,
G3
(1 + 3 Q
,
,
if = .
2 4 ),
2
Here we shall normalize the correlators (taking also into account the rescaling (5.18)), as
(+)
G4
b b1 (Q) b1 (ee0 ) (+)
M
G4 ,
e0 L
()
G4
1 b1 (Q) b1 (ee0 ) ()
M
G4 .
be0 L
(5.21)
With this normalization (5.20) becomes

()
()
2(cosh e0 P1 cosh e0 P2 )G4 (P1 |P2 , P3 , P4 ) = e0 G3 (P1 |P2 + P3 , P4 ).
(5.22)
The solutions of the mass-shell condition, restricted to diagonal momenta P3 Ze0 L allow
only P3 = 0 as a possible momentum leading to an accidental contact term. This is the tachyon
TQ/2 of no definite chirality.
In the same way one derives relations with 3 exchanged with 2 or 1 respectively. The
derivation can be repeated also with the third field taken at infinity. The collection of these identities for generic momenta imply a set of symmetry relations for the contact terms in the r.h.s.,
e.g. for = 1,
G3 (P1 , P2 + P3 , P4 ) = G3 (P1 , P2 + P4 , P3 ) = G3 (P1 , P4 + P3 , P2 ).
(5.23)
6. Solutions of the ring relations in the absence of matter (or Liouville) screening charges
In this section we will describe solutions of the ring generated functional equations in the
simplest case of only Liouville or only matter perturbation.
6.1. Solutions with matter charge conservation
In the case of Gaussian matter field (formally M = 0 = M ) the neutrality condition (4.2)
holds. The l.h.s. of the functional relations (5.12), (5.13) reduces to a difference of two terms. For
generic momenta the equations extend [25] straightforwardly to p-point correlators satisfying the
chirality rule, i.e., one of the tachyons has the opposite chirality to the chirality of the
other p 1 ones; these are in fact the only correlators comparable with the microscopic approach.
If
restrict to the resonant correlators, satisfying also the Liouville type conservation condition
we
p
s=1 s = Q, the equations simplify with only the L independent term surviving in the l.h.s.
The r.h.s. is recursively reduced to a 3-point function and the solution is a constant independent
of the momenta. In general the functions are symmetric with respect to p 1 of the charges and
296
since theyhave to reproduce as a special case the resonant amplitudes they depend only on the
p
sum = s=1 s of the Liouville charges. Taking into account the new normalization in (5.18)
the equations read,

()
()
()
G
p () L Gp + b = (p 3)b Gp1 (),
p
(6.1)
s .
s=1
Starting with the 4-point case, the solution involves an arbitrary solution of the homogeneous
equation, i.e., a periodic in b or 1/b function. We shall use as a boundary condition the known
expressions (5.15) and its dual, in which one of the Liouville charges is b or 1/b. This leaves us
with a linear function of

Q b

1
1
(2Q+ i i Pi )
i Pi .
b
= L2b
2
1
G() (1 , 2 , 3 , 4 ) = Lb
(Q)
(6.2)
The two choices of boundary conditions are correlated with the two effective actions (2.4) and
(2.5). The p 4 relation (6.1) is solved by
1
(Q)+p3
p3 b
G(+)
L
,
p () = (bL )

p3
1
b(Q)+p3
G()
,
L
p () = L
b
(6.3)
recovering formula (2.53) of [34]. The solutions of the two equations are interchanged by the
duality transformations (2.16) and (2.17), up to a sign in the second case; it disappears for the
properly normalized correlators as in (3.22).

The formulae (6.2) are valid also for the 4-point functions with i i = 4, which are constants,because of the matter charge conservation condition. These constants and the solutions
with i i = 2 in (6.2) are related with an inhomogeneous analog of the Liouville reflection
relation, in contrast with what we had for the 3-point functions in (3.19). For example,
G++++ (1 , 2 , 3 , 4 )
1
= Lb
(Q21 )
G+++ (Q 1 , 2 , 3 , 4 ) + (Q 21 )Lb
4
(Q
i=1 i )
(6.4)
The second term in the last equality compensates the contact term in the ring relation (5.12) for
the 4-point function of type (+++) in the r.h.s. of (6.4), so that the l.h.s. satisfies a homogeneous equation without contact terms, as it should. So far we have excluded from the discussion
the correlators with two equal chiralities, i.e., of the type (++). The reason is that the contact terms depend on the choice for the integrated tachyon and one obtains an inconsistent set of
relations. These correlators have been neglected in the earlier considerations, e.g. [34] and [9],
basically because of the vanishing of the unnormalized perturbative expressions, as discussed
above in the comment after (3.6). Furthermore the correlator of type (+) is also trivial
constant when determined by the action (2.4), and similarly for the correlator of type (+++)
computed with (2.5), since the corresponding functional equations are homogeneous.
The chirality rule satisfying solutions of the two type equations are related by pairs of
inhomogeneous Liouville reflections
297
G +++ (P1 , P2 , P3 , P4 ) = P1 + b = G + (P1 , P2 , P3 , P4 ) + Q

= G +++ (P1 , P2 , P3 , P4 ) + (P1 P2 )
= G + (P1 , P2 , P3 , P4 ) + (P1 + P4 e0 )
= G + (P1 , P2 , P3 , P4 ) + 2P1 e0 .
(6.5)
The matter reflections do not

make sense since they violate the charge conservation condition
(4.2). Using this condition i Pi = 2e0 , we can rewrite (6.2) as

4
1

(=1 )
2b (2Q+ i=1 i Pi )
(P1 |P2 , P3 , P4 ) = L
(e0 P1 Ps )
Q 1
G
(6.6)
s=1
i.e., in the form of (1.2), with NP1 ,P2 ,P3 = 1, if i Pi = e0 and NP1 ,P2 ,P3 = 0 otherwise.
If we restrict the momenta to the range s (Ps e20 ) > 0, s = 1 (physical for e0 > 0 and
implying 1 (P1 e20 ) > 0 as well), the correlator (6.6) reproduces the three channel expansion
formula of [34],
1
4
(
|Pi |2Q)
1 , P2 , P3 , P4 ),
G(P1 , P2 , P3 , P4 ) = L2b i=1
G(P

1 , P2 , P3 , P4 ) = 1 Q |P1 + P2 e0 | |P1 + P3 e0 | |P1 + P4 e0 | .
G(P
(6.7)
2
This formula holds irrespectively of which of the four momenta is chosen with opposite sign
since, unlike (6.6), it is symmetric in them, but at the price that it is not analytic. Vice versa, for
any choice of the signs of the combinations
{Pst := Ps + Pt e0 }s,t=1,...,4

compatible with
the conservation condition i Pi = 2e0 , the formula (6.7) recovers one of the
correlators with i i = 2. In other words (6.7) is a symmetrization over the chiralities,

1 , P2 , P3 , P4 ) =
G(P
(6.8)
t (Pt + Ps e0 ) G (t ) (P1 , P2 , P3 , P4 )

t s=t

1 if x 0
. Here t = t t = and t in G (t ) indicates as before the
0 if x < 0
chirality opposite to the remaining three.11
The permutation symmetry with respect to the four matter charges e1 , e2 , e3 , e4 is an analog
of the locality of the 4-point euclidean correlation functions, so we shall refer to formulae of
this type as local or physical. The other symmetric combination with a plus relative sign
corresponds to generically unphysical momenta i (Pi e0 ) < 0. The local correlators do not
depend on the chiralities and so they are invariant under Liouville reflections. Formula (6.7) is
reproduced in the discrete model framework. For one of the momenta coinciding with e0 (6.7)
reduces to a derivative of the physical 3-point function
where (x) =
G(P1 , P2 , P3 , e0 ) = bL G3 (P1 , P2 , P3 ) = bL L2b
(Q+
|Ps |)
(6.9)
11 We could restrict to the subdomain in which the signs of P e /2, instead of the chiralities, satisfy the chirality
i
0

e
rule. Then replacing the step function factor in the symmetrization formula (6.8) with t (t (Pt 20 )) makes the
correspondence of the two types of correlators local, i.e., depending only on the individual momenta. However the
e
subdomain is not preserved by the shifts, combinations with two positive and two negative signs of {Pt 20 } may
appear.
298
if e0 > 0 and analogously a derivative with respect to L for e0 < 0. On the level of 3-point
functions the physical tachyons are identified by the fixed chirality fields WP+ for P > 0 or WP
for P < 0. The functional identities rewritten for the correlators of these physical representatives
contain in general P -dependent powers of L , coming from a Liouville reflection as in (3.19),
whenever unphysical value P < 0 is reached. As it is clear from (6.8) this local representation
of the physical fields is not possible on the level of the 4-point function. Note that there are
other symmetric combinations
by the eight solutions of the equations. They
locally reproduced

are obtained by replacing i i Pi with i |Pi | in (6.2): one gets two combinations which are
interchanged under the transformation (2.16). What distinguishes the correlator (6.7) is that it
preserves the simple fusion rule of the underlying local matter theory in each of the s, t, u
channels so that the notion of locality of the 4-point tachyon correlation numbers matches
that of standard locality. Furthermore with the chosen normalisation (5.18) this correlator is selfdual with respect to the Liouville type transformation (2.16); in the initial normalization the
two analogs of (6.7) differ only by an overall power of b2 . If we further normalize with the
partition functions (3.21) we can define two correlators, depending only on {b2 , bPi }, which are
exchanged by (3.23)
(n)
G4 (P1 , P2 , P3 , P4 ; L , M , b)

e0 Q 1 ( 4 |Pi |4Q) eb0 Q 1

M
= 2 L2b i=1
P
1s

2b
2b
2b
s=1

1
(n)
= G 4 P1 , P2 , P3 , P4 ; L , M ,
b

1
= (n) G 4 P1 , P2 , P3 , P4 ; L , M , .
b
(6.10)
The difference identities for the correlators with definite chiralities like (6.6) do not preserve
the physical regions, neither the region determined by the set of inequalities above. Accordingly (6.7) does not satisfy globally the equations which apply by definition only to the partially
symmetric, fixed chirality correlators. One can compute directly the shift relations for the local
correlators from the explicit expression (6.7), or derive them from the initial identities.
It is instructive to compare the two types of equations. If the shift crosses the boundary of the
momenta region in which a given fixed chirality correlator represents the local one, the shifted
correlator can be replaced via pairs of Liouville reflections (6.5) by the proper local representative
in the new region. In this replacement there appear linear in the momenta terms which can be
moved to the r.h.s. and interpreted as a modification of the contact terms. By the same mechanism
any of the homogeneous relations acquires a non-trivial r.h.s., if the shift crosses the boundary
of the corresponding region. Since the coefficients in the linear relation (6.8) project to different
regions of momenta, this effectively implies that the (4-point) OPE coefficients in the analog
of (5.8) for the local correlators will be no more constants but will depend themselves on the
momenta. The rule which is extracted from the explicit expression (6.7) is that whenever the
boundary is crossed, a Liouville reflected tachyon in the OPE appears, dressed with the inverse
propagator (3.20). Namely (taking as usual b > 0) we compute from (6.7)
G4 (P1 , P2 , P3 , P4 ) + G4 (P1 + b, P2 b, P3 , P4 )

b
=
(P2s ) (P2s ) + (P2s + b) (P2s )(P2s b) .
2
s=3,4
(6.11)
299
We shall take P2s > 0, s = 3, 4, P34 > 0 so that the first correlator in the l.h.s. of (6.11) is
identified with G +++ . Then the r.h.s. of (6.11) reduces to

(P2s b) (P2s b)
b+
s=3,4
=b
(P2s + b)G2 (P23 b, P23 + b)1 .
(6.12)
s=3,4
If both shifted momenta change sign, P2s b < 0, s = 3, 4 the shifted correlator in the l.h.s.
is identified with G +++ . If only P23 b < 0, while P24 b > 0 (i.e., P13 + b < 0), this
shifted correlator is of type G + . Irrespectively of the signs of P2s b the first term in
(6.12) corresponds to the standard constant contact term. For negative P23 b the physical
tachyon in the 3-point function G3 (P1 + b, P23 b, P4 ) has to be identified with the Liouville
reflection of the tachyon W+2 +3 b/2 = WP+23 b . However the new contact term cannot be identified simply with the product of this 3-point correlator and the Liouville reflected OPE constant
C ++
b
2 2 3
Q2 3 + b2
P23 b
b
= b2 L
(see (A.32) below); rather it is related to the derivative of this
constant with respect to L .

We stress that (6.11) is just an alternative rewriting of the initial shift relation as a relation
for the local correlators; otherwise the new terms P23 + P24 2b = (P2 b) (P1 + b) (or
P23 b = (P1 + b) + P4 e0 ) in (6.12) are precisely the inhomogeneous terms of the Liouville
reflections in (6.5), needed to represent the shifted correlator in the l.h.s. by a function of type
G +++ (or of type G + ), respectively.
There are several remarks in order concerning the identity (6.11), (6.12):
(i) The appearance of P24 -dependent terms, besides P23 , serves as a symmetrization as in the
simple relation (5.23) (in which the OPE coefficients are set to 1). These terms correspond to
correlators in which the fourth tachyon is represented by an integral T4 ; the shift equation for
the symmetrised correlator does not distinguish the two situations. Or, alternatively, the relation
(6.11) represents the splitting of the local 5-point function with a ring generator into various
products of 3-times 4-point, times the inverse of a 2-point, correlators. Effectively the shift equation rewritten for the symmetrized correlator manifests the short distance expansion around all
the three points 0, 1, .12
(ii) On the other hand, once extracted from (6.12), these modified contact terms can be used
to extend the r.h.s. of the general 4-term ring relation (5.9) for n = 0. Namely for Pst > 0,
s, t = 2, 3, 4 (ensured by the physical values s Ps > e0 /2) we have
+++ (P1 , P2 , P3 , P4 )
1 + b, P2 b, P3 , P4 ) G
G(P
+++ (P1 + 2b, P2 , P3 , P4 )
+++ (P1 + b, P2 + b, P3 , P4 ) G
+G

=b+
(P2s + b)(P2s b),
P2s > 0, s = 3, 4, P34 > 0.
(6.13)
s=3,4
This extended relation can be taken as a definition of the local correlator in a range of momenta,
larger than the physical range in which it is represented by the fixed chirality correlators. Com12 This is analogous to the general discussion in [19] (see also [2022]), where functional relations for the tachyon
correlators in the c = 1 theory without interactions are derived starting from Ward identities of non-scalar currents. The
resonant amplitudes described in these works are too simple to actually make a distinction between the two types of
equations but an extension of the method might be appropriate for the problem under consideration.
300
bining (6.13) with the initial fixed chirality relation (5.9) we obtain alternatively, relabeling the
momenta,

+++ (P1 , P2 , P3 , P4 ) +
1 , P2 , P3 , P4 ) = G
(P2s )P2s ,
G(P
s=3,4
for P2s + b > 0, s = 3, 4, P34 > 0.
(6.14)
The momenta of the three fixed chirality correlators in (6.13) are in the range, in which these
correlators coincide with the local correlators so that (6.13) can be also interpreted as a shift relation for local correlators. Similarly one can derive shift relations and their duals in other regions
of momenta, neighbouring the physical range. In particular, the dual of (6.13), extending the
second equation in (5.19) for the case m = 0, implies

+ (P1 , P2 , P3 , P4 )
1 , P2 , P3 , P4 ) = G
(P2s )P2s ,
G(P
s=3,4
1
for P2s < 0, s = 3, 4, P34 < 0.
b
We shall exploit all these relations in Section 7.
(6.15)
6.2. Distribution type solution of the two term ring relations

We can also interpret the solution of the ring relation in distribution sense, accounting for the
charge conservation condition (4.2) by a -function. The correlators are expressed in terms of the
p-point multiplicities for Gaussian matter
NP1 ,P2 ,...,Pp = Np (P1 + P2 + + Pp ),

Np (P ) = P (p 2)e0 , p 3.
(6.16)
(6.17)
We interpret the 3-point multiplicity NP1 ,P2 ,P3 as the factor modifying the generic 3-point constant (3.18), i.e. as the matter part of the 3-point correlator instead of (3.9). It satisfies the second
relation in (3.25) and its dual, which are the 3-point ring relations in the absence of matter screening charges. Now (6.6) is replaced by the integral representation

1
G () (P1 |P2 , P3 , P4 ) = NP1 ,P2 ,P3 ,P4 Q P
P1 ;P2 ,P3 ,P4 ,
2
where

permutations
dP NP1 ,P2 ,P P NP ,P3 ,P4 +
.
P
P1 ;P2 ,P3 ,P4 :=
P2 P3 , P4
(6.18)
(6.19)
The analog of the local 4-point function (6.7) is obtained replacing P |P | = |Q 2| in the
three channels in (6.19).
The ring relation (5.9) remains a relation with two terms in the l.h.s. for the correlators satisfying (5.10) with fixed m = 0 and non-trivial n = 0. A second boundary condition is provided by
the correlator with 1 = 0, e1 = 1/b, i.e., the negative chirality field is given by the dual matter
charge T0

1
1
1
(Q) n
(+)
G4 (0, 2 , 3 , 4 )|e1 =1/b = M G3 (2 , 3 , 4 ) = (n + 1)Lb
M (6.20)
b
b
301
and similarly
1
b
G()
4 (0, 2 , 3 , 4 )|e1 =1/b = (bM )G3 (2 , 3 , 4 ) = (m + 1)bL
(Q) m
M .
The solutions (to be discussed in more detail below in Section 7) generalize (6.2)

1
n
(+)
b (Q) n
G4 (1 , 2 , 3 , 4 ) = L
,
M (n + 1) Q b +
b

1
1
()
b (Q) m
G4 (1 , 2 , 3 , 4 ) = L
M (m + 1) Q + mb .
b
(6.21)
(6.22)
6.3. Solution with Liouville charge conservation

One can obtain similarly
the solutions of
ring relations in the absence of Liouville screenthe
p
p
ing charges, so that s=1 s Q = 0, or s=1 s Ps = (p 2)Q. Such a constraint again goes
beyond the normalization assumptions which led us to (3.18) since (3.2) is singular, i.e., it rather
corresponds to the Coulomb gas constant C Liou obtained as a residuum of (3.2). Nevertheless the
final expression in (3.18) satisfies the (L = 0) equations for the tachyon 3-point functions and
can be taken as the solution in this case. Then the analog of (6.3) is given by a derivative with
respect to the matter constant M

p
1 e0
p3
G(+)
M b
p (1 , 2 , . . . , p ) = (bM )
i=1 ei
+p3
(6.23)
and a similar formula for the opposite chiralities. The normalized with the partition functions
L , M , b) solution (6.23) and its
ZM (L , M , b) = Z(L , M , b) and Z M (L , M , b) = Z(
dual are related to the normalized correlators (6.3), by the matter-Liouville duality (3.24), now
equivalent to {b2 , L , b, be} {b2 , M , be, b}. The analog of the formula (6.6) reads

4

Q
e
0
()
(1 , 2 , 3 , 4 ) = e0 e b =
.
G
(6.24)
1 + s
4
2
2
s=2
One can introduce also different analogs of the local correlator (6.7), now symmetric with
respect to the four Liouville momenta 1 , 2 , 3 , 4 . This correlator can be used to define another
local extension of the general ring relations, analogously to (6.13). This case is however more
speculative since we lack a selection rule of the type of Seiberg inequality and moreover we have
no independent information on the generic c < 1 n-point correlators.
7. The 4-point function for fixed number of screening charges
7.1. The fixed chirality solutions
In this section we analyse the difference equations (5.19) in the case when the total sum of
momenta is restricted by integer numbers of matter screening charges as in (5.10). Reducing
recursively with m to the two term identities for m = 0 discussed above one obtains

b1 ()
G()
4 (; m, n) L G4 + b ; m, n

n
m

l1
l1
k
l
k ()
M (L M ) G3 1 +
b, 2 + 3
b + , 4 ; m l, k
=
2
2
b
l=0 k=0
302
1
(Q) m n
= b (m + 1)(n + 1)Lb
M M ,
(7.1)
4
where = s=1 s is as before the total Liouville charge. The recursive equations for the 3-point
functions are similarly reduced to the second identity in (3.25) and its dual. In the last line of (7.1)
we have inserted again the simplest solution (3.18) of these identities.
Eq. (7.1) admits a solution generalizing (6.22), and which can be cast into a three channel
expansion form, consistent with a sl(2) sl(2) type decomposition rule:

n1+
1
()
+ mb
G (; m, n) = (m + 1)(n + 1) Q b +
b 2
2

1
n
= (m + 1)(n + 1) Q + mb +
2
b

m
n
2l

.
P1 + Ps e0 + 2kb
+
(7.2)
2
b
s=1 k=0 l=0
Let us take for definiteness = +1. The solution (7.2) reduces to (6.2) for m = 0 = n, and to
(6.22) for m = 0. For n = 0 the correlator is compared with another
solution of (5.12), namely

the solution (6.23), (6.24) with Liouville charge conservation i i Q = Q = 0, i.e., it is
consistent with a third boundary condition given by the matter charge T0+ ,13
1
(ee0 )
b
G(+) (; m, 0)|=Q = b(m + 1)m
M = (e0 e b)M
(7.3)
The boundary conditions described do not fix uniquely the arbitrariness in the solution of the
homogeneous difference equationa term of the type mn(m + 1)(n + 1)P (m, n) with an arbitrary polynomial P (m, n) is still allowed. As we shall see below, (7.2) is the solution smoothly
related to another class of solutions of (7.1), the ones with one degenerate field, which are constructed recursively starting from a boundary value.
Let us summarize. We have imposed three boundary conditions corresponding to derivatives
of 3-point correlators with respect to L , M , M for = 1 (or L , M , M for = 1).
These are the coupling constants in a three-term interaction which includes one of the Liouville
and both matter screening charges. The doubled matter interaction contributes perturbatively,
i.e., with integer powers of the screening charges.
The duality transformation (2.16) exchanges the two solutions (7.2) = 1. Pairs of inhomogeneous Liouville reflections interchange the solutions of different chiralities, generalising the
relations (6.5).
We note that there is a special case involving a non-integer number of screening charges.
For n = 0, = 1 or for m = 0, = 1, the fixed chirality solutions (7.2) of the functional
equations (5.12) and (5.13) can be written in a form which allows to extend them to arbitrary
(non-integer) values of m or n respectively,
(+)
G4
= Lb
(Q)
(ee0 ) 1
b
M
b
1
= bL M Lb
(e e0 + b)( Q b)
(Q)+1
(ee0 )+1
b
M
13 The two simultaneous restrictions on the matter and Liouville charges lead to a value P L, which implies a new
1
accidental contact term in (4.5), see Appendix A.3. The solution here is consistent with (4.5) becoming a homogeneous
relation.
()
G4

1
1
b(Q)
b(e0 e)
= L
b e0 e +
Q
M
b
b
1
b(Q)+1
b(e0 e)+1
= L M L
.
M
b
303
(7.4)
7.2. Correlators satisfying the locality requirement

Let us now look for a local 4-point function, symmetric in the four matter charges, which
reproduces for certain range of their values the fixed chirality correlators (7.2). Now the symmetry constraint has many solutions. We observe that the representation in the two last lines of
(7.2) takes the form of (1.2) with fusion multiplicities
determined by the charge conservation

condition, i.e. N(P1 , P2 , P3 ) = 1 if Pi satisfy i Pi = 2e0 2kb + 2 bl with some integers k, l
between 0 and m or n, and N (P1 , P2 , P3 ) = 0 otherwise. These are the fusion rules of the underlying local matter correlators of this type and it is natural to solve the symmetry requirement,
i.e., to determine the correlator of local tachyons so that to preserve these fusion rules. Of the
two possible such combinations we choose the one which reduces to (6.7) for m = 0 = n,
1 , P2 , P3 , P4 ; m, n)
G(P

4 m n
2l
n
1
1
P
+
P
e
+
2kb
= (m + 1)(n + 1) Q + mb +
s
0
1
2
b
2
b
s=2 k=0 l=0

n
1
= (m + 1)(n + 1) Q + mb +
2
b

m
4
n

e0 mb
n
rb
t

e1 + es 2 2 + 2b 2 2b .
r=m t=n
s=2
(7.5)
mod 2 mod 2
The symmetry under permutations of the momenta is ensured by the charge conservation condition (5.10). The duality properties of the properly normalised correlators (7.5) are analogous to
those in (6.10); to ensure that the transformation (2.16) and (2.17) become identical, we should
include a power of b under the modulus in (7.5). The shift equations satisfied by the local correlators are derived from the explicit expression (7.5).
We shall now give another argument in support of the formula (7.5).
Clearly unlike the simplest example (6.7) discussed in Section 4, we now lack a complete atlas of fixed chirality solutions to match locally (7.5) in all regions of the momenta.
Consider the case n = 0. In the physical region Pst = Ps + Pt e0 > 0, s, t = 2, 3, 4 (or
equivalently P1i + 2mb < 0, i = 1, 2, 3) the local correlator is represented by the solution
G+++ (P1 , P2 , P3 , P4 ; m, 0) in (7.2). On the other hand we can use the extended identities as
(6.14) to find a representation of the local correlator in the vicinity of any region described by
the eight fixed chirality correlators. The identities (6.13), (6.14) imply that in the extended range
of momenta the shift relation (7.1) is replaced by a relation for the local correlators, namely
1 + b, P2 b, P3 , P4 ; m, 0)
1 , P2 , P3 , P4 ; m, 0) + G(P
G(P

= 2b +
(P2s + b)(P2s b), for Pst > 0, s, t = 2, 3, 4.
(7.6)
s=3,4
The proposed correlators (7.5) do indeed satisfy (6.14) and the shift Eq. (7.6), as well as all other
similar identities. In fact these equations determine completely the local correlators for m = 1,
304
n = 0 (or the correlators for m = 0, n = 1), taking also into account the inhomogeneous Liouville
reflection relations generalizing (6.5). The solution is identical to the one prescribed by (7.5).
Then in the next step we can use this solution, as we did in the case m = 0 = n, in order to extend
further the general ring relations as identities for the local correlators, generalising (6.13). This
in particular determines the local correlator in the next to the nearest range, extending (6.14),
1 , P2 , P3 , P4 ; m, 0)
G(P
+++ (P1 , P2 , P3 , P4 ; m, 0)
=G

(P2s )P2s + (P2s 2b)(P2s + 2b) ,
+
s=3,4
for P2s + 3b > 0, s = 3, 4, P34 > 0.
(7.7)
In principle one can reproduce in this way recursively the correlators (7.5) for the two thermal cases n = 0, or m = 0. Furthermore we can combine the two types of shift relations.
1 , P2 , P3 , P4 ; 1, 0), which is represented by
Thus starting again from the local correlator G(P
+
(P1 , P2 , P3 , P4 ; 1, 0) in the range {Pst + 2b < 0, s = 2, 3, 4}, we can compute the r.h.s.
G
of the second identity in (5.19) and use the new contact terms to extend this identity for arbitrary
n and the fixed m = 1,

1
1
+ (P1 , P2 , P3 , P4 ; 1, n)
G P1 , P2 + , P3 , P4 ; 1, n G
b
b

1
1
+
P1 , P2 , P3 , P4 ; 1, n 1
+G
b
b

2
+
G
P1 , P2 , P3 , P4 ; 1, n 1
b

2
1
1
1
1
=
P2s +
+ P2s + 2b +
P2s + 2b +
,
P2s +
b
b
b
b
b
s=3,4
Pst + 2b < 0,
s = 2, 3, 4;
or,
1 , P2 , P3 , P4 ; 1, n)
G(P

+ (P1 , P2 , P3 , P4 ; 1, n) +
=G
(P2s )P2s + (P2s + 2b)(P2s + 2b) ,
s=3,4
1
for P2s + 2b < 0, s = 3, 4, P34 + 2b < 0,
(7.8)
b
etc., confirming (7.5).
One finds also symmetric with respect to the Liouville labels i tachyon correlatorsthey
preserve the fusion rules of the Coulomb gas c > 25 theory.
7.3. Distribution type solutions
Furthermore a distribution type solution generalizing (6.18) is obtained by multiplying (7.2)
with (P 2e0 +2mb 2n/b) and summing over non-negative m, n. The n-point multiplicities
are again distributions, depending only on the total momentum P , but instead of (6.17) they are
given by semi-infinite double sums of -functions. They are expressed in terms of the 3-point
305
multiplicity
N3 (P ) =
(P e0 + 2mb 2n/b).
(7.9)
m,n=0
It satisfies the two relations (3.27), while (3.26) is replaced by the difference identities
NP1 b,P2 ,P3 NP1 +b,P2 ,P3 =
n=0
3

