Nucl - Phys.B v.589

Nuclear Physics B 589 (2000) 3–37
www.elsevier.nl/locate/npe
Near-extremal correlators and vanishing

supergravity couplings in AdS/CFT
Eric D’Hoker a , Johanna Erdmenger b , Daniel Z. Freedman b,c ,
Manuel Pérez-Victoria b
a Department of Physics, University of California, Los Angeles, CA 90095, USA
b Center for Theoretical Physics, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
c Department of Mathematics, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
Received 11 April 2000; accepted 17 August 2000
Abstract
We study near-extremal n-point correlation functions P of chiral primary operators, in which the
maximal scale dimension k is related to the others by k = i ki − m with m 6 n − 3. Through order
g 2 in field theory, we show that these correlators are simple sums of terms each of which factors
into products of lower-point correlators. Terms which contain only factors of two- and three-point
functions are not renormalized, but other terms have non-vanishing order g 2 corrections.
We then show that the contributing AdS exchange diagrams neatly match this factored structure. In
particular, for n = 4, 5 precise agreement in form and coefficient is established between supergravity
and the non-renormalized factored terms from field theory. On the other hand, contact diagrams
in supergravity would produce a non-factored structure. This leads us to conjecture that the
corresponding bulk couplings vanish, so as to achieve full agreement between the structure of these
correlators in supergravity and weak-coupling field theory.  2000 Elsevier Science B.V. All rights
reserved.
1. Introduction
The AdS/CFT correspondence [1–3] has enabled the exact calculation of many
correlation functions in a strong coupling limit of N = 4 SU(N) supersymmetric Yang–
Mills theory and to the initially surprising fact that many correlators appear to be non-
renormalized strong coupling results agree with free field limits. Although it is debatable
whether a rigorous proof has been achieved, there is ample evidence from an interplay
E-mail addresses: dhoker@physics.ucla.edu (E. D’Hoker), jke@mitlns.mit.edu (J. Erdmenger),

dzf@math.mit.edu (D.Z. Freedman), manolo@pierre.mit.edu (M. Pérez-Victoria).
0550-3213/00/$ – see front matter  2000 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 3 4 - 4
4 E. D’Hoker et al. / Nuclear Physics B 589 (2000) 3–37
of arguments from AdS supergravity, order g 2 and g 4 calculations and formal non-
perturbative considerations in the field theory that non-renormalization holds for all values
of g and N and for general gauge groups.
This line of investigation is continued in the present paper in which we present new
results on the structure of near-extremal n-point correlators. We consider the chiral primary
operators Ok = Tr Xk and study order m sub-extremal n-point functions hOk Ok1 · · · Okn−1 i
with k = k1 + · · · + kn−1 − 2m and 0 6 m 6 n − 3. We shall call such correlators Enm
functions. As we will review in more detail below, previous studies strongly suggest
that (extremal) En0 and (next-to-extremal) En1 functions for n > 3 are not renormalized.
On the field theory side these correlators are characterized by the factorization of their
free-field graphs into products of two- and three-point structures. Order g 2 radiative
corrections and Yang–Mills instanton corrections vanish [4,5]. For n > 4 the contributing
exchange diagrams from Type IIB supergravity on AdS5 ×S 5 reproduce the factored space–
time form, but contact diagrams involving quartic or higher order vertices do not. Thus
supergravity En0 and En1 couplings should vanish, and this has been verified [6] for n = 4.
For near-extremality, m > 2, the situation is more complex as we now exemplify for
the case m = 2. Here the E42 correlator hO2 O2 O2 O2 i is known to be renormalized to
order g 2 in field theory [7,8] (for a recent calculation to three loops see [9]) and in the
large N strong coupling limit in supergravity [10–12]. Indeed, E42 correlators have no
special factored free-field limits, which suggests that one must examine E52 functions to
find non-renormalization. It is easy to see that the free-field graphs for the E52 function
hO4 O2 O2 O2 O2 i are of two distinct types; see Fig. 1. The first is a product of two three-
point structures while the second is a product of a two-point and four-point structure. We
show that the order g 2 radiative corrections to the first type vanish, but corrections to the
second type do not vanish as might be expected because there is a four-point sub-structure
present. Therefore the party is over as far as complete non-renormalization is concerned,
but the factored structure is preserved by radiative corrections and this suggests that we
look at the situation in supergravity.
Indeed, it is very striking that the structure found at weak coupling is exactly mirrored at
strong coupling in supergravity. In particular one of the two contributing double-exchange
diagrams in supergravity reproduces the first factored structure of free field theory both
in space–time form and with exactly the same numerical strength. The second double
exchange and both single exchange diagrams reproduce the second factored structure from
field theory, but contain a non-trivial four-point sub-structure which cannot be directly
compared with weak coupling. Full agreement in structure between supergravity and
weak field theory requires that the supergravity five-point coupling corresponding to the
s4 s2 s2 s2 s2 vertex vanishes. And indeed it must vanish for consistent decoupling of the
multiplet containing the bulk field s4 dual to O4 from the graviton multiplet. This is
required by consistent Kaluza–Klein truncation of Type IIB supergravity on AdS5 × S 5 ,
which means that the solutions to the equations of motion of N = 8 gauged supergravity
on AdS5 are exact solutions of the complete Type IIB supergravity theory on AdS5 ×S 5 [13,
14]. This property forbids terms in the full action which both contain the graviton multiplet
and are linear in a field of a higher Kaluza–Klein multiplet.
E. D’Hoker et al. / Nuclear Physics B 589 (2000) 3–37 5
We show that the example of hO4 O2 O2 O2 O2 i generalizes to all E52 functions (provided
an extra assumption about descendent couplings holds). The factored structure of weak-
coupling field theory agrees with supergravity calculations if all E52 bulk couplings vanish,
which is our prediction. There is a further extension of these results to all Enm functions with
m 6 n − 3: Supergravity reproduces the factored space–time form of the field-theoretical
calculation if bulk Enm couplings vanish. This leads us to conjecture that these couplings
do indeed vanish.
We shall now give a detailed summary of the non-renormalization properties of
N = 4 SYM correlators which have been established through study of the AdS/CFT
correspondence.
1. Two- and three-point functions of single-trace chiral operators. The first result of this
type came in [15] where a calculation of cubic couplings gk1 k2 k3 of the bulk fields sk dual to
the operators Ok was combined with results [16] for AdS three-point integrals. Operators
were normalized to have unit two-point functions, and it was observed that supergravity
results for all three-point functions hOk1 Ok2 Ok3 i agreed with free field theory. This was
soon followed by an explicit study of order g 2 radiative corrections in field theory which
were shown to vanish for both two-point functions hOk Ok i and three-point functions of
non-normalized operators [17]. A superspace calculation may be found in [18]. Recently
it was shown [19] that order g 4 contributions to hO3 O3 i vanish.
There have been many attempts to prove the non-renormalization of the three-point
functions non-perturbatively in field theory. All approaches require unproved technical
assumptions. Most convincing are the arguments using N = 2 analytic superspace [20,
21] that there are no possible superspace forms which could contribute to the derivative of
a three-point function [22] with respect to the gauge coupling (see also [23]).
2. Two- and three-point functions of other chiral operators. Related results have emerged
from studies [24] of short representations [25] of the conformal superalgebra SU(2, 2|4) of
the N = 4 SYM theory. The chiral primaries Ok are 1/2-BPS operators which transform
in the [0, k, 0] representation of the R-symmetry group SU(4). There are other 1/2-BPS
multitrace operators in the same representations, such as the projection of : Tr Xk1 Tr Xk2 :
in the [0, k1 + k2 , 0] representation, and all of them have protected dimension ∆k = k. It is
therefore of interest to check the non-renormalization properties of these operators, and it
was shown in [26] that order g 2 radiative corrections to their two- and three-point functions
vanish for all gauge groups. The same question can be asked of the 1/4-BPS operators of
dimension p + 2q in the SU(4) representation [q, p, q] and the 1/8-BPS operators of
dimension p + 2q + 3r in the representation [q, p, q + 2r]. This is work in progress. See
also [27,28].
3. Extremal and next-to-extremal functions. The study of En0,1 functions of chiral
primaries for n > 4 emerged from a peculiarity of E30 functions noted in [29]. It was shown
that the factored form of free field theory was reproduced in supergravity as the product
of a zero bulk coupling constant with an infinite AdS integral. The product was defined
by analytic continuation in the dimensions ∆k of the operators, and this procedure was
justified in a related example by careful consideration of boundary interactions. It was then
noticed [29] that En0 functions had factored space–time forms for all n > 4. All contributing
supergravity diagrams, each defined by analytic continuation, give the same factored form.
The coefficient of this form could then be shown to vanish by OPE arguments. After order
g 2 perturbative test in [4], non-perturbative N = 2 and N = 4 superspace arguments
were given to support this result and to suggest that En1 next-to-extremal correlators are
not renormalized [30]. This was then shown for any n through order g 2 [5] and in AdS
supergravity [5,6].
4. Near-extremal functions. This brings us to the situation of Enm functions for 2 6 m 6
n − 3 which we have summarized at the beginning of this introduction with supporting
arguments to be given below.
The paper is organized as follows. In Section 2 we present our results by considering
the simplest possible case, which is the correlator hO4 O2 O2 O2 O2 i. We then move on to
general Enm functions, which we discuss from the field theory point of view in Section
3 and from the supergravity side in Section 4, paying particular attention to supergravity
couplings in Section 5. There is a short conclusion and an Appendix which contains a
detailed consideration of the AdS calculations essential to our argument.
2. The correlator hO4 O2 O2 O2 O2 i
In order to illustrate the structure of En2 correlation functions in the simplest possible
situation, we study in detail the correlator hO4 (x)O2 (x1 )O2 (x2 )O2 (x3 )O2 (x4 )i. We first
calculate the order g 2 corrections in SYM and then the corresponding diagrams in AdS
supergravity. Although non-trivial radiative corrections appear at order g 2 , their factored
form is compatible with supergravity provided the bulk s4 s2 s2 s2 s2 coupling vanishes, as
required by consistent Kaluza–Klein truncation.
2.1. hO4 O2 O2 O2 O2 i in SYM
We shall use the methods of [5] (where the interested reader can find more details). We
normalize the operators as in [15]:
(2π)k
Ok (x) = √ Tr Xk (x) . (2.1)
N k/2 k
With this normalization the two point function is given by

1
Ok (x)Ok (y) = . (2.2)
(x − y)2k
At the free-field level there are two connected graphs, up to permutations of the operators
O2 (Fig. 1). Graphs a and b give the contributions

O4 (x)O2 (x1 )O2 (x2 )O2 (x3 )O2 (x4 ) a
1 1 1 1 1 1
= C (a)Q , (2.3)
(x − x1 ) (x1 − x2 ) (x2 − x) (x − x3 ) (x3 − x4 ) (x4 − x)2
2 2 2 2 2
Fig. 1. Feynman graphs contributing to the correlator hO4 O2 O2 O2 O2 i at the free-field level.

O4 (x)O2 (x1 )O2 (x2 )O2 (x3 )O2 (x4 ) b
1 1 1 1 1
= C (b) Q , (2.4)
(x − x1 )4 (x − x2 )2 (x2 − x3 )2 (x3 − x4 )2 (x4 − x)2
respectively, where C (a,b) are tensors in flavour space, and
Q = Str(T a1 · · · T a4 ) Str(T a1 T a2 ) Str(T a3 T b ) Str(T b T c ) Str(T c T a4 )
= Str(T a1 · · · T a4 ) Str(T a1 T b ) Str(T b T a2 ) Str(T a3 T c ) Str(T c T a4 )
N2 − 1
= . (2.5)
26
The symmetric trace is defined as
X 1
Str(T a1 · · · T ak ) = Tr(T aσ (1) · · · T aσ (k) ). (2.6)
perms σ
k!
In the following we suppress explicit flavour tensors. Our order g 2 calculations are valid
for all N , but for later comparison with supergravity, we write the large N limit of all non-
vanishing contributions. Thus we have, for example, for the contribution of the graph a in
Fig. 1,

O4 (x)O2 (x1 )O2 (x2 )O2 (x3 )O2 (x4 ) a
4

= O2 (x)O2 (x1 )O2 (x2 ) O2 (x)O2 (x3 )O2 (x4 ) , (2.7)
N
where
√

2 2 1
O2 (x)O2 (x1 )O2 (x2 ) = . (2.8)
N (x − x1 ) (x1 − x2 )2 (x2 − x)2
2
The coefficients in (2.7) and (2.8) incorporate both Wick combinatoric factors and the
normalization factors of (2.1).
Let us now consider the connected Feynman graphs contributing at order g 2 , which are
depicted in Fig. 2. The dashed lines denote the combination of gauge boson exchanges
and quartic scalar interactions (see [5] for details). Graphs a1, b1 vanish due to the well-
known non-renormalization theorems for the functions hO2 O2 i and hO2 O2 O2 i. These
theorems hold independently of colour contractions, see [26] and the further explanations
Fig. 2. Graphs contributing to the correlator hO4 O2 O2 O2 O2 i at order g 2 .
below Eq. (13) in [5]. All the remaining graphs but b2 vanish due to the symmetry
properties of colour indices. Let us show this explicitly for a3:
a3 ∼ Str(T a1 · · · T a4 ) Str(T b T c ) Str(T c T a4 ) Str(T d T a2 ) Str(T e T a3 )

× A(a3)f a1 bp f dep + B (a3) f a1 dp f bep + C (a3) f a1 ep f bdp
= Str(T a1 · · · T a4 )

× A(a3)f a1 a4 p f a2 a3 p + B (a3)f a1 a2 p f a4 a3 p + C (a3)f a1 a3 p f a4 a2 p , (2.9)
which vanishes as each term is a contraction of a symmetric with an antisymmetric tensor.
Graphs a2, a4, b3, b4, c1, c2 and c3 can be shown to vanish in the same way. Therefore,
the only possible contribution is that of graph b2, which factors into a two-point and a
four-point function. For large N , the contribution of b2 is given by

O4 (x)O2 (x1 )O2 (x2 )O2 (x3 )O2 (x4 ) b2
4

(1)
= O2 (x)O2 (x1 ) O2 (x)O2 (x2 )O2 (x3 )O2 (x4 ) , (2.10)
N
where the index (1) indicates order g 2 . We do not indicate the order in two- and three-
point functions as they are not renormalized. The second factor was calculated and shown
to contain logarithms in [7–9]. Therefore the full five-point correlator is renormalized at
order g 2 . Nevertheless, the contribution does factor. Hence to order g 2 , the complete E52
correlator is given by the sum of the two factored contributions (2.7) and (2.10),

O4 (x)O2 (x1 )O2 (x2 )O2 (x3 )O2 (x4 )
4

= O2 (x)O2 (x1 )O2 (x2 ) O2 (x)O2 (x3 )O2 (x4 )
N
4

+ O2 (x)O2 (x1 ) O2 (x)O2 (x2 )O2 (x3 )O2 (x4 ) + perms. (2.11)
N
The two-point and three-point structures are not renormalized. This factored space–time
structure is a consequence of the properties of N = 4 SYM. Conformal invariance alone
permits a more general structure.
2.2. hO4 O2 O2 O2 O2 i in AdS
According to the Maldacena conjecture, the same function can be calculated (at
strong coupling and large N ) using classical Type IIB supergravity on AdS5 × S 5 . The
corresponding connected Witten diagrams are shown in Fig. 3, up to permutations. We refer
to the bulk fields dual to the operator Ok as sk . There are double-exchange (a and b), single-
exchange (c and d) and contact diagrams (e). The cubic and quartic couplings we need were
calculated in [15,31,32] and [33], respectively. The main property we need for our purposes
is that the couplings of the vertices s2 s2 s4 , s2 s2 s2 s6 and s2 s2 s2 s4 vanish (although allowed
by SU(4) symmetry). By “vanishing coupling” we mean that the combination of derivative
and non-derivative terms in the vertex give a vanishing net coupling in the bulk. It is
important to note that even if the coupling vanishes in the bulk, there can be a contribution
from a surface term if the space–time integral is divergent [29]. In [29] it was shown
(for a particular case) that this boundary contribution is also obtained when couplings
and integrals are regularized by analytic continuation. Moreover, analytic continuation has
successfully produced results which agree with field theory [5,15,29]. We use this method
here as well and regularize the divergent integrals by analytic continuation in the highest
conformal dimension, k = ∆ = 4 → 4 − ε, which also implies that the vanishing couplings
give rise to factors of ε. 1 The convention of normalized two-point functions [15] requires
that AdS integrals must be divided by a product of factors
√
4N 1 k (k − 1)(k − 2)
Nk = 2 k/2 , k > 2,
π 2 (k + 1)
1 An alternative procedure would be to analyze the implications of field redefinitions removing the surface
terms, in the spirit of [6].
Fig. 3. Witten diagrams contributing to E52 correlators in the AdS/CFT correspondence.
4N 1
N2 = √ , (2.12)
π2 2 · 3
one for each external sk line 2 . For the couplings G we also use the expressions given in
[15], and the Poisson kernels Kk (x, z) are given in Appendix A.
Let us first consider the exchange of primary fields. We begin with diagram a. The
exchanged fields φ, φ 0 have to be in the representation with Dynkin labels [0, δ, 0],
[0, δ 0 , 0], respectively, with δ, δ 0 restricted to 2 and 4 in the present example. The
contribution of the exchange diagram for generic δ, δ 0 is given by

O4 (x)O2 (x1 )O2 (x2 )O2 (x3 )O2 (x4 ) a
ZZZ 5 5 5
G(4, δ, δ 0 )G(δ, 2, 2)G(δ 0 , 2, 2) d yd zd w
= K4 (x, y)Gδ (y, z)
N4 N2 4
y05 z05 w05
× K2 (x1 , z)K2 (x2 , z)Gδ 0 (y, w)K2 (x3 , w)K2 (x4 , w). (2.13)
If δ = δ 0 = 2, the vertex y is extremal, such that the corresponding coupling vanishes [15],
and it is adjacent to the boundary operator with the highest conformal dimension. It is
crucial to note here and for the subsequent discussion that in this case the integral over y
is divergent, with the dominant contribution arising when y ∼ x. As shown in detail in the
Appendix, using analytic continuation ∆ → ∆ − ε for the highest conformal dimension,
2 The normalization for k = 2 is in agreement with the discussion in [34].

Fig. 4. Non-vanishing contributions to E52 functions in AdS supergravity. Diagram d0 does not appear
for hO4 O2 O2 O2 O2 i.
both for the extremal coupling and to evaluate the y integral, we find an unambiguous finite
expression for ε → 0.
The remaining integrals over z and w are finite and, as represented in diagram a 0 of
Fig. 4, factor into two three point functions
Z

G(2, 2, 2) d5 z
O2 (x)O2 (x1 )O2 (x2 ) = K2 (x, z)K2 (x1 , z)K2 (x2 , z), (2.14)
N2 3 z05
which may be evaluated using [16] with the result
√

2 2 1
O2 (x)O2 (x1 )O2 (x2 ) = . (2.15)
N (x − x1 )2 (x1 − x2 )2 (x2 − x)2
Putting all the factors together and inserting the explicit values for the supergravity
couplings, we obtain for the five-point function contribution (2.13)

O4 (x)O2 (x1 )O2 (x2 )O2 (x3 )O2 (x4 ) a
4

= O2 (x)O2 (x1 )O2 (x2 ) O2 (x)O2 (x3 )O2 (x4 ) , (2.16)
N
which is in exact agreement in form and value with the free-field contribution (2.7). Later
we show that this is valid for general conformal dimensions as well, not only in the simple
case considered here as an introduction. This is both a non-trivial test of the AdS/CFT
correspondence and of the method of analytic continuation. Of course, this explicit
comparison is possible since the three point functions involved are not renormalized, and
there is exact agreement of the functional form between weak and strong coupling. As far
as the factors are concerned, we see that the field theory large N Wick factors agree with
the factors obtained from the supergravity couplings, the normalization of the propagators
and the integral evaluated using analytic continuation.
If δ + δ 0 > 4 in (2.13), at least one of the remaining vertices is extremal and its coupling
vanishes. The integral, however, is finite and hence the diagram vanishes.
Let us now consider diagram b, where the exchanged field φ has to be in the
representation [0, δ, 0] with δ = 2, 4, 6 and conformal dimension δ. If δ = 2, the cubic
vertex at y is again extremal and the corresponding coupling vanishes. Again, there is a
zero-times-infinity situation, such that a finite contribution emerges from the region y ∼ x:

O4 (x)O2 (x1 )O2 (x2 )O2 (x3 )O2 (x4 ) b
X ZZZ 5 5 5
1 0 0 d yd zd w
= G(4, 2, 2)G(δ , 2, 2)G(δ , 2, 2)
N4 N24 δ 0 =2,4 y05 z05 w05
× K4 (x, y)K2 (x1 , y)G2 (y, z)K2 (x2 , z)Gδ 0 (z, w)K2 (x3 , w)K2 (x4 , w)
√
2 2
1 X
= O2 (x)O2 (x1 ) 3 G(δ 0 , 2, 2)G(δ 0 , 2, 2)
N N2 δ 0 =2,4
ZZ 5 5
d zd w
× K2 (x, z)K2 (x2 , z)Gδ 0 (z, w)K2 (x3 , w)K2 (x4 , w)
z05 w05
4

= O2 (x)O2 (x1 ) O2 (x)O2 (x2 )O2 (x3 )O2 (x4 ) b . (2.17)
N
Again there is agreement between the factors in (2.17) and in (2.10). However, since the
four-point functions are renormalized, it is not possible to compare them directly: In (2.10)
we have an order g 2 contribution at weak coupling, whereas in (2.17) we have an exchange
contribution at strong coupling. Diagram b0 of Fig. 4 represents the space–time structure
of Eq. (2.17). The integrals on z and w are finite.
If δ = 4 (such that the vertex at y is next-to-extremal), the field φ 0 is in the representation
[0, δ 0 , 0] with δ 0 = 2 or δ 0 = 4. Then either the vertex at z or the vertex at w is extremal
and hence has a vanishing coupling. If δ = 6, the only possibility is δ 0 = 4 and the three
vertices are extremal. On the other hand, as shown in the appendix, the integrals are finite
in all these cases. Therefore the corresponding diagrams vanish.
The four-point factor in (2.17) corresponds to an exchange diagram. In diagram c we
can have δ = 2, 4, 6. If δ = 2 the vertex y is extremal and the diagram factors:

O4 (x)O2 (x1 )O2 (x2 )O2 (x3 )O2 (x4 ) c
√ Z 5
2 2
d z
= O2 (x)O2 (x1 ) G(2, 2, 2, 2) K2 (x, z)K2 (x2 , z)K2 (x3 , z)K2 (x4 , z)
N z05
4

= O2 (x)O2 (x1 ) O2 (x)O2 (x2 )O2 (x3 )O2 (x4 ) c , (2.18)
N
where the four-point function is the contact contribution to the E42 function. The factored
structure is shown in diagram c0 of Fig. 4. If δ = 4 there is a next-to-extremal quartic vertex
at z. The diagram gives zero since the z-integral is finite (Appendix A) and the coupling
vanishes. If δ = 6 both vertices have vanishing couplings as they are extremal and the
integral is finite, so the diagram vanishes as well.
In diagram d both δ = 2, 4 are possible. In both cases the vertex at y is sub-extremal and
the integral finite. If δ = 2, the quartic coupling at y is next-to-extremal and has a vanishing
coupling. If δ = 4 both vertices are extremal and have vanishing couplings. Therefore this
diagram does not contribute.
Let us now study the same set of diagrams when SU(2, 2|4) descendents are exchanged.
We will show that descendent contributions to diagrams a − d of Fig. 3 all vanish or give
corrections to the four-point structure in (2.17). Since the discussion is very technical,
the reader interested only in the main flow of the argument may wish to proceed to the
summary paragraph below.
SU(4) flavour symmetry restricts the quantum number of internal lines according to the
Clebsch–Gordan decomposition (q 6 k)
M M
q q−µ
[0, k, 0] ⊗ [0, q, 0] = [ν, k + q − 2µ − 2ν, ν]. (2.19)
µ=0 ν=0
In the present application [0, k, 0] and [0, q, 0] are external chiral primary fields dual to
operators of dimension k and q, respectively, while the exchanged fields in representations
[ν, j, ν] on the right side are either primaries (for ν = 0) of dimension ∆φ = j or
descendents (for ν > 0) of dimension ∆φ > j + 2ν. A descendent must descend from a
chiral primary in the representation [0, l, 0] with ∆φ > l > j + 2ν.
We shall also use a fact which follows from the unique representation of given three-
point function as an extended superspace invariant. All component bulk couplings sk sq φ,
sk sq φ 0 , sk φφ 0 are then related by extended supersymmetry to those of the scalar primaries
in the same multiplet, sk sq s̃, sk sq s̃ 0 , sk s̃ s̃ 0 , where s̃ and s̃ 0 are the chiral primaries from
which φ and φ 0 descend. In particular, descendent couplings vanish if the associated
primary vertex vanishes because of extremality. The final link of our argument is the
convergence or divergence of the AdS integrals, as discussed in the Appendix.
We start with diagram b, where SU(4) symmetry allows the combinations of exchanged
fields φ, φ 0 in the following representations:
φ = [0, 2, 0] → φ 0 = [0, 0, 0], [0, 2, 0], [1, 0, 1], [0, 4, 0], [1, 2, 1], [2, 0, 2],
φ = [0, 4, 0] → φ 0 = [0, 2, 0], [0, 4, 0], [1, 2, 1],
φ = [1, 2, 1] → φ 0 = [0, 2, 0], [1, 0, 1], [0, 4, 0], [1, 2, 1], [2, 0, 2],
φ = [0, 6, 0] → φ 0 = [0, 4, 0],
φ = [1, 4, 1] → φ 0 = [0, 4, 0], [1, 2, 1],
φ = [2, 2, 2] → φ 0 = [0, 4, 0], [2, 0, 2]. (2.20)
If φ is a chiral primary in the [0, 2, 0], while φ 0 is any descendent, the exchange diagrams
are infinite and made finite by analytic continuation of the highest dimension 4 → 4 − ε
both in the integrand and in the vanishing extremal coupling. The limit ε → 0 reduces
all diagrams to b0 of Fig. 4 and results in a correlation function of the structure (2.17),
i.e., the product of the two-point function hO2 O2 i and a radiatively corrected four-point
function hO2 O2 O2 O2 i. In all other cases in (2.20), φ has scale dimension ∆φ > 2, and the
AdS integral is finite. On the other hand at least one of the couplings vanishes because the
associated primary vertex is either extremal or forbidden by SU(4) symmetry.
For descendents in diagram a of Fig. 3, the allowed representations are
φ = [0, 0, 0] → φ 0 = [0, 4, 0],
φ = [0, 2, 0] → φ 0 = [0, 2, 0], [0, 4, 0], [1, 2, 1],
φ = [1, 0, 1] → φ 0 = [1, 2, 1], [0, 4, 0],
φ = [0, 4, 0] → φ 0 = [0, 0, 0], [0, 2, 0], [1, 0, 1], [0, 4, 0], [1, 2, 1], [2, 0, 2],
φ = [1, 2, 1] → φ 0 = [0, 2, 0], [1, 0, 1], [0, 4, 0], [1, 2, 1], [2, 0, 2],
φ = [2, 0, 2] → φ 0 = [0, 4, 0], [1, 2, 1], [2, 0, 2]. (2.21)
The case where φ and φ 0 are chiral primaries in the [0, 2, 0] is the no-descendent case
considered above. In all other cases ∆φ + ∆φ0 > 4 and the integrals are finite, while at least
one of the (associated) couplings is extremal and vanishes. Hence there are no descendent
contributions to diagram a.
In diagram c, the exchanged field φ can be in the representations [0, 2, 0], [0, 4, 0],
[1, 2, 1], [0, 6, 0], [1, 4, 1] and [2, 2, 2]. In all cases except the [0, 2, 0] primary discussed
above, we have descendent or primary fields in excited Kaluza–Klein multiplets. The
quartic couplings φs2 s2 s2 to the chiral primary of the graviton multiplet then vanish by
consistent truncation. Since the integral is finite, all of these contributions vanish.
In diagram d, φ can be in any of the representations [0, 0, 0], [0, 2, 0], [1, 0, 1],
[0, 4, 0], [1, 2, 1] and [2, 0, 2]. Integrals are always convergent. If φ comes from a chiral
primary in the [0, δ̃, 0] with δ̃ > 3, the coupling at the lower vertex vanishes because the
associated vertex of chiral primaries is forbidden or extremal (or alternatively by consistent
truncation). If, on the other hand, δ̃ = 2, φ is in the multiplet of the graviton and consistent
truncation forbids the upper vertex.
Summarizing, we have seen that the descendent-exchange diagrams only contribute to
the hO2 O2 O2 O2 i factor in the reduced diagram b0 . Observe that diagrams b0 and c0 in
Fig. 4 have the same factored structure as graphs b and b2 in the SYM calculation (Figs. 1
and 2), whereas diagram a0 in Fig. 4 has the same structure as graph a in Fig. 1. Adding all
the non-vanishing contributions, which are given by a0 , b0 and c0 , we find

O4 (x)O2 (x1 )O2 (x2 )O2 (x3 )O2 (x4 ) exchange
4

= O2 (x)O2 (x1 )O2 (x2 ) O2 (x)O2 (x3 )O2 (x4 )
N
4

+ O2 (x)O2 (x1 ) O2 (x)O2 (x2 )O2 (x3 )O2 (x4 ) AdS + perms. (2.22)
N
hO2 (x)O2 (x2 )O2 (x3 )O2 (x4 )iAdS is the full AdS four-point correlator. It includes exchange
(diagram b0 ) and contact (diagram c0 ) contributions. This function has been calculated
very recently in [10]. We note that the AdS and field theory four-point functions cannot
be compared directly since they are renormalized correlators at strong and weak coupling
respectively. Nevertheless, comparing Eqs. (2.22) and (2.11), we see that the same general
space–time structure is found in both the order-g 2 field theory calculation and the exchange
diagram contribution to the AdS calculation.
Finally, there is a contact contribution given by diagram e of Fig. 3:

O4 (x)O2 (x1 )O2 (x2 )O2 (x3 )O2 (x4 ) e
Z 5
d y
∼ G(4, 2, 2, 2, 2) K4 (x, y)K2 (x1 , y)K2 (x2 , y)K2 (x3 , y)K2 (x4 , y). (2.23)
y05
As shown in Appendix A, this integral is finite. Furthermore, it does not have the factored
form we have just discussed: it is not a product of a free function times another factor.
Hence, the entire AdS calculation of this five-point function does not give the same space–
time structure we found at order g 2 unless the coupling G(4, 2, 2, 2, 2) vanishes. On the
other hand this coupling does vanish according to consistent truncation. Therefore, if
consistent truncation holds, the space–time structure in the full AdS calculation agrees with
the field theory results. This, together with the results for extremal and next-to-extremal
functions, suggests that the factorization properties of (sub)-extremal correlators may be
extrapolated from weak to strong coupling, at least in the large N limit. This is our basic
assumption in order to derive the vanishing of certain supergravity couplings.
3. General near-extremal correlators in field theory
In this section we study near-extremal correlation functions in N = 4 SYM to order g 2

and show that all the Feynman graphs contributing to an Enm function factor into at least
n − m − 1 pieces which have only one point in common: the point at which the highest-
dimension operator is inserted. In the following, “factor” (or “piece”) should be understood
in that sense.
First, we recall the results in [4] and [5], where it was shown that extremal and next-
to-extremal n-point (En0 and En1 , respectively) functions are not renormalized to order
g 2 . This means that these correlators have a free-field form and, moreover, the overall
coefficient does not depend on g. In the extremal case, each of the non-vanishing Feynman
graphs is a product of n − 1 (free) two-point functions. The point at which the highest-
dimension operator is inserted is common to all the two-point functions. In the next-to-
extremal case there is in addition a propagator connecting two of the other operators, such
that the corresponding graphs are a product of n−2 factors: n−3 (free) two-point functions
times one (free) next-to-extremal three-point function.
Next we consider a general E52 function, hOk (x)Ok1 (x1 )Ok2 (x2 )Ok3 (x3 )Ok4 (x4 )i with
k = k1 + k2 + k3 + k4 − 4. The Feynman graphs contributing at the free-field level are
shown in Fig. 5.
The Feynman graphs that contribute at order g 2 are depicted in Fig. 6. The graphs a1–c1
are radiative corrections of the graphs in Fig. 5. Graphs with two “tadpoles” connected by
a dashed line vanish trivially due to the symmetry in the colour indices in the highest-
dimension operator, and we do not display them explicitly here or in the subsequent.
Graphs a1 and b1 vanish due to non-renormalization theorems for two-point and three-
point correlators [17]. Indeed, as was discussed in [5], the operator Ok 0 (x) entering the
relevant two-point and three-point functions is in the [0, k 0 , 0] representation of the SU(4)
Fig. 5. Graphs contributing to E52 functions at the free-field level. Diagram c is disconnected in the
particular case k = 4, ki = 2 and, therefore, was not considered in Section 2.1.
flavour group and the extra colour-group generators appearing in the trace of this operator
do not spoil the validity of the non-renormalization proofs in [17]. As in Section 2, all the
remaining graphs except b2 vanish due to the symmetries in the colour indices. We show
this explicitly for graph b4.
In order to simplify the equations we introduce the following schematic notation:
{b1 b2 · · · bs } = Str(T b1 T b2 · · · T bs T ai · · · T ai+r−1 ), (3.1)
where ai , . . . , ai+r−1 are the colour indices of the r legs of a given operator which are
attached to the highest-dimension operator, Ok (x), and b1 , . . . , bs are extra colour indices.
Moreover, we write the trace of the operator Ok (x) as
{a} = Str(T a1 · · · T ak ), (3.2)
and a trace with a commutator as, e.g.,

{[b, c]} = Str [T b , T c ]T ai · · · T ai+r−1 , (3.3)
where the trace on the r.h.s. is not symmetrized under b ↔ c. With this notation, the
contribution of graph b4 reads

b4 ∼ Af a1 bp f cdp + Bf a1 cp f bdp + Cf a1 dp f bcp
× Str(T a1 · · · T ak ) Str(T b T a2 · T ak1 ) Str(T d T ak1 +1 · · · T ak1 +k2 −1 )
× Str(T c T e T ak1 +k2 · · · T ak1 +k2 +k3 −3 ) Str(T e T ak1 +k2 +k3 −2 · · · T ak1 +k2 +k3 +k4 −4 )

≡ Af a1 bp f cdp + Bf a1 cp f bdp + Cf a1 dp f bcp
× {a}{b}{d}{ce}{e}, (3.4)
where A, B and C are functions of space–time and flavour. We show in turn that the three
terms vanish, using the identities [T a , T c ] = if acp T p and
X
r

Tr M1 · · · [Mi , N] · · · Mr = 0, (3.5)
i=1
where Mi , N is any set of matrices. The same identity holds for the symmetric trace.
Fig. 6. Feynman graphs contributing to an E52 function at order g 2 .
For the term with coefficient A, contracting one of the structure constants with a group
constant inside a trace, we find
A-term ∼ {a}{[a1, p]}{d}{ce}{e}
X
∼ {a} {p[a1 , ai ]}{d}{ce}{e}
i
= 0, (3.6)
where in the last identity we have used the fact that in each term there is a contraction of
an antisymmetric commutator, [T a1 , T ai ], times a symmetric trace, {a}. Similarly, for the
term with coefficient B we have
B-term ∼ {a}{b}{d}{[a1, p]e}{e}

X
∼ {a}{b}{d} {p[a1 , e]} + {pe[a1 , ai ]} {e}
i
∼ {a}{b}{d}{pq}f a1 eq {e}
∼ {a}{b}{d}{pq}{[a1, q]}
X
∼ {a}{b}{d}{pq} {q[a1, ai ]}
i
= 0. (3.7)
Finally, let us show that the term with coefficient C vanishes as well:
C-term ∼ {a}{b}{[a1, p]}{ce}{e}
X
∼ {a}{b} {p[a1 , ai ]}
i
= 0. (3.8)
Hence graph b4 vanishes. The essential step in the proof is to use the identity (3.5) to shift
antisymmetric combinations of indices arising from the structure constants between traces,
until a commutator [T ai , T aj ] is obtained, which vanishes when contracted with {a}. One
can prove in a similar manner that graphs a2–a4, b3, b5, c1, and d1–d7 vanish as well.
On the other hand, there are graphs of the general form b2 that give finite contributions
because they cannot be reduced to a sum of terms containing a commutator [T ai , T aj ].
Nevertheless we see that b2 has the factored form

Ok (x)Ok1 (x1 )Ok2 (x2 )Ok3 (x3 )Ok4 (x4 ) b2
1p

(1)
= kk1 (k − k1 ) Ok1 (x)Ok1 (x1 ) Ok−k1 (x)Ok2 (x2 )Ok3 (x3 )Ok4 (x4 ) , (3.9)
N
which is a product of a free two-point function and an E42 four-point function. In this
equation it is understood that the colour generators of the operators Ok1 (x) and Ok−k1 (x)
are included inside the same trace, such that the two factors are coupled by colour.
However, the colour indices within the first factor are contracted and eventually a product
of two functions which are colour singlets is obtained. Therefore all the graphs factor into
at least two pieces and the E52 function has, to order g 2 , the form

Ok (x)Ok1 (x1 )Ok2 (x2 )Ok3 (x3 )Ok4 (x4 )
1p
= k(k1 + k2 − 2)(k3 + k4 − 2)
N

× Ok1 +k2 −2 (x)Ok1 (x1 )Ok2 (x2 ) Ok3 +k4 −2 (x)Ok3 (x3 )Ok4 (x4 )
1p
+ kk1 k2 (k3 + k4 − 4)
N

× Ok1 (x)Ok1 (x1 ) Ok2 (x)Ok2 (x2 ) Ok3 +k4 −4 (x)Ok3 (x3 )Ok4 (x4 )
1p

0
+ kk1 (k − k1 ) Ok1 (x)Ok1 (x1 ) Ok−k1 (x)Ok2 (x2 )Ok3 (x3 )Ok4 (x4 )
N
+ perms, (3.10)
where

Ok1 +k2 −2p (x)Ok1 (x1 )Ok2 (x2 )
√
1 (k1 + k2 − 2p)k1 k2
= , (3.11)
N (x − x1 )2(k1 −p) (x − x2 )2(k2 −p) (x1 − x2 )2p
and the square roots account for the large N Wick factors and the normalization. The prime
in the four point function in (3.10) indicates that a factorized free-field contribution to this
four point function has been explicitly written in the second term of the equation and must
not be included again. Observe that the sum of the degrees of extremality of the factors in
each term is 2 (i.e., next-to-next-to-extremal). Schematically Eq. (3.10) can be written
E52 = E31 E31 + E20 E20 E32 + E20 E42 . (3.12)
The extension of this detailed analysis to Enm functions for n > 5 and m > 2 is very
tedious, and we limit the discussion to the demonstration of factorization properties
through order g 2 . Specifically we will argue that
• E62 functions split into at least three factors:
E62 = E20 E31 E31 + E20 E20 E42 . (3.13)
• E63 functions split into at least two factors:
E63 = E20 E53 + E31 E42 + E32 E42 + E33 E40 . (3.14)
• In the general case Enm splits into at least n − m − 1 factors:
X Y
n−m−1
Enm = Enmi i , (3.15)
{nj ,mj } i=1
P P
where n−m−1
i=1 ni = 2(n − 1) − m and n−m−1 i=1 mi = m. If m 6 n − 3 each term
factors into at least two pieces and the maximum value of ni is n − 1.
The qualification “at least” is meant to indicate that there are graphs in which more factors
occur. For instance, in graph c of Fig. 7 below, the factor E42 of the last term in (3.13) splits
in turn into E20 E32 . Even with this limited aim, the discussion is complicated and some
readers may wish to proceed to Section 4.
Let us consider six-point functions hOk (x)Ok1 (x1 ) · · · Ok5 (x5 )i. The E62 correlators have
P
k = 5i=1 ki − 4. The free-field Feynman graphs for such a function are displayed in Fig. 7.
We include graphs with tadpoles, which vanish at this order (because the primary operators
are traceless tensors in flavour space) but can be used to construct order g 2 graphs. The
graphs in Fig. 7 are similar to the ones contributing to the E52 functions studied above, but
with an additional “rainbow” of propagators connecting the highest-dimension operator to
the additional operator (at x5 in the figures). The order g 2 corrections can be classified
in three groups: graphs that factor into the E52 -graphs times a free two-point subgraph
involving the new operator, corrections inside the rainbow and graphs that connect the new
rainbow to the rest of the free-field graph. Using the results above it is clear that the graphs
of the first group have at least three factors, two of which are free-field functions. The
Fig. 7. Feynman graphs contributing to an E62 function at the free-field level.
corrections within the rainbow are corrections to a two-point function and vanish. Finally,
with the method used above for graph b4 of Fig. 6, the graphs of the third kind can be
shown to either vanish or factor into three factors, two of which are free-field functions.
The second possibility occurs only when the new rainbow is connected to one of the two
lines at the bottom of graph c. Summarizing, all graphs have at least three factors and we
find the structure in (3.13).
The Feynman graphs that contribute at the free-field level to an E63 function (k =
P5
i=1 ki − 6) are depicted in Fig. 8. It is useful to introduce the following definition: We
denote as “T graphs” those graphs which have no closed loops after removing all the lines
connected to the highest-dimension operator. A Tk graph is defined to be a T graph with k
factors. We can then distinguish four kinds of graphs in Fig. 8:
1. Graphs with q > 3 factors: d, e, f and g.
2. Graphs with one tadpole and at least two factors: i.
3. T2 graphs: a, b and c.
4. T1 graphs with one tadpole: h.
Let us consider now the order g 2 graphs, that are either self-energy corrections or can be
constructed inserting a dashed line in the graphs of Fig. 8. Self-energy corrections do not
change the number of factors of the graph and we do not need to study them, although we
observe that they cancel other corrections in such a way that two- and three-point factors
are not renormalized. From now on, we consider only dashed-line graphs. Order g 2 graphs
constructed by inserting a dashed line into the free-field graphs of the first kind obviously
have at least two factors, as a single line cannot connect three or more pieces. The same
applies to order g 2 graphs constructed from graphs of the second kind since one end of the
dashed line has to be connected to the tadpole line. Finally, we prove below for any number
Fig. 8. Feynman graphs contributing to an E63 function at the free-field level. The dots stand for other
graphs with one tadpole and two or more pieces.
of points that T graphs do not have order g 2 corrections connecting the different factors
or one factor to a tadpole. This means that order g 2 graphs constructed from graphs of the
third kind factor into two pieces and that the order g 2 graphs constructed from graphs of
the fourth kind vanish. Therefore we conclude that to order g 2 all non-vanishing graphs
factor into at least two pieces.
Furthermore, each graph has a global degree of extremality equal to 3, i.e., it is a product
Qp mi Pp Pp
i=1 Eni with i=1 ni = 6 + p − 1 and i=1 mi = 3. This property can be read directly
from the graphs in Fig. 8. Therefore, we obtain the structure written in (3.14).
The analysis of the six-point correlators can be easily generalized to Enm functions,
P
hOk (x)Ok1 (x1 ) · · · Okn−1 (xn−1 )i with k = n−1
i=1 ki − 2m. It turns out that to order g an
2
En function, m 6 n − 2, is a sum of terms that factor into at least n − m − 1 pieces. The

m
essential ingredient in the proof is the fact that the Feynman graphs contributing to an Enm
function at the free-field level fall among the following classes:
1. Graphs with q > n − m factors.
2. Graphs with one tadpole and at least n − m − 1 factors.
3. Tn−m−1 graphs.
4. Tn−m−2 graphs with one tadpole.
To see this, consider first graphs without tadpoles. These graphs have m lines that are not
connected to x. Let us add these m lines in turn. After adding one line we are left with a
Fig. 9. Relevant part of a general order-g 2 correction to a T graph.
graph with n − 2 factors. Then we add another line. If it connects the same two operators
that were joined by the first line, we obtain a non-T graph with n − 2 factors. On the other
hand, if one end of the second line is attached to an operator that were not connected to the
first line, we get a T graph with n − 3 factors. The third line can either connect different
factors or connect operators within one factor. In the first case, one gets a non-T graph with
n − 3 factors or a T graph with n − 4 factors, depending on the choice of the second line.
In the second case one gets a non-T graph with n − 2 or n − 3 factors. We can proceed
recursively and find that after adding the mth line we get a T graph with n − m − 1 factors
or a non-T graph with q > n − m factors. If the graph has one tadpole there are m + 1 lines
that are not connected to x, and the argument can be repeated with m → m + 1.
The proof that the order g 2 graphs constructed from these ones factor into n − m − 1
graphs is identical to the proof given above for n = 6. The only non-trivial ingredient is
that order g 2 graphs constructed from free-field Tq graphs factor into q parts or vanish. We
show this next.
First, we observe that the following property holds:
X
r−1 X
k
f bcp {cd1 · · · dk } ∼ f bai q {pd1 · · · dk } + f bdi q {pd1 · · · dbi · · · dk }, (3.16)
i=1 i=1
where r is the number of lines of the given operator connected to the highest dimension
operator. We see that all the terms on the r.h.s. have an f with one of the indices of the f
on the l.h.s., and one of the indices in the trace at the l.h.s. In other words: p changes, b is
replaced by ai or di and c stays the same.
Consider now a general Tp graph with p > 2 and one dashed line connecting the
“external” (i.e., not connected to the highest-dimension operator) lines of two different
factors. In Fig. 9 we show a part of this graph. There are three contributions:
1. f bcp f dep . The index b appears in the trace of the operator at x1 . Using the relation
1 1 1 1
(3.16) we obtain a sum of terms that contain f ai cq or f gi cq . This last f is in turn
contracted with an index in another trace and we can apply the same relation to obtain
2 2 2 2
terms with f ai cq or with f gi cq . We repeat the procedure for the terms that do not
contain an f with an ai until we reach an operator with only one external leg. Then
all terms contain f ai cq . Since c appears in the trace of the operator at x2 , we can
use the same method but keeping now ai fixed. Finally the expression is reduced to a
series of terms that contain f ai aj q , which is antisymmetric under ai ↔ aj . Since all
the terms also contain the symmetric trace {a}, they vanish.
2. f bdp f cep . The indices in the first f can be shifted just as above until a sum of terms
containing f ai dq is obtained. This f is contracted to the trace in y1 through the index
a a1q 1 a h1 q 1
d, and can be transformed into a sum of terms with either f i j or f i j . The first
kind of terms vanish when contracted with {a} and the second kind can be transform
1
until all terms contain f ai aj q . All these terms vanish when contracted with {a}.
3. f bep f cdp . The previous argument can be applied to this case as well.
The same can be argued if one or both ends of the dashed line were connected to a rainbow
or to a tadpole. Note that it is essential for the argument that, when the indices in one f are
shifted by using Eq. (3.16), the two new indices are different from all the indices in previous
steps. This would not occur if there were loops of external lines. For this reason the proof
only applies to T graphs. Since the different factors of a T graph cannot be connected at
order g 2 we conclude that their factored structure is preserve to this order.
Moreover, the sum of the degrees of extremality of the factors of any term in an Enm
function is equal to m. Indeed, if a Feynman graph factors into two or more functions it
can be written as Enm11Enm22 , with n1 + n2 = n + 1. In general, each factor can be further
factored. Both factors have x as a common point. From the degrees of extremality of the
P 1 −1
two factors it follows that ni=1 ki − 2m1 legs of the operator Ok (x) enter the first factor
P 2 −1
and ni=1 ki − 2m2 legs of this operator enter the second factor. Since the operator Ok (x)
has k legs,
1 −1
nX 2 −1
nX
k= ki − 2m1 + ki − 2m2
i=1 i=1
X
n−1
= ki − 2(m1 + m2 ), (3.17)
i=1
and from the m-extremality condition, m = m1 + m2 . The same procedure can be applied
to the subfactors of Enm11 and Enm22 . An alternative way of understanding this property is
noting that the degree of extremality of each factor is given by the number of lines in that
factor that are not connected to x. This proves the structure written in (3.15).
4. General near-extremal correlators in AdS
The AdS5 × S 5 supergravity cubic and quartic couplings of chiral primary fields have
been calculated in [15] and [31], respectively. In the extremal case, i.e., when the conformal
dimension of one field equals the sum of the remaining conformal dimensions, these
couplings vanish. Furthermore the requirement that the AdS amplitudes have to be finite for
Fig. 10. Non-vanishing contributions to En0 (diagram a0 ) and En1 (diagram b0 ) functions.
non-coincident points implies that the couplings of extremal n-field vertices must vanish
for any n > 3 [29]. As we show in Appendix A, there are boundary contributions and the
Witten diagrams of extremal n-point functions reduce to a product of two-point functions
connected to the point x at which the highest-dimension operator is inserted. This is
illustrated in Fig. 10. The space–time structure of these diagrams agree with the free-field
approximation. Furthermore, in [29,30] it was shown that the coefficient multiplying this
structure is not renormalized.
In [5] it was shown that the next-to-extremal couplings of n chiral primaries have to
vanish, since otherwise the AdS and the field-theoretical calculations of En1 functions
would not agree. This has been recently checked explicitly for n = 4 [6]. All non-vanishing
Witten diagrams contributing to the En1 functions then reduce to products of n−3 two-point
functions times one next-to-extremal three-point function, all of them with x as a common
point, as we show in Fig. 10. The space–time structure of the reduced diagrams agrees
with the field-theoretical calculation [5]. For E41 functions we have also checked that the
coefficient of the AdS structure agrees with the result found in field theory. The calculation
is similar to the ones we show in the appendix below.
This analysis can be extended to other Enm functions. It can be shown in general that, as
long as certain supergravity couplings vanish, the AdS calculation gives the same space–
time structure as the order g 2 calculation, up to the detailed form of the factors that are
renormalized. We have already shown this for the simplest En2 function in Section 2.2.
The calculation of a general E52 function hOk Ok1 Ok2 Ok3 Ok4 i in AdS is very similar to
the one for hO4 O2 O2 O2 O2 i and we discuss only the new features:
1. We have checked for general ki ’s that the coefficient of diagram a in Fig. 3 agrees
with the coefficient of the field-theory graph a in Fig. 5 as given by the first term in (3.10).
This calculation may be found in Appendix A below Eq. (A.14).
2. Diagram d of Fig. 3 does not vanish in general, but gives a new reduced diagram, d0 in
Fig. 4 (the same structure results from diagram b0 when the upper vertex is extremal). This
reduced diagram has the same form as graph c in Fig. 5, which gives the second term in
the field theory result (3.10). It is interesting to note that this structure appears neither in
field theory nor in AdS for the connected hO4 O2 O2 O2 O2 i. It is possible to calculate the
coefficient of diagram d0 as well, but this requires considerable processing of the results
for quartic couplings in [33].
3. The discussion of descendent exchange becomes more complicated. For diagrams in
which quartic couplings appear, consistent Kaluza–Klein truncation is not sufficient for
general ki , but it motivates the assumption 3 that a coupling with descendents vanishes
whenever the associated coupling of chiral primaries does. With this assumption, one
can easily extend the analysis of descendents in Section 2.2 and conclude that they only
contribute to four-point factors.
4. The contact diagram spoils the factorization property. Hence the result in supergravity
only agrees with the factorization we have found at weak coupling if the E52 couplings
vanish. This is an extension of consistent Kaluza–Klein truncation.
More general functions will be considered in the next section. There we reverse the
argument and derive the vanishing of certain supergravity couplings from the requirement
that the field-theoretical calculation to order g 2 and the AdS calculation give the same
factored structure.
5. Vanishing near-extremal supergravity couplings
In this section we show that Enm supergravity couplings vanish for any m 6 n − 3 if the
following assumptions hold:
1. Witten diagrams are finite for non-coincident external points.
2. The factorization of Enm functions, m 6 n − 3 that we have found (for any N ) to order
g 2 is preserved at large t’Hooft coupling in the large N limit.
3. A vertex with one or more descendents has a vanishing coupling if the coupling of
the associated chiral primary vertex vanishes.
We do not use known results about supergravity couplings in the argument. The fact that
extremal cubic and quartic couplings and next-to-extremal quartic couplings do vanish then
supports the validity of the assumptions above.
We also need the results of Appendix A, which are summarized by:
• The integral over y in a Witten diagram is divergent if and only if the vertex at y
is extremal and the highest-dimension field entering that vertex propagates into the
boundary (at point x).
• If this is the case, the divergence is logarithmic and comes from the region y ∼ x.
In analytic continuation this divergence appears as a pole that, according to the first
assumption, must be cancelled by a zero coming from the coupling. When the regulator
is removed, the diagram factors into at least two pieces.
The proof proceeds by induction in the number of points, n. The induction starts for
n = 3, m = 0. We consider an extremal three-point function. The integral is divergent and,
according to the first assumption, the coupling must vanish. Therefore, the E3m couplings
3 For scalar and vector descendents there is evidence from AdS diagrams and field theory to support this
assumption. We do not discuss it here.
m function.
Fig. 11. Witten diagrams contributing to an En+1
vanish for any m 6 3 − 3 = 0. Furthermore, the diagram factors into two two-point
functions, in agreement with the field-theoretical results.
Let now n > 3. Suppose that the Epm couplings vanish for all m 6 p − 3, p 6 n. We have
to show that the En+1m
couplings vanish for m 6 2.
Consider an En+1 function hO∆ (x)O∆1 (x1 ) · · · O∆n i, with m 6 n − 2. This means
m
P
that ∆ = ni=1 ∆i − 2m. ∆ and ∆i are the conformal dimensions of the chiral
primary operators, which coincide with the non-vanishing Dynkin labels of their SU(4)
representation. Two kinds of diagrams contribute: contact and exchange. All the exchange
diagrams are of the general form shown in Fig. 11, with 1 6 k0 6 n − 2. We consider first
diagrams that involve only primary exchanges. The exchanged fields φ1 , . . . , φr are in the
representations [0, δ1 , 0], . . . , [0, δr , 0], and have dimensions δ1 , . . . , δr , respectively. Only
the sets of fields with
X r Xn X n X
n
δj = ∆i − 2m, ∆i − (2m − 1), . . . , ∆i (5.1)
j =1 i=k0 +1 i=k0 +1 i=k0 +1
are allowed by SU(4)R . We distinguish three possibilities:

X
r X
n
1. δj = ∆i − 2m.
j =1 i=k0 +1
Xn X
r X
n
2. ∆i − 2(m − 1) 6 δj 6 ∆i − 2(m − k − r + 2).
i=k0 +1 j =1 i=k0 +1
Xr X
n
3. δj > ∆i − 2(m − k0 − r + 1).
j =1 i=k0 +1
We study the three cases in turn.

P Pk0
Case 1. Since ni=k0 +1 ∆i − 2m = ∆ − i=1 ∆i , the vertex at y is extremal and
vanishes due to the induction hypothesis. On the other hand, the integral diverges since the
field with highest dimension, ∆, propagates to the boundary. Hence the diagram factors
m
into (E20 )k0 En−k , as shown in Fig. 12.
0 +1
m function.
Fig. 12. Non-vanishing contribution to an En+1
P Pk0
Case 2. Since ni=k0 +1 ∆i − 2(m − k − r + 2) = ∆ − i=1 ∆i + 2(k − 1), the vertex
p
at y is Ek0 +2 with p 6 (k + 2) − 3 and is zero by the induction assumption. Because the
integral over y is convergent and the other possible divergences are cancelled by zeros in
the extremal couplings, the diagram vanishes (see the appendix).
Case 3. We have the inequality
 
k0 +···+kj+1
Xn Xr X
∆i − 2(m − k0 − r + 1) >  ∆i − 2(kj − 1) + 2. (5.2)
i=k0 +1 j =1 i=k0 +···+kj +1
Pk0 +k1
If δ1 > i=k0 +1 −2(k1 − 2), the subdiagram involving the points y and xk0 +1 , . . . , xk0 +k1
p
and the points in the corresponding blob is an Ek1 +1 diagram with p 6 (k1 + 1) − 3 and
p
the highest-dimension field (φ1 ) propagating into the bulk, which we denote as Ẽk1 +1 . If,
Pk0 +k1
on the contrary, δ1 6 i=k0 +1 −2(k1 − 1), the following relation holds:
 
k0 +···+kj+1
X
r X r X
δj >  ∆i − 2(kj − 1) + 2. (5.3)
j =2 j =2 i=k0 +···+kj +1
Pk0 +k1 +k2 p
If δ2 > i=k0 +k1 +1 −2(k2 − 2), the corresponding subdiagram is an Ẽk2 +1 function with
P
p 6 (k2 +1)−3. Otherwise, we have a relation similar to the one above involving rj =3 δj .
Pk0 +···+kj
Iteratively, if δj 6 i=k 0 +···+kj−1 +1
−2(kj − 1) for j = 1, . . . , r − 1, we find that
k0 +···+k
X r
δr > ∆i − 2(kr − 2), (5.4)
i=k0 +···+kr−1 +1
p
such that the corresponding subdiagram is an Ẽkr +1 function with p 6 (kr + 1) − 3. Hence,
p
at least one of the subdiagrams connected to y is an Ẽkj +1 function with p 6 (kj + 1) − 3.
The analysis of the entire diagram b can be applied to this subdiagram to show that it
vanishes. We have again three possibilities. The only difference is that in the equivalent of
case (1) the integral finite because the highest-dimension field (φ) propagates into the bulk.
More details are given in the appendix. The iteration ends when all the subdiagrams are
p
contact diagrams, which vanish because the induction hypothesis implies that the Ekj +1
coupling is zero for p 6 (kj + 1) − 3.
Therefore, the cases (2) and (3) lead to vanishing diagrams, whereas the case (1) gives
a reduced diagram with at least two factors. Consider now the case when at least one of
the fields φj is a descendent. Then, if ∆φj denotes the conformal dimension of each field,
Pr Pn
j =1 ∆φj > i=k0 +1 ∆i − 2m and the vertex at y is sub-extremal. Hence the integral over
y is finite. On the other hand, if φj is a descendent, it descends from a chiral primary field
in a representation [0, δ̃j , 0] with δ̃j > (δj )min , where (δj )min is the minimal dimension
allowed by SU(4) if φj were chiral primary. All the possibilities can be discussed just
as in the pure chiral primary case, but changing some δj by δ̃j . Using assumption 3, we
conclude that in all cases one or more of the couplings involved in the diagram vanish if
the induction hypothesis holds.
We have shown that all the exchange diagrams either vanish or factor into at least two
pieces. Finally, there is a contact diagram contributing to the En+1 m function (diagram
a in Fig. 11). In the extremal case, m = 0, this diagram has a logarithmic divergence
0
that has to be cancelled by a zero in the coupling. Hence the En+1 coupling vanishes.
If m > 1 the diagram is finite and does not have a factored structure. Since the non-
vanishing exchange diagrams factor, the non-factored contribution of the contact diagram
survives in the full function. On the other hand, we have shown in Section 3 that, to
m functions factor into at least two pieces if m 6 n − 3. According to the
order g 2 , En+1
second assumption (and the Maldacena conjecture), the AdS calculation should also give
m
a factored structure. Therefore, the En+1 couplings must vanish in order for the contact
diagram not to contribute. This completes the induction.
Finally, we observe that the first assumption can be relaxed: It is sufficient to require that
the sum of the Witten diagrams contributing to an AdS amplitude is finite. This however,
complicates the description of the proof and for that reason we have chosen here an slightly
stronger assumption.
6. Conclusions
The main results of this paper are

1. Through order g 2 , field theory graphs for near-extremal n-point functions have a
factored structure.
2. This structure is shown to be matched exactly by exchange diagrams in AdS
supergravity for the five-point function hO4 O2 O2 O2 O2 i. For more general E52
functions an additional technical assumption about descendent couplings is required
to reach the same conclusion.
3. For E52 functions we show that there are unique supergravity diagrams which match
the corresponding field theory graphs in form and coefficient.
4. For Enm functions with n > 5 and m 6 n − 3 we establish a similar but less precise
correspondence between field theory and supergravity.
5. The correspondence would be spoiled by non-factored supergravity contact diagrams

unless the associated near-extremal couplings vanish. We therefore conjecture that
Enm couplings do vanish when m 6 n − 3. For some cases this is a consequence of
consistent Kaluza–Klein truncation and it is a natural extension of known results
for n = 3, 4. It remains to explain the pattern of vanishing couplings, which is
presumably a consequence of the reduction of Type II B supergravity on the internal
space S 5 .
Acknowledgements
We thank Leonardo Rastelli for useful discussions. The research of E.D’H is supported
in part by NSF Grants No. PHY-95-31023 and PHY-98-19686 and the research of D.Z.F by
NSF Grant No. PHY-97-22072. J.E., who is a DFG Emmy Noether fellow, acknowledges
funding through a DAAD postdoctoral fellowship. M.P.V. thanks MEC for a postdoctoral
fellowship.
Appendix A
A.1. AdS integrals
Here we study the integrals that appear in the calculation of near-extremal correlation
functions via the AdS/CFT correspondence. We first give some explicit examples of
integrals that appear in E52 functions and then move to general properties of Enm integrals.
We use the methods developed in [16,35,37] for general correlators and in [29] and [5]
for extremal and next-to-extremal correlators. We use the Euclidean continuation of AdS5
whose metric is given by
!
1 X4
ds 2 = 2 dz0 2 + dzi 2 . (A.1)
z0
i=1
The scalar bulk to boundary propagator is given by [3,16]
∆
z0
K∆ (x, z) = C∆ ,
z0 2 + (Ez − xE)2
Γ (∆) 1
C∆ = 2 , ∆ > 2, C2 = . (A.2)
π Γ (∆ − 2) 2π 2
∆ > 2 is a real number. Divergences in AdS integrals can arise only when the integration
points approach the boundary. Hence we need the behaviour of the propagators when the
bulk point approaches the boundary. When z0 → 0 but zE 6→ xE ,
1
K∆ (x, z) → C∆ z0∆ . (A.3)
(Ez − xE)2∆
On the other hand, when z → x,
K∆ (x, z) → z04−∆ δ(Ez − xE). (A.4)
The explicit form of the bulk propagator Gδ (y, z) is not needed here. Its behaviour when
one of the points approaches the boundary is [35,36], in the conventions of [15],
y0 →0
Gδ (y, z) −→ C̃δ y0 δ Kδ (E
y , z),
2 δ−6 (2π) (δ + 1)2
5 (2π)5 9
C̃δ = , k > 2, C̃2 = . (A.5)
δ − 2 4N 2 δ(δ − 1) 4N 2 25
A.2. Examples of five-point integrals
We first illustrate the general properties of AdS integrals by studying the contribution of
diagram b of Fig. 3 to a general next-to-next-to-extremal five-point function. We always
assume that the five points are non-coincident. The conformal dimensions of the chiral
P
primary operators involved satisfy ∆ = 4i=1 ∆i − 4. We take φ and φ 0 to be chiral
primaries of conformal dimension δ and δ 0 , respectively. The contribution of the diagram
is
hO∆ (x)O∆1 (x1 )O∆2 (x2 )O∆3 (x3 )O∆4 (x4 )i

ZZZ
G3 (∆, ∆1 , δ)G3 (δ, ∆2 , δ 0 )G3 (∆3 , ∆4 , δ 0 ) d5 y d5 z d5 w
=
N∆ N∆1 N∆2 N∆3 N∆4 y05 z05 w05
× K∆ (x, y)K∆1 (x1 , y)Gδ (y, z)K∆2 (x2 , z)Gδ 0 (z, w)K∆3 (x3 , w)K∆4 (x4 , w),
(A.6)
with N∆ as in (2.12). The possible divergence can only arise from the region y ∼ x. Indeed,
in this region we can use Eqs. (A.3), (A.4) and (A.5) to see that the integrand is proportional
to
y0 −5 y0 4−∆ y0 ∆1 y0 δ = y0 δ+∆1 −∆−1

≡ y0−α−1 . (A.7)
The degree of divergence of the integral over y0 is given by α. The SU(4) R-symmetry
implies that δ > ∆ − ∆1 and thus α 6 0. Hence there are two possibilities: the integral is
finite if α < 0 and it is logarithmically divergent if α = 0. Since ∆ > ∆i , i = 1, . . . , 4,
a similar counting shows that the regions when one or more of the integration points
approach any xi do not give any divergence.
Let us consider first the divergent case which occurs when δ = ∆−∆1 . We regularize the
integral by analytic continuation in the highest dimension which we write as ∆ − ε, ε > 0.
The coupling of the vertex at y has to be changed accordingly. The integral is convergent
since now we have α = −ε. For small ε the integral is dominated by the contribution of
the region y ∼ x. In this region we can use Eq. (A.5) and write Eq. (A.6) as

O∆ (x)O∆1 (x1 )O∆2 (x2 )O∆3 (x3 )O∆4 (x4 ) 1
y

= I1 O∆−∆1 (x)O∆2 (x2 )O∆3 (x3 )O∆4 (x4 ) 1 (A.8)
with
Z 5
y C∆ C∆1 C̃δ N∆−∆1 d y y0 2∆−ε
I1 = G3 (∆ − ε, ∆1 , δ) , (A.9)
(x − x1 )2∆1 N∆ N∆1 y0 5 (y − x)2(∆−ε)
R

O∆−∆1 (x)O∆2 (x2 )O∆3 (x3 )O∆4 (x4 ) 1
−1
= N∆−∆1 N∆2 N∆3 N∆4 G3 (δ, ∆2 , δ 0 )G3 (∆3 , ∆4 , δ 0 )
ZZ 5 5
d zd w
× Kδ (x, z)K∆2 (x2 , z)Gδ 0 (z, w)K∆3 (x3 , w)K∆4 (x4 , w). (A.10)
z05 w05
y
R is any neighbourhood of x where the approximation (A.5) is valid. The integral in I1
leads to a pole in ε, which is cancelled by a factor of ε arising from G3 (∆ − ε, ∆1 , δ). This
is described in detail in [5]. The result when ε → 0 is
y 1p

I1 = ∆∆1 (∆ − ∆1 ) O∆1 (x)O∆1 (x1 ) . (A.11)
N
On the other hand, hO∆−∆1 (x)O∆2 (x2 )O∆3 (x3 )O∆4 (x4 )i1 has the form of an exchange
contribution to a four-point function which may be evaluated using the methods of [37].
We see that the five point function contribution (A.8) factors into a free-field two-point
function times a four-point function contribution.
As an example of a convergent integral, let us consider the case when the dimension of
the exchanged fields is δ = ∆ − ∆1 + 2 and δ 0 = ∆3 + ∆4 − 2. The integral is convergent
since α = −2. Indeed, in this simple case we can perform the integral over y and w
explicitly using the methods of [37] and find

O∆ (x)O∆1 (x1 )O∆2 (x2 )O∆3 (x3 )O∆4 (x4 ) 2
= G(∆, ∆1 , ∆ − ∆1 + 2)G(∆ − ∆1 + 2, ∆2 , ∆3 + ∆4 − 2)
1 X
∆−1
1
× G(∆3 + ∆4 − 2, ∆3 , ∆4 )a∆4 −1 bk
(x3 − x4 )2 (x − x1 )2(∆−k)
k=∆−∆1 +1
Z 5
d y
× Kk (x, y)K∆1 −∆+k (x1 , y)K∆2 (x2 , y)K∆3 −1 (x3 , y)K∆4 −1 (x4 , y),
y05
(A.12)
with finite coefficients a∆4 −1 , bk . The remaining y-integral is finite for every term in the
sum [38]. Therefore, since the second coupling is extremal and therefore vanishes, the
whole contribution vanishes.
As a final example for a five-point function contribution we consider diagram b of Fig. 3,

O∆ (x)O∆1 (x1 )O∆2 (x2 )O∆3 (x3 )O∆4 (x4 ) 3
ZZZ 5 5 5
G3 (∆, δ, δ 0 )G3 (∆1 , ∆2 , δ)G3 (∆3 , ∆4 , δ 0 ) d yd zd w
= K∆ (x, y)Gδ (y, z)
N∆ N∆1 N∆2 N∆3 N∆4 y05 z05 w05
× K∆1 (x1 , z)K∆2 (x2 , z)Gδ 0 (y, w)K∆3 (x3 , w)K∆4 (x4 , w), (A.13)
when δ = ∆1 + ∆2 − 2 and δ 0 = ∆3 + ∆4 − 2. Again the vertex at y is extremal, and
the dominant contribution arises when y ∼ x. For this case the limit (A.5) applies to both
bulk-to-bulk propagators, such that the integral factors into

O∆ (x)O∆1 (x1 )O∆2 (x2 )O∆3 (x3 )O∆4 (x4 ) 3
y

= I3 O∆1 +∆1 −2 (x)O∆1 (x1 )O∆2 (x2 ) O∆3 +∆4 −2 (x)O∆3 (x3 )O∆4 (x4 ) , (A.14)
where
y N∆1 +∆2 −2 N∆3 +∆4 −2
I3 = C̃∆1 +∆2 −2 C̃∆3 +∆4 −2 C∆
N∆
Z 5
d y y0 2∆−ε
× G(∆ − ε, δ, δ 0 ) (A.15)
y0 5 (y − x)2(∆−ε)
R
and

O∆1 +∆1 −2 (x)O∆1 (x1 )O∆2 (x2 )
Z 5
G(δ, ∆1 , ∆2 ) d w
= Kδ (x, w)K∆1 (x1 , w)K∆2 (x2 , w)
N∆1 +∆2 −2 N∆1 N∆2 w05
√
1 (∆1 + ∆2 − 2)∆1 ∆2
= , (A.16)
N (x − x1 )2(∆1 −1) (x − x2 )2(∆2 −1) (x1 − x2 )2
such that we obtain

O∆ (x)O∆1 (x1 )O∆2 (x2 )O∆3 (x3 )O∆4 (x4 ) 3
1p
= ∆(∆1 + ∆2 − 2)(∆3 + ∆4 − 2)
N

× O∆1 +∆1 −2 (x)O∆1 (x1 )O∆2 (x2 ) O∆3 +∆4 −2 (x)O∆3 (x3 )O∆4 (x4 ) (A.17)
which agrees exactly with the corresponding contribution to the field theory result (3.10).
A.3. General AdS integrals
Let us now turn to studying general En+1m functions with m 6 n − 2 (and non-coincident
points). This is relevant for the inductive proof of Section 5. For simplicity we work
with non-derivative couplings. Derivative couplings do not change the leading degree of
divergence of the integrals: if y is the integration variable, the derivatives ∂/∂y0 make the
integrand more singular as y approach the boundary, but this effect is exactly compensated
by the powers of y0 arising from the metric. On the other hand, we assume in the following
that all the extremal couplings vanish. More precisely, if analytic continuation is used, we
assume that the extremal couplings contains a zero in the regulator. This is required in order
to obtain non-divergent Witten diagrams. We consider first the contact diagram (diagram
a in Fig. 11) and regularize a possible divergence by analytical continuation in the highest
dimension, ∆. This does not affect the result if the integral is convergent. Up to couplings
the diagram gives
Z 5 Yn
d y
Ia = K∆−ε (x, y) K∆i (xi , y). (A.18)
y0 5
i=1
Pn P
As y → x the integrand is proportional to y0 i ∆i −(∆−ε)−1 , i.e., α = ∆ − ε − ni ∆i
for this diagram. Therefore, for ε = 0 the integral diverges logarithmically if the function
Fig. 13. Typical exchange diagram.
is extremal and is convergent in any sub-extremal case. Obviously the integrand is less
singular in the regions y ∼ xi , as the m-extremality condition implies that ∆i < ∆. In the
P
extremal case, ∆ = ni ∆i , the integral gives rise to a pole in ε arising from the region
y ∼ x:
Z
dy0 ε Y
n
1
Ia ∼ y0 + Ra
y0 (x − xi )2∆i
R i=1
1Y
n
1
∼ + Ra0 , (A.19)
ε (x − xi )2∆i
i=1
where Ra and Ra0 are analytic in ε. The pole is cancelled by a zero in the extremal coupling.
Hence, the analytic part R 0 is irrelevant when ε → 0, and the diagram factors into n two-
point functions.
A typical exchange diagram is shown in Fig. 13. Let us study the possible divergent
limits. We consider first the case when only one internal point, which we denote generically
as u, approaches a point x 0 in the boundary. If x 0 6= xi , i = 0, . . . , 9 (x0 ≡ x) or x 0 = xi but u
is not directly connected to xi by a bulk-to-boundary propagator, both the bulk propagators
and bulk-to-boundary propagators connected to u are proportional to positive powers of
u0 as u0 → 0. For example, if y → x2 , the integrand is proportional to y0 ∆+∆1 +δ1 +δ2 −5 .
In the worst possible case the integrand is proportional to u0 2+2+2−5 . Hence this region
does not contribute to the integral. On the other hand, if x 0 = xi and u is directly connected
to xi there is a power u0 4−∆i . For instance, if y → x the integrand is proportional to
y0 ∆1 +δ1 +δ2 −∆−1 . Therefore a logarithmic divergence appears in this region if the vertex
at u is extremal and ∆i is the largest dimension entering this vertex. Otherwise the
integral over this region is finite. For instance, if y → x the integrand is proportional to
y0 ∆1 +δ1 +δ2 −∆−1 .
Let us now allow the possibility that two bulk points u1 and u2 approach the
boundary. If they approach different points the two limits can be considered inde-
pendently. For instance, if y → x and z1 → x2 , the integrand is proportional to
0 0
y0 ∆1 +δ1 +δ2 −∆−1 (z1 )0 δ1 +δ1 +δ2 −∆2 −1 . If u1 and u2 approach the same point x 0 and none
of them is directly connected to this point, all the propagators contribute with positive
powers of (u1 )0 and (u2 )0 and the double integral over u1 and u2 is convergent. Finally, if
u1 → xi and u2 → xi and xi is directly connected to, say, u1 , there are two possibilities:
If u1 and u2 are not directly connected by a bulk propagator there is a factor of (u1 )0 4−∆i
but all the powers of (u2 )0 are positive. On the other hand, if u1 and u2 are directly con-
nected there are two factors with negative powers: (u1 )0 4−∆i (u2 )0 4−δ12 , where δ12 is the
dimension of the propagator connecting u1 and u2 . For example, if y → x and z2 → x, the
0
integrand is proportional to y0 ∆1 +δ1 +δ2 −∆−1 (z2 )0 ∆6 +∆7 +∆8 −δ2 −1 . The integral over y (z2 )
is then logarithmically divergent if the vertex at y (z2 ) is extremal. In general, a (logarith-
mic) divergence in the integral over an internal point u can only arise when the following
two conditions are fulfilled:
1. Either u approaches a point xi that is connected to u by a bulk-to-boundary
propagator or it approaches a point xi that is connected to u by a string of propagators
and all the points in those propagators also approach xi .
2. The vertex at u is extremal and the highest dimension of the fields entering the vertex
is the one of the field that connects u (directly or indirectly) with xi .
We shall use these properties in the following.
Any diagram with at least one exchange is of the general form represented by diagram b
of Fig. 11, where we have isolated the vertex at y to which the highest-dimension operator
is connected. This vertex involves (k0 + 1) bulk-to-boundary propagators and r bulk-
to-bulk propagators depending on y and one zj , j = 1, . . . , r. There are r subdiagrams
involving further tree interactions depicted by a shadowed circle. The j th subdiagram
depends on the bulk variable zj and on kj boundary variables xk0 +···+kj−1 +1 , . . . , xk0 +···+kj
which we collectively denote as x̃j . We denote the contribution from the j th subdiagram
by Dj (zj , x̃j ). Using analytic continuation, the contribution of diagram b is given by
Z 5 Yk Y r Z
d y d5 zj
Ib = K∆−ε (x, y) K ∆ i (x i , y) Gδj (y, zj )Dj (zj , x̃j ), (A.20)
y05 i=1 j =1
(zj )0 5
where we have omitted the couplings. A divergence in the integral over y can only arise
from the region y ∼ x. The degree of divergence is given by
X
k X
r
α=∆− ∆i − δj − ε. (A.21)
i=1 j =1
Therefore, the integral over y is divergent (for ε = 0) if and only if the vertex at y
P P
is extremal, rj =1 δj = ∆ − ki=1 ∆i . In this case (Case 1 in the text), the integral is
dominated by the region y ∼ x which gives rise to a pole 1/ε:
Z r Z
dy0 ε Y Y
k
1 d5 zj
Ib ∼ y0 Kδ (x, zj )Dj (zj , x̃j ) + Rb (A.22)
y0 (x − xi )∆i [(zj )0 ]5 j
R i=1 j =1
Fig. 14. Contact (a) and exchange (b) contributions to a subdiagram of the exchange diagram in
Fig. 10.
1Y Y
k r
1
∼ Hj (x, x̃j ) + Rb0 , (A.23)
ε (x − xi )∆i
i=1 j =1
where Rb and Rb0 are regular in ε and

Z
d5 zj
Hj (x, x̃j ) ≡ Kδj (x, zj )Dj (zj , x̃j ). (A.24)
(zj )0 5
Since the extremal coupling contains a factor of ε, the final result is finite and factors into
k two-point functions times a product of r functions. The j th of these functions has kj + 1
points. This structure is illustrated in Fig. 12.
P P
Consider now the case when the vertex at y is sub-extremal: rj =1 δj > ∆ − ki=1 ∆i .
Then we have α < 0 (we set ε = 0 now) and the integral over y is convergent. We also have
to study the behaviour of the integrand when some of the other internal points approach
the boundary. As we have shown above, (logarithmic) divergences only arise from extremal
vertices. They manifest as poles in analytic continuation and are cancelled by the zeros in
the corresponding extremal couplings. The diagram is then finite even if the coupling of the
vertex at y is not included. Therefore if this coupling vanishes, as in Case 2 in the text, the
diagram gives zero. Finally, for Case 3 in the text we need to show that the diagram vanishes
p p
when one of the subdiagrams is an Ẽkj +1 function (i.e., an Ekj +1 function with the bulk-to-
boundary propagator of highest dimension changed by a bulk propagator) and p 6 kj − 2.
This subdiagram can be either contact or exchange (diagrams a and b of Fig. 14). The
power counting shows that the integral over zj in the contact diagram is convergent. This
is all we need for diagram a. In diagram b we can distinguish again two possibilities:
1. If the vertex at zj is extremal with δj the highest dimension, according to the analysis
above one can only have a logarithmic divergence in the integration over zj when
y → x and zj → x. The behaviour of the integrand is then
Pr Pk 0 0
y0 l=1 δl + i=1 ∆i −∆−1−ε [(zj )0 ]ε −1 ,
where ε0 is a regulator of the zj integral. The 1/ε0 pole in the zj integral is cancelled
by a ε0 factor in the corresponding extremal coupling. Moreover, as the vertex at y is
sub-extremal the region y ∼ x of the integral over y gives a vanishing contribution.

Therefore, the region y ∼ x, z ∼ x does not contribute to diagram b. On the other
hand if y is away from x there is no divergence in the zj integral, and the zero in the
coupling ensures that the diagram vanishes.
2. If the vertex at zj is sub-extremal, the integral over zj converges. The divergences
in the remaining integrals are cancelled by zeros in the corresponding couplings.
Therefore the diagram vanishes if the coupling of the vertex at zj vanishes. As we
show in the text, if this coupling does not vanish the subdiagram contains at least one
(non-trivial) subdiagram that is a an Ẽkm0 +1 function with m 6 kj0 − 2 and the same
j
procedure can be employed to analyze it.
The iteration ends when the last Ẽ function is a contact diagram.
References
[1] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.

[2] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[3] E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[4] M. Bianchi, S. Kovacs, Phys. Lett. B 468 (1999) 102, hep-th/9910016.
[5] J. Erdmenger, M. Pérez-Victoria, Non-renormalization of next-to-extremal correlators in N = 4
SYM and the AdS/CFT correspondence, Phys. Rev. D 62 (2000) 045008, hep-th/9912250.
[6] G. Arutyunov, S. Frolov, On the correspondence between gravity fields and CFT operators,
JHEP 0004 (2000) 017, hep-th/0003038.
[7] F. Gonzalez-Rey, I.Y. Park, K. Schalm, Phys. Lett. B 448 (1999) 37, hep-th/9811155.
[8] B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Nucl. Phys. B 557 (1999) 355, hep-
th/9811172.
[9] B. Eden, C. Schubert, E. Sokatchev, Three-loop four-point correlator in N = 4 SYM, Phys.
Lett. B 482 (2000) 309, hep-th/0003096.
[10] G. Arutyunov, S. Frolov, Four-point functions of lowest weight CPOs in N = 4 SYM(4) in
supergravity approximation, Phys. Rev. D 62 (2000) 064016, hep-th/0002170.
[11] L. Hoffmann, A.C. Petkou, W. Rühl, Aspects of the conformal operator product expansion in
AdS/CFT correspondence, hep-th/0002154.
[12] G. Chalmers, K. Schalm, Nucl. Phys. B 554 (1999) 215, hep-th/9810051.
[13] M. Gunaydin, L.J. Romans, N.P. Warner, Phys. Lett. B 154 (1985) 268.
[14] M. Pernici, K. Pilch, P. van Nieuwenhuizen, Nucl. Phys. B 259 (1985) 460.
[15] S. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Adv. Theor. Math. Phys. 2 (1998) 697, hep-
th/9806074.
[16] D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Nucl. Phys. B 546 (1999) 96, hep-
th/9804058.
[17] E. D’Hoker, D.Z. Freedman, W. Skiba, Phys. Rev. D 59 (1999) 045008, hep-th/9807098.
[18] F. Gonzalez-Rey, B. Kulik, I.Y. Park, Phys. Lett. B 455 (1999) 164, hep-th/9903094.
[19] S. Penati, A. Santambrogio, D. Zanon, JHEP 9912 (1999) 006, hep-th/9910197.
[20] B. Eden, P.S. Howe, P.C. West, Phys. Lett. B 463 (1999) 19, hep-th/9905085.
[21] P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Explicit construction of nilpotent covariants
in N = 4 SYM, Nucl. Phys. B 571 (2000) 71, hep-th/9910011.
[22] K. Intriligator, Nucl. Phys. B 551 (1999) 575, hep-th/9811047.
[23] A. Petkou, K. Skenderis, Nucl. Phys. B 561 (1999) 100, hep-th/9906030.
[24] S. Ferrara, E. Sokatchev, Short representations of SU(2, 2/N) and harmonic superspace
analyticity, hep-th/9912168.
[25] V.K. Dobrev, V.B. Petkova, Lett. Math. Phys. 9 (1985) 287.
[26] W. Skiba, Phys. Rev. D 60 (1999) 105038, hep-th/9907088.
[27] M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, JHEP 9908 (1999) 020, hep-th/9906188.
[28] M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, Anomalous dimensions in N = 4 SYM theory at
order g 4 , Nucl. Phys. B 584 (2000) 216, hep-th/0003203.
[29] E. D’Hoker, D.Z. Freedman, S. Mathur, A. Matusis, L. Rastelli, Extremal correlators in the
[30] B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Phys. Lett. B 472 (2000) 323, hep-
th/9910150.
[31] G. Arutyunov, S. Frolov, Phys. Rev. D 61 (2000) 064009, hep-th/9907085.
[32] S. Lee, AdS(5)/CFT(4) four-point functions of chiral primary operators: cubic vertices, Nucl.
Phys. B 563 (1999) 349, hep-th/9907108.
[33] G. Arutyunov, S. Frolov, Scalar quartic couplings in type IIB supergravity on AdS5 × S 5 , Nucl.
Phys. B 579 (2000) 117, hep-th/9912210.
[34] I. Klebanov, E. Witten, Nucl. Phys. B 556 (1999) 89, hep-th/9905104.
[35] E. D’Hoker, D.Z. Freedman, Nucl. Phys. B 550 (1999) 261, hep-th/9811257.
[36] E. D’Hoker, S.D. Mathur, A. Matusis, L. Rastelli, The operator product expansion of N = 4
SYM and the 4-point functions of supergravity, Nucl. Phys. B 589 (2000) 38, next article in this
issue, hep-th/9911222.
[37] E. D’Hoker, D.Z. Freedman, L. Rastelli, Nucl. Phys. B 562 (1999) 395, hep-th/9905049.
[38] E. D’Hoker, D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Nucl. Phys. B 562 (1999)
353, hep-th/9903196.
The operator product expansion of N = 4 SYM and

the 4-point functions of supergravity
Eric D’Hoker a , Samir D. Mathur b , Alec Matusis c , Leonardo Rastelli d,∗
a Department of Physics, University of California, Los Angeles, CA 90095, USA
b Department of Physics, Ohio State University, Columbus, OH 43210, USA
c Department of Physics, Stanford University, Stanford, CA 94305, USA
d Center for Theoretical Physics, Massachusetts Institute of Technology, Cambridge, MA 02139 USA
Received 18 April 2000; revised 3 August 2000; accepted 11 August 2000
Abstract
We give a detailed Operator Product Expansion interpretation of the results for conformal 4-point
functions computed from supergravity through the AdS/CFT duality. We show that for an arbitrary
scalar exchange in AdSd+1 all the power-singular terms in the direct channel limit exactly match the
corresponding contributions to the OPE of the operator dual to the exchanged bulk field and of its
conformal descendents. The leading logarithmic singularities in the 4-point functions of protected
N = 4 super-Yang–Mills operators (computed from IIB supergravity on AdS5 × S 5 ) are interpreted
as O(1/N 2 ) renormalization effects of the double-trace products appearing in the OPE. Applied to
e + · · ·), this analysis leads
the 4-point functions of the operators Oφ ∼ (trF 2 + · · ·) and OC ∼ (trF F
to the prediction that the double-trace composites [Oφ OC ] and [Oφ Oφ − OC OC ] have anomalous
dimension −16/N 2 in the large N, large gYM 2 N limit. We describe a geometric picture of the OPE
in the dual gravitational theory, for both the power-singular terms and the leading logarithms. We
comment on several possible extensions of our results.  2000 Elsevier Science B.V. All rights
reserved.
1. Introduction
The study of 4-dimensional Conformal Field Theories is an old and important topic.
The AdS/CFT correspondence [1,2] provides new powerful tools to address this problem.
Difficult dynamical questions about the strong coupling behavior of the CFT are answered
by perturbative computations in anti-de Sitter supergravity. A natural set of questions
concerns the nature of the Operator Product Expansion (OPE) of the CFT at strong
coupling. Thanks to the AdS/CFT duality, we can now answer some of these questions.
∗ Corresponding author.
E-mail addresses: dhoker@physics.ucla.edu (E. D’Hoker), mathur@pacific.mps.ohio-state.edu
(S.D. Mathur), alecm@leland.Stanford.edu (A. Matusis), rastelli@ctp.mit.edu (L. Rastelli).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 2 3 - X
The prime example of an exactly conformal 4-dimensional field theory, namely the
N = 4 super-Yang–Mills theory with gauge group SU(N), is dual [1] to Type IIB
string theory on AdS5 × S 5 , with N units of 5-form flux and compactification radius
R 2 = α0 (gYM
2 N)1/2 . For large N and large ’t Hooft coupling λ ≡ g 2 N the dual string
YM
theory is approximated by weakly coupled supergravity in AdS5 background. Since the
5-dimensional Newton constant G5 ∼ R 3 /N 2 , the perturbative expansion in supergravity
corresponds to the 1/N expansion in the CFT. Correlation functions of local operators of
the CFT belonging to short multiplets of the superconformal algebra are given for N →
∞, λ → ∞ by supergravity amplitudes according to the prescription of [4,5]. While the
supergravity results for 2- and 3-point functions of chiral operators [6–9] have been found
to agree with the free field approximation, giving strong evidence for the existence of
non-renormalization theorems [10–16], 4-point functions [17–33] certainly contain some
non-trivial dynamical information. 1
The 4-point functions of the operators Oφ and OC dual to the dilaton and axion
fields were obtained in [26] through a supergravity computation, and expressed as very
explicit series expansions in terms of two conformal invariant variables. The fundamental
fields of the N = 4 theory are the gauge boson Aµ , 4 Majorana fermions λa and
6 real scalars Xi , all in the adjoint representation of the gauge group SU(N). The
operators Oφ and OC are the exactly marginal operators that correspond to changing the
gauge coupling and the theta angle of the theory. In other terms the SYM Lagrangian
has the form L ∼ (1/gYM 2 )O + (θ/8π 2 )O . It is convenient to define operators that
φ C
have unit-normalized 2-point functions, O bφ ∼ (1/N)Oφ ∼ (1/N) tr(F 2 + · · ·) and O bC ∼
e
(1/N)OC ∼ (1/N) tr(F F + · · ·).
The computation of the axion–dilaton 4-point functions required the sum of several
supergravity diagrams, weighted by the appropriate couplings in the Type IIB action on
AdS5 × S 5 ,
Z
1 5 √ 1 µν 1 2φ µν
S= 2 d z g −R + Λ + g ∂µ φ∂ν φ + e g ∂µ C∂ν C . (1.1)
2κ5 2 2
AdS5
Besides these ‘complete’ dilaton–axion 4-point functions, explicit results for arbitrary
supergravity exchange diagrams involving massive scalars, massive vectors and massless
gravitons are also available [27]. In the present paper we give a detailed OPE interpretation
of some of these results, and obtain new predictions for the strong coupling behavior of
the N = 4 SYM theory. The fact that the 5d supergravity amplitudes can be consistently
interpreted in terms of a 4d local OPE is by itself quite remarkable, and constitutes a strong
test of the AdS/CFT duality.
Let us introduce the main issues from the field theory viewpoint. By considering the limit
of a 4-point function as the operator locations become pairwise close (take a ‘t-channel’
limit x13 ≡ |Ex1 − xE3 | → 0 and x24 ≡ |E
x2 − xE4 | → 0), we expect a double OPE expansion to
hold:
1 Perturbative studies of 4-point functions in N = 4 SYM include [34–43].


X αn hOn (E
x1 )Om (Ex2 )iβm
O1 (E
x1 )O2 (E
x2 )O3 (E x4 ) =
x3 )O4 (E +∆ −∆
, (1.2)
n,m
(x13)∆1 3 m (x24 )∆2 +∆4 −∆n
at least as an asymptotic series, and hopefully with a finite radius of convergence in x13
and x24 . For simplicity we have suppressed all Lorentz and flavor structures and generically
denoted by {On } the set of primary operators Op and their conformal descendents 5k Op .
Let us take the operators in the 4-point function to be ‘single-trace’ 2 chiral primaries
tr(X(i1 . . . Xik ) ) or any of their superconformal descendents, such as Oφ ∼ tr(F 2 + · · ·).
These operators belong to short representations of the superconformal algebra 3 and their
dimensions do not receive quantum corrections. On purely field-theoretic grounds, we
expect that all the operators allowed by selection rules (most of which are not chiral) can
contribute as intermediate states to the r.h.s. of (1.2).
The AdS/CFT duality makes the interesting prediction that the non-chiral operators of
the SYM theory actually fall into two classes, 4 which behave very differently at strong
coupling:
• Operators dual to string states, like for example the Konishi operator trXi Xi , whose
dimensions become very large in the strong coupling limit (as ∼ λ1/4 );
• Multi-trace operators obtained by taking (suitably regulated) products of single-trace
chiral operators at the same point, like for example the normal-ordered product
[Oφ Oφ ]. These operators are dual to multi-particle supergravity states.
Since in the limit of large N , large λ the dual supergravity description is weakly coupled,
the dimension of a multi-particle state is approximately the sum of the dimensions of the
single-particle (single-trace in SYM language) constituents, with small corrections of order
G5 ∼ 1/N 2 due to gravitational interactions. From a perturbative analysis in the SYM
theory it is not hard to show (see Section 4.2) that for large N and small λ the anomalous
dimension of an operator like [Oφ Oφ ] is of the form ∼ (1/N 2 )f (λ) + O(1/N 4 ), where
P
f (λ) can be computed as a perturbative series f (λ) = k>1 ak λk . The AdS/CFT duality
then predicts that as λ → ∞ the function f (λ) saturates to a finite value.
A non-trivial issue is whether in the double OPE interpretation of the supergravity
amplitudes one can find any remnant of the non-chiral operators corresponding to string
states. In fact, although these operators acquire a large anomalous dimension as λ → ∞,
an infinite number of them is exchanged in the r.h.s. of (1.2) for any finite λ, and one may
worry about a possible non-uniformity of the limit. On the contrary, our analysis will lend
support to the idea that as λ → ∞ the string states consistently decouple.
Since each single-trace chiral operator O of the SYM theory is dual to some Kaluza–
Klein mode φ of supergravity, there appears to be a 1–1 correspondence between
supergravity diagrams in which φ is exchanged in the ‘t-channel’ (that is, the bulk-to-
bulk φ propagators joins the pairs O1 O3 and O2 O4 , see Fig. 1) and the contribution
to the double OPE (1.2) of the operator O and its conformal descendents 5k O. In
2 The trace is over the color group SU(N ). Single-trace operators in SYM are dual to single-particle Kaluza–
Klein states in supergravity [44].
3 We will call ‘chiral’ any operator belonging to a short multiplet.
4 Group theoretic aspects of string and multi-particle states are considered in [45–47].
Fig. 1. Scalar exchange in the ‘t-channel’.
Section 2 we prove a general theorem: 5 for any scalar exchange 6 in AdSd+1 all the
singular terms O(1/x13n ) exactly match the corresponding contributions of the conformal
block {5 O} to the double OPE (1.2) of the d-dimensional boundary CFT. We believe
k
that a similar theorem must hold for exchanges of arbitrary spin. The correspondence
between supergravity exchanges and ‘conformal partial waves’ breaks down precisely
when the double-trace operator [O1 O3 ], which has dimension ∆1 + ∆3 + O(1/N 2 ), starts
contributing to the OPE. This result implies that as λ → ∞ the singular part of the OPE
of two chiral SYM operators is entirely given by other chiral operators and their multi-
trace products. This is of course consistent with the expectation that non-chiral operators
corresponding to string states have a large dimension in this limit.
A generic feature of supergravity 4-point amplitudes is that their asymptotic expansions
contain logarithmic terms. For example, a ‘t-channel’ exchange diagram (Fig. 1) contains
as x13 → 0 a logarithmic singularity of the form

1 x13 x24
2∆1 +2∆2
log ,
x12 x12 x34
n
as well as a whole series of regular terms x13 log(x13 ). (All these terms are subleading
with respect to the power singularities O(1/x13 k ) discussed above. The expansion of the
same diagram in the limit x12 → 0 contains instead no power singularities, and the
logarithmic term is leading.) This logarithmic behavior may appear puzzling in a unitary
CFT. However, as stressed to us by Witten early in this work, logs naturally arise in the
perturbative expansion of a CFT from anomalous dimensions and operator mixing. 7
Section 3 of the paper is devoted to a general discussion of the logarithmic behavior of
CFT’s. The difference between logs that arise in a perturbative expansion of a unitary
theory and the ‘intrinsic logs’ of a non-unitary theory is emphasized. In our case the
perturbative parameter is 1/N . As already noted, we expect operators like [Oφ Oφ ] to have
anomalous dimensions of order O(1/N 2 ). The logs in the supergravity 4-point functions
arise indeed at the correct order and with the right structure to be interpreted as O(1/N 2 )
5 A similar result has been obtained in [21] in a rather different formalism.

6 We restrict for simplicity to pairwise equal external dimensions ∆ = ∆ 6 ∆ = ∆ .
1 3 2 4
7 See, e.g., [48,49] and references therein for examples of this connection between logarithmic terms in CFT
and anomalous dimensions.
renormalization effects of the double-trace composites produced in the OPE of two chiral
operators.
In Section 4, we perform a careful analysis of the leading logarithmic terms in the
supergravity correlators hOφ Oφ Oφ Oφ i and hOφ OC Oφ OC i. In order to reproduce the
structure of the supergravity logs it is crucial to take into account the mixing between
operators with the same quantum numbers, like [Oφ Oφ ] and [OC OC ]. This analysis
leads to the prediction of the strong coupling values of the anomalous dimensions of
the operators [Oφ OC ] and [Oφ Oφ − OC OC ], which are the only two operators with the
maximal UY (1) charge |Y | = 4 and thus cannot mix with any other operator of approximate
dimension 8.
In Section 5 we present our conclusions and propose some avenues for future research.
2. Supergravity exchanges versus OPE: power singularities
There is an intriguing relation [17,18] between supergravity exchange diagrams and

‘conformal partial waves’. A conformal partial wave is the contribution to the double OPE
representation (1.2) of a full conformal block, which consists of a given primary operator
Op and all its conformal descendents 5k Op . Let us take the external operators Oi in
the l.h.s. of (1.2) to be single-trace, chiral SYM operators and let us consider the partial
waves in which the intermediate primaries are also single-trace and chiral. These are the
operators which are in 1–1 correspondence with the single-particle Kaluza–Klein states
of supergravity. It is then clear that for each such conformal partial wave one can draw a
related supergravity diagram, in which the dual KK mode is exchanged in the bulk, 8 see
Fig.1.
Here we wish to compare supergravity scalar exchange diagrams in AdSd+1 with
conformal partial waves in d-dimensional CFT’s. We shall find that for a given partial wave
all the singular terms in the OPE are exactly reproduced by the corresponding supergravity
exchange. However the higher order terms are different. 9
The conformal partial wave for an intermediate scalar primary is an old result [50,51].
Consider for simplicity the 4-point function of scalar operator Oi with pairwise equal
dimensions (∆1 = ∆3 , ∆2 = ∆4 ). Introducing the conformal invariant variables
2 x2
x13 2 x2 − x2 x2
x12
1
s≡ 24
2 x2 + x2 x2
, t≡ 34 14 23
2 x2 + x2 x2
. (2.1)
2 x12 34 14 23 x12 34 14 23
The contribution from an intermediate operator O∆ and its conformal descendents can be
written as (see Appendix A for the conversion from the form in [51] to our notations):
8 For a given primary O to contribute to the double OPE, the 3-point functions hO O O i and hO O O i
p 1 3 p 2 4 p
must be non-vanishing: this condition translates on the supergravity side to the existence of cubic couplings
φ1 φ3 φp and φ2 φ4 φp .
9 This is contrary to the claim in [18] of an exact equivalence between supergravity exchanges and conformal
partial waves, but compatible with the results in [21].

1 0(∆)0 ∆ + 1 − d
O1 (x1 )O2 (x2 )O3 (x3 )O4 (x4 ) 0 = 2∆1 2∆2
2 4
2
x13 x24 0 ∆2
∞ 2
X ∆ +n 0 ∆
+n
× s 2 a ∆ +n−1 (t) 2
,
n=0
2 0 n+∆− d
2 + 1 n!
(2.2)
where

√ 0(k + 1) k+1 k 3 2
ak (t) = π F , + 1; k + ; t . (2.3)
0 k + 32 2 2 2
Here all operators are normalized to have unit two-point functions and the correlators
hO1 O3 O∆ i and hO2 O4 O∆ i are also assumed to have coefficient 1. Observe that the
singular terms in the limit x13 → 0 are given by n 6 ∆1 − (∆/2) − 1.
The supergravity ‘t-channel’ exchange diagram of Fig. 1 in which the field φ dual to
O propagates in the bulk can be expressed in a similar series expansion in terms of the
variables s and t (see Appendix A for details). Let ∆1 6 ∆2 . It is found that in the limit
x13 → 0 the terms containing power singularities O(1/x13 k ) are

1 0(∆)0 ∆ + 1 − d2
O∆1 O∆2 O∆3 O∆4 sing = 2∆ 2∆ 2 4
x13 1 x24 2 0 ∆2
∆1 − ∆
2 −1
2
X ∆ +n 0 ∆2 + n
× s 2 a ∆ +n−1 (t) , (2.4)
n=0
2 0 n + ∆ − d2 + 1 n!
in precise agreement with (2.2). The full series expansion of the supergravity diagram
is however different from (2.2), for example logarithmic terms ∼ log s arise at the first
non-singular order. Taking the limit x24 → 0 (keeping the condition ∆1 6 ∆2 ) one finds
that not all singular powers O(1/x24 k ) match, but only the terms more singular than
2∆2 −2∆1
O(1/x24 ). This is precisely the singularity expected from the contribution to the OPE
of the double-trace operator [O1 O3 ], of approximate dimension 2∆1 . Not surprisingly,
the correspondence between conformal partial waves and AdS exchanges breaks down
precisely when double-traces start to contribute.
We expect similar results for arbitrary spin exchange. Expressions for conformal partial
waves for arbitrary spin can be found for example in [52]. The exchange supergravity
diagrams for vectors of general mass and massless gravitons have been evaluated in
[20,26,27], where it was also checked that the leading power singularity reproduced the
contribution expected from OPE considerations (see (4.23) of [20] and Section 2.3 of
[26]). It would be interesting to extend the comparison to all the subleading power-singular
singular terms.
This nice holographic behavior of the AdSd+1 supergravity exchange diagrams (their
power singularities match the d-dimensional conformal OPE) can also be understood in
the following heuristic way [17]. The exchange amplitude is given by an integral over the
two bulk interaction points z and w
Fig. 2. Upper-half plane representation of an exchange integral in AdS space. In the limit x1 → x2
the z-integral is supported in a small ball (in coordinate units) close to the boundary point x1 ≈ x2 .
Z
I= [dz] [dw] K∆1 (z, xE1 )K∆3 (z, xE3 )G∆ (z, w)K∆2 (w, xE2 )K∆4 (w, xE4 ), (2.5)
where K and G stand for the boundary-to-bulk and bulk-to-bulk propagators, and
[dz], [dw] denote the invariant measures. Take for concreteness the upper-half plane
representation of AdS
1
ds 2 = 2 dz02 + (dEz)2 . (2.6)
z0
Then
∆
z0
K∆ (z, xE) ∼ .
z02 + (Ez − xE)2
It is easy to prove that as we let xE1 → xE3 , the z-integral is dominated by a small coordinate
region, {z s.t. [z02 + (Ez − xE1 )2 ] ∼ x13
2 }, which is approaching the insertion points on the AdS
boundary of the two colliding operators O(E x1 )O(Ex3 ) (see Fig. 2). This is another example
of the UV/IR connection: short distances in the field theory are probed by large distance
physics in the AdS description. As x13 → 0 we can then approximate 10
1
G∆ (z, w) −→ z∆ K∆ (Ex1 , w), (2.7)
2∆ − d 0
and we get the expected factorization of (2.5) into a trivial integral over z and an integral
over w which defines the 3-point function hO∆ O2 O4 i:
1

I −→ O∆ (E
x1 )O2 (E
x2 )O4 (E
x4 ) . (2.8)
x13 ∆1 +∆3 −∆
(The numerical coefficients work out exactly.) The replacement (2.7) gives a clear
geometric equivalent, on the supergravity side, of the operator product expansion
10 This relation can be proven by taking z → 0 in the explicit functional form of the normalized G as given
0 ∆
for example in (2.5) of [27].
−∆ −∆ +∆
O1 (E x3 ) → x13 1 3 O∆ (E
x1 )O3 (E x1 ). It is possible to compute the first few higher-order
corrections to (2.7) and to match them exactly [17] with the singular contributions to the
operator product O1 O3 of the descendents 5k O∆ .
So far we have analyzed the power singularities in the ‘direct channel’ limit of an
exchange graph, i.e., when we let approach together two operators that join to the same
bulk interaction vertex (x13 → 0 or x24 → 0 in (2.5)). A given 4-point function is obtained
by summing all the crossed-symmetric exchanges, as well as ‘quartic’ graphs (diagrams
with a single bulk interaction vertex, Eq. (A.5). Thus in analyzing the singular behavior of
a 4-point correlator we also need to consider the type of leading singularities that appear in
quartic graphs, as well as in the ‘crossed channel’ limit of an exchange graph (for example,
x12 → 0 in (2.5)). It turns out that these two cases (crossed exchanges and quartic) have the
∆1 +∆2 −∆3 −∆4
same qualitative behavior: 11 as x12 → 0, the leading asymptotic is O(1/x12 )
if ∆1 + ∆2 − ∆3 − ∆4 > 0 or log(x12) if ∆1 + ∆2 − ∆3 − ∆4 = 0, and the limit is smooth if
∆1 + ∆2 − ∆3 − ∆4 < 0. These are the singularities expected from the contribution to the
operator product O1 O2 of the composite operator [O3 O4 ], which has dimension ∆3 + ∆4
in the large N limit.
The results of this section have a clear implication for the N = 4 SYM theory: in the
limit of large N , large λ, the only singular terms in the product of two chiral operators are
given by other chiral operators and their multi-trace products (and their first few conformal
descendents).
3. On the logarithmic behavior of conformal field theories
3.1. General analysis
Let us analyze the issue of logarithms in general, for an arbitrary CFT, before returning to
the case of the N = 4 supersymmetric Yang–Mills in 4 dimensions. A CFT is characterized
by the absence of any inherent length scale. Under quite general conditions one can argue
that for primary operators of the conformal algebra the 2-point function is forced by
conformal invariance to have the form

C
O(x1 )O(x2 ) = . (3.1)
|x1 − x2 |2∆
The power law on the r.h.s. is covariant under scale transformations: if we write x = µx 0 ,
then the 2-point function will be left unchanged provided we also change the operators
from O to O0 = µ−∆ O.
One might imagine that a logarithmic dependence log(|x1 − x2 |/L) would violate
conformal invariance of the theory, since a length scale L is needed to make the argument of
the logarithm dimensionless. There exists however a class of 2d theories called logarithmic
CFTs, where logarithms do arise [53]. These theories will not be the focus of our interest
later on, so we mention them now and then exclude them from the rest of the discussion
11 These statements can be proved by the methods reviewed in Appendix A.

below. In logarithmic CFTs the dilation operator cannot be diagonalised, but (in the
simplest case) has a Jordan form instead on a pair of operators O, O. e The 2-point functions
are of the form

1
O(z1 )O(z2 ) = 0, e 1 )O(z2 ) =
O(z ,
(z1 − z2 )2∆

1 z1 − z2
e 1 )O(z
O(z e 2) = log . (3.2)
(z1 − z2 )2∆ Y
(In the above we have considered only the holomorphic parts of the operators.) Under
a dilation z = µz0 we must not only rescale the fields but also implement a shift
transformation: O0 = µ−∆ O, O e0 = µ−∆ [O e + 1 log(µ)O]. With this change of variables
2
the correlators of the rescaled theory become identical to the original ones, and so the
parameter Y in (3.2) does not represent a fundamental length in the theory, but instead
describes a certain choice for the operator O e from the subspace of operators O, O
e of the
same dimension.
These logarithmic CFTs are however not expected to be unitary. In a radial quantization
the dilation operator is the Hamiltonian. In 2d CFT’s the eigenvalues of the dilation
operator on the plane map to the eigenvalues of the time translation operator on the
cylinder, so in a unitary theory we expect that the dilation operator will be diagonalisable
and will not have a nontrivial Jordan form. Our case of interest (the 4d N = 4
supersymmetric Yang–Mills theory) is a unitary theory, and we will assume in what follows
that the dilation operator is diagonalisable on the space of fields. The analysis of the next
section will confirm this assumption.
For unitary CFTs the 3-point function of primary fields also has a standard form which
is fixed by conformal invariance

γ123
O1 (x1 )O2 (x2 )O3 (x3 ) = ∆ +∆ −∆ ∆ +∆ −∆ ∆ +∆ −∆ , (3.3)
1 2 3
x12 x23 3 1 x131 3 2
2
where xij ≡ |xi − xj |. 4-point functions are however not fixed in their functional form by
conformal invariance, though they are restricted to be a function of two cross ratios

1
O1 (x1 )O2 (x2 )O3 (x3 )O4 (x4 ) = Σ −∆ ∆ +∆ ∆ −∆ Σ −∆ f (ρ, η),
13 13 13 24
x13 x12 x1413 24 x2424 13
(3.4)
2 x2
x12 x 2 x2
η = 2 34 2
, ρ = 14 23
2 x2
, (3.5)
x13 x24 x13 24
where Σij = ∆i + ∆j , ∆ij = ∆i − ∆j . One might imagine that the expansion of f in say
η would in general contain a logarithm:
f (η, ρ) = · · · + f−1 (ρ)η−1 + f0 (ρ) + f˜0 (ρ) log η + f1 (ρ)η + · · · , (3.6)
so that the x1 → x2 limit of the 4-point function would contain a logarithmic term ∼
log x12 . But consider evaluating the 4-point function using the OPE:
X ∆ −∆ −∆
O1 (x1 )O2 (x2 ) = Op (x2 )x12p 1 2 ,
p
X ∆ −∆3 −∆4
O3 (x3 )O4 (x4 ) = Oq (x4 )x34q ,
q

X
∆ −∆ −∆ ∆ −∆ −∆
O1 (x1 )O2 (x2 )O3 (x3 )O4 (x4 ) = Op (x2 )Oq (x4 ) x12p 1 2 x34q 3 4 . (3.7)
p,q
We have written the functions arising in the OPE in symbolic form: the operators Op will
in general carry tensor indices and these can contract with the unit vector along x1 − x2 .
But the basic point that we wish to observe is the following. If Op is a conformal primary,
then the coefficient function appearing in the OPE of O1 and O2 can be deduced from the
3-point function (3.3), and contains no logarithm. If Op is a conformal descendant, then
the coefficient function will be obtained as derivatives of the coefficient function for the
corresponding primary, and so again there will be no logarithm in these functions. Similar
arguments yield that if the 2-point functions of primaries contain no logarithms, then nor
do the two point functions hOp (x2 )Oq (x4 )i appearing in the last line of the above equation.
Thus if the OPE sums in (3.7) converge, we conclude that the 4-point function does not
in fact have a logarithm in x12 in an expansion around x12 = 0. In unitary 2d CFTs such
OPE sums do yield the correct 4-point functions, and logarithms do not arise in the short
distance expansions.
Let us now consider the circumstances where we will encounter logarithms in our
analysis of CFT correlation functions. Suppose that we study a 1-parameter family of
CFTs; let us denote this parameter by a. Suppose that the theory at a particular value of
the parameter, say zero, is particularly simple. Then we may ask for the n-point correlation
functions in a series in the parameter around zero. The example that we are concerned
with is of course N = 4 SU(N) SYM, where the theory for 1/N → 0 is expected to be
simple, 12 and the correlators may be studied in a series in 1/N . Consider first the 2-point
function (3.1). Let
∆ = ∆(0) + ∆(1), (3.8)
where ∆(1) vanishes as a → 0. Then we may write
(1)

C 1 L 2∆
O(x1 )O(x2 ) = (3.9)
2∆(0) L2∆(1) x12
x12

C 1 x12
≈ 1 − 2∆ (1)
log . (3.10)
2∆(0) L2∆(1)
x12 L
Here we have introduced a length scale L to be able to write the logarithms in
dimensionless form: L may be chosen in an arbitrary way since it just defines the
normalisation of the operators, and is needed because the operators O have at a generic
value of a a dimension that is different from the one at a = 0.
Thus we see that the leading correction to the 2-point function at nonzero a has a
logarithm in x12 , with a coefficient that depends in the manner shown on the correction ∆(1)
12 We are actually interested in the double limit N → ∞, λ → ∞. Since in the supergravity analysis there is no
remnant of the λ dependence, while the 1/N expansion coincides with the perturbative expansion in powers of
the Newton constant, G5 ∼ 1/N 2 , we restrict here to the N dependence alone.
to the dimension at nonzero a. Such a correction will of course be absent if the dimensions
of the operators do not change with a. Thus in particular for our case of interest, the two
point functions of SYM chiral operators will not have such a logarithmic correction. But we
can consider composite operators made from two chiral operators, and such operators will
in general have a logarithm in their 2-point function. We will compute the 4-point function
of four chiral operators; if we then take the limit x12 → 0, x34 → 0, then we can extract
the 2-point function of the nonchiral composites, and observe the logarithmic correction.
Now consider the 3-point function. Suppose for simplicity that the operators O1 and O2
have dimensions that are protected (i.e., unchanged under variations of the parameter a).
Let the dimension of O3 be ∆3 = ∆(0) + ∆(1) . Then we have

γ123
O1 (x1 )O2 (x2 )O3 (x3 ) =
∆1 +∆2 −∆(0) ∆2 +∆(0) −∆1 ∆(0) +∆1 −∆2
x12 x x13
23
1 x13 x23
× (1)
1 − ∆ log
(1)
. (3.11)
L2∆ x12 L
As mentioned above, what we will do is compute a 4-point function of chiral operators, and
the composite operator appearing in the above equation will be obtained in a limit where
two of the chiral insertions are taken to approach each other.
In the above we have considered the case of operators that change by a multiplicative
factor under scale transformations. In a unitary CFT we can always find a basis of operators
where the action of the dilation operator is thus diagonalised. But if we are considering a
one parameter family of theories, then operators which are dilation eigenstates at say a = 0
will not generically be such in the theory with a 6= 0. We will be interested in computing
only the leading order correction to the dimensions, so we can neglect mixing among
operators that have different dimensions at a = 0 (this is a familiar story, for example, from
ordinary quantum mechanics perturbation theory), and we need to consider only mixing
among operators which are degenerate at zeroth order. Let us consider the logarithms
arising at the first order in correction away from a = 0, in the case where there are two
or more degenerate operators at a = 0.
At this point for clarity of the discussion we specialize to the case of the theory that we
are going to study — the N = 4 supersymmetric SU(N) Yang–Mills theory, studied in a
1/N expansion around the large N limit.
Let Oi be single-trace chiral operators (i.e., chiral primaries trX(i1 · · · Xik ) or any of their
supersymmetry descendents). Superconformal symmetry fixes their scaling dimension to a
value independent of N . Recall that these operators are a small subset of all the operators
that are primaries of the conformal algebra. We take the Oi to be normalized such that

δij
Oi (x)Oj (0) = 2∆ . (3.12)
|x| i
We now construct the double-trace composite operators made from pairs of these primaries.
Let Oij be given through
1
[Oii (x)]y ≡ √ Oi (x + y)Oi (x) − subtractions, (3.13)
2
[Oij (x)]y ≡ Oi (x + y)Oj (x) − subtractions (i 6= j ). (3.14)

To define these composite operators we point-split the two chiral constituents by a distance
y and subtract the power-singular contributions in their OPE. In the supergravity limit of
N → ∞ and λ large the singular part of the OPE of two single-trace chiral operators O1
and O2 is given by other single-trace chiral operators with dimension ∆ < ∆1 + ∆2 (with
an OPE coefficient O(1/N)) and by double-trace composites [O3 O4 ] with ∆3 + ∆4 <
∆1 + ∆2 (with an OPE coefficient O(1/N 2 )). These statements follow from standard large
N counting rules and from the input of the AdS/CFT duality (confirmed by the analysis of
Section 2) that non-chiral operators dual to string states have a large anomalous dimension
in this limit.
Let us denote the composite indices {ij } by α, β . . . . Also, to simplify the notation we
will usually drop the explicit dependence on the cut-off distance y and simply indicate the
double-trace composites by Oα . Since the dimensions of the chiral operators are integral,
there are clearly several composites for which the sum of the dimensions of the chiral
constituents equals the same value. For N → ∞ all these operators will have the same
dimension. Let us consider the subspace of composites {Oα } that have the same dimension
∆ in the large N limit, ∆α = ∆i + ∆j = ∆ for all α. At 1/N = 0 the 2-point function of
the composites is obviously

δαβ
Oα (x)Oβ (0) = 2∆ . (3.15)
|x|
In the above we have assumed that y x.
For 1/N small but non-zero, these composites will mix among each other under scale
d
transformations. The infinitesimal dilation operator is simply y dy , so we have
d
y Oα = Mαβ Oβ . (3.16)
dy
It will be apparent from the supergravity analysis of the next section that Mαβ is a
real symmetric matrix, as expected in a unitary theory. We can then find the dilation
eigenstates and their anomalous dimensions by solving the eigenvalue problem for M.
d
Let the eigenvectors of the dilation y dy be the operators O(A)
O(A) = V(A)
α
Oα , Oα = Vα(A) O(A) , (3.17)
d
y O(A) = ∆(A) O(A) . (3.18)
dy
Composite operators [O(A) ]y defined with a certain value of the cut-off are related to the
composites with a different cut-off y 0 by a simple rescaling
∆(A)
y
[O(A) ]y = [O(A) ]y 0 . (3.19)
y0
The orthogonal matrix
b)α (A) = Vα(A)
(V (3.20)
accomplishes the change of basis that makes M diagonal

bD
Mαβ = V bV
bT , (3.21)
αβ
b = diag{∆(1), . . . , ∆(n) }.
D (3.22)
The full conformal dimension of O(A) is
∆ + ∆(A) , ∆(A) ∆. (3.23)
From the supergravity results (or from field-theory large N counting) we have that
∆(A) = O(1/N 2 ). Since conformal invariance implies that primary operators with different
dimensions are orthogonal, we have

δAB
O(A) (x)O(B) (0) a = |y|2∆(A) 2∆+2∆ , (3.24)
|x| (A)
where the power of y reflects the fact that we have chosen to normalize the composite
operators at scale y, for all values of the parameter 1/N . We then get
(B)

hOα (x)Oβ (0)ia = Vα(A) Vβ O(A) (x)O(B) (0) a
(B) δAB
= Vα(A) Vβ |y|2∆(A) 2∆+2∆
|x| (A)

(B) δAB x
≈ Vα(A) Vβ 1 − 2∆ (A) log . (3.25)
|x| 2∆ y
We see that if we have the 4-point function of chiral operators (assume ∆i + ∆j =
∆k + ∆l = ∆)

Oi (x + y)Oj (x)Ok (y 0 )Ol (0) , y, y 0 x, (3.26)
then we get to order O(1/N 2 ) a term
1 yy 0
2∆
log ,
x x2
with coefficient
X
Vα(A) Vβ(A) ∆(A) = Mαβ , α = (ij ), β = (kl). (3.27)
(A)
Thus from the knowledge of the leading logarithmic term in the 4-point functions of chiral
operators we can directly extract the mixing matrix Mαβ of the double-trace composites,
and by solving the eigenvalue problem we can then find the dilation eigenstates and their
anomalous dimensions. Observe that for α = β, i.e., when the chiral constituents are
pairwise equal (say i = k, j = l) the logarithmic term in (3.25) appears as a O(1/N 2 )
correction to the leading power behavior, which is of order O(1) in the large N counting
and is given in (3.26) by the disconnected contribution to the 4-point function (obtained
by simply contracting the equal chiral operators). However for α 6= β, i.e., in the case of
mixing between different composites, the log naturally appears to order O(1/N 2 ) without
a O(1) power term.
Let us finally relate this discussion to the Operator Product Expansion. From the very
definition of the composite double-trace operators, we have
Oi (x)Oj (0) ∼ singular terms + [Oα (0)]x , α = (ij ). (3.28)
We can rewrite
[Oα ]x = Vα(A) [O(A) ]x

β
= Vα(A) (x/y)∆(A) [O(A) ]y = Vα(A) (x/y)∆(A) V(A) [Oβ ]y
β β
≈ Vα(A) V(A) + Vα(A) ∆(A) V(A) log(x/y) [Oβ ]y
= [Oα ]y + Mαβ log(x/y)[Oβ ]y , (3.29)
b and the relation
where in the last step we have used the orthogonality of the matrix V
2
(3.21). Thus to order O(1/N ) the OPE takes the form
Oi (x)Oj (0) ∼ singular terms + [Oα (0)]y + Mαβ log(x/y)[Oβ (0))]y ,

α = (ij ). (3.30)
This equation translates in the OPE language the renormalization and mixing of the double-
trace composites. This is the form of the OPE required by compatibility with the action of
the dilation operator. In fact, the l.h.s. of the above OPE is clearly independent from the
d
scale y, and applying y dy to the r.h.s. it is immediate to see that we get identically zero
2
(to order O(1/N )) once the mixing (3.16) is taken into account.
3.2. Logarithms in the supergravity picture
The AdS/CFT correspondence provides an interesting dual picture of the Yang–Mills

theory, and in this dual picture we have a simple and elegant pictorial way of seeing the ap-
pearance of the above logarithms. Consider a 4-point function hO(x1 )O(x2 )O(x3 )O(x4 )i.
At 1/N = 0, the supergravity theory is free, and the 4-point function factorizes into the
product of the 2-point functions obtained by pairwise contractions of the operators. At
order O(1/N 2 ), we get contributions from tree–level supergravity graphs, either of the ex-
change type (with two cubic vertices, as for example Fig. 1) or quartic graphs with a single
interaction vertex, as in Fig. 3. These connected graphs include logarithmic corrections to
the correlator, which represent corrections to the lcomposite operators generated by the
approach of two chiral primaries. To see that the logarithm indeed arises from the vicinity
of the composite operator in the supergravity diagram, we look at a typical term that arises
in the supergravity description. Consider for simplicity the quartic graph, Fig. 3. The AdS
Fig. 3. Quartic graph.

Fig. 4. A quartic graph in the upper-half plane representation of AdS space. Each annulus is e times
bigger in diameter than the one nested inside it. The z-integration receives an equal contribution from
each annular region.
space is represented as the upper half space in Fig. 4. The quartic vertex is at a location z,
which must be integrated over. In Fig. 4 we have partitioned the domain of integration of
z into annular regions, each of which is e times in diameter compared to the one nested
inside it. The integration has the form (take for simplicity all conformal dimensions to be
equal)
Z 5
d z z0∆ z0∆ z0∆ z0∆
. (3.31)
z05 (z − x1 )2∆ (z − x2 )2∆ (z − x3 )2∆ (z − x4 )2∆
Let x1 , x2 be close to the point x = 0, and consider the region of integration where z
approaches z = 0 as well. The points x3 , x4 are assumed to be far away from this region.
Then we may approximate
z0∆ z0∆ z02∆

≈ . (3.32)
(z − x3 )2∆ (z − x4 )2∆ x32∆ x42∆
If we set x1 , x2 to zero the integral becomes

Z 5
1 d z −2∆ 2∆
z z0 , (3.33)
2∆
x3 x4 2∆
z05 0
which has an equal contribution from each annulus in Fig. 4, and so can be seen to diverge
logarithmically. The actual integral we have is cut off at x1 − x2 in the UV, and at ∼ x1 − x3
in the IR, and so is of order
x13
∼ log ,
x12
which is a logarithm of the kind that we will observe in the supergravity diagrams. In more
general integrals we will have terms with singular powers in x12 as well. 13 Further the
exchange graphs like that in Fig. 1 have two vertices z, w that are integrated over, and a
contribution log |x12 | can arise when one or both of these vertices are in the vicinity of
x1 , x2 .
13 When a single variable z is being integrated and the leading singularity is more singular than a logarithm
k ; it cannot be of the form ∼ (1/x k ) log x . Logarithms can however
then this singularity is of the form ∼ 1/x12 12 12
appear at subleading orders in x12 .
4. Dilaton–axion four point functions and anomalous dimensions
In this section we will consider the supergravity results [26] for 4-point functions of the
SYM operators Oφ , OC dual to the dilaton and axion fields. Following the logic of the
previous section, we can extract information about the O(1/N 2 ) anomalous dimensions
and mixings of the operators occurring in the OPE of Oφ and OC by looking at the
logarithmic behavior of the correlators.
The expressions for the 4-point functions of normalized operators (see (C.5)), for
λ → ∞ and up to order O(1/N 2 ), are summarized in Appendix C, Eqs. (C.7) and
(C.11). The disconnected graphs give some trivial powers of the separation, of order
O(1) in the large N counting, while the connected tree-level supergravity graphs provide
O(1/N 2 ) contributions which are non-trivial functions of the cross-ratios. The leading
logarithmic asymptotics are given in (C.24)–(C.26). They have the structure expected from
the contribution to the OPE of double-trace composites of dimensions 8 + O(1/N 2 ).
Let us first discuss the simplest case, the ‘s-channel’ limit (i.e., x12 → 0, x34 → 0) of
bφ (x1 )O
the correlator hO bC (x2 )O
bφ (x3 )O
bC (x4 )i. In this limit there are no power singularities
and the logarithmic term in (C.26) is the leading contribution. Thus from (C.11), (C.12)
and (C.26) we obtain one of the leading coefficient functions of the OPE of O bφ and ObC

Obφ (x1 )O
bC (x2 ) = Aφc (x12, y) O bC (x2 ) + · · · .
bφ O (4.1)
y
Here, [Obφ O
bC ]y is the composite (double-trace) operator defined by the above equation
with Aφc given to order O(1/N 2 ) and for λ → ∞ by

A x12
Aφc (x12, y) = 1 + 2 ln . (4.2)
N y
The numerical constant A is readily determined from the logarithmic asymptotics as x12 →
bφ O
0 of hO bC Obφ O
bC i given by (C.26):
A = −16. (4.3)
The above leading behavior of the coefficient function receives corrections both in inverse
powers of λ and in the 1/N expansion. For example, tree-level stringy corrections, of order
O(α03 ) with respect to the Einstein–Hilbert action, give a O(1/(N 2 λ3/2 )) contribution. The
first quantum corrections (one loop in supergravity) are of order O(1/N 4 ).
Next, from the ‘t-channel’ limit (x13 → 0, x24 → 0) of the same correlation function
hObφ ObC O
bφ ObC i, as well as of the correlators hO
bφ O
bφ O
bφ Obφ i and hO
bC ObC ObC O
bC i, we can
b b
extract terms in the OPE of two Oφ ’s and two OC ’s. We expect on general grounds that
the OPE will assume the schematic form

O bφ (x3 ) = I + T + ∂T + · · · + Cφφ O
bφ (x1 )O bφ O
bφ
8 4 3 y
x13 x13 x13

+ Cφc O bC O
bC + CφT [T T ] + · · · , (4.4)
y

O bC (x4 ) = I + T + ∂T + · · · + Ccφ O
bC (x2 )O bφ O bφ
8 4 3 y
x24 x24 x24

+ Ccc ObC O
bC + CcT [T T ] + · · · . (4.5)
y
We have suppressed for the sake of brevity the dependence of the coefficient functions
upon the positions xi , and the Lorentz structures in the stress-energy tensor terms. Unlike
the Obφ O
bC OPE, here there are some power-singular terms, arising from the contributions
of the stress-energy tensor and its first descendents. We discussed these terms in Section 2,
where we checked that the singular powers of the ‘direct channel’ supergravity exchange
graph (in this case, the t-channel graviton exchange) exactly match the predictions of the
OPE. 14
Let us then analyze the contributions to the above OPE’s of the double-trace compos-
bφ O
ites. Clearly, the correlator hO bC O bφ ObC i determines the coefficient functions Cφc =
b b b b
Ccφ , while hOφ Oφ Oφ Oφ i determines Cφφ and hO bC O
bC O
bC ObC i determines Ccc . Since
hObφ O
bφ Obφ O
bφ i = hO
bC ObC ObC O
bC i, we immediately have that Cφφ = Ccc . Thus,

C x13
Cφφ = Ccc = 1 + 2 ln , (4.6)
N y

B x13
Cφc = Ccφ = 1 + 2 ln . (4.7)
N y
From the logarithmic asymptotics (C.24), (C.25) of the 4-point functions, we find that 15
27 24 · 13
B= , C =− . (4.8)
3·7 3·7
Now, it is clear from the above OPE’s and identification of the coefficient functions
that we lack some information on the data for operators in the OPE whose dimension
is approximately 8. For example, to compute CcT and CφT would require knowledge of
correlators hTµν T µν Oφ Oφ i and hTµν T µν OC OC i, which are not at present available. The
correlator hT T T T i, which is even more out of reach, would also be necessary. In order to
extract the dilation eigenstates and their anomalous dimensions, we would need to compute
the full mixing matrix of the operators of dimension approximately 8 in the large N , large
λ limit. There are actually several such operators that we omitted when writing down the
OPE’s above. The reason is that many more operators of dimension 8 in the free-field
approximation may be constructed by taking the product of conformal descendants, such
as [∂J ∂J ] = [O0(3) O0(3) ], or even products of fermion operators. Clearly, obtaining the
full mixing matrix is possible but very involved, since it would require computing several
4-point functions in supergravity.
Fortunately, using the invariance of the theory under the UY (1) symmetry in the
supergravity limit, it is possible to disentangle the OPE and isolate some operators that
mix in a simple way. This is done in the next section.
4.1. The use of U (1)Y symmetry and anomalous dimensions
The automorphism group U (1)Y of the conformal N = 4 supersymmetry algebra

provides a very useful tool in the organization of the operators and correlation functions in
14 Here we have a tensor rather than a scalar exchange, but we expect a completely analogous result.
15 The constant C is a half of the coefficient in (C.24) because there are 2 possible Wick contractions of the
bφ O
composite [O bφ ](x3 ) generated in the OPE (4.4) with the two remaining operators Oφ (x2 ) and Oφ (x4 ).
supergravity. UY (1) transformations are not symmetries of the full IIB string theory, nor of
the N = 4 SYM theory. However this symmetry is recovered in the supergravity limit, and
hence the AdS/CFT duality predicts that the SYM theory in the limit N → ∞, λ → ∞ is
invariant under UY (1) [11,12].
From the Table 1, it is clear that the composite operator
1
OB ≡ √ {Oφ + iOC } (4.9)
2
has maximal hypercharge Y = 2 amongst the chiral operators. In fact, it is the only single-
trace operator with this maximal hypercharge value. By the same token, the operator
[OB OB ] is the only double trace operator with the maximal hypercharge of Y = 4. For
example, the stress tensor has Y = 0 and thus [T T ] has vanishing Y as well.
As supergravity has exact U (1)Y symmetry, this guarantees that the operator [OB OB ]
does not mix in the limit of large N and large λ with any other operator at all, and that
its anomalous dimension can be read off from only the correlators already computed
above. We must also expect that the real and imaginary components of [OB OB ], which
are [Oφ Oφ − OC OC ] and [Oφ OC ], have the same anomalous dimension. This is indeed
realized in the above correlation functions.
The anomalous dimension of [Oφ OC ] can be immediately read off from (4.1). To order
O(1/N 2 ) and for λ → ∞
d
y [Oφ OC ] = γ[Oφ OC ] [Oφ OC ],
dy
A 16
γ[Oφ OC ] = 2 = − 2 . (4.10)
N N
From (4.4)–(4.7) we deduce that the action of the dilation operator on the subspace of
operators spanned by [Oφ Oφ ] and [OC OC ] is

d [Oφ Oφ ] 1 C B [Oφ Oφ ]
y = 2 . (4.11)
∂y [OC OC ] N B C [OC OC ]
As already explained, the space of operators of approximate dimension 8 is much bigger
than this two-dimensional subspace, and we do not have at present enough information to
fill the entries of the full mixing matrix. However, the operator [Oφ Oφ − OC OC ], which
is the eigenvector of this 2 × 2 matrix of eigenvalue (−B + C)/N 2 , has maximal UY (1)
charge |Y | = 4 and we can isolate its anomalous dimension
−B + C 16
γ[Oφ Oφ −OC OC ] = =− 2, (4.12)
N2 N
as expected. The fact that A = −B + C as required by UY (1) symmetry is a nice check on
our calculation.
All other double-trace operators of approximate dimension 8 will have |Y | < 4. In
the dilaton/axion sector of the theory, we only encounter operators with Y = 0, such as
[OB∗ OB ] and [Tµν T µν ]. One particular linear combination of all these Y = 0 operators is
known to be protected. This is the descendant of the protected double trace operators

Q4 QS4 tr X2 tr X2 , (4.13)
105
Table 1
56
Super-Yang–Mills operators, supergravity fields and SO(4, 2) × U (1)Y × SU(4) quantum numbers. The range of k is k > 0, unless otherwise specified
SYM operator Desc SUGRA Dim Spin Y SU(4) Lowest reps
Ok ∼ tr Xk , k > 2 – hαα aαβγ δ k (0, 0) 0 (0, k, 0) 200 , 50, 105

20, 60, 1400
(1)
Ok ∼ tr λXk , k > 1 Q ψ(α) k + 32 ( 12 , 0) 1
2 (1, k, 0)
(2)
Ok ∼ tr λλXk Q2 Aαβ k+3 (0, 0) 1 (2, k, 0) 10c , 45c , 126c
E. D’Hoker et al. / Nuclear Physics B 589 (2000) 38–74

(3)
Ok ∼ tr λλ̄Xk S
QQ hµα aµαβγ k+3 ( 12 , 12 ) 0 (1, k, 1) 15, 64, 175
(4)
Ok ∼ tr F+ Xk , k > 1 Q2 Aµν k+2 (1, 0) 1 (0, k, 0) 6c , 20c , 50c
S 4∗ , 20∗ , 60∗
(5)
Ok ∼ tr F+ λ̄Xk Q2 Q ψµ k + 72 (1, 12 ) 1
2 (0, k, 1)
(6)
Ok ∼ tr F+ λXk Q3 “λ” k + 72 ( 12 , 0) 3
2 (1, k, 0) 4, 20, 60
(7)
Ok ∼ tr λλλ̄Xk S
Q2 Q ψ(α) k + 92 (0, 12 ) 1 (2, k, 1) 36, 140, 360
2
1c , 6c , 20c0
(8)
Ok ∼ tr F+2 Xk Q4 B k+4 (0, 0) 2 (0, k, 0)
S2 h0µν 1, 6, 200
(9)
Ok ∼ tr F+ F− Xk Q2 Q k+4 (1, 1) 0 (0, k, 0)
(10)
Ok ∼ tr F+ λλ̄Xk S
Q3 Q Aµα k+5 ( 12 , 12 ) 1 (1, k, 1) 15, 64, 175
(11)
Ok ∼ tr F+ λ̄λ̄Xk 2 S
Q Q2 aµναβ k+5 (1, 0) 0 (0, k, 2) 10c , 45c , 126c
(12)
Ok ∼ tr λλλ̄λ̄Xk S2
Q2 Q h(αβ) k+6 (0, 0) 0 (2, k, 2) 84, 300, 2187
(13)
Ok ∼ tr F+ 2 λ̄X k S
Q4 Q “λ” k + 11 (0, 12 ) 3 (0, k, 1) 4∗ , 20∗ , 60∗
2 2
(14)
Ok ∼ tr F+ λλ̄λ̄Xk 3
Q QS2 ψ(α) k + 13 ( 12 , 0) 1 (1, k, 2) 36∗ , 140∗ , 360∗
2 2
(15)
Ok ∼ tr F+ F− λXk S2
Q3 Q ψµ k + 11 ( 12 , 1) 1 (1, k, 0) 4, 20, 60
2 2
S2 1c , 6c , 20c0
(16)
Ok ∼ tr F+ F− 2 Xk Q4 Q Aµν k+6 (1, 0) 1 (0, k, 0)
(17)
Ok ∼ tr F+ F− λλ̄Xk S3
Q3 Q hµα aµαβγ k+7 ( 12 , 12 ) 0 (1, k, 1) 15, 64, 175
(18)
Ok ∼ tr F+ 2 λ̄λ̄X k 4
Q QS2 Aαβ k+7 (0, 0) 1 (0, k, 2) 10c , 45c , 126c
(19)
Ok ∼ tr F+ 2 F λ̄X k
− S3
Q4 Q ψ(α) k + 15 (0, 12 ) 1 (0, k, 1) 4∗ , 20∗ , 60∗
2 2
S4 1, 6, 200
(20) 2 Xk
2
Ok ∼ tr F+ F− Q4 Q hαα aαβγ δ k+8 (0, 0) 0 (0, k, 0)
where in the tensor product of the two 20’ of tr X2 , only the representation of dimension
105 is retained.
4.2. Comparison with large N SYM calculations
The prediction for the anomalous dimension of the operator [OB OB ] to order 1/N 2
obtained from sugra holds for infinitely large value of the ’t Hooft coupling λ = gYM 2 N
on the N = 4 SYM side. As the only window to date into large ’t Hooft coupling is via
the Maldacena conjecture, we do not have any direct checks available for the values of the
anomalous dimensions or for the space–time dependence of the correlation functions from
SYM. However, it is very illuminating to reproduce the 1/N 2 dependence of the anomalous
dimension from the standard large N counting rules of field theory, and to investigate any
other consequences this may produce. We proceed by expanding N = 4 SYM in 1/N ,
while keeping the ’t Hooft coupling λ fixed and perturbatively small. The strategy will be
to isolate the general structure of the expansion, then to seek the limit where λ → ∞ and
compare with supergravity predictions.
By way of example, we concentrate on the hOφ OC Oφ OC i correlator, but the results
apply generally. First, we normalize the individual operators via their 2-point functions,
as in (C.5). To leading order in 1/N , this requires (up to numerical coefficients we do not
keep track of)
Obφ = 1 tr F F + · · · , (4.14)
N
ObC = 1 tr F Fe+ ···. (4.15)
N
To Born approximation, this normalization is easily obtained by inspection of Fig. 5: each
operator has a 1/N normalization factor, and there are two color loops, producing each
a factor of N . According to general non-renormalization results, the 2-point function is
actually independent of gYM and thus independent of the ’t Hooft coupling.
With this normalization, the disconnected graph in Fig. 6(a) contributes precisely to
order N 0 . The simplest connected Born graph (b) of Fig. 6 has a factor of 1/N 4 from
the normalizations of the 4 operators, and two color loops, so its net contribution is of
order 1/N 2 , as expected. Finally, in graph (c) of Fig. 6, we illustrate the higher order
6 . With a factor 1/N 4 from operator
perturbative contributions with a graph of order gYM
6 N = λ3 /N 2 . Thus, for fixed
normalization, and 5 color loops, its total dependence is gYM
’t Hooft coupling, the N -dependence of graphs (b) and (c) are the same, as expected. In
bφ O
Fig. 5. Born diagram for the 2-point function hO bφ i in the double-line representation.
Fig. 6. Some Feynman diagrams in the double-line representation contributing to the 4-point function
bφ O
hO bC Obφ O
bC i.
fact, all planar graphs have this same N -dependence to leading order in N for fixed ’t Hooft
coupling, and we thus expect the connected part of the correlator to behave as

bC O
O bφ O
bC Ocφ ∼ 1 f (λ), (4.16)
N2
for some function f of the ’t Hooft coupling and position variables.
The above result was established perturbatively in the ’t Hooft coupling. To compare
with supergravity results, f ought to have a finite limit as λ → ∞. The Maldacena
conjecture predicts that it does, and gives a specific value for the limit. It would be
interesting to explicitly compute the anomalous dimension of [Oφ OC ] in perturbation
theory. From the previous discussion,

1 1
γ[Oφ OC ] = 2 γ1 λ + γ2 λ2 + · · · + O . (4.17)
N N4
If the interpolation between small and large λ is a smooth cross-over, it is natural to expect
the coefficient γ1 to be negative, as supergravity predicts a negative asymptotic value for
λ → ∞.
While the N -dependence of the anomalous dimension of [OB OB ] follows simply from
large N counting rules in SYM theory, the space-time dependence of the correlators cannot
be simply inferred from SYM. Supergravity results demonstrate that to order 1/N 2 , the
4-point correlator has analytic behavior in position variables, except for a single power of
a logarithm. On the SYM side however, our perturbative treatment of the ’t Hooft coupling
prevents us from making any sensible predictions on the space-time dependence of the
correlators. While graph (b) of Fig. 6 is only power behaved, one expects graphs like
(c) to contain higher and higher powers of logarithms as larger and larger numbers of
virtual particles are being exchanged. The Maldacena conjecture predicts that somehow, as
λ → ∞, all these powers and logarithms rearrange themselves and combine into a single
logarithm.
5. Conclusions
We have shown that the supergravity results for 4-point functions of chiral SYM
operators can be successfully interpreted in terms of a 4d Operator Product Expansion.
There is a generic relation between the power-singular terms that arise in the direct channel
limit of an exchange diagram and the contributions to 4-point function of the corresponding
‘conformal partial wave’. Logarithmic singularities can be naturally understood in terms
of O(1/N 2 ) renormalization and mixing of the double-trace composites that arise in the
OPE of two single-trace chiral operators.
It should be emphasized that the very possibility of a 4d OPE interpretation is quite non-
trivial. For example, generic exchange integrals in AdS5 contain (log)2 singularities, 16
which would be impossible to interpret as O(1/N 2 ) renormalization effects. It is then
crucial that the couplings of IIB supergravity on AdS5 × S 5 do not allow this type of
processes.
We did not attempt to interpret the series expansions of the supergravity 4-point
correlators beyond the leading logarithmic term. It would be of great interest to extend
the analysis of this paper to the higher order terms. The expansion of the supergravity
amplitudes as the operator insertions become close (see, e.g., (A.17)) is given by series
with a finite radius of convergence. This could be regarded as an indication that the the 4d
SYM theory admits a convergent OPE.
A comment is in order about the issue of decoupling of operators dual to string states.
We were able to match the power-singular terms and the leading log with the contributions
to the OPE of chiral primaries and their double-trace products, so our results support the
idea that at strong ’t Hooft coupling it is consistent to ignore the string states. One would
ultimately like to show that for each possible limit as the boundary insertion points become
pairwise close, the expansion of the 4-point supergravity amplitude can be interpreted to all
orders as a convergent double OPE in terms of the subset of operators given by the chiral
operators and their multi-trace products. This would prove that as λ → ∞ this subset forms
a closed algebra.
16 This happens when the exchange integral is equal to an infinite sum of quartic graphs. It appears that all
exchanges that arise in IIB supergravity on AdS5 × S 5 can be reduced to a finite sum of quartics graphs [27].
To tackle these issues it might be more convenient to consider a 4-point function of

N = 4 chiral primaries, the simplest example being the correlator of four 200 operators
trX(i Xj ) . The field theory analysis is somewhat cleaner than in the dilaton/axion
sector since there are fewer double-trace operators of approximate dimension 4 than of
approximate dimension 8. On the supergravity side the computation involves exchanges of
the 200 , of the massless vector and of the massless graviton, as well as one quartic graph.
All the exchange integrals are available and the only missing piece of information is the
numerical value of the quartic interaction vertex of four 200 . As noticed in [54], it should
be possible to obtain this vertex directly from the 5d gauged supergravity Lagrangian.
The OPE of two 200 contains double-trace operators in the 105, 84, 200 and singlet
representations. The 105 operator is known to be protected, and it has been recently
argued [55] that the 84 should also be protected due to another shortening condition of
the N = 4 superconformal algebra. The 200 double-trace operator is not protected but
the superconformal algebra constrains its anomalous dimension to be positive. Finally the
singlet is the ‘parent’ chiral-primary operator from which the [OB OB ] descendent operator
considered in this paper is obtained by applying 8 Q’s: hence its anomalous dimension
must be the same as γ[OB OB ] . It would be nice to check these facts through an explicit
supergravity calculation.
Finally, we would like to remark that although we have confined our investigation to
the 4d N = 4 SYM theory, our methods apply to other AdS/CFT dualities in various
dimensions. In particular, the results of Section 2 imply that for any CFT that has an AdS
dual, in the limit in which the gravity approximation is valid all the singular terms in the
OPE of two protected operators are given by other protected operators and their normal-
ordered products.
Acknowledgements
It is a pleasure to thank Dan Freedman for very enjoyable collaboration in previous

projects that have led to this investigation, and for his advice in the present project
as well. We are grateful to Edward Witten for important discussions. E.D’H. gratefully
acknowledges the warm hospitality offered by the Laboratoires de Physique Théorique at
Ecole Polytechnique and at Ecole Normale Supérieure, as well as the support provided by
the Centre National de Recherche Scientifique (CNRS).
The research of E.D’H. is supported in part by NSF Grant PHY-98-19686, A.M. by NSF
grant PHY-9870115, and L.R. by D.O.E. cooperative agreement DE-FC02-94ER40818
and by INFN ‘Bruno Rossi’ Fellowship.
Appendix A. Summary of supergravity 4-point functions
Exchange amplitudes for massive scalars, massive vectors and massless gravitons with
external scalar operators were evaluated in [20,22,26,27] in general AdSd+1 space. We
summarize here the results for massive scalar amplitudes with non-derivative couplings,
and for massless graviton exchange, defined by
Z Z
(t ) √ √
Iscal = dz g dw gK∆1 (z, x1)K∆3 (z, x3 ) G∆ (z, w)
× K∆2 (w, x2 )K∆4 (w, x4 ), (A.1)
Z Z
1 √ √ µν 0
µν 0
(t )
Igrav = dz g dw g T13 (z) Gµνµ0 ν 0 (z, w) T24 (w). (A.2)
4
Here the coupling of the graviton propagator is to to a conserved stress tensor, given by
µν
T13 (z, x1 , x3 ) = Dµ K∆1 (z, x1 )Dν K∆1 (z, x3 )
1
− g µν Dρ K∆1 (z, x1 )Dρ K∆1 (z, x3 )
2

+ m21 K∆1 (z, x1 )K∆1 (z, x3 ) , (A.3)
µ0 ν 0
with an analogous expression for T24 . The superscript ‘t’ in (A.1), (A.2) indicates that
these integrals define what we call ‘t-channel’ exchanges. In our terminology, the graphs
in the s- and u-channels are obtained from those in the t-channel by letting respectively,
(x1 , x2 , x3 , x4 ) → (x1 , x3 , x2 , x4 ) and (x1 , x2 , x3 , x4 ) → (x1 , x2 , x4 , x3 ).
Adopting the methods of [27], the evaluation of these integrals does not require the
explicit form of the bulk-to-bulk propagators, but only their equations of motion:

− + m2 G∆ (u) = δ(z, w),
−Dσ Dσ Gµνµ0 ν 0 − Dµ Dν Gσ σ µ0 ν 0 + Dµ Dσ Gσ νµ0 ν 0

+Dν Dσ Gµσ µ0 ν 0 − 2 Gµνµ0 ν 0 − gµν Gσ σ µ0 ν 0

2
= gµµ0 gνν 0 + gµν 0 gνµ0 − gµν gµ0 ν 0 δ(z, w)
d −1
+Dµ0 Λµνν 0 + Dν 0 Λµνµ0 . (A.4)
Remarkably, for all cases that arise in the IIB compactification on S5,
the amplitudes can
be expressed as simple linear combinations of a finite number of 4-point quartic graphs.
It is convenient to define the integrals associated with these general quartic graphs as
follows:
D∆1 ∆3 ∆2 ∆4 (x1 , x3 , x2 , x4 )
Z d+1
d ze e∆3 (z, x3 )K
e∆2 (z, x2 )K
e∆4 (z, x4 ).
= K∆1 (z, x1 )K (A.5)
z0d+1
Here, the bulk-to-boundary propagators are given by [5,6],
∆
e z0
K∆ (z, x) = C∆ K∆ (z, x) = C∆ 2 , (A.6)
z0 + (Ez − xE)2
with the normalization
0(∆)
C∆ = d . (A.7)
π 2 0 ∆ − d2
We introduce a special short-hand notation for some special quartic graphs which will
appear frequently 17
0
W∆∆ (xi ) = D∆∆0 ∆∆0 (x1 , x3 , x2 , x4 ), (A.8)
S∆∆0 (xi ) = D∆+2∆0 ∆∆0 (x1 , x3 , x2 , x4 ).
W (A.9)
The expression for the exchange graphs are then as follows. The massive scalar amplitude
is
(t )
X
pmax
Np
Iscal = C∆1 C∆2 C∆3 C∆4 D∆13 +pp∆2 ∆4 (x1 , x3 , x2 , x4 ), (A.10)
p=pmin (x13 )∆3 −p
2

1 0 12 (∆1 + ∆3 − ∆) 0 12 (∆1 + ∆3 + ∆ − d) 0(p)0(p + ∆13 )
Np = .
4 0 p + 1 − 12 ∆ + 12 ∆13 0 p + 1 − d2 + 12 ∆ + 12 ∆13 0(∆1 )0(∆3 )
(A.11)
Here, ∆13 ≡ ∆1 − ∆3 , pmin = 12 (∆ − ∆13 ), and pmax = ∆3 − 1. The massless graviton
amplitude for d = 4 is given by
"
9 0 0
Igrav = ∆C∆ C∆0
(t ) 2 2
∆ − 1 − ∆ W∆∆
8
#
X
∆−1
0 2∆(∆ 0 )2 0 ∆(∆ 0 )2 0
+ Mk Wk∆ − S ∆ +1 − s
W W ∆ +1 , (A.12)
∆−1 k ∆−1 k
k=1
where the constants Mk are defined by
∆(3∆ − 8) 0 2 0
∆∆0 (∆0 − 2)
Mk = −(∆ ) + 4∆ + 3 + . (A.13)
8(∆ − 1)2 ∆−1
For d = 4, ∆ = ∆0 = 4 we have:
" #
25 · 33 X
3

Igrav = −
(t )
15W4 +
4 4 Sp + 32sWp .
−17Wp + 64W 5 5
(A.14)
π8
p=1
A.1. Explicit form of quartic graphs and series expansion
All 4-point functions depend non-trivially on two conformal invariant cross ratios of the
points xi . We find it convenient to choose the combinations s and t defined in (2.1) of
the text. For Euclidean positions xi , the ranges for these combinations are 0 6 s 6 1 and
−1 6 t 6 1.
q
It was shown in [26] that the 4-point quartic functions Wp may be expressed as follows
d
q (−)p+q π 2 0(p + q − d2 ) q ∂ q−1 p−1 ∂ p−1
Wp (xi ) = 2 )p (x 2 )q
s s I (s, t) , (A.15)
0(p)2 0(q)2 (x13 24
∂s ∂s
17 Up to a number of external x -dependent factors, these quantities equal the function W ∆0 (0, 0) and W ∆0 (1, 0)
i ∆ ∆
introduced in [26].
where the universal function I (s, t) is given by

Z∞ Z1
1 1
I (s, t) = dµ dλ . (A.16)
µ + s(1 − λ2 ) 1 + µ + λt
0 −1
The integral I (s, t) is perfectly convergent and produces an analytic function in s and t,
with logarithmic singularities in s and t.
(a) Direct channel series expansion

The direct channel limit is given by |x13| |x12 | or/and |x24 | |x12|, so that s, t → 0.
We find
q
Wp (xi ) ≡ Dppqq (xi )
∞
(−)p+q π 2 0(p + q − d2 ) X
d
0(k + 1)2 s k+1
= 2 )p (x 2 )q
0(p)2 0(q)2 (x13 24
0(k − p + 2)0(k − q + 2)
k=0
n o
× bk (t) − ak (t) ln s + 2ψ(k + 1) − ψ(k − q + 2) − ψ(k − p + 2)
(A.17)
and
d
X∞
(−)p+q π 2 0(p + q − 2 + 1)
d
0(k + 1)0(k + 3)s k+1
Spq+1 (xi ) =
W
40(p)0(p + 2)0(q + 1)2 (x13
2 )p (x 2 )q
24
0(k − p + 2)0(k − q + 2)
k=0
n
× b̂k (t) − âk (t) ln s + ψ(k + 1) + ψ(k + 3)
o
− ψ(k − q + 2) − ψ(k − p + 2) . (A.18)
Here the coefficient functions are given by
Z1 Z1
(1 − λ2 )k (1 − λ2 )k 1 + λt
ak (t) = dλ , bk (t) = dλ ln , (A.19)
(1 + λt)k+1 (1 + λt)k+1 1 − λ2
−1 −1
Z1 Z1
(1 − λ2 )k+1 (1 − λ2 )k+1 1 + λt
âk (t)= dλ , b̂k (t) = dλ ln . (A.20)
(1 + λt)k+1 (1 + λt)k+1 1 − λ2
−1 −1
The coefficient functions admit Taylor series expansions in powers of t with radius of
convergence 1. In particular the functions ak (t) have the following representation in terms
of hypergeometric series:

√ 0(k + 1) k+1 k 3 2
ak (t) = π F , + 1; k + ; t . (A.21)
0 k + 32 2 2 2
We have the following relations between these functions

(k + 2)âk (t) = (k + 1) 2ak (t) − ak+1 (t) , (A.22)

(k + 2)2 b̂k (t) = (k + 1)(k + 2) 2bk (t) − bk+1 (t) − 2ak (t) + ak+1 (t). (A.23)
The presentation of these series expansions is slightly formal in the sense that for k 6 q −2,
the 0(k − q + 2) function in the denominator produces a zero, while the ψ(k − q + 2)
term produces a pole, which together yield a finite result, which amounts to a pole term
in s. Its coefficient can be obtained from the formula limx→0 ψ(x − q)/ 0(x − q) =
(−)q+1 0(q + 1) for any non-negative integer q.
(b) Crossed channel series expansion

The crossed channel asymptotics is given by s → 1/2 and t → ±1. The ‘s-channel’ limit
|x12| |x13| or/and |x34| |x13 | corresponds to s → 1/2, t → −1, while the ‘u-channel’
limit |x23 | |x34| and/or |x14 | |x34 | corresponds to s → 1/2, t → +1.
Both limits were derived in [26] by first obtaining a suitable expansion for the universal
function I (s, t) and then using (A.15). This expansion may be used to evaluate the
q
logarithmic part of Wp and we obtain
d
2p−2 π 2 0(p + q − d2 )
Wp (xi )log = − ln 1 − t 2
q
0(p)0(q)(x132 )p (x 2 )q
24
XX
q−1 ∞
(−2)−` 0(k + 1)s p+q−`−1 (1 − 2s)k−p+`−q+2
× αk (t).
0(q − `)0(p − `)`!0(k + ` − p − q + 3)
`=0 k=0
(A.24)
Notice that in the crossed channel, no power singularities arise. The coefficient functions
αk (t) are given by
∞
X 0(` + 12 ) (1 − t 2 )`
αk (t) = . (A.25)
`=0
0( 12 )`! 2` + k + 1
Appendix B. Matching supergravity exchanges and partial waves
B.1. Conformal partial amplitude
Here we present the partial amplitude hO1 (x1 )O2 (x2 )O3 (x3 )O4 (x4 )i0 , which corre-
sponds to the contribution to the 4-point function of a scalar operator O∆ from the OPE
∆1 +∆2 −∆
of O1 (x1 )O3 (x3 ) ∼ O∆ (x1 )/x13 and all its conformal descendants. In a CFT, an
expression for the partial amplitude can be written as

1
O1 (x1 )O2 (x2 )O3 (x3 )O4 (x4 ) 0 = Σ −∆ ∆ +∆ ∆ −∆ Σ −∆
f0 (ρ, η),
x1313 13 x1213 24 x1413 24 x2424 13
(B.1)
where Σij ≡ ∆i + ∆j , ∆ij ≡ ∆i − ∆j and

2 x2
x12 2 x2
x14
η= 34
2 x2
, ρ= 23
2 x2
(B.2)
x13 24 x13 24
are the two cross-ratios. The function f0 (ρ, η) of the cross-ratios was obtained 18 in [51]
(Eq. (3.2)):
0(∆)
f0 (ρ, η) = η 2 (∆13 +∆24 )
1
1
2 (∆ + ∆13 ) 0 2 (∆ − ∆13 )
1
0
Z1 − 1 (∆24 +∆)
1 ρ η 2
× dσ σ 2 (∆13 −∆24 )−1 − 12 (∆13 +∆24 )−1
(1 − σ ) +
σ (1 − σ )
0
−1 !
1 1 d ρ η
×F (∆ − ∆24 ), (∆ + ∆24 ); ∆ + 1 − ; + .
2 2 2 σ (1 − σ )
(B.3)
Our goal now is to rewrite this partial amplitude in the form of an expansion in conformally
invariant variables s = 1/(2(η + ρ)) and t = (η − ρ)/(η + ρ) (see (2.1)), which will
allow a more direct comparison with the corresponding supergravity exchange diagram.
To do this, following the steps of Appendix A of [51] ((A.1) to (A.3)), we first expand the
hypergeometric function in a power series
−1 !
1 1 d ρ η
F (∆ − ∆24 ), (∆ + ∆24 ); ∆ + 1 − ; +
2 2 2 σ (1 − σ )
∞
X 1 2 (∆ − ∆24 ) n 2 (∆ + ∆24 ) n ρ(1 − σ z) −n
1 1
= , (B.4)
n! ∆ + 1 − d2 σ (1 − σ )
n=0 n
where z = 1 − η/ρ. We then perform the integral in σ, which gives
Z1
dσ σ 2 (∆13 +∆)+n−1 (1 − σ )− 2 (∆13 +∆)+n−1 (1 − σ z)− 2 (∆24 +∆)−n
1 1 1
0

0 1
+ ∆) + n 0 12 (−∆13 + ∆) + n
2 (∆13
=
0(∆ + 2n)

×F 12 (∆ + ∆13 ) + n, 12 (∆ + ∆24 ) + n; ∆ + 2n; z . (B.5)
Putting everything together, we get the expression for f0 (ρ, η):
∞ ∆+∆13 ∆−∆13 ∆+∆24 ∆−∆24
0(∆)η 2 (∆13 +∆24 ) X
1
2 2 2 2
f0 (ρ, η) = n n n n
+ + − d
1 (∆+∆ )
ρ 2 24
n=0
0(∆ 2n)(∆ 1 )
2 n n!
1
× n F 12 (∆ + ∆13 ) + n, 12 (∆ + ∆24 ) + n; ∆ + 2n; z . (B.6)
ρ
We now restrict ourselves to a case of pairwise equal dimensions, ∆13 = ∆24 = 0. We also
change to s, t variables:
18 With normalizations such that hO O i, hO O O i, hO O O i are all given by coefficient 1 times the
∆ ∆ 1 3 ∆ 1 3 ∆
appropriate conformally invariant function of coordinates. This implies that in the OPE of O1 O2 the coefficient
C13∆ of the operator O∆ is 1.
η 2t 1 4s
z=1− = , = . (B.7)
ρ t −1 ρ 1−t
Using quadratic transformation of hypergeometric function

F ∆2 + n, ∆2 + n; ∆ + 2n; t −1
2t
∆ n ∆ n 1 ∆ 1 2
= F + , + + ; + n + ; t (B.8)
(1 − t) 2 +n
∆
4 2 4 2 2 2 2
and the definition of ak (t) from (A.21) we can finally rewrite the partial amplitude as an
expansion in s and t variables as

1
O1 (x1 )O2 (x2 )O3 (x3 )O4 (x4 ) 0 = g (s, t),
2∆1 2∆2 0
(B.9)
x13 x24
∞ 2
0 (∆) 0 ∆ + 1 − d X 0 ∆
+n 1
g0 (s, t) = 2 4
2 2
s 2 ∆+n an+ ∆ −1 (t).
0 ∆2 n=0
0 ∆+1− d
2 + n n! 2
(B.10)
Note that singular terms in the limit |x13| |x12 | correspond to n 6 ∆1 − ∆/2 − 1.
B.2. Singular terms from Witten diagrams
In this subsection we will find the singular terms of a given AdS scalar exchange diagram
in the form of an expansion in s and t, and comparing with the singular terms of the
corresponding partial amplitude from (B.10) we will find an exact match. We recall the
result (A.10) that any exchange supergravity diagram reduces to a sum of quartic graphs.
For simplicity, we restrict ourselves to the case of pairwise equal dimensions, as in the
analysis of partial amplitude in the previous subsection, ∆13 = ∆24 = 0. We also recall the
expansion of a quartic graph (A.17). Let us assume that ∆1 6 ∆2 , which means that pmax 6
∆2 − 1. Upon insertion of the expansion (A.17) into (A.10) we notice that the only power-
singular terms O(1/x13 n ) in the limit x x are the ones for which k 6 ∆ − 2. Keeping
13 12 1
terms k 6 ∆2 − 2 would amount to consider also all the power singularities O(1/x24 n )
in the limit x24 x12 . However as will be apparent below only the terms that obey the
more stringent restriction k 6 ∆1 − 2 can be matched with the partial amplitude. Using the
relation
ψ(k − ∆2 + 2)
= (−1)k−∆2 +1 0 ∆2 − k − 1 , (B.11)
0 k − ∆2 + 2
n ) to each quartic
we can now write explicitly the contribution of the singular terms O(1/x13
graph (notice that p 6 ∆1 − 1):
d
(−)p π 2 0(p + ∆2 − d2 )
Dpp∆2 ∆2 sing = 2 2 2 p 2 ∆2
0 p 0 ∆2 x13 x24
∆X −2 2
1
0 k + 1 s k+1
× ak (t) (−)k+1 0 ∆2 − 1 − k . (B.12)
k=p−1
0 k−p+2
We are now in a position to extract power singularities from a scalar exchange (A.10).
Inserting (B.12) into (A.10), we get

O∆ O∆ O∆ O∆
1 2 3 4 sing
Y
4
π
d
2 1
= C(∆i ) 2 ∆2
0 ∆2 2 ∆1
x13 2
x24
i=1
∆X1 −2
2
× 0 k + 1 s k+1 (−)k+1 ak (t)0 ∆2 − 1 − k
k=0
∆X1 −1

(−)p 0 p + ∆2 − d2
× Np 2 . (B.13)
p= ∆
0 p 0 k−p+2
2
We will now evaluate the second summation in this formula. Due to the pole of the gamma
function in the denominator for k − p + 2 6 0, one can extend the p-summation to infinity.
Notice that this would not be possible if the k-summation extended up to ∆2 − 2 instead
of ∆1 − 2. This gives
∞
X (−)p 0 p + ∆2 − d2
Np 2
p= ∆
0 p 0 k−p+2
2
∆
k+1 0 2 + ∆1 − 2 0 2 + ∆2 − 2 0 ∆1 − 2 0 ∆2 − 2
∆ d d ∆ ∆
1
= (−) 2 . (B.14)
4 0 ∆1 0 k + ∆ − d + 2 0 k + 2 − ∆ 0 ∆2 − 1 − k
2 2 2
Putting this back into the expression for the scalar exchange (B.13) and shifting the
remaining k-summation to n = k − ∆/2 + 1 we finally get

O∆1 O∆2 O∆3 O∆4 sing

1 1 0 ∆2 + ∆1 − d2 0 ∆2 + ∆2 − d2 0 ∆1 − ∆2 0 ∆2 − ∆2
= 2 2
3d
4π 2 x13 2 ∆1 x 2 ∆2 0 ∆1 − d2 0 ∆2 − d2
24
∆1 − ∆
2 −1
2
X ∆ +n 0 ∆2 + n
× s 2 a ∆ +n−1 (t) . (B.15)
n=0
2 0 n + ∆ − d2 + 1 n!
Now, as in (B.10), we normalize hO∆ O∆ i, hO1 O3 O∆ i, hO2 O4 O∆ i to 1. This can be
Ok
achieved by noting that the OPE coefficient Oi Oj = Cij k ∆i +∆j −∆k
+ · · · is
xij
AhOi Oj Ok i
Cij k = , (B.16)
AhOk Ok i
where by A we denote the normalization of the corresponding correlator. To make
Cij ∆ = 1 in the double OPE of the 4-point function we multiply (B.15) by
AhO1 O3 O∆ i AhO2 O4 O∆ i
×
AhO∆ O∆ i AhO∆ O∆ i
and then we further divide by the normalization of hO∆ O∆ i to normalize this 2-point
function in the double OPE to 1. The appropriate coefficients,

AhOi Oj Ok i = −0 12 (∆i + ∆j − ∆k ) 0 12 (∆j + ∆k − ∆i ) 0 12 (∆k + ∆i − ∆j )
1 −1
×0 (∆i + ∆j + ∆k − d) 2π d 0 ∆i − d2 0 ∆j − d2 0 ∆k − d2 ,
2

0(∆) ∆ − d2
AhOi Oi i = 2 d (B.17)
π 2 0 ∆ − d2
can be found in [6]. We now arrive to the expression

2 0(∆)0 ∆ + 1 − d2
O∆1 O∆2 O∆3 O∆4 sing = 4
2 ∆1 x 2 ∆2
x13 24 0 ∆2
∆1 − ∆
2 −1
2
X 0 ∆2 + n 1
× s 2 ∆+n an+ ∆ −1 (t),
n=0
0 ∆ + 1 − 2 + n n!
d 2
(B.18)
n
which directly matches the singular terms O(1/x13 ) in (B.10).
Appendix C. Dilaton–axion correlators
Let Oφ and OC be the operators that couple to the dilaton and axion supergravity fields
with unit strength:
Z
Sint = d 4 x φ(x)Oφ (x) + C(x)OC (x). (C.1)
The kinetic terms in the 5-dimensional supergravity action are normalized as

Z
1 1 1
2
(∂φ)2 + (∂C)2 , (C.2)
2κ5 2 2
AdS5
where the gravitational coupling is given in terms of SYM parameters (setting the AdS
radius R ≡ 1) by
8π 2
2κ52 = . (C.3)
N2
The 2-point functions are then (Eq. (A.13) of [6])

4 · C4 1 3N 2 1
Oφ (x1 )Oφ (x2 ) = OC (x1 )OC (x2 ) = 2 8
= 4 8 . (C.4)
2κ5 x12 π x12
bφ = ζ Oφ , O
It is convenient to define normalized operators O bC = ζ OC , such that

bφ (x1 )O
O bφ (x2 ) = O bC (x2 ) = 1 .
bC (x1 )O (C.5)
8
x12
The normalization constant ζ is then

πκ5 π2
ζ= √ =√ . (C.6)
2 3 3N
The complete 4-point functions in the dilaton–axion sector were assembled in [26]. Since
the supergravity action is even under sign reversal of the axion field, 3 different amplitudes
enter: hObφ O
bφ O
bφ Obφ i, hObφ O
bC O
bφ O
bC i and hO bC O bC ObC O
bC i. Only graviton exchanges
contribute to the 4-dilaton amplitude. The 4-axion amplitude is given by graviton and
dilaton exchanges, but it was checked in [19] that the dilaton contributions precisely cancel
when the s, t and u channels are added together, so that

(s)
bφ O
O bφ O
bφ O
bφ = O bC O
bC O bC = I0 + 2κ 2 −1 ζ 4 Igrav
bC O + Igrav
(t )
+ Igrav
(u)
. (C.7)
5
The equality of the 4-dilaton and 4-axion amplitudes actually follows directly from the
U (1)Y symmetry of supergravity. Here, I0 is the contribution from disconnected graphs
(i)
and Igrav is the graviton exchange integral in the channel i. The t-channel graviton
exchange has been given in terms of quartic graphs in (A.14), the other channels are
readily obtained by permuting the coordinates xi . The disconnected contributions are easily
evaluated recalling the normalization (C.5):
1 1 1
I0 = 8 8
+ 8 8
+ 8 8
. (C.8)
x12 x34 x13 x24 x14 x23
The constant (2κ52 )−1 ζ 4 = π 6 /(72N 2 ) in front of the exchange graphs in (C.7) arises from
a factor 2κ52 for each bulk-to-bulk propagator, a factor (2κ52)−1 for each vertex and from
the normalization ζ of the dilaton and axion operators. As expected, the connected graphs
are O(1/N 2 ) relative to the disconnected part.
The amplitude hO bφ (x1 )O
bC (x2 )O
bφ (x3 )ÔC (x4 )i is given by the t-channel graviton
exchange, as well as axion exchanges and a quartic interaction. It was shown in [19] that
the sum of all the graphs except the graviton exchange can be conveniently expressed as
an ‘effective’ quartic graph
Z 5
d z
Iqeff = −2 K4 (z, x1 )Dµ K4 (z, x2 )K4 (z, x3 )Dµ K4 (z, x4 ) (C.9)
z05
29 · 34 4
=− 8
W4 (xi ) − 4sW45 (xi ) . (C.10)
π
We then have

(t )
bφ O
O bC O bC = I 0 + 2κ 2 −1 ζ 4 Igrav
bφ O + Iqeff . (C.11)
0 5
Here the disconnected contribution I00 is simply

1
I00 = 8 8
. (C.12)
x13 x24
Notice that the 4-point functions of axions and dilatons do not receive contributions
from exchanges of the ‘fixed scalar’ of m2 = 32 that corresponds [3] to the dilation
mode of the S 5 . The subtleties associated with the computation of 3-point functions of
two dilatons (or axions) and a fixed scalar were recently discussed in [44,56]. Although
naively the cubic coupling of 2 dilatons and a fixed scalar is absent in the 5-dimensional
supergravity Lagrangian, the corresponding 3-point function can be shown to be non-
vanishing, either through a procedure of analytic continuation in the KK level of the
dilatons [56], or by a careful dimensional reduction that takes into account the constraints
of the supergravity fields and leads to non-vanishing boundary interactions [44]. The two
procedures give identical results [44]. In the analytic continuation method, one takes the
conformal dimension of the dilatons to be 4 + . The cubic coupling dilaton–dilaton-fixed
scalar vanishes as O() in the limit → 0, but the 3-point integral diverges as O(1/), so
that the product gives a finite contribution. The procedure of analytic continuation can be
immediately used to prove that fixed scalar exchanges give no contribution to the 4-point
functions of dilatons and axions. In fact, contrary to the 3-point function case, the exchange
integral with 4 external dimensions 4 + and bulk dimension 8 is perfectly convergent
in the limit → 0 (see, e.g., [22]), whereas the cubic couplings vanish, yielding zero
contribution.
C.1. Leading logarithm asymptotics
We now determine the leading logarithmic terms in the dilaton–axion 4-point functions
in the limits as the points become pairwise close. We adopt the terminology:
(a) t-channel limit: |x13 | |x12|, |x24 | |x12|, which corresponds to s, t → 0;

(b) s-channel limit: |x12| |x13 |, |x34| |x13 |, or s → 1/2, t → −1;
(c) u-channel limit: |x23| |x34 |, |x14 | |x34 |, or s → 1/2, t → 1.
(t )
The limit of Igrav for s, t → 0 (t-channel) was obtained in [26]; its logarithmic parts are
given by
∞

(t ) 3 · 23 ln s X 4+k 0(k + 4)
Igrav = s
log π 6 x13
8 x8
24 k=0
0(k + 1)
n
× − 2(5k 2 + 20k + 16)(3k 2 + 15k + 22)ak+3(t)
o
+ (k + 4)2 (15k 2 + 55k 2 + 42)ak+4(t) . (C.13)
The analogous result for Iqeff is

∞
26 · 3 · 5 ln s X k+4
Iqeff log = 6 8 8
s (k + 1)2 (k + 2)2 (k + 3)2 (3k + 4) ak+3 (t) .
π x13 x24
k=0
(C.14)
For our purposes, we shall need only the leading logarithmic contributions of the
amplitudes in the various channels. As we shall have to use permutations on the points
xi to find the exchange amplitudes in all channels, it will be most convenient to re-express
the s- and t-dependence in terms of the xi variables. Using the numerical values a3 (0) =
32/35 and a4 (0) = 256/315, we find
27 · 3 2 x2
x13
(t )
Igrav ∼− 16
ln 4
24
t-channel, (C.15)
6
7π x12 x 12
27 · 33 2 x2
x13
Iqeff ∼ + 16
ln 4
24
t-channel. (C.16)
7π 6 x12 x12
The limits in the crossed channels, where s → 1/2 and t 2 → 1, were not obtained
explicitly in [26]. We shall now derive them here, by deriving the crossed channel limits
for the W -functions, starting from (A.24). First, we notice from the definition of αk (t) that
αk (±1) = 1/(k + 1). Next, it is clear from (A.24) that a non-trivial leading logarithmic
contribution will arise only if the orders of summation satisfy k = p + q − 2 − ` > 0.
Since in our expressions p, q > 1 in all cases, and 0 6 ` 6 q − 1, the inequality above can
always be realized and there is always a k > 0 satisfying k = p + q − 2 − `. The remaining
summation over ` is then
q π 2 0(p + q − 2)
Wp (xi ) ∼ − ln(1 − t 2 ) 2 )4 (x 2 )4
20(p)0(q)(x13 24
X
q−1
0(p + q − 1 − `)0(p + q − 1 − `)
× (−)` . (C.17)
0(q − `)0(p − `)0(p + q − `)`!
`=0
But this `-sum is precisely proportional to a hypergeometric function 3 F2 (p, p, 1 − q;

p − q + 1, p + 1; 1) evaluated at unit argument. As a result, we have
2 x 2 /x 4 )
q π2 1 ln(x12
Wp (xi ) ∼ − 34 13
s-channel, (C.18)
2 (p + q − 1)(p + q − 2) 2 )8
(x13
2 x 2 /x 4 )
q π2 1 ln(x14
Wp (xi ) ∼ − 23 12
u-channel. (C.19)
2 (p + q − 1)(p + q − 2) 2 )8
(x12
Finally, we need the asymptotic behavior of the function W S in this channel with s → 1/2
and t → 1. From (A.17), it is clear that the limit s → 1/2 is regular for any fixed t. Also,
2
the limit t 2 → 1 is regular, since both coefficients âk (t) and b̂k (t) have smooth limits.
Thus, in both the crossed s- and u-channels, the function W̄ is smooth and produces no
logarithmic contributions.
We are now ready to sum all contributions in the crossed channels. For graviton
exchange, we have
28 · 32 2 x2
x12
(t )
Igrav ∼− 2 )8
ln 4
34
s-channel, (C.20)
7π 6 (x13 x13
28 · 32 2 x2
x14
(t )
Igrav ∼− 2 )8
ln 4
23
u-channel. (C.21)
7π 6 (x12 x12
For Iqeff , we have
26 · 33 x2 x2
Iqeff ∼ − 2
ln 124 34 s-channel, (C.22)
7π 6 (x13 )8 x13
26 · 33 2 x2
x14
Iqeff ∼ − 2 )8
ln 4
23
u-channel. (C.23)
7π 6 (x12 x12
We can finally assemble the leading logarithmic asymptotics for the full amplitudes of
normalized operators. For the 4-dilaton and 4-axion amplitudes the limits in the various
channel are equivalent, so we need only quote

bC ∼ − 1 2 · 13 1 ln x13 x24
5
bφ O
O bφ O
bφ O
bφ = ObC O
bC O
bC O t-channel. (C.24)
N 3 · 7 x12
2 16 2
x12
bφ O
The amplitude hO bC O
bφ O
bC i admits the two different limits (the s- and u-channels are
equivalent):

8
bφ O
O bC O bC ∼ 1 2 1 ln x13 x24
bφ O t-channel, (C.25)
N 2 3 · 7 x12
16 2
x12

4
O bC O
bφ O bC ∼ − 2 1 ln x12x34
bφ O s-channel. (C.26)
N 2 x13
16 2
x13
References
[1] J. Maldacena, The large N limit of superconformal theories and supergravity, Adv. Theor. Math.
Phys. 2 (1998) 231, hep-th/9711200.
[2] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory
and gravity, hep-th/9905111.
[3] H.J. Kim, L.J. Romans, P. van Nieuwenhuizen, The mass spectrum of chiral N = 2 D = 10
supergravity on S 5 , Phys. Rev. D 32 (1985) 389.
[4] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string
theory, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[5] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hep-
th/9802150.
[6] D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Correlations functions in the AdS/CFT
correspondence, hep-th/980458.
[7] W. Muck, K.S. Viswanathan, Conformal field theory correlators from classical scalar field
theory on AdSd+1 , Phys. Rev. D 58 (1998) 041901, hep-th/9804035.
[8] S. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Three point functions of chiral operators in
D = 4, N = 4 SYM at large N, Adv. Theor. Math. Phys. 2 (1998) 697, hep-th/9806074.
[9] G. Chalmers, H. Nastase, K. Schalm, R. Siebelink, R-current correlators in N = 4 super-Yang–
Mills theory from anti-de Sitter supergravity, Nucl. Phys. B 549 (1999) 247, hep-th/9805105.
[10] E. D’Hoker, D.Z. Freedman, W. Skiba, Field theory tests for correlators in the AdS/CFT
correspondence, Phys. Rev. D 59 (1999) 045008, hep-th/9807098.
[11] K. Intriligator, Bonus symmetries of N = 4 super-Yang–Mills correlation functions via AdS
duality, hep-th/9811047.
[12] K. Intriligator, W. Skiba, Bonus symmetry and the operator product expansion of N = 4 super-
Yang–Mills, hep-th/9905020.
[13] P.S. Howe, E. Sokatchev, P.C. West, 3-point functions in N = 4 Yang–Mills, Phys. Lett. B 444
(1998) 341, hep-th/9808162.
[14] F. Gonzalez-Rey, B. Kulik, I.Y. Park, Non-renormalization of two point and three point
correlators of N = 4 SYM in N = 1 superspace, Phys. Lett. B 455 (1999) 164, hep-th/9903094.
[15] A. Petkou, K. Skenderis, A non-renormalization theorem for conformal anomalies, hep-

th/9906030.
[16] S. Penati, A. Santambrogio, D. Zanon, Two-point functions of chiral operators in N = 4 SYM
at order g 4 , hep-th/9910197.
[17] D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, in: Freedman’s conference lecture at
Strings ’98, Available at http://www.itp.ucsb.edu/online/strings98/.
[18] H. Liu, A.A. Tseytlin, On four-point functions in the CFT/AdS correspondence, hep-
th/9807097.
[19] D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Comments on 4-point functions in the
CFT/AdS correspondence, hep-th/9808006.
[20] E. D’Hoker, D.Z. Freedman, Gauge boson exchange in AdSd+1 , Nucl. Phys. B 544 (1999) 612,
hep-th/9809179.
[21] H. Liu, Scattering in anti-de Sitter space and operator product expansion, hep-th/9811152.
[22] E. D’Hoker, D.Z. Freedman, General scalar exchange in AdSd+1 , Nucl. Phys. B 550 (1999)
261, hep-th/9811257.
[23] G. Chalmers, K. Schalm, The large Nc limit of four point functions in N = 4 super-Yang–Mills
theory from anti-de Sitter supergravity.
[24] J.H. Brodie, M. Gutperle, String corrections to four point functions in the AdS/CFT
correspondence, Phys. Lett. B 455 (1998) 296, hep-th/9809067.
[25] E. D’Hoker, D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Graviton and gauge boson
propagators in AdSd+1 , Nucl. Phys. B 562 (1999) 330, hep-th/9902042.
[26] E. D’Hoker, D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Graviton exchange and
complete 4-point functions in the AdS/CFT correspondence, Nucl. Phys. B 562 (1999) 353,
hep-th/9903196.
[27] E. D’Hoker, D.Z. Freedman, L. Rastelli, AdS/CFT 4-point functions: how to succeed at z-
integrals without really trying, Nucl. Phys. B 562 (1999) 395, hep-th/9905049.
[28] S. Sanjay, On direct and crossed channel asymptotics of four-point functions in AdS/CFT
correspondence, Mod. Phys. Lett. A 14 (1999) 413, hep-th/9906099.
[29] T. Banks, M.B. Green, Nonperturbative effects in AdS in five-dimensions × S(5), string theory
and d = 4 SUSY Yang–Mills, JHEP 05 (1998) 002, hep-th/9804170.
[30] M. Bianchi, M.B. Green, S. Kovacs, G. Rossi, Instantons in supersymmetric Yang–Mills and D
instantons in IIB superstring theory, JHEP 08 (1998) 013, hep-th/9807033.
[31] N. Dorey et al., Multi-instanton calculus and the AdS/CFT correspondence in N = 4
superconformal field theory, hep-th/9901128.
[32] M.B. Green, Interconnections between type II superstrings, M theory and N = 4 supersymmet-
ric Yang–Mills, hep-th/9903124.
[33] V. Balasubramanian, S.B. Giddings, A. Lawrence, What do CFTs tell us about anti-de Sitter
space-times?, hep-th/9902052.
[34] S. Kovacs, A perturbative re-analysis of N = 4 supersymmetric Yang–Mills theory, hep-
th/9902047.
[35] S. Kovacs, N = 4 supersymmetric Yang–Mills theory and the AdS/SCFT correspondence, hep-
th/9908171.
[36] M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, On the logarithmic behavior in N = 4 SYM
theory, JHEP 08 (1999) 020, hep-th/9906188.
[37] M. Bianchi, S. Kovacs, Non-renormalization of extremal correlators in N = 4 SYM theory,
hep-th/9910016.
[38] W. Skiba, Correlators of short multi-trace operators in N = 4 supersymmetric Yang–Mills,
Phys. Rev. D 60 (1999) 105038, hep-th/9907088.
[39] R. Gopakumar, M.B. Green, Instantons and non-renormalisation in AdS/CFT, hep-th/9908020.
[40] B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Four-point functions in N = 4
supersymmetric Yang–Mills theory at two loops, hep-th/9811172.
[41] B. Eden, P.S. Howe, P.C. West, Nilpotent invariants in N = 4 SYM.