Pi (2n + 1)/b ,
i=1
NP1 +1/b,P2 ,P3 NP1 1/b,P2 ,P3 =
m=0
3

Pi + (2m + 1)b .
(7.10)
i=1
We define the quantity P

as in (6.19), but with the new 3-point multiplicity (7.10). Using the
properties of (7.9) one reproduces the functional relation
P
P1 ;P2 ,P3 ,P4 + P
P1 +2b;P2 ,P3 ,P4 P
P1 +b;P2 b,P3 ,P4 ) P
P1 +b;P2 +b,P3 ,P4 )

= 2b
(n + 1)
Pi 2e0 + 2mb 2n/b .
(7.11)
m,n=0
The identity is equivalent to (5.9), when projected to a fixed sum of momenta, since the irreducible part of the 4-point function satisfies the homogeneous equation. A local correlator with
P
replaced by |P |
is also obtained.
Now let us turn to the diagonal theory defined by the action (4.26). We shall look for solutions for the 4-point function assuming a diagonal (m = n) charge conservation condition
(5.10). This leads to a single sum of functions representing the 3-point multiplicity
NP1 ,P2 ,P3 N3 (P ) =

P (2k + 1)e0 .
(7.12)
k=0
The 4-point multiplicity is accordingly

NP1 ,P2 ,P3 ,P4
dP NP1 ,P2 ,P NP ,P3 ,P4

(m + 1)
m=0
4

Pi 2(m + 1)e0 .
(7.13)
i=1
Instead of (7.11) one obtains

P
P1 + e0 ;P2 ,P3 ,P4
=1
P
P1 ;P2 + e0 ,P3 ,P4
= 2e0 N3 (P1 + P2 + P3 + P4 ).
(7.14)
306
The r.h.s. of (6.18), now with the multiplicities defined in (7.12), (7.13), provides a solution of
(5.22),14
()
G4 (P1 |P2 , P3 , P4 )

1
= QNP1 ,P2 ,P3 ,P4 P
P1 ;P2 ,P3 ,P4
2
4

m
4

1
P1 + Ps (2k + 1)e0
Pi 2(m + 1)e0
=
Q(m + 1) +
2
m=0
s=2 k=0
i=1

4

e0
(m + 1)
i Q b + m
Pi 2(m + 1)e0 .
=
(7.15)
2
m=0
i=1
Note that in contrast with (7.15), in the non-diagonal theory the solution cannot be expressed
entirely in terms of the 3-point multiplicities and the inverse 2-point correlator due to the more
complicated form of the 1pi part.
From (7.15) one extrapolates the symmetric correlator
1 , P2 , P3 , P4 ; m)
G(P
=
m
4

Q
1
P1 + Ps (m + 1)e0 re0 .
(m + 1)
2
2
r=m
s=2
(7.16)
mod 2
In this case there is no underlying local matter theory to compare with, rather we preserve the
fusion rules (7.12). A formula of this type is reproduced in the microscopic approach in [29],
with the delta-functions replaced by periodic deltas.
The multiplicities introduced in this section are considered for real momenta only, but they
can be expressed in terms of meromorphic functions defined in the whole complex plane. Thus
the 3-point multiplicity (7.10) is given by the discontinuity on the real axis of a meromorphic
function,

1
e0 P
e0 P
N3 (P ) =
(7.17)
f
+ i0 f
i0 ,
4i
2
2
namely the logarithmic derivative of the double -function

f (z) z log b (z + b) = dt
0
z
ezt
bt
t/b
(1 e )(1 e
)
t
e0
2 t

1
+ 2 .
t
(7.18)
The diagonal multiplicity (7.12) is expressed as the discontinuity of (z) = z log (z).
8. Degenerate fields in the diagonal theory
The most interesting correlation functions, especially from the point of view of comparing
with the microscopic theory, are those involving four degenerate fields. In this section we solve
14 The arbitrariness in the diagonal case is fixed comparing the first term, m = 0, with (6.18), and furthermore, with the
solutions with one degenerate field, to be discussed in the next section.
307
the difference equations for the spectrum of momenta corresponding to the degenerate matter
fields (order operators) in the diagonal theory. This spectrum is given by the diagonal e0 Z of the
grid (4.4), with the point P = 0 excluded. We will assume that there is no tachyon with P = 0,
i.e., the tachyon correlation functions vanish if one of the momenta is zero.
To begin with, we will find the solutions of the diagonal ring relations with one degenerate
field P2 = (m + 1)e0 , 2 = b me0 /2, and three generic. As initial condition we take

(+)
s Q =
s Q ,
G4 (P1 , e0 , P3 , P4 ) = Nm1 ,m3 ,m4
(8.1)
s=2
s=2
where the generic solution with NP1 ,P2 ,P3 = 1 for the 3-point correlator is inserted. We solve
(5.22) recursively, under the assumption that at the point P2 = 0 the correlator vanishes. This
is achieved automatically if the fields are interpreted as linear combinations of vertex operators
anti-symmetric under a composition of matter and Liouville reflections:
(,A)
VP
(,A)
= VP Lb M b VP
= Lb M b VP .
(8.2)
Since the degenerate field is assumed anti-symmetric, the contact terms cancel, as the generic
solution (3.18) satisfies the reflection identity (3.19). We get

e0
(+)
G4 (1 , 2 , 3 , 4 ) = (m + 1) Q b + m
2
m
1
Q
(s Ps re0 ).
= (m + 1)
(8.3)
2
2
r=m
s=2
mod 2
This expression has the form (1.2), with trivial multiplicities NP ,Ps ,Pr = 1 for s, r = 2 and a nontrivial multiplicity NPt ,P2 =(m+1)e0 ,P , representing a continuation of the sl(2) decomposition rule
(5.3) to non-integer isospins; the shifts by re0 in (8.3) correspond to the weight diagram of the
irrep of dimension m + 1. The solution (8.3) also justifies the choice of the linear combination in
(7.15). The meaning of the nonnegative integer m in the two types of solutions is different, but
in both cases m + 1 counts the number of intermediate contributions in each channel. Projecting
(7.15) to a fixed charge m and inserting the value P2 = (m + 1)e0 reproduces (8.3).
Now let us consider correlators in which all tachyons correspond to degenerate fields, Pi =
m i e0
i mi e0 , mi N, i.e., i = Q
2 2 . These tachyons satisfy fusion rules given by the sl(2)
decomposition multiplicity (5.3), which is also expressed by an integral, in general
Nm1 ,...,mp
1
=
2
d sin2
0
p

sin(mi )
i=1
sin
(8.4)
i )
in terms of the characters mi ( ) = sin(m
sin . These multiplicities preserve the homogeneous identity, implied by (5.1), with respect to any pair of variables:
Nm1 +1,m2 ,...,mp + Nm1 1,...,mp = Nm1 ,m2 +1,...,mp + Nm1 ,m2 1,...,mp .
(8.5)
They are symmetric under permutations and extend to arbitrary integer values of the weights mi
by the (shifted) Weyl reflection property
Nm1 ,m2 ,...,mp = Nm1 ,m2 ,...,mp ,
so that they vanish if some mi = 0 = Pi .
308
We start with a simple example in which P2 = 2e0 , illustrating the recursive determination
of the 4-point correlators. As an initial condition we take again the correlator in the first
equality in (8.1), but instead of the generic solution Nm1 ,m3 ,m4 = 1 we take the sl(2) 3-point
multiplicity (5.2). Up to the contact terms, which we will neglect at this stage, Eq. (5.22) gives
(+)
4 (1 , b 1 e0 , 3 , 4 )
for G4 (P1 |2e0 , P3 , P4 ) = G
2

1
1
Nm1 +,m3 ,m4 Q b + e0

+
G4 1 , b e0 , 3 , 4 =
2
2
=1
= Nm1 ,2,m3 ,m4 ( Q b) + (e0 Nm1 1,m3 ,m4 + ).
(8.6)
To obtain the first term in the second line we have used the homogeneous relation (8.5) for
the 4-point multiplicity, applied for m2 = 1 and using that Nm1 ,m3 ,m4 = Nm1 ,1,m3 ,m4 . The result
should be symmetric with respect to m1 , m3 , m4 , so instead of the incomplete second term in
the last line of (8.6) we should have a symmetric expression, which vanishes if some ms = 1,
recovering (8.1). A solution to these conditions is given by

4 1 , b 1 e0 , 3 , 4 = Nm1 ,2,m3 ,m4 Q b + e0 (Nm1 ,2,m3 ,m4 1) .
G
(8.7)
2
2
The normalization of the added term is fixed to + e20 , since generically Nm1 ,2,m3 ,m4 = 2 and this
is in agreement with our previous solution (8.3) taken for m + 1 = m2 = 2. In the next step
of the recursion we take P2 = 3e0 and use the result in (8.7). Once again we recover the first
term Nm1 ,3,m3 ,m4 ( Q b) uniquely, while we get an expression for the second term which is
not symmetric, and generically should be equal to 2e0 = e0 (Nm1 ,3,m3 ,m4 1), if compared with
(8.3). The end result is a formula in which Nm1 ,2,m3 ,m4 in (8.7) is replaced by Nm1 ,m2 ,m3 ,m4 . This
formula can be cast in the form
(+) (m1 , m2 , m3 , m4 )
G
4

1
Nm1 ,m2 ,m (me0 )Nm,m3 ,m4 + permutations
QNm1 ,m2 ,m3 ,m4
=
2
m=1

e0
e0
Q
= Nm1 ,m2 ,m3 ,m4
i Q b + (Nm1 ,m2 ,m3 ,m4 1) , i = mi . (8.8)
2
2
2
i
To connect the two expressions in the second and the third lines we have used the relation

(Nm1 ,m2 ,m mNm,m3 ,m4 + permutations)
m=0
= Nm1 ,m2 ,m3 ,m4
4

mi Nm1 ,m2 ,m3 ,m4 .
(8.9)
i=1
This identity has a purely group theoretical formulation being expressed in terms of the sl(2) tensor product decomposition multiplicities (5.3) and the dimensions mi of the irreps. It is derived
using the definition (5.3). We stress that by construction the diagonal degenerate fields satisfy
closed fusion algebra, in contrast with the standard c < 1 matter quasi-rational theory.
We obtained recursively the solution (8.8) from the difference equations (5.22) without
referring to the exact form of the contact terms. Instead, we strongly used the expected symmetries of the solution and the requirement that whenever Nm1 m2 m3 m4 = m2 , the solution coincides with (8.3), derived for one degenerate and three generic momenta. In the particular case
309

P2 = (m + 1)e0 = P2 = P4 = P1 in which i Pi = 2(m + 1)e0 all the three formulae (8.8),
(8.3) and (7.15) coincide.
We shall now show that (8.8) satisfies the difference relations (5.22), but with a contact term
proportional to the difference of two sl(2) multiplicities,
[N]m1 ,m2 +m3 ,m4 := Nm1 ,m2 +m3 ,m4 Nm1 ,|m2 m3 |,m4 .
(8.10)
The quantity [N] can take values 0, 1. The second term reflects the interpretation (8.2) of the
fields. Indeed, if we represent the fields as in (8.2) and assume that the 3-point functions of
the initial fields are given by (5.3), we have finally to retain two of the four resulting contact
termsnamely the ones with positive labels, as they appear in (8.10).
To prove the above statement we need some identities for the sl(2) multiplicities. In particular
we shall exploit
Nm1 +1,m2 ,m3 Nm1 1,m2 ,m3 = Nm1 ,|m2 m3 |,1 Nm1 ,m2 +m3 ,1 .
(8.11)
This identitythe r.h.s. of which represents the deviation from the simpler relation in (3.25),
is derived using the general integral representation (8.4); two of the initially four terms in the
r.h.s. survive, as in (8.11), when the equality is restricted to positive indices, i.e., when the multiplicities of the l.h.s. are given by (5.3), as we assume throughout this section. Applying (8.11)
to both sides of the following equality

Nm1 ,m3 ,m (Nm+1,m2 ,m4 Nm1,m2 ,m4 )
m=0
(Nm+1,m1 ,m3 Nm1,m1 ,m3 )Nm,m2 ,m4
(8.12)
m=0
we obtain
Nm1 ,m3 ,m (Nm+1,m2 ,m4 Nm1,m2 ,m4 )
= [N ]m2 +m4 ,m1 ,m3 = [N ]m1 +m3 ,m2 ,m4 .
(8.13)
If m1 ms , s = 2, 3, 4, the linear combination in (8.10), (8.13) is symmetric with respect to the

three variables m2 , m3 , m4 . Indeed in this case
[N ]m1 +m3 ,m2 ,m4
1
=
2
d
0
cos m1
sin
sin ms
s=2,3,4
= [N ]m1 ,m2 +m4 ,m3 = [N ]m1 ,m2 +m3 ,m4 = [N ]m1 ,m4 +m3 ,m2 .
(8.14)
Thus choosing the largest of the labels mi , say, m1 , as the one corresponding to the negative
chirality 1 = 1, we arrive at the symmetry relation (8.14) of the type of (5.23).
We shall now check that (8.8) satisfies the ring relation with the contact term given by the
linear combination (8.10). Indeed if we compute the shifts of the function (8.8)interpreted as
G (+) (P1 |P2 , P3 , P4 )we get, using (8.5), (8.13), (8.14),
(+)
(+)
(m1 + , m2 , m3 , m4 )
(m1 , m2 + , m3 , m4 )
G
G
4
4
=1
=1

e0
[N ]m2 +m3 ,m1 ,m4 + [N ]m2 +m4 ,m1 ,m3 = e0 [N ]m2 +m3 ,m1 ,m4 ,
=
2
310
(+) (m1 , m2 + , m3 , m4 )
G
4
=1
(+) (m1 , m2 , m3 + , m4 )
G
4
=1

e0
=
(8.15)
[N]m1 ,m3 +m4 ,m2 [N ]m1 ,m2 +m4 ,m3 = 0.
2
The r.h.s. of the second relation in (8.15) vanishes due to (8.14), so that it takes the form of
the diagonal version of the homogeneous relation (5.16). The two terms in the r.h.s of the first
relation are identical and sum up to one term (which can now take the values 0, 1). We stress that
these identities hold in the region of validity of (8.14), i.e., when the field of negative chirality is
chosen to coincide with the largest of the integers mi . Otherwise the formula (8.8) is symmetric
with respect to the four labels. Eq. (8.15) is an analog of the formula
(6.7) in the sense that,
similarly to (6.7), it reproduces solutions of the ring relations with i i = 2 in certain regions
of momenta (i.e., it does not distinguish the negative chirality sign unless we specify which mi
is bigger). What simplifies here the correlator and the shift equations is that the various local
regions are determined by the individual momenta and furthermore the intermediate momenta
all have an identical sign.
(+)
e0
The first line in (8.8) extends to negative ms , so that G4 (m1 , m2 , m3 , m4 ) = ( ML ) b
+
, Tb+ , respectively.
G4 (m1 , m2 , m3 , m4 ). The values m4 = 1, m4 = 1 correspond to T1/b
Restoring the prefactor b/e0 in (5.21) we can write
(+)
3
e0 (+)
G (m1 , m2 , m3 , 1)
b 4

e0
1 L b (+)
(+)
= G4 (m1 , m2 , m3 , 1) + 2
G4 (m1 , m2 , m3 , 1).
b M
1
(bL )Lb
(Q
s=1 s )
Nm1 ,m2 ,m3 =
(8.16)
We can interpret (8.16) as a boundary condition obtained from the first two of the four terms in
the diagonal action (4.26); the differentiation with respect to L gives the linear combination in
the second line in (8.16) (taken with a prefactor 1/b due to the rescaling in (5.18)). In that sense
the action defining our correlators is given by the two positive chirality terms of the diagonal
perturbation (4.26).

The 4-point correlator with i i = 2 is constructed in a similar way, parametrizing the
momenta as Pi = i mi e0 (so that they are physical for e0 < 0),
() (m1 , m2 , m3 , m4 )
G
4

1
Nm1 ,m2 ,m (me0 )Nm,m3 ,m4 + permutations
QNm1 ,m2 ,m3 ,m4
:=
2
m=0

1 e0
= Nm1 ,m2 ,m3 ,m4
i Q (Nm1 ,m2 ,m3 ,m4 1) ,
b
2
i
e0
Q
i = + mi .
2
2
Then

1
b(Q 3s=1 s )
Nm1 ,m2 ,m3
L L
b
()
= e0 bG4 (m1 , m2 , m3 , 1)
(8.17)

()
= G4 (m1 , m2 , m3 , 1) + b2
L
M
e0 b
()
G4 (m1 , m2 , m3 , 1)
311
(8.18)
can be interpreted as a boundary value related to the two negative chirality terms in (4.26).
The solution (8.8) with be0 > 0 (or (8.17) with be0 < 0) reproduces the 4-point correlation
function of the microscopic model [29]. Duality interchanges the normalised with (3.21) correlators; effectively both (8.8) and (8.17) get multiplied by Q times the standard powers of L ,
M . On the other hand the transformations (3.24) lead to correlators of the same type, in which
the positive integers mi = 2ji + 1 parametrize the diagonal degenerate Liouville points (i.e.,
Q 2 = (2j + 1)Q, so that now the Liouville scaling dimension takes a Sugawara form
L = j (j + 1)Q2 ). These correlators are solutions of a ring relation computed with the dual
diagonal action (4.27).
9. Degenerate fields in the conventional theory
As we have discussed, in order to extend all ring relations to the whole lattice L, one needs
to know all possible additional contact terms. On the other hand when only one of the tachyons
in the correlator is degenerate, solving some of the ring equations already determines the unique
solution.
One degenerate, three generic fields
It will be convenient to shift the notation compared with (4.10), so that the matter degenerate
momenta P = e0 mb + n/b are parametrized by non-negative integers m, n Z0 . We take
()
WP2 as the degenerate tachyon, while the momenta of the remaining three operators are assumed
generic. According to the analysis in Appendix A.3 there are no additional unknown contact
terms in this case. We shall solve recursively the equations, assuming that the tachyons at the
border lines n = 1 and m = 1 have vanishing correlators.
Let us start with the thermal cases n = 0, or m = 0. As before we take as initial conditions

() 1 , b , 3 , 4 =
s Q = Q b .
G
(9.1)
4
s=2
Solving recursively (5.12) we obtain

(+) 1 , 2 = b + mb , 3 , 4
G
4
2

m

4 1 r b , 2 = b, 3 , 4 + (m + 1) mb
=
G
2
2
r=m
(9.2)
mod 2
which can be also rewritten as

(+) 1 , 2 = b + mb , 3 , 4
G
4
2

mb
= (m + 1)
s Q +
= (m + 1)( Q b)
2
s=2

m
Q mb
mb
Q
= (m + 1)
+
s
+ kb .
2
2
2
2
s=2 k=0
(9.3)
312
Similarly (5.13) gives

() 1 , 2 = 1 + n , 3 , 4
G
4
b 2b

n
1
s Q +
= (n + 1) Q
= (n + 1)
2b
b
s=2

m
Q
n
n
Q
= (n + 1)
+
s
+ kb .
2
2b
2
2b
(9.4)
s=2 k=0
The m + 1 or n + 1 terms in each of the three channels of the above expansions correspond to
the weight diagram of the sl(2) irreps of dimension m + 1, or n + 1.
In deriving these formulae we have used only one of the ring relations. In the other channels
one has to take into account the additional contact terms. If the degenerate field is represented by
an integrated tachyon, then the accidental contact term due to (4.24), taken for n = 0, precisely
compensates the generic one. Indeed (9.3), (9.4) satisfy homogeneous relations, e.g.15

(+)
b
b
(+)
1 , 2 , 3 , 4 =
1 , 2 , 3 , 4 .
G4
G4
2
2
We can compare these solutions with the ones in (7.2), extending the latter to the values
P2 = e0 mb or P2 = e0 + n/b for = 1 respectively. For these special values (7.2) coincides
with (9.3), or (9.4), and this justifies the choice in the 1pi-term in (7.2), obtained by a different
argument.
Now let us consider an arbitrary degenerate momentum P2 = e0 mb + n/b. Solving (5.9)
recursively with m we get instead of (9.2)

mb
n
(+)
, 3 , 4
1 , 2 = b +
G4
2
2b

m

b
n
mb
(+)
=
(9.5)
.
1 r , 2 = b , 3 , 4 + (n + 1)(m + 1)
G4
2
2b
2
r=m
mod 2
(+) with a negative chirality correlator (9.4) with the

To proceed further we need to identify G
4
same value of the degenerate momentum P2 . We choose

() 1 , 1 + n , 3 , 4 ,
(+) 1 , b n , 3 , 4 G
G
4
4
2b
b 2b

1 mb
m
()
(+)
, 3 , 4 G4
1 ,
1 , b + , 3 , 4
G4
(9.6)
b
2
2
so that in particular the initial condition (9.1) for n = 0 = m is preserved. Inserting (9.6) in (9.5)
and using the first equalities in (9.4), (9.3), we obtain
15 On the other hand the homogeneous equation for the correlator of type ++++ is solved by G++++ ( , = b +
1 2
mb , , ) = (m + 1)(
Q).
s
3
4
=

2
2
313

mb
n
()
, 3 , 4
G4 1 , 2 = b +
2
2b

n1+
1
= (n + 1)(m + 1) Q b +
+ mb
b 2
2

m
n

rb
n
t
Q
Q mb
+
+
s +
+
.
= (n + 1)(m + 1)
2
2
2b
2
2
2b
r=m t=n
s=2
(9.7)
mod 2 mod 2
The identification (9.6) is suggested by the comparison with (7.2)the latter coincides with (9.7)
if P2 = e0 + n/b mb.
The third line of (9.7) illustrates the general form (1.2). The first fusion multiplicity corresponds to the shift of Ps with the weight diagram of the degenerate field, i.e. P = Ps + P2
e0 + 2kb 2l/b, k = 0, . . . , m, l = 0, . . . , n, while the multiplicity depending on three generic
momenta corresponds to the trivial solution (3.18). As in Section 7 we shall choose a solution of
the symmetry requirement preserving these fusion rules since once again they correspond to the
fusion rules of the underlying local matter correlator. We obtain a symmetric in the three generic
momenta Ps formula
G4 (P1 , P2 = e0 + n/b mb, P3 , P4 )

m
n

1
n
Ps + rb +
=
(n + 1)(m + 1) Q + mb +

2
b
r=m t=n
s=2

t
.
b
(9.8)
mod 2 mod 2
To check this result let us analyse directly Eq. (6.13) for the local correlators similarly as we
did in Section 7. We rewrite (6.13) as
1 , P2 b = e0 mb, P3 , P4 )
G(P

1 , P2 + b, P3 , P4 )
1 b, P2 , P3 , P4 ) G(P
=
G(P
+b+
(P2s + b)(P2s b),
s=3,4
P2s = Ps (m 1)b > 0,
s = 3, 4.
(9.9)
We shall illustrate this identity for m = 1 in which case the last correlator corresponding to the
border momentum P2 = e0 + b drops. We start with the local counterpart of (9.1) as an initial
condition

1 , P2 = e0 , P3 , P4 ) = Q 1
G(P
|Ps |.
2
2
s=2
Then
1 , e0 b, P3 , P4 )
G(P

1 b, e0 , P3 , P4 ) + b
=
G(P
(P2s + b)|P2s b|
s=3,4
1
1
= Q + b |P1 b| |P1 + b|
2
2
314

1
(Ps + b) + |Ps b| (Ps b) + (Ps + b)
2
s=3,4
=Q+b

1
|Ps + b| + |Ps b| ,
2
for P3 , P4 > 0
(9.10)
s=2
thus confirming formula (9.8) for this particular example.

The fixed chirality formulae (9.3), (9.4) were presented in [35]. The physical correlator (9.8)
reproduces the expression found by a different method in [37], in which the locality of the underlying correlators is automatically taken into account.
Similarly one solves the equations in the case when one of the tachyons is Liouville degenerate.
Four degenerate fieldsa conjecture
When all four fields are labelled by degenerate matter representations the 3-point function
3 = 1 is to be replaced by the fusion multiplicity in (5.3), (5.5). Accordingly the initial value
G
(9.1) gets multiplied by this multiplicity. The equations themselves get more complicated due
to many additional contact terms and the possible cancellations between them. We conjecture
that the effect will be, like in the diagonal case, an expression in which the 3-point sl(2) fusion
multiplicities (5.5) determine the expansion range, while the factors n + 1 and m + 1 in (9.7) are
replaced by the 4-point sl(2) multiplicities in (5.6), symmetric under the change of sign of any of
ni
mi b
the momenta. For i = Q
2 ( 2b 2 ), i = 1, 2, 3, 4 and Pm,n = n/b mb, ni , mi , n, m N,
this leads to