[42] P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Explicit construction of nilpotent covariants
in N = 4 SYM, hep-th/9910011.
[43] B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Extremal correlators in four-
dimensional SCFT, hep-th/9910150.
[44] E. D’Hoker, D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Extremal correlators in the
AdS/CFT correspondence, in: M. Shifman (Ed.), Yuri Golfand Memorial Volume: Many Faces
of the Superworld, World Scientific, 2000, hep-th/9908160.
[45] L. Andrianopoli, S. Ferrara, ‘Nonchiral’ primary superfields in the AdSd+1 /CFT d correspon-
dence, Lett. Math. Phys. 46 (1998) 265, hep-th/9807150.
[46] L. Andrianopoli, S. Ferrara, On short and long SU(2, 2/4) multiplets in the AdS/CFT
[47] M. Gunaydin, D. Minic, M. Zagermann, Novel supermultiplets of SU(2, 2/4) and the
AdS5 /CFT 4 duality, hep-th/9810226.
[48] A.C. Petkou, Conserved currents, consistency relations and operator product expansions in the
conformally invariant O(N) vector model for 2 < d < 4, Ann. Phys. 249 (1996) 180, hep-
th/9410093.
[49] A.C. Petkou, Operator product expansions and consistency relations in a O(N) invariant
fermionic CFT for 2 < d < 4, Phys. Lett. B 389 (1996) 18, hep-th/9602054.
[50] S. Ferrara, A.F. Grillo, R. Gatto, Lett. Nuovo Cimento 2 (1071) 1363.
[51] S. Ferrara, A.F. Grillo, G. Parisi, Nuovo Cimento 19 (1974) 667.
[52] S. Ferrara, A.F. Grillo, G. Parisi, Nucl. Phys. B 49 (1972) 77.
[53] V. Gurarie, Logarithmic operators in conformal field theory, Nucl. Phys. B 410 (1993) 535,
hep-th/9303160.
[54] S. Lee, AdS(5)/CFT(4) four-point functions of chiral primary operators: cubic vertices, hep-
th/9907108.
[55] S. Ferrara, A. Zaffaroni, Superconformal field theories, multiplet shortening, and the
AdS(5)/SCFT(4) correspondence, hep-th/9908163.
[56] H. Liu, A.A. Tseytlin, Dilaton-fixed scalar correlators and AdS5 × S 5 – SYM correspondence,
JHEP 10 (1999) 003, hep-th/9906151.
Localized (super)gravity and cosmological constant

Zurab Kakushadze
C.N. Yang Institute for Theoretical Physics, State University of New York, Stony Brook, NY 11794, USA
Received 6 June 2000; accepted 9 August 2000
Abstract
We consider localization of gravity in domain wall solutions of Einstein’s gravity coupled to a
scalar field with a generic potential. We discuss conditions on the scalar potential such that domain
wall solutions are non-singular. Such solutions even exist for appropriate potentials which have no
minima at all and are unbounded below. Domain walls of this type have infinite tension, while usual
kink type of solutions interpolating between two AdS minima have finite tension. In the latter case
the cosmological constant on the domain wall is necessarily vanishing, while in the former case it
can be zero or negative. Positive cosmological constant is allowed for singular domain walls. We
discuss non-trivial conditions for physically allowed singularities arising from the requirement that
truncating the space at the singularities be consistent. Non-singular domain walls with infinite tension
might a priori avoid recent “no-go” theorems indicating impossibility of supersymmetric embedding
of kink type of domain walls in gauged supergravity. We argue that (non-singular) domain walls
are stable even if they have infinite tension. This is essentially due to the fact that localization of
gravity in smooth domain walls is a Higgs mechanism corresponding to a spontaneous breakdown
of translational invariance. As to discontinuous domain walls arising in the presence of δ-function
“brane” sources, they explicitly break translational invariance. Such solutions cannot therefore be
thought of as limits of smooth domain walls. We point out that if the scalar potential has no minima
and approaches finite negative values at infinity, then higher derivative terms are under control, and
do not affect the cosmological constant which is vanishing for such backgrounds. Nonetheless, we
also point out that higher curvature terms generically delocalize gravity, so that the desired lower-
dimensional Newton’s law is no longer reproduced.  2000 Elsevier Science B.V. All rights reserved.
1. Introduction
In the brane world scenario the standard model gauge and matter fields are assumed
to be localized on p 6 9 spatial dimensional branes (or an intersection thereof), while
gravity lives in a larger (10 or 11) dimensional bulk of space–time [1–21]. The volume
of dimensions transverse to the branes is then assumed to be finite. This is automatically
achieved if the transverse dimensions are compact. However, as was pointed out in [24,
E-mail address: zurab@insti.physics.sunysb.edu (Z. Kakushadze).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 1 4 - 9
76 Z. Kakushadze / Nuclear Physics B 589 (2000) 75–118
25], if one considers certain warped [22,23] compactifications, then the volume of the
transverse dimensions can be finite even if the latter are non-compact. Another way of
phrasing this result is that gravity is localized on a lower-dimensional subspace (that is, the
brane). 1
One motivation for considering such unconventional compactifications is the moduli
problem. In particular, the extra dimensions in such scenarios are non-compact while their
volume is finite and fixed in terms of other parameters in the theory such as those in
the scalar potential. That is, the scalars descending from the components of the higher
dimensional metric corresponding to the extra dimensions are actually massive, and their
expectation values are fixed.
Also, scenarios with localized gravity could possibly have implications for the
cosmological constant problem. In particular, in this context together with the localized
graviton zero mode one has a continuum of massive bulk graviton modes. One can
then ask whether this can somehow help avoid the usual four-dimensional field theory
arguments [30] why having vanishing (or small) cosmological constant in the absence of
supersymmetry generically seems to require gross fine-tuning. Thus, on one hand, a priori it
is not completely obvious whether the four-dimensional effective field theory arguments do
always apply in this context. On the other hand, the fact that the volume of extra dimensions
is finite seems to suggest that they should, at least at low enough energies. One of the
purposes of this paper is to get additional insights into these issues.
In this paper we consider localization of gravity arising in (D − 1)-dimensional domain
wall solutions in the system of D-dimensional Einstein gravity coupled to a single real
scalar field with a generic scalar potential. In particular, we discuss conditions on the
scalar potential such that the corresponding domain wall solutions are non-singular (in
the sense that singularities do not arise at finite values of the coordinate transverse to
the domain wall). The usual kink type of solutions are non-singular as they interpolate
between two adjacent local AdS minima of the scalar potential. Such solutions always
have vanishing (D − 1)-dimensional cosmological constant. On the other hand, we point
out that there exist other non-singular solutions (subject to the aforementioned non-
singularity conditions on the scalar potential) which do not interpolate between AdS
minima. In fact, such solutions exist even for potentials which have no minima at
all and are unbounded below. Domain walls of this type have infinite tension. The
(D − 1)-dimensional cosmological constant in such solutions can be vanishing or negative.
However, positive cosmological constant is allowed for singular solutions. In fact, we
discuss non-trivial conditions for physically allowed singularities, which arise from the
requirement that truncating the space at the singularities be consistent. In particular, as
we point out in the following, these conditions are not satisfied in some of the recently
discussed “self-tuning” solutions.
1 The idea to use warped compactifications to achieve localization of gravity was originally presented in [26].
In [24,25] a concrete realization of this idea was given in the context of one extra dimension. In [27] this was
generalized to cases with more extra dimensions. In [28] possible string theory embeddings of such a brane world
scenario were discussed. For a general discussion of localization of various fields in such backgrounds, see [29].
Z. Kakushadze / Nuclear Physics B 589 (2000) 75–118 77
One possible implication of the existence of non-singular solutions with infinite

tension is that if the corresponding potentials can be obtained from, say, D = 5 N = 2
gauged supergravity, then one would obtain a supersymmetric domain wall with localized
supergravity. In particular, a priori it is unclear whether such potentials avoid recent “no-
go” theorems, which seem to imply that potentials with more then one AdS minima cannot
be obtained in this context thus ruling out supersymmetric kink type of solutions. If not,
then it would be important to show that the aforementioned “no-go” theorems also extend
to infinite tension domain walls.
In this paper we also study the issue of stability of domain walls which localize gravity.
In particular, a priori it is not obvious why, say, infinite tension domain walls, arising in
theories with potentials which have no minima at all and are unbounded below, should be
stable. Indeed, the D-dimensional theory in such cases appears to be sick, and could even
have tachyons. Nonetheless, we argue that (non-singular) (D − 1)-dimensional solutions
are (classically) stable even for non-vanishing (D − 1)-dimensional cosmological constant
(in particular, in the latter case the corresponding domain walls are not BPS saturated). The
basic reason for stability is that in the case of smooth domain walls (that is, those without
any ad hoc δ-function “brane” sources which break D-dimensional diffeomorphism
invariance explicitly), localization of gravity is a Higgs mechanism for the graviton field in
the process of which the scalar field is eaten (or, more precisely, its corresponding modes
are) by the would-be massless (D − 1)-dimensional graviphoton which acquires non-zero
mass. In particular, we point out the importance of the full D-dimensional diffeomorphism
invariance for the self-consistency of this Higgs mechanism.
In the light of the above discussion it is appropriate to mention that discontinuous
domain walls arising in the presence of δ-function “brane” sources cannot be though of
as limits of smooth solutions. One way to understand this is to note that such “brane”
sources break translational invariance explicitly, while in the case of smooth domain walls
this breakdown is spontaneous. In particular, there is an explicit discontinuity between
these two different setups unless the (D − 1)-dimensional Planck scale is vanishing.
At the end of the paper we discuss some aspects of the cosmological constant problem
in the context of a brane world realized via a domain wall which localizes gravity. In
particular, we point out that higher derivative such as higher curvature terms do not seem
to be under control unless we consider the following type of domain walls. Consider a
potential with no local minima such that it approaches finite negative values for infinite
values of the scalar field. In this case we argue that the higher derivative terms are under
control (as long as the AdS curvature at infinity is small enough compared with the
cut-off scale for the higher derivative terms). Then provided that quantum corrections
do not modify the aforementioned behavior of the scalar potential (that is, if the scalar
potential still approaches constant negative values at infinity), one might hope to solve the
cosmological constant problem in this context as the (D − 1)-dimensional cosmological
constant in such solutions is ultimately zero (albeit it is not completely clear how to control
the local corrections which could modify the behavior of the scalar potential at infinity).
Even though the above setup might appear promising in the context of the cosmological
constant problem, we point out certain difficulties with such a scenario, which, in fact,
seem to be generic to theories with finite volume non-compact extra dimensions. In

particular, even though higher curvature terms appear to be under control as far as
preserving the desirable properties of the corresponding warped background is concerned,
they generically delocalize gravity. That is, once we include such terms, generically there is
no longer a normalizable graviton zero mode, and the usual (D − 1)-dimensional Newton’s
law is no longer reproduced. We also point out some possibilities for avoiding these
problems, which will be discussed in more detail elsewhere.
2. Setup
In this section we discuss the setup within which we will discuss solutions with localized
gravity. Thus, consider a single real scalar field φ coupled to gravity with the following
action: 2
Z
√ 4
S = MP D−2
d x −G R −
D
(∇φ) − V (φ) ,
2
(1)
D−2
where MP is the D-dimensional (reduced) Planck mass. The equations of motion read:
8
∇ 2 φ − Vφ = 0, (2)
D−2
1 4 1
RMN − GMN R = ∇M φ∇N φ − 12 GMN (∇φ)2 − GMN V . (3)
2 D−2 2
The subscript φ in Vφ denotes derivative w.r.t. φ.
In the following we will be interested in solutions to the above equations of motion with
the warped [22,23] metric of the following form:
2
dsD = exp(2A)dsD−1
2
+ dy 2 , (4)
where y ≡ x D , the warp factor A, which is a function of y, is independent of the
coordinates x µ , µ = 1, . . . , D − 1, and the (D − 1)-dimensional interval is given by
2
dsD−1 =e
gµν dx µ dx ν , (5)
with the (D − 1)-dimensional metric e gµν independent of y.
With the above ansätz, we have:

Rµν = Reµν − exp(2A) A00 + (D − 1)(A0 )2 e gµν , (6)
00 0 2

RDD = −(D − 1) A + (A ) , (7)
RµD = 0, (8)
eµν is the (D − 1)-dimensional Ricci tensor
where prime denotes derivative w.r.t. y. Also, R
corresponding to the metric e
gµν .
2 Here we focus on the case with one scalar field for the sake of simplicity. In particular, in this case we can
absorb a (non-singular) metric Z(φ) in the (∇φ)2 term by a non-linear field redefinition. This cannot generically
be done in the case of multiple scalar fields φ i , where one must therefore also consider the metric Zij (φ).
In the following we will be interested in solutions where φ depends non-trivially on y.

From the above equations it then follows that φ is independent of x µ . The equations of
motion for φ and A then become:
8 00
φ + (D − 1)A0 φ 0 − Vφ = 0, (9)
D−2
4 D−1
(D − 1)(D − 2)(A0 )2 − (φ 0 )2 + V − eexp(−2A) = 0,
Λ (10)
D−2 D−3
4 1
(D − 2)A00 + (φ 0 )2 + eexp(−2A) = 0.
Λ (11)
D−2 D−3
The first equation is the dilaton equation of motion, the second equation is that for RDD ,
and the third equation is a linear combination of the latter and the equation for Rµν . In
fact, the equation for Rµν implies that Λ e is independent of x µ and y. In fact, it is nothing
but the cosmological constant of the (D − 1)-dimensional manifold, which is therefore an
Einstein manifold, described by the metric e e is such that the
gµν . Our normalization of Λ
(D − 1)-dimensional metric e gµν satisfies Einstein’s equations
eµν − 1 e
R e = − 1e
gµν R e
gµν Λ, (12)
2 2
so that the (D − 1)-dimensional Ricci scalar is given by
e = D − 1 Λ.
R e (13)
D−3
Note that we have only two fields φ and A, yet we have three equations. However, only
two of these equations are independent. This can be seen as follows. Using the second
equation one can express φ 0 (A0 ) via A0 (φ 0 ) and V . One can then compute φ 00 (A00 ) and
plug it in the first (third) equation. This equation can then be seen to be automatically
satisfied as long as the third (first) equation is satisfied. As usual, this is a consequence of
Bianchi identities.
3. Solutions with (D − 1)-dimensional Poincaré invariance
e = 0. In
In this section we will discuss solutions of the aforementioned equations with Λ
this case the equations of motion read:
4
(D − 1)(D − 2)(A0 )2 − (φ 0 )2 + V = 0, (14)
D−2
4
(D − 2)A00 + (φ 0 )2 = 0. (15)
D−2
To study possible solutions to these equations, it is convenient to rewrite the scalar potential
V as follows:
V = Wφ2 − γ 2 W 2 , (16)
where
√
2 D−1
γ= , (17)
D−2
and W is some function of φ. Note that this is always possible (at least for a large class of
scalar potentials) if we view (16) as a differential equation for W . Also note that even if V
is a simple function of φ, generically W is a complicated function, which is due to the fact
that the differential equation (16) is highly non-linear.
The advantage of using W instead of V is that the equations of motion for φ and A can
now be written as the first order differential equations. Thus, the above two equations are
satisfied if φ and A satisfy
φ 0 = αWφ , (18)
0
A = βW, (19)
where
√
D−2 2
α≡ , β ≡ − . (20)
2 (D − 2)3/2
Here = ±1.
Here one can ask whether the equations of motion imply that φ and A must satisfy the
first order differential equations for some W . The answer to this question is positive. To
see this, it is convenient to first rewrite the equations of motion by treating A as a function
of φ (instead of y), while still treating φ as a function of y. Thus, we have
4
(φ 0 )2 1 − 14 (D − 1)(D − 2)2 (Aφ )2 = V , (21)
D−2
D−2 1
Aφφ (φ 0 )2 = − A φ Vφ − V. (22)
8 D−2
Next, let
−1
V 1 − 14 (D − 1)(D − 2)2 (Aφ )2 ≡ h2 , (23)
where h is a function of φ. Then φ satisfies the first order differential equation
φ 0 = αh. (24)
Now define W via
V = h2 − γ 2 W 2 . (25)
It then follows that A satisfies the first order differential equation
A0 = η|β|W, (26)
where η = ±1. Requiring that the equation for A00 (or, equivalently, that for Aφφ ) is
satisfied then implies that
h = −ηWφ , (27)
which is what we wished to show. Note that without loss of generality above we have
chosen η = −.
3.1. Non-singularity conditions
In this subsection we would like to discuss the conditions on W such that the
corresponding solutions do not blow up at finite values of y. More precisely, in this section
we will focus on solutions such that φ is non-singular 3 at finite y. (We will discuss singular
solutions in Section 5.) To begin with note that if V is non-singular, which we will assume
in the following, then W and Wφ should (generically) be non-singular as well. This then
guarantees that solutions are continuous for finite values of φ. However, a priori it is still
possible that φ blows up at finite values of y.
Note that the question of non-singularity of solutions can be studied without any
reference to the coupling to gravity. Indeed, the equation we would like to study is
φ 0 = αWφ . (28)
This equation arises in a non-gravitational theory described by the action
Z
4
S= d x − D
(∂φ) − V(φ) .
2
(29)
D−2
We can rewrite the potential V as
V = Wφ2 (30)
if we view it as a differential equation for W . Note that this is possible as long as V is non-
negative. Now let us study solutions which depend on y only. Then the equation of motion
for φ is given by (28), which is nothing but the BPS equation. As should be clear from our
discussion, one can write this equation for any non-negative potential V. In particular, the
theory need not be supersymmetric. However, a priori there is no guarantee that solutions
of this equation, that is, BPS solutions, satisfying certain physical requirements exist for a
given V. This is precisely the question we would like to study here.
It is convenient to divide the discussion of this question according to whether Wφ
vanishes for some value(s) of φ or its non-vanishing for all real φ. We will first discuss
the latter case.
Non-vanishing Wφ
Suppose Wφ is non-vanishing for all real φ. In a (globally) supersymmetric setup this
would be equivalent to the statement that the theory does not have a supersymmetric
vacuum. More generally, we have two possibilities. The potential V has extrema if and
only if Wφφ vanishes for some values of φ. Some of these extrema may correspond to
(local) minima (if Vφφ = 2Wφ Wφφφ > 0 at the corresponding values of φ). On the other
hand, there might be cases where the potential V has no extrema at all — this happens if not
only Wφ but also Wφφ is non-vanishing. In this case we have a runaway type of potential.
At any rate, for our discussion here it will have little relevance whether the potential has
3 We will refer to the corresponding domain walls as non-singular. However, as we will point out in the
following, some of such solutions are actually singular in the sense that the D-dimensional Ricci scalar R blows
up, but the singularities are located at y = ±∞.
local minima or not. As we will point out, non-singular BPS solutions exist in either case
provided that a certain simple criterion is satisfied.
It might be unfamiliar, and even sound surprising, that potentials with, say, runaway
behavior can have BPS solutions. An example of this type was originally discussed in [31].
Before we give our general discussion here, we would like to give an even simpler example.
Thus consider a theory with
W = ρφ, (31)
where ρ is a parameter. Note that this is a theory with a constant potential V = ρ 2 . In the
globally supersymmetric context supersymmetry is completely broken in D dimensions.
However, there are BPS solutions with only half of the supersymmetries broken. These are
given by
φ(y) = αρy. (32)
Thus, even though supersymmetry is broken if we preserve D-dimensional Lorentz
invariance, we can preserve half of the supersymmetries at the cost of breaking Lorentz
invariance, more concretely, to that of a (D − 1)-dimensional theory. If we view this
(delocalized) solution as a domain wall, then its tension is infinite. In fact, so is the
corresponding central charge in the superalgebra. Thus, for instance, in the context of
N = 1 supersymmetry in D = 4 (in which case we really have one complex scalar filed
in the corresponding chiral supermultiplet) there is a central extension of the N = 1
superalgebra with an infinite central charge [31]:
{Qα , Qβ } = Σαβ Z, (33)
where Σαβ is proportional to the area tensor in the plane perpendicular to the y direction,
and

Z = 2 W (y = L) − W (y = −L) = (const) × L → ∞. (34)
Thus, solutions of this type can be viewed as domain walls with infinite tension/central
charge. As we will see in the following, gravitational generalizations of such domain walls,
if they are non-singular, automatically localize gravity.
Next, let us discuss the general condition for such domain walls to be non-singular. That
is, we would like to find the condition under which φ does not blow up at finite values of y.
For this to be the case, it is necessary and sufficient that the function
Z
dφ
F (φ) ≡ (35)
Wφ
is unbounded at φ → ±∞. Indeed, let φ(y0 ) = φ0 be finite for some finite point y0 . Then
Z∞
dφ
= α(y∞ − y0 ), (36)
Wφ
φ0
where y∞ is the point where φ becomes ∞. If the above condition is satisfied, then the
above integral is divergent, and y∞ is either +∞ or −∞ (depending on the form of W
and sign of α). Similar considerations apply to y−∞ where φ becomes −∞. The above
condition can be reformulated directly in terms of W : W should not 4 grow faster than φ 2
for φ → ±∞ . It is clear that there exist infinite choices of W which satisfy this condition.
Let us discuss general properties of such domain walls assuming that the above non-
singularity condition is satisfied. Note that Wφ never changes sign as Wφ is non-vanishing.
This then implies that φ monotonically increases/decreases from −∞/ + ∞ to +∞/ − ∞.
That is, non-singular domain walls of this type have infinite tension (and, equivalently,
infinite central charge in the supersymmetric context).
Here the following remark is in order. Consider the cases where Wφ is non-vanishing
at finite φ but goes to zero for φ → +∞ and/or φ → −∞. If W blows up at large φ (but
more slowly than ∼ φ, so that Wφ vanishes there), then the tension of the domain wall
is infinite. However, if W approaches finite values at φ → ±∞ (which implies that Wφ
vanishes at φ → ±∞), then the tension of the domain wall is finite — it is proportional to
W (+∞) − W (−∞), which is finite in this case. The situation here is similar to that where
Wφ vanishes at finite values of φ.
Wφ vanishing at one point

Let us now discuss the case where Wφ vanishes for only one value of φ = φ0 . This
corresponds to a global minimum of the potential. (More precisely, there might also be a
runaway branch with vanishing V in the large |φ| limit.)
In such cases we can also have non-singular domain walls. First, let us study the y
dependence of φ near φ0 . We have
1
Wφ = Wφφ (φ0 )(φ − φ0 ) + Wφφφ (φ0 )(φ − φ0 )2 + · · · . (37)
2
Let us first assume that Wφφ (φ0 ) 6= 0. Then the leading y dependence of φ near φ0 is given
by
φ(y) − φ0 ∼ C exp(ay), (38)
where C is the integration constant, and a ≡ αWφφ (φ0 ). For definiteness let us assume
that a > 0. Then φ approaches φ0 at y → −∞. In the complete solution φ will then
monotonically increase/decrease to +∞/ − ∞ at y → +∞ provided that the solution
is non-singular. This is the case provided that W does not grow faster than φ 2 at φ →
+∞/ − ∞. If this condition is satisfied only for, say, φ → +∞, then we only have non-
singular solutions with φ → φ0 at y → ∓∞ and φ → +∞ at y → ±∞. If it is also
satisfied for φ → −∞, we then also have the corresponding solutions with φ → φ0 at
y → ∓∞ and φ → −∞ at y → ±∞.
Cases with Wφφ (φ0 ) = 0 can be discussed similarly. In fact, let us assume that the lowest
(non-trivial) derivative of W which is non-vanishing at φ0 is the kth derivative (k > 3).
Then the leading y dependence of φ near φ0 is given by
4 More precisely, this is correct up to usual “logarithmic” factors (that is, log(φ), log(log(φ)), etc., or, more
generally, the appropriate products thereof). Thus, for instance, the non-singularity condition on (35) is satisfied
for W = ξφ 2 log(φ).
a
(φ(y) − φ0 )k−2 ∼ , (39)
y
where a ≡ −(k − 1)!/(k − 2)αW (k) (φ0 ). Once again, we have domain walls interpolating
between φ = φ0 at y → ∓∞ and φ = ±∞ at y → ±∞ provided that the corresponding
non-singularity conditions are satisfied.
Let us consider a simple example of such domain walls. Let
1
W = ζ φ2. (40)
2
Then we have a single minimum at φ = 0. Note that the non-singularity conditions are
satisfied for this choice of W . The domain wall solutions are given by
φ(y) = C exp(αζy), (41)

where C is an arbitrary integration constant.
Wφ vanishing at multiple points

If Wφ vanishes at more then one points, call them φa , then we have familiar domain
walls (with finite tension) interpolating between adjacent vacua. However, we can also have
domain walls of the aforementioned type, which interpolate between φ = min(φa )/φ =
max(φa ) and φ = −∞/φ = +∞ provided that W does not grow faster then φ 2 at φ →
−∞/ + ∞.
Before we end this subsection, let us illustrate some of the points discussed above on the
example of the φ 4 theory. More concretely, consider the potential
V = V0 + m2 φ 2 + λφ 4 , (42)
where λ > 0. First consider the case m2 > 0. In this case we must assume that V0 > 0 so
that V is non-negative. Then the theory has one minimum, which corresponds to Wφ = 0
for V0 = 0, but for V0 > 0 Wφ is non-vanishing. In neither case, however, do we have
non-singular domain walls as the non-singularity conditions on Wφ at φ → ±∞ are not
satisfied.
Next, let us consider the case m2 < 0. Let
m2 ≡ −2λv2 . (43)
Then we can rewrite the potential as

2
V = V0 − λv4 + λ φ 2 − v2 , (44)
where we must assume that V0 > λv4 so that the potential is non-negative at the minima
φ = ±v. These correspond to Wφ = 0 for V0 = λv4 , but for V0 > λv4 Wφ is non-vanishing.
It then follows that for V0 > λv4 there are no non-singular domain walls. For V0 = λv4 ,
however, we have the well-known kink solutions interpolating between the two vacua φ =
±v. Note that in this case we do not have non-singular domain walls interpolating between
φ = ±v and φ → ±∞ (the signs are correlated) as the non-singularity conditions are not
satisfied.
Thus, for the φ 4 theory we have reproduced a known result that non-singular domain
wall solutions exist (for m2 < 0) only if we “fine-tune” V0 in terms of m2 and λ as follows:
m4
V0 = . (45)
4λ
In this particular case this is equivalent to the requirement that the (two degenerate) ground
state(s) of the theory have vanishing vacuum energy. This is because in this case non-
singularity conditions are not satisfied for non-singular infinite tension domain walls to
exist. However, as should be clear from our previous discussions, such domain walls can
exist in theories with ground state(s) with non-vanishing vacuum energy or even theories
with no ground state at all.
Normalizable modes
Before we end this subsection, we would like to discuss the condition for the existence of
normalizable modes localized on a domain wall (in the non-gravitational context discussed
in this subsection). Thus, consider small fluctuations ϕ(x µ , y) around the domain wall
background. The linearized action for ϕ reads (up to surface terms)
S[ϕ] ≡ S[φ + ϕ] − S[φ]
Z
1 2
= − 2 d D x ∂µ ϕ∂ µ ϕ + (ϕ 0 )2 + α2 Wφφ + Wφ Wφφφ ϕ 2 . (46)
α
Here φ satisfies (28). Let us define
ϕ ≡ φ 0 ω. (47)
The action for ω reads:
Z

S[ω] = − d D−1 x dy Wφ2 ∂µ ω∂ µ ω + (ω0 )2 . (48)
The zero mode is given by the configurations where ω is independent of y

ω = ω(x µ ). (49)
To have such a normalizable zero mode, the integral
Z+∞
2 y=+∞
T ≡2 dy Wφ = W
2
, (50)
α y=−∞
−∞
which is nothing but the domain wall tension, must be finite. Thus, domain walls with
infinite tension do not have normalizable zero modes in this context. On the other hand,
domain walls with finite tension do have normalizable zero modes. 5
Here the following remark is in order. The action (48) is actually exact, that is, the ω
zero mode is a free field. This is due to the fact that it describes the Goldstone mode
of the broken translational invariance in the y direction. Thus, let φ(y) be a solution
describing a non-singular domain wall. Consider the following fluctuations around this
5 As we will see in the following, once we consider the coupling to gravity, this conclusion is modified.
solution: φ(y + ω(x µ )). Then we have φ(y + ω(x µ )) = φ(y) + φ 0 (y)ω(x µ ) + O(ω2 ), so
that ω(x µ ) is precisely the linearized zero mode described above. However, if we plug
φ(y + ω(x µ )) into the action S[φ], after shifting the integration variable y (note that
integration over y is from −∞ to +∞ for non-singular domain walls), we obtain the
action S[ω] given in (48) for arbitrary ω(x µ ).
Also, the equation of motion for ω, which follows from the action (48), is given by
0
Wφ2 ∂µ ∂ µ ω + Wφ2 ω0 = 0. (51)
Let
ω(x µ , y) = e
ω(x µ )σ (y), (52)
where e
ω satisfies the (D − 1)-dimensional Klein–Gordon equation:
∂µ ∂ µ e
ω = m2 e
ω. (53)
Then we have the following equation for σ :
0
Wφ2 m2 σ + Wφ2 σ 0 = 0. (54)
In particular, a zero mode (m2 = 0) must satisfy

σ 0 = C Wφ2 , (55)
where C is a constant. Above we claimed that the zero mode corresponds to taking C = 0.
To see that C 6= 0 does not give rise to normalizable zero modes, note that in this case σ is
a monotonous function of y as can be seen from (55). Thus, it either goes to a constant at,
say, y → +∞, or is unbounded. In either case, if the domain wall tension is infinite, such
a solution is not normalizable. In fact, it is not difficult to show that such zero modes are
not normalizable even if the domain wall tension is finite.
Normalizability of other (that is, non-zero) modes can be discussed in a similar fashion.
For this it is sometimes convenient to treat σ as function of φ. We then have

σφ Wφ3 φ + Wφ m2 σ = 0, (56)
and
Z Z
1 2
||ϕ|| ≡ 2
dy ϕ = e
2
ω dφ Wφ σ 2 . (57)
α
Thus, for illustrative purposes let us consider an example where W = 12 ζ φ 2 . As we
discussed above, in this case the domain wall solution is given by
φ(y) = C exp(αζy), (58)
where C is an integration constant. The equation for σ now reads:
m2
φ 3 σφ φ
+ φσ = 0. (59)
α2 ζ 2
The general solution to this equation is given by:
σ = D+ φ −ρ+ + D− φ −ρ− , (60)
where
s
m2
ρ± = 1 ± 1− , (61)
α2 ζ 2
and D± are integration constants. Since the domain wall interpolates between φ = 0 and
φ = ±∞ (depending on the sign of C), it then follows that ||ϕ||2 is infinite for all allowed
m2 , so that there are no normalizable modes. Note that the tension of this domain wall is
infinite.
3.2. What about the warp factor?
In this subsection we would like to discuss what happens to the warp factor once
we consider the gravitational generalizations of the above domain walls. (Note that the
discussion of the behavior of φ is unchanged.) To answer this question, let us look at the
equation for A:
A0 = βW. (62)
To begin with let us discuss the behavior of A as a function of φ near the points where
Wφ vanishes. Let us refer to such a point as φ0 . First, let us focus on the case where
Wφφ (φ0 ) 6= 0. In this case we have

A0 = β W (φ0 ) + 12 Wφφ (φ(y) − φ0 )2 + · · · . (63)
Since in this case φ(y) − φ0 ∼ C exp(ay) vanishes exponentially fast for the corresponding
limit y → +∞ or y → −∞, it then follows that all the contributions to A except for that
corresponding to the W (φ0 ) term vanish exponentially as well. That is, A(y) goes to a
constant if W (φ0 ) vanishes, while in the case W (φ0 ) 6= 0 the leading behavior of A as a
function of y is given by:
A(y) ∼ βW (φ0 )y. (64)
This implies that for W (φ0 ) = 0 the factor exp(2A) in the metric goes to a constant in the
corresponding limit for y, so that the volume of the compactification in this case is actually
infinite. However, if W (φ0 ) is non-zero and has a correct sign, then exp(2A) vanishes
exponentially in the corresponding limit for y.
In fact, with appropriately chosen W such that the domain wall interpolates between two
adjacent vacua we can have the correct behavior at both y → −∞ and y → +∞ so that the
compactification volume is finite. More concretely, let Wφ vanish at two points, call them
φ1 and φ2 , and be non-vanishing for φ1 < φ < φ2 . Then if W (φ1 ) and W (φ2 ) are non-zero
with sign(W (φ1 )) = −sign(W (φ2 )) = sign(β) (that is, W must change sign between φ1
and φ2 ), A(y) goes to −∞ at both y → ±∞. Let us illustrate this with a simple example.
Consider the theory with

W = ξ φ − 13 φ 3 . (65)
The solution for φ is given by (y0 is the integration constant corresponding to the center of
the domain wall)

φ(y) = tanh αξ(y − y0 ) , (66)
while the solution for A reads (C is an integration constant)

2β 1 1
A(y) = ln cosh(αξ(y − y0 )) − + C. (67)
3α 4 cosh2 (αξ(y − y0 ))
We therefore have

4β/3α β 1
exp(2A) = exp(2C) cosh(αξ(y − y0 )) exp − . (68)
3α cosh2 (αξ(y − y0 ))
Since α and β have opposite signs, it then follows that exp(2A) falls off exponentially at
large y, and the compactification volume is indeed finite.
The cases where not only Wφ but also higher derivatives vanish at φ0 can be discussed
similarly. Suppose the lowest non-trivial derivative of W which is non-vanishing is the kth
derivative (k > 3). Then we have
1 (k)
W = W (φ0 ) + W (φ0 )(φ − φ0 )k + · · · . (69)
k!
Since in this case (φ(y) − φ0 )k−2 ∼ a/y, then it follows that in the corresponding limit for
y the contributions to A from all the terms except the W (φ0 ) term vanish. For instance, the
first subleading term proportional to W (k) (φ0 ) goes to zero as 1/y 2/(k−2). Thus, as in the
previous case, here we have a finite volume compactification only if W (φ0 ) 6= 0.
Let us now discuss the behavior of A in the cases where in the corresponding limit for y
φ goes to ±∞. (This is the case of domain walls with infinite tension.) Let us assume that
for large φ W has the following behavior:
W ∼ ξ φν , (70)
where ν 6 2 so that the non-singularity conditions are satisfied. Moreover, first let us
consider the case where ν > 0 as the analog of the ν < 0 case requires a separate treatment.
Let us discuss the ν = 2 and ν < 2 cases separately.
For ν = 2 we have Wφ ∼ 2ξ φ, and
φ(y) ∼ C exp(2αξy), (71)

where C is the integration constant. In this case we therefore have the following leading
behavior for A:
β 2 β 2
A(y) ∼ C exp(4αξy) ∼ φ , (72)
4α 4α
so that A(y) goes to −∞ exponentially fast (recall that α and β have opposite signs), and
exp(2A) goes to zero very fast, more concretely, as an exponential of an exponential.
Next, let us discuss 0 < ν < 2 cases. We then have Wφ ∼ νξ φ ν−1 , and
1 1
φ(y) ∼ [αν(2 − ν)ξ ] 2−ν y 2−ν . (73)
In this case we therefore have the following leading behavior for A:

β 2 2 β 2
A(y) ∼ [αν(2 − ν)ξ ] 2−ν y 2−ν ∼ φ , (74)
2να 2να
so for 0 < ν < 2 A(y) goes to −∞ as a power of y. Let

β 2
u≡ [αν(2 − ν)ξ ] 2−ν . (75)
να
Then we have
2
exp(2A) ∼ exp −u|y| 2−ν . (76)
Thus, exp(2A) rapidly decays to zero. In particular, for ν = 1 we have a Gaussian fall-off
for exp(2A), while φ is linear in y for large y. So, the compactification volume is finite as
long as 0 < ν 6 2.
Thus, as we see, such infinite tension domain walls localize gravity even more efficiently
than usual domain walls interpolating between adjacent minima. In the latter case the warp
factor exp(2A) decays as
exp(2A) ∼ exp(−e
u|y|), (77)
where eu ≡ |2βW (φ0 )|. On the other hand, for 0 < ν < 2 the exponent of |y| in (76) is larger
than 1, while for ν = 2 exp(2A) decays even more rapidly, namely, as an exponential of an
exponential.
Next, let us assume that for large φ W has the following behavior:
W ∼ ζ + ξ φ −σ . (78)
where σ > 0. Here we have written the leading constant term ζ which W approaches in
the infinite φ limit, and also the next-to-leading term which vanishes in this limit. We then
have Wφ ∼ −σ ξ φ −σ −1 , and
1 1
φ(y) ∼ [−ασ (2 + σ )ξ ] 2+σ y 2+σ . (79)
In this case we therefore have the following leading behavior for A:
β 2
A(y) ∼ βζy − [−ασ (2 + σ )ξy] 2+σ , (80)
2σ α
so that for ζ =
6 0 the leading behavior is
A(y) ∼ βζy, (81)
while for ζ = 0 we have
β 2
A(y) ∼ − φ . (82)
2σ α
Since the signs of α and β are opposite, in the ζ = 0 case A goes to +∞ (instead of −∞),
so that exp(2A) blows up, and the compactification volume is infinite. However, for non-
zero ζ with the appropriate sign A goes to −∞, so that exp(2A) decays exponentially, and
the compactification volume is finite (provided that exp(2A) goes to zero fast enough in
the opposite limit for y as well). Here the situation is analogous to what we have found
for the behavior of A near the points where Wφ vanishes. In particular, to have decaying
exp(2A) we need a constant term ζ in W in the corresponding limit.
Before we end this subsection, let us discuss the analog of the ν = 0 case above. Thus,
let us assume that for large φ W has the following behavior:
W ∼ ξ ln(φ). (83)
In this case we have Wφ ∼ ξ/φ, and

1 1
φ(y) ∼ (2αξ ) 2 y 2 . (84)
For definiteness let as assume that αξ > 0, so that we are dealing with the y → +∞ limit
here. The warp factor then has the following leading behavior:
1
A(y) ∼ βξy ln(y). (85)
2
Note that βξ < 0 (which follows from the fact that αξ > 0 and αβ < 0), which implies that
in the y → +∞ limit A(y) goes to −∞, and the compactification volume is finite (once
again, provided that exp(2A) goes to zero fast enough in the y → −∞ limit).
4. Killing spinors and supersymmetry
In this section we would like to discuss the aforementioned domain walls in the
supersymmetric context, where their BPS property is expected to persist to the quantum
level. In particular, such domain walls, as usual, are expected to be stable.
The domain walls we study in this paper preserve one half of the original supersymme-
tries in D dimensions. More precisely, the domain walls preserve one half of the super-
symmetries of a supersymmetric minimum of the scalar potential V if such minima are
present. However, one of the key points is that domain walls have the same number of su-
persymmetries (which act trivially on the domain wall) even if V has no supersymmetric
minima at all. In this sense such domain walls are similar to those in the previous case.
To see the aforementioned properties of domain walls explicitly, let us study the
corresponding Killing spinor equations, which (up to equivalent representations) read:
M
Γ ∂M φ − αWφ ε = 0, (86)

DM − 2 βW ΓM ε = 0.
1
(87)
Here ε is a Killing spinor, ΓM are D-dimensional Dirac gamma matrices satisfying
{ΓM , ΓN } = 2GMN , (88)
and DM is the covariant derivative

1
DM ≡ ∂M + ΓAB ωAB M . (89)
4
The spin connection ωAB M is defined via the vielbeins eA M in the usual way (here the
capital Latin indices A, B, . . . = 1, . . . , D are lowered and raised with the D-dimensional
Minkowski metric ηAB and its inverse, while the capital Latin indices M, N, . . . = 1, . . . , D
are lowered and raised with the D-dimensional metric GMN and its inverse). Furthermore,
1
ΓAB ≡ [ΓA , ΓB ] , (90)
2
where ΓA are the constant Dirac gamma matrices satisfying
{ΓA , ΓB } = 2ηAB . (91)
Finally, W , which determines the scalar potential via (16), is interpreted as the superpoten-
tial in this context.
Note that (86) comes from the requirement that the variation of the superpartner λ of the
scalar field φ vanish. On the other hand, (87) arises from the requirement that the gravitino
ψM have vanishing variations under the corresponding supersymmetry transformations.
Next, we would like to study the above Killing spinor equations in the warped
backgrounds of the form (4) with the flat (D − 1)-dimensional metric e gµν = ηµν (and
the scalar field depending only on y):

ΓD φ 0 − αWφ ε = 0, (92)
1
ε0 − βW ΓD ε = 0, (93)
2
1
∂µ ε + exp(A)Γeµ A0 ΓD − βW ε = 0. (94)
2
Here Γeµ are the constant (D − 1)-dimensional Dirac gamma matrices satisfying

Γeµ , Γeν = 2e
gµν = 2ηµν . (95)
Also, note that ΓD , which is the D-dimensional Dirac gamma matrix ΓM with M = D
(that is, the Dirac gamma matrix corresponding to the x D = y direction) is constant in this
background.
Note that (92), which is an algebraic equation for ε, comes from (86), while the two
differential equations (93) and (94) come from (87). These last two equations correspond
to the coupling to gravity, while (92) appears already in a non-gravitational system with
the scalar potential V = Wφ2 .
Since on the solution we have φ 0 = αWφ and A0 = βW , the Killing spinor satisfying
(92), (93) and (94) has positive helicity w.r.t. ΓD , and is given by
(0)
ε+ = exp(A/2)ε+ , (96)
(0)
where ε+ is a constant spinor of positive helicity.
Actually, the above conclusion that we only have Killing spinors of positive helicity
is only correct if W has non-trivial φ dependence. Thus, consider the case where W is
constant. It is clear that (92) is then satisfied for both positive as well as negative helicity
spinors (as long as φ is constant). As to (93) and (94), the general solution to this set of
equations for constant W is given by
(0)
ε = βW exp(A/2)x µ Γeµ + exp(−A/2) ε− + exp(A/2)ε+ (0)
, (97)
(0) (0)
where ε− and ε+ are constant negative respectively positive helicity spinors. Thus, for
constant W , that is, in the AdS space (for constant W we have constant V = −γ 2 W 2 < 0)
we have Killing spinors of both positive and negative helicity. However, if we consider a
domain wall for a non-constant superpotential W , then we have only the positive helicity
Killing spinor (96). Thus, such domain walls are BPS objects preserving one half of the
supersymmetries.
The BPS property of such a domain wall implies that the modes on the wall have boson-
fermion degeneracy corresponding to the unbroken supersymmetries. 6 Thus, for instance,
we expect to find a normalizable gravitino zero mode localized on the wall. In fact, this zero
mode is given by the variation of ψµ corresponding to the supersymmetry transformation
parameter for which the variations of λ and ψD vanish. Such a spinor is given by
ε = exp(A/2)χ+ , (98)
where χ+ = χ+ (x µ ) is an arbitrary spinor of positive helicity (w.r.t. ΓD ) which depends
on x µ but is independent of y. The corresponding gravitino variation is given by
δψµ = exp(A/2)∂µ χ+ , (99)
and it is not difficult to check that δψµ defined this way has precisely the correct
normalization (in terms of powers of the warp factor) for it to be the superpartner of the
four-dimensional graviton. In fact, both the Einstein and the Rarita–Schwinger terms in the
(D − 1)-dimensional action are proportional to
Z+∞
MPD−2 e D−3 ,
dy exp [(D − 3)A] ≡ M (100)
P
−∞
where MP is the D-dimensional Planck scale, while M eP is the (D − 1)-dimensional

Planck scale. Thus, supersymmetric BPS domain walls discussed in this paper localize
supergravity.
Before we end this section, we would like to make the following comment. As we
have already mentioned, in the supersymmetric context BPS domain walls are expected
to be stable. This is the case not only for the domain walls interpolating between two
supersymmetric AdS minima, but also for the infinite tension domain walls (or, more
precisely, their gravitational counterparts). That is, the D-dimensional scalar potential V
might have no minima at all, yet the corresponding BPS domain walls are supersymmetric
and stable. In particular, in this context there cannot be any tachyonic modes present, which
is guaranteed by the unbroken supersymmetries.
Finally, the following remark is in order. The above discussion applies only to the cases
where we can sypersymmetrize the system of the scalar field φ coupled to gravity. A priori
such a supersymmetrization might not exist for a given potential V (φ). In particular, it
might not be possible to couple the aforementioned bosonic system to fermions in a fashion
6 Actually, the state corresponding to the domain wall itself does not have a fermionic superpartner as the
unbroken supersymmetry acts trivially on it.
consistent with the corresponding superalgebra. However, the question of whether a given
potential can be obtained from a supergravity theory is outside of the scope of this paper.
e 6= 0
5. Solutions with Λ
In this section we will discuss solutions with non-zero (D − 1)-dimensional cosmolog-

ical constant Λ.e Thus, let us look at the equations for A and φ, namely, (10) and (11). In
particular, let us rewrite (11) as follows:
4 1 e
(φ 0 )2 exp(2A) + Λ = −(D − 2)A00 exp(2A). (101)
D−2 D−3
To begin with, let us discuss non-singular domain walls. To have a non-singular domain
eP , it is necessary that exp(A) asymptotically goes to zero. It then follows
wall with finite M
that A00 exp(2A) (as well as, say, (A0 )2 exp(2A)) asymptotically goes to zero as well. Then
the r.h.s. of (101) asymptotically goes to zero at y → ±∞, while the l.h.s. is strictly
positive definite for Λe > 0. This then implies that Λe cannot be positive for non-singular
domain walls that localize gravity.
e< 0
5.1. Non-singular domain walls with Λ
Here we can ask whether there exist non-singular domain walls with negative
cosmological constant. The answer to this question is affirmative. To construct such a
domain wall, pick A(y) such that M eP is finite. That is, at y → ±∞ exp[(D − 3)A] goes to
zero faster than 1/y. Moreover, let A(y) be such that A00 is non-positive everywhere. This
then guarantees that the r.h.s. of the equation (which follows from (11))
4 1 e
(φ 0 )2 = − Λ exp(−2A) − (D − 2)A00 (102)
D−2 D−3
is strictly positive. We can then solve for φ(y), and express y in terms of φ as φ(y) is a
monotonous function of y. Using (10) we can express the potential V as a function of y,
and, subsequently, as a function of φ. This way we can construct an infinite class of non-
singular domain walls with Λ e < 0 which localize gravity. 7
Here we would like to study some properties of such domain walls. In particular, let
us understand the asymptotic behavior of φ and A at large y. From (102) we obtain the
following leading behavior for φ 0 at large y:
4 1
(φ 0 )2 ∼ − eexp(−2A).
Λ (103)
D−2 D−3
7 A similar procedure was described in [32]. However, in [32] the second order equations of motion for φ and
A were rewritten as the first order equations. Unlike the Λ e = 0 case, for non-zero (in particular, negative) Λ e in
this process one loses a class of solutions. Thus, it is not difficult to show that, for any choice of V , the first order
equations of [32] do not have non-singular solutions with Λ e < 0 which localize gravity. More concretely, in such
solutions A0 must change sign for some finite y as A goes to −∞ for y → ±∞. However, solutions of this type
are lost in the process of taking the square root while deriving a first order equation for A0 from (10) — see below.
On the other hand, from (10) we have:

D−2
V∼ eexp(−2A).
Λ (104)
D−3
That is, at large y the potential V behaves as
V ∼ −4(φ 0 )2 , (105)
so that it is unbounded below there. This, in particular, implies that such solutions with
negative Λe cannot asymptotically approach AdS space (that is, they cannot be of the kink
type).
Let us define χ(φ) at large φ via
V (φ) ∼ −4[χ(φ)]2 . (106)

Then at large y we have
φ 0 ∼ χ(φ). (107)
Note that, for the domain wall to be non-singular, χ(φ) must not (up to usual “logarithmic”
factors — see below) grow faster than ∼ φ.
Thus, let χ(φ) ∼ ξ φ ν , where 0 < ν < 1. Then we have
1
φ(y) ∼ [(1 − ν)ξy] 1−ν , (108)
and
ν
A(y) ∼ − ln(ξy). (109)
1−ν
eP we must assume that exp[(D − 3)A] goes to zero at large y
Note that to have finite M
faster than 1/y. This then implies the following restriction on ν:
1
ν> . (110)
D−2
e
Thus, if this condition is not satisfied, then the corresponding domain walls with negative Λ
e
do not localize gravity. This is to be contrasted with the Λ = 0 case where such potentials
do give rise to walls which localize gravity. Also, note that to have finite volume in the y
direction, which is given by
Z
Vol = dy exp[(D − 1)A], (111)
we must assume that exp[(D − 1)A] goes to zero at large y faster than 1/y. This is satisfied
provided that ν > 1/D. Thus, for 1/D < ν < 1/(D − 2) the compactification volume is
finite, yet gravity is not localized.
Next, assume that χ(φ) ∼ φρ(ln |φ|). For the domain wall to be non-singular, the
function ρ(u) must be such that the integral
Z
du
(112)
ρ(u)
is unbounded at large u. The leading behavior for A is then obtained from the following
equation:
A0 = −ρ(−A). (113)
Thus, for instance, if ρ(u) ∼ ζ uσ , where 0 6 σ < 1, then
1
A ∼ −[(1 − σ )ζy] 1−σ . (114)
If ρ(u) ∼ ζ u, then
A ∼ C exp(ζy). (115)
Here C is a negative integration constant, and ζy is assumed to be positive.
First order equations

Sometimes it is convenient to rewrite the second order equations of motion for φ and
A as the first order equations. However, as we have already mentioned, for non-zero (in
particular, negative) Λ e some extra care is needed not to lose some of the solutions.
Thus, consider the case of non-singular domain walls with Λ e < 0 such that A goes to
−∞ at y → ±∞. Then it follows that at some finite y, call it y0 , A0 must change sign. It
is then not difficult to show that the first order equations for φ and A are given by:
φ 0 = αh, (116)
0
A = sign(y − y0 )βΩW, (117)
where
Wφ = sign(y − y0 )Ωh, (118)
s
D−1 Λ e
Ω = 1+ exp(−2A), (119)
D − 3 γ W2
2
V = h2 − γ 2 W 2 . (120)
Here we note that W does not vanish anywhere, and has an extremum at y = y0 . Also, φ 0
does not vanish at y = y0 , where by assumption A00 is non-positive.
5.2. Singular domain walls
By a singular domain wall we mean a solution where φ blows up at some finite value(s)
of y. We can then have domain walls that are singular on both sides (they interpolate
between y− and y+ , y± being the points where φ blows up), or domain walls that are
singular on one side only (that is, those interpolating between −∞ and y+ , or y− and
+∞). To begin with, note that if φ blows up at some finite y, call it y∗ , then A is bounded
above at y = y∗ . Indeed, let us rewrite (101) as follows:

4 1
exp(2A) 0 2
(φ ) + (D − 2)A = − 00 e
Λ. (121)
D−2 D−3
If A goes to +∞ at y = y∗ , then A00 also goes to +∞, and the above equation cannot be
satisfied.
Thus, A goes to either −∞ or a constant at y = y∗ . In either case A00 goes to −∞

there (which follows from (121)). This then implies that for domain walls singular on both
sides the compactification volume is automatically finite if we cut off the space in the y
direction at the singularity. 8 Similarly, for domain walls singular on one side only the
compactification volume is finite as long as on the non-singular side exp(2A) goes to zero
fast enough.
We now wish to show that if the scalar potential V is bounded above at the singularity,
then A necessarily goes to −∞ there (that is, it cannot go to a constant). To see this,
consider the sum of (10) and (11):
D−2
(D − 2) A00 + (D − 1)(A0 )2 + V − eexp(−2A) = 0.
Λ (122)
D−3
Now suppose V is bounded above at y = y∗ . Let us assume that A goes to a constant at
y = y∗ . It then follows from (122) that (A0 )2 goes to ∞. In fact, (A0 )2 goes to ∞ at least
as fast as −A00 . However, from (121) it follows that −A00 goes to ∞ as fast as (φ 0 )2 . We
therefore conclude that A0 blows up at least as fast as φ 0 . This, however, implies that A
cannot go to a constant at y = y∗ as φ blows up there.
Here we note that if we relax the requirement that V be bounded above at singularities,
then A could a priori go to a constant at a singularity. However, as we will point out in the
next subsection, for singularities to be physical an additional non-trivial condition must be
satisfied. As we will see in a moment, singular domain walls with the property that A goes
to a constant at at least one singularity do not satisfy this consistency condition.
5.3. Physically allowed singularities
Let us substitute the domain wall ansätz into the action S given in (1). We then obtain
e 2 ≡e
(here (∇φ) eµ φ ∇
g µν ∇ eν φ):
Z
p
S MPD−2 = d D x −e g exp[(D − 3)A] R e − 1 (∇φ)
e 2
α2
Z p
− d D−1 x −e g E[A, φ], (123)
where the energy functional E[A, φ] is given by

Z
1
E[A, φ] = dy exp[(D − 1)A] (D − 1)[2A00 + D(A0 )2 ] + 2 (φ 0 )2 + V (φ) .
α
(124)
Here the integral over y is taken from y− to y+ , where y± correspond to the edges of the
domain wall (y± can be finite or infinite). This energy functional can be rewritten as
8 Here one might suspect that not all singularities are physical. The purpose of the next subsection is to
determine the conditions for physically allowed singularities.
Z
1 0 2 0 2
E[A, φ] = dy exp[(D − 1)A] (φ ) − (D − 1)(D − 2)(A ) + V (φ)
α2
0 y+
+ 2(D − 1) A exp[(D − 1)A] . (125)
y−
The boundary term will become important in the following, so we will keep it. Here
we note that (9) and a linear combination of (10) and (11) with Λ e = 0 are the Euler–
Lagrange equations for the energy functional (125) [33]. Similarly, for non-zero Λ e the
same equations are the Euler–Lagrange equations for the functional
Z
b φ] ≡ E[A, φ] − dy D − 1 Λ
E[A, eexp[(D − 3)A]. (126)
D−3
Using the equations of motion for φ and A, it is not difficult to show that on the solution
we have
Z
y+
e
E[A, φ] = Λ dy exp[(D − 3)A] + 2 A0 exp[(D − 1)A] . (127)
y−
eµ φ = 0, and e
This then implies that on the solution, where ∇ gµν is independent of y, we
have
Z p y+
S=M e D−3 d D−1 x −e g e− Λ
R e − 2M D−2 A0 exp[(D − 1)A]
P P y−
Z p
× d D−1 x −e g, (128)
where
Zy+
eD−3 ≡ M D−2
M dy exp[(D − 3)A]. (129)
P P
y−
eP and the
This coincides with the (D − 1)-dimensional action with the Planck scale M
e
cosmological constant Λ
Z p
e
S =MeD−3 d D−1 x −e g Re− Λe (130)
P
up to the surface term

Z p
2 y
− MPD−2 A0 y+− d D−1 x −e
g, (131)
D−1
where
A ≡ exp[(D − 1)A]. (132)
Thus, to have a consistent (D − 1)-dimensional interpretation, we must require that the
aforementioned boundary contribution vanish. That is, we have an additional consistency
condition:
y
A0 y+− = 0. (133)
Note that for non-singular domain walls with finite M eP this condition is automatically
satisfied — indeed, for such walls A and, therefore, A0 asymptotically go to zero at y± =
±∞. However, for singular domain walls this condition is non-trivial. In fact, in the case
of singular domain walls A0 (y± ) need not even be finite. For the condition (133) to be
meaningful, we must then require that
A0 (y± ) are finite, and A0 (y+ ) − A0 (y− ) = 0. (134)
Note that if A goes to zero at singularities, then A0
is non-positive at y = y+ , and non-
negative at y = y− , so that the condition (133) is satisfied if and only if
A0 (y± ) = 0. (135)
As we will show in a moment, singularities where A is finite do not satisfy the condition
(134). This then implies that at physically allowed singularities A as well A0 must
necessarily vanish.
Potentials bounded above

Let us first consider the case where V is bounded above at singularities. In this case A
goes to zero at y± , and, therefore, A0 must also go to zero there.
If Λe 6 0 these conditions imply a certain restriction on the scalar potential. Thus,
multiplying both sides of (122) by exp[(D − 1)A], we obtain
D − 2 A00 D − 2 Λ e
V =− + . (136)
D−1 A D − 3 A D−1
2
Now, both A and A0 must vanish at y± . Then it follows that A00 is non-negative at y± .
This then implies that for Λe 6 0 at the singularities the scalar potential must be unbounded
below: 9
e 6 0.
V → −∞ at y = y± for Λ (137)
Note that this condition is necessary but might not always be sufficient for (135) to be
satisfied. Thus, generally the full condition (135) should be used to check whether a given
singularity is physically allowed.
Here we note that for Λe > 0 the corresponding condition on V is more detail dependent.
In particular, as we will see in the following, there exist (discontinuous) domain walls
with physically acceptable singularities and Λ e > 0 such that V is not unbounded below at
singularities.
Potentials unbounded above

Let us now consider potentials unbounded above at the singularities. Let us see if we can
satisfy the condition (134). First, the above discussion for singularities where A → −∞
applies to this case as well. In particular, solutions with Λ e 6 0 with singularities where
A → −∞ and V is unbounded above are not physically allowed.
Next, let us assume that A goes to a constant at, say, y+ . Recall that A00 must go to −∞.
In particular, (φ 0 )2 goes to ∞ as fast as −A00 . Let A0 ≡ N , and z ≡ 1/(y+ − y). As y → y+
9 This is a much stronger condition than that recently conjectured in [34] to be necessary for a singularity to be
e = 0) a singularity is physical only if V is
physical. In particular, it was argued in [34] that (for solutions with Λ
bounded above there.
we have z → +∞. Now, N 0 = −z2 Nz , where the subscript z denotes derivative w.r.t. z.
On the other hand, φ 0 = −z2 φz . This then implies that at z → +∞ we have (see (121))
D − 2 1p
φz ∼ η Nz , (138)
2 z
where η = ±1. For φ to blow up at z → +∞ (that is, at y → y+ ) it is then necessary
that Nz be a non-decreasing 10 function of z. But then it follows that N blows up at z →
+∞ at least as fast as z. That is, A0 blows up at y → y+ at least as fast as 1/(y+ − y).
Thus, the condition (134) cannot be satisfied for such domain walls (recall that A0 =
(D − 1)A0 exp[(D − 1)A], so to have finite A0 (y+ ) we must have finite A0 (y+ ) as A goes to
a constant there by assumption). Similarly, if A goes to a constant at y = y− , (134) cannot
be satisfied there. We therefore conclude that singularities where A goes to a constant are
e
unphysical (regardless of the value of Λ).
Physical interpretation
From the above discussion it follows that for singularities to be physical they must
satisfy the condition (135). Here we would like to discuss a physical interpretation of this
condition.
Thus, we would like to be able to cut off the space in the y direction at singularities. That
is, singularities correspond to boundaries of the space. On the other hand, the condition
√
(135) can be rewritten as (note that A = −G)
√ 0
−G y± = 0. (139)
That is, at a boundary the normal derivative of the volume density must vanish. This is just
as well for otherwise there would be no reason why we should not continue the space in
the y direction beyond such a boundary.
Note that if we have a domain wall which is singular on both sides, then topologically the
space is an interval, which is the same as S1 /Z2 . Similarly, if the domain wall is singular on
one side only, then topologically the space is R/Z2 . However, the actual compactification,
say, in the former case does not correspond to that on S1 /Z2 in the sense that it cannot be
viewed as an orbifold of an S1 compactification. In particular, in this case we do not have
the usual Kaluza–Klein modes (or, more precisely, their combinations invariant under the
corresponding Z2 orbifold action) arising in the S1 compactifications.
5.4. Discontinuous domain walls
In the previous subsections we discussed smooth domain walls. Here we would like to
consider discontinuous domain walls. More concretely, both φ and A could be continuous,
but need not have continuous derivatives. In particular, to have a non-singular domain wall
that localizes gravity we must make sure that A goes to −∞ at y → ±∞. Also, in the
10 More precisely, this is correct up to the usual “logarithmic” factors. For instance, let N ∼ 1/ ln2 (z). Then
z
φz ∼ 1/z ln(z), and φ ∼ ln(ln(z)), so that φ blows up at z → +∞. However, such logarithmic factors do not
affect the following discussion. In particular, in this case N ∼ z/ log2 (z), which blows up at large z.
case of singular domain walls A must go to −∞ at singularities. This implies that A0

must change sign at finite y. This can happen continuously as in the previous subsections.
However, a priori we also have the following possibility. 11 Let A0 > 0 for y < y0 , and
A0 < 0 for y > y0 . At y = y0 A0 is discontinuous. Then A00 behaves as a δ-function at
y = y0 . To be able to satisfy the equations of motion for A and φ we then must add a
source term to the action S in (1):
Stotal = S + Ssource , (140)
where
Z
√
Ssource = −MPD−2 d D−1 x −g f (φ). (141)
Σ
Here Σ is the y = y0 hypersurface,
gµν ≡ δµ M δν N GMN , (142)
and f (φ) describes the coupling of φ to the space–time defect (often referred to as a
“brane”) corresponding to the hypersurface Σ. The above source term (generically) leads
to discontinuities in A0 and φ 0 at y = y0 , and the corresponding jump conditions must be
imposed on the solution.
The equations of motion now read:
√
8 −g
∇ 2 φ − Vφ − √ fφ δ(y − y0 ) = 0, (143)
D−2 −G
1 4 1
RMN − GMN R = ∇M φ∇N φ − 12 GMN (∇φ)2 − GMN V
2 D−2 2
√
1 −g
− √ δM µ δN ν gµν f δ(y − y0 ). (144)
2 −G
As before, we are interested in solutions with the metric of the following form:
2
dsD = exp(2A)e
gµν dx µ dx ν + dy 2 , (145)
where the warp factor A, which is a function of y, is independent of the coordinates x µ ,
and the (D − 1)-dimensional metric e gµν is independent of y. Moreover, φ is independent
of x µ , but can depend non-trivially on y. With this ansätz, we have the following equations
of motion for φ and A:
8 00
φ + (D − 1)A0 φ 0 − Vφ − fφ δ(y − y0 ) = 0, (146)
D−2
4 D−1
(D − 1)(D − 2)(A0 )2 − (φ 0 )2 + V − Λeexp(−2A) = 0, (147)
D−2 D−3
4 1 1
(D − 2)A00 + (φ 0 )2 + Λeexp(−2A) + f δ(y − y0 ) = 0. (148)
D−2 D−3 2
The jump conditions read:
11 Here we should point out that one can also consider discontinuous domain walls where A0 does not change
sign at the discontinuity. Furthermore, the following discussion can be straightforwardly generalized to include
multiple discontinuities.
8 0
φ (y0 +) − φ 0 (y0 −) − fφ (φ0 )7 = 70, (149)
D−2
1
(D − 2) A0 (y0 +) − A0 (y0 −) + f (φ0 ) = 0, (150)
2
φ(y0 +) = φ(y0 −) ≡ φ0 , (151)

A(y0 +) = A(y0 −) ≡ A0 . (152)
From our previous discussions it should be clear that if we demand that A goes to −∞ at
y = y± , then A0 (y0 −) > 0 and A0 (y0 +) 6 0, so that from the jump condition on A0 we
have
f (φ0 ) > 0. (153)
Since MPD−2 f (φ0 ) is nothing but the tension of the aforementioned space–time defect
(“brane”), it then follows that gravity can be localized in this context only if this tension is
non-negative. 12
It is now clear how to obtain a solution of this type with finite compactification volume.
Start with two smooth domain walls, call them (1) and (2), which are solutions of the
equations of motion without the source term. Let these domain walls be such that A(1) →
(1) (2) (2) (1)
−∞ as y → y− , and A(2) → −∞ as y → y+ . Moreover, let y+ > y− , and assume that
(1) (2)
there is a point y0 , y− < y0 < y+ , such that φ (y0 ) = φ (y0 ) ≡ φ0 , and A(1)(y0 ) =
(1) (2)
A(2)(y0 ) ≡ A0 with some finite φ0 and A0 . Now consider the domain wall given by
(1)
φ(y) = φ (1) (y), y− < y 6 y0 , (154)
(2)
φ(y) = φ (2)
(y), y0 < y < y+ , (155)
(1)
A(y) = A(1) (y), y− < y 6 y0 , (156)
(2)
A(y) = A (y),(2)
y0 < y < y+ . (157)
This discontinuous domain wall will satisfy the equations of motion with the source term
as long as the function f (φ) is such that the following jump conditions are satisfied:
8 (2) 0 0
φ (y0 ) − φ (1) (y0 ) − fφ (φ0 ) = 0, (158)
D−2
0 0 1
(D − 2) A(2) (y0 ) − A(1) (y0 ) + f (φ0 ) = 0. (159)
2
Note that from our discussion at the beginning of this section it follows that
discontinuous domain walls with finite M eP which are non-singular or singular on one side
only cannot have positive cosmological constant. On the other hand, discontinuous domain
walls with finite MeP which are singular on both sides can a priori have either vanishing or
non-vanishing cosmological constant of either sign.
Here we can repeat the analysis of the previous subsection for discontinuous domain
walls. It is then not difficult to show that Stotal on the solution has the same form as (128), so
12 Note that this need not be the case if, say, A0 does not change sign at the discontinuity.
that we must impose the consistency condition (133). The latter is automatically satisfied
for non-singular domain walls, but for singular domain walls it is non-trivial. Here we
would like to give a simple example which does not satisfy this condition.
Examples with unphysical singularities

Thus, consider the case where the scalar potential vanishes:
V (φ) ≡ 0. (160)
Let us look for domain wall solutions with zero cosmological constant. Let us rewrite the
scalar potential in terms of W :
V = Wφ2 − γ 2 W 2 . (161)
Without loss of generality we can choose
W = ξ exp(γ φ), (162)
e = 0 are then the
where ξ is an arbitrary constant. The smooth domain walls with Λ
solutions to the equations
φ 0 = αWφ = αξ γ exp(γ φ), (163)
0
A = βW = βξ exp(γ φ). (164)
These solutions are given by:
1
φ = − ln αξ γ 2 (y∗ − y) , (165)
γ
β
A=C− 2
ln αξ γ 2 (y∗ − y) , (166)
αγ
where C is an integration constant, and y∗ corresponds to the location of the singularity.
As expected, at the singularity A → −∞ (recall that α and β have opposite signs). Note,
however, that at the other edge A → +∞, so that these smooth domain walls do not
localize gravity.
Now take two such domain walls satisfying the aforementioned properties, and construct
a discontinuous domain wall with the edges corresponding to singularities (that is, such a
domain wall is singular on both sides). This domain wall then has finite compactification
volume. However, it does not satisfy the consistency condition (133), and therefore the
singularities are unphysical. Indeed, we have (for definiteness we assume that αξ > 0, so
that y < y∗ , that is, y∗ corresponds to y+ )
A = exp[(D − 1)A] = exp[(D − 1)C]αξ γ 2 (y∗ − y), (167)
so that
A0 = − exp[(D − 1)C]αξ γ 2 . (168)
This then implies that the consistency condition (134) is not satisfied 13
for these
solutions. 14 In fact, it is not difficult to show that the same conclusion applies to
13 This observation was also made by Gary Shiu.

14 Here we note that such discontinuous domain walls were recently discussed in [35–37].
e obtained by starting with

discontinuous singular domain walls with non-vanishing Λ
V (φ) ≡ 0 and adding a source term. 15
5.5. Smooth vs. discontinuous domain walls
The purpose of this subsection is to point out that a discontinuous domain wall of the
aforementioned type cannot be thought of as a limit of a smooth domain wall. The physical
reason for this is actually quite simple. Thus, note that in the case of a smooth domain wall
eP the localization widths for the scalar field φ and gravity are related to each
with finite M
other — essentially they are both determined by the fall-off of the warp factor exp(2A).
Thus, if we take the localization width of the scalar field to zero with the intension of
reproducing the source term in (140), the localization width for gravity will also go to
zero. Then for finite D-dimensional Planck scale MP the (D − 1)-dimensional Planck
scale MeP goes to zero as well. This indicates that we cannot approximate a discontinuous
domain wall by a limit of a smooth domain wall.
To understand this point in a bit more detail, let us consider the original Randall–
Sundrum scenario [24,25], where the scalar potential is a negative constant
V = −ξ 2 . (169)
e = 0. We have the
Let us consider solutions with vanishing cosmological constant Λ
following smooth solutions (φ0 is a constant):
φ = φ0 , (170)
βξ
A= y + const, (171)
γ
which correspond to the D-dimensional AdS space–time.
Now let us add the source term with f (φ) independent of φ:
f (φ) ≡ f0 . (172)
eP (for definiteness we
We then find the following discontinuous domain walls with finite M
are assuming βξ > 0):
φ(y) = φ0 , (173)
A(y) = A0 − βξ |y − y0 |, (174)
with the following jump condition:
f0 = 4(D − 2)βξ, (175)
which implies that the “brane”√ tension must be fine-tuned to the D-dimensional
cosmological constant (ξ = −V ) for such a solution to exist. Note that this solution
is precisely the Randall–Sundrum solution up to the extra spectator scalar field φ.
We can now ask whether we can approximate this solution by a limit of a smooth
solution. The starting point, as it should become clear in a moment, is not very important
15 Such domain walls were recently studied in [38].

here as long as it correctly reproduces the features of the above solution such as the fact that
away from the “brane” we have AdS vacua. Thus, for illustrative purposes, let us consider
the kink solution discussed in Section 3.2. Now, W is given by

3ξ 1
W= φ − φ3 , (176)
2γ 3
so that in the AdS minima with Wφ = 0 (φ = ±1) we have W = ±ξ/γ , and V = −ξ 2 . The
solution for φ and A is given by

3αξ
φ(y) = tanh (y − y0 ) , (177)
2γ

2β 3αξ 1 1
A= ln cosh (y − y0 ) − + C. (178)
3α 2γ 4 cosh2 3αξ (y − y0 )
2γ
It is now clear that the Randall–Sundrum solution for A is obtained in the limit
ξ → ∞. (In this limit φ(y) = −sign(y − y0 ).) However, in this limit the D-dimensional
cosmological constant V = −ξ 2 → −∞, and the (D − 1)-dimensional Planck scale
eD−3 = 2 1 D−2
M M →0 (179)
P
D − 3 βξ P
for finite D-dimensional Planck scale MP . Note that even if we attempt to take MP to
infinity to have MeP finite, the effective D-dimensional cosmological constant M D−2 V
P
would still go to −∞. This indicates that discontinuous solutions cannot be thought of as
limits of smooth solutions.
Examples of discontinuous solutions with Λ e 6= 0

Before we end this section, we would like to discuss discontinuous solutions with non-
zero cosmological constant. In particular, let us consider the Randall–Sundrum setup. As
before this is achieved by setting φ(y) ≡ φ0 , and V = −ξ 2 . A restricted set of solutions
with Λe 6= 0 in this context were considered in [29,39,40]. Here we would like to discuss
the most general solutions of this type. 16
Thus, first consider the equations of motion for A without the source term:
D−1
(D − 1)(D − 2)(A0 )2 = ξ 2 + eexp(−2A),
Λ (180)
D−3
1 e
(D − 2)A00 = − Λ exp(−2A). (181)
D−3
Let
exp(A) ≡ U. (182)
Then we have:
D−1
(D − 1)(D − 2)(U 0 )2 = ξ 2 U 2 + e
Λ, (183)
D−3
(D − 1)(D − 2)U 00 = ξ 2 U. (184)
e were recently studied in a more general context in [41].

16 Discontinuous domain walls with non-zero Λ
The general solution to this system of equations is given by

e
Λ
U = C exp(ay) − 2 2 exp(−ay) , (185)
4ξ C
where
ξ
a≡√ , (186)
(D − 1)(D − 2)
and C is an integration constant.
Note that if Λe < 0, then we must choose C > 0, and the solution is non-singular. More-
over, U blows up at both y → ±∞, so even if we add the source term, the corresponding
e > 0 U vanishes for some finite y,
solution cannot localize gravity. On the other hand, for Λ
so that the domain wall is singular. Note that without loss of generality we can assume that
C > 0. Then we can obtain a singular domain wall with non-vanishing Λ eP
e and finite M
by adding the source term and gluing together the appropriate parts of two smooth domain
walls. Thus, consider the solution (for definiteness we assume a > 0, that is, ξ > 0)

Λe
U (y) = C exp(−a|y|) − 2 2 exp(a|y|) , (187)
4ξ C
where
p
e
Λ
C> . (188)
2ξ
The singularities (which are physical as they satisfy (135)) are located at

1 2ξ C
y± = ± ln p . (189)
a Λe
The discontinuity is located at y0 = 0, and the corresponding jump condition reads:
f0 = 4(D − 2)a coth(a|y± |) > 0. (190)
Here we note that the solutions discussed in [29,39,40] correspond to taking C −
e 2 C = 1. However, generally C is an independent integration constant. In particu-
Λ/4ξ
e is not determined by specifying ξ and f0 . However, the ratio of the effective
lar, a priori Λ
(D − 1)-dimensional cosmological constant M eD−3 Λ e to Me D−1 , that is, Λ/
eM e2 , depends
P P P
on Λ e and C via the ratio Λ/C
e 2 , which is fixed via ξ and f0 .
Next, consider the same U (y) as in (187) with
p
Λe
C< . (191)
2ξ
It then describes a non-singular domain wall which does not localize gravity. In fact, in this
case the jump condition reads:
f0 = −4(D − 2)a coth(ζ ) < 0, (192)
where
p e
1 Λ
ζ ≡ ln . (193)
a 2ξ C
Note that such a domain wall would require a negative tension “brane”.
5.6. Summary
In this subsection we would like to briefly summarize the results of this section.
(i) Non-singular smooth domain walls with A → −∞ at y → ±∞ (that is, finite
compactification volume and M eP ) cannot have Λ e > 0. On the other hand, non-singular
smooth domain walls approaching AdS vacua on at least one side can only have Λ e = 0.
(ii) Singular smooth domain walls can a priori have vanishing or non-vanishing Λ e
of either sign. Singularities with A → −∞ are physical only if the scalar potential is
unbounded below there. However, singularities where A → const, which can only arise
if the potential is unbounded above there, are unphysical. In particular, it is not consistent
to cut off the space at such singularities in the y direction.
(iv) The aforementioned conclusions also apply to discontinuous domain walls (with a
δ-function source term). However, such a domain wall cannot be thought of as a limit of a
smooth domain wall with non-zero M eP .
6. Normalizable modes
In this section we would like to discuss normalizable modes living on the domain
walls discussed in the previous sections. In particular, let us focus on non-singular domain
walls. To study normalizable modes, it is convenient to consider small fluctuations around
the background metric GMN and the scalar field φ, which we will denote by hMN
and ϕ, respectively. Using the diffeomorphism invariance of the theory we can choose
the following gauge:
hµD = hDD = 0. (194)
Instead of the remaining metric fluctuations hµν , it will be convenient to work with e
hµν
defined via
hµν = exp(2A)e
hµν . (195)
The quadratic action for e
hµν and ϕ is given by
S[e
hµν , ϕ]/MPD−2
Z np
b− Λ
p 1 e 2 o
= d D x exp[(D − 3)A] −b
g R e (2) − −e g 2 ∇ϕ
α
Z
p 1 1 1
g 2 (ϕ 0 )2 + Vφφ ϕ 2 − 2 φ 0e
− d D x exp[(D − 1)A] −e h0 ϕ
α 2 α

1 e0 2 e

0 2
+ hµν − h . (196)
4
Here we are using the following notations. First, b
g ≡ det(b
gµν ), where b gµν + e
gµν ≡ e hµν .
b
Also, R is the (D − 1)-dimensional Ricci scalar constructed from the metric b gµν . The
superscript “(2)” in the term
p
−bg Rb− Λ e (2) (197)
indicates that only the terms quadratic in e hµν should be kept in the corresponding
expression. Finally, e
h ≡ehµ , and (e
h0µν )2 ≡ (e
hµν )0 (e
hµν )0 , where the indices are lowered and
µ
raised with the metric egµν and its inverse. We also note that the boundary contributions
discussed in Section 5.3 vanish in the case of non-singular domain walls.
As in the case of non-gravitational domain walls, the above action has a much simpler
form if instead of ϕ we use the field ω defined via
ϕ ≡ φ 0 ω. (198)
Thus, the quadratic action for e

hµν and ω is given by:
D−2
S ehµν , ω MP
Z
= d D x exp[(D − 3)A]

p p e
× −bg Rb− Λ e (2) − −e 1 e 2 − D − 1 Λ ω2
g 2 (φ 0 )2 ∇ω
α D−2D−3
Z p
− d D x exp[(D − 1)A] −e g

1 0 2 0 2 1 e∗0 2 D − 2 e0 2 0 2 2

× 2 (φ ) (ω ) + hµν − h + 2γ (φ ) ω . (199)
α 4 D−1
Here
e 1 e
h∗µν ≡ e
hµν − he
gµν (200)
D−1
is the traceless part of e
hµν .
Next, we would like to study normalizable modes. Here we are assuming that the warp
factor exp(2A) goes to zero at y → ±∞ fast enough, so that the compactifications volume
as well as M eP are finite. Then normalizable modes correspond to configurations with
0
ω = 0. It is not difficult to show that the equation of motion for such ω simplifies as
follows:
e e2 ω + 2(D − 1)A00 ω.
h0 = −2 exp(−2A)∇ (201)
e
Note that this equation does not explicitly contain Λ.
Thus, the equation of motion (201) for ω = ω(x µ ) seems to imply that we have
normalizable modes satisfying the (D − 1)-dimensional Klein–Gordon equation
e2 ω = m2 ω
∇ (202)
for any m2 (including massive and tachyonic). At first this might appear to imply that the
domain wall is unstable. This is, however, not the case due to the following. Note that ω (or,
equivalently, ϕ) mixes with e
h0 , that is, the y derivative of the trace part of e
hµν . Moreover,
as we will see in a moment, there is a residual diffeomorphism invariance which respects
the gauge choice (194). These facts have important implications which we would like to
discuss next.
6.1. Residual diffeomorphism invariance
The gauge choice (194) does not fix all the gauge freedom in the system. Thus, there is
a residual diffeomorphism invariance which preserves (194). 17 For our purposes here it
will suffice to consider infinitesimal residual diffeomorphisms. Thus, we have
δhMN = ∇M ξN + ∇N ξM . (203)
In particular,
0
δhDD = 2ξD , (204)
and the gauge condition hDD = 0 is preserved if and only if
0
ξD = 0. (205)
Furthermore, we have
eµ ξD + ξµ0 − 2A0 ξµ ,
δhµD = ∇ (206)
where on the r.h.s. we have kept only the terms independent of hMN , which is consistent
with the linearized approximation. Requiring that the gauge condition hµD = 0 is
preserved, we obtain
ξµ = exp(2A)e
ξµ , (207)
where e
ξµ satisfies the following first order differential equation
e eµ ξD .
ξµ0 = − exp(−2A)∇ (208)
Finally, for the rest of the hMN components we obtain
eµ ξν + ∇
δhµν = ∇ eν ξµ + 2A0 exp(2A)ξDe
gµν , (209)
or, equivalently,
δe eµe
hµν = ∇ eνe
ξν + ∇ ξµ + 2A0 ξDe
gµν . (210)
In particular, we have
δe eµe
h = 2∇ ξµ + 2(D − 1)A0 ξD , (211)
from which it follows that
δe e2 ξD + 2(D − 1)A00 ξD .
h0 = −2 exp(−2A)∇ (212)
The last result implies that for any choice of ω in (201) (where ω is assumed to be
independent of y), the corresponding e h is a pure gauge, and the corresponding gauge
parameter is given by
ξD = ω. (213)
That is, ω is not a propagating degree of freedom, but a gauge parameter, and, up to a gauge
transformation, e h is independent of y. This result has a simple physical interpretation to
which we now turn.
17 This residual gauge invariance was discussed in the context of the Randall–Sundrum model in [42], and also
in [43].
6.2. Localization of gravity as a Higgs mechanism
The above analyses suggest that localization of gravity is actually a Higgs mechanism
for the graviton. In particular, in the D-dimensional language we have the fields hMN
and ϕ with D(D − 3)/2 and 1 physical degrees of freedom, respectively. In the
(D − 1)-dimensional language together with the scalar ϕ we have the fields hµν , hµD
and hDD corresponding to the (D − 1)-dimensional graviton, a graviphoton and a scalar,
respectively. If we turn off the coupling to gravity, there is a ϕ zero mode (which
might or might not be normalizable in this context) corresponding to the Goldstone
mode of the broken translational invariance in the y direction. However, once we gauge
diffeomorphisms, that is, once we consider the coupling to gravity, the ϕ zero mode is
eaten by the graviphoton, which now becomes massive. The scalar hDD is also massive —
the volume of the compactification, which is finite, is not a free parameter but is completely
fixed in terms of the parameters in the scalar potential V . This is analogous to what
happens, say, in the abelian Higgs model, where a U (1) gauge field is coupled to a single
complex scalar field. Once the latter acquires an expectation value, the angular part is
eaten by the gauge field, which therefore becomes massive, while the radial part acquires a
mass due to a non-trivial scalar potential. However, as we will see in a moment, the Higgs
mechanism that takes place in the localization of gravity has certain novel features, which
is due to important differences between gravity and an ordinary gauge theory.
6.3. Zero and massive modes
To analyze normalizable modes, consider the standard factorized form of the corre-
sponding field, which is e
hµν in this case:
e
hµν (x λ , y) = ζµν (x λ )Σ(y). (214)
Using the fact that ω in (201) is a gauge parameter, without loss of generality we can set it
to zero, which implies eh0 = 0, and, consequently, either Σ 0 = 0, or ζµ = 0. In the former
µ
case we have a zero mode

e
hµν (x λ , y) = const × ζµν (x λ ), (215)
where ζµν satisfies the equation of motion for the massless (D − 1)-dimensional graviton,
which is nothing but the Euler–Lagrange equation obtained by varying the quadratic
(D − 1)-dimensional action
Z p

e
Se hµν ≡ M e D−3 d D−1 x −b g Rb− Λ e (2) (216)
P
w.r.t. e
hµν (or, equivalently, ζµν ). In this case the zero mode is quadratically normalizable,
that is, the (D − 1)-dimensional Planck mass M eP is finite.
0
Next, if Σ 6= 0, then using (199) it is not difficult to show that Σ(y) must satisfy the
equation
0
exp[(D − 1)A]Σ 0 + m2 exp[(D − 3)A]Σ = 0, (217)
where m2 is a constant. As to ζµν , it then satisfies the equation of motion of a (D − 1)-

dimensional symmetric traceless (recall that e
h = 0 in this case) two-index tensor field with
mass squared m2 6= 0 propagating in the background metric e gµν . Using (217) the action
e
for hµν can be written as
Z
p 1 2
e
Sehµν ≡ MPD−2 d D−1 x dy exp[(D − 3)A] −bg Rb− Λ e (2) − m2 e h∗µν ,
4
(218)
and the norm of e
hµν is proportional to
Z
2
e
hµν ∼ dy exp[(D − 3)A]Σ 2 . (219)
We would now like to show that massive modes with m2 > 0 are plane-wave normalizable
(in contrast to the zero mode, which is quadratically normalizable), while tachyonic modes
with m2 < 0 are not normalizable.
To see this, let us make the coordinate transformation y → z so that the metric takes the
form:

dsD2
= exp(2A) e gµν dx µ dx ν + dz2 . (220)
That is,
dy = exp(A) dz, (221)
where we have chosen the overall sign so that z → ±∞ as y → ±∞. In this new
coordinate system (217) reads:
Σzz + (D − 2)Az Σz + m2 Σ = 0, (222)
where the subscript z denotes derivative w.r.t. z. Let

1 b
Σ ≡ exp − (D − 2)A Σ. (223)
2
b reads:
The equation for Σ

b 1 1 2 b
Σzz + m − (D − 2)Azz − (D − 2) (Az ) Σ = 0.
2 2
(224)
2 4
Let us assume that m2 6= 0. Then for large z the m2 term is dominant in the square brackets
in (224) for
Az = A0 exp(A) → 0, (225)

Azz = A00 + (A0 )2 exp(2A) → 0 (226)
b has the following
for large y (that is, large z). This then implies that for m2 > 0 Σ
asymptotic behavior at large z:
b ∼ C1 sin(mz) + C2 cos(mz),
Σ(z) (227)
b the norm of e
where C1 , C2 are some constants. On the other hand, in terms of Σ hµν (219)
is given by
Z Z
dy exp[(D − 3)A]Σ = dz Σ
2 b2 . (228)
This implies that massive ehµν modes are plane-wave normalizable (but not quadratically
normalizable). Thus, we have a continuum of massive bulk modes just as in the Randall–
Sundrum model.
What about the tachyonic modes? It is not difficult to see that these are not normalizable.
Indeed, (224) can be written as [32]
(QQ† )Σ b
b = m2 Σ, (229)
where
1
Q ≡ ∂z + (D − 2)Az , (230)
2
1
Q† = −∂z + (D − 2)Az . (231)
2
Evidently (229) does not have normalizable solutions with m2 < 0 as it is nothing but
Schrödinger’s equation of a supersymmetric quantum mechanics with the Hamiltonian
H ≡ QQ† [32].
Before we end this section, let us make the following remark. The existence of the
quadratically normalizable zero mode as well as plane-wave normalizable massive bulk
modes in the above setup is similar to what happens in the original Randall–Sundrum
model. However, localization of gravity in the case of smooth domain walls arises via
a spontaneous breakdown of diffeomorphism invariance, while in the Randall–Sundrum
model as well as the cases where we have discontinuous domain walls diffeomorphism
invariance is explicitly broken by the “brane”. That is, in the former case we have Higgs
mechanism, while in the latter case we do not, and, in particular, there is no massless
graviphoton to begin with. This is precisely the underlying reason why discontinuous
domain walls cannot be thought of as limits of smooth domain walls.
7. The upshot
In the previous sections we have studied possible types of domain wall solutions of
(1) which localize gravity. In this section we would like to discuss some aspects of the
cosmological constant problem within the context of a brane world realized via such
a domain wall. In particular, for the reasons which should be clear from our previous
discussions, we will focus on smooth domain walls.
To begin with, let us consider non-singular domain walls with infinite tension. In
particular, suppose the corresponding scalar potential is such that it does not have local
minima and is unbounded below. Then it is clear that no solution that does not localize
gravity is consistent. However, as we saw in the previous sections, subject to non-
singularity conditions on the scalar potential, there are (classically) consistent domain wall
solutions which localize gravity. This implies that the theory will ultimately 18 end up
choosing such a background. In such cases we therefore have spontaneous localization of
gravity via the Higgs mechanism discussed in subsection B of the previous section.
Had it been the case that domain wall solutions with infinite tension could only have
vanishing (D − 1)-dimensional cosmological constant, the aforementioned property a
priori might have been considered as a possible solution to the cosmological constant
problem. However, as we saw in the previous sections, infinite tension domain walls can
have negative cosmological constant. Could we then conclude that such domain walls solve
“half” of the cosmological constant problem as they do not allow Λ e > 0?
Here we would like to point out that such a conclusion might be premature. The reason
is twofold. First, to have a non-singular domain wall, the scalar potential must satisfy
certain conditions discussed in the previous sections. Quantum corrections can a priori
modify the potential in such a way that these conditions are no longer satisfied. For singular
domain walls, on the other hand, non-zero cosmological constant of either sign is a priori
allowed. Nonetheless, one could in principle argue that (at least in some sense) the subset
of potentials that satisfy non-singularity conditions has non-zero measure, so this particular
objection might be avoidable.
There is, however, another point here, which could potentially be more problematic.
Quantum corrections modify not only the scalar potential, but generically also introduce
higher derivative, in particular, higher curvature terms, which become important in such
backgrounds. Thus, for instance, the D-dimensional Ricci scalar is given by:

R=R eexp(−2A) − (D − 1) 2A00 + D(A0 )2 , (232)
where R e is the (D − 1)-dimensional Ricci scalar. Even if R e = 0 on the solution (that is,
if the (D − 1)-dimensional cosmological constant is vanishing), the D-dimensional Ricci
scalar blows up at large y unless A0 goes to a constant there. 19 In the case of infinite
tension domain walls A0 diverges at large y, and so does the Ricci scalar R, so that higher
curvature (or, more generally, higher derivative) terms cannot be ignored. What is worse,
in this case it is not even clear if there exists a controlled approximation scheme for taking
into account such terms. The underlying reason for this is that such domain walls are
actually singular but the singularities are located at y → ±∞ (that is, at infinite distances
from the “core” of the domain wall). In particular, a priori there does not seem to be a
reason to believe that any conclusions about allowed values of the (D − 1)-dimensional
cosmological constant would not be modified by higher curvature terms. In fact, as we will
point out in the following, in this context it is not even clear whether such domain walls
do indeed localize gravity. Note that the above discussion also applies to singular domain
walls with singularities located at finite values of y.
The above discussion suggests that we consider solutions where the D-dimensional
Ricci scalar R goes to constant values at large y. Such solutions should have vanishing
18 Here we should point out that a priori there might also exist consistent localized solutions of codimension
r > 1 if we have r scalar fields. In such a case the corresponding theory might also be able to choose such
backgrounds.
e 6= 0, then R ultimately blows up at large y as long as A goes to −∞ there.
19 Note that if on the solution R
(D − 1)-dimensional cosmological constant, and A0 should go to constant values at large y.

In particular, if the cut-off scale for higher derivative terms in the action is of order M∗ ,
then for such solutions we have a controlled approximation scheme for including higher
curvature terms as long as
|R| ∼ (A0 )2 M∗2 . (233)
In fact, from our previous discussions it should be clear that such solutions must necessarily
interpolate between two adjacent local AdS minima of the scalar potential. However, if
such minima are located at finite values of φ, then a priori there exist other solutions with
non-zero cosmological constant. 20 These solutions have the property that φ “overshoots”
the values corresponding to the local AdS minima. To avoid this, one could consider
potentials which have no local minima but approach constant negative values at large φ.
7.1. Runaway potentials
Thus, we are lead to consider the following setup. Let the scalar potential, such that
it has no local minima, be of the runaway type, that is, it approaches constant values at
large φ. More precisely, let these values be negative, and satisfy the condition

V (φ → ±∞) M 2 , (234)
∗
with M∗ defined as above. Then the corresponding domain walls necessarily have
vanishing cosmological constant. In fact, it is not difficult to see that this statement holds
even if we take into account higher curvature (or, more generally, higher derivative) terms
as these are now under control. 21 We therefore conclude that for such domain walls (at
least in some sense — see below) the higher curvature corrections do not change the
conclusion that the cosmological constant must vanish. There is, however, the following
issue here. A priori there does not seem to be a reason to believe that quantum corrections
will not modify the behavior of the scalar potential at large φ. In particular, if instead
of constant values the potential goes to +∞ or −∞ at large φ, then the conclusion that
the cosmological constant must vanish would be modified. 22 Note that supersymmetry
locally broken in the “core” of the domain wall does not seem to help here — since the
compactification volume is actually finite, supersymmetry breaking in the bulk is non-
vanishing as it is only suppressed by the volume factor.
At any rate, if the runaway behavior of the scalar potential is not lifted by quantum
corrections, 23 then such domain walls could a priori lead to a solution to the cosmological
constant problem.
20 If the scalar potential is bounded below, then before taking into account higher derivative terms such solutions
are necessarily singular and have positive cosmological constant. Otherwise one can a priori have singular or non-
singular solutions with negative cosmological constant. However, we should mention that these conclusions might
be modified by higher curvature terms.
21 More precisely, this is correct subject to the convergence of the |R|/M 2 expansion.
∗
22 Actually, if the potential is of the aforementioned runaway type on at least one side, it is enough to ensure
vanishing of the cosmological constant.
23 Here we should mention that typically it is not easy to lift runaway behavior perturbatively.
Here we would like to discuss an explicit example of such a domain wall. Since the
domain wall has vanishing Λ,e it is convenient to express the scalar potential in terms of W
via (16). Thus, consider the example where
W = ξ tanh(φ). (235)
The scalar potential reads:

1
V = ξ2 − γ 2
tanh2
(φ) , (236)
cosh4 (φ)
and it approaches a constant value V → −γ 2 ξ 2 at φ → ±∞. The solution for φ and A is
given by
2φ + sinh(2φ) = 4αξ(y − y0 ), (237)
β
A=C+ cosh(2φ), (238)
4α
where y0 and C are integration constants. Note that |φ| grows logarithmically for large |y|,
whereas A linearly goes to −∞. In particular, higher derivative terms are either vanishing
at large y, or subleading provided that ξ 2 M∗2 .
7.2. Delocalization of gravity
The discussion in the previous subsection might appear promising in the context of the
cosmological context problem. There is, however, an additional issue here, which we would
like to discuss next.
The point is that higher curvature terms appear to delocalize gravity. Indeed, let us, say,
consider adding to (1) a generic term of the form
Z
√
ζ d D x −G R k , (239)
where ζ has dimension of (mass)D−2k . We have

ek exp(−2kA) + · · · ,
Rk = R (240)
e The contribution of the
where the ellipses stand for terms containing lower powers of R.
ek term to the action then reads:
R
Z p
ζ d D−1 x dy exp[(D − 1 − 2k)A] −e gRek . (241)
Assuming that A goes to −∞ at large y, for large enough k the corresponding y integral
Z
dy exp[(D − 1 − 2k)A] (242)
diverges, which implies that gravity is no longer localized if we include such a higher
curvature term. In fact, for D = 5 delocalization of gravity generically occurs already
at the four-derivative level, that is, once we include R 2 , RMN R MN and RMNST R MNST
terms. In particular, it is not difficult to show that once we include higher curvature terms
there is no normalizable graviton zero mode, and we no longer have (D − 1)-dimensional
gravity localized on the wall, in particular, (D − 1)-dimensional Newton’s law is no longer

reproduced. 24
As we have already mentioned, in the case of runaway potentials considered in the
previous subsection we have control over higher derivative terms as far as the background
is concerned. However, what we see here is that we do not have control over the fluctuations
around such a background. The reason for this is that at large y we approach the horizon of
the AdS space where gravity is strongly coupled. It is unclear whether there is any “non-
perturbative” approach to this problem, but in any case the requirement that one reproduce
the (D − 1)-dimensional Einstein gravity at low energies seems reasonable. 25
Before we end this subsection, we would like to make the following remark. If we forget
about higher curvature terms beyond four derivatives, then for D = 5 there is an exception
to the aforementioned observation that gravity is delocalized. In particular, consider adding
the Gauss–Bonnet term to the 5-dimensional action
Z
√
ζ d 5 x −G R 2 − 4RMN R MN + RMNST R MNST . (243)
The corresponding term containing four derivatives w.r.t. the 4-dimensional coordinates
x µ then reads:
Z p 2
ζ d 4 x dy −e e − 4R
g R eµν R
eµν + Reµνσ τ R
eµνσ τ . (244)
Note that the integral over y is divergent — there is no warp factor in this expression.
However, this term is harmless as long as the four-dimensional space–time is topologically
trivial. Indeed, the Gauss–Bonnet term in four dimensions is a total derivative.
Note that if we considered adding the Gauss–Bonnet term in D > 5, then the analog
of the aforementioned term would be normalizable as the y integral is now non-trivially
weighted:
Z
dy exp[(D − 5)A]. (245)
Note, however, that if we expand around the (D − 1)-dimensional Minkowski space, the
term quadratic in e
hµν is a total derivative [53,54] (albeit the full Gauss–Bonnet term in
D − 1 > 4 dimensions no longer is). This implies that the (D − 1)-dimensional graviton
zero mode is unaffected even if D > 5.
24 Here we could ask how is gravity modified. However, generically inclusion of four-derivative terms leads to
negative-norm states in the Hilbert space of the theory thus violating unitarity.
25 Here we would like to point out a possible way out of this difficulty. If the warp factor A goes to +∞ instead
of −∞ at large y, then in, say, D = 5 terms with more than four derivatives are now normalizable. On the other
hand, one then must arrange for the lower-derivative terms to be normalizable as well, which a priori does not
seem impossible. Such a framework is interesting as here the volume of the fifth dimension is actually infinite.
As was pointed out in [44–46], in the supersymmetric context this could have important implications for the
cosmological constant problem. In particular, supersymmetry locally broken on a “brane” (e.g., the “core” of a
smooth domain wall) is not transmitted to the bulk as supersymmetry breaking in the bulk is suppressed by the
compactification volume, which is infinite. For a recent discussion of infinite volume extra dimension scenarios,
see [44,47–52].
Discontinuous domain wall solutions in the presence of the Gauss–Bonnet term were
recently studied in [55]. Smooth domain walls in this context appear to have some
interesting properties which will be discussed elsewhere [56]. 26
7.3. Comments
Before we end our discussion, we would like to make a few comments. First, the
aforementioned difficulties with the higher derivative terms might be under better control if
we consider a supersymmetric setup. However, attempts to construct non-singular smooth
supersymmetric domain walls in the context of, say, D = 5 N = 2 gauged supergravity
have not been successful so far (see, e.g., [57–60] 27 ). In particular, typically one searches
for solutions interpolating between two local AdS minima of the scalar potential. However,
as was argued in [62,63], on general grounds potentials with more then one AdS minima
are not expected to exist in D = 5 N = 2 gauged supergravity. 28
A possible way around this difficulty could be to consider potentials with, say, no local
minima at all. As we discussed in the previous sections, such potentials admit non-singular
solutions subject to certain conditions. Understanding whether such potentials can arise in
D = 5 N = 2 gauged supergravity (that is, if they avoid the general “no-go” arguments of
[62,63]) is beyond the scope of this paper. At any rate, it would be interesting if one could
construct a supersymmetric infinite tension domain wall, or else prove that such solutions
do not exist in D = 5 N = 2 gauged supergravity. 29
The second comment concerns the “self-tuning” approach of [35,36] to the cosmological
constant problem. In [66] it was argued that fine-tuning in these solutions is hidden in
singularities. In [37] it was argued that in such solutions one either has a naked singularity
or fine-tuning. Here we would like to point out that, as we have already mentioned, the
singularities in these solutions are unphysical. In particular, it does not seem consistent
to truncate the space in the y direction at such singularities. At any rate, the self-tuning
property would only imply that the four-dimensional cosmological constant is not affected
by the quantum effects on the “brane”. However, once bulk contributions are included, one
seems to lose control over the cosmological constant. In fact, a setup where pure brane
contributions to the cosmological constant vanish at least in the conformal/large N limit
has been known [67–71], but as usual there too one loses control over the cosmological
constant once the coupling to gravity is included.
26 Here we should point out the following issue arising in this context. As we have already mentioned, if we
expand the (D-dimensional) Gauss–Bonnet term around a flat background, the term quadratic in hMN is a total
derivative. In particular, it does not lead to violation of unitarity unlike a generic four-derivative action. However,
in a non-trivial warped background the linearized theory has a non-trivial propagator. Nonetheless, we expect that
if gravity is localized, then the corresponding Higgs mechanism ensures absence of propagating negative-norm
states.
27 In a recent paper [61] a construction of such a domain wall was reported. However, as was pointed out in [61],
the corresponding solution only exists if one ad hoc freezes one of the scalar fields whose dynamics is such that
the domain wall is actually singular.
28 Note, however, that such domain walls have been constructed within the framework of D = 4 N = 1
supergravity — see [64] for a review.
29 For a recent analysis of general D = 5 N = 2 gauged supergravity models, see [65].
Acknowledgements
I would like to thank Gregory Gabadadze, Martin Roček, Gary Shiu, Stefan Vandoren
and Peter van Nieuwenhuizen for discussions, and especially Gia Dvali for valuable
discussions and collaboration at some stages of this work. This work was supported in
part by the National Science Foundation. I would also like to thank Albert and Ribena Yu
for financial support.
References
[1] V. Rubakov, M. Shaposhnikov, Phys. Lett. B 125 (1983) 136.

[2] A. Barnaveli, O. Kancheli, Sov. J. Nucl. Phys. 52 (1990) 576.
[3] J. Polchinski, Phys. Rev. Lett. 75 (1995) 4724.
[4] P. Hor̆ava, E. Witten, Nucl. Phys. B 460 (1996) 506.
[5] P. Hor̆ava, E. Witten, Nucl. Phys. B 475 (1996) 94.
[6] E. Witten, Nucl. Phys. B 471 (1996) 135.
[7] J. Lykken, Phys. Rev. D 5 (1996) 3693.
[8] G. Dvali, M. Shifman, Nucl. Phys. B 504 (1997) 127.
[9] G. Dvali, M. Shifman, Phys. Lett. B 396 (1997) 64.
[10] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263.
[11] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Rev. D 59 (1999) 086004.
[12] K.R. Dienes, E. Dudas, T. Gherghetta, Phys. Lett. B 436 (1998) 55.
[13] K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 537 (1999) 47, hep-ph/9807522.
[14] Z. Kakushadze, Nucl. Phys. B 548 (1999) 205.
[16] Z. Kakushadze, T.R. Taylor, Nucl. Phys. B 562 (1999) 78.
[17] Z. Kakushadze, Phys. Lett. B 434 (1998) 269.
[19] I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257.
[20] G. Shiu, S.-H.H. Tye, Phys. Rev. D 58 (1998) 106007.
[21] Z. Kakushadze, S.-H.H. Tye, Nucl. Phys. B 548 (1999) 180.
[22] M. Visser, Nucl. Lett. B 159 (1985) 22.
[23] P. van Nieuwenhuizen, N.P. Warner, Commun. Math. Phys. 99 (1985) 141.
[24] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370.
[25] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690.
[26] M. Gogberashvili, Europhys. Lett. 49 (2000) 396, hep-ph/9812296.
[27] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, N. Kaloper, Phys. Rev. Lett. 84 (2000) 586.
[28] H. Verlinde, hep-th/9906182.
[29] B. Bajc, G. Gabadadze, hep-th/9912232.
[30] S. Weinberg, Rev. Mod. Phys. 61 (1986) 1.
[31] G. Dvali, M. Shifman, Phys. Lett. B 454 (1999) 277.
[32] O. DeWolfe, D.Z. Freedman, S.S. Gubser, A. Karch, hep-th/9909134.
[33] K. Skenderis, P.K. Townsend, Phys. Lett. B 468 (1999) 46.
[34] S.S. Gubser, hep-th/0002160.
[35] N. Arkani-Hamed, S. Dimopoulos, N. Kaloper, R. Sundrum, hep-th/0001197.
[36] S. Kachru, M. Schulz, E. Silverstein, hep-th/0001206.
[37] C. Csaki, J. Erlich, C. Grojean, T. Hollowood, hep-th/0004133.
[38] S. Kachru, M. Schulz, E. Silverstein, hep-th/0002121.
[39] T. Nihei, Phys. Lett. B 465 (1999) 81.
[40] N. Kaloper, Phys. Rev. D 60 (1999) 123506.

[41] S. Nojiri, O. Obregon, S.D. Odintsov, hep-th/0005127.
[42] J. Garriga, T. Tanaka, hep-th/9911055.
[43] S.B. Giddings, E. Katz, L. Randall, JHEP 0003 (2000) 023.
[44] G. Dvali, G. Gabadadze, M. Porrati, hep-th/0002190.
[45] E. Witten, hep-ph/0002297.
[46] G. Dvali, hep-th/0004057.
[47] R. Gregory, V.A. Rubakov, S.M. Sibiryakov, hep-th/0002072.
[48] R. Gregory, V.A. Rubakov, S.M. Sibiryakov, hep-th/0003045.
[49] C. Csaki, J. Erlich, T.J. Hollowood, hep-th/0002161.
[50] C. Csaki, J. Erlich, T.J. Hollowood, hep-th/0003020.
[53] B. Zwiebach, Phys. Lett. B 156 (1985) 315.
[54] B. Zumino, Phys. Rept. 137 (1986) 109.
[55] I. Low, A. Zee, hep-th/0004124.
[56] O. Corradini, Z. Kakushadze, to appear.
[57] K. Behrndt, M. Cvetič, hep-th/9909058.
[58] K. Behrndt, M. Cvetič, hep-th/0001159.
[59] R. Kallosh, A. Linde, hep-th/0001071.
[60] K. Behrndt, S. Gukov, hep-th/0001082.
[61] K. Behrndt, hep-th/0005185.
[62] M. Wijnholt, S. Zhukov, hep-th/9912002.
[63] G.W. Gibbons, N.D. Lambert, hep-th/0003197.
[64] M. Cvetič, H. Soleng, Phys. Rept. 282 (1997) 159.
[65] A. Ceresole, G. Dall’Agata, hep-th/0004111.
[66] S. Forste, Z. Lalak, S. Lavignac, H.P. Nilles, hep-th/0003152.
[67] S. Kachru, E. Silverstein, Phys. Rev. Lett. 80 (1998) 4855.
[68] A. Lawrence, N. Nekrasov, C. Vafa, Nucl. Phys. B 533 (1998) 199.
[69] M. Bershadsky, Z. Kakushadze, C. Vafa, Nucl. Phys. B 523 (1998) 59.
[71] P.H. Frampton, C. Vafa, hep-th/9903226.
Three point functions and the effective Lagrangian

for the chial primary fields in D = 4 supergravity
on AdS2 × S 2
Julian Lee
Institute of Physics, University of Tokyo, Komaba, Meguro-ku, Tokyo153, Japan
Received 17 May 2000; accepted 9 August 2000
Abstract
We consider the D = 4, N = 8 supergravity on AdS2 × S 2 space. We obtain the truncated
Lagrangian for the bosonic chiral primary fields, and compute the tree level three-point correlation
functions.  2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 04.65.+e; 11.15.-q; 04.50.+h

Keywords: Ads; Supergravity; Correlation functions; Compactification
1. Introduction
Among all known examples of the AdS/CFT correspondence [1–4], the least understood
is the AdS2 /CFT 1 case. The D = 1 conformal field theory (CFT), or conformal quantum
mechanics (CQM), has not been formulated and therefore no quantitative comparison
between the two sides of the duality has been made. However, there have been conjectures
[5–7] that the dual CFT is given by the n-particle, N = 4 superconformal Calogero model,
which has yet to be constructed for arbitrary n [8]. By going to the second-quantized
formulation, this becomes a (1 + 1)-dimensional field theory. (See also Refs. [9–12].)
As a first step toward investigating this conjecture we consider the supergravity side
of the duality. The low energy limit of type II(A or B) string theory on T 6 is the
N = 8 supergravity theory of Cremmer and Julia [13]. Off-shell this theory has an SO(8)
symmetry while on-shell the symmetry is enhanced to an E7(7) duality. We consider
the fluctuations of the fields around a classical configuration called Bertotti–Robinson
solution, which gives the AdS2 × S 2 spacetime. This solution is the near-horizon limit
of four D-branes intersecting over a string. The equations for the fluctuations of the fields
E-mail address: jul@hep1.c.u-tokyo.ac.jp (J. Lee).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 2 8 - 9
120 J. Lee / Nuclear Physics B 589 (2000) 119–133
were considered up to linear order in Refs. [14,15] to obtain the physical spectrum of the
theory. 1
In this paper, we compute the cubic interactions among the bosonic chiral primary fields.
Such computations were done for D = 10 IIB supergravity on AdS5 × S 5 and D = 6
supergravity on AdS3 × S 3 in Refs. [18] and [19], respectively. The computation involves
the standard elements of previous works, namely nonlinear field redefinitions, which then
lead to three point functions of chiral operators. The results have some similarities in form
with the examples in the higher dimensional AdS spaces. It would be interesting to see
whether one can make computations in a supersymmteric Calogero model, to be compared
with these results.
2. Supergravity in D = 4
In this section, we review following [14,15] the equations of motion of D = 4, N = 8

supergravity. We mainly follow the four dimensional formalism and notation of Ref. [14].
We are interested in the equation of motion for the metric gµ̂ν̂ , vectors Bµ̂AB , and scalars
WABCD , where the capital roman letters are SU(8) indices. The scalars are packaged in a
E7(7) matrix V parametrizing the coset E7(7) /SU(8) [13],
!
−1
Q[C δ D] Pµ̂ABCD
µ̂[A B]
∂µ̂ V V = . (1)
P̄µ̂ABCD Q̄[A δ B]
µ̂[C D]
P ’s parametrize the coset manifold and can be expressed in terms of the scalar fields W ,
and Q’s are the SU(8) gauge fields, but for our purposes it is enough to note that the SU(8)
gauge symmetry can be fixed to the so called symmetric gauge where V = exp(X),

0 WABCD
X = S ABCD (2)
W 0
and WABCD is complex, completely antisymmetric in A, B, C, D, and satisfies the
constraint
WS ABCD = 1 ABCDEF GH WEF GH . (3)

24
We will use the following notation for indices: µ̂, ν̂ = 0, 1, 2, 3 are D = 4 coordinates,
λ, µ, ν, . . . = 0, 1 are AdS2 coordinates and α, β, γ , . . . = 2, 3 are S 2 coordinates.
The bosonic part of the D = 4 supergravity action is:

1 1 µ̂ν̂ 1
L= eR(ω, e) + eFµ̂MNν̂ (B)H̃ MN (B, V ) − eP µ̂ABCD P̄ µ̂ABCD
, (4)
4 8 24
where
ν̂ = 2∂[µ̂ Bν̂]
Fµ̂AB AB
(5)
1 See also Refs. [16,17] for the spectrum of the minimal D = 4, N = 2 supergravity on AdS × S 2 and the
2
D = 10, IIA supergravity on “quasi” AdS2 × S 8 , respectively.
J. Lee / Nuclear Physics B 589 (2000) 119–133 121
µ̂ν̂
H̃MN are defined in the Appendix A. For our purpose, we only need the leading expansion
of the H̃ and P in W ABCD :

GMN H̃ (B) = −GMN 1 + W + W S + W2 + W S2
µ̂ν̂ µ̂ν̂MN µ̂ν̂

1 S 1S S S3
− WW W− W WW + W3 + W Gµ̂ν̂P Q
3 3 MNP Q

+ iGMN S S2
µ̂ν̂ W − W + W − W + O W G̃µ̂ν̂P Q + O W 4 ,
2 3
MNP Q

Pµ̂ABCD = ∇µ̂ WABCD + O W 3 . (6)
3. The equations of motion for the bosonic chiral primary fields
We consider the classical configuration called Bertotti–Robinson solution [20,21],

1
ds 2 = − dx02 + dz2 + dΩ22 ,
z2
Rµλνσ = −(gµν gλσ − gµσ gλν ), Rαγβδ = (gαβ gγ δ − gαδ gγβ ),
12
F̄αβ = αβ , AB
F̄αβ =0 (A 6= 1, 2),
WABCD = 0, (7)
which breaks the SU(8) internal symmetry into SU(6) × SU(2). The bulk fields of interest
are the fluctuations about this background,
gµ̂ν̂ = ḡµ̂ν̂ + hµ̂ν̂ ,

ν̂ = F̄
Fµ̂AB + 2∇[µ̂ bν̂]
ABµ̂ν̂ AB
(8)
and WABCD . One can consider the classical equations for these fluctuations to linear order
and organize the spectrum into the multiplets of SU(6) × SU(2) [14,15]. In this paper we
expand the equations of motion up to the second order in these fluctuations to obtain the
interaction terms.
To find the chiral primary fields on AdS2 we expand the fields in spherical harmonics on
S 2 . The expansions are quite simple in this case as all harmonic functions on the 2-sphere
can be expressed in terms of just the scalar spherical harmonics Ylm . l is the quantum
number labeling the Casimir of the representation,
∇α ∇ α Ylm = −l(l + 1)Ylm . (9)
The expansions of the bosonic fluctuations are then given by (denoting the l, m indices
collectively by I )
X
hµν = HµνI
YI , (10)
X
hµα = I
B1µ ∇α YI + B2µ
I
eαβ ∇ β YI , (11)
X
hαβ = φ1I ∇α ∇β YI + φ2I e(α γ ∇β) ∇γ YI + φ3I gαβ YI , (12)
X
bµAB = bµ(I )AB YI , (13)
X
bαAB = b1(I )AB ∇α YI + b2(I )AB eαβ ∇ β YI , (14)
X
WABCD = I
WABCD YI . (15)
Before substituting the expansions into the equations of motion we can first simplify the
expansions by fixing gauge symmetries by imposing 2
φ1I = φ2I = B1µ

I
= b1(I )AB = 0. (16)
At the linearized level, the bosonic fields can be decomposed into the eigenstates of the
AdS2 Laplacian [14,15],
φ I = 2lT I + 2(l − 1)T̃ I ,

b(I )[12] = T I − T̃ I ,
a (I )[12] ≡ µν ∇µ bν(I )[12] = (l + 1)S I + l S̃ I ,
B I ≡ µν ∇µ BνI = −2S I + 2S̃ I ,
a (I )[MN] ≡ µν ∇µ bν(I )[MN] = (l − 1)U (I )[MN] + (l + 2)Ũ (I )[MN] (M, N 6= 1, 2),
(W − W S )(I )12MN = i∇x2 (U (I )[MN] − Ũ (I )[MN] ),
b(I )[MN] = V (I )[MN] + Ṽ (I )[MN] (M, N 6= 1, 2),
(W + W S )(I )12MN = −lV (I )[MN] + (l + 1)Ṽ (I )[MN] , (17)
where these fields satisify the equations of the form:

∇x2 − l(l − 1) AI + QIA = 0, (18)

∇x2 − (l + 1)(l + 2) ÃI + Q̃IÃ = 0. (19)
Here, A and Ã stand for T , S, U, V and T̃ , S̃, Ũ , Ṽ , respectively, and QA and QÃ are the
second order corrections. Also, we note that the fields Hµν are not independent degrees of
freedom and they are completely determined by the equations [14–16]
I
∇x2 + 2 − l(l + 1) Hµν − 2∇(µ ∇ λ Hν)λI
+ ∇µ ∇ν − gµν ∇x2 + 1 − l(l + 1) H I

+gµν ∇ λ ∇ ρ Hλρ
I
+ 2 ∇µ ∇ν − gµν ∇x2 − 1 − 12 l(l + 1) φ I
= 4gµν l(l + 1)bI + (higher order), (20)

∇ ν Hµν
I
− ∇µ H I − ∇µ φ I = −4∇µ bI + (higher order), (21)

∇x2 + 4 φ I + ∇x2 − 1 − l(l + 1) H I − ∇ µ ∇ ν Hµν
I
= 4l(l + 1)bI + (higher order), (22)

H = (quadratic order).
I
(23)
2 Since we project out only the physical modes, actually it is enough to impose φ = φ = 0. From now on, we
1 2
write φ3 → φ, B2µ → Bµ , b2 → b.
By making an ansatz, one can easily find the solution to the equations above:
1
I
Hµν = −2l(l − 1)gµν T I + 4∇µ ∇ν T I + (higher order corrections). (24)
l+1
Since we are interested only in the three-point function of chiral primary fields, only
QT , QS , QU , QV are of interest. We also put any non-chiral primary fields appearing in
Q’s to be zero, and substitute (24) for Hµν , neglecting the higher order corrections. The
detailed form of the Q’s are rather complicated and not interesting at this stage. They are
of the generic form:
QA = α∇ µ ∇ν B∇µ ∇ν C + β∇ µ B∇µ C + γ BC + · · · , (25)
where A, B, C are chiral primary fields. We see that there are terms involving derivatives
of the fields in Q’s. These terms can be removed by nonlinear redefinitions of the fields,
which does not change the equations (19) at the linear level. It is easy to see that this can
be done by redefining [18,19]
α 1
A→A− (∇B) · (∇C) − α + (1 + Γ )β B · C, (26)
2c 2c
where Γ ≡ 12 (l(l − 1) − l1 (l1 − 1) − l2 (l2 − 1)) with l, l1 , l2 being the total angular
momentum quantum numbers of A, B, C, respectively. After these redefinitions, the linear
term

∇x2 − l(l − 1) A (27)
generates an additional term which removes the derivative terms, and QA becomes:

QA = γ + Γ (β + (1 + Γ )α) BC · · · . (28)
We then get
(l(l 2 − 1) + l1 (l12 − 1) + l2 (l22 − 1))
QT = αα1 α2 (Σ 2 − 1)C̃(I ; I1 , I2 )T 2
2l(2l + 1)(l − 1)(l1 + 1)(l2 + 1)
(l(l 2 − 1) − l1 (l12 − 1) − l2 (l22 − 1))C̃(I ; I1 , I2 )
− αα1 α2 (Σ 2 − 1)S 2
2l(2l + 1)(l − 1)
(l1 + l2 )(l1 − 1)(l2 − 1)
+ αα1 α2 (Σ 2 − 1)C̃(I ; I1 , I2 )U AB UAB
4l(2l + 1)(l − 1)
(l1 + l2 )
+ αα1 α2 (Σ 2 − 1)C̃(I ; I1 , I2 )V AB VAB , (29)
4l(2l + 1)(l − 1)
(l(l 2 − 1) + l1 (l12 − 1) − l2 (l22 − 1))

QS = αα1 α2 (Σ 2 − 1)C̃(I ; I1 , I2 )S (1) T (2)
2l(l 2 − 1)(2l + 1)(l2 + 1)
(l1 − l2 )(l2 − 1)
− αα1 α2 (Σ 2 − 1)C̃(I ; I1 , I2 )V (1) U (2) , (30)
4l(l 2 − 1)(2l + 1)
(l − l2 )
U =
QAB αα1 α2 (Σ 2 − 1)C̃(I ; I1 , I2 )S (1) V (2)AB
2l(l − 1)(2l + 1)
l1 − 1 (1) (2)
− αα1 α2 (Σ 2 − 1)C̃(I ; I1 , I2 ) 12ABCDEF UCD VEF
8l(l − 1)(2l + 1)
(l + l1 )(l1 − 1)
+ αα1 α2 (Σ 2 − 1)C̃(I ; I1 , I2 )U (1)AB T (2) , (31)
2l(l − 1)(2l + 1)(1 + l2 )
(l2 − l)(l2 − 1)
V =
QAB αα1 α2 (Σ 2 − 1)C̃(I ; I1 , I2 )S (1) U (2)AB
2l(2l + 1)
(l1 − 1)(l2 − 1)
− αα1 α2 (Σ 2 − 1)C̃(I ; I1 , I2 ) 12ABCDEF UCD UEF
16l(2l + 1)
1
+ αα1 α2 (Σ 2 − 1)C̃(I ; I1 , I2 ) 12ABCDEF VCD VEF
16l(2l + 1)
(l + l1 )
+ αα1 α2 (Σ 2 − 1)C̃(I ; I1 , I2 )V (1)AB T (2), (32)
2l(2l + 1)(1 + l2 )
at the quadratic level, where α ≡ l1 +l2 −l, α1 ≡ l +l2 −l1 , α2 ≡ l +l1 −l2 , Σ ≡ l +l1 +l2 ,
and
Z
C̃(I ; I1 , I2 ) ≡ YI∗ YI1 YI2 . (33)
We also used the abbreviation A(i) ≡ AIi and A2 ≡ AI1 AI2 for any field A.
4. The effective action and the three-point function
The equations in the previous section can be derived from the truncated Lagrangian

1 1 MN µ̂ν̂ 1
L= eR(ω, e) + eFµ̂ν̂ (B)H̃MN (B, V ) − ePµ̂ABCD P̄ µ̂ABCD
4 8 24

(2l + 1)l l − 1
2 (2l + 1)l l 2 − 1
= T ∇x − l(l − 1) T +
2
S ∇x2 − l(l − 1) S
4 2
l(2l + 1) AB 2 l(l − 1)2 (2l + 1) AB 2
+ V ∇x − l(l − 1) VAB + U ∇x − l(l − 1) UAB
4 4
l1 l12 − 1 + l2 l22 − 1 + l3 l32 − 1
+ α1 α2 α3 (Σ 2 − 1)C(I1 , I2 , I3 )T 3
3(l1 + 1)(l2 + 1)(l3 + 1)

l1 l12 − 1 + l2 l22 − 1 − l3 l32 − 1
+ α1 α2 α3 (Σ 2 − 1)C(I1 , I2 , I3 )S (1) S (2) T (3)
(l3 + 1)
(2) (3)
− (l1 − l2 )(l2 − 1)α1 α2 α3 (Σ 2 − 1)C(I1 , I2 , I3 )V (1)AB UAB S
1 (1) (2) (3)
+ α1 α2 α3 (Σ 2 − 1)C(I1 , I2 , I3 ) 12ABCDEF VAB VCD VEF
24
l1 + l2 (2)
+ α1 α2 α3 (Σ 2 − 1)C(I1 , I2 , I3 )V (1)AB VAB T (3)
2(l3 + 1)
(l1 − 1)(l2 − 1) (1) (2) (3)
− α1 α2 α3 (Σ 2 − 1)C(I1 , I2 , I3 ) 12ABCDEF UAB UCD VEF
8
(l1 − 1)(l2 − 1)(l1 + l2 ) (2) (3)
+ α1 α2 α3 (Σ 2 − 1)C(I1 , I2 , I3 )U (1)AB UAB T . (34)
2(l3 + 1)
with
Z
C(I1 , I2 , I3 ) ≡ Y I1 Y I2 Y I3 . (35)
Here the normalizations were fixed by directly substituting the expression (17) into the
lagrangian (4) and evaluating the leading terms for some fields.
To compute from (34) the 2- and 3-point functions of chiral primary operators of the
boundary theory, we apply the formulae derived, for instance, in Ref. [22]. From Eq. (17)
and the correction factor in Eq. (95) of Ref. [22], we read off the tree-level two-point
functions to be 3

(2l + 1)(l 2 − 1) 1 Γ (l + 1) δ I1 I2
T I1 (x)T I2 (y) = (2l − 1) ,
2 π 1/2 Γ (l − 1/2) |x − y|2l

I 1 Γ (l + 1) δ I1 I2
S 1 (x)S I2 (y) = (2l + 1)(l 2 − 1) 1/2 (2l − 1) ,
π Γ (l − 1/2) |x − y|2l

(I )AB 1 Γ (l + 1)
U 1 (x)U (I2 )CD (y) = (l − 1)2 (2l + 1) 1/2
π Γ (l − 1/2)
(δAC δBD − δAD δBC )δ I1 I2
× (2l − 1) ,
|x − y|2l

(I )AB 1 Γ (l + 1)
V 1 (x)V (I2 )CD (y) = (2l + 1) 1/2
π Γ (l − 1/2)
(δAC δBD − δAD δBC )δ I1 I2
× (2l − 1) . (36)
|x − y|2l
From Eq. (25) of the same paper we derive that

(1)
T (x)T (2) (y)T (3) (z)
Q α
25 i Γ 2i + 1 Γ 12 Σ + 32 l1 l12 − 1 + l2 l22 − 1 + l3 l32 − 1
=− Q3
i Γ (li − 1/2)(li + 1) |x − y| |y − z| |z − x|
π α3 α1 α2
×C(I1 , I2 , I3 ),

S (1) S (2) T (3)
Q α
25 i Γ 2i + 1 Γ 12 Σ + 32 l1 l12 − 1 + l2 l22 − 1 + l3 l32 − 1
=− Q3
i Γ (li − 1/2) (l3 + 1)|x − y| |y − z| |z − x|
π α3 α1 α2
×C(I1 , I2 , I3 ),
Q 1

25
αi
i Γ 2 + 1 Γ 2 Σ + 2 (l1 − l2 )(l2 − 1)
3
V (1)AB
U (2)CD (3)
S = Q3
π i Γ (li − 1/2) |x − y|α3 |y − z|α1 |z − x|α2

× δ AC δ BD − δ AD δ BC C(I1 , I2 , I3 ),
3 Although the following results are the correlation functions of the chiral primary operators of the boundary
theory which couple to the bulk fields S, T , U, V we will still denote them by S, T , U, V for the simplicity of the
notations.
Q

(1) (2) (3) 2 5
i Γ αi
+1 Γ 1
2Σ + 3
VAB VCD VEF = − Q3 2 2
Γ (li − 1/2) |x − y|α3 |y
π i − z|α1 |z − x|α2
× C(I1 , I2 , I3 )12ABCDEF ,
Q

(1) (2) 2 6 (l1 + l2 ) i Γ αi
+1 Γ 1
2Σ +
(δAC δBD − δAD δBC )
3
VAB VCD T (3) = − Q3
2
2
(l3 + 1)
π i Γ (li − 1/2) |x − y| 3 |y − z|α1 |z − x|α2
α
× C(I1 , I2 , I3 ),
Q

(1) (2) (3)
α
25 (l1 − 1)(l2 − 1) i Γ 2i + 1 Γ 12 Σ + 32
UAB UCD VEF = Q3
i Γ (li − 1/2) |x − y| |y − z| |z − x|
π α3 α1 α2
× C(I1 , I2 , I3 )12ABCDEF ,
Q

α
26 (l1 − 1)(l2 − 1)(l1 + l2 ) i Γ 2i + 1 Γ 12 Σ + 32
U (1)AB
U (2)CD
T (3)
=− Q
π (l3 + 1) 3i Γ (li − 1/2) |x − y|α3 |y − z|α1 |z − x|α2

× δ AC δ BD − δ AD δ BC C(I1 , I2 , I3 ). (37)
The normalizations of the fields can be fixed by demanding that the two point functions are

δ I1 I2
A(I1 ) (x)A(I2 ) (y) = (38)
|x − y|2l
for any two chiral primary fields AI . After rescaling the fields in order to satisfy this
condition, the normalized three-point function are given by:
Q 1
(1) 27/2
αi
i Γ 2 + 1 Γ 2Σ + 2
3
T (x)T (y)T (z) = − 1/4 qQ
(2) (3)

π 3
Γ (l + 3/2)Γ (l + 2)(l + 1) l 2 − 1
i i i i i

l1 l12 − 1 + l2 l22 − 1 + l3 l32 − 1
× C(I1 , I2 , I3 ),
|x − y|α3 |y − z|α1 |z − x|α2
Q 1
(1) (2) (3) 25/2

αi
i Γ 2 + 1 Γ 2Σ + 2
3
S S T = − 1/4 qQ
π 3
i Γ (li + 3/2)Γ (li + 1) li − 1 (l3 + 1)
2

l1 l12 − 1 + l2 l22 − 1 + l3 l32 − 1
× C(I1 , I2 , I3 ),
|x − y|α3 |y − z|α1 |z − x|α2

V (1)AB U (2)CD S (3)
Q αi 1
i Γ 2 + 1 Γ 2 Σ + 2 (l1 − l2 )C(I1 , I2 , I3 ) δ
3 AC δ BD − δ AD δ BC
22
= 1/4 qQ ,
π 3
Γ (l + 3/2)Γ (l + 1) l 2 − 1 |x − y|α3 |y − z|α1 |z − x|α2
i i i 3
Q
(1) (2) (3) 22

αi
iΓ 2 +1 Γ
2 Σ + 2 C(I1 , I2 , I3 )12ABCDEF
1 3
VAB VCD VEF = − q ,
π 1/4 Q3 Γ (l + 3/2)Γ (l + 1) |x − y|α3 |y − z|α1 |z − x|α2
i i i
Q
(1) (2) (3) 27/2 (l1 + l2 ) i Γ α2i + 1 Γ 12 Σ + 32

VAB VCD T = − 1/4 qQ 2
π (l3 + 1) 3
i Γ (li + 3/2)Γ (li + 1) l3 − 1
(δAC δBD − δAD δBC )C(I1 , I2 , I3 )
× ,
|x − y|α3 |y − z|α1 |z − x|α2
Q 1
(1) (2) (3) 22

αi
i Γ 2 + 1 Γ 2 Σ + 2 C(I1 , I2 , I3 )12ABCDEF
3
UAB UCD VEF = 1/4 q Q ,
π 3
i Γ (li + 3/2)Γ (li + 1) |x − y| |y − z| |z − x|
α3 α1 α2
Q
(1)AB (2)CD (3) 27/2 (l1 + l2 ) i Γ α2i + 1 Γ 12 Σ + 32

U U T = − 1/4 q
π (l + 1) Q3 Γ (l + 3/2)Γ (l + 1) l 2 − 1
3 i i i 3
(δ AC δ BD − δ AD δ BC )C(I1 , I2 , I3 )
× . (39)
|x − y|α3 |y − z|α1 |z − x|α2
5. Conclusions
In this paper, we gave the calculation of three point interactions of chiral primaries in 4D
supergravity. In order to remove the derivatives from the interactions, we used nontrivial
redefinitions of the fields which solved the linear equation of motion for chiral primaries.
It is after removing the derivative terms that we deal with a standard field theory in two
dimensions. These redefinitions were also needed for the cases of higher dimensional AdS
spaces [18,19]. Deeper reasons behind these redefinitions were discussed in Ref. [23,24].
In Section 4, we derived the three point interactions. The factors appear which are similar
to the results in higher dimensional AdS spacetimes. In order to make a useful statement
on AdS2 /CFT 1 correspondence, similar computation has to be made in the boundary
conformal quantum mechanics dual to this theory. There has been a conjecture that this
dual quantum mechanics is given by a supersymmetric extension of the Calogero model [5–
7]. It would be interesting to see whether this is true. However, since a many-body quantum
mechanics becomes a (1 + 1)-dimensional field theory after the second quantization, one
might directly obtain this field theory starting from the Lagrangian (34) by appropriate
truncation. This interesting issue, along with the evaluation of the interaction terms for the
fermions, are left for future studies.
Acknowledgements
I give special thanks to Antal Jevicki for many helpful discussions and suggestions. I also
thank Tamiaki Yoneya and Sangmin Lee for useful discussions. This work was supported
by JSPS through Institute of Physics, University of Tokyo.
Appendix A. Vector and scalar part of the Lagrangian
As mentioned in the text, the SU(8) gauge symmetry can be fixed to the so called
symmetric gauge where V = exp(X) and

0 W
X = S ABCD ABCD . (A.1)
W 0
Expanding in W , we get the expressions
!
S EF CD + O(W 4 )
δAB CD + 12 WABEF W WABCD + O(W 3 )
V= ,
S ABCD + O(W 3 )
W S ABEF WEF CD + O(W 4 )
δ AB CD + 12 W
(A.2)
and
!
S EF CD + O(W 4 )
δAB CD + 12 WABEF W −WABCD + O(W 3 )
−1
V = .
S ABCD + O(W 3 )
−W S ABEF WEF CD + O(W 4 )
δ AB CD + 12 W
(A.3)
From these expressions one easily gets
!
S S
2 ∇µ (W W ) − (∇µ W )W + O(W ) ∇µ W + O(W 3 )
1 4
∂µ̂ V V −1 = .
∇µ WS + O(W 3 ) S S
2 ∇µ (W W ) − (∇µ W )W + O(W )
1 4
(A.4)
Comparing with Eq. (1), we see that

Pµ̂ABCD = ∇µ̂ WABCD + O W 3 , P̄ µ̂ABCD = ∇µ̂ W ABCD + O W 3 . (A.5)
H̃ (B, V ) is defined by the equation

−iGµ̂ν̂ − Hµ̂ν̂ −1 iG̃µ̂ν̂ − H̃µ̂ν̂
= (V V )
†
, (A.6)
iGµ̂ν̂ − Hµ̂ν̂ −iG̃µ̂ν̂ − H̃µ̂ν̂
where it is to be understood that the matrices are multiplied by contracting the SU(8)
indices, which I did not write explicitly. 4 G̃, H̃ denotes the dual fields,
1 1
G̃µ̂ν̂ = µ̂ν̂ ρ̂ σ̂ Gρ̂ σ̂ , H̃ µ̂ν̂ = µ̂ν̂ ρ̂ σ̂ Hρ̂ σ̂ . (A.7)
2 2
Expanding in W , we get
! !
Gµ̂ν̂ + iHµ̂ν̂ 1 + 2W W S + O(W 4 ) S W + O(W 5 )
−2W − 43 W W
=
Gµ̂ν̂ − iHµ̂ν̂ S − 4W
−2W S S S W + O(W 4 )
3 W W + O(W ) 1 + 2W
5
!
G̃µ̂ν̂ − iH̃µ̂ν̂
× . (A.8)
G̃µ̂ν̂ + iH̃µ̂ν̂
By adding the first and second row, we get
4 In this paper, I use only SU(8) indices, in contrast to Ref. [14] where E indices were also used.
7

G=− 1−W −W S + WW S +W SW − 2 W
SW WS − 2WW
S W + O W 5 H̃
3 3

S + W + WW S −W 2
SW − W SW W 2
S + WW S W G̃.
+ i −W (A.9)
3 3
Solving in terms of H , we finally get

GAB H̃ (B) = −GAB 1 + W + W S + W2 + W
S2
µ̂ν̂ µ̂ν̂AB µ̂ν̂

1 S 1S S S3
− WW W− W WW + W3 + W Gµ̂ν̂CD
3 3 ABCD

+ iGAB W − S + W2 − W
W S 2 + O(W 3 ) G̃µ̂ν̂CD + O W 4 .
µ̂ν̂ ABCD
(A.10)
Appendix B. Spherical harmonics
We list some formulas about spherical harmonics needed for the calculations in the
text. The spherical hamonics on S 2 are very simple in that there are only scalar spherical
harmonics, which we denote by YI = Ylm . They are normalized so that
Z
Yl1 m1 Yl2 m2 = δl1 l2 δm1 m2 . (B.1)
The explicit form is given by

s
1 (2l + 1)(l − m) imφ m
Ylm ≡ e Pl (cos θ), (B.2)
2 π(l + m)
where Plm (x) is the associated Legendre polynomial
(−1)m m/2 dm+l 2 l
Plm (x) ≡ l
1 − x2 m+l
x −1 , (B.3)
2 l! dx
although this explicit form was not really used in the text. More important is the formula
for the integral of three spherical harmonics with derivatives, expressed in terms of
C(I1 , I2 , I3 ). They are as follows 5 :
Z
A(1; 2, 3) ≡ Y1 ∇Y2 ∇Y3
1
= l2 (l2 + 1) + l3 (l3 + 1) − l1 (l1 + 1) C(1, 2, 3), (B.4)
Z2
B(1; 2, 3) ≡ Y1 ∇α ∇β Y2 ∇ α ∇ β Y3
1
= l1 (l1 + 1) − l2 (l2 + 1) − l3 (l3 + 1)
4
× 2 + l1 (l1 + 1) − l2 (l2 + 1) − l3 (l3 + 1) C(1, 2, 3), (B.5)
5 Again, we use the abbreviation i for I ≡ (l , m ).

i i i
Z
D(1; 2, 3) ≡ ∇α ∇β Y1 ∇ α Y2 ∇ β Y3
1
= l3 (l3 + 1) + l1 (l1 + 1) − l2 (l2 + 1)
4
× l3 (l3 + 1) + l2 (l2 + 1) − l1 (l1 + 1) C(1, 2, 3), (B.6)
Z
G(1, 2, 3) ≡ ∇α ∇β Y1 ∇β ∇γ Y2 ∇ γ ∇ α Y3
1
= −l13 (l1 + 1)3 − l23 (l2 + 1)3 − l33 (l3 + 1)3
8
− 2l12 (l1 + 1)2 − 2l22 (l2 + 1)2 − 2l32 (l3 + 1)2
+ 4l1 l2 (l1 + 1)(l2 + 1) + 4l2 l3 (l2 + 1)(l3 + 1) + 4l3 l1 (l3 + 1)(l1 + 1)
+ l12 l2 (l1 + 1)2 (l2 + 1) + l1 l22 (l1 + 1)(l2 + 1)2 + l22 l3 (l2 + 1)2 (l3 + 1)
+ l2 l32 (l2 + 1)(l3 + 1)2 + l32 l1 (l3 + 1)2 (l1 + 1) + l3 l12 (l3 + 1)(l1 + 1)2

− 2l1 l2 l3 (l1 + 1)(l2 + 1)(l3 + 1) . (B.7)
Proof.
(a) By integrating by parts, one gets
Z Z
A(1; 2, 3) = − Y1 ∇ Y2 Y3 − ∇α Y1 ∇ α Y2 Y3
2
= l2 (l2 + 1)C(1, 2, 3) − A(3; 1, 2). (B.8)
By permuting the indices and adding the resulting equations, we get Eq. (B.4).
(b) By integrating by parts, we get

Z
B(1; 2, 3) = Y1 ∇α ∇β Y2 ∇ α ∇ β Y3
Z Z
= − ∇α Y1 ∇ α ∇ β Y2 ∇β Y3 − Y1 ∇α ∇β ∇ α Y2 ∇ β Y3
Z
= ∇α Y1 ∇ α Y2 ∇y2 Y3 + ∇β ∇α Y1 ∇ α Y2 ∇ β Y3
− Y1 ∇ γ Y2 ∇ β Y3 Rβγ − Y1 ∇ β ∇y2 Y2 ∇β Y3
= −l3 (l3 + 1)∇α Y1 ∇ α Y2 Y3 − ∇β ∇α ∇ β Y1 ∇ α Y2 Y3 − ∇β ∇α Y1 ∇ β ∇ α Y2 Y3
− Y1 ∇ β Y2 ∇β Y3 + l2 (l2 + 1)Y1 ∇ β Y2 ∇β Y3
= −l3 (l3 + 1)A(3; 1, 2) − Rαγ ∇ α Y1 ∇ γ Y2 Y3
− ∇α ∇y2 Y1 ∇ α Y2 Y3 − B(3; 1, 2)

+ A(1; 2, 3) l2 (l2 + 1) − 1

= l1 (l1 + 1) − l3 (l3 + 1) − 1 A(3; 1, 2) − B(3; 1, 2)

+ A(1; 2, 3) l2 (l2 + 1) − 1 (B.9)
which gives

B(1; 2, 3) + B(3; 1, 2) = l2 (l2 + 1) − 1 A(1; 2, 3)

+ l1 (l1 + 1) − l3 (l3 + 1) − 1 A(3; 1, 2). (B.10)
On the other hand, we have

Z Z
B(1; 2, 3) = − ∇ α Y1 ∇α ∇β Y2 ∇ β Y3 − Y1 ∇ α ∇β ∇α Y2 ∇ β Y3
Z Z
= ∇ ∇ Y1 ∇α ∇β Y2 Y3 + ∇ α Y1 ∇ β ∇α ∇β Y2 Y3
α β
Z
− Rβγ Y1 ∇ γ Y2 ∇ β Y3 − Y1 ∇ β ∇y2 Y2 ∇β Y3
= B(3; 1, 2) + Rαγ ∇ α Y1 ∇ γ Y2 Y3
+ ∇α Y1 ∇ α ∇y2 Y2 Y3 − Y1 ∇ β Y2 ∇β Y3 + l2 (l2 + 1)Y1 ∇β Y2 ∇ β Y3
= B(3; 1, 2) + ∇ α Y1 ∇α Y2 Y3
− l2 (l2 + 1)∇α Y1 ∇ α Y2 Y3 − Y1 ∇ β Y2 ∇β Y3 + l2 (l2 + 1)Y1 ∇β Y2 ∇ β Y3

= B(3; 1, 2) + 1 − l2 (l2 + 1) A(3; 1, 2) − A(1; 2, 3) (B.11)
which is

B(1; 2, 3) − B(3; 1, 2) = 1 − l2 (l2 + 1) A(3; 1, 2) − A(1; 2, 3) . (B.12)
Adding (B.10) and (B.12) gives Eq. (B.5).
(c)
Z
D(1; 2, 3) = ∇α ∇β Y1 ∇ α Y2 ∇ β Y3
Z
= − ∇β Y1 ∇y2 Y2 ∇ β Y3 − ∇ β Y1 ∇ α Y2 ∇α ∇β Y3
Z
= l2 (l2 + 1) ∇β Y1 Y2 ∇ β Y3
+ Y1 ∇β ∇α Y2 ∇ β ∇ α Y3 + Y1 ∇α Y2 ∇ β ∇ α ∇β Y3
Z
= l2 (l2 + 1) ∇β Y1 Y2 ∇ β Y3 + Y1 ∇β ∇α Y2 ∇ β ∇ α Y3
+ Rαγ Y1 ∇ α Y2 ∇ γ Y3 + Y1 ∇α Y2 ∇ α ∇y2 Y3

= l2 (l2 + 1)A(2; 3, 1) + B(1; 2, 3) + 1 − l3 (l3 + 1) A(1; 2, 3) (B.13)
which is equivalent to Eq. (B.6).
(d)
Z
G(1, 2, 3) = ∇α ∇β Y1 ∇β ∇γ Y2 ∇ γ ∇ α Y3
Z
= − ∇γ ∇α Y1 ∇ α ∇ β ∇ γ Y2 ∇β Y3
Z

− 1 − l1 (l1 + 1) ∇γ Y1 ∇ β ∇ γ Y2 ∇β Y3
Z
δ
= − ∇γ ∇α Y1 Rα β γ ∇δ Y2 + ∇ β ∇α ∇ γ Y2 ∇β Y3
Z

+ l1 (l1 + 1) − 1 ∇γ Y1 ∇ β ∇ γ Y2 ∇β Y3
Z

= l1 (l1 + 1) − 1 D(2; 3, 1) − ∇y2 Y1 ∇β Y2 ∇ β Y3
Z Z
+ ∇ β ∇ α Y1 ∇α Y2 ∇β Y3 − ∇γ ∇α Y1 ∇ β ∇α ∇ γ Y2 ∇β Y3

= l1 (l1 + 1) − 1 D(2; 3, 1) + l1 (l1 + 1)A(1; 2, 3) + D(1; 2, 3)
Z
− l3 (l3 + 1)B(3; 1, 2) + ∇ β ∇γ ∇α Y1 ∇ α ∇ γ Y2 ∇β Y3

= l1 (l1 + 1) − 1 D(2; 3, 1) + l1 (l1 + 1)A(1; 2, 3) + D(1; 2, 3)
Z

− l3 (l3 + 1)B(3; 1, 2) + R β αγ δ ∇δ Y1 + ∇α ∇ β ∇γ Y1 ∇ α ∇ γ Y2 ∇β Y3
= l1 (l1 + 1)D(2; 3, 1) + l1 (l1 + 1)A(1; 2, 3) − l3 (l3 + 1)B(3; 1, 2)
+ l2 (l2 + 1)A(2; 3, 1) + l2 (l2 + 1)D(1; 2, 3) (B.14)
which is Eq. (B.7).
References
[1] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231.

[2] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105.
[3] E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253.
[4] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Phys. Rep. 323 (2000) 183.
[5] G. Gibbons, P.K. Townsend, Phys. Lett. B 454 (1999) 187.
[6] P.K. Townsend, The M(atrix) Model/AdS2 correspondence, in: Proceedings of the 3rd Puri
workshop on Quantum Field Theory, Quantum Gravity and Strings, Puri, Orissa, India,
December 9–19, 1998, hep-th/9903043.
[7] A. Jevicki, T. Yoneya, Nucl. Phys. B 411 (1994) 64.
[8] V. Akulov, M. Kudinov, Phys. Lett. B 460 (1999) 365.
[9] A. Strominger, JHEP 9901 (1999) 007.
[10] J. Maldacena, J. Michelson, A. Strominger, JHEP 9901 (1999) 007.
[11] T. Nakatsu, N. Yokoi, Mod. Phys. Lett. A 14 (1999) 147.
[12] M. Cadoni, S. Mignemi, Nucl. Phys. B 557 (1999) 165.
[13] E. Cremmer, B. Julia, Nucl. Phys. B 159 (1979) 141.
[14] S. Corley, JHEP 9909 (1999) 001.
[15] J. Lee, S. Lee, Nucl. Phys. B 563 (1999) 125.
[16] J. Michelson, M. Spradlin, JHEP 9909 (1999) 029.
[17] Y. Sekino, T. Yoneya, Nucl. Phys. B 570 (2000) 1307.
[18] S. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Adv. Theor. Math. Phys. 2 (1998) 697.
[19] M. Miahilescu, JHEP 0002 (2000) 007.
[20] B. Bertotti, Phys. Rev. 116 (1959) 1331.
[21] I. Robinson, Bull. Acad. Polon. Sci. 7 (1959) 351.
[22] D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Nucl. Phys. B 546 (1999) 96.
[23] H. Nastase, D. Vaman, The AdS–CFT correspondence, consistent truncations and gauge
invariance, hep-th/0004123.
[24] H. Nastase, D. Vaman, P. van Nieuwenhuizen, Phys. Lett. B 469 (1999) 96.
String theory on AdS3 as discrete light-cone

Liouville theory
Yasuaki Hikida a , Kazuo Hosomichi b , Yuji Sugawara a,∗
a Department of Physics, Faculty of Science, University of Tokyo Bunkyo-ku,
Hongo 7-3-1, Tokyo113-0033, Japan
b Yukawa Institute for Theoretical Physics, Kyoto University, Kyoto 606-8502, Japan
Received 11 May 2000; revised 11 July 2000; accepted 11 August 2000
Abstract
We investigate (super) string theory on AdS3 background based on an approach of free field
realization. We demonstrate that this string theory can be reformulated as a string theory defined
on a linear dilaton background along the transverse direction (“Liouville mode”) and compactified
onto S 1 along a light-like direction.
Under this reformulation we analyze the physical spectrum as that of a free field system, and
discuss the consequences when we turn on the Liouville potential. The discrete light-cone momentum
in our framework is naturally interpreted as the “winding number” of the long string configuration
and is identified with the spectral flow parameter that is introduced in the recent work by Maldacena
and Ooguri (hep-th/0001053).
Moreover we show that there exist infinite number of the on-shell chiral primary states possessing
the different light-cone momenta and the spectral flow consistently acts on the space of chiral
primaries. We observe that they are also chiral primaries in the sense of space–time (or the conformal
theory of long string) and the spectrum of space–time U (1)R charge is consistent with the expectation
from the AdS3 /CFT 2 -duality. We also clarify the correspondence between our framework and the
symmetric orbifold theory of multiple long string system (K. Hosomichi, Y. Sugawara, JHEP 9907
(1999) 027, hep-th/9905004).  2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 11.25.Hf; 11.25.Pm

Keywords: String theory; Conformal field theory; Ads/CFT correspondence; Free field realization
1. Introduction
Study of string theory on AdS3 background has been a subject of great importance
mainly for the following two reasons: firstly, it is a non-trivial example of completely
E-mail addresses: hikida@hep-th.phys.s.u-tokyo.ac.jp (Y. Hikida), hosomiti@yukawa.kyoto-u.ac.jp
(K. Hosomichi), sugawara@hep-th.phys.s.u-tokyo.ac.jp (Y. Sugawara).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 2 9 - 0
Y. Hikida et al. / Nuclear Physics B 589 (2000) 134–166 135
solvable string theories on a curved background with the Lorentzian signature [1–7].
The second reason, which is comparably newer than the first one, is the possibility of
understanding the AdS/CFT-duality [8–11] at a stringy level [12–14].
Although the string theory on this background (with NS B-field) is believed to be
described by a simple conformal field theory, SL(2; R)–WZW model [12], there are still
subtleties. Especially two different theoretical grounds have been proposed with respect to
the set up of the Hilbert space of quantum states;
b
(1) The Hilbert space is defined to be a representation space of current algebra SL(2; R).
(2) The Hilbert space is defined to be the Fock space of free fields which should be
identified with the string coordinates in some suitable parametrizations of the group
manifold SL(2; R).
The first ground is based on the standard prescription of two dimensional conformal field
theory (CFT 2 ). It is well-known that the WZW model for a compact group actually has
such a Hilbert space: the physical Hilbert space should be made up of a finite number
of the unitary (integrable) representations of current algebra. Since we now have a non-
compact group SL(2; R), the situation becomes rather non-trivial. If k ≡ k 0 − 2 (k 0 means
b
the level of SL(2; R)) is equal to a negative rational value, we have a finite number of the
“admissible” representations 1 which contain a rich structure with many singular vectors
[15]. However, string theory on AdS3 background corresponds to the cases of positive k,
in which we cannot have any unitary representation of SL(2; b R) (except for the trivial
representation). There are infinite number of non-unitary representations some of which
have a few singular vectors (generically, no singular vectors). The best we can expect is
that the BRST condition successfully eliminates all negative norm states from the physical
Hilbert space. Many discussions have been given concerning the no-ghost theorem along
this line [1–5]. With respect to the discrete series the no-ghost theorem was proved under
the assumption of a truncation of quantum number “j ” parametrizing the second Casimir
[16].
In this traditional approach to CFT 2 from the representation theory of current algebra,
primary states are naively characterized by the two quantum numbers j , m. However, as
was carefully discussed by Bars [6], there is a subtle point if we recall that we are working
in a σ -model with a three-dimensional non-compact target space. One should keep it in
mind that under the limit of weak curvature; k → +∞, AdS3 space (the universal covering
of AdS3 , strictly speaking) may be replaced by a flat background R2,1 . In this sense it may
be more natural that we have three conserved momenta characterizing the physical states,
and the second definition of Hilbert space is likely to be more appropriate, as was claimed
in the works [6,7,17].
b
More recently, based on the representation theory of SL(2; R), Maldacena and Ooguri
claimed [18] that one should take the Hilbert space enlarged by the spectral flow and
proved the extended no-ghost theorem. On physical ground it may be plausible that the
b
1 The admissible representations are not necessarily unitary representations of SL(2; R) with k = −q/p. But
these define the good-natured conformal blocks in the similar manner as the familiar (p, q) minimal model with
c = 1 − 6(p − q)2 /pq [15].
136 Y. Hikida et al. / Nuclear Physics B 589 (2000) 134–166
third momentum we mentioned above is related to such an enlargement of Hilbert space.

One of the main purposes of this paper is to clarify the relation between them, namely, the
role of spectral flow in the argument of [18] and that of the third momentum of zero-modes
[6,7,17] (discrete light-cone momentum in the context of this paper).
Another aim of this work is to manifest further the role of “space–time Virasoro algebra”
introduced in [12]. It is inspired by the asymptotic isometries of Brown–Henneaux [19,20]
and is understood to describe the conformal symmetry of the long string sector [21,22]. The
generators of this algebra are most conveniently realized as operators acting on the Fock
space of free fields (the Wakimoto’s ϕ, β, γ system in the usual treatment). This is one
of the reasons why we take the free field realization rather than the abstract representation
theory of current algebra.
This paper is organized as follows: in Section 2 we start with reformulating the bosonic
and superstring theories on AdS3 background by a free field realization. With the help of
some field redefinitions we show that the AdS3 string theory can be described by a string
theory on a linear dilaton background (along the transverse direction) and with a light-like
compactification. We further demonstrate that the space–time conformal algebra given in
[12] has a quite simple form analogous to the DDF operators [23] in this framework of
“discrete light-cone Liouville theory”.
In Section 3 we analyze the physical spectrum in our framework. As that of a free
field system we will reproduce the spectrum proposed in [6,7,17]: only the principal
series is allowed due to the unitarity. We also comment on the outcome of turning on
the Liouville potential term, which should be a marginal perturbation. Such an interaction
term breaks the translational invariance along the radial direction (in other words, makes
the “screening” of the extra zero-mode momentum from the view points of the SL(2; R)
current algebra), and hence the “bound string states” possessing the imaginary momentum
along this direction may appear in the physical spectrum. The space–time Virasoro
operators play a role as the spectrum generating algebra, and we will observe that one
b
must include the representations of SL(2; R) which are broader than those given in [18] in
order to make the full space–time Virasoro algebra act successfully on the physical Hilbert
space.
In Section 4 we further investigate the spectrum for superstring cases that give rise to
space–time N = 2, 4 SCFTs. We present the complete set of on-shell chiral primaries. We
will find that there are infinite number of on-shell chiral primary states with the different
light-cone momenta, and the spectral flows act naturally among them. They become the
chiral primaries also in the sense of space–time and have the space–time U (1)R charges in
agreement with the expectation of AdS3 /CFT 2 -duality. As a byproduct we also clarify
the relationship with the symmetric orbifold CFT describing the multiple long strings
discussed in [24].
We will summarize the main results of our analyses and present some discussions in
Section 5.
2. Reformulation of AdS3 string theory as the discrete light-cone Liouville theory
2.1. Bosonic string on AdS3 × N
Through this paper we shall consider the universal covering of the AdS3 space with the
Lorentzian signature so that the time direction is non-compact. We start with the following
worldsheet Lagrangian for the (quantum) bosonic string on AdS3 with NS B-filed [12]
q q
L = ∂ϕ ∂ϕ¯ − 2 R (2) ϕ + β ∂γ ¯ + β̄∂ γ̄ − β β̄ exp − 2 ϕ , (2.1)
k k
where R (2) denotes the curvature on the worldsheet. Throughout this paper we shall only
focus on the physics at the near boundary region ϕ ∼ +∞,
q in which we can consider the
interaction term (“screening charge term”) β β̄ exp(− 2k ϕ) as a small perturbation. By
dropping this term simply we obtain the free conformal field theory
1 1
T = − ∂ϕ∂ϕ − √ ∂ 2 ϕ + β∂γ , (2.2)
2 2k
1
ϕ(z)ϕ(0) ∼ − ln z, β(z)γ (0) ∼ . (2.3)
z
q
We will later discuss the effect of restoring this interaction term β β̄ exp(− 2k ϕ) on the
physical spectrum.
The SL(2; R) symmetry in this free system is described by the Wakimoto representation
[25]
j − = β,
q
j 3 = βγ + k
2 ∂ϕ, (2.4)
√
j + = βγ 2 + 2k γ ∂ϕ + (k + 2)∂γ ,
b
which generates the SL(2; R) current algebra of level k + 2


 (k + 2)/2
 j 3 (z)j 3 (0) ∼ −
 ,

 z2
 ±1 ±
j 3 (z)j ± (0) ∼ j (0), (2.5)

 z



 k+2 2
 j + (z)j − (0) ∼ 2 − j 3 (0).
z z
By using the standard bosonization of β, γ [26]
β = i∂ve−u−iv , γ = eu+iv ,
u(z)u(0) ∼ − ln z, v(z)v(0) ∼ − ln z, (2.6)
we can rewrite the currents (2.4)
j − = i∂ve−u−iv ,
q
j 3 = −∂u + k2 ∂ϕ,
√
j + = eu+iv k∂(u + iv) + 2k ∂ϕ + i∂v . (2.7)
Moreover it is convenient to introduce the following new variables;

r r
2 k
Y :=
0
iu − iϕ,
k+2 k+2
r r
k+2 k k
Y := −
1
v+ √ iu + iϕ,
2 2(k + 2) k+2
r
k
ρ := (u + iv) + ϕ. (2.8)
2
Since this field redefinition is an SO(2, 1)-rotation, we simply have
Y 0 (z)Y 0 (0) ∼ ln z, Y 1 (z)Y 1 (0) ∼ − ln z, ρ(z)ρ(0) ∼ − ln z, (2.9)
and any other combinations have no singular OPEs.

b
In these variables the SL(2; R)k+2 currents are given by
r
k+2
j3 = i∂Y 0
2
r r √
± k+2 k
∂ρ e∓ 2/(k+2) i(Y +Y ) ,
0 1
j = − i∂Y ±
1
(2.10)
2 2
and also the stress tensor is rewritten as
1 2 1 2 1 1
T = ∂Y 0 − ∂Y 1 − (∂ρ)2 − √ ∂ 2 ρ, (2.11)
2 2 2 2k
which of course has the correct central charge c = 3 + 6k .
In this way we have found that the bosonic string theory on AdS3 can be realized by
two free bosons Y 0 , Y 1 (with
q no background charge) and a “Liouville mode” ρ with the
b
background charge Q ≡ 2 . The essentially same realizations of SL(2; R) were used
k
in several works [7,27,28]. It was suggested in [29] that the fields Y 0 , Y 1 , ρ roughly
corresponds to the global coordinates of AdS3 space, and in the similar sense one might
suppose that the Wakimoto fields ϕ, β, γ correspond to the Poincaré coordinates.
The definitions of hermitian conjugations of these free fields are standard
†
Y i (z) = Y i (1/z), (ρ(z))† = ρ(1/z) − Q ln z. (2.12)
(Recall that the Liouville field ρ has a background charge.) One can easily verify that the
3 , j ±† = j ∓
hermitian conjugations of the current generators take the usual forms jn3† = j−n n −n
under (2.12). This is an advantage of this free field realization (2.10) compared with the
Wakimoto representation in which the rules of hermitian conjugation are not simple.
Notice also that the OPE of Y 0 with itself has the wrong sign (2.9) and thus Y 0 should
correspond to the time-like coordinate. This fact is consistent with the usual interpretation;
j03 ∼ energy.
There is a subtle point with respect to the realizations of currents (2.10): j ± (z) are
not necessarily defined as local operators on the whole Fock space of Y 0 , Y 1 , ρ. To
overcome this difficulty it is natural to assume the following light-like compactification.

Let us introduce the light-cone coordinates
1
Y± = √ Y0 ±Y1 , (2.13)
2
and assume the periodic identification
2
Y − ∼ Y − + 2πR, R= √ . (2.14)
k+2
Such a prescription is known as the name “discrete light-cone quantization”, and have been
applied to the studies of M(atrix) theory with finite N [30–32].
In this case, the conjugate momentum of Y − is quantized as
∂Y + 2n
≡ P + + PS+ = , n ∈ Z, (2.15)
∂τ R
and the winding mode of Y − should take
∂Y −
≡ P − − PS− = mR, m ∈ Z. (2.16)
∂σ
Since Y + remains non-compact, there is no winding mode along this direction
∂Y +
≡ P + − PS+ = 0, (2.17)
∂σ
and thus we obtain P + = PS+ = n/R (n ∈ Z).
By using these facts, we can concretely write the “tachyon” vertex operators
Sj,m̄,p (z̄), where the left mover is defined by
Vj,m,m̄,p (z, z̄) ≡ Vj,m,p (z)V
√ √ r
k+2 2m + k+2 − 2
Vj,m,p = exp p− √ iY + piY − jρ , (2.18)
2 k+2 2 k
and the winding condition (2.16) means that m − m̄ ∈ Z. The corresponding Fock vacuum
|j, m, pi has the following properties;

k+2
j0 |j, m, pi = m −
3
p |j, m, pi, (2.19)
2
±
j∓p |j, m, pi = (m ± j )|j, m ± 1, pi, (2.20)
±
j∓p+n |j, m, pi = 0 (∀n > 1), (2.21)

1 k+2 2
L0 |j, m, pi = − j (j − 1) + mp − p |j, m, pi. (2.22)
k 4
Namely, j , m mean the quantum numbers appearing in the usual SL(2; b R) theory and p
corresponds to the label of “flowed representation” of [18]. In fact, the spectral flow in the
context of [18] is defined as the following transformations

 3 k+2p
j (z) −→ j 3 (z) + ,
2 z (2.23)
 ± ∓p ±
j (z) −→ z j (z).
In the system of Y 0 , Y 1 , ρ, this is simply the momentum shift

r
k+2
Y −→ Y − p
0 0
i ln z, (2.24)
2
and Y 1 , ρ remain unchanged.
The global SL(2; R) algebra {j03 , j0± } is manifestly BRST invariant. We can immediately
extend this algebra to the “space–time Virasoro algebra”
 I r

 k+2

 L0 = −j0 ≡ −
3
i∂Y 0 ,
2
I r r (2.25)

 k+2 k − √2n iY +

 Ln = i∂Y − n
1
∂ρ e k+2 (n 6= 0),
2 2
which actually generates the Virasoro algebra on the Fock space over the vacuum |j, m, pi
c 3
[Ln , Lm ] = (n − m)Ln+m + n − n δn+m,0 , (2.26)
12
where c = 6(k + 2)p.
We here make a few comments: firstly, the Virasoro operators Ln are well-defined as
local operators on the whole Fock space compatible with the light-like compactification
(2.14). Secondly, it is natural to regard Ln (n 6= 0) as analogs of the DDF operators
[23] along the ρ-direction. In fact, Ln (n 6= 0) is no other than theH unique solution
for the BRST condition among the operators having the form ∼ (A∂ρ + B∂Y +
+ C∂Y − ) exp[− √2n k+2
iY + ], A 6= 0 (up to BRST exact terms and some overall constant,
of course). In the next section we will make use of such DDF like operators in order to
construct the complete set of the physical states. As the last comment, we should point
out that Ln are BRST equivalent to the space–time Virasoro operators constructed in [12].
It is straightforward to confirm that the quantum number p precisely coincides with the
H
“winding of γ ”; γ −1 dγ = p.
2.2. Superstring on AdS3 × S 1 × N /U (1)
Let us try to extend our previous results to the superstring cases. We start with the general
superstring vacua AdS3 × S 1 × N /U (1) studied in [33,34], which are compatible with the
worldsheet N = 2 SUSY. The most familiar example AdS3 × S 3 × M 4 (M 4 = T 4 or K3)
is nothing but a special example of these backgrounds, and we can readily apply the results
in this subsection to that case too.
First of all, to fix the notations we summarize the worldsheet properties of this
superstring model;
• AdS3 sector (j A , ψ A )
To extend to the superstring case, we introduce free fermions in the adjoint representation
1 2
ψ 3 (z)ψ 3 (0) ∼ − , ψ + (z)ψ − (0) ∼ ,
z z
±
ψ = ψ ± iψ .
1 2
(2.27)
b
The total SL(2; R) currents are given by
i A B C
J A = j A + jFA = j A − BC ψ ψ , A, B, C = 1, 2, 3, (2.28)
2
where the fermionic currents jFA have the level −2. The fermionic currents jFA can be
written by free fermions as
1
jF± = ±ψ ± ψ 3 , jF3 = ψ + ψ − . (2.29)
2
This sector has an N = 1 superconformal symmetry given by
r
2 1 + − 1 − + 1 + − 3
GSL(2,R) = ψ j + ψ j −ψ j − ψ ψ ψ ,
3 3
(2.30)
k 2 2 2
and the central charge is
3(k + 2) 3
c= + . (2.31)
k 2
• S 1 sector (Y, χ )
We have a scalar field Y parametrizing S 1
Y (z)Y (0) ∼ − ln z, (2.32)
and its fermionic partner χ
1
χ(z)χ(0) ∼ . (2.33)
z
This sector has the simplest N = 1 superconformal symmetry
GS 1 = χi∂Y, (2.34)
with the central charge
3
c= . (2.35)
2
• N /U (1) sector
We require that this sector has an N = 2 superconformal symmetry described by the
currents
TN /U (1), G±
N /U (1) , JN /U (1) . (2.36)
The relation between N = 2 and N = 1 superconformal current is
GN /U (1) = G+ −
N /U (1) + GN /U (1) . (2.37)
Because of the criticality condition, the central charge of this sector should be equal to
6
c=9− , (2.38)
k
and the U (1)R current satisfies
3 − 2k
JN /U (1)(z)JN /U (1)(0) ∼ . (2.39)
z2
We can realize the N = 2 superconformal symmetry on the worldsheet in this system.

We choose the U (1)R current as
JR = JR1 + JR2 + JN /U (1) , (2.40)
where
1 2
JR1 = ψ + ψ − + J 3 , (2.41)
2 k
JR2 = χψ 3 . (2.42)
According to the charge of this current the N = 1 superconformal current splits into two
terms
G = GSL(2,R) + GS 1 + G+ −
N /U (1) + GN /U (1)
≡ G+ + G− , (2.43)
where
G± = G± ± ±
1 + G2 + GN /U (1) , (2.44)
1
G± ± ∓
1 = √ ψ j , (2.45)
2k

± 1 1 1 3
G2 = √ χ ∓ ψ 3
√ i∂Y ± √ J . (2.46)
2 2 k
The energy–momentum tensor is also decomposed as follows;
T = T1 + T2 + TN /U (1), (2.47)
1 A 1
T1 = (j jA + J 3 J 3 ) − ψ + ∂ψ − − ∂ψ + ψ − , (2.48)
k 4
1 1 1 1
T2 = − J 3 J 3 − (∂Y )2 − χ∂χ + ψ 3 ∂ψ 3 . (2.49)
k 2 2 2
It may be worthwhile to mention that the superconformal generators {Ti , G± i , JRi } (anti-)
commute among the different sectors. Furthermore {T1 , G±
1 , JR1 } has the same expression
as that of the Kazama–Suzuki coset model for SL(2; R)/U (1) [35,36].
The BRST charge QBRST is defined in the standard manner
I

QBRST = c T − 12 (∂φ)2 − ∂ 2 φ − η∂ξ + ∂cb + ηeφ G − bη∂ηe2φ , (2.50)
where φ, η, ξ are the familiar bosonized superghosts [26].

Now let us try to reformulate this superstring model as the discrete light-cone Liouville
theory as in the case of bosonic string. Our goal is the N = 2 Liouville theory [37,38] with
the light-like compactification; R+ × S− 1 × R × S 1 × N /U (1). To this aim we need to
ρ
perform further field redefinitions.
As a preliminary we bosonize the fermions ψ ±
√
ψ ± = 2 e±iH1 , (2.51)
where H1 (z)H1 (0) ∼ − ln z, and the radius of compact boson H1 should be 1. Let Y0 , Y1
be as given in (2.8), and define
r r
k+2 0 2
X :=
0
Y + H1 ,
k k
r r
2 k 2
X := − √
1
Y +
0
Y −
1
H1 , (2.52)
k(k + 2) k + 2 k
r
2
H10 := Y 0 + Y 1 + H1 .
k+2
Since this is again an SO(2, 1) rotation, we have the OPEs
X0 (z)X0 (0) ∼ ln z, X1 (z)X1 (0) ∼ − ln z, H10 (z)H10 (0) ∼ − ln z, (2.53)
and all the non-diagonal OPEs vanish. We also rewrite
X2 := Y, Ψ 2 := χ, Ψ 0 := ψ 3 ,

1 0
Ψ ± ≡ − √ Ψ 1 ± iΨ ρ := e±iH1 . (2.54)
2
After all, we have changed the system of

ϕ, β, γ , Y, ψ ± , ψ 3 , χ (2.55)
into the system of new free fields

ρ, X0 , X1 , X2 , Ψ ± , Ψ 0 , Ψ 2 . (2.56)
In these new variables the energy–momentum tensors (2.48), (2.49) are rewritten as
1 2 1 1 1
T1 = − ∂X1 − (∂ρ)2 − √ ∂ 2 ρ − (Ψ + ∂Ψ − − ∂Ψ + Ψ − ), (2.57)
2 2 2k 2
1 2 1 2 1 1
T2 = ∂X0 − ∂X2 + Ψ 0 ∂Ψ 0 − Ψ 2 ∂Ψ 2 . (2.58)
2 2 2 2
The U (1)R currents (2.41), (2.42) become
JR1 = Ψ + Ψ − − Qi∂X1 , (2.59)
JR2 = −Ψ Ψ . 0 2
(2.60)
q
Q is the background charge of Liouville mode ρ and in this case Q = 2k .
The N = 2 superconformal currents (2.45), (2.46) now take the following forms
1 ± Q
G±1 = −√ Ψ i∂X1 ± ∂ρ ∓ √ ∂Ψ ± , (2.61)
2 2
1 1
G±2 = −√ Ψ ∓ Ψ
0 2
× √ i∂ X0 ± X2 . (2.62)
2 2
It is also convenient to rewrite the total super current
G = −Ψ 0 i∂X0 + Ψ 1 i∂X1 + Ψ 2 i∂X2 + Ψ ρ i∂ρ + Qi∂Ψ ρ . (2.63)
Notice that {T1 , G±
JR1 } (2.48), (2.45), (2.41) have been now transformed into
1,
the expressions of superconformal algebra in the N = 2 Liouville theory [37,38] as
we mentioned before. The essential part of this field redefinition is the identification
between SL(2; R)/U (1) Kazama–Suzuki model and the N = 2 Liouville theory (see the
Appendix B of [39], and also Ref. [40]) and it was claimed in [41] that these two theories
are related by a T-duality.
As in the bosonic case, we introduce the light-cone coordinates
1
X± = √ X0 ± X1 , (2.64)
2
and assume the compactifications
4π
X− ∼ X− + √ , H10 ∼ H10 + 2π. (2.65)
k
√
These are indeed consistent with the previous compactifications Y − ∼ Y − + 4π/ k + 2,
H1 ∼ H1 + 2π , because we can obtain from (2.52)
 r

 − 2 + k+2 −
X = √ Y + Y + √2 H1 ,
k(k + 2) k k
(2.66)

 2
 H10 = √ +
Y + H1 .
k+2
We likewise introduce the tachyon vertices compatible with this light-like compactifica-
tion
√ √ r
k 2m k 2
Vj,m,p = exp p − √ iX+ + piX− − jρ , (2.67)
2 k 2 k
and the corresponding Fock vacua satisfying

k
J03 |j, m, pi = m − p |j, m, pi, (2.68)
2
±
J∓p |j, m, pi = (m ± j )|j, m ± 1, pi, (2.69)
±
J∓p+n |j, m, pi = 0 (∀n > 1), (2.70)

1 k 2
L0 |j, m, pi = − j (j − 1) + mp − p |j, m, pi. (2.71)
k 4
The total SL(2; R) currents are also rewritten in the new coordinates;
r
k
J =
3
i∂X0 , (2.72)
2
r r
± k k + −
√ 0 ± ∓ √2 iX+
J = − i∂X ±
1
∂ρ − Ψ Ψ ± 2 Ψ Ψ e k . (2.73)
2 2
To close this section, we present the space–time superconformal algebra in our new
variables, which again has the forms reminiscent of the DDF operators.
First, the space–time Virasoro algebra is given (in the (−1)-picture) by
 r I

 k
 L0 = − e−φ Ψ 0 ,
r I 2 (2.74)

 +
 Ln = k e−φ e−n √k iX Ψ 1 + niΨ ρ
2
(n 6= 0).
2
We again mention that LHn (n 6= 0) is the unique solution of the BRST constraint among the
operators of the form ∼ e−φ exp[−n √2 iX+ ](AΨ ρ + BΨ 0 + CΨ 1 ), A 6= 0.
k
The space–time U (1)R current is given by
√ I −n √2 iX+ 2
Jn = 2k e−φ e k Ψ . (2.75)
To construct the space–time super currents we must introduce the spin fields. According
to [33,34], we bosonize the “deformed U (1)R current” on the worldsheet;

JR0 := JR − Qi∂ X0 + X2
√
≡ i∂H10 − i∂H2 − i 3 ∂H3 − Qi∂ X0 + X1 , (2.76)
where we set
i∂H10 = Ψ + Ψ − (as defined before),
i∂H2 = Ψ Ψ , 0 2
√
−i 3 ∂H3 = JN /U (1) − Qi∂X2 . (2.77)
(The combined current JN /U (1) − Qi∂X2 actually has the Schwinger term ∼ 3/z2 .)
The
practical reason why we do so is as follows: JR has non-trivial OPEs with the vertex
operators such as exp[−n √2 iX+ ], and thus the DDF like operators including the spin
k
fields (see (2.81)) made up of JR do not nicely behave under the BRST transformation. In
contrast, we can rather simply solve the BRST condition for the vertices associated to the
current JR0 , since it has no singular OPE with exp[−n √2 iX+ ].
k
Now the spin fields should take the form (up to cocycle factors)
√
exp 2i (1 H10 + 2 H2 + 3 3 H3 ) exp 4 2i Q X0 + X1 (i = ±1). (2.78)
The GSO projection leaves only a half of them satisfying
Y
3
i = −1, (2.79)
i=1
and we use the notation
√
S 3 1 = exp 2i (1 H10 + 2 H2 + 3 3 H3 ) (2.80)
to express the spin fields allowed
H by the GSO condition. We can explicitly verify that the
fermion vertices of the type exp[− φ2 ] exp[ 2i 4 Q(X0 + X1 )]S 3 1 are BRST invariant, if
H
and only if 4 = −1 , namely the vertices of the type exp[− φ2 ] exp[− √2 21 iX+ ]S 3 1 .
k
They define the space–time N = 4 SUSY (the global part of the space–time N = 2
superconformal symmetry in NS sector). We can work out more general fermion vertices
of the type
I
X
∼ exp − φ2 exp −r √2 iX+ A3 ,1 S 3 1 r ∈ 12 + Z . (2.81)
k
3 ,1
The BRST condition uniquely (up to BRST exact terms and an overall normalization)
determines the coefficients A3 ,1 and we finally obtain the physical vertices
I

Gr± = k 1/4 exp − φ2 exp −r √2 iX+ r + 12 S ±+ − r − 12 S ±− ,
k

r∈ 1
2 +Z . (2.82)
They generate the N = 2 superconformal algebra (in NS sector) together with Ln , Jn , and
the central charge is equal to 6kp on the Fock space over the vacuum |∗, ∗, pi. In fact, it
is a straightforward calculation to check that these space–time superconformal generators
(2.74), (2.75), (2.82) coincide with the ones constructed in [33,34] up to BRST exact terms.
3. Analyses on spectra of physical states
In this section we shall investigate the spectrum of physical states in our reformulation
of AdS3 string theory. The unitarity of physical Hilbert space is an important problem. We
first analyze the physical spectrum as that of a free field theory, namely without taking the
Liouville potential term into account, and later discuss the effect of turning on this term.
3.1. Spectrum as free field theory
First, we analyze the spectrum of bosonic string on AdS3 × N . Let us recall that the
Fock vacuum is defined from the tachyon vertex
|j, m, pi = lim Vj,m,p (z)|0i, (3.1)
z→0
and we denote the corresponding Fock space as Fj,m,p from now on. We also define “out
state” as
hj, m, p| = lim h0|Vj,m,p (z)z2hj,m,p , (3.2)
z→∞
where
1 k+2 2
hj,m,p = − j (j − 1) + mp − p . (3.3)
k 4
Using the momentum conservation and taking account of the existence of background
charge along the ρ-direction, we obtain
h1 − j, −m, −p|j, m, pi 6= 0, (3.4)
and the other combinations vanish. Notice that we must use the following hermitian
conjugation
h1 − j, −m, −p| = (|j, m, pi)† (3.5)
to discuss the unitarity.
We shall qhere neglect the Liouvilleq
potential term (that is, the “screening charge term”
R R
β β̄ exp[− ϕ] ∼ ∂Y ∂Y
2 1 ¯ exp[− 2 ρ] in the σ -model action (2.1)). This means that
1
k k
the Hilbert space of physical states should be defined as the BRST cohomology on the
Fock spaces of the free fields Y 0 , Y 1 , ρ properly tensored by the Hilbert space of N sector.
Deciding the physical spectrum is a rather simple problem as in the usual string theory
on the flat Minkowski space (at least as long as we only take the primary states in the
N sector in constructing the physical states). However, there is one non-trivial point: the
existence of background charge in the ρ-direction. As is well-known, the BRST constraint
eliminates two longitudinal degrees of freedom in the case of Minkowski background:
one is eliminated by the BRST condition itself and another becomes BRST exact. From a
physical point of view this aspect is closely related to the fact that the first excited states
(the graviton states) become mass-less, and thus one of the polarization vectors must be
light-like. On the other hand, in our case of linear dilaton theory we have a mass gap
originating from the background charge of ρ and so the first excited states become massive.
This implies that one of the two longitudinal modes does not decouple from the physical
Hilbert space, since all the polarization vectors are space-like.
To make this point clearer, let us consider a simple example. We are here given
P ρ
one transverse oscillator i∂ρ = n αn /zn+1 and two longitudinal oscillators i∂Y ± =
P ± n+1
± n αn /z . The simplest candidate of the first excited states is
α−1 |p+ , p− , pρ i ⊗ c1 |0igh ,
ρ
(3.6)
where the on-shell condition is given by
1 1
p+ p− + (pρ )2 + = 0. (3.7)
2 4k
√
(The relation of the momenta with our previous convention
q is given by p+ = −( k + 2/2)
√ √
× p, p− = −( k + 2/2)p − (2m/ k + 2), pρ = i 2k (j − 12 ).) Now we assume
p+ p− 6= 0. The BRST transformation of this candidate (3.6) yields a non-vanishing term
due to the background charge. We must cancel it by mixing the longitudinal modes to
recover the BRST invariance. After some simple calculation we find two independent
solutions

Q ±
|p+ , p− , pρ i ⊗ c1 |0igh .
ρ
α−1 + i ± α−1 (3.8)
p
In the usual free string theory the general physical states are created by making the
transverse DDF operators act on suitable Fock vacuum (“allowed states”). The above
simple observation suggests that, in our case of AdS3 , we must make use of the two
independent DDF operators that are not purely transversal. Some candidates for the
suitable DDF operators are already given by Satoh [7] 2
I
± Q ± n ±
Bn := i ∂ρ + ∂ ln ∂Y exp ± iY , (3.9)
2 p
which are BRST invariant and satisfy the commutation relation of a free boson
± ±
Bm , Bn = mδm+n,0 ,
± +
Bm |p , p− , pρ i = 0 (∀m > 1). (3.10)
2 In [7] the other candidates for the DDF operators are also proposed. However, they include the ghost number
current explicitly in their expressions and are not BRST invariant. Y.S. should express his great thanks to Dr. Satoh
for the discussion about this point.
Moreover, Bn± act on the Fock space defined by |p+ , p− , pρ i as the operators ∼ αn + · · · ,
ρ
±
as is expected, and it is easy to check that B−1 actually give rise to the first excited states
discussed above (3.8). So the reader might suppose that we can naively use Bn+ , Bn− to
construct the complete set of physical states. But this is not the case. Bn+ and Bm − are not
mutually local and thus we can never use them at the same time. The best we can do is to
use only one of them, Bm + , which is compatible with the light-like compactification (2.14).
We rewrite it as
I
(p) Q + n 2 +
An = i ∂ρ + ∂ ln ∂Y exp − √ iY ,
2 p k+2
(p) (p)
Am , An = mδm+n,0 , (3.11)
which are defined as local operators on F∗,∗,p (p 6= 0).
Now, the question we must solve is as follows: what is the missing DDF operator that
can compensate (3.11)? As we already suggested, the answer is the space–time Virasoro
operators. Let Ln be the space–time Virasoro operators defined in (2.25). L pn are well-
ρ
defined as local operators on the Fock space Fj,m,p and behave as ∼ αn + · · · . Therefore
they are the candidates of the missing DDF operators. Alternatively we shall define
(p) k+2 2
Ln := pL pn − p − 1 δn,0 , (3.12)
4
which are shown to generate a Virasoro algebra with the central charge c = 6(k + 2)
(p) (p)
(irrespective of the value p). Furthermore, Ln are mutually local with Am and satisfy
the commutation relation
r
(p) (p) (p) k 1
Ln , Am = −mAm+n + iαn δn+m,0 , α ≡
2
1− . (3.13)
2 k
It is also convenient to introduce improved Virasoro operators
(p) (p) 1 X (p) (p) (p) 1
L̃n := Ln − : A−(n+m) Am : −iαnAn − α2 δn,0 , (3.14)
2 m 2
(p)
which are defined so that they commute with {Am } and generate the Virasoro algebra
with c = 23 − 6k ≡ cN . This value of the central charge is quite expected. One can show
(p)
that the DDF operators L̃n correspond to the energy–momentum tensor of N sector in
the light-cone gauge.
In this way, we have found that the physical Hilbert space should be spanned by the
states having the forms
(p) (p) (p) (p)
L̃−n1 L̃−n2 · · · A−m1 A−m2 |j, m, pi ⊗ · · · (ni > 1, mi > 1). (3.15)
In order to complete our discussion we must confirm the linear independence of the
states of the type (3.15). Happily, this is very easy to prove in our case. The Virasoro algebra
(p)
{L̃n } has the central charge greater than 1 for sufficiently large k, and thus the Kac deter-
minant does not vanish, as is well-known in the representation theory of Virasoro algebra.
We can now present the complete list of physical states. This spectrum is specified
by the momenta of the Fock space Fj,m,p ⊗ F Sj,m̄,p previously defined. The light-like
(p) (p)
compactification (2.14) implies that p ∈ Z, and also L̃0 − L̃0 ≡ p(L0 − L̄0 ) ∈ pZ (the
(p)† (p)
“level matching condition”). Since A0 = A0 holds and we have
r
(p) 2 1
A0 |j, m, pi = i j− |j, m, pi, (3.16)
k 2
the value of j allowed by the unitarity is j = 12 + is (s ∈ R), at least when p 6= 0. It
corresponds to the principal continuous series of unitary representation of SL(2; R). Also
in the case of p = 0 we can show that only the principal series is permitted from the
unitarity, as we will observe below.
To avoid unnecessary complexity we shall only take the primary states in the N sector
in constructing the physical states, and assume the conformal weights hN of these primary
states are non-negative. It is not difficult to construct more general physical states including
the descendant states in the N sector, if we are concretely given the unitary CFT model
describing this sector.
We discuss the p = 0 and p 6= 0 cases separately.
(1) p = 0.
(p) (p)
In this case, the DDF operators of the types An , Ln are ill-defined. But we must require
the space–time Virasoro operators {Ln } (with central charge 0) should define an unitary
representation, since they are well-defined as local operators even in this sector. Ln simply
maps a Fock vacuum to another Fock vacuum and so the representation space is given by
⊕r∈Z C|j, m + r, 0i with arbitrary fixed values of m ∈ R, j . We can obtain
h1 − j, −m − r, −p|L−n Ln |j, m + r, pi

n 2 1 2
= m+r + −n j −
2
h1 − j, −m − r − n, 0|j, m + r + n, 0i.
2 2
(3.17)
Since the conformal weight (in the sense of worldsheet CFT) must be real at least, j should
take the values j ∈ R or j = 12 + is (s ∈ R). If j − 12 ∈ iR (principal series), the coefficient
of R.H.S in (3.17) is always positive and we have an unitary representation of {Ln }. On the
other hand, if j ∈ R, we can always find r ∈ Z for which this coefficient becomes negative
as long as we choose n to be sufficiently large. This means that the cases of j ∈ R cannot
be unitary representations of {Ln }, and hence we must rule out these sectors.
The general physical states with p = 0 are written as
1
+ is, m, 0 ⊗ |hN i ⊗ c1 |0igh ⊗ 1 + is, m̄, 0 ⊗ |hN i ⊗ c̄1 |0igh
2 2
m, m̄ ∈ R, m − m̄ ∈ Z, (3.18)
where |hN i is the primary state with conformal weight hN in the N sector and |0igh is the
vacuum of the ghost system. The on-shell condition is given by

1 2 1
s + + hN = 1. (3.19)
k 4
If k > 1/4, we can always solve the on-shell condition (3.19) for hN = 0. These physical
states are tachyons whose mass-squared are lower than the Breitenlohner–Freedman bound
[42,43]. Such an instability in bosonic string theory is not surprising, and we later observe
that the GSO projection successfully eliminates these tachyonic states in the superstring
case.
There is one comment: if we took account of the unitarity of the representation only
of SL(2; R) (that is, {Ln }, n = 0, ±1), many representations would survive in the sector
j ∈ R: the discrete series Dj± and the exceptional series Ej,α , as is well-known and many
readers might expect. It is crucial in the above argument to take the full Virasoro algebra
{Ln } in place of the SL(2; R) subalgebra.
(2) p 6= 0.
As we already discussed, in this sector j must be equal to j = 12 + is (s ∈ R), and the
(p) (p)
physical Hilbert space is generated by the actions of the DDF operators {A−n }, {L̃−n }
(n ∈ Z>1 ) on the on-shell Fock vacua. We must discuss the positivity of the norm of such
(p)
physical states. Obviously A−n create only positive norm states, and do not lead to any
(p)
constraint. However, the Virasoro generators {L̃n } give rise to a non-trivial constraint
for unitarity. Since this Virasoro algebra has the central charge c > 1, the condition for
(p)
the unitarity means that the L̃0 -eigenvalue of the Fock vacuum | 12 + is, m, pi is non-
negative. (We again assume k is sufficiently large.) It is easy to show that this unitarity
condition is equivalent to a simple inequality hN > 0 thanks to the on-shell condition

1 2 1 k+2 2
s + + mp − p + hN = 1. (3.20)
k 4 4
(p)
This equivalence is not surprising, since L̃0 corresponds to the stress tensor for N sector
in the light-cone gauge. In this way we conclude that the no-ghost theorem for this sector
is trivially satisfied as long as the internal CFT N is unitary. This result is consistent with
those of [6,7], although we here take a different convention of free field representation: Our
convention diagonalizes the time-like current j 3 (corresponding to the energy operator).
On the other hand, those given in [6,7,17] diagonalize one of the space-like currents. We
remark that the light-cone momentum p plays a role similar to that of the extra zero-mode
momentum emphasized in [6,7,17].
(p)
One can find that the L0 -eigenvalue (not L̃0 ) of the on-shell Fock vacuum, which
corresponds to the space–time energy, is bounded below

hN (k − 1)2 k + 2 1 hN − 1 k + 2
L0 > + + p− ∼ + p, (3.21)
p 4kp 4 p p 4
(for a sufficiently large value k). This means that this sector corresponds to the long string
states in the sense of [18] and belongs to the continuous spectrum above the threshold
energy ∼ k4 p discussed in [21,22].
In summary, the general physical states are written as

L̃−n1 L̃−n2 · · · A−m1 A−m2 · · · 12 + is, m, p ⊗ |hN i ⊗ c1 |0igh
(p) (p) (p) (p)

⊗L̃(p) −n̄1 L̃(p) −n̄2 · · · A(p) −m̄1 A(p) −m̄2 · · · 12 + is, m̄, p ⊗ |h̄N i ⊗ c̄1 |0igh
n1 , n2 , · · · > 1, m1 , m2 , · · · > 1,
n̄1 , n̄2 , · · · > 1, m̄1 , m̄2 , · · · > 1, (3.22)
where the on-shell conditions are

1 2 1 k+2 2
s + + mp − p + hN = 1,
k 4 4

1 2 1 k+2 2
s + + m̄p − p + h̄N = 1, (3.23)
k 4 4
and the “level matching condition” is given as
X X X X
ni + mj − mp = n̄i + m̄j − m̄p(modp). (3.24)
i j i j
The superstring case AdS3 × S1× N /U (1) is similarly analyzed. The unitarity of the
physical Hilbert space is derived from the unitarity in the N = 2 SCFT describing N /U (1)
sector. We here only discuss how tachyonic states in the p = 0 sector are eliminated by the
GSO projection.
The tachyon vertex operators have slightly different expressions as compared with the
bosonic case
√ √ r
k 2m + k − 2
Vj,m,p = exp p − √ iX + piX − jρ . (3.25)
2 k 2 k
q
−i 2 qX 2
Together with the vertex for the S1
direction e k we construct the Fock vacuum
|j, m, p, qi such that

k
J03 |j, m, p, qi = m − p |j, m, p, qi, (3.26)
2
±
J∓p |j, m, p, qi = (m ± j )|j, m ± 1, p, qi, (3.27)
±
J∓p+n |j, m, p, qi = 0 (∀n > 1), (3.28)

1 k 2 q2
L0 |j, m, p, qi = − j (j − 1) + mp − p + |j, m, p, qi, (3.29)
k 4 k

0 2q
JR0 |j, m, p, qi = + p |j, m, p, qi. (3.30)
k
In the p = 0 sector the argument similar to the bosonic case leads to the following
physical states:
1
+ is, m, 0, q ⊗ |hN , qN i ⊗ ce−φ |0igh
2

⊗ 12 + is, m̄, 0, q̄ ⊗ |h̄N , q̄N i ⊗ c̄e−φ̄ |0igh
m, m̄ ∈ R, m − m̄ ∈ Z, (3.31)
with the on-shell conditions

1 2 1 q2 1
s + + + hN = ,
k 4 k 2
2
1 2 1 q̄ 1
s + + + h̄N = . (3.32)
k 4 k 2
Naively we can solve the on-shell conditions as in the bosonic case and they are tachyonic
states except for s = 0. However, we can show that the GSO projection eliminates such
tachyonic states, as is expected. From the on-shell conditions (3.32) and the assumption
hN > 12 |qN |, which is derived from the unitarity in the N /U (1)-sector, we obtain the
inequality

1 1 q 2 |qN | 1
+s +2
+ 6 . (3.33)
k 4 k 2 2
We should define the GSO condition with respect to the deformed U (1)R current JR0 and
it reads as 2q
k + qN = 2l + 1 (l ∈ Z). First we assume qN > 0. If l > 0, substituting qN =
2q
2l + 1 − k into the above inequality (3.33), we obtain

1 2
s + q−
2
+ 2lk 6 0, (3.34)
2
which leads to n = 0, q = 12 , s = 0. In the case of l < 0, q 6 k2 (2l + 1) must hold. We thus
obtain

1 1 1 q 2 |qN | s 2 1
> + s2 + + > + , (3.35)
2 k 4 k 2 k 2
which leads to s = 0, again. Therefore the tachyonic states whose mass-squared are lower
than the BF bound are successfully eliminated by the GSO projection. We can repeat the
same analysis when qN < 0.
3.2. Physical spectrum under the existence of Liouville potential
In the previous argument only the principal series was allowed. In the physical sense it
was a quite natural result, because we regarded the system as a free system and thus all the
momenta should be real.
Now, let us try to turn on the Liouville potential term (or the screening charge term in the
worldsheet action (2.1)). In this case we can expect some physical states with an imaginary
momentum along the ρ-direction describing the bound states (“bound string states” in the
terminology of [18]).
Unfortunately, a rigorous treatment of the quantum Liouville theory as an interacting
theory is quite non-trivial. Instead we shall here take the operator contents as free fields
and treat the Liouville potential as a small perturbation.
Recalling the σ -model action (2.1), this perturbation term may be identified with the
operator ∼ S SS, where
I r r I r
2 k+2 2
S = β exp − ϕ ≡− i∂Y exp −
1
ρ (3.36)
k 2 k
b
is no other than the familiar screening charge which commutes with all modes of SL(2; R)
currents. As for the spectrum generating operators, we may as well expect that at least
(p)
L±1 (∼ L±p ), L0 remain the good DDF operators, since they commute with the screening
charge (3.36).
On the other hand, because such an interaction breaks the translational invariance along
the ρ-direction, the ρ-momenta ∼ i(j − 12 ) loses its meaning. However the second Casimir
∼ j (j − 1) remains well-defined as a conserved quantity characterizing the physical states
even under the interacting theory. This is nothing but the standard argument of “screening
out” of the extra zero-mode momentum in the free field representation of CFT 2 [15,25,
44]. We may expect the bound string states possessing the imaginary ρ-momenta as long
(p)
as their second Casimirs take real values. In this way we can no longer regard A0 as a
(p)
good DDF operator. Moreover, we must also rule out the non-zero modes An (n 6= 0),
(p) (p)
because we have the following commutation relations A0 ∼ [(L−1 ) , An ] + const, for
n
n > 0, and A0 ∼ [(L1 )−n , An ] + const, for n < 0.

(p) (p)
(p)
It is a subtle problem whether the other modes of Virasoro operators Ln (n 6= 0, ±p)
remain the members of the spectrum generating algebra, since they also do not commute
with the screening charge (3.36). However, it may be plausible to admit these operators
from the point of view of the AdS3 /CFT 2 correspondence or the arguments of Brown–
Henneaux [19]. The fact that only the SL(2; R) generators L0 , L±1 commute with the
screening charge and the other modes do not is supposed to reflect the following fact: in the
argument of [19] the true isometry generates only the SL(2; R) and the other modes merely
correspond to the asymptotic isometries, which can be regarded as symmetries only near
the boundary. We shall now propose that the DDF operators suitable for the interacting
(p) (p) (p)
theory including (3.36) are {Ln } rather than those for the free system {L̃n , An }. This
claim is consistent with the analyses based on the light-cone gauge for the long string
configuration [22,28]. Hence our assertion is likely to be consistent at least with the
spectrum of the long string located near the boundary. (One should keep it in mind that
our assumption of small Liouville perturbation is valid only for such a configuration of
worldsheet.)
Based on this assumption we now present the complete physical spectrum as the
interacting theory. For the principal series j = 12 + is, the results are similar to those of
free fields. The case of p = 0 is the same as before, and when p 6= 0, only we have to do
(p) (p) (p)
is to replace the DDF operators {An , L̃m } in the expression of (3.22) by {Ln }. They
kp
likewise belong to a continuous spectrum above the threshold energy ∼ 4 .
A crucial difference is the existence of the physical states with j ∈ R as we already
suggested. To discuss the unitarity of this sector we again assume p > 0, and the cases
(p)
of p < 0 can be analyzed in the same way. The unitarity condition means that the L0 -
eigenvalue of the Fock vacuum should be non-negative. Thanks to the on-shell condition
1 k+2 2
− j (j − 1) + mp − p + hN = 1, (3.37)
k 4
we can immediately obtain the following inequality for the unitarity

1 1 2 (k − 1)2
j− 6 hN + . (3.38)
k 2 4k
We must also restrict the range of j as j > 1/2 because of the normalizability of wave
function (see [45,46]). Especially, in the case of hN = 0 we obtain the unitarity condition
1 k
<j 6 . (3.39)
2 2
These physical states do not propagate along the radial direction ρ, and are supposed to
correspond to the bound string states in the argument of [18]. In fact, we can evaluate the
space–time energy for this sector

k+2 1 < < hN − 1 k+2
p− ∼ L0 ∼ + p, (3.40)
4 p p 4
which is consistent with the result given in [18].
The physical states are summarized as follows;
(p) (p)
L−n1 L−n2 · · · |j, m, pi ⊗ |hN i ⊗ c1 |0igh
(p) (p)
⊗L−n̄1 L−n̄2 · · · |j, m̄, pi ⊗ |h̄N i ⊗ c̄1 |0igh
n1 , n2 , · · · > 1, n̄1 , n̄2 , · · · > 1, (3.41)
where the on-shell conditions are
1 k+2 2
− j (j − 1) + mp − p + hN = 1,
k 4
1 k+2 2
− j (j − 1) + m̄p − p + h̄N = 1, (3.42)
k 4
and the “level matching condition” is given as
X X
ni − mp = n̄i − m̄p(modp). (3.43)
i i
We here remark that the positive (negative) energy (L0 > 0) physical states should have
p > 0 (p < 0). This fact will be important in the discussions in the next section about the
interpretation of the long string theory.
To close this section we compare the above spectrum with the result of [18]. For this
b
purpose we must clarify which representations of SL(2; R) we should choose.
Let us assume p < 0. Going back to the Wakimoto free fields ϕ, β, γ we introduce
ϕ
the “Wakimoto module” Wj,m,p which is defined as the Fock space generated by α−n−1 ,
ϕ
βp−n , γ−p−n (n > 0) out of the vacuum |j, m, pi for m 6= j , and by α−n−1 , βp−n−1 ,
γ−p−n (n > 0) for m = j (corresponding to the flowed discrete series D b+ (p) in [18]).
j
Wj,m,p is obviously a subspace of Fj,m,p (and they are not isomorphic). Moreover, at
least under the restriction 12 < j 6 k2 (3.39), we can show that Wj,m,p can be identified
b
with some (reducible, in general) SL(2; R)-module, since we have no singular vectors in
the corresponding Verma module (except for the Fock vacua themselves). It is easy to see
that
Y (p) M
L−ni |j, m, pi ∈ Wj,m+ pr ,p . (3.44)
i r∈Z
(p)
Therefore we can successfully realize the actions of DDF operators Ln within the
L
(reducible) representations corresponding to r∈Z Wj,m+ pr ,p .
For p > 0, the essentially same argument works by introducing the “inverse Wakimoto
representation”;
 r

 k
 j = β̃ γ̃ −
3
∂ϕ,
2 (3.45)
+

 j = β̃, √
 −
j = β̃ γ̃ 2 − 2kγ̃ ϕ̃ − (k + 2)∂ γ̃ ,
or more explicitly,
 s

 2k

 ϕ̃ = ρ − Y +,

 (k + 2)


 r r
k+2 k 2 + (3.46)
 β̃ = − i∂Y +
1
∂ρ exp − √ iY ,

 2 2 k+2



 2

 γ̃ = exp √ iY + .
k+2
b
Consequently our choice of the representations of SL(2; R) is much larger than that of
[18] for the cases of j ∈ R, although the enlargement of the Hilbert space by the spectral
flow in [18] is incorporated into our setup as the discrete light-cone momentum p. By
construction our physical Hilbert space also contains no ghosts. In the analysis in [18]
m must take discrete values related with a fixed j (belonging to the discrete series of
b
SL(2; R) transformed by the spectral flow); j − m ∈ Z. On the other hand, in our analysis
m is arbitrary and independent of j as long as it satisfies the on-shell condition. This is
natural from our starting point: the σ -model (2.1) rather than the abstract representation
theory of affine Lie algebra, and thus j and m (and of course p, too) should correspond to
independent momenta along the different directions.
Furthermore, D b+(p) and D b−(p−1) are identified in [18], since they are equivalent
j (k+2)/2−j
b
as an irreducible representation of SL(2; R). Nevertheless they should be distinguished
from our viewpoint, because they possess the different light-cone momenta p. Especially,
the standard discrete series D b+ (lowest weight representations), D b− (highest weight
j j
representations) are realized in the sectors p = −1, p = 1, respectively, since the sectors
p = 0, j ∈ R are excluded in our analysis.
We also mention that the above unitarity condition (3.39) is analogous to the result
given in [16,18], but it is a slightly stronger condition. The unitarity bound proposed in
[16] reads 12 < j < k+2 1 k+1
2 and the one given in [18] reads 2 < j < 2 in our convention.
3
This disagreement originates from the different choices of the representations mentioned
above. In fact, if one choose to restrict the value of m as j − m ∈ Z when solving the
on-shell condition, one can show the no-ghost theorem under the assumption 12 < j < k+2 2
b+ (p) = D
3 The unitarity bound for [18] is stronger than that of [16] due to the identification D b− (p−1)
j (k+2)/2−j
mentioned above. Moreover, the same range of j was proposed in a different context [41] by requiring good
behaviors of the two point functions of the non-normalizable primary operators. Such two point functions nicely
behave, too, under our constraints (3.39), since it is more stringent than that of [41].
same as [16] rather than (3.39). 4 It is not yet clear whether our choice of the momenta m,
independent of j , is completely valid even in the rigid treatment as an interacting theory,
or nothing but an artifact originating from the free field approximation. However, we again
emphasize that our setup of physical Hilbert space admits the whole actions of DDF
(p)
operators {Ln } (and necessarily also with the space–time Virasoro algebra {Ln }). This
fact is found to be consistent with the several results about the long string sectors given by
the light-cone gauge approach [22,24,28], as we will comment in the next section. In fact,
one can readily find that our choice of m so as to be independent of j is crucial in order to
ensure the equivalence with the spectrum in the light-cone gauge after solving the on-shell
condition. In this sense we believe that our physical spectrum is valid at least for the long
string configuration near the boundary, which is well described by the light-cone gauge
approach. In order to justify this spectrum beyond the near boundary region we will have
to carry out a further analysis with the Liouville interaction term treated more precisely.
The extension of the above arguments to superstring examples is not so difficult and we
do not present it here. We instead focus on the spectrum of on-shell chiral primaries of
superstring on AdS3 × S 1 × N /U (1) in the next section.
4. Chiral primaries and spectral flow
In this section we further study the spectrum in the superstring cases. We especially
investigate an important class of observables: chiral primaries. In other words, we shall
concentrate on the topological sector of superstring vacua on AdS3 × S 1 × N /U (1) [47].
They are significant from the perspective of AdS3 /CFT 2 -duality. Although we have not
yet achieved the complete understanding of this duality, the study of their spectrum will
certainly clarify some aspects of it.
Through this section we only deal with the left moving parts of objects, and it is easy to
complete our discussions by taking also the right movers.
4.1. Background with space–time N = 4 SUSY
We first discuss the most familiar superstring vacua with space–time N = 4 SUSY;
AdS3 × S 3 × T 4 ∼
= AdS3 × S 1 × SU (2)/U (1) × T 4 , (4.1)
where SU (2)/U (1) means that the Kazama–Suzuki model [35,36] for this coset with c =
3 − N6 (N − 2 is equal to the level of (bosonic) SU (2)–WZW model describing the S 3
sector), which is identified with the N = 2 minimal model of AN−1 type and we denote
it by MN from now on. The criticality condition gives k = N , and this background is
regarded as the near horizon limit of NS5/NS1 system, as is well-known [8–12].
4 To be precise, under this restriction j − m ∈ Z we must take L as the DDF operators rather than L (p)
n n =
pLn/p + · · ·, (recall (3.44)) and hence the unitarity here means that Ln -descendants should not include any
negative norm states.
Let |Φl i (l = 0, 1, . . . , N − 2) be the chiral primary states in the MN sector with hN =

qN
2 = 2Nl
. We must look for the chiral primary states in the total system by tensoring |Φl i
with suitable vertex operators in the AdS3 × S 1 sector. In the notation of previous section,
namely,
√ √ r r
k 2m + k − 2 2
|j, m, p, qi ≡ lim exp p − √ iX + piX − jρ − qiX2 |0i,
z→0 2 k 2 k k
(4.2)
the possible candidates for the desired vertices are written as follows;

j, j, p, j − k p , +
Ψ−1/2 j, −(j − 1), p, −(j − 1) − k2 p . (4.3)
2
They are primary states with respect to T (z), G(z) and also satisfy

G+ + +
−1/2 j, j, p, j − 2 p = G−1/2 Ψ−1/2 j, −(j − 1), p, −(j − 1) − 2 p = 0.
k k
(4.4)
First we consider |j, j, p, j − k2 pi ⊗ |Φl i ⊗ c1 e−φ |0igh . The on-shell condition leads to
N −l
j= . (4.5)
2
(The GSO condition is automatically satisfied, since we have the relation h = Q 2 .)
+
Similarly, for the second candidate Ψ−1/2 |j, −(j − 1), p, −(j − 1) − 2 pi ⊗ |Φl i ⊗
k
c1 e−φ |0igh we can solve the on-shell condition and obtain

l
j =1+ . (4.6)
2
From now on we denote the first type of chiral primary (4.5) as |l, p ; 1i and the second
type (4.6) as |l, p ; 2i. Namely, we set

N − l N − l N(p − 1) + l
|l, p; 1i := , , p, − ⊗ |Φl i ⊗ ce−φ |0igh , (4.7)
2 2 2

+ l
l Np + l
|l, p; 2i := Ψ−1/2 2 + 1, − , p, − ⊗ |Φl i ⊗ ce−φ |0igh . (4.8)
2 2
They are both normalizable and satisfy the unitarity condition (3.38).
Remarkably one can find that (4.7), (4.8) are also chiral primaries with respect to the
space–time superconformal algebra {Ln , Gr± , Jn }, as suggested in [47]. They satisfy
1 1
L0 |l, p; 1i = J0 |l, p; 1i = {l + N(p − 1)}|l, p ; 1i, (4.9)
2 2
1 1
L0 |l, p; 2i = J0 |l, p; 2i = (l + Np)|l, p; 2i. (4.10)
2 2
Since the light-cone momentum p can now take an arbitrary integer, we have infinite
number of on-shell chiral primaries. All of them have the same U (1)R charge in the sense
of worldsheet because of the the on-shell condition. But they have the different U (1)R
charges with respect to the space–time conformal algebra.
Let us study the action of the spectral flow on these states. A natural extension of the
spectral flow (2.23) to the superstring case is given by
r
k
Up X 0
(z)U−1
p = X (z) − p
0
i ln z,
2
r
k
Up X2 (z)U−1p = X (z) + p
2
i ln z, (4.11)
2
b
and the other fields should remain unchanged by the spectral flow. The total SL(2; R)-
currents are transformed by them as follows;
kp
Up Jn3 U−1
p = Jn +
3
δn,0 ,
2
±
Up Jn± U−1
p = Jn∓p . (4.12)
We can show that
Ur |l, p; ii = |l, p + r; ii. (4.13)
Namely, the spectral flow maps an on-shell chiral primary to another on-shell chiral
primary. This is a general feature. In fact, it is a straightforward calculation to check that
Up G+ (z)U−1 +
p = G (z), (4.14)
r r
k k 0
Up QBRST U−1
p = QBRST − lim pc i∂ X0 + X2 + pηeφ Ψ +Ψ2 .
z→0 2 2
(4.15)
Because the correction term in (4.15) vanishes when acting on an arbitrary on-shell chiral
primaries, the spectral flow Up preserves the on-shell condition in the space of chiral
primaries. One should remark that Up does not transform all the physical states among
themselves. Indeed the physical states which are not the chiral primaries are transformed
into off-shell states by Up .
Turning our attention to the space–time conformal algebra, we obtain
1
{G − +
1 , G 1 } = L0 − J0
− 2
2 2
I r
k
=− i∂ X0 + X2 , (4.16)
2
and this identity is unchanged by the spectral flow. This means that the spectral flows are
closed in the space of the space–time chiral primaries, which is consistent with the above
observation.
Next we present some remarks from the view points of AdS3 /CFT 2 -duality and the long
string theory given in [22]. Although our understanding of them by string theory is not yet
complete, we believe the following remarks are useful to clarify some important aspects of
them.
Tracing back to the procedure of our field redefinitions, one can find that |l, p = 1; 1i
can be identified with the space–time chiral primary states given in [48] (see also [49]). 5
5 They correspond to |ω0 i in the notation of [48], which contain the trivial cohomology in the T 4 sector. We
l/2
can also consider more general space–time chiral primary states that contain higher cohomologies of T 4 , as given
in [48]. But they are not chiral primaries in the sense of worldsheet.
It has the following structure

|l, 1; 1i = lim Ol (z)|0, 1; 1i, (4.17)
z→0
where

l
Ol (z) := V− l ,− l ,0,− l Φl (z) ≡ exp √ iX0 + iX1 + iX2 + ρ Φl (z) (4.18)
2 2 2 2N
naturally corresponds to the chiral operator in the
q light-cone gauge formalism of the long
string theory given in [22]. |0, 1; 1i ≡ exp[− N2 (iX1 + ρ)]ce−φ |0i has the maximal
j -value j = N/2 ≡ k/2 and is the same as the “space–time vacuum” (or “long string
vacuum”) presented in [48]. It satisfies
Ln |0, 1; 1i = 0 (∀n > −1), (4.19)

Gr± |0, 1; 1i = 0 (∀r > −1/2), (4.20)
Jn |0, 1; 1i = 0 (∀n > −1), (4.21)
√ I
k
− i∂X+ |0, 1; 1i = |0, 1; 1i, (4.22)
2
H
where the last line simply means that p ≡ γ −1 ∂γ = 1, and these identities hold up to
BRST-exact terms.
As discussed in [22] (see also [45,46]), the chiral operator Ol corresponds to a non-
normalizable state. Its wave function is divergent near the boundary, where the Coulomb
branch CFT is weakly coupled. On the other hand, thanks to the existence of |0, 1; 1i, the
chiral primary state |l, 1; 1i itself becomes normalizable state vanishing exponentially at
large ρ, as expected from the observation about the Higgs branch tube in [22]. In this sense
the interpretation of |0, 1; 1i as the long string vacuum may be natural.
bl (x) ≡
It may be also useful to define explicitly the space–time chiral primary operator O
P
b
n Ol,n /x
n+l/2 by introducing the vertex operators
I

Obl,n := V l 1 −φ
− ,n,0,− l Φl Ψ + Ψ e ,
0
(4.23)
2 2
bl (x) actually behaves as a

where n runs over all (half-)integers if l/2 is an (half-)integer. O
chiral primary operator with respect to the space–time SCA. For example, we obtain

bl,n = l − 1 m − n O
Lm , O bl,m+n , (4.24)
2
which means that O bl (x) is a primary operator with conformal weight h = l . We can further
2
show that

Obl,n |0, 1; 1i = 0 ∀n > − l (4.25)
2
and also obtain the “operator-state correspondence”

b l |0, 1; 1i = |l, 1; 1i
O (4.26)
l,− 2
(up to the picture changing and an overall constant).

For p > 1 we can consider more general chiral primaries with the higher space–time
U (1)R -charges
b l |0, p; 1i.
|l, p; 1i = Up−1 |l, 1; 1i = Ol (0)|0, p; 1i = O (4.27)
l,− 2
As we observed above (4.9), the spectrum of J0 charge is l +N(p −1), l = 0, 1, . . . , N −2,

p > 1.
We have not yet known the suitable interpretation of the “graviton-like” chiral primary
states |l, p; 2i in the context of AdS3 /CFT 2 correspondence. We only point out that they
do not seem to have the forms such as (4.26), and so it might be plausible to suppose that
they do not have any counterparts in the boundary theory, as long as our identification of
|0, 1; 1i with the space–time vacuum is justified. In any case we will need a further analysis
to give a more definite statement about this problem.
The following aspect may be worthwhile to point out. Here the chiral primaries with the
higher space–time U (1)R -charges appear in the sector with higher light-cone momenta p
(or by taking account of the degrees of freedom of spectral flows). On the other hand, in
the analysis of [24], they correspond to the Zp -twisted sector of the symmetric orbifold
theory, which describes the sector of long string with the “length” p as in the Matrix string
theory [51–53]. It suggests a remarkable correspondence between the spectral flow in the
covariant gauge formalism and the twisted sector of the symmetric orbifold in the light-
cone gauge formalism [22,28].
To address the precise correspondence between them we should work on the second
quantized framework. It is quite reasonable from the viewpoints of AdS3 /CFT 2 corre-
spondence, since the boundary CFT should also contain multi-particle excitations. We shall
now focus on the physical states with positive energies, which should have the light-cone
momenta p > 0 as we found in Section 3. The physical Hilbert space of the first quan-
tized string states, which was studied in our previous analyses, is then decomposed to
L
H = p>0 Hp , where Hp denotes the sector with the light-cone momentum p > 0. The
Hilbert space in the (free) second quantized theory can be roughly written as
∞
M
b=
H (H)⊗n . (4.28)
n=0
(To be precise, we must make some (anti-)symmetrization to assure the correct statistics
in this and the expressions given below.) The second quantized space Hb has a natural
decomposition with respect to the total light-cone momentum
M
Hb= Hbp . (4.29)
p>0
bp is decomposed to the subspaces of the forms

Obviously H
Hp1 ⊗ Hp2 ⊗ · · · ⊗ Hpn ⊗ H0 ⊗ H0 ⊗ · · · ,

!
X
n
p1 > p2 > . . . > pn > 1, pi = p . (4.30)
i=1
The p = 0 Hilbert space H0 only contains tachyons, which are eliminated by the GSO
projection, as was already shown. 6 Therefore we can neglect the H0 factors, and can
explicitly write down
M
p M
bp =
H ⊗ni=1 Hpi . (4.31)
Pn
i=1 pi =p
n=1
Now let us consider the system of Q5 NS5 and Q1 NS1. The NS5 charge Q5 (≡ N)
appears in the worldsheet action of AdS3 -string theory, but Q1 does not. It only appears
in the string coupling, which is stable under the near horizon limit, and hence we cannot
find this effect in the first quantized theory. However, in the second quantized theory, it
is quite natural to identify the NS1 charge Q1 with the total light-cone momentum p in
the expression of H bp . Hence we propose that the physical Hilbert space of this NS5–
NS1 system should be defined as H b . Notice that it has the structure characterized
PQ1
by the various partitions {pi (> 1); i pi = Q1 } which is consistent with the expected
correspondence to the symmetric orbifold theory. Clearly this system can be decomposed
to the subsystems of various long strings with the “lengths” (or “windings”) pi (1 6 pi
P
6 Q1 i pi = Q1 ), as observed in [14,22].
It is interesting to present the spectrum of the “single-particle chiral primaries” in this
framework. Let p 6 Q1 be a positive integer. We obtain the required states as
|l, p; 1i ⊗ |0, 1; 1i ⊗ · · · ⊗ |0, 1; 1i, (4.32)
| {z }
(Q1 − p)-times
which satisfies L0 = 12 J0 = (l + Q5 (p − 1))/2 as expected. Since l, p run over the ranges

0 6 l 6 Q5 − 2, 0 6 p 6 Q1 , this spectrum is completely in agreement with the result of
[24], in which the (multiple) long string CFT was analyzed using the symmetric orbifold
theory. Quite remarkably, this has the upper bound ∼ Q1 Q5 /2 which is expected from the
AdS3 /CFT 2 -duality [50], as already commented in [24].
Notice that there are the missing states corresponding to the J0 -charge Q5 p − 1
(1 6 p 6 Q1 ). They are (formally) written as
|Q5 − 1, p; 1i ⊗ |0, 1; 1i ⊗ · · · = OQ5 −1 (0)|0, p; 1i ⊗ |0, 1; 1i ⊗ · · · , (4.33)
and “OQ5 −1 (z)” is no other than the missing chiral operator discussed in [22], which
should correspond to the cohomology with a delta function support at the small instanton
singularity. In this sense these missing states (4.33) are supposed to be the natural
generalizations to the cases of 1 < p 6 Q1 of the one discussed in [22], in which only
the p = 1 sector was treated.
The above observation implies that the first quantized Hilbert space Hp (p 6 Q1 )
precisely corresponds to the Zp -twisted sector in the SQ1 -orbifold theory as we already
suggested. The relation
6 The fact that the physical Hilbert space of “short string sector” H is vacant is not a contradiction. It rather
0
means that only the non-normalizable physical operators can appear and the operator-state correspondence fails
in the short string sector as suggested in [13].

1 (p) Q5 1
Ln = Lpn + p− δn,0 (4.34)
p 4 p
indeed confirms this identification. On the Hilbert space Hp , the space–time Virasoro
(p)
algebra Ln has the central charge c = 6pQ5 , and the DDF operators Ln generate the
Virasoro algebra with c = 6Q5 . This relation (4.34) is the same as the well-known formula
to define the conformal algebra describing the Zp -twisted sector of the symmetric orbifold.
It is easy to define the tensor product representation of space–time conformal algebra with
L
c = 6Q1 Q5 on the second quantized Hilbert space H bQ1 ≡ P (⊗Hpi ) including
i pi =Q1
the conformal invariant vacuum
⊗Q1
|0, 1; 1i ⊗ |0, 1; 1i ⊗ · · · ⊗ |0, 1; 1i ∈ H1 . (4.35)
| {z }
Q1 times
Such a correspondence, which was essentially suggested in [24,28], is quite expected from
our standpoint as the discrete light-cone theory fitted to the spirit of Matrix string [30–
32,51–53]. Recall that our setup of physical Hilbert space in section 3 allows the action
(p)
of Ln = pLn/p + · · · , and moreover we must impose the “level matching condition”
(p) (p)
L0 − L0 ∈ pZ onto the Hilbert space Hp . These facts are crucial to establish the above
correspondence to the symmetric orbifold.
One should keep in mind the following fact: one can also construct the representation
with c = 6Q1 Q5 on the first quantized Hilbert space HQ1 that is the subspace of H bQ1
describing the single long string with the maximal length Q1 . However, HQ1 cannot
include the conformal invariant vacuum. Recall that L0 |0, p; 1i 6= 0, unless p = 1. More
precisely speaking, we can show that, in our setup of the first quantized Hilbert space
the BRST-invariant state with the properties (4.19), (4.20), (4.21) and non-zero p is
possible only if p = 1, and the solution is unique (up to BRST exact terms and an
overall normalization), |0, 1; 1i, as suggested in [24,48]. This fact leads us to the only one
possibility of the conformal invariant vacuum (4.35). The large Hagedorn density suited
for c = 6Q1 Q5 , which may reproduce the correct entropy formula of black-hole, should
be attached to HbQ , not to HQ , since HQ does not include the vacuum state such that
1 1 1
L0 = 0 (see the discussions given in [45,46,54]).
4.2. Background with space–time N = 2 SUSY
In principle it is not difficult to generalize the above analysis on chiral primaries to more
general superstring vacua with space–time N = 2 SUSY [33,34].
We first give a rather generic argument. Consider superstring theory on AdS3 × S 1
× N /U (1), where N /U (1) is an arbitrary N = 2 SCFT of center 9 − 6/k. As in the
N = 4 case, we can construct two series of chiral primary states from chiral primaries Vj
of conformal weight j/2k in the N /U (1)-sector:

k − j k − j k(p − 1) + j
|j, p; 1i := , , p, − ⊗ |Vj i ⊗ ce−φ |0igh , (4.36)
2 2 2

+ j + 2 j kp + j
|j, p; 2i := Ψ−1/2 2 , − 2 , p, − 2 ⊗ |Vj i ⊗ ce−φ |0igh . (4.37)
They have the light-cone momentum p and the following conformal weight:
1 j + k(p − 1)
L0 |j, p; 1i = J0 |j, p; 1i = |j, p; 1i,
2 2
1 j + kp
L0 |j, p; 2i = J0 |j, p; 2i = |j, p; 2i. (4.38)
2 2
Note that, if we take as N /U (1) an arbitrary N = 2 SCFT of center 9 − 6/k, the
conformal weight h = j/2k of Vj runs within the range 0 6 h 6 3 − 2/k. However, it
is only if 0 6 h 6 1/2 − 1/k that the chiral primary states are in the spectrum allowed
from unitarity and normalizability.
Let us consider a specific example. Take as N /U (1) the N = 2 minimal model which
we denote by MN as before. It was proposed in [55] that the superstring theory on this
background is marginally equivalent to the non-critical superstring theory [37,38] which
is holographically dual to the decoupled theory based on the AN−1 -singular CY 4 . In this
case the criticality condition leads to k = N+1 N
.
Let |Φl i (l = 0, 1, . . . , N − 2) be again the chiral primary states of weight l/2N in the
MN sector. We obtain the following chiral primaries;

N −l N −l N(p − 1) + l

|l, p; 1i := , , p, − ⊗ |Φl i ⊗ ce−φ |0igh , (4.39)
2(N + 1) 2(N + 1) 2(N + 1)

+
l l Np + l
|l, p; 2i := Ψ−1/2 + 1, − , p, −
2(N + 1) 2(N + 1) 2(N + 1)
⊗ |Φl i ⊗ ce−φ |0igh . (4.40)
In this way we have again infinite number of on-shell chiral primaries possessing the
following space–time U (1)R charges;

l N
J0 |l, p; 1i = + (p − 1) |l, p; 1i, (4.41)
N +1 N +1

l N
J0 |l, p; 2i = + p |l, p; 2i. (4.42)
N +1 N +1
Unfortunately, |l, p; 1i is non-normalizable and |l, p; 2i does not satisfy the unitarity
constraints (3.38). Hence we cannot consider the chiral primary states within the physical
Hilbert space. We must only treat these chiral primaries as operators and cannot expect the
operator-state correspondence. Nevertheless, they may be regarded as an important class of
operators in the context of AdS3 /CFT 2 -duality, or more general holographic dualities [55,
56]. In particular the non-normalizable chiral primaries |l, p; 1i (“tachyon-like operators”)
may be important, because they possess the momentum structures which can be regarded
as natural generalizations of those of the scaling operators in the space–time conformal
theory proposed in [55]. Since the light-cone momentum p runs over an infinite range, we
can obtain the infinite tower of space–time chiral operators for each of the chiral operators
in the “matter sector” Φl . This aspect may be interesting, since they look like analogues
of “gravitational descendants” in the theory of two dimensional gravity. We must make
further studies to gain more precise insights about these objects. In addition, the roles of
the graviton-like operators |l, p; 2i are again unclear. More detailed argument for them will
be surely important, although it is beyond the scope of this paper.
5. Conclusions and discussions
In this paper we have studied the spectrum of the physical states in string theory on AdS3
based on a free field realization. We have found that the system is quite simply described
as a linear dilaton theory with a light-like compactification, which we called as “discrete
light-cone Liouville theory”. Our key idea is to utilize the DDF operators according to the
traditional approach to string theory on flat backgrounds. We have two independent sets
(p) (p)
of DDF operators; An , L̃n . This situation is similar to the non-critical string, although
we started with the critical string on AdS3 × N background. In fact, we can easily find
(p) (p)
that a suitable linear combination of An and L̃n gives the “longitudinal DDF operator”
utilized in [57].
Regarding the system as a free theory with no Liouville interaction term (screening
charge term), the physical spectrum contains only the principal series j = 12 +is as asserted
(p) (p)
in [6,7,17], and the physical states are generated by the DDF operators {An , L̃n }
mentioned above.
However, once we turn on the Liouville potential, the story becomes rather non-trivial. In
this interacting theory the translational invariance along the Liouville direction is broken.
The physical Hilbert space is expected to be spanned only by the SL(2; b R) currents, rather
+ −
than the whole oscillators of the string coordinates ρ, Y , Y , because the interaction term
b
commutes only with the SL(2; R) currents. In this situation the spectrum generating algebra
(p) (p) (p)
becomes {Ln } in place of {An , L̃n }, and the physical states possessing the imaginary
ρ-momenta (j ∈ R) are also allowed. Physically they correspond to the bound string states
of [18] that are trapped inside the AdS3 -space.
It may be worthwhile to mention that only the physical Hilbert space as the interacting
Liouville theory may be consistent with the microscopic evaluation of the black hole
(p) (p) (p)
entropy. In the free system the DDF operators should be {An , L̃m } and L̃m (which
are identified with the Virasoro operators in N -sector under the light-cone gauge) have a
small central charge c = 23 − 6k . (An important discussion related to such a counting of
physical states was given in [54].) On the other hand, after turning on the Liouville potential
(p)
we claimed that the full Virasoro generators {Ln } are well-defined (and it is also crucial
(p)
that {An } should be discarded). Taking further account of the second quantized Hilbert
space they seem to generate sufficiently many physical states with the Hagedorn density
that can reproduce the correct entropy. We would like to study this problem in more detail
elsewhere.
In the study of superstring examples, we have presented the complete set of on-shell
chiral primaries. There exist infinite number of such operators and the spectral flows
naturally act on them. Moreover, to describe the well-known Q5 (≡ k) NS5–Q1 NS1
system we made use of the second quantized framework. The Hilbert space of the multiple
L
long string system was given as the form H bQ1 = P
pi =Q1 (⊗i Hpi ), where Hp denotes
i
the first quantized physical Hilbert space of the sector with the light-cone momentum
p(> 0). This space reproduce the spectrum of chiral primaries same as that given by the
symmetric orbifold theory [24], and among other things, we have successfully obtained
the upper bound ∼ Q1 Q5 /2 consistent with the prediction of AdS3 /CFT 2 correspondence
[50].
It may be also worth pointing out that our reformulation of superstring on AdS3
× S 1 × N /U (1) has the same field contents as those of the non-critical string that is
holographically dual to a singular Calabi–Yau compactification (especially, the cases of
CY 4 ) proposed in [55]. The only difference between these models is the existence/absence
of the light-like compactification. In [55] it was discussed that these two backgrounds
can be interpolated by some marginal deformation. It may be an interesting problem to
clarify the rigid correspondence between them. In particular, our analyses on general chiral
operators in Section 4 will be readily generalized to the cases of such non-critical string
theories.
Acknowledgement
Y.S. would like to thank I. Bars and Y. Satoh for helpful discussions.
The work of K.H. is supported in part by Japan Society for Promotion of Science under
the Postdoctoral Research Program (]12-02721). The work of Y.S. is supported in part by
Grant-in-Aid for Encouragement of Young Scientists (]11740142) and also by Grant-in-
Aid for Scientific Research on Priority Area (]707) “Supersymmetry and Unified Theory
of Elementary Particles”, both from Japan Ministry of Education, Science, Sports and
Culture.
References
[1] J. Balog, L. O’Raifeartaigh, P. Forgacs, A. Wipf, Nucl. Phys. B 325 (1989) 225.
[2] P. Petropoulos, Phys. Lett. B 236 (1990) 151.
[3] N. Mohammedi, Int. J. Mod. Phys. A 5 (1990) 3201.
[4] I. Bars, D. Nemeschansky, Nucl. Phys. B 348 (1991) 89.
[5] S. Hwang, Nucl. Phys. B 354 (1991) 100.
[6] I. Bars, Phys. Rev. D 53 (1996) 3308, hep-th/9503205.
[7] Y. Satoh, Nucl. Phys. B 513 (1998) 213, hep-th/9705208.
[8] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.
[9] S. Gubser, I. Klebanov, A. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[10] E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[11] O. Aharony, S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, hep-th/9905111, for a review and a
complete list of references.
[12] A. Giveon, D. Kutasov, N. Seiberg, Adv. Theor. Math. Phys. 2 (1998) 733, hep-th/9806194.
[13] J. de Boer, H. Ooguri, H. Robins, J. Tannenhauser, JHEP 9812 (1998) 026, hep-th/9812046.
[14] D. Kutasov, N. Seiberg, JHEP 9904 (1999) 008, hep-th/9903219.
[15] D. Bernard, G. Felder, Commun. Math. Phys. 127 (1990) 145.
[16] J. Evans, M. Gaverdiel, M. Perry, Nucl. Phys. B 535 (1998) 152, hep-th/9806024.
[17] I. Bars, C. Deliduman, D. Minic, hep-th/9907087.
[18] J. Maldacena, H. Ooguri, hep-th/0001053.

[19] J. Brown, M. Henneaux, Commun. Math. Phys. 104 (1986) 207.
[20] A. Strominger, JHEP 9802 (1998) 009, hep-th/9712251.
[21] J. Maldacena, J. Michelson, A. Strominger, JHEP 9902 (1999) 011, hep-th/9812073.
[22] N. Seiberg, E. Witten, JHEP 9904 (1999) 017, hep-th/9903224.
[23] E. Del Giudice, P. Di Vecchia, S. Fubini, Ann. Phys. 70 (1972) 378.
[24] K. Hosomichi, Y. Sugawara, JHEP 9907 (1999) 027, hep-th/9905004.
[25] M. Wakimoto, Commun. Math. Phys. 104 (1986) 605.
[26] D. Friedan, E. Martinec, S. Shenker, Nucl. Phys. B 271 (1986) 93.
[27] H. Ishikawa, M. Kato, Phys. Lett. B 302 (1993) 209.
[28] M. Yu, B. Zhang, Nucl. Phys. B 551 (1999) 425, hep-th/9812216.
[29] S. Mizoguchi, JHEP 0004 (2000) 014, hep-th/0003053.
[30] L. Susskind, hep-th/9704080.
[31] A. Sen, hep-th/9709220.
[32] N. Seiberg, Phys. Rev. Lett. 79 (1997) 3577, hep-th/9710009.
[33] A. Giveon, M. Roček, JHEP 9904 (1999) 019, hep-th/9904024.
[34] D. Berenstein, R.G. Leigh, Phys. Lett. B 458 (1999) 297, hep-th/9904040.
[35] Y. Kazama, H. Suzuki, Nucl. Phys. B 321 (1989) 232.
[36] Y. Kazama, H. Suzuki, Mod. Phys. Lett. A 4 (1989) 235.
[37] D. Kutasov, N. Seiberg, Phys. Lett. B 251 (1990) 67.
[38] D. Kutasov, Lecture given at ICTP Spring School, Trieste, hep-th/9110041.
[39] T. Eguchi, Y. Sugawara, Nucl. Phys. B 577 (2000) 3, hep-th/0002100.
[40] I. Antoniadis, S. Ferrara, C. Kounnas, Nucl. Phys. B 421 (1994) 343, hep-th/9402073.
[41] A. Giveon, D. Kutasov, JHEP 9910 (1999) 034, hep-th/9909110.
[42] P. Breitenlohner, D. Freedman, Phys. Lett. B 115 (1982) 197.
[43] P. Breitenlohner, D. Freedman, Ann. Phys. NY 144 (1982) 249.
[44] Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 240 [FS12] (1984) 312.
[45] D. Kutasov, N. Seiberg, Nucl. Phys. B 358 (1991) 600.
[46] N. Seiberg, Prog. Theor. Phys. Suppl. 102 (1990) 319.
[47] Y. Sugawara, Nucl. Phys. B 576 (2000) 265, hep-th/9909146.
[48] K. Hosomichi, Y. Sugawara, JHEP 9901 (1999) 013, hep-th/9812100.
[49] D. Kutasov, F. Larsen, R. Leigh, Nucl. Phys. B 550 (1999) 183, hep-th/9812027.
[50] J. Maldacena, A. Strominger, JHEP 9812 (1998) 005, hep-th/9804085.
[51] R. Dijkgraaf, E. Verlinde, H. Verlinde, Nucl. Phys. B 500 (1997) 43, hep-th/9703030.
[52] L. Motl, hep-th/9701025.
[53] T. Banks, N. Seiberg, hep-th/9702187.
[54] S. Carlip, Class. Quant. Grav. 15 (1998) 3609, hep-th/9806026.
[55] A. Giveon, D. Kutasov, O. Pelc, JHEP 9910 (1999) 035, hep-th/9907178.
[56] O. Aharony, M. Berkooz, D. Kutasov, N. Seiberg, JHEP 9810 (1998) 004, hep-th/9808149.
[57] R. Brower, Phys. Rev. D 6 (1972) 1655.
Toroidal compactification in string theory

from Chern–Simons theory
P. Castelo Ferreira ∗ , Ian I. Kogan 1 , Bayram Tekin 2
Department of Physics — Theoretical Physics, University of Oxford, 1 Keble Road, Oxford OX1 3NP, UK
Received 19 April 2000; revised 14 June 2000; accepted 21 June 2000
Abstract
A detailed study of the charge spectrum of three-dimensional Abelian Topological Massive
Gauge Theory (TMGT) is given. When this theory is defined on a manifold with two disconnected
boundaries there are induced chiral Conformal Field Theories (CFT’s) on the boundaries which can
be interpreted as the left and right sectors of closed strings. We show that Narain constraints on
toroidal compactification (integer, even, self-dual momentum lattice) have a natural interpretation in
purely three-dimensional terms. This is an important result which is necessary to construct toroidal
compactification and heterotic string from Topological Membrane (TM) approach to string theory.
We also derive the block structure of c = 1 Rational Conformal Field Theory (RCFT) from the three-
dimensional gauge theory.  2000 Elsevier Science B.V. All rights reserved.
PACS: 11.10.Kk; 11.15.Tk; 11.25.Hf; 11.25.Sq

Keywords: TM; TMGT; Strings; CFT; Compactification
1. Introduction
Inherent to all known perturbative string theories are the two-dimensional conformal
field theories. Moreover, keeping in mind the fact that a given string theory can be
considered as a two-dimensional sigma model (quantum field theory) which has spacetime
as its target space, it is hard to over emphasize the importance of 2D conformal field
theories. Most of the claimed beauties of string theories undoubtly reside in the fact that
one has free and conformal theories on the worldsheet. For the closed string case, which
we exclusively work within this paper, an important feature of the the worldsheet theory
is that left and right moving degrees of freedom decouple from each other. One can then
formulate the theory by considering two chiral conformal field theories, holomorphic and
∗ Corresponding author. E-mail: pcastelo@thphys.ox.ac.uk

1 kogan@thphys.ox.ac.uk
2 tekin@thphys.ox.ac.uk
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 0 7 - 7
168 P.C. Ferreira et al. / Nuclear Physics B 589 (2000) 167–195
anti-holomorphic, living on the same two-dimensional surface Σ. The features of the

non-chiral CFT, and the corresponding string theory, depend on how one combines the
holomorphic and anti-holomorphic theories. This part of the story is well known. In this
paper we continue to develop the idea of topological membrane approach to string theory
which was first set forward in [1,2] and worked out in many subsequent papers, where the
left and right moving sectors of closed strings live on two different surfaces ΣL and ΣR .
In other words, we are giving a finite physical thickness (be it spatial or temporal) to the
string worldsheet which becomes now a three-manifold.
Topological membrane idea is based on the results of Witten [3] and Moore and
Seiberg [4] (see also [5–8]), Chern–Simons theory defined on a three-manifold with a
boundary is equivalent to 2D conformal field theory, gauged chiral WZWN model, living
on the boundary. This is essentially the first example of a theory where the currently
popular holographic principle works [9,10]. The chain of reasoning is as follows: the
3D bulk theory is equivalent to the boundary conformal theory and the later underlines
a 10D string theory. So the hope is that at the very fundamental level of string theory lies a
three-dimensional quantum field theory. Although it is rather speculative at this stage this
formulation might provide us with an off-shell formulation of string theory as well as new
relations between different types of strings from a more fundamental theory.
The actual story is not as simple as it is laid above. Chern–Simons theory in the bulk
is not enough to reproduce string theory. See [2] for more complete details. Besides
the fact that we must induce all aspects (moduli, Liouville fields, gravitational dressing,
etc.) of 2d gravity [11–15] there is a problem of obtaining the spectrum of toroidal
compactification [16,17] and heterotic string [18,19] from the original pure Chern–Simons
theory [3,4]. This is due to the impossibility to have different right and left charge spectrum
and to induce holomorphic CFT in one boundary without inducing anything in the other.
This problem is solved by adding a Maxwell term in the action and considering CS as
the low energy limit of TMGT [20–22]. In this way monopoles emerge in the theory and
induce new processes [23] which lead to local charge non-conservation that permit a right
charge to have a different value than a left charge. The important point is that bulk effects
(monopoles) allow us to have winding modes in the boundary theory. Imposing suitable
boundary conditions [24,25] one can construct heterotic strings in this framework.
With the Maxwell term the bulk theory of course becomes non-topological and the
holographic principle partially works. The spectrum of the topological massive gauge
theory has a mass gap between its excited states and the degenerate ground state. But
still the Hilbert subspace of the ground state of Abelian TMGT is exactly equivalent to the
Hilbert space of the full CS theory. Moreover in order to define the quantum CS theory
the Maxwell term has to be introduced as an ultraviolet regulator, which is set to zero after
renormalization [26]. So in any case one needs the Maxwell term.
In [27] the charge non-conservation induced by monopoles was discussed and and these
ideas were applied to T-duality and mirror symmetry in [28]. Nevertheless the problem
of building the correct left–right spectrum has never been properly solved. The main goal
of this paper is to determine the charge lattice structure allowed by TMGT. Essentially
we show that, upon imposing suitable boundary conditions, the theory demands a charge
P.C. Ferreira et al. / Nuclear Physics B 589 (2000) 167–195 169
lattice which is exactly of the form of string theory momentum lattice. From the non-
perturbative dynamics of three-dimensional gauge theory we derive the Narain lattice
spectrum. We consider here both bosonic and heterotic string theories.
In Section 2 we start with a short review of 3D TMGT and some related topics necessary
for the present work, namely some results in conformal theories and string theory. Next we
move on to the main part of this paper, Section 3, a model which describes the dynamics
of charges propagating in the 3D bulk theory is built. We arrive to some well know results
in string theory but derived purely from the dynamics of the bulk 3D theory. Namely the
mass spectrum of toroidally compactified closed string theory emerges. In Section 4 we
study a relevant issue of gluing both 2D boundaries of the 3D manifold in order to get
a single conformal field theory in the boundary. In Section 5 the underlying structure of
conformal block structure of the c = 1 compactified bosonic RCFT and the corresponding
fusion rules are found as a result of monopole-instanton induced interactions in the bulk.
Finally in Section 6 the spectrum of heterotic string and possible backgrounds are rederived
in the light of these new results.
2. A short review on TMGT
There are several ways to derive CFT from CS theory. We choose the path integral
approach first suggested by Ogura [29] (see also [30–32]). In this section we review some
features of the topologically massive gauge theory defined on a three-dimensional flat
manifold with a-boundary. In order to clarify some points we present some arguments
derived from canonical formalism. We present a list of the induced chiral boundary
conformal field theories and we carefully analyze the Gauss law structure of the gauge
theory with compact gauge group U (1) given by the action
Z √
−g(3) µν k µνλ
S = d 2 z dt − F Fµν + Aµ ∂ν Aλ , (1)
4γ 8π
M
where M = Σ × [0, 1] and Σ is a 2-dimensional compact Euclidean manifold with a
complex structure denoted by (z, z̄). The time-like coordinate takes values in the compact
domain t ∈ [0, 1]. The indices run over µ = 0, i with i = z, z̄.
Under an infinitesimal variation of the fields A → A + δA the action changes by
Z √ "Z #t =1
−g(3) k µλν
δS = ∂µ F +
µν
∂µ Aλ δAν − i
Π δAi , (2)
γ 4π
M Σ t =0
where
1 0i k ij
Πi = F − ˜ Aj (3)
γ 8π
is the canonical momenta conjugate to Ai . Note that the 2d antisymmetric tensor ˜ ij is
induced by the 3d antisymmetric tensor and metric
− 0ij
˜ ij = √ . (4)
−g(3)
When referring to the usual 2d antisymmetric tensor we use the notation ij without the
tilde.
In order the theory to have classical extrema it is necessary to impose suitable boundary
conditions for which the second term in the variation of the action vanishes. Let us assume
that the boundary of M has two pieces, which are Σ(t =0) = ΣL and Σ(t =1) = ΣR . L and
R denote left and right. On each of the boundaries, up to gauge transformations, one can fix
one or both fields Az and Az̄ . In doing so one should add an appropriate boundary action
SB = SBL + SBR such that the new action S + SB has no boundary variation, hence well
defined classical extrema. Note that upon canonical quantization we have to impose the
corresponding equal time commutation relations of Π and A
¯ ¯
[Π i (z), Aj (z0 )] = g i j δ (2) (z − z0 ). (5)
Our convention for the metric is g zz̄ = g z̄z = 2. So fixing Az we fix Π z̄ as well and the
same holds for the Az̄ and Π z components.
On each components of Σ the possible choices of boundary conditions and the boundary
actions are
boundary conditions left bound. action right bound. action
N: δAz = δAz̄ = 0, SBL = 0, SBR = 0,

Z Z
C: δAz̄ = 0, SBL = − Π z Az , SBR = Π z Az , (6)
ΣZR ΣZL
C̄ : δAz = 0, SBL = − Π z̄ Az̄ , SBR = Π z̄ Az̄ .

ΣR ΣL
There are nine allowed choices: NN , NC, CC, C C̄, and so on. The first letter denotes
the type of left boundary conditions and the second one the right type. The N boundary
condition stands for Non-Conformal or Non-Dynamical, C stands for Conformal and C̄
for anti-Conformal and are related to the kind of CFT we obtain in the boundaries when
we choose them. Note the importance of the F 2 term, it gives the theory four independent
canonical coordinates, opposed to the two of the pure Chern–Simons theory (where one
of the A’s is canonically conjugate to the other). This fact allow us to fix one of the A’s
and corresponding Π without fixing the other two allowing us to have different boundary
conditions in opposite boundaries. For further details we refer the reader to [24,25]. This
topic will be addressed again in Section 4.
By Faddeev–Popov procedure we can fix the gauge Aµ = Āµ + ∂µ Λ. The action and
path integral factorizes as
Z Z

Z = DĀ1FP δ F (Ā) ei(S̄+S̄B )[Ā] DλDχeiδ S̄L [χ]+iδ S̄R [λ] , (7)
where in the boundaries the gauge parameters Λ(t = 0) = χ and Λ(t = 1) = λ become
dynamical degrees of freedom which decouple from the bulk theory.
Now let us find the allowed charges in this theory. The compactness of the gauge group
has to be taken into account and it turns out to be of fundamental importance. Defining the
electric and magnetic fields as
1 0i k ij
Ei = F = Πi + ˜ Aj ,
γ 8π (8)
B = ij ∂i Aj .
(For the sake of notational simplicity we have omitted the complex conjugation on the
indices.) The commutation relations read
i k
E (z), E j (z0 ) = −i ij δ (2) (z − z0 ),
i 4π (9)
E (z), B(z0 ) = −i ij ∂j δ (2) (z − z0 ).
If there is an external charge the Gauss law is
k
∂i E i + B = ρ0 . (10)
4π
In the quantum theory this equation needs to be satisfied by the physical states. So
following [27] we can define the generator of time independent gauge transformations U
( Z )
k
U = exp i Λ(z) ∂i E i + B − ρ0 . (11)
4π
Σ
Σ stands for a generic fixed time slice of M. Since the gauge group is compact Λ is
identified with an angle in the complex plane such that
ln(z) = ln|z| + i(Λ(z) + 2πn),
(12)
∂i Λ(z) = −ij ∂j ln|z|,
where the second equation follows from the Cauchy–Riemann equations. This last
condition on Λ will restrict the physical Hilbert space in the compact theory [33–36].
Let us define a new operator
( Z )
k
V (z0 ) = exp −i d 2 z E i + ij Aj ik ∂k ln|z0 − z| − Λ(z0 − z)ρ0 . (13)
4π
Σ
The physical states of the theory must be gauge invariant (under U operator) as well
eigenstates of this new local operator. Using the identity ∂k ∂k ln|z| = 2πδ(z) and the
commutation relations (9) for E and B we obtain

B(z), V n (z0 ) = 2πnV n (z0 )δ (2) (z − z0 ). (14)
This means that the operator V creates a pointlike magnetic vortex at z0 with magnetic flux
Z
B = 2πn, n ∈ Z. (15)
Σ
As stated before the F 2 term is fundamental for the existence of these tunneling
processes that hold local charge non-conservation, see [23,33–36] for further details.
Instantons in three dimensions are the monopoles in four dimensions. So in the rest of
the paper we will make use of the terms instanton and monopole without any distinction.
Using the functional Schrödinger representation Π i = iδ/δAi and imposing the
condition that phase acquired by a physical state under a gauge transformation be single
valued we obtain the charge spectrum
Z
k k
q =m+ B = m + n, m, n ∈ Z. (16)
8π 4
Σ
As it can be seen from the above formula for a generic, non-integer, value of k the allowed
charges are non-integers. In principle in the compact Abelian theory one expects that the
charges are quantized as it happens in the Maxwell theory. But the existence of the Chern–
Simons term changes this picture, the charges are quantized trough m and n dependence but
we consider k itself not quantized. As will be explained below this fact is only compatible
(and demands) with the existence of a non-compact gauge sector in the theory.
A charge q propagating in the bulk can interact with one monopole with flux (15)
changing by an amount
k
1q = n. (17)
2
The path of a charge in the bulk can be thought as a Wilson line. The phase induced by
the linking of two Wilson lines carrying charges q1 and q2 is

2
hWq1 Wq2 i = exp 2πi q1 q2 l , (18)
k
where l is the linking number between the lines. The above computation was done in the
limit of vanishing Maxwell term and with the assumption that there is no self-linking for
the individual Wilson lines. The connection between the boundary CFT’s and the bulk
theory can be achieved by noting that the bulk gauge fields become pure gauges in the
boundary and the Wilson lines in the boundary are none other than the vertex operators in
ΣL and ΣR with momenta q
n Z o

VR,qR = exp −iq Aν dx ν = exp{−iqR λ},
n Z o ΣR (19)

VL,qL = exp −iq Aν dx ν = exp{−iqLχ}.
ΣL
The conformal dimensions of these vertices is
qR2
∆R = ,
k (20)
qL2
∆L = .
k
k appears because the induced action in the boundary is that of a chiral string action
multiplied by k. To see the Wilson line and vertex operator correspondence let us obtain
the above result from the bulk theory. Consider two Wilson lines carrying charges q and
−q propagating from one boundary to the other corresponding to two vertex insertion in
the boundary with momenta q and −q, see Fig. 1.
Fig. 1. Two charges propagating through the bulk are represented as two Wilson lines. By a rotation
of one charge around the other we induce one linking (l = 1) in the bulk.
Two point correlation of these two vertices follows as usual

1
VL,q (z1 )VL,−q (z2 ) = ,
z12 2∆

1 (21)
VR,q (z̄1 )VR,−q (z̄2 ) = ¯ .
z̄212∆
A rotation of one charge (vertex) around the other in one boundary induces a phase of
−4πi∆ in (21). In the bulk this rotation induces one linking of the Wilson lines (l →
l + 1), so (18) gives an Aharonov–Bohm phase change of −4πiq 2/k. Identifying these two
phases we conclude that (20) follows. Let us recall that in terms of the three-dimensional
theory Polyakov [37] pointed out that ∆ is the transmuted spin of a charged particle which
exists because of the interaction with Chern–Simons term. So we identify the conformal
dimension of the boundary fields with the transmuted spin of the bulk charges. For later
use let us generalize this result two three charges q1 , q2 and q3 = −q1 − q2 . We will always
take the net charge to be zero. Three point correlation of three vertices reads

1
Vq1 (z1 )Vq2 (z2 )Vq3 (z3 ) = +∆ −∆ +∆
, (22)
z12 ∆1 2 3 z13 ∆1 3 −∆2 z ∆2 +∆3 −∆1
23
where the charge-conformal dimension identifications are
2
∆1 + ∆2 − ∆3 = − q1 q2 ,
k
2
∆1 + ∆3 − ∆2 = q1 (q1 + q2 ), (23)
k
2
∆2 + ∆3 − ∆1 = q2 (q1 + q2 ).
k
The anti-holomorphic three point correlation follows trivially. It is clear that if any one
of the vertices is rotated around one of the other two; the phase factor induced in the three-
point function will be the same as (18) which comes from the linking of Wilson lines in
the bulk. From the point of view of CFT the sum of conformal factors must be integer in
order to have single valued OPE’s. This result is going to be derived from the bulk theory
in the next section independent of the boundary. By a simple argument let us show that
we can only have integer conformal dimensions for each vertex. At the level of the two
point correlation functions we can have two integer ∆’s or two half-integer ∆’s. For three
point functions it is necessary to have three integer ∆’ or two integer and one half-integer
∆’s. One can then consider the four point function and decompose it to two pairs which
are well separated from each other. In this case the conformal dimensions can be integers
and half-integers. Suppose that at least two of the vertices have half-integer conformal
dimensions. Then one vertex from a pair can be adiabatically moved to the other pair and
a three point function with a total half-integer conformal dimension will be formed. But
this is not permitted as argued above. For this reason we exclude the existence of vertices
with half-integer conformal dimension. The same arguments follow for the charges from
the point of view of the 3d bulk theory.
Let us remember that we have two independent chiral conformal field theories on two
different surfaces up to this point. In order to get a non-chiral CFT (as in string theory)
we have to identify in some way the two boundaries. The most obvious way to define
non-chiral vertex operators is
VqR ,qL (z, z̄) = VR,qR (z)VL,qL (z̄) (24)
with the conformal dimension

1 2
∆ R + ∆L = qR + qL2 . (25)
k
Note that we have a little bit of difference in the nomenclature. In most of the literature ∆R
and ∆L are called conformal weights and ∆R − ∆L is called the spin and the sum (25) is
called the conformal dimension. Here we call ∆R and ∆L conformal dimensions as well
when we refer to the chiral vertices. From now on we will change our notation to ∆ = ∆R
and ∆¯ = ∆L , the same for the charges.
For generic k and q’s, ∆ in (20) and the sums in (23) are not integers. This is not a
problem, actually is very welcome since it is simply the statement that we need something
else in our theory, a new non-compact gauge group sector. Or in terms of string theory,
non-compactified dimensions. So in addition to the action (1) with compact gauge group
U (1) we will consider the following action with non-compact gauge group U (1)D
Z
1 µν M k 0 δMN µνλ M
SD = d 2 z dt − 0 fM fµν + aµ ∂ν aλN , (26)
4γ 8π
M
where N and M run from 0 to D − 1. We could take generally a coupling tensor KMN 0 ,
0 0
for the purposes of this work is enough to consider the above KMN = k δMN . We will
consider the actions (1) and (26) together. A generic (meaning that not necessarily the
tachyon operators we considered before) non-chiral vertex of the boundary CFT is of the
following form
Y Y 0
s0 ri M
s
∂z ζ ∂z̄ ζ ri M
∂z Ω ∂z̄ Ω exp −i (q + q̄)ζ + pM Ω M . (27)
i i
The fields ζ and Ω M correspond to the gauge parameters of A and a M , respectively. Levels
of the vertex operators are integers defined as
X
L = s+ ri ,
X
i (28)
L̄ = s 0 + ri0 .
i
The exponential part may be represented as the bulk Wilson line propagating from
boundary to boundary
n Z o
WD = exp −i qAν + pM aνM dx ν . (29)
The mass shell condition for the boundary vertex is

pM pM = −mass2 . (30)
The mass spectrum in CFT’s is built out of the allowed values for the conformal dimension
of the fields or the operators. In particular the vertex operators (27) have conformal
dimensions
q 2 p2
∆= + 0 + L,
k k (31)
q̄ 2 p 2
∆¯ = + 0 + L̄.
k k
Due to conformal invariance one has ∆ = ∆¯ = 1. To get the usual string theory
normalization one should replace k 0 = 4/α = k/R 2 and take the average of both equations
(31)
(q 2 + q̄ 2 )k 0 k 0
mass2 = −p2 = + (L + L̄ − 2)
2k 2
m2 R 2 n2 2
= 2 + 2 + (L + L̄ − 2). (32)
R α α
Subtracting one equation from the other in (31) one gets
q 2 − q̄ 2
+ L − L̄ = nm + L − L̄ = 0. (33)
k
Let us call (32) and (33) the mass shell condition and spin condition, respectively. For
further details see [38]. We used the explicit forms of the charges corresponding to
momentum spectrum of string theory, in our normalization
k
q = m + n,
4
k
q̄ = m − n. (34)
4
But remember that the allowed charges in the theory are of the q-form. In principle q̄ is not
related to q and it should be of the form q̄ = m0 + n0 k/4. As it will be explained in detail in
the rest of the paper we have the following picture in mind. A charge q is inserted in one of
the boundaries and it goes through the bulk interacting with the gauge fields until it reaches
the other boundary. During its journey through the bulk its interactions with (in)finite many
monopole-instantons induce a change of its charge 1q = Nk/2. The the charge emerges
in the other boundary as q̄ = q + Nk/2 = m + (n + 2N)k/4. If N = −n we have q̄ =
m − nk/4, this is what we are looking for. But why N must be −n? How do the monopoles
know a priori they have to interact by this amount with the initial charge q? In principle it
could exist any other process holding N 6= −n that would generate some charge q̄ leading
us in disaster. We would get the pair (q, q̄) with no correspondence with any physical
sensible momentum pair (PR , PL ) of string theory. So we have to show that the full 3d-
amplitudes q → q̄ in the theory corresponding to unwanted processes q̄ 6= m − nk/4 are
vanishing! This crucial property shows that seemingly independent chiral field theories
on different boundaries are actually related when the non-perturbative excitations in the
bulk are taken into account. This fact is closely related to considering the Maxwell Chern–
Simons theory instead of the pure Chern–Simons and the presence in the bulk of monopole
processes. In the next section we shall prove that the theory indeed has this property.
3. Propagation in the bulk
Our aim in this section is to investigate the changes that a charge undergo when it travels
from one boundary to the other. As it will be shown the bulk theory imposes, independent
of what the boundary conditions are, restrictions for the charges. We are assuming irrational
k in this section and in Section 5 the case of rational k will be addressed.
We are dealing with a closed manifold M with zero total charge on it. The instantons can
induce a charge non-conservation locally only. So let us insert a pair of charges Q1 = −Q2
in one boundary, say ΣL . They will travel through the bulk and emerge in the opposite
boundary, ΣR as two new charges Q̄1 = −Q̄2 . In their paths the charges can interact with
the bulk fields. Their paths can link but we assume there is no self-linking and the charges
do not interact with each other.
Translating this to a more formal language. We have two open Wilson lines, each of
them have their ends attached to different boundaries. The Wilson line represents the path
of a charged particle. Leaving aside the perturbative interactions in the bulk, there are
charge non-conserving interactions with the monopole-instantons. We will represent these
interactions as insertions of instantons along the Wilson lines. Each instanton on the path
of the particle changes the sign of the particle as mentioned before. So the physical picture
is that we have a Wilson line on which the charge varies from point to point. But this
variation is not random and it is induced by instantons. One can in principle consider
chains of Wilson lines which is essentially the same picture. In between two insertions the
two lines (minimal Wilson line segments with constant charges on them) can link to each
other changing the correlator of the lines according to (18). So we assume that we can
separate the insertions of instantons from the linkings of the lines. Or an other way to put
is that insertions of instantons are done at the ends of the Wilson line segments.
A generic propagation of two such charges with j instanton interactions and j + 1 sets
of linkings in the bulk is represented as a planar diagram in Fig. 2. The insertions are
represented by crosses and the linkings by horizontal lines. The full physical picture is
obtained by taking the limit j → ∞ which allows infinitely many instanton insertions on
a line. Since we assume global charge conservation, Q̄1 − Q1 = Q2 − Q̄2 and the upper
P
charges are the result of all the monopole contributions Q̄1 = Q1 + k/2Ni . So we have
to impose the condition
X
j X
j
Ni = N = − Ni0 . (35)
i=1 i=1
Each pair of charges can be formally interpreted as a quantum state (Q1 , −Q1 ) and its
propagation from one boundary to the other boundary can be considered as a time evolution
to the state (Q̄1 , −Q̄2 ). Then for each initial Q1 and final Q̄1 we can built a transition
coefficient C[Q1 , N] which is to be interpreted as the square root of probability.

k k
(Q1 , −Q1 ) → C[Q1 , N] Q1 + N, −Q1 − N , (36)
2 2
where we already make use of condition (35). C is a function of Q1 and N only but in
order to find it we need to take into account all the possible processes in the bulk, monopole
interactions and linkings. To be able to extract some information about the boundary let us
insert a new variable l, which denotes the total linking number. Up to normalization factors
C can be written as
Fig. 2. Two and three charges propagation through the bulk. The N ’s, M’s, and N 0 ’s represent the
flux of instantons and the l’s, d’s and k’s the linking number of the Wilson lines between insertions.
∞ Z
X
C[Q1 , N] = dl δ(l˜ − l) c[Q1 , N, l] (37)
˜
l=−∞
with the definition

X
j
l = l0 + li . (38)
i=1
It is now about time to find how the infinitely many bulk processes contribute to this
coefficient. We know that interacting with the monopoles labeled by i the pair of charges
will go through the following transition (q, q 0 ) → (q + k/2Ni , q 0 + k/2Ni0 ) according
to (17). Between each monopole insertions the linking will induce a phase change of
4πiqq 0li /k according to (18). The total change of the charges is already taken in account
through N dependence on C. But the processes that we are summing over must be selected
in order to fulfill the requirement that the total monopole contribution is N and that the
total linking number is l. The coefficient c is then the sum of the phases
X X
c[Q1 , N, l] = exp 2πiΦ[Na , Nb0 , li , l0 ] , (39)
Na ,Nb0 li ,l0
where the sums over Na and Nb0 stand for all the allowed configurations obeying (35) and
the sum over li and l0 for the ones obeying (38). The phase, Φ, for each process is simply
the sum of the several phases induced at each step i (see Fig. 2) along the propagation in
the bulk
( ! ! )
2 X j
kX
i
kX 0
i
Φ= −Q1 l0 +
2
Q1 + Nb −Q1 + Na li . (40)
k 2 2
i=1 b=1 a=1
Writing the last term of the sum in i using (35), i.e., for i = j , and expanding the products
we obtain
( 2 !
2 X j
k
Φ = −Q1 l0 − Q1
2 2
li + Q1 − Q1 + N
2
lj
k 2
j −1
i=1 ! ) (41)
kX X X X
i i i
k
+ Q1 (Na0 − Na ) + Na0 Nb li .
2 2
i=1 a=1 a=1 b=1
Note that the phase is no longer Nj dependent, we replaced it by N dependence. We further

need to get the l dependence of c. In the following discussion we are going to investigate
the cases for which the coefficient is non-vanishing. Bearing in mind that we are dealing
with formal series (which are divergent in general) different ways of factorizing the sums
is safer. So we will depict three different factorizations below, c will factorize into a l
dependent phase and an independent phase factor. Unless otherwise stated from now on
the index i will run from 1 to j − 1.
P
The most obvious way to factorize the sum is to eliminate l0 = l − li − lj . Then (39)
factorizes as

2
c[Q1 , N, l] = exp −2πi Q21 l
k
X∞ X∞ jY−1 X
∞ ∞
X (42)
× exp 2πiΦ00 [Na , Nb , li , lj ] .
Na =−∞ Nb0 =−∞ i=1 li =−∞ lj =−∞
The indices a and b run from 1 to j − 1. Since the phase Φ00 is the sum of several phase
changes φ we can rewrite the second factor as
−1 X ∞
! ∞
!
X jY X
exp{2πiφi li } exp{2πiφj lj } , (43)
Na ,Nb0 i=1 li =−∞ lj =−∞
where
Xi
kX 0X
i i
φi = Q1 (Na0 − Na ) + Na Nb ,
2
(a=1 2 )
a=1 b=1 (44)
2 k
φj = Q1 − Q1 + N
2
.
k 2
Now we can consider each of the sums over lj and the li ’s independently. If one of them is
zero c will be vanishing. Note that we didn’t worry too much about normalization factors
but each of these phase sums must be normalized to a number between 0 and 1 in order
that one has the interpretation of C 2 as a probability. Further, if we take the limit of j → ∞
these sums will become integrals. We can investigate if they are zero or not by using the
identity
∞
X ∞
X
exp{2πi φ q} = δ(φ − p). (45)
q=−∞ p=−∞
The sum over delta functions is zero if φ is not an integer. The conclusion is then that, for
every i, φi must be an integer. Summing over lj we obtain the restriction
( 2 )
2 k
Q1 − Q1 + N
2
∈ Z. (46)
k 2
Replacing Q1 by its form (16), expanding and getting rid of the even integer term 2mN
which does not change anything, the former condition reads
k
N(n + N) ∈ Z. (47)
2
For irrational k the only two solutions are
N = 0, (48)
N = −n. (49)
The physical meaning of these results will be given a little later. The remaining conditions
in the φi ’s will allow us to build the intermediate processes, these will be addressed in the
end of this section.
P
Now an other way to carry out the sums is to eliminate lj = l − l0 − li . Then (39)
factorizes as
( 2 )
2 k
c[Q1 , N, l] = exp −2πi Q1 + N l
k 2
∞
X ∞
X −1 X
jY ∞ ∞
X (50)

× exp 2πiΦj0 [Na , Nb , li , l0 ] .
Na =−∞ Nb0 =−∞ i=1 li =−∞ l0 =−∞
Performing the sum over l0 we obtain the same conditions (46) and (47) on N . There
are extra 2N and N 2 factors but they do not change the coefficient c. We can check this
explicitly by replacing Q1 by its form (16). Apart from an irrelevant factor of 4πimN one
gets the same result as for Φ 0 0 .
An other way of factorization is as the following. Write (41) in terms of l ±
l0+ = l0 + lj , l0− = l0 − lj (51)
P
and take into account that l0+ = l − li . Then (39) factorizes as
( 2 ! )
2 k l
c[Q1 , N, l] = exp −2πi Q1 + Q1 + N
2
k 2 2
X∞ X∞ X∞ X∞ (52)

× exp 2πiΦ+ 0
[Na , Nb , li , l0− ] .
Na =−∞ Nb0 =−∞ li =−∞ l − =−∞
0
Performing the sum over l0− we obtain a different condition

( 2 )
2 k
Q1 − Q1 + N
2
∈ 2Z. (53)
k 2
Expanding Q1 this gets to
k
N(n + N) ∈ 2Z. (54)
2
For generic k ∈ R we obtain the same solutions as (46) and apparently this condition turns
out to hold the same results of (47). It will become clear that this last result is different
from the previous two cases. As before the extra N and N 2 /2 factors do not change the
coefficient.
It is now time to explain the physical implications of the three previous results. The
phase factors in front of l are none other than the conformal dimensions (20) of the
vertex operators inserted in the boundary CFT’s. When we eliminated the sum over l0
we obtained (42). Extracting the l phase factor we get
2 2
Q = 2∆ = ∆1 + ∆2 . (55)
k 1
Upon the elimination of lj sum we obtained the factorization (50). The phase factor reads
2 2
Q̄ = 2∆¯ = ∆¯ 1 + ∆¯ 2 . (56)
k 1
And for the (52) factorization where we eliminated l0+ sum, we obtain
1 2
Q1 + Q̄21 = ∆ + ∆.¯ (57)
k
Note that we cannot extract any other factors. If we try to eliminate any other
combination of l0 and lj we would get c = 0. Eliminating some combination of li ’s we
will always end up extracting one of the above factors or get again c = 0. The solutions for
the monopole interactions (63) and (65) computed below assure this result.
In order C not to be zero it is necessary to get integer conformal dimensions as stated
in Section 2. Consider then adding extra non-compact gauge fields. One Wilson line (29)
carries now charges from all the U (1)’s. Summing over l for the three previous cases we
obtain the conditions

p2
r = 2 ∆ + 0 ∈ Z,
k

p2
¯
s = 2 ∆ + 0 ∈ Z, (58)
k
2p2
t = ∆ + ∆¯ + 0 ∈ Z.
k
Subtracting the second equation from the first we obtain
2 2
¯ =
r − s = 2(∆ − ∆) Q1 − Q̄21 = −2mN, (59)
k
where N = 0, −n according to the allowed solutions (48) and (49). Averaging the first two
conditions we get the third one as long as we identify
t = (r + s)/2. (60)
According to the discussion in Section 2 r and s have to be even in order the full conformal
dimension to be integer. Making the identifications
r
= 1 − L,
2 (61)
s
= 1 − L̄,
2
we retrieve the mass shell condition (32). And (59) becomes the spin condition (33).
Below we explore the possible intermediate processes by performing the sums over each
of the li ’s. We obtain a chain of conditions φi ∈ Z which allow us to build the full set of
possible diagrams contributing to the transitions. Starting with i = 1 we find
n(N10 − N1 ) + 2N1 N10 = 0, (62)
which has two possible solutions
a: N1 = 0 = N10 ,
(63)
b: N1 = −n = −N10
that we name a and b.
For i = 2 we have
n(N10 − N1 ) + n(N20 − N2 ) + 2N1 N10 + 2N2 N20 + 2N1 N20 + 2N2 N10 = 0. (64)
Choosing a for N1 and N10
we have again the same previous solutions for N2 and N20 .
Choosing b we obtain again a or a new solution
b̃ : N2 = n = −N20 (65)
that we name b̃. If we choose b̃ we will find for i = 3 an a or b solution. So no new
solutions emerge.
Note that a solution a will reduce the diagram to one of type i − 1, we can join the
linkings before and after that monopole since it does not change the charge. Without loss
of generality we will disregard the a solution in the following discussion.
By induction the ith condition reads then
n(1 − 2#b + 2#b̃)(Ni0 − Ni ) + 2Ni Ni0 = 0 (66)
with the restriction that #b − #b̃ takes only the values 0 and 1. Then we will obtain a chain
of alternating solutions bb̃ . . . bb̃bb̃ . . . as the only possible solution. Some diagrams are
presented in Fig. 3.
The above results are consistent with global charge conservation. And moreover they
show that local charge violation is quite restricted. given some charge q in the first
boundary it can change to q̄ and return to its previous form q again as many times as
it wants but it can become nothing else.
Lets go on to consider now 3 charges propagating from one boundary to the other (or
equivalently evolving in time)
Fig. 3. Some of the possible diagrams for two charge propagation, i = 2 and i = 3.
Q1 → Q̄1 ,
Q2 → Q̄2 , (67)
Q3 = −Q1 − Q2 → Q̄3 = −Q̄1 − Q̄2 ,
as pictured in Fig. 2. The phase is know
(
2
Φ3 = Q1 Q2 l0 − Q2 (Q1 + Q2 )d0 − Q1 (Q1 + Q2 )k0
k
j −1
X
+ Q̄21(i) li − Q̄1(i) Q̄3(i) di − Q̄2(i) Q̄3(i) ki (68)
i=1
)
+ Q̄1 Q̄2 lj − Q̄2 (Q̄1 + Q̄2 )dj − Q̄1 (Q̄1 + Q̄2 )kj ,
where we defined
kX
i
k
Q̄1(i) = Q1 + Ni , Q̄1 = Q1 + N,
2 2
a=1
kX
i
k
Q̄2(i) = Q2 + Mi , Q̄2 = Q2 + M, (69)
2 2
a=1
kX 0
i
k
Q̄3(i) = Q1 + Q2 − Ni , Q̄3 = Q1 + Q2 − N 0 .
2 2
a=1
Now we can choose to factor out some combination of l0 , d0 , k0 , lj , dj and kj . But it
is now clear from the previous discussion of two charges propagation what factorizations
we should look for. So we are only going to eliminate all the 0’s, all the j ’s or some
combination of 0’s and j ’s for the three linkings. Defining the total linking numbers l =
P P P
l0 + li + lj , d = d0 + di + dj and k = k0 + ki + kj we can replace the l0 , d0 and k0
dependence and obtain the phases appearing in the correlation of three fields given in (22)
for z12 , z23 and z13 on l, d and k, respectively. Considering extra non-compact gauge
fields with charges p1 , p2 and p3 and summing over the previous variables we obtain the
conditions that the combinations of conformal dimensions in (22) must be integer. The
conformal dimensions are to be read considering the extra non-compact gauge fields and
charges ∆ = Q2 /k + p2 /k 0 and ∆¯ = Q̄2 /k + p2 /k 0 . From now on let us use this definition
of conformal dimensions. As discussed in Section 2 from the point of view of the boundary
CFT’s these factors must indeed be integer to insure that the three point OPE are single
valued.
In terms of the individual vertices it is not so clear what these conditions mean. Without
loss of generality and in order to clarify it let us replace the sums over l, d and k by the
sums over three new variables l+ , d+ and k+ such that l = l+ + k+ , d = d+ + l+ and
k = k+ + d+ . Summing over these new variables one obtains the following conditions
2∆1 , 2∆2 , 2∆3 ∈ Z. (70)
In a similar way we may change lj , dj and kj dependence by l+ , d+ and k+ and upon
summation obtain the respective conditions
2∆¯ 1 , 2∆¯ 2 , 2∆¯ 3 ∈ Z. (71)

Or using the definition (51) for l0± , d0± and k0± and replacing l0+ , d0+ and k0+ dependence by
l+ , d+ and k+ in the same way as before, after performing the sums in these last variables,
we get the conditions
∆1 + ∆¯ 1 , ∆2 + ∆¯ 2 , ∆3 + ∆¯ 3 ∈ Z. (72)
Similarly to the previous discussion for two charge propagation these conditions turn out
to be equivalent to the mass shell and spin conditions of string theory.
The fundamental differences arrive in the remaining conditions. Taking the first two
cases (70) and (71), where we replaced the 0’s and j ’s linking number sums, and summing
over (l+j , d+j , k+j ) and (l+0 , d+0 , k+0 ) we obtain in both cases three conditions. The first
two are redundant, they correspond to (46) which has been obtained from the two charges
propagation, the third one is new
4
(Q1 Q2 − Q̄1 Q̄2 ) ∈ Z. (73)
k
Nevertheless the solutions for irrational k end up by being of the same kind as before.
There are two solutions

N = 0, N = −n1 ,
(74)
M = 0, M = −n2 .
For (72) case, where we replaced the sum over +’s linking number, we obtain, upon
− − −
summation over (l+0 , d+0 , k+0 ), the conditions two conditions (53) which were obtained
in the two charges case. The third one is again new
2
(Q1 Q2 − Q̄1 Q̄2 ) ∈ Z. (75)
k
The solutions for the full diagrams are very similar to the ones of two charges. With
the charges oscillating simultaneous between Q1 , Q2 and Q̄1 = Q1 − n1 k/2, Q̄2 = Q2 −
n2 k/2. For four charges there are no new conditions.
After all these algebra let us summarize what we obtained. We started with a theory
which, at the perturbative level, has the allowed charges of the form Q = m + (k/4)n.
We showed, under certain assumptions that, if the non-perturbative processes are taken
into account, the charge spectrum is modified (restricted) quite drastically. The allowed
left/right pair of charges are in one-to-one correspondence with the processes which have
non-vanishing quantum amplitudes. All these processes can be organized in a lattice Γ
with elements l = (Q, Q̄ = Q + (k/2)N) with N either 0 or −n. Moreover a Lorentzian
product ◦ of signature (+,−) emerge naturally from (46), (53) and (75) defined as
2
l ◦ l 0 = (QQ0 − Q̄Q̄0 ). (76)
k
And we can go even further and extract very important properties of the lattice, it is even
due to (53)
2
l ◦ l = (Q2 − Q̄2 ) ∈ 2Z. (77)
k
and integer due to (75)

2
l ◦ l 0 = (QQ0 − Q̄Q̄0 ) ∈ Z. (78)
k
Note that (73) is redundant since it is necessarily obeyed if (75) is.
By inspection, for N = −n the lattice is self-dual as well. But not for N = 0. In the next
section we explain how to exclude N = 0 case and obtain what we seek. The lattice is then
exactly the one for the bosonic string theory with one compact dimension. It is the Narain
lattice for compactification in S 1 . For toroidal compactification of several dimensions is
enough to consider several compact U (1). This situation is well described in Section 6.
4. Boundary conditions
We still need to understand how we can select between N = 0 and N = −n cases. The
key is to consider different combinations of the boundary conditions introduced in (6).
Further, to obtain a CFT in the boundary (whether it is chiral or not) as an effective 2D
theory of the 3D TMGT we have to identify the two boundaries ΣL and ΣR in some way.
These are the mechanisms that allow us to build several string theories out of the same 3D
theory. Note that this constructions are only possible due to the Maxwell term in the action
and the existence of monopoles.
We learned in the last section that the bulk theory only allows the charge q = m + kn/4
to become q̄ = m − kn/4 or keep it to be the same charge q. What is actually n? Let us
go back to (16), 2πn is the flux of the magnetic field. So upon some kind of boundary
identification our theory only admits the ones that hold one of the conditions
Z Z
N = 0, d 2 zL B = d 2 zR B,
ΣZL ΣR Z
(79)
N = −n, d zL B = −
2
d 2 zR B.
ΣL ΣR
Let us consider a map (coordinate transformation) that transforms a vector in ΣR into

another vector in ΣL , giving us the identification rule. There will be two kinds of maps.
One which maintain the relative orientation of both 2d boundaries, let us call it parallel
(//), and the other kind that reverses the relative orientation, let us call it perpendicular
(⊥). The names are chosen by the relative identification of the axes from boundary to
boundary as pictured in Fig. 4.
Note that the induced ˜ does not change from boundary to boundary. From the 3d
point of view we are reversing time in one boundary and swapping the complex space
coordinates. From the point of view of the bulk nothing special is happening.
Let us define the difference of fluxes as
Z Z Z Z
δφ = B − B = d z ∂i Aj − d 2 z ij ∂i Aj .
2 ij
(80)
R L ΣR ΣL
Fig. 4. Boundary identification with the same orientation (parallel; //) and reversed orientation
(perpendicular; ⊥).
Given our boundary conditions we can evaluate this difference explicitly. Let us write what
the magnetic field is for each of the allowed boundary conditions
Z
C : φC = − ∂z̄ Az ,
Z (81)
C̄ : φC̄ = ∂z Az̄ .
In these equations we took the magnetic field definition B = ∂z Az̄ − ∂z̄ Az and imposed
the respective boundary conditions. Taking into account the boundary identifications and
labeling the different combinations of boundary conditions accordingly we compute the
flux difference
CC// : δφ = 0, Z
CC⊥ : δφ = − (∂z̄ Az + ∂z Az̄ ) = −2π(2n),
Z (82)
C C̄// : δφ = − (∂z̄ Az + ∂z Az̄ ) = −2π(2n),
C C̄⊥ : δφ = 0.
Note that for parallel type of identifications the fields and integrals are summed without
any relative change. For the perpendicular type we have to change the space indices of the
fields to their conjugates and the measure in the integral changes sign (in one boundary,
say the right one). The results are summarized in Fig. 5.
So the conclusion is that Q = Q̄ corresponds to CC// , C̄ C̄// , C C̄⊥ or C̄C⊥ and Q − Q̄
= (k/2)n to CC⊥ , C̄ C̄⊥ , C C̄// or C̄C// boundary conditions. Note that in principle there
may exist other choices of the maps which would give different boundary theories. We
leave this to a future work.
Fig. 5. Boundary identifications for several combinations of boundary conditions. The fat arrow
represents magnetic flux and the thin one boundary orientation.
Note that the spectrum for NC or N C̄ boundary conditions must be obtained by

truncating the spectrum of CC⊥ or C C̄// . We simply pick Q̄ = 0 for the N boundary,
for instance
k
m= n (83)
4
and the charges in the other boundary become
k
Q = n. (84)
2
Of course for irrational k the only solution for the condition (83) is m = n = 0. We end
up with a very poor and empty theory in the boundary. In Section 6 it will be clarified in
which cases condition (83) allows some dynamics in the boundary. Trying to truncate the
spectrum of CC// or C C̄⊥ we will set straight away Q = Q̄ = 0 since the charges are equal
in both boundaries killing the hope of finding any dynamics in the boundaries.
We choose from now on to work with CC⊥ or C C̄// type of boundary conditions since
they are the ones which give us the desired spectrum in the boundaries CFT’s. Further, as
we just explained, N boundary conditions can easily be obtained from them.
5. RCFT’s and fusion rules
A CFT is rational when its infinite set of primary fields (vertex operators) can be
organized in a finite number N of families usually called primary blocks. In each of those
blocks we can choose one minimal field that is the generator of that family. There is an
algebra between these families, or if we want, between the N minimal fields called the
fusion algebra. The fusion rules define this second algebra of fields.
pholomorphic part of the c = 1 RCFT of
In what follows we are only going to discuss the
a bosonic field φ living in a circle of radius R = 2p0 /q. Here p0 and q are integers. The
vertex operators are
r R
VQc = exp(2πiQc φ), Qc = +s , r, s ∈ Z, (85)
R 2
where Qc are the charges (or momenta) of the theory. The conformal dimensions of these
vertex operators are ∆ = Q2c /2. There are N = 2p0 q primary blocks (or families). The
generators VQλ of such families are chosen such that their conformal dimensions are the
lowest allowed by the theory. In terms of their charges these are
λ
Qλ = √ , λ = 0, 1, 2, . . . , N − 2, N − 1. (86)
N
√ to call them from now on primary charges. In this way Qλ runs from 0 to
We are going
√
N − 1/ N . The remaining fields of the theory are obtained by successive products of
the generators. The charges in a family λ are
√
Qλ,L = Qλ + L N, L ∈ Z. (87)
The generators form the fusion algebra given by the fusion rules
Qλ + Qλ0 = Q(λ+λ0 ) mod N . (88)
We took the liberty to express it in terms of the charges. Formally it is expressed in terms of
fields or vertex operators Vλ × Vλ0 = V(λ+λ0 ) mod N . This simply means we pick two primary
charges out of two families and add them. We will obtain a charge from a different family,
but not necessarily the primary one. The fusion rules pick the primary charge of that new
family. For further details on these subjects see [39].
For our normalization we have the following relations
k = 2R 2 , Q = RQc . (89)
In what follows we are going to explain how all this emerges with some naturalness from
the bulk theory. Take then k = 2p/q to be a rational number. Note that in the previous
discussion 2p0 = p, so take even p. By inspection we check that the (m, n) space is
divided in diagonal bricks containing p × q charges. They are distributed in diagonal layers
of identically valued bricks as symbolically pictured in Fig. 6. To check this diagonal
structure explicitly take a generic charge labeled by the pair (m, n), it can be represented
by any other pair (m − pn0 /2, n + qn0 )
p p p
Q=m+ n = m − n0 + (n + qn0 ). (90)
2q 2 2q
This simply represents a diagonal translation of n0 bricks in the figure.
Note that this choice of the brick shape is not unique, we could take some other choice
as long as it has dimension pq (e.g., parallelogram of sides p and q) and the following
results would hold nevertheless. The reason for this particular choice is in order to get a
direct parallel with the usual result of RCFT’s. Further we consider one of such bricks to
be the primary brick. It corresponds to the minimal charges (chosen to be positive) allowed
by the theory built out of the lowest pairs of integers (r, s)
p λ p
Qrs = r + s= , λ = rq + s = 0, 1, 2, . . . , pq − 2, pq − 1. (91)
2q q 2
Examples for some values of k are given in Fig. 7. The geometric rule to organize the
charges inside the brick is to order them in ascending order by their distance to the upper
diagonal line of slope k/4.
Fig. 6. (m, n) structure for even p. The dashed region is the primary brick (r, s), some examples are
presented in Fig. 7. The numbers inside the blocks represent L1Q shift of the charge values as given
by (95). Blocks with the same L have the same charge entries. In each brick there is exactly one
element of level L belonging to each family.
So far we just reproduced the well know charge structure of RCFT’s, it is now time to
return to the bulk theory and justify it. Take again condition (53), replacing k, it reads now
p
nN − N 2 ∈ Z (92)
2q
which can be reexpressed as
N(n − N) = 0 mod q. (93)
The solutions for (93) can be easily computed to be
N = 0 mod q,
(94)
N = −n + 0 mod q,
Fig. 7. Primary charges distribution in the (m, n) plane, they constitute the primary brick (r, s). The
rule to order the charges is by their distance to the (thin) line connecting the charges Q = 0 (filled
dots) of slope k/4 = p/2q.
Fig. 8. Charge structure for k = 2p. Only even values are considered in the n axis.
which are equivalent to (46) and (47). There is one important lesson to take from this
result, besides the previous allowed monopole-instanton process n → −n which is charge
dependent, there is a new charge independent one:
1Q = ±p. (95)
This is actually the physical process that spans each of the families! To obtain the charges
of some family in terms of m and n we can think on shift either m → m + pL or n →
n + 2qL. This is due to an infinite degeneracy in the (m, n) plane of the charges values
expressed by (90). Fig. 8 presents the distribution of charges for k = 2p. In this simplified
case the families are simply organized along the m axis for even values of 2n. This structure
is repeated by shifts on the m axis of value p.
For odd p all the structure is similar but with bricks of dimension p × 2q, take again a
generic charge Q
pn p(n + 2 n0 )
Q= +m= + mp − pn0 . (96)
2q 2q
Because p is odd, same valued charges correspond to pairs related by (m, n) → (m − pn0 ,
n+2qn0). The slope of the brick is the same k/4 but they are twice bigger in the n direction.
All the rest works in the same way. This would correspond to a boson field living on a circle
√
of radius R = p/q.
6. Lattices and heterotic string
Considering an action of the type (1) with compactified U (1)D gauge group (replacing
k by a generic tensor KI J ) we have
Z √
−g(3) µν I GI J + BI J µνλ I
S = d z dt −
2
FI Fµν + Aµ ∂ν Aλ ,J
(97)
4γ 8π
M
GI J and BI J stand for the symmetric and antisymmetric parts of KI J . Using the same
procedure outline in Section 2 we build the monopole-instanton which induces a charge
change of
GI J J
1QI = N . (98)
2
Using the Schrödinger picture we get the charge spectrum
KI J J
QI = mI + n . (99)
4
Note that the antisymmetric part BI J is present in the charge but not in the monopole
effects. This is due to the fact that it is manifested only at the level of the boundary. Note
that in the action the term corresponding to BI J is a total derivative and can be completely
integrated out to the boundary
Z " Z Z #
1
BI J Aµ ∂ν Aλ =
µνλ I J
− BI J ˜ Ai Aj + BI J ˜ Ai Aj .
ij I J ij I J
(100)
2
M ΣL ΣR
Let us analyze first the case for BI J = 0. The previous discussion of Section 4 follows
in the same fashion. Choosing CC⊥ or CC// boundary conditions we have the desired
relative spectrum in each boundary, that is, Q = m + Kn/4 and Q̄ = m − Kn/4. This
means that every monopole contribution of the form (98) has exactly N J = −nJ .
What about if BI J 6= 0? Let us return to our boundary identifications, For RCC⊥ the
R 2 of the integrals in opposite boundaries change their relative sign, say R d z →
measure 2
− R d z, due to the 2d measure changing sign. The fields in one boundary (say ΣR ) are
swapped, Ai ↔ Aj . Note that we are not changing neither BI J nor ˜ ij in the right integral,
they are induced from the bulk and do not change by the boundary identifications. Thus
this transformation has no effect in the BI J term
Z Z
BI J ij AIi AJj → BI J ij AIi AJj . (101)
ΣL ΣL
For CC// nothing changes either. This means that BI J term does not change sign under
any of our boundary identifications.
So we obtain the left/right spectrum
GI J + BI J J
QI = n + mI ,
4 (102)
−GI J + BI J J
Q̄I = n + mI
4
such that the charge difference QI − Q̄I = GI J nJ /2 is indeed the monopole contribution.
We obtain a lattice l = (Q, Q̄) with the Lorentzian product of signature (+, −) defined
as
l ◦ l 0 = 2G(−1)I J (QI Q0 J − Q̄I Q̄0J ), (103)
where G−1 stands for the inverse of GI J . The signature of the product has D plus and D
minus. The properties of this lattice are the same as the ones for the previous D = 1 case
and follow in a similar way from the bulk theory as presented in Section 3. The lattice is
integer, even and self-dual.
Let us analyze which lattices do exist for CN boundary conditions. As explained before
(Section 4) they can be obtained by truncating the lattices where N J = −nJ . Imposing
then CN is equivalent to truncating the lattice by choosing Q̄ = 0. Similarly to (83), this
means we are selecting elements of our lattice such that
GI J − BI J J
mI = n . (104)
4
This sublattice has elements l = (Q, 0) with Q built out of n and m obeying (104).
Once Q̄ = 0 the Lorentzian product (103) becomes, for this particular sublattice, simply
2GI J QI QJ . So it becomes Euclidean for this particular choice of elements. But the
properties of being even, integer and self dual are inherited from the full lattice. It is a
known fact that the only even integer self dual Euclidean lattices are of dimension 0 mod 8.
Of dimension 8 there is only one, Γ 8 , the root lattice of E8 . From (104) we can derive the
spectrum of allowed Q’s
GI J J
QI = n . (105)
2
Note that the resulting spectrum is independent of BI J . Due to the spectrum (105) and
by closure of the lattice under the defined product (103) we are obliged not to restrict n.
Anyhow, as stated before, the charge lattice must be the root lattice of E8 , it is clear that
the only choice that holds this result is G = 2C. C is the standard Cartan matrix of this
group (the diagonal elements have the value 2 and the off diagonal −1). The corresponding
Dynkin diagram is pictured in Fig. 9. Let us return to (104). It must be a realizable condition
for every n, Thus we have no choice but to impose GI J − BI J = 0 mod 4.
We are left with
GI J = 2CI J ,
(106)
BI J = (2 + 4r)CI J , J > I, r ∈ Z.
There are two lattices of dimension 16 obeying these properties: Γ 8 × Γ 8 with elements
in the root lattice of E8 × E8 and Γ 16 . We retrieve the well known results of string theory.
Note that in the first works [18] of non-interacting heterotic string there is no mention
to the tensor BI J (the free spectrum depends solely on GI J ). Only after the introduction
of interactions and an effective action for the theory [19] it becomes clear that there is a
need for BI J . For r = 0 we retrieve the result stated in [40] and references there in up to
normalization of 1/4 on KI J .
Fig. 9. Dynkin diagram of E8 .

7. Conclusions
Most of the constraints on string theories exist because of their intimate relations to 2D
geometry, 2D conformal field theories and their symmetries. One of these constraints is the
modular invariance which puts severe restrictions on the spectrum of the theories. Modular
invariance, as employed by Narain in the context of toroidal compactification, reduces the
physical problem of finding the allowed momenta and winding modes in string theory to
the problem of constructing lattices on a pseudo-Euclidean space RdR ,dL on which physical
invariants are modular invariant functions. These lattices turn out to be self-dual, integer
and even lattices.
In this paper we have reproduced the results of Narain compactification starting from
3D gauge dynamics without referring to modular invariance at all. The fact that 3D
topologically massive gauge theory carries full information about the lattices of string
theory is quite fascinating. Our approach and the modular invariance construction are
compatible and moreover as we will comment on it below they are logically connected.
But there is still an element of surprise here because of the crucial difference between the
ideology of these two approaches. To see this let us summarize what we have obtained.
We started with a TMGT defined on a three manifold with two disconnected boundaries.
Gauge degrees of freedom became dynamical on the boundaries generating chiral CFT’s.
Each boundary of the three manifold is interpreted as a “chiral worldsheet” of string theory,
meaning that the left and right modes are physically separated. This is the frame work of
topological membrane theory. We have considered the compact Abelian theory which has
a discrete charge spectrum of the form Q = m + kn/4. A particle with charge Q is inserted
at one boundary and it travels through the bulk interacts with all possible charge violating
instantons on its path and links with other charges and emerges as a new charge Q̄ =
m−kn/4 on the other boundary. The path of the charged particle is represented by a Wilson
line in the bulk theory. After taking care of all the linkings and instanton interactions it is
quite an amusing result to obtain that the emerging charge in the other boundary is of the
Q̄ form and nothing else. The Aharonov–Bohm phases of the linkings interfere in a way
that for a particle which does not have the charge ±Q̄ there is zero probability to emerge in
the other boundary. Moreover a natural self-dual Lorentzian lattice structure emerges from
the linkings of the Wilson lines as it was shown in Section 4. The connection between
our approach and the modular invariance arguments was not worked out in this paper but
naively one can see that modular transformations in the boundaries will yield linkings of
the Wilson lines in the bulk. This is why we have gotten the same results.
There are a couple of important things which need to be addressed in future. One of them
is the bulk interpretation of A-D-E classification [41] of modular invariant partition func-
tions in the boundary. An other important issue is to study non-orientable surfaces which
appear for non-oriented strings. We hope to address these issues in forthcoming publica-
tions.
Acknowledgements
We would like to thank Arshad Momen and Martin Schvellinger for useful discussions.
PCF would also like to thank Alex Kovner for many useful discussions and Gerald
Dunne for some remarks on the manuscript. The work of PCF is supported by PRAXIS
XXI/BD/11461/97 grant from FCT (Portugal). The work of IK and BT is supported by
PPARC Grant PPA/G/0/1998/00567. The work of IK is also supported by EU Grant
FMRXCT960090.
References
[1] I.I. Kogan, Phys. Lett. B 231 (1989) 377.

[2] S. Carlip, I.I. Kogan, Mod. Phys. Lett. A 6 (3) (1991) 171–181.
[3] E. Witten, Commun. Math. Phys. 121 (1989) 351–399.
[4] G. Moore, N. Seiberg, Phys. Lett. B 220 (1989) 422.
[5] S. Elitzur, G. Moore, A. Schwimmer, N. Seiberg, Nucl. Phys. B 326 (1989) 108.
[6] M. Bos, V.P. Nair, Phys. Lett. B 223 (1989) 61, Int. J. Mod. Phys. A 5 (1990) 959.
[7] J.M.F. Labastida, A.V. Ramallo, Phys. Lett. B 227 (1989) 92; Phys. Lett. B 228 (1989) 214.
[8] W. Ogura, Phys. Lett. B 229 (1989) 61.
[9] G. ’t Hooft, Salamfest, e-print archive: gr-qc/9310026, 284–296.
[10] L. Susskind, J. Math. Phys. 36 (1995) 6377–6396.
[11] S. Carlip, I.I. Kogan, Phys. Rev. Lett. 64 (1990) 148; Phys. Rev. Lett. 67 (1991) 3647.
[12] I.I. Kogan, Phys. Lett. B 256 (1991) 369; Nucl. Phys. B 375 (1992) 362.
[13] G. Amelino-Camelia, I.I. Kogan, R.J. Szabo, Nucl. Phys. B 480 (1996) 413–456; Int. J. Mod.
Phys. A 12 (1997) 1043–1052.
[14] I.I. Kogan, R.J. Szabo, Nucl. Phys. B 502 (1997) 383–418.
[15] I.I. Kogan, A. Momen, R.J. Szabo, JHEP 9812 (1998) 013.
[16] K.S. Narain, Phys. Lett. B 231 (1989) 377.
[17] K.S. Narain, M.H. Samardi, E. Witten, Nucl. Phys. B 279 (1987) 369.
[18] D. Gross, J. Harvey, E. Martinec, R. Rohm, Nucl. Phys. B 256 (1985) 253.
[19] D. Gross, J. Harvey, E. Martinec, R. Rohm, Nucl. Phys. B 267 (1986) 75.
[20] R. Jackiw, S. Templeton, Phys. Rev. D 23 (1981) 2291.
[21] J. Schonfeld, Nucl. Phys. B 185 (1981) 157.
[22] S. Deser, R. Jackiw, S. Templeton, Phys. Rev. Lett. 48 (1982) 975, Ann. Phys. (NY) 140 (1982)
372.
[23] K. Lee, Nucl. Phys. B 373 (1992) 735.
[24] L. Cooper, I.I. Kogan, Phys. Lett. B 383 (1996) 271–280.
[25] L. Cooper, Ph.D. Thesis, 1997.
[26] P. Ning-Tan, B. Tekin, Y. Hosotani, Phys. Lett. B 388 (1996) 611–620; Nucl. Phys. B 502 (1997)
483–515.
[27] L. Cooper, I.I. Kogan, K.M. Lee, Phys. Lett. B 394 (1997) 67–74.
[28] L. Cooper, I.I. Kogan, R.J. Szabo, Ann. Phys. 268 (1998) 61–104.
[29] W. Ogura, Phys. Lett. B 229 (1989) 61.
[30] S. Carlip, Nucl. Phys. B 362 (1991) 111.
[31] M. Asorey, F. Falceto, S. Carlip, Phys. Lett. B 312 (1993) 477–485.
[32] M.C. Ashworth, Mod. Phys. Lett. A 10 (1995) 2749–2755.
[33] E.C. Marino, Phys. Rev. D 38 (1988) 3194.
[34] A. Kovner, B. Rosenstein, D. Eliezer, Nucl. Phys. B 350 (1991) 325.
[35] A. Kovner, B. Rosenstein, Int. J. Mod. Phys. A 7 (1992) 7419.
[36] I.I. Kogan, A. Kovner, Phys. Rev. D 51 (1995) 1948.