1
1
()
G4 (1 , 2 , 3 , 4 ) = NP1 ,P2 ,P3 ,P4 bNm1 ,m2 ,m3 ,m4 + Nn1 ,n2 ,n3 ,n4
2
b

n
mb NPm,n ,P3 ,P4

NP1 ,P2 ,Pm,n
b
m,n=1
+ permutations{2, 3, 4}

1
= NP1 ,P2 ,P3 ,P4 Q b +
b(Nm1 ,m2 ,m3 ,m4 1)
2

1+
+
(9.11)
(Nn1 ,n2 ,n3 ,n4 1) .
2b
In the last equality we used (8.9). The conjectured local correlator is given by a formula as in the
first line of (9.11) with intermediate momenta ( nb mb) replaced by | nb mb|.
10. Summary and discussion
In this paper we reported the results of our study of 2d quantum gravity, or non-critical bosonic
string theory, with generic non-rational values of the matter central charge (1.1).
The main point of our investigation is the systematic study of the effects of including matter interactions in the 2d string. Conventionally one adds to the Gaussian action the two matter
screening charges, which together with the Liouville ones serve as interaction terms. Motivated
by the comparison with a discrete, microscopic approach, to be discussed in a subsequent paper [29], we introduced and studied also another deformation of the Liouville theory, defined
by the interaction action (4.26). While in the first, conventional theory, the c < 1 (matter) and
315
c > 25 (Liouville) parts factorize before moduli integration, there is no such factorization in the
second theory, which we called diagonal.
To construct the tachyon correlators we have adopted and extended the ground ring approach
introduced long ago [1418]. In this approach the matter-Liouville factorization of the integrand
of the 4-point tachyon correlators (in the conventional theory) is not directly exploited, and so
the precise realization of any of these c < 1 and c > 25 correlators is not a priori required.
In particular no assumption on the existence of a fully consistent non-rational matter theory is
made. Indeed such a theory has not been rigorously established in the conventional theory, and
does not exist in the second, diagonal theory. For our purposes it was sufficient to derive a 3-point
generic matter OPE constant, formula (3.8), a c < 1 analog of the Liouville DOZZ formula, (3.2)
which extends the DotsenkoFateev Coulomb gas constant. The ground ring method is based
on the derivation of functional equations for the tachyon correlators, using the module action
(operator product) of the fundamental ground ring elements on a(0, 0)-form tachyon W in the
presence of integrated tachyons. The OPE coefficients of the ground ring action are determined
by well defined free field correlators, computed either by using the matter-Liouville Coulomb
gas representation, or exploiting the factorization into known c < 1 and c > 25 Coulomb gas
correlators.
The explicit 3-point OPE coefficients in (4.5), (4.6) confirm the ground ring structure conjectured in [24]. The functional relations for the 3-point functions are closely related to a standard
identity for the tensor product decomposition multiplicities of sl(2) finite dimensional irreps,
which are reproduced as a particular case.16 Besides those in the non-rational case one has more
3-point solutions and some were used as a building block in the construction of the 4-point solutions we have described. The diagonal theory admits an action of the ground ring generated by
the new deformations of the product of ring generators a a+ . The result (4.28) is an effective
projection of the ring action to a diagonal sl(2) type identities.
What complicates the case of n-point functions, n > 3, are the additional contact terms in the
functional relations due to the fact that the fourth, etc. field, given by an integrated tachyon T ,
serves as a new screening charge. Thus, besides the two operator terms in (4.5), (4.6) which
correspond to perturbations by the screening charges in the interaction actions (2.4), (2.5), there
are other channels in the OPE of a ring generator and a tachyon W . These OPE terms account for
the effect of the QBRST -exact terms, skipped in the r.h.s. of (4.5), (4.6). We have computed two
series of 4-point OPE coefficients, (4.21), (4.24), sufficient for the class of tachyon correlators
we consider. The diagonal model is more restrictive on the content of the operator products and
in particular leaves less room for contact terms.
We have found basically two types of 4-point solutions of the functional equations (5.8). Apart
from a particular example, both involve an integer number of some of the screening charges. We
have presented in more detail the solutions with matter screening charges, however, because
of the symmetry of the ring identities, in the conventional non-diagonal theory some of these
solutions have Liouville analogs as well.
The first class of solutions, (7.2), appears for generic values of the four tachyon momenta,
such that their sum is restricted by a matter charge conservation, thus generalizing the tachyon
correlators for Gaussian matter of [34]. The arbitrariness in the solutions of the homogeneous
equations, or, effectively, in the determination of the 1pi part in (1.2), is partially fixed by com16 We stress that these multiplicities would not be allowed if the formal matter Liouville factorisation was taken
too literary.
316
parison with 4-point functions in which one of the tachyons is a screening charge. The choice
of these boundary conditions corresponds to the type of interaction action, which otherwise
enters the definition of the correlators rather formally. To fix the remaining arbitrariness we have
required that different classes of solutions are related to each other, whenever their partial wave
expansions, as in (1.2), are comparable.
The second class of solutions found in Sections 8 and 9 represents 4-point functions in which
one field (9.7), or all fields (formulae (8.8), (8.17) in the diagonal theory) correspond to a degenerate Virasoro representation. For the correlators with four degenerate fields in the conventional
theory we only give a conjecture. The problem is complicated by the unknown additional contact
terms at degenerate values.
The equations we have derived and studied apply by definition to the correlators satisfying
the chirality rule. Besides these fixed chirality, and hence partially symmetric, 4-point functions
we have described also correlators symmetric with respect to the four (or the three generic)
matter charges. We interpreted this symmetry as tachyon locality. Until this point locality of
the underlying c < 1 and c > 25 correlators is only partially exploited in the computation of the
OPE coefficients. In the simplest example in Section 6 the set of fixed chirality solutions serves
as a local basis for the local correlators. Then the original equations are rewritten equivalently
as equations for these symmetrised correlators. To fix in general the arbitrariness in the solution
of the symmetry requirement we have exploited the fact that all our solutions admit the channel
decomposition form (1.2). Our universal choice was to preserve the fusion multiplicities in the
symmetric counterpart of this expressionthe formal rule is to replace the inverse propagator P
with |P |. This choice indirectly takes into account the locality of the underlying c < 1 theory,
since these fusion rules correspond to the ones manifested by the local correlators of that theory.
Furthermore the local correlators (6.7), (7.5), (9.8) are invariant under Liouville reflections.
The symmetric correlators (as well as their analogs, symmetrised with respect to the Liouville labels) do not satisfy globally the original ring relations, rather satisfy shift equations with
modified and momenta dependent inhomogeneous terms. We have proposed an alternative recursive derivation of these equations for the local correlators starting with the simplest case of
Section 6.1. It also yields a full set of local representatives of the symmetric tachyon correlators,
extending the set of fixed chirality correlators obtained as solutions of the initial equations.
Our treatment of the local correlators remains however rather phenomenological and the
direct derivation of their equations is still an open problem which requires an extension of the
Coulomb gas based technique we had mostly exploited. Conceptually this is important since it
is natural to interpret the local correlators as the true physical ones, while the sets of partially
symmetric, fixed chirality correlators, though basic in our construction, should be considered
rather as auxiliary objects. This is confirmed by the matrix model approach [29] (formula (7.16))
and also by the comparison with the recent paper [37], in which the underlying matter and Liouville theories are explicitly exploited in the computation of the 4-point function with one matter
degenerate and three generic fields: the local correlator (9.8) coincides with the expression in [37]
computed by this more constructive method. A notable exceptional case, avoiding these problems
and confirmed by the discrete model in [29], are the 4-point functions (8.8) of four degenerate
fields in the diagonal theory.
Our analysis has been restricted so far to the bulk quantities. However, as it is well known from
the studies in the rational matter and the generic Liouville BCFT, the bulk 3-point correlators,
i.e., the (properly normalized) OPE coefficients, give information about the boundaries, since the
matrices diagonalizing them are closely related to the disc 1-point functions, as briefly discussed
in Appendix B. In Appendix A.5 we have also computed some chiral OPE coefficients in the
317
presence of matter charges, including the four OPE coefficients of the boundary ground ring,
which has a similar to (4.5), (4.6) two-term structure. The functional relations for the boundary
tachyon correlators, generalizing the trivial matter case [18,25], will be discussed elsewhere, see
also the paper [36] which appeared meanwhile, which deals with this problem too, but in the
minimal string theory.
Acknowledgements
We thank Al. Zamolodchikov for numerous valuable discussions. We also thank S. Alexandrov, A. Belavin, Vl. Dotsenko, V. Fateev, V. Schomerus, M. Stanishkov and J.-B. Zuber for the
interest in this work and for useful comments. Special thanks to Paolo Furlan for the careful reading of the manuscript, the many useful remarks on it, as well as for the explicit computation included in formulae (4.18), (4.19). I.K.K. thanks the Institute for Advanced Study, Princeton, and
the Rutgers University for their kind hospitality during part of this work. V.B.P. acknowledges
the hospitality of Service de Physique Thorique, CEA-Saclay and the University of Northumbria, Newcastle. This research is supported in part by the European Community through RTN
EUCLID, contract HPRN-CT-2002-00325, MCRTN ForcesUniverse, contract MRTN-CT-2004005104, MCRTN ENRAGE, contract MRTN-CT-2004-005616, MCRTN ENIGMA, contract
MRTN-CT-2004-005652, and by the Bulgarian National Council for Scientific Research, grant
F-1205/02.
Appendix A. Coulomb gas computations
In this appendix we shall compute some matrix elements of the type

|c1 c0 dzn Vnn (zn ) dz2 V22 (z2 )a (z)cV11 (z1 )|0
free

=
Cn

C2

n

i ei 1
1
1 e1
1
b
z z1
b zi z
i=2
n
1
|Vn (zn ) V( b , b ) (z)V1 (z1 )|0
free
2
2

1
1
1
=

M z
L 2
L z
M

M
L
z z1
b2
b
(A.1)
and their volume integral counterparts, which determine the OPE coefficients of the ring generator a with the tachyon fields. Everywhere here V , or V(e,) denote unnormalized products
of vertex operators, with no relation necessarily of the type in (2.7) on the pair of matter and
Liouville charges (e, ). There is a similar formula for the other generator.
Conventions:

1
2
(x1 )(x2 ) = log x12
= (x1 )(x2 ) ,
2

an
an
i
i
(+)
zn + iq
zn ,
(z) = (z) + () (z) := a0 log z +
n
n
2
2
n>0
n>0
[an , am ] = nn,m ,
[a0 , q] = i,
(+)
()
2(e e ) ()
(+)
V(e1 ,1 ) (z1 )V(e2 ,2 ) (z2 ) = z12 1 2 1 2 V(e2 ,2 ) (z2 )V(e1 ,1 ) (z1 ),
|z1 | > |z2 |.
318
Let us also recall some ghost field correlation functions. The correlators of the ghost b, c
fields decouple as the full correlators factorize. The 2-point function is computed in the vacua
0|c1 c0 c1 and |0
, normalizing 0|c1 c0 c1 |0
= 1,

b(z1 )c(z2 ) = 0|c1 c0 c1
bk z1k2
cm z2m+1 |0
k=1
m=1

1 z2 p
1
= 0|c1 c0 c1 |0
=
,
z1
z1
z12
|z1 | > |z2 |.
(A.2)
p=0
The 3-point ghost c function is

0|c(z1 )c(z2 )c(z3 )|0
= 0|
1
3

ki +1
cki zi
|0
= z12 z13 z23 .
(A.3)
i=1 ki =1
The 3-point function with the insertion of one field :bc: reads
0|c1 c(z2 ):bc(z):c(z1 )|0

p+1
s+1
m1
= 0|c1
cs z2
z
bk cm+k
cmk bk
cp z1 |0
s=0
= 0|c1 c(z2 )c(z1 )|0
k=2
1
1
z2 z z z1

,
k=1
|z2 | > |z| > |z1 |,
p=1
(A.4)
while
1
(A.5)
, |z| > |z1 |.
z z1
The last formula is used in (A.1) producing the shifts by 1. In particular it leads to the last term
in the matter-Liouville factorized expression in the last line, where one is using the representation
of the ring operator in terms of derivatives,

1
bi
z + bc(z) eb :,
a (z) = :e
(A.6)
b2
0|c1 c0 :bc(z):c(z1 )|0
=
meaning action of the derivative to the right minus action to the left.
A.1. 3-point volume integral matrix elements
We start with some bulk correlators, most of which have been already computed [1618]. In
these examples we shall use the first representation in (A.1), while in the next subsection we will
exploit the matter-Liouville factorized expression in the second line.
The factor c1 c 1 c0 c 0 is denoted (cc)1 (cc)0 . Consider first the matrix element

|(cc)1 (cc)0 a (x0 ) ccV (x1 )

e b 2 2 b(+e 1 )

b
x01
= e + b2 , b2
b
if = 1,
0
=
(A.7)
( b1 (Q2+b)
2Q 2
if = 1
( b ) = ( 1 (Q2)
b
319
and = Q + b2 , recovering
a W = W b ,
a+ W+ = W +
1
2b
(A.8)
a W+ = a+ W = 0.
(A.9)
Our next example is

2

d x2
(cc)1 (cc)0 a (x0 ) ccV (x1 )V22 (x2 ) free

= e + e2 + b2 , + 2 b2

2
d x2 e b 1
2 e2 1 2 2 b(+e) 2 b(2 +e2 ) 2 2ee2 22
=
x
x02
x21
+
.

b
z01
b
z02 01
(A.10)
In the three of the four possible cases this integral vanishes for generic momenta, either due to
factors (1), or, because of sign compensation of the various terms. It survives only for = 2 =
1 producing the constant
(b(Q 2( + 2 b2 ))
(b(22 b)) (b(2b 2 22 ) + 1)
=
(b(b 2) + 1)
(b(Q 2)) (b(Q 22 ))
which precisely provides the leg factor normalization of the three tachyons, thus recovering the
first formula in (4.17).17 In agreement with the BRST invariance, both in (A.7) and the integrated
(A.10) only the combination satisfying the mass-shell condition survives, while all the other
terms, possible in the analogous pure matter or Liouville 3-point matrix elements, now cancel
out automatically, due to the effect of the raising prefactor in the ring generator. In particular
choosing 2 = b or 2 = 0 one recovers the generic two term action in the first line in (4.5).
Finally for 2 = 3 = 1 and 2 + 3 = b there is a double integral matrix element
2 2

d x2
d x3
+
)(x1 )V+2 (x2 )Vb
(x3 ) free
(cc)1 (cc)0 a (x0 )(ccV
2
( b1 (Q 2 b))
( b1 (Q 2)) (b(Q 22 )) (b(Q 2(b 2 )))
(A.11)
This constant reproduces again the relevant leg factors and thus we obtain for the normalized
fields the first of the relations (4.25).
A.2. The general 3-point constant
In general accounting for all possible matter and Liouville screening charges one computes
the 3-point function of the ring generator with two tachyons using the representation in the last
line in (A.1). The result for the 2d integral (for the unnormalized tachyons) is proportional to
the product of the 3-point c < 1 Coulomb gas OPE constant computed in [32] and its c > 25
counterpart given by an analytic continuation of the same formula,
17 In particular the non-generic value + b = Q corresponds to a tachyon of no definite chirality, for which the
2
2
2
numerator and equally the compensating leg factor, become singular.
320
0| ccV(e30 e3 ,Q3 ) ()(cc)0 a ccV(e22 ,2 ) |0

b
k2
b
= c(2 , 3 )C Matt , e2 , e0 e2 + k1 b +
2
2
b

b
s
b
2
C Liou , 2 , Q 2 + s1 b +
,
2
2
b
where

b
b
3 + 2 Q
3 2 +
2
2

2
b
b
+ 3 2
3 + 2 Q +
2
2

1
b
b
= 4
e 3 + e2 e0
e3 e 2 +
2
2
b

2

b
b
.
e3 + e2 e0 +
+ e3 e2
2
2
c(2 , 3 ) =
(A.12)
1
b4
(A.13)
Here e3 = e2 + b2 k1 b + kb2 , 3 = 2 b2 + s1 b + sb2 and the four integers s1 , s2 , k1 , k2 the
, T0+ , T0 , are restricted by (2.7) depending on the

number of screening charges of type Tb+ , T1/b
combination of chiralities, i.e.,
1 + 3
1
(A.14)
+ b3 b2 = (s1 + 3 k1 )b + (s2 3 k2 ) .
2
b
The OPE coefficient for the normalized tachyons is given, by the r.h.s., of (A.12) times the ratio
of leg factors, i.e.,

b
a W22 =
c(2 , 3 )C Matt , e2 , e0 e3
2
3 ,3

b
(b2 (Q 22 )) 3
W .
C Liou , 2 , Q 3
(A.15)
2
(b3 (Q 23 )) 3
(3 2 )e2 + b
The coefficient in the r.h.s. is examined either using directly the expressions of the two c < 1
and c > 25 Coulomb gas constants, or by exploiting the compact formula (3.8) for the matter
constant and the relation of C Liou to C Liou in (3.2), regularizing 1 = e1 = b/2 + . We get
that the overall constant goes to zero like 2 , unless one of the four factors
1
2 3 2 2 2
e13
e0 e12
(e123 e0 )2 e23
vanishes as well.
The values e3 = e2 b/2 are equivalent to k2 = 0, k1 = 0, 1. When the chirality is preserved,
2 = 3 , we obtain taking into account the condition (A.14) two solutions for each sign
k 2 = 0 = s2 ,
k1 = 1 s1 = 0, 1,
k 2 = 0 = s2 ,
k1 = s1 = 0, 1,
for 2 = 1 = 3 ,
for 2 = 1 = 3 ,
altogether leading to the generic OPE relations (4.9), (4.5).

For e3 = e2 b/2 but 2 = 3 = 1 the values of e2 become restricted by (A.14)
e0 2e2 = P2 = Q 22 = s1 b +
s2 b b
,
b
2 2
(A.16)
321

b
b
(e3 , 3 ) = e2 , Q 2 2 .
2
2
(A.17)
The resulting (e3 , 3 ) correspond to the Liouville reflected counterparts of the two terms in (4.5);
they have to be added whenever the momenta take the special discrete values in (A.17). These
values include the Liouville degenerate points (with the plus sign in (4.14)). Similar formula
arises for 2 = 1 = 3 .
One gets a non-zero expression also for e3 = e0 e2 b/2. The combination of chiralities
2 = 3 reproduces the matter reflected points occurring for
k2 b b
+ ,
e0 2e2 = P = k1 b +
b
2 2

b
b
(e3 , 3 ) = e0 e2 , 2 2
2
2
(A.18)
with s2 = 0, s1 = 0, 1, for 2 = 1, and s2 = 0, s1 = 1, 0 for 2 = 1, respectively. These values

include the matter degenerate momenta (with the positive sign in (4.10)). Finally for e3 = e0
e2 b/2 and 2 = 3 there are two series of solutions, corresponding to both matter and Liouville
reflections
b b
k2
e0 2e2 = k1 b + ,
2 2
b

b
b
(e3 , 3 ) = e0 e2 , Q 2 2 ,
2
2
(A.19)
with s1 + k1 = 1, s2 = k2 for 2 = 1 and k2 = 0 = s2 , s1 = k1 for 2 = 1.

2
2)
and the final result simplifies
In all cases the constant in (A.13) becomes c(2 , 3 ) = (Q2
b2
to powers of L , M and b, e.g., in the case (A.18), e0 2e2 = Q 22 = k1 b k2 /b we have
k1
b
(+)
C b 2
2 ,
k2
2
b
= M2
b

(Q 2)2 (b(Q 2))
b
Matt b
=
, e2 , e 2
C
2
2
b2
( b1 (Q 2 + b))
(A.20)
using that C Liou ( b2 , 2 , Q 2 + b2 ) = 1, while C Matt ( b2 , e2 , e2 b2 ) is determined by matter

reflection as in (3.16) from (3.14). This constant is finite for positive integers k1 , k2 = 0, which
are the values of the degenerate representations in (4.10). Similar formulae hold for the other
cases in (A.17), (A.18) and the constituent matter or Liouville Coulomb gas constants C Matt ,
C Liou are finite for the degenerate values of the momenta. On the other hand in the last case
(A.19), which involves values on the boundary of the degenerate regions appear, there may appear
singularities in the constants or the leg factors. In all these considerations we have assumed that
b2 is generic, non-rational.
In contrast to the above result in the diagonal case the 3-point function computed with k1 ,
+
k2 , s1 , s2 tachyons of type Tb , T1/b

, Tb+ , T1/b
respectively, is more severely restricted. We have
s2 +k2
e3 = e2 e20 + (k1 + k2 )e0 , 3 = 2 Q
2 + (s1 + k1 )b + b and the condition (A.14) is replaced
by
322
e0
Q
3 + b3 b2
2
2

1
= s1 + k1 + 3 (k1 + k2 ) b + s2 + k2 3 (k1 + k2 ) .
(A.21)
b
Thus in the generic case 2 = 3 we have s1 = 1, s2 = k1 = 0 = k2 , or k2 = 1, s2 = k1 = 0 = s1 for
2 = 1 and s2 = 1, s1 = k2 = 0 = k1 , or k1 = 1, s1 = k2 = 0 = s2 ,for 2 = 1. These solutions
all involve one of the interaction terms in (4.26). The action of a a+ can be understood as the
composition of the free field formulae (A.8), (A.9), (4.17) for the two generators
(3 2 )e2 +
a+ a T+1 W+ = a+ W +
= W +
+1 b2
+1 Q

2

+
+
+
a+ a Tb W = W e0 , a+ a T 1+ W+ = W +
(A.22)
e0
2
and
a+ T1 a W = a+ T1 W b = W
2
+1 Q
2

a+ T 1 a W = W e0 ,
a+ Tb a W = W
b
e0
2
(A.23)
If the chirality is inversed there are more solutions. However, restricting to diagonal momenta
the only solutions are P = e0 , 0. Since the tachyon of P = 0 has no definite chirality, all these
solutions effectively fit the generic formula (4.28).
A.3. Mass-shell restrictions on the contact terms
The contact terms in the difference equations for the 4-point tachyon correlators are determined by the OPE coefficients computed by the 4-point Coulomb gas functions
1
4
()(cc)0 a W22 T33 |0
,
0|WQ
4
((Q 24 )b4 ) ((24 Q)b4 )
b
s2
b
k2
4 = 2 + 3 + s1 b + ,
(A.24)
e 4 = e 2 + e 3 + k1 b + .
2
b
2
b
As in the computation of the 3-point OPE coefficients the denominator comes from the leg factors
in the trivial 2-point matrix element, cf. (3.20),
C
(2 3 4 ) 4
b2 2 3
4
4
0|WQ
()(cc)0 W44 |0
= 0|c(z)zc(z)z WQ
(z, z )W44 (z , z )|0
.
4
4
If 2 coincides with one of the four values of the screening charges (A.24) reduces to a 3-point
function. The mass-shell condition (2.7) implies
1 + 4
b + b4 b2 b3
2
1
= (s1 + 4 k1 )b + (s2 4 k2 ) .
(A.25)
b
For generic momenta P2 , P3
/ L and P2 + P3
/ L the only solution of the mass-shell condition
occurs for 2 = 3 = 4 = 1, with k2 = s2 = k, k1 + s1 = 0, whence k1 = 0 = s1 . The case k = 0,
corresponds to (4.17) while k 1 to (4.18), (4.19).
N1
The relation (A.25) is generalized to a product W22 T33 TN1

in the N -point analog of the
Coulomb gas correlator (A.24), which contains N 3 1 integrated tachyons. If the momenta
(4 2 )e2 + (4 3 )e3 +
323
are generic, i.e, Pi1 + + Pis

/ L for any partial sum, the mass shell condition implies i =
N , i = 2, 3, . . . , N 1. Hence for N = 1 one gets s1 + k1 + N 4 = 0, k2 = s2 , while for
N = 1, the constraint is k1 = s1 , k2 + s2 + N 3 = 0. In both cases there are no solutions
for N 3 > 1. As a consequence, there are no new contact terms coming from two or more
integrated tachyons in the equations for the n-point functions, n 5 for those generic values of
the
If the given chiralities i , i = 2, . . . , N 1 are all identical, imposing the condition
Nmomenta.
1
P
=

L
forces N to be of the same sign which again implies that there are no solutions of
i
i=2
the mass
shell condition for N > 4. Similarly for 2 = 1 and i = 1, i = 3, . . . , N 1 imposing
P2 = L, N1
i=3 Pi = L excludes any contact terms. These properties are taken into account in
writing Eq. (6.1).
Apart from the above series of contact terms for 2 = 3 = 4 , which occurs for generic
momenta, there are various possibilities taking place for particular values of 2 or 3 , determined
by the choice of the signs i in (A.25). E.g.,
2 = 1 = 3 = 4 ,
2 = 1 = 3 = 4 ,
e0 2e2 = P2 = (k2 s2 )/b (s1 + k1 )b,

e0 2e3 = P3 = (k2 + s2 + 1)/b (k1 s1 )b.
(A.26)
In particular these conditions admit solutions for P2 or P3 which correspond to the degenerate
matter values when the interaction involves the two matter charges. Another example of (A.25)
is given by 2 = 3 = 1 = 4 which may occur for generic values of P2 and P3 but their sum
restricted by P2 + P3 = (s1 k1 )b + (s2 + k2 + 1)/b.
For the N -point generalization of (A.24) with p = N 3 > 1 integrated tachyons, there are
p+2
further possibilities. E.g., the second condition in (A.26) is replaced by i=3 Pi = (k2 + s2 +
p)/b (k1 s1 )b. etc. The simplest example with p = 2 and no screening charges was computed
in (A.11).
In the diagonal theory described by the action (4.26) the mass-shell condition for the 4-point
function (A.24) is again more restrictive. We have e4 = e2 + e3 e20 + (k1 + k2 )e0 , 4 = 2 +
3 Q
2 + (s1 + k1 )b + (s2 + k2 )/b and in the generic case 2 = 3 = 4 = 1 the condition admits
the unique solution k1 = 0 = k2 = s1 = s2 . It reproduces the OPE in the first lines in (A.22) and
(A.23). The analog of the second example in (A.26) is
e0 2e3 = P3 = (s2 + 2k2 + k1 )/b (k2 s1 )b.
Restricting to diagonal values P3 = ke0 , we obtain s1 + k1 + k2 + s2 = 0, i.e., the only possible
value is P3 = 0, which is beyond the degenerate matter range.
The above conditions on the momenta are kinematical. As in the analysis of the general 3point function (A.12), further restrictions appear from the fusion rules of the degenerate fields
dictated by the 3-point constants in the decompositions of the 4-point matter and Liouville functions. In the next section we consider some examples.
A.4. 4-point OPE coefficients
We derive here the 4-point OPE coefficients in (4.18) and (4.24).
The vertex part of the function (A.24) for
b n
b n
(A.27)
+ ,
4 = 2 + 3 + ,
i = 1, i = 2, 3, 4
2 b
2 b
n
is realized by a 2n-multiple integral coming from the power of screening charges (T0 T1/b
) .
We shall use instead the matter-Liouville factorization formula as in the last line in (A.1), with
e4 = e2 + e3 +
324
an alternative realization of each of the two types of correlators. Since one of the fields is the
simplest degenerate field, these c < 1 and c > 25 correlators are standard, given by sums of
products of hypergeometric functions. In particular in our example only one of the OPE channels
survives. E.g., the matter matrix element is given for the values ei in (A.27) by the product of
blocks with intermediate charge e2 + b/2,
e0 e4 |Ve3 (x3 )V b (x0 )Ve2 (x2 )|0
M
2
2 M (e4 )M (e2 )M (e3 )M (b/2)
= x32

z02
2 F1 n, (2e2 + 2e3 + 2b)b + n 1; (2e2 + b)b;
(same, with z z )
z32

b
b
C Matt , e2 , e0 e2
2
2

b
b n
C Matt e2 + , e3 , e0 e2 e3
fM (z)fM (z),
2
2 b
be2 be3
z02
z30
.
fM (z) =
(A.28)
z32
z32
The expression for the Liouville correlator Q 4 |V3 (x3 )V b (x0 )V2 (x2 )|0
L is analo2
gous, with bei , b2 replaced by bi , b2 , etc., while the constants C Matt are replaced by the
c > 25 Coulomb gas constants C Liou . The first of the OPE constants in both cases is trivial,
C Matt ( b2 , e2 , e0 e2 b2 ) = 1 = C Liou ( b2 , 2 , Q 2 + b2 ). The hypergeometric function is the
same, using that ei = i b, and it reduces to a finite series of n + 1 terms. We then apply the
derivatives with respect to z0 and z 0 term by term. Using that the difference of the powers of z0i ,
i = 2, 3 from the matter and the Liouville functions is a constant b(i ei ) = b2 one gets simply
an overall factor

2
1
1
d x3
fM (z)fL (z)
z0i
z02
i

2
z02
2 F1 n, (22 + 23 2b)b + n 1; (22 b)b;
(same, with z z )
z32

2

2 u d 2 x3
z02
= x02
2 F1 n, w + n 1; u;
z32

2
2 v1 2 w
z 02
x30
x32
2 F1 n, w + n 1; u;
z 32

2
(n)k (w + n 1)k
=
3 F2 (n, w + n 1, u + k; w + k, u; 1)
k!(w)k
k
(v) (w + 1)
(u + 1)
wn+1
(v)n
= n!
(u)n (w)n w 2n + 1
2
(v) (w + 1)
=: C(2 , 3 ; n)
(u + 1)
where
u := (22 b)b,
v := (23 b)b,
w := u v = (b 24 )b + 2n.
(A.29)
325
Altogether one obtains for the values in (A.27) the OPE coefficient in the nth term in (4.21)
(+++)
( L M )n = C b
2 2 3
(b(Q 22 )) (b(Q 23 ))
= C(2 , 3 ; n)
(b(Q 24 ))

b
b
C Matt e2 + , e3 , e0 e4 C Liou 2 , 3 , Q 4 .
2
2
(A.30)
This result has been derived for generic values of the momenta. If however 2 takes a degenerate
p
value, 2 = b mb
2 2 , m, p Z0 , one of the Coulomb gas constants in (A.30) vanishes, the
other becomes singular. The vanishing of the constant reflects a null vector factorization so it is
natural to resolve this ambiguity for such fields by restricting the validity of (A.30) to the values
of n avoiding the singularity, i.e., n p. On the other hand we can still keep (A.30) for any n for
fields with degenerate labels but not obeying the factorization conditions.
This derivation generalizes to other cases discussed in the previous section. In particular let
us consider the case described by the second line in (A.26)
(k1 s1 + 1)b k2 + s2
,
2
2b
b
s2
1
4 = 2 + 3 + s1 b + = e4 + .
2
b
b
1
2 = e2 + ,
b
3 = e3 + b =
Examining the product C Matt (e2 + b2 , e3 , e0 e4 )C Liou (2 + b2 , e3 , Q 4 ) one observes that

for generic values of 2 it becomes singular only if = = 1, k1 = 0, s1 1, or = = 1,
s1 = 0, k1 1; alternatively for these values the corresponding products C Matt C Liou are finite (as
well as each of the two constants itself). This implies that in each of these two cases only one
of the four possible products of matter and Liouville blocks survives in (A.24). Furthermore the
hypergeometric functions corresponding to the matter and Liouville local correlators are identical
again and the differences of overall powers of z0i are b2 as before. E.g., in the case s1 = 0, the
chiral factor fM (z)2 F1 in (A.28) is replaced by

zb(e0 2 ) (1 z)be3 2 F1 s2 , b(2e3 2e2 ) + s2 + 1; 1 + b(e0 2e2 ); z

2
2
= zb2 b (1 z)b3 b 2 F1 s2 , (22 + 23 2b)b + s2 1; (22 b)b; z .
Then all the steps in the derivation of (A.29) are repeated with n replaced by s2 or k2 respectively.
We summarize these results by relations analogous to (A.30) for s1 = 0, k1 1, and k1 = 0,
s1 1, respectively:
(+)
kM1 kM2 sL2 = C b
2 2 3
= C(2 , 3 ; s2 )
( b1 (Q 22 )) (b(Q 23 ))
( b1 (Q 24 ))

b
b
Matt
Matt b
, e2 , e0 e2 +
C
e 2 , e 3 , e 0 e4 C
2
2
2

b
C Liou 2 , e3 , Q 4 ,
2
326
(+)
kM2 sL1 sL2 = C b
2 2 3
= C(Q 2 , 3 ; k2 )
( b1 (Q 22 )) (b(Q 23 ))
( b1 (Q 24 ))

b
b
b
Matt
Liou
e2 + , e3 , e 0 e 4 C
, 2 , Q 2
C
2
2
2

b
C Liou 2 + , e3 , Q 4 .
2
(A.31)
Denoting s2 + k2 = n, s2 k2 = 2s n and k1 = m + 1, or s1 = m + 1, respectively we arrive at