[37] A.M. Polyakov, Mod. Phys. Lett. A 3 (1988) 325.
[38] J. Polchinski, String Theory, Cambridge Univ. Press.
[39] P. Di Francesco, P. Mathieu, D. Sénéchal, Conformal Field Theory, Springer.
[40] A. Giveon, M. Porrati, E. Rabinovici, Phys. Rep. 244 (1994) 77–202.
[41] A. Cappelli, C. Itzykson, J.B. Zuber, Nucl. Phys. B 280 (1987) 445–465; Commun. Math.
Phys. 113 (1987) 1.
Marginal and relevant deformations of N = 4

field theories and non-commutative
moduli spaces of vacua
David Berenstein, Vishnu Jejjala, Robert G. Leigh ∗
Department of Physics University of Illinois at Urbana-Champaign, Urbana, IL 61801, USA
Received 11 May 2000; accepted 20 June 2000
Abstract
We study marginal and relevant supersymmetric deformations of the N = 4 super Yang–Mills
theory in four dimensions. Our primary innovation is the interpretation of the moduli spaces of
vacua of these theories as non-commutative spaces. The construction of these spaces relies on the
representation theory of the related quantum algebras, which are obtained from F -term constraints.
These field theories are dual to superstring theories propagating on deformations of the AdS5 × S 5
geometry. We study D-branes propagating in these vacua and introduce the appropriate notion
of algebraic geometry for non-commutative spaces. The resulting moduli spaces of D-branes
have several novel features. In particular, they may be interpreted as symmetric products of non-
commutative spaces. We show how mirror symmetry between these deformed geometries and
orbifold theories follows from T-duality.
Many features of the dual closed string theory may be identified within the non-commutative
algebra. In particular, we make progress towards understanding the K-theory necessary for
backgrounds where the Neveu–Schwarz antisymmetric tensor of the string is turned on, and we
shed light on some aspects of discrete anomalies based on the non-commutative geometry.  2000
Elsevier Science B.V. All rights reserved.
Keywords: D-branes; AdS/CFT; Non-commutative geometry; K-theory
1. Introduction
The study of the deformations of the N = 4 U (M) supersymmetric theory in four

dimensions is of interest from several points of view. This theory is superconformally
invariant, and it has been known for some time that exactly marginal deformations of
∗ Corresponding author. E-mail: rgleigh@uiuc.edu

E-mail addresses: berenste@pobox.hep.uiuc.edu (D. Berenstein), vishnu@pobox.hep.uiuc.edu (V. Jejjala).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 9 4 - 1
D. Berenstein et al. / Nuclear Physics B 589 (2000) 196–248 197
this theory exist (Ref. [1] and references therein) and should be described by interacting
superconformal field theories (CFT). These CFT’s are largely unexplored.
In the large M-limit, the deformations of the N = 4 theory have a nice description in
terms of a supergravity dual [2–4] and are obtained by the addition of operators which
modify the boundary conditions at infinity. Each of the renormalizable deformations are
reflected in the AdS/CFT correspondence through backgrounds for massless and tachyonic
excitations, including both RR and NS fields [5].
Among the marginal deformations, of particular interest is the q-deformation
Wq = tr(φ1 φ2 φ3 − qφ2 φ1 φ3 ) (1.1)
which is a deformation of the superpotential by the symmetric invariant preserving N = 1
supersymmetry and a U (1)3 global symmetry. For special values of q these theories are
described by the near-horizon geometries of orbifolds with discrete torsion [6,7]. It was
conjectured in Ref. [7] that these orbifold theories are related by mirror symmetry to string
theories on S 5 -deformations of AdS5 × S 5 .
There are also relevant deformations which carry the theory away from the ultraviolet
fixed point CFT. In some cases, the infrared theory is of interest — a prime example being
the deformation by rank-one mass terms [1,8]. From the supergravity point of view, the
renormalization group flow is encoded as a dependence of the background on the radial
scaling variable [9,10].
With a rank-three mass matrix, the field theory has been analyzed in many pa-
pers [11–15]. More recently [16], the supergravity duals of these theories have been an-
alyzed. There it was noticed that 5-brane sources resolve the would-be singularity in the
dual supergravity background, an application of the dielectric effect [17].
In this paper, we will begin an exploration of these field theories obtained by marginal
and relevant deformations of the N = 4 theory. The analysis will concentrate on the
classical vacua, particularly those aspects which depend upon holomorphic quantities. We
introduce a new way of thinking about these moduli spaces that should be of quite general
applicability.
Normally, the vacua of a supersymmetric gauge theory are parameterized using gauge
invariant holomorphic polynomials in the fields. This is attractive because of the gauge
invariance, but it is also unwieldy. As M increases, the number of independent invariants
increases dramatically. The F -term constraints on vacua are given, on the other hand,
directly in terms of holomorphic matrix equations, and the proposal centres around
using this description directly. Matrix variables have a number of technical advantages,
principally that the analysis is independent of their dimension, M. The main problem with
this approach is gauge invariance — the D-term constraints must be applied separately.
Given this, the F -terms can be thought of as a set of constraints on the algebra of M × M
matrices. Generally, this is a non-commutative algebra. There is a technical simplicity to
the choice of renormalizable superpotentials, namely that the constraints are quadratic, and
this simplifies the algebraic analysis significantly. The constrained algebras that appear
here in some cases bear some resemblance to algebras considered in the literature on
quantum groups (see, for example, Refs. [18,19]).
198 D. Berenstein et al. / Nuclear Physics B 589 (2000) 196–248
A related problem is the behavior of D-branes in dual descriptions of these field theories.
For small deformations, these duals are close to AdS5 × S 5 , and therefore the moduli
space of D-branes also has a description in this framework. The moduli space of probe
D-branes is roughly a symmetric product space; indeed, classically, we can think of a
single D-brane as moving on the moduli space of the corresponding field theory, and the
moduli space for multiple branes can often be related to the direct product of this space,
modded out by the permutation group. Realizing all aspects of the field theory analysis in
these dual descriptions gives insight into many non-trivial aspects of D-brane geometry. In
particular, a clear understanding of these issues reveals a T-duality transformation which
realizes mirror symmetry [7] between near-horizon geometries and orbifold theories.
In the present context, these remarks lead to the notions of non-commutative moduli
spaces of vacua, and moduli spaces of D-brane configurations are symmetric products of
a non-commutative space. We construct these notions algebraically; for example, points
in the non-commutative space correspond to irreducible representations of the algebra or
equivalently to maximal ideals with special properties. As we show later, these notions have
some tremendous advantages over the standard points of view. In particular, it is often the
case that we can think of a commutative subring (built out of the center of the algebra)
as a sort of coarse view of the full moduli space. In fact, the phenomenon of D-brane
fractionation at singularities follows precisely this rule: the fractionation is present in a
commutative description, but from the full non-commutative point of view, the fractional
nature is more readily understandable.
It is clear that from a string theory point of view, it is the open strings that see
this non-commutative structure directly [20]. Closed strings appear naturally within this
framework as single trace operators [3], and thus should see only the commutative part of
the space [21]. The remnant of non-commutativity in the closed string sector is the presence
of twisted states.
In general, when one studies more general configurations of D-branes, they should
correspond to algebraic geometric objects and classes in K-theory [22–24]. Because of
our emphasis, one needs to develop a non-commutative version of algebraic geometry
and K-theory. K-theory in this context is provided by the algebraic K-theory of the
non-commutative ring. We give the rudimentary structure of such a definition of
algebraic geometry; this definition apparently differs from others given in the mathematics
literature [25–27], but we believe our proposal is more natural, as dictated by string theory.
Our version of non-commutative geometry is clearly different than that which has been
recently studied extensively (see, for example, [20,21,28–31] and citations thereof). In
that case, the non-commutativity occurs in the base space of a super Yang–Mills theory,
whereas here it is in the moduli space, namely the directions transverse to a brane. In many
cases, the boundary state formalism (for a review, see Ref. [32]) is convenient to describe
D-branes, but we do not use that technology here. It would be interesting to generalize our
discussion to that formalism, although it is not clear to us how to solve for the boundary
states in the absence of a spectrum-generating algebra.
The paper is organized as follows. In Section 2, we review the structure of marginal
and relevant deformations of the N = 4 theory, and discuss the interpretation of
supersymmetric vacua in terms of non-commutative geometry. (We focus throughout on

U (M) gauge groups.) We also review the map between the superpotential deformations
and supergravity backgrounds. In Section 3, we give our construction of non-commutative
algebraic geometry and the resulting K-theory. This section is mathematically intensive; in
order that the casual reader may skip this section if desired, we provide at the beginning,
an overview of the key structures. In Section 4, we investigate the vacua of various field
theories, using the non-commutative formalism. We begin with the q-deformed theory, and
then investigate this theory with (a) a single mass term, (b) a mass term and a linear term,
(c) three arbitrary linear terms, and (d) three mass terms. In each case, we work out the
representation theory. We also consider the general case, and in particular consider the
effects of the other independent marginal deformation. The general case is quite difficult,
but we are able to identify a few interesting properties.
In Section 5, we turn our attention to string theory. For the q-deformed theory, the field
theory predicts new branches in moduli space for arbitrarily small values of q − 1. To
realize this branch in string theory, we need to consider BPS states corresponding to D5-
branes with 3-brane charge in the deformed backgrounds; the physics here is reminiscent of
the dielectric effect [17], but is more general. The new branch of moduli space is identified
as a D5-brane wrapped on a degenerating 2-torus. One obtains a natural 2-torus fibration
of the 5-sphere; T-duality on this torus leads to the mirror orbifold theory.
In Section 6, we consider the problem of identifying closed string physics directly from
the field theory description. Closed string states are naturally identified with single trace
operators. In this section, we also note several features of interest, including connections
to quantum groups and the K-theory of the non-commutative geometry.
In Section 7, we make some final remarks and indicate avenues for further research.
2. Field theory deformations
Our first objective will be to analyze the marginal and relevant deformations of the N =
4 super Yang–Mills (SYM) theory in four dimensions, with gauge group U (M). As usual,
we write this in terms of an N = 1 SYM theory with three adjoint chiral superfields φi ,
i = 1, 2, 3, coupled through the superpotential
W = g tr([φ1 , φ2 ]φ3 ). (2.1)
If we choose to preserve N = 1 SCFT, there is a moduli space of marginal

deformations [1], given by a general superpotential of the form

λ
Wmarg = a tr φ1 φ2 φ3 − qφ2 φ1 φ3 + φ13 + φ23 + φ33 . (2.2)
3
The Yang–Mills coupling g measures how strongly interacting the theory is and is a
function of a, q, λ such that each of the β-functions are zero. The structure of the moduli
space of vacua depends only on q, λ.
We will also consider relevant deformations of the form
X
Wrel = c1 tr φ12 + c2 tr φ22 + φ32 + ζj tr(φj ). (2.3)
j
For q 6= 1, general quadratic polynomials may always be brought to this form after a change
of variables.
The vacua of the theory are found by solving the F -term constraints
∂W
= 0. (2.4)
∂φj
In the present cases, these are quadratic matrix polynomial equations in the φj
φ1 φ2 − qφ2 φ1 = −λφ32 − 2c2 φ3 − ζ3 , (2.5)

φ2 φ3 − qφ3 φ2 = −λφ12 − 2c1 φ1 − ζ1 , (2.6)
φ3 φ1 − qφ1 φ3 = −λφ22 − 2c2 φ2 − ζ2 . (2.7)
These matrix equations are independent of M. In general, solutions will consist of a
collection of points, but at special values of parameters, we get a full moduli space of
vacua.
The equations (2.5)–(2.7) are a quite general class of relations. Note in particular that
when q = 1, we have a Poisson bracket structure, whereas if in addition λ = ζi = 0, we
find SU(2) commutation relations. When q 6= 1 and/or λ 6= 0, with cj = ζj = 0, the algebra
is that of a quantum plane [18]. When q 6= 1 and λ = cj = 0, these are q-deformations of
Heisenberg algebras.
The moduli space of vacua is usually parameterized in terms of gauge invariant
polynomials in the fields φj . This has the feature that the non-holomorphic D-term
constraints are automatically satisfied. The down side is that for large M, the number
of polynomials required becomes very large, and when perturbations are present, the
description of the space becomes quite complicated. Instead, we will choose to describe
the moduli space of vacua directly in terms of matrix variables. This has the virtue that the
equations are independent of M, as noted. Thus instead of considering the moduli space of
vacua as an algebraic variety, for general values of parameters, we should think of this as a
non-commutative algebraic variety.
Understanding the vacua of these theories then is equivalent to understanding the non-
commutative geometry defined by the relations (2.5)–(2.7). The φj can be thought of as
the generators of the corresponding non-commutative algebra. M × M matrices which
satisfy the relations are an M-dimensional representation of the abstract algebra. The
general problem at hand then is to study the representation theory of the algebra. The basic
representations of interest are those that are irreducible; given a finite set of such solutions
(φ1i , φ2i , φ3i ) labeled by i, then
M
φ̃k = φki (2.8)
i
is also a solution of the matrix equations.
It is important to keep in mind however that we must also consider the D-term
constraints. It is well-known [33] that for every solution of the F -term constraints, there is
a solution to the D-terms in the completion of the orbit of the complexified gauge group
SL(M, C). If the solution occurs at a finite point in the orbit, then we get a true vacuum.
If it occurs in the completion of the orbit (at infinity in the complexified gauge group), we
need to check that the solution does not run away to infinity.
2.1. Relation to deformations in AdS5 × S 5 geometry
Since the N = 4 theory is related to superstring theory on AdS5 × S 5 , the marginal

deformations correspond to deformations of S 5 . In particular, these are related to massless
states [3,5] in the 5-dimensional supergravity. They transform in the 45 of SU(4)
R-symmetry and are related to vevs for harmonics of RR and NSNS fields, F(3) RR and
NS
H(3) , along the 5-sphere. Similarly, the relevant deformations correspond to tachyonic
excitations of the 5-dimensional supergravity, and transform in the 10 of SU(4).
When q is a root of unity, we will often for convenience say that q is rational. In this
case, the q-deformation is known to be dual to the near-horizon geometry of D-branes on
an orbifold with discrete torsion, C3 /(Zn × Zn ).
The moduli space of vacua of the field theory is the moduli space of D-branes. Because
of the RR and NSNS backgrounds, D-branes move on a non-commutative space [20]. For
small enough deformations, we expect the AdS5 × S 5 geometry to be a close approximation
and we can interpret the eigenvalues of matrices as the positions of D-branes, à la matrix
theory [34].
Note that the superpotentials that we are considering are single trace operators.
This suggests that these operators correspond to effects that may be seen in classical
supergravity. This may be understood by looking at how background couplings to D-branes
behave at weak coupling. The leading effect comes from a disk diagram as shown in Fig. 1
where V is the background vertex.
Multiple trace operators would then correspond to string loop diagrams, and are
therefore suppressed by powers of gstr . 1 It is not clear that the string generates these
effects perturbatively, but to avoid them we could work at weak string coupling. It is also
possible that a non-perturbative non-renormalization theorem might keep multiple trace
operators equal to zero, at least up to some number of derivatives.
Fig. 1. Tadpole calculation of superpotential.
1 We assume a large but finite number of branes, so that relations between traces of finite matrices appear only
for very irrelevant perturbations.
3. Non-commutative algebraic geometry
3.1. Overview
As discussed above, we usually think of the moduli spaces of vacua as varieties,

namely commutative algebraic geometric objects. Because the F - and D-term constraints
may be recast in matrix form, it is more convenient to think of the same space as a
non-commutative object. Although the physical problem to solve is the same and the
space of solutions is the same, the non-commutative interpretation invokes extra structure
(namely, the commutative algebra of individual matrix elements is organized into the non-
commutative algebra of matrices). Because D-brane solutions are associated with algebraic
geometric objects in general [35] (and see [36] for a review), we need a formulation of non-
commutative algebraic geometry which captures D-brane physics correctly.
Several different versions of non-commutative algebraic geometry have been discussed
in the mathematics literature [25–27], but none of these seem natural in the present context.
In this section, our aim is to describe a definition of algebraic geometry appropriate to the
moduli spaces of vacua of D-branes in the field theory limit.
We want to understand the holomorphic structure of the moduli space, so we will only
concentrate on the F -term equations and will assume that given a solution to the F -terms,
there is a solution to the D-term equations. From the physics point of view, we confine
ourselves to those properties which are protected by supersymmetry; from a mathematical
viewpoint, these are holomorphic structures and can be described in terms of algebraic
geometry. Ordinarily, non-commutative geometries are related to C ∗ -algebras [37], which
include the adjoint operation; this is not a natural operation in a holomorphic framework,
and we will discard it for the considerations of this paper. In supersymmetric theories,
the holomorphic and anti-holomorphic features couple only through D-terms, and these
effects are in general not protected by supersymmetry. In discarding the D-terms, we lose
information about the metric in moduli space, but not topological features. Thus we need
a framework where we can do non-commutative algebraic geometry without C ∗ -algebras.
In the rest of this subsection, we give a brief outline of the mathematics involved, and its
physical interpretations. For the reader who wishes to skip the details of the mathematics,
this overview should suffice, and one may proceed to Section 4.
The building blocks for solutions are the finite-dimensional irreducible representations
of the algebra, as in Eq. (2.8). In the non-commutative algebraic geometry that we will
describe, these are defined to be points [37,40]. In matrix theory, non-commuting matrices
are interpreted as extended objects; here, these are considered point-like objects. In orbifold
theories, D-branes in the bulk are also considered to be point-like even though they can be
built out of fractional branes.
In our construction the center (the commutative subalgebra of the Casimir operators)
plays a pivotal role. In particular, one expects [21] that for a given background in
string theory, there are two descriptions, a commutative version relevant to closed strings
and a non-commutative version for open strings. In our best-understood examples, the
commutative space is the algebraic geometry associated to the center. By Schur’s lemma,
on every irreducible representation the Casimirs are proportional to the identity. Because of
this, one finds a map between non-commutative points and commutative points. For “good”
algebras, the non-commutative geometry covers the commutative geometry; in this case,
we will say that the non-commutative algebra is semi-classical. In this case we can think
of the commutative space as a coarse-grained version of the full non-commutative space.
For semi-classical algebras, our interpretation of irreducible representations as points is
equivalent to the point-like properties of D-branes in orbifolds. In other cases, the non-
commutative geometry may have little relation to the commutative geometry and it is not
clear that one should interpret the D-brane states as pointlike.
A general solution of the F -term constraints is a direct sum of irreducible represen-
tations, and thus the natural non-commutative structure is an unordered finite collection
of points. For commutative algebras, we would interpret this as a symmetric product
space, and we will carry over this name in the non-commutative case. This symmetric
product structure leads directly to an interpretation of D-brane fractionation at a singu-
larity, whereby an irreducible representation can be continuously deformed and becomes
reducible at a certain point. In this sense, in the non-commutative version, single-particle
and multi-particle states are continuously connected. In commutative geometry on the other
hand, this process would be singular.
The remainder of the formal discussion deals with an extension of this construction to
subvarieties, sheaves, and the algebraic K-theory of the ring, relevant to an understanding
of extended D-branes. The discussion presented here lays the outline for non-commutative
algebraic geometry; a full account will appear elsewhere [38].
3.2. Preliminaries: points and topology
Consider an associative algebra A over the complex numbers C, generated by a set

of operators subject to some relations. (In all the examples we consider, we will have
three generators and a set of quadratic relations.) As is standard, the non-commutative
algebra should be thought of as a ring of functions on some affine non-commutative
space [37]. Ring homomorphisms should correspond to holomorphic maps between affine
non-commutative geometries. We will assume that all rings are Noetherian (so that any
ideal always has a finite basis) and that the algebra is polynomial.
Given the matrix equations (2.5)–(2.7), we want to find solutions in terms of M × M
matrices (with unspecified M). That is, we are interested in representations of the algebra,
and we will assign a geometrical space to these solutions.
An element a ∈ A is central if it commutes with every other element in A; that is, a is a
Casimir of the algebra. We usually think of the Casimir operators as sufficient to define a
representation (e.g., as in the finite-dimensional representations of SL(2, C)) and thus will
pay particular attention to the center of the algebra, denoted ZA.
Since ZA is commutative, we can associate an ordinary commutative space to it, which
is the general philosophy behind algebraic geometry (see, for example, Ref. [39]). We
interpret the center of the algebra as a coarse description of the full non-commutative
geometry. The picture we have is that there is a map between the non-commutative
geometry to the commutative one which forgets some of the structure (namely the functions
that do not commute). The natural inclusion i : ZA → A is to be thought of as the pullback
of functions from the commutative space to the non-commutative space.
To describe the non-commutative space, we need to define the notions of points and
open sets and to impose a topology. Loosely speaking, a point will be a solution of
the constraint equations in finite matrices (just as points in varieties are solutions of the
equations defining the variety). In commutative algebra, points are interpreted as maximal
ideals of the algebra, and we want to incorporate both of these notions in our definition of
a point.
Let us now be precise. A representation R of dimension M of the algebra A is an
algebra homomorphism µ from A to the algebra of M × M matrices (i.e., µ respects
the addition, product, and multiplication by scalars). For the map to be well defined, the
relations (F -term constraints) must be satisfied in terms of the M × M matrices.
A representation is irreducible if there is no linear subspace of CM which is invariant
under multiplication by all the elements of the image of the algebra µ(A). The
representation is reducible otherwise. If a representation is irreducible, then the map is
such that we have an exact sequence
µ
A → MM (C) → 0 (3.1)
with the map µ defined by the representation of the algebra. The kernel of this map is
a double-sided ideal I of A to which the representation is associated, and we have the
isomorphism
A/I ∼ MM (C). (3.2)
This isomorphism is non-canonical (two representations are identified if they lie in the
same orbit of the group GL(M, C) by similarity transformations).
The space associated to an algebra is constructed from the irreducible representations
of A as follows. To each irreducible representation of finite dimension M, there is an
associated ideal in the algebra A, namely the ideal I = ker(µ). I is a double-sided
maximal ideal and is declared to be a point. This definition is borrowed from [37,40], but
without the C ∗ -algebra framework. In general, one would also allow infinite-dimensional
representations of the algebra; in that case, we would need some sense of convergence
for sequences. For our physical problem, we are interested only in finite-dimensional
representations, and so we will simply discard this possibility. As a result, we have a
space which is better behaved from an algebraic standpoint. Irreducible representations are
considered equivalent if they are related by a change of basis (i.e., by orbits of the group
GL(M, C)). This equivalence is the fact that we have in supersymmetric field theories a
complexified gauge group, and each point has an associated maximal double-sided ideal I
of the algebra A such that A/I is non-canonically isomorphic to the algebra of M × M
matrices. The variety associated to A will be labeled MA .
A closed set is defined in terms of an arbitrary double-sided ideal I 0 in A, as one does
to define the Zariski topology of a space. A closed set is the collection of points (given by
maximal ideals I) which contain I 0 . By definition, points are closed sets, as one takes the
maximal ideal I associated to the point.
The union of two closed sets corresponds to the double-sided ideal I = I1 ∩ I2 , and
the intersection of two closed sets corresponds to the double-sided ideal I1 + I2 , which
is the direct sum of the ideals. Indeed, direct sums may be extended to an infinite number
of ideals, so arbitrary intersections and finite unions of closed sets are closed and define
a topology on the set of points. This should be thought of as a model for the definition of
the geometry, and the construction mimics the construction of algebraic varieties over C
as much as possible.
Note that the definition of a point has the following technical property. A point is Morita-
equivalent to a point in a commutative algebraic variety. This is important for K-theory
considerations, which we return to in a later subsection.
3.3. Naturalness of symmetric spaces
So far, we have defined a non-commutative space together with some topology. Given
these definitions, additional structure naturally emerges, as we now discuss.
When one has the ring of functions of a variety, one can pull back functions between
maps of varieties. Thus, a map between non-commutative spaces will correspond to ring
homomorphisms. If we take two rings A and B and consider a ring homomorphism
ϕ : A → B, it will correspond to a continuous map from MB → MA .
Now consider a point x ∈ MB . By construction, it corresponds to an irreducible
representation of B in M × M matrices for some M. We label the corresponding
representation rx , i.e., a homomorphism µx : B → MM (C). Thus we have a diagram of
maps
ϕ µx
A → B → MM (C). (3.3)
We wish to find the image of the point x in MA . By composing arrows, we get a
natural representation of A in the ring of M × M matrices. The natural image of x is the
kernel of the composition map, but it is important to note that the corresponding M × M
representation of A might be reducible. In general, to find irreducible representations
associated to a given reducible one, we need to consider the composition series of the
representation R.
That is, given a reducible representation R, there is an invariant linear subspace R 0 , and
we have an exact sequence of vector spaces:
0 → R 0 → R → R/R 0 → 0. (3.4)
By construction the dimensions of R 0 and R/R 0 are smaller than that of R. If any of
R 0 , R/R 0 is reducible, we repeat the procedure. Eventually, we obtain a collection of
irreducible representations of A, and any two such decompositions contain the same
irreducibles.
The image of x should be considered a positive sum of points in MA with multiplicities
given by the number of times a particular irreducible representation of B appears. Thus the
map is multivalued, given the description used so far. If we wish to make such maps single-
valued, we can either restrict the choice of maps, or modify the definition of a point. The
latter possibility is most natural, and there is an obvious choice. We should consider instead
a new space consisting of the free sums of points with coefficients in Z+ . These sums of
points are generated by the points of M such that the sums are finite. We should consider
the maps between two such spaces as linear transformations between the two formal sums.
Notice that the formal positive sums of n points in a commutative variety M corresponds
to the symmetric product Mn /Sn . Hence the natural object to understand in this version of
non-commutative algebraic geometry is the symmetric product of the space MA . This is
also the framework necessary for matrix theory and matrix string theory [34,41], and this is
why we find it a very appealing aspect of the construction. We will denote this symmetric
product space by SMA . MA is a subset of its symmetric space, and it is the set which
generates the formal sums.
There is a grading present here which gives a notion of the degree of a point, which we
denote deg(x). There is a natural map from the formal sums of points to Z; namely, for each
irreducible representation x ∈ MA , we consider the map that assigns to x the dimension
of the representation that x is associated with (that is, the character trµx 1). This extends by
linearity to the symmetric product space, and the maps between the sums of points are such
that they are degree-preserving. Indeed, given any function on the space (an element of the
ring a ∈ A), we consider the invariant of the function a at the point x, trµx a. The trace is
linear, and independent of the choice of basis for the local matrix ring and can therefore be
extended to direct sums of representations (i.e., to the space SMA ).
Each positive sum of points of MA is associated with a representation of the ring A,
which is the direct sum of irreducible representations. To each element in the ring, we
associate a character in the representation. Consider the vector space associated to a
representation on which the matrix ring acts as a left module of the ring of functions.
Because the associated representations of points are only well defined up to conjugation,
we should impose the same constraint on the modules; namely, we want isomorphism
classes of modules for the ring A. This is very reminiscent of algebraic K-theory, and we
will expand on this idea later, making the connection precise.
Now, although we have found it natural to extend MA to SMA , it is not the case that
SMA automatically inherits a topology from that of MA . Rather, we should repeat the
construction of a Zariski topology, by giving a definition of closed sets. For any function
a ∈ A and for every complex number z, we define the following set
Z = {p ∈ SMA | trRp (a) = z} (3.5)
to be closed. These sets form a basis of closed sets for the topology of SMA . This topology
coincides with the natural topology in the commutative algebra of functions generated by
traces of operators, which are the polynomials in the gauge invariant superfields, and thus
gives us the same topological information that we would want for the moduli space if we
just consider the ring of holomorphic functions on the moduli space with values in C. For
later use, we define the support of a character as
Supp(tr a) = {p ∈ SMA | trRp a 6= 0}, (3.6)
where the overline denotes the closure operation.

We would like to be able to say that the space is foliated by sets of degree m (which will
count the D-brane charge of the point). Because the Zariski topology is coarse, we must
then additionally declare that the sets defined by deg(x) = m are both open and closed.
Now recall that for two different points in an algebraic variety V , there is some function
on the variety which distinguishes them. Only a finite number of these functions is needed
to determine a point exactly; the ring of polynomials (with relations) is finitely generated.
This construction is also sufficient to determine a collection of n unordered points of the
variety. By examining nth order polynomials in a function f , we can determine the values
that f takes at the n points. If the values are different for all the points for some function f ,
then one can use f as a coordinate, and one has a collection of n non-overlapping algebraic
subsets of V , with one point chosen from each one. Thus we can construct a function
which vanishes at all but one of the subsets, which we call f1 and by multiplying f1 by
all the basis functions of the ring associated to V , we can identify one of the points. The
procedure can be repeated if no one function is able to tell them all apart, and then we get
the multiplicities of the points.
In the non-commutative case, we say that we can always distinguish two irreducible
representations by some collection of characters (traces) of the ring A. Thus there is a
given finite number of functions with which we can distinguish n points.
Recall that we wish to think of the non-commutative symmetric space as a refined
version of the commutative space. Topologically, the two spaces are the same. It is clear
that, at least locally, given the characters of enough elements of the ring, we can fully
reconstruct a representation by holomorphic matrices on the commuting variables. This
endows the symmetric space locally with a holomorphic vector bundle structure.
3.4. The role of the center
Now let us apply the above construction to maps between the spaces SMA and SMZA .
Consider in particular the inclusion map ZA → A, which is the pullback of functions on
MZA to MA .
We want to know the image of a point p in MA . Given the point p, there is an associated
i µp
M-dimensional irreducible representation µp . Consider composing the maps ZA → A →
MM (C). As the last map is onto, if a ∈ ZA, it commutes in the image of the composition
of maps and by Schur’s lemma is proportional to the identity. The representation associated
to p splits into M identical copies of a single representation of ZA, namely, into M copies
of a single point. We write this as
p 7→ M p̄, (3.7)
where p̄ is the associated maximal ideal of ZA, which is the kernel of the inclusion map.
We will call this the natural map, as it respects degree. Notice that we can also define
a map between the symmetric spaces p 7→ p̄ which forgets the degree. In this case, the
image of a point is a point, so we can restrict the maps to MA and MZA . We will call
this the forgetful map.
The center of the algebra will play an important role in the physics. We wish to restrict
the maps between rings A and B in the following way. Note that we have the following
diagrams
A B SMB SMA (3.8)
ZA ZB SMZB SMZA
We require that these diagrams be commutative; namely, the ring homomorphism A →
B induces a map ZA → ZB, and consequently SMZB → SMZA . Thus, we want the
map to be such that central elements are central not just in the subalgebra of the image of
A but in the algebra B itself.
Now we are at a point where we can build a category for the non-commutative algebraic
geometry. The objects will be rings (A) with the center identified and the inclusion map
singled out. 2 The allowed maps between rings are such that they produce commuting
squares
A B (3.9)
ZA ZB
with the upwards arrows the natural inclusion maps.
The non-commutative space is a contravariant functor from this category of rings to a
category of ‘symmetric spaces’ as we have defined previously (including the degree map
and the degree-preserving property). Thus we have the diagram
SMB SMA (3.10)
SMZB SMZA
Together with the center preserving property, the forgetful map induces
MZB → MZA (3.11)
This is a map of commutative algebraic varieties, to which we can apply intuition. It is also
clear that this map between varieties has all the data required to specify the map between
their symmetric spaces. With the topology of these spaces, all the arrows are continuous
maps, and composition of maps is a map that respects the properties of the category.
To make a full connection with algebraic geometry, we want to be able to glue rings on
open sets. This should be done by a process of localization. These details will be left for a
future publication [38].
It is useful to notice that all the irreducible representations of the center may not appear
when we consider the projection map, SMA → SMZA . If most 3 do appear, then we will
call the algebra semi-classical, because to the points in the variety associated to the center,
we can lift to points in the non-commutative variety. The non-commutative variety covers
2 It is appropriate then to use the larger notation (A) ∼ (A, ZA, i : ZA → A).
3 That is, an open set of M
ZA in the Zariski topology.
the commutative one and this notion will be important from several perspectives below.
In particular, there are applications involving D-branes in which phenomena on orbifold
spaces are more precisely described by non-commutative geometry.
3.5. D-brane fractionation
The technology developed so far contains some interesting aspects of D-brane physics.
In particular, we wish to show that a D-brane fractionates as we move to a singular point of
a non-commutative algebraic variety. In fact we define singular points via this process of
fractionation. We will consider in this subsection D-branes which correspond to points in
MA , and the degree of the point is identified with the D-brane charge. The moduli space
of supersymmetric configurations of D-branes is identified with SMA .
Let R be an irreducible representation of dimension M in A. Consider its image in
SMA as a single point of degree M. Because SMA is an algebraic variety, it will consist
of several components or branches. The branch of SMA where R is located is a closed
set of some complex dimension d which is not a closed subset of any set with larger local
dimension.
On this branch, we can define a local function which is the dimension of the commutant
ZR of the representation R. As we move along the branch, this function is semi-continuous
— it may jump in value on closed sets.
Clearly, the sets with dim(ZR) > 1 are closed. In this case, we have at least two linearly
independent matrices which commute with everything in the image of A, and thus the
representation cannot be irreducible. For irreducible representations, dim(ZR) must be
unity. Thus, if we start at a point on a branch of the variety that is irreducible, as we
continuously deform along it, we can reach a special point as a limit point, where the
representation becomes reducible.
Parametrize this deformation by z; on the symmetric product space we have the process
lim x(z) = x1 + · · · + xn (3.12)

z→z0
if z0 is such a limit point, and where n is the number of irreducible representations that R(z)
splits into. Then, we say that the D-brane has fractionated, and there may be additional
branches that intersect that point, corresponding to separating the fractional branes.
From the point of view of the center of the algebra, each element is proportional to the
identity throughout the branch of the symmetric space, and thus there is no splitting seen in
SMZA . In this sense, the non-commutative geometry is a finer description of the D-brane
moduli space than the associated commutative geometry.
In the cases where there are branches corresponding to separating the branes at z0 , if we
think in terms of the forgetful map, we would have a single point splitting into n points.
From the point of view of the commutative algebraic variety, there is a jump in dimension
as we go from one branch to the next; this is naturally associated with a singularity. We
will see explicit examples of this in Section 4.
3.6. Higher-dimensional branes
So far, we have considered D-branes that are point-like on the moduli space. We
would also like to identify more general brane configurations, such as those wrapped on
holomorphic subspaces; hence we need to construct such objects algebraically. It is natural
to consider coherent sheaves [35]: these are the modules over the ring A which locally
have a finite presentation and are well-behaved when considered from the commutative
standpoint. Extended BPS brane solutions usually correspond to stable sheaves, given some
appropriate notion of stability. Moreover, they are also well-behaved as far as K-theory is
concerned. However, in order to define these structures, it is most convenient to have a
semi-classical ring. This does not mean that there is no useful way to define these objects
for more general rings, but on some rings where the points are discrete there is no obvious
notion of an extended object. For the rest of this section, we will assume that we are indeed
working in a semi-classical ring.
As the ring is semi-classical, we can try to construct extended objects by first building
them over a holomorphic subspace of the commutative structure, and then try to lift them
up to the non-commutative geometry. In the commutative case, a D-brane corresponds
to a coherent sheaf with support on a commutative subvariety. For any notion of non-
commutative sheaf, it must be the case that it is also a coherent sheaf over the commutative
ring. In the commutative case, the D-brane is a module over the ring ZA, such that if
ZI is the ideal corresponding to the support of the sheaf, the module action of ZA factors
through ZA/ZI, which is considered to be the coordinate ring of the closed set associated
to the ideal ZI.
On ‘good’ varieties we always have a presentation of a sheaf S as the right-hand term of
some exact sequence
ZAm → ZAn → S → 0 (3.13)
that is, as a module with n generators with relations induced by the images of ZAm .
We want to mimic this construction for the non-commutative version of the D-brane.
Note that in the non-commutative case, we have a choice of left-, right- or bi-modules of
the algebra A. However, physically, we need to consider only bi-modules, as both ends of
open strings end on a D-brane. That is, gauge transformations (which act locally) act both
on the left and right, and therefore the algebra has to be able to accommodate both types
of actions on the modules. Referring to the bi-module as R, we want them to arise from
exact sequences
Am → An → R → 0 (3.14)
in analogy to Eq. (3.13). This defines the coherent sheaves over A.
locally 4
The annihilator of a bi-module R is defined as the largest ideal I of A such that IR =
RI = 0. This is a double-sided ideal, and thus defines a closed set, in the topology of MA .
For any point p in MA which does not belong to I, I + Ip is equal to A (because Ip is
maximal). We will refer to this ideal I as Ann(R).
4 Recall that we are not concerned with gluing.

We can find how a bi-module restricts to a closed subset (described by an ideal I) by

noticing that if R is a bi-module over A, then R/(IR + RI) is a bi-module over A/I. For
maximal Ip the restriction is zero if Ann(R) 6⊂ Ip . We can define the support of a sheaf to
be the set of points such that
Ann(R) ⊂ Ip . (3.15)
We study the sheaves locally by restricting to a point. We will look at two different
notions of the rank of a sheaf. Each non-commutative point is Morita equivalent to a
commutative point. This tells us that modules over the functions restricted to a point p (the
ring of n × n matrices) behave just as vector spaces over C. Thus the sheaf restricted to a
point is a bi-module over the ring of deg(p) × deg(p) matrices, and thus R|p is isomorphic
to (A|p )k , for some k. One possible definition of non-commutative rank of a sheaf at the
point p is just k. However, as seen from the commutative standpoint, the dimension of the
representation associated to (A|p )k is k deg p, and this then serves as another definition of
rank, which we will refer to as the commutative rank.
As usual, rank is upper semi-continuous (the points where rank(R)|p > m form a closed
set). The rank can jump in value on some closed subset, and this is interpreted in terms of an
additional D-brane of smaller dimension stuck to the brane, as follows from the anomalous
couplings of D-branes [42].
For the non-commutative points, we have to take into account that a limit set of a
collection of points might be a sum of points. Consider the trivial bi-module of A,
namely A. The non-commutative rank (equal to 1) does not jump under the fractionation
process. The commutative rank on the other hand, does jump at this singularity. The non-
commutative rank then is the natural definition of rank for non-commutative algebras.
Note, however, that if we look just at the center ZA, the commutative rank is the natural
definition, as it does not jump in a splitting process; we have
X
deg(p) = deg pi . (3.16)
i
With these definitions, a D-brane is a coherent sheaf over both the non-commutative
ring and the commutative sub-ring.
3.7. K-theory interpretation
We have seen that our approach to non-commutative geometry has led us to some
definitions of D-brane states. We now want to add K-theory to the discussion. Because
we have a ring and we have bi-modules, we get automatically a K-theory associated to
this structure, namely, the algebraic K-theory of the ring A (see [43] for example). From a
mathematical standpoint this is review material, and we will just glimpse at the dynamics
in terms of brane–antibrane systems [23,44].
Indeed, let us start with the construction of the symmetric space. We had formal sums
of points which makes the non-commutative geometry a semigroup. We can make it into
a group by adding minus signs, and a rule for cancellation. This group is the equivalent of
zero-chains of points.
Now, the idea of the group structure is to understand that p + (−p) = 0. So if p is a

point, we interpret it as a point like D-brane, and −p is an anti-D-brane. The cancellation
law of the group is the statement that a D-brane anti-D-brane pair can be created from the
vacuum.
Given these minus signs, the degree function now maps to the integers and gives us
an invariant, which is a group homomorphism of Abelian groups, deg : ⊕p → Z. This
number is the total D-brane charge of a configuration. It is possible to give a topology to
finite formal sums of points by the same construction we used based on characters. The
extra ingredient to make p − p = 0 is to add minus signs for the anti-D-brane.
Thus dynamically p − p = 0 is the statement that any character of p − p is the same as
a character of zero, and thus the configurations are connected. When we create a D-brane
anti-D-brane pair we can separate them if there are moduli available, and thus the process
is continuous in this topology. Dynamical information would include the energy required
for this process. (A generic point is such that this energy is much less than the energy
required to move off of the moduli space of sums of points. A non-generic point is where
the mass matrix has zero eigenvalues.)
The idea now is to define the K-theory of points as a homotopy invariant which respects
p
the additivity of branes. On one hand, we have the mathematical definition of the K0 -
theory of points as the formal abelian group of homotopy classes of finite-dimensional
representations of the algebra A, such that if a, b are such representations, then the
K-theory classes associated to a ⊕ b, a, b satisfy
K(a ⊕ b) = K(a) + K(b) (3.17)
and if one has a homotopy between the two representations a ∼ b, then K(a) = K(b).
Indeed, to the point p we associate the bi-module A|p , and this is thought of as the
skyscraper sheaf over p. With this extra relation this is part of the K-theory of bi-modules
of the algebra.
The other way to define K-theory is to say K(a) = K(b) if there is c such that a ⊕ c ∼
b ⊕ c. Both of these definitions agree.
There are a few possible choices of K-theory depending on the type of modules one
chooses. As we have stated above, in this paper we are interested in finitely presented bi-
modules over the ring A. Generically, one defines the K-theory associated to projective bi-
modules of the algebra. The K-theory of projective bi-modules is the same as the K-theory
of finitely presented bi-modules as long as every bi-module admits a projective resolution
(this is true, for example, in smooth manifolds where every vector bundle is projective over
the coordinate ring of the manifold). Thus, as long as there is a long exact sequence
0 → P1 → · · · → Pk → M → 0 (3.18)
with the Pi projective, then the K-theory class of M is defined. This requires the ring to be
regular. Whether or not we will always get regular rings in string theory in this framework
is not clear. (Singular varieties are not regular as commutative algebras, yet they do appear
in string theory.)
The K0 -theory is defined as the set of formal sums of bi-modules modulo homotopy,
and modulo the relations
K(b) = K(a) + K(c) (3.19)
whenever there is a short exact sequence
0→a →b→c→0 (3.20)
of the bi-modules we have described. We think of this as the statement that b − a − c = 0.
If a module M admits a projective resolution as in (3.18) then it is a simple exercise to
show that
K(P1 ) − K(P2 ) + · · · − (−1)k K(Pk ) + (−1)k K(M) = 0. (3.21)
Physically, we say the dynamics of brane–anti-brane configurations is such that given a
short exact sequence (3.20), the process
X →X±a ∓b±c (3.22)
is allowed, namely a − b + c carries no D-brane charge. In particular, the exact sequence
0→a →a→0→0 (3.23)
will allow any of the two processes
X → X + a + (−a) → X, (3.24)
which correspond to the creation of brane–anti-brane pairs. Of course, the real dynamics
of these processes is not available to us, but the topology of allowed transitions is correctly
reproduced.
Note that taking tensor products of bi-modules is locally a good operation (at each point
we are taking a tensor product of finite-dimensional spaces, and we get a finite-dimensional
space). Thus the K-theory is not just an additive group but we have a multiplication as well,
and this permits us to do intersection theory (that is, we can count strings when branes
intersect). This type of information can often be enough to calculate topological quantities
in string theory.
As a final comment, we have to give some warnings to the reader. The K-theory we have
constructed here is the one associated to the holomorphic algebra, and thus is a version
which is relevant for the algebraic geometry. This K-theory is an invariant of a holomorphic
space which is much finer than the topological K-theory, and thus contains a lot more
information. The K-theory which is relevant for D-brane charge is the one associated
to both the holomorphic and anti-holomorphic structures, namely, the K-theory of a C∗
algebra of which A is a subalgebra. This C∗ algebra includes the D-term constraints,
and by theorems on existence of solutions [33] the geometric space associated to the C∗
algebra has just as many non-commutative points as the one associated to A. So just as
in commutative cases, the non-commutative holomorphic data parametrize the full variety.
The K-theory of the two algebras does differ. The holomorphic K-theory is therefore more
appropriate to count BPS states, rather than just account for the D-brane charge.
This is the end of the mathematical preliminaries. We believe that we have presented a
fairly general account of how applications of these techniques might be pursued. We will
see that this approach is not just a big machine which describes things we already knew in
a complicated manner. Indeed, once we have examined the examples in the next sections,
it will be clear that the formulation brings sound intuition and gives a very nice picture of
how string geometry behaves.
4. Examples
In this section, we consider a variety of examples in order to build a picture of the generic
behavior of the geometry, which is not present in the simplest case. The presentation is
given in terms of the language of Section 3; the reader will find it necessary to read, at least,
the overview in Section 3.1. In the first few examples, we first calculate the commutative
algebra of the center which reproduces the string geometry associated with the field theory
(e.g., the orbifold). We then attempt to build the irreducible representations of the full non-
commutative algebra by exploiting knowledge of the center. A posteriori, the structures that
we find here and the relations to the physics of D-branes in these geometries discussed in
later sections, motivates the formal constructions of Section 3. In more general examples,
the calculation of the center is difficult, and we present only partial results.
4.1. Orbifolds with discrete torsion: the q-deformation
Our first example to study will be orbifolds with discrete torsion. In particular, we
consider the orbifold C3 /Zn × Zn with maximal discrete torsion. To construct the
low energy effective field theory of a point-like brane one can use a quiver construc-
tion [45] with projective representations of the orbifold group [6]. The use of projective
representations was justified in Ref. [46]. The algebraic variety associated to the orbifold
singularity is given by the solutions of one complex equation in four variables,
xyz = wn . (4.1)
As Douglas showed [6,47], the theory has N = 1 supersymmetry in four dimensions and
consists of a quiver with one node, gauge group U (M), three adjoint superfields and a
superpotential
Wq = tr(φ1 φ2 φ3 ) − q tr(φ2 φ1 φ3 ) (4.2)
with q a primitive nth root of unity. This theory can be obtained by a marginal deformation
of the N = 4 supersymmetric field theory as shown in [1,7], and as such, when studied
under the AdS/CFT correspondence, displays a duality between two totally different near-
horizon geometries, describing the same field theory.
The F -term constraints are given by
φ1 φ2 − qφ2 φ1 = 0, (4.3)
φ2 φ3 − qφ3 φ2 = 0, (4.4)
φ3 φ1 − qφ1 φ3 = 0. (4.5)
We will often write these using the q-commutator notation [φ1 , φ2 ]q = 0, etc. These
equations are exactly the type of relations seen in the algebras related to quantum
planes [18], and have been very well studied. Let us analyze the algebra using the tools
described in Section 3.
Because of the F -term constraints, we can always write any monomial in ‘standard
order’
k
φ1k1 φ2k2 φ33 . (4.6)
We associate to this monomial the vector (k1 , k2 , k3 ).
Note that if an element commutes with φ1 , φ2 , φ3 , then it commutes with any of the
monomials, and thus is an element of the center of the algebra. Monomials may be
multiplied, and up to phases, we have
(k1 , k2 , k3 ) · (s1 , s2 , s3 ) ∼ (k1 + s1 , k2 + s2 , k3 + s3 ). (4.7)
Because of the phases, generators of the center are monomials.
It is easy to see that (k1 , k2 , k3 ).φ1 = φ1 .(k1 , k2 , k3 )q k3 −k2 , so that k3 = k2 mod n for
(k1 , k2 , k3 ) to be in the center. Similarly one proves k2 = k1 mod n and thus the center is
given by the condition
nX o

ZA = (k1 , k2 , k3 )k1 = k2 = k3 mod n . (4.8)
This is a sub-lattice of the lattice of monomials, and it is generated by the vectors

(1, 1, 1), (n, 0, 0), (0, n, 0), (0, 0, n). Call w = (1, 1, 1), x = (n, 0, 0), y = (0, n, 0) and
z = (0, 0, n). Clearly we have the relation
(−w)n + xyz = 0 (4.9)
so we see the orbifold space is described by the center of the algebra. The singularities
occur along branches where two of x, y, z are zero.
Now that we have the commutative points, let us consider the non-commutative points
of the geometry. We should consider the irreducible finite-dimensional representations of
the algebra. Because x, y, z are central, on an irreducible representation of the algebra they
act by multiples of the identity.
Suppose at least two of x, y, z are non-zero (say x, y). In this case (1, 0, 0) and
(0, 1, 0) are invertible matrices. By a linear transformation, we can diagonalize (1, 0, 0).
Consider an eigenvector |ai of (1, 0, 0) with eigenvalue a. We see that |qai ≡ (0, 1, 0)|ai
is an eigenvector of (1, 0, 0) with eigenvalue qa. Thus we get a collection of states
|ai, |qai, . . ., |q n−1 ai constructed as |ai, (0, 1, 0)|ai, . . ., (0, n − 1, 0)|ai. This sequence
terminates (and thus the representation is of dimension n) because (0, n, 0) is central, and
q n = 1. A set of matrices which satisfies these conditions is
(1, 0, 0) = aP , (4.10)
(0, 1, 0) = bQ (4.11)
with P and Q defined by 5

  0
1 0 0 ... 0 0 ... 0 1
0 q 0 ... 0  1 0 ... 0 0
   
 
Q= 0
2 ...
P = 0 0 q 0 ,  0. 1. . .. . 0 . (4.12)
. .   .. .. .. 
.
.. . . .. ..
 .. .. . . .  .. .
0 0 0 ... q n−1 0 0 ... 1 0
As w = (1, 1, 1) is central, it is proportional to the identity. It trivially follows that
(0, 0, 1) = cQ−1 P −1 .
Notice that our solutions are parameterized by three complex numbers, namely a, b, c.
It is easy to see that x = a n I , y = bnI and z = −(−c)n I , w = abcI , and that one can
cover the full orbifold with these solutions, except for the singularities (where two out of
the three x, y, z are zero). Notice also that the covering is done smoothly, so any two points
can be connected by a path which does not touch the singularities.
If we label the representation by R(a, b, c), it is easy to see that R(a, b, c) is
equivalent under a similarity transformation to R(qa, q −1b, c) and R(qa, b, q −1c). Thus
the eigenvalues of the center completely describe the representation. That is, for any
commutative point which is non-singular, we have a unique non-commutative point of
degree n sitting over it.
Let us now analyze the case where two of the three x, y, z are zero. Then w = 0 as well,
and we are along one of the singular branches of the orbifold. Let us assume that x 6= 0;
then (1, 0, 0) is invertible, and can be diagonalized. On the other hand (0, 1, 0).(0, 0, 1) =
(0, 0, 1).(0, 1, 0) = (0, n, 0) = (0, 0, n) = 0 in the representation. Given any vector v in
the representation, v0 ≡ (0, n − 1, 0)v is annihilated by (0, 0, 1) and (0, 1, 0), and any
other vector obtained by multiplying with (1, 0, 0) enjoys this same property. Thus given
a representation, we find a sub-representation where both (0, 0, 1) and (0, 1, 0) act by
zero. As (1, 0, 0) is invertible it can be diagonalized in this subrepresentation. Clearly the
representation is irreducible only if it is one-dimensional, and determined by the eigenvalue
of (1, 0, 0), which is a free parameter that we call a. The value of x is a n , and for each point
in the singular complex line y = z = 0 we find n irreducible representations of the algebra,
except at the origin. The same result holds when we go to any of the other complex lines
of singularities. We label these representations by R(a, 0, 0), etc.
Here R(a, 0, 0) is not equivalent to R(qa, 0, 0). They are equivalent as far as the
commutative points are concerned, because both of these representations have the same
characters over the center of the algebra. But as far as the non-commutative points are
concerned, the characters of the non-central element (1, 0, 0) differ. That is
trR(a,0,0)(1, 0, 0) = a. (4.13)
Thus we have two distinct points. It is also clear that any one of these representations can
be continuously connected to any other.
5 We have changed basis compared to Ref. [6].

Fig. 2. Quiver diagram for the An−1 singularity.
These smaller representations are not regular for the C3 /Zn × Zn orbifold and may be
identified with the fractional branes. Notice also that
trR(a,b,c)(1, 0, 0) = a tr P = 0 (4.14)
so that this character is different from zero only at the classical singularity. This is the
primary reason for adopting the convention for the support of a character in Eq. (3.6).
To summarize, for each point in the classical moduli space we have at least one point
in the non-commutative space which sits over it. This is an example of a semi-classical
geometry (see Section 3). The commutative singular lines are covered by an n-fold non-
commutative complex plane branched at the origin.
Now consider what happens when we bring a point from the regular part of the orbifold
towards the singularity. The representation behaves in this limit as
lim R(a, b, c) = R(a, 0, 0) ⊕ R(qa, 0, 0) ⊕ · · · ⊕ R(q n−1 a, 0, 0). (4.15)

b,c→0
In our description of moduli space, this corresponds to the branes becoming fractional at
the orbifold fixed lines, as we have discussed previously. Indeed, once we reach this point
we can separate the fractional branes, and the non-commutative symmetric product is the
right tool for describing the moduli space in full.
We can also see the quiver of the singularity type by consideration of this same limit.
Indeed, we assign a node to each irreducible representation in the right-hand side of
Eq. (4.15). We draw an arrow between any two nodes appropriate to the non-zero entries
in Eqs. (4.12) and we obtain Fig. 2 which is indeed the quiver diagram of the orbifold in
the neighborhood of a point in the singular complex line.
Thus the singularities can be said to be locally quiver. From the point of view of the
center of the algebra, the nodes of the quiver are at the same point, but they are distinct
in the non-commutative algebra. The behavior of the field theory near the singularities is
precisely what we would get from the orbifold analysis.
Recall that the commutative singular lines are covered by n non-commutative branches.
The monodromies of the quiver diagram are encoded in this structure, and thus
their calculation is geometrically obvious. This compares quite favorably to the rather
cumbersome procedure employed in [7]. Indeed, we can change a → ωa for ω = e2πi/n .
This results in a permutation of the factors appearing on the right-hand side of Eq. (4.15).
This permutation is the monodromy.
4.2. Adding one mass term
Next, we consider a relevant deformation of the last theory, obtained by the addition of
a single mass term. This theory is a q-deformed version of the theory which flows in the
infrared to an N = 1 conformal field theory [1,8]. The superpotential is

m 2
W = tr φ1 φ2 φ3 − qφ2 φ1 φ3 + φ3 . (4.16)
2
Again, we assume that q is an nth root of unity. The F -term constraints are given by
[φ1 , φ2 ]q = −mφ3 , (4.17)

[φ2 , φ3 ]q = 0, (4.18)
[φ3 , φ1 ]q = 0. (4.19)
As in the previous case, we look for the center of the algebra to obtain the commutative
manifold. It is easy to see that z = φ3n is still in the center. We can also show that
[φ1n , φ2 ] = φ1n φ2 − qφ1n−1 φ2 φ1 + qφ1n−1 φ2 φ1 − q 2 φ1n−2 φ2 φ12 + · · ·

= φ1n−1 (−mφ3 ) + qφ1n−2 (−mφ3 )φ1 + · · ·
X
n−1
= −mφ1n−1 φ3 q 2r , (4.20)
r=0
which vanishes, apart from at the special values q = ±1. Similarly one proves that φ2n is
central, away from q = ±1. For now, we will assume that q 2 6= 1, and return to these cases
later. Thus we have at least three central variables x = φ1n , y = φ2n , z = φ3n .
The variable w is modified by the presence of the mass term. Consider the commutator
[φ1 φ2 φ3 , φ1 ] = φ1 φ2 φ3 φ1 − qφ1 φ2 φ1 φ3 + qφ1 φ2 φ1 φ3 − φ1 φ1 φ2 φ3

= mφ1 φ32 .
This result may be rewritten as a commutator for q 6= ±1, and thus we see that
m
w = φ1 φ2 φ3 + φ2 (4.21)
1 − q2 3
is central.
The four variables x, y, z, w are related by
n
m
xyz = −(−w) − n
z2 . (4.22)
1 − q2
This is a deformation of the complex structure of (4.9). It is easy to see that we now
have singularities at w = xz = yz = xy + 2tz = 0 with t = (m/(1 − q 2 ))n . Thus, the
singularities are at xy = 0, w = 0, z = 0, and so we have two lines of singularities x = 0
and y = 0. The mass term has resolved one of the three complex lines of singularities (for
q 2 6= 1).
It is easy to check that a general solution is of the form
φ1 = aP , (4.23)
−1 −1 −1
φ2 = −bP Q − cP Q , (4.24)
−1
φ3 = dQ (4.25)
with P , Q defined as in (4.12), and where a, b, c, d are numbers satisfying
ac(1 − q 2 ) = md. (4.26)
One then gets x = a nI , y = −(bn + cn )I , z = d n I and w = −abdI . Note that this
representation has been chosen such that φ1 is diagonal at the singularity y = 0, z = 0.
Because we have a three complex parameter solution of the equations, we at least cover
an open patch of the commutative variety, and we are again in a semi-classical ring. Indeed,
we cover everything by finite matrices except x = 0, as then c is infinite.
A patch which does cover x = 0 is given by
φ1 = −aP Q−1 − cP Q, (4.27)
−1
φ2 = bP , (4.28)
−1
φ3 = dQ (4.29)
with ab(1 − q 2 ) = mqd. This will be a good description for y 6= 0. The two patches cover
the two lines of singularities. There is still the closed set x = y = 0 which is not covered
by either patch. We can find solutions for this set by taking φ3 diagonal and making an
ansatz for φ1 which is upper triangular with entries just off-diagonal and φ2 a similar
lower triangular matrix. The dimension of this representation is also n and depends on one
complex parameter, namely the eigenvalues of φ3 .
On approaching the singularity y = z = 0 from the bulk, we again get a split set of
irreducible representations as follows:
lim R(a, b, d) = R(a, 0, 0) ⊕ R(qa, 0, 0) ⊕ · · · . (4.30)
b,d→0
4.2.1. Comments on the infrared CFT

This case is also very interesting from the field theory perspective because by adding
one mass term to a theory with three adjoints, we obtain a nontrivial conformal field theory
in the infrared.
On the moduli space, we have the following U (1) symmetries
a, b, c → λγ a, λγ −1 b, λγ −1 c, (4.31)
d → λ d, 2
(4.32)
where we refer to the parameterization for x 6= 0. The transformation given by γ is an
ordinary U (1), while λ is the U (1)R symmetry, that of the superpotential in the infrared.
The R charges can be chosen in such a way that the superpotential is invariant at the
infrared fixed point. Indeed, one can integrate out φ3 and one finds a theory in the infrared
with a quartic superpotential
1
tr(φ1 φ2 − qφ2 φ1 )2 . (4.33)
2m
This superpotential is a marginal deformation of the infrared theory. Note that we also have,
in the infrared, a Z2 symmetry φ1 ↔ φ2 which ensures that the anomalous dimensions of
φ1 and φ2 are equal. (In the ultraviolet, this Z2 symmetry is absent, as we would also have
to simultaneously exchange q → q −1 and rescale m.) This symmetry exchanges the two
singular complex lines of the commutative moduli space.
4.2.2. Special cases: q = ±1

Let us return to discuss the moduli space for the cases q = ±1 from the algebraic point
of view.
For q = 1, which is the N = 4 theory with one mass term, the moduli space is the set of
solutions to
[φ1 , φ2 ] = −mφ3 (4.34)
with all other commutators vanishing. For an irreducible representation, φ3 is central and
thus a constant. Because the commutator of φ1 , φ2 is a constant, we get the Heisenberg
algebra, and the only finite-dimensional representations are those with φ3 = 0. Thus the
moduli space is a commutative space consisting of the symmetric product of the complex
plane, C2 . Notice that this space is of complex dimension two and not complex dimension
three as in the generic case studied above. Indeed, in this case the center of the algebra is
generated by φ3 . Because φ3 = 0 on the moduli space, we can actually relax the condition
for an element being central: we can take φ1 , φ2 as central elements, which makes the
moduli space commutative.
Indeed, this is a case where the algebra is not semi-classical. The variety associated to
the center is the algebra of C. The non-commutative space is C2 , which projects to the
origin of C. The two have almost nothing in common.
As far as the commutative variety is concerned, the moduli space is a point. Notice that
in this case when we integrate out the field φ3 we get the correct dimension of the moduli
space by counting fields. This does not happen for generic q.
For q = −1, we can find the two-dimensional solution
φ1 = aσ1 , φ2 = bσ2 , φ3 = 0 (4.35)
plus two one-dimensional branches where either φ1 or φ2 is zero.

Here, the center is generated by φ32 . Indeed, it can be shown that these solutions exhaust
the list of irreducible representations of the q = −1 algebra. This result follows from the
fact that [zx, y] ∼ z2 , so a finite-dimensional representation must have z2 = 0. 6
The lesson to be learned from these special examples is that the commutative and non-
commutative spaces may contain little or no information about each other when the center
of the associated algebra is small. In this case, the full algebra is an infinite-dimensional
vector space over the center, and by considering only finite-dimensional representations,
we miss a lot of information.

6 Nilpotent possibilities, such as φ ∼ 0 a
3 are ruled out by D-terms.
0 0
4.3. One mass term and a linear term
We can easily modify the previously studied cases by adding a linear term to the
superpotential

m 2
W = tr φ1 φ2 φ3 − qφ2 φ1 φ3 + φ3 + ζ3 φ3 . (4.36)
2
Note that by a field redefinition of φ3 , this is equivalent to adding a mass term
ζ3
m (q − 1) tr φ1 φ2 . We will see that the usual intuition for mass terms fails in this case,
namely, that the moduli space is not destroyed by the quadratic terms. On the other hand, if
we had added tr φ1 φ2 for q = 1, we would indeed expect the space of vacua to be reduced
to a set of points.
It is straightforward to show that x = φ1n , y = φ2n , and z = φ3n are central, and that
m ζ3
w = φ1 φ2 φ3 + φ2 + φ3 (4.37)
1 − q2 3 1 − q
is also central, provided that q 6= ±1.
The relation between the central elements is
n n
m ζ3
xyz = −(−w) − n
z +
2
z (4.38)
1 − q2 q −1
and a generic solution of the equations is provided by
φ1 = aP , (4.39)
−1 −1 −1 −1
φ2 = −bP Q + cP − dP Q , (4.40)
−1
φ3 = eQ (4.41)
with ac(1 − q) = −ζ3 , ad(1 − q 2 ) = me, and x = a n , y = −bn + cn − d n , z = en ,

w = −abe. The singularities now occur at z = w = 0 and
n
ζ3
xy = . (4.42)
q −1
The two complex lines of singularities that met at the origin when ζ3 = 0 are now replaced
by a single C∗ , a cylinder. In the parameterization above, this corresponds to b, d, e = 0.
We see that the non-commutative C∗ is an n-fold cover of the cylinder without branch
points, and again the monodromies of the cover are manifest, since we chose φ1 diagonal.
This is again a semi-classical ring.
In addition, there are finite-dimensional representations which may be thought of as
deformations of SU(2) representations. These occur for x = y = 0 and cover regions not
captured by the parameterization above. Some solutions give rise to isolated fractional
branes at x = y = 0. A similar effect occurs in Section 4.5 and we will return to a full
discussion there.
The values q = ±1 are special, as in previous cases, in the sense that singularities occur,
and the non-commutative algebra is not semi-classical.
4.4. Three linear terms
Consider the superpotential

X
W = tr(φ1 φ2 φ3 ) − q tr(φ2 φ1 φ3 ) + (q − 1)ζi tr φi . (4.43)
i
This case was studied in Ref. [6] using gauge invariant variables. Our conclusions will be
consistent with that analysis.
For convenience, we have rescaled the ζ parameters by a factor of (q − 1). The F -terms
give
[φ1 , φ2 ]q = (1 − q)ζ3, [φ2 , φ3 ]q = (1 − q)ζ1 , [φ3 , φ1 ]q = (1 − q)ζ2 .
(4.44)
A possible parameterization is
ζ3 ζ1 ζ2
φ1 = aP − Q−1 P , φ2 = −bP −1 Q + Q, φ3 = cQ−1 + P −1 . (4.45)
b c a
Note that x1 = φ1n , x2 = φ2n and x3 = φ3n are central, while the fourth central variable takes
the form
w = φ1 φ2 φ3 − ζ1 φ1 − qζ2 φ2 − ζ3 φ3 . (4.46)
In the given basis, we find x1 = a n − (ζ
3 /b)n ,
x2 = −bn + (ζ 1 /c)n , x3 = cn + (ζ 2 /a)
n and
−w = abc + q ζ1abcζ2 ζ3
.
These four variables are related on the moduli space by
X
w
x1 x2 x3 − ζi xi + 2β Tn −
n n
= 0, (4.47)
2β
i
where β ≡ (qζ1 ζ2 ζ3 )1/2 and Tn (x) = cos(n cos−1 x) is the nth Chebyshev polynomial of
the first kind.
4.5. Three mass terms
Next, we consider a rank 3 mass term of the form

1 X
W = tr(φ1 φ2 φ3 ) − q tr(φ2 φ1 φ3 ) + m tr φi2 . (4.48)
2
i
This superpotential has a Z2 × Z2 symmetry that changes two of the φi → −φi , and a Z3
cyclic symmetry that permutes the φi . This is the remnant of the SU(4)R symmetry group
of the N = 4 SYM theory. The group generators do not commute with each other, and this
symmetry is enhanced to SU(2) when q = 1. Thus the symmetry is a subgroup of SU(2)
which contains a Z2 × Z2 and a Z3 subgroup. These are the symmetries of the tetrahedron,
b6 , and since they arise from the SU(4) R-symmetry they are chiral.
E
This superpotential yields the F -flatness conditions (cyclic on j , mod 3)
[φj , φj +1 ]q = φj +2 , (4.49)
where we have rescaled the fields in order to eliminate a factor of m.
We wish to find representations of this algebra; we will not immediately assume that
q is a root of unity. There is a certain class of solutions which may be thought of as
deformations of representations of SL(2, C).
Note that (for q 6= 1) there is a one-dimensional representation
1
φj = . (4.50)
1−q
Higher-dimensional representations may always be constructed as φi = 1−q 1
I , but this is
clearly reducible. An irreducible 2-dimensional representation (for q 6= −1) is given by
−i
φj = σj , (4.51)
q +1
where the σj are the Pauli matrices. We can construct higher-dimensional irreducible
representations by making the following ansatz: we suppose that one of the fields, φ3 ,
is diagonal and traceless, and that the other two fields only have non-zero elements just off
the diagonals. (For q = 1, these reduce to standard M-dimensional SL(2, C) generators.)
We have not been able to construct a proof that all such irreps may be obtained this way.
These are the representations which respect the discrete chiral symmetry of the system, and
are all obtained from the deformation of the representations of SL(2, C). The eigenvalues
will thus be paired ±αk and will be the same for all three matrices because the symmetries
are respected.
The explicit forms for the representation matrices fall into two classes, with dimensions
M = 2p and M = 2p + 1, the analogues of half-integer and integer spins.
For M = 2p, one finds
b`
(φ1 )k` = δk+1,` ak + δk−1,` , (4.52)
a`
b`
(φ2 )k` = iq k−p δk+1,` ak − iq p−k+1δk−1,` , (4.53)
a`
(φ3 )k` = iαk δk` (4.54)
and we have bp+j = bp−j for j = 1, 2, . . . , p − 1, and αp+n = −αp−n+1 for n =
1, 2, . . . , p.
The ak ’s may all be set to, say, unity, by SL(M) transformations. The bj ’s are determined
recursively by the formula 7
q
− 1+q σ2(p−j ) [q] + bj −1 (1 + q 2(p−j )+3)
bj = , b0 = 0, (4.55)
q 2 (1 + q 2(p−j )−1)
for j = 1, 2, . . . , p − 1. The recursion relation is solved by
q(q 4p − q 2k )(q 2k − 1)
bk,p = (4.56)
(q 2 − 1)2 (q 2p + q 2k−1)(q 2p + q 2k+1 )
and notice that all singularities (poles and zeroes) happen for q a root of unity.
7 We have defined σ [q] = 1 + q + q 2 + · · · + q x .

x
All three matrices have the eigenvalues

1
±αn = ± σ2(p−n) [q]. (4.57)
q p−n (1 + q)
for n = 1, 2, . . . , p.
When M = 2p + 1, we have instead
b`
(φ1 )k` = δk+1,` ak + δk−1,` , (4.58)
a`
b`
(φ2 )k` = iq k−p−1/2δk+1,` ak − iq p−`+1/2δk−1,` , (4.59)
a`
(φ3 )k` = iαk δk` , (4.60)
where bp+n = bp−n+1 and
−qσp−n[q 2 ] + bn−1 (1 + q 2(p−n+2))
bn = , b0 = 0, (4.61)
q 2 (1 + q 2(p−n) )
for n = 1, 2, . . . , p. The recursion relation is solved by
q(q 2k − 1)(q 4p − q 2k−2 )
bk,p = (4.62)
(q 2 − 1)2 (q 2p + q 2k−2)(q 2p + q 2k )
and again we see that all singularities happen for roots of unity. We also have αp+r+1 =
−αp−r+1 for r = 0, 1, . . . , p and the eigenvalues of each matrix are in this case
σp−n [q 2]
0, ±αn = ± (4.63)
q (M−2n)/2
for n = 1, 2, . . . , p.
Note that the solutions that we have written here are not D-flat. However, by standard
theorems, there exists such a solution, which is an SL(M) transformation of the stated
solutions. Still, we must be careful in drawing conclusions based on these solutions. In
particular, there are apparent singularities at special values of q. We will analyze this point
further in Section 4.5.2.
4.5.1. Finding more solutions

So far, we have found representations of the algebra which in the limit q → 1
reduce to finite-dimensional representations of the SL(2, C) algebra. We also noted an
additional one-dimensional representation which becomes singular in this limit, and
therefore corresponds to a vacuum of the theory, which goes to infinity in the limit. This
additional solution is characterized by the property tr φ1 6= 0, whereas for all the other
solutions tr φ1 = 0.
We should ask if there are more irreducible representations of this algebra, that we have
not found above. The answer must be yes, because for q → −1 many of the solutions
which correspond to irreducible representations of SL(2, C) go away to infinity (the
eigenvalues of the matrices are rational functions of q with finite numerator and q + 1
in the denominator. Thus they are infinitely far away in field space, and do not describe
vacua of the theory.)
We can construct additional irreps that do not disappear in the q → −1 limit as follows.
The discrete subgroup of SU(2) has a three-dimensional representation in terms of Pauli
matrices, which suggests the following ansatz for the representations.
The following satisfy the algebra (4.49)
φ1 = φ10 ⊗ (−iσ1 ), (4.64)
φ2 = φ20 ⊗ (−iσ2 ), (4.65)
φ3 = φ30 ⊗ (−iσ3 ), (4.66)
if we have
0 0
φj , φj +1 −q = φj0 +2 . (4.67)
Thus, if we know solutions for a given q, we generate solutions for −q in this way. These
representations are reducible. We will refer to the the irreducible representations obtained
in this way as twisted. There are two cases to consider, ‘half integer’ spin and ‘integer spin’
representations.
The integer spin representations have each eigenvalue repeated twice, including zero and
are split into two irreducible representations with eigenvalues for φ3 in the succession
±iα1 → ∓iα2 → ±iα3 → · · · → 0 → · · · → ∓iα2 → ±iα1 . (4.68)
These satisfy tr(φ3 ) 6= 0, and tr φ1,2 = 0, as these are off-diagonal. The broken Z2
exchanges these two representations. By acting with the Z3 symmetry we get a total of
six new representations for each even-spin irreducible representation of SU(2).
The ‘half-integer’ cases satisfy tr(φ1,2,3) 6= 0. One can clearly see a splitting into two
irreducible representations, but because there is no eigenvalue 0, this splitting into two is
reducible and in total we get four new representations of the algebra. One of these is a Z3
singlet, and the other three form a triplet.
4.5.2. Interpreting the singularities

In this section, we will study properties of representations. In general there are two
classes of representations, irreducible and reducible. In the reducible case, there is no mass
gap (classically) as some part of the gauge group is unbroken (apart from the decoupled
U (1)). The case of irreducible representations are potentially more interesting as they
confine magnetic degrees of freedom. We will exploit S-duality to find dual configurations
that are electrically confining. Note that as we have not been able to prove that all
irreducible representations are accounted for, we cannot be sure that we see all of the vacua.
For the sake of the present argument, we will assume that the classification is complete and
try to extract conclusions about the non-perturbative behavior of the theory.
The representations we have found are all matrices which are rational functions of q.
From the solutions (4.56), (4.62), we see that there are poles at roots of unity, q n = 1.
In the case where we have zeroes and not poles, one observes that as we take the limit
to an appropriate root of unity, the matrix decomposes in block-diagonal form. Thus the
representation becomes reducible in the limit, and we get various copies of the same type
representations (of lower dimension). These singularities are interpreted in the field theory
as having enhanced gauge symmetry, because the commutant of the representation is larger.
If one pictures the vacua of fixed rank as a covering of the q-plane, we have branch points
at some roots of unity.
There are other singularities at roots of unity in the denominators of the fields φ2,3 .
As these are not singularities in the eigenvalues of the matrices, it is not clear that these
are singular solutions. This may correspond to an unfortunate choice of basis for the
representation.
Considering that the roots of unity are special, in the sense that they are related to
orbifolds with discrete torsion which have a very nice semi-classical geometry associated
to them, and also considering that in the limit q → ±1 an infinite family of solutions to the
vacua disappear (in this case there are singularities in the eigenvalues of the matrices), it is
plausible that these are actually bona-fide singularities and the vacua go to infinity. As we
will see, at these values of q, there are moduli spaces of vacua and this is how we interpret
the singularities.
Let us begin with a discussion of q = ±1. First, we know that at q = 1 all of the states
which break the chiral symmetry disappear. Thus we get a jump in the Witten index at this
special value. It is also the case that here for some representations one sees no signal of
the eigenvalues of the matrices being badly behaved, but it is true that we get poles in the
off-diagonal elements.
Let us now discuss q = −1, paying particular attention to discrete chiral symmetry
breaking. For U (M), M even, the q-deformed SU(2) representations move off to infinity
at q = −1, and thus all the irreducible representations come from the ‘half integer’ twisted
case. Thus the Higgs vacua break the Z2 × Z2 subgroup completely, and the vacuum has
an unbroken Z3 subgroup. Each of the four vacua have the Z3 embedded differently.
For U (M), M odd, there are irreducible representations of either integer or half-integer
twisted type. Thus, some of the Higgs vacua break the group to an unbroken Z3 as in
the previous case, and some leave an unbroken Z2 if they are constructed from the ‘integer
spin’ type representations. In addition, the q-deformed representations survive for q → −1
but are reducible (the matrix elements bk,p → 0).
Notice that in the previous arguments we have used only the perturbative symmetries of
the theory. We believe that the SL(2, Z) S-duality of N = 4 SYM is realized and perhaps
enlarged in the present case in some way. We will not address that interesting question
here; instead, we confine ourselves to a few remarks based on SL(2, Z) alone.
Because of the SL(2, Z) symmetry, at the N = 4 point we can make a map of
gauge invariant operators between the different dual theories. Thus we can follow the
deformations of the theory for any S-dual configuration of the N = 4 theory we start with.
Because of the symmetries preserved by the superpotential, changing from one dual
picture to another keeps the general form of the Lagrangian invariant. Thus we have a map
between couplings (g, q) → (g 0 , q 0 ), and m → m0 (g, q). Because at the roots of unity the
theory is special (many vacua collide), the roots of unity must be preserved by the S-duality
action on the space of field theories, thus the most general holomorphic transformation that
keeps q = 1 fixed and the structure of the singularities is of the form q → q ±1 .
Given a vacuum that disappears at a root of unity, let’s say a Higgs vacuum, any of
the vacua related to it by S-duality also disappear. For q = −1 and M even, the trivial
vacuum is S-dual to the q-deformed Higgs vacuum which moves off to infinity as q →
−1, and thus the trivial vacuum is also removed. For M odd, again the trivial vacuum
is S-dual to the q-deformed Higgs vacuum. The latter is reducible, and thus does not
appear to have a mass gap; we conclude that the trivial vacuum is not confining. There
are still the twisted representations, and thus at q = −1, confinement implies (discrete)
chiral symmetry breaking.
If q is a more general nth root of unity, even though we get poles in the bk,p , we have not
been able to find any gauge invariant chiral quantity which becomes singular. This suggests
that the poles are obtained from a coordinate singularity. In any case, there seems to be an
upper bound on the number and dimension of irreducible representations, as each of these
general representations seems to decompose into irreducibles of smaller rank. The bound
is given in terms of n.
Thus as q goes to a root of unity, we can obtain enhanced gauge symmetry. If we do an
S-duality transformation and use some more general combination of electric and magnetic
condensates, there will still be an upper bound on the dimension of irreducibles and thus
no mass gap.
The upper bound on the irreducibles also suggests that one can construct a large center
for the algebra. Indeed, one can take the direct sum of all the irreducible representations
of the algebra we have constructed. If there are no more irreducible representations, this
is a finite-dimensional reducible representation, and the subalgebra of the φi which is the
inverse image of the center of the representation is a large center for the full algebra.
Experience with the example in Section 4.3 suggests that in this case one might actually
get a moduli space of vacua. As we have argued that we get a finite number of discrete
vacua, let us now show that there is a moduli space for roots of unity q n = 1 with n > 2.
We would want the moduli space to be built out of the P , Q matrices in some simple
fashion. Let us choose φ1 to be diagonal. Without the mass deformation, the φi contain
P , Q, P −1 Q−1 . Indeed, one can see that only with the powers P ±1 , Q±1 can one get a
single factor of q in the commutation relations, and there is a potential to get a cancellation
of terms. Thus we take
φ1 = a1 P + a2 P −1 . (4.69)
Because of the symmetry between P , Q, we also take
φ2 = a3 Q + a4 Q−1 (4.70)
and the q commutation relations are as follows
φ1 φ2 − qφ2 φ1 ∼ P Q−1 + QP −1 (4.71)
so we take
φ3 = a5 P Q−1 + a6 QP −1 . (4.72)
The parameters are related by
a1 a4 (1 − q 2 ) = ma5 , (4.73)
−1
q a5 a2 (1 − q ) = ma4
2
(4.74)
and thus it follows that
qm2
a5 a6 = a3 a4 = a1 a2 = (4.75)
(1 − q 2 )2
so apart from factors depending on m, q, a2i+1 a2i ∼ 1. This cuts the number of variables
from six down to three, and (4.73) provides one more constraint. Thus we are left with
a two parameter solution of the F -term constraints. More surprisingly, these also solve
the D-term constraints. These representations are inequivalent as one can show that the
gauge invariant vacuum expectation value tr(φ1n ) is not independent of the ai . For q = ±1,
Eq. (4.75) shows that the ai are singular, and tr(φ12 ) is singular for q = −1, thus this branch
of moduli space does not appear at these roots of unity, and one only has isolated vacua.
2
One can also explicitly show that for q 3 = 1, the element xi = φi3 + mq φi is central.
Thus here one gets a large center, as we have four Casimir operators and one relation. The
fourth Casimir is of the form
w = Aφ1 φ2 φ3 + α1 φ12 + α2 φ22 + α3 φ32 (4.76)
and it is invariant under the full discrete group of symmetries of the potential. A Casimir
of this form exists for all q, and when m = 0 it is the familiar φ1 φ2 φ3 ; it also reduces to
the quadratic Casimir of the SU(2) algebra when q → 1.
The commutative space associated to the algebra is again a deformation of the C3 /Z3 ×
Z3 orbifold, and it is three complex dimensional. We have only found a two parameter
solution of the equations; we believe that this is because we chose a very special form for
the solutions, and not necessarily because the ring fails to be semi-classical.
4.6. The general superpotential
In this section we will try to make progress towards understanding the general
deformation, Eqs. (2.2), (2.3). Solving for the center of the general algebra and also finding
the most general finite-dimensional irreducible representations of the algebra can be quite
difficult. There are some cases which are worth singling out among these, because at
least we can find some partial solutions to the moduli space problem. We have also seen
that semi-classical rings are better behaved than others, as they lead to nice commutative
geometries. Finding all possible semi-classical geometries from our sets of constraints
is very important as they might correspond to the behavior of D-branes at new dual
singularities (not necessarily orbifolds with discrete torsion) which can be connected to
AdS5 × S 5 . Of particular importance are configurations with conformal invariance, as they
might provide new non-spherical horizons [48,49]. Our analysis is quite incomplete due
to the difficulties of the algebraic program involved, but some general comments will be
made here.
As the deformations are taken to zero, the algebra looks like a Poisson algebra if
we interpret commutators as Poisson brackets. Because we have three variables, and
Poisson manifolds are foliated by symplectic manifolds (which are of even dimension),
the symplectic form in the full algebra is degenerate and therefore there is at least one
constant of motion. This suggests that there is at least one element of the center which can
be easily computed. For the q-deformations, the element of the center w = φ1 φ2 φ3 exists
for arbitrary values of q, which suggests that the element of the center is a polynomial
of degree less than or equal to three, depending on the chosen perturbation. For marginal
deformations, it is indeed of degree three, as will be shown later; for the deformation by
three mass terms it is quadratic (the Casimir of the SU(2) algebra).
Because we have commutators we can think of the algebra as deformation-quantization
of the Poisson structure. This suggests that we can standard order operators and establish a
correspondence between the full algebra, and the algebra of three commuting variables.
In standard constructions, this is given by formal power series expansions in a small
parameter h̄. As we have argued before, we want to avoid infinite power series, and rather
give an explicit solution which shows that the constraints can be standard ordered in some
open set. In order to do this, we separate at each order the polynomials in φ1 , φ2 , φ3 which
can be considered as standard ordered.
A choice of standard ordering is important. If we want to find elements of the center, we
need to check that their commutators are zero for all the generators of the algebra. Without
standard ordering a given expression, it is very hard to decide if it is zero or not in the
algebra.
By using the constraints, an arbitrary polynomial operator O can be re-ordered into
standard ordered form up to small corrections. We write this as
O = Oso + h̄O0 . (4.77)
Oso is a linear combination of standard ordered monomials, and it is polynomial in h̄.
Similarly, we can expand O0 ∼ ai M i , where the ai are polynomial in h̄ and the M i are
a collection of non-standard ordered monomials. Because of the form of the algebras, the
degree of O0 as a polynomial in the variables of the algebra is smaller than or equal to the
degree of O. Taking all the possible non-standard ordered monomials of degree less than
or equal to some fixed number g, we obtain a matrix equation
M i = Mso
i
+ h̄aji M j . (4.78)
Now h̄ is a small parameter, so the matrix
j j j
Ai = δi − h̄ai (4.79)
is finite-dimensional and invertible. Hence, any non-standard ordered operators may be
written as linear combinations of the standard ordered operators, where the coefficients are
rational functions in the deformation parameters, the denominators coming from A−1 .
Since the parameters are complex, more generally we need only worry about the
possibility of poles in this construction. At such poles, one of two things can happen. Either
the basis for standard ordered polynomials is badly chosen (e.g., the elements become
linearly dependent), or there is a true obstruction to standard ordering independent of the
basis. This second possibility can happen, if we take q = 0 for example.
Thus, in principle we can proceed order-by-order in the degree of polynomials to find

central elements. Every element of the algebra can be written in standard ordered form,
and as the degree of the element is preserved or lowered by the commutation relations, it is
a matter of linear algebra to calculate the elements of a given order which are in the center.
Although the procedure is well-defined, it is not efficient, as we need to calculate the
matrix A at each order to resolve this problem. Thus a general solution of how the center
depends on the parameters is at best difficult to calculate. Also notice that in the examples
we have studied, there is no upper bound in degree for elements of the center.
In some cases, we may find a large center; that is, the center is generated by more than
one element of the algebra. If the center is large enough, then we may obtain a semi-
classical algebra.
Let us consider the case where the algebra A is a finitely generated module over its
center, with generators ei . We can choose one of the generators to be the identity in the
ring, and the others will satisfy a multiplication rule of the type
ei · ej = fij k ek (4.80)
with fij k ∈ ZA. On a given irreducible representation of the algebra, the elements of the
center can be treated as numbers, and thus we can argue that we have a family of algebras
parametrized by the algebraic variety corresponding to the center of the algebra.
Because of the form of Eq. (4.80), we can see that given a vector in the representation
of the algebra, its orbit under the action of the ei is finite-dimensional. Thus there is an
upper bound on the dimensions of the irreducible representations. We can imagine that this
upper bound is realized by the branes living in the bulk, and that any other representation
with smaller dimension is a fractional brane of some sort. The finite-dimensionality of the
irreps suggests that the ring is semi-classical, although we have no proof of this assertion.
The semi-classical rings that we have studied all have this property, and this suggests that
the two conditions might be equivalent.
Let us now consider a few more examples.
4.6.1. General marginal deformations

As an example of the general difficulties that one faces, let us consider a general marginal
deformation of the N = 4 theory. The superpotential is given by

λ
W = tr φ1 φ2 φ3 − qφ2 φ1 φ3 + φ13 + φ23 + φ33 (4.81)
3
and the equations we need to solve for the moduli space are (cyclic)
[φj , φj +1 ]q = −λφj2+2 . (4.82)
This algebra is homogeneous, and thus if we were able to find a non-trivial irrep, we could
scale it to zero: this implies that the moduli space is connected.
For λ = 0 the element of the center that is always present is w = φ1 φ2 φ3 , and this
suggests that the element of the center is cubic in general. Indeed, a direct calculation
shows that
(1 − q)φ1 φ2 φ3 − λφ13 + qλφ23 − λφ33 (4.83)
is central.
Let us first consider one-dimensional irreps. These will satisfy (cyclic)
(q − 1)φj φj +1 = λφj2+2 . (4.84)
A non-trivial solution will have the φj all non-zero complex numbers. We can easily see
that this is only solvable provided that
(q − 1)3 = λ3 (4.85)
so (q − 1)/λ is a cube root of unity. Given λ and q satisfying these constraints, one can find
solutions to the equations where φ1 and φ2 are equal up to cube roots of unity, and then φ3
is determined from the other two. Thus we get three complex lines meeting at the origin,
reminiscent of the moduli space for q-deformations. Indeed when λ and q are related in
this way, there is a linear change of basis of the fields which returns the superpotential
to a q-deformation. Therefore we have new semi-classical rings, but they are related by a
change of basis to the ones we already know.
Another thing that we can do is exploit the Z3 symmetry which permutes φ1 , φ2 , φ3 . Set
φ1 = aP , φ2 = bQ, φ3 = cP −1 Q−1 ; for the P , Q matrices associated to the cube roots of
unity, we also have φ32 ∼ φ1 φ2 , so three-dimensional irreps of the algebra may exist.
In this case we want to find solutions to (cyclic on a, b, c)
ab(qω − 1) = ωλc2 (4.86)
with ω a cube root of unity. One can see that this gives us the constraint
(qω − 1)3 = λ3 . (4.87)
For λ → 0 we associate the geometry to C3 /(Z3 × Z3 ) which happens to be one of the
orbifolds one can realize globally on a three-dimensional complex torus.
If for λ 6= 0 we can find a large center, we might be able to compute the full geometry
of moduli space, and treat this solution as a fractional brane. Notice that if this is the case,
it does not correspond to a C3 /Zn × Zn singularity, as the fractional branes in that case
behave differently. Further exploration of this model will be left for future work [38].
5. D-branes in near-horizon geometries
So far, we have mainly discussed moduli spaces of vacua and how to include extended
objects in the discussion. The analysis has been done directly in the field theory. Now we
will try to understand the background and the moduli space that the D-branes realize from
the AdS/CFT perspective.
The field theory is understood as being dual to the near-horizon geometry of a brane
configuration. Because the moduli space is of the form of a symmetric product (built out of
smaller components), one can think of adding these small component D-branes as probes
in the near horizon geometry and testing how the field theory moduli space is realized on
these probes.
We will carefully compare the field theory marginal and relevant deformations to the
corresponding deformations of AdS5 × S 5 geometry. In doing so, we uncover and solve
several puzzles. In particular, there are new branches of moduli space in the field theory
which open up for arbitrarily small values of q − 1, as discussed in Section 4.1. Uncovering
this structure in the string theory will have several bonuses. This non-generic branch is
realized by wrapping a 5-brane on a 2-torus and using this information, we will argue that
the mirror symmetry [7] between deformed 5-spheres and orbifolds can be understood as a
standard T-duality operation. The two supergravity descriptions are valid in different areas
of parameter space. It also becomes clear in this analysis that there is no sense in which
the field theories are dual to supergravity on a space; rather, string theory is absolutely
necessary for a consistent duality.
We have seen that the moduli space of vacua has very non-trivial behavior in response
to the deformations. In particular, the somewhat artificial separation between the center of
the algebra and other elements of the algebra is very subtle in field theory. This will be
addressed later and we will find a satisfactory solution. If we look at the same construction
from the AdS5 × S 5 , each of these perturbations is in the bulk of the S 5 geometry, and there
is no reason to single out any special elements of the algebra.
5.1. Effects of the background on D-branes
It is important to notice that as seen from the AdS5 × S 5 perspective when one deforms
the theory, the D-brane moduli space changes drastically. To first approximation, this is
because the added potential localizes the D-branes to the ‘fixed planes’ of the deformation.
But even for very small deformations q ∼ 1, we can find rational solutions of q n = 1 for
large n, and thus the moduli space has non-generic behavior for a large enough number of
branes; indeed, we need n such branes to find extra components of moduli space. These
new branches can be seen from (4.15), and predict that the D-branes are going to be
uniformly distributed on a circle. We take the eigenvalues of matrices to determine the
coordinates of the D-branes, as in matrix theory [34].
If the branes are point-like then the open string states stretching between them would be
massive and one would not find the new branch in moduli space. However, this is clearly
inconsistent with our field theory results, and thus we are motivated to find a satisfactory
solution within string theory. Note that these extra components of moduli space do not just
appear in the vicinity of the origin; rather, they extend to infinity with the rest of D-brane
moduli space.
The resolution of these issues bears close resemblance to recent results of Myers [17]
concerning the dielectric properties of branes in background fields. Since the deformations
of the field theory superpotential correspond to non-zero vevs of fields on the 5-sphere,
we do indeed expect these phenomena to occur. Roughly speaking, the D3-branes should
be thought of as D5-branes on R4 × S 2 , where, as we show below, the S 2 is contained
in S 5 . In order to find new branches of the moduli space, we want to argue that there
are configurations which support massless open string modes, and topologically this will
happen when different spheres intersect each other. Thus their centers can be separated,
and we can still have massless string states stretching between them.
Now let us begin by analyzing in some detail the map between superpotential
deformations and vevs. This material is of course not new, but is included here for
completeness.
The q − 1 and m deformations correspond to background values for magnetic potentials
RR and H NS . The mass deformation is not marginal, and will therefore depend on the
F(3) (3)
radial direction of AdS5 . The field τ = C + ie−φ gives the gauge coupling, and will be
kept constant. The field G(3) = F(3) − hτ iH(3) is related directly to the superpotential
deformations. The harmonic in the 10 of SU(4) is a tachyon state in the AdS, thus
this perturbation blows-up in the infrared. The marginal cubic operators correspond to a
harmonic of G(3) in the 45 of SU(4). In this case, there will be no dependence on the radial
direction of AdS as the associated scalar is massless in five dimensions; this fact guarantees
that we preserve the conformal group to leading order.
Let us now specialize to the marginal deformations. As explained in Ref. [17], D3-
branes in the presence of RR background fields pick up a dipole moment for higher brane
charge, and become extended in two additional dimensions. The simplest topological
shape, and the one with the lowest energy, is a 2-sphere centered at the position of the
D3-brane. Since we are considering a weakly coupled string theory regime, we should
take these to be D5-branes [16]. More precisely, the F(3) background is dual to F(7)
which couples to a D5-brane. F(7) has support on R4 × D3 , where D3 is the 3-disk with
S 2 boundary. The D3-branes are extended in the R4 , which in near-horizon geometry is
contained in AdS5 . We thus write F(7) = F e(3) ∧ dV ol4 , and integrating, we can normalize
it such that
Z Z
F(7) = F e(3) . (5.1)
R4 ×D 3 D3
The 3-disk extends along the radial direction of AdS5 plus two directions along the the S 5 .
As a result, we can write
e(3) = dρ ∧ C
F e(2) . (5.2)
As such, if the effect were solely due to the dielectric effect it is hard to understand how
the D-branes can have massless states at different angles along the S 5 , as the stretching
happens mostly in the radial direction.R The D3-brane charge of this 5-brane is obtained
S 2 F = n.
1
from a flux through the 2-sphere, 2π
As follows from Ref. [5], there will also be a background H(3)NS turned on in the presence
of the superpotential deformations. If we expect some energy contribution from the integral
NS over the disk, then the 2-sphere prefers to be stretched along the 5-sphere, because
of H(3)
NS
H(3) does not have any component along the AdS directions. In general, then, the radius of
the disk D3 is oriented partially in the radial direction of AdS5 and partially in S 5 , as there
are two competing effects deforming the branes.
We want to look for configurations where D3-branes are intersecting in the sense
of intersections of their S 2 ’s. This is where we can expect massless string states, at
least topologically. The H NS deformation is the one that gives us the deformation of
the D-branes in the appropriate direction. We will assume that these configurations are
supersymmetric and that probes do not affect the background.
For rational q, the moduli space has a scaling direction, which follows from the
conformal invariance: in the language of Section 4.1, we have
a, b, c → ta, tb, tc. (5.3)
This is reflected in the near-horizon geometry by the fact that if we move D3-branes
along the radial direction of AdS5 , they simply rescale-in particular, if we have intersecting
branes, they remain intersecting as we perform this motion.
For relevant deformations, such as a mass term, the RR and NS backgrounds grow as
we move in along the AdS5 , and thus we expect the 2-spheres to grow in size along the
5-sphere as we go to the infrared. As in this case the H(3) fields will also have a radial
component, then both types of fields H(3) and F e(3) help the 2-sphere to grow along the
S 5 and the radial direction. Eventually, the 2-spheres will be of comparable size to the
5-sphere, and at this point, the notion of point-like D-branes loses any meaning. To avoid
these issues and for ease of calculation, we will treat only marginal deformations in the
following sections.
5.2. Size and configurations of 5-branes
First, let us find the expected size r of an S 2 associated with a D3-brane. We will assume
that this S 2 is small compared to R5 , the radius of S 5 , but that it is large enough that
we can neglect its self-interaction (from opposite sides of the S 2 ). We will show that for
small deformations, the size grows linearly with the potential. To this effect, we do a probe
calculation. We have a geometry which is almost AdS5 × S 5 generated by some D-branes
which are at the origin, and we have a small extra D-brane which turns into a sphere
on which we are going to do our analysis. Because conformal invariance is preserved by
the marginal deformations, there can be no dependence on the AdS radial direction in the
physical quantities of interest, apart from setting the scale of the physics. We can therefore
work in a local frame and ignore redshift factors, etc.
The DBI action for a D5-brane determines the energy
Z p
µ5
EDBI = d 2 det(G − B + 2πα0 F )
g
Z Z
− µ5 F e(3) − µ5 (2πα0 F − B) ∧ C4 . (5.4)
D3
The metric G scales as r 2 , whereas F behaves as r 0 (by the flux quantization [50]). By
expansion of the DBI part for small r, we find an energy of the form
α
E = E(D3) + r 4 − βr 3 + o(r 5 ). (5.5)
g
We write
Z Z
NS
H(3) = cNS r 3 , e(3) = cR r 3 .
F (5.6)
D3 D3
The field strengths are constant over the disk. We will do the analysis ignoring the five-
form field strength. At the end, we will compensate for this omission. The general features
of the result should not depend on how far we are from the origin. This is how we can
justify this omission.
From the energy (5.4), we see that the D3-brane charge is given by the coupling to C4
Z
1
Q3 = n − 2 0 B (5.7)
4π α
S2
cNS 3
=n− r . (5.8)
4π 2 α0
The expansion of the energy in powers of r now reads

µ3 cR 3 1
E= Q3 − r + r + ··· ,
4
(5.9)
g 4π 2 α0 (4π 2 α0 )2 2n
where Q3 is a constant plus small corrections in r 3 . The result is minimized at
3 2
hri ' (cNS + gcR ) 4π 2 α0 n. (5.10)
2
The energy at this radius satisfies

µ3 1
E= Q3 1 + hri3 (3cNS − gcR ) + · · · . (5.11)
g 4Q3
This result is puzzling, since it suggests that n D3-branes extend to a single S 2 of radius
proportional to n, as opposed to a sphere wrapped n times around the solution for a single
brane. This result is wrong from several points of view. First, this solution cannot give
enhanced U (n) gauge symmetry, as there are no massless states apparent, and suggests a
totally different picture of the moduli space, very different for each value of n. We must be
more careful in interpreting Eq. (5.11).
We interpret the branes as a black hole in the supergravity which is almost pointlike as
far as the S 5 is concerned. One minimizes the energy (5.9) and then compares the ratios
of energy to D3-brane charge of two configurations to determine which may be BPS. In
fact, n D5-branes wrapping an S 2 of radius r have lower E/Q3 than a single D5-brane
wrapping an S 2 of radius nr. This indicates that the former configuration has the better
chance of being BPS.
The stabilization mechanism which impedes the spheres from shrinking further is that
the flux of F is quantized. This mechanism has been found when studying D-branes from
the boundary state formalism for group manifolds [50], but it is clear that it should be a
general phenomenon for D-branes in non-trivial H(3) NS backgrounds.
Here we see also that the RR charge for the large sphere is not quantized in general as it
gets an anomalous defect proportional to H 4 n3 . These can be meta-stable boundary states,
and in group manifolds these can be calculated exactly [50], where a similar defect in the
brane charge quantization condition occurs. When HNS = 0 the D-brane charge is related
to K-theory, and then we expect a quantization condition. This puzzle was recently solved
in Ref. [51] where there is a back reaction from the bulk which contributes to the 3-brane
charge. Thus, Eq. (5.11) is incomplete as it does not take into account the energy associated
to this back reaction. However, the ratio E/Q3 on the horizon of the brane probe is exact,
being the local tension divided by charge. Here the BPS D-branes behave better, as we get
an anomalous charge which is proportional to the D-brane number.
5.3. Large n branches
We now want to find the new branches of moduli space for q n = 1, by finding the
geometric configuration into which it can be deformed. Because the marginal deformation
preserves the conformal group, in the near-horizon geometry, the H(3) NS lies entirely along
the 5-sphere, and thus the D-brane becomes spherical along 5-sphere directions.
For the q-deformation, this means that the D-branes grow in size linearly with respect
to q − 1, which is the small parameter. This is the important point of the calculation in
the previous section. Consider a configuration of n of these 2-spheres, distributed around
a circle in S 5 so that they touch each other. The value of n is proportional to (q − 1)−1 , in
accordance with q n = 1.
In order for the D-branes to touch, we need to know the shape of the 2-spheres well. For
|q| = 1, one finds massless states between 2-spheres which are at the same distance from
the origin in AdS space. To see this, we can calculate the masses of the off-diagonal states
from the superpotential
tr φ1 φ2 φ3 − q tr φ2 φ1 φ3 (5.12)
with

a 0
φ1 = . (5.13)
0 b
These masses are proportional to a − qb and b − qa, and thus in order to have massless
states for |b| = |a| we need |q| = 1.
In order to get the 2-spheres to touch when they are at different radii, the dielectric
effect on the D-brane has to be included, as it is responsible for extending the D-brane
in the radial direction. We now want to argue that the D5-branes laying flat on the S 5
actually do touch at another point. The reason why this is important is that moving apart a
pair of 2-spheres on S 5 might make it impossible for them to touch again. Because of the
geometric setup, if we consider two D3-branes at the same location and we move one with
respect to the other in moduli space (of one real dimension on the S 5 ), we will get two
NS as we
2-spheres. Because the solution of the linearized supergravity equations rotates H(3)
move along this one parameter, the spheres become linked on the S 5 . This is explicitly
shown in Fig. 3.
As the spheres are unlinked when they are very far from each other, they necessarily pass
through a point where they touch. This is, topologically, the place where the extra states
become massless. With the dielectric effect turned on the spheres are tilted with respect to
Fig. 3. Knotted spheres on S 5 .
the S 5 and that is why they touch at different values of their radial position. The tuning
required to make the D-branes lie flat on the S 5 is precisely the action of removing the
dielectric effect on the D-branes, and corresponds to one real condition on a one complex
parameter deformation of the theory.
Thus, we arrive at a configuration of spherical D5-branes which touch at points. Now
there are configurations, for rational q, with n spheres where each touches the next one
and they stack on a circle. This is the configuration where the new branch of moduli space
opens up, as in Eq. (4.15). This structure should be thought of as a 2-torus with n pinches.
Indeed the massless states at the intersection of the D-branes are such that they resolve the
pinching points into tubes, as in Fig. 4.
This resolution of these configurations is equivalent to moving onto the new branches of
moduli space.
A pinching torus with n nodes is also exactly the degeneration which produces fractional
branes in an ALE singularity or on an elliptically fibered Calabi–Yau in F-theory. Thus this
configuration of branes seems to be the right one to deform into the extra branches of
moduli space for the rational values of q.
Notice also that this semi-classical torus is reflected also in Eq. (4.12), where we see
a realization of a non-commutative torus algebra via clock and shift operators. Thus the
non-commutative geometry description of the moduli space knows that the D-brane in
AdS5 × S 5 is shaped like a torus, and that when we deform to the degeneration, we split
the torus into n spheres (fractional branes), as required by the ALE singularity type of the
orbifold in moduli space.
The picture presented above is meant as a topological argument for the branches of
moduli space in string theory. These arguments rely upon a few technical assumptions,
which we think are reasonable. We have assumed that the different D-branes do not affect
each other and that they intersect at supersymmetric angles. Although it would be nice to
Fig. 4. Resolving the pinched Riemann surface.

assert this, as it would make our whole construction a purely topological argument, there
is no natural complex structure for the spheres which would guarantee this property, and
we have to rely on a dynamical mechanism instead. For completeness we should study the
possibility that the 2-spheres might interact strongly with each other near the intersection
point. In that case, the 2-spheres would develop a throat between them; so, topologically,
we have a sphere, with a line bundle of degree two to count the number of D3-branes.
When we move in moduli space we deform the line bundle and the metric. For a non-
generic bundle, one can get extra states which are massless, and these would be the extra
massless modes one needs. Of course, because the field theory tells us that the massless
states are there, we believe that these constructions are sensible.
A second point which needs to be made is that although we made an argument based on
D5-branes, by the SL(2, Z) duality we can make an argument with any (p, q)-5 brane.
Thus the fact that the RR and NS mix in the near-horizon geometry is necessary to
implement the SL(2, Z) duality on the field theory space of deformations as we change the
string coupling g and make different (p, q)-strings light. The reason we get a description
purely in terms of D-branes is that we are using weakly coupled string theory, and for
any other brane with NS5-brane charge the fundamental strings cannot end on it. This
ambiguity in the description has also been found in Ref. [16]. In their case, only one
configuration would be such that the supergravity degrees of freedom were weakly coupled
through most of the geometry.
5.4. Mirror symmetry
We have seen that the construction of moduli space suggests a two torus fibration of the
five sphere. This torus can be made explicit by using the following invariant coordinates
r12 = |φ1 |2 , r22 = |φ2 |2 , r32 = |φ3 |2 , w = φ1 φ2 φ3 (5.14)
P
indeed, ρ = ri is the radial direction in AdS5 and w is equal to r1 r2 r3 except for a
2 2
phase. We get a total of three real coordinates on the S 5 , and we are left with two phases
to determine, which are the arguments of φ1 /φ2 and φ1 /φ3 . These two phases determine
the two-torus fibration on the S 5 , and the fibration is independent of how many branes are
stacked together to get the new branches of moduli space.
Note that the T 2 so obtained may have n nodes (related to the D3-brane charge) but
the T 2 may wrap m times around the S 5 before closing. The latter clearly corresponds
to 5-brane charge. In Ref. [7] a mirror symmetry was noted between string theory on a
deformed 5-sphere and an orbifold theory. We are now in position to demonstrate that this
mirror symmetry may be obtained by T-duality 8 on the near-horizon geometry, where the
T-duality is taken fiberwise on the T 2 fibration.
The T-duality acts on the 2-torus that we have described above. Explicitly, the charges
(m, n) transform as a doublet under the SL(2, Z) T-duality. Choose a mapping that takes
(m, n), with m, n relatively prime, to (0, 1); this mapping will single out a point-like D3-
brane on the mirror. This is achieved by the matrix
8 This is expected from the work of Strominger et al. [52].


a m
M= , (5.15)
b n
where a, b are fixed numbers, modulo m, n, respectively. The torus with complexified
Kähler form K = B + iA is taken to a torus with a different value of K. Explicitly, we
have
aK + m
K → K0 = . (5.16)
bK + n
The area of the torus goes to zero and BNS is smooth at the singularities (where only
one phase remains). We can examine the effect of the transformation on K near this limit.
Indeed, we get that in the dual torus
m
K0 → , (5.17)
n
which signals a constant B-field of strength m/n. This value is quantized and its fractional
part corresponds to the discrete torsion phase.
As the area of the two torus is not constant, if we start without HNS then upon the T-
duality, we will get a varying BNS flux through the dual torus, and thus we have generated
an HNS in the T-dual geometry. If we want to cancel this quantity, there is a choice of BNS
which makes Re(K 0 ) = m/n constant over the dual fibration. This determines explicitly
the HNS field needed to perform the marginal deformation on the field theory, from q = 1
to a given value of q.
Notice that at the singularities we have the allowed degeneration into fractional branes
from the splitting of (m, n) → n(m/n, 1). Thus the T-dual fibration has singularities of the
An−1 type. As the fractional branes can be connected to each other in moduli space, we get
a circle of such singularities and the monodromies around that circle are exactly the ones
associated to orbifolds with discrete torsion.
Thus we have both the fractional B-field on the T-dual torus, and the monodromies of
the singularities so that we can identify the T-dual geometry as the orbifold with discrete
torsion.
As we have a T-duality description of the relation between the two compactifications,
if the Kähler form is generically large in one setup, it is small in the other. It is
therefore necessary to understand which description can be accounted for by supergravity
calculations at a given point.
This question can be answered in AdS5 × S 5 . If we want n D-branes to become one of
these 2-tori, then q n ∼ 1, and as we saw before n ∼ 1/H . The calculation of the D-brane
action was done in string units, thus the D-branes are generically of a size commensurate
with the string scale.
When we go to the supergravity regime, the string length is related to the supergravity
background by the relation
1
ls ∼ √
4
lp . (5.18)
gN
√
In the configurations that we have discussed, we have nr ∼ R5 ∼ 4 gN in string units.
Now, in order for α0 -corrections to be small, we must have r ∼ H . `s , which implies
p
n& 4
gN . (5.19)
Thus if we want n relatively small, AdS5 × S5is a poor description of the geometry unless
√
the total number of branes N is such that gN n.
4
Similarly, for the S 5 /Γ to be large, we need a very large number N of D branes. Indeed,
the size of S 5 /Γ is of order
r ∼ lp /n. (5.20)
For the supergravity approximation to be valid here, we should require that twisted sector
states are massive; this is the condition ls lp /n. As a result, the crossover region is at
√
the same place, n ∼ 4 gN . Thus, in the orbifold frame, we need to have N large enough
so that lp > nls , whereas for the deformed 5-sphere, we needed N small enough so that
nls > lp .
For a general q-deformation which is not a root of unity, no supergravity description
will be good, and one is forced to take into account all of the stringy corrections to the
supergravity equations of motion in order to determine the background.
It is also clear that the strength of the perturbation in string units needed to change the
value of q at the boundary is related to the number of branes in the configuration. Thus
the limit is not uniform in supergravity. In this sense, it is hard to separate vevs from
expectation values, as the supergravity boundary conditions are changed drastically when
we change the number of branes.
6. Closed strings and K-theory
So far we have described features of the moduli space of vacua for point-like (in the
sense of non-commutative geometry) D-branes in deformed geometries. We want now to
present a more complete picture of the field theory. This will have two aspects. First, we
discuss the chiral ring of the field theory which has a clear interpretation in terms of the
supergravity background and we give an interpretation in terms of the algebra itself. The
second point that we wish to address is some of the features of extended branes which
are accessible by topological considerations. In particular, this involves a somewhat more
detailed understanding of K-theory and of discrete anomalies.
6.1. Closed strings for near-horizon geometry
Next, we will use ideas from the geometry/field theory correspondence to describe the
physics of closed strings from the field theory point of view. This closed string theory
is to be thought of as the dual string theory to the field theory of some D-branes near
a singularity. Our aim is to understand the open string — closed string duality a little
better, and how one might expect to realize it in the field theory. We have dealt with four-
dimensional field theories so far in the classical regime. Our purpose is now to extract a
closed string theory out of the quantum dynamics of the field theory.
The near-horizon geometry will have certain boundary conditions which control the
superpotential, and some additional set of boundary conditions which specify the vacuum.
That is, there are two contributions to the boundary conditions: those that decay sufficiently
fast are related to the moduli of the branes, and those that decay more slowly are related
to changes in the superpotential. To fully specify the field theory, we need in addition
the correlation functions of operators. First, though, we need an identification of those
operators.
We take the closed string states to be single trace operators in the field theory. This is in
accordance with the AdS/CFT correspondence [2–4] in that closed string states are gauge
invariant operators in the field theory. The idea is to restrict ourselves now to the chiral ring
of the field theory for simplicity, and because in all of our analysis we have kept only the
parts which are protected by supersymmetry.
Let us assume first that we have a conformal field theory, and that its associated algebra
is semi-classical (e.g., orbifolds with discrete torsion). We will exploit the following idea:
the vevs of the closed string states (corresponding to states that decay quickly enough at the
AdS boundary) are generated by the stack of D-branes being at different locations in the
moduli space [48]. With the asymptotic values one reconstructs the near-horizon geometry
of a set of parallel D-branes by summing over holes [53] with given boundary conditions.
Thus we can identify different tadpoles of the string states by motion in the moduli space of
vacua. The right question to ask is what region of moduli space gives a vev to an operator.
We will combine this knowledge with the identification of the chiral ring for some
geometries. Let us review a few results from Ref. [7]. In that paper it was noticed that
for orbifolds with discrete torsion, one could see the twisted and untwisted string states
in the near-horizon geometry as coming from traces of different chiral operators. We will
review the case of the orbifold C3 /Zn × Zn with maximal discrete torsion.
Chiral operators come in two types

O(k1 , k2 , k3 ) = tr φ1k1 φ2k2 φ3k3 (6.1)
with k1 = k2 = k3 mod(n), which are untwisted states, and
Oj (k) = tr φjk (6.2)
which are twisted states so long as k 6= 0 mod (n).
The constraint on the kj for untwisted states is familiar from Eq. (4.8). That is, the
center of the algebra is associated with the untwisted states. This shows why the center of
the algebra is so important to understand the geometry. Namely, the algebraic geometry
of the center of the algebra is the geometry that the closed string sector sees. Here again
we see that the geometry of the closed strings is commutative, as in Ref. [21]. The non-
commutativity of the moduli space appears from the closed string theory point of view
because we have twisted sectors.
Notice that in (4.13) it is clear that it is the fractional branes which give vevs to the
twisted sector strings. This is just as it should be, as we always think of coupling twisted
sector strings to fractional branes living at the singularities of the classical space. Although
we have discussed chiral operators here, it is more generally possible to distinguish twisted
and untwisted states. As well, the same statement may be made if we do not have a
conformal field theory.
In the case of AdS5 × S 5 , the F - and D-term constraints give us a commutative geometry.
Thus the center is the whole algebra, and every closed string state is untwisted and lives in
the bulk.
We saw in Section 4 that the behavior under mass deformations was special for q =
±1. From our analysis, we can now see why this is the case. Namely, for q = ±1, the
mass perturbation is untwisted, and therefore affects the bulk of moduli space. For any
other rational q, the mass perturbation is twisted, and we expect that it will only affect the
vicinity of a singularity.
6.2. Chiral ring revisited and quantum groups
Let us analyze the chiral ring in more detail. We have already learned that twisted and
untwisted states are associated with traces of central (non-central) elements of the algebra,
respectively.
States in the chiral ring are made by taking traces of holomorphic elements of the
algebra. There are two steps for this construction. First we need a description of the
elements of the algebra, and then we need to interpret the properties of the trace.
Any operator (for the deformations we have studied) can always be written in monomial
ordered form for a small enough deformation, as we shown in Section 4.6. The difference
between two possible orderings is given by F -terms and therefore they correspond to
derivatives of other fields. In a conformal theory, these would be descendants and not
primaries. For the topological chiral ring, we set all F -terms to zero, so the operators
are identified as traces of elements of the algebra.
Let us consider the case where we have a conformal field theory in the ultraviolet.
Because we have an algebra described by quadratic relations, we have a quantum
hyperplane geometry [18]. The operators with the same conformal dimension are
homogeneous. On every quadratic algebra of the type described, there is an associated
quantum group acting on the algebra. The states of the same degree are associated to the
representations of this quantum group. This suggests that there might be a relation between
operators in the closed string theory and representations of the quantum group. If this is
indeed the case, then the fusion rules of the closed string operators will be related to the
fusion rules of the representations of the quantum group algebra. This relation would give
testable predictions for 3-point functions in the deformed AdS5 ×S 5 supergravity. Quantum
groups have also made an appearance in near-horizon geometry in the work of Ref. [54] in
connection with the stringy exclusion principle [55].
We do have to remember that we associate an operator to an element of the algebra, and
that it is not the element of the algebra itself which is the gauge invariant operator. The
association is by taking
O(a) = tr(a). (6.3)
Because of the cyclic property of the trace, we need the following rule
O(ab) = tr(ab) = tr(ba) = O(ba) (6.4)

thus the map from the algebra to the operators factors through
A → A/[A, A] (6.5)
as a vector space. It is the class [a] in A/[A, A] that matters, and not a itself.
The space
A/[A, A] = H H0(A) (6.6)
is actually a homology group of the Hochschild complex [56] and suggests that the chiral
ring is in general a cohomology group of the non-commutative space (so long as we have
some sort of Poincaré duality). Because of our knowledge of Calabi–Yau manifolds, we can
think of the chiral ring as a ring of deformations of a non-commutative complex structure,
because we have found a relation with homology. Indeed, for a non-compact orbifold space
the ring of deformations of the complex structure is infinite-dimensional because of the
non-compactness, and it is associated to a cohomology group of the manifold H 2,1(M).
This suggests that orbifolds with discrete torsion may be better understood as a non-
commutative Calabi–Yau space.
6.3. K-theory
Let us now make a few remarks about K-theory. To this effect we will review some of
the results of Section 4.
Let us analyze the results of the q-deformation for rational q. There we found two
types of finite-dimensional representations: the representation of a non-commutative point
associated to the bulk and some other representations which correspond to fractional branes
at a singularity.
The set of non-singular points are all connected, and thus each point defines the same
K-theory class. On going to the singularity, the points would split as
M
lim Rreg = i
Rsing , (6.7)
i
where the subscript indicates that the point belongs to the regular part of the variety, or the
singular part.
It so happens that the R i are homotopic to each other. That is, they can be deformed
continuously into each other. If q n = 1, then there are n representations on the right-hand
side of (6.7). In K-theory we thus have
K(Rreg ) = nK(Rsing ) (6.8)
and the K-theory of points is generated by the K-theory class of a single singular point.
p
Thus K0 (A) = Z.
If we add one mass deformation, we find two coordinate patches that cover all of
the variety except for a single complex line. This complex line is the complex line of
singularities that was resolved by the deformation. One can also find solutions that cover
this line of singularities. One still has two complex lines of singularities meeting at the
origin, and the K-theory of points is still Z. In both of these cases the degree of a point is
enough to determine its K-theory class.
For the other rings, we find different phenomena. There are isolated points which
correspond to fractional branes which cannot be connected to other singular points. For
a rank three mass deformation and q = ±1 or q not a root of unity, the moduli space is
completely destroyed and the number of finite-dimensional irreducible representations of
the algebra is infinite. These are examples of rings which are not semi-classical, and in
these cases the K-theory of points consists of an infinite number of copies of Z, one for
each irreducible representation. In the other cases, for q a root of unity, the number of
isolated points is finite.
The reason why the K-theory of points is not preserved under the deformations of the
algebra relies on the fact that this is the K-theory appropriate to algebraic geometry, and
not real geometry. This stems from the fact that we are restricting ourselves to the moduli
space of vacua, and we are forbidding transitions that go between the different components
in moduli space. This is only appropriate if we are studying BPS objects, so this K-theory
would serve to count BPS states, and not brane charges.
The full K-theory that we would need to understand brane-charge properly requires the
inclusion of anti-holomorphic data and is less refined. This new K-theory would be the
algebraic K-theory of the C∗ algebra associated to the string compactification. That is, the
holomorphic K-theory construction gives too many K-theory classes, and does not give
classes for the objects which cannot be represented in the holomorphic setting (e.g., odd-
dimensional D-branes).
The second statement that we want to make in K-theory has to do with extended classes.
Indeed, based on discrete anomalies, the orbifolds with discrete torsion have a different
K-theory than the commutative one [9,23,57,58] associated to the ordinary orbifold. Our
K-theory of points reproduces this result. We can also see the anomaly for extended objects.
Consider the orbifold with discrete torsion C3 /Zn × Zn , and consider trying to wrap
a brane along the singular complex line x = y = 0. This is a holomorphic subspace of
the manifold. For the brane to cover the complex line, we need to have a lift to the non-
commutative geometry, but the non-commutative geometry covers the singular complex
line by an n-fold cover. Thus if we write a brane solution which would cover the singular
complex line only once (which corresponds to a sheaf of rank 1), this solution would
correspond to a fractional brane. The lifting of this solution is obstructed, because if one
lifts a point and does an analytic continuation, the brane would be broken in the non-
commutative space. Indeed, we need a sheaf of rank 1 in the non-commutative sense, and
this is a sheaf of commutative rank n. Thus the brane charge is quantized in units of n
larger than in the standard orbifold, just as expected from the discrete anomaly. We believe
that the non-commutative analysis makes the calculation of the anomaly more transparent.
As a final point, note that in principle, we have defined a K-theory that is capable of
extending to B NS 6= 0. As seen from the AdS/CFT, the deformations corresponding to
the superpotentials are obtained by addition of antisymmetric tensors to the background.
Our results suggest that the K-theory necessary to study these background is the algebraic
K-theory of a non-commutative algebra.
7. Conclusions
In this paper, we have studied relevant and marginal deformations of the N = 4 SYM
theory from a non-commutative algebraic point of view. The moduli space looks like a
symmetric product of a non-commutative geometry. This is interesting because it implies
that D-branes may be considered as independent to a certain extent in the weakly coupled
regime. This symmetric space captures well the phenomena of D-brane fractionation at
singularities. Our approach has led us to the beginnings of a new definition of non-
commutative algebraic geometry, which is still under investigation. The center of the
algebra plays an important role in this construction, and, indeed, in a string theory picture
the commutative subalgebra is related to closed strings, while the full non-commutative
algebra is needed for open strings. When studied from the AdS/CFT point of view, the field
theories that we studied present new dualities between distinct near-horizon geometries.
These dualities are realized by T-duality of a 2-torus fibration of the 5-sphere. Different
choices of T-duality lead to different dual near-horizon geometries. These results imply
that AdS/CFT is inherently a stringy phenomenon, as they exhibit T-dualities which are
not symmetries of classical supergravities. In order to understand this duality, we have
constructed the D-brane configurations which realize the moduli space. We have found that
the point-like D3-branes of the AdS5 × S 5 become non-commutative 5-branes wrapping
the torus fibration.
The non-commutative geometric framework suggests a natural formulation of K-theory
appropriate to holomorphic data, and this is successful in reproducing the physics of
discrete anomalies. This suggests that in general backgrounds the K-theory appropriate
to D-brane charge is that derived from non-commutative algebra.
Our work suggests several avenues for future research. In particular, it would be
of interest to understand the general problem of classifying what we have termed
semi-classical algebras. A thorough understanding of this problem should provide new
backgrounds in which D-branes can propagate, and would shed light on the existence of
other dualities in near-horizon geometries.
In more generality, one should study the full problem of non-commutative algebraic
geometry, including global questions. With a precise notion of gluing and compactness
for example, we could entertain the idea of non-commutative Calabi–Yau and their stringy
geometry.
There are also interesting questions concerning non-perturbative effects, which we
would need to understand S-dualities for example. We also must be concerned about the
possibility of non-perturbative effects modifying our results, through, for example, the
appearance of multitrace operators in the superpotential.
For relevant deformations, there will be renormalization group flows which are reflected
in near-horizon geometry. It would be of interest to construct these flows for the examples
that we have studied, particularly since one expects that stringy corrections become
important in the infrared. The study of correlation functions should also be of interest,
with a possible connection to the representation theory of quantum groups.
Generalizations of our work to more complicated quivers is possible and will be explored
elsewhere.
Acknowledgements
We wish to thank M. Ando, D. Grayson, A. Hashimoto, A. Jevicki, D. Kutasov,

J. Maldacena, A. Strominger and C. Vafa for discussions. We are especially indebted
to M. Strassler for a critical reading of this paper. DB thanks Harvard University for
hospitality. Work supported in part by U.S. Department of Energy, grant DE-FG02-
91ER40677 and an Outstanding Junior Investigator Award.
References
[1] R.G. Leigh, M.J. Strassler, Exactly marginal operators and duality in four-dimensional N = 1
supersymmetric gauge theory, Nucl. Phys. B 447 (1995) 95–136, hep-th/9503121.
[2] J. Maldacena, The large-N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231–252, hep-th/9711200.
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253–291,
hep-th/9802150.
[4] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string
[5] H.J. Kim, L.J. Romans, P. van Nieuwenhuizen, The mass spectrum of chiral N = 2 d = 10
supergravity on S 5 , Phys. Rev. D 32 (1985) 389.
[6] M.R. Douglas, D-branes and discrete torsion, hep-th/9807235.
[7] D. Berenstein, R.G. Leigh, Discrete torsion, AdS/CFT and duality, JHEP 01 (2000) 038, hep-
th/0001055.
[8] P.C. Argyres, K. Intriligator, R.G. Leigh, M.J. Strassler, On inherited duality in N = 1 d = 4
supersymmetric gauge theories, hep-th/9910250.
[9] D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, Renormalization group flows from
holography supersymmetry and a c-theorem, hep-th/9904017.
[10] L. Girardello, M. Petrini, M. Porrati, A. Zaffaroni, Novel local CFT and exact results on
perturbations of N = 4 super Yang–Mills from AdS dynamics, JHEP 12 (1998) 022, hep-
th/9810126.
[11] C. Vafa, E. Witten, A strong coupling test of S duality, Nucl. Phys. B 431 (1994) 3–77, hep-
th/9408074.
[12] R. Donagi, E. Witten, Supersymmetric Yang–Mills theory and integrable systems, Nucl. Phys.
B 460 (1996) 299–334, hep-th/9510101.
[13] M.J. Strassler, Messages for QCD from the superworld, Prog. Theor. Phys. Suppl. 131 (1998)
439, hep-lat/9803009.
[14] N. Dorey, An elliptic superpotential for softly broken N = 4 supersymmetric Yang–Mills
theory, JHEP 07 (1999) 021, hep-th/9906011.
[15] N. Dorey, S.P. Kumar, Softly-broken N = 4 supersymmetry in the large-N limit, JHEP 02
(2000) 006, hep-th/0001103.
[16] J. Polchinski, M.J. Strassler, The string dual of a confining four-dimensional gauge theory, hep-
th/0003136.
[17] R.C. Myers, Dielectric-branes, JHEP 12 (1999) 022, hep-th/9910053.
[18] Y.I. Manin, Quantum groups and non-commutative geometry, Université de Montréal, Centre
de Recherches Mathématiques, Montreal, PQ, 1988.
[19] A. Klimyk, K. Schmüdgen, Quantum groups and their representations, Springer-Verlag, Berlin,
1997.
[20] A. Connes, M.R. Douglas, A. Schwarz, Non-commutative geometry and matrix theory:
compactification on tori, JHEP 02 (1998) 003, hep-th/9711162.
[21] N. Seiberg, E. Witten, String theory and non-commutative geometry, JHEP 09 (1999) 032, hep-
th/9908142.
[22] R. Minasian, G. Moore, K-theory and Ramond–Ramond charge, JHEP 11 (1997) 002, hep-
th/9710230.
[23] E. Witten, D-branes and K-theory, JHEP 12 (1998) 019, hep-th/9810188.
[24] G. Moore, E. Witten, Self-duality, Ramond–Ramond fields, and K-theory, hep-th/9912279.
[25] M. Artin, J.J. Zhang, Non-commutative projective schemes, Adv. Math. 109 (2) (1994) 228–
287.
[26] A.L. Rosenberg, Non-commutative Algebraic Geometry and Representations of Quantized
Algebras, Kluwer Academic, Dordrecht, 1995.
[27] M. Kapranov, Non-commutative geometry based on commutator expansions,
math.AG/9802041.
[28] A. Hashimoto, N. Itzhaki, Non-commutative Yang–Mills and the AdS/CFT correspondence,
Phys. Lett. B 465 (1999) 142, hep-th/9907166.
[29] J.M. Maldacena, J.G. Russo, Large N limit of non-commutative gauge theories, JHEP 09 (1999)
025, hep-th/9908134.
[30] M.V. Raamsdonk, N. Seiberg, Comments on non-commutative perturbative dynamics, hep-
th/0002186.
[31] S. Minwalla, M.V. Raamsdonk, N. Seiberg, Non-commutative perturbative dynamics, hep-
th/9912072.
[32] M.R. Gaberdiel, Lectures on non-BPS Dirichlet branes, hep-th/0005029.
[33] M.A. Luty, I. Washington Taylor, Varieties of vacua in classical supersymmetric gauge theories,
Phys. Rev. D 53 (1996) 3399–3405, hep-th/9506098.
[34] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M-theory as a matrix model: a conjecture,
Phys. Rev. D 55 (1997) 5112–5128, hep-th/9610043.
[35] J.A. Harvey, G. Moore, On the algebras of BPS states, Commun. Math. Phys. 197 (1998) 489,
hep-th/9609017.
[36] M.R. Douglas, Topics in D-geometry, Class. Quant. Grav. 17 (2000) 1057, hep-th/9910170.
[37] A. Connes, Non-commutative Geometry, Academic Press, San Diego, 1994.
[38] D. Berenstein, V. Jejjala, R.G. Leigh, Work in progress.
[39] R. Hartshorne, Algebraic Geometry, Springer-Verlag, New York, 1977, Graduate Texts in
Mathematics, No. 52.
[40] G. Landi, An introduction to non-commutative spaces and their geometry, hep-th/9701078.
[41] R. Dijkgraaf, E. Verlinde, H. Verlinde, Matrix string theory, Nucl. Phys. B 500 (1997) 43, hep-
th/9703030.
[42] M.B. Green, J.A. Harvey, G. Moore, I-brane inflow and anomalous couplings on D-branes,
Class. Quant. Grav. 14 (1997) 47–52, hep-th/9605033.
[43] J. Rosenberg, Algebraic K-theory and Its Applications, Springer-Verlag, New York, 1994.
[44] A. Sen, Tachyon condensation on the brane antibrane system, JHEP 08 (1998) 012, hep-
th/9805170.
[45] M.R. Douglas, G. Moore, D-branes, Quivers, and ALE Instantons, hep-th/9603167.
[46] J. Gomis, D-branes on orbifolds with discrete torsion and topological obstruction, hep-
th/0001200.
[47] M.R. Douglas, B. Fiol, D-branes and discrete torsion, II, hep-th/9903031.
[48] I.R. Klebanov, E. Witten, AdS/CFT correspondence and symmetry breaking, Nucl. Phys. B 556
(1999) 89, hep-th/9905104.
[49] D.R. Morrison, M.R. Plesser, Non-spherical horizons, I, Adv. Theor. Math. Phys. 3 (1999) 1,
hep-th/9810201.
[50] C. Bachas, M. Douglas, C. Schweigert, Flux stabilization of D-branes, hep-th/0003037.
[51] W. Taylor, D2-branes in B fields, hep-th/0004141.
[52] A. Strominger, S.-T. Yau, E. Zaslow, Mirror symmetry is T-duality, Nucl. Phys. B 479 (1996)
243–259, hep-th/9606040.
[53] D. Berenstein, R.G. Leigh, Superstring perturbation theory and Ramond–Ramond backgrounds,
Phys. Rev. D 60 (1999) 106002, hep-th/9904104.
[54] A. Jevicki, S. Ramgoolam, Non-commutative gravity from the AdS/CFT correspondence,
JHEP 04 (1999) 032, hep-th/9902059.
[55] J. Maldacena, A. Strominger, AdS(3) black holes and a stringy exclusion principle, JHEP 12
(1998) 005, hep-th/9804085.
[56] J.-L. Loday, Cyclic Homology, Springer-Verlag, Berlin, 1992, Appendix E by María O. Ronco.
[57] E. Witten, Baryons and branes in anti-de Sitter space, JHEP 07 (1998) 006, hep-th/9805112.
[58] A. Kapustin, D-branes in a topologically nontrivial B-field, hep-th/9909089.
Electroweak symmetry breaking

and extra dimensions
Hsin-Chia Cheng a,∗ , Bogdan A. Dobrescu b,c , Christopher T. Hill a,b
a Enrico Fermi Institute, The University of Chicago, Chicago, IL 60637, USA
b Theoretical Physics Department, Fermi National Accelerator Laboratory, Batavia, IL 60510, USA
c Institute for Theoretical Physics, University of California, Santa Barbara, CA 93106, USA
Received 23 December 1999; revised 6 June 2000; accepted 16 June 2000
Abstract
Electroweak symmetry can be naturally broken by observed quark and gauge fields in various
extra-dimensional configurations. No new fundamental fields are required below the quantum
gravitational scale (∼ 10–100 TeV). We examine schemes in which the QCD gauge group alone, in
compact extra dimensions, forms a composite Higgs doublet out of (t, b)L and a linear combination
of the Kaluza–Klein modes of tR . The effective theory at low energies is the Standard Model. The
top-quark mass is controlled by the number of active tR Kaluza–Klein modes below the string scale,
and is in agreement with experiment.  2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Mj; 12.60.-i; 12.60.Rc; 11.10.Kk
1. Electroweak asymmetry and extra dimensions
There are two major experimental observations which are not explainable solely in terms
of the SU(3)C × SU(2)W × U (1)Y gauge interactions and the three generations of quarks
and leptons: the electroweak symmetry breaking and the existence of gravity. It is now
widely believed that a quantum theory of gravity necessitates a spacetime dimensionality
greater than four. In this paper we show that the extra spatial dimensions, compactified at
the ∼ TeV scale, also provide simple and natural mechanisms for electroweak symmetry
breaking without the introduction of explicit Higgs fields.
We will argue that the Standard Model is the effective theory emerging, below the
compactification scale, from a higher-dimensional SU(3)C × SU(2)W × U (1)Y gauge
theory with three generations of quarks and leptons and no fundamental Higgs field.
E-mail addresses: hcheng@theory.uchicago.edu (H.-C. Cheng), bdob@fnal.gov (B.A. Dobrescu),
hill@fnal.gov (C.T. Hill).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 0 1 - 6
250 H.-C. Cheng et al. / Nuclear Physics B 589 (2000) 249–268
A composite Higgs doublet arises naturally in the presence of certain strongly coupled
four-quark operators. For concreteness, we will take these to involve typically the left-
handed top-bottom doublet (ψL ) and a vector-like quark [1–5], but we anticipate many
possible variations of this particular arrangement. These particular four-quark operators
are always induced by QCD in compact dimensions, via the exchange of the Kaluza–Klein
(KK) excitations of the gluons [6]. Hence, the KK-gluons are effective “colorons” [7,8]
and their effects can be quite large because the higher-dimensional QCD coupling constant
increases above the compactification scale. The strength of these contact interactions
depends on the ratio of the compactification scale, Mc , and the scale Ms of the underlying
quantum gravitational effects. For Mc in the TeV range [9–11], Ms has to be around
10–100 TeV such that the quantum gravitational effects cut-off the non-renormalizable
higher-dimensional gauge interactions. Hence, the measured weakness of the gravitational
interactions has to be explained by a modification of gravity at short-distance, for instance
as proposed in Refs. [12–15].
Indeed, the dependence of four-quark operator coefficients on the Ms /Mc ratio allows
us to give a nice connection with string/M theory if one assumes that the gauge couplings
unify at the string scale [16]. Due to the power-law running of the gauge couplings in
extra dimensions [17,18], the value of the unified higher-dimensional coupling, g4+δ (Ms ),
and the Ms /Mc ratio are determined almost exclusively by the number δ of compact
dimensions accessible to the Standard Model gauge bosons. For δ > 2, g4+δ (Ms ) is of
order one in Ms units, corresponding to a string coupling of order one. This is in accord
with the argument based on dilaton stability [19] that string theory is in the truly strong-
coupling regime. Furthermore, the large value of g4+δ implies that the strength of the four-
quark operators induced by KK-gluon modes is non-perturbative, and may indeed bind a
composite Higgs.
The only remaining ingredient for a viable theory of dynamical electroweak symmetry
breaking is the above-mentioned vector-like quark. In four dimensions, a composite Higgs
doublet may be bound out of the ψL and the right-handed top field, tR [20–23]. However,
the Yukawa coupling of the Higgs doublet to its constituents is typically large, so that the
top quark mass is too large (unless the theory is fine-tuned to nearly exact criticality, and
the scale of the new interactions is taken to the GUT scale; alternatively, the measured top
quark mass forces√ the VEV of this Higgs doublet to be smaller than the Standard Model
Higgs VEV, v/ 2 where v ≈ 246 GeV is the electroweak scale).
On the other hand, if a new vector-like fermion is introduced with the same quantum
numbers as tR , it can then become the appropriate constituent of the Higgs boson together
with ψL . The physical top mass is given in this case by a smaller eigenvalue of a mass
matrix involving the vector-like and top quarks [1]. Therefore, such a seesaw mechanism
√
neatly accommodates both the measured top quark mass and a Higgs VEV of v/ 2.
It is quite striking that the Kaluza–Klein modes of the tR have exactly the quantum
numbers of this requisite vector-like quark. A key point of this paper is that the role of the
vector-like quark can be naturally played by the tower of KK modes of the tR . Therefore,
compact extra dimensions appear to provide everything needed for a dynamical seesaw
H.-C. Cheng et al. / Nuclear Physics B 589 (2000) 249–268 251
model of electroweak symmetry breaking. 1 Remarkably, however, while the vector-like

excitations are required, the seesaw mechanism is no longer needed here, because the top
Yukawa coupling is automatically suppressed by the (square-root of ) number of active KK
modes of the tR with masses below Ms . Moreover, for typical ratios of Ms to the mass of
the first quark KK excitation, the top Yukawa coupling computed to leading order in 1/Nc
is between ∼ 0.7 and ∼ 1.4. Thus, the Standard Model value for the top Yukawa coupling
(∼ 1) is a natural consequence of our framework.
In Section 2 we first discuss chirality and anomaly cancellation in the case of one extra
dimension. In Section 3 we present a detailed model of electroweak symmetry breaking
valid below the quantum gravity scale which does not require any new field beyond the
SU(3)C × SU(2)W × U (1)Y gauge fields and the three generations of fermions, in a higher-
dimensional configuration. We study the low energy effects of this model in Section 4.
Finally, our conclusions are summarized in Section 5.
2. Chirality and anomaly cancellation on a thick brane
In order to present the properties of the KK excitations of the tR , we start with a

general discussion of fermions in five dimensions. The tR may be the zero-mode of a
five-dimensional fermion only if the gluons and hypercharge gauge boson propagate in
the fifth dimension. Therefore the extra dimension has to be compact, with a radius below
∼ (3 TeV)−1 [10,11,16].
2.1. Chirality from orbifold projection
A constraint on the compactification of the extra dimension comes from the requirement
that the tR is a chiral, two-component fermion. The Lorentz group in five dimensions
SO(4, 1) has only one spin-1/2 representation which turns out to be non-chiral. The
fermions have four components, and the set of gamma matrices is formed of the usual
four-dimensional ones, γ µ , µ = 0, 1, 2, 3, and of iγ5 . Therefore, a chiral zero-mode of
a five-dimensional fermion may exist only if SO(4, 1) is broken. This can be done by
compactifying the fifth dimension on an orbifold, or by imposing boundary conditions on
the compact fifth dimension to distinguish the left- and right-handed components of the
five-dimensional fermion.
Consider the four-dimensional Minkowski spacetime, with coordinates x µ , and one
additional transverse spatial dimension, with coordinate y. A simple way of breaking
SO(4, 1) while preserving the four-dimensional Lorentz invariance and allowing chiral
fermions is to compactify the fifth dimension, y, on an S 1 /Z2 orbifold, i.e., a circle of
radius R = L/π with the identification y → −y. Five-dimensional fields are classified to
be even or odd under Z2 parity. In terms of KK decomposition, the zero modes of the
1 Other studies of electroweak symmetry breaking in extra dimensions without a fundamental Higgs doublet
can be found in [24,25].
odd fields are projected out. The assignment of opposite Z2 parity to the left- and right-
handed components of the five-dimensional fermion, χL (x, y) = −χL (x, −y), χR (x, y) =
χR (x, −y), leaves massless only one four-dimensional right-handed chiral fermion.
Equivalently, one may start by considering a five-dimensional spacetime with boundaries
along the fifth dimension at y = 0 and y = L. A four-component fermion field, χ(x, y),
is defined on this space as a solution to the five-dimensional Dirac equation which obeys
some conditions at y = 0, L. The simplest chiral boundary conditions,
PL χ(x, 0) = PL χ(x, L) = 0,
∂ ∂
PR χ(x, 0) = PR χ(x, L) = 0, (2.1)
∂y ∂y
where PL,R = (1 ∓ γ5 )/2, lead to the quantization of momentum in the y direction, and
give rise to the same KK decomposition as the S 1 /Z2 orbifold projection discussed above.
These boundary conditions may result from the interactions between the bulk fields and
the four-dimensional fields living on the branes located at y = 0, L. This in fact could be a
physical explanation for the S 1 /Z2 orbifold projection. In this paper, however, we do not
attempt to derive a theory valid at any energy scale, but rather we study an effective field
theory in a compact higher-dimensional spacetime defined below an ultra-violet cut-off
Ms . We therefore impose only four-dimensional general covariance, and assume that the
physics above Ms does not generate unwanted operators.
The complete set of orthogonal functions on the [0, L] interval consistent with the
boundary condition on χL (corresponding to odd fields under y → −y) is given by
r
2 πjy
sin , j > 1. (2.2)
L L
All these functions cancel on the boundaries, so that they indeed do not include a zero-
mode on the compact interval, [0, L]. On the other hand, the boundary conditions for χR
(corresponding to even fields) allow a complete set of orthogonal functions on [0, L],
r r
1 2 πjy
, cos , j > 1, (2.3)
L L L
which includes a zero-mode. The zero-mode of χR is identified as the right-handed top
quark in the weak eigenstate basis, tR . As a result, the decomposition of χ in KK modes is
chiral:
(
1 √ X j πjy
χ(x, y) = √ tR (x) + 2 PR χR (x) cos
L L
j >1
)
j πjy
+ PL χL (x) sin . (2.4)
L
A consequence of the boundary conditions (2.1) is that there is no fermion mass term in
the five-dimensional Lagrangian. Nevertheless, the Dirac equation,

∂
γ ∂µ + iγ5
µ
χ(x, y) = 0, (2.5)
∂y
includes a γ5 term so that it cannot be decomposed in separate equations for the left- and
right-handed fermions. It is straightforward to derive the fermion propagator for this five-
dimensional spacetime with the above boundary conditions:
h0| χ(x 0 , y 0 ) χ̄(x, y) |0i
Z
d 4 k ik µ (x−x 0)µ 2 X πjy 0 πjy 0
= e cos PR + sin PL
(2π)4 L L L
j >0

γ µ kµ + γ5 πj/L πjy πjy i
× µ sin PR + cos PL . (2.6)
k kµ − (πj/L)2 L L 1 + δj 0
We will use this propagator in Section 3.2 to derive the Higgs potential.
2.2. Chiral anomalies
Next we study what happens when the χ fermion transforms under some gauge
symmetry. This is necessary in order to show that the model of electroweak symmetry
breaking presented in the next section is anomaly-free.
Under the S 1 /Z2 orbifold projection discussed in the previous subsection, the ordinary
four-dimensional spacetime components of the gauge fields Aµ must be even while
the fifth components A5 must be odd, so that they have consistent interactions with
the χ fermion. Hence the fifth component A5 has no zero mode, and its KK modes
become the longitudinal components of the heavy Aµ modes. The zero-mode of A5 may
also be eliminated by imposing boundary conditions rather than an orbifold projection.
If the gauge fields propagate in the fifth dimension only on the 0 6 y 6 L interval,
then the appropriate boundary conditions are given by A5 (x, 0) = A5 (x, L) = 0 and
∂Aµ (x, 0)/∂y = ∂Aµ (x, L)/∂y = 0. Although the graviton need not propagate at y > L
or y < 0, we refer loosely to the [0, L] interval as a “thick brane” because 1/L is smaller
than the string scale.
The five-dimensional Lorentz-invariant gauge theories have no chiral anomalies because
the fermion representation is vector-like. However, the boundary conditions considered
above prevent the existence of a χL zero-mode, which raises the question of anomalies.
The Jχa,r ≡ χ̄γ a T r χ current has an anomaly given by
1
Da Jχa,r = 2
µνλρ Tr T r ∂µ Aν ∂λ Aρ + 12 Aν Aλ Aρ , (2.7)
24π L
0 0
where Aµ = −ig5 Arµ T r is the gauge field, the trace is over the products of group
generators T r , Da is the covariant derivative, and 0123 = 1. The index a runs from 0
to 4, with ∂4 ≡ ∂/∂y. Throughout this paper we use latin (greek) indices to denote the
components of five(four)-dimensional vectors.
Naively, one may think that this anomaly spoils the gauge invariance. It turns out,
however, that the anomaly in this five-dimensional theory is more subtle. This is because
the action may include a Chern–Simons term on the [0, L] interval:
L − y abcde
LCS (A) = 2
Tr Fab Fcd Ae − Fab − 25 Aa Ab Ac Ad Ae , (2.8)
96π L
where F is the gauge field strength. In the presence of the Chern–Simons term, the gauge
current becomes the sum of the fermion current and the Chern–Simons current. As a result,
the divergence of the total gauge current cancels everywhere on the open interval (0, L):
a,r
Da Jχa,r + JCS = 0. (2.9)
Hence, the gauge theory with a Chern–Simons term is well defined (i.e., non-anomalous)
in the bulk of the fifth dimension. This is to be contrasted with the gauge anomaly in four
dimensions, which cannot be canceled by any counterterm in the action.
The physical interpretation of anomaly cancellation in the bulk of our five-dimensional
theory is similar with that given in Ref. [26] for the case of domain wall fermions in 2 + 1
dimensions. In the present case, the anomaly due to tR on the [0, L] interval produces
gauge charges which are collected by the Chern–Simons current and transported towards
the boundary. Therefore, in the bulk there is charge conservation. At the boundary, though,
the charges are lost, so that the five-dimensional theory with only one zero-mode fermion
is indeed ill-behaved due to the anomaly. This can be seen by computing the variation of
the action under a gauge transformation:
Z ZL Z

δ 4
d x dy i χ̄γ Da χ + LCS = L
a
d 4
x Da Jχa,r αr , (2.10)
y=0
0
where αr is the gauge transformation parameter.
Therefore, there is need for other fermions such that the overall anomaly cancels, and
the five-dimensional theory reduces to a non-anomalous four-dimensional gauge theory
at scales below π/L. For simplicity we will assume that tR is the only fermion with KK
excitations below the string scale Ms . This is implemented in the effective field theory
below Ms by localizing all the Standard Model fermions with the exception of tR at certain
positions in the fifth dimension. Evidently, the anomaly cancellation matches well in the
effective theory below the compactification scale, where the only fermions present are the
four-dimensional three generations of quarks and leptons.
The microscopic implication of anomaly cancellation in this case is that the charges
which are driven by the Chern–Simons current (2.8) to the boundary are brought by another
Chern–Simons current to the location of the other third generation fermions where they are
absorbed by the corresponding four-dimensional anomalies. For example, a left-handed
fermion located at y = y0 and z = 0 requires a Chern–Simon term with a step function
shape,
θ(y − y0 ) − 1 abcde
Tr Fab Fcd Ae − Fab − 25 Aa Ab Ac Ad Ae , (2.11)
96π 2
to be added to LCS (A). As a result, the right-hand side of Eq. (2.10) vanishes and the theory
is gauge invariant.
We emphasize that the five-dimensional gauge theory is non-renormalizable. The gauge
coupling has mass dimension (−1/2), and it blows up at some scale ∼ Ms . Therefore, any
five-dimensional gauge theory should be seen only as an effective field theory which at
the scale Ms is replaced by a more fundamental framework, such as string or M theory.
The Chern–Simons terms discussed here are supposed to be produced within the theory
that introduces the physical cut-off Ms .
Another possibility is that all third generation fermions are defined on the [0, L] interval
with chiral boundary conditions similar with those of χ . In this case the overall Chern–
Simons current vanishes and the anomalies are canceled exactly as in the four-dimensional
Standard Model. However, this would imply that all third generation fermions have KK
excitations, which would complicate the analysis of the model presented in the next section.
To keep the discussion simple, we will not investigate this possibility here.
3. A model: right-handed top and QCD in extra dimensions
In this section we show that the dynamics in extra dimensions allows us to construct a
model of dynamical electroweak symmetry breaking without the need for a fundamental
Higgs field.
Consider a (4 + δ)-dimensional spacetime with the four-dimensional flat spacetime
extended in the x µ , µ = 0, . . . , 3, directions, and extra spatial dimensions with coordinates
y and z1 , . . . , zδ−1 . Only some of the observed fields propagate in the extra dimensions.
The simplest configuration is that where the gluons propagate in all these dimensions,
the tR is the zero mode of a fermion, χ , which is fixed at z = 0 but propagates on the
[0, L] interval in the y dimension, and the ψL = (t, b)L is fixed at z = 0 and y = y0 . We
choose δ > 3 such that the effects of the gluons with momentum in the z dimensions are
non-perturbative when the Ms scale is sufficiently large. 2 We will assume that the gluons
propagate on intervals of size L and Lz in the y and z1 , . . . , zδ−1 dimensions, respectively,
with Lz < L. In Fig. 1 we sketch the extra-dimensional configuration.
As mentioned in the previous section, it is convenient to assume that all other quarks
and the leptons are localized on four-dimensional slices of (4 + δ)-dimensional spacetime,
so that we do not have to worry about their KK modes. Furthermore, if the left- and right-
handed fermions (other than tR and ψL ) are split in the extra dimensions [27], then they
Fig. 1. The profile of the compact space. The x-coordinates of the flat three-dimensional space are
transverse to the plane of the page. The z1 , . . . , zδ−1 coordinates are depicted collectively as one
axis. The gluons propagate inside the rectangle, the χ propagates along the y axis, on the thick line,
and the ψL is located at the point marked on the y axis.
2 Note that in the case of a single compact dimension, the four-quark operators induced by the tree level
exchange of an infinite tower of gluon KK modes are finite.
cannot acquire large masses. Note that this splitting does not produce the kind of flavor-
changing neutral currents discussed in [11] provided the light fermions of same chirality
are located at the same places.
The U (1)Y gauge bosons have to propagate in the y dimension because χ carries
hypercharge. The SU(2)W gauge bosons must propagate in the fifth dimension only
if different weak-doublet fermions are localized at different places. Note that if gauge
coupling unification is imposed, then it is preferable to have the SU(2)W × U (1)Y gauge
bosons propagating in the same space as the gluons.
3.1. The five-dimensional theory
After compactifying and integrating over the z dimensions, √ we find a tower of KK

modes of the gluons, which are grouped in levels of masses π K/Lz with K a positive
integer, and degeneracies DK (DK = 0 for some values of K, see Ref. [16]). These gluon
KK modes are five-dimensional fields whose effects at energies below their masses are
approximately described by four-quark operators.
At scales between π/Lz and the string scale, Ms , the dynamics includes both light
gluon KK modes and four-quark operators induced by the heavier gluon KK modes.
Although each gluon KK mode is weakly coupled, the number of gluon KK modes may
be sufficiently large [6] such that the loop expansion breaks down. In order to analyze the
effects of this non-perturbative theory below some scale Λ, we approximate the dynamics
of the gluons with momentum in the z dimensions by a five-dimensional effective theory
with four-quark operators. The matching between the five-dimensional low-energy theory
and the (4 + δ)-dimensional theory is likely to require the scale Λ of the four-quark
operators to be somewhere between π/Lz and Ms .
By imposing that the loop expansion parameter [16] becomes of order one at Ms , we
can estimate the separation between π/Lz and Ms . For δ > 3, the density of KK modes is
large and it turns out that Ms is only about twice π/Lz . Therefore, the uncertainty in Λ is
not worrisome.
The relevant piece of the five-dimensional Lagrangian density, involving the four-
dimensional ψL (x µ ) field and the five-dimensional χ(x µ , y) and massless gluon fields
is given at the scale Λ by

L5 (x µ , y) = δ(y − y0 )i ψ̄L γ µ Dµ ψL + χ̄ iγ µ Dµ − γ5 Dy χ
1
− 2 Tr F ab Fab + LCS (G) + Lint . (3.1)
2g5
F ab is the gluon field strength, LCS (G) is the Chern–Simons term for the gluon field [see
Eqs. (2.8) and (2.11)], and D is the covariant derivative:
Dµ = ∂µ − Gµ ,
∂
Dy = − Gy , (3.2)
∂y
with Gµ,y = −ig5 Grµ,y T r being five-dimensional gluon fields (the zero modes from the
KK expansion in the z directions) polarized in the x µ and y directions, respectively. The
five-dimensional strong coupling constant, g5 , has dimension (mass)−1/2 .
The Lint part of the L5 Lagrangian includes the four-quark operators induced by gluon
KK mode exchange. Although the SU(3)C interactions are flavor universal, the four-
quark operators contained in Lint are not, because different quark fields are assumed to
be localized at different positions in the extra dimensions. For example, all SU(2)W singlet
quark fields other than tR and its excitations may be localized at z = zR > 0, and the
SU(2)W doublet quarks of the first two generations may be localized at z = zL > 0 with
zL 6= zR . In this case the terms from Lint that could lead to large quark masses in the low
energy theory, namely the left-right current-current terms, are exponentially suppressed
unless they involve only ψL and χ .
The four-quark operators involving ψL (x µ ) and χ(x µ , y), obtained by integrating out
the five-dimensional gluon KK excitations, are given by
cg52 2 2
Lint (x, y) = − δ(y − y0 ) ψ̄L γ µ T r ψL + χ̄γ µ T r χ + χ̄γ5 T r χ , (3.3)
2Λ2
where c 1 is a dimensionless coefficient obtained by summing over the effects of the
gluon KK modes, and T r are SU(3)C generators.
These four-quark operators may be Fierz transformed, with the result

cg52 5
Lint (x, y) = 2 δ(y − y0 )(ψ̄L χ)(χ̄ψL ) + (χ̄χ) − 3 (χ̄γ5 χ)
2 1 2
+ · · · , (3.4)
Λ 16
where the ellipsis stand for vectorial and tensorial four-quark operators, which are
irrelevant at low energies.
3.2. The five-dimensional effective potential
The operators shown in Lint provide attractive interactions which give rise to bound
states: a four-dimensional weak-doublet complex scalar, H (x µ ), and a five-dimensional
gauge singlet real scalar, ϕ(x µ , y). These composite scalars are propagating degrees of
freedom only below the compositeness scale. According to our approximation in which
the full KK mode dynamics is described at low energy by a five-dimensional theory with
four-quark operators, the compositeness scale is identified with Λ.
At the compositeness scale the composite scalars are non-propagating, and the four-
quark operators may be replaced by Yukawa interactions between the scalars and their
constituents. The first two terms shown in (3.4) are equivalent with
hq i q Λ2 2
Lc [Λ] = −δ(y − y0 ) cg52 (χ̄ψL )H + Λ2 H † H − 58 cg52 (χ̄χ)ϕ − ϕ , (3.5)
2
as can be seen by integrating out H (x µ ) and ϕ(x µ , y). The last term in Eq. (3.4) gives rise
to a five-dimensional pseudo-scalar. However, the coefficient of this term is suppressed by
the factor of 1/3, such that the pseudo-scalar is not sufficiently deeply-bound to be relevant
at energies below the compositeness scale.
At scales µ < Λ, the Yukawa interactions induce kinetic terms for the scalars:
h q i
Lc [µ] = δ(y − y0 ) ZH (µ)Dν H † Dν H − cg52 (ψ̄L χ)H
q
+ Zϕ (µ)∂ a ϕ∂a ϕ − 5 2
8 cg5 (χ̄χ)ϕ − V (µ). (3.6)
The wave function renormalization ZH can be determined by computing the self-energy
of the weak-doublet in the large-Nc limit (see Fig. 2):
Z
cg 2 X cos2 (πjy0 /L) d 4k −i
ZH (µ) = 2Nc 5 . (3.7)
L
j >0
1 + δj 0 (2π) k kµ k kν − (πj/L)2
4 µ ν
The integral is logarithmic divergent, and has to be cut-off at Λ. The sum over the momenta
in the fifth dimension is also divergent, and is cut-off at nKK ≈ ΛL/π . The integral has also
an infrared cut-off at µ.
The wave function renormalization for the ϕ scalar has a more complicated form, due to
the two χ propagators involved [see Eq. (2.6)]. Keeping only the leading divergent piece,
we find
Z
5 cg52 X d 4k −i
Zϕ (µ) ≈ Nc . (3.8)
4 L (2π)4 k ν kν − (πj/L)2 2
j >0
Note that the wave function renormalization for ∂ϕ/∂y is somewhat arbitrary (it can be
absorbed in the mass term for ϕ), and does not have to be the same as Zϕ (µ). In Lc [µ] we
have chosen these two wave function renormalizations to be the same for convenience.
The scalar potential includes mass and quartic terms,

λ̃H λ̃0 L † eH
V (µ) = δ(y − y0 ) (H †H )2 + H H ϕ2 + M 2
H †H
2 2
λ̃ϕ L 4 Mϕ 2 e 2
+ ϕ + ϕ , (3.9)
4! 2
as well as higher-dimensional terms which we will ignore. The mass parameters computed
in the large-Nc limit are given by
Z
cg 2 X cos2 (πjy0 /L) d 4k i
eH
M 2
(µ) = Λ2 − 4Nc 5 ,
L 1 + δj 0 (2π)4 k ν kν − (πj/L)2
j >0
Fig. 2. Large-Nc contributions to the composite scalar self-energies. The vertical lines are
four-dimensional fields localized at y = y0 , and the curved or slanted lines are five-dimensional
fields. The external lines represent the H and ϕ, while in the loops run the ψL and χ quarks.
Z
e 5 cg52 X d 4k i
Mϕ (µ) ≈ Λ − Nc
2 2
. (3.10)
2 L (2π)4 k ν kν − (πj/L)2
j >0
In the expression for Meϕ2 we have kept again only the leading divergent piece.
In the large-Nc limit, the leading contribution to the dimensionless quartic coupling,
λ̃H , is given by a quark loop with alternating χ and ψL propagators (Fig. 3). Therefore,
the result can be written as a double sum over the χ momenta in the fifth dimension:
2 2 X
cg5 2 πj1 y0 2 πj2 y0
λ̃H (µ) = 8Nc fj1 j2 cos cos , (3.11)
L L L
j1,2 >0
where we have defined

Z
1 d 4k −i
fj1 j2 ≡ . (3.12)
(1 + δj1 0 )(1 + δj2 0 ) (2π)4 [k ν kν − (πj1 /L)2 ][k ρ kρ − (πj2 /L)2 ]
The coefficients of the quartic terms involving ϕ have mass dimension −1. The factors of
L are introduced in Eq. (3.9) such that the λ̃0 and λ̃ϕ quartic couplings are dimensionless:
2 2 X
5 cg5 dj1 j2 dj3 j2 πj1 y0 πj3 y0
λ̃0 (µ) ≈ Nc fj1 j3 cos cos ,
4 L (1 + δj2 0 ) L L
j1,2,3 >0
2 2 X
75 cg5 dj1 j2 dj3 j2 dj3 j4 dj1 j4
λ̃ϕ (µ) ≈ Nc fj1 j2 . (3.13)
128 L (1 + δj3 0 )(1 + δj4 0 )
j1,2,3,4 >0
When all the ϕ fields from the ϕ 4 interaction have momentum π/L in the y direction, we
obtain:
dj1 j2 ≡ δj2 ,j1 +1 − δj2 ,j1 −1 + δj2 ,1−j1 . (3.14)
To evaluate all these parameters, we assume that the number of the KK modes in the y
direction, nKK , is large enough so that we can approximate the sums over KK states by
integrals. The expressions obtained for the parameters are given in the appendix.
The kinetic terms in Lc [µ] [see Eq. (3.6)] may
p be canonically normalized by redefining
√
the scalar fields: H → H ZH and ϕ → ϕ Zϕ . In this case, the terms in the scalar
potential have the same form as in Eq. (3.9), but with appropriately normalized coefficients.
We denote the new parameters by dropping the tilde from the corresponding symbols used
in Eq. (3.9). The squared-masses are given by
Fig. 3. Large-Nc contributions to the λ̃H , λ̃0 and λ̃ϕ quartic couplings. The lines represent fields as
explained in the caption of Fig. 2.
e2
M 2Λ2 4π 2
2
MH = H
≈ − F3 0 ,
(y )
ZH F1 (y0 ) nKK Nc cgs2
eϕ2
M 2Λ2 32π 2
Mϕ2 = ≈ − F4 . (3.15)
Zϕ F2 5nKK Nc cgs2
We have used here the four-dimensional SU(3)C gauge coupling, gs , obtained in terms of
the five-dimensional coupling by integrating over the y dimension:
g5
gs = √ . (3.16)
L
The dependence of MH 2 on the position y of the ψ doublet is encoded in the F (y )
0 L 1,3 0
functions, which are symmetrical under the y0 → L − y0 reflection. F1 (y0 ) and F3 (y0 )
have values of order one, with maxima on the boundary and minima at y = L/2. F2 and F4
are constant functions on the [0, L] interval because the ϕ mass is induced by interactions
which conserve momentum in the y dimension. These functions are given in terms of
divergent sums and integrals and depend on the cut-off procedure. In the appendix we
estimate them in the continuum limit with a specific cut-off.
Similarly, the quartic couplings may be written as follows:
λ̃H 32π 2 F5 (y0 )
λH = ≈ ,
ZH2 Nc [F1 (y0 )]2
λ̃ϕ F6 (y0 )
λ0 = ≈ λϕ ,
3ZH Zϕ F1 (y0 )
λ̃ϕ 16π 2
λϕ = 2
≈ . (3.17)
Zϕ nKK Nc F2
Like the other F -functions written in the appendix, F5,6 (y0 ) ∼ 1. Note that λH is enhanced
by an nKK factor compared with the other quartic couplings.
4. Four-dimensional phenomenology
The squared-mass parameters from the five-dimensional potential may turn negative if
the four-quark operators induced by gluon KK modes are strong enough. Therefore, the
four-dimensional field, H , and the five-dimensional real scalar, ϕ, may acquire VEVs. The
composite weak-doublet H may be identified with the Standard Model Higgs doublet. In
this section we discuss the scalar spectrum and its phenomenological implications, and we
estimate the top quark mass.
4.1. Higgs boson mass
First, we consider the case in which ψL is located at the boundary (y0 = 0). An
inspection of the squared-masses computed in the large-Nc limit [see Eq. (3.15)] reveals
that only MH 2 should become negative because the coupling in the χ̄ψH channel is
stronger than the coupling in the χ̄χϕ channel. In addition, the four-dimensional quartic
coupling involving both H and ϕ vanishes in this case because the ϕ has a zero wave
function on the boundary. This implies that there is no mixing between H and ϕ. Therefore,
the ϕ has no effect on the Higgs potential in this case. The H acquires a VEV while the
KK modes of ϕ have masses above the compactification scale.
The effective theory below the compactification scale is given by the Standard Model.
The compositeness of the Higgs doublet is not manifest at low-energy. However, as a
remnant of the strong dynamics that binds the Higgs, the quartic Higgs coupling is large,
√
λH 1. The Higgs boson mass Mh0 , given at tree level by v λH (v), appears to be above
1 TeV. The tree level estimate, though, should not be taken too seriously due to the large
λH . Because the theory that gives rise to the composite Higgs boson is unitary (above the
Ms scale, the unitarity should be enforced by quantum gravitational effects), the Higgs
mass is below the bound imposed by the unitarity of the W W scattering cross section in
the Standard Model. Once the non-perturbative corrections to Mh0 are included, we expect
Mh0 ∼ O(1/2) TeV. Generically, when the ψL is at y0 = 0, the Higgs boson is a broad
resonance.
Note that such a heavy Higgs boson is perfectly compatible with the electroweak
precision data. The often quoted upper bound on the Higgs boson based on the fit to the
electroweak data is valid only if there are no fields or interactions beyond the Standard
Model [28–31]. In our case, however, there are KK excitations of the Standard Model gauge
bosons and tR , with masses in the TeV range. In their presence, a heavier Higgs boson is
not only allowed, but potentially preferred by the fit to the data. This has been shown
in the context of extra dimensions in Ref. [32]. Specifically, the shift in the electroweak
observables due to the mixing of the W and Z with their KK excitations reproduces that
due to a light Higgs boson (when the Higgs is trapped on a (3 + 1)-dimensional wall, like
in our case). Furthermore, when a vector-like quark identical with our KK modes of tR is
added to the Standard Model, the fit to the electroweak data yields a heavy Higgs for a
vector-like quark mass around 5 TeV [4]. Of course, when the vector-like quark is much
heavier, or equivalently the compactification scale in our model is increased, one recovers
the Standard Model in the decoupling limit. Therefore, the y0 = 0 case is consistent with
the electroweak precision data only if the compactification scale is not above O(10 TeV).
In the other case, where the ψL fermion doublet is located in the middle of the interval
occupied by χ , i.e., y0 ∼ L/2, both H and ϕ may develop VEVs. (Note that Eqs. (3.15)
imply that for F3 (y0 ) ≈ 5F4 /8 both MH 2 and M 2 turn negative at some particular value
ϕ
2
of nKK cgs .) Since the Higgs VEV is below the compactification scale, it is appropriate to
integrate first over the fifth dimension, and only afterwards to minimize the potential. The
five-dimensional real scalar decomposes in a tower of KK modes
r
2X πjy
ϕ(x , y) =
µ µ
ϕj (x ) sin . (4.1)
L L
j >1
It is likely that only one or the first few modes of ϕ are light enough to have a significant
mixing with the H .
For simplicity, we consider that the Higgs field mixes with one ϕ mode. The four-
dimensional potential may be obtained readily from Eq. (3.9):

λH 2 πy0
V4 = (H H ) + λ0 (y0 ) sin
† 2
H † H ϕ12
2 L

λϕ 4 1 π2 2
+ ϕ1 + MH H H +
2 †
Mϕ + 2 ϕ1 .
2
(4.2)
16 2 L
After the scalar potential is minimized and the scalar fields shifted, we find the following
mass terms for the two light neutral degrees of freedom:
 
2 2 πy0  
 λH v 2λ0 (y0 )vu sin  h
1 L
(h, φ) 
  ,
 (4.3)
2 2 πy0 1 2 φ
2λ0 (y0 )vu sin λϕ u
L 2
where v ≈ 246 GeV and u is the ϕ1 VEV.
In the limit λH v2 λϕ u2 , the mixing of h and φ decreases the Higgs boson mass:

8[λ0 (y0 )]2 4 πy0
Mh20 ≈ λH v2 1 − sin . (4.4)
λH λϕ L
For y0 = L/2, the Mh0 decreases by ∼ (70/nKK)%. This value is derived using the λH
given in Eq. (3.17). As argued before, we expect that the quantum corrections actually
drive λH smaller, which would lead to an enhancement of the change in Mh0 due to
mixing. If more ϕ modes participate in the mixing, the decrease in Mh0 becomes even
more significant. Perhaps the Higgs boson may be driven close to the current LEP limit.
Unfortunately, it is hard to study the scalar spectrum in general, with all KK modes
included, especially given that the parameters of the full effective potential are not
accurately known.
In the other limit, where λH v2 λϕ u2 , the h − φ mixing may be ignored. The Higgs
boson remains heavy, but the physical φ 0 scalar could be very light. Its mass,
r
λϕ
Mφ ≈ u
0
, (4.5)
2
is not constrained by the consistency of the model. The experimental lower bounds on a
neutral scalar which couples only to the top quark and the Higgs boson are quite weak [5].
It is therefore possible that the Higgs boson decays into φ 0 pairs, giving rise to unusual
signals at future collider experiments [33].
4.2. Top-quark mass prediction
We can now predict the top-quark mass as a function of the number of KK modes and
the position y0 of the ψL doublet. The fermion couplings to the composite scalars are
given by Eq. (3.6). Upon normalization of the scalar kinetic terms and integration over the
y dimension, the Yukawa couplings become:
nKK
X 1/2
2 πjy0 j
−ξt cos χ̄R ψL H
1 + δj 0 L
j =1
X
nKK
j j
− ξχ (δj3 ,j1 +j2 − δj3 ,j1 −j2 + δj3 ,j2 −j1 )χ̄L1 χR2 ϕj3 + h.c. (4.6)
j1,2,3 =1
Note that the Yukawa couplings of the Higgs doublet depend on the position in the fifth
dimension. The zero-mode of χ , namely tR , has a Yukawa coupling to H given by
√
2 2π
ξt = √ . (4.7)
Nc nKK F1 (y0 )
The Yukawa couplings of the ϕ KK modes are position-independent due to momentum
conservation at the χ̄χφ vertex:
2π
ξχ = √ . (4.8)
Nc nKK F2
The fermion masses for the tL component of ψL and the KK modes of χ form a
(nKK + 1) × (nKK + 1) matrix. There are two contributions to the elements of this mass
matrix. First, the Yukawa interactions give contributions determined by replacing the
H and ϕ j scalars with their VEVs in Eq. (4.6). Second, the kinetic term of the five-
dimensional χ field yields the usual KK mass terms:
X
nKK
πj j j
χ̄L χR . (4.9)
L
j =1
In the case where y0 = 0, the fermion mass matrix is easy to write:

 √  
ξt v/ 2 ξt v ξt v ··· tR

 0 π/L 0 ···  1 
  χR 
t¯L , χ̄L1 , χ̄L2 , . . .    2  + h.c. (4.10)
 0 0 2π/L · · ·   χR , 
··· ··· ··· ··· ···
The top-quark mass (in the limit where we ignore the small mixing of the top with the
charm and up) is given by the lowest eigenvalue of the above mass matrix. It is amusing
that this matrix has the same form as the one for neutrino masses given in Ref. [34]. Note
that our assumption that the KK-gluons in the z dimensions may be integrated out below
the cut-off scale Λ (see Section 3) is legitimate provided Λ π/L. Thus, to be consistent
we must impose nKK > 10. Expanding in (vL/π)2 1, and taking nKK 1, we find
ξt v
mt ≈ √ 1 − 32 (ξt vL)2 . (4.11)
2
For L . 1 TeV−1 , the second term gives a small correction (. 1/nKK) to mt . Therefore,
the top mass is predicted in terms of nKK :
600 GeV
mt ≈ √ . (4.12)
nKK
The measured top mass can be used to determine the number of top KK modes:
nKK ≈ 12. (4.13)
The number of top KK modes is related to the cut-off scale Λ ≈ nKK π/L, which is of the
order of the string scale Ms . If the first KK modes have a mass of a few TeV, then the above
prediction determines the scale of quantum gravity Ms ∼ 30 TeV.
Furthermore, given that a cut-off scale significantly above ∼ 50 TeV would require
excessive fine-tuning (we assume that the theory is not supersymmetric below the string
scale), we find a naturalness upper bound nKK . 20. Therefore, instead of using the
measured top mass to determine the number of KK modes, we may reverse the argument
and determine the typical values of the top mass in our model. For 10 . nKK . 20, we
find a range, 130 GeV . mt . 190 GeV, which within the theoretical uncertainties is in
agreement with the measured value.
When the ψL is placed in the middle of the thick brane occupied by χ , i.e., y0 ∼ L/2,
some of the ϕ KK modes may acquire VEVs, as discussed in Section 4.1. Therefore, the
fermion mass matrix becomes more complicated to analyze. If only the first ϕ KK mode
has a non-zero VEV, u, and u π/L, then the top mass may be computed as in the y0 = 0
√
case. The only notable difference is √ that mt is enhanced by a factor of F1 (0)/F1 (y0 ).
This factor reaches its maximum of 2 at y0 = L/2. It appears that the upper end of the
interval for nKK is preferred in this case.
In the more general case, where the VEVs of some ϕj are comparable with the
compactification scale, one could imagine that the preferred value of the string scale is
lower, Ms ∼ 10 TeV. In such a situation our estimates would no longer be reliable, but the
qualitative picture of a composite Higgs doublet bound out of ψL and a tower of tR KK
modes might remain valid.
Finally, we emphasize that the masses of the light quark and leptons may easily be
accommodated in our scenario. For example, certain four-quark operators presumed to be
generated at the string scale with coefficients of order one in Ms units, give rise in the
low-energy effective theory to the Standard Model Yukawa couplings [5].
5. Conclusions
Electroweak symmetry breaking remains the foremost problem facing elementary

particle physics at this moment. We expect to come to understand it in scientific detail
in the next decade with the Tevatron and the LHC.
We find it remarkable that the ingredients needed for a dynamical explanation
of the origin of the electroweak scale, which we often have previously invoked in
model building attempts (e.g., topcolor, vector-like fermions, strong coupling Nambu–
Jona-Lasinio dynamics, etc.), are seemingly presented automatically in theories with
extradimensions at the ∼ TeV scale.
In this paper we have explicitly constructed a “demo-model” of the dynamics in
which the only fundamental fields below the string scale are the SU(3)C × SU(2)W ×
U (1)Y gauge bosons and the three generations of quarks and leptons, living in a higher-
dimensional compact space.
Strong dynamics comes from the existing QCD gauge group, which has a large coupling
strength above the compactification scale, due to the large number of KK modes. The KK
mode gluons act like degenerate octets of colorons which, via exchange, give rise to four-
fermion operators. Thus follows an NJL approximation to the dynamics induced by these
operators.
We find that various attractive channels lead to the formation of scalar bound-states. The
Higgs doublet channel corresponds to χ̄ψL where χ is the right-handed top quark field
which we take to live in the bulk. While χ has a chiral zero-mode by construction, which
is the tR , the Higgs doublet emerges as a bound-state involving a linear combination of
the active KK modes inherent in χ . In the effective theory the large number of active
KK modes, nKK , controls the dynamics, and naturally leads to a tachyonic mass term
for the Higgs at low energies, and thus electroweak symmetry breaking. We also expect
various gauge-singlet composite bosons to form in channels such as χ̄χ , which somewhat
complicate the discussion of the low energy spectroscopy. A low mass Higgs boson may
emerge through mixing between the primary composite Higgs and the extra composite
singlets.
Our model is largely intended to illustrate what can happen in the extra-dimensional
theories. It is hardly unique. The only selection criterion seems to be the assignment
of Standard Model fields to the world-brane or into the bulk, in various dimensional
configurations. We believe that, once the brane/bulk field assignments are made in this
manner, much of the dynamics we describe is forced to happen. New strong dynamics is
therefore natural and expected to occur in these theories. The experimental confirmation of
a strongly interacting Higgs sector beyond the Standard Model would, though not “imply”,
nonetheless lend support to the notion of extra dimensions at the TeV scale.
Acknowledgements
We would like to thank Bill Bardeen, Jeffrey Harvey, and Martin Schmaltz, for valuable
comments. H.-C. Cheng is supported by Department of Energy Grant DE-FG02-90ER-
40560 and by a Robert R. McCormick Fellowship. The research of B. A. Dobrescu is
supported by NSF Grant PHY94-07194, and by DOE Grant DE-AC02-76CH03000.
Appendix A. Effective potential parameters
In this appendix we give the formulae for the parameters of the low-energy effective
Lagrangian in the continuous approximation by replacing sums of the KK states with the
momentum integrals in the fifth direction.
Cutting off the integrals at Λ and replacing ΛL/π by nKK , we find the following wave
function renormalizations at low-energy (∼ L−1 )
Nc cg52
ZH ≈ nKK 2 F1 (y0 ),
16π 2L
Nc cg52 5
Zϕ ≈ nKK F2 . (A.1)
16π 2L 4
Likewise, we find the parameters from the five-dimensional effective potential (see
Section 3.2):

Nc cg52
eH
M 2
≈ Λ2 1 − nKK 4F3 0 ,
(y )
16π 2 L

N cg 2
Meϕ2 ≈ Λ2 1 − nKK c 5 5 F4 ,
16π 2 L 2

Nc cg52 2
λ̃H ≈ n2KK 8F5 (y0 ),
16π 2 L

Nc cg52 2
λ̃0 ≈ nKK 5F6 (y0 ),
16π 2 L

Nc cg52 2 75
λ̃ϕ ≈ nKK F2 , (A.2)
16π 2 L 8
where the F -functions are defined by
Z1 Z1
1
F1 (y0 ) = 2
p dp 2
dq cos2 (qΛy0) ,
p2 (p2 + q 2)
0 0
Z1 Z1
1
F2 = 2
p dp 2
dq
(p2 + q 2 )2 ,
0 0
Z1 Z1
1
F3 (y0 ) = 2
p dp 2
dq cos2 (qΛy0) ,
p2 + q2
0 0
Z1 Z1
1
F4 = 2
p dp 2
dq ,
p2 + q 2
0 0
Z1 Z1 Z1
1
F5 (y0 ) = 2
p dp 2 2
dq cos (qΛy0) dq 0 cos2 (q 0 Λy0 ) b,
(p2 + q 2 )(p2 + q 02 )
0 0 0
Z1 Z1
1
F6 (y0 ) = p2 dp2 dq cos2 (qΛy0) . (A.3)
(p2 + q 2 )2
0 0
For ψL localized at the boundary, we have

π
F1 (0) = + ln 2 ≈ 2.26,
2
π
F6 (0) = F2 = + ln 2 ≈ 1.48,
4
1 π 1
F3 (0) = F4 = + − ln 2 ≈ 0.63,
3 6 3
F5 (0) ≈ 2.71. (A.4)
If ψL is localized in the middle of the [0, L] interval and Λy0 1, the cos2 (qΛy0) weigth
factor averages to 1/2, and therefore
1
F1,3,6 (y0 ∼ L/2) ≈ F1,3,6 (0),
2
1
F5 (y0 ∼ L/2) ≈ F5 (0). (A.5)
4
References
[1] B.A. Dobrescu, C.T. Hill, Electroweak symmetry breaking via top condensation seesaw, Phys.
Rev. Lett. 81 (1998) 2634, hep-ph/9712319.
[2] R.S. Chivukula, B.A. Dobrescu, H. Georgi, C.T. Hill, Top quark seesaw theory of electroweak
symmetry breaking, Phys. Rev. D 59 (1999) 075003, hep-ph/9809470.
[3] G. Burdman, N. Evans, Flavor universal dynamical electroweak symmetry breaking, Phys. Rev.
D 59 (1999) 115005, hep-ph/9811357.
[4] H. Collins, A. Grant, H. Georgi, The phenomenology of a top quark seesaw model, Phys. Rev.
D 61 (2000) 055002, hep-ph/9908330.
[5] B.A. Dobrescu, Minimal composite Higgs model with light bosons, hep-ph/9908391.
[6] B.A. Dobrescu, Electroweak symmetry breaking as a consequence of compact dimensions,
Phys. Lett. B 461 (1999) 99, hep-ph/9812349; Higgs compositeness from top dynamics and
extra dimensions, hep-ph/9903407.
[7] C.T. Hill, Topcolor: top quark condensation in a gauge extension of the standard model, Phys.
Lett. B 266 (1991) 419.
[8] R.S. Chivukula, A.G. Cohen, E.H. Simmons, New strong interactions at the tevatron?, Phys.
Lett. B 380 (1996) 92, hep-ph/9603311.
[9] I. Antoniadis, A possible new dimension at a few tev, Phys. Lett. B 246 (1990) 377.
[10] C.D. Carone, Electroweak constraints on extended models with extra dimensions, Phys. Rev.
D 61 (2000) 015008, hep-ph/9907362.
[11] A. Delgado, A. Pomarol, M. Quiros, Electroweak and flavor physics in extensions of the
standard model with large extra dimensions, JHEP 0001 (2000) 030, hep-ph/9911252, and
references therein.
[12] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, The hierarchy problem and new dimensions at a
millimeter, Phys. Lett. B 429 (1998) 263, hep-ph/9803315.
[13] I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, New dimensions at a millimeter to
a Fermi and superstrings at a TeV, Phys. Lett. B 436 (1998) 257, hep-ph/9804398.
[14] L. Randall, R. Sundrum, A large mass hierarchy from a small extra dimension, Phys. Rev.
Lett. 83 (1999) 3370, hep-ph/9905221.
[15] J. Lykken, L. Randall, The shape of gravity, hep-th/9908076.
[16] H.-C. Cheng, B.A. Dobrescu, C.T. Hill, Gauge coupling unification with extra dimensions and
gravitational scale effects, Nucl. Phys. B 573 (2000) 597, hep-ph/9906327.
[17] T.R. Taylor, G. Veneziano, Strings and D = 4, Phys. Lett. B 212 (1988) 147.
[18] K.R. Dienes, E. Dudas, T. Gherghetta, Extra spacetime dimensions and unification, Phys. Lett.
B 436 (1998) 55, hep-ph/9803466; Grand unification at intermediate mass scales through extra
dimensions, Nucl. Phys. B 537 (1999) 47, hep-ph/9806292.
[19] M. Dine, N. Seiberg, Is the superstring weakly coupled?, Phys. Lett. 162B (1985) 299.
[20] W.A. Bardeen, C.T. Hill, M. Lindner, Minimal dynamical symmetry breaking of the standard
model, Phys. Rev. D 41 (1990) 1647.
[21] Y. Nambu, BCS mechanism, quasi supersymmetry, and fermion masses, in: Z. Ajduk et al.
(Eds.), Proc. of the XI Warsaw Symposium on Elementary Particle Physics, May, 1988,
World Scientific, 1989; in: Bando, Muta, Yamawaki (Eds.), Quasisupersymmetry, bootstrap
symmetry breaking and fermion masses, Proc. of the 1988 International Workshop on New
Trends in Strong Coupling Gauge Theories, Nagoya, Japan, World Scientific, 1989; Bootstrap
symmetry breaking in electroweak unification, EFI-89-08, 1989.
[22] V.A. Miransky, M. Tanabashi, K. Yamawaki, Mod. Phys. Lett. A 4 (1989) 1043; Phys. Lett.
B 221 (1989) 177.
[23] W.J. Marciano, Phys. Rev. Lett. 62 (1989) 2793.
[24] N. Arkani-Hamed, S. Dimopoulos, New origin for approximate symmetries from distant
breaking in extra dimensions, hep-ph/9811353.
[25] A.B. Kobakhidze, The top quark mass in the minimal top condensation model with extra
dimensions, hep-ph/9904203.
[26] C.G. Callan, J.A. Harvey, Anomalies and fermion zero modes on strings and domain walls,
Nucl. Phys. B 250 (1985) 427.
[27] N. Arkani-Hamed, M. Schmaltz, Hierarchies without symmetries from extra dimensions, Phys.
Rev. D 61 (2000) 033005, hep-ph/9903417.
[28] R.S. Chivukula, N. Evans, Triviality and the precision bound on the Higgs mass, Phys. Lett.
B 464 (1999) 244, hep-ph/9907414.
[29] J.A. Bagger, A.F. Falk, M. Swartz, Precision observables and electroweak theories, Phys. Rev.
Lett. 84 (2000) 1385, hep-ph/9908327.
[30] L. Hall, C. Kolda, Electroweak symmetry breaking and large extra dimensions, Phys. Lett. B 459
(1999) 213, hep-ph/9904236.
[31] R. Barbieri, A. Strumia, What is the limit on the Higgs mass?, Phys. Lett. B 462 (1999) 144,
hep-ph/9905281.
[32] T.G. Rizzo, J.D. Wells, Electroweak precision measurements and collider probes of the standard
model with large extra dimensions, Phys. Rev. D 61 (2000) 016007, hep-ph/9906234.
[33] B.A. Dobrescu, G. Landsberg, K.T. Matchev, Higgs boson decays to CP-odd scalars at the
tevatron and beyond, hep-ph/0005308.
[34] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, J. March-Russell, Neutrino masses from large
extra dimensions, hep-ph/9811448.
Local anomaly cancellation, M-theory orbifolds

and phase-transitions
Michael Faux a,∗ Dieter Lüst b , Burt A. Ovrut c
a Departments of Mathematics and Physics, Columbia University, 2990 Broadway, New York, NY 10027, USA
b Institut für Physik, Humboldt Universität, Invalidenstraße 110, 10115 Berlin, Germany
c Department of Physics, University of Pennsylvania, Philadelphia, PA 19104-6396, USA
Abstract
In this paper we consider orbifold compactifications of M-theory on S 1 /Z2 × T 4 /Z2 . We discuss
solutions of the local anomaly matching conditions by twisted vector, tensor and hypermultiplets
confined on the local orbifold six-planes. In addition we consider phase-transitions among different
solutions which are mediated by M-theory fivebranes which touch the local orbifold planes and are
converted there to gauge instantons.  2000 Elsevier Science B.V. All rights reserved.
PACS: 04.65.+e; 12.60.J; 11.25.-w; 11.10.Kk

Keywords: M-theory; Orbifolds; Strings; Anomalies
1. Introduction
M-theory harbors a broad spectrum of phenomena which can be systematically

probed by analyzing anomalies in effective quantum field theories. In the case of
orbifold compactifications of eleven-dimensional supergravity, a wide range of topological
restrictions can be resolved, and the states localized on orbifold planes determined, by
imposing factorization criteria on anomaly polynomials. The basic paradigm was espoused
in [1,2] by analyzing the S 1 /Z2 compactification. This was successfully applied to the
T 5 /Z2 compactification in [3,4]. In each of these cases, the orbifold planes comprise
isolated, non-intersecting submanifolds. In more general situations, the orbifold planes
can intersect, which gives rise to a number of novel features. In this paper, we describe
local anomaly cancellation on S 1 /Z2 × T 4 /ZM compactifications of M-theory, with
M = 2, 3, 4 or 6. These correspond to special points in the moduli space of S 1 /Z2 × K3
E-mail address: faux@math.columbia.edu (M. Faux).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 1 3 - 7
270 M. Faux et al. / Nuclear Physics B 589 (2000) 269–291
compactifications. In such situations, the orbifold planes do intersect. These issues were
first discussed in [5] and later in [6,7]. Here, we greatly extend the analysis of local anomaly
cancellation in orbifolds of this type, particularly emphasizing results pertaining to the
simplest of these cases, corresponding to M = 2.
In our previous work [5], we described the general features of S 1 /Z2 ×T 4 /ZM M-theory
orbifolds. In this paper, we expand those results, incorporating two subtle technical points
which were not addressed in complete generality in our previous work. The first is an issue
pertaining to the periodicity of the four-form G that has recently been more thoroughly
described in [8]. The second issue concerns the precise normalization of the CGG term
in the supergravity action. Each of these impinge numerically on our analyses, both in [5]
and in this current paper, by changing the overall coefficient of the anomaly inflow due
to the classical variation of the CGG term. We treat this coefficient as a parameter, to be
determined by consistency arguments, in a manner similar to the approach described in [9].
In order that M-fivebranes have unit magnetic charge, we choose a scale for the three-
form potential
R C such that, upon integration over dimensions transverse to a fivebrane, we
obtain dG = 1. This leaves only one coefficient in the basic Chern–Simons interactions
of the effective field theory unconstrained by supersymmetry and by the requirement
of fivebrane anomaly cancellation. However, this one parameter is uniquely fixed by
the additional requirement of consistent orbifold compactifications and gives rise to the
particular coefficient cited below for the CGG inflow anomaly. As a result of these two
described changes we now find a whole class of solutions of the local anomaly matching
conditions, which in general consist of vector, hyper and tensor multiplets confined on
the local orbifold six-planes and which also, for global consistency requirements, contain
fivebranes, free to move in the 11-dimensional bulk (see also Ref. [6] for discussion of
solutions of the local anomaly equations).
This situation is interesting since it is the simplest scenario which involves fivebrane-
mediated phase transitions in which the gauge group is nontrivially influenced by local
tensor couplings. Specifically, if a fivebrane hits one of the local orbifold six-planes it
will be described by a torsion free sheaf being equivalent to a small gauge instanton. Due
to the presence of this instanton the original gauge group at the local orbifold plane will
be broken to some subgroup, where also the number of tensor and hypermultiplets gets
changed. In this way elements of a certain class of orbifold compactifications are related to
each other by fivebrane-mediated phase transitions, where the associated magnetic charges
at the six-dimensional orbifold planes are changing by one unit. In addition we also
discuss the possibility of phase transitions with half-integer change of magnetic charge
at the local six-planes. We suggest that these phase transitions are due to fivebranes which
split at interconnecting ten-planes and are then transmuted to half-integrally charged gauge
instantons.
The orbifold compactifications which we discuss in this paper are of interest for other
reasons as well. For example, in a series of papers, the S 1 /Z2 orbifold was compactified
on smooth Calabi–Yau threefolds producing realistic “brane universe” theories of particle
physics and cosmology [10–14]. In future work, we will explore both the formal and
phenomenological aspects of different M-theory orbifold compactifications.
M. Faux et al. / Nuclear Physics B 589 (2000) 269–291 271
2. S 1 /Z2 × T 4 /ZM orbifolds
The S 1 /Z2 × T 4 /ZM orbifolds each involve a pair of ten-dimensional hyperplanes

fixed under the Z2 projection, which we denote by α, and a set of distinct seven-
dimensional hyperplanes fixed under the ZM projection, which we denote by β. Each of the
β-planes transversally intersects each of the α-planes once, at particular six-dimensional
hyperplanes invariant under both α and β. Chiral anomalies are induced on the α-planes
and (separately) on the αβ-planes, due to localized chiral projections of fields. Cancellation
of the ten-dimensional α-plane anomalies is uniquely accomplished by an additional ten-
dimensional E8 Yang–Mills supermultiplet on each of the two α-planes, as is well-known.
In this paper, we concern ourselves with the six-dimensional αβ-plane anomalies.
Although some of our discussion will be more general, it is helpful to have a specific
orbifold in mind to help visualize the basic geometric setting. Our prototype is the
simplest of the orbifolds described above, namely, the S 1 /Z2 × T 4 /Z2 orbifold. In this
case, spacetime has topology R6 × T 5 and each of the five compact coordinates takes
values on the interval [−π, π] with the endpoints identified. The nontrivial projections
are α : (x µ , x i , x 11 ) → (x µ , x i , −x 11 ) and β : (x µ , x i , x 11 ) → (x µ , −x i , x 11 ), where x µ
parameterizes the six noncompact dimensions, while x i and x 11 parameterize the T 4 and
S 1 factors, respectively. The element α leaves invariant the two ten-planes defined by
x 11 = 0 and x 11 = π , while β leaves invariant the sixteen seven-planes defined when the
four coordinates x i individually assume the values 0 or π . Finally, αβ leaves invariant the
thirty-two six-planes defined when all five compact coordinates individually assume the
values 0 or π . The αβ six-planes coincide with intersections of the α ten-planes with the
β seven-planes. The global structure is depicted in Fig. 1.
There are several magnetic and electric sources for G necessary to resolve chiral
anomalies in these orbifolds. The basic Chern–Simons terms include the CGG interaction
and also the higher-derivative GX7 interaction, where X8 ≡ dX7 is the eight-form
describing the worldvolume anomaly generated by the fivebrane zero modes. The
worldvolume anomaly is cancelled by inflow mediated by the GX7 interaction, provided
the fivebrane acts as a magnetic source for G, in the sense described above. Next, each
Fig. 1. The global structure of orbifold planes in the S 1 /Z2 × T 4 /Z2 orbifold. Horizontal
lines represent the two ten-dimensional α-planes, while vertical lines represent the sixteen
seven-dimensional β-planes. The thirty-two six-dimensional αβ-planes are represented by the solid
dots and coincide with the intersection of the α-planes and the β-planes. The in the figure indicates
a “wandering” fivebrane.
of the two α-planes supports a ten-dimensional E8 super-Yang–Mills multiplet. They also

provide magnetic sources for G due to the presence of terms δ (1)I4 in the dG bianchi
identity, where δ (1) is a one-form brane-current localized on the α-plane and I4 is a
four-form polynomial involving the Lorentz-valued curvature R and the local E8 field-
strength F .
The seven-dimensional
R β-planes provide electric sources for G via Chern–Simons
interactions δ (4) GY30 , where δ (4) is a four-form brane-current localized on the seven-
plane and Y4 = dY30 is a gauge-invariant four-form polynomial. This polynomial involves
the curvature R and also a field strength F associated with additional adjoint super-
gauge fields localized on the seven-plane (with the gauge group determined by anomaly
cancellation in a manner which we will describe). This coupling gives rise to an “I-
brane” effect via interplay with the ten-dimensional magnetic source (involving I4 ). This
contributes additional inflow localized on the six-dimensional intersection of the ten-
dimensional α-plane and the seven-dimensional β-plane. 1
The magnetic and electric sources described so far are encapsulated by the following
three polynomials

1 1 1
2 2
X8 (R) = tr R 4
− tr R ,
(2π)3 4! 8 32

1 1
I4 (R, F ) = − tr R 2
+ tr F 2
,
16π 2 2

1 1
Y4 (R, F) = − η tr R 2 + ρ tr F 2 . (1)
4π 32
The precise forms of X8 and I4 are fixed by fivebrane consistency and ten-dimensional
anomaly cancellation, respectively. The polynomial Y4 is parameterized by two rational
coefficients η and ρ. These are determined by further requirements described below.
Finally, the six-dimensional αβ-planes carry a magnetic charge. This appears in the dG
Bianchi identity as a term gδ (5) , where δ (5) is a five-form brane-current localized on the
αβ-plane and g is a rational magnetic charge subject to a quantization condition. For the
case M = 2, the magnetic charge g should be quarter-integer, as explained in [5]. The
Bianchi identity for the four-form field strength G is, therefore, given by
X
2 X
2f X
N5
dG = I4(i) δ (1)10 + gi δ (5)6 + δ (5)6 , (2)
Mi Mi Wi
i=1 i=1 i=1
where we have included all of the magnetic sources described above. The manifold Mi10
is the ith α-plane, while Mi6 is the ith αβ-plane and Wi6 is the worldvolume of the ith
fivebrane. 2 In most of this paper, our expressions apply to a particular αβ-plane. Hence,
the label i is implicit but omitted.
1 See Section 5 of [5] for a description of this effect.