(4.24). The case in the first line of (A.26) can be analysed similarly.
On the other hand the alternative multiple integral representation of any of these 4-point
correlators is not of the type in [32]. Comparing with (A.31) one effectively computes these
non-standard integrals.
+
replaced by the tachyon W4 (or by
One can compute also the 4-point correlator with WQ
4
WQ
) so that the labels i (or the labels ei ) satisfy a Liouville (matter) reflected version of
4
the respective charge conservation condition. This does not change the hypergeometric functions
and it remains to use (3.15) or (3.16), respectively. The result is
24 Q
b
b 2 L
(++)Q4
b2 2 3
=C
(+++)4
b2 2 3
2e4 e0
b
= b M
2

(++)4
C b
2 2 3
L
M
24 Q
b
(+++)Q4
.
b2 2 3
(A.32)
Thus the reflection properties of the underlying 4-point Liouville and matter correlators ensure
the validity of (3.19) on the level of these particular string 4-point correlators.
A.5. Some chiral OPE coefficients
Now we consider a few chiral matrix elements, some of which have been computed in [18,25].
The chiral analog of the simplest matrix element (A.7) reads for |z0 | > |z1 |

0
c1 c0 a (z0 ) cV (z1 ) = 2Q
( 1 (Q2+b)
b = (b 1 (Q2)
if = 1,
if = 1.
(A.33)
We recognize in the r.h.s. the leg factor normalization exploited in the boundary theory, which is
obtained replacing in (2.6) (x) (x).
The fields are radially ordered as above, accordingly the bounds on the integrals are given
by the arguments of the neighbouring fields, the utmost left one being at +, the utmost right
onein . E.g., let us look at the chiral analog of (A.10) for = 1 = 2 . We choose |z0 | > |z1 |
and send these two arguments to 1 and 0, respectively. The coordinate z2 is floating and we can
collect the result for the three possible insertions of the integral by writing the linear combination
with coefficients indicating the contours of integration

dz2 c1 c0 a (z0 ) cV+ (z1 )V+2 (z2 )
ci,j
(ij )
327
Cij
(b(Q 2 22 + b))
=
(b(Q 2))(b(Q 22 ))

sin b(Q 2 22 + b)
1
c,1
+ c1,0
sin b(Q 2) sin b(Q 22 ))
sin b(Q 22 )

1
+ c0,
.
sin b(Q 2)
(A.34)
The overall constant reproduces the chiral leg factor normalization of the three fields. Taking
2 = b or 2 = 0 the r.h.s. can be identified with a (linear combination of) Liouville or matter
matrix elements of three chiral vertex operators (CVO). The intermediate states are described by a
proper choice of the coefficients in (A.34), as it has been done in the boundary Liouville case [5];
the boundary fields are linear combinations of CVO.18 Each of the two constants determine the
corresponding OPE of CVO, i.e., has the meaning of a particular Liouville or matter fusing matrix
elements in a certain gauge. The parametrization of the Liouville or matter intermediate states
1/2
1/2
(boundary conditions) is taken as zL ( ) = L cos b(Q 2 ) or zM (a) = M cos b(e0
2a), with

b
c,1 = zL (3 ),
c1,0
= zL 3
,
c0, = zL (1 ),
2

b

c,1
(A.35)
= zM (a3 ),
c1,0
= zM a3
= zM (a1 ).
,
c0,
2
Up to the overall leg factor normalization this gives for the constants in the r.h.s.
b
2
CL b ,+ b
3 2
3 1
2
2 sin b(
= L2

CaM b ,e b
3
1
2
= M
b
2
a3
e
a1
b
2
(3 + 1 Q)) sin b(
sin b(Q 2)
b
2
(3 1 ))
b
2
(a3 + a1 e0 )) sin b(e +

sin b(e0 2e)
b
2
(a3 a1 ))
(A.36)
2 sin b(e +
(A.37)
and the case without screening charges (A.33) corresponds to (a product) of trivial constants
b
2
e2
M
L
2
Ca ,e+ b
= 1 = C , b
.
3 1
2
a3 a1
2
2
2
The first of these expressions (A.36) has been derived in [25] combining the formulae in [18]; it
differs by an overall constant from the boundary Liouville constant computed in [5]. The analytic
continuation of the latter is similarly related to the matter constant (A.37); vice versa (A.37) is
obtained from (A.36) via the analytic continuation formula (3.24). Finally let us look at the chiral
18 See [47] for the precise meaning of this statement in the rational case.
328
analog of the double integral matrix element (A.11) with the various possible positions of the two
inserted vertex operators. E.g. for 2 = b we obtain for |z0 | > |z1 |

dz2 dz3 c1 c0 a (z0 )(cV )(z1 )Vb+ (z2 )V0+ (z3 )
C2
C3
b
sin b(Q 2) L
2
C
b
2 ,+ 2
3
sin 2b
( b1 (Q 2))(bQ)(be0 )
b
e
CaM,e b 2
,
2
a
a
2
3
1
2 ( b1 (Q 2 b))
(A.38)
where the ratio of leg factors is extracted in the r.h.s. of the first line.
These formulae will be applied to the boundary tachyon operators at generic values of momenta. Whenever the relation is applied to the left, i.e., with the opposite order of the fields, there
appears an overall minus sign. The four constants, (A.36), (A.37) for plus chirality, and 1 in
(A.33), and the constant in the second line in (A.38) for minus chirality, correspond to the four
OPE coefficients in the bulk identities (4.5). One can write down a formula analogous to (4.9),
expressing any of the four OPE coefficients in terms of a product of the corresponding Liouville
and matter boundary OPE constants.
The formula (A.34) for arbitrary 2 , but keeping one of the terms, provides the simplest
contact term in the boundary counterpart of the recurrence relation (5.12). It remains to compute
the boundary 4-point functions determining the analogs of (4.21), (4.24) and hence of the contact
terms in (5.9). The details will be presented elsewhere.
Appendix B. More 3-point solutions and boundary CFT interpretation
We have encountered several examples of solutions of the ring relations for the 3-point
functions described by the various fusion multiplicities NP1 ,P2 ,Pn . They generate n-point multiplicities which can be cast formally into the general form

NP1 ,P2 ,...,Pn = d(a)
Pi (a),
i

P +b (a) + P b (a) = 2 cosh b P P (a) = fb (a)P (a),
= 1.
(B.1)
The relation in (B.1) is a sufficient condition ensuring the validity of the homogeneous relations (3.27); analogous identity holds in the diagonal case.
This formula is specialized by certain range of the variable a, dual to the spectrum of momenta P , and by some choice of the measures in the two spaces, the characters P (a) and
ia(i e0 Pi ) with
the
pfunction fb (a). E.g., the simplest example (6.16) corresponds to Pi (a) = e
=
p
2,
and
f
(a)
=
2
cos
ba.
Another
explicit
solution
of
the
homogeneous
relations
b
i=1 i
(3.27) is given by a formula dual to (5.5),

e0 2
NP1 ,P2 ,P3 =
P1 (m, n)P2 (m, n)P3 (m, n),
4 sin me0 b sin n
b
m=0 n=0
sin mP b sin nP /b
P (m,n) =
= P (m, n),
sin me0 b sin ne0 /b

fb (am,n ) = 2 cos b nb mb .
(B.2)
329
The characters satisfying the relation in (B.1) coincide up to a normalization with the
P
tachyon disk 1-point functions VP
a L 2b P (a), with a boundary label a,

()
()

()
()
L VP b a + VP +b a = a VP a = L fb (a) VP a .
(B.3)
In the first equality we have used (4.5) (with M set to 1). The second is a version of the standard
bulkboundary equations, yet to be established in this context. It implies [24] that the eigenvalue
of the operator a is identified up to a power of L with the function fb (a) in (B.1) for any
solution NP1 ,P2 ,P3 of the ring relations (3.27); in [24] this reasoning has been used in the rational
case, assuming the validity of the OPE relations (4.5), (4.6).
The solution (B.2) provides in the boundary CFT interpretation an example of boundaries
parametrized by the degenerate matter (or Liouville, as in [6]) representations; the two transformations (3.24) preserve the formula for the characters inverting the sign (m, n) (m, n) or
(m, n) (m, n) in agreement with the c < 1 versus c > 25 parametrizations (4.10) and (4.14).
In this case taking P = e0 and = 1 the first (or the second) term in (B.3) disappears, respec(+)
(+)
tively, so that we have for a = am,n that L Ve0 b
a = a Ve0
a = L a
a . This determines
the 1-point function a
a .
Finally let us mention another symmetric under the change of sign P P solution of
(3.27), represented as in (B.1),
NP1 ,P2 ,P3 4

3
1
i=1 cosh Pi t
dt
t2
sinh bt sinh bt
0

3 2

1
Sb 12 Sb 13 .
log Sb (123 Q)Sb Q 23
1
(B.4)
The formula applies to complex values of the momenta Pi . This 3-point function is similar to
the density (P1 ) which appears in the disk partition function [2,48], with the two boundary
parameters replaced by the two momenta P2 , P3 . It is interpreted as the derivative of the log of a
particular fusion matrix element. The diagonalizing matrix here is a disc bulk 1-point function
cosh(Q 2)t, analogous to the solution in [5].
References
[1] H. Dorn, H.J. Otto, Two and three point functions in Liouville theory, Nucl. Phys. B 429 (1994) 375, hepth/9403141.
[2] A.B. Zamolodchikov, Al.B. Zamolodchikov, Structure constants and conformal bootstrap in Liouville field theory,
Nucl. Phys. B 477 (1996) 577, hep-th/9506136.
[3] J. Teschner, On the Liouville three-point function, Phys. Lett. B 363 (1995) 65, hep-th/9507109.
[4] B. Ponsot, J. Teschner, Liouville bootstrap via harmonic analysis on a noncompact quantum group, hep-th/9911111;
B. Ponsot, J. Teschner, ClebschGordan and RacahWigner coefficients for a continuous series of representations
of Uq (sl(2, R)), Commun. Math. Phys. 224 (2001) 3, math.QA/0007097.
[5] V. Fateev, A.B. Zamolodchikov, Al.B. Zamolodchikov, Boundary Liouville field theory. I: Boundary state and
boundary two-point function, hep-th/0001012.
[6] Al.B. Zamolodchikov, A.B. Zamolodchikov, Liouville field theory on a pseudosphere, hep-th/0101152.
[7] B. Ponsot, J. Teschner, Boundary Liouville field theory: Boundary three point function, Nucl. Phys. B 622 (2002)
309, hep-th/0110244.
[8] K. Hosomichi, Bulk-boundary propagator in Liouville theory on a disc, JHEP 0111 (2001) 044, hep-th/0108093.
[9] P. Ginsparg, G. Moore, Lectures on 2D gravity and 2D string theory (TASI 1992), hep-th/9304011.
[10] P. Di Francesco, P. Ginsparg, J. Zinn-Justin, 2D gravity random matrices, Phys. Rep. 254 (1995) 1, hep-th/9306153.
330
[11] J. Polchinski, What is string theory, Lectures presented at the 1994 Les Houches Summer School Fluctuating
Geometries in Statistical Mechanics and Field Theory, hep-th/9411028.
[12] I. Klebanov, String theory in two-dimensions, hep-th/9108019.
[13] A. Jevicki, Developments in 2d string theory, hep-th/9309115.
[14] E. Witten, Ground ring of two-dimensional string theory, Nucl. Phys. B 373 (1992) 187, hep-th/9108004.
[15] I.R. Klebanov, A.M. Polyakov, Interaction of discrete states in two-dimensional string theory, Mod. Phys. Lett. 6
(1991) 3273, hep-th/9109032.
[16] D. Kutasov, E. Martinec, N. Seiberg, Ground rings and their modules in 2D gravity with c 1 matter, Phys. Lett.
B 276 (1992) 437, hep-th/9111048.
[17] S. Kachru, Quantum rings and recursion relations in 2D quantum gravity, Mod. Phys. Lett. A 7 (1992) 1419, hepth/9201072.
[18] M. Bershadsky, D. Kutasov, Scattering of open and closed strings in (1 + 1)-dimensions, Nucl. Phys. B 382 (1992)
213, hep-th/9204049.
[19] E. Witten, B. Zwiebach, Algebraic structures and differential geometry in 2D string theory, Nucl. Phys. B 377
(1992) 55, hep-th/9201056.
[20] I.R. Klebanov, Ward identities in two-dimensional string theory, Mod. Phys. Lett. A 7 (1992) 723, hep-th/9201005.
[21] E. Verlinde, The master equation of 2D string theory, Nucl. Phys. B 381 (1992) 141, hep-th/920202.
[22] I.R. Klebanov, A. Pasquinucci, Correlation functions from two-dimensional string Ward identities, Nucl. Phys.
B 393 (1993) 261, hep-th/9204052.
[23] I.R. Klebanov, J. Maldacena, N. Seiberg, D-brane decay in two-dimensional string theory, JHEP 0307 (2003) 045,
hep-th/0305159.
[24] N. Seiberg, D. Shih, Branes, rings and matrix models on minimal (super)string theory, JHEP 0402 (2004) 021,
hep-th/0312170.
[25] I.K. Kostov, Boundary ground ring in 2D string theory, Nucl. Phys. B 689 (2004) 3, hep-th/0312301.
[26] D.V. Boulatov, V.A. Kazakov, The Ising model on random planar lattice: The structure of phase transition and the
exact critical exponents, Phys. Lett. B 186 (1987) 379.
[27] Al.B. Zamolodchikov, Talk delivered at the Workshop, Liouville theory and matrix models, RIKEN, Wako, Saitama,
June 2005.
[28] I.K. Kostov, O(N ) vector model on a planar random lattice: Spectrum of anomalous dimensions, Mod. Phys. Lett.
A 4 (1989) 217.
[29] I.K. Kostov, V.B. Petkova, Non-rational 2d quantum gravity: II. Target space CFT, hep-th/0609020.
[30] I.K. Kostov, Strings with discrete target space, Nucl. Phys. B 376 (1992) 539, hep-th/9112059.
[31] S. Higuchi, I.K. Kostov, Feynman rules for string field theories with discrete target space, Phys. Lett. B 357 (1995)
62, hep-th/9506022.
[32] Vl.S. Dotsenko, V.A. Fateev, Four point correlation functions and the operator algebra in the two-dimensional
conformal invariant theories with the central charge c < 1, Nucl. Phys. B 251 (1985) 691.
[33] Al.B. Zamolodchikov, The three-point function in the minimal Liouville gravity, Theor. Math. Phys. 142 (2005)
183;
Al.B. Zamolodchikov, On the three-point function in minimal Liouville gravity, hep-th/0505063.
[34] P. Di Francesco, D. Kutasov, World sheet and space time physics in two-dimensional (super) string theory, Nucl.
Phys. B 375 (1992) 119, hep-th/9109005.
[35] I.K. Kostov, V.B. Petkova, Bulk correlation functions in 2D quantum gravity, Theor. Math. Phys. 146 (1) (2006)
108, hep-th/0505078.
[36] A. Basu, E.J. Martinec, Boundary ground ring in minimal string theory, Phys. Rev. D 72 (2005) 106007, hepth/0509142.
[37] V. Pokrovsky, A. Belavin, Al.B. Zamolodchikov, Moduli integrals, ground ring and four-point function in minimal
Liouville gravity, in: Polyakovs string: Twenty five years after, hep-th/0510214.
[38] F. David, Conformal field theories coupled to 2D gravity in the conformal gauge, Mod. Phys. Lett. A 3 (1988) 1651.
[39] J. Distler, H. Kawai, Conformal field theory and 2D quantum gravity or whos afraid of Joseph Liouville?, Nucl.
Phys. B 321 (1989) 509.
[40] N. Seiberg, Notes on quantum Liouville theory and quantum gravity, Prog. Theor. Phys. Suppl. 102 (1990) 319.
[41] B.H. Lian, G.J. Zuckerman, New selection rules and physical states in 2D gravity: Conformal gauge, Phys. Lett.
B 254 (1991) 417.
[42] V. Schomerus, Rolling tachyons from Liouville theory, JHEP 0311 (2003) 043, hep-th/0306026.
[43] A. Strominger, T. Takayanagi, Correlators in time like bulk Liouville theory, Adv. Theor. Math. Phys. 7 (2003) 369,
hep-th/0303221.
331
[44] Vl. Dotsenko, V. Fateev, Operator algebra of two-dimensional conformal theories with central charge c 1, Phys.
Lett. B 154 (1985) 291.
[45] P. Furlan, A.Ch. Ganchev, V.B. Petkova, Remarks on the quantum group structure of the rational c < 1 conformal
theories, Int. J. Mod. Phys. A 6 (1991) 4859.
[46] C. Imbimbo, S. Mahpatra, S. Mukhi, Construction of physical states of non-trivial ghost number in c < 1 string
theory, Nucl. Phys. B 375 (1992) 399.
[47] V.B. Petkova, J.-B. Zuber, The many faces of Ocneanu cells, Nucl. Phys. B 603 (2001) 449, hep-th/0101151.
[48] J. Teschner, Remarks on Liouville theory with boundary, hep-th/0009138.
Boundary field theory approach to the renormalization

of SQUID devices
Domenico Giuliano a, , Pasquale Sodano b
a Dipartimento di Fisica, Universit della Calabria and I.N.F.N., Gruppo collegato di Cosenza,
Arcavacata di Rende I-87036, Cosenza, Italy

b Progetto Lagrange, Fondazione C.R.T. and Fondazione I.S.I., Dipartimento di Fisica, Politecnico di Torino,
Corso Duca degli Abruzzi 24, Torino, Italy 1
Received 5 August 2006; received in revised form 13 December 2006; accepted 26 February 2007
Abstract
We show that the quantum properties of some Josephson SQUID devices are described by a boundary
sine-Gordon model. Our approach naturally describes multi-junction SQUID devices and, when applied to
a single junction SQUID (the rf-SQUID), it reproduces the known results of Glazman and Hekking. We provide a detailed analysis of the regimes accessible to an rf-SQUID and to a two-Josephson junction SQUID
device (the dc-SQUID). We then compute the normal component of the current-response of a SQUID device to an externally applied voltage and show that the equation describing the current-voltage characteristic
function reduces to well-known results when the infrared cutoff is suitably chosen. Our approach helps in
establishing new and interesting connections between superconducting devices, quantum Brownian motion,
fermionic quantum wires and, more generally, quantum impurity problems.
PACS: 03.70.+k; 11.25.Hf; 85.25.Dq; 85.25.Cp
Keywords: Boundary conformal field theories; Superconducting quantum interference devices (SQUIDs)
E-mail address: giuliano@fis.unical.it (D. Giuliano).

1 Permanent address: Dipartimento di Fisica, Universit di Perugia and I.N.F.N., Sezione di Perugia, Via A. Pascoli,
06123 Perugia, Italy.

doi:10.1016/j.nuclphysb.2007.02.015
D. Giuliano, P. Sodano / Nuclear Physics B 770 [FS] (2007) 332370
333
1. Introduction
Quantum effects in Josephson junctions have been by now investigated for quite a long time,
both experimentally [1] and theoretically [2]: the most important manifestation of quantum fluctuations in these systems being the macroscopic quantum tunneling of the phase across a current
biased junction and the consequent observation of quantum phase slips events at a bias rather
close to the Josephson critical current [1].
Early studies of quantum phase fluctuations in Josephson systems have been carried using
either the CaldeiraLeggett [3] or the electromagnetic environment [4] models: within this framework, it has been shown that macroscopic quantum tunneling causes a finite voltage to appear
for any finite current, leading to a nonlinear IV characteristic which, in the limit of zero temperature, reads as V = AI with depending on the impedance of the superconducting leads
and A being proportional to the bare tunnel matrix element between two adjacent minima of the
potential [5]. The dual result for a voltage biased junction shows that the dc-current in such a
junction is proportional to the square of the unrenormalized Josephson energy [6].
As already pointed out in Ref. [7], in the above mentioned approaches the effective boundary conditions for the quantum fluctuations of the environment modes do not depend on the
Josephson energy of the junction. While this assumption is perfectly legitimate for weak fluctuations of the phase of the order parameter across the junction, better care should be used if
these fluctuations are strong as it may well happen for one-dimensional superconductors, where
phase fluctuations diverge logarithmically with the length of the system, opening the undesirable
option that the Josephson energy could average to zero for diverging random fluctuations of the
phase.
In their seminal paper Glazman and Hekking [7] showed that a finite renormalized Josephson
energy may arise since the junction itself affects the fluctuations of the environment. They considered an rf-SQUID, i.e. a thin superconducting loop containing a Josephson junction, phase-biased
by threading a flux ext through the loop. The environment modes of the loop consist of plasmon
modes propagating through a superconducting wire with a sound-like dispersion law [8]. Their
analysis was based on a perturbative renormalization group approach exploiting the similarity
of the effective action, emerging from integrating out the fluctuations away from the junction,
with the one used to analyze the quantum Brownian motion in a periodic potential [9]. They
found how the Josephson coupling EJ is renormalized when high energy degrees of freedom are
integrated out and showed that the renormalization leads to values of the Josephson critical current which are smaller than one would expect from the mean field theory result Jc0 = G/2e,
where G is the conductance of the junction and is the superconducting gap in the loop. The
RG approach is supplemented by an instanton analysis [10] needed to compute the effect of the
macroscopic quantum tunneling on the phase dependence of the Josephson current for relatively
large loops.
Generalized SQUID devices are much studied today mainly because of their relevance for
the implementation of flux qubits [11]: it is by now widely accepted that multi-junction SQUID
devices [12] are the most promising realization of a flux qubit due to their reduced size, which
renders them less sensitive to decoherence effects [13] induced by the interaction with the environment. As it is well known, an rf-SQUIDs [14] may be represented as a two level system
only when the superconducting loop is rather large and, thus, unfortunately more sensitive to
interactions with the environment.
In this paper we evidence that all the non-perturbative renormalization effects induced by
quantum fluctuations of the phase of the superconducting order parameter [7] are described by
334
the two-boundary sine-Gordon model [15], which becomes then the pertinent description of the
dynamics of the fluctuation field and of a Josephson junction with arbitrary nominal strength EJ .
As we shall see, the quantum behaviors of rf-SQUIDs and of multi-junction SQUID devices,
may be analyzed within the framework of boundary field theories [16].
A two boundary sine-Gordon model description of SQUID devices is possible since the dynamics of the fluctuations of the phase of the superconducting wire may be described as a spinless
TomonagaLuttinger liquid (TLL) [17]. In fact, for a superconducting wire of length 2L and
cross-sectional area S, the Lagrangian describing the phase fluctuations of the order parameter
may be written as [7]
L
L=
dx

h 2 2 h 2 ns S 2
,
2ec t
4m
x
(1)
where ns is the superfluid density, m is the electron mass, 1/ec is the characteristic inverse
charging energy per unit length of the loop:
1

,
= 2
ec
8e ln(R/a)
(2)
where is the dielectric constant of the medium surrounding the loop, while R is the distance
from a metallic screen, which, together with a, fixes the leads capacitance.
Upon setting h = 1,
if one defines the Luttinger parameter g = 2 ns S/(mec ) and the
plasmon velocity vpl = ns Sec /(4m), one immediately sees that the Lagrangian (1) may be
written as the one describing a spinless Luttinger liquid [18]2 :
g
L=
2
L
dx
L

1 2
2
vpl
.
vpl t
x
(3)
Physically, 1/g is the dimensionless zero-frequency impedance of the superconducting wire [7,
20]. Due to its dependence on S, g may take values in a rather wide interval of real numbers: for
g > 1, Eq. (3) describes a TLL with an attractive interaction, while, for g < 1 the interaction is
repulsive. The value g = 1 describes the very special case of an essentially free theory. Different
superconducting wires may realize different values of g: for instance, for rather clean aluminum
wires, one can easily attain values of g such that g > 1 and, thus, realize a wire for which the
dynamics of the phase fluctuations is described by an attractive TLL [7].
For g > 1, the two boundary sine-Gordon model description of a SQUID device allows to
make very general statements regarding the regimes accessible to the system for a finite loopsize L. Namely, there will be a perturbative (in EJ ) weak coupling regime accessible when the
relevant Josephson couplings are small and a non-perturbative strong coupling regime accessible
when EJ becomes large; most importantly, there will be a renormalization group invariant length
scale L such that for L < L the SQUID device is in the perturbative weak coupling regime,
while it is in the non-perturbative strong coupling regime for L > L . Intuitively speaking, L is
(2 times) the radius for which the ratio between the Josephson energy and the magnetic energy
due to the loop self-inductance equals one; the existence of an healing length L is generic
2 Our normalization of g renders the boundary interaction marginal when g = 1. It differs by the one used in Ref. [19]
by a factor of 1/2.
335
to a boundary sine-Gordon model describing a TLL [21]. At variance, for g < 1, the theory lies
within the perturbative regime at any L. Therefore, there is no healing length in this case and
no universal behavior is then attainable. In the following, we shall show that these properties are
shared by a variety of SQUID devices.
We investigate also the stability of the two (weakly and strongly coupled) regimes by looking
at the relevance/irrelevance of the operators describing the boundary interactions. It turns out
that these operators are either vertex operators or instantons (phase slips) of the plasmon field ,
depending on if the system is within the weakly, or the strongly coupled regime. As evidenced
in the following, the vertex operators are a relevant perturbation of the weakly coupled regime
for g > 1, while they are irrelevant for g < 1. On the other hand, the instantons are an irrelevant
perturbation of the strongly coupled regime for g > 1, while they are relevant for g < 1.
It emerges the picture that, for g > 1, it always exists an healing length L such that, for
L > L , the Josephson coupling affects the dynamics of the plasmon field as a boundary condition: in this regime it is impossible to disentangle the Josephson junction from the environmental
modes. For g < 1, instead, this situation is never attainable and the Josephson coupling has to be
always regarded as a quantum impurity embedded in a superconducting wire.
The strongly and weakly coupled regimes may be probed by looking either at the dcconductance, or at the behavior of the Josephson current as a function of the magnetic flux
treading the superconducting loop. The two-boundary sine-Gordon approach provides a systematic and powerful method to evaluate the relevant currents. In fact, the correlators of the plasmon
fluctuation field(s) may be exactly computed and this allows for a systematic derivation of
the leading corrections to the Josephson current for all values of the applied flux = ext /0
(0 = h/(2e)), which, for = , yields the results of the instanton analysis carried in Ref. [7];
in addition, we shall show that it also allows for a systematic derivation of the response of a
SQUID device to an applied external voltage. Since applying a finite voltage V introduces a
new energy scale e V , with e = 2e, the dc-current response of the system strongly depends
on the ratio between the new energy scale, and the intrinsic scale provided by the level spacing
s = 2vpl /L of the plasmon field(s) . Our computation of the dc-current clearly shows that,
when e V > s, one gets the well known KaneFisher formula [9], while, for s > e V , one gets
the results obtained by Glazman and Hekking [7]; in addition, when e V > s, the dc-current
dominates over the Josephson current.
Boundary field theories have by now become relevant in several different contexts. In condensed matter theory, they are mostly generalizations of quantum impurity models, which may
be described by using the TLL paradigm [17]; for instance, boundary interactions appear in the
analysis of the Kondo problem [22], in the study of a one-dimensional conductor in presence
of an impurity [23], in the derivation of tunneling between edge states of a Hall bar [24], in
the study of quantum Brownian motion in a periodic potential [25] and in the analysis of the
phases accessible to networks of quantum wires [19] and Josephson junctions [21,26]. The TLL
paradigm evidences that many interactions are simply diagonalizable in the basis of pertinent
collective bosonic modes, and that non-diagonalizable interactions correspond to exactly solvable Hamiltonians, such as sine-Gordon models [27]. Recently, boundary field theories have been
investigated in the context of string theories. For instance, in studying tachyon instabilities [28],
one is faced with the fact that the space of interacting string theory [29] is mapped onto the space
of boundary perturbations of conformal field theories [30], and that the renormalization group
flow determined by boundary perturbations may be identified with tachyon condensation [31].
Affleck and Ludwig [32] showed that the boundary entropy g is decreasing along the renor-
336
malization group trajectories, triggered by the boundary interactions. It is remarkable that this
entropy has been proposed as a measure of entanglement for boundary states [32].
The paper is organized as follows. Section 2 shows that quantum behaviors of an rf-SQUID
are described by a two-boundary sine-Gordon model. There, we shall analyze the weak and the
strong coupling regime accessible to the device for a finite size of the superconducting loop.
We show that, for g > 1, there is a renormalization group invariant scalethe healing length
L [7] marking the crossover between the weakly coupled and the strongly coupled regime.
We first analyze the weakly coupled fixed point corresponding to mixed boundary conditions
(i.e. Dirichlet at the inner boundary and Neumann at the outer boundary) and we determine the
scale dimension of the pertinent boundary operators and the scaling equations for the running
coupling constants using renormalization group (RG) methods; this allows for a perturbative
computation of the beta-function for all the couplings. Computation of the canonical ground
state energy shows thatat the weakly coupled fixed pointthe Josephson current has a sine
dependence on the flux treading the loop. Then, we analyze the strong coupling fixed point, corresponding to Dirichlet boundary conditions at the outer and inner boundaries of the rf-SQUID.
Evaluation of the canonical ground state energy shows thatat the strongly coupled Dirichlet
fixed pointthe Josephson current exhibits a saw-tooth dependence on the external flux.
Due to the zero modes, appearing in the strong coupling regime as a consequence of the
Dirichlet boundary conditions, the ground state is doubly degenerate when = + 2k: when
these levels are confined away from the rest of the spectrum, the rf-SQUID realizes a qubit (see,
for instance, Ref. [11]). Our analysis points out that this realization is possible only for g > 1
and for sizes of the superconducting loop bigger than the healing length, and, thus, for a rather
big self-inductance of the device. Finally, we analyze the effects of quantum phase slips on the
ground state degeneracy of the rf-SQUID at the strong coupling fixed point: an explicit instanton
computation shows that, for g > 1, the degeneracy is lifted by an amount proportional to the
instanton fugacity, and that the instantons smoothen the edges of the saw-tooth phase-current
relation.
In Section 3 we analyze a SQUID-device involving only two Josephson junctions, of arbitrary
strength EJ,1 , EJ,2 (the dc-SQUID). Under the assumption that the two plasmon fields living in
the two branches of the superconducting loop (see Fig. 1) are described by TLLs corresponding
to the same g, we show that also the dc-SQUID is described by a two-fields generalization of the
two boundary sine-Gordon model. We show that also here, for g > 1, there is an unstable weak
coupling fixed point and a stable one at strong coupling and that the crossover from one to the
other is characterized by an healing length L , whose value depends, of course, only on the smallest of the two Josephson couplings. When g < 1, the only stable fixed point is at weak coupling.
In Section 4 we analyze the normal component of the current-response of the rf- and dcSQUIDs to an applied external voltage. There, we evidence that a computation of the timeindependent current using the -field(s) correlators derived from the two-boundary sine-Gordon
model yields the well-known KaneFisher formula when the infrared cutoff is e V > s [9], as
well as the GlazmanHekking result [7], when the infrared cutoff is s > e V . It is remarkable
that, as a result of the boundary field theory approach to superconducting SQUID devices developed in this paper, one gets a current-voltage equation valid for any choice of the infrared cutoff.
Furthermore, for a dc-SQUID, the boundary field theory approach evidences the possibility of
new remarkable interference effects (even in the absence of an external applied magnetic flux (!))
between the currents flowing across the two junctions.
Section 5 is devoted to our conclusions and final remarks.
337
In order to be pedagogical and self-contained, a few pertinent mathematical details are summarized in Appendices A and B.
2. The rf-SQUID
From Eq. (3) one may easily obtain the Hamiltonian Hrf for an rf-SQUID (i.e. for a superconducting wire cut by a Josephson junction with nominal Josephson coupling EJ and pierced by a
magnetic flux ). By neglecting the charging energy of the junction [7], Hrf is given by
Hrf H0 + HB