2 The pervasive use of the label i is merely convenient, and does not imply any specific correlation between
these manifolds.
The projection β can independently break the ten-dimensional E8 gauge groups on the
αβ-planes to maximal subgroups. Since a ten-dimensional vector supermultiplet decom-
poses into one six-dimensional N = 1 vector and one six-dimensional hypermultiplet, the
breaking pattern will be characterized by an integer number VB of vector multiplets and
another integer number HB of hypermultiplets, each transforming according to some repre-
sentation R of the residual maximal subgroup of E8 . The projection β necessarily removes
half of the E8 degrees of freedom. But the identity of which half depends on how β acts
on the E8 root lattice.
Chiral projection of the supergravity fields results in further contributions to the local
αβ anomalies. These derive from “untwisted” fields comprising one universal N = 1
tensor multiplet and some number h of hypermultiplets. The value of h depends on which
ZM orbifold is being considered. For the cases M = 2 and M = 3, we have h = 4 and
h = 2, respectively. Furthermore, since the local anomaly due to the supergravity and
residual E8 fields arises from the coupling of fields which are not themselves localized
on the αβ-planes, it involve fractions which reflect the multiplicies of the fixed planes. We
parameterize this by another integer f corresponding to the number of β-planes associated
with the orbifold in question. For the cases M = 2 and M = 3, we have f = 16 and f = 9,
respectively. These correspond to the respective sixteen and nine fixed-planes in the Z2
and Z3 orbifold limits of the K3 manifold. 3
Finally, we allow for as yet unspecified N = 1 supermatter localized on each αβ-plane.
We call this matter “twisted”, since it is analogous to twisted sector matter in superstring
orbifolds. This matter assembles into nT tensor multiplets, nV vector multiplets and nH
hypermultiplets, and involves an as yet undetermined “twisted” gauge group G. e The vector
multiplets transform in the adjoint representation, while the hypermultiplets transform in
an unspecified representation R.e
We focus on a particular αβ-plane, and assemble each of the contributions to the
six-dimensional anomaly localized on this plane. There are three classical (inflow)
contributions, described by the polynomials
I8 (CGG) = −πgI4 (R, F )2 ,
I8 (GX7 ) = −gX8 (R),
I8 (I B) = −I4 (R, F ) ∧ Y4 (R, F). (3)
The first two arise
R from the variation of the CGG and GX7 terms. The third arises from
the variation of δ GY30 and describes the “I-brane” anomaly. 4 There are also three
(4)
quantum contributions
1 (3/2) (1/2)
I8 (SG) = IGRAV (R) − (1 + h)IGRAV (R) ,
2f
1 (1/2) (1/2) (1/2)
I8 (G) = (VB − HB )IGRAV (R) + IMIXED (R, F )R + IGAUGE(F )R ,
f
3 The Z and Z orbifolds involve additional subtlety which we will not discuss in this paper.
4 6
4 The necessity for the “I-brane” contribution in an orbifold context was first recognized in [5], and was inspired
by an analogous effect on intersecting D-branes, introduced in [16].
e = (nV − nH − nT )I (1/2) (R) − nT I (3-form) (R)

I8 (G) GRAV GRAV
(1/2) (1/2)
+ IMIXED (R, F)R
e + IGAUGE (F)R
e. (4)
The factors IGRAV , IMIXED and IGAUGE which appear in the quantum anomalies describe
one-loop gravitational, mixed and pure-gauge anomalies. They are attributable to the type
of chiral fields with spin indicated by the superscripts. These are determined by index
theorems and are listed explicitly in Appendix C of [5]. The anomaly I8 (G) describes the
local anomaly involving whatever subgroup G ⊂ E8 is left unbroken by β on the relevant
αβ-plane. The subscript R indicates the representation content of the E8 residual subgroup.
(1/2) (1/2)
Hence, the traces over the gauge factors in IMIXED (R, F )R and IGAUGE(F )R are traces
over R. Similar comments apply to the αβ twisted sector with gauge group G, e where the
e
representation content is indicated by the subscript R. 5
There are numerous unspecified parameters involved in the six contributions in (3)
and (4). To begin with, there are ten integers characterizing the global geometry of the
orbifold, the local E8 breaking pattern and multiplicities in the twisted and untwisted
spectra. For instance, the orbifold geometry is partially encoded in the number f of β-fixed
planes and the number h of universal untwisted hypermultiplets. The magnetic charge g
is an integer times a basic quantization unit (which also reflects the orbifold geometry).
The two parameters η and ρ describe electrical charges of the seven-planes, as defined in
Eq. (1). The E8 breaking pattern is encoded in the multiplicities VB and HB . Finally, the
local twisted spectrum is encoded in the multiplicities nV , nH and nT .
Furthermore, there are six sets of rational parameters which characterize the representa-
tion content of the matter fields. These are given by “representation indices”, which allow
one to relate traces over a given representation to traces over the fundamental representa-
tion. The relevant representation indices are denoted I2 (R), I2,2 (R) and I4 (R), and are
defined by
trR F 2 = I2 (R) tr F 2 ,
trR F 4 = I2,2 (R)(tr F 2 )2 + I4 (R) tr F 4 , (5)
where all traces on the right-hand side are over the fundamental representation. Some
useful representation indices are listed in Table 1. 6
Remarkably, each and every one of the above parameters can be resolved, in a sense to
be made clear, by imposing necessary factorization properties on the net anomaly obtained
by summing the six contributions listed in (3) and (4). That result describes the total local
anomaly on a particular αβ-plane due to the sources which we have described so far. Since
the total local anomaly must vanish, there are two possibilities. The first is that the net
anomaly vanishes identically. The remaining possibility is that this net anomaly is non-
5 The technical aspects involved in determining (3) and (4) are described in [5]. The determination of I (CGG)
8
is more subtle than indicated in that paper, however, for reasons mentioned previously. See [8] for a more thorough
description of this one term. Note that the twisted gauge group G e can include factors which coincide with some
factors in the E8 residual group G.
6 The evaluation of the representation indices for arbitrary representations of the classical Lie groups is a
complicated problem. See [15] for a discussion of this issue.
Table 1
Some useful representation indices
G R I4 (R) I2,2 (R) I2 (R)
E8 248 0 9 30
E7 133 0 1/6 3
56 0 1/24 1
SO(16) 120 8 3 14
128 −8 6 16
16 1 0 1
SU(2) 3 0 8 4
2 0 1/2 1
zero, but is cancelled by a contribution arising from special couplings of local tensor fields
which, thereby, provide a local Green–Schwarz mechanism.
Local tensor fields reside in twisted N = 1 tensor multiplets on the αβ-plane. The two-
form fields in such a multiplet are anti-self-dual, in the sense that the associated gauge-
invariant three-form field-strength satisfies H = − ∗ H . As a result of this, a special Chern–
R
Simons interaction δ (5) H Z30 , where δ (5) is the five-form brane-current with support on
the αβ-plane and Z4 = dZ30 is a gauge-invariant four-form polynomial, will lead to two
consequences. First, due to the anti-self-duality, the electric coupling described by the
Chern–Simons interaction implies a magnetic coupling described by the Bianchi identity
dH = Z4 . The magnetic coupling involves the same polynomial as the electric coupling
because anti-self-duality is equivalent to an electric-magnetic duality. 7 Secondly, the
Chern–Simons coupling generates an inflow anomaly described by I8 (GS) = Z4 ∧ Z4 ,
where GS designates Green–Schwarz. Note that this polynomial is a perfect square due
to the duality. This is, in turn, a consequence of N = 1 supersymmetry. Thus, the net
anomaly can be cancelled by specialized local tensor dynamics, provided that the net
anomaly reduces to a sum of perfect squares with one term for each available tensor
field.
3. Local anomaly cancellation
As a result of the above discussion, a program for analyzing the local anomaly on a
given αβ-plane becomes apparent. First, we assemble the net anomaly by summing the six
terms in (3) and (4). We will call this result I8 . Then we sequence through the possibilities
nT = 0, 1, 2, . . . , in each case imposing that I8 satisfies the appropriate factorization
requirement. For the case nT = 0, since there are no local tensors, we require I8 = 0.
For the case nT = 1, we impose that I8 is proportional to a complete square, that is,
I8 ∝ (Z4 )2 . (In this second case, we simultaneously determine the form of the electric
7 This is easy to see by taking the exterior derivative of the Euler–Lagrange equation ∗H = −Z , and then
3
replacing ∗H with −H .
and magnetic couplings of the local tensor.) For the case nT = 2, we require that I8 is
the sum of two perfect squares. And so forth. These factorization requirements prove to
be marvelously restrictive. For each choice of nT , there results unambiguous values for
each and every one of the previously unspecified geometric and topological parameters,
including the values of the magnetic charge, electric charges, and the identity of the gauge
groups and representation content of the twisted sectors.
Since M-fivebranes carry unit magnetic charge, as well as the zero mode fields described
previously, we infer that a fivebrane moving onto (or off of) a given αβ-plane deposits
(or removes) charge and twisted modes in the process of doing so. Since the fivebrane
carries one N = 1 tensor multiplet and one hypermultiplet, we expect that the solutions
to our factorization constraints will assemble into hierarchies linked by incrementing the
magnetic charge and the local tensor and hyper multiplicities as g → g + 1, nT → nT + 1
and nH → nH + 1. This is precisely what we find. Mathematically, this can be understood
as follows. A single fivebrane touching an orbifold fixed plane is described by a singular
object called a “torsion free sheaf” [17]. This sheaf carries one extra unit of magnetic
charge and has one tensor multiplet and one hypermultiplet as zero modes, identical to the
analogous fivebrane data. This accounts for the increase in each of these quantities by unity
when the fivebrane is moved to a fixed plane.
At least in some cases, the torsion free sheaf can be shown to be the singular “small
instanton” limit of a smooth gauge instanton [18]. Smoothing out the sheaf into an
instanton represents a true phase transition, where the fivebrane data disappears and is
replaced by a vector bundle [10–14]. In this process, the unit magnetic charge of the
sheaf is replaced by a unit increase in the second Chern number of the vector bundle.
The zero modes of the vector bundle are, in general, quite different from those of
the torsion free sheaf. Most importantly, the appearance of a smooth, nontrivial vector
bundle signals the breakdown of the original twisted sector gauge group to a smaller
group G ⊂ E8 . Therefore, after this phase transition, we expect to have a smaller twisted
sector gauge group with identical topological charge but different numbers of tensor and
hypermultiplets. Be that as it may, this theory remains anomaly free. As we will see, locally
anomaly free orbifold planes do exist that could be related to each other through small
instanton phase transitions. Thus, factorization of the local anomaly polynomial yields an
extra bonus.
In addition to the torsion free sheaf transitions described earlier, we expect another
interesting grouping of our factorization solutions. In this second grouping, we expect
small instanton transitions between sets of solutions with identical local magnetic charge,
but with different numbers of zero modes and different local gauge groups.
For clarity, let us recapitulate the fivebrane transitions discussed in the previous two
paragraphs. If a fivebrane moves to a particular αβ-plane, it should increment the local
magnetic charge by one, and add one local tensor and one local hypermultiplet to the
associated twisted spectrum. We would then attribute one unit of the total local magnetic
charge to the latent magnetic charge of the fivebrane, now interpreted as a torsion-free
sheaf. The associated tensor would be available to help mediate anomaly cancellation
through a local Green–Schwarz mechanism. Now, assume this configuration is a small
instanton and can be deformed to a smooth vector bundle. Then, since the instanton does
not have a tensorial zero mode, there would be one less tensor available to mediate the
anomaly cancellation. The anomaly polynomial should, therefore, reconfigure so as to
ensure continued anomaly cancellation, but with a modified factorization criterion. Thus,
by classifying independent solutions to the factorization requirements into sets involving
identical values of g but different numbers of tensor fields, one can infer such nontrivial
phase transitions.
We begin our analysis by considering a particular αβ plane, corresponding to one of the
solid dots in Fig. 1, representing the unique six-dimensional intersection of a particular ten-
dimensional α-plane and a particular seven-dimensional β-plane. To be more concrete, we
focus on one of the intersection points on the lower of the two α-planes in Fig. 1, so that the
local geometry is depicted as in Fig. 2. In Fig. 2, the horizontal line represents the α-plane,
the vertical line represents the β-plane and, finally, the point of intersection represents the
αβ-plane. The intersection supports the local anomaly in which we are interested and has
magnetic charge g. The α-plane supports E8 super Yang–Mills fields, as described above.
This E8 group is, in general, broken on the αβ-plane to some subgroup depending of the
action of β on the E8 root lattice. In the vertex-diagrams below, we indicate, to the right
of the horizontal lines, the subgroup of E8 left unbroken at the intersection. Thus, Fig. 2
indicates a scenario in which the full E8 is left unbroken. This figure also indicates the
presence of additional gauge structure with group G e7 localized on the β-plane and further
e
gauge structure on the αβ-plane with group G6 . These correspond to additional seven-
dimensional and six-dimensional fields, respectively.
The only multiplet in seven dimensions is a vector multiplet, which transforms in the
adjoint of Ge7 . This decomposes into one six-dimensional N = 1 vector multiplet and one
six-dimensional hypermultiplet. 8 An important fact is that the seven-dimensional fields
are chirally projected by α onto the embedded six-dimensional αβ-plane. It, thereby,
contributes to the local αβ anomaly.
e7 does not coincide with some factor of the broken subgroup of E8 , then six-
If G
dimensional gauge invariance dictates that the hypermultiplet is the part projected out by α.
Thus, the Ge7 gauge fields remain at the intersection to enforce local G e7 invariance. The Ge7
adjoint gauginos fill out a six-dimensional N = 1 vector multiplet, which is indicated by
the V next to the downward arrow in Fig. 2. This tells us that the “vector” part survives the
α projection as we move down the vertical line in that diagram and land on the intersection
Fig. 2. The local geometry near a particular αβ-plane, showing some of the data to be resolved by
anomaly factorization.
8 As a point of interest, this is the same decomposition enjoyed by a ten-dimensional vector multiplet.
point. Since the N = 1 vector multiplet is chiral, the G e7 gauginos will contribute to the
αβ anomaly. However, since this anomaly results from the coupling of fields not localized
on the αβ-plane, this anomaly must be divided by two (since there are two αβ-planes
embedded within a given β-plane) compared to a similar anomaly due to six-dimensional
e7 adjoint gauginos.
G
e7 coincides with a factor in the unbroken subgroup of E8 , then the G
If G e7 gauge fields
on the αβ-plane may be supplied by the ten-dimensional gauge fields which survive
the β projection. Consequently, the projection α should remove the “vector” part of
the seven-dimensional adjoint matter, so that the other part, corresponding to an adjoint
hypermultiplet, survives the α projection. We indicate this alternate situation by an H next
to the downward arrow in the corresponding diagram. In this case, the surviving hyperinos
would contribute to the local anomaly. This anomaly would include a division by two
compared to a similar anomaly due to six-dimensional G e7 adjoint hyperinos, for reasons
identical to those described in the preceeding paragraph. 9
A third possibility is that the group G e7 is broken, by β, to some maximal subgroup
e
H ⊂ G7 on the αβ-plane. In this case, the seven-dimensional fields would decompose into
various six-dimensional fields transforming according to representations determined by the
appropriate branching rule. These would include fields transforming in the adjoint of H and
other fields transforming in other representations of H. The vector part of the adjoint fields
would survive the β projection while the hypermultiplet part of the remaining fields would
survive. We indicate this hybrid situation by replacing the V in Fig. 2 with the relevant
subgroup H ⊂ G e7 which survives the projection.
In resolving the factorization criteria necessary to explain local anomaly cancellation
on a given αβ-plane, one requires factors of two which can only be explained by seven-
dimensional matter in the manner we have just described. Thus, the identity of seven-
dimensional matter is indicated by anomaly cancellation on an embedded sub-plane. This
is interesting because the seven-plane itself cannot support a local anomaly since it is odd-
dimensional. (The situation is analogous to the fact that eleven-dimensional supergravity
is needed by the E8 super-gauge multiplets on the two α-planes to render those ten-planes
anomaly-free.)
There is one more subtle factor of one-half which needs explanation. This relates to
hypermultiplets in the six-dimensional twisted sector. It is possible for r hypermultiplets
to transform according to a 2r-dimensional representation R e of the gauge group G, e
provided the representation is “pseudoreal” in a sense to be clarified. In this case, the
4r scalar fields assemble into r quaternions represented as φiα , where i = 1, 2 is an
index which spans the 2 represention of an Sp(1) automorphism of the supersymmetry
e The group G
algebra and α = 1, . . . , 2r spans R. e acts as δφ α = θ a (Ta )α , where (Ta )α are
i β β
antihermitian generators. There exists a real invariant tensor ρ αβ = ραβ which, by suitable
field redefinition, can be put into block-diagonal form ρ = diag(iσ2 · · · iσ2 ), where σ2 is the
second Pauli matrix. The representation is pseudoreal if (Ta∗ ) = −ρTa ρ. See [19,20] for a
9 The possibility of alternatively projecting out the vector or hypermultiplet parts of the seven-dimensional
fields on the six-dimensional planes was also mentioned in [6].
more comprehensive discussion of hypermultiplets. In this case, we refer more properly to

2r half-hypermultiplets, since the number of hypermultiplets is half the dimensionality of
the representation. Each such half-hypermultiplet then contributes one-half of the anomaly
which we would normally attribute to 2r antichiral spinors transforming in R e via naive
application of index theorems.
Next, we should explain how to algebraically characterize the possible branching
patterns describing the projection of the two E8 factors by β. The simplest possibility
is the one indicated in Fig. 2, where the relevant E8 factor remains unbroken. In terms of
six-dimensional N = 1 multiplets, the ten-dimensional E8 vector multiplet decomposes
into one vector multiplet and one hypermultiplet. In this case, it is the hypermultiplet
components which are projected out by β on the αβ-plane. This leaves us with the gauge
fields necessary to enforce local E8 invariance on the αβ-plane. This also fixes two of
our parameters, (VB , HB ) = (248, 0). In this case, all E8 traces which appear in both the
inflow anomalies (3) and the quantum anomalies (4) are each taken at face value. Thus,
using the representation indices in Table 1, we can use the results Trace248 F 2 ≡ 30 tr F 2
and Trace248 F 4 = 9(tr F 2 )2 to express the traces which appear in the quantum anomalies
(4) in terms of the fundamental (tr) traces. 10
More generally, the E8 factors will be broken by β, on the αβ-planes, to some maximal
subgroup with a branching pattern which can be found from the tables in [21]. In this case,
when we determine our anomaly polynomial I8 by adding up the six contributions in (3)
and (4), we replace the various E8 traces by traces over the relevant representations of the
residual subgroup. We then relate these to traces over fundamental representations of the
factors in this subgroup by using representation indices, such as those listed in Table 1.
However, there is a subtle difference between the inflow contributions (3) and the quantum
contributions (4) which should be properly accounted for, and which we now describe.
Since the inflow anomalies are classical expressions, we can apply the group theoretic
reduction directly on the traces which appear in (3). However, in the general case, the six-
dimensional quantum anomalies derive from both chiral and antichiral fields. The chiral
fields, which satisfy Γ7 ψ = ψ, appear in the six-dimensional N = 1 vector multiplets.
The antichiral fields, which satisfy Γ7 ψ = −ψ, occur in hypermultiplets. Since chiral and
antichiral fields contribute one-loop anomalies with opposite sign, there is an extra minus
sign associated with all quantum anomalies arising from hypermultiplet couplings. We
illustrate this with two explicit examples.
As a first example, we choose the breaking pattern E8 → E7 × SU(2). In this case,
we have the branching rule 248 = (133, 1) ⊕ (1, 3) ⊕ (56, 2). We determine that the
surviving six-dimensional fields comprise 133 N = 1 vector multiplets transforming as
the adjoint of E7 , another three vector multiplets transforming as the adjoint of SU (2) and
112 hypermultiplets transforming as a bifundamental representation. Thus, (VB , HB ) =
(136, 112). In this case, we reduce the E8 traces which occur in the inflow anomaly as
follows
10 Note that the traces which appear in the inflow anomaly (3), through the implicit dependence of (1), are
fundamental traces to begin with.
tr F 2 = 1
30 Tr248 F
2

= 30 Tr133 Fa + 2 tr56 Fa
1 2 2
+ Tr3 Fb2 + 56 tr2 Fb2
= 6 tr Fa + 2 tr Fb ,
1 2 2
(6)
where we have used the indices in Table 1. In (6), the subscripts a and b denote E7 and
SU(2), respectively. In the final line, we have dropped the labels from the fundamental 56
and 2 traces. Thus, to describe the inflow anomaly in the case where the E8 factor is broken
by β to E7 × SU(2), we substitute the identity (6) for the tr F 2 in the inflow anomaly (3). In
the quantum anomaly, on the other hand, hypermultiplets and vector multiplets contribute
with opposite signs. As a result, when computing the local one-loop anomaly in the case
where E8 → E7 × SU(2), we we should replace the factor trace F 2 which appears in
(1/2)
IMIXED using the following 11
trace F 2 = Tr248 F 2
= Tr133 Fa2 − 2 tr56 Fa2 + Tr3 Fb2 − 56 tr2 Fb2
= tr Fa2 − 52 tr Fb2 . (7)
This derivation differs from (6) by the minus signs on terms relating to hypermultiplet
couplings. As described above, these minus signs reflect the antichirality of hyperinos.
(1/2)
Similar comments apply to the term trace F 4 which appears in IGAUGE. In total, to describe
the case E8 → E7 × SU(2) we should make the following replacements
tr F 2 = 16 tr Fa2 + 2 tr Fb2 ,
trace F 2 = tr Fa2 − 52 tr Fb2 ,
trace F 4 = 1
12 (tr Fa )
2
− 20(tr Fb )2 − 6 tr Fa2 ∧ tr Fb2 . (8)
The first of these should be substituted in the classical (inflow) anomaly, while the
second two should be substituted in the quantum anomaly. Note our mnemonic that traces
which appear in the quantum anomaly are designated “trace”, whereas classical traces are
abbreviated “tr”.
As a second example, we choose the breaking pattern E8 → SO(16). In this case,
we have the branching rule 248 → 120 ⊕ 128. We determine that the surviving six-
dimensional fields comprise 120 N = 1 vector multiplets transforming as the adjoint and
128 hypermultiplets transforming as the spinor of SO(16). Thus, (VB , HB ) = (120, 128).
In this case, we reduce the E8 traces which occur in the inflow anomaly and in the quantum
anomaly in a manner similar to that described in our previous example, making use of the
representation indices in Table 1. The appropriate reductions are
tr F 2 = tr Fa2 ,
trace F 2 = −2 tr Fa2 ,
trace F 4 = −3(tr Fa )2 + 16 tr Fa4 , (9)
11 See Eq. (C.2) of [5] for the explicit polynomial corresponding to I (1/2) , as well as all of the other quantum
MIXED
anomaly polynomials referred to in this paper.
where the subscript a now denotes SO(16). Once again, the first of these should be
substituted in the classical (inflow) anomaly, while the second two should be substituted in
the quantum anomaly.
Using the three distinct cases which we have so far addressed, corresponding to the
choices where β breaks E8 to E8 , E7 × SU(2) or SO(16), we have enough data to
completely determine an interesting set of solutions to our anomaly factorization problem.
These solutions conform to our expectations by assembling into hierarchies as described
previously.
On a given αβ-plane, such as that depicted by the intersection point in Fig. 1, the net
six-dimensional anomaly I8 is determined by adding up all six terms in (3) and (4). One
then substitutes identities, such as (8) and (9), relevant to the particular E8 breaking pattern
being considered, in the manner explained above. What results is a polynomial with terms
proportional to the each of tr R 4 , (tr R 2 )2 , tr R 2 ∧ tr F 2 , tr R 2 ∧ tr F 2 , (tr F 2 )2 , tr F 2 ∧ tr F 2 ,
tr F 4 , and tr F 4 , where F stands generically for factors in the residual group G ⊂ E8
which survives the β projection (we have denoted these Fa and Fb above) and F stands
generically for factors in any gauge group associated with twisted matter.
Note that we have allowed for twisted fields which are either six- or seven-dimensional.
In the former case, we refer to N = 1 fields living exclusively on the αβ-plane
under consideration. In the latter, we refer to vector adjoint supermultiplets living on
the seven-dimensional β-plane which intersects this αβ-plane. The seven-dimensional
fields will contribute to the anomaly with a tell-tale factor of two, as described above.
Computationally, we accommodate both of these cases simultaneously by formally
allowing nT tensors, nV vectors and nH hypermultiplets, where nV and nH can assume
half-integral values. The (formal) appearance of half-integer numbers of multiplets then
indicates that the associated matter is seven-dimensional.
Keeping the twisted matter arbitrary, we determine I8 by adding up all six contributions
in (3) and (4). For any choice of G ⊂ E8 , we reduce the various E8 traces to G traces
according to the scheme described above. This process provides us with a provisional form
of the local anomaly. It remains provisional since the twisted contribution remains to be
resolved. Nevertheless, we can extract our first bits of useful information. Since anomaly
cancellation is possible only if I8 either vanishes identically or reduces to a sum of perfect
squares, it follows that any nonfactorizable terms in I8 must vanish. The vanishing of the
tr R 4 term requires
1 1
nH − nV = 30g − 29nT + (244 − h) + (VB − HB ). (10)
2f f
This constraint is the local version of the global constraint NH − NV + 29NT = 273, where
NH , NV and NT are the total number of hyper, vector and tensor multiplets in the entire
orbifold, including all twisted and untwisted contributions. The relationship between the
local constraint (10) and the global version is described in [5]. Note that (10) is invariant
when g → g + 1, nT → nT + 1, and nH → nH + 1, consistent with expectations described
previously.
Henceforth, we concentrate on the S 1 /Z2 × T 4 /Z2 orbifold. In this case, we have
(f, h) = (16, 4), as described above, so that (10) becomes
15 1
nH − nV = 30g − 29nT + + (VB − HB ). (11)
2 16
Since the left-hand side must be either an integer or a half-integer, it follows that (VB −
HB )/16 must be integer or half-integer as well, since in this case g is quantized in
quarter-integer units. Each of the three E8 breaking patterns which we have addressed,
G = E8 , E7 × SU(2) and SO(16), corresponding to (VB , HB ) = (248, 0), (136, 112) and
(120, 128), respectively, respect this constraint. We restrict our study to these three
breaking patterns.
The systematic analysis of local anomaly cancellation proceeds as outlined above. We
first seek solutions with no twisted tensor multiplets, so that nT = 0. In this case, I8 must
vanish identically. We consider each of the three possible E8 breaking patterns described
above, using the relevant values of VB and HB in each case. For these three possibilities,
when nT = 0 Eq. (11) reduces to the constraints indicated in Table 2. Thus, if there are
no local twisted tensor multiplets, anomaly cancellation implies the indicated correlations
between the local magnetic charge and the multiplicities of twisted hyper and vector
multiplets. Note that, in each case, extra twisted matter is required since the indicated
multiplicities cannot be made to vanish with any properly quantized choice of g. The
challenge in the case nT = 0 is not only to identify the multiplicities of twisted states,
but also to identify the twisted gauge groups, the representation content of the twisted
matter, as well as values of g, η and ρ which satisfy the restrictions in Table 2 (which
ensures cancellation of the local tr R 4 anomaly). Furthermore, these choices must provide
for complete cancellation of all other terms in the full polynomial I8 . Satisfying all of these
requirements is a highly restrictive demand.
We first look for a “basic” solution where G = E8 . In this case, we find a unique solution
to all of our constraints. This solution requires g = −3/4 and G e7 = SU(2), which is broken
as SU(2) → U (1) on the αβ-plane. In this case, the three adjoint SU(2) fields provide
one six-dimensional vector multiplet and two hypermultiplets on the αβ plane. There are
no further twisted fields. Thus, the gauge structure on the αβ-plane is E8 × U (1), under
which the twisted fields transform as follows
Vectors: 1
2 1(0) , nV = 1/2,
2 1(+1) ⊕ 2 1(−1) , nH = 1/2 + 1/2 = 1,

1 1
Hypers:
Tensors: none, nT = 0,
Table 2
Constraints linking the local magnetic charge and the
multiplicities of twisted hyper and vector multiplets
when nT = 0 for three possible E8 breaking patterns
G nH − nV
E8 30g + 23
E7 × SU(2) 30g + 9
SO(16) 30g + 7
Fig. 3. Vertex diagram corresponding to the “basic” solution in the S 1 /Z2 × T 4 /Z2 orbifold, as
described in the text.
where the U (1) charges are indicated in the parenthetical subscripts. This solution also
requires (η, ρ) = (1, 0). 12 In this case, the anomaly vanishes completely, as it should
since there are no tensors to provide a local Green–Schwarz mechanism. The factor of
one-half accompanying the twisted field representions indicate that these describe seven-
dimensional fields living on the β-plane, contributing via a chiral projection onto the
embedded αβ-plane. We represent this local solution with the diagram shown in Fig. 3. In
Fig. 3, the twisted gauge group SU(2) is indicated next to the vertical line, signifying that
this corresponds to Ge7 . The magnetic charge associated with the intersection is indicated
by the −3/4. There is no local U (1) anomaly because the charges cancel, leaving a net
U (1) charge of zero.
Our basic solution has a two notable aspects. First, using the multiplicities listed above,
we compute nH − nV = 1/2, which is precisely the value specified in Table 2 for the case
of unbroken E8 and for the choice g = −3/4. The second notable aspect concerns the
magnetic charge. For reasons described in [5], we attribute a topological significance to
this number. Specifically, g corresponds to an E8 instanton number (associated with an
instanton residing on the αβ-plane) minus the local contribution due to the nontrivial Euler
character of the K3 manifold. (We attribute the second of these to a local gravitational
instanton, associated with a pointlike version of the ALE space needed to blow up
the orbifold.) Since the Euler number of K3 is 24, we divide this evenly over the 32
αβ-planes, so that the local gravitational contribution to g should be exactly −3/4.
Since we are considering the case of unbroken E8 , we assume there are no local gauge
instantons. Therefore, the only contribution to g should be the gravitational result of
−3/4. It is gratifying that this number is required by our independent anomaly cancellation
requirements.
When we impose nT = 0 and I8 = 0 on the cases where β breaks E8 to G = E7 × SU(2)
or G = SO(16), we also find unique solutions with specialized values of g and with specific
twisted matter content. These solutions are described by the diagrams in the left-hand
column of Fig. 4. Note that, in each case, we require seven-dimensional SU(2) vector
multiplets. In the G = E7 × SU(2) case, the seven-dimensional SU(2) factor is identified
with the SU(2) factor in G. This identification is indicated by the asterix on the two SU(2)
factors in the relevant diagram. As described previously, under these circumstances, α
projects out the “vector” component of the seven-dimensional matter but preserves the
“hyper” component. This is represented by the H next to the arrow in the same diagram.
In contrast to our “basic” solution, both of these new solutions involving local breakdown
12 The requirement that ρ = 0 and the need for U (1) gauge factors in the “basic” solution to S 1 /Z × T 4 /Z
2 2
orbifolds was also discussed in [6].
Fig. 4. A collection of consistent orbifold vertices for the S 1 /Z2 × T 4 /Z2 orbifold for three possible
E8 breaking patterns and for the two possibilities nT = 0 and nT = 1. In the middle two diagrams,
the asterix signifies that the indicated SU(2) factors are identified.
of E8 require (η, ρ) = (1, 1). The nonvanishing of ρ indicates that the β-planes support
nonzero SU(2) electric charge in these cases. The physics of this observation may prove
interesting.
We interpret the β-induced E8 symmetry breakdown as a reflection of instantons
residing on the αβ-plane. Since there are two classes of gauge instantons which
could reside on the αβ-planes, one contributing integer magnetic charge and the other
contributing half-integer magnetic charge, we infer that solutions could exist with g taking
values at half-integer increments greater than the “basic” value of −3/4. This is precisely
what we find; the nT = 0 solutions for G = E7 × SU(2) and G = SO(16) require g = −1/4
and g = +1/4, corresponding to half-integer and integer valued instantons, respectively.
In the next phase of our systematic search for local anomaly-free vertices, we study the
cases with nT = 1. In these cases, we impose that the anomaly factorizes as a complete
square, I8 ∝ (Z4 )2 , so that it can be cancelled by dynamics involving the self-dual tensor
in the local twisted spectrum. Eq. (11), which enforces the vanishing of the tr R 4 term
in I8 , is still important since tr R 4 cannot factorize. Once again, we consider each of the
three E8 breaking patterns discussed above, using the relevant values of VB and HB in
each case. For these three possibilities, when nT = 1, Eq. (11) reduces to the constraints
indicated in Table 3. Thus, if there is one local twisted tensor multiplets then local anomaly
factorization implies the indicated correlations between the local magnetic charge and the
multiplicities of twisted hyper and vector multiplets. Note that, in each case, extra twisted
matter is still required since the indicated multiplicities cannot be made to vanish with any
Table 3
Constraints linking the local magnetic charge and the
multiplicities of twisted hyper and vector multiplets
when nT = 1 for three possible E8 breaking patterns
G nH − nV
E8 30g − 6
E7 × SU(2) 30g − 20
SO(16) 30g − 22
properly quantized choice of g. The challenge, in the case nT = 1, is not only to identify
multiplicities of twisted states, but also the twisted gauge groups, the represention content
of the twisted fields, as well as values of g, η and ρ which can satisfy the restrictions in
Table 3 (which ensures cancellation of the local tr R 4 anomaly). Furthermore, they must
provide the appropriate factorization of I8 . Satisfying these requirements is, again, a highly
restrictive demand.
As before, we start with the case where E8 remains unbroken. We again find a unique
solution to our constraints, but this time with g = +1/4. Once again G e7 = SU(2), and the
seven dimensional gauge group is broken by β on the αβ-plane as SU(2) → U (1). The
only change to the twisted spectrum in the analogous nT = 0 solution is the addition of
one singlet hypermultiplet to the local six-dimensional spectrum. Thus, the twisted fields
transform under E8 × U (1) as follows
Vectors: 1
2 1(0) , nV = 1/2,
2 1(+1) ⊕ 2 1(−1) ⊕ 1(0), nH = 1/2 + 1/2 + 1 = 2,

1 1
Hypers:
Tensors: 1(0), nT = 0,
where the U (1) charges are again indicated in the parenthetical subscripts. This solution
also requires (η, ρ) = (1, 0) The factors of one-half indicate that these fields describe
a projection of a seven-dimensional multiplet, living on the β-fixed plane, via a chiral
projection onto its boundary. This solution is shown in the upper right-hand diagram in
Fig. 4. In our diagrams, we indicate the presence of a twisted tensor multiplet by an . In
this case, the anomaly does not vanish, but is given by
1 3 2
I8 = − 3
tr R 2 − 2 tr F 2 . (12)
(2π) 4 16
Since this is proportional to a perfect square, it can be removed by a local Green–Schwarz
mechanism mediated by the anti-self-dual tensor in the twisted tensor multiplet.
When we impose nT = 1 and I8 ∝ (Z4 )2 on the cases where β breaks E8 to G =
E7 × SU(2) or G = SO(16), we also find unique solutions with specialized values of g
and specific twisted matter content. The set of nT = 1 solutions corresponding to each
of the our three choices for G are described by the diagrams in the right-hand column of
Fig. 4. In each of these cases, the presence of a twisted tensor multiplet is indicated by
the on the relevant vertex. In these cases, the anomaly I8 does not vanish but, rather, is
Fig. 5. A two step process in which a fivebrane which has moved to a vertex smoothly deforms to an
integrally-charged instanton.
given by a complete square. Therefore, the twisted tensor involves interesting dynamics.
Notably, the cases where E8 is broken require ρ = 1. This is in contrast to the situation
involving unbroken E8 , where ρ = 0.
It is useful to compare the top right vertex-diagram in Fig. 4 with the lower left vertex-
diagram in that same figure. Since these two vertices have identical magnetic charge,
we infer a transition whereby the fivebrane connects smoothly to an instanton, locally
breaking E8 to SO(16). A subtle point concerns the electric charge of the associated
seven-dimensional β-plane (the vertical line in these diagrams). In the unbroken (E8 )
phase we require ρ = 0, so the seven-plane is not electrically charged. However, in the
broken (E7 × SU(2) or SO(16)) phases we require ρ = 1. This process is shown by the
two-diagram sequence depicted in Fig. 5.
4. The global structure
Now that we have tabulated some consistent local solutions, as classified in Fig. 4, we
can attempt to assemble these into a coherent global orbifold. There are extra constraints
on this procedure, however, which need to be taken into account. The first of these is
implied by the exactness of dG and follows from integrating the Bianchi identity (2) over
the five compact dimensions. Since this region has no boundary, the left-hand side of the
integrated version of (2) vanishes due to Stokes theorem since the integrand is a total
derivative. This implies that the net magnetic charge of the entire orbifold is zero. Without
loss of generality, we can concentrate all of the magnetic sources either on the αβ-planes
or on fivebrane worldvolumes. We therefore determine that
X
32
N5 + gi = 0, (13)
i=1
where N5 is the number of fivebranes not residing on αβ-planes. Note that N5 is necessarily
a positive integer.
There is a unique “basic” global configuration satisfying this constraint for which
neither of the E8 factors is broken at any intersection. In this case, each αβ-plane carries
a magnetic charge of −3/4. Since there are thirty-two αβ-planes, the proper magnetic
balance is minimally achieved by including 24 fivebranes distributed randomly in the bulk
of the orbifold. This situation is depicted by the first diagram in Fig. 6.
The diagrams in Fig. 6 depict a portion of the orbifold in which only four of the thirty-
two αβ-planes are shown. (The entire orbifold would be represented as in Fig. 1.) One
Fig. 6. A global picture of a phase transition in the S 1 /Z2 × T 4 /Z2 orbifold mediated by fivebranes.
should think of these ladder-diagrams as assemblies of the individual vertex-diagrams

represented in Fig. 4. Thus, vertical lines represent seven-dimensional β-planes and
horizontal lines represent ten-dimensional α-planes. It is further understood that each
of the two ten-planes supports local E8 matter. The explicit factors of E8 shown in
the first diagram of Fig. 6 indicate that the ten-dimensional E8 matter is completely
unbroken on each of the four αβ intersections shown in that diagram. Similarly, any
explicit group shown at a vertex in a ladder-diagram indicates the subgroup G ⊂ E8 which
remains unbroken by β at that indicated vertex. The ’s indicate fivebranes. Each has a
worldvolume which fills the six noncompact dimensions extending out of the plane of the
diagram and carries unit magnetic charge.
The fivebranes are free to move about the diagrams. Notably, due to the consistent vertex
indicated at the top of the right-hand column in Fig. 4, the fivebranes are free to move to,
and wrap, any of the vertices in the first diagram of Fig. 6. Such a procedure is shown in
the second diagram in Fig. 6, where the arrow indicates that the fivebrane has moved to,
and wrapped, the indicated vertex. Note that, in this process, the local magnetic charge
is increased by one unit, from −3/4 to +1/4. In addition, the twisted spectrum at that
vertex is augmented by one tensor and one hypermultiplet. All of this is consistent with the
absorption of a fivebrane. Transitions of this type can occur, without further constraints, at
any vertex of a global orbifold configuration.
We would now like to consider the case where a fivebrane moves to a vertex and
metamorphizes into a gauge instanton via a small instanton phase transition. As discussed
above, this results in the breaking of E8 to one of its subgroups. Previously in this paper,
such transitions have been analyzed at a single vertex only. However, within the context
of a global orbifold configuration, it is necessary to insure that such a phase transition
is compatible with the structure of the surrounding vertices. This puts additional, rather
strong, constraints on the allowed phase transitions. The pertinent issue involves the
electric charge ρ of the β-planes. Since the unbroken (E8 ) phases correspond to ρ = 0 and
the broken phases (E7 × SU(2) or SO(16)) correspond to ρ = 1, and since ρ is associated
with an entire seven-dimensional β-plane (i.e., an entire vertical line in one of our ladder-
diagrams), it would seem that instanton transitions on vertices should only occur in pairs.
According to this interpretation, the lone nT = 1 vertex shown in the second diagram in
Fig. 6 can smoothly connect to an instanton only if another fivebrane first moves to the
complementary six-plane on the top of the ladder-diagram, as depicted in the third diagram
in Fig. 6. This enables each of the fivebranes to then smoothly connect to instantons,
through processes of the sort shown in Fig. 6, simultaneously turning on an electric charge
ρ = 1 on the interpolating seven-plane. Thus, the sequence depicted in Fig. 5 describes a
fivebrane-mediated transition which smoothly connects an E8 × E8 phase of the moduli
space to an SO(16) × SO(16) phase.
It is puzzling that the value of the electric parameter ρ, which is ostensibly derived
from a seven-dimensional Chern–Simons coupling, changes from zero to one when the
described transitions take place. This is puzzling because if ρ is a mere coupling constant,
its value should not be subject to intermittent change. (On the other hand, we are able
to justify a related change in the magnetic charge g, since this has an understandable
topological origin, which allows us to resolve such a change in the manner described
previously.) We suppose that this issue has an interesting resolution. This curiosity was
independently noticed and commented on in [6]. For the time being we allow a situation-
dependent ρ as an allowed rule. We hope to discuss this issue further in a future paper.
It is less clear how the half-integer instantons can emerge via smooth transitions
involving fivebranes. One picture which suggests itself, however, is the following. We can
imagine a fivebrane moving to one of the ten-planes, which smoothly connects in moduli
space to a small ten-dimensional E8 gauge instanton. Such a small instanton could then
grow until it encompasses each of two fixed six-planes within the ten plane. We could
imagine that this instanton then splits into two half-integer instantons, one localized on
each of the two adjacent αβ-planes. This process is shown by a sequence of diagrams in
Fig. 7. We abbreviate this transition by the two-step sequence depicted in Fig. 8. We can
describe this process heuristically by saying that half of a fivebrane has wrapped each of
the two involved αβ-planes. If such a thing were possible, then we could describe another
phase transition as indicated by the sequence of diagrams in Fig. 9. This depicts a fivebrane-
mediated transition which smoothly connects an E8 × E8 phase of moduli space to an
SO(16) × E7 × SU(2) phase.
Fig. 7. A fivebrane spawning a pair of half-integrally charged gauge instantons in a process involving
an intermediate integrally charged instanton which grows to encompass two adjacent αβ-planes.
Fig. 8. A two-step abbreviation of the process depicted in Fig. 7. Heuristically, half of a fivebrane is
wrapping each of the indicated vertices.
Fig. 9. An global picture of a phase transition in the S 1 /Z2 ×T 4 /Z2 orbifold, mediated by fivebranes.
5. Conclusions
We have described some of the technology needed to resolve microscopic consistency

issues in M-theory orbifolds. We have applied these techniques particularly to the S 1 /Z2 ×
T 4 /Z2 orbifold, and have presented an interesting set of consistent vertices, describing the
twisted states residing on orbifold planes at intersecting points. By assembling consistent
vertices we are able to build up consistent global orbifolds which describe different phases
of moduli space. Our construction rather nicely indicates the possibility of phase transitions
involving M-fivebranes as mediators. There remain intriguing conceptual issues which we
intend to discuss and hope to resolve in forthcoming papers.
By resolving a larger set of consistent local vertices, thereby enlarging the number of
diagrams in Fig. 4, we should be able to significantly enrich our understanding of the
possible allowed global configurations, and of the implied interconnectedness amongst
phases in the moduli spaces associated with various orbifold compactifications indicated
by the transitions which our diagrams suggest. It should prove interesting to apply our
techniques to other orbifolds as well.
Acknowledgements
This work is supported by the European Commission TMR programme ERBFMRX-

CT96-0045, and the work of B.A.O. is supported by the Alexander von Humboldt-
foundation. We would like to acknowledge helpful emails from Luis Alvarez-Gaume,

Paul Aspinwall, Bernard de Wit and Bert Schellekens, and helpful comments by Vadim
Kaplunovsky. In addition, we are grateful to Stefan Theisen for pointing out an error
in the previous version of this paper pertaining to the pseudoreality of a hypermultiplet
representation.
References
[1] P. Hořava, E. Witten, Heterotic and type I string dynamics from eleven dimensions, Nucl. Phys.
B 475 (1996) 94–114, hep-th/9510209.
[2] P. Hořava, E. Witten, Eleven-dimensional supergravity on a manifold with boundary, Nucl.
Phys. B 460 (1996) 506–524, hep-th/9603142.
[3] K. Dasgupta, S. Mukhi, Orbifolds of M-theory, Nucl. Phys. B 465 (1996) 399–412, hep-
th/9512196.
[4] E. Witten, Five-branes and M-theory on an orbifold, Nucl. Phys. B 463 (1996) 383–397, hep-
th/9512219.
[5] M. Faux, D. Lüst, B.A. Ovrut, Intersecting orbifold planes and local anomaly cancellation in
M-theory, Nucl. Phys. B 554 (1999) 437–483, hep-th/9903028.
[6] V. Kaplunovsky, J. Sonnenschein, S. Theisen, S. Yankielowicz, On the duality between
perturbative heterotic orbifolds and M-theory on T 4 /ZN , hep-th/9912144.
[7] K. Meissner, M. Olechowski, Anomaly cancellation in M-theory on orbifolds, hep-th/0003233.
[8] A. Bilal, J.P. Derendinger, R. Sauser, M-theory on S 1 /Z2 : new facts from a careful analysis,
hep-th/9912150.
[9] E. Witten, On flux quantization in M-theory and the effective action, J. Geom. Phys. 22 (1977)
1–13, hep-th/9609122.
[10] A. Lukas, B.A. Ovrut, K.S. Stelle, D. Waldram, The Universe as a domain wall, Phys. Rev. D 59
(1999) 086001.
[11] R. Donagi, A. Lukas, B.A. Ovrut, D. Waldram, Non-perturbative vacua and particle physics in
M-theory, JHEP 9905 (1999) 018.
[12] A. Lukas, B.A. Ovrut, K.S. Stelle, D. Waldram, Cosmological solutions of Horava–Witten
theory, Phys. Rev. D 59 (1999) 086001.
[13] R. Donagi, B.A. Ovrut, D. Waldram, Moduli spaces of fivebranes on elliptic Calabi–Yau
threefolds, JHEP 9911 (1999).
[14] M. Braendle, A. Lukas, B.A. Ovrut, Heterotic M-theory cosmology in four and five dimensions,
hep-th/0003256.
[15] T. van Ritbergen, A.N. Schellekens, J.A.M. Vermaseren, Group theory factors for Feynman
diagrams, Int. J. Mod. Phys. A 14 (1999) 41–96, hep-th/9802376.
[16] M. Green, J.A. Harvey, G. Moore, I-brane inflow and anomalous couplings on D-branes, Class.
Quant. Grav. 14 (1997) 47–52, hep-th/9605033.
[17] B.A. Ovrut, T. Pantev, J. Park, Small instanton transitions in heterotic M-theory, hep-
th/0001133.
[18] E. Witten, Phase transitions in M-theory and F-theory, hep-th/9603150.
[19] E. Bergshoeff, E. Sezgin, A. van Proeyen, Superconformal tensor calculus and matter couplings
in six dimensions, Nucl. Phys. B 264 (1986) 653–686.
[20] B. de Wit, P.G. Lauwers, A. van Proeyen, Lagrangians of N = 2 supergravity-matter systems,
Nucl. Phys. B 255 (1985) 569–608.
[21] R. Slansky, Group theory for unified model building, Phys. Rep. 79 (1981) 1–128.
Dirichlet branes on a Calabi–Yau three-fold

orbifold
Bogdan Stefański jr. ∗
Department of Applied Mathematics and Theoretical Physics, Centre for Mathematical Sciences,
Wilberforce Road, Cambridge CB3 0WA, UK
Received 7 June 2000; accepted 20 June 2000
Abstract
The D-brane spectrum of a Z2 × Z2 Calabi–Yau three-fold orbifold of toroidally compactified
type IIA and type IIB string theory is analysed systematically. The corresponding K-theory groups
are determined and complete agreement is found. New kinds of stable non-BPS D-branes are found,
whose stability regions are far more complicated than the previously discussed non-BPS D-branes.
The decay channels of non-BPS D-branes beyond their stability regions are identified. Finally the
T-dual orbifold is analysed and a suitable K-theory is found.  2000 Elsevier Science B.V. All rights
reserved.
PACS: 11.25.-w; 11.25.Sq
1. Introduction
The study of non-BPS D-branes 1 has provided a better understanding of several

aspects of string theory. The classification of D-branes in terms of K-theory [4–6] can
be physically understood by thinking of D-branes as solitonic solutions of the tachyon
field of higher-dimensional brane–antibrane systems [5,7–9]. The study of the tachyon
potential using string field theory [10–13] has given strong evidence to the conjecture that
a brane–antibrane system annihilates into the vacuum. Due to the presence of a single
decay mode unstable non-BPS D-branes have been interpreted as sphaleron solutions of
string theory [14] and some understanding of bosonic D-branes [15] has now also been
developed. The confinement process of the unbroken gauge group boson on the world-
volume of annihilating brane–antibrane systems [5,16] has been better understood [17–19].
Another aspect of recent developments has involved the study of stable non-BPS
D-branes through the boundary state formalism [20,21]. This was first considered in a
∗ b.stefanski@damtp.cam.ac.uk
1 For some recent reviews see [1–3].
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 1 0 - 7
B. Stefański jr. / Nuclear Physics B 589 (2000) 292–314 293
specific context in [7,22]. Over the past year the boundary state formalism has been used
to construct many more examples of non-BPS D-branes [23–33]. These constructions
have tested the connection between D-branes and K-theory [25,29,34] as well as testing
S-dualities beyond the BPS constraints. In particular the dualities between heterotic and
type II theories [24] and heterotic and type I theories [35–37] have been investigated. Due
to the fact that certain non-BPS D-branes are the lightest states carrying a particular charge
it has been possible to identify their duals and compare regions of stability [36] as well as
interaction properties [37] in the dual theories.
The integrally (rather than torsion) charged non-BPS D-branes have so far been
constructed on Z2 orbifolds 2 and have been interpreted in the blow-up [23] as coming
from certain volume minimising non-supersymmetric cycles of the manifold which in the
orbifold limit shrink to one-cycles. Here we discuss a Z2 × Z2 orbifold of types IIA and
IIB. In particular we study BPS and non-BPS D-branes on such an orbifold. The orbifold
is a particular limit of a Calabi–Yau three-fold, and as such preserves N = 2, D = 4
supersymmetry. The action of the two generators g1 and g2 will be taken as
g1 (x 0 , . . . , x 9 ) = (x 0 , x 1 , x 2 , x 3 , x 4 , −x 5 , −x 6 , −x 7 , −x 8 , x 9 ), (1.1)
g2 (x , . . . , x ) = (x , x , x , −x , −x , x , x , −x , −x , x )
0 9 0 1 2 3 4 5 6 7 8 9
(1.2)
and we label g3 = g1 g2 . Thus both Z2 ’s have fixed points. This orbifold has been studied
previously [38–42]. In particular in [39] fractional D-branes on it were discussed and it was
noted that there are two kinds of fractional D-branes, those that live at the fixed hyperplanes
of one of the gi (we shall refer to these as singly fractional D-branes) and those that live at
the fixed hyperplanes of all gi (we shall refer to these as totally fractional D-branes).
In this paper we use the boundary state formalism to investigate the full D-brane
spectrum, BPS and non-BPS on this orbifold and study the stability regions of non-BPS
b
D-branes on the orbifold. We give a boundary state description of both kinds of fractional
D-branes as well as bulk D-branes wrapping the special Lagrangian three-cycle of the
Calabi–Yau orbifold. We find non-BPS b D-branes similar to those of [29]. We also find
that there are new kinds of non-BPS b D-branes on this orbifold, and we refer to these
too as truncated b D-branes [29]. The truncated b D-branes break all supersymmetry and
are charged under twisted R–R fields. The relevant K-theory groups are evaluated and
complete agreement between the D-branes we have constructed and K-theory is found.
Some of the new b D-branes have very unusual domains of stability and besides coupling
to twisted R–R sectors they couple to certain twisted NS–NS sectors as well. We investigate
the decay channels of the various b D-branes. Besides the usual decays into brane–antibrane
pairs of BPS fractional D-branes there are other decay channels in which non-BPS
b
D-branes decay into one another. There are also certain decay channels whose decay
products are unknown. When the decay products are known the masses and charges of
the relevant objects are the same at critical radii, indicating that the process is a marginal
deformation in the conformal field theory [8,43,44].
2 In [23,30] this had been generalised somewhat to a Calabi–Yau Z × Z0 orbifold, where Z0 is freely acting.
2 2 2
294 B. Stefański jr. / Nuclear Physics B 589 (2000) 292–314
The construction of boundary states describing the D-branes is given in Section 2, as

well as the appendices. In Section 3 we compute the relevant equivariant K-theory groups
and find complete agreement with Section 2. In Section 4 we consider the compact orbifold
and we discuss in detail the minimal charge configurations of the D-branes. The stability
regions of the truncated non-BPS b D-branes are analysed in Section 5. The new kind of
truncated bD-brane exhibits some remarkable stability properties as given in Eq. (4.8).
In Section 6 we identify most of the decay channels for the various b D-branes. Finally,
Section 7 discusses D-branes on the T-dual orbifold. A new K-theory group suitable for
such an orbifold is defined.
2. Non-compact orbifold
In this section we discuss the non-compact orbifold, while later on the directions
x 3, . . . , x 8
will be compactified on circles. We first briefly discuss the massless spectrum,
and then construct relevant D-brane boundary states.
The Z2 × Z2 orbifold discussed here has been studied in the past [38,39]. In particular
its closed string spectrum has been worked out in some detail [38]. Here we point out some
of the points of present importance. Each gi , i = 1, 2, 3, gives rise to a twisted sector. Both
the twisted NS and R sectors are massless and have zero modes, those of the R sector
are in the directions unaffected by gi , while the zero modes in the NS sector are in the
directions inverted by gi . The lowest lying states in the twisted R–R sector transform as a
tensor product of SO(4) (which contains the spacetime SO(2)), and are further required to
be invariant under the remaining orbifold projections. The twisted NS–NS sector ground
states transform as bispinors of an internal SO(4), and have to be invariant under the other
elements of the orbifold group. They give rise to four-dimensional scalars. We denote the
gi -twisted sectors by NS–NS,T gi and R–R,T gi .
Boundary states corresponding to physical D-branes consist of linear combinations
of boundary states from the various closed string sectors which are GSO and orbifold
invariant. We refer to a D-brane as of type (r; s) = (r; s1 , s2 , s3 ), if it is a (r + s1 + s2 + s3 )-
brane, and extends along r + 1, s1 , s2 , s3 of the directions (x 0 , . . . , x 2 , x 9 ), (x 3 , x 4 ),
(x 5 , x 6 ), (x 7 , x 8 ), respectively. 3 Due to the presence of fermionic zero modes in the
various sectors requiring GSO and orbifold invariance places restrictions on r and the si .
These conditions are analysed in Appendix B.
A boundary state which corresponds to a physical D-brane has to satisfy certain
consistency criteria. In particular the string with endpoints on the D-brane has to be a
suitably projected open string. It turns out that in the Z2 × Z2 orbifold there are four kinds
of combinations of boundary states from the various closed string sectors which correspond
to physical D-branes. These include a bulk D-brane
|B(r; s)iNS−NS + ε|B(r; s)iR−R , (2.1)
where ε = ±1, and a fractional D-brane
3 With this convention g acts trivially on the s directions.

i i
X
3

|B(r; s)iNS−NS + ε|B(r; s)iR−R + εi |B(r; s)iNS−NS,Tgi + ε|B(r; s)iR−R,Tgi ,
i=1
(2.2)
for which εi εj = εk for i 6= j 6= k 6= i; both these objects are BPS as can be seen for
example from the vanishing of the cylinder amplitudes computed in Appendix A. In the
above ε, εi indicate the sign of the various R–R charges. Further there are truncated non-
BPS bD-branes charged under one type of twisted R–R field
|B(r; s)iNS−NS + ε|B(r; s)iR−R,Tgi , i = 1, 2, 3. (2.3)
Their boundary state is very similar to that of the b

D-branes in [29]. Finally there is a second
type of truncated non-BPS b D-brane whose boundary state is
|B(r; s)iNS−NS + εi |B(r; s)iR−R,Tgi

+εi εj |B(r; s)iNS−NS,Tgk + εj |B(r; s)iR−R,Tgj ,
i, j, k = 1, 2, 3, s.t. ij k 6= 0. (2.4)
This is a new kind of b D-brane which couples to a twisted NS–NS sector as well as to
twisted R–R sectors.
A second consistency condition is that a string beginning on any one of the above
branes and ending on a different one must describe an open string. In [29] this condition
ensured that for any given (r, s) if a fractional and truncated (r, s)-branes existed then
the fractional brane would be of smaller mass and charge rendering the truncated brane
unstable. Applying the condition to the branes above we again conclude that if a fractional
and truncated brane exist for some (r; s) then it is the fractional object that is minimal and
stable. Similarly one shows that if for some (r; s) both truncated objects exist it is the one
in Eq. (2.4) that is minimally charged and so is fundamental.
Given these consistency conditions the D-brane spectrum can then be determined. The
detailed analysis can be found in Appendix B. Bulk and fractional D-branes exist for (r; s)
of the form
(r; 0, 0, 0), (r; 0, 0, 2), (r; 0, 2, 0), (r; 2, 0, 0),

(r; 0, 2, 2), (r; 2, 0, 2), (r; 2, 2, 0), (r; 2, 2, 2), (2.5)
where r is even/odd for type IIA/IIB. The elementary objects above are totally fractional
branes which live on the fixed point of the whole orbifold group (x 3 = · · · = x 8 = 0) and
are charged under all three twisted R–R sectors. The corresponding K-theory group should
be Z⊕4 . The totally fractional branes correspond to a single brane in the covering space.
Two such D-branes with the same bulk and gi -twisted R–R charges but opposite remaining
charges can come together to form a singly fractional brane, which is stuck to the fixed
plane of gi . Two singly fractional branes with opposite twisted R–R charge can come
together and form a bulk brane. Further bulk D-branes exist for (r; 1, 1, 1). Since these are
charged under the untwisted R–R field, the corresponding K-group should be Z.
Truncated b
D-branes with a boundary state given by Eq. (2.3) exist for (r; s) of the form 4
(r; 0, 1, 1), (r; 1, 0, 1), (r; 1, 1, 0), (r; 2, 1, 1), (r; 1, 2, 1), (r; 1, 1, 2),
(2.6)
with r even/odd for type IIA/IIB. These bD-branes are stuck at the fixed points of the
gi under which they are charged, and one expects the corresponding K-group to be Z.
Truncated b
D-branes with a boundary state given by Eq. (2.4) exist for (r; s) of the form
(r; 0, 0, 1), (r; 1, 0, 0), (r; 0, 1, 0), (r; 2, 0, 1), (r; 1, 2, 0), (r; 2, 1, 0),
(r; 0, 2, 1), (r; 1, 0, 2), (r; 0, 1, 2), (r; 2, 2, 1), (r; 1, 2, 2), (r; 2, 1, 2).
(2.7)
Here r is even/odd for type IIA/IIB. The basic such branes are stuck at the fixed points of
the whole orbifold group and are charged under two twisted R–R fields suggesting that the
K-group should be Z ⊕ Z. However, as with the fractional branes there are also b D-branes
with the above (r; s) which are charged under only gi -twisted R–R fields and as such are
only stuck to the fixed points of gi .
3. K-theory analysis in uncompactified theory
The K-groups relevant to this orbifold are the Z2 × Z2 -equivariant K-groups [45] with
compact support

KZ∗2 ×Z2 Ra;b,c,d , (3.1)
where a = 0, . . . , 4 and b, c, d = 0, 1, 2. In the above the directions a are left invariant
by G = Z2 × Z2 , c and d are inverted by the first Z2 and b, c are inverted by the second
Z2 . These groups exhibit complex Bott periodicity in a, b, c and d thus the answer depends
only on the parity of a, b, c and d. Further there is a symmetry between b, c, d. For example
KG∗ (Ra;b,c,d ) = K ∗ (Ra;c,b,d ). As a result we need only compute very few terms. Since the
G
representation ring of Z2 × Z2 is Z⊕4 we have

∗
Z⊕4 , a + ∗ even,
KG R a;b,c,d
= (3.2)
0, a + ∗ odd,
for b, c, d even. Fractional branes in the decompactified theory can be charged under three
twisted R–R fields as well as the untwisted R–R field, the above K-groups confirm the
presence of all fractional branes. Since g1 acts trivially on R0;1,0,0 we have
∗

KG R0;1,0,0 = KZ∗2 R0,1 ⊗ R[Z2 ], (3.3)
where on the right-hand side Z2 inverts the line R0,1 . The right-hand side groups have been
obtained in [25,29]. Thus we have for c, d even

∗
Z⊕2 , a + ∗ even,
KG R a;1,c,d
= (3.4)
0, a + ∗ odd.
4 We disregard here the tachyon that arises when considering decompactified s > 0 truncated branes [29], as
i
on compactification the D-brane will stabilise for certain values of the compactification radii.
Similar results hold for permutations of 1, c, d. This is in agreement with the presence of
truncated b
D-branes in (2.7). In order to compute KG ∗ (R0;1,1,2k ) we re-write it as
∗
∗

KG R0;1,1,0 = KG R0;1,0,0 × D1 , Ra;1,0,0 × S 0 , (3.5)
where S 0 ⊂ D1 ⊂ R0;0,1,0 are the zero-sphere (two points) and the one-disc (the interval).
By homotopy equivalence we have
∗
∗
KG X × D1 = KG (X), (3.6)
for any X and

∗

KG R0;1,0,0 × S 0 = KZ∗2 R0,1 , (3.7)
where on the right-hand side Z2 inverts the real line R0,1 . We may now use the long exact
sequence
−1
· · · → KG R0;1,0,0 × S 0 → KG R0;1,0,0 × R0;0,1,0

→ KG R0;1,0,0 × D1 → KG R0;1,0,0 × S 0

→ KG
1
R0;1,0,0 × R0;0,1,0 → KG 1
R0;1,0,0 × D1 → · · · (3.8)
which in turn becomes

0 → KG R0;1,1,0 → Z ⊕ Z → Z → KG
1
R0;1,1,0 → 0. (3.9)
It is not difficult to see that the map Z ⊕ Z → Z is surjective. This gives for d even

∗
Z, a + ∗ even,
KG R a;1,1,d
= (3.10)
0, a + ∗ odd.
Together with the permutations of 1, 1, d this confirms the presence of truncated b

D-branes
in Eq. (2.6). Finally we have
∗
∗

KG R0;1,1,1 = KG R0;1,1,0 × D1 , Ra;1,1,0 × S 0 , (3.11)
where S 0 ⊂ D1 ⊂ R0;0,0,1. Using Eq. (3.6) the results above and

∗

KG R0;1,1,0 × S 0 = KZ∗2 R0,2 , (3.12)
the following sequence is exact

0 → KG R0;1,1,1 → Z → Z ⊕ Z → KG
1
R0;1,1,1 → 0. (3.13)
One can show using this exact sequence that [46]

∗
Z, a + ∗ odd,
KG Ra;1,1,1 = (3.14)
0, a + ∗ even.
This confirms the presence of bulk BPS branes with s1 = s2 = s3 = 1. We have
demonstrated complete agreement between K-theory and D-branes on this orbifold.
4. The compactified orbifold
From now on we take the directions x 3 , . . . , x 8 to be compactified on circles of radii

R3 , . . . , R8 . This introduces new fixed points at x i = πRi and gives rise to new twisted
sectors — in total 48 twisted R–R sectors and 48 twisted NS–NS sectors. D-branes with
si > 0 will be charged under the twisted sectors over which they stretch. The structure of
the boundary states and in particular the normalisation is described in Appendix A. The
allowed charges for various D-branes are restricted by factorisation. In particular a brane
charged under many R–R charges cannot have an arbitrary choice of minimal positive and
negative charges. This was already encountered in [29] where for example a b D(r, 2)-brane
in the In (−1) orbifolds was charged under four twisted R–R fields but the minimally
Fl
charged branes had to have an even number of negative charges. This behaviour is typical
and we encounter it here as well. The restriction arises as a result of cylinder-annulus
consistency. The process of checking what sign freedom one has in the closed string sector
in order to factorise on a consistent open string amplitude is laborious, here we summarise
the results.
In the following subsections we shall discuss the D-brane spectrum of the compactified
orbifold, paying attention to the allowed charges, as well as stability regions and
decay products of the non-BPS branes. For simplicity we concentrate on the fractional
D(r; 0, 0, 0)- and D(r; 0, 0, 2)-branes and truncated b D(r; 0, 0, 1)- and b
D(r; 0, 1, 1)-branes.
The extension of these results to the other branes is obvious.
4.1. The D(r; 0, 0, 0)-brane
The fully fractional D(r; 0, 0, 0)-brane’s consistent boundary state is given by

|D(r; 0, 0, 0), ε, εi i = |D(r; 0, 0, 0)iNS−NS + ε|D(r; 0, 0, 0)iR−R
X
3

+ εi |D(r; 0, 0, 0)iNS−NS,Tgi + ε|D(r; 0, 0, 0)iR−R,Tgi ,
i=1
(4.1)
with ε3 = ε1 ε2 = ±1. If we denote by [a; b, c, d] the four charges under which the D-brane
is charged (with a corresponding to the bulk charge, and b, c, d to the twisted charges) the
allowed configurations of minimal charge are
[1; 1, 1, 1], [1; 1, −1, −1], [1; −1, 1, −1], [1; −1, −1, 1],
(4.2)
[−1; −1, −1, −1], [−1; −1, 1, 1], [−1; 1, −1, 1], [−1; 1, 1, −1].
From this it is easy to see that a singly fractional brane is a sum of two totally
fractional branes. For example, [2; 0, 2, 0] = [1; 1, 1, 1] + [1; −1, 1, −1] or [−2; 0, 2, 0] =
[−1; −1, 1, 1] + [−1; 1, 1, −1].
4.2. The D(r; 0, 0, 2)-brane
We take the fully fractional D(r; 0, 0, 2)-brane to extend along x 7 and x 8 and to be fixed
at x 3 = · · · = x 6 = 0. Its’ boundary state is
|D(r; 0, 0, 2), ε, εi , θj i
X
= ei(θ7 w7 +θ8 w8 ) |D(r; 0, 0, 2), w7, w8 iNS−NS
w7 ,w8

+ ε|D(r; 0, 0, 2), w7, w8 iR−R
X
+ εi |D(r; 0, 0, 2)iNS−NS,Tgi,1 + ε|D(r; 0, 0, 2)iR−R,Tgi,1
i∈{1,2}

+ εi eiθ7 |D(r; 0, 0, 2)iNS−NS,Tgi,2 + ε|D(r; 0, 0, 2)iR−R,Tgi,2

+ εi eiθ8 |D(r; 0, 0, 2)iNS−NS,Tgi,3 + ε|D(r; 0, 0, 2)iR−R,Tgi,3

+ εi ei(θ7 +θ8 ) |D(r; 0, 0, 2)iNS−NS,Tgi,4 + ε|D(r; 0, 0, 2)iR−R,Tgi,4
X
+ ε1 ε2 ei(θ7 w7 +θ8 w8 ) |D(r; 0, 0, 2), w7, w8 iNS−NS,Tg3
w7 ,w8

+ ε|D(r; 0, 0, 2), w7, w8 iR−R,Tg3 . (4.3)
From this it is apparent that D(r; 0, 0, 2)-branes can only have an even number of negative
R–R,T g1 and R–R,T g2 charges, and further that the sign of the R–R,T g3 charge is fixed
by the other twisted charges. A D(r; 0, 0, 2)-brane with all positive R–R,T g1 and bulk
R–R charges, but negative remaining charges (take ε = ε1 = +1, θ7 = θ8 = 0, ε3 = −1)
can be added to a D(r; 0, 0, 2)-brane with all positive charges, to give a brane which is
only charged under bulk R–R and R–R,T g1 fields, and so is a singly fractional brane (it
can move off the x 3 = x 4 = 0 fixed plane).
b 0, 0, 1)-brane
4.3. The D(r;
We consider the b D(r; 0, 0, 1)-brane to stretch along the x 8 direction and to be fixed at
x3= · · · = x = 0. In order to be consistent with the open string partition function the
7
b
D(r; 0, 0, 1)-brane boundary state must be written in the form
|D̂(r; 0, 0, 1), εi , θi
X
eiθw |D̂(r; 0, 0, 1), wiNS−NS + ε1 |D̂(r; 0, 0, 1)iR−R,Tg1,1
w
+ ε1 eiθ |D̂(r; 0, 0, 1)iR−R,Tg1,2 + ε2 |D̂(r; 0, 0, 1)iR−R,Tg2,1
X
+ ε2 eiθ |D̂(r; 0, 0, 1)iR−R,Tg2,2 + ε1 ε2 eiθw |D̂(r; 0, 0, 1), wiNS−NS,Tg3 .
w
(4.4)
Denoting by [a, b; c, d] the four charges a bD(r; 0, 0, 1)-brane carries (two from the
g1 -twisted sector, say a, b and two from the g2 twisted sector, say c, d), the allowed
b
D(r; 0, 0, 1)-branes with minimal charge are
[1, 1; 1, 1], [−1, −1; 1, 1], [1, 1; −1, −1], [1, −1; 1, −1],
(4.5)
[−1, 1; 1, −1], [1, −1; −1, 1], [−1, 1; −1, 1], [−1, −1; −1, −1].
A singly charged b D(r; 0, 0, 1)-brane with minimal charge comes from [1, 1; 1, 1] +
[1, 1; −1, −1] = [2, 2; 0, 0] or [1, −1; 1, −1] + [1, −1, −1, 1] = [2, −2; 0, 0].
Since the stability region of this bD-brane is so different from the non-BPS branes
previously encountered we discuss this case in detail. The partition function for a string
with both end-points on this b
D-brane is
Z
dt 1 + (−1)F g1 1 + (−1)F g3 −2t Ho
TrNS−R e
2t 2 2
Z 8
−(r+1)/2 f3 (q̃) − f2 (q̃)
8
Vr+1 dt
= (2t)
4(2π)r+1 2t f18 (q̃)
X Y
7 X
e−2t πn8 /R8 e−2t πwm Rm
2 2 2 2
×
n8 ∈Z m=3 wm ∈Z
Z
Vr+1 dt f 4 (q̃)f44 (q̃) Y X −2t πw2 R 2
− (2t)−(r+1)/2 34 e m m
(2π)r+1 2t f1 (q̃)f24 (q̃) m=3,4 w ∈Z
m
Z
Vr+1 dt f 4 (q̃)f44 (q̃) Y X −2t πw2 R 2
− (2t)−(r+1)/2 34 e m m
(2π)r+1 2t f1 (q̃)f24 (q̃) m=5,6 wm ∈Z
Z
Vr+1 dt f (q̃)f4 (q̃) X −2t πw2 R 2 X −2t πn2 /R 2
4 4
+ (2t)−(r+1)/2 34 e 7 7 e 8 8
(2π)r+1 2t f1 (q̃)f24 (q̃) w ∈Z
7 n ∈Z
Z
8
Vr+1 dt
≈ (2t)−(r+1)/2q̃ −1
4(2π)r+1 2t
2 2 X
7
2 2 2
× q̃ 2(n8 /R8 ) + q̃ 2(w7 R7 ) + q̃ 2(wi Ri ) q̃ 2(n8 /R8 ) + q̃ 2(w7 R7 )
i=3
2(w3 R3 )2 +2(w5 R5 )2 2 +2(w R )2
+ q̃ + q̃ 2(w3 R3 ) 6 6
!
2(w4 R4 )2 +2(w5 R5 )2 2(w4 R4 )2 +2(w6 R6 )2
+ q̃ + q̃ . (4.6)
We have expanded to relevant order in the last line of the amplitude. It is then clear that the
tachyon cancels provided 5
1 √
R7 > √ , R8 6 2, (4.7)
2
1 1 1 1
R32 + R52 > , R42 + R52 > , R32 + R62 > , R42 + R62 > . (4.8)
2 2 2 2
Note that by fixing say R3 , R4 to suitable values R5 , R6 can take on any value and the
truncated D-brane will still be stable.
In order to analyse the decay products of this b
D-brane we take it to carry a unit of positive
twisted R–R charge in each of the four sectors under which it is charged (two R–R,T g1
and two R–R,T g2 ). As can be seen from the above equations there are three distinct decay
5 It is straightforward to see how this generalises to other b D(0; 2k, 2k 0 , 1)-branes. For √
example for
√a
b
D(0; 2, 0, 1)-brane stretching along the directions x 3 , x 4 , x 8 the stability region is R7 > 1/ 2, R8 6 2,
(1/R32 ) + R52 > 1/2, 1/R42 + R52 > 1/2, (1/R32 ) + R62 > 1/2, (1/R42 ) + R62 > 1/2.
√ √
channels: R8 can increase beyond 2, R7 can decrease below 1/ 2, and the bounds for
R3 , R4 , R5 , R√
6 in Eq. (4.8) can be violated.
For R8 > 2 the b D(r; 0, 0, 1)-brane decays, conserving mass and charge, into a pair
of totally fractional D(r; 0, 0, 0)-branes, with opposite bulk and R–R,T g3 charge. In the
notation of Eq. (4.1) one of the fractional D-branes has ε = ε1 = ε2 = ε3 = 1, and the
other has ε = ε1 = −ε2 = ε3 = −1. The easiest way to compare the mass and charges
of the original truncated brane with those of the decay products is by perusal of the
relevant normalisation constants computed in Appendix A. The normalisation constants
of the b
D(r; 0, 0, 1)-brane are (cf. Eqs. (A.36) and (A.37))
Vr+1 1 R8
Nt 2d,U = r+1 Q7 , (4.9)
(2π) 64 m=3 Rm
Vr+1 23 1
Nt 2d,Tg1,i = , i = 1, 2, (4.10)
(2π)r+1 64 R3 R4
Vr+1 23 1
Nt 2d,Tg3,i = , i = 1, 2, (4.11)
(2π)r+1 64 R5 R6
while those of the two D(r; 0, 0, 0)-branes (cf. Eqs. (A.32) and (A.33))
Vr+1 1 1
Nf,U
2
=4 r+1 Q8 , (4.12)
(2π) 128 m=3 Rm
Vr+1 24 1
Nf,Tg
2
1 ,i
= , i = 1, 2, (4.13)
(2π)r+1 128 R3 R4
Vr+1 24 1
Nf,Tg
2
3 ,i
= , i = 1, 2. (4.14)
(2π)r+1 128 R5 R6
The factor of 4 comes from the fact that there are two BPS branes. These agree at the
critical radius. √
The second decay channel occurs when we decrease R7 below 1/ 2. The decay
products are a pair of totally fractional D(r; 0, 0, 2)-branes. In the notation of Section 4.3
one of the D-branes has , 1 , 3 = +1, and θ7 , θ8 = 0, while the other brane has , 1 ,
3 = −1 and θ7 , θ8 = 0. Comparing the normalisation constants confirms that charge and
mass are conserved in the decay.
The third decay channel seems much more complicated. Naively one might expect a
decay into a brane–antibrane combination of fractional branes stretching along x 8 and one
of x 3 , x 4 , x 5 , x 6 . However, there are no such fractional branes. If the decay is to be into
branes with two internal directions, then by considering the Dirac quantisation condition
these will have to be fractional branes, which presumably will be somehow bent away from
the axes, and not have a boundary state description of the type analysed in this paper.
It is also possible to analyse the stability of the b D(r; 0, 0, 1)-brane charged only under,
say, the g1 -twisted R–R sector. Its stability region is
1 √
R5 , R6 , R7 > √ , R8 6 2, (4.15)
2
and the R7 and R8 decay channels follow easily from the decay channels of the
b
D(r; 0, 0, 1)-brane charged under the g1 and g2 -twisted R–R sectors discussed above.
b 0, 1, 1)-brane
4.4. The D(r;
Finally, we discuss a b
D(r; 0, 1, 1)-brane which we take to extend along x 8 and x 6 , and
be fixed at x = x = 0 and to consist of two identical copies at (x 3 , x 4 ) and (−x 3 , −x 4 ),
5 7
as explained in Appendix A. The allowed boundary states are of the form

|b
D(r; 0, 1, 1), εi , θj i
X
= ei(θ6 w6 +θ8 w8 ) |b
D(r; 0, 1, 1), w6, w8 iNS−NS
w6 ,w8
+ ε1 |b
D(r; 0, 1, 1)iR−R,Tg1,1 + ε1 eiθ6 |b
D(r; 0, 1, 1)iR−R,Tg1,2
+ ε1 e |D(r; 0, 1, 1)iR−R,Tg1,3 + ε1 ei(θ6 +θ8 ) |b
iθ8 b
D(r; 0, 1, 1)iR−R,Tg1,3 . (4.16)
Hence, such truncated branes can only carry an even number of negative twisted R–R
charges. The stability of this brane is very similar to those encountered in Z2 orbifolds, we
simply state the results here. It is stable for
1 √
R5 , R7 > √ , R6 , R8 6 2. (4.17)
2
√
The b D-brane can decay in two different√ ways: by increasing R6 or R8 beyond 2, or
by decreasing R5 or R7 below 1/ 2. In the first the decay is into a pair of singly
charged b D(r; 0, 0, 1)-branes, which carry the same charges as√the b D(r; 0, 1, 1)-brane. It
is straightforward to check that at the critical radius (R6 = 2) mass and charge √ are
conserved. The second decay channel corresponds to decreasing R5 below 1/ 2. The
b
D-brane decays into a pair of singly charged b D(r; 0, 2, 1)-branes, whose charges are
b
the same at the four fixed points on which the D(r; 0, 1, 1)-brane ends and opposite at
the other four fixed points. Again both mass and charge are conserved at the transition.
The decay products of the b D(r; 0, 1, 1)-branes are different from the decay products
previously encountered. In particular the brane does not decay into a brane–antibrane pair
of separately BPS objects, rather the decay is into other non-BPS objects.
5. T-duality
In this section we discuss the T-dual of the Z2 × Z2 orbifold. We shall T-dualise along
the x 5 direction. The orbifold we consider will then be a Z2 × Z2 orbifold where the
generators are g1 = I5678 (−1)Fl and g2 = I3478 , where Fl is the left moving spacetime
fermion number. One imposes GSO and orbifold invariance of the boundary states in the
different closed string sectors. As in the I4 (−1)Fl orbifolds there is an ambiguity regarding
which part of the twisted sector states to keep: the I4 (−1)Fl odd states or the I4 (−1)Fl
even states? For the case of the invariance under gi for the gi -twisted sector we follow the
prescription of [29]. In particular we require the g1 - and g3 -twisted sectors to be odd under
g1 and g3 , respectively, while the g2 -twisted sectors are to be even under g2 . As a result
shall require the g1 -twisted sector to be g2 even and g3 odd, the g2 -twisted sector to be
g1 and g3 even and the g3 -twisted sector to be g2 even and g1 odd. This ensures that the
Table 1
Restrictions on r and si as a result of requiring GSO and orbifold invariance for boundary states in
the various closed string sectors for type IIB
GSO g1 g2 g1 g2
NS–NS – – – –
R–R r + s1 + s2 + s3 odd s2 + s3 odd s1 + s3 even s1 + s2 odd
NS–NS,T g1 s2 + s3 odd s2 + s3 odd s3 even s2 odd
R–R,T g1 r + s1 even – s1 even s1 even
NS–NS,T g2 s1 + s3 even s3 even s1 + s3 even s1 even
R–R,T g2 r + s2 odd s2 odd – s2 odd
NS–NS,T g3 s1 + s2 odd s2 odd s1 even s1 + s2 odd
R–R,T g3 r + s3 even s3 even s3 even –
group relation g1 g2 = g3 is satisfied in each sector, and that fractional branes exist. With
these clarifications one obtains Table 1 for type IIB
With these restrictions it is not difficult to identify bulk, fractional and both kinds of
truncated branes. As before if two kinds of branes can exist for a given (r; s) it is the frac-
tional rather than truncated (or the truncated with the NS–NS twisted sector rather than
the other truncated brane) that will be fundamental and of minimal charge. For type IIB
we then have bulk wrapped branes for r, s1 , s3 odd and s2 even (only charged under the
untwisted R–R sector), and bulk and fractional branes (latter charged under four kinds
of charges in the decompactified theory) for r, s1 , s3 even and s2 odd. Truncated branes
charged under R–R,T g1 and R–R,T g2 fields exist for r, s1 even and s2 , s3 odd. Truncated
branes charged under R–R,T g1 and R–R,T g3 fields exist for r, s1 , s2 , s3 all even. Trun-
cated branes charged under R–R,T g2 and R–R,T g3 fields exist for r, s3 even and s1 , s2
odd. Truncated branes charged only under R–R,T g1 exist for r, s1 , s2 even and s3 odd.
Truncated branes charged only under R–R,T g2 exist for r even and s1 , s2 , s3 odd. Trun-
cated branes charged only under R–R,T g3 exist for r, s2 , s3 even and s1 odd. For type IIA
in the above the parity of r changes.
One can develop a K-theory understanding of this. We define a new kind of K-theory
which we call KZ∗ ×Z± . Elements of this are pairs of isomorphism classes of bundles
2 2
(E, F ), which are equivariant under the action of the first Z2 and we are given an
isomorphism between (E, F ) and (g ∗ (E), g ∗ (F )) where in this case g = I5678 and g ∗ (E)
is the pullback of E by g. Further we are given an isomorphism which maps (E, F ) to
(h∗ (F ), h∗ (E)), where h∗ (E) is the pullback of E by h, the generator of the geometric part
of the second Z2 (in other words in this case h = I3478 ). A version of Hopkins’ formula
then is

KZ∗ ×Z± Ra;b,c,d ∼ = KZ∗2 ×Z2 Ra+1;b,c+1,d , (5.1)
2 2
where the second group is just the usual equivariant K-group. We have computed the latter
in Section 3, and using the above relation exact agreement between K-theory and the string
results follows. The stability and decay of the the truncated branes can easily be obtained
from the results of the T-dual scenario described in detail in the previous sections.
Acknowledgements
I am grateful to D.-E. Diaconescu, M.B. Green, F. Quevedo, A. Sen, B. Totaro and in

particular to M.R. Gaberdiel and G. Segal for many helpful discussions and insights. B.S. is
supported by the Cambridge Commonwealth Trust.
Appendix A. Construction and normalisation of boundary states
In this appendix we determine the normalisation constants of the boundary states for the
orbifold theories under consideration.
A.1. The uncompactified case
In each (bosonic) sector of the theory we can construct the boundary state
∞ ∞
!
X 1 µ X µ
|B(r; s), k, ηi = exp α Sµν α̃−l + iη
ν ν
ψ−m Sµν ψ̃−m
l −l
l>0 m>0
×|B(r; s), k, ηi , (0)
(A.1)
where, depending on the sector, l and m are integer or half-integer, and k denotes the
momentum of the ground state. We shall always work in light-cone gauge with light-
cone directions x 0 and x 9 ; thus µ and ν take the values 1, . . . , 8. We shall also drop the
dependence on α0 .
The parameter η = ± describes the two different spin structures [20,21], and the matrix
S encodes the boundary conditions of the Dp-brane which we shall always take to be
diagonal
S = diag(−1, . . . , −1, 1, . . . , 1), (A.2)
where p + 1 entries are equal to −1, 7 − p entries are equal to +1, and p = r + s1 + s2 + s3 .
If there are fermionic zero modes, the ground state in (A.1) satisfies additional conditions
discussed in the following appendix.
In order to obtain a localised D-brane, we have to take the Fourier transform of the above
boundary state, where we integrate over the directions transverse to the brane,
Z Y
µ 0 9
|B(r, s), y, ηi = dk µ eik yµ dk 0 eik y0 dk 9 eik y9 |B(r; s), k, ηi, (A.3)
µ transverse
y denotes the location of the boundary state, and in the gi -twisted sector the momentum
integral only involves transverse directions that are not inverted by the action of gi . In the
following we shall typically consider (without loss of generality) the case y = 0 in which
case the boundary state is denoted by |B(r; s), ηi.
The invariance of the boundary state under the GSO-projection always requires that the
physical boundary state is a linear combination of the two states corresponding to η = ±.
Using the conventions of Appendix B in [29], these linear combinations are of the form
1
|B(r; s)iNS−NS = |B(r; s), +iNS−NS − |B(r; s), −iNS−NS , (A.4)
2

|B(r; s)iR−R = 2i |B(r; s), +iR−R + |B(r; s), −iR−R , (A.5)
1
|B(r; s)iNS−NS,Tgi = |B(r; s), +iNS−NS,Tgi + |B(r; s), −iNS−NS,Tgi , (A.6)
2

|B(r; s)iR−R,Tgi = i |B(r; s), +iR−R,Tgi + |B(r; s), −iR−R,Tgi , (A.7)
where, depending on the theory in question, these states are actually GSO-invariant
provided that r and si satisfy suitable conditions discussed in Section 2 (i = 1, 2, 3). The
normalisation constants have been introduced for later convenience.
In order to solve the open-closed consistency condition the actual D-brane state is a
linear combination of physical boundary states from different sectors. There are three
elementary cases to consider, fully fractional, and the two truncated D-branes. (cf.
Eqs. (2.2)–(2.4)) In the fully fractional case, the D-brane state can be written as

|D(r; s)i = Nf,U |B(r; s)iNS−NS + |B(r; s)iR−R
X
3

+ i Nf,T gi |B(r; s)iNS−NS,Tgi + |B(r; s)iR−R,Tgi , (A.8)
i=1
where = ± determines the sign of the charge with respect to the untwisted R–R sector
charge, while i = ±, i = 1, 2, 3 determines the sign of the charge with respect to the
R–R,T gi charge. For consistency when i 6= j 6= k 6= i we have i = j k . The closed
string cylinder diagram is then of the form
Z
A = dlhB(r; s)|e−lHc |B(r; s)i
Z
1 2 f38 (q) − f48 (q) − f28 (q)
= Nf,U dl l (p−9)/2
2 f18 (q)
X
3 Z
1 (r+si −5)/2 f34 (q)f24 (q) − f24 (q)f34 (q)
+ Nf,T
2
gi dl l , (A.9)
i=1
2 f14 (q)f44 (q)
where the functions fi are defined as in [28], q = e−2πl , and the closed string Hamiltonian
is given by
8 X
X ∞ ∞
X
µ µ µ µ µ µ
Hc = πk + 2π
2
α−l αl + α̃−l α̃l + m µ
ψ−m ψm + ψ̃−m ψ̃m
µ
µ=1 l>0 m>0
+ 2πCc . (A.10)
Here the constant Cc is −1 in the NS–NS sector, and zero in all other sectors. The corre-
sponding open string amplitude is obtained by the modular transformation t = 1/2l, q̃ =
e−πt ,
Z !
dt −(p+1)/2 f38 (q̃) − f28 (q̃) − f48 (q̃)
A=2 (7−p)/2 2
Nf,U t
2t f18 (q̃)
Z !
X 3
dt −(r+si +1)/2 f34 (q̃)f44 (q̃) − f44 (q̃)f34 (q̃)
+ 2 Nf,Tgi
(3−r−si )/2 2
t .
i=1
2t f14 (q̃)f24 (q̃)
(A.11)
This is to be compared with the open string one-loop diagram,
Z
dt 1 + (−1)F 1 + g1 + g2 + g3 −2t Ho
TrNS−R e
2t 2 4
Z
Vp+1 −(p+7)/2 dt −(p+1)/2 f38 (q̃) − f48 (q̃) − f28 (q̃)
= 2 t
(2π)p+1 2t f18 (q̃)
X3
Vr+si +1 −(r+si +3)/2
+ 2
(2π)r+si +1
i=1
Z
dt −(r+si +1)/2 f34 (q̃)f44 (q̃) − f44 (q̃)f34 (q̃)
× t , (A.12)
2t f14 (q̃)f24 (q̃)
where Vp+1 is the (infinite) (p + 1)-dimensional volume of the brane, whilst Vr+si +1 is the
volume of the projection onto the directions unaffected by gi . The open string Hamiltonian
is given by
"∞ ∞
#
X 8 X µ µ X µ
Ho = πp + π
2
α−l αl + mψ−m ψm + πCo ,
µ
(A.13)
µ=1 l>0 m>0
where, in the R sector, l and m run over the positive integers for NN and DD directions,
and over positive half integers for ND directions. In the NS sector, the moding of the
fermions (and therefore the values for m) are opposite to those in the R sector. Co is zero
in the R sector and is (4 − t)/8 in the NS sector, where t is the number of ND directions.
Comparison of Eqs. (A.12) and (A.11) then gives
Vp+1 1
Nf,U
2
= , (A.14)
(2π)p+1 128
Vr+si +1 1
Nf,T
2
= . (A.15)
gi
(2π)r+si +1 8
Consider next the singly fractional D-branes charged under the R–R,T gi field. These
can be thought of as superpositions of two totally fractional D-branes with opposite twisted
charges in the other two twisted sectors. These pairs can move off the fixed points of gj for
j 6= i, such that they lie at positions x µ and −x µ for the directions transverse to the brane
fixed by gj and not by gi . In particular their boundary states will now look like

|D(r; s)i = Nf,U |B(r; s), x µ iNS−NS + |B(r; s), −x µ iNS−NS

+ |B(r; s), x µ iR−R + |B(r; s), −x µ iR−R
+i Nf,Tgi |B(r; s), x µ iNS−NS,Tgi + |B(r; s), −x µ iNS−NS,Tgi

+ |B(r; s), x µ iR−R,Tgi + |B(r; s), −x µ iR−R,Tgi . (A.16)
The above normalisation is consistent with the fact that the cylinder diagram for such
D-branes factorises on
Z
dt 1 + (−1)F 1 + gi −2t Ho
2 TrNS−R e . (A.17)
2t 2 2
Such singly fractional D-branes correspond to, in the covering space, two type II D-branes
stuck at the fixed point of gi . There are four kinds of open strings stretching between two
such D-branes so one would expect a factor of four in front of the trace above. However, the
orbifold identifies certain strings leaving only two independent ones (namely the string that
has endpoint on the same brane and the string that has ends on opposite branes). Similarly
a bulk D-brane is a configuration of four totally fractional D-branes whose twisted charges
cancel. It’s boundary state is a sum over four NS–NS and four R–R boundary states at
positions mapped to one another under the orbifold action. The cylinder diagram for such
a D-brane factorises on
Z
dt 1 + (−1)F −2t Ho
4 TrNS−R e . (A.18)
2t 2
The analysis for the case of the truncated b D-branes is similar. The truncated D-brane of
the type given by Eq. (2.3) charged only under the R–R,T g1 sector, say, has a boundary
state

|b
D(r; s)i = Nt s,U |B(r; s), x µ iNS−NS + |B(r; s), −x µ iNS−NS

+ Nt s,Tg1 |B(r; s), x µ iR−R,Tg1 + |B(r; s), −x µ iR−R,Tg1 , (A.19)
where = ± determines the sign of the R–R,T g1 sector charge and x µ denotes some of the
directions x 3 , x 4 which are transverse to the b
D-brane. The closed string tree diagram now
only produces some of the terms of (A.9), and the corresponding open string amplitude is
Z
dt 1 + g1 (−1)F −2t Ho
4 TrNS−R e
2t 2
Z
Vp+1 −(p+3)/2 dt −(p+1)/2 f38 (q̃) − f28 (q̃)
=4 2 t
(2π)p+1 2t f18 (q̃)
Z
Vr+s1 +1 (1−s1−r)/2 dt −(r+s1 +1)/2 f44 (q̃)f34 (q̃)
−4 2 t . (A.20)
(2π)r+s1+1 2t f14 (q̃)f24 (q̃)
Comparison with the corresponding closed string calculation then gives
Vp+1 1
Nt 21,U = , (A.21)
(2π)p+1 8
Vr+s1 +1
Nt 21,Tg1 = 2. (A.22)
(2π)r+s1 +1
The truncated D-brane of the type given in Eq. (2.4) is charged under the R–R,T gi and
R–R,T gj sectors (i 6= j ) has a boundary state given by
|b
D(r; s)i = Nt d,U |B(r; s)iNS−NS + i Nt d,Tgi |B(r; s)iR−R,Tgi
+ j Nt d,Tgj |B(r; s)iR−R,Tgj + i j Nt d,Tgk |B(r; s)iNS−NS,Tgk , (A.23)
where i = ± determines the sign of the R–R,T gi sector charge and j 6= k 6= i with k =
1, 2, 3. The closed string tree diagram produces only some of the terms of (A.9), and the
corresponding open string amplitude is
Z
dt 1 + gi (−1)F 1 + gj (−1)F −2t Ho
TrNS−R e
2t 2 2
Z
Vp+1 −(p+5)/2 dt −(p+1)/2 f38 (q̃) − f28 (q̃)
= 2 t
(2π)p+1 2t f18 (q̃)
X Z
Vr+sα +1 −(r+sα +1)/2 dt −(r+sα +1)/2 f44 (q̃)f34 (q̃)
− ρ(α) 2 t ,
(2π)r+sα +1 2t f14 (q̃)f24 (q̃)
α∈{i,j,k}
(A.24)
where ρ(i) = ρ(j ) = −ρ(k) = 1. Comparison with the corresponding closed string
calculation then gives
Vp+1 1
Nt 22,U = , (A.25)
(2π)p+1 64
Vr+sα +1 1
Nt 22,Tgα = . (A.26)
(2π)r+sα +1 4
A.2. The compactified case
The construction in the compactified case is essentially the same as in the above
uncompactified case; however there are the following differences.
1. In the localised boundary state (A.3) the integral over compact transverse directions
is replaced by a sum
Z X
ν ν
dk ν eik yν −→ eim yν /Rν , (A.27)
mν ∈Z
where Rν is the radius of the compact x ν direction.

2. In the two untwisted sectors, the ground state is in addition characterised by a winding
number wν for each compact direction that is tangential to the world-volume of the brane.
In the gi twisted sectors, the ground states will also be characterised by winding numbers in
each of the si directions tangential to the world-volume of the brane. The localised bound
state (A.3) then also contains a sum over these winding states
X µ
eiθ wµ , (A.28)
wµ
where θ µ is a Wilson line; as required by orbifold invariance, θ µ ∈ {0, π}.

3. For general si , the contribution in the twisted sectors consists of a sum of terms that
are associated to 2si of the 16 different twisted sectors that define the endpoints of the
world-volume of the brane in the internal space. For convenience we may assume that one
of the 2si fixed points is always the origin.
4. The open and closed string Hamiltonians, Ho and Hc , each acquire an extra term
P
1/4π( µ wµ2 ).
Let us now construct in more detail the boundary state for a fully fractional D(r; s)
brane. This is of the form

|D(r; s)i = Nf,U |B(r; s)iNS−NS + ε|B(r; s)iR−R
s
X
3 X
2i

+ εi Nf,Tgi eiθαi |B(r; s)iNS−NS,Tgαi + |B(r; s)iR−R,Tgαi ,
i=1 αi =1
(A.29)
where αi labels the different fixed points between which the brane stretches (where we
choose the convention that Tgi1 is the twisted sector at the origin), and θαi is the Wilson
line that is associated to the difference of the fixed point αi and the origin. The closed
string tree diagram is now
Z
Ac = dlhB(r; s)|e−lHc |B(r; s)i
Z 8
1 2 f (q) − f28 (q) − f48 (q)
= Nf,U dl l (r−3)/2 3
2 f18 (q)
s1 +s
Y 2 +s3 X −s2 −s3
6−s1Y X
−lπRj2m wj2m
e−lπ(nkm /Rkm )
2
× e
m=1 wjm ∈Z m=1 nkm ∈Z
X
3 Z
2s1 +s2 +s3 −si f34 (q)f24 (q) − f24 (q)f34 (q)
+ Nf,Tg
2
dl l (r−3)/2
i=1
2 i
f14 (q)f44 (q)
Ysi X −lπRj2m wj2m
Yi
2−s X
e−lπ(nkm /Rkm ) ,
2
× e (A.30)
m=1 wjm ∈Z m=1 nkm ∈Z
where Rjm , m = 1, . . . , s1 + s2 + s3 , are the radii of the circles that are tangential to
the world-volume of the brane, and Rkm , i = 1, . . . , 6 − s1 − s2 − s3 , are the radii of
the directions transverse to the brane. Upon the substitution t = 1/2l, using the Poisson
resummation formula (see, for example, [7,28]), this amplitude becomes
Q6−s1 −s2 −s3
Rkm (7−r)/2
Ac = Nf,U
2 m=1
Qs1 +s2 +s3 2
m=1 Rjm
Z
dt −(r+1)/2 f38 (q̃) − f28 (q̃) − f48 (q̃)
× t
2t f18 (q̃)
s1 +s
Y 2 +s3 X −s2 −s3
6−s1Y X
e−2t πnjm /Rjm e−2t πwkm Rkm
2 2 2 2
×
m=1 njm ∈Z m=1 wkm ∈Z
X3 Q2−si
Rkm (1−r)/2 s1 +s2 +s3 −si 2
+ Qm=1
si 2 2 Nf,T gi
i=1 m=1 Rjm
Z
dt −(r+1)/2 f34 (q̃)f44 (q̃) − f44 (q̃)f34 (q̃)
× t
2t f14 (q̃)f24 (q̃)
Y
si X Yi
2−s X
e−2t πnjm /Rjm e−2t πwkm Rkm .
2 2 2 2
×
This is to be compared with the open string amplitude

Z
dt 1 + (−1)F 1 + g1 + g2 + g3 −2t Ho
TrNS−R e
2t 2 4
Z
Vr+1 −(r+1)/2 dt −(r+1)/2 f38 (q̃) − f48 (q̃) − f28 (q̃)
= 2 t
8(2π)r+1 2t f18 (q̃)
s1 +s
Y 2 +s3 X −s2 −s3
6−s1Y X
e−2t πnjm /Rjm e−2t πwkm Rkm
2 2 2 2
×
X
3 Z
Vr+1 dt −(r+1)/2 f34 (q̃)f44 (q̃) − f44 (q̃)f34 (q̃)
+ 2−(r+1)/2 t
2(2π)r+1 2t f14 (q̃)f24 (q̃)
i=1
Ysi X Yi
2−s X
e−2t πnjm /Rjm e−2t πwkm Rkm .
2 2 2 2
× (A.31)
By comparison this then fixes the normalisation constants as

Qs1 +s2 +s3
Vr+1 1 Rjm
Nf,U =
2
r+1
m=1
Q6−s1 −s2 −s3 , (A.32)
(2π) 128 Rkm
m=1
Qsi
Vr+1 24−(s1 +s2 +s3 −si ) m=1 Rjm
Nf,Tg
2
= Q2−si . (A.33)
i
(2π)r+1 128 Rk
m=1 m
The extension of this to the singly fractional and bulk D-branes is obvious.
The analysis for the truncated D-branes is almost identical. The boundary state of a
truncated D-brane of the type given in Eq. (2.3) is the truncation of (A.29) to the untwisted
NS–NS and the gi -twisted R–R sectors with the addition of boundary states at mirror
positions as in Eq. (A.19). The open string amplitude contains also only the corresponding
terms. Furthermore, since the projection operator is now 6 4 × 12 (1 + gi (−1)F ) each of the
terms that appears is sixteen times as large as in the fractional case above. This implies that
the relevant normalisation constants are given as
6 Cf. Eq. (A.20).

Qs1 +s2 +s3

Vr+1 1 Rjm
Nt 21,U = Q
m=1
, (A.34)
(2π)r+1 8 6−s1 −s2 −s3 Rkm
m=1
Qsi
Vr+1 24−(s1+s2 +s3 −si ) m=1 Rjm
Nt 21,Tgi = r+1 Q2−si . (A.35)
(2π) 8 Rkm
m=1
The second type of truncated bD-brane (Eq. (2.4)) contains the untwisted NS–NS , twisted
R–R,T gi , R–R,T gj and NS–NS,T gk boundary states. The projection operator is
4 (1 + gi (−1) )(1 + gj (−1) ) thus fixing the normalisations to be
1 F F
Qs1 +s2 +s3

Vr+1 1 Rjm
Nt 2,U =
2
r+1
m=1
Q6−s1 −s2 −s3 , (A.36)
(2π) 64 Rkm
m=1
Qsα
Vr+1 24−(s1+s2 +s3 −sα ) m=1 Rjm
Nt 2,Tgα =
2
r+1 Q2−sα , (A.37)
(2π) 64
m=1 Rkm
where α ∈ {i, j, k}.
Appendix B. Consistency conditions of boundary states
In this appendix we discuss the invariance of the boundary states under the GSO and
orbifold actions. Since the action of the orbifold group is purely geometrical the GSO
projection is the same in twisted and untwisted sectors namely we have
1 e
NS–NS : 1 + (−1)F 1 + (−1)F , (B.1)
4
1 e
R–R : 1 + (−1)F 1 ∓ (−1)F , (B.2)
4
for types IIA and IIB, respectively.
The GSO invariance of each of the sectors’ boundary states was computed in detail
in [29]. Here we simply restate those results in terms of r and si . We shall work in the
light-cone gauge with the directions x 0 , x 9 always Dirichlet [47]. In the untwisted NS–NS
sector it is easy to see that
N
|B(r; s)iNS−NS = |B(r; s), +iNS−NS − |B(r; s), −iNS−NS (B.3)
2
is GSO invariant for all r, si . The exact values of the normalisation constants of the
boundary states will depend on the kind of D-brane the boundary state is a part of, and
are computed in the appendix. In the untwisted R–R sector
4iN
|B(r; s)iR−R = |B(r; s), +iR−R + |B(r; s), −iR−R (B.4)
2
is a GSO invariant boundary state if r +s1 +s2 +s3 is even/odd for type IIA/B, respectively.
For the NS–NS,T g1 sector we find that

|B(r; s)iNS−NS,Tg1 = NT |B(r; s), +iNS−NS,Tg1 + |B(r; s), −iNS−NS,Tg1 (B.5)
is a GSO invariant boundary state provided s2 + s3 is even, while

|B(r; s)iR−R,Tg1 = iNT |B(r; s), +iR−R,Tg1 + |B(r; s), −iR−R,Tg1 (B.6)
is GSO invariant for r + s1 even/odd for type IIA/IIB. Similarly in the NS–NS,T g2 sector
s1 + s3 is to be even and for the R–R,T g2 sector r + s2 has to be even/odd for type IIA/IIB.
Finally for the NS–NS,T g3 s1 + s2 is to be even and for the R–R,T g2 sector r + s3 has to
be even/odd for type IIA/IIB.
Next one requires the boundary states to be invariant under gi . As usual [29] this
places no restrictions on the untwisted NS–NS sector, and in the untwisted R–R sector
it requires for s1 + s2 , s2 + s3 and s1 + s3 to be all even. The R–R,T gi boundary state
has no restrictions placed on it by gi since it has no zero modes in the directions which
gi inverts, while the NS–NS,T gi boundary state’s invariance under gi is equivalent to the
GSO condition on that boundary state [29].
New non-trivial restrictions arise by requiring the gi -twisted sector’s boundary state be
invariant under gj , j 6= i. We consider the invariance of NS–NS,T g1 and R–R,T g1 under
g2 in some detail. The other conditions will follow in a similar way. Since the NS–NS,T
g1 sector has zero modes in the directions x 7 and x 8 (as well as x 5 and x 6 ) g2 will have a
non-trivial representation on these zero-modes, given by 7
Y √ µ Y √ µ
g2 = 2 ψ0 2 ψ̃0 . (B.7)
µ=7,8 µ=7,8
This operator squares to one. It is not difficult to see that
g2 |B(r; s), ηi0NS−NS,Tg1 = (−1)s3 |B(r; s), ηi0NS−NS,Tg1 (B.8)
and hence that s3 has to be even. 8
Similarly since the R–R,T g2 sector has zero modes in
3 4
directions x and x , g2 has a non-trivial representation on this sector as
Y √ µ Y √ µ
g2 = 2 ψ0 2 ψ̃0 , (B.9)
µ=3,4 µ=3,4
and so
g2 |B(r; s), ηi0R−R,Tg1 = (−1)s1 |B(r; s), ηi0R−R,Tg1 . (B.10)
Thus s1 has to be even. 9 Performing a similar analysis for the other twisted sectors one
finds that in type IIB the following boundary states are GSO and orbifold invariant
|B(r; s)iNS−NS , for all r and si ,
|B(r; s)iR−R , for either r odd and si all even or r even and si all odd,
|B(r; s)iNS−NS,Tgi , for sj , sk even, j, k 6= i,
|B(r; s)iR−R,Tgi , for r odd and si even.
For type IIA the conditions are the same for si but r has to be even.
7 We use the same convention here for g as was used for I in [29]. Since this is a supersymmetric theory this
2 2
can be viewed as a condition for supersymmetry preservation by the orbifold action.
8 Note that in the notation of Eqs. (B.4) and (B.5) Appendix B in [29] for the NS–NS,T g state a = b = 1.
1
9 Following Eqs. (B.12) and (B.13) of [29] â = b̂ = 1 for the R–R,T g ground state.
1
References
[1] A. Sen, Non-BPS states and branes in string theory, APCTP winter school lectures, hep-
th9904207.
[2] L. Lerda, R. Russo, Stable non-BPS states in string theory: a pedagogical review, hep-
th9905006.
[3] J.H. Schwarz, TASI lectures on non-BPS D-brane systems, hep-th9908144.
[4] R. Minasian, G. Moore, K-theory and Ramond–Ramond charge, JHEP 11 (1997) 002, hep-
th9710230.
[5] E. Witten, D-branes and K-theory, JHEP 9812 (1998) 019, hep-th9810188.
[6] P. Hořava, Type IIA D-branes, K-theory, and matrix theory, Adv. Theor. Math. Phys. 2 (1998)
1373, hep-th9812135.
[7] A. Sen, Stable non-BPS bound states of BPS D-branes, JHEP 9808 (1998) 010, hep-th9805019.
[8] A. Sen, Tachyon condensation on the brane–antibrane system, JHEP 9808 (1998) 012, hep-
th9805170.
[9] A. Sen, Universality of the tachyon potential, JHEP 9912 (1999) 027, hep-th9911116.
[10] A. Sen, B. Zwiebach, Tachyon condensation in string field theory, JHEP 0003 (2000) 002, hep-
th9912249.
[11] N. Berkovits, The tachyon potential in open Neveu–Schwarz string field theory, hep-th0001084.
[12] N. Berkovits, A. Sen, B. Zwiebach, Tachyon condensation in superstring field theory, hep-
th0002211.
[13] N. Moeller, W. Taylor, Level truncation and the tachyon in open bosonic string field theory,
hep-th0002237.
[14] J.A. Harvey, P. Hořava, P. Kraus, D-sphalerons and the topology of string configuration space,
JHEP 0003 (2000) 021, hep-th0001143.
[15] J.A. Harvey, P. Kraus, D-branes as unstable lumps in bosonic open string field theory, hep-
th0002117.
[16] M. Srednicki, IIB or not IIB, JHEP 9808 (1998) 005, hep-th9807138.
[17] P. Yi, Membranes from five-branes and fundamental strings from Dp-branes, Nucl. Phys. B 550
(1999) 214, hep-th9901159.
[18] A. Sen, Supersymmetric world-volume action for non-BPS D-branes, JHEP 9910 (1999) 008,
hep-th9909062.
[19] O. Bergman, K. Hori, P. Yi, Confinement on the brane, hep-th0002223.
[20] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Loop corrections to superstring equations of
motion, Nucl. Phys. B 308 (1988) 221.
[21] J. Polchinski, Y. Cai, Consistency of open superstring theories, Nucl. Phys. B 296 (1988) 91.
[22] O. Bergman, M.R. Gaberdiel, Stable non-BPS D-particles, Phys. Lett. B 441 (1998) 133, hep-
th9806155.
[23] A. Sen, BPS D-branes on non-supersymmetric cycles, JHEP 9812 (1998) 021, hep-th9812031.
[24] O. Bergman, M.R. Gaberdiel, Non-BPS states in heterotic — type IIA duality, JHEP 9903
(1999) 013, hep-th9901014.
[25] S. Gukov, K-theory, reality, and orientifolds, hep-th9901042.
[26] M. Frau, L. Gallot, A. Lerda, P. Strigazzi, Stable non-BPS D-branes in type I string theory,
Nucl. Phys. B 564 (2000) 60, hep-th9903123.
[27] D.-E. Diaconescu, J. Gomis, Fractional branes and boundary states in orbifold theories, hep-
th/9906242.
[28] M.R. Gaberdiel, A. Sen, Non-supersymmetric D-brane configurations with Bose–Fermi
degenerate open string spectrum, JHEP 9911 (1999) 008, hep-th9908060.
[29] M.R. Gaberdiel, B. Stefański jr., Dirichlet branes on orbifolds, hep-th9910109.
[30] M. Mihailescu, K. Oh, R. Tatar, Non-BPS branes on a Calabi–Yau threefold and Bose–Fermi
degeneracy, JHEP 0002 (2000) 019, hep-th9910249.
[31] R. Russo, C.A. Scrucca, On the effective action of stable non-BPS branes, hep-th9912090.
[32] S. Mukhi, N.V. Suryanarayana, D. Tong, Brane–antibrane constructions, JHEP 0003 (2000)
015, hep-th0001066.
[33] E. Eyras, S. Panda, The spacetime life of a non-BPS D-particle, hep-th0003033.
[34] O. Bergman, E.G. Gimon, P. Horava, Brane transfer operations and T-duality of non-BPS states,
JHEP 9904 (1999) 010, hep-th/9902160.
[35] A. Sen, Stable non-BPS states in string theory, JHEP 9806 (1998) 007, hep-th9803194.
[36] T. Dasgupta, B. Stefański jr., Non-BPS states and heterotic — type I’ duality, Nucl. Phys. B 572
(2000) 95, hep-th9910217.
[37] L. Gallot, A. Lerda, P. Strigazzi, Gauge and gravitational interactions of non-BPS D-particles,
hep-th0001049.
[38] C. Vafa, E. Witten, On orbifolds with discrete torsion, J. Geom. Phys. 15 (1995) 189, hep-
th9409188.
[39] M.R. Douglas, D-branes and discrete torsion, hep-th9807235.
[40] B.R. Greene, D-brane topology changing transitions, Nucl. Phys. B 525 (1998) 284, hep-
th9711124.
[41] S. Mukhopadhyay, K. Ray, Conifolds from D-branes, Phys. Lett. B 423 (1998) 247, hep-
th9711131.
[42] M.R. Douglas, B. Fiol, D-branes and discrete torsion II, hep-th9903031.
[43] A. Sen, SO(32) spinors of type I and other solitons on brane–antibrane pair, JHEP 9809 (1998)
023, hep-th9808141.
[44] J. Majumder, A. Sen, Vortex pair creation on brane–antibrane pair via marginal deformation,
hep-th0003124.
[45] G. Segal, Equivariant K-theory, Inst. Hautes Etudes Sci. Publ. Math. 34 (1968) 129.
[46] G. Segal, Private communication.
[47] M.B. Green, M. Gutperle, Light-cone supersymmetry and D-branes, Nucl. Phys. B 476 (1996)
484, hep-th9604091.
Bulk fields in dilatonic and self-tuning

flat domain walls
Donam Youm
Theory Division, CERN, CH-1211, Geneva 23, Switzerland
Received 24 February 2000; revised 20 June 2000; accepted 21 August 2000
Abstract
We study the Kaluza–Klein zero modes of massless bulk fields with various spins in the
background of dilatonic and self-tuning flat domain walls. We find that the zero modes of all
the massless bulk fields in such domain wall backgrounds are normalizable, unlike those in the
background of the non-dilatonic domain wall with infinite extra space of Randall and Sundrum. In
particular, gravity in the bulk of dilatonic domain walls is effectively compactified to the Einstein
gravity with vanishing cosmological constant and nonzero gravitational constant in one lower
dimensions for any values of dilaton coupling parameter, provided the warp factor is chosen to
decrease on both sides of the domain wall, in which case the tension of the domain wall is positive.
However, unexpectedly, for the self-tuning flat domain walls, the cosmological constant of the
zero mode effective gravity action in one lower dimensions does not vanish, indicating the need
for additional ingredient or modification necessary in cancellation of the unexpected cosmological
constant in the graviton zero mode effective action.  2000 Elsevier Science B.V. All rights reserved.
1. Introduction
Recently, theories with extra spatial dimensions have been actively studied as solutions
to the hierarchy problem in particle physics. In such scenario, all the fields of Standard
model are assumed to live within the worldvolume of a brane, whereas gravity can freely
propagate in the extra space as well as within the worldvolume of the brane. Although
the earlier proposal [1,2] attempts to solve the hierarchy problem by assuming large
enough extra space, the hierarchy problem is recast into the problem of large ratio between
the (fundamental) TeV Planck scale and the compactification scale of the extra spatial
dimensions. Later proposal [3–5] by Randall and Sundrum (RS) relies on the exponentially
decreasing warp factor in the metric of the non-factorizable spacetime for solving the
hierarchy problem. The RS model does not require large enough extra dimensions for
E-mail address: donam.youm@cern.ch (D. Youm).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 3 1 - 9
316 D. Youm / Nuclear Physics B 589 (2000) 315–336
solving the hierarchy problem; it is the decreasing warp factor that accounts for much
smaller electroweak scale compared to the Planck scale. Another novelty of the RS model
is that the extra dimensions can even be infinite in size because gravity in the bulk of the
RS domain wall is effectively localized [4] around the domain wall.
In our previous works [6,7], we showed that the RS type scenario can be extended to
the dilatonic domain walls. In fact, one is bound to consider dilatonic domain walls if
one wishes to embed the RS type scenarios within string theories, since most of the five-
dimensional domain walls in string theories are dilatonic. We showed that the warp factor
decreases [6] (and becomes zero at finite distance away from the wall) and the gravity in
the bulk of the dilatonic domain walls is effectively compactified to the Einstein gravity
with vanishing cosmological constant and nonzero gravitational constant in one lower
dimensions [7], if the cosmological constant is negative (Λ = −2Q2 /1 > 0, therefore
1 < 0, in our convention for the action). In this paper, we show that even for the positive
cosmological constant (Λ < 0 in our convention and therefore 1 > 0) the bulk gravity is
effectively compactified to the Einstein gravity with vanishing cosmological constant and
nonzero gravitational constant in one lower dimensions, provided the warp factor is chosen
to decrease on both sides of the wall and the extra spatial dimension is cut off through
the introduction of another domain wall. Note, for any values of the dilaton coupling
parameter (therefore, for any values of 1), the warp factor can be chosen to be increasing
or decreasing within the finite allowed extra spatial coordinate interval around the wall.
(In our previous works, we considered the case when the warp factor decreases [increases]
for 1 < 0 [1 > 0].) Therefore, the RS type scenario can be realized for dilatonic domain
walls with any values of the dilaton coupling parameter, so for any dilatonic domain walls
in string theories.
The RS scenario relies on the fine-tuning of the domain wall tension, whose value is
determined by the five-dimensional cosmological constant. It is pointed out [8] that the
fine-tuned value of the domain wall tension is required by supersymmetry. However, so far
it has been observed [8–14] that supersymmetry rather requires increasing warp factor at
least on one side of the wall, instead of decreasing warp factor on both sides. Furthermore,
it is not guaranteed that the fine-tuned value of the domain wall tension does not receive
quantum corrections after the SUSY breaking. Recently, new type of domain wall solutions
which do not suffer from such problem were constructed [15,16]. These solutions, called
“self-tuning flat domain wall”, are obtained within the model with bulk dilaton but without
bulk cosmological constant term. Novelty of such domain wall solutions is that a static
domain wall solution with the Poincaré invariance in one lower dimensions exists for any
values of the domain wall tension. So, even if the quantum effect corrects the domain
wall tension, the Poincaré invariance is not disturbed. With a choice of the warp factor
that has singularity at finite distance away from the wall on both sides, one therefore
would expect that such domain wall effectively compactifies the five-dimensional gravity to
four-dimensional gravity with vanishing cosmological constant regardless of the quantum
corrections on the domain wall tension.
It is the purpose of this paper to study the Kaluza–Klein (KK) zero modes of the massless
fields with various spins in the bulk of the dilatonic and the self-tuning flat domain walls.
D. Youm / Nuclear Physics B 589 (2000) 315–336 317
(The previous works on bulk fields in the non-dilatonic domain wall can be found for
example in Refs. [17–24].) Although the brane world scenarios assume that only gravity
lives in the bulk and the remaining fields are confined within the brane worldvolume, it
would be interesting to study various fields in the bulk. One of the reasons is that if we
want to embed the brane world scenarios within string theories, we have to consider bulk
fields compactified from ten or eleven dimensions, unless we want to truncate them ad hoc
just for the purpose of letting only gravity (and the dilaton for the dilatonic domain wall
case) live in the bulk. We find that the KK zero modes of massless fields of various spins in
the bulk of the dilatonic and the self-tuning flat domain walls are normalizable, whereas the
KK zero modes of only the massless spin-0 and spin-2 fields in the bulk of non-dilatonic
RS domain wall with infinite extra spatial dimensions [4,5] are normalizable. In general,
we find that the KK zero modes for the integer spin bulk fields are independent of the extra
spatial coordinate, whereas those of the half-integer spin fields depend on it. An unexpected
result is that, contrary to the claims in Refs. [15,16], the KK zero mode effective action (in
one lower dimensions) for the gravity in the bulk of the self-tuning flat domain wall has
non-vanishing cosmological constant term. This seems to be an indication of the need
for additional yet unknown ingredient in the picture of the self-tuning flat domain wall
necessary in cancellation of the unexpected cosmological constant term in the effective
action.
The paper is organized as follows. In Section 2, we discuss the dilatonic and the
self-tuning flat domain walls. We reparametrize the dilatonic domain walls, which we
previously studied, in terms of the bulk cosmological constant. We rederive the self-tuning
flat domain walls with different parametrization (from the one in Ref. [16]) which we find
more convenient. In Section 3, we obtain the KK zero modes for the bulk fields with various
spins and their effective actions in one lower dimensions.
2. Domain wall solutions
In this section, we discuss the dilatonic domain walls that we studied previously
and the self-tuning flat domain wall solutions that are constructed in Refs. [15,16].
Although detailed derivation is given in Ref. [16], we rederive the self-tuning flat
domain wall solutions because we wish to use different parametrization of the solutions,
which we find more convenient. Also, even if only five-dimensional domain walls
are phenomenologically of interest, we derive the domain wall solutions in arbitrary
dimensions just for the purpose of generality and because such solutions might be useful
for other studies.
Generally, the total action is the sum of the D-dimensional action Sbulk in the bulk of the
domain wall and the (D − 1)-dimensional action SDW on the domain wall worldvolume:
S = Sbulk + SDW . (1)
For the domain wall solutions under consideration in the paper, the D-dimensional action
contains the bulk action for the domain wall solutions:
Z
1 √ 4
Sbulk ⊃ dDx −G RG − ∂M φ∂ M φ + e−2aφ Λ , (2)
2
2κD D−2
and the (D − 1)-dimensional action contains the following worldvolume action for the
domain wall solutions:
Z
√
SDW ⊃ − d D−1 x −γ f (φ), (3)
where γ is the determinant of the induced metric γµν = ∂µ XM ∂ν XN GMN on the domain
wall worldvolume, M, N = 0, 1, . . . , D − 1 and µ, ν = 0, 1, . . . , D − 2. Note, for the
purpose of following the notation of our previous works [6,7], we are using the different
convention for the sign of the cosmological constant Λ and the sign and the value of the
dilaton coupling parameter a from that in Ref. [16]. So, in our case, the positive [the
negative] Λ (with a = 0) corresponds to the AdS space [dS space]. We are interested in
finding solutions with the Poincaré invariance in (D − 1) dimensions. The general Ansätze
for the fields with such symmetry are

GMN dx M dx N = e2A(y) −dt 2 + dx12 + · · · + dxD−2
2
+ dy 2, (4)
and φ = φ(y).
The Λ 6= 0 6= a case, considered in Ref. [16], is nothing but the extreme dilatonic domain
wall solutions studied in our previous works [6,7]. In this case, f (φ) in the worldvolume
action has to take a specific form f (φ) = σDW e−aφ with the energy density σDW of the
domain walls taking the fine-tuned value determined by the bulk cosmological constant
Λ and the dilaton coupling parameter a. It is observed in Ref. [6] that the extra spatial
coordinate y terminates at finite nonzero value due to the singularity. It is explicitly
shown in Ref. [7] that the gravity in the bulk of the dilatonic domain wall is effectively
compactified to the Einstein gravity in one lower dimensions with nonzero gravitational
constant and vanishing cosmological constant. The solution, reparametrized in terms of
the bulk cosmological constant Λ and the dilaton coupling parameter a, has the following
form:
8 1
e2A = (Ky + 1) (D−2)2 a2 , φ= ln(Ky + 1) + C, (5)
a
where C is an integration constant and
s
(D − 2)a 2 −aC D−2
K =± e Λ. (6)
2 4(D − 1) − a 2 (D − 2)2
The requirement of the term inside the square root in Eq. (6) to be positive fixes the sign
of the cosmological constant Λ to be positive [negative] for a 2 < 4(D − 1)/(D − 2)2
[a 2 > 4(D − 1)/(D − 2)2 ]. This is in accordance with the expression Λ = −2Q2 /1,
where 1 ≡ (D − 2)a 2 /2 − 2(D − 1)/(D − 2), that we used in our previous works [6,
7]. The sign ± in Eq. (6) has to be chosen such that the spacetime metric (4) with the warp
factor in Eq. 5) has a singularity at finite y, if we want the gravity in the bulk of the domain
wall to be effectively compactified. So, we choose the plus [minus] sign for the region
y < 0 [y > 0]. With this choice of the signs, the energy density of the wall, determined by
solving the boundary condition at y = 0, takes the following positive value:
s
4 D−2
σDW = 2 Λ. (7)
κD 4(D − 1) − a 2 (D − 2)2
With a choice of the same signs on both sides of the wall, σDW = 0. With a choice such that
there is no singularity at y 6= 0, σDW is negative with the same absolute value as Eq. (7).
These properties of the extreme dilatonic domain wall are essentially what we discussed
in our previous works [6,7], except the choice of the sign ± in Eq. (6). In our previous
works, we considered the case when the metric has the singularity at finite y for 1 < 0, but
no singularity at finite y for 1 > 0. However, we see from the above that there is another
choice of signs which gives the opposite singularity properties. So, the correct statement
has to be that for any values of the dilaton coupling parameter a, one can choose the
warp factor to be decreasing on both sides of the wall by choosing the sign ± in Eq. (6)
such that K < 0 [K > 0] for y > 0 [y < 0], in which case σDW > 0 and the there are
singularities on both sides of the wall. However, we note that for the 1 > 0 case (or the
a 2 > 4(D − 1)/(D − 2)2 case), the cosmological constant Λ has the opposite sign, i.e., the
bulk spacetime is dS-like.
Now, we consider the Λ = 0 case. In deriving the solutions, we choose the static
gauge for the domain wall worldvolume action, so γµν = δµM δνN GMN . With the (D − 1)-
dimensional Poincaré invariant Ansätze for the fields, the equation of motion for the dilaton
φ takes the following form:
4 1
(D − 1)A0 φ 0 + φ 00 = f 0 (φ)δ(y), (8)
D − 2 κD
2
and the Einstein’s equations can be brought to the following forms:

1 2
(D − 2)(D − 1)(A0 )2 − (φ 0 )2 = 0, (9)
2 D−2

1 00 4 0 2
(D − 2)A + (φ ) + f (φ)δ(y) = 0, (10)
2
κD D−2
where the primes on φ and A mean differentiations with respect to y and the prime on f (φ)
means differentiation with respect to φ. Eq. (9) implies the following relation between φ
and A:
2 1
A0 = η √ φ0 (η ≡ ±1). (11)
D−2 D−1
So, Eqs. (8) and (10) can be brought to the following forms:

4 1 2 √
η D − 1 (φ ) + φ = f 0 (φ)δ(y),
0 2 00
(12)
D − 2 κD 2 D−2

1 2 00 4 0 2
η √ φ + (φ ) = −f (φ)δ(y). (13)
κD2
D−1 D−2
In the bulk (y 6= 0), Eqs. (12) and (13) take the same form and one can solve one of
them to obtain the general solution for φ. In Ref. [16], the solution is parametrized in the
following way:

D−2 1 2 √
φ=η √ ln D − 1 y + c + d, (14)
2 D−1 D−2
but we find it more convenient to parametrize the solution in the following way:
D−2 1
φ=η √ ln(Ky + 1) + C. (15)
2 D−1
Then, from Eq. (11) we obtain the following standard form of the metric warp factor:
2
e2A = (Ky + 1) D−1 . (16)
Note, Ref. [16] considers the possibility of having an arbitrary constant factor in front of
this warp factor, which results from the integration constant in integrating Eq. (11) with
respect to y as well as from the integration constant C in Eq. (15). However, such constant
factor can be absorbed by rescaling the (D − 1)-dimensional coordinates x µ in the metric.
The only effect of such constant factor on the KK zero modes of massless bulk fields is
values of their normalization constants.
Note, η’s in the above equations and the dilaton solution can take any (independent)
signs on each side of the domain wall. So, we have two sets of Eqs. (12) and (13), one for
each side, with η replaced by η+ [η− ] for the set of equations in the region y > 0 [y < 0].
And we have the following expressions for the warp factor:
( 2
(K+ y + 1) D−1 , y > 0,
e =
2A
2 (17)
(K− y + 1) D−1 , y < 0,
and the dilaton

 D−2 1

 η+ √ ln(K+ y + 1) + C, y > 0,
φ=
2 D−1 (18)

 D−2 1
 η− √ ln(K− y + 1) + C, y < 0,
2 D−1
where we have imposed the continuity of φ at y = 0, i.e., φ(0+ ) = φ(0− ). In the following,
we impose the boundary conditions at y = 0 on the general solutions (17) and (18) to
determine the integration constants K± in terms of the parameters in the actions (2) and (3)
and the other integration constant.
First, we consider the case when η’s in the above equations and the solutions on the
two sides of the domain wall have the opposite signs. Namely, we choose η+ = −η−
= η, where η = ±1. By imposing the boundary conditions on φ at y = 0, we obtain the
following expressions for K± :
√
1 2 D−1 0 D−1
K+ = κD η f (C) − f (C) ,
2 2 D−2
√
1 2 D−1 0 D−1
K− = κD η f (C) + f (C) . (19)
2 2 D−2
With a choice f (φ) = σDW ebφ , K± take the following forms:

√
1 2 D−1 D−1
K+ = κD σDW e bC
η b− ,
2 2 D−2
√
1 2 D−1 D−1
K− = κD σDW ebC η b+ . (20)
2 2 D−2
From these expressions for K± , we see that a non-trivial solution does not exist when
√
b = ±2 D − 1/(D − 2). As we will see in the following, non-trivial solutions for this
case exist when η’s on the two sides of the wall have the same signs.
Second, we consider the case when η’s on the two sides of the domain wall have the
same signs. Namely, we choose η+ = η− = η, where η = ±1. By imposing the boundary
conditions on φ at y = 0, we obtain the following relation between K+ and K− :
√
D−1 2 0 D−1 2
K+ − K− = η κD f (C) = − κ f (C). (21)
2 D−2 D
With a choice f (φ) = σDW ebφ , the relation becomes:
√
D−1 2 D−1 2
K+ − K− = η κD bσDW ebC = − κ σDW ebC , (22)
2 D−2 D
from which we see that b is fixed to take the following values:
√
D−1
b = −2η . (23)
D−2
Note, we have freedom of choosing any values of K± as long as the constraint (21) or (22)
is satisfied.
We now comment on novelty [15,16] of the domain wall solution with Λ = 0. First of
all, as we can see from the explicit expressions for K± , the (D − 1)-dimensional Poincaré
invariant solution exists for any values of the energy density σDW of the domain wall (and
the parameter b). (In fact, the independent free parameters 1 , which can take any values,
of the Λ = 0 solutions are the parameters of f (φ), i.e., σDW and b for the f (φ) = σDW ebφ
case, and C.) So, σDW (and b) needs not be fine-tuned and the quantum correction on
σDW does not disturb the (D − 1)-dimensional Poincaré invariance. Second, in general
K+ 6= −K− , namely the solution has no Z2 invariance under y → −y. The Z2 invariance
is not possible for the η+ = −η− case, but one can choose the values of K± such that
the Z2 invariance is achieved for the η+ = η− case. So, the Z2 invariant domain wall
solution obtained in Ref. [15] is a special case of the self-tuning flat domain wall solution
with η+ = η− constructed in Ref. [16]. All of these special features of the self-tuning flat
domain wall solutions are manifestly recognizable, if we use our parametrization (15) of
the solutions, instead of the one (14) used in Ref. [16]. Also, the solutions take simple and
attractive forms with our parametrization.
1 Another free parameter, namely the integration constant which results from integrating Eq. (11) with respect
to y, can be removed by rescaling the (D − 1)-dimensional coordinates x µ , as we mentioned previously.
3. The Kaluza–Klein zero modes of bulk fields
In this section, we study the KK zero modes of massless fields in the bulk of various
domain walls discussed in the previous section. Although we are interested in the KK
zero modes of massless bulk fields, we shall first obtain general equations satisfied by
any KK modes of massive (for the spin-0 and spin-1 cases) bulk fields, and then we shall
restrict ourselves to the special case of the KK zero modes of massless bulk fields. In
this section, we consider the five-dimensional domain walls, only, since only these are
phenomenologically of interest. So, in the following we rewrite the explicit domain wall
solutions specifically for the D = 5 case:
• Dilatonic domain wall:
The warp factor and the dilaton for the domain wall with singularities at finite y (therefore,
decreasing warp factor) on both sides of the wall are given by
8 1
e2A = (1 − K|y|) 9a2 , φ= ln(1 − K|y|) + C, (24)
a
where the parameter K and the tension σDW of the wall take the following fixed values
determined by the bulk cosmological constant Λ:
r r
3 3 4 3
K = a 2 e−aC Λ, σDW = Λ. (25)
2 16 − 9a 2 κ52 16 − 9a 2
• Self-tuning flat domain wall:
The warp factor and the dilaton are given by
(
1
(K+ y + 1) 2 , y > 0,
e =
2A
1
(K− y + 1) 2 , y < 0,

 3
 η+ ln(K+ y + 1) + C, y > 0,
φ= 4 (26)

 η− 3 ln(K− y + 1) + C, y < 0.
4
First, for the η+ = −η− = η case, the parameters K± are given in general by
1
K+ = κ52 ηf 0 (C) − 43 f (C) ,
2
1
K− = κ52 ηf 0 (C) + 43 f (C) . (27)
2
With a choice f (φ) = σDW ebφ ,
1
K+ = κ52 σDW ebC ηb − 43 ,
2
1
K− = κ52 σDW ebC ηb + 43 . (28)
2
Second, for the η+ = η− = η case, the parameters K± are in general constrained to satisfy
4
K+ − K− = ηκ52 f 0 (C) = − κ52 f (C). (29)
3
With a choice f (φ) = σDW ebφ ,

4
K+ − K− = ηκ52 bσDW ebC = − κ52 σDW ebC . (30)
3
3.1. Scalar field
The action for the bulk scalar field Φ(x µ , y) is

Z
1 √
Sbulk ⊃ d 4 x dy −G GMN ∂M Φ∂N Φ − m2 Φ 2 , (31)
2
where GMN is the metric (4) for the domain wall solution. From this action, we obtain the
following the equation of motion for the scalar:
√ √
∂M −G GMN ∂N Φ + −G m2 Φ = 0, (32)
which takes the following form after the metric (4) is substituted:

e2A ηµν ∂µ ∂ν Φ + ∂y e4A ∂y Φ + m2 e4A Φ = 0. (33)
To consider the KK mode of Φ with mass mn , only, we decompose Φ as
Φ(x µ , y) = ϕn (x µ )fn (y), (34)
and require ϕn (x µ ) to satisfy the following Klein–Gordon equation for a scalar with mass
mn in flat four-dimensional spacetime:
µν
η ∂µ ∂ν + m2n ϕn = 0. (35)
Then, the equation of motion (33) for Φ reduces to the following form of Sturm–Liouville
equation:

∂y e4A ∂y fn + m2 e4A fn = m2n e2A fn . (36)
The operator L = ∂y (e4A ∂y ) + m2 e4A is self-adjoint, provided the boundary condition
fn∗ e4A fm0 |y=a = 0 is satisfied, where a 6 y 6 b is the interval in which the domain wall
y=b
metric (4) is well-defined. In this case, the eigenvalues m2n are real and the eigenfunctions
fn with different eigenvalues are orthogonal to each other with respect to the weighting
Ra
function w(y) = e2A , i.e., b dyfm∗ (y)fn (y)w(y) = 0 for m2m 6= m2n .
In term of a new y-dependent function fñ = e2A fn , Eq. (36) takes the following form of
the Schrödinger equation:
d 2 fñ
− + V (y)fñ = m2 fñ (37)
dy 2
with the potential

V (y) = 2 A00 + 2(A0 )2 + 12 m2n e−2A . (38)
From now on, we shall be interested in only the zero mode (m0 = 0) of the massless bulk
scalar (m = 0).
First, we consider the dilatonic domain wall solution (24). The potential (38) in the
Schrödinger equation (37) takes the following form:
8K 2 8 − 9a 2 16K
V (y) = − δ(y). (39)
81a 4 (1 − K|y|)2 9a 2
The Z2 invariant solution to the Schrödinger equation that satisfies the boundary condition
8
f˜00 (0+ )− f˜00 (0− ) = − 16K
9a 2 f˜0 (0) has the form f˜0 (y) ∼ (1−K|y|) 9a2 . So, the KK zero mode
is constant: f0 (y) = e−2A f˜0 (y) = constant. This constant zero mode is normalizable 2
with respect to the weighting function w(y) = e2A . The normalization constant is N0 =
q R 1/K
( 12 + 9a4 2 )K, i.e., −1/K dy f0∗ (y)f0 (y)w(y) = 1 with f0 (y) = N0 and w(y) = e2A .
Indeed, by substituting the KK zero-mode Φ = ϕ0 f0 into the action (31) for the massless
bulk scalar (m = 0), we obtain the following action for the massless scalar in the flat four-
dimensional spacetime:
Z Z
1/K Z
1 √ 1
d x dy −G GMN ∂M Φ∂N Φ =
4
dy e2A f02 dx 4 ηµν ∂µ ϕ0 ∂ν ϕ0
2 2
−1/K
Z
1
= dx 4 ηµν ∂µ ϕ0 ∂ν ϕ0 . (40)
2
As we pointed out in the previous section, had we chosen to have a constant factor
(due to the integration constants) in the warp factor e2A , the normalization constant N0
would have
R had dependence on the constant factor (as can be seen from the normalization
relation dyf02 e2A = 1), but the effective action (40) for the KK zero mode in one lower
dimensions does not depend on the constant factor. This generally holds for the case of
the self-tuning flat domain walls and for other bulk fields to be discussed in the following
subsections.
Second, for the self-tuning flat domain wall solution (26), the potential takes the
following form: 3


 1 K+ 2

 − , y > 0,

 4 (K+ y + 1)2


K+ − K−
V (y) = δ(y), y = 0, (41)