L

g
1 (x) 2
2
(x)
=
dx
+ vpl
EJ cos (L) (L) ,

2
vpl
t
x
2L
L
(4)
where H0 and HB describe the bulk and boundary Hamiltonian, respectively.
x
The flux may be accounted for by a redefinition of the field as 2L
. By
1
introducing the symmetric and the antisymmetric phase fields e (x) = [(x) + (x)] and
o (x) =
1 [(x) (x)],
2
with 0 x L, one gets
L

g

j (x) 2
1 j (x) 2
dx
+ vpl
Hrf =
EJ cos 2o (L) + .
2
vpl
t
x
j =e,o
(5)
The field e (x) fully decouples from the interaction term in Eq. (5). The Hamiltonian for the
field o (x) (which will be referred to as (x), from now on), on the other hand, is the Hamiltonian for a two-boundary sine-Gordon model [15], with Dirichlet boundary condition at the
inner boundary, i.e., (0, t) = 0, and dynamical boundary conditions at the outer boundary. The
latter are obtained by requiring that the energy functional is conserved, yielding [21]

gvpl (L, t)
+ 2EJ sin 2(L, t) + = 0.
(6)
x
For EJ 0, one has Neumann boundary conditions at the inner boundary, i.e., (L,t)
= 0,
x
while, for EJ , one gets Dirichlet boundary conditions, 2(L, t) = + 2k, k Z.

As it will be clearer in the following, the effective two-boundary sine-Gordon model provides
the relevant renormalizations of an rf-SQUID: namely, it gives the renormalized value (E J ) of
the Josephson energy and yields the correct functional dependence of the Josephson current, as
a function of the flux .
As we shall see in the following sections, as L , there are two relevant fixed points,
namely, the weakly coupled Neumann fixed point (EJ = 0), and the strongly coupled Dirichlet fixed point (EJ = ). Our analysis will evidence the existence of an healing length L =
gvpl /EJ , providing the size of a loop for which the ratio between the Josephson energy and
the magnetic energy due to the loop self-inductance equals 1, such that, for g > 1, at L = L the
device crosses over from the weakly to the strongly coupled regime.
2.1. The weakly coupled fixed point
At the weakly coupled fixed point (EJ 0), the field (x, t) obeys Dirichlet boundary
conditions at the inner boundary and Neumann boundary conditions at the outer boundary. As
338
a consequence, (x, t) takes the mode expansion

1 (n)
(x, t) =
sin(kn x)eikn ut ,
g
n
(7)
n=0
with [(n), (n )] = nn+n ,0 , and kn = L (n + 12 ).

The ground state of the bosonic theory, |GS
, is defined in terms of the oscillatory modes as
(n)|GS
= 0,
n > 0.
(8)
Normal ordering the Josephson energy with respect to |GS

, leads to
HJ =
EJ
2
2a
L
1
g
:ei
2(L)+i
: + :ei
2(L)i
:,
(9)
where the column : : denotes, as usual, normal ordering.

The thermodynamic limit (i.e., large L) is attained once the scaling equations for the pertinent running couplings have been obtained using the renormalization group (RG) approach (see
Appendix A).
The RG equations are determined by the requirement that the partition function Z is independent of the cutoff. In order to compute Z, one should Wick-rotate the field and resort to the
imaginary-time Feynman path-integral formalism. The partition function is, then, given by

Z = D eS0 T eSB ,
(10)
where T is the imaginary time ordering operator, S0 is the free Euclidean action
g
S0 =
2
L
d

1 2
2
dx
+ vpl
,
vpl
x
(11)
and SB is the boundary interaction action at imaginary times

E J
SB =
d :ei[ 2(i )+] : + :ei[ 2(i )+] : ,
2
(12)
with (i ) = (L, i ).
1
In Eq. (12), the renormalized coupling constant E J = EJ (2a/L) g has been introduced. Of
course, after this redefinition of the Josephson coupling, there is not anymore a cutoff dependence
of the interaction operator [33].
The
boundary action in Eq. (12) contains only the vertex operators V1 ( ) = ei
:exp[i 2(i )]:. Due to the mixed boundary conditions on , the operator (i ) contains
no zero modes and, thus, there are no selection rules on the KacMoody charge carried by the
vertex operators. As a consequence, the Operator Product Expansions (O.P.E.s) may generate
additional vertices, with higher periodicity in , given by

Vn ( ) = eni :exp i 2n(i ) :.

(13)
339
The scaling dimension of Vn , hn , is derived from the correlation function T [Vn ( )Vn ( )]
which, for , is approximatively given by (see Appendix B)

2n2
g

vpl

.
( )
T Vn ( )Vn ( ) tanh
4L

(14)
At short imaginary time distances (vpl | |/L 1), Eq. (14) becomes

2n2
g
vpl

,
| |
T Vn ( )Vn ( )
L
(15)
leading to hn = n2 /g.
As we shall see in the following, despite the fact that hn depends on n, due to nonlinearities,
the higher harmonics in Eq. (13) are irrelevant operators, when g < 1, while they become relevant
for g > 1.
The O.P.E. yields the pertinent fusion rules for the vertex operators
hn hn +hn+n

vpl
Vn ( )Vn ( )
(16)
Vn+n ( ).
( )
L
Introducing the pertinent counterterms, one may easily show that the effective boundary action
looks as
1
SB,Eff =
2

E n ein Vn ( ) + E n ein Vn ( ) ,
d
(17)
n=1
n2
where the renormalization condition is that all the coupling strengths E n = (2a/L) g En , with
n = 1, vanish at a given reference length scale L0 .
Setting the running couplings gn as

gn =
L
2a
1 n2
g
a n En
n ,
vpl
(18)
from the fusion rules given in Eq. (16), one is able to derive, in principle, the -function for all
the couplings gn .
For the sake of simplicity, we shall restrict our attention to the first two relevant terms, proportional to the running couplings g1 , g2 . The RG equations are then

dg1
1
(19)
= 1 (g1 , g2 ) = 1
g 1 + g1 g 2 ,
d ln(L/L0 )
g
and

dg2
4
= 2 (g1 , g2 ) = 1
g2 + (g1 )2 ,
d ln(a/a0 )
g
(20)
the flux fully decouples from Eqs. (19), (20) and, thus, the scaling near the Neumann fixed
point is independent of .
For g < 1, g1 and g2 are both irrelevant. Thus, the Neumann fixed point is stable and the
theory is perturbative in the boundary interaction with effective, size-dependent couplings given
340
by
(1 1 )
g
L
,
g2 (L) 0.
g1 (L) g1 (L0 )
(21)
L0
On the other hand, when g > 1, g1 becomes relevant since it grows with L as
(1 1 )
g
L
.
g1 (L) g1 (L0 )
(22)
L0
By inserting Eq. (22) into the nonlinear term of Eq. (20), one gets

1 2
g
(g 2 (L))
L
g2 (L) 1
(23)
1
,
1 + 2/g
L0
where the integration constants have been chosen so as to be consistent with the renormalization
condition g2 (L0 ) = 0.
From Eq. (23), one sees that the nonlinear term in Eq. (19) makes g2 (L) increase, as soon as
one starts the scaling flow, even when 1 4/g < 0. This is a remarkable result, since it implies
that, as soon as the first harmonics in sets in, all the others follow, due to nonlinearities in the
RG equations.
At the Neumann fixed point, the Josephson current is perturbative in E J : of course, as L
increases, our results are reliable only for g < 1. As a function of the flux , I [], is given by

1
1
ZEff []
I [] =
(24)
lim
ln
,
c0
Z0
where ZEff [] is the -dependent partition function, given by

ZEff [] = D eS0 T eSB,Eff .
(25)
To the second order in gn , one gets

1

g2 (L)
2a g
I [] = e EJ
(26)
sin(2) ,
sin() +
L
g1 (L)
which explicitly gives the contribution of the second harmonics to the Josephson current.
For g > 1, Eq. (23) shows that, for L , the ratio between the running coupling constants
of the first two harmonics is given by
1 1
g
g2 (L)
L
(27)
,
g1 (L)
L0
showing the relevance of the boundary interaction, since higher harmonics become more and
more relevant, as L . The scaling equations (18)(21) cease to be valid at the renormalized
healing length L for which g1 (L ) 1, i.e., when

g
vpl g1
,
L = L = 2a
(28)
aEJ
yielding that the renormalized Josephson coupling E = E J (L ) is given by
J
E J =
aEJ
vpl
1
g1
EJ ,
(29)
341
which is the energy where the level spacing of the plasmon field becomes of the order of the
1
Josephson energy, i.e., 2vpl /L (2a/L ) g EJ [7].
2.2. The strongly coupled fixed point
In this subsection, we analyze the properties of the rf-SQUID in the limit in which
E J /(2vpl /L) is 1: HJ is now the dominating potential term, and the field (L, t) takes
values corresponding to a minimum of the boundary energy. Therefore, at both boundaries (x)
must obey to Dirichlet boundary conditions, given by
(0, t) = 0,
(30)
2(L, t) = + 2k (k Z).
Both boundary conditions are satisfied if one chooses for (x, t) the mode expansion

1 (n)
2x
P
sin(kn x)eikn vpl t ,
(x, t) =
L
g
n
(31)
n=0
where P is the zero mode operator and kn = ( L n). From Eq. (30), one obtains the set of possible
eigenvalues of P :

1
Pk =
(32)
+k .
2
2
As a consequence, there will be an infinite set of ground states of the oscillatory modes, corresponding to the possible eigenvalues of P , and denoted by |Pk , {0}
.
Once Hrf has been evaluated for the particular solution given in Eq. (31), one obtains the
effective Hamiltonian at the Dirichlet fixed point, HD ,

vpl
2
(n)(n) ,
2gP + 2
HD =
(33)
L
n=1
from which one gets that the partition function, at the strongly coupled fixed point

2
H
gvpl
1
D
=
exp
+k
,
ZD [] = Tr e
(q)
L
2
(34)
kZ

vpl
n
with (x) =
n=1 (1 x ), and q = exp[ L ], leading to the well-known saw-tooth-like
dependence on of the Josephson current as [7], since
1 ln ZD []
[],

I [] = lim
(35)
of course, [] is the integer part of (in units of 2 ).

At strong coupling, the degeneracy among the minima of the boundary term is removed by
gv
(0)
+ k)2 , selecting only one eigenvalue of the zero mode

the magnetic energy, Ek = L pl ( 2
(0)
(0)
of the field (x, t), except when = + 2k, where Ek = Ek+1 . When these two levels are
confined away from the others, the rf-SQUID may operate as a qubit [11]. For L > L , however,
the oscillatory modes of the plasmon field determine a renormalization of the parameters of the
effective two-level system, as it will be discussed in the following.
342
2.3. Renormalization of the Josephson current for the rf-SQUIDs

(0)
At strong coupling and for = , the ground state is twofold degenerate, that is, E0 =
(0)
E1 . This degeneracy is removed by phase slips (instantons), of amplitude 2 , connecting the
ground states.
In order to describe phase slips within the framework of the two-boundary sine-Gordon model,
one has to introduce the dual field of (x, t), (x, t), defined by
(x, t) vpl (x, t)
=
,
x
2g
t
2g (x, t) (x, t)
=
.
vpl
t
x
(36)
Due to the duality between and , the effective Hamiltonian given in Eq. (33) may be usefully
rewritten as
1
HD =
4(2g)
L
0

1 2
2
dx
+ vpl
.
vpl t
x
From the algebra of the bosonic fields, one easily derives [19] the commutation relations

(x, t)
(x, t)
, (y, t) =
, (y, t) = 2i(x y).
x
x
(37)
(38)
In order to construct the leading correction to the Dirichlet fixed point, one may use the Delayed
Evaluation of Boundary Conditions (DEBC) approach, introduced in Ref. [19]: for this purpose,
one has first to consider the most general primary field of the bulk theory, namely, a vertex operator of the form Vn,n (x, t) = :exp[i n (x, t) + i n (x, t)]:, and then to obtain the boundary
2
perturbation, Vn,(B)
n (t), by evaluating it at x = L, using the pertinent boundary conditions for the
fields (x, t) and (x, t). Since (x, t) obeys to the Dirichlet boundary conditions at x = 0 and
(B)
at x = L, it will not contribute to Vn,n (t). At variance, (x, t), obeys to the Neumann boundary
conditions at both boundaries; namely, one has that
(0, t) (L, t)
=
=0
x
x
implying the following mode expansion for (x, t),
(39)

2ut
n
(n)
(x, t) = 0 +
2gP + 2i g
cos
x eikn t ,
L
n
L
(40)
n=0
with 0 = (qR qL )/ 2 and kn = n/L.

As a consequence of Eq. (40), one finds
(i ) = (L, i ) = 0 +
n
2iu
(n)
2gP + i2 g
(1)n e L
L
n
(41)
n=0
and thus, the generic boundary perturbation at the strongly coupled fixed point, H B , may be
written as a linear combination of the vertices Vn = V0,(B)
n as
H B =
n :e
i n (L)
2
:.
(42)
343
At the Dirichlet fixed point, the partition function Z D [], including the boundary interaction in
Eq. (42), is written as

D [] = Tr e
Z

HD
T exp
d n :e
i n (i )
2

:
(43)
n =0 0
with (i ) = (L, i ).
Eq. (43) reduces to Eq. (34) when all the n s are set to zero.
Following the same steps used in Section 2.1, one may derive the scaling dimension of Vn ,
hn , from the large-L-limit of the two point correlation function; namely,

2g n 2
vpl

,
| |
T Vn (i )Vn (i ) (0)
L

(44)
where . . .
(0) denotes thermal averaging defined by the free Hamiltonian with Dirichlet boundary conditions for (x, t) at both boundaries.
Eq. (44) implies that hn = g n 2 and that the leading perturbation at the strongly coupled fixed
point is given by
i (L)
i (L)
2 : .
HB = Y :e 2 : + :e
(45)
The physical meaning of the operators Vn may be inferred from the commutation relation

P , :e
i n(i
1
: =
2
L
dx
0

(x, i ) i n (t)
n i n (i )
: = :e 2
:,
, :e 2
x
2
(46)
i (i )
2. Thus, the boundary fields :e 2
:
which shows that Vn changes the eigenvalue of P by n/
describe instanton/antiinstanton trajectories between the two ground states that are degenerate at
= and the parameter Y may be interpreted as the instanton fugacity.
A dimensionless running coupling YInst (L) is defined as
YInst (L) = Y L1g .
(47)
From Eq. (47), one sees that, for g < 1 (i.e., when the Neumann fixed point is stable), instantons
are a relevant perturbation, since YInst (L) scales as ( LL )1g . However, the scaling ceases to be
valid as YInst (L) 1.
For g > 1, using the boundary perturbation operator introduced in Eq. (45), it is straightforward to compute the leading corrections to the Josephson current. For instance, when 0, one
may approximate the partition function in Eq. (43) as

gvpl 2

exp[ 4L
]
2
D []
Z
d1 d2
P0 , {0} 1 + 2Y
n
n>0 [1 q ]
0

1
1
T :cos (i1 ) ::cos (i2 ) : P0 , {0} .
2
2
(48)
344
By inserting in Eq. (48) the explicit form of the correlator

(i1 )
(i2 )
P0 , {0}T :cos
::cos
: P0 , {0}
2
2
=
gvpl
2L (1 2 )]
,
vpl
[2 sinh[ 2L |1 2 |]]2g
cosh[
(49)
one gets the partition function Z[],

gvpl 2
exp[ 4L
]
L 2
Y [1 2g]
1+
ZD []
n
vpl
n>0 [1 q ]

[g(1 2
[g(1 + 2
)]
)]
+
.
[1 g(1 + 2
)] [1 g(1 2
)]
(50)
Computing the logarithmic derivative of Eq. (50) with respect to and dividing it by , one
obtains

[g(1 2
[g(1 + 2
)]
)]
e gvpl
2L [1 2g] 2
Y
e
+
I []
2L
2vpl
[1 g(1 + 2
)] [1 g(1 2
)]

e gvpl
(51)
Y 2 G[g] ,
2L
with
2L [1 2g]
G[g] = e
4 2 vpl

[g(1 + 2
[g(1 2
)]
)]
2
.
+
)] [1 g(1 2
)] =0
2 [1 g(1 + 2
(52)
For small , Y , HB provides a small correction to the slope of the Josephson current. As we shall
see in detail in the following subsection, for = and g > 1, instantons at the Dirichlet fixed
point remove the degeneracy between the minima and smooth the edges in the saw-tooth-like
Josephson current. For g < 1, of course, instantons are irrelevant operators.
2.4. Instanton contribution to the Josephson current at the Dirichlet fixed point for =
In the previous section, it has been shown that, for = , the state |P0 , {0}
becomes degenerate with |P1 , {0}
. This corresponds to a discontinuous jump of e gvpl /L in the Josephson
current, between the values corresponding to the states that are degenerate at = . Since instantons connecting the two states should remove the degeneracy, for Y = 0 one should expect
that the Josephson current becomes a continuous function of at = .
In order to compute the Y -dependent corrections to the ground state energy, it is useful to
partition the Hilbert space into subspaces on which HB is diagonal. If |Pk ,
denotes a generic
state in the sector corresponding to the eigenvalue Pk of the zero-mode operator, one defines the
states belonging to the above subspaces as
1
|,
= |P0 ,
|P1 ,
.
2
Thus, the partition function at = may be approximated as
345

ZD [] = Tr|,
eHD T exp
d HB (i )
0

+ Tr|+,
eHD T exp

d H B (i )
(53)
where Tr|,
denotes tracing over the sector of the Fock space spanned by the states |,
.
At low temperature, Eq. (53) may be approximated as

H

D T exp
D [] , {0}e
Z
d H B (i ) , {0}

H

D T exp
d H B (i ) +, {0} ,
+ +, {0}e
(54)

where |, {0}
are, of course, the ground states of both subspaces. By expanding T exp[ 0 d
H B (i )] in a power series of Y , one gets

H

D T exp
d H B (i ) , {0}
, {0}e

gvpl 1
= exp
d1 dm , {0}T H B (i1 ) H B (im ) , {0} .
4L
m!
m=0
0
0
(55)
When computing the partition function, one has to sum over contributions from connected, as
well as from disconnected diagrams. By applying Wicks theorem, one obtains

, {0}T H B (i1 ) H B (im ) , {0}
=
M

, {0}H B (i1 ), {0}
j =1

, {0}T H B (ij1 )H B (ij2 ) , {0}
, {0}H B (ii ), {0}
j1 =j2 =1
+ .
i=j1 ,j2
(56)
Since , GS|H B (i )|, GS

= Y , it is straightforward to evaluate, using Eqs. (55), (56), the
leading Y -dependent contribution to the energy of the states |, {0}
; namely E is given by

[ g2 ]
[ 32 g]
gvpl
L 2
E =
(57)
,
Y [1 2g]
Y
+
g
4L
vpl
[1 32 g] [1 2 ]
and, thus, even to the first order in Y , instantons remove the degeneracy between the ground
states.
346
The observable consequence of the removal of the ground state degeneracy is the smoothing
down of the Josephson current, which becomes a continuous function of the applied flux when
it crosses the value = .
In order to see this effect, one has to resum over instanton contributions to the partition function, by setting = + , with / 1. In order to obtain a meaningful result, one has first to
sum over the instanton contributions, and then to compute the leading corrections in .
When = + , the low temperature partition function may be approximated as

Tr|a,
eHD T exp S[Y, ] ,
Z D [ + ] =
(58)
a=
with

S[Y, ] = 2Y
0

gvpl
(i )
g 2
d cos
P +
.
+
L
4L
2
(59)
When = 0, one obtains the result of the previous section. When computing the leading corrections in , one has to notice that P mixes the two subspaces labeled by ; namely
1
, |P |,
= .
2 2
(60)
D [], appears only to the order 2 , and

Thus, the leading -dependent contribution to Eq. (59), Z
is given by

gvpl 2
(m + 1)(m + 2) gvpl 2

D []e 4L
Z D [] = Z
(m + 2)!
2L
a=
m=0
m

a, {0}(P )2 a, {0}

di a, {0} H B (ij ) a, {0}

i=1 0

gvpl 2 2
(m + 1)(m + 2)
m+1
m
1 + (1)
(Y )
(m + 2)!
2L
2Y
m=0

gvpl 2 2
D []
=Z
(61)
exp a Y +
.
2L
2Y
a=
gvpl
D []e 4L 2
=Z
To the leading order in , the ground state energy derived from Eq. (61) is given by

gvpl 2 2
gvpl 2
+
.
EGS [] = EGS [ = 0] Y
2L
2Y
4L
(62)
Eq.
(62) may be regarded as the lowest order term in the expansion of the function
gv
Y 2 + ( 2Lpl )2 . Although our approach allows to compute, in principle, also higher-order
corrections, for our purposes it is enough to notice that, already to the leading order in , one
gets that the current behaves as a continuous straight line, with no sharp, discontinuous, jumps,
at = 0 (i.e., = ).

gvpl
gvpl 2
EGS []
I =e
(63)

.

2L
2L
Y
347
Eq. (63) is not only consistent with the results of previous analyses of rf-SQUIDS [7,34], but also
with a strong coupling expansion of the ground state energy of the two-boundary sine-Gordon
model derived in Ref. [15] using the Bethe ansatz approach, once one of the two couplings is
set to , and a strong coupling expansion in inverse powers of the other coupling has been
performed. This makes us confident that the smoothing effect of the Josephson current due to
instantonsexactly computed in Ref. [15] for g = 1is effective for any value of the Luttinger
parameter g.
3. The dc-SQUID
A dc-SQUID is realized as a superconducting loop, interrupted by two Josephson junctions, of
nominal values EJ,1 and EJ,2 . It is possible to generalize to the dc-SQUID the renormalization
group analysis developed so far for the rf-SQUID, obtaining that, also in this case, the only
alleged fixed points are a weakly coupled (Neumann), and a strongly coupled (Dirichlet) one.
The device is drawn in Fig. 1: it is made out of two superconducting wires, connected by two
Josephson junctions, of nominal values EJ,1 , EJ,2 , with a magnetic flux piercing the loop.
1 (x), 2 (x) are the plasmon fields in the arms connecting the two junctions.
When, in the following section, we investigate the transport properties of a dc-SQUID, the
leads to which the device drawn in Fig. 1 is connected, are described by two additional plasmon
fields L (x) and R (x).
The Hamiltonian describing the dynamics of 1 (x), 2 (x) is given by

L
j 2
g
1 j 2
dx
+ vpl
,
Hdc,0 =
2
vpl t
x
(64)
j =1,2 L
while the Josephson energy of the two junctions is described now by the boundary interaction
Hamiltonian Hdc,J given by

EJ,2 cos 2 (L) 1 (L)

, (65)
Hdc,J = EJ,1 cos 2 (L) 1 (L) +
2
2
with = /0 .
Again, it is most convenient to introduce the even- and odd-fields, j,e/o (x), as
1
j,e/o (x) = j (x) j (x) ,
2
0 x L.
Fig. 1. Sketch of the dc-SQUID device with the leads.
(66)
348
By definition, the j,e (x)s obey to Neumann boundary conditions at x = 0, while the j,o (x)s
obey to Dirichlet boundary conditions; namely, one has
1,e (0, t) 2,e (0, t)
=
= 0,
x
x
1,o (0, t) = 2,o (0, t) = 0.
(67)
The boundary Hamiltonian Hdc,J gets contributions only from two linear combinations of the
fields e , o . Defining
1
X(x) = 2,e (x) 1,e (x) ,
2
1
(x) = 2,o (x) 1,o (x) ,
2
(68)
it is easy to realize that the device is described by the reduced Hamiltonian Hdc given by
g
Hdc =
2

L
dx
0

2

2
1 X 2
X 2
1
+ vpl
+ vpl
+
vpl t
x
vpl t
x

EJ,2 cos X(L) (L)

.
EJ,1 cos X(L) + (L) +
2
2
(69)
Hdc in Eq. (69) may be regarded as the two-field generalization of the two boundary sine-Gordon
Hamiltonian describing the properties of an rf-SQUID. By construction, at the inner boundary,
= 0, while (x) obeys to Dirichlet
x = 0, X(x) obeys to Neumann boundary conditions, X(0,t)
x
boundary conditions, (0, t) = 0. The boundary conditions at the outer boundary, x = L, are
obtained by requiring that the energy functional is conserved; namely, by

gvpl X(L, t)
+ EJ,1 sin X(L, t) + (L, t) +
x
2

= 0,
+ EJ,2 sin X(L, t) (L, t)
2

gvpl (L, t)
+ EJ,1 sin X(L, t) + (L, t) +
x
2

= 0.
EJ,2 sin X(L, t) (L, t)
(70)
2
For EJ,1 = EJ,2 = 0, X and obey to Neumann boundary conditions at x = L: X(L,t)
=
x
(L,t)
= 0. For EJ,1 , EJ,2 , the fields obey to Dirichlet boundary conditions, X(L, t) +
x
(L, t) 2 = 0 (mod 2), X(L, t) (L, t) + 2 = 0 (mod 2).
In the following, we shall use Hdc to derive the renormalization of the Josephson energies and
the functional form of the Josephson current in both the weakly coupled and the strongly coupled
regimes accessible to a dc-SQUID.
3.1. The weakly coupled fixed point
At the fixed point EJ,1 = EJ,2 = 0, X(x) obeys to Neumann boundary conditions at both
boundaries, while (x) satisfies mixed boundary conditions (namely, Dirichlet boundary conditions at x = 0 and Neumann boundary conditions at x = L). The mode expansion of the field
X(x, t) is, then, given by

vpl t PX
+i
X(x, t) = qX +
L g
1 X (n)
cos(kn x) exp(ikn vpl t),
g
n
349
(71)
n=0
with [X (n), X (n )] = nn+n ,0 ; qX is the constant zero-mode operator and kn = L n. PX is, of

course, the dual zero-mode operator, whose eigenvalues are the winding numbers.
The mixed boundary conditions for the field (x, t) are satisfied by setting

1 a (n)
sin[Kn x]eiKn vpl t ,
(x, t) =
(72)
g
n
n=0
with [ (n), (n )] = nn+n ,0 , and Kn = L (n + 12 ).

By taking into account Eqs. (71), (72), the boundary interaction may be normal ordered as in
Section 2. In particular, one obtains

: EJ,2 :cos X(L) (L)

:,
Hdc,J = EJ,1 :cos X(L) + (L) +
(73)
2
2
1
g
with E J,j = ( 2a
L ) EJ,j , j = 1, 2.
Again, the scaling equations for the pertinent running couplings are obtained from the requirement that the partition function Z is independent of the cutoff. Z is given by

Z = DX D eS0 T eSB ,
(74)
with
g
S0 =
2

d
dx
g
+
2
L
0
L
d

1 X(x, i ) 2
X(x, i ) 2
+ vpl
vpl
x

1 (x, i ) 2
(x, i ) 2
dx
+ vpl
,
vpl
(75)
and

E J,1
SB =
d ei V1,1 (i ) + ei V1,1 (i )
2
0
E J,2
d ei V1,1 (i ) + ei V1,1 (i ) ,
2
(76)
while the vertex operators Va,b ( ) are now given by

:,
Va,b (i ) = :exp i bX(i ) + a(i ) a
2
a, b = 1.
(77)
To determine the scaling dimensions of the relevant operators, one needs to compute the
correlators of the vertices written in Eq. (77). They can be computed as discussed in Appendix B;
350
however, for the purpose of deriving the perturbative RG equations, one only needs the shortdistance limit of the vertex correlators given by

(aa 1)
g

vpl

T Va,b (i )Va ,b (i )
b+b ,0 .
| |
L
(78)
Eq. (78) shows that the scaling dimension of Va,b is ha,b = 1/g.
Since ( ) does not have a zero mode, the O.P.E.s generate, just as it happens for an rfSQUID, higher periodicity terms, according to
ha,b ha ,b +ha+a ,b+b

vpl
Va,b (i )Va ,b (i )
(79)
Va+a ,b+b (i ).
| |
L
To the second order in the couplings E J,1 , E J,2 , Eq. (79) requires to add to SB an extra counterterm given by

SB = 2
d V2,0 (i ) + V2,0 (i ) .
(80)
The running coupling constants G1 , G2 are now defined as

G1 (L) =
L
2a
1 1
g
aEJ,1
,
vpl

G2 (L) =
L
2a
1 1
g
aEJ,2
.
vpl
(81)
Since the vertices V2,0 (i ) = ei :exp[2i( )]: are less relevant than the V1,1 s, they do
not contribute to the evaluation of the leading terms of the perturbative functions for G1 , G2 ,

d ln G1 (L)
1
= 1 (G1 , G2 ) = 1
G1 (L),
d ln(L/L0 )
g
and

d ln G2 (L)
1
= 2 (G1 , G2 ) = 1
G2 (L).
d ln(L/L0 )
g
(82)
As for the rf-SQUID, one finds that both couplings are irrelevant for g < 1; namely, the Neumann
fixed point is infrared stable, and the theory is perturbative in the couplings E J,1 , E J,2 .
1 1
When g > 1, instead, both couplings Gj (L) increase, when L increases, as (L/L0 ) g . Thus,
just as for the rf-SQUID, the theory becomes non-perturbative as soon as L L , with L
g
vpl
) g1 , being EJ,min the smaller coupling, between EJ,1 and EJ,2 .
2a( aEJ,min
To evaluate the Josephson current for g < 1, we compute the partition function for finite .
The result is

Z[] = Z0 T exp
d E J,1 cos X(i ) + (i ) +
2
0

+ EJ,2 cos X(i ) (i )

(83)
,
2
with
Z0 =

1q
n=0

1
n+ 12
1
1 q n+1

2gvpl 2
L m
351
(84)
mZ
vpl
and q = e L .
Since the field X(i ) has a constant zero-mode term, a perturbative expansion in the boundary
interaction yields nonzero contributions to the partition function starting only from the second
order in E J,1 , E J,2 , implying that, at low temperatures, the leading -dependence is given by

1
d1 d2 T Hdc,J (1 )Hdc,J (2 )
Z[] = Z0 1 +
2
0

= Z0
1 + E J,1 E J,2 cos()

d

1
2
g
[2 cosh( u
4L )]

2L
= Z0 1 +
EJ,1 E J,2 B cos() ,
u
with
B=
[ g1 ]
2
2 g [1 + g1 ]
(85)
(86)
as a result, the Josephson current is given by

I [] =
1
ln Z[]
2L
lim
= e
B EJ,1 E J,2 sin().
c0
(87)
The functional form of I [] is similar to Eq. (24), but it is now proportional to E J,1 E J,2 . For
g < 1, Eq. (87) is valid for any value of L, since the quantum corrections are perturbative in
E J,1 E J,2 . When g > 1, instead, higher harmonics operators become more and more relevant as
L increases, inducing a crossover to a saw-tooth-like functional form of the Josephson current,
as L approaches L .
3.2. The strongly coupled fixed point
In this subsection the strong coupling (EJ,j /(2vpl /L) 1) regime of a dc-SQUID is analyzed and the explicit functional forms of the partition function and of the Josephson current are
derived. To compute the partition function with Dirichlet-like boundary conditions at the outer
boundary, X(L, t) + (L, t) + 2 = 0 (mod 2), and X(L, t) (L, t) 2 = 0 (mod 2), one
notices that
(L, t) = (n1 n2 ),
X(L, t) = (n1 + n2 ),
(88)
2
with n1 , n2 relative integers.
Eqs. (88), together with the boundary conditions at the inner boundary, lead to

1
n (n) i n ut
2
sin
x
e L ,
(x, t) = P x
(89)
L
g
L
n
n=0
352
with [a (n), a (n )] = nn+n ,0 , kn = n/L; the 0-mode operator of the field , P , has the
eigenvalues

m
P =
(90)
+
(m = n1 n2 ).
4
2
Similarly, one obtains

X(x, t) = qX + i

1
1
X (n) i (n+ 1 )ut
2
cos
e L
,
n+
x
g
L
2
n + 12
(91)
nZ

with [X (n), X (n )] = nn+n ,0 , Kn = L (n + 12 ), qX = g2 (n1 + n2 ).
As for the rf-SQUID, the strongly coupled fixed point is described in terms of the (now two!)
dual fields X (x, t) and (x, t), whose mode expansion is given by

1
X (n) i (n+ 1 )ut

2
sin
e L
,
n+
x
X (x, t) = 2 PX 2 g
(92)
1
L
2
n
+
2
nZ
where PX is the canonical conjugate operator of qX , and

2
n (n) i n ut

(x, t) = 0 +
cos
gP ut + i2 g
x
e L .
L
L
n
(93)
n=0
Following the same steps used in Section 2.2, one has that
HD [X , ] =
2vpl g
X (n)X (n 1) + (n) (n) ,
(P )2 +
L
(94)
n=0
from which it is straightforward to compute the partition function at the Dirichlet fixed point:

ZD = Tr eHD [X , ]

vpl g
2
1
1
exp
.
=
(95)
n+ 12
2L
2
1 q n+1
1
q
kZ
n=0
Eq. (95) shows that, for = 2k + , with k integer, the ground state is characterized by a
nonzero value of P , proportional to []. Tunneling events from this ground state to the nearest
(in energy) ground states are suppressed by the exponential factor exp[vpl g/2L].
Taking the logarithmic derivative of Eq. (95) one easily derives the Josephson current, which
is given by

1 ln ZD [] e gvpl
=
[] .

4L
I [] = e lim
(96)
From Eq. (95), one sees that the degeneracy among the minima of the Josephson energy is re2gv
(0)
(0)
(0)
moved by the magnetic energy Em

= L pl ( 4
+ m2 )2 . Since Em = Em+1 , for = + 2m
also a dc-SQUID may operate as a qubit between these two levels. When the couplings are large,
but finite, phase slips in the plasmon field (instantons) will induce a renormalization of the physical parameters, as we shall show in the following subsection.
353
3.3. Instantons and boundary perturbations at the Dirichlet fixed point

In order to construct the leading boundary perturbations at the strongly coupled fixed point,
one may use again the DEBC-method [19], applied to the primary bulk operators involving the
dual fields X , . A generic, imaginary time, bulk operator at the strongly coupled fixed point
may be written as V, (i ) = :exp{i[X (i ) + (i )]}:, where the imaginary time boundary
fields are given by
X (n) (n+ 1 )u

2
X (i ) = X (L, i ) = 2 PX 2 g
(1)n
e L
,
1
n
+
2
nZ
(97)
and by
(i ) = (L, i ) = 0 +
(n) n u
2

(1)n
2gP iu + i2 g
e L .
L
n
(98)
n=0
From the commutation rules between the zero modes of the two fields, one obtains

qX , V, (i ) = 2V, (i ),
P , V, (i ) = 2V, (i ).
(99)
Thus, V, (i ) changes the eigenvalue of qX by 2 and the eigenvalue of P by 2. Using

Eqs. (97)(99), one may write the leading boundary operators as

Va,b (i ) = :exp
(100)
aX (i ) + b (i ) :, a, b = 1,
2

VX, (i ) = :exp iX (i ) :,
(101)
V, (i ) = :exp i (i ) :.
V1,1 (i ) and V1,1 (i ) (V1,1 (i ) and V1,1 (i )) change by 2 the phase at junction 1 (2),
while they leave the phase at junction 2 (1) unchanged and thus they describe phase slip operators
at junction 1 (2). Similarly, VX, (i ) (V, (i )) change the eigenvalue of the overall (relative)
phase X(L) ((L)) by 2 leaving the eigenvalue of (L) (X(L)) unchanged: thus, VX, (i )
do not affect the Josephson current across the ring, while V, (i ) provide corrections to the
Josephson current when = 2k + . In the following we shall refer to VX, (i ) (V, (i )) as
q- ( -)instanton operators.
The leading boundary perturbation at the strongly coupled fixed point is given by an arbitrary
linear combinations of the operators listed above, as

H dc,J = Y1 V1,1 (i ) + V1,1 (i ) + Y2 V1,1 (i ) + V1,1 (i )

+ YX VX,+ (i ) + VX, (i ) + Y V,+ (i ) + V, (i ) .

(102)
From the mode expansion of the dual fields, it is straightforward to compute the (lowtemperature) correlators among the boundary vertex operators appearing in Eq. (102). For
|| < , for instance, one has that

T V1,1 (i )V1,1 (i ) = T V1,1 (i )V1,1 (i ) =
| sinh[
gvpl

4L ( )
vpl
4L (
)]|2g
, (103)
| cosh[ 4Lpl ( )]|4g

T V+,X (i )V,X (i ) =
,
v
| sinh[ 4Lpl ( )]|4g
(104)
354
and

T V+, (i )V, (i ) =
e
|2 sinh[
gvpl

L ( )
vpl
2L (
)]|4g
(105)
Eqs. (103)(105) imply that the fugacity of the q- and -instantons and of the phase slips at the
junctions scale with the SQUID size as

12g
1g
L
L
,
Y1(2) (L) = Y1(2) (vpl Tx )
.
YX( ) (L) = YX( ) (vpl Tx )
vpl Tx
vpl Tx
(106)

The instanton size Tx in Eq. (106) is estimated to be of the order of gL/[vpl (EJ,1 + EJ,2 )]
[7].
Eq. (106) shows that, for g < 1, phase slips at the two junctions provide the most relevant
perturbation to the Dirichlet fixed point, rendering the strongly coupled fixed point not infrared
stable. For finite , L and for = 2k + , however, the phase slips are always suppressed by
the magnetic energy associated to the eigenvalue of P . At variance, when g > 1, -instantons
are an irrelevant perturbation which, for L < L,1 , L,2 (i.e., of the healing lengths associated to
the two junctions), only smooths down the edges of the saw-tooth shape of the Josephson current:
for L , the ground state exhibits a discrete Z-symmetry.
4. Transport in SQUID devices
To probe the different regimes attainable by SQUID devices, it is most useful to look at their
dc-conductance. To get conduction of current across a SQUID device, one should be able to
connect it to two leads, enabling to apply a biasing voltage V . This can be achieved easily for a
dc-SQUID while, for the purposes of this section, it is most convenient to regard the rf-SQUID
as an inhomogeneous chain of Josephson junctions connected to two bulk superconducting leads
at fixed phase difference and at finite biasing voltage V ; the chain is made by junctions of
equal strength with a weak link of nominal strength EJ located at its center. For V = 0, this
inhomogeneous chain mimics the response of an rf-SQUID to an external magnetic flux [21].
In this section we study the transport properties of both devices, evidencing the different current (normal and Josephson) response at weak and at strong coupling. Our boundary field theory
approach well reproduces the known results of Refs. [7,9] and, for a dc-SQUID, evidences interesting interference effects between the current flowing through the two junctions.
4.1. The Josephson junction chain at the weakly coupled fixed point
The relevant dynamics of a Josephson junction chain with a weak link connected to two bulk
superconductors at fixed phase difference has been already analyzed in Refs. [21,35]. The
properties of this device may be described by the bosonic field (x) = 1 [R (x) L (x)],
2
where R (x) and L (x) are the fields describing the phase of each junction in the right and
the left half-chain, respectively. The current operator across the junction is then given by j (t) =
ge v
pl (L,t)
x . Energy conservation provides the dynamical boundary conditions for (x, t) at
2
the outer boundary, which is given by

vpl g (L, t)
(107)
+ 2E J sin 2(L, t) + = 0;
using Eq. (107), the current operator may be written as

(t) = (L, t) .
j (t) = e E J sin 2(t) +
355
(108)
At the weakly coupled fixed point (EJ = 0), (x, t) satisfies the Neumann boundary conditions
at the outer boundary; namely, (L,t)
= 0.
x
In order to apply a voltage V to the device, one may bias at voltage V /2 and V /2 each one
of the two leads. In the small E J -limit, it is safe to assume that each chain is at equilibrium with
its own lead, and that the two half chains are disconnected. Accordingly, the voltage bias V is
introduced by adding to the total Hamiltonian the term
e V g
HV =
vpl
L
dx
,
t
(109)
which can be accounted for by a mere shifting of the field , as

(x, t) (x, t) e V t.
(110)
As a result of Eq. (110), the Josephson current acquires an explicit dependence on time and, to
the first order in E J , reads
IJ [, t] = e E J sin[ + e V t].
(111)
In addition to the Josephson current, there is a normal time-independent dc-current which, as it

will be shown in the following, becomes the leading contribution for a pertinent choice of the
infrared cutoff leading to the celebrated KaneFisher formula [9].
Since the normal current is independent of time, it may be computed using the imaginary time
formalism. From Eq. (110), one obtains

j (i ) = e E J :sin 2(i ) + + e V i :,
(112)
with

(i ) =
2
(n) (n+ 1 )vpl
2
(1)n
e L
.
g n
n + 12
(113)
From Eqs. (112), (113), it follows that the leading contribution to the normal current appears
only at the second order in E J , since
e
I (E J )2
4i
d T :exp 2(i ) + + ie V :

:exp 2(i ) ie V :

e 2
d T :exp 2(i ) ie V :
(EJ )
4i
0

:exp 2(i ) + + ie V :

sinh[ax]
2e (E J )2 L
2 2L
dx 2
I[a],
=
= e (EJ )
2
vpl
vpl
g sinh g (x)
2
0
(114)
356
with
a=
2Le V
vpl
(115)
and [36]

I[a] =
dx
0
sinh[ax]
2
[2 sinh(x)] g
[1 g2 ] [ a2 + g1 ]
4
[1
a
2
g1 ]
[ a2 + g1 ]
[1 +
a
2
g1 ]
(116)
From Eq. (114), one sees that the dc-current depends crucially on the ratio a between the energy
window defined by the applied voltage, e V , and the level spacing characteristic of the device,
vpl /(2L). In fact, if e V < vpl /L, one may expand Eq. (114) to first order in a, getting
I=
e (E J )2 L [1 g2 ]
2vpl
BV ,
(117)
with
[ z + 1 ]
[ 2z + g1 ]
2
g
,
B=
z [1 2z g1 ] [1 + 2z g1 ] z=0
(118)
which yields the (normal) dc-linear response, to an applied voltage V , of a junction with renor1
g
malized strength E J = ( 2a
L ) ; this is expected, since, in this limit, vpl /L provides the infrared
cutoff. If, instead, vpl /L e V , one may compute the current in a large-a expansion, using the
1
Stirling formula, [z] 2ez zz 2 , which yields

2
2 2Le V g 1
e (E J )2 L
sin
.
1
I=
(119)
vpl
g
g
vpl
Eq. (119) is the celebrated KaneFisher formula for the current across a constriction in a spinless
Luttinger liquid. Our result is, after all, not surprising, since the wires connecting the junctions
have been regarded as one-dimensional spinless Luttinger liquids, and the junctions as boundary
2
interactions. It should be noticed, however, that the normal current is proportional to (e V ) g

since e V is the pertinent infrared cutoff when an external bias voltage is applied to the device.
At variance, in the computation of the Josephson current, the pertinent cutoff is provided by the
plasmons level spacing s = 2vpl /L. Thus, in the limit where s e V , our analysis shows that
the leading contribution to the total current across the junction is given by the normal current.
4.2. The Josephson junction chain at the strongly coupled fixed point
In this subsection, the current across the Josephson junction chain is computed at zero phase
difference as E J /(u/L) . Here, it is most convenient to resort to the dual formulation,
based on the dual field (x, t). The total Hamiltonian (including the leading boundary perturbation) is given by
1
H [] =
4(2g)
L
0

i (L)
1 2
2
i (L)
dx
+u
: + :e 2
:,
Y :e 2
u t
x
(120)
357
where the dual field (x, t) obeys to Neumann boundary conditions at x = L. Thus, the mode
expansion of the field (i ) = (L, i ) is given by
(n) n u
2

(1)n
e L .
(i ) = (L, i ) = 0 + i 2gP u + i2 g
(121)
L
n
n=0
Since at the strongly coupled fixed point, the two half chains cannot be regarded as isolated from
each other, one must apply a voltage V by biasing the right-handed chiral mode of (x, t), with
respect to the left-handed ones. This corresponds to adding to H [] a voltage-dependent term
given by
e V
H V =
2
L
dx
0
e V
=
x
2vpl
L
dx
.
t
(122)
The imaginary time current operator is

SE [; V , a]
j (i ) =
,
a( )
a=0
(123)
where SE [; V , a] is the Euclidean action with a source term for the current, given by
1
SE [; V , a] =
4(2g)

d
0

2Y
0
ie
2L
L
dx
0

1 (x, i ) 2
(x, i ) 2
+ vpl
vpl

L
(i )
(x, i )
e V
d :cos
d dx
:i
2vpl
2
0
L
d
dx a( )
(i )
.
(124)
By shifting the field according to

e gvpl
(x, i )
(x, i )
A( ),

2ige V i
L
2
2
(125)
with A ( ) = a( ), one finds that the current is given by

e gvpl
SE [; V , a]
g(e )2
(i )
j( ) =
=
V
:.
V
+
4Y
:sin
ie

a( )
L
2
a=0
(126)
From Eq. (126) one sees that, apart from the constant term g(e ) V , the current may be computed
as in the previous section, provided that one substitutes 4/g with g and E J with Y . Thus, to the
second order in Y one obtains

g(e )2
2e Y 2 L
2Le V 2g1
I=
(127)
.
V+
sin[g] [1 2g]
vpl
As it happens also at the weakly coupled fixed point, the celebrated power-law dependence of I
on the applied voltage V emerges when 2Le V /(vpl ) 1.
358
4.3. The dc-SQUID at the weakly coupled fixed point

For a dc-SQUID, it is most convenient to represent the leads as two quantum wires with the
same Luttinger parameters g and vpl , with plasmon fields respectively given by L and R .
The two leads are assumed to be connected to the dc-SQUID with smooth junctions, that is,
three-wire junctions with no backscattering in any arm. Thus, one should describe the SQUIDs
arm with four fields, rather than with two; namely u,L , u,R for the left-upper and for the
right-upper arm of the ring, d,L , d,R for left-lower and for the right-lower arm of the ring,
respectively. Since the connections between leads and SQUIDs arm are assumed to be ideal,
they are described by the strongly coupled fixed point of the three quantum wire Y-junction
studied in Ref. [19], with no concatenated flux. Thus, by centering the Y-junction at x = 0, the
boundary conditions are given by

a (x, t) + u,a (x, t) + d,a (x, t) x=0 = 0,
x
which embodies the current conservation at the connection, and by
2a (0, t) + u,L (0, t) + d,a (0, t) = u,a (0, t) d,a (0, t) = 0,
(128)
(129)
with a = L, R.
The relevant fields for the Josephson junction dynamics are given by the odd-parity combinations
1
u/d (x, t) = u/d,R (x, t) u/d,L (x, t) ,
2
1
(x, t) = R (x, t) L (x, t) ,
(130)
2
which are related to the fields (x, t), X(x, t) introduced in Section 3 by
1
1
X(x, t) + (x, t) ,
d (x, t) = X(x, t) (x, t) .
(131)
2
2
As one sees from Eqs. (128), (129), introducing the contacts does not affect at all the boundary
condition of (x, t) at the inner boundary (x = 0). The boundary conditions for X(x, t), instead,
explicitly depend on the applied voltage bias, whose effect may be accounted for by including in
the total Hamiltonian a pertinent source term.
At the weakly coupled fixed point, biasing the left-hand lead at a dc-voltage V /2 and the
right-hand lead at a voltage V /2, corresponds to adding to the SQUIDs Hamiltonian a voltage
dependent term given by
u (x, t) =
HV =

L
L
g 2e V
R (x) L (x)
X(x)
ge V
dx
dx
=
,
4
t
t
2
t
L
(132)
which is equivalent to a shift of (x, t), which is linear in time, namely, the field (x, t) should
be replaced by (x, t) e V t. From Eqs. (128), (129), (132) one gets that also u/d (x, t) must
be shifted as
u/d (x, t) u/d (x, t)
e V
t.
2
(133)
359
As a consequence, one sees that X(x, t) takes the additional contribution, given by
X(x, t) X(x, t) e V t,
(134)
which introduces in the boundary interaction Hamiltonian an explicit dependence on , given by

Hdc,J Hdc,J (i ) = EJ,1 cos X(i ) + (i ) + + ie V

2

E J,2 cos X(i ) (i ) + ie V ,

(135)
2
at finite V , the current operators across the two junctions are then given by

j1 (i ) = e EJ,1 sin X(i ) + (i ) + + ie V ,

2
and

j2 (i ) = e EJ,2 sin X(i ) (i ) + ie V .

2
(136)
(137)
As in the previous section, the current across the two junctions is computed from a perturbative
expansion in E J,1 , E J,2 . One obtains the following results.
Current across the junction 1: I1 ,
I1 = I1,1 + I1,2 .
I1,1 is given by
(1)

d T :sin X(i ) + (i ) + + ie V :
2
0

:cos X(i ) + (i ) + + ie V :
2

sinh[e V ]
e
2
,
d
= (EJ,1 )
2
vpl
2
g
[2
sinh(
)]
2L
0
I1,1 = e (E J,1 )
which, for 4Le V /(vpl ) 1 yields

2

2 4Le V g 1
2e (E J,1 )2 L
I1,1
1
sin
.
vpl
g
g
vpl
(2)

d T :sin X(i ) + (i ) + + e V :
2
0

:cos X(i ) (i ) + e V :
2
I1,2 = e E J,1 E J,2
(138)
360
e E J,1 E J,2

d
0
ee
V +i
[cosh(
2
vpl
g
4L )]
ee
V i
[cosh(
2
vpl
g
4L )]

4e E 1 E J,2 L i
e J[a] ei J[a] ,
vpl
(139)
with

J[a] =
eax
dx
where a = 4e V L/(vpl ) and

J[a] = 2
(140)
cosh g (x)
2
g 1
i( a2 + g1 )

a 1
2
B 1;
;1
,
2 g
g
(141)
z
where the incomplete-beta function is defined as B[z, , ] = 0 dt t 1 (1 t)1 .
For large a and 1 2/g > 0, one may consider the asymptotic expansion of the incompletebeta function [36],

2
2
2
2
a 1
1
g
+
;1
i exp i a i
1 a
1 ,
B 1;
2 g
g
2
2 g
g
(142)
from which

2e E J,1 E J,2 L
2
2e V L
1
+
cos() cos
I1,2
vpl
g
vpl
g
2

2e V L
4Le V g 1
+ sin() sin
(143)
+
.
vpl
g
vpl
Current across the junction 2: I2 ,
I2 = I2,1 + I2,2 .
The two contributions are computed below:
(1)
(2)
d T :sin X(i ) + (i ) + ie V :
2
0

:cos X(i ) + (i ) + ie V :
2

2
2e (E J,2 )2 L
2 4Le V g 1
sin
.
1
vpl
g
g
vpl

I2,1 = e EJ,1 E J,2 d T :sin X(i ) + (i ) + e V :
2
0

:cos X(i ) (i ) + + e V :
2
I2,2 = e (E J,2 )2
(144)

4e E 1 E J,2 L
2
2e V L
1
+
cos() cos
vpl
g
vpl
g
2

2e V L
4Le V g 1
+ sin() sin
+
.
vpl
g
vpl
361
(145)
The net transport current across the dc-SQUID is obtained as the sum of the currents across
the two junctions, that is

2e L
2
2
2
1
(EJ,1 ) + (EJ,2 ) sin
I = I 1 + I2 =
u
g
g

2
2e V L
4Le V g 1
+ EJ,1 EJ,2 cos() cos

(146)
+
.
vpl
g
u
Eq. (146) contains an interference contribution (proportional to cos()), representing the cumulative effect of a bias voltage applied to junction 1 on the current flowing across junction 2,
and of a bias voltage applied to junction 2 on the current flowing across junction 1. This is a
remarkable result of the application of the boundary sine-Gordon techniques to SQUID devices.
Treating the plasmon modes as modes of the dynamical fields, allows for an explicit computation
of the interference contributions to the dc-current.
The circulating current across the dc-SQUID, J , may be computed as the difference between
the currents flowing across the two junctions, IJ = I1 I2 . The result is

2
2
2e V L 4Le V g 1
4e E J,1 E J,2 L
1
+
,
sin() sin
J=
vpl
g
vpl
g
vpl
(147)
which, again, comes from pertinently taking into account correlation functions of the plasmon
modes computed using the boundary sine-Gordon theory.
4.4. The dc-SQUID at the strongly coupled fixed point
At the strongly coupled fixed point, as long as one neglects phase slips at the junctions, one
should not expect any resistance across the junctions. As a consequence, the external voltage V
is added by biasing the left chiral mode of the plasmon field at each lead with respect to the right
chiral mode, with opposite biases in the two contacts. As discussed in Section 4.2 for the chain,
this is accounted for by shifting the field (x, t) according to
(x, t) (x, t)
e V
x.
vpl
(148)
Due to Eqs. (128), (129), (148), one has that, at finite V , also u/d (x, t) must be shifted as
u/d (x, t) u/d (x, t)
e V
x.
2vpl
(149)
Eqs. (131) imply that