 2



 1 K− 2

− , y < 0.
4 (K− y + 1)2
The solution f0 (y) to the Schrödinger equation satisfying the boundary condition f˜00 (0+ )−
f˜00 (0− ) = K+ −K
2
− ˜
f0 (0) is f˜0 (y) ∼ (K+ y + 1)1/2 for y > 0 and ∼ (K− y + 1)1/2 for
y < 0. So, as in the dilatonic domain wall case, the KK mode zero mode is constant:
2
R ∞In the 2case of Rthe non-dilatonic RS domain wall, the KK zero mode f0 = constant is also normalizable:
2A ∼ ∞ dy e−2k|y| < ∞.
−∞ dy f0 e −∞
3 We define A0 (y) at y = 0 as A0 (0) ≡ lim A(ε)−A(−ε) 00
ε→0+ ε−(−ε) , and similarly for A (0). And we used the fact
that a + bδ(y) = bδ(y) at y = 0 for any finite constant a.
f0 (y) = e−2A f˜0 (y) = constant. The normalized form of the zero mode is f0 (y) = N0 =
q
3K+ K−
2(K+ −K− ) . Similarly as in the dilatonic domain wall case, one obtains the action for the
massless dilaton ϕ0 in the four-dimensional flat spacetime by substituting the zero mode
field Φ = ϕ0 f0 into the bulk action (31). In this case, the integration interval for the extra
spatial coordinate is −1/K− 6 y 6 −1/K+ .
3.2. Abelian gauge field
The action for the bulk abelian gauge field AM (x µ , y) is

Z
√
Sbulk ⊃ dx 4 dy −G − 14 GMN GRS FMR FNS + 12 m2 GMN AM AN , (42)
from which we obtain the following equation of motion for AM :

1 √
√ ∂M −G GMN GRS FNS + m2 GRS AS = 0. (43)
−G
√
By taking the divergence (defined as ∇M V M ≡ √ 1 ∂M ( −G V M )) of this equation, one
−G
has m2 ∇M AM = 0. For m 6= 0, one obtains the gauge condition ∇M AM = 0 on a massive
AM . By using this gauge condition, one can eliminate one of the five components of AM ,
which we choose as Ay . Then, the gauge condition ∇M AM = 0 along with the gauge
choice Ay = 0 and the five-dimensional metric GMN of the form (4) implies ηµν ∂µ Aν = 0.
In the case of massless (m = 0) bulk Abelian gauge field AM , one can also choose
the gauge Ay = 0 = ηµν ∂µ Aν by using the gauge degrees of freedom. In the Ay = 0
= ηµν ∂µ Aν gauge, the equation of motion (43) for AM takes the following form:
µν
η ∂µ ∂ν + ∂y e2A ∂y + m2 e2A Aρ = 0. (44)
To consider the KK mode of the bulk Abelian gauge field Aρ with mass mn , only, we
decompose Aρ as

Aρ x µ , y = aρ(n) x µ fn (y), (45)
and require aρ(n) to satisfy the following Proca equation for an Abelian gauge field with
mass mn in the Lorentz gauge in flat four-dimensional spacetime:
µν
η ∂µ ∂ν + m2n aρ(n) = 0. (46)
Then, the equation of motion (44) reduces to the following Sturm–Liouville equation:

∂y e2A ∂y fn + m2 e2A fn = m2n fn . (47)
So, the KK modes with different masses are orthogonal to each other with respect to
the weighting function w(y) = 1, provided the boundary condition fm∗ e2A fn0 |y=b = 0 is
y=b
satisfied.
By using a new y-dependent function fñ = eA fn , one can bring Eq. (47) to the following
Schrödinger equation form:
d 2 fñ
− + V (y)fñ = m2 fñ (48)
dy 2
with the potential

V (y) = A00 + (A0 )2 + m2n e−2A . (49)
From now on, we consider only the zero mode (m0 = 0) of the massless bulk abelian gauge
field (m = 0).
First, for the dilatonic domain wall solution (24), the potential (49) in the Schrödinger
equation (48) takes the following form:
4K 2 4 − 9a 2 8K
V (y) = − 2 δ(y). (50)
81a (1 − K|y|)
4 2 9a
The solution to the Schrödinger equation that satisfies the boundary condition f˜00 (0+ ) −
4
8K ˜
f˜00 (0− ) = − 9a ˜
2 f0 (0) is f0 (y) ∼ (1 − K|y|)
9a 2 . So, the KK zero mode is constant: f0 (y) =
√
e−A f˜0 = constant. The normalized form of the zero mode is f0 (y) = N0 = K/2. We
(0)
obtain the following effective action for the massless abelian gauge field aµ (with the field
(0) (0) (0)
strength fµν = ∂µ aν − ∂ν aµ ) in the four-dimensional flat spacetime by substituting the
(0)
zero mode field Aµ = aµ f0 into the action (42) for the bulk massless abelian gauge field
(m = 0):
Z Z
1/K Z
1 √ 1
− dx dy −G GMN GRS FMR FNS = −
4
dy f02 dx 4 ηµν ηρσ fµρ
(0) (0)
fνσ
4 4
−1/K
Z
1
=− dx 4 ηµν ηρσ fµρ
(0) (0)
fνσ . (51)
4
Second, for the self-tuning flat domain wall solution (26), the potential is given by


 3 K+2
−
 , y > 0,

 16 (K+ y + 1)2


K+ − K−
V (y) = δ(y), y = 0, (52)

 4



 3 K−2

− , y < 0.
16 (K− y + 1)2
The solution f˜0 (y) to the Schrödinger equation satisfying the boundary condition f˜00 (0+ )−
f˜00 (0− ) = K+ −K
4
− ˜
f0 (0) is f˜0 (y) ∼ (1 + K+ y)1/4 for y > 0 and ∼ (1 + K− y)1/4 for
y < 0. So, as in the dilatonic domain wall case, the KK zero mode is y-independent:
−A ˜
p0 (y) = e f0 = constant. The normalized form of the KK zero mode is f0 (y) = N0 =
f
K+ K− /(K+ − K− ). Similarly as in the case of the dilatonic domain wall, we obtain the
(0)
four-dimensional effective action for the massless abelian gauge field aµ in flat spacetime
by plugging the zero mode field Aµ = aµ(0)f0 into the bulk action.
Note, in the case of the non-dilatonic domain wallR of the original RS model [3–5], the
∞
KK zero mode f0 = constant is not normalizable: −∞ dy f02 = ∞ for the warp factor
e2A = e−2k|y| which is defined on −∞ < y < ∞. So, the four-dimensional effective action
for the massless abelian gauge field cannot be obtained, unless one restricts the allowed
values of y, for example, by regarding the extra dimension to be a segment S 1 /Z2 as in the
first RS model [3].
3.3. Spinor field
The action for the bulk spin-1/2 fermion Ψ (x µ , y) is 4

Z √
Sbulk ⊃ dx 4 dy G i Ψ SΓ M DM Ψ, (53)
where DM ≡ ∂M + 14 ωMAB γ AB is the gravitational covariant derivative on a spinor. Here,

ωMAB is the usual spacetime spin-connection. The convention for the spacetime vector
indices are M, N, P [A, B, C] for the five-dimensional curved [flat-tangent] spacetime
indices, µ, ν, ρ [α, β, γ ] for the four-dimensional curved [flat-tangent] spacetime indices,
and y and 5, respectively, for the curved and flat extra spatial indices. The flat space gamma
matrices γ A satisfying {γ A , γ B } = 2ηAB are give by γ A = (γ α , iγ 5 ). The curved space
gamma matrices Γ M ≡ EA M γ A satisfy {Γ M , Γ N } = 2GMN , where E M is the inverse of
A
A
the Fünfbein EM .
From the above action, we obtain the following equation of motion for the bulk spinor Ψ :

i Γ M ∂M + 14 Γ M ωMAB γ AB Ψ = 0, (54)
which reduces to the following form after the metric (4) is substituted:

i e−A γ α ∂α + iγ 5 ∂y + 2i∂y Aγ 5 Ψ = 0. (55)
To consider the KK mode with mass mn , only, we decompose the bulk spinor Ψ =
Ψ R + Ψ L as Ψ R,L (x µ , y) = ψnR,L (x µ )fnR,L (y) and require ψn = ψnR + ψnL to satisfy
the following Dirac equation for a spinor with mass mn in flat four-dimensional spacetime:
α
iγ ∂α − mn ψn = 0. (56)
Here, Ψ R,L ≡ 12 (1 ± γ 5 )Ψ and similarly for ψnR,L . Then, the equation of motion (55)
reduces to the following form:

∂y + 2∂y A fnR,L = ±mn e−A fnL,R . (57)
In the following, we study the KK zero mode (m0 = 0).
First, for the dilatonic domain wall solution (24), the KK zero mode is
− 8
f0 (y) ∼ (1 − K|y|) 9a 2 . (58)
We check the normalizability of the KK zero mode by plugging the zero mode field Ψ =
ψ0 (x µ )f0 (y) into the bulk action:
Z √ Z
1/K Z
4
dx dy SΓ DM Ψ =
G iΨ M
dy e3A f02 dx 4 i ψ̄0 γ α ∂α ψ0 . (59)
−1/K
µν =
4 Note, for bulk fermions, we use the mostly negative metric signature convention, i.e., η
diag(1, −1, −1, −1, −1).
2
We see that the KK zero mode is normalizable, q provided a > 4/9. The8 KK zero mode
−
including the normalization factor is f0 (y) = ( 12 − 9a2 2 )K (1 − K|y|) 9a2 .
Second, for the self-tuning flat domain wall solution (26), the KK zero mode is
(
(K+ y + 1)− 2 , y > 0,
1
f0 (y) ∼ (60)
(K− y + 1)− 2 , y < 0.
1
One can show that this KK zero mode is normalizable by plugging the zero mode field
Ψ = ψ0 (x µ )f0 (y) into the bulk action, as we did for the dilatonic
q domain wall case. The
3K+ K−
normalization factor for the KK zero mode in this case is N0 = 4(K+ −K− ) .
For
R∞ the non-dilatonic
R∞ RS domain wall, the zero mode f0 (y) ∼ e2k|y| , is not normalizable,
i.e., −∞ dy e3A f02 ∼ −∞ dy ek|y| = ∞, unless one restricts the allowed values of y within
a finite interval.
3.4. Gravitino
The action for the bulk gravitino ΨM is

Z
√ 1
Sbulk ⊃ d 4 x dy −G Ψ SM Γ MNP DN ΨP . (61)
2
So, the equation of motion for the gravitino is
Γ MNP DN ΨP = 0. (62)
We choose the gauge Ψy = 0.
To consider the KK zero mode of Ψµ , we decompose it as

Ψµ x ν , y = ψµ(0) (x ν )f0 (y), (63)
and require ψµ(0) to satisfy the following Rarita–Schwinger equation for the massless
gravitino in flat four-dimensional spacetime:
γ αβδ ∂β ψδ(0) = 0, (64)
(0) (0)
along with the gauge conditions ∂ α ψα = 0 = γ α ψα . Then, the equation of motion (62)
takes the following form:
[∂y + ∂y A]f0 = 0. (65)
First, for the dilatonic domain wall solution (24), the KK zero mode is
− 4
f0 (y) ∼ (1 − K|y|) 9a 2 . (66)
To check the normalizability of the KK zero mode, we substitute the zero mode field Ψµ =
ψµ(0) f0 into the bulk action:
Z Z
1/K Z
√ 1 S MNP 1 (0)
4
d x dy −G Ψ MΓ DN ΨP = dy eA f02 dx 4 ψ̄α(0) γ αβδ ∂β ψδ .
2 2
−1/K
(67)
We see that the KK zero mode (66) is normalizable and its normalization factor is N0 =
√
K/2.
Second, for the self-tuning flat domain wall solution (26), the KK zero mode is
(
(K+ y + 1)− 4 , y > 0,
1
f0 (y) (68)
(K− y + 1)− 4 , y < 0.
1
Similarly as in the dilatonic domain wall case, one can show that this KK zero mode is
normalizable.
p The normalization factor of the KK zero mode f0 (y), in this case, is N0 =
K+ K− /(K+ − K− ).
For
R∞ the non-dilatonic
R ∞RS domain wall, the zero mode f0 (y) ∼ ek|y| is not normalizable,
i.e., −∞ dy e f0 ∼ −∞ dy e
A 2 k|y| = ∞, unless one restricts the allowed values of y within
a finite interval.
3.5. Graviton
In our previous works, we observed [6] that the RS type scenario can be extended
to the extreme dilatonic domain walls and showed [7] that indeed the dilatonic domain
walls effectively compactify gravity to one lower dimensions by calculating the graviton
KK zero mode effective action. Later, this was further confirmed [25] by explicitly
constructing the normalizable graviton KK zero modes. Although the explicit graviton
KK zero modes are given in Ref. [25], we shall calculate them again since we are using
different parametrization of solution, which we find to be more convenient.
For the purpose of studying the KK zero mode of graviton, it is more convenient to
consider the domain wall metric in conformally flat form. (The reason is that in such
coordinate frame the metric perturbation in the RS gauge around the domain wall metric
takes the form of the Schrödinger equation (76) in the below.) This is achieved by
transforming the extra spatial coordinate to new one, which we denote as z. The conformal
factor for the domain wall metric and the dilaton are 5
8
S 9a2 −4 , 9a S
C(z) = (1 + K|z|) φ= 2 ln(1 + K|z|) + C, (69)
9a − 4
for the dilatonic domain wall, and
(
(1 + KS+ z) 23 , z > 0,
C(z) =
(1 + KS− z) 23 , z < 0,

η ln(1 + K S+ z) + C, y > 0,
φ= + (70)
η− ln(1 + KS− z) + C, y < 0,
S and K
for the self-tuning flat domain wall, where K S± are defined as
r
S ≡ 4 − 9a e−aC
2 3
K Λ, KS± ≡ 3 K± . (71)
6 16 − 9a 2 4
5 Making use of the invariance of conformally flat form of the domain wall metric Ansatz under the translation
in z, we choose the coordinate such that y = 0 corresponds to z = 0. Then, the requirement C(z = 0) =
e2A(y=0) = 1 fixes the conformal factor to take the following forms.
To study the KK modes of graviton in the bulk background of the domain walls, we
consider the following small fluctuation around the domain wall metric:

GMN dx M dx N = C(z) (ηµν + hµν ) dx µ dx ν + dz2 , (72)
where metric perturbation hµν (x ρ , z) is assumed to satisfy the transverse traceless
µ
gauge condition h µ = 0 = ∂ µ hµν . The (µ, ν)-component of the Einstein equations is
approximated, to the first order in hµν , to

3 ∂z C
2x + ∂z +
2
∂z hµν = 0, (73)
2 C
where 2x ≡ ηµν ∂µ ∂ν . To consider the KK mode with mass mn , only, we decompose
(n) (n) (n) (n)
hµν as hµν (x ρ , z) = ĥµν (x ρ )fn (z) and require ĥµν to satisfy 2x ĥµν = m2n ĥµν . Then,
the linearized Einstein equation (73) reduces to the following form:

3 ∂z C
∂z +
2
∂z + mn fn = 0.
2
(74)
2 C
Had we used y as the extra spatial coordinate, the equation (74) satisfied by the KK mode
fn with mass mn would have taken the following form of the Sturm–Liouville equation:

∂y e4A ∂y fn + m2n e2A fn = 0, (75)
from which we know that the KK modes are orthogonalized with the respect to the
weighting function w(y) = e2A , provided the boundary condition fm∗ e4A fn0 |y=a = 0 is
y=b
satisfied, or with respect to the weighting function w(z) = C 3/2 if z is used as the extra
spatial coordinate.
In terms of a new z-dependent function defined as fñ ≡ C 3/4 fn , Eq. (74) takes the
following form of the Schrödinger equation:
d 2 fñ
− + V (z)fñ = m2n fñ , (76)
dz2
with the potential
00 0 2
3 C C
V (z) = 4 − . (77)
16 C C
(0)
In the following, we study the KK zero mode (m0 = 0), for which ĥµν satisfies the
linearized vacuum Einstein equation 2x ĥ(0)
µν = 0 in the Lorentz gauge.
First, for the dilatonic domain wall solution (69), the potential (77) in the Schrödinger
equation (76) takes the following form:
6K S2 10 − 9a 2 S
12K
V (z) = − δ(z). (78)
S 2 4 − 9a 2
(9a 2 − 4)2 (1 + K|z|)
Note, the δ-function potential is always attractive, implying that the KK zero mode solution
can always be supported. The solution f˜0 to the Schrödinger equation satisfying the
S − 4−9a2 . So, the
6
S
boundary condition f˜0 (0+ ) − f˜0 (0− ) = − 12K 2 f˜0 (0) is f˜0 ∼ (1 + K|z|)
0 0 4−9a
−3/4 f˜ = constant. The normalization factor

KK zero mode is z-independent: q f0 (z) = C 0
for the KK zero mode is N0 = ( 2 + 9a 2 )K.
1 4
Second, for the self-tuning flat domain wall solution (70), the potential has the following
form:

 1 S+
K 2

−

 S+ z + 1)2
, z > 0,

 4 (K

1 S S− )δ(z), z = 0,
V (z) = (K+ − K (79)

 2



 1 S−
K 2

− , z < 0.
4 (KS− z + 1)2
As in the dilatonic domain wall case, the δ-function potential is always attractive. The
solution f˜0 (z) to the Schrödinger equation satisfying the boundary condition f˜00 (0+ ) −
0 2
S+ − K
f˜0 (0− ) = 1 (K S− )f˜0 (0) is f˜0 (z) ∼ (K
S+ z + 1)1/2 for z > 0 and ∼ (K
S− z + 1)1/2 for
z < 0. So, as in the dilatonic domain wall case, the KK zero mode is z-independent:
= C −3/4f˜0 = constant. The normalization constant for the KK zero mode f0 is
f0 (z) q
N0 = 32 KK++−KK−
−
.
In the following, we obtain the KK zero mode effective actions for the graviton in the
bulk backgrounds of the dilatonic and the self-tuning domain walls. Since we have shown
that the KK zero mode hµν = ĥ(0) µν f0 of the graviton is independent of the extra spatial
coordinate z, we consider the following form of the bulk metric:

GMN dx M dx N = C(z) gµν (x ρ ) dx µ dx ν + dz2 , (80)
where the conformal factor C is given by Eq. (69) or (70). The dilaton φ(z) is given by
Eq. (69) or (70). A useful formula for obtaining the KK zero mode effective actions is
0 2
√ √ 3 C 00 C
−G RG = −g C 2 Rg − 4 + . (81)
C C
Here, RG and Rg are, respectively, the Ricci scalars for the metrics GMN and gµν . If we
instead use the KK zero mode bulk metric of the following form:
GMN dx M dx N = e2A(y)gµν (x ρ ) dx µ dx ν + dy 2 , (82)
then the corresponding formula would be

√ √
−G RG = −g e2A Rg − 4e4A 2A00 + 5(A0 )2 , (83)
but the resulting expression for the effective actions will be the same.
First, we consider the effective action in the bulk of the dilatonic domain wall. In our
previous work [7], we obtained the effective action for the case where K defined in Eq. (6)
is positive [negative] in the region y < 0 [y > 0] for the 1 < 0 case, only. In this case, the
tension σDW of the wall takes the positive value given by Eq. (7) with D = 5. In this paper,
we shall assume that K defined in Eq. (6) is positive [negative] in the region y < 0 [y > 0]
even for the 1 > 0 case, as well as for the 1 < 0 case. Substituting the Ansätze 6 for the
fields given by Eqs. (69) and (80) into the total action, we obtain the following:
Z
1 √
S = 2 d 5 x −G RG − 43 ∂M φ∂ M φ + e−2aφ Λ
2κ5
Z
√
− σDW d 4 x −γ e−aφ
Z S2 −2 S
1 √ 12 20(16 − 9a 2)K 64K
= 2 d x dz −g $ 9a −4 Rg −
4 2
$ + δ(z)
2κ5 (4 − 9a )
2 2 4 − 9a 2
Z
√
+ e−2aC Λ$ −2 − d 4 x −g e−aC σDW , (84)
where from Eqs. (71) and (25) we see that the cosmological constant and the tension of the
wall can be expressed as
12(16 − 9a 2) 2aC S2 1 24 S
Λ= e K , σDW = eaC K, (85)
(4 − 9a 2 )2 2
κ5 4 − 9a 2
S After the integration over z (with the integration interval −∞ < z < ∞
and $ ≡ 1 + K|z|.
2
for a < 4/9 and K S−1 < z < −K S−1 for 4/9 < a 2 < 16/9), we see that all the extra terms
cancel out and we are left with the term for the four-dimensional general relativity (with
vanishing cosmological constant) with the gravitational constant given by
r
8 + 9a 2 S 2 −aC 8 + 9a
2 3
κ4 =
2
Kκ = e Λ κ52 . (86)
2(4 − 9a 2) 5 12 16 − 9a 2
A troublesome case is the a 2 > 16/9 case (with the integration interval K S−1 < z
< −K S−1 ), in which the integration on the $ −2 terms in the action (84) diverges, whereas
the effective four-dimensional gravitational constant κ42 remains to have the nonzero value
given by Eq. (86). (This is the case which normally would have been discarded because
of the positive cosmological constant (negative Λ in our convention).) On the other hand,
the graviton KK zero mode gµν (x ρ ) in the dilatonic domain wall background (69) satisfies
the four-dimensional vacuum Einstein’s equation with vanishing cosmological constant
term. So, it seems to be contradictory that we do not reproduce the action for the four-
dimensional general relativity with vanishing cosmological constant by integrating the
action (84) with respect to the extra spatial coordinate z. One can avoid infinity in the
effective action by truncating the extra spatial dimension through the introduction of
another domain wall in the region between z = 0 and |z| = −K S−1 . This seems to be
reasonable because |z| = −K S (or |y| = K ) corresponds to the curvature singularity,
−1 −1
which has to be avoided in the brane world scenario unless there is reasonable physical
significance associated with the singularity. Then, the total action has the following form:
6 Such Ansätze are consistent with the five-dimensional equations of motion as long as the zero mode metric
gµν (x ρ ) satisfies the Ricci flat condition. The Ricci flat condition is equivalent to the condition that the zero
mode gµν (x ρ ) satisfies the four-dimensional Einstein equations with vanishing cosmological constant.
Z Zz0
1 √
S= 2 d x4
dz −G RG − 43 ∂M φ∂ M φ + e−2aφ Λ
2κ5
−z0
Z q Z p
− σh d 4x −γ h e−aφh − σv d 4 x −γ v e−aφv , (87)
z=0 z=z0
where
r
4 3
σh = −σv = 2 Λ,
κ5 16 − 9a 2
8
h
γµν = gµν , v
γµν = (1 + KzS 0 ) 9a2 −4 gµν ,
9a S 0 ) + C.
φh = C, φv = 2 ln(1 + Kz (88)
9a − 4
Going through the similar calculation as in Eq. (84), except that there will be an additional
δ-function term at |z| = z0 in the integrand of the bulk action, one will find after the
z-integration that all the extra terms are cancelled and one is left only with the curvature
term. The effective four-dimensional gravitational constant in this case is given by
8 + 9a 2 Sh 9a 2 +8 i
2 −1
=
κ42 K 1 − $09a −4 κ52
2(4 − 9a )
2
r h 9a 2 +8 i−1
8 + 9a 2 3
= e−aC Λ 1 − $ 9a 2 −4
κ52 , (89)
12 16 − 9a 2 0
S 0 . Note, even if we introduced the additional domain wall to cure the

where $0 = 1 + Kz
problem of diverging effective action for the a 2 > 16/9 case, this result with the additional
domain wall is valid for any values of a.
So, we see that gravity in the bulk of the dilatonic domain wall is effectively compactified
to one lower dimensions, for any values of the dilaton coupling parameter a, provided the
extra space is truncated through the introduction of another domain wall in the case of
a 2 > 16/9. For this to happen, one has to choose the sign ± in the expression for K in
Eq. (6) such that the warp factor for the domain wall metric has singularity at a finite
value of y on both sides of the wall and therefore the warp factor decreases on both sides
of the domain wall. This choice of signs corresponds to the conformal factor of the form
(69) (for the D = 5 case) when the metric tensor is transformed to take a conformally flat
form. Also, in this case, the tension of the wall takes the positive value given by Eq. (7).
For other choices of sign ± in the expression for K in Eq. (6), the gravity in the bulk is
not effectively compactified to the Einstein gravity with zero cosmological constant in one
lower dimensions and the tension of the wall is either zero or negative.
Next, we consider the effective action in the bulk of the self-tuning flat domain wall.
Substituting the Ansätze for the fields given by Eqs. (70) and (80) into the total action, we
obtain the following:
Z Z
1 √ √
S = 2 d 5 x −G RG − 43 ∂M φ∂ M φ − d 4 x −γ f (φ)
2κ5
Z Z
1 √ √
= 2 4 S+ − K
d x dz −g $± Rg − 83 (K S− )δ(z) − d 4x −g f (C), (90)
2κ5
where in the second line we choose the plus [the minus] sign in $± ≡ 1 + K S± z for the
S S
z > 0 [z < 0] region and f (C) = (K− − K+ )/κ5 . Note, a term coming from RG which
2
would potentially have diverged after the z-integration is cancelled by another diverging
term − 43 ∂M φ∂ M φ and we are left only with the δ-function term coming from RG . After
the integration over z (with the integration interval −KS−−1
< z < −KS+−1
), we obtain the
following effective action:
Z S+ − K S−
1 √ K
S = 2 d 4 x −g S+ K
Rg − 43 K S− . (91)
2κ5 S
2K+ K−S
Unexpectedly, the four-dimensional effective action for the graviton zero mode has a
cosmological constant term. This is contradictory, because the graviton KK zero mode
gµν (x ρ ) in the bulk of the self-tuning flat domain wall satisfies the four-dimensional
vacuum Einstein equations with vanishing cosmological constant and the effective four-
S+ K
2K S− 2
dimensional gravitational constant has a nonzero value given by κ42 = K S− κ5 . It is
S+ −K
also inconsistent with the fact that the self-tuning flat domain wall solution has the
four-dimensional Poincaré invariance, which requires zero cosmological constant in four
dimensions. 7 Ref. [26], which appeared in the preprint archive after the present work,
independently observes such unexpected nonzero cosmological constant term in the
effective action and resolves the problem by introducing extra domain walls at the naked
singularities 8 and fine-tuning their tensions to cancel out the undesirable cosmological
constant term. So, such resolution spoils nice self-tuning property of the self-tuning flat
domain walls. This seems to indicate that there is some other ingredient missing or some
inconsistency in the scenario of self-tuning flat domain wall necessary in reproducing four-
dimensional effective theory with vanishing cosmological constant without loosing nice
self-tuning property. (It seems to be important to obtain the expect form of the effective
action, because anyway in the RS model we usually consider a part of the complete
effective action to determine the four-dimensional effective gravitational constant.) It is
important to note that the existence of a solution with the Poincaré invariance in one
lower dimensions and the fact that the graviton zero mode in the bulk of such solution
satisfies the four-dimensional vacuum Einstein equation with zero cosmological constant
do not necessarily mean that the effective action in one lower dimensions has the right
expected form. An example is the non-dilatonic domain wall with exponentially increasing
warp factor. The graviton zero mode in such background satisfies the four-dimensional
vacuum Einstein equations with zero cosmological constant, but the four-dimensional
gravitational constant is zero if we take the extra spatial dimension to be infinite in size.
7 I would like to thank Prof. Kallosh for pointing out inconsistency between the existence of the domain
wall solution with the four-dimensional Poincaré invariance and the non-vanishing cosmological constant in the
effective action, after the first version of this paper appeared in the preprint archive.
8 The author feels that the extra introduced domain walls should rather be placed between the singularities
and the self-tuning flat domain wall, because the singularities of the self-tuning flat domain wall are problematic
regions where the induced metric vanishes, the dilaton blows up and the spacetime curvature diverges.
Additional condition that the four-dimensional gravitational constant should be nonzero

seems to be also insufficient. As we have seen the case of the dilatonic domain walls with
positive cosmological constant (Λ < 0 in our convention), although the four-dimensional
gravitational constant is nonzero, as well as the KK zero mode graviton in the domain
wall bulk satisfies the four-dimensional vacuum Einstein equations with zero cosmological
constant, the effective action diverges unless additional domain wall is introduced. One
possible solution to the problem might be to consider all the massive KK modes of bulk
graviton to cancel out the unwanted cosmological constant, but on the other hand the
graviton zero mode itself is a consistent solution to the five-dimensional equations of
motion provided the zero mode gµν (x ρ ) satisfies the Ricci flat condition.
References
[1] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, The hierarchy problem and new dimensions at a
millimeter, Phys. Lett. B 429 (1998) 263, hep-ph/9803315.
[2] I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, New dimensions at a millimeter to
a Fermi and superstrings at a TeV, Phys. Lett. B 436 (1998) 257, hep-ph/9804398.
[3] L. Randall, R. Sundrum, A large mass hierarchy from a small extra dimension, Phys. Rev.
Lett. 83 (1999) 3370, hep-ph/9905221.
[4] L. Randall, R. Sundrum, An alternative to compactification, Phys. Rev. Lett. 83 (1999) 4690,
hep-th/9906064.
[5] J. Lykken, L. Randall, The shape of gravity, hep-th/9908076.
[6] D. Youm, Solitons in brane worlds, hep-th/9911218.
[7] D. Youm, A note on solitons in brane worlds, hep-th/0001166.
[8] K. Behrndt, M. Cvetič, Supersymmetric domain wall world from D = 5 simple gauged
supergravity, hep-th/9909058.
[9] K. Skenderis, P.K. Townsend, Gravitational stability and renormalization-group flow, Phys.
Lett. B 468 (1999) 46, hep-th/9909070.
[10] A. Chamblin, G.W. Gibbons, Supergravity on the brane, hep-th/9909130.
[11] R. Kallosh, A. Linde, M. Shmakova, Supersymmetric multiple basin attractors, JHEP 11 (1999)
010, hep-th/9910021.
[12] M. Cvetič, H. Lu, C.N. Pope, Domain walls and massive gauged supergravity potentials, hep-
th/0001002.
[13] R. Kallosh, A. Linde, Supersymmetry and the brane world, hep-th/0001071.
[14] K. Behrndt, M. Cvetič, Anti-de Sitter vacua of gauged supergravities with 8 supercharges, hep-
th/0001159.
[15] N. Arkani-Hamed, S. Dimopoulos, N. Kaloper, R. Sundrum, A small cosmological constant
from a large extra dimension, hep-th/0001197.
[16] S. Kachru, M. Schulz, E. Silverstein, Self-tuning flat domain walls in 5d gravity and string
theory, hep-th/0001206.
[17] W.D. Goldberger, M.B. Wise, Bulk fields in the Randall–Sundrum compactification scenario,
Phys. Rev. D 60 (1999) 107505, hep-ph/9907218.
[18] W.D. Goldberger, M.B. Wise, Modulus stabilization with bulk fields, Phys. Rev. Lett. 83 (1999)
4922, hep-ph/9907447.
[19] W.D. Goldberger, M.B. Wise, Phenomenology of a stabilized modulus, hep-ph/9911457.
[20] H. Davoudiasl, J.L. Hewett, T.G. Rizzo, Bulk gauge fields in the Randall–Sundrum model, hep-
ph/9911262.
[21] A. Pomarol, Gauge bosons in a five-dimensional theory with localized gravity, hep-ph/9911294.
[22] B. Bajc, G. Gabadadze, Localization of matter and cosmological constant on a brane in anti-de
Sitter space, hep-th/9912232.
[23] Y. Grossman, M. Neubert, Neutrino masses and mixings in non-factorizable geometry, hep-
ph/9912408.
[24] S. Chang, J. Hisano, H. Nakano, N. Okada, M. Yamaguchi, Bulk standard model in the Randall–
Sundrum background, hep-ph/9912498.
[25] C. Gomez, B. Janssen, P. Silva, Dilatonic Randall–Sundrum theory and renormalization group,
hep-th/0002042.
[26] S. Forste, Z. Lalak, S. Lavignac, H.P. Nilles, A comment on self-tuning and vanishing
cosmological constant in the brane world, Phys. Lett. B 481 (2000) 360, hep-th/0002164.
AdS box graphs, unitarity and operator product

expansions
L. Hoffmann, L. Mesref, W. Rühl ∗
Department of Physics, Theoretical Physics, University of Kaiserslautern,
Postfach 3049, 67653 Kaiserslautern, Germany
Received 12 July 2000; accepted 11 August 2000
Abstract
We develop a method of singularity analysis for conformal graphs which, in particular, is
applicable to the holographic image of AdS supergravity theory. It can be used to determine the
critical exponents for any such graph in a given channel. These exponents determine the towers of
conformal blocks that are exchanged in this channel. We analyze the scalar AdS box graph and show
that it has the same critical exponents as the corresponding CFT box graph. Thus pairs of external
fields couple to the same exchanged conformal blocks in both theories. This is looked upon as a
general structural argument supporting the Maldacena hypothesis.  2000 Elsevier Science B.V. All
rights reserved.
PACS: 11.15.Tk; 11.25.Hf; 11.25.Pm

Keywords: AdS/CFT; Box graph; Unitarity
1. Introduction
The AdS/CFT correspondence [1–5] connects N = 4 supersymmetric SU(N) Yang–

Mills theory in four dimensions at large N and strong ’t Hooft coupling λ = gYM
2 N with
type IIB supergravity on the AdS5 × S background based on a perturbatively defined

5
action. The correspondence works by comparison of series expansions in powers of

1/N 2 . At leading order many predictions have been verified, and at next order, results
such as concerning anomalies, nonrenormalization theorems and 1/N 2 -corrections to field
dimensions for composite fields and structure constants of the SYM4 field algebra have
been obtained [6–14].
E-mail addresses: hoffmann@physik.uni-kl.de (L. Hoffmann), lmesref@physik.uni-kl.de (L. Mesref),
ruehl@physik.uni-kl.de (W. Rühl).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 1 7 - 4
338 L. Hoffmann et al. / Nuclear Physics B 589 (2000) 337–355
In this context the evaluation of AdS graphs, that represent the holographic image of
the AdS perturbation expansion in powers of α 0 /R 2 = λ−1/2 , confronts us with serious
technical problems whose difficulty goes much beyond the corresponding CFT flat space
graphs. Partly with techniques developed first for CFT in flat space, the exchange graph
was calculated and studied in a series of works [15–17]. The results of all such calculations
were finally expressed in terms of generalized hypergeometric functions. However, in some
cases the field dimensions had to be specialized to small natural numbers.
Due to these difficulties, we advocate another approach in this work. We present Green
functions as multiple “Mellin–Barnes integrals” 1 over a meromorphic function Φ. This
function Φ is defined as the integral over a positive function on a compact domain. Usually
one would expand this integral into a series of ratios of gamma functions, so that Φ obtains
poles from the gamma functions and from the divergence of the series. The latter are
difficult to work out. 2 Thus we would like to extract the poles of Φ by another method.
The relevant poles of Φ, namely those to the right of the Mellin–Barnes contours, originate
from the divergence (infinity) of the integrand at certain faces or intersections of faces of
the regular polyhedral integration domain. So guessing them is not difficult. These poles
form sequences which are integrally spaced and tend to +∞. Of course at the end all
Mellin–Barnes integration contours are shifted to +∞, so that we find series expansions
again.
In Section 2 we discuss this method and typical results from the point of view of unitarity
of Green functions and operator product expansions. Important information on the structure
of the field algebra is obtained this way. Since AdS/CFT correspondence also implies
(supposedly) a correspondence between both field algebras (all orders of 1/N 2 included),
the AdS conformal field theory as the holographic picture of supergravity and flat space
CFT must therefore already show a partial correspondence on the level of the meromorphic
functions Φ. We demonstrate that this is in fact true for the box graph.
In Section 3 we study once again the exchange graph as a simple example of the
previously developed method.
In Section 4 we treat the box graph with arbitrary field dimensions 3 with our method.
We do not give all the details of the lengthy analysis.
A few remarks are added in Section 5.
2. Critical exponents and unitarity
We discuss here the connection between unitarity, operator product expansions and the
“critical exponents”, that we shall introduce now. Consider a four-point function in flat
space CFTd (Fig. 1). Its Green function G can be split into a covariant multiplier and an
e
invariant function G
1 Inverse Mellin transforms and Barnes integrals are equivalent.

2 Except for the functions, say, F (1) and F (1) almost nothing is known.
2 1 3 2
3 This arbitrariness is essential.
L. Hoffmann et al. / Nuclear Physics B 589 (2000) 337–355 339
Fig. 1. An unspecified four-point function of CFTd .

− 1 (µ1 +µ2 −µ3 +µ4 ) − 1 (µ1 −µ2 +µ3 −µ4 )
G(x1 , x2 , x3 , x4 ) = x12
2 2 2
x13 2
1 (µ1 −µ2 −µ3 +µ4 ) −µ4

× x23
2 2 2
x34 e v),
G(u, (2.1)
where
xij = xi − xj (2.2)
and
2 x2
x14 2 x2
x13
u= 23
2 x2
, v= 24
2 x2
(2.3)
x12 34 x12 34
are conformal invariant variables. If we intend an operator product expansion in the channel
x14 → 0, x23 → 0
we must let
u → 0, v → 1. (2.4)
e v) can in turn be decomposed as
The function G(u,
X
K
e v) =
G(u, uγk Fk (u, v), (2.5)
k=1
where Fk are holomorphic functions in the neighborhood of (2.4) and possess the Taylor
expansion
∞
X un (1 − v)m (k)
Fk (u, v) = cmn . (2.6)
n! m!
m,n=0
The γk are the “critical exponents”. Of course, the γk are, due to possible changes in
the covariant multiplier (2.1), defined up to a common additive constant. So, what is the
physical information encoded in these exponents?
Consider the exchange of the scalar field of dimension δ in the channel 4
(1, 4) ←→ (2, 3)
as described by Fig. 2 where the dimension δ is assumed to be generic. Note, that the CFTd
4 We call the exchange channel the “direct” channel.

Fig. 2. Scalar field exchange in the direct channel.
Fig. 3. Representation of a full three vertex by three unique vertices.
covariant vertex functions

Z Yn
−µi
dy (y − xi )2 (2.7)
i=1
are necessarily “unique”, i.e., they satisfy the condition

X
n
µi = d. (2.8)
i=1
A full vertex, such as in Fig. 2, can always be resolved in three unique vertices (Fig. 3)
in an unambiguous fashion. This fact can be readily used to compute the Green function
corresponding to Fig. 2. The result is explicitly known [18] and can be represented as
X
2
e v) =
G(u, uγk Fk (δ; u, v) (2.9)
k=1
with
1
γ1 = (δ − µ1 − µ4 ), (2.10)
2
1
γ2 = (d − δ − µ1 − µ4 ) (2.11)
2
and, after an appropriate renormalization,
F2 (δ; u, v) = F1 (d − δ; u, v). (2.12)
On the other hand, for the holographic image of the AdS exchange graph Fig. 4, termed
Fig. 4. The Witten exchange graph.
“Witten graph”, we obtain

X
3
eW (u, v) =
G uγk FW,k (δ; u, v) (2.13)
k=1
with
2γ1 + (µ1 + µ4 ) = µ1 + µ4 , (2.14)

2γ2 + (µ1 + µ4 ) = µ2 + µ3 , (2.15)
2γ3 + (µ1 + µ4 ) = δ. (2.16)
Of course (2.10) and (2.16) are identical. On the other hand there are striking differences.
In CFT jargon the k = 2 term in (2.11) is called “shadow term” of the k = 1 term. Its
appearance is a consequence of conformal harmonic analysis on Rd and the equivalence
of scalar representations with dimension δ and d − δ. Only if
d
δ6 +1 (2.17)
2
a scalar field of dimension d − δ exists and we have two equivalent formulations of the
same CFTd : each external leg of a Green function with dimension δ can by amputation be
transformed into a leg with dimension d − δ and vice versa. This shadow term is absent
in (2.13). Instead, there are two terms k ∈ 1, 2, which are obviously connected with the
exchange of some fields of dimension
µ1 + µ4 + n (µ2 + µ3 + n), n ∈ N0 .
Now we remember that unitarity of the S-matrix in perturbative quantum field theory is
usually formulated by Cutkosky’s rule [19–21]: cutting a graph (Fig. 5) through internal
lines and replacing these by a sum over the corresponding states (with appropriate
normalization) gives an absorptive part of the original Green function. In CFT, we can
reduce these states by operator product expansion to states created by conformal blocks of
fields. In Fig. 2 there is one conformal block, namely the conformal field of dimension
δ and all its derivative fields. Due to the reducibility of the conformal representation
transported by the bulk-to-bulk propagator in Fig. 4, the part k = 3 (Eq. (2.16)) contains
Fig. 5. Cutkosky cut through three internal lines of a graph.
an infinite number of tensor fields besides the scalar field of dimension δ. The part k = 1
(Eq. (2.14)) involves an infinite number of conformal tensor fields of rank l with dimension
µ1 + µ4 + l + 2t
and their derivative fields. In fact, there are two parameters l (rank) and t (twist) to label
all blocks exchanged. The same is true for k = 2. Thus we conclude that each critical
exponent corresponds to an infinite tower of conformal blocks, that this tower is determined
by a Cutkosky cut acting on internal and external lines and that 2γk + µ1 + µ4 is in fact
the dimension of the lowest dimensional scalar field in the tower, which in turn can be
understood as “composite field” of the fields belonging to the lines cut. Thus the difference
between CFTd and AdSd+1 theory is in the exchange graphs:
1. There is no shadow term in AdSd+1 ;
2. There are terms from cutting external lines in AdSd+1 . As was argued [17] in
the shadow term in CFTd and the external line terms in AdSd+1 are necessary to
guarantee analytic behavior in the crossed channel.
Indeed, it turns out that such differences between CFTd and AdSd+1 seem to arise only in
the exchange graphs 5 in the direct channel.
Next we consider a CFTd box graph with four unique vertices (Fig. 6). The uniqueness
conditions imply certain constraints on the dimensions of the external and internal fields,
e.g.,
X
4
µ1 + µ3 = µ2 + µ4 = 2d − λi . (2.18)
i=1
This box graph Green function is explicitly known [22] and

X
3
e v) =
G(u, uγk Fk (u, v) (2.19)
k=1
with
γ1 = 0, (2.20)
5 The usual notation is “one-particle reducible graphs”.

Fig. 6. The CFTd box graph with unique vertices.
Fig. 7. The AdSd+1 box graph with generic dimensions.
1
γ2 = (λ1 + λ3 − µ1 − µ4 ), (2.21)
2
1
γ3 = (µ2 + µ3 − µ1 − µ4 ). (2.22)
2
Now we have Cutkosky cuts through the external pairs of lines as well. The pair (1, 4) gives
rise to the exponent γ1 and the pair (2, 3) to γ3 . We note that the box graph with non-unique
vertices (full vertices) has not been calculated yet. Since the critical exponents determine
the towers of conformal blocks that are coupled to the pairs of external and internal fields in
the direct channel, they enter the structure of the field algebra. If the Maldacena conjecture
in the strong version is correct, the large λ 1/N -expanded SYMd with gauge group SU(N)
and N = 4 supercharges has the same field algebra as the holographic image of the AdSd+1
supergravity with coupling constants of order 1/N k , k ∈ 2N. Therefore the results (2.19)–
(2.22) should hold in the case of the Witten graph Fig. 7 as well. We shall prove in the
sequel, that this is correct indeed.
3. The singularity analysis of conformally covariant Green functions
We aim at a direct determination of the critical exponents γk (2.5) before attempting

(k)
the explicit evaluation of integral representations. The Taylor coefficients cmn (2.6) are
then finally represented as integrals which eventually can be evaluated numerically.
Since analytic continuation of the integral representations in the parameters (field and
space dimension) off the domain of absolute convergence is always tacitly understood,
the integrals must necessarily be transformed into absolutely convergent expressions by
subtraction regularization methods before the numerics can be performed. The method of
analyzing conformal Green functions developed by us consists of several steps:
1. We derive a multi-parametric Mellin–Barnes integral representation, where the
integrand Φ depends meromorphically on the Mellin–Barnes parameters and the field and
space dimensions. This function Φ is itself given as an integral of a positive function over
a compact polyhedral domain Kn in Rn with possible zeros and infinities on the boundary
of Kn . Kn is the n-dimensional generalization of the regular tetrahedron K3 or the regular
triangle K2 . Kn is bounded by (n + 1) faces Kn−1 , which intersect in edges Kn−2 , etc.
2. If the integrand is +∞ on a face or a lower-dimensional intersection Kr , then poles
may appear in the Mellin–Barnes parameters on the “right” side of the Mellin–Barnes
contours.
3. Two Mellin–Barnes parameters are connected with the kinematical variables u and
1 − v (2.3) by the powers
uσ1 (v − 1)σ2 . (3.1)
The pole positions of Φ in σ2 lie in N0 and the shift of the σ2 integration contour to +∞
gives simple power series in 1 − v. The pole positions in σ1 lie in different sequences
[
{γk + N0 } (3.2)
k
which leads us to the series representations (2.5), (2.6).
4. Since the zero of an analytically continued integral is difficult to recognize (zeros can
only arise after analytic continuation since the original integrand is a positive function) the
list of candidates for the exponents {γk } is generally too long. We can reduce this list by
different arguments, e.g., a “beta-function argument” and a symmetry argument.
As a nontrivial example of describing our method, we choose the holographic image
of the AdSd+1 graph in Fig. 4. Due to conformal invariance, a Green function can be
completely reconstructed if three of its n > 3 variables are fixed to the values, say
x1 = 0, x2 = ∞, x3 arbitrary unit vector.
We shall exploit this fact by letting x3 → ∞, but keeping translational and scale invariance
µ
2 −1µ e
lim x32 3 G(x1 , x2 , x3 , x4 ) = x12 G(u, v) (3.3)
x3 →∞
with
2
x14 2
x24
u= 2
, v= 2
(3.4)
x12 x12
and
1
1µ = (µ1 + µ2 − µ3 + µ4 ). (3.5)
2
Denoting the bulk variables of AdSd+1 by w1 , w2 , w3 , . . . and boundary variables by xE, yE,
zE, . . . we have for the bulk-to-boundary propagators (i ∈ {1, 2, 3, 4}) [15,16]
µi
w0
Kµi (w, xi ) = cµi , (3.6)
w02 + (w
E − xE)2
where

Γ (µi ) 1
cµi = νi = µi − d (3.7)
π d/2 Γ (νi ) 2
and
µ3 µ
lim x32 Kµ3 (w, x3 ) = cµ3 w0 3 . (3.8)
x3 →∞
For the bulk-to-bulk propagator we use the Mellin–Barnes integral representation [15,16]
Z
+i∞
01 Γ (λ + 2s) 1
Gλ (w, w ) = dsΓ (−s)eiπs
2πi Γ (ν̃ + s + 1) 2π d/2
−i∞
λ+2s
w0 w00
× (3.9)
E − wE0 )2
w02 + w0 20 + (w
with ν̃ = λ − 12 d. The graph of interest (Fig. 4) is, up to coupling constants, factorials and
symmetry factors, represented by the integral
Z Y Y
dµ(w) dµ(w0 ) Gλ (w, w0 ) Kµi (w, xi ) Kµj (w0 , xj ), (3.10)
i∈{1,4} j ∈{2,3}
where dµ is the invariant AdSd+1 measure

E
dw0 d w
dµ(w) = . (3.11)
w0d+1
The integration starts by using a Γ -function auxiliary integration for each denominator
in (3.6), (3.9)
Z∞
1 1
dt t µ−1 e−t kxk
2
= (3.12)
kxk 2µ Γ (µ)
0
distributing the parameters {ti }1,2,4 to the Kµi and r to Gλ . Then w0 and w00 can be
integrated giving
Y 1 − 1 αi
Γ 12 αi ηi 2 (3.13)
2
i∈{1,2}
with
η1 = r + t1 + t4 , η2 = r + t2 (3.14)
and
α1 = µ1 + µ4 + λ + 2s − d, α2 = µ2 + µ3 + λ + 2s − d. (3.15)
E w
The w, E 0 integration is Gaussian and gives
2 d/2
π
exp χ T A−1 χ − D (3.16)
det A
with

η1 −r
A= , (3.17)
−r η2
X
D= ti xi2 , (3.18)
i

t1 x1 + t4 x4
χ= . (3.19)
t2 x2
The exponent (3.16) can be written as a quadratic form
1 X
− βij (xi − xj )2 , (3.20)
det A
i<j
where
β12 = rt1 t2 , β14 = η2 t1 t4 , β24 = rt2 t4 . (3.21)
Since we aim at an expression of the type (2.6), we can use (3.4) to write (3.20) as

2 β0 β14 β24
−x12 1+ u+ (v − 1) (3.22)
det A β0 β0
with
β0 = β12 + β24 = rt2 (t1 + t4 ). (3.23)

Following Symanzik [23], the second and third term in (3.22) are represented by Mellin–
Barnes integrals
Z
+i∞
−x 1
e = dσ Γ (−σ )x σ . (3.24)
2πi
−i∞
Finally we perform one integration by introducing scaled parameters

X
T =r + ti , ti = T τi , i ∈ {1, 2, 4}, r = T ρ, (3.25)
i∈{1,2,4}
so that

det A = T 2 ρ(1 − ρ) + τ2 (1 − ρ − τ2 ) . (3.26)
The remaining parameter integrals can then be summed up into a meromorphic function.
Φ(σ1 , σ2 , s)
Z
µ −1 µ2 −1
= Γ (−σ1 )Γ (−σ2 )Γ (−s) dτ1 dτ2 dτ4 dρδ(1 − τ1 − τ2 − τ4 − ρ)τ1 1 τ2
K2
µ −1 −1µ−σ2
×τ4 4 ρ λ+2s−1 (1 − τ2 )− 2 α1 (ρ + τ2 )− 2 α2 ρτ2 (τ1 + τ4 )
1 1
σ2
σ1 τ4 − 1 d+1µ
× τ1 τ4 (ρ + τ2 ) ρ(1 − ρ) + τ2 (1 − ρ − τ2 ) 2 (3.27)
τ1 + τ4
and this enters a threefold Mellin–Barnes integral
e v) = 1 Γ (µ3 )
G(u, 3d Q4 (2πi)−3
8π 2 i=1 Γ (νi )
Z+i∞
ZZ
Γ ( 1 α1 )Γ ( 12 α2 )Γ (1µ + σ2 + σ2 )
× dσ1 dσ2 ds 2
Γ (ν̃ + s + 1)
−i∞
× eiπs uσ1 (v − 1)σ2 Φ(σ1 , σ2 , s) (3.28)
with 1µ = 12 (µ1 + µ2 − µ3 + µ4 ), see (3.5). Such a representation (3.27), (3.28) of any
four-point function for CFTd or AdSd+1 field theory is the starting point for our singularity
analysis, leading to the critical exponents.
In this particular case we can simplify the integral representation (3.27) by integrating
over ξ in
τ1 = τ ξ, τ4 = τ (1 − ξ ), (3.29)
Φ(σ1 , σ2 , s)
Γ (µ1 + σ1 )Γ (µ4 + σ1 + σ2 )
= Γ (−σ1 )Γ (−σ2 )Γ (−s)
Γ (µ1 + µ4 + 2σ1 + σ2 )
Z
× dτ dτ2 dρ δ(1 − τ − τ2 − ρ)
K2
µ −1µ−σ −1
× τ µ1 +µ4 −1µ+σ1 −1 (1 − τ )− 2 α2 +σ1 τ2 2 (1 − τ2 )− 2 α1
1 1
1
− 1 d+1µ
× ρ λ+2s−1µ−σ1 −1 ρ(1 − ρ) + τ2 (1 − ρ − τ2 ) 2 . (3.30)
Here σ2 has vanished from the integral into the factor in front. Except for the factor
Γ (−σ2 ), there is no pole to the right of the σ2 Mellin–Barnes contour. This a general
feature since (see (3.23)) in Kn
β24 β24
06 = 6 1. (3.31)
β0 β12 + β24
There are obviously poles from the faces τ2 = 0 and ρ = 0 in σ1 , arising from the Mittag–
Leffler expansion
∞
X
∼ (−1)nδ (n) (t)
t µ−1 Θ(t) = (3.32)
poles only n!(µ + n)
n=0
with positions −µ ∈ N0 . Including the poles in σ1 from the factor Γ (−σ1 ), we have three
possibilities: (n ∈ N0 )
(1) σ1 = n, (3.33)
(2) from τ2 = 0: σ1 = µ2 − 1µ + n, (3.34)
(3) from ρ = 0: σ1 = λ + 2s − 1µ + n. (3.35)
In the cases (1) and (2) we get the critical exponents
γ1 = 0, (3.36)
1
γ2 = (µ2 + µ3 − µ1 − µ4 ) (3.37)
2
whereas case (3) necessitates knowledge of the pole positions in s. One possibility is that
these poles are produced by Γ (−s), then
σ1 = λ − 1µ + n, (3.38)
γ3 = λ − 1µ. (3.39)
There is another candidate for poles in σ1 , namely the intersection of the faces (2) and (3):
τ2 = ρ = 0, τ = 1. (3.40)
We use the parameters
ρ = ωψ, τ2 = ω(1 − ψ), τ = 1 − ω. (3.41)
The behavior of the integrand at w → 0 is given by

Z
dω ω(µ2 −1µ−σ1 )+(λ+2s−1µ−σ1)+(− 2 α2 +σ1 )+(− 2 d+1µ)−1 .
1 1
(3.42)
0
The exponent is
1 1
λ + s − (µ1 + µ4 ) − σ1 − 1 (3.43)
2 2
and gives rise to poles in σ1 at
1 1
(4) τ2 = ρ = 0: σ1 = λ + s − (µ1 + µ4 ) + n. (3.44)
2 2
If the s poles are from Γ (−s), we get from (3.44)
1
σ1 = (λ − µ1 − µ4 ) + n, (3.45)
2
1
γ4 = (λ − µ1 − µ4 ). (3.46)
2
But there exist other s-poles. If we consider (3.44), set n = 0 and insert the delta function
following from (3.32) into (3.30), there remains the ψ-integral, see (3.41)
Z1

dψ ψ λ+2s−1µ−σ1 −1 (1 − ψ)µ2 −1µ−σ1 −1 σ
1 = 2 λ+s− 2 (µ1 +µ4 )
1 1
0
1
= Γ 1 (µ2 + µ3 − λ) − s Γ 12 (λ + µ3 − µ2 ) + s (3.47)
Γ (µ3 ) 2
which shows, that there exist relevant s-poles from the first factor in the numerator. For
arbitrary n in (3.44) the poles lie at
1
s + n = (µ2 + µ3 − λ) + n0 , n0 ∈ N0 . (3.48)
2
If we consider (3.35) at n = 0 and insert it together with the delta function (3.32) into
(3.30), then the integral turns into a beta-function
Z1
(µ2 −1µ−σ1 −1)+(− 21 α2 +σ1 )+(− 12 d+1µ)
dτ2 τ2
0
×(1 − τ2 )(µ1 +µ4 −1µ+σ1 −1)− 2 α1 +(− 2 d+1µ)

1 1
1
= Γ 12 (µ2 − µ3 − λ) − s Γ 12 (λ − µ2 + µ3 ) + s . (3.49)
Γ (0)
The denominator is unchanged if we let n in (3.35) assume arbitrary values from N0 . Thus
the denominator of the beta-function lets the singularity (3.35) vanish, implying that (3.38),
(3.39) do not exist either. Only in exceptional cases do we get control over the zeros when
we can perform an integral completely. Often the integral is a beta-function, then we call
our way of proof “the beta-function argument”. More effort is needed to evaluate integrals
in terms of functions p+1 Fp (1) in which case the zeros are also controllable.
A simple but surprisingly powerful argument to eliminate whole sequences of poles
comes from the symmetry of the graph (Fig. 4). We define this symmetry to consist of
those mappings of the graph on itself:
(a) which lead to the same graph after an appropriate relabelling of the external
coordinates and the field dimensions;
(b) leave u and v invariant.
In the case of Fig. 4, this leads to a group Z2 × Z2 , generated by the reflections
S1 : 1 ←→ 2, 3 ←→ 4,
S2 : 1 ←→ 4, 3 ←→ 2. (3.50)
While the Green function G(x1 , x2 , x3 , x4 ) is invariant under Z2 × Z2 by definition, the
e is not. Let Si (G)
invariant function G e denote the function obtained by applying (3.50) to
e
the dimensions in G, then from (2.1) we obtain
e v) = uδi vi Si (G)(u,
G(u, e v) (3.51)
with
1 1
S1 : δ1 = (µ2 + µ3 − µ1 − µ4 ), 1 = (µ1 + µ3 − µ2 − µ4 ), (3.52)
2 2
S2 : δ2 = 0, 2 = 1 . (3.53)
Inserting (3.51) into (2.5), we see that the labels {k} of γk are submitted to a representation
of Z2 × Z2 : Si → σi , so that:
Si (γk ) + δi = γσi (k) , (3.54)
Si (Gk )v = Gσi (k) .
i
(3.55)
Holomorphy of Gk at v = 1 is obviously not touched by (3.55). Applying (3.54) to the
graph Fig. 4, we find
σ1 (1) = 2, σ1 (2) = 1,
σ2 (1) = 1, σ2 (2) = 2 (3.56)
and
σ1,2 (4) = 4, (3.57)
whereas γ3 does not fit into any representation.
4. The AdS box graph
Now we turn to the box graph Fig. 7. In terms of bulk-to-bulk propagators Gλ and bulk-
to-surface propagators Kµ , the Green function is given by the integral
Z Y
4
G(x1 , x2 , x3 , x4 ) = dµ(wi ) Kµi (xi , wi )Gλi (wi , wi+1 ), w5 = w1 . (4.1)
i=1
Again we consider the limit (3.3), (3.8). Due to the four bulk-to-bulk propagators, the
e v) has the form of a sixfold Mellin–Barnes integral 6
invariant Green function G(u,
Z Z
+i∞ ZZ Z Z (Y
+i∞ 4
)
1 Γ (µ ) Γ ( 12 αi )
e v) =
G(u, Q
3
(2πi)−6
dσ1 dσ2 dsi
28 π 2d 4i=1 Γ (νi ) Γ (ν̃i + si + 1)
−i∞ i=1 −i∞
P
× Γ (1µ + σ1 + σ2 )e iπ
u (v − 1) Φ(σ1 , σ2 , s1 , s2 , s3 , s4 ),
i i σ1
s σ2
(4.2)
where
αi = µi + λi + 2si + λi−1 + 2si−1 − d (λ0 = λ4 , s0 = s4 ) (4.3)
and the meromorphic function Φ is given by
Φ(σ1 , σ2 , s1 , s2 , s3 , s4 )
Z ! !
Y
2 Y
4 Y
4
µ −1
Y
4
λ +2s −1 − 1 αj
= Γ (−σi ) Γ (−sj ) dτi τi i dρj ρj j j j 2
i=1 j =1 K6 i=1(6=3) j =1
!
X
4 X
4
−1µ−σ1 −σ2
ρj δ(A)− 2 +1µ f0
d
×δ 1 − τi − f1σ1 f2σ2 .
i=1(6=3) j =1
6 All the techniques and notations used are the same as in the preceding section.
Here
i = τi + ρi + ρi−1 (τ3 = 0, ρ0 = ρ4 ) (4.4)
and the remaining functions f0 , f1 , f2 and δ(A) can be represented best with the help of
elementary symmetric polynomials
S2 (1, 2, 3) = ρ1 ρ2 + ρ1 ρ3 + ρ2 ρ3 ,
S3 (1, 2, 3, 4) = ρ1 ρ2 ρ3 + ρ1 ρ3 ρ4 + ρ2 ρ3 ρ4 + ρ1 ρ2 ρ4 , (4.5)
namely,

f0 = τ2 τ1 τ4 S2 (1, 2, 3) + (τ1 + τ4 )S3 (1, 2, 3, 4) , (4.6)

f1 = τ1 τ4 S3 (1, 2, 3, 4) + τ2 ρ4 (ρ2 + ρ3 ) , (4.7)

f2 = τ2 τ4 S3 (1, 2, 3, 4) + τ1 ρ2 ρ3 , (4.8)
δ(A) = τ1 τ2 τ4 (ρ2 + ρ3 ) + τ1 τ2 S2 (2, 3, 4) + τ1 τ4 S2 (1, 2, 3)
+ τ2 τ4 (ρ1 + ρ4 )(ρ2 + ρ3 ) + (τ1 + τ2 + τ4 )S3 (1, 2, 3, 4). (4.9)
The function δ(A) originates from the determinant in the Gaussian integration. It is obvious
that
f2
06 6 1 on K6 (4.10)
f0
so that the only relevant poles in σ2 arise from Γ (−σ2 ).
The analysis of the pole positions in σ1 is rather involved. In the sequel n0 ∈ N0 holds
throughout. There is one face of type K5 producing a singularity:
(I) τ2 = 0: poles appear at
1
σ1 = (µ2 + µ3 − µ1 − µ4 ) + n0 , (4.11)
2
1
γ1 = (µ2 + µ3 − µ1 − µ4 ). (4.12)
2
There are two faces of K4 type leading to poles.
(II) ρ1 = ρ2 = 0: we introduce the parameters
ρ1 = ρξ, ρ2 = ρ(1 − ξ ) (4.13)
and let ρ → 0. This gives pole positions
σ1 = λ1 + λ2 + 2(s1 + s2 ) − 1µ + n0 . (4.14)
If the pole positions of s1 , s2 are chosen from N0 , we get
γ2 = λ1 + λ2 − 1µ. (4.15)
The other case is
(III) ρ1 = ρ3 = 0: this case is treated analogously to case (II). We find poles at
σ1 = λ1 + λ3 + 2(s1 + s3 ) − 1µ + n0 . (4.16)
If the pole positions of s1 , s3 are from N0 , we find
γ3 = λ1 + λ3 − 1µ. (4.17)
Now we come to the intersections K5 ∩ K4 and K4 ∩ K04 of K3 type.

(IV) ρ1 = ρ2 = ρ3 = 0: we choose as parameters
X
ρi = ρξi , i ∈ {1, 2, 3}, ξi = 1 (4.18)
i
and let ρ → 0. Pole positions are

1 1
σ1 = λ1 + (λ2 + λ3 ) + 2s1 + s2 + s3 − (µ1 + µ2 + µ4 ) + n0 . (4.19)
2 2
If the {si }3i=1 have poles in N0 , we get
1 1
γ4 = λ1 + (λ2 + λ3 ) − (µ1 + µ2 + µ4 ). (4.20)
2 2
(V) ρ1 = ρ2 = τ2 = 0: we choose as parameters
ρi = ρξi , i ∈ {1, 2}, τ2 = ρξ3 (4.21)

and let ρ → 0. The poles of σ1 appear at
1
σ1 = (λ1 + λ2 − µ1 − µ4 + µ3 ) + s1 + s2 + n0 . (4.22)
2
Provided the poles of s1 , s2 are in N0 , we find
1
γ5 = (λ1 + λ2 − µ1 − µ4 + µ3 ). (4.23)
2
However, if we perform some of the integrations after insertion of the delta function
δ (0) (ρ) corresponding to the pole (4.28) by (3.32), we obtain a beta function with
denominator Γ (−2n0 ). So these poles (V) cancel completely.
(VI) ρ1 = ρ3 = τ2 = 0: we proceed as in the case (V) and get as pole positions
1 1
σ1 = λ1 + λ3 + (µ2 + µ3 − µ1 − µ4 ) − d + 2(s1 + s3 ) + n0 (4.24)
2 2
which, if the poles of s1 , s3 are in N0 , gives
1 1
γ6 = λ1 + λ3 + (µ2 + µ3 − µ1 − µ4 ) − d. (4.25)
2 2
(VII) Finally, there is one K2 face: τ2 = ρ1 = ρ2 = ρ3 = 0: coordinates are
X
ρi = ρξi , i ∈ {1, 2, 3}, τ2 = ρχ4 , ξi = 1 (4.26)
i
and we let ρ → 0. We get the pole positions

1
σ1 = (λ1 + λ3 − µ1 − µ4 ) + s1 + s3 + n0 . (4.27)
2
If s1 , s3 have poles in N0 , we obtain
1
γ7 = (λ1 + λ3 − µ1 − µ4 ). (4.28)
2
The symmetry group of the graph Fig. 7 is the same as that of Fig. 4: Z2 × Z2 . It acts on
the λi as
S1 (λi ) = λi , i ∈ 1, 3, S1 (λ2 ) = λ4 , S1 (λ4 ) = λ2 , (4.29)

S2 (λi ) = λi , i ∈ 2, 4, S2 (λ1 ) = λ3 , S2 (λ3 ) = λ1 . (4.30)
This rules out all γ ’s, except γ1 , γ7 and of course γ0 = 0, which originates from the σ1
poles of Γ (−σ1 ). Thus the AdS box graph has the same critical exponents as the CFT box
graph Fig. 6.
The poles in s4 are all from Γ (−s4 ). The other variables (s1 , s2 , s3 )produce poles of the
function Φ (4.4) that can be ordered in (triple) sequences

(ν1 + n1 , ν2 + n2 , ν3 + n3 ), νi fixed, ni ∈ N0 running . (4.31)
In Tables 1 and 2 we list all possible triples (ν1 , ν2 , ν3 ) and their connection (“origin”) with
the σ1 singularities (I)–(VII). The entries in the tables originating from the cases (VI) or
(VII) are marked by (∗). The corresponding ni runs over 12 N0 (not N0 ).
Table 1
Sequences of {s1 , s2 , s3 } poles contributing to uγ1 F1 (u, v), γ1 = 12 (µ2 + µ3 − µ1 − µ4 )
Origin ν1 ν2 ν3
(I) 0 0 0
(II) 1 (µ − λ − λ ) 0 0
2 2 1 2
(II) 0 1 (µ − λ − λ ) 0
2 2 1 2
(II), (IV), (VII) 0 1 (µ − λ − λ )∗ 1 (µ + µ − λ − λ )
2 2 1 2 2 2 3 1 3
(II), (IV), (VII) 1 (µ − λ − λ )∗ 0 1 (µ + λ − λ )∗
2 2 1 2 2 3 2 3
(II), (IV), (VII) 1 (µ + µ − λ − λ ) 1 (λ − λ − µ )∗ 0
2 2 3 1 3 2 3 2 3
(VI) − 12 (λ1 + λ3 ) + 14 d∗ 0 0
(VI) 0 0 − 12 (λ1 + λ3 ) + 14 d∗
(VII) 1 (µ + µ − λ − λ ) 0 0
2 2 3 1 3
(VII) 0 0 1 (µ + µ − λ − λ )
2 2 3 1 3
(VII) 0 1 (3µ − d − λ − λ ) 1 (µ + µ − λ − λ )
2 2 1 2 2 2 3 1 3
(VII) 1 (3µ − d − λ − λ ) 0 1 (λ − λ − 2µ + µ + d)
2 2 1 2 2 2 3 2 3
(VII) 1 (µ + µ − λ − λ ) 1 (−λ + λ + 2µ − µ − d) 0
2 2 3 1 3 2 2 3 2 3
Table 2
Sequences of {s1 , s2 , s3 } poles contributing to uγ7 F7 (u, v), γ7 = 12 (λ1 + λ3 − µ1 − µ4 )
Origin ν1 ν2 ν3
(IV) 0 1 (µ − λ − λ ) 0
2 2 1 2
(VII) 0 0 0
(VII) 0 1 (−λ − 2λ + λ + µ − µ )∗ 0
4 1 2 3 2 3
(VII) 0 1 (λ − λ + 2λ + µ − 2µ − d)∗ 0
2 1 2 3 2 3
5. Concluding remarks
We have proved that for the box graphs of CFTd and AdSd+1 supergravity, we obtain
the same critical exponents, namely those which are determined from the “Cutkosky rule”
with external lines included. We suggest that this behavior is also shown by other one-
particle-irreducible graphs. Each critical exponent γk belongs to one or more sequence of
poles in the Mellin–Barnes parameters (s1 , s2 , s3 ), each of which is generated by a triple
(k)
(ν1 , ν2 , ν3 ) (see (4.31)), and each sequence contributes to the coefficient cmn in (2.6). The
larger the number of γk , ν1 , ν2 , ν3 that are nonzero, the smaller the number of remaining
integrations. More details on this can be found in [24].
Acknowledgements
The authors thank A.C. Petkou for interesting discussions during the initial stages and
one of us (W.R.) thanks the staff of the Werner-Heisenberg-Institut in Munich for their
hospitality during the final stage of this work.
References
[1] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231–252, hep-th/9711200.
[2] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from noncritical string
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 105–291,
hep-th/9805028.
[4] J.L. Petersen, Introduction to Maldacena Conjecture on AdS/CFT, hep-th/9902131.
[5] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory
and gravity, hep-th/9905111.
[6] A. Bilal, C.-S. Chu, A note on the Chiral Anomaly in the AdS/CFT correspondence and 1/N 2
correction, hep-th/9907106.
[7] A. Bilal, C.-S. Chu, Testing the AdS/CFT correspondence beyond the large N , hep-th/0003129.
[8] A. Petkou, K. Skenderis, A non-renormalization theorem for conformal anomalies, hep-
th/9906030.
[9] M. Bianchi, S. Kovacs, Non-renormalization of extremal correlators in N = 4 SYM theory,
hep-th/9910016.
[10] B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Extremal correlators in four-
dimensional SCFT, hep-th/9910150.
[11] E. D’Hoker, S.D. Mathur, A. Matusis, L. Rastelli, The operator product expansion of N = 4
SYM and the 4-point functions of supergravity, hep-th/9911222.
[12] G. Arutyunov, S. Frolov, A.C. Petkou, The operator product expansion of the lowest weight
CPOs in N = 4 SYM4 at strong coupling, hep-th/0005182.
[13] S. Penati, A. Santambrogio, D. Zanon, Two-point functions of chiral operators in N = 4 SYM
at order g 4 , hep-th/9910197.
[14] S. Penati, A. Santambrogio, D. Zanon, Correlation functions of chiral primary operators in
perturbative N = 4 SYM, hep-th/0003026.
[15] H. Liu, Scattering in anti-de Sitter space and operator product expansion, hep-th/9811152.
[16] D. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Correlation functions in the CFTd /AdSd+1
[17] L. Hoffmann, A. Petkou, W. Rühl, Aspects of the conformal operator product expansion in
[18] V.K. Dobrev, G. Mack, V.B. Petkova, S.G. Petrova, I. Todorov, Harmonic analysis on the n-
dimensional conformal group and its applications to conformal quantum field theory, Lecture
Notes in Physics, Vol. 63, Springer-Verlag, Berlin, 1977.
[19] R.E. Cutkosky, J. Math. Phys. 1 (1960) 429.
[20] M. Veltman, Physica 29 (1963) 186.
[21] C. Itzykson, J.-B. Zuber, Quantum Field Theory, McGraw-Hill, New York, 1980, Section 6-3-4.
[22] K. Lang, W. Rühl, The critical O(N ) σ -model at dimension 2 < d < 4 and order 1/N 2 : operator
product expansions and renormalization, Nucl. Phys. B 377 (1992) 371.
[23] K. Symanzik, On calculations in conformal invariant field theories, Lett. Nuovo Cimento 3
(1972) 734.
[24] L. Hoffmann, PhD Thesis, University of Kaiserslautern, to appear.
Electroweak Bloch–Nordsieck violation at

the TeV scale: “strong” weak interactions? ✩
Marcello Ciafaloni a , Paolo Ciafaloni b,∗ , Denis Comelli c
a Dipartimento di Fisica, Università di Firenze e INFN, Sezione di Firenze, I-50125 Florence, Italy
b INFN, Sezione di Lecce, Via per Arnesano, I-73100 Lecce, Italy
c INFN, Sezione di Ferrara, Via Paradiso 12, I-35131 Ferrara, Italy
Received 10 April 2000; revised 31 May 2000; accepted 8 August 2000
Abstract
Hard processes at the TeV scale exhibit enhanced (double log) EW corrections even for inclusive
observables, leading to violation of the Bloch–Nordsieck theorem. This effect, previously related
to the non-abelian nature of free EW charges in the initial state (e− e+ , e− p, pp, . . .), is here
investigated for fermion initiated hard processes and to all orders in EW couplings. We find that
the effect is important, especially for lepton initiated processes, producing weak effects that in some
cases compete in magnitude with the strong ones. We show that this (double log) BN violating effect
has a universal energy dependence, related to the Sudakov form factor in the adjoint representation.
The role of this form factor is to suppress cross section differences within a weak isospin doublet, so
that at very large energy the cross sections for left-handed electron–positron and neutrino–positron
scattering become equal. Finally, we briefly discuss the phenomenological relevance of our results
for future colliders.  2000 Elsevier Science B.V. All rights reserved.
PACS: 11.15.-q; 12.15.Lk; 12.38.Cy

Keywords: Bloch–Nordsieck violation; Resummation of electroweak corrections
1. Introduction
Enhanced electroweak corrections at the TeV scale have been recently investigated by
various authors [1–6] starting from the observation, made by two of us [1], that double
and single logarithms of Sudakov type are present and sizeable in fixed angle fermion
antifermion annihilation processes at NLC energies. These effects produce, namely,
✩
Work supported in part by E.U. QCDNET contract FMRX-CT98-0194 and by MURST, Italy.
E-mail addresses: ciafaloni@fi.infn.it (M. Ciafaloni), paolo.ciafaloni@le.infn.it (P. Ciafaloni),
comelli@fe.infn.it (D. Comelli).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 0 8 - 3
360 M. Ciafaloni et al. / Nuclear Physics B 589 (2000) 359–380
energy-growing corrections ∝ αW log2 (s/M 2 ), the weak scale M ∼ 90 GeV providing

a physical cutoff for infrared and collinear divergences.
In a recent paper [7] we have pointed out a different but related effect, which is peculiar
of electroweak interactions. Due to the non-abelian nature of electroweak charges in the
initial state, the double logs persist at inclusive level, thus leading to violation of the
Bloch–Nordsieck theorem [8–10], in the sense that the dependence on the IR cutoff M
is not washed out when summing real and virtual corrections, as is usually the case. The
peculiar aspect which makes such double logs observable is symmetry breaking itself,
which generates the physical cutoff M on one hand, and allows the preparation of initial
states as free abelian charges on the other. 1
In fact, the BN theorem violation was initially pointed out for QCD [11,12], where it
would imply genuine IR divergences at partonic level, but was found eventually to have
no physical consequences because of the color averaging forced by coupling to colorless
hadrons. For instance, non factorized IR singular terms do occur in Drell–Yan processes,
but at next-to-leading level only and turn out to be higher twist [13–19]. The “preconfine-
ment” features, pointed out at various stages [16,20,21] mean that free quark asymptotic
states make no sense, even at perturbative level, because of form factors analogous to the
ones discussed here. The situation is different of course in the EW case: the analogous of
color averaging would mean for instance averaging over the cross sections for νe+ and
e− e+ ; this is meaningless from an experimental point of view.
In this paper we investigate the structure of double log EW corrections to all orders
for light fermion initiated hard processes, and we characterize them by a universal energy
dependence, related to the EW Sudakov form factor in the adjoint representation. The
core of such analysis is provided by two lines of thought. One, explained in Section 3, is
based on the observation that only W contributions are BN violating, the Z and γ ones
being canceled between virtual and real emission terms. The energy dependence is then
universal because the W couples universally to left handed fermions and can be calculated
to all orders by a simple Feynman diagram technique.
The other line of thought, explained in Section 5 following the coherent state formalism
of Section 4, is based on the isospin structure of the overlap matrix, describing the squared
matrix element. With a proper choice of indices, the singlet projection provides the isospin
average (i.e., for instance, σνe+ + σe− e+ ), and the vector projection the cross sections
difference (i.e., σνe+ − σe− e+ ). Only the latter is suppressed by a Sudakov form factor,
which refers to the adjoint representation, and is thus universal. The latter approach is made
completely rigorous by the use of the general coherent state formalism [22–25] introduced
for cancellation theorems in Section 2, and by the important observation that the photon
scale plays no role for the (inclusive) BN violating terms. This fact follows from the explicit
cancellation of Z and γ contributions found by the diagrammatic approach, so that the
effect starts, and the symmetry is effectively restored at, the same W threshold M.
1 For instance, an electron is prepared at low energy M as an abelian (QED) charge. As it is accelerated at
energies M, its charge acquires a fully non-abelian character.
M. Ciafaloni et al. / Nuclear Physics B 589 (2000) 359–380 361
Applications to physical processes are analyzed in Section 6 and discussed in Section 7.

We limit ourselves to lepton and/or quark large angle scattering as trigger process and
we classify the results according to the initial state, which is provided by the accelerator.
Important corrections are found, especially for lepton initiated processes, as in the case of
eē (NLC) or ep and ēp. Roughly speaking, we find in this case a 50 percent reduction
in hadron beams compared to lepton beams, mostly due to the hadrons acting as weak
isospin mixtures in the initial state. Another important feature is the strong dependence of
the BN violating effects on the polarization of the colliding beams, of particular relevance
for NLCs [26,27].
2. Cancellation theorems and Bloch–Nordsieck violation
We consider the structure of soft interactions accompanying a hard SM process, of type

I I I I
α1 p1 , α2 p2 → α1F p1F , α2F p2F , . . . , αnF pnF , (1)
where α, p denote isospin/color and momentum indices of the initial and final states, that
we collectively denote by {αI pI } and {αF pF }. The S matrix for such a process can be
written as an operator in the soft Hilbert space HS , that collects the states which are almost
degenerate with the hard ones, in the form

S = UαFF βF as , as† SβHF βI (pF , pI )UαII βI as , as† , (2)
where U F and U I are operator functionals of the soft emission operators as , as† .
Eq. (2) is supposed to be of general validity [23–25,28–30], because it rests essentially
on the separation of long-time interactions (the initial and final ones described by the U ’s),
and the short-time hard interaction, described by S H . The real problem is to find the form of
the U ’s , which is well known in QED [22], has been widely investigated in QCD [23–25],
and is under debate in the electroweak case [5,6]. Their only general property is unitarity
in the soft Hilbert space HS , i.e.,
† †
Uαβ Uβα 0 = Uαβ Uβα 0 = δαα 0 . (3)
The key cancellation theorem satisfied by Eq. (2) is due to Lee, Nauenberg and Kinoshita
[31,32], and states that soft singularities cancel upon summation over initial and final soft
states which are degenerate with the hard ones:
f ∈∆(p
XF )
hf |S|ii2 = TrH U I † S H † U F † U F S H U I
S
i∈∆(pI )

= TrαI S H † (pF , pI )S H (pF , pI ) , (4)
where ∆(pI , pF ) denote the sets of such soft states, and we have used the unitarity
property (3).
Although general, the KLN theorem is hardly of direct use, because it involves the sum
over the initial degenerate set, which is not available experimentally. In the QED case,
however, there is only an abelian charge index, so that UI commutes with S H † S H , and
cancels out by sum over the final degenerate set only. This is the BN theorem: observables
which are inclusive over soft final states are infrared safe.
If the theory is non-abelian, like QCD or the electroweak one under consideration, the
BN theorem is generally violated, because the initial state interaction is not canceled, i.e.,
by working in color space,
X
hf |S|ii2 = S h0|U I † 0 S H † S H 0 U I |0iS
α β I I β β βI αI
I I
f ∈∆(pF )

= SH †SH αI αI
+ 1σαI , (5)
where the αI indices are not summed over, and 1σαI is, in general, nonvanishing and IR
singular.
Fortunately, in QCD the BN cancellation is essentially recovered because of two
features: (i) the need of initial color averaging, because hadrons are colorless, and (ii)
the commutativity of the leading order coherent state operators (U l ) for any given color
indices [23–25]:
l
U l = U l as − as† , Uαβ , Uαl 0 β 0 = 0. (6)
We obtain therefore
X l† X
Uα β 0 S H † S H β 0 β Uβl I αI = S H † S H β 0 β Uβl I αI Uαl† β 0 = Trcolor S H † S H , (7)
I I I I I I I I
color color
thus recovering an infrared safe result (for subleading features, see Refs. [14–19]).
In the electroweak case, in which M provides the physical infrared cutoff, there is no
way out, because the initial state is prepared with a fixed non-abelian charge. Therefore
Eq. (5) applies, and double log corrections ∼ αW log2 (s/M 2 ) must affect any observable
associated with a hard process, even the ones which are inclusive over final soft bosons.
This fact is at first sight surprising, if one is used to think of electrons and protons as
abelian, rather than non-abelian, charges. In fact, it is usually implicitly assumed that
inclusive observables depend only on energy and on running couplings, while the double
logs represent an explicit M (infrared cutoff) dependence. Here we emphasize precisely
this presence of EW double logs, and we investigate their magnitude.
3. Lowest order calculation and picture of higher orders
In order to compute the uncanceled double logs, we first notice that, according to Eq. (5),
only initial state interactions need to be taken into account. Here we give a diagrammatic
account of the calculations, by considering EW corrections to the overlap matrix OH ≡
†
SH SH , in which soft bosonic lines are emitted and/or absorbed on initial lines only. In the
following sections we shall give a more rigorous treatment, based on the coherent state
approach that we have introduced previously.
We start treating the lowest order soft EW contributions to 1σ ≡ σ − σ H for a basic
hard process involving a hard scale E MW ∼ MZ ≡ M, and two massless fermions.
We consider here the case for two L initial fermions, both carrying non-abelian isospin
Fig. 1. Unitarity diagrams for (a) virtual and (b) real emission contributions to lowest order initial
state interactions in the Feynman gauge. Sum over gauge bosons a = γ , Z, W and over permutations
is understood.
(SU(2)) indices; a more general treatment is demanded to Section 5. Since we only

consider inclusive processes, a sum over degenerate final states is understood, and we drop
the superscript indicating initial states: αiI → αi . In isospin space, the hard cross section
structure is then defined by the so-called hard overlap matrix, describing the squared matrix
†
element: hβ1 β2 |SH SH |α1 α2 i ≡ (OH )β1 β2 ,α1 α2 (see Fig. 1). While for cross sections we
always have αi = βi , we leave open the possibility that αi 6= βi and see OH as an operator
in isospin space with four indices.
Since pure QED corrections, at energies below M, cancel out automatically as noticed
before, we limit ourselves to bosonic energies M w E, for which the gauge bosons
γ , Z, W have all approximately the same momenta, with M acting as the only IR cutoff.
This amounts eventually to setting the effective photon scale λ = M, and neglecting
symmetry breaking effects at the inclusive double log level we are working. We would
like to stress again the fact that this holds only for completely inclusive quantities, while
for (partially) exclusive observables the presence of a new scale λ 6= M is unavoidable and
may lead to symmetry breaking effects, a topic which is currently under discussion [5,6].
At lowest order, the above assumption is easily verified, because in the hard (Born)
matrix element, since we work in a limit in which all kinematical invariants are much
bigger than gauge bosons masses: |s|, |t|, |u| M 2 , the full gauge symmetry SU(3) ⊗
SU(2) ⊗ U (1) is restored. Boson emission and absorption is then described by the external
(initial) line insertions of the eikonal currents
µ
p pµ
Jaµ = g 1 t1a − t10a + 2 t2a − t20a ,
kp1 kp2
µ
µ 0 p1 0
p2µ 0

J0 = g Y1 − Y1 + Y2 − Y2 . (8)
kp1 kp2
Here a = 1, 2, 3 is the SU(2) index, we work in the unbroken basis A0 = cW γ − sW Z,
A3 = sW A + cW Z, g and g 0 being the usual electroweak couplings with sW /cW = g 0 /g.
The isospin operator t1 (t10 ) acts on the α1 (β1 ) index, and so on. The currents (8) are
conserved (k · J = 0) because of charge conservation at Born level:
t1a + t2a = t10a + t20a , Y1 + Y2 = Y10 + Y20 . (9)
Since both Y and t3 are diagonal matrices, the A0 , A3 (or γ , Z) contributions to the
cross section cancel out automatically between virtual and real emission terms (ti3 = ti03 ,
Yi = Yi0 ). The W contributions are instead provided by:
2p1 p2 X
g2 (t 1 − t 01 ) · (t 2 − t 02 ), t1 · t2 ≡ t1a t2a , (10)
(kp1 )(kp2 ) a
where we have kept, for notational convenience, the vanishing A3 contribution, and the
charge factor can be replaced, because of the conservation (9), by
(t 1 − t 01 ) · (t 2 − t 02 ) = −(t 1 − t 01 )2 = 2t 1 · t 01 − t 21 − t 02
1. (11)
The latter expression provides the charge computation in the axial gauge, because in
this gauge the W emission and absorption takes place on the same leg, for both virtual
(−2t 21 ) and real emission (2t 1 · t 01 ) contributions. We can see from Fig. 1 the reason for
noncancellation between virtual and real one loop corrections: the crucial point is that
while γ , Z emission does not change the initial state, W emission does. Then, in the W
case virtual corrections (Fig. 1a) are proportional to σeē , while real corrections (Fig. 1b) are
of opposite sign but proportional to σν ē 6= σeē , giving rise to a non complete cancellation.
In order to simplify our considerations, let us note that, because of isospin conservation,
we must have (see (44))
σν ν̄ = σeē , σν̄e = σν ē . (12)
We can therefore limit ourselves to the cross sections σα ≡ σα ē where α = ν, e is a single
isospin index. Since real W emission changes the isospin index, while virtual corrections
don’t, we can summarize the first order calculations based on (8) and (10) in the form

1σα1 = LW (s) − δαβ + (τ1 )αβ σβH , (13)
where τ1 is the customary Pauli matrix, the superscript 1 denotes the order of correction,
and
ZE
g2 d3 k 2p1 p2 αW 2 E
2 g2
LW (s) = = log , αW = (14)
2 2wk (2π)3 (kp1 )(kp2 ) 4π M2 4π
M
is the eikonal radiation factor for W exchange.
From Eq. (13) we get in particular:

1σe1 = LW (s) σνH − σeH = −1σν1 , (15)
which was the main result of [7], leading also to 1σe1 + 1σν1 = 0, i.e., to cancellation upon
isospin averaging.
The insertion mechanism just explained can be iterated to higher orders in the strong
ordering region M w1 w2 · · · E, by noticing that the current (8), after
cancellation of the γ , Z contributions takes up only a W part with only one cutoff (M),
which cannot induce symmetry breaking. Since the action of real W emission is simply
that of interchanging ν and e indices, we end up with a two channel problem where the
energy evolution Hamiltonian — derived from Eq. (13) — commutes at different energies.
The outcome is a simple exponentiation of the first order result:
∞
X
σα (s) = σαH + 1σαn = exp LW (s)(τ1 − 1) αβ σβH , (16)
n=0
or, by simple algebra,
σνH + σeH σ H − σeH −2LW (s)
σe,ν = ∓ ν e . (17)
2 2
This means that on average the double logs cancel out (σeē + σν ē = σeHē + σνHē ), while
the ν-beam and e-beam difference σν ē − σeē decreases exponentially with the universal
exponent 2LW (s), and vanishes eventually at infinite energy.
Apart from the important phenomenological implications of Section 6, the above results
have a nice theoretical interpretation in a non-abelian framework, which will be illustrated
in the following sections. First, the exponent 2LW (s) can be related to the form factor in the
adjoint representation, which is in fact a typically non-abelian quantity; because of gauge
invariance the exponent is also universal, i.e., the same for any fermion doublet in the initial
state. A similar relationship was noticed for QCD in [13,16]. Furthermore, the asymptotic
equality of neutrino and electron beam cross sections is related to the idea that in the unbro-
ken limit (M → 0), non-abelian charges are actually not observable as asymptotic states,
as already noticed in QCD in connection with preconfinement [20,21] and factorization
violating contributions [16]. In fact, if there were no symmetry breaking (infinite 2LW (s)),
Eq. (17) would imply that any coherent superposition of electron and neutrino states is
projected by the non-abelian interaction into an incoherent mixture with equal weights.
In the following sections we shall further elaborate on the heuristic result of Eq. (17), by
providing a more general classification, and a proof based on the coherent state operator
formalism adopted for the cancellation theorems of Section 2.
4. Leading coherent state operators
The evaluation of BN violating terms — just pictured in the diagrammatic approach —

rests on the separation of initial and final state soft interactions, that can be performed
on the basis of the time evolution in a given reference frame, e.g., the c.m. frame of
the initial state. This method, first applied by Faddeev and Kulish [22] to QED, has
been extended to (unbroken) non-abelian theories in Refs. [23–25], by constructing the
large time (asymptotic) Hamiltonian as an infinite series of progressively subleading
contributions.
Here we limit ourselves to describe such analysis at the leading (double log) level. This
does not mean that subleading contributions are unimportant; it just means that the state
of the art in a broken theory is not sophisticated enough, at present, to allow confidence in
all-order subleading evaluations, even for the inclusive quantities that we investigate here.
In particular, collinear factorization theorems should be carefully revised, partly because
of the very presence of the uncanceled double logs that we are computing.
4.1. Quantum electrodynamics
According to Eq. (2), the process is separated into a hard scattering matrix SH , involving
the scale E λ (λ is the IR cutoff), and in a soft interaction, described by U I and U F ,
involving photon frequencies w such that E w λ. In describing the soft interaction,
the hard (initial and final) fermions are treated as external currents. Thus, the interaction
Hamiltonian describing the evolution in the soft Hilbert space Hs is simply:
X ZE

d[k] p̂i Aµ (k)e−i(p̂i k)t + h.c.
µ
Hs (t) = ei
i λ
Z

≡ dν h+ (ν)e−iνt + h− (ν)eiνt , (18)
X ZE
µ
h+ (ν) = ei d[k] p̂i Aµ (k)δ(ν − p̂i k), (19)
i λ
d 3k µ µ
where d[k] = 2wk (2π)3
Ei p̂i = pi (i = 1, . . . , nI ) denote the momenta of the hard
,
particles in the initial (or final) state, k µ = (wk , k) is the photon momentum, and
Aµ (k) (A†µ (k)) denote the photon annihilation (creation) operators. The “energy transfer”
variable ν = p̂i k is therefore conjugated to the time variable t in this approach, and
represents physically the quantity of energy transferred in an elementary vertex.
The key feature of Eq. (18) is the (universal) eikonal coupling of charged hard particles
to photons, proportional to their velocities p̂µ . At amplitude level, this implies the insertion
formulas with the eikonal current
X pµ
J µ (k) = ei i , (20)
kpi
i
whose non-abelian counterpart has been given in Eq. (8).
The in (out) coherent state operators occurring in Eq. (2) are obtained from the soft
Hamiltonian (18) by computing the time-ordered evolution operator before (after) the hard
scattering. In the QED case, the calculation is simplified by the fact that Hs is linear in the
Aµ , and has therefore c-number commutators at non-equal times. For instance we obtain,
by standard methods [22], the initial state operator
" ZE #
dν
UI = U (0, −∞) = e exp
iφC
h+ (ν) − h− (ν) ,
ν
λ
ZE
dν
φC = 2π h+ (ν), h− (ν) . (21)
ν
λ
Let us now consider for simplicity the case of an initial state of two particles 1 and 2
with opposite charges 2 and relative velocity v12 . By using the explicit form of the h’s we
obtain:
ZE ZE
µ 2
UI = U1 U2 = exp d[k] J12(k) Aµ (k) − Aµ (k) =: UI : exp d[k] J12 (k) ,
†
λ λ
µ µ
α µ p1 p2
φC = , J12 = e − . (22)
2v12 kp1 kp2
Since the Coulomb phase φC doesn’t give physical effects for the processes considered
here, we shall drop it from now on.
Note that the operator UI is factorized in the particle indices i = 1, 2 and is a functional
µ
of the combination Aµ (k) − A†µ (k) ≡ i Πµ (k) only. For this reason the currents Ji can be
freely added in the exponent, and unitarity is trivial.
A less trivial, but straightforward step is the normal ordering quoted in Eq. (22), which
emphasizes the Sudakov form factor. In the massless limit we obtain
" ZE #
2p1 p2 α E2
F12 ≡ h0|U (1)U (2)|0i = exp −e 2
d[k] ≈ exp − log2 2 .
(kp1)(kp2 ) 4π λ
λ
(23)
From the same normal ordering formula one can obtain the (Poissonian) properties of soft
photon radiation, the energy resolution form factors, and so on.
4.2. Non-abelian coherent states
The asymptotic Hamiltonian has been constructed [23–25] in the (unbroken) SU(N)
gauge theory also. The difficulty, in such case, is that soft bosons can be either primaries
(i.e., emitted directly by the fast incoming particles) or secondaries (i.e., emitted by the
primary ones which have higher energy). If strong ordering in energies is assumed, the soft
Hamiltonian has nevertheless a simple form, as follows:
ZE "
X
tia p̂i e−i(p̂i k)t
µ
Hs (t) = g d[k]
λ i
ZE ! #
0 0 0µ −i(k̂ 0 k)t
+ d[k ] ρ (k )k̂ e
a
Aµa (k) + h.c. , (24)
wk
ρ a
(k) = −A†µb (k)(T a )bc Aµ
c (k), (T a )bc = if bac .
In Eq. (24), the second term in round brackets represents the secondary emission by pri-
mary bosons with energies wk 0 wk , which occurs again through an eikonal vertex k̂ 0µ ≡
2 For general charge configurations, a gauge invariant picture requires considering initial and final state
operators all together.
k 0µ /wk 0 . Because of the assumption of energy ordering, we are limiting ourselves to the
leading coherent state operators, and we refer to [23–25] for the subleading contributions.
At this level of accuracy, we can compute the evolution operator by assuming strong
ordering in the energy transfers also (for details, see [23–25]). By indicating with Pω the
energy ordering operator (smaller energies act first), and by introducing the field Πµa (k) ≡
−i(Aµa (k) − A†µa (k)), we obtain
"ZE #
dν
Us (0, −∞) = Pν exp h+ (ν) − h− (ν) ≡ VsE UIl
ν
λ
" ZE ZE ! #
X µ
pi k 0µ
0 0
= Pω exp ig d[k] tia + d[k ] ρ (k ) 0 Πµa (k) ,
a
kpi kk
λ i w
" ZE ZE #
k 0µ
VsE ≡ Pω exp ig d[k] d[k 0 ] ρ a (k 0 ) 0 Πµa (k) . (25)
kk
λ wk
Here we face the problem of explicitating the recursive emission of secondaries. We

treat the full path-ordered exponential in Eq. (25) as a two-potential problem, one of
which produces the purely soft operator Vs , which carries no isospin/color indices, and
the other one is computed in the interaction representation of the first, and provides
the properly called coherent state operator UIl , carrying the isospin/color indices of the
incoming particles. The result of this procedure is
" ZE #
X p
µ
UIl ≡ Vs† Us (0, −∞) = Pω exp ig d[k] tia i Πµa
wk
(k) , (26)
kpi
λ I
" ZE µ #
X pf
UFl ≡ Us† (0, −∞)Vs = P̄ω exp −ig d[k] tfa wk
Πµa (k) , (27)
kpf
λ F
where P̄ω is the energy anti-ordering operator (bigger energies act first) and where the
w can be made more explicit as follows
“dressed” field Πµa

w
Πµa ≡ Vsw† Πµa Vsw = UAω ab (k̂)Πµb (k), (28)
where (UAω )ab (k̂) denotes the coherent state operator in the adjoint representation, for a
boson wk̂ µ , having the matrix properties (recall (Ta )bc = ifbac )
UA∗ = UA , UAT = UA† = UA−1 . (29)
Eq. (28) was proved in Refs. [23–25], by showing that the l.h.s. and the r.h.s. satisfy the
same evolution equation in the w variable.
Eqs. (26)–(29) show that non-abelian coherent states operators are defined in a quite
nonlinear manner. Eq. (26) exponentiates up to energy E the dressed fields Π ω , which in
turn are expanded in terms of a coherent state at a lower energy w < E by Eq. (28), which
in turn, etc. In other words non-abelian soft radiation is described by a “coherent state
of lower energy coherent states” whose structure is obtained recursively by a branching
process [23–25], by now incorporated in some QCD event generators [33,34].
For our purpose here, the important point is that we are now able to explicitate the soft
part of the process (1) as follows
S = Us† (0, +∞)SH Us (0, −∞) = UFl SH UIl . (30)
Here the operator Vs defined in Eq. (25), being purely soft, commutes with SH and drops
out by unitarity, and UFl , UIl are provided by Eqs. (26) and (28) at leading level.
The leading operator U l — despite its nonlinearity — is still a functional of the fields
Πµa (k) only, and involves therefore only commuting quantities in the Fock space. It
follows, therefore, that commutativity for any given color indices holds:
l
Uβα , Uβl 0 α 0 = 0 (31)
as already used in Section 2, and that factorization with respect to the hard particles holds
also
Uβα
l
(1, . . . , n) = Uβl 1 α1 (1) Uβl 2α2 (2) · · · Uβl n αn (n). (32)
Finally, one should note that the coherent state in the adjoint representation regulates the
evolution equation of t a matrices:

U E† (p)ta U E (p) = UAE ab (p)tb , (33)
for an arbitrary isospin/color representation of the particle p. In fact, the coherent state
operators satisfy the Schrödinger-like equation
Z
∂ E pµ E
U (p) = tb 1E b U E
(p), 1 E
b = ig d[k] δ(wk − E) Πµb (k). (34)
∂E kp
Therefore, by combining the U and U † equations, the l.h.s. of Eq. (33) satisfies the
equation
∂
Tr U †E ta U E td = ifabc 1E †E E
b Tr U tc U td , (35)
∂E
which is the same as the one satisfied by the r.h.s., by direct use of (34) in the adjoint
representation.
4.3. Form factor exponentiation
The Sudakov (singlet) form factor can be defined similarly to Eq. (23) (with opposite
initial charges), for any given color representation of particles 1 and 2:

h0| U E† (2)U E (1) αβ |0i = δαβ F12 (E, λ). (36)
However, due to the nonlinearity of Eqs. (26) and (28) the normal ordering is no longer
straightforward, and was worked out by an evolution equation method in Refs. [23–25].
Since the energy ordered exponential satisfies the Schrödinger-like equation (34), one
first computes the energy derivative of (36), using also definition (28), as follows (wK = E)

∂ g dΩK µ E
Tr(1) F12 = h0|Tr U E† (2) 3
J12 ta UA ab (K̂)
E∂E 2(2π)

× Aµb (K) − A†µb (K) U E (1) |0i, (37)
µ µ µ
where J12 = p1 /Kp1 − p2 /Kp2 . Then, one commutes Aµb (K) to the right and A† to
w †
the left. Since [Aµb (K), Aνak (k)] = 0 for wk < E this procedure singles out the upper
E
frequency in U and yields
E pµ
Aµb (K), U E (1) = g UA cb 1 tc U E (1), (38)
Kp1
µ
with a similar relation involving A† and p2 . Finally, by using the unitarity relation
X
UAE ab (i) UAE cb (i) = δac , (i = 1, 2) (39)
b
and t 2i = CF (or CA , depending on the particle’s representation) we obtain

Z !
∂ g 2 CF (A) 2
F12 (E, λ) = dΩK J12 (K) F12 (E, λ), (40)
E∂E 2(2π)3
wK =E
which is the evolution equation for the form factor we were looking for. In the double log
approximation, (40) yields
2 2
g CF (A) 2E
F12 (E, λ) = exp − log (41)
16π 2 λ2
as expected. Note that, in the SU(2) isospin case, CA = 2 and the exponent in Eq. (41)
becomes just 2LW (s).
The essential point of this derivation rests on the unitarity relation (39), which means
the cancellation of all correlation effects due to the non-abelian structure. This leads to
Eq. (40), which could be naively derived by the “external line insertion rule” for virtual
corrections.
5. Bloch–Nordsieck violation to all orders
5.1. Isospin structure of the hard overlap matrix
In this section we discuss the general structure of the hard overlap matrix, describing the
†
squared matrix element: OH ≡ SH SH . We consider the case of a generic process with two
partons in the initial state and we work in a limit in which all kinematical invariants are
much bigger than gauge bosons masses: |s|, |t|, |u| M 2 . Then, the full gauge symmetry
SU(3) ⊗ SU(2) ⊗ U (1) is restored, and the structure of the overlap matrix in isospin space
is fixed by the SU(2) symmetry.
Left particles carry non-abelian SU (2) charges while right particles don’t, so we need
to consider three cases:
Fig. 2. Soft dressing of the hard S-matrix SH is described by the coherent state operator U I (a). At
leading order, the latter is factorized into leg operators U (i) (b).
• When both initial particles are right-handed, and therefore do not carry any non-
abelian weak charge nor any isospin index, the overlap matrix is simply a number O H
in isospin space (still depending of course on the quantum numbers of the involved
particles).
• Next possibility is that one particle is left polarized and the other one is right polarized.
In this case the hard overlap matrix carries two (left) isospin indices Oβα H .
• The case of two left initial fermions, is of course the most complicated one; the hard
overlap matrix carries in this case four isospin indices OβH1 β2 ,α1 α2 (see Fig. 2).
While for cross sections we always have αi = βi , we leave open the possibility that αi 6= βi ,
and see O H as an operator in isospin space. The form of the overlap matrix is severely
restricted by the requirement of SU(2) symmetry. In the three cases discussed above, we
have:
RR : O H = A0 , RL, LR : H
Oβ,α = B0 δβα ,
LL : OβH1 β2 ,α1 α2 = C0 δβ1 α1 δβ2 α2 + C1 4tβa2 α2 tβa1 α1 . (42)
The last expression, where the indices α1 , β1 (and α2 , β2 ) are grouped together, corre-
sponds to a t-channel decomposition in singlet and vector components (see Fig. 3). The
case where the initial particle on leg 2 is an antiparticle, thus belonging to the conjugate
representation t ∗ = t T , is correspondingly:
ŌβH1 β2 ,α1 α2 = C̄0 δβ1 α1 δβ2 α2 + C̄1 4tαa2 β2 tβa1 α1 (particle–antiparticle). (43)
Let us consider as an example a generic hard cross section involving a left e− and a left e+
(which we indicate with e and ē), and ν, ν̄. 3 We have:
σeHē = σνHν̄ ∝ Ō11,11

H
= Ō22,22
H
= C̄0 + C̄1 , (44a)
σeHν̄ = σνHē ∝ Ō12,12
H
= Ō21,21
H
= C̄0 − C̄1 . (44b)
3 In our notation a particle and its antiparticle share the same isospin index: 1 for ν, ν̄ and 2 for e, ē.
Fig. 3. Singlet and vector decomposition of the overlap matrix OH in the t -channel. The adjoint
coherent state dressing is depicted also.
5.2. Resummed energy dependence from coherent states
We now proceed to “dress” the hard matrix element with soft interactions. In the case of
right initial particles, since weak interactions become purely abelian in this case, it should
be clear from the above discussion that no BN violating effect is present. If one particle is
L and the other one is R, the dressing by soft interactions is described by (5):
dress †
H
Oαβ → Oαβ =S h0|Uαα 0 Oα 0 β 0 Uβ 0 β |0iS .
H
(45)
But then, since by SU(2) symmetry Oαβ H = B δ , and because of the unitarity
0 αβ
property (3), also in this case no BN violating effect is present and the dressed overlap
matrix is equal to the hard one. In the remaining of this section we discuss the interesting
case of two left initial fermions. As we have seen, the dressing in this case is described by
a coherent state operator U I such that (see Fig. 2):
dress
OβH1 β2 ,α1 α2 → Oβ1 β2 ,α1 α2 =S h0|UβI †β 0 0 (OH )β10 β20 ,α10 α20 UαI 0 α 0 ,α1 ,α2 |0iS , (46)
1 2 ,β1 ,β2 1 2
where |0iS is the soft vacuum. At the leading log level, U I † is factorized (Section 4.2) into
two leg operators:
UαI 0 α 0 ,α = Uα(1) (2)†

0 α Uα α 0 (particle–antiparticle),
1 2 1 ,α2 1 1 2 2
UαI 0 α 0 ,α = Uα(1) (2)

0 α Uα 0 α (particle–particle), (47)
1 2 1 ,α2 1 1 2 2
where we take into account that antiparticles live in the conjugate representation, so that in
the particle–antiparticle case we have U (2) → U (2)∗ . Putting together (43), (46), (47) we
obtain, for the particle–antiparticle case, the dressed overlap operator:
Ōβ1 β2 ,α1 α2 ≡ C̄0 (s)δβ1 α1 δβ2 α2 + C̄1 (s)4tβa1 α1 tαa2 β2


= C̄0 δα1 β1 δα2 β2 + 4C̄1S h0| U (1)† t a U (1) β1 α1
U (2)† t a U (2) α2 β2
|0iS .
(48)
By using twice Eq. (33) that relates the coherent states in the fundamental representation
with the one in the adjoint representation, we obtain:

S h0| U t U β α U (2)†t a U (2) α β |0iS = S h0| UA(2)† UA(1) ab |0iS tβa1 α1 tαb2 β2
(1)† a (1)
1 1 2 2

= FA s, M 2 tβa1 α1 tαa2 β2 , (49)
where the definition (36) of the form factor has been used, so that we obtain:
C̄0 (s) = C̄0 , C̄1 (s) = C̄1 e−CA LW (s) = C̄1 e−2LW (s). (50)
From these expressions we obtain the final results for the dressed cross sections (see Fig. 4):
σ11 = σ22 = C̄0 (s) + C̄1 (s) = C̄0 + C̄1 e−2LW (s)
(σ11 + σ12 )H (σ11 − σ12 )H −2LW (s)
= + e , (51a)
2 2
σ12 = σ21 = C̄0 (s) − C̄1 (s) = C̄0 − C̄1 e−2LW (s)
(σ11 + σ12 )H (σ11 − σ12 )H −2LW (s)
= − e , (51b)
2 2
which reproduce Eq. (17), and the relative effects in double log approximation:
H H
1σ σ11 − σ11
H σ11 − σ12 1 − e−2LW (s)
≡ H
= H
, (52a)
σ 11 σ11 σ11 2
H H
1σ σ12 − σ12
H σ12 − σ11 1 − e−2LW (s)
≡ H
= H
. (52b)
σ 12 σ12 σ12 2
An analogous treatment for the particle–particle case allows one to conclude that Eqs. (51),
(52) hold also for this case with the obvious replacement C̄i → Ci . These final results are
therefore completely general: σ11 stands for any cross section with two incoming particles
(or one particle and one antiparticle) with isospin index 1, which means for instance
Fig. 4. σ12 and σ11 as a function of energy. The vertical scale is arbitrary.
σν ν̄ , σuu , σuū and so on. While in all these cases the expressions for the coefficients C0 and
C1 (or C¯0 and C¯1 ) that describe the hard cross section are of course different in general, the
expression for the dressed cross sections is always the same and is described by Eqs. (51).
Notice the appearance, as anticipated, of the adjoint Casimir CA = 2 in (51). This means
that the energy dependence of the effect we are discussing is universal, i.e., the same for
any fermion doublet in the initial state. Note however that the relative effect does depend
on the structure of the hard cross sections, as one can see from (52). We will discuss several
cases of phenomenological interest more in detail in Section 6.
6. Applications to simple processes
We consider inclusive observables associated to a large angle hard scattering process

(|s| ∼ |t| ∼ |u| M 2 ) involving massless fermions and antifermions in the initial and
final states. As we have seen, a BN violating effect is present only if there are two non-
abelian charges in the initial state. This means that big noncancellations are present only
with two left fermions in the initial state, while the effect is absent in the RR and RL
cases. In turn, this implies a strong dependence on the physical polarization of the initial
beams: a maximal effect if the beams are both polarized L, no effect in all other cases, and
somewhere in between for unpolarized beams. This is particularly important since one of
NLCs features is the possibility of having highly polarized beams [26,27].
We discuss in this section some cases that we think are/will be phenomenologically
relevant for NLCs (Section 6.1), and for electron–hadron (Section 6.2) and hadron–hadron
(Section 6.3) colliders. Given the master formulas (51), and since the energy dependence
is universal as already noticed, it is clear that only the (tree level) hard cross sections need
to be discussed. We expect large effects when LL contributions dominate the hard cross
section, and/or when there is a big difference between σ12 and σ11 (see (52)). The latter
is the case, for instance, in pure q q̄ s-channel annihilation where σ11 is of order αS2 while
σ12 is flavor changing and is thus electroweak (Section 6.3). Similarly, in e+ e− → hadrons
(Section 6.1), the effect is pretty large because this process is dominated by L components.
6.1. l l¯ → q q̄ (s-channel) annihilation
This kind of process is typically relevant for NLC. The Born (hard) amplitude is simply
described by an s-channel annihilation involving only weak interaction and has the form
(a = 1, 2, 3):
Mβ1 β2 ,α1 α2 ∼ g 0 2 yY δβ1β2 δα1 α2 + g 2 tβa1 β2 tαa1 α2 , (53)
where y (Y ) denote the initial lepton (final quark) hypercharges. Expression (53) is the
s-channel analogue of the singlet-vector t-channel decomposition (42). By squaring and
summing over final states, we obtain the overlap matrix t-channel components, as defined
in (42), for the case of (initial) L fermions:
1 1 X 3 1 X
C1L = − g 4 + g 0 4 yL2 YL2 , C0L = g 4 + g 0 4 yL2 YL2 , (54a)
16 2 16 2
1 X 1 X
C0R = g 0 4 yL2 YR2 , C1R = g 0 4 yL2 YR2 , (54b)
2 2
where CiL(R) refers to final L (R) quarks. For the case of initial R leptons we obtain:
X X
04 2 04 2
AR0 = δij g yR YR2 , 0 = δij g yR
AL YL2 , (54c)
where the factor δij in (54c) takes into account that the contribution ∝ g 0 4 is present
only for the case of an initial particle and its own antiparticle. The sums are over final
P P
states hypercharges, namely YR2 = 4/9 + 1/9 and YL2 = 1/36 + 1/36 for quarks. The
unpolarized cross section is then found by putting in the usual phase space factor, and by
using the resummed energy evolution in Eq. (50). We find:
dσij Nc Nf h R
= A0 + C0L ± C1L e−2LW (s) (1 + cos θ)2
d cos θ 128πs i
R −2LW (s)

+ AL0 + C0
R
± C 1 e (1 − cos θ)2
, (55)
where Nf is the number of quark families, Nc is the number of colors and the + (−) sign
refers to σ11 (σ12 ).
The terms proportional to g 0 4 in Eq. (55) are pretty small, being suppressed by a
g 04 /g 4 = tW
4 factor. In the limit g 0 → 0 only L components contribute to the cross section:
2
dσij dσijLL 2
πNc Nf αW 1 + cos θ
≈ = 3 ∓ e−2LW (s) . (56)
d cos θ d cos θ 32s 2
Note that, for instance, dσνHē > dσeHē . Therefore, the BN violating corrections increase the
physically relevant eē cross section, that reaches in the asymptotic limit of very high energy
the isospin average. From (56) we also see that in the g 0 → 0 limit, angular and energy
dependences are factorized. Therefore in this limit the forward–backward asymmetry for
eē is equal to the tree level value of 3/4. However, the g 0 4 terms are not completely
negligible, producing a relative correction to AFB of about 1.8% at 1 TeV, the dependence
on energy being given by the by now familiar universal behavior.
From (55) and taking into account the g 0 4 terms also, we obtain the relative effect for
eL ēL → hadrons:
H
1σeē L σν ē − σeHē L 1 − e−2LW (s)
= ≈ 0.8LW (s). (57)
σeHē σeHē 2
The effect for the unpolarized cross section is slightly reduced (we give also the first order
QCD corrections):

1σ EW αW 2 s 1σ QCD αS
' 0.58LW (s) = 0.58 log , ' . (58)
σ eē 4π M2 σ eē π
That is, radiative corrections to e+ e− → hadrons of weak origin are bigger, at the TeV
scale, than strong QCD corrections (see Fig. 5).
Fig. 5. Resummed double log EW corrections to e+ e− → hadrons and strong corrections (dashed
line) up to 3 loops. The dash-dotted line is for a LL polarized beam, while the continuous line is for
an unpolarized beam. A numerical value M = MW = 80 GeV has been used.
6.2. t-channel scattering
Here we consider processes in which the Born term is a t-channel scattering diagram
involving only weak interactions. This involves processes like lq and l q̄ scattering where
l is a lepton and q a quark. We also discuss briefly the case for νµ e scattering. The hard
cross section is described in this case by the amplitude
Mβ1 β2 ,α1 α2 ∼ g 0 2 yY δβ1α1 δβ2 α2 + g 2 tβa1 α1 tβa2 α2 , (59)
where y, α1 (Y, α2 ) refer to lepton (quark) indices. By squaring and by summing over final
states we obtain the projections of the overlap matrix for this case:

1 2 02 21 3
C1 = g g yL YL ∓ g
L
, C0L = g 4 + g 0 4 yL2 YL2 , (60a)
2 4 16
B0R = g 0 4 yR2 YL2 , B0L = g 0 4 yL2 YR2 , (60b)
AR
0 = g 0 4 yR2 YR2 . (60c)
This time, differently from Eqs. (54), (55), the convention is that the upper index is for the
initial lepton chirality. So B0R indicates an initial state with a R lepton and a L quark, and
¯ case.
so on. The upper (lower) sign refers to the lq (lq)
The energy and angular dependence of the resummed cross section can be found as
before:
dσij Nc s 2 h R L −2LW (s)
i
= A 0 + C L
0 ± C 1 e (1 + cos θ)2
+ 4 B0
R
+ 4 B L
0 . (61)
d cos θ 128 πs t 2
We can neglect the terms ∝ g 0 4 in (60), keeping the terms ∝ g 2 g 0 2 that are suppressed only
by a factor g 0 2 /g 2 = tW
2 and therefore not negligible in general, and thus obtaining:
2 Nc π (1 + cos θ)
dσij 2
= αW
d cos θ 2 s (1 − cos θ)2

3 1 1
× + (−1)i+j ∓ + yL YL tW
2
e−2LW (s) , (62)
16 8 2
¯ case.
where the − sign refers to the lq and the + sign to the lq
Although the BN violating corrections are regulated by the universal Eqs. (51), it should
H and σ H .
be clear from (52) that their relative effect is dependent on the magnitude of σ12 11
In particular, in the case of left polarized beams one has:
1σēu 1σēd
H
' 2.98 LW (s), H
' −0.75 LW (s),
σēu σēd
1σeu 1σed
H
' −0.84 LW (s), H
' 5.4 LW (s), (63a)
σeu σed
1σeν̄µ 1σνµ ē
H
= H ' 10.6 LW (s). (63b)
σeν̄µ σνµ ē
The magnitude of the effect in the electron–muon antineutrino scattering in (63b) is

given by the ∝ g 0 2 g 2 terms in (62). In fact, the cross sections for ēu and ēνµ are equal in
the g 0 → 0 limit. However, the different hypercharges for νµ , u and the signs in formulas
(60), (61) conspire to produce a particularly small hard cross section in the ēνµ case.
Considering unpolarized beams, it is interesting to discuss the dependence of the effect
on the scattering angle. As we can see from Fig. 6, there is no effect for cos θ = −1, the
reason being that the LL component does not contribute to the cross section in this limit.
The effect then increases greatly with the angle. A cutoff on the maximum value of cos θ
is necessary, since we always work in the limit 1 − | cos θ| M 2 /s.
dσ (ν ē) and O = dσ (uē) at √s = 1 TeV and √s = 5 TeV. A

Fig. 6. Relative effects for O = d cos θ µ d cos θ
numerical value M = MW = 80 GeV has been used.
The physical process proceeds via a proton; then, since the effect has opposite signs for
u and d quarks in (63b), the overall effect gets diminished. For instance, taking the simplest
picture of a hadron constituted only by valence quarks, we obtain for polarized left handed
beams:
1σep 21σeu + 1σed 1σēp
≈ ' −0.39LW (s), ≈ 0.5LW (s). (64)
H
σep 2σeu + σed H
σēp
This means that for hadron beams the relative effect is about one-half that for lepton beams
in Eq. (57).
6.3. q q̄ scattering
In the q q̄ case the overlap matrix OH contains, besides s-channel and t-channel
contributions, also interference terms in the identical quarks case, and is characterized by
the additional presence of QCD contributions, regulated by the strong coupling constant
αS (s), where s M 2 denotes the hard process scale.
In order to have a preliminary understanding, we neglect EW contributions to OH ,
and we note that EW BN violating corrections are absent of course in the gluon–gluon
scattering cases, but also in the gluon–quark case (see (45) and discussion thereafter). In the
example of pure qL q̄L s-channel annihilation on the other hand, we expect the possibility
of having big relative effects at the parton level, because of the hierarchy between σ12 and
σ11 . For instance, σuHd̄ vanishes in the limit considered here, while σuHū is of order αS2 , so
that we obtain (in the LL case):
1
σuL ūL = σdL d̄L = σuHL ūL 1 + e−2LW (s) , (65)
2
1
σuL d̄L = σdL ūL = σuHL ūL 1 − e−2LW (s) (66)
2
so that the corrections to, say, the cross section for uL ūL is of the order of the cross
section itself. However, one should be cautious here because the observable effect with
hadron beams is not really large. Considering for instance the case of valence quarks in pp̄
collisions we first take into account that BN violating effects involve only left particles, so
that the unpolarized parton level cross sections are:
−2LW (s)
H 1+e 1
σuū = σd d̄ = σuL ūL + σuR ūR = σuū + , (67a)
4 2

1 − e−2LW (s)
σud̄ = σd ū = σuL d̄L = σuHū . (67b)
4
Next, a further suppression of the effect comes about because of a partial cancellation
between uū and ud̄ channels. In fact, since scattering occurs with probability 5/9 in a uū
or d d̄ configuration and 4/9 in a ud̄ or d ū configuration, we obtain:

5 4 e−2LW (s) − 1 1σpp̄ 1
σpp̄ ≈ σuū + σud̄ = σpHp̄ 1 + , H
≈ − LW (s) (68)
9 9 20 σpp̄ 10
and about twice as much in the pure LL channel. Therefore, the relative effect is reduced
to the one percent range at the TeV threshold.
7. Outlook
The outcome of this paper is that electroweak corrections become pretty large at
the TeV scale, even for inclusive processes, because of uncancelled double logarithmic
enhancements involving the effective coupling LW (s). This leads to a sort of early
unification within the Standard Model itself, because strong and EW corrections are to
be considered together much before the respective running couplings become comparable.
From a theoretical point of view, the effect above is due to both the non-abelian nature of
the SU(2) component of the standard model and to symmetry breaking itself, which allows
the initial states to be prepared as abelian charges. It follows that all initial states carrying
nontrivial weak isospin are affected by the uncancelled double logs. For initial fermions
we find the following features:
(i) Only left-handed doublets are affected. This means that the BN violating effect is
strongly polarization dependent, and may lead to nontrivial angular dependence in
some cases (Fig. 6).
(ii) The effect is particularly important for purely leptonic beams, for which its size is
directly provided by LW (s), which is about 7% at the TeV threshold. For instance,
+ −
in the case of eL eL → hadrons, despite some reduction by exponentiation and
running coupling effects, EW corrections are already 5.2%, compared to 3% strong
corrections (Fig. 5).
(iii) There is a composite state reduction of the effect for hadron beams, which act
in the hard process as mixtures of partonic isospin states (recall that the effect
vanishes for a mixture with equal weights). The suppression is of about 50% in the
¯ scattering processes considered in Section 6.2. It is even stronger for hadron–
lp, lp
hadron beams, due to both initial isospin averaging and to QCD being flavor blind.
Nevertheless, sizeable effects in the percent range are still expected in the pp̄ case
(Section 6.3). The more relevant pp case requires a detailed analysis, in which both
structure functions and boson initiated processes will presumably play a role.
Undoubtedly, quantitative estimates of the effects presented here are needed for the
planning of future accelerators. This implies not only a more detailed analysis of Born cross
sections, but also an extension to subleading corrections which is far from being trivial [5,
6]. What we learn here is that even at inclusive level we need to revise the factorization
theorems we are used to in QCD, in order to cope with fixed initial flavor and to disentangle
the enhanced SM corrections from new physics.
References
[1] P. Ciafaloni, D. Comelli, Phys. Lett. B 446 (1999) 278.

[2] M. Beccaria, P. Ciafaloni, D. Comelli, F.M. Renard, C. Verzegnassi, Phys. Rev. D 61 (2000)
011301.
[3] M. Beccaria, P. Ciafaloni, D. Comelli, F.M. Renard, C. Verzegnassi, Phys. Rev. D 61 (2000)
073005.
[4] J.H. Kuhn, A.A. Penin, hep-ph/9906545.
[5] P. Ciafaloni, D. Comelli, Phys. Lett. B 476 (2000) 49.
[6] V.S. Fadin, L.N. Lipatov, A.D. Martin, M. Melles, Phys. Rev. D 61 (2000) 094002.
[7] M. Ciafaloni, P. Ciafaloni, D. Comelli, hep-ph/0001142, Phys. Rev. Lett. 84 (2000) 4810.
[8] F. Bloch, A. Nordsieck, Phys. Rev. 52 (1937) 54.
[9] V.V. Sudakov, Sov. Phys. JETP 3 (1956) 65.
[10] D.R. Yennie, S.C. Frautschi, H. Suura, Ann. Phys. 13 (1961) 379.
[11] R. Doria, J. Frenkel, J.C. Taylor, Nucl. Phys. B 168 (1980) 93.
[12] G.T. Bodwin, S.J. Brodsky, G.P. Lepage, Phys. Rev. Lett. 47 (1981) 1799.
[13] A.H. Mueller, Phys. Lett. B 108 (1981) 355.
[14] W.W. Lindsay, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 214 (1983) 61.
[15] P.H. Sorensen, J.C. Taylor, Nucl. Phys. B 238 (1984) 284.
[16] S. Catani, M. Ciafaloni, G. Marchesini, Phys. Lett. B 168 (1986) 284.
[17] J.C. Collins, D.E. Soper, G. Sterman, Nucl. Phys. B 261 (1985) 104.
[18] G.T. Bodwin, Phys. Rev. D 31 (1985) 2616.
[19] G.T. Bodwin, Phys. Rev. D 34 (1986) 3932, Erratum.
[20] D. Amati, G. Veneziano, Phys. Lett. B 83 (1979) 87.
[21] A. Bassetto, M. Ciafaloni, G. Marchesini, Nucl. Phys. B 163 (1980) 477.
[22] P.P. Kulish, L.D. Faddeev, Teor. Mat. Fiz. 4 (1970) 153; Theor. Math. Phys. 4 (1970) 745.
[23] M. Ciafaloni, Phys. Lett. B 150 (1985) 379.
[24] S. Catani, M. Ciafaloni, G. Marchesini, Nucl. Phys. B 264 (1986) 588.
[25] M. Ciafaloni, in: A.H. Mueller (Ed.), Perturbative Quantum Chromodynamics, 1989, p. 491.
[26] E. Accomando et al., Phys. Rep. 299 (1998) 1.
[27] Y. Sugimoto, KEK-Proceedings-97-2.
[28] F. Low, Phys. Rev. 110 (1958) 974.
[29] S. Weinberg, Phys. Rev. B 140 (1965) 516.
[30] V.N. Gribov, Sov. J. Nucl. Phys. 5 (1967) 399.
[31] T. Kinoshita, J. Math. Phys. 3 (1962) 650.
[32] T.D. Lee, M. Nauenberg, Phys. Rev. 133 (1964) 1549.
[33] G. Marchesini, B.R. Webber, Nucl. Phys. B 238 (1984) 1.
[34] G. Marchesini, B.R. Webber, Nucl. Phys. B 330 (1990) 261.
Higher twist distribution amplitudes of the nucleon

in QCD
V. Braun, R.J. Fries, N. Mahnke, E. Stein ∗
Institut für Theoretische Physik, Universität Regensburg, D-93040 Regensburg, Germany
Abstract
We present the first systematic study of higher-twist light-cone distribution amplitudes of the
nucleon in QCD. We find that the valence three-quark state is described at small transverse
separations by eight independent distribution amplitudes. One of them is leading twist-3, three
distributions are twist-4 and twist-5, respectively, and one is twist-6. A complete set of distribution
amplitudes is constructed, which satisfies equations of motion and constraints that follow from
conformal expansion. Nonperturbative input parameters are estimated from QCD sum rules.  2000
Keywords: QCD; Nucleon; Power corrections; Distribution amplitudes
1. Introduction
Hard exclusive processes are coming to the forefront of high energy nuclear and particle
physics. This is already visible at TJNAF, COMPASS, HERMES that go for more and
more exclusive channels, and it makes the core of the ELFE proposal. One main reason for
that is the growing understanding that exclusive reactions provide unmatched opportunities
to study the hadron structure, as demonstrated by recent interest to deeply virtual Compton
scattering (DVCS) and hard diffractive meson production. All future plans also call for very
high luminosity and would therefore be perfectly suited for the investigation of exclusive
and semi-exclusive reactions with and without polarization.
The classical theoretical framework for the calculation of hard exclusive processes
in QCD was developed in [1–14], see [15–17] for a review. This approach introduces
a concept of hadron distribution amplitudes as fundamental nonperturbative functions
E-mail addresses: vladimir.braun@physik.uni-regensburg.de (V. Braun), rainer.fries@physik.uni-
regensburg.de (R.J. Fries), nils.mahnke@physik.uni-regensburg.de (N. Mahnke), eckart.stein@physik.uni-
regensburg.de (E. Stein).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 1 6 - 2
382 V. Braun et al. / Nuclear Physics B 589 (2000) 381–409
describing the hadron structure in rare parton configurations with a minimum number of
Fock constituents (and at small transverse separations). The distribution amplitudes are
equally important and to a large extent complementary to conventional parton distributions
which correspond to one-particle probability distributions for the parton momentum
fraction in an average configuration. The recent data from CLEO [18] and E791 [19]
provided, for the first time, quantitative information on the pion distribution amplitude.
For a theorist, the main challenge is to make the QCD description of hard exclusive reac-
tions fully quantitative. Although the leading-twist QCD factorization approach correctly
reproduces the power dependence of exclusive amplitudes on the large momentum, there
exist many indications that soft end-point and higher-twist corrections might dominate over
the naive QCD prediction over the range of accessible values of Q2 . For mesons, the study
of preasymptotic corrections to exclusive reactions has been pursued intensively and in
different directions. In particular, meson distribution amplitudes of higher twist have been
studied in detail in [15,20–23]. For baryons, the corresponding task is more complicated
and received less attention in the past. Soft end-point contributions to several hard exclusive
reactions involving nucleons have been estimated using various models and were found to
be large [24–29]. To the best of our knowledge, hard higher-twist corrections and nucleon
mass corrections have never been addressed in the literature. In this paper we present the
first systematic study of nucleon distribution amplitudes of higher twist.
The notion of higher twist contributions to the distribution amplitudes comprises a broad
spectrum of effects of different physical origin: first, contributions of “bad” components
in the wave function and in particular of components with “wrong” spin projection;
second, contributions of transverse motion of quarks (antiquarks) in the leading twist
components; third, contributions of higher Fock states with additional gluons and/or
quark–antiquark pairs. Finally, one can speak of hadron mass effects, similar to target
mass corrections in deep inelastic scattering. The relative significance of these corrections
depends on the hadron state in question and the particular hard process. There exists an
important conceptual difference between mesons and baryons. For mesons, all effects due
to “bad” components in the quark–antiquark wave functions can be rewritten in terms
of higher Fock state components thanks to the QCD equations of motion. Since quark–
antiquark–gluon matrix elements between the vacuum and the meson state are numerically
small (see [20–23] for the existing estimates), contributions of “bad” components in
two-particle wave functions are small as well, and to a large extent dominated by
meson mass corrections [20–23]. For baryons, on the contrary, QCD equations of
motion are not sufficient to eliminate the higher-twist three-quark states in favor of the
components with extra gluons, so that the former present genuine new degrees of freedom.
Moreover, vacuum-to-baryon matrix elements of higher-twist three-quark operators are
large, comparable to the ones of leading twist. One may expect, therefore, that higher-twist
effects in baryon wave functions are dominated by “bad” components of three-quark states
rather than extra gluons. A systematic study of such states as well as taking into account
nucleon mass corrections presents the subject of this work.
We find that a generic three-quark matrix element on the light-cone can be parametrized
in terms of eight independent distribution amplitudes. In particular, there exists one
V. Braun et al. / Nuclear Physics B 589 (2000) 381–409 383
amplitude of leading twist-three that is familiar from earlier studies [9–14,30–33], three
distribution amplitudes of twist-four and twist-five, respectively, and one distribution
amplitude of twist-six. Following the approach of [20–23], we further expand all
distribution amplitudes in contributions of operators with increasing conformal spin.
The conformal expansion corresponds, physically, to the separation of longitudinal and
transverse degrees of freedom, and the coefficients of this expansion present the relevant
nonperturbative parameters. We find that the higher-twist distribution amplitudes with the
lowest conformal spin (asymptotic wave functions) involve two nonperturbative matrix
elements that are well known from the QCD sum rule studies of the nucleon [34–36]. The
leading corrections to the asymptotic wave functions involve three more parameters (for
all twists) that we estimate in this work.
The outline of the paper is as follows. The general classification of twist-three
distribution amplitudes is worked out in Section 2. One starts with 24 invariant functions
in a general decomposition of a three-quark light-cone operator. By isospin symmetry
the number of independent amplitudes is reduced to eight amplitudes that determine the
three-quark proton state completely. A convenient representation of these independent
amplitudes is derived in terms of chiral fields.
Section 3 is devoted to the conformal expansion. We identify the nonperturbative
parameters that describe the higher-twist distribution amplitudes up to next-to-leading
order in the conformal expansion and work out the relations between them, as given by
isospin constraints and equations of motion.
In Section 4 we present particular models for the higher-twist distributions based on
QCD sum rules and summarize.
Our work encloses a number of appendices. Appendix A gives the Fierz transformation
rules used to derive the isospin relations. Appendix B summarizes the determination
of nonperturbative parameters in the QCD sum rule approach. Appendix C contains a
“handbook” of nucleon distribution amplitudes, where we collect all expressions needed
in practical calculations.
2. General classification
2.1. Lorentz structure
The notion of hadron distribution amplitudes in general refers to hadron-to-vacuum

matrix elements of nonlocal operators built of quark and gluon fields at light-like
separations. In this paper we will deal with the three-quark matrix element
0 j0 0
h0|εij k uiα (a1 z)[a1z, a0 z]i 0 ,i uβ (a2 z)[a2z, a0 z]j 0 ,j dγk (a3 z)[a3 z, a0 z]k 0 ,k |P (P , λ)i,
(2.1)
where |P (P , λ)i denotes the proton state with momentum P , P 2 = M 2 and helicity λ.
u, d are the quark-field operators. The Greek letters α, β, γ stand for Dirac indices, the
Latin letters i, j , k refer to color. z is an arbitrary light-like vector, z2 = 0, the ai are real
numbers. The gauge-factors [x, y] are defined as
" Z1 #

[x, y] = P exp ig dt (x − y)µ A tx + (1 − t)y ,
µ
(2.2)
0
and render the matrix element in (2.1) gauge-invariant. To simplify the notation we will not
write the gauge-factors explicitly in what follows but imply that they are always present.
Taking into account Lorentz covariance, spin and parity of the nucleon, the most general
decomposition of the matrix element in Eq. (2.1) involves 24 invariant functions:
j
4h0|εij k uiα (a1 z)uβ (a2 z)dγk (a3 z)|P i
= S1 MCαβ (γ5 N)γ + S2 M 2 Cαβ (/zγ5 N)γ + P1 M(γ5 C)αβ Nγ + P2 M 2 (γ5 C)αβ (/zN)γ

+ V1 (P
/ C)αβ (γ5 N)γ + V2 M(P/ C)αβ (/zγ5 N)γ + V3 M(γµ C)αβ γ µ γ5 N γ

+ V4 M 2 (/zC)αβ (γ5 N)γ + V5 M 2 (γµ C)αβ iσ µν zν γ5 N γ + V6 M 3 (/zC)αβ (/zγ5 N)γ

+ A1 (P
/ γ5 C)αβ Nγ + A2 M(P/ γ5 C)αβ (/zN)γ + A3 M(γµ γ5 C)αβ γ µ N γ

+ A4 M 2 (/zγ5 C)αβ Nγ + A5 M 2 (γµ γ5 C)αβ iσ µν zν N γ + A6 M 3 (/zγ5 C)αβ (/zN)γ

+ T1 P ν iσµν C αβ γ µ γ5 N γ + T2 M zµ P ν iσµν C αβ (γ5 N)γ

+ T3 M(σµν C)αβ σ µν γ5 N γ + T4 M P ν σµν C αβ σ µ% z% γ5 N γ

+ T5 M 2 zν iσµν C αβ γ µ γ5 N γ + T6 M 2 zµ P ν iσµν C αβ (/zγ5 N)γ

+ T7 M 2 (σµν C)αβ σ µν /zγ5 N γ + T8 M 3 zν σµν C αβ σ µ% z% γ5 N γ , (2.3)
where Nγ is the nucleon spinor, C the charge conjugation matrix and σµν = (i/2)[γµ , γν ].
The factor 4 on the l.h.s. is introduced for later convenience. Each of the 24 functions Si ,
Pi , Ai , Vi , Ti depends on the scalar product P · z.
The invariant functions in Eq. (2.3) do not have a definite twist yet. For the twist
classification, it is convenient to go over to the infinite momentum frame. To this end we
introduce the second light-like vector
1 M2
pµ = Pµ − zµ , p2 = 0, (2.4)
2 p·z
so that P → p if the nucleon mass can be neglected M → 0. Assume for a moment that
the nucleon moves in the positive ez direction, then p+ and z− are the only nonvanishing
components of p and z, respectively. The infinite momentum frame can be visualized as
the limit p+ ∼ Q → ∞ with fixed P · z = p · z ∼ 1, where Q is the large scale in the
process. Expanding the matrix element in powers of 1/p+ introduces the power counting
in Q. In this language twist counts the suppression in powers of p+ . Similarly, the nucleon
spinor Nγ (P , λ) has to be decomposed in “large” and “small” components as
1
Nγ (P , λ) = / )Nγ (P , λ) = Nγ+ (P , λ) + Nγ− (P , λ),
p/z + /zp
(/ (2.5)
2p · z
where we have introduced two projection operators
p
/ /z z/p
/
Λ+ = , Λ− = , (2.6)
2p · z 2p · z
that project onto the “plus” and “minus” components of the spinor. Note the useful relations
2pz −
/ N(P ) = MN + (P ),
p /z N(P ) = N (P ), (2.7)
M
/ N(P ) = MN(P ).
that follow readily from the Dirac equation P p
Using the explicit expressions for N(P ) it is easy to see that Λ + N = N + ∼ p+ ,
p
while Λ− N = N − ∼ 1/ p+ . To give an example of how such power counting works,
decompose the Lorentz structure in front of V1 in Eq. (2.3) in terms of light-cone vectors

/ C)αβ (γ5 N)γ = (/pC)αβ γ5 N + γ + (/pC)αβ γ5 N − γ
(P
M2 M2
+ (/zC)αβ γ5 N + γ + (/zC)αβ γ5 N − γ . (2.8)
2p · z 2p · z
The first structure on the r.h.s. is of order (p+ )3/2 , the second of order (p+ )1/2 , the
third of order (p+ )−1/2 and the fourth (p+ )−3/2 , respectively. These contributions are
interpreted as twist-3, twist-4, twist-5 and twist-6, respectively, and it follows that the
invariant function V1 contributes to all twists starting from the leading one. Such an effect
is familiar from deep inelastic scattering, where the twist-3 structure function g2 (x, Q2 )
receives the so-called Wandzura–Wilczek contribution related to the leading-twist structure
function g1 (x, Q2 ).
The twist classification based on counting of powers of 1/p+ is mathematically similar
to the light-cone quantisation approach of [37]. In this language, one decomposes the quark
fields contained in the matrix element Eq. (2.3) in “plus” and “minus” components q =
q + + q − in the same manner as done above with the nucleon spinor Eq. (2.5). The leading
twist amplitude is identified as the one containing three “plus” quark fields while each
“minus” component introduces one additional unit of twist. Up to possible complications
due to isospin, one expects, therefore, to find eight independent three-quark nucleon
distribution amplitudes: one corresponding to the twist-3 operator (u+ u+ d + ), three
related to the possible twist-4 operators (u+ u+ d − ), (u+ u− d + ), (u− u+ d + ), three more
amplitudes of twist-5 of the type (u− u− d + ), (u− u+ d − ), (u+ u− d − ) and one amplitude of
twist-6 having the structure (u− u− d − ).
Either way, distribution amplitudes of definite twist correspond to the decomposition
of Eq. (2.3) in different light-cone components. After a simple algebra, we arrive at the
following definition of light-cone nucleon distribution amplitudes:
j
4h0|εij k uiα (a1 z)uβ (a2 z)dγk (a3 z)|P i

= S1 MCαβ γ5 N + γ + S2 MCαβ γ5 N − γ + P1 M(γ5 C)αβ Nγ+ + P2 M(γ5 C)αβ Nγ−
V3
+ V1 (/ pC)αβ γ5 N + γ + V2 (/ p C)αβ γ5 N − γ + M(γ⊥ C)αβ γ ⊥ γ5 N + γ
2
V4 M 2
+ M(γ⊥ C)αβ γ ⊥ γ5 N − γ + V5 (/zC)αβ γ5 N + γ
2 2pz
M2
+ V6 (/zC)αβ γ5 N − γ + A1 (/ pγ5 C)αβ Nγ+ + A2 (/pγ5 C)αβ Nγ−
2pz
A3 A4
+ M(γ⊥ γ5 C)αβ γ ⊥ N + γ + M(γ⊥ γ5 C)αβ γ ⊥ N − γ
2 2
M2 M 2
+ A5 (/zγ5 C)αβ Nγ+ + A6 (/zγ5 C)αβ Nγ−
2pz 2pz

+ T1 (iσ⊥p C)αβ γ ⊥ γ5 N + γ + T2 (iσ⊥ p C)αβ γ ⊥ γ5 N − γ
M M
+ T3 (iσp z C)αβ γ5 N + γ + T4 (iσz p C)αβ γ5 N − γ
pz pz
M 2 M2
+ T5 (iσ⊥ z C)αβ γ ⊥ γ5 N + γ + T6 (iσ⊥ z C)αβ γ ⊥ γ5 N − γ
2pz 2pz
T7 0 T8 0
+ M (σ⊥ ⊥0 C)αβ σ ⊥ ⊥ γ5 N + γ + M (σ⊥ ⊥0 C)αβ σ ⊥ ⊥ γ5 N − γ , (2.9)
2 2
where an obvious notation σpz = σ µν pµ zν etc. is used as a shorthand and ⊥ stands for the
projection transverse to z, p, e.g., γ⊥ γ ⊥ = γ µ gµν
⊥ γ ν with g ⊥ = g −(p z +z p )/pz.
µν µν µ ν µ ν
By power counting we identify three twist-3 distribution amplitudes V1 , A1 , T1 ,
nine twist-4 and twist-5, respectively, and three twist-6 distributions, see Table 1. Each
distribution amplitude F = Vi , Ai , Ti , Si , Pi can be represented as
Z P
F (ai p · z) = Dx e−ipz i xi ai F (xi ), (2.10)
P
where the functions F (xi ) depend on the dimensionless variables xi , 0 < xi < 1, i xi = 1
which correspond to the longitudinal momentum fractions carried by the quarks inside the
nucleon. The integration measure is defined as
Z Z1
Dx = dx1 dx2 dx3 δ(x1 + x2 + x3 − 1). (2.11)
0
Comparison of the expansions in Eqs. (2.3) and (2.9) then leads to the expressions for
the invariant functions Vi , Ai , Ti , Si , Pi in terms of the distribution amplitudes. For scalar
and pseudoscalar distributions we have:
Table 1
Twist classification of the distribution amplitudes in Eq. (2.9)
twist-3 twist-4 twist-5 twist-6
Vector V1 V2 , V3 V4 , V5 V6
Pseudo-vector A1 A2 , A3 A4 , A5 A6
Tensor T1 T2 , T3 , T7 T4 , T5 , T8 T6
Scalar S1 S2
Pseudo-scalar P1 P2
S1 = S1 , 2p · z S2 = S1 − S2 ,
P1 = P1 , 2p · z P2 = P2 − P1 , (2.12)
for vector distributions:
V1 = V1 , 2p · zV2 = V1 − V2 − V3 ,
2V3 = V3 , 4p · zV4 = −2V1 + V3 + V4 + 2V5 ,
4p · zV5 = V4 − V3 , (2p · z)2 V6 = −V1 + V2 + V3 + V4 + V5 − V6 , (2.13)
for axial vector distributions:
A1 = A1 , 2p · zA2 = −A1 + A2 − A3 ,
2A3 = A3 , 4p · zA4 = −2A1 − A3 − A4 + 2A5 ,
4p · zA5 = A3 − A4 , (2p · z)2 A6 = A1 − A2 + A3 + A4 − A5 + A6 , (2.14)
and, finally, for tensor distributions:
T1 = T1 , 2p · zT2 = T1 + T2 − 2T3 ,
2T3 = T7 , 2p · zT4 = T1 − T2 − 2T7 ,
2p · zT5 = −T1 + T5 + 2T8 , (2p · z)2 T6 = 2T2 − 2T3 − 2T4 + 2T5 + 2T7 + 2T8 ,
4p · zT7 = T7 − T8 , (2p · z)2 T8 = −T1 + T2 + T5 − T6 + 2T7 + 2T8 .
(2.15)
2.2. Symmetry properties
Not all of the 24 distribution amplitudes in Eq. (2.9) are independent. First of all,
each distribution amplitude itself has definite symmetry properties. The identity of the
two u-quarks in the proton together with symmetry properties of quark operators and
γ -matrices implies that the vector and tensor distribution amplitudes are symmetric,
whereas scalar, pseudoscalar and axial-vector distributions are antisymmetric under the
interchange of the first two arguments:
Vi (1, 2, 3) = Vi (2, 1, 3), Ti (1, 2, 3) = Ti (2, 1, 3),
Si (1, 2, 3) = −Si (2, 1, 3), Pi (1, 2, 3) = −Pi (2, 1, 3),
Ai (1, 2, 3) = −Ai (2, 1, 3). (2.16)
Similar relations hold for the “calligraphic” structures in Eq. (2.3).
In addition, the full matrix element in Eq. (2.9) has to fulfill the symmetry relation
j
h0|εij k uiα (1)uβ (2)dγk (3)|P i + h0|εij k uiα (1)ujγ (3)dβk (2)|P i
j
+ h0|εij k uiγ (3)uβ (2)dαk (1)|P i = 0, (2.17)
that follows from the condition that the nucleon state has isospin 1/2:

1 1 j
T2− + 1 h0|εij k uiα (1)uβ (2)dγk (3)|P i = 0, (2.18)
2 2
where
1
T2 = (T+ T− + T− T+ ) + T32 , (2.19)
2
and T± are the usual isospin step-up and step-down operators. Applying the set of Fierz
transformations detailed in Appendix A, we end up with the following six equations:
2T1 (1, 2, 3) = [V1 − A1 ](1, 3, 2) + [V1 − A1 ](2, 3, 1),

[T3 + T7 + S1 − P1 ](1, 2, 3) = [V3 − A3 ](3, 1, 2) + [V2 − A2 ](2, 3, 1),
2T2 (1, 2, 3) = [T3 − T7 + S1 + P1 ](3, 1, 2)
+ [T3 − T7 + S1 + P1 ](3, 2, 1),
[T4 + T8 + S2 − P2 ](1, 2, 3) = [V4 − A4 ](3, 1, 2) + [V5 − A5 ](2, 3, 1),
2T5 (1, 2, 3) = [T4 − T8 + S2 + P2 ](3, 1, 2)
+ [T4 − T8 + S2 + P2 ](3, 2, 1),
2T6 (1, 2, 3) = [V6 − A6 ](1, 3, 2) + [V6 − A6 ](2, 3, 1). (2.20)
The first relation in Eq. (2.20) is familiar from [30–33] and expresses the tensor nucleon
distribution amplitude of the leading twist in terms of the vector and axial vector
distributions. Since the latter have different symmetry, they can be combined together to
define the single independent leading twist-3 nucleon distribution amplitude
Φ3 (x1 , x2 , x3 ) = [V1 − A1 ](x1, x2 , x3 ), (2.21)
which is well known and received a lot of attention in the past. The rest of the relations
in Eq. (2.20) are new and constrain the number of independent distribution amplitudes of
higher twist.
In particular, nine distributions amplitudes of twist-4 (cf. Table 1) can be reduced to
three independent distributions that we choose to be
Φ4 (x1 , x2 , x3 ) = [V2 − A2 ](x1 , x2 , x3 ),

Ψ4 (x1 , x2 , x3 ) = [V3 − A3 ](x1 , x2 , x3 ),
Ξ4 (x1 , x2 , x3 ) = [T3 − T7 + S1 + P1 ](x1 , x2 , x3 ). (2.22)
It is easy to check that all nine distributions V2 , A2 , T2 , V3 , A3 , T3 , T7 , S1 , P1 can be

restored from the three above, taking into account the isospin relations in the second and
the third lines in Eq. (2.20).
Similarly, we introduce three independent twist-5 distribution amplitudes
Φ5 (x1 , x2 , x3 ) = [V5 − A5 ](x1 , x2 , x3 ) ,

Ψ5 (x1 , x2 , x3 ) = [V4 − A4 ](x1 , x2 , x3 ) ,
Ξ5 (x1 , x2 , x3 ) = [T4 − T8 + S2 + P2 ](x1 , x2 , x3 ). (2.23)
Finally, one distribution amplitude exists of twist-6:
Φ6 (x1 , x2 , x3 ) = [V6 − A6 ](x1, x2 , x3 ). (2.24)

2.3. Representation in terms of chiral fields
The physics interpretation of the distribution amplitudes Φ(xi ) is most transparent in

terms of quark fields of definite chirality
1
q ↑(↓) = (1 ± γ5 )q. (2.25)
2
Projection on the state where the spins of the two up-quarks are antiparallel u↑ u↓ singles
out vector and axial vector amplitudes, while parallel spins u↑ u↑ or u↓ u↓ correspond
to scalar, pseudoscalar and tensor structures. We are lead to the classification of the
distribution amplitudes in terms of the quarks light-cone components and spin projections,
summarized in Table 2. The leading twist-3 distribution amplitude can be defined as
(cf. [38]):
↑ ↓ ↑
h0|εij k ui (a1 z)C/zuj (a2 z) /zdk (a3 z)|P i
Z P
1
= − pz/zN ↑ Dx e−ipz xi ai Φ3 (xi ). (2.26)
2
Twist-4 distributions allow for the following representation:
↑ ↓ ↑
h0|εij k ui (a1 z)C/zuj (a2 z) p
/ dk (a3 z)|P i
Z P
1
= − pz/ p N ↑ Dx e−ipz xi ai Φ4 (xi ),
2
ij k ↑ ↓ ↓
h0|ε ui (a1 z)C/zγ⊥p / uj (a2 z) γ ⊥ /zdk (a3 z)|P i
Z P
= −pzM/zN ↑ Dx e−ipz xi ai Ψ4 (xi ),
Table 2
Eight independent nucleon distribution amplitudes that enter the expansion in Eq. (2.9)
Lorentz-structure Light-cone projection Nomenclature
twist-3 (C /z) ⊗ /z u+ + +
↑ u↓ d↑ Φ3 (xi ) = [V1 − A1 ](xi )
twist-4 (C /z) ⊗ p
/ u+ + −
↑ u↓ d↑ Φ4 (xi ) = [V2 − A2 ](xi )
/ ) ⊗ γ ⊥ /z
(C /zγ⊥p u+ − +
↑ u↓ d↓ Ψ4 (xi ) = [V3 − A3 ](xi )
/ /z) ⊗ /z
(Cp u− + +
↑ u↑ d↑ Ξ4 (xi ) = [T3 − T7 + S1 + P1 ](xi )
twist-5 / ) ⊗ /z
(Cp u− − +
↑ u↓ d↑ Φ5 (xi ) = [V5 − A5 ](xi )
/ γ⊥ /z) ⊗ γ ⊥p/
(Cp u− + −
↑ u↓ d↓ Ψ5 (xi ) = [V4 − A4 ](xi )
/) ⊗ p
(C /zp / u+ − −
↑ u↑ d↑ Ξ5 (xi ) = [T4 − T8 + S2 + P2 ](xi )
twist-6 /) ⊗ p
(Cp / u− − −
↑ u↓ d↑ Φ6 (xi ) = [V6 − A6 ](xi )
↑ ↑ ↑
h0|εij k ui (a1 z)C/p z/uj (a2 z) /zdk (a3 z)|P i
Z P
1 ↑
= pzM/zN Dx e−ipz xi ai Ξ4 (xi ) (2.27)
2
and, similar, for twist-5:
↑ ↓ ↑
h0|εij k ui (a1 z)C/
p uj (a2 z) /zdk (a3 z)|P i
Z P
1 2 ↑
= − M /zN Dx e−ipz xi ai Φ5 (xi ),
4
↑ ↓ ↓
p γ⊥ /zuj (a2 z) γ ⊥p
h0|εij k ui (a1 z)C/ / dk (a3 z)|P i
Z P
= −pzM/ p N ↑ Dx e−ipz xi ai Ψ5 (xi ),
↑ ↑ ↑
h0|εij k ui (a1 z)C/zp / dk (a3 z)|P i
/ uj (a2 z) p
Z P
1
= pzM/ p N ↑ Dx e−ipz xi ai Ξ5 (xi ). (2.28)
2
Finally, the twist-6 distribution amplitude is written as
↑ ↓ ↑
h0|εij k ui (a1 z)C/ / dk (a3 z)|P i
p uj (a2 z) p
Z P
1
= − M 2p / N ↑ Dx e−ipz xi ai Φ6 (xi ). (2.29)
4
In the rest of the paper we study this set of distribution amplitudes in some more
detail. Final expressions for all the 24 distribution amplitudes in Eq. (2.9) are collected
in Appendix C.
3. Conformal expansion
The conformal expansion of light-cone distribution amplitudes is the field-theoretic

analogue to the partial wave expansion in quantum mechanics. The idea, in both cases,
is to use the symmetry of the problem to introduce a set of separated coordinates. In
quantum mechanics, spherical symmetry of the potential allows to separate dependence
on radial coordinates from angular ones. All angular dependence is included in spherical
harmonics which form an irreducible representation of the symmetry group O(3), while the
dependence on radial coordinates is governed by a one-dimensional Schrödinger equation.
In the same spirit conformal symmetry [39] of the QCD Lagrangian can be used to
study the distribution amplitudes as it allows to separate longitudinal degrees of freedom
from transverse ones [20,21,38,40–48]. The dependence on longitudinal momentum
fractions is taken into account by a set of orthogonal polynomials that form an irreducible
representation of the collinear subgroup SL(2, R) of the conformal group which describes
Möbius transformations on the light-cone. Transverse coordinates are replaced by the
renormalization scale, the dependence on which is governed by the renormalization group.
Since the renormalization group equations to leading logarithmic accuracy are driven by
tree-level counterterms, they have the conformal symmetry. As a consequence, components
in the distribution amplitudes with different conformal spin, dubbed conformal partial
waves, do not mix under renormalization to this accuracy.
Conformal spin of the quark is defined as
1
j = (l + s), (3.1)
2
where l = 3/2 is the canonical dimension of the quark field and s = ±1/2 is the quark
spin projection on the light-cone. The spin projection operators are in fact the same as
used to separate the “plus” and “minus” components of a spinor in Eq. (2.6). The “plus”
component of the quark field q + corresponds to s = 1/2 and, therefore, j = 1, while the
“minus” component q − has s = −1/2 and j = 1/2. For multiquark states, we are left with
the classical problem of summation of spins, with the difference that the group is in our
case noncompact. The distribution amplitude corresponding to the lowest conformal spin
jmin = j1 + j2 + j3 of the three-quark system is equal to [20,21,44]
0[2j1 + 2j2 + 2j3 ] 2j1 −1 2j2 −1 2j3 −1
Φas (x1 , x2 , x3 ) = x x2 x3 . (3.2)
0[2j1 ]0[2j2 ]0[2j3 ] 1
Contributions with higher conformal spin j = jmin + n, n = 1, 2, . . . , are given by Φas
multiplied by polynomials which are orthogonal over the weight function Eq. (3.2).
A suitable orthonormal basis of such “conformal polynomials” has been constructed
in [38].
In this section we consider the conformal expansion of nucleon distribution amplitudes
taking into account contributions of leading and next-to-leading conformal spin (i.e., “S”-
and “P”-waves). The expansions can easily be extended to arbitrary spin. One reason why
we do not present complete expansions in this paper is that we do not have the tools to
estimate the corresponding additional parameters. The second reason is that for yet higher
spins one has to take into account contributions of four-particle distributions involving an
extra gluon that we do not consider here.
3.1. Leading twist-3 distribution amplitude
At leading twist, there is only one independent distribution amplitude corresponding to

the light-cone projection (C/z) ⊗ /z and therefore involving three “plus” quark fields , see
Table 2. The conformal expansion then reads

Φ3 (xi , µ) = 120x1x2 x3 φ30 (µ) + φ3− (µ)(x1 − x2 ) + φ3+ (µ)(1 − 3x3 ) + · · · . (3.3)
Here µ is the renormalization scale. For the discussion of the shape of the distribution it is
often convenient to rewrite Eq. (3.3) to factor out the overall normalization:

Φ3 (xi , µ) = 120x1x2 x3 φ30 (µ) 1 + e
φ3− (µ)(x1 − x2 ) + e
φ3+ (µ)(1 − 3x3) + · · · . (3.4)
The relation between φ3± and φ e± is obvious.
3
The statement of conformal symmetry is that the coefficients φ ± (µ) do not mix with
φ 0 (µ) under renormalization since they have different conformal spin: j = 4 and j = 3,
respectively. By an explicit calculation one obtains [49–53]
αs (µ2 )
φ30 (µ2 ) = L2/3b φ30 (µ1 ), L≡ , (3.5)
αs (µ1 )
where b = 11 − 2/3nf , and
1 + 1 −
e
φ3+ (µ2 ) = 3L20/9b + L8/3b e
φ3 (µ1 ) + L20/9b − L8/3b e
φ3 (µ1 ),
4 4
3 + 1 −
e
φ3− (µ2 ) = L20/9b − L8/3b e
φ3 (µ1 ) + L20/9b + 3L8/3b e
φ3 (µ1 ). (3.6)
4 4
Numerical estimates for the coefficients are available from QCD sum rules [30–33]: 1
φ30 (µ = 1 GeV) = (5.3 ± 0.5) × 10−3 GeV2 ,

e
φ + (µ = 1 GeV) = 1.1 ± 0.3,
3
e
φ3− (µ = 1 GeV) = 4.0 ± 1.5. (3.7)
More sophisticated models suggested in [30–33] involve in addition contributions of
second order polynomials related to the operators with conformal spin-5. The estimates
of the corresponding coefficients are, however, less reliable and have large errors.
3.2. Higher-twist distribution amplitudes
The conformal expansion of the higher-twist distribution amplitudes defined in Section 2

is equally straightforward. With the help of the general expression in Eq. (3.2) one obtains
for twist-4:

Φ4 (xi ) = 24x1x2 φ40 + φ4− (x1 − x2 ) + φ4+ (1 − 5x3 ) ,

Ψ4 (xi ) = 24x1x3 ψ40 + ψ4− (x1 − x3 ) + ψ4+ (1 − 5x2 ) ,

Ξ4 (xi ) = 24x2x3 ξ40 + ξ4− (x2 − x3 ) + ξ4+ (1 − 5x1 ) , (3.8)
for twist-5:

Φ5 (xi ) = 6x3 φ50 + φ5− (x1 − x2 ) + φ5+ (1 − 2x3 ) ,

Ψ5 (xi ) = 6x2 ψ50 + ψ5− (x1 − x3 ) + ψ5+ (1 − 2x2) ,

Ξ5 (xi ) = 6x1 ξ50 + ξ5− (x2 − x3 ) + ξ5+ (1 − 2x1 ) , (3.9)
and for twist-6:

Φ6 (xi ) = 2 φ60 + φ6− (x1 − x2 ) + φ6+ (1 − 3x3 ) . (3.10)
At this point the expansion introduces 21 new parameters. Our next task is to find out how
many of the parameters actually are independent and which are connected by equations of
motion.
The normalisation of all distribution amplitudes and, therefore, the asymptotic wave
functions are determined by matrix elements of a local three-quark operator without
derivatives. The Lorentz decomposition of a local three-quark matrix element is much
1 In notations of [30–33] φ 0 ≡ f . The given numbers correspond to the last reference in [30–33].
3 N
simpler compared to the general parametrisation in Eq. (2.3) and involves only four
structures:
j
4h0|εij k uiα (0)uβ (0)dγk (0)|N, P i

/ C)αβ (γ5 N)γ + V30 M(γµ C)αβ γ µ γ5 N γ
= V10 (P

+ T10 P ν iσµν C αβ γ µ γ5 N γ + T30 M(σµν C)αβ σ µν γ5 N γ . (3.11)
From isospin constraints it follows that in addition V10 = T10 . Thus there exist only three
independent constants.
Remarkably, these three parameters are well known and can be obtained from existing
estimates of the following three matrix elements:

h0|εij k ui (0)C/zuj (0) γ5 /zd k (0)|P i = fN pz/zN(P ),

h0|εij k ui (0)Cγµ uj (0) γ5 γ µ d k (0)|P i = λ1 MN(P ),

h0|εij k ui (0)Cσµν uj (0) γ5 σ µν d k (0)|P i = λ2 MN(P ). (3.12)
The parameter fN = V10 enters already at the level of twist-3 and determines the
normalisation of the leading twist distribution amplitude Eq. (3.7). The other two
parameters λ1 = (V10 − 4V30 ) and λ2 = 6(V10 − 4T30 ) correspond to the nucleon coupling
to the two possible independent nucleon interpolating fields that are widely used in
calculations of dynamical characteristics of the nucleon in the QCD sum rule approach.
The operator corresponding to λ1 was introduced in [34] while the one corresponding
to λ2 was advertised in [35,36]. The QCD sum rules summarized in Appendix B yield the
following estimates:
λ1 (µ = 1 GeV) = −(2.7 ± 0.9) × 10−2 GeV2 ,

λ2 (µ = 1 GeV) = (5.1 ± 1.9) × 10−2 GeV2 . (3.13)
Anomalous dimensions are the same for both currents [54]
λ1 (µ2 ) = L2/b λ1 (µ1 ), λ2 (µ2 ) = L2/b λ2 (µ2 ). (3.14)

Note that one overall sign in the determination of vacuum-to-nucleon matrix elements
is arbitrary as it corresponds to arbitrary (unphysical) overall phase of the nucleon
wave function. The relative signs between the couplings fN , λ1 , λ2 , etc. are, however,
well defined and can be determined from suitable nondiagonal correlation functions, see
Appendix B for details. We choose fN to be real and positive; then, it turns out that λ1 is
negative and λ2 positive. The fact that λ1 and λ2 have opposite signs is known from [36],
the negative relative sign between λ1 and fN is a new result.
The coefficients in Eqs. (3.8), (3.9), (3.10) corresponding to the operators of leading
conformal spin then read
1
φ30 = φ60 = fN , φ40 = φ50 = (λ1 + fN ),
2
1 1
ξ40 = ξ50 = λ2 , ψ4 = ψ5 = (fN − λ1 ).
0 0
6 2
Note that the normalization of twist-3 and twist-6 distribution amplitudes are equal,
and similarly for twist-4 and twist-5 distributions. The constants φ4,5 0 and ψ 0 involve
4,5
both the leading twist matrix element fN Eq. (3.12) and the higher-twist contribution
∼ λ1 . The former is analogous to the Wandzura–Wilczek-type contribution to the higher-
twist distribution amplitudes studied in [22,23] for the ρ-meson, and the latter defines
the “genuine” higher-twist correction. Note that λ1 , λ2 fN that is in agreement with
our expectation (see introduction) that the matrix elements of higher-twist three-quark
operators are large.
The remaining contributions of next-to-leading conformal spin are related to operators
with one derivative. In much the same way as before and as elaborated in Appendix B the
14 unknown parameters are reduced by isospin symmetry and equations of motion to five
dimensionless parameters V1d , Au1 , f1d , f2d , f1u which we define in the following way:
→
−
h0| u(0)C/zu(0) γ5 /z(iz D d)(0)|P i = fN V1d (pz)2 /zN(P ),
↔
h0| u(0)C/zγ5 iz D u(0) /zd(0)|P i = −fN Au1 (pz)2 /zN(P ),
→
−
h0| u(0)Cγµ u(0) γ5 γ µ /z(iz D d)(0)|P i = λ1 f1d (pz)M/zN(P ),
→
−
h0| ua (0)Cσµν u(0) γ5 σ µν /z(iz D d)(0)|P i = −λ2 f2d (pz)M/zN(P ),
↔
h0| u(0)Cγµ γ5 iz D u(0) γ µ /zd(0)|P i = −λ1 f1u (pz)M/zN(P ). (3.15)
↔
Here, we have used the shorthand notation for the left–right derivative iz · D =
→ ←
− −
iz · ( D − D ) and for brevity omitted colour indices. The first two matrix elements V1d
and Au1 are leading twist-3 and were estimated in [30–33]: 2
V1d (µ = 1 GeV) = 0.23 ± 0.03,

Au1 (µ = 1 GeV) = 0.38 ± 0.15. (3.16)
We have used these estimates above in Eq. (3.7): φ3−
e
= (21/2)Au1 and e φ3+ =
(7/2)(1 − 3V1d ). The other three parameters are genuine higher-twist and are estimated
in Appendix B using QCD sum rules. We obtain:
f1d (µ = 1 GeV) = 0.6 ± 0.2,

f2d (µ = 1 GeV) = 0.15 ± 0.06,
f1u (µ = 1 GeV) = 0.22 ± 0.15. (3.17)
The remaining coefficients expressed in terms of the above parameters read for twist-4:
5
φ4− = λ1 1 − 2f1d − 4f1u + fN 2Au1 − 1 ,
4
+ 1
φ4 = λ1 3 − 10f1d − fN 10V1d − 3 ,
4
− 5
ψ4 = − λ1 2 − 7f1d + f1u + fN Au1 + 3V1d − 2 ,
4
2 In notation of [30–33] V d = V (0,0,1) , Au = 2A(0,1,0) .