X(x, t) X(x, t)
e V
x,
vpl
(x, t) (x, t),
(150)
362
which is equivalent to adding to the total Hamiltonian a voltage dependent contribution given by
e V g
H V =
L
0
e V g
X(x)
=
dx
x
2vpl
L
dx
X (x)
.
(151)
The current operators across the two junctions are now written as

e X (i ) (i )
j1/2 (i ) = i
= jX (i ) j (i ).
2
(152)
An useful representation may be derived by adding to the total Euclidean action SE [X , ,

aX , a ] the source term given by
SESource
ie
=
2L
L
d

(i )
X (i )
+ a ( )
,
dx aX ( )
(153)
and by taking the functional derivatives of SE [X , , aX , a ] with respect to aX , a , after shifting the dual fields according to
X (x, i ) X (x, i ) 2i
e gvpl
AX ( ) ige V ,
L
(154)
and
(x, i ) (x, i ) 2i
e gvpl
A ( ) + i ,
L
2L
(155)
with A X ( ) = aX ( ), A ( ) = a ( ).
The Euclidean action in the dual representation is then given by
SE =

a=X,
1
4(2g)

d
0

2Y1
L
dx
0

a (x, i ) 2
1 a (x, i ) 2
+ vpl
vpl

1
X (i ) + (i ) + i ge V vpl
d cos
2
4L

2e gvpl
1
+i
A ( ) + AX ( ) 2Y2 d cos
X (i ) + (i )
L
2
0

gv
2e
pl
A ( ) AX ( )
i ge V + vpl
+i
4L
L
1
+
4(2g)

d
0

L
dx
aX ( )
2ge V + 2e gvpl
L
2

a ( )
2
.
+ vpl
+ 2e gvpl
L
4L
(156)
363
As a result of Eq. (156), one finds that the current operators jX , j may be written as

e gvpl Y1
g(e )2
1
jX (i ) =
V + 4i
sin
X (i ) + (i ) + i ge V + vpl
L
2
4L

e gvpl Y2
1
2i
(157)
sin
X (i ) + (i ) i ge V + vpl
,
L
2
4L
and by
j (i ) =

e gvpl
e gvpl Y1
1
+ 4i
sin
X (i ) + (i ) + i ge V vpl
4L
L
2
4L

e guY2
1
+ 4i
(158)
sin
X (i ) + (i ) i ge V + vpl
.
L
2
4L
From Eqs. (157), (158), one sees that, to the lowest order in the phase slip contribution (Y1 =
Y2 = 0), the stationary component of the circulating current reduces to the value of the Josephson
gu
2
current at zero voltage, e4L
, while the conduction current is given by g(e ) V , just as for a
single chain at the strongly coupled fixed point.
Higher-order corrections, including the effects of phase slips, may be computed following the
same steps used in the previous section. The result is

[ a2 + g]
[ a2 + g]
4e g(Y1 )2
jX =
[1 2g]
[1 a2 g] [1 + a2 g]

[ a2+ + g]
[ a2+ + g]
4e g(Y2 )2
+
(159)
,
[1 2g]
[1 + a2+ g] [1 a2+ g]
with
a =
2ge V L u
,
u
(160)
and by

[ a2 + g]
[ a2 + g]
4e g(Y1 )2
j =
[1 2g]
[1 a2 g] [1 + a2 g]

[ a2+ + g]
[ a2+ + g]
4e g(Y2 )2
.
[1 2g]
[1 + a2+ g] [1 a2+ g]
(161)
For 0 and for |a | 1, one may expand Eqs. (159), (161) to the first order in a . In the
symmetric case (Y1 = Y2 ), this provides a correction to the slope of both the dc-current and of
the stationary component of the Josephson current, so that one gets

(e )2 g
16Y 2 gL
jX
(162)
1+
C [1 2g] V ,
2
u
and
j

e gvpl
16Y 2 gL
1+
C [1 2g] ,
4L
u
(163)
364
with

[z + g]
[z + g]
C=
x [1 z + g] [1 + z g] z=0
(164)
Eqs. (162), (163) show in a rather simple context the way in which relevant perturbation may
affect both the Josephson and the normal currents across the device.
5. Comments and conclusions
We provided a framework where two-boundary sine-Gordon models may be used to investigate the relevant non-perturbative features, occurring in the renormalization of the Josephson
couplings EJ,i of a multi-junction SQUID. As pointed out in Ref. [7], this is needed sincefor
large fluctuations of the phase of the order parameterthe effective boundary conditions on the
plasmon fields depend on the Josephson energy of the junction. Our analysis shows that, when
g > 1 and L > L , the Josephson couplings affect the dynamics of the plasmon field only as a
boundary term, while, for L < L , they can be regarded as quantum impurities in a superconducting loop. For g < 1, instead, the Josephson couplings have to be always regarded as quantum
impurities.
The boundary field theory approach to SQUID devices proposed in this paper turns out to be
very powerful in providing a systematic procedure for computing not only the Josephson currents (and the leading corrections induced by the relevant perturbations), but also the dc-currents
flowing in a SQUID device, due to externally applied voltages. For pertinent choices of the infrared cutoff, the results of the boundary field theory well reproduce what has been obtained in
Refs. [7,9], allowing, in addition, to appreciate new and closer connections between the theories
of Josephson superconducting devices, quantum Brownian motion, fermionic quantum wires,
and quantum impurity problems. Furthermore, for dc-SQUID devices, our analysis points out
the existence of new remarkable interference effects between the currents flowing through each
junction.
We investigated here SQUID-devices with one or two junctions. For these systems, we have
found two stable regimes characterized by stable fixed points of the pertinent RG equations,
driven by the strength of the bare Josephson energy. It would be interesting to ascertain if superconducting loops with more than two junctions may exhibit new (in addition to the by now
well known weakly and strongly coupled fixed points) renormalization group fixed points at a
finite Josephson coupling. Furthermore, it could be instructive to investigate the properties of a
multi-junction SQUID device, for which g = 1, since the boundary sine-Gordon model provides
an exact solution for this value of the Luttinger parameter. We feel that multi-junction superconducting loops may become an interesting laboratory for testing the physical properties of field
theories describing interacting Luttinger liquids with pertinent boundary interactions.
Acknowledgements
We thank Ian Affleck, Leonid Glazman, Hubert Saleur and Gordon W. Semenoff for inspiring discussions and useful correspondence at various stages of this project. We benefited from
conversations with Maria Cristina Diamantini, Gianluca Grignani, Francesco P. Mancini, Arianna Montorsi, Mario Rasetti, Arturo Tagliacozzo and Andrea Trombettoni. This work has been
partly supported by the MIUR National Project Josephson Networks for Quantum Coherence
and Information (grant No. 2004027555).
365
Appendix A. Scaling properties of boundary interaction strengths

In this appendix we shall derive the scaling laws for boundary interaction operators.
Let us consider a one-dimensional system described by the Euclidean action
S = S0

=1
(A.1)
d ( ),
0
where S0 , the action for K independent massless KleinGordon fields, {j }, with pertinent
boundary conditions, is given by:

L
K
j 2
g
1 j 2
S0 =
d dx
+ vpl
.
4
vpl
j =1 0
(A.2)
The boundary operators { ( )} are functionals of the fields {j } with scaling dimension h . As
a consequence, for , one gets

G ( ) = ( ) ( ) 0 ,
1
,
[2vpl ( )/L]2h
(A.3)
where . . .
0 denotes averaging with respect to the free action S0 , while L is the size of the
system.
It is well known that, in order to introduce scale invariant interaction terms, one needs to
define dimensionless coupling constants, given by g (a) = a 1h , where is the coupling
constant appearing in Eq. (A.1), and a is a short distance cutoff. To first order in the coupling
strengths, the renormalization group equations are given by
dg (a)
= [1 h ]g (a).
d ln(a/a0 )
(A.4)
Higher order corrections to the renormalization group equations come from nontrivial shortdistance fusion rules of the s, which are represented by the O.P.E.s

C
( ) ( )
(A.5)
( ).
h +h h
k | |
If the O.P.E.s coefficients C are different than zero, the scaling equations get contributions
which are of second order in the couplings. To derive them, one should start from the partition
function

M

1+h
Z = Z exp
(A.6)
g (a) d a
( )
.
=1
Upon introducing the cutoffs a and L and by expanding Eq. (A.6) up to the third-order in the
running coupling strengths, one gets
L/v

pl L/v
pl
M

Z
a
1+
g g
d
d a 1+h a 1+h
( ) ( ) 0
Z
vpl

, =1
366

L
u
g g g
, , =1
L/v
pl
d
0
L/v
pl
d a 1+h a 1+h a 1+h

a
a

( ) ( ) ( ) 0 .
vpl
vpl
(A.7)
If one rescales the cutoff, a (1 + )a, with 1, additional contributions to the coupling
constant renormalization arise, as a result of configurations where the arguments of two fields lie
between a and (1 + )a. The ensuing renormalization is derived by means of the identity

a
a

(1 + )
(1 + )

vpl
vpl

a
a
a
a
a

vpl
vpl
vpl
vpl
vpl

a
a
+
(A.8)

.
vpl
vpl
Thus, the third-order contribution to Eq. (A.7) renormalizes the second-order term with
2
M

, =1
L/v
L/v

pl
pl
M

a

g g
[g C, , ]
d
d
vpl

1+h 1+h
(A.9)
( ) ( ).
By setting = ln(a/a0 ) in Eq. (A.6), it is straightforward, but tedious, to derive explicit expression for the nonlinear terms appearing in the RG equations for the running coupling strengths;
the final result is
M

dg (a)
C, , g (a)g (a).
= [1 h ]g (a) +
d ln(a/a0 )

(A.10)
, =1
The RG equations in Eq. (A.10) may be rewritten as

C[{g}]
dg
= 0,
d ln(a/a0 )
g
where C[{g}] is given by
N

1
1
C {g} =
[1 h ]g2 +
2
3
=1
(A.11)
C, , g g g .
(A.12)
, , =1
The RG fixed points coincide with the extrema of the function C[{g}], that is, with the set of
values of {g}, {g }, such that
C[{g }]
(A.13)
= 0, = 1, . . . , M.
g
Due to its properties, C may also be identified with the boundary entropy of the system [32].
We would like to point out that the thermodynamics limit may be achieved either by sending
either a or L to . In fact, sending a to amounts to cut off high momenta contributions,
which amounts to send the size of the system to . This implies that scaling may be realized
either using a/a0 , or L/L (as it has been done in this paper), as scaling parameters.
367
Appendix B. Boundary vertex operators

As an example of correlation functions computed with both and L finite, in this appendix we derive the correlators of the normal-ordered vertices VX,1 ( ) = :exp[iX(i )]:, and
V,1 (i ) = :exp[i(i )]: introduced in Section 3, where X(x, t) obeys Neumann boundary
conditions at both boundary x = 0 and x = L, while (x, t) obeys Dirichlet boundary conditions
at the inner boundary, Neumann at the outer boundary.
The partition function ZX = Tr[HX ], where HX is the Hamiltonian for the field X, has
been derived in Eq. (84). The result is

2gvpl 2
1
ZX =
(B.1)
exp
m .
(q)
L
mZ
The two-vertex correlation function, GX (1 , 2 ), is then given by

GX (1 , 2 ) =

1
Tr T :eiX(i1 ) ::eiX(i2 ) : eHX .
ZX
(B.2)
As it happens for the partition function, also GX is factorized into a contribution from the oscillatory modes, times a contribution from the zero modes. The trace over the oscillatory modes
yields
Tosc =
(B.3)
Tn ,
n=1
where

1 a(kn ) vpl kn 1
1 a(kn ) vpl kn 1
Tn = Tr T exp i
e
e
exp i
g
n
g n

2 aX (kn ) vpl kn 2
2 aX (kn ) vpl kn 2 aX (kn )aX (kn )
exp i
.
e
e
exp i
q
g
n
g n
(B.4)
The trace in Eq. (B.4) may be computed by resorting to a coherent state decomposition, which
uses the basis of coherent states |xn
, defined by
aX (kn )|xn
=
nxn |xn
|xn
= e
|n |2
2
xn aX (kn )
n
|0
,
and the decomposition of the identity given by

(xn )2
1
dxn e 2 |xn
xn |.
I=
The derivation is tedious, but straightforward [37]. The final result is

2

1
4
qn
Tn =
exp
1 qn
gn 1 q n

vpl kn |1 2 |
vpl kn |1 2 |
2
e
2q n
e
exp
+
.
g(1 q n )
n
g(1 q n )
n
(B.5)
(B.6)
(B.7)
368
As a consequence, one gets

qn
1
4
Tn =
Tosc =
exp
(q)
g
n(1 q n )
n=1
n=1

1 (ekn vpl |1 2 | )n 1 q n (ekn vpl |1 2 | )n

+
.
exp
g
(1 q n )
g
(1 q n )
n=1
(B.8)
n=1
Using the Jacobis triple product identity [37], one finally obtains
Tn =
n=1
ivpl (1 2 ) vpl 1
|i L ) g
1 (
1
2L
,
v
(q)
(0|i pl )
1
(B.9)
where 1 ( | ) is Jacobis 1 elliptic function, defined as

1 2
1
(1)n e (n+ 2 ) e2i (n+ 2 ) .
1 ( | ) = i
(B.10)
nZ
By tracing over the zero mode and applying Poissons resummation formula, one eventually gets

GX (1 , 2 ) = 3

2 ) vpl g2
|i L )
1 (ivpl (12L
u u
,
i

v
2gL 2gL
(0|i pl )
1
(B.11)
2L
where the 3 -function is defined as

n2
3 ( | ) =
ei 2 e2n .
(B.12)
nZ
By performing similar computations, one obtains

1
Tr T :ei(i1 ) ::ei(i2 ) : eH
ZX

2 ) vpl g2
|i L )
4 (ivpl (12L
u u
= 3
,
i

v
pl

2gL 2gL
(0|i
)
G (1 , 2 ) =
where the 4 -function is defined as

n2
4 ( | ) =
(1)n ei 2 e2in .
(B.13)
2L
(B.14)
nZ
When taking the limit , the correlators in Eqs. (B.11), (B.13) give the ones that we have
used in the paper.
References
[1] For a classical review, see M.H. Devoret, D. Esteve, C. Urbina, J. Martinis, A. Cleland, J. Clarke, in: Yu. Kagan,
A.J. Leggett (Eds.), Quantum Tunnelling in Condensed Media, Elsevier, Amsterdam, 1992.
[2] G. Schoen, A.D. Zaikin, Phys. Rep. 198 (1990) 237.
[3] A.O. Caldeira, A.J. Leggett, Ann. Phys. (N.Y.) 149 (1983) 374.
[4] M.H. Devoret, D. Esteve, H. Grabert, G.L. Ingold, H. Pothier, C. Urbina, Phys. Rev. Lett. 64 (1990) 1824;
S.M. Girvin, L. Glazman, M. Jonson, D.R. Penn, M.D. Stiles, Phys. Rev. Lett. 64 (1990) 3183.
369
[5] H. Grabert, U. Weiss, Phys. Rev. Lett. 54 (1985) 1605;

M.P.A. Fisher, A.T. Dorsey, Phys. Rev. Lett. 54 (1985) 1609;
A.T. Dorsey, M.P.A. Fisher, M.S. Wartak, Phys. Rev. A 33 (1986) 1117.
[6] D.V. Averin, Yu.A. Nazarov, A.A. Odintsov, Physica B 165166 (1990) 945;
G. Falci, V. Bubanja, G. Schoen, Europhys. Lett. 16 (1991) 109;
G. Falci, V. Bubanja, G. Schoen, Z. Phys. B 85 (1991) 451;
G.L. Ingold, H. Grabert, U. Eberhardt, Phys. Rev. B 50 (1994) 395.
[7] F.W.J. Hekking, L.I. Glazman, Phys. Rev. B 55 (1997) 6551.
[8] J.E. Mooij, G. Schoen, Phys. Rev. Lett. 55 (1985) 114;
O. Buisson, P. Xavier, J. Richard, Phys. Rev. Lett. 73 (1994) 3153.
[9] M.P.A. Fisher, W. Zwerger, Phys. Rev. B 32 (1985) 6190;
C.L. Kane, M.P.A. Fisher, Phys. Rev. Lett. 68 (1992) 1220;
C.L. Kane, M.P.A. Fisher, Phys. Rev. B 46 (1992) 15233.
[10] A.I. Vainshtein, V.I. Zakharov, V.A. Novikov, M.A. Shifman, Sov. Phys. Usp. 25 (1982) 195;
P.W. Anderson, G. Yuval, Phys. Rev. Lett. 23 (1969) 89;
D.R. Hamann, Phys. Rev. Lett. 23 (1969) 95;
G. Yuval, P.W. Anderson, Phys. Rev. B 1 (1970) 1522;
P.W. Anderson, G. Yuval, D.R. Hamann, Phys. Rev. B 1 (1970) 4464.
[11] Y. Makhlin, G. Schoen, A. Shnirman, Rev. Mod. Phys. 73 (2001) 357.
[12] J.E. Mooij, T.P. Orlando, L. Levitov, L. Tian, C.H. van der Wal, S. Lloyd, Science 285 (1999) 1036;
M.V. Feigelman, V.B. Geshkenbein, L.B. Ioffe, G. Blatter, J. Low Temp. Phys. 118 (2000) 805;
L.B. Ioffe, V.B. Geshkenbein, M.V. Feigelman, A.L. Fauchere, G. Blatter, Nature 398 (1999) 679;
C.H. van der Wal, A.C.J. ter Haar, F.K. Wilhelm, R.N. Schouten, C.J.P.M. Harmans, T.P. Orlando, S. Lloyd, J.E.
Mooij, Science 290 (2000) 773;
G. Blatter, V.B. Gesjkenbein, L.B. Ioffe, Phys. Rev. B 63 (2001) 174511.
[13] T.P. Orlando, J.E. Mooij, L. Tian, C.H. van der Wal, L.S. Levitov, S. Lloyd, J.J. Mazo, Phys. Rev. B 60 (1999)
15398.
[14] R.F. Voss, R.A. Webb, Phys. Rev. Lett. 47 (1981) 265;
J.M. Martinis, M.H. Devoret, J. Clarke, Phys. Rev. B 35 (1987) 4682;
J. Clarke, A.N. Cleland, M.H. Devoret, D. Esteve, J.M. Martinis, Science 239 (1988) 992;
R. Rouse, S. Han, J.E. Lukens, Phys. Rev. Lett. 75 (1995) 1614;
P. Silvestrini, V.G. Palmieri, B. Ruggiero, M. Russo, Phys. Rev. Lett. 79 (1997) 3046;
J.R. Friedman, V. Patel, W. Chen, S.K. Tolpygo, J.E. Lukens, Nature 406 (2000) 43.
[15] A. Le Clair, G. Mussardo, H. Saleur, S. Shorik, Nucl. Phys. B 453 (1995) 581;
J.-S. Caux, H. Saleur, F. Siano, Nucl. Phys. B 672 (2003) 411;
J.-S. Caux, H. Saleur, F. Siano, Phys. Rev. Lett. 88 (2002) 106402.
[16] J. Cardy, in: J.P. Francoise, G. Naber, S. Tsun Tsou (Eds.), Boundary Conformal Field Theory, in: Encyclopedia of
Mathematical Physics, Elsevier, New York, 2006.
[17] J.M. Luttinger, J. Math. Phys. 4 (1963) 1154;
S. Tomonaga, Prog. Theor. Phys. 5 (1950) 544.
[18] H.J. Shulz, G. Cuniberti, P. Pieri, in: G. Morandi, P. Sodano, V. Tognetti, A. Tagliacozzo (Eds.), Field Theories for
Low-Dimensional Correlated Systems, Springer, Berlin, 2000.
[19] C. Chamon, M. Oshikawa, I. Affleck, Phys. Rev. Lett. 91 (2003) 206403;
C. Chamon, M. Oshikawa, I. Affleck, J. Stat. Mech. 0602 (2006) P008.
[20] K.A. Matveev, L.I. Glazman, Physica B 189 (1993) 266.
[21] D. Giuliano, P. Sodano, Nucl. Phys. B 711 (2005) 480.
[22] I. Affleck, Nucl. Phys. B 336 (1990) 517;
I. Affleck, A.W.W. Ludwig, Nucl. Phys. B 352 (1991) 849;
I. Affleck, A.W.W. Ludwig, Nucl. Phys. B 360 (1991) 341;
I. Affleck, A.W.W. Ludwig, Phys. Rev. Lett. 67 (1991) 161;
I. Affleck, A.W.W. Ludwig, Phys. Rev. B 48 (1993) 7292.
[23] C.L. Kane, M.P.A. Fisher, Phys. Rev. B 46 (1992) 1220.
[24] P. Fendley, A.W.W. Ludwig, H. Saleur, Phys. Rev. Lett. 74 (1995) 3005;
P. Fendley, A.W.W. Ludwig, H. Saleur, Phys. Rev. B 52 (1995) 8934;
C.L. Kane, M.P.A. Fisher, Phys. Rev. B 46 (1992) 15233.
[25] I. Affleck, M. Oshikawa, H. Saleur, Nucl. Phys. B 594 (2001) 535;
H. Yi, C.L. Kane, cond-mat/9602099.
370
[26] I. Affleck, J.S. Caux, A.M. Zagoskin, Phys. Rev. B 62 (2000) 1433.
[27] F.D.M. Haldane, Phys. Rev. Lett. 45 (1980) 1358;
J.L. Black, V.J. Emery, Phys. Rev. B 23 (1981) 429;
M.P.M. den Nijs, Phys. Rev. B 23 (1981) 6111;
P. Gueret, N. Blanc, R. Germann, H. Rothuizen, Phys. Rev. Lett. 68 (1992) 1896.
[28] K. Bardakci, A. Konechny, hep-th/0009214;
S.A. Harvey, D. Kutasov, E.J. Martinec, hep-th/0003101.
[29] A. Sen, JHEP 9912 (1999) 027.
[30] A. LeClair, M.E. Peskin, C.R. Preitschopf, Nucl. Phys. B 317 (1989) 411;
J.A. Harvey, D. Kutasov, E.G. Martinec, G. Moore, hep-th/0111154.
[31] E. Gava, K.S. Narain, N.H. Sarmadi, Nucl. Phys. B 504 (1997) 214;
A. Sen, JHEP 9809 (1998) 013;
A. Sen, Int. J. Mod. Phys. A 14 (1999) 4061.
[32] I. Affleck, A.A. Ludwig, Phys. Rev. Lett. 67 (1991) 161.
[33] S. Coleman, Phys. Rev. D 11 (1975) 2088;
D.J. Amit, Y.Y. Goldshmidt, G. Grinstein, J. Phys. A 13 (1980) 585.
[34] K.A. Matveev, A.I. Larkin, L.I. Glazman, Phys. Rev. Lett. 89 (2002) 096802.
[35] L.I. Glazman, A.I. Larkin, Phys. Rev. Lett. 79 (1997) 3736.
[36] M. Abramowitz, Handbook of Mathematical Functions, Applied Mathematics Series, vol. 55, National Bureau of
Standards, United States, 1964.
[37] See, for example, M.B. Green, J.H. Schwartz, E. Witten, Superstring Theory, vol. 2, Cambridge Univ. Press, 1988.
The chiral ring and the periods of the resolvent

Frank Ferrari
Service de Physique Thorique et Mathmatique, Universit Libre de Bruxelles and International Solvay Institutes,
Campus de la Plaine, CP 231, B-1050 Bruxelles, Belgium
Abstract
The strongly coupled vacua of an N = 1 supersymmetric gauge theory can be described by imposing
quantization conditions on the periods of the gauge theory resolvent, or equivalently by imposing factorization conditions on the associated N = 2 SeibergWitten curve (the so-called strong-coupling approach).
We show that these conditions are equivalent to the existence of certain relations in the chiral ring, which
themselves follow from the fact that the gauge group has a finite rank. This provides a conceptually very
simple explanation of why and how the strongly coupled physics of N = 1 theories, including fractional
instanton effects, chiral symmetry breaking and confinement, can be derived from purely semi-classical
calculations involving instantons only.
1. Introduction
When a four-dimensional N = 2 supersymmetric gauge theory is deformed into an N = 1
theory, many interesting strong coupling effects are expected to occur, like confinement, chiral
symmetry breaking and the creation of a mass gap. Whereas the solution of the parent N = 2 theory is governed by semi-classical instanton effects [1], most vacua of the corresponding N = 1
theory are strongly coupled and cannot be described in semi-classical terms. For example, chiral
observables vacuum expectation values are typically given by fractional instanton series (that
is to say, series for which the expansion parameter is a fractional power of the usual instanton
factor).
Yet, it has been known for a long time that a very simple and consistent description of the
strongly coupled N = 1 vacua could be given in terms of the underlying N = 2 theory [1]. This
E-mail address: frank.ferrari@ulb.ac.be.
doi:10.1016/j.nuclphysb.2007.02.018
372
F. Ferrari / Nuclear Physics B 770 [FS] (2007) 371383
so-called strong coupling approach is based on the fact that at low energy, the parent N = 2
theory is governed by a free Abelian gauge theory. Assuming that the N = 1 theory creates a
mass gap, the N = 2 moduli must be frozen at the singularities of the N = 2 moduli space when
the N = 1 deformation is turned on. This is so because the only way a free Abelian gauge theory
can have a mass gap is through the usual Higgs mechanism, and this mechanism can only occur
when charged fields, that are only present at the singularities, condense. This strong coupling
procedure then implies confinement and chiral symmetry breaking [1], and, when combined
with the generalized Konishi anomaly equations [2], essentially fixes all the correlators of chiral
operators in the theory.
A second, equivalent description of N = 1 theories can be given in the context of the gauge
theory/matrix model correspondence [3,4]. This is an elegant approach that allows to derive most
of the known exact results in the field, including the SeibergWitten solution of N = 2 superYangMills [1,5], and many non-perturbative effects on the space of vacua of N = 1 theories
[69]. On the matrix model side of the correspondence, the most general solution depends on
a set of parameters, the filling fractions, that can be chosen arbitrarily. On the gauge theory
side, however, these filling fractions correspond to the gluino condensates SI , and thus must be
fixed, non-perturbative functions of the parameters (couplings in the tree-level superpotential and
dynamically generated scale or gauge coupling constant).
Understanding the basic principles that fix the filling fractions SI in the gauge theory is a major challenge. The original conjecture was that the filling fractions are determined by extremizing
a certain superpotential, the so-called DijkgraafVafa glueball superpotential WDV (SI ),
WDV
= 0.
SI
(1.1)
This proposal is well motivated by using the gauge/string correspondence (in the dual string
formulation, the DijkgraafVafa superpotential is a flux superpotential [10]), but is difficult to
understand from the field theory point of view. Eq. (1.1) look like complicated dynamical constraints, consistently with the idea that a field theoretic proof would involve a deep understanding
of the non-perturbative gauge dynamics [2]. Actually, we are going to show that a conceptually
simple justification can be found.
A very nice property of Eq. (1.1), pointed out in [9], is that they are mathematically equivalent
to a set of quantization conditions for the periods of the one-form R dz, where R is the gauge
theory resolvent defined by
R(z) = tr
1
.
zX
(1.2)
The adjoint chiral superfield X in the above formula is the superpartner of the vector superfield
in the N = 2 theory. A short proof of this statement is given for example in [11]. More precisely,
the generalized Konishi anomaly equations [2] imply that R dz is a meromorphic differential on
a certain hyperelliptic curve C. If we write down the equation for C in the form
C:
y2 =
d

z aI z aI+ ,
I =1
(1.3)
373
Fig. 1. The non-compact two-sheeted Riemann surface C, with the contours I and I J used in the main text.
and define the contours as in Fig. 1, the quantization conditions read1

R dz 2iZ,
(1.4)
I J
R dz 2iZ.
(1.5)