1 1
1
ψ4+ = − λ1 −2 + 5f1d + 5f1u + fN 2 + 5Au1 − 5V1d ,
4
− 5
ξ4 = λ2 4 − 15f2d ,
16
+ 1
ξ4 = λ2 4 − 15f2d , (3.18)
16
for twist-5:
5
φ5− = λ1 f1d − f1u + fN 2Au1 − 1 ,
3
+ 5
φ5 = − λ1 4f1d − 1 + fN 3 + 4V1d ,
6
− 5
ψ5 = λ1 f1d − f1u + fN 2 − Au1 − 3V1d ,
3
5
ψ5+ = λ1 −1 + f1u + fN 1 + Au1 + V1d ,
3
5
ξ5− = − λ2 f2d ,
4
+ 5
ξ5 = λ2 2 − 3f2d , (3.19)
12
and for twist-6:
1
φ6− = λ1 1 − 4f1d − 2f1u + fN 1 + 4Au1 ,
2
+
φ6 = λ1 1 − 2f1d + fN 4V1d − 1 . (3.20)
With these relations, the construction of higher-twist nucleon distribution amplitudes is

complete to our accuracy. Note that numerical values of the parameters given in Eqs. (3.16),
(3.13) are strongly correlated within the QCD sum rule approach, and the given errors
should not be added together in the sums (3.18)–(3.20). We expect that the accuracy of the
QCD sum rule calculation of all coefficients in the distribution amplitudes is of the order
of 50%. In Table 3 we have collected the numerical values for the expansion parameters.
Table 3
Numerical values for the expansion parameters defined in Eq. (3.8) to Eq. (3.10). The values are
given in units of 10−2 GeV2
φi0 φi− φi+ ψi0 ψi− ψi+ ξi0 ξi− ξi+
twist-3: i = 3 0.53 2.11 0.57

twist-4: i = 4 −1.08 3.22 2.12 1.61 −6.13 0.99 0.85 2.79 0.56
twist-5: i = 5 −1.08 −2.01 1.42 1.61 −0.98 5.02 0.85 −0.95 3.29
twist-6: i = 6 0.53 3.09 0.49
Fig. 1. Twist-3 distribution amplitude Φ3 (xi ).
Fig. 2. Twist-4 distribution amplitudes Φ4 (xi ) in the first line to the left, Ψ4 (xi ) in the first line to
the right, and Ξ (xi ) in the second line.
In Fig. 1 we have presented the leading twist distribution amplitude Φ3 (xi ), in Fig. 2 the
three distribution amplitudes of twist-4 are plotted. Note that the scale on the z-axis is not
the same for each plot.
4. Summary and conclusions
We have carried out a systematic study of the higher-twist light-cone distribution

amplitudes of the nucleon in QCD and found that a generic three-quark matrix element
on the light-cone can be parametrized in terms of eight independent nucleon distribution
amplitudes. In particular, we have identified one distribution of leading twist-3, three of
twist-4, three of twist-5 and one of twist-6. The shape of the distribution amplitudes
at asymtotically large momentum transfers is found and is dictated by the conformal
symmetry of the QCD Lagrangian. In order to quantify the corrections, we attempt to
expand the distribution amplitudes in contributions of local conformal operators and
keep contributions of the next-to-leading conformal spin (“P”-waves). The coefficients
in this expansion present the necessary nonperturbative input. Some of them are related
by QCD equations of motion, so that to our accuracy we end up with eight independent
parameters.
From this amount, three parameters determine the leading twist amplitude and are
known already from the analysis in [30–33]; two more — the normalisations of the
higher-twist asymptotic distributions — can be directly inferred from QCD sum rule
phenomenology of the nucleon. The remaining three parameters determine the deviation
of the higher-twist distribution amplitudes from their asymptotic form and are estimated in
this work.
To avoid confusion, we remind that our work does not present the complete analysis
of all existing higher-twist distributions, but of its subset related to three-quark operators
without extra gluon fields. As explained in the introduction, we believe that the three-
quark contributions considered in this paper dominate higher-twist corrections because of
large matrix elements. Indeed, we found that the matrix elements of higher-twist three-
quark operators are of the same order or larger than those of the leading twist, see
Section 3. We stress that the situation in the nucleon case is quite different to that of
higher twist-corrections in the meson case studied earlier [20–23]. The reason is that in
the meson case “bad” components of quark fields always can be eliminated in favour
of “good” components and an additional gluon. The resulting matrix elements turn out
to be numerical small. In the nucleon case quark fields with “minus” projection give
rise to genuine three-quark higher-twist effects that cannot be attributed to higher Fock
components in the wave function, involving an additional gluon.
A detailed summary of the distribution amplitudes is presented in Appendix C. In this
paper we do not consider phenomenological applications, but expect that our results are
relevant for the studies of a wide range of interesting physical processes.
One obvious application can be found in the calculation of the nucleon form factors.
While the magnetic form factor of the proton GM (Q2 ) allows for the leading twist
description [1–8,15], it is known that the Pauli formfactor F2 (Q2 ) is higher-twist,

suppressed by an additional power of Q2 . The description of F2 therefore calls for the
input of higher-twist distribution amplitudes presented in this paper.
Twist-4 distribution amplitudes of the nucleon are, probably, the most interesting and
in general are related to various spin asymmetries in exclusive processes. Indeed, while
the total helicity is conserved in processes involving leading twist distribution amplitdues
[13], this is not the case for nonleading twist contributions. Thus spin-sensitive transition
formfactors will provide a testing ground for higher-twist effects.
Last but not least, the distribution amplitudes presented in this paper can immediately
be applied to calculations of exclusive reactions at moderate momentum transfers in the
framework of QCD light-cone sum rules [56–58]. This approach allows to study soft non-
factorisable contributions to hard exclusive reactions in a largely model-independent way,
see, e.g., [59–61].
Acknowledgements
This work was supported by the DFG, project 920585 and Graduiertenkolleg “Physik
der starken Wechselwirkung”.
Appendix A. Fierz transformations
In this section we give the Fierz transformation rules necessary to derive the isospin
constraints Eq. (2.20) from the symmetry requirement Eq. (2.17).
We write the definition of distribution amplitudes in Eq. (2.9) in a shorthand notation as
j
h0|εij k uiα (a1 z)uβ (a2 z)dγk (a3 z)|P (P , λ)i
= S1 (s1 )αβ,γ + S2 (s2 )αβ,γ + P1 (p1 )αβ,γ + P2 (p2 )αβ,γ + V1 (v1 )αβ,γ
1 1
+ V2 (v2 )αβ,γ + V3 (v3 )αβ,γ + V4 (v4 )αβ,γ + V5 (v5 )αβ,γ + V6 (v6 )αβ,γ
2 2
1 1
+ A1 (a1 )αβ,γ + A2 (a2 )αβ,γ + A3 (a3 )αβ,γ A4 (a4 )αβ,γ + A5 (a5 )αβ,γ
2 2
+ A6 (a6 )αβ,γ + T1 (t1 )αβ,γ + T2 (t2 )αβ,γ + T3 (t3 )αβ,γ + T4 (t4 )αβ,γ
1 1
+ T5 (t5 )αβ,γ + T7 (t7 )αβ,γ + T8 (t8 )αβ,γ . (A.1)
2 2
The small letters, e.g., (s1 )αβ,γ = (C)αβ Nγ stand for the Lorentz structures
(Γ C)αβ (Γ 0 N)γ , etc. By means of the following Fierz transformation
(Γ C)αβ (Γ 0 N)γ

1 1
= Cγβ (Γ Γ 0 N)α + (γ5 C)γβ (Γ γ5 Γ 0 N)α + pC)γβ (Γ /zΓ 0 N)α
(/
4 p·z
1
+ / Γ 0 N)α + (γ⊥ C)γβ (Γ γ ⊥ Γ 0 N)α
(/zC)γβ (Γ p
p·z
1 1
− pγ5 C)γβ (Γ /zγ5 Γ 0 N)α −
(/ / γ5 Γ 0 N)α
(/zγ5 C)γβ (Γ p
p·z p·z
1
− (γ⊥ γ5 C)γβ (Γ γ ⊥ γ5 Γ 0 N)α + (σpz C)γβ (Γ σzp Γ 0 N)α
(p · z)2
1 0 1
+ (σ⊥⊥0 C)γβ (Γ σ ⊥⊥ Γ 0 N)α + (iσ⊥p C)γβ (Γ γ ⊥ /zΓ 0 N)α
2 p·z

1
+ (iσ⊥z C)γβ (Γ γ ⊥p / Γ 0 N)α (A.2)
p·z
all Lorentz structures are brought into the same form, i.e., we apply for the expansion of
the third contribution in Eq. (2.17) the following transformations of twist-3 structures
1
(v1 )γβ,α = (v1 − a1 − t1 )αβ,γ ,
2
1
(a1 )γβ,α = (−v1 + a1 − t1 )αβ,γ ,
2
(t1 )γβ,α = −(v1 + a1 )αβ,γ , (A.3)
and similar transformations for twist-4

1 1
(s1 )γβ,α = s1 + p1 − 2v2 − 2a2 + v3 − a3 − t3 − 2t2 + t7 ,
4 2 αβ,γ

1 1
(p1 )γβ,α = s1 + p1 + 2v2 + 2a2 − v3 + a3 − t3 − 2t2 + t7 ,
4 2 αβ,γ

1 1
(v2 )γβ,α = −s1 + p1 + v3 + a3 − t3 − t7 ,
4 2 αβ,γ

1 1
(a2 )γβ,α = −s1 + p1 − v3 − a3 − t3 − t7 ,
4 2 αβ,γ
1
(v3 )γβ,α = (2s1 − 2p1 + 4v2 − 4a2 − 2t3 − t7 )αβ,γ ,
4
1
(a3 )γβ,α = (−2s1 + 2p1 + 4v2 − 4a2 + 2t3 + t7 )αβ,γ ,
4
1
(t2 )γβ,α = (−2s1 − 2p1 − 2t3 + t7 )αβ,γ ,
4
1 1
(t3 )γβ,α = −s1 − p1 − 2v2 − 2a2 − v3 + a3 + t3 − 2t2 − t7 ,
4 2 αβ,γ
1
(t7 )γβ,α = (2s1 + 2p1 − 4v2 − 4a2 − 2v3 + 2a3 − 2t3 + 4t2 + t7 )αβ,γ . (A.4)
4
Relations for twist-5 and twist-6 are identical up to the substitutions: {v1 , a1 , t1 } →
{v6 , a6 , t6 } and {s1 , p1 , v2 , v3 , a2 , a3 , t2 , t3 , t7 } → {s2 , p2 , v5 , v4 , a5 , a4 , t5 , t4 , t8 }. By a
simple substitution, one also gets the necessary Fierz transformations from (. . .)αγ ,β →
(. . .)αβ,γ required in the second contribution in Eq. (2.17). Putting everything together we
derive three equations for the twist-3 amplitudes from the coefficients v1 , a1 , t1 :

0 = 2T1 (1, 2, 3) − V1 (1, 3, 2) + A1 (1, 3, 2) − V1 (3, 2, 1) − A1 (3, 2, 1) (t1 )αβ,γ

+ 2V1 (1, 2, 3) + 2V1 (1, 3, 2) + 2A1 (1, 3, 2) − 2T1 (1, 3, 2)

+ V1 (3, 2, 1) − A1 (3, 2, 1) − 2T1 (3, 2, 1) (v1 )αβ,γ

+ 2A1 (1, 2, 3) + 2V1 (1, 3, 2) + 2A1 (1, 3, 2) + 2T1 (1, 3, 2)

− V1 (3, 2, 1) + A1 (3, 2, 1) − 2T1 (3, 2, 1) (a1 )αβ,γ . (A.5)
Using the symmetry properties Eq. (2.16) it is easy to see, however, that all three above
equations are in fact identical to the one given in Eq. (2.20). In the similar way one can
derive an overcomplete set of nine twist-4 equations that all can be reduced to the two
equations relating twist-4 amplitudes in Eq. (2.20). The derivation for the remaining twist-5
and twist-6 equations is similar.
Appendix B. Nonperturbative parameters
In Section 3.2 we have demonstrated that the normalisation of the asymptotic

distribution amplitudes of all twists involves three independent parameters. In the first part
of this appendix we consider leading corrections to the asymptotic distribution amplitudes,
related to contributions of three-quark conformal one extra derivative, and find that they
involve five new parameters.
In the second part, we work out QCD sum rule estimates of the higher-twist matrix
elements. Partially the results can be adapted from existing sum rule calculations
containing nucleon currents [34,36,55], partially they are new.
B.1. Equations of motion
The coefficients of “P”-wave contributions in the conformal expansion of distribution

amplitudes are given by matrix elements of operators containing one derivative. In case the
derivative acts on the down quark we have the general decomposition:
j
4h0|εij k uiα (0)uβ (0)[iDλ dγ ]k (0)|N, P i

= V1d Pλ (P
/ C)αβ (γ5 N)γ + V20 M(P/ C)αβ (γλ γ5 N)γ + V3d Pλ M(γµC)αβ γ µ γ5 N γ

+ V40 M 2 (γλ C)αβ (γ5 N)γ + V50 M 2 γ µ C αβ (iσµλ γ5 N)γ

+ T1d Pλ P ν iσµν C αβ γ µ γ5 N γ + T20 M P ν iσλν C αβ (γ5 N)γ

+ T3d Pλ M(σµν C)αβ σ µν γ5 N γ + T40 M Pν σ µν C αβ (σµλ γ5 N)γ

+ T50 M 2 (iσµλ C)αβ γ µ γ5 N γ + T70 M 2 (σµν C)αβ σ µν γλ γ5 N γ (B.1)
yielding 11 parameters.
Acting with one derivative on the up-quarks picks up the structures of γ -matrices that
are odd under transposition:
↔ j
4h0|εij k uiα (0)i D λ uβ (0)dγk (0)|N, P i
= S1u Pλ MCαβ (γ5 N)γ + S20 M 2 Cαβ (γλ γ5 N)γ
+ P1u Pλ M(γ5 C)αβ Nγ + P20 M 2 (γ5 C)αβ (γλ N)γ

+Au1 Pλ (P
/ γ5 C)αβ Nγ + A02 M(P/ γ5 C)αβ (γλ N)γ + Au3 Pλ M(γµ γ5 C)αβ γ µ N γ
αβ
+ A04 M 2 (γλ γ5 C)αβ Nγ + A05 M 2 γ µ γ5 C (iσµλ N)γ (B.2)
and introduces 9 parameters. Hence, altogether there are 20 unknown numbers. The
equations of motions yield, however, a number of constraints:
h0|εij k ui (0)Cγ% uj (0) γ λ[iDλ dγ ]k (0)|N, P i = 0,
h0|εij k ui (0)Cγ λ uj (0) [iDλdγ ]k (0)|N, P i = Pλ h0|εij k ui (0)Cγλ uj (0) dγk (0)|N, P i,
h0|εij k ui (0)Cσαβ uj (0) γ λ[iDλ dγ ]k (0)|N, P i = 0,
k
h0|εij k ui (0)Ciσαβ uj (0) iDβ dγ (0)|N, P i
= P β h0|εij k ui (0)Ciσαβ uj (0) d k (0)|N, P i
↔ ij
− h0|εij k u(0)Ci D α u(0) dγk (0)|N, P i,
k
h0|εij k ui (0)Ciγ5 σαβ uj (0) iDβ dγ (0)|N, P i
= P β h0|εij k ui (0)Cγ5 iσαβ uj (0) d k (0)|N, P i
↔ ij
− h0|εij k u(0)Cγ5 i Dα u(0) dγk (0)|N, P i, (B.3)
which translate to
V1d = 4V20 + 2V3d , −3V50 = V3d + V40 ,
V10 − V30 = V1d − V3d + 4V40 − V20 ,
0 = T3d + T50 , T20 = −T1d + 4T3d + 3T40 ,
T10 − 2T30 − S20 = T1d − 2T3d − T40 + 3T50 + 6T70 ,
T10 − 2T30 − S1u = T1d − 3T20 − 2T3d − T40 ,
−2T3d + 2T40 = −2T30 − P1u , 2T30 − P20 = 2T3d − 2T40 − 6T70 . (B.4)
The two relations in the first line in Eq. (B.4) follow from the first equation in Eq. (B.3), the
relations in the second line from the second equation, the third line from the third equation,
fourth and fifth line from the fourth equation and the last line from the last equation. In a
similar way one observes that the axial structures have to fulfill the constraints
↔ ij
h0|εij k u(0)Cγ % γ5 D % u(0) dγk (0)|N, P i = 0,
↔ ↔ ij
h0|εij k u(0)C γλ i D % −γ% i D λ γ5 u(0) dγk (0)|N, P i
ij
i α ← − →
−
= h0|ε ij k
u(0)C σλ% γ i D α + γ σλ% i D α γ5 u(0) dγk (0)|N, P i
α
2
α
= −iελ%αδ P h0|εij k ui (0)Cγ δ uj (0)dγk (0)|N, P i
k
− h0|εij k ui (0)Cγ δ uj (0) iDα dγ (0)|N, P i , (B.5)
where from it follows that

Au1 + Au3 + A02 + 4A04 = 0,
Au3 − A02 = V20 + V30 − V3d , 2A05 = V20 + V30 − 2V50 − V3d , (B.6)
where the relation in the first line is derived from the first equation in Eq. (B.5). Combining
the constraints from the equations of motion with the eight isospin relations obtainable
from Eq. (2.20) we obtain an overcomplete set of 20 equations. Choosing V1d , Au1 , V3d ,
Au3 and T3d as the independent parameters and solving these equations, we can express the
remaining parameters as
1 d 1
V20 = V − 2 V3d , V40 = 4 V10 − 3 V1d − 4 V30 + 2 V3d ,
4 1 16
1 Vd 3 V3d
V50 = −4 V10 + 3 V1d + 4 V30 − 18 V3d , A02 = Au3 − 1 − V30 + ,
48 4 2
1
A04 = −4 Au1 − 8 Au3 + V1d + 4 V30 − 6 V3d ,
16
1
A5 =
0
4 V10 + 3 V1d + 20 V30 − 18 V3d ,
48
1 u
T2 = A1 + 6 Au3 + 8 T3d − V10 − 2 V1d − 6 V30 + 12 V3d ,
0
T50 = −T3d ,
8
1
T40 = −Au1 + 2 Au3 − 8 T3d + V10 − 2 V1d − 2 V30 + 4 V3d ,
8
1
T70 = 3 Au1 − 6 Au3 − 16 T30 + 48 T3d + V10 + 6 V1d + 6 V30 − 12 V3d ,
96
1
S20 = 3 Au1 + 10 Au3 − 16 T30 + 16 T3d + 9 V10 − 2 V1d − 10 V30 + 20 V3d ,
16
1
S1 = 3 Au1 + 10 Au3 − 8 T30 + 16 T3d + V10 − 2 V1d − 10 V30 + 20 V3d ,
u
4
1
P1 = Au1 − 2 Au3 − 8 T30 + 16 T3d − V10 + 2 V1d + 2 V30 − 4 V3d ,
u
4
1
P20 = −Au1 + 2 Au3 + 16 T30 − 16 T3d + 5 V10 − 2 V1d − 2 V30 + 4 V3d ,
16
1
T1d = −Au1 + V10 − V1d . (B.7)
2
Finally, the parameters V1d , Au1 , V3d , Au3 and T3d can be expressed by the reduced matrix
elements of the operators defined in Eq. (3.15). We have
1 d
V1d = fN V1d , Au1 = fN Au1 , λ1 f1d = V − 6V3d ,
2 1
1
λ1 f1u = −2Au1 − 12Au3 + V1d + 4V30 − 6V3d ,
2

λ2 f2d = 2 −2Au3 − 8T3d + V1d + 2V30 − 4V3d + 2T1d . (B.8)
The first two parameters appear already at the level of twist-3 and are known from [30–33].
The others will be calculated in the next subsection.
B.2. QCD sum rule estimates
The QCD sum rule estimates for λ1 , λ2 are derived from the consideration of the two-
point correlation function
Z
q + M)
(/
i d4 x eiqx h0|ηi (x)η̄i (0)|0i = |λi |2 M 2 2 + ···, (B.9)
M − q2
where h0|η1 (0)|P i = Mλ1 N(P ) and h0|η2 (0)|P i = Mλ2 N(P ) 3 are the local three-
quark operators defined in Eq. (3.12). The dots refer to excited states and the continuum
contribution. Taking into account vacuum condensates up to dimension-8 one obtains the
sum rules [36,55]:

b
2(2π) M |λ1 | = exp M /MB MB6 E3 s0 /MB2 + MB2 E1 s0 /MB2
4 2 2 2 2
4

a2 m20
+ 4− 2 (B.10)
3 MB
and

2 |λ2 |
2 b 2
2(2π) M 4
= exp M 2
/MB2 MB E3 s0 /MB + MB E1 s0 /MB ,
6 2 2
(B.11)
6 4
where

2 X 1
n−1
s0 k
En s0 /MB2 = 1 − e(−s0 /MB ) . (B.12)
k! MB2
k=0
Here and below MB is the Borel parameter. We use the Borel window 1 GeV2 6 MB2 6
√
2 GeV2 , with the continuum threshold s0 ∼ 1.5 GeV and values of the condensates
normalized at µ = 1 GeV [22,23]:
a = −(2π)2 hq̄qi ∼ 0.55 GeV3 ,

2 αS 2
b = (2π) G ∼ 0.47 GeV4 ,
π
hq̄gGqi
m20 = ∼ 0.8 GeV2 . (B.13)
hq̄qi
With these inputs, we find |λ1 | = 0.027 ± 0.009 GeV2 , |λ2 | = 0.051 ± 0.019 GeV2 .
Note that the above sum rules only fix the absolute value of the parameters λ1 and λ2 .
Relative phases between different nucleon-to-vacuum matrix elements can be computed
by investigating suitable nondiagonal correlation functions. To determine the relative sign
between fN and λ1 we consider the correlation function
Z
q + M)
(/
i d4 x eiqx h0|εij k ui (x)C/zuj (x) γ5 /zd k (x) η̄1 (0)|0i = fN λ∗1 M/z 2 + ···.
M − q2
(B.14)
Taking the ratio of the corresponding sum rule and the sum rule Eq. (B.10) we obtain
3 Note that our normalisation of λ differs by a factor M from the standard one.
i
fN 1 (2MB2 E1 (s0 /MB2 ) − m20 )aM

=− , (B.15)
λ1 3 M 6 E3 (s0 /M 2 ) + b M 2 E1 (s0 /M 2 ) + a 2 (4 − m2 /M 2 )
B B 4 B B 3 0 B
which is real and negative. We have chosen fN to be positive according to the standard
choice in [30–33] and λ1 to be negative. Note also that the above sum rule leads to a value
fN = 5.6 × 10−3 GeV2 consistent with standard estimates. A similar calculation of the
nondiagonal correlation function
Z
q + M)
(/
i d4 x eiqx h0|η1 (x)η̄2 (0) + η2 (x)η̄1 (0)|0i = λ1 λ∗2 + λ2 λ∗1 M 2 2 + ···
M − q2
(B.16)
was performed in [36]. The resulting sum rule

−2(2π)4 M 3 λ1 λ∗2 + λ2 λ∗1

2 2 32αs a 2
= 2aeM /MB 12MB4 E2 s0 /MB2 − 6m20 MB2 E1 s0 /MB2 + (B.17)
27πMB2
shows that the relative sign between λ2 and λ1 is negative.
To estimate the remaining parameters we consider the nondiagonal correlation functions
involving the last three higher-twist operators in Eq. (3.15) with either η̄1 (0) or η̄2 (0):
Z
→ k
−
i d4 x eiqx h0|εij k ui (x)Cγµ uj (x) γ5 γ µ iz D d (x) η̄1 (0)|0i
f d |λ1 |2 M 2 q · z/
q
= 1 + ···,
M2 − q2
Z
↔ ij
i d4 x eiqx h0|εij k u(x)Cγµ γ5 iz D u(x) γ µ d k (x) η̄1 (0)|0i
f u |λ1 |2 M 2 q · z/
q
= 1 + ···,
M −q2 2
Z
→ k
−
i d4 x eiqx εij k h0| ui (x)Cσµν uj (x) γ5 σ µν iz D d (x) η̄2 (0)|0i
f2d |λ2 |2 M 2 q · z/
q
= + ···. (B.18)
M −q2 2
The dots refer, again, to contributions of excited states and different Lorentz structures that
we do not consider. Expressing |λi |2 by the expression in Eqs. (B.10) and (B.11) we obtain
the following sum rules:
a2 m20
3 6
10 MB E3 s0 /MB2 + b9 MB2 E1 s0 /MB2 + 4−
3 MB2
f1d = ,
a2 m20
MB6 E3 s0 /MB2 + b4 MB2 E1 s0 /MB2 + 4−
3 MB2
2
5a 2 m0
1 6
10 MB E3 s0 /MB2 + b8 MB2 E1 s0 /MB2 + 18 M 2
f1u =
B
,
a2 m20
MB6 E3 s0 /MB2 + b4 MB2 E1 s0 /MB2 + 4−
3 MB2

8 6
5 MB E3 s0 /MB2 − 4b 2 2
9 MB E1 s0 /MB
f2d = . (B.19)
6MB6 E3 s0 /MB2 + 3b 2
2 MB E1 s0 /MB
2
Inserting the numerical values of parameters, we end up with the following estimates for
the higher-twist matrix elements:
f1d (µ = 1 GeV) = 0.6 ± 0.2,

f2d (µ = 1 GeV) = 0.15 ± 0.06,
f1u (µ = 1 GeV) = 0.22 ± 0.15. (B.20)
Appendix C. Handbook of nucleon distribution amplitudes
In this appendix we give a complete list of all 24 nucleon distribution amplitudes.

The chiral structure of our basic set of eight independent distribution amplitudes is
summarized in Table 2 in the text. The corresponding representations for the remaining
16 distributions are collected in Table A and Table B. The set in Table A contains the
amplitudes that can be directly obtained from Eq. (2.26) to Eq. (2.29) by flipping spin-
projections of the two up-quarks. Table B contains four distributions that are obtainable
from the previous ones by exchanging light-cone projections, and also T1 , T2 , T5 and T6
that cannot be read off from the previous expressions in a straightforward way and therefore
are given below:
↑ ↑ ↓
h0|εij k ui (a1 z)Ciσ⊥z uj (a2 z) γ ⊥ /zdk (a3 z)|P i
Z P
= −2pz/zN ↑ Dx e−ipz xi ai T1 (xi ),
Table A
Eight nucleon distribution amplitudes that can be obtained from Table 1 by flipping the spin
projections of the two up quarks
twist-3 (C /z) ⊗ /z u+ + +
↓ u↑ d↑ [V1 + A1 ](xi )
twist-4 (C /z) ⊗ p
/ u+ + −
↓ u↑ d↑ [V2 + A2 ](xi )
/ ) ⊗ γ ⊥/z
(C /z γ⊥p u+ − +
↓ u↑ d↓ [V3 + A3 ](xi )
/ /z) ⊗ /z
(Cp u− + +
↓ u↓ d↑ [T3 + T7 + S1 − P1 ](xi )
twist-5 / ) ⊗ /z
(Cp u− − +
↓ u↑ d↑ [V5 + A5 ](xi )
/ γ⊥ /z) ⊗ γ ⊥p/
(Cp u− + −
↓ u↑ d↓ [V4 + A4 ](xi )
+ − −
/) ⊗ p
(C /zp / u↓ u↓ d↑ [T4 + T8 + S2 − P2 ](xi )
twist-6 / ) ⊗ p/
(Cp u− − −
↓ u↑ d↑ [V6 + A6 ](xi )
↓ ↓ ↓
h0|εij k ui (a1 z)Ciσ⊥z uj (a2 z) γ ⊥p
/ dk (a3 z)|P i
Z P
= −2pz/ p N ↑ Dx e−ipz xi ai T2 (xi ),
↓ ↓ ↓
h0|εij k ui (a1 z)Ciσ⊥p uj (a2 z) γ ⊥ /zdk (a3 z)|P i
Z P
= −M 2 /zN ↑ Dx e−ipz xi ai T5 (xi ),
↑ ↑ ↓
h0|εij k ui (a1 z)Ciσ⊥p uj (a2 z) γ ⊥p
/ dk (a3 z)|P i
Z P
= −M 2p / N ↑ Dx e−ipz xi ai T6 (xi ). (C.1)
Note also that for scalar and tensor contributions the following representations can be
given:
↑ ↑ ↑
h0|εij k ui (a1 z)Cuj (a2 z) /zdk (a3 z)|P i
Z P
1
= M/zN ↑ Dx e−ipz xi ai [S1 + P1 ](xi ),
2
↑ ↑ ↑
h0|εij k ui (a1 z)Ciσzp uj (a2 z) /zdk (a3 z)|P i
Z P
1 ↑
= pzM/zN Dx e−ipz xi ai [T3 − T7 ](xi ). (C.2)
2
In the following subsections we give the complete set of distribution amplitudes starting
from twist-3 to twist-6. The numerical values of the expansion parameters can be obtained
from Table 3 in the text.
Table B
The remaining eight nucleon distribution amplitudes that enter the expansion in Eq. (2.9)
twist-3 (Ciσ⊥z ) ⊗ γ ⊥ /z u+ + +
↑ u↑ d↓ T1 (xi )
twist-4 (Ciσ⊥z ) ⊗ γ ⊥p
/ u+ + −
↓ u↓ d↓ T2 (xi )
/ ) ⊗ /z
(C /zp u+ − +
↑ u↑ d↑ −[T3 − T7 − S1 − P1 ](xi )
/ ) ⊗ /z
(C /zp u+ − +
↓ u↓ d↑ −[T3 + T7 − S1 + P1 ](xi )
twist-5 (Ciσ⊥p ) ⊗ γ ⊥ /z u− − +
↓ u↓ d↓ T5 (xi )
/ /z) ⊗ p
(Cp / u− + −
↑ u↑ d↑ −[T4 − T8 − S2 − P2 ](xi )
− + −
/ /z) ⊗ p
(Cp / u↓ u↓ d↑ −[T4 + T8 − S2 + P2 ](xi )
twist-6 (Ciσ⊥p ) ⊗ γ ⊥p
/ u− − −
↑ u↑ d↓ T6 (xi )
C.1. Twist-3 distribution amplitudes

V1 (xi , µ) = 120x1x2 x3 φ30 (µ) + φ3+ (µ)(1 − 3x3) ,
A1 (xi , µ) = 120x1x2 x3 (x2 − x1 )φ3− (µ),

1
T1 (xi , µ) = 120x1x2 x3 φ30 (µ) + φ3− − φ3+ (µ)(1 − 3x3 ) . (C.3)
2

V2 (xi , µ) = 24x1x2 φ40 (µ) + φ4+ (µ)(1 − 5x3 ) ,
A2 (xi , µ) = 24x1x2 (x2 − x1 )φ4− (µ),
h
V3 (xi , µ) = 12x3 ψ40 (µ)(1 − x3 ) + ψ4− (µ) x12 + x22 − x3 (1 − x3 )
i
+ ψ4+ (µ)(1 − x3 − 10x1x2 ) ,

A3 (xi , µ) = 12x3(x2 − x1 ) ψ40 (µ) + ψ4+ (µ) + ψ4− (µ)(1 − 2x3 ) ,
h
T3 (xi , µ) = 6x3 ξ40 + φ40 + ψ40 (µ)(1 − x3 )

+ ξ4− + φ4− − ψ4− (µ) x12 + x22 − x3 (1 − x3 )
i
+ ξ4+ + φ4+ + ψ4+ (µ)(1 − x3 − 10x1x2 ) ,
h
T7 (xi , µ) = 6x3 −ξ40 + φ40 + ψ40 (µ)(1 − x3 )

+ −ξ4− + φ4− − ψ4− (µ) x12 + x22 − x3 (1 − x3 )
i
+ −ξ4+ + φ4+ + ψ4+ (µ)(1 − x3 − 10x1x2 ) ,

T2 (xi , µ) = 24x1x2 ξ40 (µ) + ξ4+ (µ)(1 − 5x3) ,
h
S1 (xi , µ) = 6x3 (x2 − x1 ) ξ40 + φ40 + ψ40 + ξ4+ + φ4+ + ψ4+ (µ)
i
+ ξ4− + φ4− − ψ4− (µ)(1 − 2x3) ,
h
P1 (xi , µ) = 6x3 (x2 − x1 ) ξ40 − φ40 − ψ40 + ξ4+ − φ4+ − ψ4+ (µ)
i
+ ξ4− − φ4− + ψ4− (µ)(1 − 2x3) . (C.4)

V5 (xi , µ) = 6x3 φ50 (µ) + φ5+ (µ)(1 − 2x3 ) ,
A5 (xi , µ) = 6x3 (x2 − x1 )φ5− (µ),
h
V4 (xi , µ) = 3 ψ50 (µ)(1 − x3 ) + ψ5− (µ) 2x1x2 − x3 (1 − x3 )
i
+ ψ5+ (µ) 1 − x3 − 2 x12 + x22 ,

A4 (xi , µ) = 3(x2 − x1 ) −ψ50 (µ) + ψ5− (µ)x3 + ψ5+ (µ)(1 − 2x3 ) ,
3h 0
T4 (xi , µ) = ξ5 + ψ50 + φ50 (µ)(1 − x3 )
2
+ ξ5− + φ5− − ψ5− (µ) 2x1x2 − x3 (1 − x3 )
i
+ ξ5+ + φ5+ + ψ5+ (µ) 1 − x3 − 2 x12 + x22 ,
3h 0
T8 (xi , µ) = ψ5 + φ50 − ξ50 (µ)(1 − x3 )
2
+ φ5− − ψ5− − ξ5− (µ) 2x1x2 − x3 (1 − x3 )
i
+ φ5+ + ψ5+ − ξ5+ (µ) 1 − x3 − 2 x12 + x22 ,

T5 (xi , µ) = 6x3 ξ50 (µ) + ξ5+ (µ)(1 − 2x3 ) ,
3 h
S2 (xi , µ) = (x2 − x1 ) − ψ50 + φ50 + ξ50 (µ) + ξ5− + φ5− − ψ50 (µ)x3
2 i

+ ξ5+ + φ5+ + ψ50 (µ)(1 − 2x3 ) ,
3 h
P2 (xi , µ) = (x2 − x1 ) ψ50 + ψ50 − ξ50 (µ) + ξ5− − φ5− + ψ50 (µ)x3
2 i

+ ξ5+ − φ5+ − ψ50 (µ)(1 − 2x3 ) . (C.5)

V6 (xi , µ) = 2 φ60 (µ) + φ6+ (µ)(1 − 3x3 ) ,
A6 (xi , µ) = 2(x2 − x1 )φ6− ,

1 − +
T6 (xi , µ) = 2 φ6 (µ) + φ6 − φ6 (1 − 3x3 ) .
0
(C.6)
2
References
[1] V.L. Chernyak, A.R. Zhitnitsky, JETP Lett. 25 (1977) 510.

[2] V.L. Chernyak, A.R. Zhitnitsky, Yad. Fiz. 31 (1980) 1053.
[3] A.V. Efremov, A.V. Radyushkin, Phys. Lett. B 94 (1980) 245.
[4] A.V. Efremov, A.V. Radyushkin, Teor. Mat. Fiz. 42 (1980) 147.
[5] G.P. Lepage, S.J. Brodsky, Phys. Lett. B 87 (1979) 359.
[6] G.P. Lepage, S.J. Brodsky, Phys. Rev. D 22 (1980) 2157.
[7] V.L. Chernyak, V.G. Serbo, A.R. Zhitnitsky, JETP Lett. 26 (1977) 594.
[8] V.L. Chernyak, V.G. Serbo, A.R. Zhitnitsky, Sov. J. Nucl. Phys. 31 (1980) 552.
[9] G.P. Lepage, S.J. Brodsky, Phys. Rev. Lett. 43 (1979) 545.
[10] G.P. Lepage, S.J. Brodsky, Phys. Rev. Lett. 43 (1979) 1625, Erratum.
[11] V.A. Avdeenko, V.L. Chernyak, S.A. Korenblit, Yad. Fiz. 33 (1981) 481.
[12] S.J. Brodsky, G.P. Lepage, A.A. Zaidi, Phys. Rev. D 23 (1981) 1152.
[13] S.J. Brodsky, G.P. Lepage, Phys. Rev. D 24 (1981) 2848.
[14] A.I. Milshtein, V.S. Fadin, Yad. Fiz. 35 (1982) 1603.
[15] V.L. Chernyak, A.R. Zhitnitsky, Phys. Rep. 112 (1984) 173.
[16] S.J. Brodsky, G.P. Lepage, in: A.H. Mueller (Ed.), Perturbative Quantum Chromodynamics,
World Scientific, Singapore, 1989, p. 93.
[17] G. Sterman, P. Stoler, Ann. Rev. Nucl. Part. Sci. 47 (1997) 193.
[18] J. Gronberg et al., Phys. Rev. D 57 (1998) 33.
[19] D. Ashery, For the E791 Collaboration, hep-ex/9910024.
[20] V.M. Braun, I.E. Filyanov, Z. Phys. C 48 (1990) 239.
[21] P. Ball, JHEP 9901 (1999) 010.
[22] P. Ball, V.M. Braun, Y. Koike, K. Tanaka, Nucl. Phys. B 529 (1998) 323.
[23] P. Ball, V.M. Braun, Nucl. Phys. B 543 (1999) 201.
[24] A.V. Radyushkin, Nucl. Phys. A 532 (1991) 141.
[25] V.M. Belyaev, A.V. Radyushkin, Phys. Rev. D 53 (1996) 6509.
[26] P. Kroll, M. Schurmann, P.A. Guichon, Nucl. Phys. A 598 (1996) 435.
[27] A.V. Radyushkin, Phys. Rev. D 58 (1998) 114008.
[28] M. Diehl, T. Feldmann, R. Jakob, P. Kroll, Eur. Phys. J. C 8 (1999) 409.
[29] P. Kroll, Nucl. Phys. A 666/667 (2000) 3.
[30] V.L. Chernyak, I.R. Zhitnitsky, Nucl. Phys. B 246 (1984) 52.
[31] I.D. King, C.T. Sachrajda, Nucl. Phys. B 279 (1987) 785.
[32] V.L. Chernyak, A.A. Ogloblin, I.R. Zhitnitsky, Sov. J. Nucl. Phys. 48 (1988) 536.
[33] V.L. Chernyak, A.A. Ogloblin, I.R. Zhitnitsky, Z. Phys. C 42 (1989) 583.
[34] B.L. Ioffe, Nucl. Phys. B 188 (1981) 317.
[35] Y. Chung, H.G. Dosch, M. Kremer, D. Schall, Nucl. Phys. B 197 (1982) 55.
[36] A.V. Kolesnichenko, Yad. Fiz. 39 (1984) 1527.
[37] J.B. Kogut, D.E. Soper, Phys. Rev. D 1 (1970) 2901.
[38] V.M. Braun, S.E. Derkachov, G.P. Korchemsky, A.N. Manashov, Nucl. Phys. B 553 (1999) 355.
[39] G. Mack, A. Salam, Ann. Phys. 53 (1969) 174.
[40] S.J. Brodsky et al., Phys. Lett. B 91 (1980) 239.
[41] S.J. Brodsky et al., Phys. Rev. D 33 (1986) 1881.
[42] Yu.M. Makeenko, Sov. J. Nucl. Phys. 33 (1981) 440.
[43] Th. Ohrndorf, Nucl. Phys. B 198 (1982) 26.
[44] I.I. Balitsky, V.M. Braun, Nucl. Phys. B 311 (1989) 541.
[45] D. Müller, Phys. Rev. D 49 (1994) 2525.
[48] A.V. Belitsky, D. Müller, Nucl. Phys. B 537 (1999) 397.
[49] G.P. Lepage, S.J. Brodsky, Phys. Rev. Lett. 43 (1979) 545.
[50] G.P. Lepage, S.J. Brodsky, Phys. Rev. Lett. 43 (1979) 1625, Erratum.
[51] M.E. Peskin, Phys. Lett. 88B (1979) 128.
[52] K. Tesima, Nucl. Phys. B 202 (1982) 523.
[53] S.-L. Nyeo, Z. Phys. C 54 (1992) 615.
[54] A.A. Pivovarov, L.R. Surguladze, Nucl. Phys. B 360 (1991) 97.
[55] D.B. Leinweber, Phys. Rev. D 51 (1995) 6383.
[56] I.I. Balitsky, V.M. Braun, A.V. Kolesnichenko, Nucl. Phys. B 312 (1989) 509.
[57] V.M. Braun, I.E. Filyanov, Z. Phys. C 44 (1989) 157.
[58] V.L. Chernyak, I.R. Zhitnitsky, Nucl. Phys. B 345 (1990) 137.
[59] P. Ball, V.M. Braun, Phys. Rev. D 55 (1997) 5561.
[60] P. Ball, V.M. Braun, Phys. Rev. D 58 (1998) 094016.
[61] V.M. Braun, A. Khodjamirian, M. Maul, Phys. Rev. D 61 (2000) 073004.
Monopole Chern–Simons term: charge–monopole

system as a particle with spin
Mikhail S. Plyushchay a,b
a Departamento de Física, Universidad de Santiago de Chile, Casilla 307, Santiago 2, Chile
b Institute for High Energy Physics, Protvino, Russia
Received 17 April 2000; revised 26 June 2000; accepted 11 August 2000
Abstract
The topological nature of Chern–Simons term describing the interaction of a charge with magnetic
monopole is manifested in two ways: it changes the plane dynamical geometry of a free particle
for the cone dynamical geometry without distorting the free (geodesic) character of the motion,
and in the limit of zero charge’s mass it describes a spin system. This observation allows us to
interpret the charge–monopole system alternatively as a free particle of fixed spin with translational
and spin degrees of freedom interacting via the helicity constraint, or as a symmetric spinning top
with dynamical moment of inertia and “isospin” U(1) gauge symmetry, or as a system with higher
derivatives. The last interpretation is used to get the twistor formulation of the system. We show that
the reparametrization and scale invariant monopole Chern–Simons term supplied with the kinetic
term of the same invariance gives rise to the alternative description for the spin, which is related to the
charge–monopole system in a spherical geometry. The relationship between the charge–monopole
system and (2 + 1)-dimensional anyon is discussed in the light of the obtained results.  2000
PACS: 45.50.-j; 03.65.-w; 11.15.-q; 04.60.Kz; 05.30.Pr; 02.40.-k
1. Introduction
The Dirac charge–monopole system [1] was one of the first models with which the
importance of topology for physics was realized. The model was investigated in various
aspects classically and quantum mechanically [2–18], and the Dirac’s quantization of the
charge–monopole constant was understood naturally in terms of the underlying topological
fibre bundle structure [6–8,12–16]. In addition, an interesting analysis of the Dirac
quantization condition relies on 3-cocycles [17,18]. Another very seminal interplay of
physics and topology is related to the Chern–Simons (CS) “effects” [19–36], whose one
E-mail address: mplyushc@lauca.usach.cl (M.S. Plyushchay).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 3 0 - 7
414 M.S. Plyushchay / Nuclear Physics B 589 (2000) 413–439
of the basic physical aspects established by Deser, Jackiw and Templeton [19,20] consists
in quantization of coupling in field theory analogous to Dirac quantization in quantum
mechanics.
The spin particle nature of the charge–monopole system was observed by Jackiw,
Rebbi, Hasenfrantz and ’t Hooft in the context of the “spin from isospin” field theoretical
mechanism [2,3], but the first results on the nature of this system anticipating its
interpretation as a particle with spin were obtained by Poincaré (see Ref. [9]). In this
paper we exploit the CS nature of the charge–monopole coupling term to interpret the
charge–monopole system as a particle with spin from the opposite side, not appealing to
its field theoretical origin but treating it classically and quantum mechanically as a system
with finite number of degrees of freedom. More specifically, we observe that the charge–
monopole coupling term having a nature of (0 + 1)-dimensional CS term manifests its
topological nature in two ways. First, the CS term changes the plane dynamical geometry
of a free particle for the cone dynamical geometry of a charged particle without distorting
the free (geodesic) character of the motion. Second, in the limit of zero charge’s mass
the CS term describes a spin system. This observation allows us to interpret the charge–
monopole system alternatively:
(i) As a free particle of fixed spin with translational and spin degrees of freedom
interacting via the helicity constraint; in such interpretation, noncommuting
gauge-covariant momenta (velocities) are the analogs of the Foldy–Wouthuysen
coordinates of the Dirac particle;
(ii) As a symmetric spinning top with dynamical moment of inertia and “isospin” U(1)
gauge symmetry;
(iii) As a system with higher derivatives; this interpretation is the charge–monopole
analog of another known classical equivalence [37] between the relativistic scalar
massive particle in a background of the constant homogeneous electromagnetic
field and higher derivative model of relativistic particle with torsion [38] underlying
(2 + 1)-dimensional anyons [39–45];
(iv) As an analog of the massless particle with nonzero spin, that naturally leads to the
twistor formulation for the charge–monopole system.
We also show that the reparametrization and scale invariant CS term supplied with the
kinetic term of the same invariance gives rise to the alternative description for the spin.
The partially gauge fixed version of such spin model corresponds to the charge–monopole
system in a spherical geometry. The obtained results are used to discuss the relationship
between the charge–monopole system and (2 + 1)-dimensional anyon.
The paper is organized as follows. Section 2 is devoted to discussion of the classical
theory of the charge–monopole system in the context of its similarity to the 3D free
particle. Here we observe the geodesic character of the charge’s motion and demonstrate
that being reduced to the fixed level of the angular momentum integral, it is described by
the Lagrangian corresponding to the 2D non-relativistic particle in the planar gravitational
field of a point-like source [46–48] carrying simultaneously a nontrivial magnetic flux.
This, in particular, explains why the charge–monopole system and 2D charged particle
moving in the field of the point magnetic vortex have the same “hidden” or “dynamical”
M.S. Plyushchay / Nuclear Physics B 589 (2000) 413–439 415
SO(2, 1) symmetry revealed by Jackiw [10,49]. We obtain the general solution to the
canonical equations of motion, compare the charge–monopole system with the 3D free
particle from the point of view of integrals of motion and their Lie–Poisson algebras, and,
finally, interpret the charge–monopole system as a reduced E(3) system. In Section 3 we
discuss the classical and quantum theory of the spin represented in the form of the (0 + 1)-
dimensional topological theory given by the CS charge–monopole action corresponding to
the limit of zero charge’s mass. In Section 4 we construct the description of the charge–
monopole system as a particle with spin. We start with the free particle system with spin,
whose value is fixed by the charge–monopole constant. Then we switch on interaction
between translational and spin degrees of freedom by introducing the helicity constraint
which freezes spin degrees of freedom and provides finally the physical equivalence of
the extended model to the initial charge–monopole system. In Section 5 the system is
interpreted as a spinning symmetric top. In this picture the initial U(1) electromagnetic
gauge symmetry is changed for the U(1) gauge symmetry generated by the “isospin”-fixing
constraint, which makes the rotations about the top’s symmetry axis to be unobservable.
The spinning top picture is used in Section 6 to get the higher derivative form for the CS
term, which, in turn, is employed in Section 7 for constructing the twistor formulation
for the charge–monopole system proceeding from its analogy to the 4D massless particle
with spin. In Section 8 we discuss the charge–monopole system in a spherical geometry
and find the alternative formulation for the spin. Here we also observe that the SO(2, 1)
symmetry of the charge–monopole system [10,49] can be treated formally as a relic of
the reparametrization invariance surviving the Lagrangian gauge fixing procedure applied
to the Euclidean relativistic version of the model. Section 9 contains discussion of the
relationship between the charge–monopole system and (2 + 1)-dimensional anyon, and in
the last section we present some concluding remarks.
2. Charge–monopole dynamics and 3D free particle
2.1. Lagrangian formalism
A non-relativistic particle of unit mass and electric charge e in the field of magnetic
monopole of charge g is described by the Lagrangian
1
L = ṙ2 + eAṙ, (2.1)
2
with a U(1) gauge potential A(r) defined by the relations
ri p
∂i Aj − ∂j Ai = Fij = ij k Bk , Bi = g 3 , r = r2 . (2.2)
r
The case of arbitrary mass m can be obtained from (2.2) via the transformation of time and
charge: t → m−1/2 t, e → m1/2 e. In definition (2.2) it is assumed that the point r = 0 is
excluded, i.e., the configuration space of the system, M = R3 − {0}, is diffeomorphic to
(0, ∞) × S 2 and inherits a nontrivial topology of a two-sphere S 2 . It is well known that the
electromagnetic potential (2.2) gives a connection of the monopole U(1) fibre bundle being
a nontrivial Hopf bundle over S 2 , and the problems with Dirac strings of singularity can be
escaped by covering the configuration space M with two charts [34,50]. These topological
complications, however, do not play any role for us under treating the classical dynamics
of the system, but will reveal themselves at the quantum level.
Equations of motion following from Lagrangian (2.1) result in the Lorentz force law,
ν
r̈ = − 3 (r × ṙ), ν = eg, (2.3)
r
which implies that instead of the orbital angular momentum vector L = r × ṙ, the vector
J = L − νn, n = r r −1 , (2.4)
is the conserved angular momentum of the system. The nontrivial term −νn can be
understood as the electromagnetic angular momentum p produced by both the electric charge
and magnetic monopole (see Ref. [9]). Note that J = J2 is restricted from below by the
modulus of the charge–monopole coupling constant ν, J > |ν|. Due to the relation Jn =
−ν, the trajectory of the particle lies on the cone. The cone’s axis is given by the vector J
and its half-angle is
cos γ = −νJ −1 . (2.5)
Since the force f = −νr −3 (r × ṙ) is orthogonal to r and to the velocity ṙ, it is perpendicular
to the cone. Therefore, the particle performs a free motion on the cone. This can also
be observed directly as follows. Taking into account the conservation of the angular
momentum J and that in spherical coordinates the charge–monopole coupling term takes
a form Lint = ν cos ϑ ϕ̇ (see Eq. (3.9)), we can choose the system of coordinates with axis
ϑ = 0 directed along the vector J. Then in accordance with Eq. (2.5), cos ϑ = −νJ −1 ,
ϑ̇ = 0, and Lagrangian (2.1) is reduced to L = 12 (ṙ 2 + (1 − ν 2 J −2 )r 2 ϕ̇ 2 ) − ν 2 J −1 ϕ̇. After
transformation
p
t → t 0 = αt, α = 1 − ν 2 J −2 , (2.6)
we obtain finally the following form for the charge–monopole Lagrangian:
1 −2 2
L= α ṙ + r 2 ϕ̇ 2 − αν 2 J −1 ϕ̇. (2.7)
2
The last term is the total time derivative of the topologically nontrivial angular variable. It
is the reduced form of the charge–monopole interaction term having a nature of (0 + 1)-
dimensional CS term [29–33]. As we shall see, quantum mechanically this will give rise
to the Dirac quantization condition for ν, but classically the total derivative can be omitted
without changing the equations of motion. The topologically nontrivial term corresponds
exactly to the 2D term describing the interaction of the charge e with a (singular) point
vortex carrying the magnetic flux [49] Φ = −2παν 2 J −1 e−1 . Without the last term, Eq.
√ particle of unit mass on the cone given by the relations x =
(2.7) is a Lagrangian of a free
r cos ϕ, y = r sin ϕ, z = r α−2 − 1, r > 0, 0 6 ϕ 6 2π , with 0 < α < 1, that confirms
our statement on a free (geodesic) motion of the charge over the cone (2.5). Together with
the pointed out nature of the total derivative term, this, as we shall see, explains why the
charge–monopole system and 2D charged particle moving in the field of the point magnetic
vortex have the same dynamical SO(2, 1) symmetry [10,49].
Since the conical metric ds 2 = α−2 (dr)2 + r 2 (dϕ)2 corresponds to the metric produced
by the point mass [46–48,51], we can say that the classical motion of the charge in the
field of magnetic monopole (reduced to the fixed value of the integral J) is equivalent to
the classical motion of a particle in a planar gravitational field of a point massive source
carrying simultaneously magnetic flux Φ = −2παν 2 J −1 e−1 .
From the form of transformation (2.6) and Lagrangian (2.7) it is clear that the case J =
|ν| is singular and should be treated as a limit case, i.e., we have to assume that J > |ν|. We
shall discuss this peculiarity of the charge–monopole system in different aspects in what
follows.
In the case of a 3D free particle (ν = 0, M = R3 ), the motion is characterized by the
coordinate r and by the conserved linear momentum p. Alternatively, with the appropriate
choice of the origin of the system of coordinates, the particle’s motion can be characterized
by the unit vector n and by the conserved orbital angular momentum L supplemented with
the canonically conjugate scalars r and pr = pr r −1 . Since Ln = 0, for a given L the
particle’s trajectory is in the plane orthogonal to the orbital angular momentum. So, we
conclude that classically the topological nature of (0 + 1)-dimensional charge–monopole
CS term is manifested in changing the global structure of the dynamics without distorting
its local free (geodesic) character: the “plane dynamical geometry” of the free particle
(e = 0) is changed for the free “cone dynamical geometry” of the charged particle. We
shall discuss the relation between the two systems in the context of (dynamical) integrals
of motion in Section 2.3.
2.2. Canonical formalism: solutions to the equations of motion
To solve the equations of motion in general form and analyze in more detail the
system’s dynamics, we turn to the canonical formalism. The Hamiltonian corresponding
to Lagrangian (2.2) is H = 12 P2 , where P = p − eA is a classical analog of the gauge-
covariant derivative defined via the momentum p canonically conjugate to r. This gives
rise to the Poisson brackets
ν
{Pi , Pj } = 3 ij k rk , {ri , Pj } = δij , {ri , rj } = 0, (2.8)
r
and to the equations of motion
ν
ṙ = P, Ṗ = − L, (2.9)
r3
with L = r × P. From Eq. (2.9) we find that the vectors n and L evolve according to
the equations L̇ = νr −2 (L × n), ṅ = r −2 (L × n), and, as a consequence, precess about
the integral J = L − νn with the same frequency: L̇ = r −2 (J × L), ṅ = r −2 (J × n).
The vectors J and n obeying the relations n2 = 1 and Jn = −ν, together with the two
scalars r2 and Pr form the complete set of observables in terms of which r and P can be
completely “restored”. The equations of motion for the scalar observables have a simple
form (r2 )˙= 2Pr, (Pr)˙= P2 = 2H , being exactly the same as in the case of a free particle.
Their integration gives
(Pr)(t) = (Pr)(t0 ) + P2 · (t − t0 ),
r2 (t) = P2 · (t − t0 )2 + 2(Pr)(t0 ) · (t − t0 ) + r2 (t0 ). (2.10)
The minimal charge–monopole distance corresponds to the moment of time for which
Pr = 0, and is given by the relation r2min = L2 P−2 . To complete the integration of equations
of motion, we solve the equation ṅ = r −2 (J × n) and get
n(t) = −νJ −1 j + n⊥ (t),

n⊥ (t) = n(t0 ) + jνJ −1 cos τ (t) + j × n(t0 ) sin τ (t), (2.11)
p
τ = J L−1 tan−1 Pr L−1 , L = J2 − ν 2 , (2.12)
where j = JJ −1 , and evolution of Pr is given by Eq. (2.10). Zero value of the angular
function τ (t) corresponds to the point of perihelion (r = rmin ) with respect to which the
trajectory is symmetric. The vector n⊥ is the projection of n to the plane orthogonal to
the angular momentum J, and Eqs. (2.11), (2.12) give the classical scattering angle of the
particle’s motion projected into the plane perpendicular to J as a function of it:
⊥
ϕscat = τ (+∞) − τ (−∞) = πJ L−1 . (2.13)
Eq. (2.11) together with Eq. (2.10) and relation P = ṙ give a complete solution to the
equations of motion (2.9).
In correspondence with Eqs. (2.5), (2.13), in the limit J → ∞ the cone over which the
particle moves is close to the plane: γ → π/2 and ϕscat ⊥ → π . In another limit, J → |ν|,
the cone is degenerated into a half-line, γ → 0(π) for ν < 0 (> 0), whereas the number of
⊥ /2π] is infinite, where [·] is the integer part. Therefore, the case
full rotations N = [ϕscat
J = |ν| (L = 0) with P 6= 0, which corresponds to the motion of the charge with constant
velocity along a straight half-line defined by n(t0 ) in the direction to or from the monopole
is not a proper limit: such a trajectory corresponds to the half of the trajectory with J →
∞ (L → ∞) but not to the limit J → |ν|. This supports our conclusion of the previous
subsection on a necessity to treat the values of the angular momentum to be confined to the
domain |ν| < J < ∞.
2.3. Integrals of motion and their algebra
Let us compare the charge–monopole system with a 3D free particle from the point of
view of integrals of motion and corresponding Lie–Poisson algebras formed by them.
For the 3D free particle (ν = 0), the first integrals of motion are linear, p, and angular,
L = r × p, momenta forming the set of 5 (pL = 0) algebraically independent conserved
quantities not depending explicitly on time. With respect to the Poisson brackets, they form
the algebra of Euclidean group E(3). Algebraically, the Hamiltonian is a square of the
vector p, H = 12 p2 , but it generates the symmetry of time translations being independent
from the symmetries of space translations and rotations generated by p and L. This is
not the only independent symmetry which can be generated via constructing algebraically
dependent quantities from L and p. Indeed, the vector Q = p × L is the analog of the
Laplace–Runge–Lenz vector of the Coulomb–Kepler system, which can be treated as a
generator of the corresponding canonical symmetry transformations. The integrals L and
Q, LQ = 0, form together with the Hamiltonian the nonlinear algebra:
{Li , Vj } = ij k Vk , {Qi , Qj } = −2H ij k Lk ,
{H, Vi } = 0, Vi = Li , Qi .
p
The renormalized (at p 6= 0) vector R ≡ Q/ p2 and the angular momentum vector L
form √the Lorentz algebra so(3, 1). One can construct another vector, K ≡ (L × R)L−1 ,
L = L2 , which together with R and L provides us with the complete set of the three
orthogonal vectors, RL = KL = RK = 0, of the same norm, R2 = K2 = L2 . They form
the following nonlinear algebra:
{Li , Vj } = ij k Vk , {Ri , Rj } = {Ki , Kj } = −ij k Lk ,
{Ri , Kj } = δij L, (2.14)
where Vi = Li , Ri , Ki . One notes that the scalar L rotates the set of vectors R and K:
{L, Ri } = −Ki , {L, Ki } = Ri . (2.15)
Therefore, the vector integral K also forms the so(3, 1) algebra with the orbital angular
momentum and is (non-canonically) conjugate to R (see the last Poisson bracket relation in
Eq. (2.14)). The rotated about L vectors R0 = R cos ϕ + K sin ϕ and K0 = K cos ϕ − R sin ϕ,
ϕ = const, possess exactly the same set of properties as R and K.
There are also the so called dynamical integrals of motion [10] depending explicitly on
time and they appear as follows. Solving the equations of motion ṙ = p, ṗ = 0, one gets
r = p · (t − t0 ) + X, X ≡ r(t0 ). One can treat X = r − p · (t − t0 ) as a vector integral of
motion dependent explicitly on time:
d ∂X
X = {X, H } + = 0.
dt ∂t
It generates the transformations xi → xi − εi (t − t0 ) corresponding to the Galilei boosts.
The integrals of motion p and X satisfy the Heisenberg algebra {Xi , Xj } = {pi , pj } = 0,
{Xi , pj } = 1 · δij . Algebraically, the vector X is equivalent to the vector integral not
containing the explicit dependence on time, X⊥ = r − p(pr) · p−2 = Q · p−2 , X⊥ p = 0, and
to the scalar D = Xp = rp − p2 · (t − t0 ) being a dynamical integral of motion generating
the time dilations [10,49]. One notes that the angular momentum L can also be treated as
an integral algebraically dependent on X and p: L = X × p = X⊥ × p. On the other hand,
from the equations of motion it follows that
d 2 d
r = 2pr, (pr) = p2 .
dt dt
Integration of the second equation gives rise to the dynamical integral D and the subsequent
integration of the first equation gives, similar to Eq. (2.10),
r2 (t) = p2 · (t − t0 )2 + 2D · (t − t0 ) + r2 (t0 ).
The last relation can be rewritten in the form of the dynamical integral of motion
R = r2 (t0 ) = r2 − p2 · (t − t0 )2 − 2D · (t − t0 ),
which generates the time special conformal transformations [10,49]. It can be represented
equivalently as

R = L2 + D2 p−2 . (2.16)
The integrals D, 2H and R form the same algebra as the scalars pr, p2 and r2 (the latter set
is reduced to these integrals at the initial moment t = t0 ), which is the so(2, 1) ∼ sl(2, R)
algebra [52–57]:
{J0 , J1 } = J2 , {J0 , J2 } = −J1 , {J1 , J2 } = −J0 , (2.17)
where J0 = 14 (2H + R), J1 = 14 (2H − R), J2 = 12 D, with the Casimir central element
C = −J02 + J12 + J22 = − 14 L2 6 0. The dynamical integral D together with H commutes
in the sense of Poisson brackets with the vector integrals L, R and K, whereas the integral
R due to Eqs. (2.15), (2.16) has nontrivial Poisson bracket relations with R and K.
Let us turn to the charge–monopole system, where instead of the orbital angular
momentum, the vector J = L − νn is conserved. Since the equations of motion for the
scalar variables r2 and Pr look exactly as the corresponding equations for the free particle
(with the change of p for P), the charge–monopole system has the set of the scalar
dynamical integrals of motion of the same form [10],
D = Pr − P2 · (t − t0 ),
R = r2 − P2 · (t − t0 )2 − 2D · (t − t0 ). (2.18)
Though Pi are characterized by the nontrivial Poisson brackets (2.8), the scalar dynamical
integrals (2.18) and the Hamiltonian generate the same so(2, 1) algebra (2.17) as in a free
case with the Casimir central element
1
C = −J02 + J12 + J22 = − J2 − ν 2 < 0. (2.19)
4
The free particle’s vector integrals of motion R and K also have analogs in the charge–
monopole system. These are given by the vectors
J2 J
R ≡ N√ , K ≡ J × N√ , (2.20)
J2 − ν2 J − ν2
2
where the vector integral N = (n + νJ −1 j) cos τ − (j × n) sin τ can be identified with the
vector n⊥ (t0 ) (see Eq. (2.11)). Vector N satisfies the relations NJ = 0, N2 = 1 − ν 2 J −2 ,
and {Ni , Nj } = −ν 2 J −4 ij k Jk , {Ni , r2 } = {Ni , Pr} = {Ni , P2 } = 0. From these relations
we find that RJ = KJ = RK = 0, R2 = K2 = J2 , and that the Poisson bracket algebra of
the complete set of orthogonal vectors J, R and K is given by Eq. (2.14) with L changed
for J. This set of vector integrals is in involution with the dynamical scalar integral of
motion D and with H , but the vectors R and K have nontrivial Poisson bracket relations
with the dynamical integral R = (D2 + J2 − ν 2 )/(2H ) (see Eq. (2.15)).
Therefore, the dynamical geometry similarity of the charge–monopole system to the

3D free particle discussed in Section 2.1 also reveals itself in existence of similar sets of
integrals of motion (depending and not depending explicitly on time), which form between
themselves the same (nonlinear) Lie–Poisson algebras.
2.4. Charge–monopole as a reduced E(3) system
Like a 3D free particle, the charge–monopole system may be treated as a reduced E(3)
system. To get such an interpretation, let us pass over from the Hamiltonian variables r and
P to the set of variables n, J, r and Pr = Pr r −1 . They have the following Poisson brackets:
{r, Pr } = 1, {r, n} = {r, J} = {Pr , n} = {Pr , J} = 0, (2.21)

{Ji , Jj } = ij k Jk , {Ji , nj } = ij k nk , {ni , nj } = 0. (2.22)
Poisson brackets (2.22) correspond to the algebra of generators of the Euclidean group
E(3) with Ji being a set of generators of rotations and ni identified as generators of
translations. The quantities n2 and Jn lying in the center of e(3) algebra, {n2 , ni } =
{n2 , Ji } = {nJ, ni } = {nJ, Ji } = 0, are fixed in the present case by the relations
n2 = 1, nJ = −ν. (2.23)
In terms of the introduced variables, the Hamiltonian of the system takes the form
1 (J × n)2
H = Pr2 + . (2.24)
2 2r 2
Therefore, the charge–monopole system can be treated as the E(3) system reduced by
the conditions (2.23) fixing the Casimir elements and supplemented by the independent
canonically conjugate variables r and Pr . It is the second relation from Eq. (2.23) that
encodes the topological difference between the charge–monopole and the 3D free particle
cases: for ν 6= 0, the space given by the spin vector J is homeomorphic to R3 − {0}, (J >
|ν|), whereas for ν = 0 the corresponding space R3 is topologically trivial. In the next
section we shall discuss the physical consequences of the nontrivial topological structure
of the charge–monopole system.
In accordance with Eqs. (2.23), one could treat the vector L = J + νn, Ln = 0, as an
orbital angular momentum. However, the Poisson bracket relations {Li , Lj } = ij k (Lk +
νnk ) following from (2.22) and restriction L2 > 0 corresponding to J > |ν| prevent such
interpretation. Nevertheless, as we shall see, it is possible to treat L as an orbital angular
momentum in extended physically equivalent formulation of the model.
3. Monopole Chern–Simons term and spin
This section contains mainly the known results which are necessary for the self-
contained presentation of the subsequent analysis.
The integrand in action corresponding to the the charge–monopole interaction term, θ =

eṙA(r) dt, can be treated as a differential one-form, θ = eA(r) dr. Then the relations (2.2)
defining the monopole vector potential are equivalent to the relation
ν
dθ = 3 ij k ri drj ∧ drk . (3.1)
2r
The right-hand side of Eq. (3.1) is the gauge-invariant curvature two-form,
1
dθ = eF, F = Fij dri ∧ drj (3.2)
2
and, consequently, the gauge-non-invariant one-form θ has a sense of (0 + 1)-dimensional
CS term [29–34]. The two-form (3.1) can be represented equivalently as
ν
dθ = ij k ni dnj ∧ dnk , ni = ri r −1 . (3.3)
2
Via the (local) parametrization by the spherical angles, n = n(ϑ, ϕ), this can be treated as
the differential area of a two-sphere multiplied by the charge–monopole coupling constant:
dθ = ν d(cos ϑ) ∧ dϕ. If the vector n(t) describesH some Rclosed curve R Γ on the sphere
(i.e., if n(t1 ) = n(t2 )) 1 , the Stokes theorem gives Γ θ = S+ dθ = − S− dθ , where Γ =
∂S+ = −∂S− , S+ ∪ S− = S 2 . Within the path-integral quantization scheme, the alternative
representations for the same charge–monopole interaction term in the action can differ only
in 2πn, n ∈ Z,
Z Z ! Z
dθ − − dθ = ν d cos ϑ ∧ dϕ = 4πν = 2πn, (3.4)
S+ S− S2
and we arrive at the Dirac quantization condition for the charge–monopole coupling
constant: 2ν = n ∈ Z. Defining the dependent variables si = −νni , one can represent (3.3)
equivalently as
1
ωspin = dθ = − ij k si dsj ∧ dsk , si si = ν 2 . (3.5)
2s2
The two-form (3.5) is closed and nondegenerate, and can be treated as a symplectic form
corresponding to the symplectic potential θ . If we drop out the kinetic term in the charge–
monopole action (that corresponds to taking the charge’s zero mass limit, m → 0, [30]),
we get the CS action
Zt2
S= θ, (3.6)
t1
describing the spin system. Indeed, for any function f on the sphere S 2 , the symplec-
tic form (3.5) defines the Hamiltonian vector field Xf on the cotangent bundle T ∗ S 2 :
iXf ω(Y ) = ω(Xf , Y ) = df (Y ). This gives Xf = ∂a f ωab ∂/∂x b , and defines the corre-
sponding Poisson brackets, {f, g} = −ω(Xf , Xg ). Here x a , a = 1, 2, are the local coordi-
nates on S 2 and ωab are the elements of the matrix inverse to the symplectic matrix ωab ,
1 According to Eqs. (2.11), (2.12) and relation (Pr)˙= P2 = const, such a closed curve is smooth.
ωspin = 12 ωab dx a ∧ dx b . Taking into account Eq. (3.5), we get {si , sk } = ij k sk , si si = ν 2 .
These relations define the classical spin system with fixed spin modulus. Geometric quan-
tization [58–62] applied to such a system (for the details see Refs. [41,42]) leads to the
same Dirac quantization of the parameter ν, |ν| = j , j = 1/2, 1, 3/2, . . ., and results in
(2j + 1)-dimensional representation of su(2),
1 − z2 d
s1 = + j z,
2 dz
1 + z2 d d
s2 = i − ij z, s3 = z − j, (3.7)
2 dz dz
realized on the space of holomorphic functions with the basis ψjk ∝ zj +k , k = −j, −j +
1, . . . , +j , s3 ψjk = kψjk , and scalar product
ZZ
2j + 1 ψ1 (z)ψ2 (z) 2
(ψ1 , ψ2 ) = d z.
π (1 + |z|2 )2j +2
The classical relation s2 = ν 2 is changed for the quantum relation s2 = j (j + 1). Here
the complex variable z is related to the spherical angles via the stereographic projection
z = tan(ϑ/2) eiϕ from the north pole, or via z = cot(θ/2)e−iϕ for the projection from the
south pole. In both cases the symplectic two-form is represented as
dz̄ ∧ dz
ωspin = 2iν . (3.8)
(1 + z̄z)2
Geometrically, the obtained spin system is a Kähler manifold [34] with Kähler potential
∂2 ∗
K = 2iν ln(1 + z̄z): ωspin = ∂z∂ z̄ K dz̄ ∧ dz, z̄ = z . Locally, in spherical coordinates the
spin Lagrangian is given by
Lspin = ν cos ϑ ϕ̇, (3.9)
and in terms of global complex variable the Lagrangian (3.9) takes the form
˙
z̄ż − z̄z
Lspin = iν . (3.10)
1 + z̄z
The appearance of the two stereographic projections for the spin system (3.6) reflects
the above mentioned necessity to work in two charts in the case of the initial (m 6= 0)
charge–monopole system to escape the problems with Dirac string singularities. In terms
of globally defined independent variables z, z̄ no gauge invariance left in the spin system
given by the Lagrangian (3.10) but it is hidden in a fibre bundle structure reflected, in
particular, in the presence of two charts [58].
We conclude that the charge–monopole interaction term has a nature of (0 + 1)-
dimensional Abelian CS term, that leads to the Dirac quantization of the charge–monopole
coupling constant, |ν| = j . In the limit of zero charge’s mass the total charge–monopole
Lagrangian is reduced to the Lagrangian (3.10) in terms of independent variables z and
z̄, which describes the spin-j system. In what follows, we consider other possibilities to
describe spin system proceeding from its nature associated with the monopole CS term.
4. Charge–monopole as a particle with spin
The spin nature of the charge–monopole CS term and observed free character of the
charge’s dynamics allow us to get the alternative description for the charge–monopole
system as a free particle of fixed spin with translational and spin degrees of freedom
interacting via the helicity constraint. To find such a description, we forget for the moment
that the spin system has been obtained via the identification si = −νni (in the limit m → 0),
and simply start with its Hamiltonian description given by the symplectic form (3.5) and
corresponding brackets {si , sk } = ij k sk . The canonical Hamiltonian of the spin theory
given by the first order action (3.6) or by the Lagrangian (3.10) is equal to zero. Let us
extend such a pure spin system by adding to it independent translational degrees of freedom
described by the particle’s coordinates ri and canonically conjugate momenta pi , i.e., we
suppose that the corresponding symplectic two-form is
ω = dpi ∧ dri + ωspin . (4.1)
Moreover, let the dynamics of the system is given by the Hamiltonian H = 1 2
2p . Then
this Hamiltonian and relations (3.5), (4.1) specify the nonrelativistic free particle with
internal degrees of freedom describing spin of fixed value. In such a system we have 6 + 2
independent phase space variables instead of 6 variables in the initial charge–monopole
system. We can reduce such an extended system to the initial system by introducing into it
one first class constraint. Having in mind the identification si = −νni , for the initial system
(2.1), let us postulate the helicity constraint
χ ≡ sr + νr ≈ 0. (4.2)
This constraint can be interpreted as a constraint introducing interaction between
translational and spin degrees of freedom. But the constraint (4.2) is not conserved by
the Hamiltonian, 12 {p2 , χ} 6= 0, and for consistency of such a theory we have to modify the
latter. For the purpose, let us find the complete set of gauge-invariant variables commuting
in the sense of Poisson brackets with constraint (4.2). Since we have 6 + 2 phase space
variables and one constraint, there are 6 independent gauge-invariant variables. They are ri
and
1
Πi ≡ pi − 2 ij k rj sk , (4.3)
r
{ri , χ} = 0, {Πi , χ} ≈ 0. Therefore, it is natural to change H = 12 p2 for its gauge-invariant
analog, H = 12 52 , and to take the sum of it and constraint (4.2) multiplied by an arbitrary
function λ = λ(t),
1
H = 52 + λ · (sr + νr), (4.4)
2
as a total Hamiltonian [63]. The Poisson brackets for the gauge-invariant variables are
rs
{Πi , Πj } = − 4 ij k rk , {Πi , rj } = δij , {ri , rj } = 0. (4.5)
r
Taking into account constraint (4.2), the Poisson brackets between Πi and Πj coincide
with the Poisson brackets (2.8) between Pi and Pj . Identifying Πi with Pi , the dynamics
generated by the Hamiltonian (4.4) for the gauge-invariant variables ri , Πi is exactly the
same as the dynamics in the initial charge–monopole system (2.1), and the physical content
of the extended system (4.4) is the same as that of the initial system (2.1). In particular, the
vector
J=r×p+s (4.6)
is the integral of motion, whose components satisfy the Poisson bracket relations {Ji , Jj } =
ij k Jk , and generate rotations. It can be represented equivalently as J = r × 5 + (ns)n,
and on the constraint surface (4.2) takes the form J ≈ r × 5 − νn, which coincides with
(2.4) with identified Πi and Pi . With the help of relation (4.6), the Hamiltonian (4.4) can
be written down equivalently (cf. with Eq. (2.24)),
1 (J × n)2
H = pr2 + + λ(Jn + ν), (4.7)
2 2r 2
where pr = pr r −1 is the momentum canonically conjugate to the radial variable r. Since
Ln = 0, L = r × p, the difference of the present Hamiltonian interpretation of the charge–
monopole system as a particle with spin from the reduced E(3) system from Section 2.4
is that here the helicity is fixed weakly by the constraint (4.2), whereas there it was fixed
strongly by the second relation from Eq. (2.23). As a consequence, here the components
of the angular momentum vector L form the so(3) algebra, {Li , Lj } = ij k Lk , but they,
unlike the total angular momentum vector J, are not physical variables: {Li , χ} 6= 0. In
the present interpretation, we have additional spin variables si restricted by the condition
s2 = ν 2 , but the only physical observable constructed from them is the combination sn,
which is fixed by the helicity constraint. The important comment is in order here. Strictly
speaking, in correspondence with the discussion above on the necessity of restriction J 2 >
ν 2 , we have to suppose that the relation s2 = ν 2 has to be treated only in the sense of the
limit s2 = ν 2 + ε2 , ε → 0. This will not only correspond to the specified restriction on J 2
in accordance with relations (4.2) and (4.6), but is necessary for the consistent treatment
of the extended model as a constrained system. Indeed, in correspondence with general
theory of gauge systems [63], only in this case the constraint (4.2) can be treated as a
good constraint condition, which in the two-dimensional spin phase subspace given by si ,
{si , sj } = ij k sk , s2 = ν 2 + ε2 , specifies one-dimensional physical subspace (on which it
will act transitively).
The Lagrangian corresponding to the described extended system is
1 1
L = ṙ2 − 2 (r × ṙ) · s − λ · (rs + νr) + Lspin , (4.8)
2 r
with Lspin chosen in the form (3.10) and λ treated as a Lagrange multiplier.
We conclude that the system (4.8) describing the nonrelativistic particle of spin ν with
interacting translational and spin degrees of freedom (see the second Ls-coupling term and
the third constraint term in Lagrangian) is classically equivalent to the charge–monopole
system (2.1).
The quantum theory of the charge–monopole system in such interpretation is obvious.
As we have seen, the quantization of the spin variables results in the Dirac condition, ν =
j , = + or −, and gives rise to the corresponding (2j + 1)-dimensional representation of

the su(2) with spin operators (3.7) acting on the space of holomorphic functions. Choosing
the representation diagonal in ri and realizing pj as differential operators, pj = −i∂/∂rj ,
Pj
one can work on the space of functions of the form Ψ j (r, z) = k=−j ψk (r)zj +k . The
quantum analog of the classical constraint takes the form of the quantum condition
j
separating the physical subspace, (sn + j )Ψphys (r, z) = 0. It is necessary to note that
quantum mechanically the above mentioned necessity of the regularization s2 = ν 2 + ε2
is taken into account automatically. Indeed, since the eigenvalues of the operators s2 and
(sn)2 are separated in a necessary way, s2 = j (j + 1), (sn)2 = j 2 , one can say that the
quantization “cures” the classical system.
5. Charge–monopole system as a symmetric top
In this section we show that the charge–monopole system can also be interpreted as a
reduced symmetric spinning top with dynamical tensor of inertia and “isospin” U(1) gauge
symmetry. To this end, we return to the spin symplectic form (3.3), and supplement the unit
vector n ≡ e3 with the two vectors e1 and e2 forming together the oriented orthonormal set
of vectors,
ea eb = δab , e1 × e2 = e 3 . (5.1)
As a consequence of basic relations (5.1), the vectors ea , a = 1, 2, 3, satisfy also the

completeness relation
j
eai ea = δ ij . (5.2)
Using Eqs. (5.1), (5.2), one can represent the two-form (3.3) as
dθ = ν de1 ∧ de2 , (5.3)
from which we get another representation for the one-form,

ν
θ = (e1 de2 − e2 de1 ). (5.4)
2
Taking into account the relation ė23 = (ė1 e3 )2 + (ė2 e3 )2 , we get the alternative Lagrangian
for the charge–monopole system,
1 1 ν
L = ṙ 2 + r 2 (ė1 · e1 × e2 )2 + (ė2 · e1 × e2 )2 + (e1 ė2 − e2 ė1 )
2 2 2
λ1 2 λ2 2
− e −1 − e − 1 − λ12 e1 e2 , (5.5)
2 1 2 2
with λ1 , λ2 and λ12 being Lagrange multipliers, variation over which gives the Lagrangian
constraints e21 − 1 = 0, e22 − 1 = 0 and e1 e2 = 0. With these constraints, the equations of
motion for r = re3 , e3 = e1 × e2 , following from (5.5) coincide with the charge–monopole
Lagrange equations (2.3). To prove the complete equivalence of the model (5.5) to the
initial system (2.1), we pass over to the Hamiltonian formalism. The canonical Hamiltonian
corresponding to Lagrangian (5.5) is
1 1 λ1 2 λ2 2
Hc = pr2 + 2 (J × e3 )2 + e −1 + e − 1 + λ12 e1 e2 , (5.6)
2 2r 2 1 2 2
where e3 = e1 × e2 and the total angular momentum is J = e1 × p1 + e2 × p2 with
p1,2 being the momenta canonically conjugate to e1,2 . Note that in terms of the velocity
phase space variables the total angular momentum is reduced to J = r 2 (e3 × ė3 ) − νe3 in
correspondence with Eqs. (2.4). The application of Dirac–Bergmann theory to the system
(5.5) results in the following complete set of constraints:
π1 ≈ 0, π2 ≈ 0, π12 ≈ 0, (5.7)
e21 − 1 ≈ 0, p1 e1 ≈ 0, e22 − 1 ≈ 0,
p2 e2 ≈ 0, e1 e2 ≈ 0, p1 e2 + p2 e1 ≈ 0, (5.8)
p1 e2 − p2 e1 + ν ≈ 0, (5.9)
with the momenta π1 , π2 and π12 canonically conjugate to Lagrange multipliers.
Constraints (5.7) mean that the Lagrange multipliers are pure gauge variables and can
be completely excluded, e.g., by supplementing (5.7) with the gauge conditions λ1 ≈ 0,
λ2 ≈ 0, λ12 ≈ 0. The constraints (5.8) form the subset of second class constraints, whereas
(5.9) is the first class constraint. Reduction of the symplectic two-form of the system, ω =
dp1 ∧ de1 + dp2 ∧ de2 + dpr ∧ dr, to the surface of second class constraints (5.8) is given
by
1
ω = d(J × ea ) ∧ dea + dpr ∧ dr, (5.10)
2
where the summation over a = 1, 2, 3 is assumed. The two-form (5.10) is the symplectic
form on the reduced phase space which is described by the basis of vectors ea , a = 1, 2, 3,
subject to conditions (5.1) as strong relations, by the angular momentum vector J and by
the canonically conjugate radial variables r and pr . From (5.10) we obtain the following
Poisson–Dirac brackets on the reduced phase space:
i j i j
ea , eb = 0, J , ea = ij k eak ,
i j
J , J = ij k J k , {r, pr } = 1, (5.11)
and all other brackets for radial variables r and pr to be equal to zero. The remaining
constraint (5.9) takes the form
χ = Je3 + ν ≈ 0, (5.12)
and the total Hamiltonian of the system is reduced to the canonical Hamiltonian extended
by the first class constraint multiplied by an arbitrary function λ(t):
1 1
H = pr2 + 2 (J × e3 )2 + λ · (Je3 + ν). (5.13)
2 2r
To understand the physical sense of the obtained system, let us define the scalar quantities
Ia ≡ −Jea , Ia Ia = J2 . They satisfy the following Poisson bracket relations:

{Ia , Ib } = abc Ic , Ia , ebi = abc eci , Ia , J i = 0. (5.14)
Since Ia commute with Ji and satisfy su(2) algebra, one can treat them as components
of the isospin vector. On the other hand, due to the basic relations (5.1), the quantities eai
can be treated as the elements of the group SO(3). Then the quantities J i have a sense
of the basis of the left-invariant vector fields on this group whereas the quantities −Ia
can be identified with the basis of the right-invariant vector fields [34,64]. Due to the
relations (5.14), the first class constraint (5.12) generates gauge transformations of the
system which have a sense of SO(2) ∼ U (1) isospin rotations generated by I3 . Under
such transformations the vectors J and e3 are invariant. The electromagnetic U(1) gauge
invariance of the curvature form (3.2) corresponding to the CS charge–monopole coupling
term is changed here for the described SO(2) ∼ U (1) gauge invariance of the two-form
(5.3) corresponding to the CS form (5.4). Since (J × e3 )2 = I12 + I22 , the system given by
the Hamiltonian (5.13) and brackets (5.11) can be identified as a symmetric spinning top
with the symmetry axis given by e3 and dynamical moment of inertia I1 = I2 = r 2 . The
isospin U(1) gauge symmetry generated by the constraint (5.12) means that the rotation
about the symmetry axis e3 is of pure gauge nature and, so, is not observable.
The formal difference of the symmetric spinning top system from the reduced E(3)
system discussed in Section 2.4 is that here the condition Jn = Je3 = −ν appears in
the form of first class constraint (weak equality) unlike the strong equality in the case
of the reduced E(3) system. The physical content of both systems is exactly the same.
Indeed, besides the radial variables r and pr , the only observable quantities (commuting
in the sense of Dirac–Poisson brackets with the constraint (5.12)) are the vector of angular
momentum J and the unit vector e3 ≡ n. So, the E(3) system can also be treated as a
symmetric spinning top reduced by the action of the first class constraint (5.12).
The system is quantized as follows. In correspondence with the classical relations (5.1),
(5.2) and the sense of the complete orthonormal set of vectors ea , they can be parametrized
by the three Euler angles α, β, γ , and we can choose a representation diagonal in these
angle variables and in the radial variable r. Then the components of the operator J
are realized in the form of linear differential operators of angular variables [65,66]. An
arbitrary state can be decomposed over the complete basis of Wigner functions,
X j
Ψ (r, α, β, γ ) = ψj,s,k (r)Ds,k (α, β, γ ),
j,s,k
where j takes either integer, j = 0, 1, 2, . . . , or half-integer, j = 1/2, 3/2, . . . , values,

j j j j j
and s, k = −j, −j + 1, . . . , +j , J2 Ds,k = j (j + 1), J3 Ds,k = sDs,k , I3 Ds,k = kDs,k .
The quantum analog of the first class constraint (5.12) is transformed into the equation
separating the physical states:
(I3 − ν)Ψphys = 0. (5.15)
This equation has nontrivial solutions only for ν = n/2, n ∈ Z, i.e., we arrive once again
at the Dirac quantization condition, and the corresponding solutions of Eq. (5.15) have the
form
X j
0 j
Ψphys(r, α, β, γ ) = ψs, n (r)Ds, n (α, β, γ ), (5.16)
2 2
j,s
where prime means that in the sum j takes the values j = |n/2|, |n/2| + 1, . . . . Once
again, let us note that here the quantization separates in the necessary way the value of
the quantized parameter |ν| = |n/2| and possible eigenvalues of the angular momentum
operator in the physical subspace: J2 = j (j + 1) > ν 2 .
It is interesting to note that in the present “spinning top” picture for the charge–monopole
system the total angular momentum is represented in terms of the isospin angular momen-
tum as J = −ea Ia , whereas from the point of view of the field theoretical mechanism
[2,3], only spin part of the total angular momentum vector of the charge–monopole system
is created from the isospin degrees of freedom. The seeming contradiction is explained by
the constraint (5.15) prescribing the total angular momentum to take the values starting
from the minimal value |n/2| = |ν|, which can be interpreted as the “internal” spin of the
charge–monopole system, and in this sense here spin is also created by isospin.
6. Higher-derivative form of CS term
The charge–monopole system can also be described by the Lagrangian with CS term
represented in a higher-derivative form. To show this, we write down the CS form (5.4)
as θ = −νe2 de1 . Identifying the vectors e3 and e1 with the unit vectors n = r r −1 and
ṅ · |ṅ|−1 , respectively, and taking into account that e2 = e3 × e1 , the charge–monopole
coupling term can be written down in the equivalent higher-derivative form
ν
Lint = − 2 n · (ṅ × n̈). (6.1)
ṅ
Simple geometrical consideration shows that the higher derivative term (n × ṅ) · n̈/ṅ2 has
a sense of angular velocity of rotation of the vector ṅ about the vector n. The total charge–
monopole Lagrangian is given by
1 r
L = ṙ2 − ν (r × ṙ) · r̈. (6.2)
2 (r × ṙ)2
Naively, the equations of motion following from Lagrangian (6.2),
∂L d ∂L d2 ∂L
− + 2 = 0, (6.3)
∂r dt ∂ ṙ dt ∂ r̈
are the third order differential equations and their equivalence to the second-order
equations (2.3) is not obvious. Let us prove the equivalence of equations of motion (6.3) to
Eq. (2.3). First we note that the higher-derivative term (6.1) satisfies the following relations:
∂Lint ∂Lint ∂Lint
r· = 0, ṙ · = −Lint , r̈ · = +Lint ,
∂r ∂ ṙ ∂ r̈
∂Lint ∂Lint ∂Lint ∂Lint
ṙ · = (rṙ)r −2 Lint , r· =r· = ṙ · = 0. (6.4)
∂r ∂ ṙ ∂ r̈ r̈
Multiplying the equations (6.3) subsequently by the vectors r, ṙ and r × ṙ, and using Eqs.
(6.4), we arrive at the three equations r̈r = 0, r̈ṙ = 0, r̈ · (r × ṙ) = − rν3 (r × ṙ)2 . Since r,
ṙ − r(rṙ)r −2 and r × ṙ form the complete basis of orthogonal vectors, we conclude that the
equations of motion (6.3) are equivalent to the obtained system of three scalar equations,
which, in turn, is equivalent to one vector equation (2.3). Therefore, the coupling of the
charge to the magnetic monopole can be alternatively described by the scale invariant
higher derivative term (6.1).
It is worth noting that analogously to the present system, the relativistic massive particle
in (2 + 1) dimensions coupled to the external constant homogeneous electromagnetic field
turns out to be classically equivalent [37] to the higher derivative R model of relativistic
particle with torsion given by the action [38,41,42] Stor = − (m + ακ) ds, where
ds 2 = −dxµ dx µ , κ is a scale invariant torsion of the particle’s world trajectory, κ =
µνλ xµ0 xν00 xλ000 /(x 00 )2 , xµ0 = dxµ /ds, and α is a parameter. Such a model underlies (2 + 1)-
dimensional anyons [39,41,42], and in the last section we shall discuss the relationship
between the charge–monopole system and anyons.
7. Twistor description of the charge–monopole system
The helicity constraint appearing in the charge–monopole system interpreted as a parti-

cle with spin is analogous to the helicity-fixing constraint in the case of massless particle
with spin. The latter system admits, in particular, the twistor description [67–71]. Using
this observation, one can get the twistor formulation for the charge–monopole system.
In the twistor approach for the massless particle, the corresponding energy–momentum
vector is treated as a “composite” object constructed from the twistor (even spinor)
variables, and the helicity fixing constraint generates the U(1) transformations for twistors.
By analogy, let us introduce mutually conjugate even complex variables za and z̄a = za∗ ,
a = 1, 2, forming two conjugate spinors. With them, we represent the charge coordinate
vector r as a composite vector,
ϕi ≡ ri − zσi z̄ ≈ 0, (7.1)
where σi is the set of Pauli matrices. Introducing the momenta Pa , P̄a canonically
conjugate to the spinor variables, {za , Pb } = δab , {z̄a , P̄b } = δab , we construct the
generators of rotations (the total angular momentum vector),
i
Ji = (r × p)i + (zσi P − P̄σi z̄), (7.2)
2
forming the su(2) algebra: {Ji , Jj } = ij k Jk . Here p is the vector canonically conjugate
to r. Following the twistor approach, we also introduce the constraint
i
χ ≡ (z̄P̄ − zP) − ν ≈ 0, (7.3)
2
which is in involution with the constraints (7.1) and generates the U(1) gauge transforma-
tions for the spinor variables, za → za0 = eiγ za , z̄a → z̄0a = e−iγ z̄a , Pa → Pa0 = e−iγ Pa ,
P̄a → P̄ 0a = eiγ P̄a , with γ = γ (t) being a parameter of transformation. Taking into ac-
count the identity
i
σab i
σcd = δab δcd − 2ac bd , (7.4)
with ab = −ba , 12 = 1, we get the relation r ≈ z̄z as the consequence of the constraint
(7.1). This relation and Eq. (7.2) allow us to represent the constraint (7.3) in the equivalent
form of the helicity-fixing constraint,
χ̃ ≡ Jr + νr ≈ 0. (7.5)
Since the total phase space described by ri , pi , za , z̄a , Pa and P̄a is 14-dimensional, and
we have 4 first class constraints (7.1) and (7.3), there are only 6 independent physical
phase space degrees of freedom like in the charge–monopole system. They are given by
the observables having zero Poisson brackets with all the constraints. Such independent
variables are ri and
1
Πi ≡ pi + zσi P + P̄σi z̄ , (7.6)
2z̄z
for which we have the Poisson bracket relations {ri , rj } = 0, {ri , Πj } = δij and
ν
{Πi , Πj } = (z̄z)−2 ij k (r × 5 − J)k ≈ 3 ij k rk . (7.7)
r
The weak equality means the equality on the surface of constraints (7.1) and (7.3). Physical
variables Πi correspond here to the charge–monopole variables Pi . The Hamiltonian of the
charge–monopole system in the twistor formulation can be taken as the linear combination
of 52 and of the first class constraints,
1
H = 52 + ρi ϕi + λχ, (7.8)
2
with ρi = ρi (t), λ = λ(t) being arbitrary functions (Lagrange multipliers). Direct
calculation shows that the equations of motion generated by this Hamiltonian are reduced
to r̈ ≈ −νr −3 (r × ṙ), i.e., the Hamiltonian (7.8) and constraints (7.1), (7.3) give the
alternative twistor description for the charge–monopole system.
The quantum theory of the system in the twistor approach is the following. It is natural
to choose the representation diagonal in r, za and z̄a , and realize the canonically conjugate
momenta in the form of differential operators. In accordance with constraint (7.1), the
physical states have to be of the form Ψphys = δ (3)(r − zσ z̄) · ψ(z, z̄). Then like in the
model of massless particle with spin [69], the quantum analog of the constraint (7.3),

1 ∂ ∂
z̄a − za − ν Ψphys = 0, (7.9)
2 ∂ z̄a ∂za
and the requirement of single-valuedness of the wave functions results in the quantization
of the charge–monopole constant, 2ν = n, n ∈ Z. Finally, the physical states subject to Eq.
(7.9) will be described by the wave functions of the form
Ψphys = δ (3) (r − zσ z̄) · ψphys(z, z̄),
∞
X X
ψphys (z, z̄) = Ckab · (za )k (z̄b )n+k , (7.10)
k=−∞ a,b=1,2
where Ckab are constants. The action of quantum analogs of classical observables r and 5
on physical wave functions (7.10) is reduced to
rΨphys = δ (3) (r − zσ z̄) zσ z̄ ψphys (z, z̄),

∂ ∂
5Ψphys = −iδ (3) (r − zσ z̄) · (2z̄z)−1 zσ + σ z̄ ψphys(z, z̄). (7.11)
∂z ∂ z̄
By inverse Legendre transformation, analogously to the case of the massless particle
with spin [69], one can construct the Lagrangian corresponding to Hamiltonian (7.8).
Instead of realizing such constructions, we note here that the list of independent observable
quantities and the action of their quantum analogs on the physical states indicate on the
possibility to exclude r and p as independent variables and to construct the theory realized
only in terms of spinor variables. To get the corresponding twistor form of the Lagrangian,
it is convenient to start from the higher-derivative Lagrangian (6.2) by putting in it r = zσ z̄.
The remarkable feature of such a twistor representation is that its substitution into the CS
˙ − z̄ż)/z̄z, and we arrive
term of the higher derivative form (6.1) reduces the latter to −iν(z̄z
at the Lagrangian
2 ˙
1 d z̄ż − z̄z
L= (zσ z̄) + iν . (7.12)
2 dt z̄z
Let us show that (7.12) describes the charge–monopole system. From the definition of the
canonical momenta Pa = ∂L/∂ ża , P̄a = ∂L/∂ z̄˙ a , we find that the constraint (7.3) is the
primary constraint for the system (7.12) and the total Hamiltonian is
1
H= π π̄ + λχ, (7.13)
2zz̄
where we have introduced the notation πa = Pa − iν z̄a (z̄z)−1 , π̄a = P̄a + iνza (z̄z)−1 .
The constraint (7.3) is conserved by the Hamiltonian and, as a consequence, there are
no secondary constraints. Therefore, the constraint (7.3) is the first class constraint and
the number of physical phase space degrees of freedom is equal to 8 − 2 = 6. The
corresponding 6 independent observables weakly commuting in the Poisson bracket sense
with the constraint (7.3) are
1
ri = zσi z̄, Πi = (P̄σi z̄ + zσi P). (7.14)
2zz̄
These quantities satisfy the following Poisson bracket relations: {ri , rj } = 0, {ri , Πj } = δij ,
{Πi , Πj } ≈ νr −3 ij k rk , where the weak equality means the equality on the surface of the
constraint (7.3). Obviously, they have the sense of the charge–monopole variables ri and Pi
represented in the composite form (7.14) in terms of twistor variables. The variables Πi are
nothing else as the classical analogs of the quantum operators corresponding to observables
(7.6), whose action is reduced to the surface of the constraints (7.1), i.e., they are the
classical analogs of operators (7.11). The conserving su(2) generators are represented
here in the form J = r × 5 − νn with n = r r −1 and ri , Πi given by Eq. (7.14). Direct
verification shows that the Hamiltonian (7.13) generates the correct equations of motion,
r̈ = −νr −3 (r × ṙ). It is worth noting that the system (7.13) can be treated as a reduction of
the Hamiltonian system (7.8) [63,64] to the surface of the constraints (7.1) supplied with
the gauge conditions pi − 12 (zz̄)−1 (zσi P + P̄σi z̄) ≈ 0.
To conclude the discussion of the twistor formulation for the charge–monopole system,
let us show how the limit of zero mass described here by the Lagrangian
˙
z̄ż − z̄z
L = iν (7.15)
z̄z
gives rise to the pure spin system. Here, as before, it is assumed that the initial configuration
space is C2 − {0}. This system can be quantized by the method advocated in Ref. [72], but
the symplectic two-form corresponding to Lagrangian (7.15) is singular, ω = ωab dz̄a ∧
dzb , ωab = 2iν(z̄z)−1 (δab − za z̄b · (z̄z)−1 ), ωab z̄a = ωab zb = 0, that complicates the
analysis. There is a more short way to reveal a spin nature of the system (7.15) by showing
its equivalence to the system (3.10). For the purpose we note that in the regions where
either z1 6= 0 (chart U1 ) or z2 6= 0 (chart U2 ), one can define the complex coordinate Z as
Z = z2 /z1 or Z = z1 /z2 , respectively. Then Lagrangian (7.15) is represented equivalently
as L = iν(Z̄ Ż − Z̄Z)(1 ˙ + Z̄Z)−1 − 2ν ϕ̇, where ϕ is the phase of the coordinate za
in the chart Ua , a = 1, 2. The first term coincides exactly with the Lagrangian (3.10)
corresponding to the pure spin system. The second total derivative term is not important
classically as well as quantum mechanically since its contribution to the action, 1S =
−4νπ , for periodical trajectories is trivialized, 1S = 0(mod 2π), if we take into account
the quantization condition 2ν ∈ Z. Therefore, the system (7.15) is equivalent to the pure
spin system (3.10).
The equivalence of the system (7.15) to the system (3.10) is encoded in its gauge
symmetries. In correspondence with the above mentioned degeneracy of the symplectic
form, the system (7.15) possesses two gauge invariances: its action is invariant under the
transformations za → ρza , ρ = ρ(t) > 0, generated by the constraint ψ = P̄ z̄ + zP ≈ 0,
and za → eiϕ za , ϕ = ϕ(t) ∈ R, generated by the helicity constraint of the form (7.3). This
means that the points za and ζ za , ζ ∈ C, ζ 6= 0, in configuration space are physically
equivalent and should be identified, za ∼ ζ za . As a result, the configuration space of the
system is the projective complex plane CP 1 = (C2 − {0})/ ∼, on which the coordinate Z
introduced above plays a role of the “inhomogeneous” coordinate [34].
8. Charge–monopole in a spherical geometry and spin
Let us restrict the motion of the charge in the monopole field to the sphere r2 = 1.
This can be done by treating the relation r2 − 1 = 0 as a Lagrangian constraint, L =
2 ṙ + eṙA(r) − 2 (r − 1), whose condition of conservation generates the Hamiltonian
1 2 λ 2
constraint pr = 0. The reduction to the surface of these second class constraints excludes
the radial variables r and pr , and we arrive at the Hamiltonian describing the charge on the
sphere with the monopole in its center, 2 H = 12 (J × n)2 . Here the dynamical variables J
and n satisfy the Poisson bracket relations (2.22) and are subject to the conditions (2.23),
i.e., the reduced system is a pure reduced E(3) system. The same procedure of reduction can
2 See also Refs. [73–75], where some aspects of the charge–monopole system in a spherical geometry were
discussed.
be realized in a spinning top picture, i.e., by adding the condition r2 − 1 = 0 to Lagrangian

(5.5) as a Lagrangian constraint, or by putting r = 1 directly in Lagrangian (5.5):
1 ν
L = ṅ2 + (e1 ė2 − e2 ė1 ), (8.1)
2 2
where we suppose that n = e1 × e2 and vectors e1 and e2 are orthonormal. In the case
of higher derivative treatment of the charge–monopole system the corresponding reduced
Lagrangian is
1 ν
L = ṅ2 − 2 n · (ṅ × n̈), (8.2)
2 ṅ
whereas in the twistor picture the Lagrangian takes the form
˙ 2 z̄) + iν(z̄ż − z̄z),
L = −2(żσ2 z)(z̄σ ˙ z̄z = 1. (8.3)
The condition z̄z = 1 can be omitted by normalizing appropriately the Lagrangian, i.e., by
multiplying the first and second terms by (z̄z)−2 and (z̄z)−1 , respectively.
The CS term is not changed by the reduction procedure due to its scale-invariance. But
this term in addition is reparametrization invariant, whereas the total Lagrangian has no
reparametrization invariance. We can√ change the non-invariant second order in velocity
1 2
term 2 ṅ for the first order term ṅ2 , that gives rise to the reparametrization invariant
action. Let us analyze the physical content of such reparametrization action considering,
e.g., the modification of the Lagrangian (8.1),
p ν
L = γ ṅ2 + (e1 ė2 − e2 ė1 ), (8.4)
2
where γ > 0 is a dimensionless parameter. The canonical Hamiltonian of the system (8.4)
is equal to zero, and from the Hamiltonian point of view, the difference of the
system (8.4) from the system (8.1) consists in the presence of the constraint J2 − κ 2 ≈ 0,
κ 2 ≡ γ 2 + ν 2 , generating reparametrizations in addition to the isospin U(1) gauge
symmetry generated by the constraint (5.12). The system (8.4) has the same physical
content as the spin system of Section 3 given by only the CS term. Indeed, applying the
quantization scheme of Section 5, the physical subspace is separated here by Eq. (5.15)
and by (J2 − κ 2 )Ψphys(α, β, γ ) = 0. These two equations have nontrivial solutions only
when the parameters κ and ν are quantized: κ 2 = j (j + 1) and ν = k, where k, |k| 6 j ,
is integer (half-integer) for j integer (half-integer), and the corresponding physical state
Pj j
is Ψphys(α, β, γ ) = s=−j Cs Ds,k (α, β, γ ), where Cs are constants. This wave function
describes an arbitrary state of fixed spin j in the form alternative to the holomorphic
functions of Section 3. Therefore, the reparametrization-invariant system (8.4) is equivalent
to the spin system described by only the CS term.
The charge–monopole system in spherical geometry can be treated as a partially gauge
fixed version of the reparametrization and scale invariant spin system (8.4). To see this, we
rewrite the Lagrangian (8.4) in equivalent form by introducing the einbein v:
ṅ2 v 2 ν
L= + γ + (e1 ė2 − e2 ė1 ). (8.5)
2v 2 2
Reparametrization invariance of the system (8.5) can be fixed (locally, see Ref. [64]) via
introducing the appropriate gauge-fixing conditions for the constraint J2 − κ 2 ≈ 0, and for
the constraint pv ≈ 0, where pv is the momentum canonically conjugate to v. On the other

hand, introducing only the condition v = 1, we obtain the partially gauge fixed version of
the system (8.5) [63]. The Lagrangian (8.5) with v = 1 is the Lagrangian (8.1) shifted for
the inessential constant. Therefore, the charge–monopole system in a spherical geometry
can formally be treated as a partially gauge fixed version of the spin system (8.5).
Analogously, we can change the kinetic term 12 ṙ2 in the initial charge–monopole
√
Lagrangian (2.1) for the reparametrization invariant term ṙ2 , the latter is a kinetic term
of relativistic particle in 3D Euclidean space. As a result, we get the reparametrization
invariant action
Z
ṙ2 v
S = Lr dt, Lr = + + eAṙ. (8.6)
2v 2
Its partial gauge fixed version (v = 1) will give the Lagrangian coinciding up to inessential
constant with the initial Lagrangian (2.1). From this point of view, the time translation,
the time dilation and special conformal transformation symmetries produced canonically
by the so(2, 1) generators H , D and R [10,49], can be treated as a relic of the
reparametrization symmetry of the system (8.6) surviving the described formal Lagrangian
gauge fixing procedure.
9. Charge–monopole system and anyons
Let us discuss the relationship between the charge–monopole system and (2 + 1)-
dimensional anyons in the light of the obtained results. Earlier, the analogy with the
charge–monopole system played an important role in constructing the theory of anyons
as spinning particles [43–45].
We have observed that the charge–monopole system in many aspects is similar to the
3D free particle. In (2 + 1) dimensions, spin is a (pseudo)scalar and, as a consequence, the
anyon of fixed spin has the same number of degrees of freedom as a spinless free massive
particle. The relationship between the charge–monopole and anyon can be understood
better within the framework of the canonical description of these two systems. As we
have seen, the charge–monopole system essentially is a reduced E(3) system. In the case
of anyons, the E(3) group is changed for the Poincaré group ISO(2,1). The translation
generators of the corresponding groups are n and pµ , the latter being the energy–
momentum vector of the anyon, and the corresponding Casimir central elements are fixed
by the relations n2 = 1 and
p2 + m2 ≈ 0, (9.1)
where m is a mass of the anyon. The rotation (Lorentz) generators are given by J and by
Jµ = µνλ x ν pλ + Jµ , (9.2)
where Jµ are the translation invariant so(2, 1) generators satisfying the algebra of the
form (2.17), {Jµ , Jν } = µνλ J λ , and subject to the relation Jµ J µ = −α2 = const. The
representation (9.2) for the anyon total angular momentum vector is, obviously, the analog
of the relation J = L + s appearing under interpretation of the charge–monopole system as

a particle with spin. In the charge–monopole system, the second Casimir element of E(3)
is fixed either strongly, Jn = −ν, or in the form of the weak relation Jn + ν ≈ 0 (see Eqs.
(4.2), (4.6)). In the anyon model, spin also can be fixed either strongly, J p = Jp = −αm,
or in the form of the weak (constraint) relation
χa ≡ Jp + αm ≈ 0. (9.3)
When the helicity is fixed strongly, for the charge–monopole system the symplectic
form corresponding to the Poisson brackets (2.8), has a nontrivial contribution describing
noncommuting quantities Pi being the components of the charge’s velocity: ω = dPi ∧
dri + 2rν 3 ij k ri drj ∧ drk . In the anyon case, the strong spin fixing gives rise to the nontrivial
Poisson structure for the particle’s coordinate’s [43–45],
{xµ , xν } = α(−p2 )−3/2 µνλ pλ . (9.4)
On the other hand, when we treat the charge–monopole system as a particle with spin
(helicity is fixed weakly), one can work in terms of the canonical symplectic structure for
the charge’s coordinates and momenta (see Eq. (4.1)), but the canonical momenta p are
not observables due to their non-commutativity with the helicity constraint, whereas the
gauge-invariant extension of p given by Eq. (4.3) plays the role of the non-commuting
quantities Pi . Exactly the same picture takes place in the case of anyon when its spin
is fixed weakly: the coordinates xµ commute in this case, {xµ , xν } = 0, but they have
nontrivial Poisson brackets with the spin constraint (9.3), whereas their gauge-invariant
extension, Xµ = xµ + p12 µνλ pν J λ (cf. with Eq. (4.3)), {Xµ , χa } ≈ 0, are non-commuting
and reproduce the Poisson bracket relation (9.4). Like in the anyon case, the advantage of
the extended formulation for the charge–monopole system (when we treat it as a particle
with spin), is in the existence of canonical charge’s coordinates ri and momenta pi . Within
the initial minimal formulation (given in terms of ri and gauge-invariant variables Pi ),
the canonical momenta pi are reconstructed from Pi only locally, pi = Pi + eAi , due to
the global Dirac string singularities hidden in the monopole vector potential. Having in
mind the gauge invariant nature and non-commutativity of Pi or their analogs Πi from
the extended formulation, we conclude that they, like anyon coordinates Xµ , are the spin–
monopole’s analogs of the Foldy–Wouthuysen coordinates of the Dirac particle [45,76].
The discussed classical so(2, 1) symmetry can be quantized in an abstract way

proceeding from the classical relations (2.17) and (2.19). In such a way the infinite-
dimensional unitary half-bounded sl(2, R) representations of the discrete series Dα+ will
be obtained, which are characterized by the quantum Casimir element C = −α(α − 1),
0 < α ∈ R and by the eigenvalues j0 = α + n, n = 0, 1, . . . , of the operator J0 [54]. Since
α > 0 is arbitrary, such a quantization procedure of the so(2, 1) symmetry algebra does
not introduce any restrictions for the charge–monopole coupling constant, and does not fix
correctly the spectrum of the operator J2 . The latter information, as we saw, is encoded in
the corresponding so(3) algebra and classical condition J2 > ν 2 . The quantization of the
parameter ν and the quantum spectrum of J2 could be obtained in principle by applying the
geometric quantization to the classical E(3) system from Section 2.4. On the other hand, the
same information could be extracted from the quantization of classical so(3, 1) symmetry
of the charge–monopole system described in Section 2.3 with taking into account the
classical relation J2 > ν 2 . We are going to consider the geometric quantization of so(3, 1)
charge–monopole symmetry elsewhere. The observation of the similarity between the
charge–monopole and the 3D free particle systems realized in Section 2.3 in the context
of the vector integrals of motion has been applied recently in Ref. [77] for explaining the
nature of the nonstandard fermion–monopole supersymmetry [78].
In Sections 4 and 5 we have discussed the two different interpretations of the charge–
monopole system as a particle with spin and as a spinning top system. It is interesting to
find the corresponding map between these pictures at the Hamiltonian level. In the first
picture, the “rotational” part of the phase space is given by the spin vector s, s2 = ν 2 , and
by the orbital angular momentum L and associated unitary vector n. In the spinning top
picture we have the angular momentum vector J and the set of orthonormal vectors ea . The
vector J of the second formulation is identified with the total angular momentum vector
L + s from the first formulation, and the vector e3 giving the symmetry axis of the top is
naturally identified with n. Therefore, to find the mapping between the two formulations,
it is sufficient to construct from the variables s, L and n the orthonormal vectors e1,2 ,
e1 × e2 = n, satisfying the necessary Poisson bracket relations. Unfortunately, we did not
succeed in realization of such a construction.
It seems interesting to investigate analogously other systems of particles (strings) in the
background of external gauge or gravitational fields from the point of view of their possible
alternative description as the particles (strings) with internal reduced degrees of freedom.
Acknowledgements
I am grateful to J. Zanelli and D. Sorokin for interest to this work and helpful discussions.
I thank J. Alfaro and J. Gamboa for bringing Refs. [5,9] to my attention and M. Bañados
for discussion of some related issues. The work has been supported by the grant 1980619
from FONDECYT (Chile) and by DICYT (USACH).
References
[1] P.A.M. Dirac, Proc. R. Soc. London A 133 (1931) 60.

[2] R. Jackiw, C. Rebbi, Phys. Rev. Lett. 36 (1976) 1116.
[3] P. Hasenfratz, G. ’t Hooft, Phys. Rev. Lett. 36 (1976) 1119.
[4] A.S. Goldhaber, Phys. Rev. Lett. 36 (1976) 1122.
[5] D. Boulware, L. Brown, R. Cahn, S. Ellis, C. Lee, Phys. Rev. D 14 (1976) 2708.
[6] T.T. Wu, C.N. Yang, Nucl. Phys. B 107 (1976) 365.
[7] T.T. Wu, C.N. Yang, Phys. Rev. D 14 (1976) 437.
[8] T.T. Wu, C.N. Yang, Phys. Rev. D 16 (1977) 1018.

[9] P. Goddard, D.I. Olive, Rep. Prog. Phys. 41 (1978) 1357.
[10] R. Jackiw, Ann. Phys. (N.Y.) 129 (1980) 183.
[11] P.A. Horvathy, Int. J. Theor. Phys. 20 (1981) 697.
[12] J.L. Friedman, R.D. Sorkin, Phys. Rev. D 20 (1979) 2511.
[13] A.P. Balachandran, G. Marmo, B.S. Skagerstam, A. Stern, Nucl. Phys. B 162 (1980) 385.
[14] A.P. Balachandran, G. Marmo, B.S. Skagerstam, A. Stern, Gauge Symmetries and Fibre
Bundles, Lecture Notes in Physics, Vol. 188, Springer-Verlag, Berlin, 1983.
[15] F. Zaccaria, E.S.G. Sudarshan, J.S. Nilsson, N. Mukunda, G. Marmo, A.P. Balachandran, Phys.
Rev. D 27 (1983) 2327.
[16] A.P. Balachandran, Wess–Zumino Terms and Quantum Symmetries, Preprint SU-4428-361,
1987.
[17] R. Jackiw, Phys. Rev. Lett. 54 (1985) 159.
[18] R. Jackiw, Phys. Lett. B 154 (1985) 303.
[19] S. Deser, R. Jackiw, S. Templeton, Ann. Phys. 140 (1982) 372.
[20] S. Deser, R. Jackiw, S. Templeton, Phys. Rev. Lett. 48 (1982) 975.
[21] E. Cremmer, B. Julia, J. Scherk, Phys. Lett. B 76 (1978) 409.
[22] W. Siegel, Nucl. Phys. B 156 (1979) 135.
[23] R. Jackiw, S. Templeton, Phys. Rev. D 23 (1981) 2291.
[24] J. Schonfeld, Nucl. Phys. B 185 (1981) 157.
[26] E. Witten, Commun. Math. Phys. 121 (1989) 351.
[27] M. Bañados, C. Teitelboim, J. Zanelli, Phys. Rev. Lett. 69 (1992) 1849.
[28] M. Bañados, M. Henneaux, C. Teitelboim, J. Zanelli, Phys. Rev. D 48 (1993) 1506.
[29] G. Dunne, R. Jackiw, C. Trugenberger, Phys. Rev. D 41 (1990) 661.
[30] G. Dunne, R. Jackiw, Nucl. Phys. Proc. Suppl. C 33 (1993) 114.
[31] P.S. Howe, P.K. Townsend, Class. Quant. Grav. 7 (1990) 1655.
[32] M. Reuter, Phys. Rev. D 42 (1990) 2763.
[33] M. Asorey, J. Geom. Phys. 11 (1993) 63.
[34] M. Nakahara, Geometry, Topology and Physics, Hilger, Bristol, 1990.
[35] C. Nash, Topology and physics: a historical essay, hep-th/9709135.
[36] J.M. Labastida, Chern–Simons gauge theory: ten years after, hep-th/9905057.
[37] M.S. Plyushchay, Mod. Phys. Lett. A 10 (1995) 1463.
[38] A.M. Polyakov, Mod. Phys. Lett. A 3 (1988) 325.
[39] M.S. Plyushchay, Phys. Lett. B 248 (1990) 107.
[40] M.S. Plyushchay, Int. J. Mod. Phys. A 7 (1992) 7045.
[42] M.S. Plyushchay, Nucl. Phys. B 362 (1991) 54.
[43] B.S. Skagerstam, A. Stern, Int. J. Mod. Phys. A 5 (1990) 1575.
[44] R. Jackiw, V.P. Nair, Phys. Rev. D 43 (1991) 1933.
[45] J.L. Cortes, M.S. Plyushchay, Int. J. Mod. Phys. A 11 (1996) 3331.
[46] S. Deser, R. Jackiw, G. ’t Hooft, Ann. Phys. (N.Y.) 152 (1984) 220.
[47] S. Deser, R. Jackiw, Commun. Math. Phys. 118 (1988) 495.
[48] R. Jackiw, Planar gravity, Lectures given at SILARG VII, Cocoyoc, Mexico, December 1990,
in: Diverse Topics in Theoretical and Mathematical Physics, World Scientific, Singapore, 1995,
pp. 155–180.
[49] R. Jackiw, Ann. Phys. 201 (1990) 83.
[50] N. Steenrod, The Topology of Fibre Bundles, Princeton Univ. Press, Princeton, NJ, 1951.
[51] E.S. Moreira, Phys. Rev. A 58 (1998) 1678.
[52] V. De Alfaro, S. Fubini, G. Furlan, Nuovo Cimento A 34 (1976) 569.
[53] J. Beckers, J. Harnad, M. Perroud, P. Winternitz, J. Math. Phys. 19 (1978) 2126.
[54] M.S. Plyushchay, J. Math. Phys. 34 (1993) 3954.

[55] V.P. Akulov, A.I. Pashnev, Theor. Math. Phys. 56 (1983) 862.
[56] P. Claus, M. Derix, R. Kallosh, J. Kumar, P.K. Townsend, A. Van Proeyen, Phys. Rev. Lett. 81
(1998) 4553.
[57] G. Papadopoulos, Conformal and superconformal mechanics, hep-th/0002007.
[58] B. Kostant, Quantization and unitary representations, in: Lecture Notes in Modern Analysis and
Applications III, Lecture Notes in Mathematics, Vol. 170, Springer-Verlag, 1970, pp. 87–208.
[59] A.A. Kirillov, Geometric quantization, in: Encyclopedia of Mathematical Sciences, Vol. 4,
Springer-Verlag, 1990, pp. 138–172.
[60] D.J. Simms, N.M.J. Woodhouse, Lecture Notes on Geometric Quantization, Lecture Notes in
Physics, Vol. 53, Springer-Verlag, 1976.
[61] J.-M. Souriau, Structure of Dynamical Systems, a Symplectic View of Physics, Birkhäuser,
1997.
[62] G. Jorjadze, J. Math. Phys. 38 (1997) 2851.
[63] M. Henneaux, C. Teitelboim, Quantization of Gauge Systems, Princeton Univ. Press, 1992.
[64] M.S. Plyushchay, A.V. Razumov, Int. J. Mod. Phys. A 11 (1996) 1427.
[65] M.S. Plyushchay, G.P. Pron’ko, A.V. Razumov, Theor. Math. Phys. 67 (1986) 576.
[67] D.P. Sorokin, V.I. Tkach, D.V. Volkov, Mod. Phys. Lett. A 4 (1989) 901.
[68] D.P. Sorokin, V.I. Tkach, D.V. Volkov, A.A. Zheltukhin, Phys. Lett. B 216 (1989) 302.
[70] I.A. Bandos, Sov. J. Nucl. Phys. 51 (1990) 906.
[71] A.S. Galperin, P.S. Howe, K.S. Stelle, Nucl. Phys. B 368 (1992) 248.
[72] L. Faddeev, R. Jackiw, Phys. Rev. Lett. 60 (1988) 1692.
[73] V.P. Akulov, D.P. Sorokin, I.A. Bandos, Mod. Phys. Lett. A 3 (1988) 1633.
[74] V.P. Akulov, D.P. Sorokin, I.A. Bandos, Sov. J. Nucl. Phys. 47 (1988) 724.
[75] M. Stone, Nucl. Phys. B 314 (1989) 557.
[76] L.L. Foldy, S.A. Wouthuysen, Phys. Rev. 78 (1950) 29.
[78] F. De Jonghe, A.J. Macfarlane, K. Peeters, J.W. van Holten, Phys. Lett. B 359 (1995) 114.
Orientifolds and twisted boundary conditions

Arjan Keurentjes
Instituut-Lorentz for theoretical physics, Universiteit Leiden, P.O. Box 9506,
NL-2300 RA, Leiden, The Netherlands
Received 12 April 2000; revised 5 July 2000; accepted 16 August 2000
Abstract
It is argued that the T-dual of a cross-cap is a combination of an O + and an O − orientifold plane.
Various theories with cross-caps and D-branes are interpreted as gauge-theories on tori obeying
twisted boundary conditions. Their duals live on orientifolds where the various orientifold planes
are of different types. We derive how to read off the holonomies from the positions of D-branes in
the orientifold background. As an application we reconstruct some results from a paper by Borel,
Friedman and Morgan for gauge theories with classical groups, compactified on a 2- or 3-torus with
twisted boundary conditions.  2000 Elsevier Science B.V. All rights reserved.
1. Introduction
The advent of D-branes and orientifolds in string theory gave new tools to study gauge
theories. The vacua of gauge theories with classical groups, compactified on tori with
commuting holonomies can be straightforwardly described by configurations of D-branes
and orientifold planes (see [17] and references quoted there). For theories with unitary
or symplectic groups, holonomies are specified in the fundamental representation, for
theories with orthogonal groups the holonomies are in the vector representation. All these
representations still have a non-trivial centre.
When fields are invariant under the centre, typical for conventional open string
perturbation theory, one can allow for holonomies that commute up to an element of the
centre, as was first considered for gauge theory by ’t Hooft [10] in terms of so-called
twisted boundary conditions. But it was only relatively recent that this extra freedom was
first considered in the context of string theory. In [24] Witten studied the case of SO(4N)
on 2- and 3-tori, which can be described by a configuration of D-branes on an orientifold.
In the appendix of the same paper, Witten used a construction involving D-branes on
an orientifold to show that for orthogonal groups the moduli space of flat connections
E-mail address: arjan@lorentz.leidenuniv.nl (A. Keurentjes).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 2 2 - 8
A. Keurentjes / Nuclear Physics B 589 (2000) 440–460 441
on the 3-torus with periodic boundary conditions has an extra component not considered
before, and that this seems to solve an old problem concerning the computation of the
Witten index [23]. Motivated by this, various authors [2,11–14] have subsequently shown
the necessary existence of extra vacuum components for exceptional groups, so as to
solve the Witten index problem for these groups, for which no D-brane construction
is available. The authors of [2] also included the case of general boundary conditions,
likewise demonstrating a richer structure than considered earlier in the computation of the
Witten index with twisted boundary conditions [23]. It has remained a challenge to translate
all these results into configurations of D-branes and orientifold planes, which allowed
Witten to make his discovery for the orthogonal groups [24]. Here we close the circle
by addressing this translation for all classical groups, with arbitrary boundary conditions.
The fact that in standard open string perturbation theory, all representations are
conjugate to the adjoint (and therefore invariant under the centre) is believed to be false
outside perturbation theory [21,22,25]. It is argued that the full gauge group is actually
Spin(32)/Z2 , and that also spinorial representations occur. But even configurations that
would be consistent for Spin(32)/Z2 -gauge theory, can be shown to be inconsistent for the
string theory by more subtle arguments [1]. However, in this paper we wish to elucidate the
underlying gauge group structure, which we stress is interesting in its own right. It is also
an essential step towards a cleaner, and more complete derivation of the string consistency
conditions governing orientifolds. We should note however, that there will be important
modifications to our results on allowed gauge groups and representations in orientifold
compactifications in D < 10 upon imposition of the string consistency conditions.
The main part of this paper is devoted to a discussion of compactification with
orthogonal and symplectic groups on 2- and 3-tori, with twisted boundary conditions
(as previously analysed in [2,19]), but we will also discuss U (n) theory with various
boundary conditions. After one T-duality, these theories correspond to configurations of
branes and orientifold fixed planes on the Möbius strip, the Klein bottle, and tori that
are not rectangular. We derive how to T-dualise the Möbius strip and Klein bottle in the
direction orthogonal to the first T-duality. This leads to orientifolds with fixed planes of
different type, much like in [24]. With these methods every possible flat connection for
symplectic gauge groups or orthogonal gauge groups allows a translation in terms of a
configuration of D-branes on an orientifold. Holonomies and possible enhanced symmetry
groups are easily read off from the configuration.
We will open with a discussion of a U (n)-theory on a 2-torus with special boundary
conditions: along one of the directions of the 2-torus there is a holonomy that is not an
element of U (n).
2. The T-dual of a cross-cap
We start by considering U (n) theory on a circle. U (n) can be embedded in an open string
theory by attaching Chan–Paton charges on the ends of oriented strings. Compactifying this
string theory on a circle and applying a T-duality transformation, we obtain a configuration
442 A. Keurentjes / Nuclear Physics B 589 (2000) 440–460
Fig. 1. The Klein bottle: a double cover of the Klein bottle, arrows indicating the direction of
identifications (middle); an attempt to draw the standard representation of the Klein bottle, obtained
by taking the lower half of the double cover as fundamental domain (right); the cylinder with two
cross-caps, obtained by taking the left half of the fundamental domain (left). We have also drawn an
example of a brane in all three pictures (depicted twice on the double cover) as it is positioned after
the first T-duality.
of n D-branes that are transverse to a dual circle, each intersecting the dual circle in one
point. The location of the D-branes is controlled by the holonomy Ω1 along the circle in the
original theory. We are interested in configurations with discrete symmetries. The discrete
symmetries of the circle are the shift symmetries, shifting the circle by an angle 2πq with
q a rational number, and the order 2 reflection symmetry. Only for specific choices of the
original holonomy will the D-brane configuration respect one or some of these symmetries.
In this section, we are interested in the reflection symmetry.
The reflection on the circle has two fixed points, which will be taken to be at X = 0 and
X = πR (X being the coordinate along the circle, and 2πR its circumference). This can
always be arranged: in the original theory we had a holonomy in U (n), which is locally
equivalent to U (1) × SU(n). The holonomy for the U (1)-factor can be chosen arbitrarily,
since it does not couple to anything. In the dual theory this corresponds to an overall
translation, which we use to set the coordinates of the fixed points to the above values.
Now compactify in addition on another circle of radius R 0 with holonomy Ω2 along this
circle. The standard formalism assumes holonomies that can be diagonalised within the
group. In case we have the above Z2 symmetry, we may consider a holonomy that includes
the Z2 reflection. Gluing the circle to a reflected circle upon going around the second cycle,
one does not obtain a 2-torus, but a Klein bottle. Instead of a non-trivial line bundle over
a circle, we will represent the Klein bottle here as a cylinder of length πR, circumference
4πR 0 , bounded by two cross-caps at the end, the cross-cap being an identification over half
the period of the circle.
The D-branes are wrapped around the cylinder, parallel to the cross-caps. There are two
possibilities, controlled by the holonomy Ω1 in the original theory. D-branes in the bulk
(away from the cross-caps) represent branes that were reflected into their image. In this
representation, D-brane and image are represented as one brane (which is in this sense a
brane pair). In the original theory there can also be D-branes at the fixed point(s) of the Z2 -
reflection. In this representation of the Klein bottle, they are located at the cross-cap. Under
a smooth deformation of the original holonomy, only even numbers of D-branes can move
away from the cross-cap. Hence for U (n) with n odd there is at least one brane fixed under
the Z2 reflection and therefore stuck to a cross-cap. For n even there are two possibilities:
the number of branes at each cross-cap is either even or odd. In the latter situation there is
at least one brane at each cross-cap.
The above is reminiscent of the situation for orientifold planes. For orientifolds, a brane
and an image brane on the double cover are mapped to a brane-pair in the orientifold.
There is also the possibility of single branes being stuck at an orientifold plane (in the
case of O − -planes. By O − we denote the orientifold plane that gives orthogonal gauge
symmetry, and O + is an orientifold plane that gives symplectic gauge symmetry).
The above configuration can be interpreted in terms of the original gauge theory. U (n)
is locally U (1) × SU(n), and the U (1) background is fixed. For n > 2, SU(n) possesses
an outer automorphism, which, in a suitable representation, corresponds to complex
conjugation. We will be working in the fundamental representation, and denote complex
conjugation as C with action
C : U → U ∗, U ∈ SU(n). (1)
One can extend this action to U (n) as C also has a simple action on U (1), and now one
may also extend to n 6 2. One normally considers holonomies taking values in the gauge
group, which corresponds to combining a translation in space with the action of an inner
automorphism (i.e., a conjugation) on the group. One may also consider a holonomy that
corresponds to an outer automorphism, and this is precisely what we are doing in the above.
The outer automorphism C can be combined with an inner one, say conjugation with a
group element A. To avoid ambiguities we require that AC = CA, which is true if A is
real, that is A ∈ O(N). The holonomy Ω2 combines the action of C with conjugation
with A, and we denote it as Ω2 = AC, with A in the fundamental representation of
U (n), and C the operator that implements complex conjugation. The holonomy Ω1 is an
“ordinary” holonomy, and we write Ω1 = B, with B an element of U (n) in the fundamental
representation. Ω1 should commute with Ω2 , which is solved by taking B commuting
with A and B ∈ O(N). Continuous variation of the U (1)-background is incompatible
with complex conjugation; in the D-brane picture this corresponds to the fact that a global
translation on the D-branes is incompatible with the reflection for generic cases.
By conjugation with O(n) matrices, we may transform A and B to a block diagonal
form with 2 × 2 blocks of the form

cos φ − sin φ
(2)
sin φ cos φ
on the diagonal, and some 1’s and −1’s as remaining diagonal entries. In the following, we
will take A and B to be of this standard form.
We wish to T-dualise the cylinder with the two cross-caps in the direction of the circle.
Ignoring the cross-caps one would roughly expect this to lead to a dual theory on a cylinder.
The inclusion of the cross-caps can be analysed by examining the symmetries of the
original theory.
A and B are elements of the vector representation of O(n), and in particular their
eigenvalues occur in pairs: if exp iφ is an eigenvalue, then so is exp(−iφ). The ordering
is unimportant as there are symmetries that allow the exchange of exp iφ and exp(−iφ),
for every φ separately. If A and B where holonomies for an O(n)-theory in an orientifold

description this symmetry would be simply the orientifold projection itself. This suggests
that also for this U (n)-theory the dual should be some orientifold.
The radius of the dual theory is expected to be 1/(2R 0 ), half the “normal” radius. The
coordinates of the D-branes in this theory reflect the eigenvalues of the holonomies in the
original theory. Naively mapping these onto the dual circle suggests a circle of radius 1/R 0 ,
which seems to lead to a contradiction. The resolution to this paradox lies in the presence
of the operator C. If Ωi are the holonomies for a certain theory, then the holonomies
Ωi0 = gΩi g −1 with g some element of U (n) represent the same theory. Consider the set
of diagonal matrices with entries ±1, ± i on the diagonal that commute with the A and B.
Taking g to be a specific element from this set has the effect

g : (B, AC) → gBg −1 , gACg −1 = B, Ag 2 C .
This leaves Ω1 and Ω2 in standard form, but with A replaced by Ag 2 . Hence in this
construction, A and Ag 2 have to be identified. g 2 is an element of O(n) that commutes
with A. By a suitable choice of g 2 , any eigenvalue exp iφ of A can be mapped to − exp iφ =
exp i(φ + π). Therefore the periods of the circle and the eigenvalues of the holonomies
match, and the dual theory is indeed an orientifold. Now we examine the orientifold planes.
To find maximal symmetry groups we set B to either 1 or −1. If we set A = 1, then
the surviving symmetry group is the subgroup of U (n) that is invariant under C, which
is O(n). For another maximal symmetry group, assume n to be even for a moment and
take A to be of block diagonal form with 2 × 2 blocks

0 1
, (3)
−1 0
on the diagonal, and call this matrix J . The unbroken symmetry group is then the subgroup
of U (n) of matrices U that commute with J C. C transforms U → U ∗ , but as U is unitary,
U ∗ = (U −1 )T , where T is for transposed. We may then rewrite the invariance condition to
U T J U = J, (4)
which, together with the unitarity condition defines the symplectic group. It is obvious how
to generalise to arbitrary holonomies, and odd n: a holonomy with k blocks diag(1, 1) and
k 0 blocks (3) gives rise to O(k) × Sp(k 0 )-symmetry, completed with some U (m)-factors,
whenever m eigenvalues not equal to ±1, ± i coincide.
The above U (n) theory on a Klein bottle is thus T-dual to an orientifold T 2 /Z2 , where
two of the four orientifold planes are of O − -type and two are of O + -type. The holonomy
B = ±1 distinguishes two parallel configurations of one O + - and one O − -plane, whereas
in the theory on the Klein bottle it distinguished the two parallel cross-caps. As a rule
of thumb one may therefore state that the dual of the cross-cap is a configuration of one
O + - and one O − -plane. This fits with the usual charge assignments: opposite charges for
the O + - and O − -plane versus no charge for the cross-cap. The original theory may have
had isolated D-branes at the cross-caps. In the dual theory the isolated branes should be
located at the O − -planes, since the O + -planes cannot support isolated branes. Examining
the holonomies that will lead to such a situation indeed shows this to be the case.
These ideas are independent of whether D-branes are static in the background, or used as
“probes”. The above orientifold background is identical to a IIB-orientifold encountered
in 24, but consistency requires absence of D-branes [26,7]. This suggests to regard this
model as a “U (0)-theory” with a holonomy that includes complex conjugation. Its duality
to IIA on a Klein bottle is obvious from the above. Considering various limits one may
also reach other theories discussed in [24] and [7].
We interpreted the Klein-bottle theory as created by combining a translation with an
outer automorphism (complex conjugation). Outer automorphisms can always be divided
even if not combined with a translation. Dividing a U (n) group by its outer automorphism
will give a symplectic or orthogonal theory, where the ambiguity comes from the fact that
an outer automorphism may be combined with an inner automorphism to give another
outer automorphism. In our case one may consider, instead of C, an operator AC with A
an element of U (n). One should require A to commute with C and therefore A ∈ O(n).
Consistency also requires that AC acting on the group squares to the identity. The group
action on itself is always in the adjoint representation, and hence we have the possibilities
A∗ A = (A−1 )T A = ±1, so A is either symmetric or antisymmetric. One may now copy a
standard textbook derivation [17] to show that this leads to either symplectic or orthogonal
groups. The reasoning is parallel to that for orientifolds, so one may interpret the intro-
duction of an orientifold plane as quotienting the gauge group by an outer automorphism.
With this point made, which is not stressed in the literature, we may also say that in the
above a translation is combined with an orientifold action, as the theory in the last chapter
of [24] was originally motivated.
3. Twisted boundary conditions on the 2-torus
3.1. Twist in unitary groups
The U (n)-gauge group allows a second form of twist. The circle also has discrete shift
symmetries by angles 2πqR, with q a rational number, which can be chosen on the interval
[0, 1). Choose a configuration of the n D-branes that respects one or some of these shifts.
In that case the number q is a multiple of 1/n. Now compactify on a second circle with
a holonomy that includes the shift over 2πqR. This results in a theory on the 2-torus, not
with n D-branes, but with k = gcd(qn, n) branes, wrapped n/k times around the torus.
This theory is naturally interpreted as a U (n)-theory with twisted boundary conditions
[10]. We will not have much new to say on this theory, but mention it for completeness,
and to point out some effects that are encountered in other theories as well.
Let X1 and X2 be the coordinates transverse, respectively parallel to the branes. Then
this 2-torus is R2 with coordinates (X1 , X2 ), quotiented by a lattice generated by the
vectors
e1 = 2π(qR1 , R2 ), e2 = 2π(R1 , 0). (5)
Now transform to an SL(2, Z)-equivalent form. Let n0 = n/k. Then n0 and qn0 are integer,
and gcd(qn0 , n0 ) = 1. Hence the equation
n0 a + qn0 b = 1, a, b ∈ Z
has a solution, which can be found using Euclid’s algorithm. The solution is not unique as
a → a + mqn0 , b → b − mn0 with integer m gives another solution. Use this arbitrariness
to select a b such that 0 6 b < n0 . Then change the fundamental domain of the torus by
using the SL(2, Z) transformation
0 0
x n b x
= , (6)
y0 −qn0 a y
where (x, y) ∈ Z are coordinates for the lattice vectors xe1 + ye2 . Under this transforma-
tion the basis vectors transform as
2π(qR1 , R2 ) → 2π(0, n0 R2 ), 2π(R1 , 0) → 2π(R1 /n0 , bR2 ). (7)
On this fundamental domain only k D-branes (which are in a sense configurations of
n0 -tuples of branes) are visible. This is analogous to the two different representations of
the Klein bottle in the previous section. The set-up is the one considered in [8], which is
argued to lead to a Yang–Mills theory on a non-commutative torus in a suitable limit.
We may T-dualise our original theory back to an open string theory (with Neumann
boundary conditions) on a torus, using the standard methods [9]. The resulting theory has
a non-zero B-field (with B = q in appropriate units) in the background, as our original
theory does not live on a square torus.
Combining the discrete shift over 2πqR, with the Z2 reflection does not lead to anything
new. The resulting transformation is of the form
X → −X + 2πqR,
which is just a Z2 -reflection, but with other fixed points. This is related to the U (n) theory
of the previous section by a trivial translation.
3.2. Twist in symplectic groups on the 2-torus
Symplectic groups can be realised in string theories by combining the Chan–Paton

construction with the gauging of world sheet parity [17]. This gives a theory of unoriented
strings. To complete the description of the theory one has to prescribe how world sheet
parity acts on the Chan–Paton matrices. If the reflection of the world sheet is combined
with the action of an anti-symmetric matrix on the Chan–Paton indices, the resulting theory
will have symplectic gauge symmetry.
Compactifying this theory on a circle of radius R1 and T-dualising leads to an oriented
string theory, living on an interval I = S 1 /Z2 of size (2R1 )−1 , bounded by two O + -planes.
For an Sp(k)-theory there will be k D-brane pairs distributed along the interval. The two
O + -planes do not allow any freely acting shift. We will instead assume that the D-branes
are distributed in a configuration that is invariant under the reflection that exchanges the two
O + -planes. For odd k one brane-pair is fixed in the middle of the interval. Now compactify
on another circle of radius R2 with a holonomy that implements the Z2 -reflection. The
resulting compactification manifold is a Möbius strip with an O + -plane as edge. For k
Fig. 2. The Möbius strip: a double cover of the Möbius strip, arrows indicating the direction of
identifications, fat lines the edges (middle); the standard representation of the Möbius strip, obtained
by taking the lower half of the double cover as fundamental domain (right); the cylinder with one
cross-cap, obtained by taking the left half as fundamental domain (left).
even half of the D-branes are exchanged with the other half on going around the circle.
For k odd half of (k − 1)/2 pairs are exchanged with another (k − 1)/2 and one brane-pair
is fixed by the Z2 reflection. Another representation of the Möbius strip, is a cylinder of
diameter 2R2 and length (4R1 )−1 . One end of the cylinder ends in a cross-cap, the other
end is formed by a single O + -plane. On the cylinder there are k/2 D-branes pairs for k
even, and (k − 1)/2 for k odd in which case there is a brane-pair stuck at the cross-cap.
This theory is a U (2k)-theory as described in Section 2 with an extra orientifold plane
inserted. The mirror symmetry of the orientifold plane turns the Klein bottle into a Möbius
strip. Take the circle that is dual to the circle of radius R1 and choose coordinates as
follows: we will take the orientifold planes at X = 0 and X = π/R1 , and the fixed points of
the Z2 -reflection at X = π/2R1 and X = 3π/2R1 . The description from the U (n) theory
has to be slightly modified, as the fixed points of the Z2 are no longer located at X = 0 and
X = πR, as before. The action of the Z2 -reflection can be interpreted in the original theory
as accomplished by the operator (−1)C, which is complex conjugation combined with
multiplying by (−1). In symplectic theories one projects onto states invariant under J C,
with J the matrix composed of 2 × 2 blocks of the form (3), and the invariance condition
is (4). In the orientifold projected theory, the operator (−1)C is identified with (−1)Ad J ,
which has as action “conjugate with J and multiply with −1”. Multiplying by −1 is not
an outer automorphism of Sp(k) (in fact, the symplectic groups do not possess any outer
automorphism at all), and it can be realised by conjugation, as we will show later.
With the appropriate symmetries realised, we can pass from the U (n)-theory to the
symplectic theory as follows. We argued that the U (n)-theory had as its holonomies
(Ω1 , Ω2 ) = (B, AC). Replace the operator C by (−1)C, and then perform the orientifold
projection. The resulting holonomies are then (Ω1 , Ω2 ) = (B, A(−1)Ad J ). Ω1 and Ω2
do not commute, but anticommute. Their eigenvalues can be read of from B and A, but we
have to find a way to implement the action of −1.
Anticommutativity of the holonomies is allowed in symplectic theories, provided
all representations of Sp(k) have trivial centre (this is the case for all representations
one encounters in Sp(k) string perturbation theory. These are the adjoint, which is
the symmetric two-tensor; a (k(2k − 1) − 1)-dimensional representation which is the
antisymmetric tensor with an extra singlet removed; and the singlet). This theory may also
be analysed by the methods of [19]. Here we will reproduce the results of such an analysis
by a different method.
The T-dual theory to the Möbius strip is an orientifold T 2 /Z2 , with the size of the T 2
being (2R1 )−1 × (2R2 )−1 (one fourth of the usual size, compare with [24]). At the four
fixed points we find orientifold fixed planes. The original O + -plane splits into two O + -
planes intersecting the torus at a point. The cross-cap will dualise into one O + -plane and
one O − -plane, so we have a total of 3 O + -planes and 1 O − -plane. On the dual we have
k/2 D-brane pairs at arbitrary positions if k is even. If k is odd, there are (k −1)/2 D-branes
whose positions can be chosen freely. The remaining brane pair was stuck at the cross-cap,
so in the dual picture there is an isolated brane at an orientifold plane, which should be
the O − .
The corresponding holonomies can be read of as follows. A brane pair in the bulk
has two coordinates, and each corresponds to four eigenvalues λi , −λi , λ−1 −1
i , −λi with
λi = exp(2πiXi /Ri ), Xi and Ri being the coordinate and the radius of the corresponding
dimension (the O − -plane is located at (X1 /R1 , X2 /R2 ) = (1/4, 1/4), the remaining O +
at (0, 0), (1/4, 0), (0, 1/4)). Corresponding to these eigenvalues we have 2 × 2 blocks on
the diagonals of the holonomies of the form

λ1 0 0 −λ2
, , (8)
0 −λ1 −λ2 0
where the left block appears in one of the holonomies and the other, resulting from
multiplying a diagonal block with a block of the form (3) in the other holonomy. There
is a second set of blocks with (λ1 , λ2 ) replaced by (λ−1 −1
1 , λ2 ). For a single brane located at
−
the O -plane we get blocks with (λ1 , λ2 ) = (i, i). One easily verifies that this prescription
leads to anticommuting elements in the fundamental representation of the symplectic
group.
On the orientifold T 2 /Z2 one should introduce a B-field which is half-integer
valued [27]. For orthogonal groups this is well known, and it is usually deduced from a
path-integral argument [20]. It may also be deduced from duality. The Möbius strip we
used may be described as the torus T 2 quotiented by the lattice generated by 2π(0, 2R2)
and 2π(R1 /2, R2 ), quotiented by an orientifold action that takes (X1 , X2 ) → (−X1 , X2 ).
Omitting the orientifold for a moment, we see that the torus is skew, implying a half-integer
value for the B-field in its dual [9]. The same reasoning applies to a Möbius strip, where the
edge is formed by an O − -plane instead of O + -plane. This corresponds to an orthogonal
theory without vector structure, as described in [24], and reproduced by our analysis later.
The resulting orientifolds describe the moduli space of compactifications of Sp(k)
theories with twisted boundary conditions. As a check consider the cases k = 1 and k = 2,
since as Sp(1)/Z2 = SU(2)/Z2 = SO(3), and Sp(2)/Z2 = SO(5) these results should be
reproduced by other orientifolds. Sp(1) with twist corresponds to an orientifold with 3
O + -planes, and a single D-brane stuck to the O − -plane. The resulting configuration allows
no continuous gauge freedom, in accordance with the standard description of SU(2) with
twist. The single D-brane at the O − fixed plane gives O(1) = Z2 residual symmetry; this
should be interpreted as the symmetry of the centre of SU(2) which is the only symmetry
of SU(2) that survives the twist.
For k = 2 the dual description consists of a single D-brane-pair on the orientifold with 3
O + - and 1 O − -planes. The rank of the unbroken group is 1, and generically it is U (1). At
the O − -plane this is enhanced to O(2), while at any of the three O + -planes it is enhanced
to Sp(1) = SU(2). We will see in the next section that this nicely agrees with the orientifold
description of the O(5) orientifold corresponding to the Z2 -twisted case.
For higher k the analysis is similar. For k even the generic unbroken group is U (1)k/2 ,
which can be enhanced to U (k/2) at a generic position at the orientifold, to Sp(k/2) at one
of the three O + -planes or O(k) at the O − point. For k odd this analysis can be copied
while replacing k by k − 1, with the exception that at the O − O(k) symmetry is possible
because of the brane already present there.
3.3. Twist in orthogonal groups on the 2-torus
Twist in the orthogonal groups gives a more involved situation and we can distinguish
several possibilities. Every orthogonal group has a two-fold cover, so the resulting Spin-
group has at least a Z2 centre. Compactification on a two torus with twist in this Z2 will
lead to absence of “spin-structure”: fields in the spin representation are not allowed since
the holonomies will not commute in this representation.
For SO(N)-theories with N -odd this is all one can do apart from compactification with
periodic boundary conditions. For N even, SO(N) already has a non-trivial centre and
the above mentioned Z2 is just a subgroup of the whole centre. For N divisible by 4, the
centre of Spin(N) is Z2 × Z2 . Z2 × Z2 allows three Z2 subgroups (basically each of the Z2
factors, and a diagonal embedding). Of these two are related by the outer automorphism
of the Spin(N)-groups with N even. Hence there are two options for twisting by a Z2 :
the already above mentioned Z2 leading to compactification without spin structure, and
a second one, named “compactification without vector structure”. The latter is named so
because in this compactification the vector representation is not an allowed one.
For Spin(N) with N even but not divisible by 4, the centre is Z4 . The previously
mentioned Z2 is generated by the order 2 element in Z4 . It is also possible to twist by
an element of Z4 generating the whole centre. We will call this compactification without
vector structure, since in this case the vector representation is not an allowed one either.
Note however that this twist forbids any representation with a non-trivial centre, so the
spin representation should be absent as well. The only representations allowed in this case
are conjugate to the adjoint. The results from this section may also be derived with the
methods of [19]. We will follow a different route.
3.3.1. No spin structure

Absence of spin-structure does not forbid the vector representation, so one can use an
ordinary orientifold T 2 /Z2 with 4 O − -planes. The topological non-triviality has to be
treated with the technique of Stiefel–Whitney classes, following the appendix of [24].
Absence of spin-structure implies that the second Stiefel–Whitney class is non-vanishing.
We will keep on demanding that the first Stiefel–Whitney class vanishes (which implies
SO(N)-symmetry, not only O(N)), and hence the total Stiefel–Whitney class should
be w = 1 + ω2 , with ω2 the 2-form on the 2-torus. For SO(N) with N odd, this is
accomplished by placing 3 branes at the three non-trivial orientifold fixed points, placing
(N − 3)/2 pairs at arbitrary points. For SO(N) with N even one distributes 4 branes over
all orientifold fixed points and has (N/2 − 2) pairs at arbitrary locations. These are the
only solutions.
It is instructive to compare the cases of SO(3) and SO(5) without spin structure to
the analysis for Sp(1) and Sp(2) with twist, as Spin(3) = Sp(1) and Spin(5) = Sp(2).
For SO(3), 3 isolated D-branes are located at three orientifold planes, and there is no
continuous gauge freedom at all, just like in the Sp(1) case. There are discrete symmetries
O(1)3 = Z32 , but these just correspond to the 8 diagonal O(3)-matrices. Of these, 4 are not
elements of SO(3), and of the remaining 4, 3 lift to elements that anticommute with the
holonomies in Spin(3) = Sp(1). Only the identity remains, which lifts to the two centre
elements of Sp(1), which is how the Z2 discrete symmetry there is recovered.
For SO(5), we have 3 orientifold planes occupied by one brane each, and a pair of branes
at an arbitrary point. Generically the unbroken symmetry is U (1), which can be enhanced
to O(3) at any of the three points where an orientifold plane with brane is sitting. At the
remaining orientifold plane U (1) is enhanced to O(2). Stressing again that SO(3) is the
double cover of Sp(1), we see that this is exactly the same as the Sp(2) orientifold with
twist.
Further easy examples are SO(4) and SO(6) without spin structure. For SO(4) there is
no residual gauge symmetry. The Z2 associated with vector structure acts as twist in both
SU(2)-factors of Spin(4), eliminating all gauge freedom. SO(6) without spin structure is
equivalent to Spin(6) = SU(4) with Z2 -twist. From the above description this gives a rank 1
subgroup, which can be enhanced to O(3) at 4 orientifold-planes. This coincides with the
SU(4)-description, where SU(2) is a maximal symmetry group.
3.3.2. No vector structure

The case of absence of vector structure was already analysed by Witten [24] for O(4N).
Here we present an analysis from a different point of view, which also nicely extends to
the case of O(4N + 2).
Compactifying a string theory with orthogonal gauge symmetry SO(2k) on a circle,
and T-dualising along this circle, gives a theory on the interval I = S 1 /Z2 . The interval
is bounded by two O − -planes, and on the interval we have k pairs of D-branes. We will
assume that the O − -planes do not contain any isolated D-branes, since if both of them
would be occupied we do not have SO(2k) but O(2k)-symmetry, and if only one of them
would be occupied this would indicate O(n)-symmetry with n odd (actually, for n odd the
following construction is impossible, which is a reflection of the fact that SO(n) with n
odd allows only one kind of twist).
Again the only possible discrete symmetry is Z2 reflection symmetry, and we will
henceforth assume that this is realised. Compactifying on an extra circle, with a holonomy
implementing this reflection leads again to a theory on the Möbius strip, this time with
O − -planes on the boundary. We again go to the representation in which the Möbius strip
is a cylinder, bounded on one end by an O − -plane, and on the other end by the cross-cap.
Notice that for k odd, there is a pair of D-branes fixed by the reflection and, on the cylinder
it has to be located at the cross-cap. Half the number of the remaining D-branes are visible
in this representation. Since we are restricting to SO(n)-configurations, we can repeat the
whole discussion presented for symplectic groups, with the difference that the orientifold
planes we insert here will not give symplectic but orthogonal symmetry. From the
geometric picture one may again deduce anticommutativity of the holonomies Ω1 and Ω2 .
We may now T-dualise as before, and obtain an orientifold with two O − -planes from
the original O − -plane, and an O + - and O − -plane from the cross-cap. This explains
the relation between the IIA-theory on a Möbius strip, that is discussed in [16], and
the IIB orientifold in [24]. Both feature in a network of theories describing duals of
the CHL-string [3]. The strong coupling limit of IIA theory on the Möbius strip gives
M theory on a Möbius strip, which has another weak coupling description as a heterotic
E8 × E8 string, with a holonomy interchanging the two E8 -factors [4]. Via heterotic
duality, this corresponds to a compactification of the Spin(32)/Z2 -string without vector
structure as described in [15,24]. In the latter paper, the strong coupling description of the
heterotic Spin(32)/Z2 -string without vector structure is derived to be the IIB-orientifold
with a single O + - and 3 O − -planes. The Z2 projection causing the reduction of rank
of the gauge group is geometrical in the IIA and M theory descriptions (compare with
similar constructions in [5,6,18]). The T-duality derived here closes the circle, and relates
the mechanism for rank reduction in the IIB-theory to the more transparent one of the
IIA-theory.
We may represent the parts of the holonomies corresponding to branes in the bulk in
the same way as in the symplectic case. These can be conjugated to matrices in the real
vector representation of O(n). Readers who prefer the real representation of O(n) should
substitute for each brane in the bulk the following 4 × 4 blocks
 
cos φ1 0 − sin φ1 0
 0 − cos φ1 0 sin φ1 
 ,
 sin φ1 0 cos φ1 0 
0 − sin φ1 0 − cos φ1
 
0 − cos φ2 0 sin φ2
 − cos φ2 0 sin φ2 0 
  (9)
 0 − sin φ2 0 − cos φ2 
− sin φ2 0 − cos φ2 0
as can be derived straightforwardly. The parameters φi = Xi /2πRi follow from the
coordinates of the D-branes on the torus.
For k odd, there was an odd number of D-brane pairs at the cross-cap, and hence the O − -
plane coming from the cross-cap has to contain an odd number of branes. The other two
orientifold points also contain isolated branes. This is possible because the even number of
branes that should be on the edge of the Möbius strip (to ensure SO(2k)-symmetry), may
translate into odd numbers of branes at each of the corresponding O − -planes in the dual
theory.
Consider a single brane stuck to the O − -plane that came from dualising the cross-cap.
As can be seen from the geometric picture, this corresponds to a block diag(i, −i) in the
holonomy Ω1 , or equivalently, a block

0 −1
1 0
in the real representation of O(n). Demanding anticommutativity with a 2 × 2 block in the
holonomy Ω2 leads to the unique solution diag(1, −1) (in the real representation, up to
conjugation with an element of SO(2)). One may also consider a single brane stuck to one
of the other O − -planes. This defines a block diag(1, −1) in the holonomy Ω1 . Demanding
anticommutativity with a second 2 × 2 block leads to two inequivalent possibilities, being