1
The quantization conditions (1.5) might seem obvious, because 2i
I R dz may be interpreted
+
as giving the number of eigenvalues of the matrix X in the cut [aI , aI ]. However, as we explain
in the next section, (1.5) is non-trivial from a fully non-perturbative point of view. Actually, both
the quantization conditions (1.4) and (1.5) appear on an equal footing in the arguments that we
present in this paper.
So we have two simple, elegant and physically well-motivated procedures to derive the solution of N = 1 gauge theories that are deformations of parent N = 2 gauge theories. It is known
that these procedures are mathematically equivalent. Naively, it is a priori very difficult to justify
these approaches from first principles. This might seem inevitable, since they are at the basis of
the derivation of strongly coupled effects that cannot be described in semi-classical terms.
The main result of the present work is to provide an extremely simple argument, from first
principles, proving directly the quantization conditions (1.4) and (1.5), or equivalently the validity of the strong coupling approach. The main idea of the proof is to concentrate on relations in
the chiral ring that must exist because the gauge group is of finite rank. The existence of these
relations is a basic difference with the associated matrix model, for which the size of the matrix
is infinite. The main point is that the conditions (1.4) and (1.5) are simply equivalent to a particular form of the constraints. Since the constraints are operator relations, they must remain true
in all the vacua of the deformed N = 1 theory if they are established in the N = 2 limit. This is
possible if and only if (1.4) and (1.5) are satisfied in all the N = 1 vacua, or equivalently if and
only if the N = 1 vacua are described by the usual factorized SeibergWitten curves.
We give full details in the case of the N = 2 U(N ) theory deformed by a superpotential term
tr Wtree (X), including when Nf 2N flavors of quarks are present. However, our arguments
are completely general. The same ideas actually apply as well to N = 1 theories that are not
necessarily deformations of N = 2 theories.
The paper is organized as follows. In Section 2, we discuss the constraints in the chiral ring
that follow from the finiteness of the number of colors N . These constraints are at the basis of
the main argument that is explained in Section 3. Finally, in Section 4 we present some open
problems and discuss the relations of the present work with [11].
1 The conditions (1.4) are equivalent to W /S = W /S . As explained in Section 4 of [11], the missing
DV
DV
I
J
equation can then be derived from a standard Ward identity.
374
A note on notations: in the following, we consider expectation values of chiral operators O

in various vacua. The symbol O is used when we do not need to specify a particular vacuum,
typically in expressions that are valid in all the vacua. Relations valid in all the vacua are also
often noted as operator relations2 without brackets. On the other hand, in vacuum-dependent
equations, we always specify explicitly in which vacuum |0 we are working, by using the symbol
0|O|0.
2. On the relations in the chiral ring
Let us consider the U(N ) theory with N = 1 supersymmetry and one adjoint chiral superfield X. The generalization to the theory with flavors is discussed in Section 3.4.
The chiral ring of the theory is generated by the operators [2]
1
1
(2.1)
vk =
tr W W X k ,
tr W X k ,
4
16 2
where W is the N = 1 super-field strength. It is well known that for any N N matrix X, the
traces tr X k for k > N can be expressed in terms of the tr X k for 1 k N . These relations take
the form
uk = tr X k ,
uk =
uN+p = Pcl,p (u1 , . . . , uN ),
p 1,
(2.2)
where the Pcl,p are homogeneous polynomials of degree p + N in the u1 , . . . , uN (uk being
of degree k). Let us emphasize that these relations are simply identities, that follow from the
finiteness of the rank of the gauge group. There are also similar relations relating the uN+p and
the vN+p for p 0 to the u0 , . . . , uN1 , u1 , . . . , uN and v0 , . . . , vN1 , u1 , . . . , uN respectively,
but we do not need them for our analysis.
Non-perturbatively, the relations (2.2) may be modified. The new relations must be consistent
with the U(1)A and U(1)R symmetries of the theory. These symmetries act on the operators, the
parameters gk (defined in (3.9)) and the instanton factor 2N as
U(1)A
U(1)R
uk
uk
vk
2N
gk
k
0
k
1
k
2
2N
0
k 1
2
(2.3)
The U(1)R symmetry implies that the uk cannot mix with the other operators uk or vk , and that
the quantum version of the relations (2.2) cannot depend on the couplings gk . In other words,
we can restrict ourselves (and this will turn out to be sufficient for our purposes) to the sector of
the chiral ring with zero U(1)R charge. This sector is itself a ring A, generated by the uk . The
relations in A can take the general form

uN+p = Pp u1 , . . . , uN ; 2N , p 1,
(2.4)
where the Pp are polynomials of U(1)A charge p + N that go to the classical polynomials Pcl,p
when 2N goes to zero.3
2 By operator relations, we always mean operator relations in the chiral ring.
3 Note that only the instanton factor 2N can enter, by 2 periodicity in the angle, because (2.4) is an operator
relation and is thus valid in all the vacua. Fractional instanton effects do occur in N = 1 theories, but only in relations
that are valid in a particular vacuum (or a particular set of vacua).
375
A fact we would like to emphasize is that the quantum corrections in (2.4) do not represent a
deformation, in any sensible mathematical sense, of the classical chiral ring. The classical chiral ring Acl (in the sector of zero R-charge we are interested in) is simply the polynomial algebra
generated by the uk for 1 k N . The quantum version A of this ring must be commutative
(since only bosonic variables are present) and generated freely by the same elements u1 , . . . , uN .
This implies that
A = Acl = C[u1 , . . . , uN ].
(2.5)
A more abstract way to understand this is to note that because the quantum theory can be seen
as a smooth deformation of the classical theory obtained by turning on the instanton factor 2N ,
the possible deformations of Acl can be studied using the standard deformation theory based
on Hochschild cohomology. It is an elementary result that there is no possible non-trivial deformation of a polynomial algebra that preserves commutativity (see for example [12] for an
elementary exposition). The conclusion is that the zero R-charge sector of the chiral ring is not
quantum corrected.4
So what is the interpretation of the relations (2.4)? They actually represent the definitions
of what we call the uN+p for p 1. These definitions are non-dynamical, and a priori can be
completely arbitrary (as long as they are consistent with symmetries and the classical limit).
This freedom has actually been used in some instances in the literature (for example to match
results obtained by different methods [13,14]). In particular, it is perfectly consistent to define
the uk by the classical relations (2.2), even in the full quantum theory. However, depending on
the context, a natural non-perturbative definition of what is called uN+p for p 1 may involve
quantum-corrected relations of the form (2.4).
For the purpose of our investigations, the natural variables are such that the anomaly equations, that were derived in perturbation theory in [2], take the same form in the full quantum
theory: for all n 1,

N
(2.6)
vq1 uq2 + uq1 uq2 = 0,
gk un+k+1 + 2
q1 +q2 =n
k0
gk un+k+1
+2
k0
k0
gk vn+k+1 +
q1 +q2 =n
vq1 uq2 = 0,
vq1 vq2 = 0.
(2.7)
(2.8)
q1 +q2 =n
Note that this assumption does not mean that the equations do not get non-perturbative corrections, but rather that the non-perturbative corrections can be absorbed in a proper definition of
the variables. We also expect that this definition of the variables is the same as the one that enters
naturally in the context of instanton calculus [14].
Lacking a detailed non-perturbative analysis of the anomaly equations, we shall allow the
relations (2.4) to take the most general possible form a priori. This implies an interesting subtlety
that has been overlooked in previous works. It is clear that the gauge theory resolvent (1.2) does
depend explicitly on the particular definitions of the higher moments uN+p . In particular, there
4 The full chiral ring can be deformed, because for example elements that are nilpotent classically, like the glueball
operator, are not in the quantum theory [2]. However these deformations are not related to the existence of the quantumcorrected relations (2.4).
376
are many consistent definitions, with relations of the form (2.4), that violate the quantization
conditions (1.5). A nice feature of our approach is that we do not need to assume (1.5), and both
quantization conditions (1.4) and (1.5) will be derived at the same time.
3. Relations and the quantization of the periods
3.1. Picking a suitable vacuum
An important point is that the precise form of the constraints (2.4) can be derived from a
purely semi-classical analysis. This is possible because we can always find a vacuum that is both
arbitrarily weakly coupled and suitable to fix the relations (2.4) unambiguously.
For example, the R-symmetry implies that the polynomials Pp in (2.4) cannot depend on the
couplings in the tree-level superpotential. We can thus choose the latter at our convenience. We
pick a degree N + 1 superpotential such that

(x) =
Wtree
N

gk x k = g N
k=0
N
(x aI ) = gN PN (x).
(3.9)
I =1
Classically, the gauge theory has several vacua, depending on the numbers NI 0 of eigenvalues
of the adjoint field X that are taken to be equal to aI . These classical vacua are denoted by
|N1 , . . . , NN and correspond to a pattern of gauge symmetry breaking U(N ) U(N1 )
U(NN ). Let us focus on the Coulomb vacuum
|C = |1, . . . , 1,
(3.10)
U(1)N .
This vacuum can be made arbitrarily weakly

in which the low energy gauge group is
coupled by going to the region |aI aJ | || in the space of parameters. Moreover, in this vacuum, the expectation values C|u1 |C, . . . , C|uN |C are independent, unconstrained variables.5
Equivalently, we can take the gN 0 limit of the N = 2 theory, in which case the u1 , . . . , uN
are moduli. This means that if we can find polynomials Pp such that

C|uN+p |C = Pp C|u1 |C, . . . , C|uN |C; 2N , p 1,
(3.11)
for arbitrary values of the C|uk |C, then we know automatically that
Pp = Pp .
(3.12)
The above reasoning is quite powerful: we are able to derive operator relations by studying the
theory in a particular vacuum. This is a basic feature of our method. It is made possible by the
fact that we know a priori that operator equations of the form (2.4) must exist.
3.2. Example
Let us use the above idea to compute the polynomials Pp in the theory with no flavor.
Eqs. (2.6) and (2.8) can be easily solved (using in particular uk = 0 by Lorentz invariance) to
5 This is not true in general. For example, for a vacuum |0 with an unbroken gauge group and for which all the
eigenvalues of X are equal classically, we have automatically 0|uk |0 = N 1k 0|u1 |0k at the perturbative level. This
shows that there is only one independent variable. In the non-perturbative theory, the relations are modified by fractional
instanton effects, but the number of independent variables do not change.
377
yield a general formula for the gauge theory resolvent expectation value, valid in any vacuum [2].
For the degree N + 1 tree level superpotential (3.9), and introducing degree N 1 polynomials
QN 1 and RN1 (whose precise forms depend on the particular vacuum under consideration),
we have

QN1 (z)
R(z) =
(3.13)
.
PN (z)2 RN1 (z)
Note that since

N
,
R(z)
z z
(3.14)
we know that QN1 (z) = N zN1 + in all the vacua.

A useful property of the Coulomb vacuum for the theory with no flavor is that the U(1)A
symmetry (2.3) implies that the C|uk |C cannot get quantum corrections for k 2N 1. This is
so because in the Coulomb vacuum the quantum corrections are entirely generated by instantons,
and the instanton factor 2N has U(1)A charge 2N . We thus obtain
C|uk |C =
N

aIk ,
1 k 2N 1,
(3.15)
I =1
which is equivalent to the following asymptotic condition,

PN (z)
(3.16)
+ O 1/z2N +1 .
PN (z)

Plugging (3.13) into (3.16), multiplying by PN2 RN1 and expanding at large z immediately
yield

QN1 (z) = PN (z) + O 1/z2 .
(3.17)
C|R(z)|C =
Since both QN1 and PN are polynomials, we must have

QN1 = PN .
(3.18)
Taking this result into account, (3.16) implies that
1
PN (z)2 RN1 (z)

1
+ O 1/z3N .
PN (z)
(3.19)
Inverting this relation, taking the square and expanding at large z then yields
PN (z)2 RN1 (z) = PN (z)2 + O(1),
(3.20)
or equivalently that RN1 must be a constant r,

RN1 (z) = r.
(3.21)
The relation (3.16), and thus (3.18) and (3.21), are of course valid only in the Coulomb vacuum.
We have thus achieved our goal: all the polynomials Pp in (2.4) can be expressed in terms of
the constant r, by expanding C|R(z)|C at large z. For example, the first non-trivial correction
is obtained for PN and reads
PN = Pcl,N +
Nr
.
2
(3.22)
378
Clearly r must be proportional to 2N and, at the expense of rescaling , we can always choose
r = 42N , which is the standard convention.6
3.3. Encoding the form of the relations
In the theory with no flavor, we have been able to compute the polynomials Pp by using a
simple symmetry argument. This would not be the case for more general theories, for example
when a large number of flavors are present. However, the only important point for us is that it is
always possible to do this calculation in a purely semi-classical context (see also Section 4 for
another possible way to derive the relations).
This being said, let us now show that these relations are equivalent to a simple algebraic
equation satisfied by the quantum characteristic function
F (z) = det(z X).
(3.23)
This algebraic relation follow from a simple procedure7 to compute the polynomials uN+p = Pp
recursively. The characteristic function F admits a simple expansion in terms of the uk of the
form

F (z) = zN
(3.24)
Fk zNk ,
k1
where the Fk = uk /k + are polynomials in the uq s of U(1)A charge k. The Fk s can be

computed explicitly by writing
F (z) = det(z X) = zN etr ln(1X/z) = zN e
k1 uk /(kz
k)
(3.25)
and expanding at large z. At the classical level, F is a polynomial of degree N . We thus have
Fk = 0 for all k > N . Since Fk is the sum of uk /k plus terms that depend only on the uq for q < k,
this yields convenient recursion relations that determine all the polynomials Pcl,p . Quantum
mechanically, we can find F in the Coulomb vacuum from the formula
C|R(z)|C =
PN (z)
PN (z)2 42N
(3.26)
derived in the previous subsection, by integrating the relation

d
ln F (z) = R(z)
dz
and using

F (z) zN .
z
(3.27)
(3.28)
This yields
C|F (z)|C =
1
PN (z) + PN (z)2 42N .
2
(3.29)
6 The identification of the constant r with 42N is straightforward in the present case, but it plays no rle in the
following, and in particular is not needed to prove the quantization conditions (1.4) and (1.5).
7 This trick has appeared several times in the literature, for example in [15] and [11].
379
In particular, C|F (z)|C is not a polynomial anymore, but it satisfies a simple quadratic equation
2N
(3.30)
= PN (z).
C|F (z)|C
By expanding at large z, this equation yields the C|uk |C for k > N as a function of the
C|u1 |C, . . . , C|uN |C. To see how this works in details, let us write

1
F (z) = zN
(3.31)
Fk zNk ,
Fk zNk ,
= zN +
F (z)
C|F (z)|C +
k1
k1
and plug these expansions into (3.30). Since the right-hand side is a polynomial, all the terms
with negative powers of z must cancel in the left-hand side, yielding
C|Fp+N |C = 2N C|FpN |C
for all p 1,
(3.32)
with the convention that F0 = 1 and Fk = 0 if k < 0. Since Fk = uk /k + depends only on the
uq for q k, (3.32) generates recursively all the relations (2.4).
Now comes the main point of our argument. Eqs. (3.32) not only provide a simple way to fix
unambiguously the relations (2.4),8 they are actually equivalent to them. Since (2.4) is valid in all
the vacua, it must be so for (3.32). In other words, we have shown that Eqs. (3.32) are operator
relations valid in all vacua,
Fp+N = 2N FpN ,
p 1,
(3.33)
because they simply correspond to a convenient rewriting of the operator relations (2.4). Moreover, since the couplings gk cannot appear in (2.4), we know that (3.33) must be true for any
tree-level superpotential tr Wtree (X), not necessarily of the form (3.9).
Using the operator relations (3.33), we deduce that F (z) + 2N /F (z) has no negative
powers of z in its large N expansion, not only in the Coulomb vacuum but also in all the other
vacua of the N = 1 theory. Equivalently, this implies that

2N
= P (z)
F (z) +
F (z)
(3.34)
is a polynomial in all vacua. The precise form of P (z) = zN + depends on the particular
vacuum because the operator relations (3.32) do not constrain the positive powers in z in the
left-hand side of (3.34).
The fundamental point in our argument is that the algebraic equation (3.34) is not dynamical,
but rather acts as a generating equation for the relations (2.4). It is very important that the algebraic equation satisfied by F contains only this purely kinematical information. Again, this is
why we can derive that the equation must be valid in all the vacua of the gauge theory.
3.4. The quantization conditions
We now have all the necessary ingredients to show that the quantization conditions (1.4) and
(1.5) must always be valid. First of all, from (3.27) and (3.34) we find

1
F (z) = P (z) + P (z)2 42N ,
(3.35)
2
8 Here we use the fact, already emphasized in Section 3.1, that the variables C|u |C, . . . , C|u |C are independent
N
1
in the Coulomb vacuum.
380

R(z) =
P (z)
.
(3.36)
P (z)2 42N
Both F and R are thus meromorphic functions defined on the same Riemann surface C,
independently of the vacuum under consideration. Eqs. (3.35) and (3.36) have a form similar to
(3.29) and (3.26), but the degree N polynomial P is vacuum-dependent and is not equal to PN in
can be any polynomial, not necessarily of the form (3.9). Consistency
general. In particular Wtree

of degree d actually immediately
with the anomaly equations (2.6) and (2.8) for arbitrary Wtree
implies the existence degrees d 1, N d and d d polynomials d1 , FNd and Hdd
respectively such that

Wtree
(z)2 (z) = Hdd (z)2 y 2 ,
P (z)2 42N = FNd (z)2 y 2 .
(3.37)
These are the factorization equations (3.37) at the basis of the strong coupling approach to N = 1
gauge theories! In particular, the curve
Y 2 = P (z)2 42N
(3.38)
is the SeibergWitten curve of the N = 2 theory obtained in the Wtree 0 limit.

The fact that F and R are defined on the same Riemann surface is also all we need to derive the quantization conditions (1.4) and (1.5). Indeed, in general, integrating (3.27) taking into
account (3.28) yields
z

R(z ) dz ,
F (z) = lim N
(3.39)
0 exp
0
where 0 is a point on the first sheet (the sheet for which (3.28) is valid) of the Riemann surface.
This formula shows that F (z) is generically a multivalued function on the curve C on which
R(z) is well-defined, because one must specify the contour from the point at infinity 0 to z to
do the integral in (3.39). The integral representation also shows that F (z) will be single valued
if and only if the quantization conditions

R dz 2iZ,
R dz 2iZ,
(3.40)
I J
are satisfied. We have thus completed the proof of (1.4) and (1.5).
3.5. Generalization to the case with flavors
Let us now add Nf = 2N flavors of fundamental and antifundamental quarks and antiquarks
q . The other cases Nf < 2N can be obtained by integrating out some of the flavors.
Qq and Q
The tree-level superpotential has the form
W = tr Wtree (X) +
2N

q (X mq )Qq .
Q
(3.41)
q=1
The most general classical vacuum |NI ; q is specified by the numbers of eigenvalues of the
matrix X, NI 0 and Q = 0 or 1, that are equal to the I th extremum of Wtree and mq respectively [9]. In particular,

NI +
q = N.
(3.42)
I
381

The pattern of gauge symmetry breaking in a vacuum |NI ; q is U(N ) I U(NI ).
The anomaly equations [16] and associated matrix model [17] for this theory are well known,
and we shall not repeat the details here (a very detailed discussion is included in [4]). The chiral
ring relations are still of the general form (2.4), but the instanton factor is now
h = e2i
(3.43)
and the polynomials Pp can depend symmetrically on the masses mq (but cannot depend on the
couplings in Wtree ).
We can compute the Pp semi-classically by choosing Wtree of the form (3.9) and going to the
Coulomb vacuum |C = |NI = 1; q = 0. Symmetries are no longer enough when Nf = 2N to
fix completely the solution to the anomaly equations in this vacuum, but we can rely on explicit
instanton calculations [18]. Introducing the polynomial
U (z) =
2N
(z mq ),
(3.44)
q=1
it can be shown that the characteristic function (3.23) satisfies

C|F (z)|C +
h U (z)
= P|C (z)
C|F (z)|C
(3.45)
for some degree N polynomial P|C . The conditions obtained by expanding at large z and writing
that the negative powers in z in the left-hand side of (3.45) must vanish determine recursively the
polynomials Pp . Conversely, the operator relations (2.4) then implies that

h U (z)
= P (z)
F (z) +
F (z)
(3.46)
in all the vacua, for a certain vacuum-dependent polynomial P . This in turn yields the quantization conditions (1.4) and (1.5) (or the appropriate factorization of the associated SeibergWitten
curves) in full generality.
4. Conclusion and open problems
In this paper, we have obtained an extremely simple interpretation of the quantization conditions (1.4) and (1.5). These conditions simply encode the precise form of vacuum-independent
operator relations between the chiral observables uk = tr X k . Since the relations are completely
fixed by studying a weakly coupled Coulomb vacuum, the validity of (1.4) and (1.5) in all the
vacua of the N = 1 theory, including the strongly coupled confining vacua, can be derived from
a purely semi-classical analysis. It is particularly startling that such a conceptually simple understanding can be achieved. It yields in particular a straightforward justification from first principles
of the well-known strong coupling approach.
A natural question is whether the same ideas can be used to derive the quantization conditions in the Coulomb vacuum as well, in the most general cases, and independently of the
semi-classical approximation. This seems plausible, because it is absolutely not obvious a priori
that the solutions to the anomaly equations can be consistent with the existence of vacuumindependent relations between the variables. This consistency requirement does not arise in the
similar-looking loop equations of the planar matrix model, because in this case all the variables
are independent. In the gauge theory, it is natural to conjecture that consistency can be achieved
382
if and only if the conditions (1.4) and (1.5) are satisfied in the Coulomb vacuum (and thus in all
the other vacua by the arguments above).
An outstanding open problem is to provide a non-perturbative proof of the generalized Konishi anomalies. The existing derivations are made in perturbation theory, with a fixed classical
background gauge field [2]. Contrary to a statement often made in the literature, the equations do
get non-perturbative corrections, but it is believed that these corrections can be made implicit by
suitably defining the variables, as explained in Section 2. Since the equations must be valid in all
the vacua, it is enough to make a proof in the context of instanton calculus, and this is presently
under investigation.
In [11], it was also shown, from another point of view, that the quantization conditions (1.4)
and (1.5) are not dynamical but rather follow from general consistency conditions. In this respect, [11] and the present work share the same philosophy. The argument of [11] was based
on the analysis of the gauge invariance of the operator F (z) = det(z X) for all values of z.
Gauge invariance turns out to be consistent with the analytic continuation in z if and only if the
conditions (1.4) and (1.5) are satisfied. However, gauge invariance of det(z X) can be achieved
only for some particular definitions of the variables uN+p , and thus is not trivial a priori (this
is similar to the fact that the quantization conditions (1.5) are not trivial a priori). Again, an
explicit non-perturbative definition of det(z X) can certainly be given in the context of the
instanton calculus. As suggested in [11], the standard representation of the determinant in terms
of a fermionic integral is then likely to make gauge invariance manifest. Combined with the nonperturbative analysis of the anomaly equations, the arguments of [11] would then provide a new
and elegant way to sum up instanton series explicitly, independently of the localization methods
used in [18].
Acknowledgements
This work is supported in part by the Belgian Fonds de la Recherche Fondamentale Collective (grant 2.4655.07), the Belgian Institut Interuniversitaire des Sciences Nuclaires (grant
4.4505.86), the Interuniversity Attraction Poles Programme (Belgian Science Policy) and by
the European Commission FP6 programme MRTN-CT-2004-005104 (in association with V.U.
Brussels). The author is on leave of absence from Centre National de la Recherche Scientifique,
Laboratoire de Physique Thorique de lcole Normale Suprieure, Paris, France.
References
[1] N. Seiberg, E. Witten, Nucl. Phys. B 426 (1994) 19, hep-th/9407087;
N. Seiberg, E. Witten, Nucl. Phys. B 430 (1994) 485, Erratum;
N. Seiberg, E. Witten, Nucl. Phys. B 431 (1994) 484, hep-th/9408099.
[2] F. Cachazo, M.R. Douglas, N. Seiberg, E. Witten, JHEP 0212 (2002) 071, hep-th/0211170.
[3] R. Dijkgraaf, C. Vafa, Nucl. Phys. B 644 (2002) 3, hep-th/0206255;
R. Dijkgraaf, C. Vafa, Nucl. Phys. B 644 (2002) 21, hep-th/0207106;
R. Dijkgraaf, C. Vafa, A perturbative window into non-perturbative physics, hep-th/0208048.
[4] F. Ferrari, Supersymmetric gauge theories, matrix models and geometric transitions, Phys. Rep., in press.
[5] F. Cachazo, C. Vafa, N = 1 and N = 2 geometry from fluxes, hep-th/0206017.
[6] F. Ferrari, Phys. Rev. D 67 (2003) 85013, hep-th/0211069.
[7] F. Ferrari, Phys. Lett. B 557 (2003) 290, hep-th/0301157.
[8] F. Cachazo, N. Seiberg, E. Witten, JHEP 0302 (2003) 042, hep-th/0301006.
[9] F. Cachazo, N. Seiberg, E. Witten, JHEP 0304 (2003) 018, hep-th/0303207.
[10] F. Cachazo, K. Intriligator, C. Vafa, Nucl. Phys. B 603 (2001) 3, hep-th/0103067.
383
[11] F. Ferrari, JHEP 0606 (2006) 039, hep-th/0602249.

[12] M. Penkava, P. Vanhaecke, J. Algebra 227 (2000) 365;
M. Penkava, P. Vanhaecke, Commun. Contemp. Math. 3 (2001) 393.
[13] N. Dorey, V.V. Khoze, M.P. Mattis, Phys. Lett. B. 396 (1997) 141, hep-th/9612231.
[14] N. Dorey, T.J. Hollowood, V.V. Khoze, M.P. Mattis, Phys. Rep. 371 (2002) 231, hep-th/0206063.
[15] P. Svrcek, JHEP 0404 (2004) 036, hep-th/0308037.
[16] N. Seiberg, JHEP 0301 (2003) 061, hep-th/0212225.
[17] R. Argurio, V.L. Campos, G. Ferretti, R. Heise, Phys. Rev. D 67 (2003) 65005, hep-th/0210291.
[18] N. Nekrasov, Adv. Theor. Math. Phys. 7 (2004) 831, hep-th/0206161;
N. Nekrasov, SeibergWitten prepotential from instanton counting, in: Proceedings of the International Congress
of Mathematicians (ICM 2002), hep-th/0306211;
N. Nekrasov, A. Okounkov, SeibergWitten theory and randon partitions, hep-th/0306238.

Nucl - Phys.B v.770

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Nucl - Phys.B v.770

Загружено:

Авторское право:

Доступные форматы

NUCLEAR PHYSICS B

Journal devoted to the experimental and theoretical study of the fundamental

Nuclear Physics B 770 (2007) 146

Moduli stabilization in non-geometric backgrounds

Received 24 November 2006; accepted 17 January 2007

E-mail address: walcher@ias.edu (J. Walcher).

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

2.1. The 19 /Z3 Gepner model

by a Z3 action which sends (x1 , x2 , x3 ) (x1 , x2 , x3 ). This corresponds to a

K. Becker et al. / Nuclear Physics B 770 (2007) 146

Z3 orbifold group defining our LG model,

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

In our present setup, which is non-geometric, there is no straightforward notion of orientifold

K. Becker et al. / Nuclear Physics B 770 (2007) 146

Fig. 1. The piece of cake picture of Lagrangian (A-type) D-branes in LG model x 3 .

as can be seen from the figure.

K. Becker et al. / Nuclear Physics B 770 (2007) 146

equivalently represented by the chiral ring C[x]/x 2 = 1, x

K. Becker et al. / Nuclear Physics B 770 (2007) 146

or, restricted to those [n] with representatives with all ni = 0, 1,

The same result can also be expressed in a form similar to (2.21)

2.3.4. B-branes and matrix factorization

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

, yielding a minimal basis. It is also not hard

2.4. Charges of O-planes

with k = 1, we notice that (g 2 0 )2 = g and the eigenvalues of g 2 0 are (2 , . . . , 2 ) for

O-plane, we here simply mean the abstract world-sheet concept.

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

where a is a basis of harmonic (2, 1) forms, leads to the conclusion

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

ij qLi qR + c.c. HRR ,

K. Becker et al. / Nuclear Physics B 770 (2007) 146

malization of the criticality of the potential, namely the condition

where ti = ti = 0 is the solution. Quantum corrections to i K evaluated at ti = ti = 0 will affect

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

3.4. Tadpole cancellation condition

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

(Again, these are represented by permutation branes.)

K. Becker et al. / Nuclear Physics B 770 (2007) 146

The Hodge decomposition of

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

K. Becker et al. / Nuclear Physics B 770 (2007) 146

000100110 000101001 + 000101010 + 001000101 001000110

= +000000101 000000110 000001001 + 000001010 + 000010101

+ 001011010 001100101 + 001100110 + 001101001 001101010 ,

010100110 010101001 + 010101010 ,

= 000010101 + 000010110 + 000011001 000011010 + 000100101

and finally, for

000010000 010010000 100010000 001010000 + 111010000 ,

= 010000001 100000001 110000001 001000001 011000001

ij qLi qR + c.c. HRR ,