0 −1 0 1
, .
1 0 1 0
These two possibilities correspond to the two O − -planes that came from dualising the
original O − -plane. The point is now that occupying 1 or 2 of the O − -planes by an odd
number of branes corresponds to holonomies in O(n), but occupying all three at once
with an odd number of single branes does give holonomies in SO(n). This also gives the
interpretation for the orientifold with 3 O − -planes and one O + -plane where not all of the
O − -planes are occupied; these represent O(k) configurations that cannot be represented
in SO(k). We see that if we demand SO(n)-symmetry, and occupy one O − -plane with an
odd number of branes, we have to occupy all O − -planes by an odd number of branes.
On the resulting dual orientifold we have k/2 pairs of D-branes if k is even, or (k − 3)/2
if k is odd. For k even we obtain back the description of [24], with the possibility of O(k)
symmetry at 3 planes, and Sp(k/2)-symmetry at 1 plane. For k odd we have the possibility
of O(k − 2) at three planes, and Sp((k − 3)/2) at one plane.
It is again instructive to look at a few examples. k = 1 is impossible in SO(2). k = 2
corresponds to SO(4)-theory without vector structure. Since Spin(4) is SU(2) × SU(2),
this corresponds to twist in one of the SU(2) factors, and arbitrary holonomies in the
other SU(2)-factor. In the orientifold description, we have the possibility of enhanced
symmetries O(2) and Sp(1). Sp(1) = SU(2) obviously corresponds to the unbroken second
factor. The O(2)’s correspond to situations in which the holonomies in the second factor
are (1, iσ 3 ), (iσ 3 , 1), (iσ 3 , iσ 3 ). The extra parity transformation is due to the fact that iσ 3
anticommutes with the elements iσ 1,2 , which lifts to the double cover as commutation
symmetry.
k = 3 gives SO(6) without vector structure. This corresponds to Spin(6) = SU(4) with
Z4 -twist. The absence of remaining gauge freedom is completely in agreement with the
SU(4) description.
k = 4 gives SO(8) without vector structure, which due to triality should be equivalent
to SO(8) without spin structure. We certainly do find Sp(2) = Spin(5) symmetry in both
descriptions. O(4) is less visible in the above description of SO(8) without spin structure,
but this is due to the fact that the parity in O(4) is actually not a real symmetry (compare to
the O(2) symmetry found for k = 2), and Spin(4) = SU(2) × SU(2) can also be obtained
outside orientifold fixed planes. The asymmetry in the two description is then due to the
fact that they represent different projections from the same moduli space.
4. Compactifications on a 3-torus
4.1. Commuting triples for orthogonal and symplectic groups
For compactifications on a 3-torus with periodic boundary conditions, the vacua are
classified by its three holonomies, which should be 3 commuting elements in the gauge
group.
Finding such a triple for a symplectic theory amounts to placing a number of D-brane
pairs on an orientifold T 3 /Z2 where the fixed points are all O + -planes. A similar thing
can be done for orthogonal theories, with the difference that on the fixed points of the Z2
orientifold action one introduces O − -planes, which can also support single D-branes. Not
all such configurations correspond to flat connections with periodic boundary conditions.
Alternative boundary conditions can be described by this orientifold, as long as the
holonomies commute in the vector representation of O(N). For a periodic connection one
should however demand that the holonomies commute in Spin(N), which is a more severe
restriction. To solve which configurations of O(N) holonomies lift to Spin(N) holonomies,
one may calculate the Stiefel–Whitney class for the configurations [24]. Only if the Stiefel–
Whitney class is trivial the configuration can be lifted to Spin(N).
Solving for these requirements Witten found new vacua for orthogonal gauge theory
with periodic boundary conditions on a 3-torus [24] whose existence was explained from
the group theory point of view in [2,11–14]. A configuration with 7 D-branes on the 7 fixed
points excluding the origin of an orientifold, and N pairs at arbitrary positions parametrises
a periodic connection for SO(2N + 7)-theory. Similarly, a configuration with 8 D-branes
distributed over all 8 fixed points, and N pairs at arbitrary positions parametrises a
periodic connection of SO(2N + 8)-theory. Besides these there are always the flat periodic
connections that are smoothly connected to the configuration where all three holonomies
are equal to the identity. For SO(N) with N even, this gives a configuration of only D-brane
pairs on the orientifold. For N odd there is a single isolated D-brane at the origin of the
orientifold. Forgetting about the D-brane pairs for a moment, we see that for each theory
there are two solutions, one in which a number k of orientifold planes is occupied by single
branes (k being 0 or 1) and one where (8 − k) orientifold planes are occupied by single
branes. This is a useful observation for later.
4.2. c-triples for symplectic groups
For symplectic groups on the 3-torus one may also choose non-periodic boundary
conditions. This can be done in various ways, but by using SL(3, Z) transformations on
the torus, all possibilities are isomorphic to one standard form. We can choose the standard
form to have twist between the holonomies in the 1 and 2 direction, and the third holonomy
commuting with the former two. Following [2], we call a triple of such holonomies a
c-triple, where c denotes that the three holonomies only commute up to a (non-trivial)
centre element of the gauge group.
From our analysis for twist in symplectic gauge theories on the 2-torus, one easily
deduces that the corresponding orientifold description has 6 O + -planes and 2 O − -planes.
The two planes with 3 O + and 1 O − , are distinguished by the eigenvalue ±1 in the
third holonomy. Eigenvalues for the third holonomy can be read of in the usual way,
with the remark that their multiplicities should be doubled. A configuration for the 2-torus
may therefore be imported in either of these planes, corresponding to choosing the third
holonomy in Sp(k) to be ±1, which are the two elements of the centre of Sp(k). One
quickly deduces that there are always 2 disconnected possibilities for placing the D-branes
in this orientifold background.
First suppose Sp(k) symmetry with k even. For the description with twist on a 3-torus,
this should give k/2 pairs of D-branes in the above orientifold background. There are two
possibilities to distribute the D-branes. First, one can have k/2 pairs at arbitrary locations
on the orientifold. But one can also split one pair, put one D-brane on one O − -plane and the
other on the other O − -plane, and have the remaining (k/2 − 1)-pairs at arbitrary locations.
Both possibilities are legitimate, since Sp(k) is simply connected. The conclusion is thus,
that Sp(k)-theory on a 3-torus with twisted boundary conditions has a moduli space of 2
components, one with a rank k/2 unbroken gauge group, and one with a rank k/2 − 1
unbroken gauge group. We can now perform a Witten index count for this theory, as also
performed in [2]. The two components will contribute k/2 + 1 and k/2 to the index giving
the total value k + 1 in agreement with both the periodic boundary conditions case, and the
infinite volume case [23].
For k is odd the procedure should also be clear. One can place (k −1)/2 pairs of D-branes
at arbitrary points in the orientifold background. The single D-brane that is left can go on
either of the two O − -planes. These are inequivalent possibilities, and hence also in this
case the moduli space consists of 2 components. Each of these components contributes
(k + 1)/2 to a Witten index calculation, giving also the correct result k + 1 [2].
Again we check the k = 1 and k = 2 cases. According to the above, Sp(1)-theory
with twist on a 3-torus gives a moduli space of 2 components. On each component the
gauge group is completely broken. This is as it should be as Sp(1) = SU(2) which, when
compactified on a 3-torus with twist has a moduli space that looks like this. We again
have the remaining O(1) = Z2 symmetry corresponding to the centre of SU(2), which
commutes with everything.
Perhaps more interesting is the Sp(2)-case, where we have one component for which the
gauge group is completely broken, and another where a rank 1 group survives. The rank
one gauge group is generically U (1), but can be enhanced to O(2) at two planes or Sp(1)
at six other planes. This coincides with the description we will find for SO(5) = Sp(2)/Z2 ,
without spin structure.
4.3. c-triples for orthogonal groups
For orthogonal groups on a 3-torus there are more possibilities for the boundary
conditions. Like in the case for the 2-torus, we can have absence of either spin- or
vector structure. By SL(3, Z)-transformations on the torus, we can again arrange that the
holonomies for the 1- and 2-direction are the ones that do not commute (in the Spin-cover
of the group), while the third holonomy does commute with the other two.
In the case of SO(4n) there is however a new possibility. For Spin(4n) the centre of the
gauge group is not cyclic but a product of cyclic groups, being Z2 × Z2 . Call the generator
of the first Z2 zs (with s for spin), and the generator of the second Z2 zc (c being the
standard notation for the second spin-representation). Also define zv = zs zc . This notation
is motivated by the fact that identifying zv ∼ 1 gives the vector representation.
We can now also impose the following twist conditions on the holonomies:
Ω1 Ω2 = zs Ω2 Ω1 , Ω2 Ω3 = zc Ω3 Ω2 , Ω3 Ω1 = zv Ω1 Ω3 . (10)
This can be thought of as a standard form. SL(3, Z)-transformations result in an isomorphic
moduli-space. We will call this case “spin nor vector”-structure, and treat it separately.
4.3.1. No spin structure

This is the easiest case, provided we use some previously obtained knowledge. From our
description of orthogonal theories on a 2-torus without spin structure, a particular case for
the 3-torus can be obtained as follows.
For SO(k) with k odd, place 3 single D-branes at three O − -planes within one plane
within the orientifold T 3 /Z2 leaving the fixed plane at the origin empty, and place the
others in pairs at arbitrary points at the orientifold. For k even one should place 4 single
D-branes at 4 orientifold fixed planes within one plane of T 3 /Z2 .
For k > 4 there is always a second possibility. Remember from Section 4.1 and
[24] that a configuration of 8 D-branes distributed at all orientifold planes has a trivial
Stiefel–Whitney class. We may “add” this orientifold configuration to another as follows.
Take a specific configuration of D-branes at the orientifold. This has a certain Stiefel–
Whitney class, which can be thought of as providing a topological classification for the
configuration. Now adding 8 more D-branes at the orientifold fixed points will not affect
the Stiefel–Whitney class. This is so because the Stiefel–Whitney class of the 8 D-branes
is trivial, and the Stiefel–Whitney class of the “new” configuration may be simply obtained
by multiplying the class of the “old” configuration with that of the added configuration (it
is important to realise that Stiefel–Whitney classes are Z2 valued, and that −1 = 1 mod 2,
so there is no ordering ambiguity). One may also add or delete any pair of D-branes without
affecting the class, also because of its Z2 nature.
We thus obtain the following possibilities: For SO(k) with k odd, we had 3 single
D-branes at three O − -planes. Adding the 8 D-branes and reducing modulo 2, we obtain a
configuration of 5 D-branes with the same topological classification as the previous one.
The 5 D-branes are precisely at the orientifold planes that were not occupied previously,
and in a sense one could speak of a Z2 -complement. One can add pairs of D-branes to
again obtain an SO(k) configuration.
For k even one had 4 single D-branes at 4 O − -planes. Taking the Z2 -complement, we
get an inequivalent configuration with 4 single D-branes at the other 4 O − -planes, with
the same topological classification. Of course, afterwards we must add pairs of D-branes
to acquire SO(k).
We will now discuss several cases. k = 3 corresponds to SO(3) = SU(2)/Z2 with twist
on the 3-torus. The SU(2) description has two components. The SO(3)-description has also
two components, but these cannot be distinguished by their holonomies. In a particular
representation, the SU(2) holonomies read
Ω1 = iσ3 , Ω2 = iσ1 , Ω3 = ±1, (11)
but ±1 in SU(2) are both projected to the same element of SO(3) being the identity.
k = 4 gives SO(4) which gives two orientifolds, but some thought will reveal that also
in this case there are twice as many components in moduli space. Using that Spin(4) =
SU(2) × SU(2), the no-spin-structure condition amounts to twisting both SU(2) factors
simultaneously. For the third holonomy one has then 4 possibilities, being any combination
of plus or minus the identity in each SU(2)-factor. These 4 possibilities project to only two
sets of holonomies in SO(4), and hence two orientifold descriptions.
k = 5 gives us SO(5) which is interesting because we should be able to reproduce the
Sp(2)-results here. SO(5) without spin-structure gives two orientifolds. On one we have
5 fixed D-branes and hence no residual gauge symmetry. On the other we have 3 fixed
D-branes and a pair wandering freely. Possible enhanced gauge symmetries are O(3) at
three points, and O(2) at five points. However, all but one of the “parity” symmetries
(corresponding to elements with det = −1) in these O(n) groups are “fake” in the
sense that they correspond to elements that anticommute in Spin(5). The remaining O(2)
corresponds to the O(2)’s we encountered in the Sp(2) case, and the SO(3)’s map to the
Sp(1)-unbroken subgroups in Sp(2). That the multiplicities of these enhanced symmetry
groups are only half of those encountered in the Sp(2) description reflects the fact that
SO(5) is a double cover of Sp(2), which also translates to the fact that the moduli space
of Sp(2)-triples is a double cover of the space of SO(5)-triples. The moduli space for the
gauge theory is the moduli space of Sp(2)-triples, as every set of SO(5) holonomies has
two inequivalent realisations in terms of gauge fields. Note however that here the number
of components in moduli space agree; the two SO(5)-components are double covers of two
Sp(2) components, not of 4 Sp(2) components.
k = 6 gives SO(6) whose spin cover is SU(4). Here there are two equal dimension
components in moduli space, both with a rank 1 gauge group which can be enhanced
to SO(3) at 4 points.
It is easy to perform the Witten index count for k > 5 [2]. In these cases we have always
two components of the moduli space and two corresponding orientifold representations.
For k even, both components contribute k/2 − 1, for a total of k − 2. For k odd, one
component contributes (k − 1)/2 whereas the second contributes (k − 3)/2 for a total of
k − 2. Of course these answers are as they should be.
4.3.2. No vector structure

From the analysis for the 2-torus we deduce that O(2k) without vector structure on
a 3-torus corresponds to an orientifold background with 6 O − -planes and 2 O + -planes.
Again eigenvalues for the third holonomy can be read off in the usual way, except that
their multiplicities should be doubled. One obvious solution to the boundary conditions is
to import the solution for the 2-torus here.
For k even we have seen that a particular solution is given by placing all D-brane
pairs at arbitrary points. For the second solution we take as before the Z2 -complement.
We have not defined how the operation of “Z2 -complement” acts on the O + -planes
but this is not hard to guess. Since O + -planes cannot support isolated D-branes, they
should remain empty. Hence the second solution has all O − -planes occupied by one D-
brane, and k − 3 pairs at arbitrary points. A way to see this is as follows. The smallest
group for which the configuration with six isolated branes exists is SO(12). One can take
the SO(6) holonomies Ω1 and Ω2 that gave “no vector structure” on the 2-torus (these
are unique up to gauge transformations), to construct the SO(12)-holonomies Ω1 ⊕ Ω1 ,
Ω2 ⊕ Ω2 and 1 ⊕ −1, where 1 stands for the identity in SO(6). That these SO(12)-
matrices satisfy the required boundary conditions is obvious, and liftings to Spin(12) can
be constructed from the liftings of the SO(6)-holonomies to Spin(6) = SU(4). That SO(12)
is the smallest group allowing these extra solutions can also be deduced with the techniques
from [2].
For k is odd we have 3 O − -planes within one plane occupied by D-branes. A priori
one has two possible planes, and actually both give distinct solutions. Note that also in
these cases the solutions are each others Z2 -complement. For k odd all components of the
moduli space are isomorphic, as also follows form the analysis of [2].
SO(4) without vector structure on a 3-torus gives only one solution, since there are
simply not enough D-branes to realise the second one. This gives a rank 1 unbroken gauge
group which can be enhanced to Sp(1). A naive calculation of the Witten index would give
half of the right answer, but as before, the moduli space consists of 2 components that
cannot be distinguished by their holonomies in SO(4). We therefore have to multiply the
naive value for the index by two, again obtaining the right answer.
SO(6) without vector structure gives two solutions, but from SU(4)-analysis one
expects four. Again this is due to the fact that inequivalent solutions exist that cannot be
distinguished by their holonomies in SO(6). Notice that the gauge group is completely
broken, which is as it should be.
A Witten index calculation is straightforward for these theories. For SO(4N + 2) there
are always 4 components [2] (projected to two orientifolds) that are all isomorphic. Each
orientifold has 2N + 1 branes on it, of which 3 are stuck. The rest should organise in
N − 1 pairs, giving a rank N − 1 gauge group and contribution N to the Witten index. With
4 components one obtains the total value 4N , which is indeed the dual Coxeter number for
these theories.
For SO(4N) one has 2 components (1 orientifold), where no branes are stuck and N pairs
move freely, giving a contribution of 2N + 2 to the index. For N > 3 this is not sufficient,
but for these cases there exist 2 more components (1 orientifold) having 6 stuck branes,
and N − 3 pairs at arbitrary points. The total adds up to 2(N + 1) + 2(N − 2) = 4N − 2,

the right answer.
4.3.3. Spin nor vector structure

For SO(4k) there is the possibility of holonomies satisfying Eq. (10). In some sense this
should encompass both the case of no spin as well as no vector structure. The orientifold
background is as in the case without vector structure, T 3 /Z2 with 2 O + -planes and 6
O − -planes. On top of this 2k branes should be distributed.
A clue on the D-brane configuration can be found from T-dualising in the direction of
the line connecting the two O + -planes. This gives a theory on the product of a circle and
an orientifold T 2 /Z2 with 3 O − ’s and 1 O + -plane. This orientifold corresponded to an
orthogonal theory without vector structure on the 2-torus. In our case, the gauge group
is SO(4k), and we know that if there are branes at an orientifold plane, their number
should be even. Translating back to the orientifold T 3 /Z2 , the pair of orientifold planes
corresponding to one O − -plane in T 2 /Z2 will be occupied either both by an even number
of branes, or both by an odd number of branes. This leaves 8 possibilities, 2 of which can
be quickly discarded as they correspond to a SO(4k)-theory without vector, but with spin
structure. The remaining possibilities have thus either 2 or 4 branes stuck at O − -planes,
and hence 2k − 2, respectively 2k − 4 D-branes in the bulk.
T-dualising in another direction, along a line connecting an O + - and an O − -plane
will lead to an orientifold of the form ((T 2 /Z2 ) × S 1 )/Z2 . One can represent this by
an orientifold T 2 /Z2 with 4 O − -planes, quotiented by a Z2 -reflection in one of the
points halfway on the line between 2 O − -planes (the multiple possibilities are related by
SL(2, Z)-transformations). This reflection has a second fixed point. Over the orientifold
(T 2 /Z2 )/Z2 one erects a circle everywhere, except at the two fixed points of the Z2 -
reflection where it is replaced by a cross-cap. On the orientifold T 2 /Z2 we should have
absence of spin structure, meaning that all O − -planes are occupied by an odd number of
branes (remember that 4k is even). This translates to occupancy of both O − -planes in the
quotient (T 2 /Z2 )/Z2 . We now have two possibilities; either there are an odd number of
branes at both cross-caps, or there are an even number. For the orientifold T 3 /Z2 , these
two possibilities translate into the situations with two O − -planes occupied (cross-caps
occupied by even number of branes), and four O − -planes occupied (cross-caps occupied
by odd number of branes). Thus both possibilities can be realised.
The 3 possibilities of occupying 2 O − -planes are related by SL(2, Z), as are the 3 possi-
bilities of occupying 4 O − -planes. Only one of each set of possibilities solves the boundary
conditions (10) (as these are not SL(2, Z) invariant), and actually, the two possibilities are
each others Z2 complement, as before.
We therefore have two orientifolds representing SO(4k)-theory on a 3-torus with spin
nor vector structure. Each orientifold background is of the form T 3 /Z2 , with 6 O − -
planes, and 2 O + -planes. One orientifold has 2 O − -planes occupied, and the other has
the remaining 4 O − -planes occupied.
SO(4) with spin nor vector structure has only 2 branes on the dual orientifold,
so only one out of the two possibilities mentioned can be realised. This corresponds
to 4 components on the moduli space, all consisting of a single point. This can be seen
as follows. Spin(4) = SU(2) × SU(2), and therefore we may write Spin(4) holonomies as
SU(2)-pairs. The holonomies obeying the boundary conditions are (up to conjugation)
Ω1 = (iσ3 , iσ3 ), Ω2 = (± iσ1 , ± iσ3 ), Ω3 = (iσ1 , iσ1 ). (12)
There are four inequivalent possible choices for the signs in Ω2 (one cannot change a sign
in Ω2 by conjugation without changing some sign in the other holonomies).
For SO(4k) (k > 1) and larger groups each of the two orientifold descriptions represents
two components. Each of the two orientifold descriptions again represents 2 components.
For SO(4k), the two components contribute k, respectively k − 1 to the Witten index.
Taking into account the correct multiplicities, gives the correct answer 4k −2 for the Witten
index [2].
5. Conclusions
We have shown how to construct orientifold configurations describing compactifications

of gauge theories with classical groups on a 2- or 3-torus. Conversely these results show
how various orientifolds should be interpreted as gauge theories. The methods can be
extended to higher-dimensional tori.
Notice that for two-dimensional orientifolds T 2 /Z2 any configuration of O + - and O − -
planes is possible. For 3-dimensional orientifolds T 3 /Z2 we only find even numbers of
O + - and O − -planes. There is a simple argument why odd numbers of O + -planes are not
allowed [1].
Some of these configurations are of immediate interest for consistent string-theories.
We already mentioned that the U (n)-theory on a 2-torus, with holonomy with outer
automorphism has precisely the same background as a certain IIB orientifold, though
consistency requires the absence of D-branes, and hence the absence of gauge symmetry
(so n = 0 in a sense). For the type I string theory on a 3-torus and its duals, which are
argued to have Spin(32)/Z2 as its gauge group, it seems that there are four configurations
of interest, two describing periodic boundary conditions, and two describing absence of
vector structure. In a subsequent publication [1] it will be argued however that two of the
four configurations actually suffer a relatively subtle inconsistency, invalidating the naive
application of group theory methods in string theory.
Acknowledgements
We would like to thank Pierre van Baal and Jan de Boer for useful discussions and
reading a draft version of this article.
References
[1] J. de Boer, R. Dijkgraaf, K. Hori, A. Keurentjes, J. Morgan, D. Morrison, S. Sethi, in

preparation.
[2] A. Borel, R. Friedman, J.W. Morgan, Almost commuting elements in compact Lie groups,
math/9907007.
[3] S. Chaudhuri, G. Hockney, J.D. Lykken, Phys. Rev. Lett. 75 (1995) 2264, hep-th/9505054.
[4] S. Chaudhuri, J. Polchinski, Phys. Rev. D 52 (1995) 7168, hep-th/9506048.
[5] S. Chaudhuri, D. Lowe, Nucl. Phys. B 459 (1996) 113, hep-th/9508144.
[6] S. Chaudhuri, D. Lowe, Nucl. Phys. B 469 (1996) 21, hep-th/9512226.
[7] A. Dabholkar, J. Park, Nucl. Phys. B 477 (1996) 701, hep-th/9604178.
[8] M.R. Douglas, C. Hull, JHEP 02 (1998) 008, hep-th/9711165.
[9] A. Giveon, M. Porrati, E. Rabinovici, Phys. Rep. 244 (1994) 77, hep-th/9401139.
[10] G. ’t Hooft, Nucl. Phys. B 153 (1979) 141.
[11] V.G. Kac, A.V. Smilga, Vacuum structure in supersymmetric Yang–Mills theories with any
gauge group, hep-th/9902029 v.3.
[12] A. Keurentjes, A. Rosly, A.V. Smilga, Phys. Rev. D 58 (1998) 081701, hep-th/9805183.
[13] A. Keurentjes, JHEP 05 (1999) 001, hep-th/9901154.
[14] A. Keurentjes, JHEP 05 (1999) 014, hep-th/9902186.
[15] W. Lerche, R. Minasian, C. Schweigert, S. Theisen, Phys. Lett. B 424 (1998) 53, hep-
th/9711104.
[16] J. Park, Phys. Lett. B 418 (1998) 91, hep-th/9611119.
[17] J. Polchinski, TASI lectures on D-branes, hep-th/9611050.
[18] J. Schwarz, A. Sen, Phys. Lett. B 357 (1995) 323, hep-th/9507027.
[19] C. Schweigert, Nucl. Phys. B 492 (1997) 743, hep-th/9509161.
[20] A. Sen, S. Sethi, Nucl. Phys. B 499 (1997) 45, hep-th/9703157.
[21] A. Sen, JHEP 09 (1998) 023, hep-th/9808141.
[22] A. Sen, JHEP 10 (1998) 021, hep-th/9809111.
[24] E. Witten, JHEP 02 (1998) 006, hep-th/9712028.
[25] E. Witten, JHEP 12 (1998) 019, hep-th/9810188.
[26] C. Angelantonj, M. Bianchi, G. Pradisi, A. Sagnotti, Y. Stanev, Phys. Lett. B 387 (1996) 743.
[27] M. Bianchi, G. Pradisi, A. Sagnotti, Nucl. Phys. B 376 (1992) 365.
Noncommutative SO(n) and Sp(n) gauge theories

L. Bonora a,c , M. Schnabl a,c , M.M. Sheikh-Jabbari b , A. Tomasiello a,∗
a Scuola Internazionale Superiore di Studi Avanzati, Trieste, Via Beirut 2, 34014 Trieste, Italy
b The Abdus Salam International Centre for Theoretical Physics, Strada Costiera 11, 34014 Trieste, Italy
c INFN, Sezione di Trieste, Italy
Abstract
We study the generalization of noncommutative gauge theories to the case of orthogonal and
symplectic groups. We find out that this is possible, since we are allowed to define orthogonal
and symplectic subgroups of noncommutative unitary gauge transformations even though the gauge
potentials and gauge transformations are not valued in the orthogonal and symplectic subalgebras
of the Lie algebra of antihermitean matrices. Our construction relies on an anti-automorphism of
the basic noncommutative algebra of functions which generalizes the charge conjugation operator
of ordinary field theory. We show that the corresponding noncommutative picture from low energy
string theory is obtained via orientifold projection in the presence of a non-trivial NSNS B field.
 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 11.15.-q

Keywords: Noncommutative field theories; Orthogonal and symplectic gauge groups; Orientifold
1. Introduction
It is well-known that a constant NSNS two form B field background can be gauged away
in the perturbative type II string theories. The addition of D-branes drastically modifies this
conclusion. The novel point is that the components of a constant B field background which
are parallel to a Dp-brane can not be gauged away anymore [1–5]. Studying the open
strings ending on D-branes with B field turned on, it has been shown that the worldvolume
of these branes become noncommutative [2–8]. In addition, by computing open string
states scattering amplitudes and extracting the massless poles contributions, one can show
that the low energy effective theory describing the system is the noncommutative U (1)
(NCU(1)) theory [9,10].
E-mail addresses: bonora@sissa.it (L. Bonora), schnabl@sissa.it (M. Schnabl), jabbari@ictp.trieste.it
(M.M. Sheikh-Jabbari), tomasiel@sissa.it (A. Tomasiello).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 2 7 - 7
462 L. Bonora et al. / Nuclear Physics B 589 (2000) 461–474
In the zero B field background we know that the low energy effective theory of n
coincident Dp-branes is a (p + 1)-dimensional supersymmetric (with 16 supercharges)
U (n) theory. The U (1) part of this U (n) basically represents the interactions of D-brane
open strings with the bulk closed strings (supergravity fields). This U (1) part, which
is usually called the center of mass U (1), decouples from the open strings dynamics
and hence effectively we find an SU(n) theory. However, for n coincident D-branes
in a constant B field background, the above argument is modified by the fact that the
brane worldvolume is a noncommutative Moyal plane. In this case we deal with a
noncommutative version of U (n) theory, namely, NCU(n). This theory is obtained by
replacing the usual products of fields by the star (Moyal) product [11]. However, in this
case separating the center of mass (noncommutative) U (1) is impossible. Intuitively, this
corresponds to the fact that when we turn on a B field, the open string left and right movers
(holomorphic and anti-holomorphic parts) contribute unequally, but in the closed strings
sector (bulk fields) left and right movers always appear on an equal footing. So, in view
of the open–closed strings interactions, we cannot decouple U (1) modes. They contribute
a part, which, as expected, depends on the background B field. Apart from this string
theoretic reasoning, one can also understand this point through a gauge theory argument.
Suppose we start with the usual SU(n) gauge theory and make it noncommutative by
replacing the products with star products. Then one can show that this theory will not
be consistent: the noncommutative gauge transformations will not close to form a group.
In other words, by performing a noncommutative SU(n) gauge transformation we create
extra terms which can consistently appear only in a noncommutative U (n) theory [12].
As one can see from the above discussion, trying to define a noncommutative gauge
theory corresponding to subgroup of U (n) and a string/brane theory configuration that
corresponds to it, does not look very promising, at first sight. However we will show below
that, to a certain extent, it is possible to find consistent noncommutative extension for
gauge theories corresponding to certain subgroups of U (n). The main observation is that
it is possible to define gauge transformations that close to form a subgroup of the group of
NCU(n) gauge transformations even though the corresponding gauge potentials and gauge
transformations are not valued in a classical Lie subalgebra of the unitary Lie algebra u(n).
We handle this problem from two different points of view, one relying on purely gauge field
theory considerations and the other on string theory. From the gauge theory point of view,
we show that it is possible to impose constraints on the gauge potentials and the gauge
transformation so that when the deformation parameter vanishes we recover the ordinary
orthogonal and symplectic gauge theories. An essential role in defining the constraints is
played by an operator which is a generalization of the charge conjugation operators for
gauge theories.
From a string theory point of view, the analogous operator is that which in the ordinary
cases (without background B field) is the worldsheet parity (possibly plus spatial parity)
which is responsible for orientifold projections. So, a natural framework from a string
theory point of view seems to be an orientifold in the presence of D-branes with a B field
switched on.
L. Bonora et al. / Nuclear Physics B 589 (2000) 461–474 463
The paper is organized as follows. In Section 2, we briefly review the noncommutative

gauge theories and show that, in general, the concept of noncommutative gauge group
should be modified compared to commutative theories. We will argue that one should
abandon the naive picture of connections taking their values in a Lie algebra of the
gauge group. The generalization we need is based on an anti-automorphism, r, in the
corresponding C∗ -algebra of functions. We show that this r map together with the matrix
transposition is a realization of the charge conjugation operator we need to extract SO and
Sp parts out of U (n).
In Section 3, we present our string theoretic arguments. First we discuss the issue of
orientifold planes with a particular step function-like B field, which passing through the
O-plane changes the sign and is zero on the plane. Putting Dp-branes in this background
parallel to Op-planes, we study the supersymmetries preserved by this system and we show
that it is stable. Then, we proceed to find the low energy effective action for open strings
attached to such Dp-branes. We show that actually it coincides with the results of our field
theory arguments. Last section is devoted to remarks and further discussion.
2. SO(n) and Sp(n): noncommutative realization
2.1. Preliminary discussion
Before we describe our proposal, it is useful to illustrate what are the problems
connected with the realization of noncommutative orthogonal and symplectic groups. This
subsection is pedagogical in nature and the reader may wish to skip it. The definition
and treatment of noncommutative orthogonal and symplectic groups start in the next
subsection.
First, let us discuss what we mean by gauge group in noncommutative geometry. Since
we have in mind the case of Rd , let us use the Moyal approach (even though some of the
remarks which follow are independent of this assumption) and define our starting complex
algebra Aθ as the vector space C ∞ (Rd ) endowed with product
i ← →
µν
f ∗ g(x) ≡ f (x)e 2 θ ∂µ ∂ν g(x). (2.1)
Let now A be the connection (the rank is not necessarily 1; the indices are understood),
which transforms under gauge transformation as AU = U −1 ∗ (d + A∗)U , where U −1 ∗
U = 1. The hermiticity condition A† = −A is preserved if U † = U −1 ; these U ’s are called
unitary automorphisms (of the module on which the gauge theory is constructed), but they
are not functions on Rd valued in U (n). It is true instead that the connection A and the
infinitesimal gauge transformations λ are u(n)-valued functions on Rd . Indeed, they are
both antihermitean: being δA = dλ + A ∗ λ − λ ∗ A, the antihermiticity condition on A is
preserved if also λ† = −λ holds, thanks to the relation
(f ∗ g)† = g † ∗ f † , (2.2)
which is a generalization of the ∗-algebra property to matrix-valued functions. For the
same reason F , the field strength,
Fµν = ∂[µ Aν] + Aµ ∗ Aν − Aν ∗ Aµ , (2.3)

is antihermitean too.
Now, when dealing with orthogonal and symplectic groups, something worse happens.
The connection is no longer a function from Rd to Lie(G); as we briefly discuss below,
this is impossible to accomplish. But, what we said above should convince the reader that
in noncommutative this is not such a dramatic loss: finite gauge transformations are just
unitary automorphisms of the module, and their infinitesimal counterparts are U (n)-valued
for the simple reason that there is no product in the antihermiticity condition. Our theory
is a noncommutative gauge theory which reduces in the commutative limit to the desired
one, and, what is most important, seems to have a physical origin in string theory.
Let us now discuss in some detail why one has to abandon the Lie(G)-valuedness of
the gauge potential. From the description above, one would hope to generalize SO(n) and
Sp(n) gauge groups in a simple way: just replace the complex (because of the i in (2.1))
algebra we started with by a real (respectively, quaternionic) one. Let us illustrate such an
approach with the example Sp(1) = SU(2). Could we find a deformation of the algebra
A ≡ C ∞ (Rd , H) of functions from Rd to H (quaternions), we would be done. Elements of
A can be expressed as f = fi τi . The idea of tensoring the usual Moyal product (2.1) with
the matrix one (a definition like (fi τi ) ∗ (gj τj ) ≡ (fi ∗ gj ) τi τj ) is clearly too naive: this
would not be a product in A, since it does not close. If we start, indeed, from real fi and
gj , we get a complex fi ∗ gj (recall the i in (2.1), which cannot be dropped because it is
needed for the property (2.2) and hence for gauge symmetry).
Alternatively one can exhibit true products. The easiest try would be to consider the i
in (2.1) as one of the imaginary quaternions i, j , k. In this way the property (2.2) remains
valid, but what is lost is associativity. Although this initial try is wrong, it illustrates the
idea. However, we have checked to first order that there is no associative deformation
compatible with (2.2). Even more, suppose one wants to take a pragmatic point of view
and accept to live with non-associative products. Then one may look directly for deformed
Moyal brackets which satisfy Jacobi identity. It turns out that at first order such brackets
do not exist.
In conclusion, if we want to find a realization of noncommutative orthogonal and
symplectic groups, we seem to be obliged to give up the familiar idea of Lie(G)-valued
connections.
2.2. Noncommutative SO(n) and Sp(n)
We are now ready to illustrate our proposal for the generalization of orthogonal and
symplectic group.
2.2.1. The r anti-automorphism

To start with, we will work in a setting in which θ has to be thought of as a parameter.
Accordingly, we will consider Aθ as an algebra of (possibly formal) power series in θ .
This algebra has an anti-automorphism r defined by
(·)r : f (x, θ) 7→ f r (x, θ) ≡ f (x, −θ). (2.4)
This map reduces to the identity on the generators x µ and reverses the order in the product:
µ µ µ µ
(x1 ∗ · · · ∗ xn )r = (xn )r ∗ · · · ∗ (x1 )r .
2.2.2. Orthogonal and symplectic constraints

First of all, we consider our groups as subgroups of U (n). In other words we
keep the usual antihermiticity condition on the u(n)-valued connections A and gauge
transformations λ. To fix our conventions we will use Greek letters for spacetime indices
and i and j for matrix (group) indices. Here, for later use, we write down explicitly the
hermiticity condition:
A∗ij (x, θ) = −Aj i (x, θ),

λ∗ij (x, θ) = −λj i (x, θ). (2.5)
Our defining condition for the NCSO(n) connections and gauge transformations is to
take the gauge connections and transformations satisfying the following constraints:
Arij (x, θ) = −Aj i (x, θ),

λrij (x, θ) = −λj i (x, θ). (2.6)
Let us comment on these constraints. First of all, it is easy to see that they are preserved
by gauge transformations. One can see it componentwise. Alternatively, rewrite (2.6) in
the concise form A = −(At )r and λ = −(λt )r , i.e., t is the matrix transposition. Define
( (·)t )r ≡ (·)rt ; one can show that the rt map enjoys the (2.2) property, with rt replacing
†. The proof is now formally similar to the usual one for U (n): (λ ∗ A − A ∗ λ)rt =
Art ∗ λrt − λrt ∗ Art = −(λ ∗ A − A ∗ λ).
The second comment we wish to make is that the constraints we introduced are natural
if one recalls that in noncommutative gauge theories the map −(·)rt is nothing but complex
conjugation; our theory is the charge-conjugation invariant version of the usual one. More
explicitly, as discussed in [13], indeed the charge conjugation operator is
Ac = −Art. (2.7)
One can write an explicit solution of (2.6) as:

1 1
Aµ = Aµ − Art
µ = Aµ + Acµ . (2.8)
2 2
This notation may be ambiguous and we hasten to specify that when (2.8) is used we
understand that Aµ transforms with gauge parameter Λ = 12 (λ − λrt ). 1 More precisely,
our Aµ enjoys the noncommutative gauge transformations generated by Λ:
Aµ → A0µ = U∗−1 (Λ) ∗ Aµ ∗ U∗ (Λ) − U∗−1 (Λ) ∗ ∂µ U∗ (Λ), (2.9)
where
1 In the ordinary commutative case, this is the way to ‘reduce’ a unitary connection to an orthogonal one, [14],
Proposition 6.4.
1
U∗ (Λ) ≡ 1 + iΛ − Λ ∗ Λ + · · · ,
2
U∗−1 (Λ) = U∗ (−Λ), U∗−1 ∗ U∗ = 1. (2.10)
As we see it is immediate that our NCSO(n) gauge fields are charge conjugation
invariant.
Thirdly, we anticipated above that under the (2.6), connections and gauge parameters do
not turn out to be so(n)-valued. Nevertheless (2.6) introduces restrictions on the matrix
functions Aij . To see what they are, let us write (2.6) more explicitly
Aij (x, θ) = −Aj i (x, −θ),
λij (x, θ) = −λj i (x, −θ). (2.11)
Inserting a power expansion in θ for A
µ µνρ
Aµ (x, θ) = A0 (x) + iθνρ A1 (x) + · · · , (2.12)
we see that (2.6) implies that A0 , A2 , . . . , are antisymmetric and A1 , A3 . . . , symmetric.
The hermiticity condition (2.5) imposes that all the coefficients A0 , A1 , . . . , be real. The
same conclusions hold for the power expansion of λ.
Up to now, A0 , A1 , . . . , are unrestricted, except for the just mentioned constraint.
However, if we want to make connection with string theory, A1 , A2 , . . . , are expected not
to introduce new degrees of freedom, but to be functionally dependent on A0 . The simplest
proposal is to regard them as given by the Seiberg–Witten map [11]:
µ i µ µ
Aµ (A0 ) = A0 − θ νρ A0ν , ∂ρ A0 + F0ρ + O θ 2 (2.13)
4
(the presence of i is due the fact that Seiberg and Witten use hermitean connections rather
than antihermitean ones, as we do). This is indeed consistent: the term linear in θ is
symmetric if the constant part is antisymmetric. In fact, one can also see that the next
term is antisymmetric, and so on; so we have complete accord with (2.12).
This is related to the further subtle issue of fixing θ to a particular value. In this case, of
course the approach we have taken so far — considering θ as a formal parameter — loses
its validity, and the very definition of r is in jeopardy. However, thanks to the fact that
A1 , A2 , . . . , depend on A0 , even when one puts θ to a particular value, A is not the most
general U (n) field; our constraint becomes more involved but is still there. If we invert the
map to obtain A0 (A), the constraint can be formulated simply as
A0 (A) = −At0 (A). (2.14)
So we could say that our theory is the image of the Seiberg–Witten map restricted to the
SO(n) case.
It is now easy to introduce similar definitions for noncommutative Sp(n). One imposes
in this case the condition J Ar = −At J , where J = ⊗ Idn , where = iσ2 . This constraint
is preserved by gauge transformations that satisfy the same condition.
One could think that the group SU(n) could be tackled in a similar way: by defining a
constraint like Tr(A + Art ) = Tr(A + Ar ) = 0. However, this would not be gauge invariant.
So, even by using the r map, it is not possible to define a NCSU (n) gauge theory.
2.2.3. Field theory

To define a Yang–Mills NCSO(n) theory, let A = A(x, θ) satisfy the constraint (2.6).
The action is the usual one
Z
1 µν
S =− d d x Fij Fj iµν , (2.15)
4
where F is defined as
Fµν = ∂[µ Aν] + Aµ ∗ Aν − Aν ∗ Aµ . (2.16)
The action (2.15) is naturally gauge invariant under NCSO(n) and positive. It reduces to
the usual one for SO(n) in the θ = 0 case.
It is rather straightforward to introduce matter fields in this context in a coherent way.
For example, suppose we want to introduce fermions in the adjoint representation. Let us
consider a generalization of the Seiberg–Witten map to such fields.
Let ψ0 be an ordinary spinor in the adjoint representation, which therefore under an
ordinary gauge transformations transforms as follows
δλ0 ψ0 = [ψ0 , λ0 ]. (2.17)
Then it is reasonable to postulate the following noncommutative gauge transformation for
the corresponding noncommutative field:
δλ ψ = ψ ∗ λ − λ ∗ ψ, (2.18)
where λ = λ0 + λ0 (λ0 , A0 ), A = A0 + A0 (A0 ) and ψ = ψ0 + ψ 0 (ψ0 , A0 ); the primed fields
are first order in θ . We want to find a function ψ(ψ0 , A0 ) which transform as (2.18) when
the corresponding ψ0 transform as (2.17). This amounts to satisfying the equation
ψ(ψ0 , A0 ) + δλ ψ(ψ0 , A0 ) = ψ(ψ0 + δλ0 ψ0 , A0 + δλ0 A0 ). (2.19)
The solution to first order in θ is
i
ψ(ψ0 , A0 ) = ψ0 − θ µν {A0µ , ∂ν ψ0 } + 12 {[ψ0 , A0µ ], A0ν } + O θ 2 . (2.20)
2
It is easy to see that our noncommutative orthogonal constraint
ψ rt = −ψ (2.21)
is consistent with this map. Therefore such spinors form a representation of NCSO(n).
In a similar way one can introduce also the fundamental representation. The action terms
containing these matter fields are the usual one with ordinary product replaced by the
noncommutative one, and will not be written down here.
3. Orientifolds and B field
We want now to derive the gauge theory we described above from a brane configuration
in string theory in the limit α0 → 0.
In the commutative case, gauge theories with orthogonal or symplectic groups are
realized as low energy effective actions of branes on orientifold planes in type I theories.
Since noncommutativity is achieved by a non-zero B field, and this vanishes on the

orientifold plane, one may deem our search hopeless. As an example let us consider type
IB theory, in which case the gauge groups of the branes are the ones we are looking for.
However a standard argument tells us that the B field is absent, except for a quantized
background [15–18]; for a comment on the quantized B and noncommutativity see [19].
Our proposal however is more subtle.
For the sake of definiteness, we will consider type IA theory. This can be obtained in
two ways: as T-dual of type IB, or as an orientifold of type IIA theory. In the second way,
it is from the very beginning a 10d theory if the initial IIA is; in the first, it is of course
compactified in at least one direction, and one can make contact with the other approach
by sending this radius to infinity. Either way, we obtain a 10D theory (and no quantization
on B). First we start with a brief review of some issues in IA theory. The symmetry of IIA
theory by which one orientifolds is P9 · Ω, where P9 : x 9 → −x 9 is a spacetime reflection
and Ω : σ → π − σ is worldsheet parity [20]. So, the orientifold plane is an eight-plane
located at x 9 = 0. Physics in the x 9 > 0 region is locally the same as the IIA one. However,
strings always have an image on the other side, and as a consequence spacetime fields are
reflected as

Φ x 1 , . . . , x 8 , x 9 = ±Φ x 1 , . . . , x 8 , −x 9 , (3.1)
the sign is determined by the Ω parity. So, in particular, the RR charges of image branes
have a relative ± according to their dimensionality. To obtain the gauge groups we are
looking for, namely SO and Sp groups, we have to put branes and their mirrors on the
orientifold plane (otherwise gauge symmetry would be broken to [U (n/2)]2 ); so, as far
as we are concerned only branes whose mirrors have the same RR charge survive — the
others meet their antibranes and annihilate. The surviving ones are 0-, 4-, and 8-branes;
the gauge group on them is SO, Sp and SO, respectively. From the low energy effective
theory point of view, as stated in the literature, [25], the orientifold projection corresponds
to charge conjugation operator.
Having specified that we have branes stuck on the orientifold plane, let us now analyze
more in detail the consequence of a (special) background B field. Since we are really
interested in its components parallel to the orientifold, we set Bµ9 = 0, µ = 1, . . . , 8. As
for the remaining components, bearing in mind that the B field is odd under worldsheet
parity [21] from (3.1) we learn that Bµν (x 1 , . . . , x 8 , x 9 ) = −Bµν (x 1 , . . . , x 8 , −x 9 ). So, we
will consider a configuration Bµν = bµν f (x 9 ), where f is odd in x 9 . It is certainly true
that the B field is zero on the orientifold, so this would seem hopeless; but strings which
end on the branes can stretch also outside, and the usual statement that their interaction
with B is a boundary term is, in general, true only when B is constant. The interaction
term equals (Σ is the worldsheet of the open string)
Z Z Z

Bµν dx ∧ dx = d Bµν x dx − dBµν ∧ x µ dx ν
µ ν µ ν
Σ Σ Σ
Z Z
= Bµν x µ dx ν − ∂ρ Bµν dx ρ ∧ x µ dx ν . (3.2)
∂Σ Σ
In the usual B = constant case, the first term of the final expression is the boundary term
which is responsible for noncommutativity, while the second vanishes. In our case, the
situation is different: the first term is now zero, due to the vanishing of the B field on the
orientifold plane (branes are on the orientifold, so ∂Σ ⊂ O8), but the second is
Z
∂9 f (x 9 )bµν x µ dx ν ∧ dx 9 . (3.3)
Σ
We choose f to be the step function (x 9 ), which is in fact the easiest field configuration
one can think of in this case. The factor ∂9 f (x 9 ) becomes a δ(x 9 ), and this makes the
integral to reduce of dimensionality and concentrate on the orientifold {x 9 = 0}:
Z
bµν x µ dx ν . (3.4)
Σ∩O8
Now, as Σ ∩ O8 ⊃ ∂Σ, this provides a boundary term which has exactly the form of the
one which usually accounts for noncommutativity.
This situation would seem at first to be strangely discontinuous with respect to what
happens as soon as one separates branes; in that case, one would be tempted intuitively to
consider boundary conditions which vary. For instance, for a string going from one brane
to its image, one would write boundary conditions (gµν ∂n + (y)bµν ∂y )x ν = 0, where
the worldsheet is to be thought of as the upper half plane, y is a coordinate on the real
axis, and ∂n denotes normal derivative. This is not right: one has to remember that both
P9 and Ω have been applied. In detail: the ΩP9 symmetry is expected to leave the closed
string background Bµν = (x 9 )bµν invariant; to do so, since x 9 is reversed, the parameter
b should be reversed as well. The operation z → −z̄, b → −b is compatible with the usual
boundary conditions (gµν ∂n + bµν ∂t )x ν = 0, and not with the naive ones.
One can see that all our arguments can also be extended to the case with lower-
dimensional orientifold planes. On the contrary they cannot be extended to the type IB
case. However we do not exclude that noncommutative gauge theories may arise in this
case too.
3.1. Supersymmetry argument
In this subsection we give another argument, based on supersymmetry, for the stability
of a system of Dp-branes on top of an orientifold plane (Op-plane with p 6 8) in the
presence of a step function-like B field. To illustrate our reasoning we first consider the
zero B field case.
Suppose that QL and QR represent the 32 supercharges of type II theories, i.e., QL
and QR have the same (or opposite) ten-dimensional chiralities for IIB (or IIA) theory.
Introducing a Dp-brane, half of these supersymmetries are preserved. The corresponding
conserved generators are given by the following linear combination of QL and QR , [21]
Q = L QL + R QR , (3.5)
where
Γ 01...p L = R . (3.6)
Now, let us consider an Op-plane parallel to the Dp-brane. To study the stability of
this system one way is to check whether the supercharges (3.5), or a portion of them,
are preserved under the orientifold projection. The Op-plane we are interested in is
characterized by invariance under ΩPT operator, where PT is the parity in all the directions
transverse to the Op-plane. Under Ω, QL and QR , which correspond to supersymmetries
of closed string left and right movers are interchanged. On the other hand under PT ,
Eq. (3.6) is reversed, namely Γ 01...p R = L . So, altogether
(ΩPT ) Q = Q.
In other words the presence of the parallel Op-plane does not break supersymmetry any
further and exactly the same supersymmetry as for a Dp-brane is preserved. Hence the
whole system is stable. One can extend the above argument to an Op-plane parallel to a
D(p − 4)-brane. Again this system is stable (it preserves some supersymmetry) however
in this case 8 supercharges survive.
For the cases with non-zero B field, we follow a similar discussion. For definiteness let
us consider a rank one B field, which is non-zero along (p − 1)th and pth directions;
generalization to other cases is straightforward. The portion (half) of supersymmetry,
which is preserved by the Dp-brane, is given by [22]

1 b
Γ 01...p √ −√ Γ p−1,p L = R . (3.7)
1 + b2 1 + b2
Now, we introduce the Op-plane, which again acts by ΩPT projection. Noting that
(i) Γ 01...p and (1 − Γ p−1,p b) are commuting,
(ii) (1 − Γ p−1,p b)(1 + Γ p−1,p b) = (1 + b2 )1 and
(iii) under ΩPT , b is reflected to −b,
ΩPT , acting on (3.7), will lead to the same equation with L and R interchanged.
Therefore, there are 16 supercharges invariant under the ΩPT transformation. Hence, our
system formed by parallel Dp-branes in the background B field introduced earlier, in the
presence of a parallel Op-plane, is stable.
The above argument can also be understood in a more intuitive way. A Dp-brane
in a b field (say bp−1,p ) background can be treated as a bound state of a Dp- and
D(p − 2)-branes, with (p − 2)-branes having their worldvolume along the 01 . . . p − 2
directions, while bp−1,p represents their distribution density in the p − 1 and p directions
[23,24]. If we put this system in front of an Op-plane, as above, (p −2)-branes are reflected
to anti-(p − 2)-branes; however since at the same time we also change the b to −b, the
image of a (p, p − 2) bound state remains the same object, which is known to preserve 16
supercharges.
Another argument (from which in fact (3.7) could follow) that supersymmetry is
preserved can be obtained via a T-duality transformation in a direction parallel to the
branes. Suppose for example, in the O8 case, that our p-branes extend in the 3 and 4
directions, and that there is a non-vanishing b34 ≡ b field. Let us perform a T-duality in the
direction 3; by considering the boundary conditions before and after the transformation, it
is easy to see that the branes get transformed to lower-dimensional tilted branes, which
extend in the direction x 4 − bx 3 . It is less trivial to understand what happens to the
orientifold plane: to see what happens, we have to consider the transformation ΩP9 and
interpret it in terms of the dual coordinates xD :
T3 µ
xµ xD
ΩP9 ΩP\
T3
ΩP9 x µ ΩP9 x µ D
.
The map ΩP\ is what we have to find, and is what defines the orientifold on the dual
side. Let us consider the action on x 3 ; in this case xD 3
= x 3 (z) − x 3 (z̄). The map ΩP9
acts on the original expansion by sending z → −z̄, b → −b (and of course x 9 → −x 9 ),
3
and the ΩP\ map, on xD , does the same but with an extra overall minus sign. This is not,
however, the map ΩP9 P3 . On this side of the duality, ΩP9 P3 (contrary to ΩP\ ) would
indeed not touch b, since it is a number not a field as seen from ΩP9 P3 : it is in fact the
angular coefficient by which the brane is tilted. To understand what the map really is, we
have to act on the mixed coordinates x 3 + bx 4, x 4 − bx 3 . These have an expansion which
contains only pure combinations (z−n − z̄−n ), (z−n + z̄−n ), respectively, and so in terms
of the latter it is easily seen that ΩP\ is ΩP9 Px 3 +bx 4 . This means that the orientifold plane
is tilted along the x 4 − bx 3 coordinate, and so is parallel to the brane. The mirror brane is
now necessarily also parallel to them, and the whole system preserves supersymmetry.
3.2. Correlation functions
To show that actually the field theories we described above arise as low energy effective
actions on branes, we will now follow similar steps to the usual U (n) case.
Low energy effective actions are found by computing scattering of strings corresponding
to the various effective fields. In our case, incoming states have to be accompanied by
orientifold projectors (1 + ΩPT )/2. This means that any vertex V|si has to be changed in
the combination 1/2(V|si + VΩPT |si ), where the second term is the vertex that creates the
ΩPT |si state. Let us specialize to the gauge bosons, in which case the vertex, in the −1
picture, is V (z) = ξ ij · (ψ + ψ̄)eikx , where we define ξ ij µ ≡ ξ µ λij . In the B = 0 case,
since ΩPT (ξ · b−1/2|0, ki) = −ξ t · b−1/2|0, ki, correlation functions become

1/2(V − V t )(y1 ) . . . 1/2(V − V t )(yn ) , (3.8)
and give the usual result: amplitudes are obtained by substituting in the usual ones ξ with
ξ − ξ t , i.e., keeping only the antisymmetric part of ξ .
In the present case, the analogy keeps working: now correlation functions after
orientifolding are obtained from the ones before by the rule ξ → ξ −ξ rt . Thus, for instance,
for the gauge three point function the result is proportional to

Tr (ξ1 − ξ1rt ) · p2 (ξ2 − ξ2rt ) · (ξ3 − ξ3rt ) + (ξ2 − ξ2rt ) · p3 (ξ3 − ξ3rt ) · (ξ1 − ξ1rt )
i 1 µν 2
+(ξ3 − ξ3rt ) · p1 (ξ1 − ξ1rt ) · (ξ2 − ξ2rt ) e− 2 pµ θ pν + (1 ↔ 2); (3.9)
inner products are understood with respect to the open string metric. This is the same
amplitude one finds starting from a noncommutative gauge theory, but with the additional
constraint ξ = −ξ rt ; thus it coincides with the field theory we have suggested.
As in theories arising from a non-orientifold case, there are two descriptions of the
system that are equivalent at least in a perturbative sense, one is noncommutative and
the other commutative, [11]; the commutative one in the present case is an ordinary SO(n)
gauge theory. So it is reasonable that, as we said, our theory is the image of a commutative
SO(n) gauge theory under the Seiberg–Witten map.
4. Discussions and remarks
In this paper we have tackled the problem of noncommutative gauge theories, for
groups other than U (n). We have argued that to obtain the noncommutative extension of
a gauge theory, in general, the usual interpretation of gauge symmetries as local internal
symmetries should be modified. More precisely one must focus on an “NC Lie-algebra of
gauge transformations”, rather than on the corresponding algebra of spacetime independent
transformations. Elaborating on the NCU(n) group of gauge transformations, we have
showed that actually one can extract some NC-subgroups of it. In particular we have
discussed NCSO(n) and NCSp(n) gauge theories. As it is clear from our construction,
in the commutative limit, θ → 0, we recover the usual SO and Sp theories. Physically our
method is based on the charge conjugation operation in gauge theories. Noting the fact
that the SO(n) subgroup of the commutative U (n) theory can be constructed by simply
choosing the gauge field to be in the charge conjugation invariant subgroup, it follows
that one must restrict the gauge transformations to the same subgroup. The same idea also
works for the noncommutative case. But, of course, in the NCU(n) case one has to consider
the proper charge conjugation operator [13]. Therefore, as we see, for the particular case of
NCSO(2), this theory is not equivalent to NCU(1), although in the commutative case they
are the same theory. The main difference between these two may be that the first, NCSO(2),
is invariant under the charge conjugation, but the other is not. So, in this way it seems
more reasonable to consider the (NCSO(2) × fermions) as the proper noncommutative
version of QED. From this example we also learn that given a commutative gauge theory,
its noncommutative extension is not unique. Another special case is NCSp(1). This theory
can be treated as the noncommutative version for an SU(2) gauge theory, and since there is
no consistent way of finding noncommutative deformed SU theories in general, NCSp(1)
seems to be very interesting.
From the string theory side of our construction, since the orientifold projection
corresponds to the charge conjugation operator in the low energy effective theory, we
have guessed that a string theory environment that might give rise to effective NCSO(n)
and NCSp(n) theories should contain an orientifold plane in a B field background. Then,
calculating the corresponding scattering amplitudes, we have checked that this is indeed
the case. This provides a further support to our definition of NCSO(n) and NCSp(n) gauge
theories.
Naturally there are lots of interesting issues related to these theories, e.g., studying
renormalizability, which we do not deal with here but postpone to future research.
However, we would like to end this paper with some remarks. The first concerns chiral
anomalies. In [26–28] the problem of anomaly cancellation in noncommutative U (n)
Yang–Mills theories was analyzed. It was shown that anomaly cancellation can occur only
by matching anomaly coefficients from opposite chiralities. The main reason (although not
the only one) is that the U (1) factor does not decouple as in ordinary theories, therefore
we cannot define a noncommutative SU(n) theory (as discussed in the introduction).
This fact motivated in part the research reported in this paper. One question we would
like to answer is: do there exist noncommutative (non-U (n)) gauge theories in which a
more subtle cancellation mechanism for anomalies exists, with, in particular, anomaly-free
representations for chiral fermions? In this paper we have introduced new noncommutative
gauge theories with more general gauge groups. However, as far as anomaly cancellation
is concerned, the situation is no better, for example, in orthogonal theories than in U (n)
theories. The reason is that connections and gauge transformations have a non-trivial
symmetric part. Therefore we are led back to the same conclusion as for noncommutative
U (n) theories. It would seem that noncommutative Yang–Mills theories are definitely more
anomalous than ordinary theories.
A second remark concerns the implications of SO(n) being the Lorentz group in
commutative n-dimensional spacetimes. Recently, its noncommutative extension has been
considered by gauging the NCU(1, D − 1) instead of the corresponding SO symmetry
[29]. As a result, a complexified gravity theory was found. Using our definition of
noncommutative SO(n) gauge theory one can take, in regards to this problem a different
attitude. One can address the formulation of gravity theories on the Moyal plane, by using
the NCSO(1, D − 1) gauge group of transformations. This may lead to a more reasonable
gravity theory, since it must correspond to the usual gravity theory when θ → 0. However,
we expect that again in this case we will deal with some complexified gravity.
Acknowledgements
We are grateful to M. Bianchi, C.S. Chu, E. Gava and T. Krajewski for useful
discussions. M.M. Sh.-J. would like to thank N. Nekrasov for helpful remarks and
K. Narain and A. Kashani-Poor for discussions. The research of L.B., M.S. and A.T. was
partially supported by EC TMR Programme, grant FMRX-CT96-0012, and by the Italian
MURST for the program “Fisica Teorica delle Interazioni Fondamentali”. The work of
M.M. Sh.-J. was partly supported by the EC contract no. ERBFMRX-CT 96-0090.
References
[1] E. Witten, Bound states of strings and p-branes, Nucl. Phys. B 460 (1996) 335, hep-th/9510135.
[2] Y.-K.E. Cheung, M. Krogh, Noncommutative geometry from 0-branes in a background B field,
Nucl. Phys. B 528 (1998) 185, hep-th/9803031.
[3] F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, Mixed branes and M(atrix) theory on noncommu-
tative torus, hep-th/9803067.
[4] F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, Noncommutative geometry form strings and
branes, JHEP 02 (1999) 016, hep-th/9810072.
[5] C.-S. Chu, P.-M. Ho, Noncommutative open strings and D-branes, Nucl. Phys. B 550 (1999)
151, hep-th/9812219.
[6] F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, Dirac quantization of open string and noncommu-
tativity in branes, Nucl. Phys. B 576 (2000) 578, hep-th/9906161.
[7] C.-S. Chu, P.-M. Ho, Constrained quantization of open strings in background B and
noncommutative D-branes, Nucl. Phys. B 568 (2000) 447, hep-th/9906192.
[8] M.M. Sheikh-Jabbari, A. Shirzad, Boundary conditions as Dirac constraints, hep-th/9907055.
[9] M.R. Douglas, C. Hull, D-branes and noncommutative torus, JHEP 9802 (1998) 008, hep-
th/9711165.
[10] M.M. Sheikh-Jabbari, Super Yang–Mills theory on noncommutative torus from open strings
interactions, Phys. Lett. B 450 (1999) 119, hep-th/9810179.
[11] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 09 (1999) 032, hep-
th/9908142.
[12] A. Armoni, Comments on perturbative dynamics of noncommutative Yang–Mills theory, hep-
th/0005208.
[13] M.M. Sheikh-Jabbari, Discrete symmetries (C, P , T ) in noncommutative field theories, hep-
th/0001167, to appear in PRL.
[14] S. Kobayashi, K. Nomizu, Foundations of Differential Geometry, Vol. I, New York, 1963.
[15] M. Bianchi, A. Sagnotti, Phys. Lett. B 247 (1990) 517.
[16] M. Bianchi, A. Sagnotti, Nucl. Phys. B 361 (1991) 519.
[17] M. Bianchi, E. Gava, J.F. Morales, K.S. Narain, D-strings in unconventional type I vacuum
configurations, Nucl. Phys. B 547 (1999) 96, hep-th/9811013.
[18] Z. Kakushadze, Geometry of orientifolds with NSNS B-flux, hep-th/0001212.
[19] C.-S. Chu, Noncommutative open string: neutral and charged, hep-th/0001144.
[20] E.G. Gimon, J. Polchinski, Consistency conditions for orientifolds and D-manifolds, Phys. Rev.
D 54 (1996) 1667, hep-th/9601038.
[21] J. Polchinski, TASI lectures on D-branes, hep-th/9611050.
[22] M.M. Sheikh-Jabbari, Noncommutative super Yang–Mills theories with 8 supercharges and
brane configurations, hep-th/0001089.
[23] M.M. Sheikh-Jabbari, More on mixed boundary conditions and D-branes bound states, Phys.
Lett. B 425 (1998) 48, hep-th/9712199.
[24] H. Arfaei, M.M. Sheikh-Jabbari, Mixed boundary conditions and brane-string bound states,
Nucl. Phys. B 526 (1998) 278, hep-th/9709054.
[25] M.B. Green, J.H. Schwarz, E. Witten, Superstring theory, Cambridge Monographs on
Mathematical Physics, Cambridge Univ. Press, Cambridge, 1987, p. 469.
[26] F. Ardalan, N. Sadooghi, Axial anomaly in noncommutative QED on R4 , hep-th/0002143.
[27] J.M. Gracia-Bondia, C.P. Martin, Chiral gauge anomalies on noncommutative R4 , Phys. Lett.
B 479 (2000) 321, hep-th/0002171.
[28] L. Bonora, M. Schnabl, A. Tomasiello, A note on consistent anomalies in noncommutative YM
theories, hep-th/0002210.
[29] A.H. Chamseddine, Complexified gravity in noncommutative spaces, hep-th/0005222.
How to integrate divergent integrals:

a pure numerical approach to complex loop
calculations
F. Caravaglios 1
LPT, Bât. 210, Univ. Paris-Sud, F-91405, Orsay, France
Received 6 April 2000; accepted 20 June 2000
Abstract
Loop calculations involve the evaluation of divergent integrals. Usually [G. ’t Hooft, M. Veltman,
Nucl. Phys. B 44 (1972) 189] one computes them in a number of dimensions different than four where
the integral is convergent and then one performs the analytical continuation and considers the Laurent
expansion in powers of ε = n − 4. In this paper we discuss a method to extract directly all coefficients
of this expansion by means of concrete and well defined integrals in a five-dimensional space. We
by-pass the formal and symbolic procedure of analytic continuation; instead we can numerically
compute the integrals to extract directly both the coefficient of the pole 1/ε and the finite part.  2000
1. Introduction
Feynman diagrams computations are often a lengthy and hard task. The complexity
of typical computations grows as a factorial with the number of external legs and/or
loops involved in the processes. In the Standard Model the main difficulties arise from
QCD corrections and the production of jets in the final states; but it is not difficult to
imagine extensions of the Standard Model, in which this problem could seriously limit
our capabilities of studying and understanding new physical phenomena; in particular if
they are involved (in addition to αs ) by new rather strong couplings, which give rise to
multiparticle/multiloop amplitudes. This issue motivates a rich and increasing activity.
Considerable progresses have been achieved in the past [2]: new methods inspired by
string theory [3–7], helicity amplitudes [8–10], different recursive algorithms to calculate
QCD dual amplitudes [12,13], or the scattering matrix elements of generic processes
with arbitrary initial/final states [14]. Very useful approximations have been proposed
1 francesco.caravaglios@th.u-psud.fr
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 9 1 - 6
476 F. Caravaglios / Nuclear Physics B 589 (2000) 475–486
[15–19], when the exact matrix elements are unknown. In loop computations interesting
simplifications [20–32] can be used.
While analytical methods can provide us in some specific processes with very simple
results, a general approach for generic Lagrangian and final states cannot avoid the use
of the computer power. A possible strategy is to implement the Feynman rules into some
automatic code with the help of some familiar packages for symbolic manipulation of
algebraic formulae. However, this way to proceed becomes ineffective, when the computer
has to manage very complex formulae (as often it is the case).
This suggests a second strategy which tries exploiting the computer power with pure
numerical algorithms and thus avoiding lengthy symbolic manipulation. In the recent past
these algorithms have been successfully used for the computation of several tree level
amplitudes both in QCD [33,34] and in electroweak processes [35–37]. In this paper we
discuss the possibility to apply pure numerical techniques [38] in loop calculations.
The problem of performing loop computations can be divided in two parts. The first
part is similar to tree level computations [14]: from a given Lagrangian one has to
produce an explicit function of virtual and real momenta which is equivalent to the sum
of all the Feynman diagrams contributing to the process. Once this task is accomplished,
loops calculations have an additional problem with respect to tree level ones: integrals
over virtual loop momenta are affected by ultraviolet and/or infrared divergences. These
divergences require the introduction of a regularization procedure which usually obliges
us to perform all virtual loop integrations in an analytical and essentially symbolic fashion
(while in tree level calculations one usually performs the integration over the real momenta
by numerical Monte Carlo methods). This is a serious difficulty in any numerical approach.
Here we address this last issue. We will define a numerical procedure to perform virtual
loop integrations even when these include divergences.
Let us consider a function f [lµ , pµ ] of the external momenta pµ and of the virtual
momenta lµ . The n-dimensional integration [1] is performed in a region of complex values
of n where the integral is convergent; then one considers the analytical continuation to
values of n where the integral is not well defined. In general for n ' 4 the analytical
continuation can have a pole 1/ε (ε = n − 4). We can use the Laurent expansion around
ε ' 0 to define some functionals Ik [f ] such that 2
Z
dn l f [lµ ] = I−1 [f ]ε−1 + I0 [f ] + I1 [f ]ε1 + I2 [f ]ε2 + · · · , (1)
Since the n-dimensional integration is well defined for any function f , also the functionals
Ik [f ] are unambiguously defined from Eq. (1). For instance, I0 [f ] coincides with the
ordinary four-dimensional integral
Z
d4 l¯ f [lµ ] (2)
if f is a convergent function. While the left-hand side of the (1) is essentially a formal
definition which does not lead us to an obvious numerical integration procedure, each
2 At one loop.
F. Caravaglios / Nuclear Physics B 589 (2000) 475–486 477
Ik [f ] can be written in terms of well defined and convergent integrals. In the following we
show how to build explicit integral representations for the functionals Ik [f ]. In the next
sections, we explain two different definitions (even if equivalent) of the Ik [f ]; in the final
part some very simple numerical examples will be given in order to emphasize the practical
purposes of the method.
2. Method of integration A
Let us assume that we have to perform an integral in n > 4 dimensions. The virtual
momentum lµ has n components. We call l¯µ the components with µ 6 4 and l˜µ the
components with µ > 4. Then we can rewrite the integral
Z Z Z
d l f [lµ , pµ ] = d l f [lµ , pµ ] d l = d4 l¯ f [lµ , pµ ]l˜ε−1 dl˜ dΩε .
n 4¯ ε˜
(3)
dΩε is the solid angle in the subspace (µ > 4) of dimension ε orthogonal to the four-vector
l¯µ . Note that f [lµ , pµ ] is invariant under rotations in this subspace, since the external
momenta do not have components µ > 4. We can exploit this invariance as follows. First
we rotate the vector l˜µ such that l˜5 6= 0 and lñ = 0 for any n > 5. Thus we can write lµ =
√
l¯1 , l¯2 , l¯3 , l¯4 , t with t = (l5 )2 + (l6 )2 + · · · + (ln )2 and omit all the components with
µ > 5. We have reduced our n-dimensional space into a five-dimensional space. Second,
we can perform the integral in dΩε and we obtain from Eq. (3)
Z∞Z
π ε/2 4¯
d l f [lµ , pµ ] t ε/2−1 dt. (4)
Γ 2ε
0
Now the integral in dt can be integrated by parts

Z∞ ε Z
π ε/2 4t 1+ 2 4¯ d2
d l [lµ , pµ ] dt. (5)
Γ 2ε ε(2 + ε) dt 2
0
ε
The function f depends on t through the fifth component of lµ . The function t 1+ 2 has a
cut in the positive real axis of t (Re(t) > 0 and Re(t) = 0), in fact the rotation t → tei2π
gives t 1+ 2 → t 1+ 2 +iπε . Therefore, we can apply the following identity
ε ε
Z∞ I
ε 1 ε
t 1+ 2 f [t] dt = t 1+ 2 f [t] dt, (6)
1 − eiπε
0
where the contour of the last integration in t must contain all the poles and singularities
lying in the complex t plane and must avoid the cut on the positive real axis (see Fig. 1).
Then the integral (5) above becomes
I ε Z
1 π ε/2 4t 1+ 2 4¯ d
2
d l f [l ,
µ µp ] dt. (7)
(1 − eiπε ) Γ ε2 ε(2 + ε) dt 2
Now we can make the expansion in powers of ε (the Euler constant and log(π) have been
reabsorbed into a redefinition of the scale µ as in the MS) and we obtain the final result
I Z
it 2 d2
+ −1 + log[−t] d4 l¯ 2 f [lµ , pµ ] dt. (8)
2π ε dt
In the above expression each integral is now convergent, due to the action of the derivative
d2 /dt 2 appearing inside the integral in d4 l. ¯ Thus one can separately calculate the singular
and the non singular part, simply evaluating well defined and convergent integrals. We can
check the result above in a specific example and see how the above formula can be used
for practical calculations.
Suppose that we have to evaluate the following divergent integral
Z
l2
dn l 2 . (9)
l + m2
In our approach lµ is not a n-dimensional object but it is a more concrete five-dimensional
√
vector with components (l¯1 , l¯2 , l¯3 , l¯4 , t ); thus l 2 = lµ l µ = l¯ 2 + t with l¯2 equal to the
squared length of the real four-vector (l¯1 , l¯2 , l¯3 , l¯4 ) and t a complex number. If we are
interested in the finite part, we can apply the non singular term in Eq. (8) and we have
I
it d2 l¯2 + t
−1 + log[−t] d4 l¯ dt. (10)
2π dt 2 l¯2 + m2 + t
After the derivative d2 / dt 2 the integral in d4 l¯ is convergent, 3 and this yields
I
it π 2 m2
−1 + log[−t] − dt. (11)
2π t + m2
The integral over the contour in the complex plane is equal to the residue at the pole t ∼
−m2

−π 2 −1 + log[m2 ] m4 . (12)
Our final result is correct. The advantage of the procedure above is clear: Eqs. (10), (11) are
well defined and concrete expressions, any step can be done in a pure numerical way. Note
that the integral in d4 l¯ must be done after the derivative d2 / dt 2 , and before the integral
in dt; otherwise the integral in d4 l¯ would be non convergent. In practice, if the above
procedure is done numerically, one should replace the integral in dt with a sum over a
finite set of points ti ; for the convergence of all integrals, it is enough to choose the ti in
this set with a non zero imaginary part.
3. Method of integration B
The formula (8) is a transparent and compact expression, which makes manifest
some properties of the dimensional regularization; for instance, the invariance under
the translations l¯µ → l¯µ + pµ is obvious, since it comes directly from the translational
3 We consider only value of t with non-zero imaginary part. See the discussion at the end of this section.
invariance of the integral in d4 l,¯ which is convergent (after the derivative in d2 /dt 2 ).
However in certain numerical calculations the (8) could be not efficient. We discuss a
different method which is more powerful in practical calculations, even if it requires a
slightly more involved formula.
There are several different ways of rewriting the (8); each of them depending on the way
we choose to parameterize the virtual momentum lµ . Here we do not intend to make an
exhaustive study, we will simply describe a quite general procedure to obtain well defined
integral representations for the Ik in (1).
First we observe that each Ik [f ] is a linear operator
Ik [f1 + f2 ] = Ik [f1 ] + Ik [f2 ], (13)
Ik [λf ] = λIk [f ]. (14)

It is quite natural to think that the linearity (13), (14) implies that the Ik are authentic
integrals or sum of integrals. In fact we can introduce a simple trick to build concrete
integral representations for generic linear functionals.
Consider the space of functions which admits a Laurent expansion around a point t0 .
They are defined by a set of an through the expansion 4
∞
X
f [t] = an (t − t0 )n . (15)
n=−∞
A linear functional I [f ] acting on this set of functions can be written

" ∞ # ∞ ∞
X X X
I [f ] = I an (t − t0 )n = an I (t − t0 )n = an bn (16)
n=−∞ n=−∞ n=−∞
where bn =I [(t − t0 )n ] is a real (or complex) number. Then we can use the identity
I
1 (t − t0 )n
dt = δnk (17)
2πi (t − t0 )k+1
C
for any n, k = 0, ±1, ±2, . . . if the integral contour C is a small circle, in the complex
plane of t, around the point t0 . Using the (17), the (16) can be written
I ∞
! ∞
!
1 X X bk
I [f ] = an (t − t0 )n
dt
2πi n=−∞ k=−∞
(t − t0 )k+1
I
1
= f [t]w[t] dt (18)
2πi
with
∞
X
bk
w[t] = . (19)
k=−∞
(t − t0 )k+1
4 We also assume that the series converges strongly enough to justify the steps below.
This is an integral representation of the linear functional I . This simple trick can be used to
build explicit and compact integral representations for any linear functional, whose action
on each term of a Laurent expansion is known.
Let us apply it to our problem. We start defining the angular integration in n = 4 + ε
dimensions. The analytical integration of any tensor with k (even) indices and constructed
with the n-dimensional vector lµ yields 5
Z
(2 + ε)!!
lµ̄ lν̄ lᾱ · · · lβ̄ lρ̄ lσ̄ dΩ 4+ε = 2π 2
(2 + k + ε)!!

× gµ̄ν̄ gρ̄ σ̄ · · · gᾱ β̄ + · · · + gµ̄ᾱ gρ̄ ν̄ · · · gβ̄ σ̄ l 2(k/2). (20)
dΩ 4+ε = sin2+ε [γ ] · · · sin2 [θ]sin[φ] dγ dθ dφ dδ (0 < γ , θ, . . . , φ < π; 0 < δ < 2π) is
the solid angle. It is understood that the indices (µ̄, ν̄, . . .) above are contracted with some
external momenta and/or other tensorial indices (like the spin of the external particles),
since the scattering matrix is Lorentz-invariant. Thus 6 only indices with µ̄ 6 4 are relevant
in our problem. This has been emphasized by the small bar on the Greek indices.
We look for an integral definition which reproduces the (20). To begin the procedure we
parameterize lµ with an array of five components 7
lµ → t x cos[θ], x sin[θ] cos[φ], x sin[θ] sin[φ] cos[δ],
p
x sin[θ] sin[φ] sin[δ], 1 − x 2 (21)
θ , φ and δ are the three angles in the space of four dimensions. t and x are two complex
variables which will be integrated over a complex contour as explained below. The need
of the auxiliary variable x will be clear in a moment. Suppose that we integrate over the
four-dimensional solid angle dΩ 4 = sin2 [θ]sin[φ] dθ dφ dδ all the tensors built with the
array (21) in place of an n-dimensional lµ . We get
Z
(2)!!
lµ̄ lν̄ lᾱ · · · lβ̄ lρ̄ lσ̄ dΩ 4 = 2π 2
(2 + k)!!
× (gµ̄ν̄ gρ̄ σ̄ · · · gᾱ β̄ + · · · + gµ̄ᾱ gρ̄ ν̄ · · · gβ̄ σ̄ )t k x k . (22)
Comparing this result with the (20) we see that tensorial structure is identical, but the
overall factor is not correct. The term of order ε0 can be obtained simply setting x = 1 and
t 2 = l 2 . Instead for the term of order ε, it would be enough to replace x k with a suitable
factor 8
ε
x k → bk = 1 − γE − ψ[2 + k/2] (23)
2
for any even k.
5 Here we assume that the overall factor π ε/2/Γ [n/2] has been reabsorbed into a redefinition of the scale µ,
as prescribed by the MS scheme.
6 We remind that only l has components > 4, and factors l l µ = l 2 are not relevant for the angular integration.
µ µ
7 In practice this means that any vector (including external momenta, etc.), must be written as an array of five
components, but we keep in mind that only lµ has a non zero fifth component.
8 The function ψ is the well known derivative d/dx log[Γ [x]]; it comes out when we expand the factorial in
the (20) and take only the first order in ε.
This can be achieved by means of a linear functional I such that I [x k ] = bk . In fact,

following the recipe above (Eqs. (15)–(19)) we can build the integral below
I X I
ε bk ε x + x 3 log[1 − 1/x 2 ]
I [f ] = f [x] dx = f [x] dx, (24)
2πi x k+1 2πi 2(−1 + x 2 )
k=even
where the integral contour is a circle around the singularity x ∼ 0. The integrand above has
a cut in the segment of the real axis between −1 and 1. It is easy to see that in the limit
of a path of integration very close to this cut, the logarithm can be approximated by its
imaginary part. The real part does not contribute to the full integral, and the logarithm can
be replaced by ±iπ . Taking also into account that f [x] = f [−x] we can rewrite the (24)
Z1
x3
I [f ] = ε f [x] dx. (25)
1 − x2 +
0
Here the notation (· · ·)+ means that
Z1 Z1

h[x] +
f [x] dx = h[x] f [x] − f [1] dx, (26)
0 0
and this makes the integral convergent near x ∼ 1.
Finally we are able to write a compact formula for the angular integration in 4 + ε
dimensions
Z Z Z1 Z !
x3
f [lµ ] dΩ 4+ε = f [lµ ] dΩ 4 +ε f [lµ ] dx dΩ 4
x=1 1 − x2 +
0
+ ε2 (· · ·). (27)
The last step is to define the integration in dl 2 = dt 2 . Again we follow the trick (15)–
(19): we have to compute the bσ in order to get the function w[t 2 ]. By definition, we know
that for any σ > 2
Z
1 1
bσ = I = t 2(1+ε/2) dt 2 2 σ
(t 2 + m2 )σ (t + m2 )
1 1
= + ε(· · ·), (28)
(−2 + σ )(−1 + σ ) (m2 )σ −2
for σ = 1, 2
m2
b1 = 2 + m2 log[m2 ] + ε(· · ·),
ε
2
b2 = − − log[m2 ] + 1 + ε(· · ·) (29)
ε
and for σ = 0, −1, −2, . . . the integral is zero (bσ = 0). If the integration contour in the t 2
complex plane is a small circle around the pole singularity t 2 = −m2 and replacing the bn
in the (19) with the bσ above we get
∞
X

2 σ 2
w[t ] =
2
bσ +1 t + m
2
= −t 2
+ log[−t 2 ] (30)
ε
σ =0
and
Z I
2 1
t 2(1+ε/2) dt 2 f [t 2 ] = − f [t 2 ]t 2 dt 2
ε 2πi
I
1
− f [t 2 ]t 2 log[−t 2 ] dt 2 +ε(· · ·). (31)
2πi
The integral contour is closed, it must contain all the singularities in the complex plane of
the function f [t 2 ], except for the singularity in t 2 = 0 and the cut for real and positive t 2
in all integrals where the logarithmic function appears (see, for example, Fig. 1). We will
comment on this contour of integration later on.
Combing the angular integration (27) and the (31) we get the full formula
Z Z
dt 2 I−1
f [lµ ] dn l = f [lµ ]t 2(1+ε/2) dΩ 4+ε = + I0 + I1 ε + · · · , (32)
2 ε
where
I
1
I−1 [f [lµ ]] = f [lµ ]t 2 dt 2 dΩ 4 , (33)
2πi x=1
I
1
I0 [f [lµ ]] = f [lµ ]t 2 log −t 2 dt 2 dΩ 4
4πi x=1
Z1 I
1 x3
+ f [lµ ]t 2 dt 2 dx dΩ 4 . (34)
2πi 1 − x2 +
0
Fig. 1. Integration contour in the t 2 complex plane. The thick line in the positive real axis represents
the cut of the logarithmic function. The path A is very close to this cut and the logarithm can be
replaced there by its imaginary part. Λ2uv and Λ2ir are, respectively, the ultraviolet and infrared cut-off.
Clearly, the integrals above are well defined and can be evaluated numerically. 9 For
example, to evaluate the first integral of I0 in the expression (34), the contour of integration
in dt 2 can be chosen as in Fig. 1. Namely we can divide the contour in three paths. The
path A is very close to the cut of the logarithm. This function can be replaced with ±iπ
(the sign depends if we are above or below the cut). This yields the following integral
2
ZΛuv !
1
P1 = f [lµ ]t 2 dt 2 dΩ 4 . (35)
2
x=1
Λ2ir
Note that this is an ordinary four-dimensional integral with an infrared and an ultraviolet
cut-off: in fact after setting x = 1, the array (21) may be regarded as a genuine four-vector.
Then we have the paths B and C, that are two circles with radius Λ2uv and Λ2ir , respectively
I !
1
P2 = f [lµ ]t log[−t ] dt dΩ 4
2 2 2
4πi
B x=1
I !
1
+ f [lµ ]t log[−t ] dt dΩ 4
2 2 2
. (36)
4πi
C x=1
These integrals can be seen as some counterterms which cancel the Λuv and Λir depend-
ence of the integral (35). The sum is equivalent to the finite part (in the limit ε→0) of the
dimensionally regularized integral.
Through the linearity (13), (14), this method can be generalized to two (or more)
loop calculations. If qµ and lµ are the two loop virtual momenta, then we can rewrite
the integration variables dn l = l 2(1+ε/2) dl 2 dΩ4+ε /2 and dn q = dqk q⊥2(1+ε/2) dq⊥ dΩ3+ε .
qk is the component of qµ parallel to lµ , q⊥ is the total length of the components of qµ
orthogonal to lµ . One then defines two six-dimensional vectors lµ and qµ , analogously
to (21), as functions of five angles θ, φ, δ, φ 0 , δ 0 , two auxiliary variables x and y (needed
for the integrations dΩ4+ε and dΩ3+ε ) and the complex variables l 2 , qk , q⊥ . Applying
the trick (15)–(19) one obtains the function w[l 2 , qk , q⊥ ]. Needless to say, this contains
dilogarithms in addition to some logarithms. From this guidelines, one gets the analogue
of the (34), for two loop calculations [39].
4. Some numerical examples
In order to make clear the practical purposes of these methods, we discuss here some
very simple examples. Suppose that we want to integrate
9 Here all integrals are understood to be in Euclidean space. In some kinematical regions, the integration (even
if convergent) needs to be regularized by the prescription m2 → m2 + iε. Also for some infrared singularities
the (34) must be rearranged differently or one should use the method A. The practical use of these methods in
various realistic situations certainly demands a much more extensive discussion, in this letter we simply present
the general idea, a more complete and systematic study will be done elsewhere (see Ref. [39]).
Z
1
dn l (37)
(l + p)2 + m2
with m2 = 2 and p2 = 1. Our numerical approach must reproduce the analytical result
π2 2
2 m + π 2 m2 log[m2 ] + · · · . (38)
ε
Suppose that we want to extract the finite part 10 π 2 m2 log[m2 ]. We apply the second
method B described above. We use the five-dimensional array (21) for lµ . Then the
integrand is a function of x, t 2 , θ, φ, δ. The finite part of the (37) is obtained computing
the two integrals in I0 (Eq. (34)). For the first one we choose the contour of integration
as in Fig. 1: the numerical values of Λuv and Λir have to be chosen in such a way that
the contour A+B+C contains all the physical singularities in the complex t 2 plane. Since
the integral is infrared convergent we can take the limit Λir → 0. Instead, in the ultraviolet
region, we can cut the integral at Λ2uv = 9. We also must set x = 1. Then the integral (35)
becomes
Z9
1 t2
P1 = dt 2 dΩ4 = 51.929. (39)
2 t 2 + 2t cos[θ] + 3
0
One can recognize in the integral above the ordinary four-dimensional integration with a
cut-off Λ2uv . Then the integration in dt 2 follows the path B
I
1 1 t2
P2 = dt 2 dΩ4 log[−t 2 ] 2 = −43.1815. (40)
2πi 2 t + 2t cos[θ ] + 3
B
This completes the first integral in I0 . The second one in Eq. (34) contains a non trivial
integration in the variable x, which must be performed using the prescription (26). The
contour of integration in dt 2 is rather simple: there is no logarithm, and no cut in the real
axis; then we do not need to follow the path A. There is no infrared singularity and also
the path C vanishes. The only non trivial contribution comes from the path B, a circle of
radius Λ2uv . This gives
Z1 I
1 1 x3
P3 = t 2 2
dt dx dΩ 4 = 4.935. (41)
i2π t 2 + 2xt cos[θ ] + 3 1 − x2 +
0 B
Note in the denominator the appearance of the variable x, which comes out when we take
the scalar product l · p with lµ from Eq. (21) and pµ in the direction µ = 1. The sum of
the three contributions yields

1
I0 = P1 + P2 + P3 = 13.6825 ' π 2 m2 log[m2 ], (42)
(l + p)2 + m2
10 It is more instructive to discuss the finite part, since the integral contour in (33) is trivial, and extracting the
singular part is straightforward.
in perfect agreement with the analytical expression. The method can also be applied for
infrared divergent integrals. The integral
Z
1 1
dn l 6 (43)
l (l + p)2 + m2
is ultraviolet convergent and infrared divergent. Thus we choose Λuv = ∞ and Λir = 1/4.
We get
Z∞
1 1 1
P1 = t 2 dt 2 dΩ 2 = 10.843, (44)
2 t + 2t cos[θ] + 3 t 6
1/4
I
1 1 1
P2 = dt 2 dΩ4 log[−t 2 ] = −12.1255,
2πi 2t 4 t 2 + 2t cos[θ ] + 3
C
Z1 I
1 1 x3
P3 = t 2 2
dt dx dΩ 4 = −0.1825. (45)
i2π t 2 + 2xt cos[θ] + 3 1 − x2 +
0 B
The sum of the three integrals gives

1 1
I0 6 = P1 + P2 + P3 = −1.465, (46)
l (l + p)2 + m2
very close to the exact result
1 2
− π 1 + log[81/4] . (47)
27
5. Conclusions
In the dimensional regularization of loop integrals, usually one defines the integral as a
function of the number of dimensions in a region of n, where the integral is convergent.
Then one makes an analytic continuation to obtain a definition of the integral also in regions
of n where the integral is divergent. This procedure is clear and unambiguous, however it
can only be applied in pure analytic (and symbolic) calculations. It cannot be converted
into an obvious numerical procedure. One is obliged to perform symbolic calculations,
which sometimes becomes lengthy and/or unaffordable. In this paper we have shown how
to set up a different approach, where each coefficient the Laurent expansion in powers of
ε in Eq. (1) can be written in terms of concrete and convergent integrals (well) defined
in a five-dimensional space (see Eq. (8) and Eqs. (32), (33) and (34)). Instead of abstract
n-dimensional vectors we have to deal with “concrete” five-dimensional vectors, and we
have to perform integrations over a compact space. This formulation has the advantage to
allow us the evaluation of all integrals through pure numerical methods. Clearly further
work is needed, to generalize the above result to more loops and to prove the efficiency
of this technique in realistic physical problems. However, we believe that the simplicity of
the numerical approach makes it very promising.
Acknowledgements
I would like to thank P. Nason for very helpful discussions, and Prof. R. Ferrari and the
Theoretical Physics Department of Milano University for its kind hospitality.
References
[1] G. ’t Hooft, M. Veltman, Nucl. Phys. B 44 (1972) 189.

[2] M.L. Mangano, S.J. Parke, Phys. Rept. 200 (1991) 301.
[3] M.L. Mangano, S.J. Parke, Z. Xu, in: M. Greco (Ed.), Proceedings of Les Rencontres de
Physique de la Vallee d’Aoste, La Thuile, Italy, Edition Frontieres, 1987, p. 513.
[4] F.A. Berends, W.T. Giele, Nucl. Phys. B 294 (1987) 700.
[5] M. Mangano, S. Parke, Z. Xu, Nucl. Phys. B 298 (1988) 653.
[6] M. Mangano, S.J. Parke, Nucl. Phys. B 299 (1988) 673.
[7] M. Mangano, Nucl. Phys. B 309 (1988) 461.
[8] S.J. Parke, T.R. Taylor, Phys. Rev. Lett. 56 (1986) 2459.
[9] P. De Causmaecker, R. Gastmans, W. Troost, T.T. Wu, Nucl. Phys. B 206 (1982) 53.
[10] R. Kleiss, W.J. Stirling, Nucl. Phys. B 262 (1985) 235.
[11] Z. Xu, D. Zhang, L. Chang, Nucl. Phys. B 291 (1987) 392.
[14] F. Caravaglios, M. Moretti, Phys. Lett. B 358 (1995) 332.
[15] Z. Kunszt, W.J. Stirling, Phys. Rev. D 37 (1988) 2439.
[16] C.J. Maxwell, Nucl. Phys. B 316 (1989) 321.
[17] M. Mangano, S. Parke, Phys. Rev. D 39 (1989) 758.
[18] F.A. Berends, W.T. Giele, H. Kuijf, Nucl. Phys. B 333 (1990) 120.
[19] R. Kleiss, H. Kuijf, Nucl. Phys. B 312 (1989) 616.
[20] G. Passarino, M. Veltman, Nucl. Phys. B 160 (1979) 151.
[21] Z. Bern, D.A. Kosower, Phys. Rev. Lett. 66 (1991) 1669.
[22] Z. Bern, D.A. Kosower, Nucl. Phys. B 362 (1991) 389.
[23] Z. Bern, D.A. Kosower, Nucl. Phys. B 379 (1992) 451.
[24] Z. Bern, D.C. Dunbar, Nucl. Phys. B 379 (1992) 562.
[25] Z. Bern, G. Chalmers, L. Dixon, D.A. Kosower, Phys. Rev. Lett. 72 (1994) 2134.
[26] Z. Bern, A.G. Morgan, Nucl. Phys. B 467 (1996) 479.
[27] Z. Bern, L. Dixon, D.A. Kosower, Ann. Rev. Nucl. Part. Sci. 46 (1996) 109.
[28] A.I. Davydychev, V.A. Smirnov, Nucl. Phys. B 554 (1999) 391.
[29] J. Fleischer, M.Yu. Kalmykov, Phys. Lett. B 470 (1999) 168.
[30] A.I. Davydychev, Acta Phys. Pol. 28 (1997) 841, and references therein.
[31] A.I. Davydychev, J.B. Tausk, Nucl. Phys. B 465 (1996) 507.
[32] L. Brücher, J. Franzkowski, A. Fink, D. Kreimer, Acta Phys. Pol. 28 (1997) 835.
[33] P. Draggiotis, R.H. Kleiss, C.G. Papadopoulos, Phys. Lett. B 439 (1998) 157.
[34] F. Caravaglios, M.L. Mangano, M. Moretti, R. Pittau, Nucl. Phys. B 539 (1999) 215.
[35] F. Caravaglios, M. Moretti, Z. Phys. C 74 (1997) 291.
[36] M. Moretti, Nucl. Phys. B 484 (1997) 3.
[37] G. Montagna, M. Moretti, O. Nicrosini, F. Piccinini, Eur. Phys. J. C 2 (1998) 483.
[38] D.E. Soper, hep-ph/9910292.
[39] F. Caravaglios, work in progress.
Algebraic generalization of the Ginsparg–Wilson

relation
Kazuo Fujikawa ∗
Department of Physics, University of Tokyo, Bunkyo-ku, Tokyo 113, Japan
Received 18 April 2000; revised 22 May 2000; accepted 20 June 2000
Abstract
A specific algebraic realization of the Ginsparg–Wilson relation in the form γ5 (γ5 D) + (γ5 D)γ5 =
2a 2k+1 (γ5 D)2k+2 is discussed, where k stands for a non-negative integer and k = 0 corresponds to
the commonly discussed Ginsparg–Wilson relation. From a view point of algebra, a characteristic
property of our proposal is that we have a closed algebraic relation for one unknown operator
D, although this relation itself is obtained from the original proposal of Ginsparg and Wilson,
γ5 D + Dγ5 = 2aDγ5 αD, by choosing α as an operator containing D (and thus Dirac matrices).
In this paper, it is shown that we can construct the operator D explicitly for any value of k. We first
show that the instanton-related index of all these operators is identical. We then illustrate in detail
a generalization of Neuberger’s overlap Dirac operator to the case k = 1. On the basis of explicit
construction, it is shown that the chiral symmetry breaking term becomes more irrelevant for larger k
in the sense of Wilsonian renormalization group. We thus have an infinite tower of new lattice Dirac
operators which are topologically proper, but a large enough lattice is required to accommodate a
Dirac operator with a large value of k.  2000 Elsevier Science B.V. All rights reserved.
1. Introduction
Recent developments in the treatment of fermions in lattice gauge theory are based on a
Hermitian lattice Dirac operator γ5 D which satisfies the Ginsparg–Wilson relation [1] 1
γ5 D + Dγ5 = 2aDγ5 D, (1.1)

where the lattice spacing a is utilized to make a dimensional consideration transparent, and
γ5 is a Hermitian chiral Dirac matrix. An explicit example of the operator satisfying (1.1)
∗ fujikawa@www-hep.phys.s.u-tokyo.ac.jp
1 To be precise, the general relation γ D + Dγ = 2aDγ αD, where α is a local operator, has been proposed in
5 5 5
Ref. [1], although the authors in Ref. [1] analyzed “only the simplest case where the matrix α is proportional to the
unit matrix in Dirac space”. With this qualification in mind, we refer to (1.1) as the “ordinary Ginsparg–Wilson
relation” in this paper. The original Ginsparg–Wilson relation is more general as stated above.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 9 5 - 3
488 K. Fujikawa / Nuclear Physics B 589 (2000) 487–503
and free of species doubling has been given by Neuberger [2]. The relation (1.1) led to an
interesting analysis of the notion of index in lattice gauge theory [3]. This index theorem
in turn led to a new form of chiral symmetry, and the chiral anomaly is obtained as a non-
trivial Jacobian factor under this modified chiral transformation [4]. This chiral Jacobian
is regarded as a lattice generalization of the continuum path integral [5]. The very detailed
analyses of the lattice chiral Jacobian have been performed [6–8]. It is also possible to
formulate the lattice index theorem in a manner [9] analogous to the continuum index
theorem [10,11]. An interesting chirality sum rule, which relates the number of zero modes
to that of the heaviest states, has also been noticed [12].
In this paper we discuss a generalization of the relation (1.1), which is characterized
by a non-negative integer k. It is shown that the explicit construction of an infinite tower
of lattice Dirac operators which satisfy the index theorem is possible, but a large enough
lattice is required to accommodate a Dirac operator with a large value of k.
2. Generalized algebra and its representation
We discuss a generalization of the algebra (1.1) to the form 2
γ5 (γ5 D) + (γ5 D)γ5 = 2a 2k+1(γ5 D)2k+2 , (2.1)

where k stands for a non-negative integer and k = 0 corresponds to the ordinary Ginsparg–
Wilson relation. When one defines
H ≡ γ5 aD, (2.2)
(2.1) is rewritten as
γ5 H + H γ5 = 2H 2k+2, (2.3)
or equivalently
Γ5 H + Γ5 H = 0, (2.4)
where we defined
Γ5 ≡ γ5 − H 2k+1 . (2.5)
Note that both of H and Γ5 are Hermitian operators.
We now discuss a general representation of the algebraic relation (2.4) following
the analysis in Appendix of Ref. [13]. (In Ref. [13], the algebra was normalized as
γ5 (γ5 D)+(γ5 D)γ5 = a(γ5 D)2 , but here we use the normalization (2.1) to simplify various
expressions.) The relation (2.4) suggests that if
2 This relation is obtained from the proposal in Ref. [1], γ D + Dγ = 2aDγ αD, by choosing α as an operator
5 5 5
containing D itself (and thus Dirac matrices). From a view point of algebra, the original construction in [1]
contains two unknown operators and one relation. In our construction, we have a closed algebraic relation for one
unknown operator D, which allows a neat analysis of representation in this section. This specific algebraically
closed realization, which is characterized by a non-negative integer, has not been discussed in Ref. [1].
K. Fujikawa / Nuclear Physics B 589 (2000) 487–503 489
H φn = aλn φn , (φn , φn ) = 1, (2.6)

with a real eigenvalue aλn for the Hermitian operator H , then
H (Γ5 φn ) = −aλn(Γ5 φn ). (2.7)

Namely, the eigenvalues λn and −λn are always paired if λn 6= 0 and (Γ5 φn , Γ5 φn ) 6= 0.
We also note the relation, which is derived by sandwiching the relation (2.3) by φn ,
(φn , γ5 φn ) = (aλn )2k+1 for λn 6= 0. (2.8)

Consequently,

(aλn )2k+1 = |(φn , γ5 φn )| 6 ||φn ||||γ5 φn || = 1. (2.9)
Namely, all the possible eigenvalues are bounded by
1
|λn | 6 . (2.10)
a
We thus evaluate the norm of Γ5 φn

(Γ5 φn , Γ5 φn ) = φn , γ5 − H 2k+1 γ5 − H 2k+1 φn

= φn , 1 − H 2k+1γ5 − γ5 H 2k+1 + H 2(2k+1) φn

= 1 − (aλn )2(2k+1)

= 1 − (aλn )2 1 + (aλn )2 + · · · + (aλn )4k , (2.11)
where we used (2.8). By remembering that all the eigenvalues are real, we find that φn is a
“highest” state
Γ5 φn = 0 (2.12)
only if

1 − (aλn )2 = (1 − aλn )(1 + aλn ) = 0 (2.13)
P
for the Euclidean positive definite inner product (φn , φn ) ≡ x φn† (x)φn (x).
We thus conclude that the states φn with λn = ± a1 are not paired by the operation Γ5 φn
and
1
γ5 Dφn = ± φn , γ5 φn = ±φn (2.14)
a
respectively. These eigenvalues are in fact the maximum or minimum of the possible
eigenvalues of H /a due to (2.10).
As for the vanishing eigenvalues H φn = 0, we find from (2.4) that H γ5 φn = 0, namely,
H [(1 ± γ5 )/2]φn = 0. We thus have
γ5 Dφn = 0, γ5 φn = φn or γ5 φn = −φn . (2.15)

To summarize the analyses so far, all the normalizable eigenstates φn of γ5 D = H /a are
categorized into the following 3 classes:
(i) n± (“zero modes”),

γ5 Dφn = 0, γ5 φn = ±φn , (2.16)
(ii) N± (“highest states”),
1
γ5 Dφn = ± φn , γ5 φn = ±φn , respectively, (2.17)
a
(iii) “paired states” with 0 < |λn | < 1/a,
γ5 Dφn = λn φn , γ5 D(Γ5 φn ) = −λn (Γ5 φn ). (2.18)
Note that Γ5 (Γ5 φn ) ∝ φn for 0 < |λn | < 1/a.
We thus obtain the index relation [3,4]
X
Tr Γ5 ≡ (φn , Γ5 φn )
n
X X X
= (φn , Γ5 φn ) + (φn , Γ5 φn ) + (φn , Γ5 φn )
λn =0 0<|λn |<1/a |λn |=1/a
X
= (φn , Γ5 φn )
λn =0
X
= φn , γ5 − H 2k+1 φn
λn =0
X
= (φn , γ5 φn )
λn =0
= n+ − n− = index, (2.19)
where n± stand for the number of normalizable zero modes with γ5 φn = ±φn in the
classification (i) above. We here used the fact that Γ5 φn = 0 for the “highest states” and that
φn and Γ5 φn are orthogonal to each other for 0 < |λn | < 1/a since they have eigenvalues
with opposite signatures.
On the other hand, the relation Tr γ5 = 0, which is expected to be valid in (finite) lattice
theory, leads to (by using (2.8))
X
Tr γ5 = (φn , γ5 φn )
n
X X
= (φn , γ5 φn ) + (φn , γ5 φn )
λn =0 λn 6=0
X
= n+ − n− + (aλn )2k+1 = 0. (2.20)
λn 6=0
In the last line of this relation, all the states except for the “highest states” with λn = ±1/a
cancel pairwise for λn 6= 0. We thus obtain a chirality sum rule [12] n+ − n− + N+ −
N− = 0 or
n+ + N+ = n− + N− , (2.21)
where N± stand for the number of “highest states” with γ5 φn = ±φn in the classifica-
tion (ii) above. These relations show that the chirality asymmetry at vanishing eigenvalues
is balanced by the chirality asymmetry at the largest eigenvalues with |λn | = 1/a. It was
argued in Ref. [13] that N± states are the topological (instanton-related) excitations of the
would-be species doublers.
All the n± and N± states are the eigenstates of D, Dφn = 0 and Dφn = (1/a)φn ,
respectively. If one denotes the number of states in the classification (iii) above by 2N0 ,
the total number of states (the dimension of the representation) N is given by
N = 2(n+ + N+ + N0 ), (2.22)
which is expected to be common to all the algebraic relations in (2.1) and to be a constant
independent of background gauge field configurations.
We note that all the states φn with 0 < |λn | < 1/a, which appear pairwise with
λn = ±|λn |, can be normalized to satisfy the relations
1/2
Γ5 φn = 1 − (aλn )2(2k+1) φ−n ,
1/2
γ5 φn = (aλn )2k+1 φn + 1 − (aλn )2(2k+1) φ−n . (2.23)
Here φ−n stands for the eigenstate with an eigenvalue opposite to that of φn . These states
φn cannot be the eigenstates of γ5 since |(φn , γ5 φn )| = |(aλn )2k+1 | < 1.
We have thus established that the representation of all the algebraic relations (2.1) has
a similar structure. In the next section, we show that the index n+ − n− is identical to all
these algebraic relations if the operator γ5 D satisfies suitable conditions.
3. Chiral Jacobian and the index relation
The Euclidean path integral for a fermion is defined by

Z Z
Dψ̄ Dψ exp ψ̄Dψ , (3.1)
where
Z X
ψ̄Dψ ≡ ψ̄(x)D(x, y)ψ(y), (3.2)
x,y
and the summation runs over all the points on the lattice. The relation (2.4) is rewritten as
γ5 Γ5 γ5 D + DΓ5 = 0, (3.3)
and thus the Euclidean action is invariant under the global “chiral” transformation [4]
X
ψ̄(x) → ψ̄ 0 (x) = ψ̄(x) + i ψ̄(z)γ5 Γ5 (z, x)γ5 ,
z
X
0
ψ(y) → ψ (y) = ψ(y) + i Γ5 (y, w)ψ(w), (3.4)
w
with an infinitesimal constant parameter . Under this transformation, one obtains a

Jacobian factor
Dψ̄ 0 Dψ 0 = J Dψ̄ Dψ (3.5)

with
J = exp[−2i Tr Γ5 ] = exp[−2i(n+ − n− )], (3.6)

where we used the index relation (2.19).
We now relate this index appearing in the Jacobian to the Pontryagin index of the gauge
field in a smooth continuum limit by following the procedure in Ref. [9]. We start with

(γ5 D)2 (H /a)2
Tr Γ5 f = Tr Γ5 f = n+ − n− . (3.7)
M2 M2
Namely, the index is not modified by any regulator f (x) with f (0) = 1 and f (x) rapidly
going to zero for x → ∞, as can be confirmed by using (2.19). This means that you can
use any suitable f (x) in the evaluation of the index by taking advantage of this property.
We then consider a local version of the index

(γ5 D)2 (γ5 D)2
tr Γ5 f (x, x) = tr γ5 − H 2k+1
f (x, x), (3.8)
M2 M2
where trace stands for Dirac and Yang–Mills indices; Tr in (3.7) includes a sum over
the lattice points x. A local version of the index is not sensitive to the precise boundary
condition, and one may take an infinite volume limit of the lattice in the above expression.
We now examine the continuum limit a → 0 of the above local expression (3.8). 3 We
first observe that the term

2k+1 (γ5 D)2
tr H f (3.9)
M2
goes to zero in this limit. The large eigenvalues of H = aγ5 D are truncated at the value
∼ aM by the regulator f (x) which rapidly goes to zero for large x. In other words, the
2
global index of the operator Tr H 2k+1f ( (γM
5 D)
2 ) ∼ O(aM)
2k+1 .
We thus examine the small a limit of

(γ5 D)2
tr γ5 f . (3.10)
M2
The operator appearing in this expression is well regularized by the function f (x), and we
evaluate the above trace by using the plane wave basis to extract an explicit gauge field
dependence. We consider a square lattice where the momentum is defined in the Brillouin
zone
π 3π
− 6 kµ < . (3.11)
2a 2a
We assume that the operator D is free of species doubling; in other words, the operator
D blows up rapidly (∼ a1 ) for small a in the momentum region corresponding to species
3 This continuum limit corresponds to the so-called “naive” continuum limit in the context of lattice gauge
theory.
doublers. The contributions of doublers are eliminated by the regulator f (x) in the above
expression, since
4
(γ5 D)2 1 1
tr γ5 f ∼ f →0 (3.12)
M2 a (aM)2
for a → 0 if one chooses f (x) = e−x , for example.
We thus examine the above trace in the momentum range of the physical species
π π
− 6 kµ < . (3.13)
2a 2a
We obtain the limiting a → 0 expression

(γ5 D)2
lim tr γ5 f (x, x)
a→0 M2
π
Z2a
d4 k −ikx (γ5 D)2 ikx
= lim tr e γ5 f e
a→0 (2π)4 M2
− 2a
π
ZL
d4 k −ikx (γ5 D)2 ikx
= lim lim tr e γ 5 f e
L→∞ a→0 (2π)4 M2
−L
ZL
d4 k −ikx / 2 ikx
(iγ5D)
= lim tr e γ5 f e
L→∞ (2π)4 M2
−L
2
/
D
≡ tr γ5 f , (3.14)
M2
where we first take the limit a → 0 with fixed kµ in −L 6 kµ 6 L, and then take the limit
L → ∞. This procedure is justified if the integral is well convergent. 4 We also assumed
that the operator D satisfies the following relation in the limit a → 0
4 To be precise, we deal with an integral of the structure
Z
π/2a Z
π/2a ZL Z−L
dx fa (x) = dx fa (x) + dx fa (x) + dx fa (x),
−π/2a L −L −π/2a
where fa (x) depends on the parameter a. (A generalization to a 4-dimensional integral is straightforward.) We

R π/2a R −L
thus have to prove that both of lima→0 L dx fa (x) and lima→0 −π/2a dx fa (x) can be made arbitrarily
small if one lets L to be large. A typical integral we encounter in lattice theory has a generic structure
Z
π/2a
2 2 2 2
lim dx e−[sin ax+(1−cos 2ax) ]/(a M )
a→0
−π/2a
Z
π/2a −π/2a
Z
2 2 2 2 2 2 2 2
= lim dx e−[sin ax+(1−cos 2ax) ]/(a M ) + lim dx e−[sin ax+(1−cos 2ax) ]/(a M )
a→0 a→0
π/2a −π/2a
Deikx h(x) → eikx (−/k + i/

∂ − gA
/ )h(x)

= i(/
∂ + igA/ ) e h(x) ≡ iD
ikx
/ eikx h(x) (3.15)
π π
for any fixed kµ , (− 2a < kµ < 2a ), and a sufficiently smooth function h(x). The function
h(x) corresponds to the gauge potential in our case, which in turn means that the gauge
potential Aµ (x) is assumed to vary very little over the distances of the elementary lattice
spacing.
Our final expression (3.14) in the limit M → ∞ reproduces the Pontryagin number in
the continuum formulation

lim tr γ5 f (D/ 2 /M 2 )
M→∞
Z
d4 k −ikx
= lim tr e / 2 /M 2 )eikx
γ5 f (D
M→∞ (2π)4
Z
d4 k ig µ ν
= lim tr γ5 f (ikµ + Dµ ) /M +
2 2
γ , γ Fµν /M 2
M→∞ (2π)4 4
Z
d4 k ig µ ν
= lim tr M 4 γ 5 f (ik µ + D µ /M) 2
+ γ , γ Fµν /M 2
, (3.16)
M→∞ (2π)4 4
where the remaining trace stands for Dirac and Yang–Mills indices. We also used the
relation
ig
D/ 2 = Dµ Dµ + γ µ , γ ν Fµν , (3.17)
4
and the rescaling of the variable kµ → Mkµ .
By noting tr γ5 = tr γ5 [γ µ , γ ν ] = 0, the above expression (after expansion in powers of
1/M) is written as (with 1234 = 1)
2 Z
1 ig µ ν d4 k 00
/ 2 /M 2 ) = tr γ5
lim tr γ5 f (D γ , γ Fµν 4
f −kµ k µ
M→∞ 2! 4 (2π)
g2
= tr µναβ Fµν Fαβ , (3.18)
32π 2
where we used
Z
π/2a
2 2 2 2
+ lim dx e−[sin ax+(1−cos 2ax) ]/(a M )
a→0
−π/2a
Z
π/2a
2 2 2 2
= lim dx e−[sin ax+(1−cos 2ax) ]/(a M )
a→0
−π/2a
ZL
2 2
= lim dx e−x /M
L→∞
−L
and satisfies the above criterion, if one chooses the regulator f (x) = e−x : here is an arbitrary small fixed
parameter, and the left-hand side of this relation stands for a conventional lattice calculation and the right-hand
side stands for a continuum calculation.
Z Z∞
d4 k 00 1 1
f −kµ k µ = f 00 (x)x dx = (3.19)
(2π)4 16π 2 16π 2
0
with x = −kµ k µ > 0 in our metric.

When one combines (3.7) and (3.18), one reproduces the Atiyah–Singer index theorem
(in continuum R 4 space) [10,11]. We note that a local version of the index (anomaly) is
valid for Abelian theory also. The global index (3.7) as well as a local version of the index
(3.8) are both independent of the regulator f (x) provided [5]

f (0) = 1, f (∞) = 0, f 0 (x)x x=0 = f 0 (x)x x=∞ = 0. (3.20)
We have thus established that the lattice index in (3.7) for any algebraic relation in (2.1)
is related to the Pontryagin index in a smooth continuum limit as
Z
g2
n+ − n− = d4 x tr µναβ Fµν Fαβ (3.21)
32π 2
by assuming the quite general properties of the basic operator D only: the basic relation
(2.1) with Hermitian γ5 D and the continuum limit property (3.15) without species doubling
in the limit a → 0. This shows that the instanton-related topological property is identical
for all the algebraic relations in (2.1), and the Jacobian factor (3.6) in fact contains the
correct chiral anomaly. (We are implicitly assuming that the index (3.7) does not change
in the process of taking a continuum limit.) Our result is naturally consistent with the
calculation of chiral anomaly by different methods in [1] and [3].
4. Explicit example of the lattice Dirac operator with k = 1
We now discuss an explicit construction of the lattice Dirac operator which satisfies the
generalized algebraic relation (2.1) with k = 1, though a generalization to an arbitrary k is
straightforward as is described in Section 5 later. For this purpose, we first briefly review
the construction of the Neuberger’s overlap Dirac operator for the ordinary Ginsparg–
Wilson relation.
We start with the conventional Wilson fermion operator DW defined by
1
DW (x, y) ≡ iγ µ Cµ (x, y) + B(x, y) − m0 δx,y ,
a
1
Cµ (x, y) = δx+µ̂a,y Uµ (y) − δx,y+µ̂a Uµ† (x) ,
2a
r X
B(x, y) = 2δx,y − δy+µ̂a,x Uµ† (x) − δy,x+µ̂a Uµ (y) ,
2a µ
Uµ (y) = exp[iagAµ (y)], (4.1)
where we added a constant mass term to DW for later convenience. The parameter r
stands for the Wilson parameter. Our matrix convention is that γ µ are anti-Hermitian,
(γ µ )† = −γ µ , and thus C
/ ≡ γ µ Cµ (n, m) is Hermitian
/† =C
C /. (4.2)
The operator D introduced by Neuberger [2], which satisfies the conventional Ginsparg–
Wilson relation (1.1), has an explicit expression
" # " #
1 HW 1 1
aD = 1 + γ5 q = 1 + DW q , (4.3)
2 HW 2 2 †
DW DW
†
where DW = γ5 HW is the Wilson operator defined above, and HW is Hermitian HW =
HW .
The physical meaning of this construction becomes more transparent if one considers
(naive) near continuum configurations specified by a small a limit with the parameters r/a
and m0 /a kept finite. We can then approximate the operator DW by [14,15]
DW ' iD
/ + Mn (4.4)
for each species doubler, where the mass parameters Mn stand for M0 = −m0 /a and one
of
2r m0 4r m0
− , (4, −1); − , (6, 1);
a a a a
6r m0 8r m0
− , (4, −1); − , (1, 1), (4.5)
a a a a
for n = 1 ∼ 15; we denoted (multiplicity, chiral charge) in the bracket for species doublers.
Here we used the relation valid in the near continuum configurations for the physical
species, for example,
X sin akµ r X m0 m0
DW (k) = γµ + (1 − cos akµ ) − ' γ µ kµ − (4.6)
µ
a a µ a a
in the momentum representation with vanishing gauge field.

In a symbolic notation, one can then write the overlap Dirac operator as
" #
X
15
1 1
aD ' 1 + (iD
/ + Mn ) q |nihn|,
2 D/ 2 + M2
n=0 n
" #
X
15
1 1
aγ5 D ' n
/ + Mn ) q
(−1) γ5 1 + (iD |nihn|. (4.7)
2 / 2 + M2
D
n=0 n
Here we explicitly write the projection |nihn| for each species doubler. If one chooses the
mass parameters so that
m0
M0 = − < 0, Mn > 0 for n 6= 0, (4.8)
a
namely,
0 < m0 < 2r, (4.9)
and if one lets all the mass parameters |Mn | become large, one obtains

1 iD / /2
1D
aγ5 D ' γ5 + for n = 0,
2 |M0 | 2 M02

1 / 1D
iD /2
aγ5 D ' (−1)n γ5 2 + − for n 6= 0. (4.10)
2 Mn 2 Mn2
If one chooses m0 to satisfy
2a|M0| = 2m0 = 1 (4.11)

one recovers the correctly normalized continuum Dirac operator for the physical species
and γ5 D ' (−1)n γ5 a1 for unphysical species doublers. In particular, the first relation in
(4.10) can then be written as
H ≡ aγ5D ' γ5 aiD

/ + γ5 (γ5 aiD)
/ 2 (4.12)
which ensures the conventional Ginsparg–Wilson relation in the leading order. These
properties become important in the following discussion.
4.1. Generalized algebra with k = 1
We now come back to the generalized algebra (2.1) with k = 1
H γ5 + γ5 H = 2H 4, (4.13)
where H = aγ5 D and Γ5 = γ5 − H 3 . This algebraic relation implies that
γ5 H 2 = [γ5 H + H γ5 ]H − H [γ5H + H γ5] + H 2 γ5 = H 2 γ5 . (4.14)

Namely, the algebraic relation (4.13) is equivalent to the two relations
H 3 γ5 + γ5 H 3 = 2H 6 , γ5 H 2 − H 2 γ5 = 0. (4.15)
If one defines H(3) ≡ H 3 , the first relation of (4.15) becomes
H(3)γ5 + γ5 H(3) = 2H(3)

2
(4.16)
with Γ5 = γ5 − H(3), which is identical to the conventional Ginsparg–Wilson relation (1.1).
We utilize this property to construct a solution to (4.15). Note that the operator Γ5 is
identical in these three ways of writing in (4.13), (4.15), and (4.16).
The physical condition for the operator H in (4.13) in the near continuum configuration
is (cf. (4.12))
H ' γ5 aiD
/ + γ5 (γ5 aiD)
/ 4 (4.17)
and thus H(3) in (4.16) should satisfy

H(3) ' γ5 aiD / 4 3 ' (γ5 aiD)
/ + γ5 (γ5 aiD) / 3 + γ5 (γ5 aiD)
/ 6 (4.18)
as can be confirmed by noting γ5D / + Dγ
/ 5 = 0. Here only the leading terms in chiral
symmetric and chiral symmetry breaking terms respectively are written.
One can thus construct a solution for H(3) by
" #
1 (3) 1
H(3) = γ5 1 + DW q , (4.19)
2 (3) † (3)
(DW ) DW
(3)
where we defined DW by 5
3
(3) m0
DW ≡ i(C
/ )3 + (B)3 − . (4.20)
a
The operators C/ , B and the parameter m0 /a are the same as in the original Wilson fermion
operator (4.1). By rewriting (4.19) as
" #
1 (3) 1
H(3) = γ5 1 + γ5 HW q (4.21)
2 (3) (3)
HW HW
(3) (3) (3) †
in terms of the Hermitian HW ≡ γ5 DW = HW and comparing it with (4.3), one can
confirm that our operator H(3) satisfies the relation (4.16). The condition (4.18) is satisfied
by noting

DW(3)
/ 3 + Mn(3) 3
' i(D) (4.22)
in the near continuum configuration, where the mass parameters are given by
3
(3) 3 m0
M0 ≡− ,
a
3 3 3 3
3 2r m0 4r m0
Mn(3) ≡ − , − ,
a a a a
3 3 3 3
6r m0 8r m0
− , − for n 6= 0. (4.23)
a a a a
Although we have the same condition on the parameters as before
0 < m0 < 2r (4.24)
to avoid the species doublers, the value of m0 itself is now required to satisfy
2(m0 )3 = 1 (4.25)
to ensure the properly normalized physical condition (4.18).
4.2. Reconstruction of H from H(3)
We now discuss how to reconstruct H , which satisfies (4.13), from H(3) defined above.
The basic idea is to take a real cubic root of H(3) as
5 It is also possible to use D (3) ≡ i(C )3 + (B − m /a)3 , or any suitable (ultra-local) operator which satisfies
/
W 0
(3) (3)
γ5 DW = (γ5 DW )† and (4.22).
H = (H(3))1/3 (4.26)
in such a manner that H thus obtained satisfies the second constraint in (4.15). For this
purpose, we first recall the essence of the general representation of the algebra (2.1)
analyzed in Section 2, which is applicable to (4.16) as well.
If one defines the eigenvalue problem
H(3)φn = (aλn )3 φn , (φn , φn ) = 1 (4.27)
one can classify the eigenstates into the 3 classes:
(i) n± (“zero modes”),
H(3)φn = 0, γ5 φn = ±φn , (4.28)
(ii) N± (“highest states”),
H(3)φn = ±φn , γ5 φn = ±φn , respectively, (4.29)
(iii) “paired states” with 0 < |(aλn )3 | < 1,
H(3)φn = (aλn )3 φn , H(3)(Γ5 φn ) = −(aλn )3 (Γ5 φn ), (4.30)
where
Γ5 = γ5 − H(3) . (4.31)
Note that Γ5 (Γ5 φn ) ∝ φn for 0 < |(aλn )3 | < 1.
We obtain the index relation
X X
Tr Γ5 ≡ (φn , Γ5 φn ) = (φn , γ5 φn ) = n+ − n− = index, (4.32)
n λn =0
where n± stand for the number of normalizable zero modes in the classification (i) above.
We also have a chirality sum rule
n+ + N+ = n− + N− , (4.33)
where N± stand for the number of “highest states” in the classification (ii) above.
If one denotes the number of states in the classification (iii) above by 2N0 , the total
number of states (the dimension of the representation) N is given by
N = 2(n+ + N+ + N0 ), (4.34)
which is expected to be common to all the fermion operators defined on the same lattice.
Also, all the states φn with 0 < |(aλn )3 | < 1, which appear pairwise with (aλn )3 =
±|(aλn )3 |, can be normalized to satisfy the relations
1/2 1/2
Γ5 φn = 1 − (aλn )6 φ−n , γ5 φn = (aλn )3 φn + 1 − (aλn )6 φ−n , (4.35)
where φ−n stands for the eigenstate with an eigenvalue opposite to that of φn .
Based on these general results in Section 2, we first observe that the index n+ − n− in
(4.32) is identical to the index of the expected solution of (4.13), although H(3) satisfies
(4.18). This observation is based on the relation
X
n+ − n− ≡ φn , Γ5 f (H(3))2 /(aM)6 φn , (4.36)
n
which is valid for any regulator with f (0) = 1. One can perform the same analysis as in
(3.7) in Section 3: The basic ingredient is the condition (4.18) for a physical momentum
region in the smooth continuum limit and the absence of species doublers. The calculation
analogous to (3.14) then gives
6 2
/
D /
D
n+ − n− = lim Tr γ5 f 6
= lim Tr γ 5 g (4.37)
M→∞ M M→∞ M2
with g(x) ≡ f (x 3 ) and g(0) = 1. The right-hand side of this relation shows that the present
index is identical to the index of the general operator in (2.1), which includes an expected
solution of (4.13). Due to the chirality sum rule (4.33), we also obtain the same value of
N+ − N− as for an expected solution of (4.13).
The agreement of the index of H(3) with the index of the expected solution H of (4.13)
suggests that we can define H operationally by
H φn ≡ aλn φn (4.38)
by using the same set of eigenfunctions and (the cubic roots of) eigenvalues
{φn }, {aλn } (4.39)

as for H(3) in (4.27). Note that the operator Γ5 = γ5 − H(3) = γ5 − H 3 , which reverses the
signature of eigenvalues of “paired states” and defines the index, is consistently chosen to
be identical for (4.16) and for (4.38). 6
We can then confirm the second constraint in (4.15) and the defining algebraic relation
(4.13) for any “paired state” φn ,
2
H γ5 − γ5 H 2 φn = H 2 γ5 φn − γ5 (aλn )2 φn
1/2
= H 2 (aλn )3 φn + 1 − (aλn )6 φ−n
1/2
− (aλn )2 (aλn )3 φn + 1 − (aλn )6 φ−n
=0 (4.40)
and
[Γ5 H + H Γ5 ]φn = Γ5 (aλn )φn − aλn (Γ5 φn ) = 0, (4.41)

where we used the relations in (4.35) and the definition (4.38). For “zero modes” and
the “highest states”, which are the eigenstates of γ5 , the condition [H 2 γ5 − γ5 H 2 ]φn = 0
obviously holds, and the relation [Γ5 H + H Γ5]φn = 0 is also confirmed.
The general representation of the algebra (4.13) is obtained from the standard
representation, which is defined by H in (4.38), γ5 in (4.35), and the state vectors {φn }
in (4.39), by applying a suitable unitary transformation.
6 This means that an explicit calculation of the chiral Jacobian (and chiral anomaly) for the theory defined
by (4.13) is performed by Tr Γ5 = T r(γ5 − H(3) ) in terms of H(3) in (4.19).
5. Discussion
When one considers the algebraic relation with a constant R

γ5 (γ5 D) + (γ5 D)γ5 = 2Ra 2k+1 (γ5 D)2k+2 (5.1)
instead of (2.1), one can eliminate the parameter R by a scale transformation
D → D0 = R 1/(2k+1)D. (5.2)
The path integral
Z Z
0
Dψ̄ Dψ exp ψ̄D ψ (5.3)
is equivalent to
Z Z
Dψ̄ Dψ exp ψ̄Dψ (5.4)
after absorbing the parameter R 1/(2k+1) into ψ̄, at least in a well regularized lattice path
integral. Consequently, the parameter R and also the factor a 2k+1 do not have an intrinsic
physical significance. 7
In contrast, the power of (γ5 D)2k+2 in the right-hand side of (5.1) has an intrinsic
physical meaning. One may recall the near continuum expressions (4.12) and (4.17)
H ' γ5 aiD
/ + γ5 (γ5 aiD)
/ 2 for k = 0,
H ' γ5 aiD
/ + γ5 (γ5 aiD)
/ 4
for k = 1, (5.5)
respectively. The first terms in these expressions stand for the leading terms in chiral
symmetric terms, and the second terms in these expressions stand for the leading terms in
chiral symmetry breaking terms. This shows that one can improve the chiral symmetry 8
by choosing a large parameter k.
The Dirac operator for such a general value of k is constructed by rewriting (2.1) as a
set of relations (see (4.14))
H 2k+1 γ5 + γ5 H 2k+1 = 2H 2(2k+1), H 2 γ5 − γ5 H 2 = 0, (5.6)
with H = aγ5D. The first of these relations (5.6) becomes identical to the ordinary
Ginsparg–Wilson relation (1.1) if one defines H(2k+1) ≡ H 2k+1 . One can construct a
solution to (5.6) by following the prescription in Section 4
" #
1 (2k+1) 1
H(2k+1) = γ5 1 + DW q , (5.7)
2 (2k+1) † (2k+1)
(DW ) DW
where
7 However, when one includes a Yukawa interaction, for example, this scaling argument need to be refined.
8 To avoid the misunderstanding, we note that the improvement of chiral symmetry here is meant in the sense
of Wilsonian renormalization group. The chiral symmetry breaking term becomes more irrelevant for larger k,
and this should be interesting from a view point of regularization of field theory in general. Also, the approach to
the continuum Dirac operator is controlled by two parameters, for example, by letting k → large and a → small
simultaneously.
2k+1
(2k+1) m0
DW ≡ i(C2k+1
/) +B 2k+1
− . (5.8)
a
The operator H is then finally defined by (in the representation where H(2k+1) is diagonal)
H = (H(2k+1))1/2k+1 (5.9)
in such a manner that the second relation of (5.6) is satisfied. This condition is in deed
satisfied as a generalization of (4.40) in the representation where H(2k+1) is diagonal. We
use the relation (2.23) in this proof. Also the conditions 0 < m0 < 2r and
2m2k+1
0 =1 (5.10)
ensure a proper normalization of the Dirac operator H .
However, one need to use a large enough lattice to accommodate the operator H with
a large k, since the operator (5.8) correlates lattice points far apart from each other for a
large k. An explicit analysis of the locality property of our operator H as in Ref. [16] is
left as an important problem. In the context of lattice simulation, it would be interesting to
see how the chiral properties are modified if one uses the operator with k = 1, which has
been analyzed in detail in this paper, instead of the conventional overlap Dirac operator
with k = 0. To detect the possible effects of k 6= 0 in a reliable way, it is expected that one
would have to consider a sufficiently large lattice and those observables which are sensitive
to low energy excitations.
As for the chiral fermions on the lattice, our general algebra (2.1) satisfies the
decomposition
(1 + γ5 ) (1 − γ̂5 ) (1 − γ5 ) (1 + γ̂5 )
D= D + D (5.11)
2 2 2 2
with
γ̂5 ≡ γ5 − 2a 2k+1(γ5 D)2k+1 , (γ̂5 )2 = 1, (5.12)
by noting γ5 (γ5 D)2 = (γ5 D)2 γ5 . This decomposition has the same form as for the overlap
operator D satisfying the ordinary Ginsparg–Wilson relation. It is thus expected that one
can apply the same considerations as in Refs. [17] and [18] to our general Dirac operator
also. In particular, the fermion number non-conservation of the chiral theory defined by
Z Z Z Z
(1 + γ5 ) (1 − γ̂5 )
Dψ̄ Dψ exp ψ̄DL ψ ≡ Dψ̄ Dψ exp ψ̄ D ψ (5.13)
2 2
follows from the fermion number transformation
ψ → eiα ψ, ψ̄ → ψ̄e−iα . (5.14)
If one remembers that the functional spaces of the variables ψ and ψ̄ are specified by the
projection operators (1 − γ̂5 )/2 and (1 + γ5 )/2, respectively, the Jacobian factor for the
transformation (5.14) is given by [17]

(1 + γ5 ) (1 − γ̂5 )
J = exp iα Tr −
2 2

= exp iα Tr γ5 − (γ5 aD) 2k+1
= exp iα[n+ − n− ] , (5.15)
where the index is defined in (2.19).

In conclusion, we have shown that the general idea of Ginsparg and Wilson can
be precisely realized as a closed algebraic relation (2.1) and it admits of the explicit
construction of an infinite tower of new lattice Dirac operators as a generalization of the
overlap Dirac operator. This should be interesting in the context of the regularization of
field theory in general.
Acknowledgement
The present work was initiated when I was visiting at Center for Subatomic Structure of
Matter (CSSM), University of Adelaide. I am grateful to David Adams and T.-W. Chiu for
stimulating discussions, and to Anthony Williams and David Adams for their hospitality at
CSSM.
References
[1] P.H. Ginsparg, K.G. Wilson, Phys. Rev. D 25 (1982) 2649.

[2] H. Neuberger, Phys. Lett. B 417 (1998) 141; Phys. Lett. B 427 (1998) 353.
[3] P. Hasenfratz, V. Laliena, F. Niedermayer, Phys. Lett. B 427 (1998) 125, and references therein.
[4] M. Lüscher, Phys. Lett. B 428 (1998) 342.
[5] K. Fujikawa, Phys. Rev. Lett. 42 (1979) 1195; Phys. Rev. D 21 (1980) 2848; Phys. Rev. D 22
(1980) 1499 (Erratum).
[6] Y. Kikukawa, A. Yamada, Phys. Lett. B 448 (1999) 265.
[7] D.H. Adams, Axial anomaly and topological charge in lattice gauge theory with overlap-Dirac,
hep-lat/9812003.
[8] H. Suzuki, Prog. Theor. Phys. 102 (1999) 141.
[9] K. Fujikawa, Nucl. Phys. B 546 (1999) 480.
[10] R. Jackiw, C. Rebbi, Phys. Rev. D 16 (1977) 1052.
[11] M. Atiyah, R. Bott, V. Patodi, Invent. Math. 19 (1973) 279.
[12] T.W. Chiu, Phys. Rev. D 58 (1998) 074511.
[13] K. Fujikawa, Phys. Rev. D 60 (1999) 074505.
[14] L. Karsten, J. Smit, Nucl Phys. B 183 (1981) 103.
[15] N. Kawamoto, J. Smit, Nucl. Phys. B 192 (1981) 100.
[16] P. Hernandez, K. Jansen, M. Lüscher, Nucl. Phys. B 552 (1999) 363.
[17] M. Lüscher, Nucl. Phys. B 549 (1999) 295; Nucl. Phys. B 568 (2000) 162.
[18] H. Neuberger, Phys. Rev. D 59 (1999) 085006.
QCD corrections up to order αs2 to polarized quark

production in e+e− -annihilation
V. Ravindran a , W.L. van Neerven b,∗
a Mehta Research Institute of Mathematics and Mathematical Physics, Chhatnag Road, Jhusi,
Allahabad-211019, India
b Instituut-Lorentz, University of Leiden, PO Box 9506, 2300 RA Leiden, The Netherlands
Abstract
We present the calculation of the order αs2 contributions to the cross section e+ + e− → q̄ + q
where the incoming leptons as well as one of the outgoing (anti) quarks are longitudinally polarized.
The computation is carried out for massless quarks so that it can be applied to light flavour production
(u, d, s). Unfortunately the massless quark approach is not valid for heavy flavour production like
c, b, t even in the case when the centre of mass energy Q is much larger than the quark mass m. This
is in contrast to unpolarized scattering where this approach works rather well for Q m. The reason
for this can be attributed to the anomalous terms which are characteristic of polarized coefficient
functions. Furthermore we also computed the order αs corrections to the longitudinal, transverse
and normal polarizations of heavy flavours with m 6= 0. The latter have been presented earlier in the
literature except for some contributions which are shown here for the first time. It turns out that the
corrections to the longitudinal and transverse (in the plane) polarization are rather small. However
the order αs corrections to the normal (out of the plane) polarization are large so that second order
contributions (for m 6= 0) are needed to get a better determination of this quantity.  2000 Elsevier
Science B.V. All rights reserved.
PACS: 12.38.-t; 12.38.Bx; 13.65.+i; 13.88.+e

Keywords: Electron–positron collisions; polarization; Heavy flavour production; QCD corrections
1. Introduction
Quark production in electron positron annihilation provides us with additional informa-

tion about the constants appearing in the standard model of the electroweak and strong
E-mail address: neerven@lorentz.leidenuniv.nl (W.L. van Neerven).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 2 0 - 4
508 V. Ravindran, W.L. van Neerven / Nuclear Physics B 589 (2000) 507–544
interactions. This in particular holds for heavy flavour production. An example is the elec-
troweak mixing angle θW which has been very accurately extracted from the forward–
backward asymmetry measured for bottom quarks at LEP and SLC. Besides a more accu-
rate determination of these constants heavy quark production can also reveal some signa-
tures of new physics. They might be observable when linear colliders are built [1] which
are giving access to much larger energies than those available until now. These energies
are large enough to observe top–anti-top production which will be one of the issues under
study. In particular of interest is the polarization of this quark because the spin can be eas-
ily measured from its decay products. This is possible because the mass of the top quark
is so big that it can decay electroweakly before it undergoes hadronization. The measure-
ment of the polarization provides us with new tests of the standard model and it might
even signal new interactions which are not predicted by this model. An example is given in
[2] where the effect of anomalous chromoelectric couplings of the gluon to the top quark
on the longitudinal polarization is studied. Another important quantity is the polarization
which is perpendicular to the plane spanned by the momenta of the incoming electron
and outgoing quark. This so-called normal polarization is a time reversal odd observable
which has implications for the observation of CP-violation in the neutral current sector.
Polarized heavy quark production has been studied on the Born level in [3–5]. In most of
these calculations the top quark spin is decomposed in the helicity basis. However one can
also make other choices like the beamline basis and the off-diagonal basis. In [6] one has
shown that choosing the off-diagonal basis the top and anti-top quarks are produced in one
unique spin configuration only which depends on the helicities of the incoming leptons.
QCD corrections to quark production have been computed in several papers [4,7–16]. The
corrections in the case of massive quarks are rather complicated even in first order. In the
case of the helicity basis the most of them are done by one group only see [12–15] except
for a few corrections which will be presented in this paper. The QCD corrections in the case
of the off-diagonal basis have been calculated in [16] as far as the soft and virtual gluon
corrections are concerned. However the contribution due to hard gluon bremsstrahlung has
not been calculated in this basis. Therefore the authors in [16] choose an alternative by
putting one of the quarks in a special spin configuration whereas the other (anti-)quark
and the gluon were taken to be inclusive. Furthermore one has neglected the width of the
Z-boson and the normal polarization was not considered. Because of the importance of
polarized quark production one should have an independent check on the calculations in
the literature. Therefore we want to repeat the computations done in [12–15] and include
some order αs corrections which were not considered before. Because of the complexity
of the expressions for the first order corrections we try to write the results as compact as
possible. The procedure is akin to the one followed in [17] in which the forward–backward
asymmetry and the shape parameter could be written in a compact form. The same was
also achieved for the longitudinal and transverse cross section in [18]. The main problem
is to reduce the number of Spence functions because they lead to unnecessarily long ex-
pressions. Exploiting several relations like the Hill-identity [19–21] one can reduce them
to a minimum. Furthermore we will present the cross section in such a way that it holds for
the longitudinal, transverse and normal polarization at the same time. Finally we concen-
V. Ravindran, W.L. van Neerven / Nuclear Physics B 589 (2000) 507–544 509
trate on the second order corrections to the longitudinal polarization of massless quarks.
They can be derived from the first moment of the timelike coefficient functions computed
in [22–26]. In the case of unpolarized scattering [17,18] the massless quark approach can
be also applied to heavy quarks provided the centre of mass (CM) energy is much larger
than the mass of the quark. These second order estimates are very useful [27] as long as
the exact calculations are not available because the latter are very difficult to compute. Un-
fortunately the massless quark approach does not work for heavy flavour production in the
case of polarized scattering. In this approach the coefficient functions are computed under
the condition that the quark mass m is put to zero at the start of the calculation like in [17,
18] (for the first order see [29–32]). One can also compute these functions in the so-called
massive quark approach with m 6= 0 and taking the limit m → 0 afterwards. In the calcula-
tion of the polarized coefficient functions both approaches lead to different results contrary
to what we have seen for unpolarized scattering. In other words the zero mass limit does
not commute with the integrations. This anomaly is due to chiral symmetry breaking when
the quark becomes massive and it was discovered for the first Bjorken sum rule which is
given by the first moment of the longitudinal spin structure function g1 (x, Q2 ) in [33–35].
For timelike processes like e+ e− collisions it was discussed in [13,15,28]. Here we would
like to emphasize that this anomaly does not affect the longitudinal polarization of the light
flavours even if we regularize the collinear divergences by giving the light quark a fictitious
mass. It turns out that this anomaly also appears in the quark operator matrix element so
that it will be removed by mass factorization (see [33,34]). Hence we obtain the same result
as for m = 0 where one can use n-dimensional regularization. However for heavy flavours,
where the mass has a definite meaning, a subtraction via the quark operator matrix element
is not justified so that these anomalous terms are retained in the QCD corrections. There-
fore in the case of heavy quarks the polarized coefficient functions have to be calculated
for non-zero masses. These calculations are far from trivial because one cannot apply the
tricks which were so successful for the calculation in [36,37] of the order αs2 corrections
to σtot (e+ e− → hadrons) with massive quarks. This is because the timelike coefficient
functions cannot be written as the imaginary part of a forward scattering amplitude which
is an essential ingredient for the computations in [36,37]. The paper will be organized as
follows. In the Section 2 we define the kinematics and give an outline of the calculation
procedure. We also present the formulae for the spin dependent and spin independent parts
of the cross section. In Section 3 we show the results for the radiative corrections com-
puted up to first order in the strong coupling constant αs for massive quarks. Here we also
include the contributions which are proportional to the width of the Z-boson. The correc-
tions will be extended for the longitudinal polarization of massless quarks up to order αs2 .
In Section 4 we discuss the effects of these corrections on the longitudinal, transverse and
normal polarization of the detected quark. The long expressions for the order αs corrected
quark structure functions Wi (x, Q2 , m2 ) are presented in Appendix A. After integrating
these functions over the Bjorken variable x we could express them into a compact form in
Appendix B.
2. Kinematics for polarized e+ e− -annihilation
Polarized heavy quark production in electron–positron annihilation is given by the

following process
e+ (qe+ , λe+ ) + e− (qe− , λe− ) → V (q) → H(p, s) + “X”, (2.1)
where H denotes the heavy quark and “X” represents any inclusive multi-partonic state
containing the heavy anti-quark and the light (anti-)quarks and gluons. Further λi , qi
denote the spin and momenta of the incoming leptons and s, p stand for the spin and
momentum of the heavy quark H detected in the final state. If we denote the momentum of
the virtual vector boson V (V = γ , Z) by q, the centre of mass energy Q is defined by
q 2 = Q2 = (qe+ + qe− )2 . (2.2)
The differential cross section corresponding to reaction (2.1) equals
1 X
dσ = dP S Lµν P
(V1 V2 ) (V1 V2 )
M µν,(V1 V2 )
2Q2
V1 V2
1 γ ) (γ γ ) µν,(γ γ )
≡ 2
dP S L(γ
µν P M + L(ZZ)
µν P
(ZZ) µν,(ZZ)
M
2Q

+ L(γ
µν P
Z) (γ Z) µν,(γ Z)
M + L(γ
µν P
Z)∗ (γ Z)∗ µν,(γ Z)∗
M . (2.3)
(V V )
Here dP S denotes the multi-parton phase space including the heavy quark and Lµν1 2
(V V )
and Mµν1 2 are the leptonic and partonic matrix elements, respectively. Further P (V1V2 )
represents the product of the propagators corresponding to the vector bosons V1 and V2 .
The leptonic tensors are given by

µν = 4παQe (1 − λe+ λe− )lµν + (λe+ − λe− )lµν ,
L(γ γ) 2 ˜
4πα h 2 2
µν = 2 2
L(ZZ) −2geV geA (λe+ − λe− ) + geV + geA (1 − λe+ λe− ) lµν
cw sw
2
i
+ −2geV geA (1 − λe+ λe− ) + geV + geA (λe+ − λe− ) l˜µν ,
2
4πα
µν = −
L(γ Z)
Qe −geV (1 − λe+ λe− ) + geA (λe+ − λe− ) lµν
cw sw

+ −geV (λe+ − λe− ) + geA (1 − λe+ λe− ) l˜µν , (2.4)
with
µ µ
l µν = qe+ qeν− + qeν+ qe− − qe+ qe− g µν ,
l˜µν = i µναβ qe+ α qe− β . (2.5)
Notice that the polarizations of the positron and the electron indicated by λe+ and λe−
respectively are defined by
1
vλe+ (qe+ )v̄λe+ (qe+ ) = q/ e+ (1 + λe+ γ5 ),
2
1
uλe− (qe− )ūλe− (qe− ) = q/ e− (1 − λe− γ5 ), (2.6)
2
where the incoming leptons are taken to be massless. In the case λi = 1 the leptons are
polarized along the direction of their momenta. When λi = −1 the leptons are polarized
opposite to the direction of their momenta.
On the Born level the vertex for the coupling of the vector boson V to the fermions will
be denoted by

V ,(0)
Γa,µ = i vaV + aaV γ5 γµ , V = γ , Z,
γ γ
va = −eQa , aa = 0,
e e
vaZ = − gV , aaZ = gA . (2.7)
cw sw a cw sw a
The electroweak coupling constants are given by
α = e2 /4π, cw = cos θW , sw = sin θW ,

1 1
gaV = Ta3 − sw2 Qa , gaA = − Ta3 . (2.8)
2 2
The electroweak charges for the leptons are equal to
1
Qa = 0, Ta3 = , a = νl , ν̄l , l = e, µ, τ,
2
1
Qa = −1, Ta = − ,
3
a = e − , µ− , τ − , (2.9)
2
and for the quarks we obtain
2 1
Qa = , T3a = , a = u, c, t,
3 2
1 1
Qa = − , T3a = − , a = d, s, b. (2.10)
3 2
The squared propagators P (V1 V2 ) are given by
1 1
Pγ γ = , P ZZ = ,
Q4 (Q2 − MZ2 )2 + MZ2 ΓZ2
(Q2 − MZ2 ) + iMZ ΓZ
Pγ Z = , (2.11)
Q2 {(Q2 − MZ2 )2 + MZ2 ΓZ2 }
where we have introduced a finite width ΓZ for the Z-boson.
The differential cross section for the inclusive reaction Eq. (2.1) can be written as
Z1 X
dσ 1
= 2
dx Lµν P
(V1 V2 ) (V1 V2 ) µν,(V1 V2 )
W ,
dΩ 2Q √
ρ (V1 V2 )
2p · q 4m2
dΩ = d cos θ dφ, x= , ρ= . (2.12)
Q2 Q2
Here θ is the polar angle of the outgoing quark with respect to the beam direction of the
electron and m denotes the mass of the heavy quark. The partonic tensor can be expressed
into structure functions Wi(V1 V2 ) (i = 1–11) in the following way

Nc 2 µν (V1 V2 ) pµ pν (V1 V2 ) s · q (V V )
W µν,(V1 V2 ) = 2
Q g W1 + 2
W2 + mg µν 2 W3 1 2
8π Q Q
s · q (V V ) p µ s ν + s µ pν
(V V )
+ mpµ pν 4 W4 1 2 + m W5 1 2
Q Q2
1 s ·q
+ 2 i µναβ pα qβ W6(V1 V2 ) + mi µναβ pα qβ 4 W7(V1 V2 )
Q Q
m µναβ (V1 V2 ) m µναβ
+ 2 i pα sβ W8 + 2 i qα sβ W9(V1 V2 )
Q Q
pµ s ν − s µ pν (V1 V2 ) m
+m W10 + 4 i pµ ναβγ pα qβ sγ
Q2 Q

(V V )
+ pν µαβγ pα qβ sγ W11 1 2 , (2.13)
where Nc denotes the number of colours (in QCD Nc = 3). Finally we have to specify the
E the spin vector is given by
spin of the quark H(p, s). In the rest frame, i.e., p = (m, 0)
b ),
s = (0, W b = (W
W b2, W
b1, W b 3 ), b 2 = 1,
with W (2.14)
so that s 2 = −1 and s · p = 0. In the CM frame of the electron positron pair we introduce
the notations
Q Q
qe− = (1, 0, 0, 1), qe+ = (1, 0, 0, −1),
2 2
p = E, |p| E n̂ n = (sin θ cos φ, sin θ sin φ, cos θ ),
q
1 1
E = Qx, |p|
E = Q x 2 − ρ. (2.15)
2 2
In this frame the spin four-vector of the quark can be found after an appropriate Lorentz
transformation so that it becomes equal to
b b )n̂
n̂ · W b + |p|
E 2 (n̂ · W
s = |p| E ,W , with n̂2 = 1. (2.16)
m m(m + E)
Note that the spin four-vector in the CM frame depends on the energy of the quark and
hence depends on the integration variable x in Eq. (2.12). The computation of the cross
section in Eq. (2.12) involves the contraction of the symmetric and antisymmetric parts
(V V )
of the leptonic tensor Lµν1 2 with W µν,(V1 V2 ) . The contraction with the symmetric part
equals
Z
dx lµν W µν,(V1 V2 )

Nc Q4 1
= 1 + cos2 θ vqV1 vqV2 T1 + aqV1 aqV2 T2
8π 2 8
1
+ sin2 θ vqV1 vqV2 T3 + aqV1 aqV2 T4
4
1 V1 V2 1 V1 V2
+ vq aq + vq aq T7 + vq aq − vq aq T8 + vq vq T9 ,
V2 V1 V2 V1 V1 V2
2 2
(2.17)
and for the antisymmetric parts we get

Z
Nc Q4 1 V1 V2 1
˜
dx lµν W µν,(V1 V2 )
= vq aq + vqV2 aqV1 T5 cos θ + vqV1 aqV2
8π 2 4 4

− vq aq T6 cos θ + vq vq T10 + aqV1 aqV2 T11
V2 V1 V1 V2
1
+ vqV1 aqV2 + vqV2 aqV1 T12
2
1
+ vqV1 aqV2 − vqV2 aqV1 T13 . (2.18)
2
In the expressions above Ti (i = 1–6) represent the unpolarized structure functions. The
polarized structure functions can be decomposed as follows

T7 = T7,T WT cos θ + WL T7,L1 1 + cos2 θ + T7,L2 sin2 θ ,

T8 = T8,T WT cos θ + T8,L WL 1 + cos2 θ ,
T9 = T9,N WN cos θ sin θ,
T10 = T10,T WT + T10,L WL cos θ,
T11 = T11,T WT + T11,L WL cos θ,
T12 = T12,N WN sin θ,
T13 = T13,N WN sin θ. (2.19)
Notice that the polarized structure functions Ti (i = 7–13) are expanded in terms of the
longitudinal (WL ), the transverse WT and the normal polarization WN of the quark. They
are defined by
b,
WL = n̂ · W WT = W b 3 − n̂ · W
b cos θ,
WN = Wb 2 cos φ − W
b 1 sin φ. (2.20)
If we decompose the partonic structure functions as follows
(V1 V2 ) v2 a2 {v ,a }
Wi = vqV1 vqV2 Wi q + aqV1 aqV2 Wi q + vqV1 aqV2 + aqV1 vqV2 Wi q q
[v ,a ]
+ vqV1 aqV2 − aqV1 vqV2 Wi q q , (2.21)
one can express the quantities Ti into integrals over the partonic structure functions Wi .
For the unpolarized structure functions we obtain
Z1
v2
T1 = −4 dxW1 q x, Q2 , m2 , (2.22)
√
ρ
Z1
a2
T2 = −4 dxW1 q x, Q2 , m2 , (2.23)
√
ρ
Z1
αx2 vq2 v2
T3 = dx W2 x, Q2 , m2 − 2W1 q x, Q2 , m2 , (2.24)
√
2
ρ
Z1
αx2 aq2 aq2
T4 = dx W x, Q , m − 2W1 x, Q , m
2 2 2 2
, (2.25)
√
2 2
ρ
Z1
{vq ,aq }
T5 = 2 dx αx W6 x, Q2 , m2 , (2.26)
√
ρ
Z1
[vq ,aq ]
T6 = 2 dx αx W6 x, Q2 , m2 . (2.27)
√
ρ
The results for the longitudinal polarized structure functions are given by
Z1
αx {vq ,aq }
T7,L1 = dx − W3 2
x, Q , m 2
, (2.28)
√
2
ρ
Z1
αx {vq ,aq } α 3 {v ,a }
T7,L2 = dx − W3 x, Q2 , m2 + x W4 q q x, Q2 , m2
√
2 8
ρ

αx {v ,a }
+
xW5 q q x, Q2 , m2 , (2.29)
2
Z 1
αx [v ,a ]
T8,L = dx − W3 q q x, Q2 , m2 , (2.30)
√
2
ρ
Z1
αx2 vq2 ρ vq2 x vq2
T10,L = dx W x, Q , m − W8 x, Q , m − W9 x, Q , m
2 2 2 2 2 2
,
√
4 7 4 2
ρ
(2.31)
Z1
αx2 aq2 ρ a2 x a2
T11,L = dx W7 x, Q2 , m2 − W8 q x, Q2 , m2 − W9 q 2
x, Q , m 2
.
√
4 4 2
ρ
(2.32)
Similar expressions are found for the transverse polarized structure functions
Z1
√
ρ {vq ,aq }
T7,T = dx − αx W5 2
x, Q , m 2
, (2.33)
√
2
ρ
Z1 √
ρ [v ,a ]
T8,T = dx − αx W5 q q x, Q2 , m2 , (2.34)
√
2
ρ
Z1 √ √
ρ vq2 ρ vq2
T10,T = dx − xW8 x, Q , m −
2 2 2
W x, Q , m 2
, (2.35)
√
4 2 9
ρ
Z1 √ √
ρ a2 ρ aq2
T11,T = dx − xW8 q x, Q2 , m2 − W9 x, Q2 , m2 , (2.36)
√
4 2
ρ
and the results for the normal polarized structure functions can be expressed into the form
Z1 √
i ρ 2 vq2
T9,N = dx 2
α W x, Q , m 2
, (2.37)
√
8 x 11
ρ
Z1 √
i ρ {v ,a }
T12,N = dx αx W10q q x, Q2 , m2 , (2.38)
√
2
ρ
Z1 √
i ρ [vq ,aq ]
T13,N = dx αx W10 2
x, Q , m 2
, (2.39)
√
2
ρ
with
q
αx = x 2 − ρ. (2.40)
Analogous to deep inelastic lepton–hadron scattering the leading contributions to the

unpolarized spin structure functions Wi with i = 1, 2, 6 in Eqs. (2.22)–(2.27) are of type
twist two. The same holds for the polarized structure function W3 . However W4 , W5 and
W7 , W8 , W9 , W10 , W11 contain twist two as well as twist three contributions. Notice that
W8 is due to the fact that the electroweak currents are not conserved. Furthermore the twist
three part cancels in the combinations
xW4 + 4W5 , −xW7 + 2W9 , (2.41)
which means that the leading contributions in m2 /Q2 to the longitudinal structure
functions Ti,L in Eqs. (2.28)–(2.32) are of twist two only. Notice that W8 in the equations
above is multiplied by ρ = 4m2 /Q2 so that this term vanishes in the limit m → 0. The
transverse parts Ti,T in Eqs. (2.33)–(2.36) receive contributions from twist two as well as
twist three. The same also applies to the normal parts Ti,N in Eqs. (2.37)– (2.39). A second
feature of the above equations is that in the limit m → 0 all transverse and normal parts
vanish whereas the longitudinal parts Ti,L (i = 7, 10, 11) and the unpolarized quantities
Ti (i = 1–5) tend to non-zero values. After having carried out the integration over x in
Eq. (2.12) the differential cross section can be written as
dσ dσU dσL dσT dσN
(λe+ , λe− , W ) = + WL + WT + WN , (2.42)
dΩ dΩ dΩ dΩ dΩ
where U represents the unpolarized cross section with respect to the outgoing quark. The
four cross sections on the right-hand side of Eq. (2.42) can be decomposed according to
the vector bosons which appear in the intermediate state, i.e.,
(γ γ ) (γ Z)
dσk dσk dσ (ZZ) dσk
(λe+ , λe− ) = Nc α 2 + k + ,
dΩ dΩ dΩ dΩ
k = U, T , L, N. (2.43)
The results for the photon–photon interference term can be written as 1

(γ γ )
dσU 1 1 2
=Q P
2 γγ
Q2e Q2q (1 − λe+ λe− ) 1 + cos θ T1 + sin θ T3 ,
2
(2.44)
dΩ 8 4
(γ γ )
dσL
= Q2 P γ γ Q2e Q2q (λe+ − λe− ) cos θ T10,L , (2.45)
dΩ
(γ γ )
dσT
= Q2 P γ γ Q2e Q2q (λe+ − λe− )T10,T , (2.46)
dΩ
(γ γ )
dσN
= Q2 P γ γ Q2e Q2q (1 − λe+ λe− ) cos θ sin θ T9,N . (2.47)
dΩ
The Z–Z interference term receives contributions from
P ZZ h
(ZZ)
dσU 2 2
= Q2 4 4
− 2geV geA (λe+ − λe− ) + geV + geA (1 − λe+ λe− )
dΩ cw sw
1 2 2 2 2
× 8 1 + cos2 θ gqV T1 + gqA T2 + 14 sin2 θ gqV T3 + gqA T4
2 2
+ 2geV geA (1 − λe+ λe− ) − geV + geA (λe+ − λe− )
i
× 12 gqV gqA cos θ T5 , (2.48)
P ZZ h V A
(ZZ)
dσL 2 2
= Q2 4 4
2ge ge (λe+ − λe− ) − geV + geA (1 − λe+ λe− )
dΩ cw sw
V A
× gq gq 1 + cos2 θ T7,L1 + sin2 θ T7,L2
2 2
+ − 2geV geA (1 − λe+ λe− ) + geV + geA (λe+ − λe− )
2 2 i
× cos θ gqV T10,L + gqA T11,L , (2.49)
P ZZ h V A
(ZZ)
dσT 2 2
= Q2 4 4
2ge ge (λe+ − λe− ) − geV + geA (1 − λe+ λe− )
dΩ cw sw
V A
× gq gq cos θ T7,T
2 2
+ − 2geV geA (1 − λe+ λe− ) + geV + geA (λe+ − λe− )
2 2 i
× gqV T10,T + gqA T11,T , (2.50)
1 Very often one replaces α by G M 2

√
F Z 2 sw 2 cw 2 /π (see, e.g., [4,13]).
dσN(ZZ) P ZZ h 2 2
= Q2 4 4 − 2geV geA (λe+ − λe− ) + geV + geA (1 − λe+ λe− )
dΩ cw sw
V2
× gq cos θ sin θ T9,N
2 2
+ 2geV geA (1 − λe+ λe− ) − geV + geA (λe+ − λe− )
i
× gqV gqA sin θ T12,N . (2.51)
The photon-Z interference term consists of the following parts
dσU
(γ Z)
Re P γ Z h
= 2Q2 2 2
Qe q ge (1 − λe+ λe− ) − ge (λe+ − λe− )
Q V A
dΩ cw sw
1
× 8 (1 + cos2 θ )gqV T1 + 14 sin2 θ gqV T3
i
+ − geV (λe+ − λe− ) + geA (1 − λe+ λe− ) 14 gqA cos θ T5
Im P γ Z h
+ 2Q2 2 2
Qe Qq geV (λe+ − λe− ) − geA (1 − λe+ λe− )
cw sw
i
× 14 gqA cos θ Im T6 , (2.52)
dσL
(γ Z)
Re P γ Z h
= 2Q2 2 2
Qe Qq − geV (1 − λe+ λe− ) + geA (λe+ − λe− )
dΩ cw sw

× 12 gqA 1 + cos2 θ T7,L1 + sin2 θ T7,L2
i
+ geV (λe+ − λe− ) − geA (1 − λe+ λe− ) gqV cos θ T10,L
Im P γ Z h
+ 2Q2 2 2
Qe Q q geV (1 − λe+ λe− ) − geA (λe+ − λe− )
cw sw
1 A i
× 2 gq (1 + cos2 θ ) Im T8,L , (2.53)
dσT
(γ Z)
Re P γ Z h
= 2Q2 2 2
Qe Qq − geV (1 − λe+ λe− ) + geA (λe+ − λe− )
dΩ cw sw
1 A
× 2 gq cos θ T7,T
i
+ geV (λe+ − λe− ) − geA (1 − λe+ λe− ) gqV T10,T
Im P γ Z h
+ 2Q2 2 s2
Qe Q q ge
V
(1 − λe + λe− ) − ge (λe+ − λe− )
A
cw w
1 A i
× 2 gq cos θ Im T8,T , (2.54)
dσN
(γ Z)
Re P γ Z h
= 2Q2 2 2 Qe Qq geV (1 − λe+ λe− ) − geA (λe+ − λe− )
dΩ cw sw
V
× gq cos θ sin θ T9,N
i
+ − geV (λe+ − λe− ) + geA (1 − λe+ λe− ) 12 gqA sin θ T12,N
Im P γ Z h
+ 2Q2 2 2
Qe Qq geV (λe+ − λe− ) − geA (1 − λe+ λe− )
cw sw
1 A i
× gq sin θ Im T13,N . (2.55)
2
The Born contributions (zeroth order in αs ) to the unpolarized quantities denoted by Ti(0)
(0) (0)
(i = 1, 2, 3, 5) can be found in [4,13,17]. The longitudinal T7,L 1
, Ti,L (i = 10, 11) and
(0) (0)
the transverse polarized structure functions T7,T and T10,T were calculated in [4]. In the
last reference one also finds the Born contribution to the normal polarized quantity given
(0)
by ImT13,N . Notice that the quantities Ti not mentioned above all vanish in the Born
approximation. The first order QCD contributions to T9,N and T12,N are also computed
in [4] but the corrections to the other structure functions Ti were neglected. The latter are
computed in [12,13,15] (longitudinal) and [14] (transverse and normal). The exceptions
are the order αs contributions to Im T6 and Im T8,L which will be presented in this paper
for the first time. The second order QCD corrections are not known yet but for light quarks
like u, d, s they can be computed for the longitudinal polarized structure functions Ti,L
(i = 7, 10, 11) and the results are shown in the next section. Notice that the transverse and
normal polarized structure functions Ti,T and Ti,N vanish for massless quarks.
3. Computation of the corrections up to order αs2
The partonic structure tensor defined in Eq. (2.13) is represented by the following
perturbation series
∞
X
αs (µ2 ) k (V1 V2 ),(k)
(V1 V2 )
Wµν = Wµν , (3.1)
4π
k=0
where µ denotes the renormalization scale. The same expression for the perturbation series
(V V )
holds for the the structure functions Wi 1 2 and the functions Ti in Eqs. (2.21)–(2.39). The
structure tensor is determined by the following reaction
S(p1 ) + l(p2 ) · · · l(pn ),
V (q) → H (p) + H (3.2)
where l(pi ) (i = 2, 3, . . . , n) represent the momenta of the light partons ((anti-)quarks and
gluons) in the final state. If the matrix element of the process above is given by M µν,(V1 V2 )
then the partonic tensor is obtained by integration over the multi-partonic phase space
which also includes the heavy anti-quark, i.e.,
n Z
!
Y d 3 pi X n
Wµν(V1 V2 )
= 4 (4)
(2π) δ q −p− pj Mµν (V1 V2 )
. (3.3)
(2π) 3 2p0
i=1 i j =1
The computation of the phase space integrals is carried out in the rest frame of the vector
boson V and proceeds in the way as is given in [22–24] where the integrals are performed
up to order αs2 in the case of massless quarks. It can be easily extended for massive quarks
up to order αs . However in second order the integrals for massive quarks become very
tedious and we are only able to present the results for massless quarks.
(V V ),(0)
In the Born approximation the zeroth order structure functions Wi 1 2 are deter-
mined by the following process
S(p1 ).
V (q) → H (p) + H (3.4)
In this case the phase space integral is trivial and the partonic structure functions, presented
in Eq. (A.1), are proportional to δ(1 − x). From Eqs. (2.22)–(2.39) we obtain for the
unpolarized structure functions
(0) (0) (0) 1 (0) (0)
T1 = β, T2= β 3, T3 = βρ, T4 = 0, T5 = β 2,
2
(0)
p
Im T6 = 0, β = 1 − ρ. (3.5)
The longitudinal polarized quantities in the Born approximation are given by
(0) β2 (0) (0) (0) β

T7,L1 = − , T7,L2 = 0, Im T8,L = 0, T10,L = − ,
4 4
(0) 1
T11,L = − β 3. (3.6)
4
The transverse polarized quantities equal
(0) β2 √ (0) (0) β√ (0)

T7,T =− ρ, Im T8,T = 0, T10,T =− ρ, T11,T = 0. (3.7)
4 4
For the normal polarized quantities we get
(0) (0) (0) β2 √

T9,N = 0, T12,N = 0, Im T13,N = ρ,
p 4
β = 1 − ρ. (3.8)
The order αs corrections originate from the one-loop contributions to the Born reaction in
Eq. (3.4) and the gluon bremsstrahlung process
S(p1 ) + g(p2 ).
V (q) → H (p) + H (3.9)
To facilitate the calculation we split the partonic tensor into a virtual (VIRT), a soft gluon
(SOFT) and a hard (HARD) gluon part, i.e.,
(V1 V2 ),(1)
Wµν = Wµν
(V1 V2 ),VIRT
+ Wµν
(V1 V2 ),SOFT
+ Wµν
(V1 V2 ),HARD
. (3.10)
Starting with the virtual contribution the order αs corrected vector boson quark vertex reads

Γa,µ = i γµ (1 + C1 )vaV + γ5 γµ (1 + C1 + 2C2 )aaV
V ,(1)

(2pµ − qµ ) qµ
+ C2 va + γ5
V
C3 a a ,
V
V = γ , Z. (3.11)
2m 2m
Here the functions Ci are given by
2
αs 2−ρ λ
Re C1 = CF − ln(t) + 2 ln − 4 − 3β ln(t)
4π β m2

2−ρ 1 2 2π 2
+ − ln (t) + 2 ln(t) ln(1 − t) + 2Li2 (t) + ,
β 2 3

αs ρ
Re C2 = − CF ln(t) ,
4π β

αs ρ
Re C3 = CF (ρ − 3) ln(t) − 2ρ ,
4π β
2
αs 2−ρ λ 2−ρ
Im C1 = πCF − ln 2
− 3β + − ln(t)
4π β m β

+ 2 ln(1 − t) ,

αs −ρ
Im C2 = πCF ,
4π β

αs ρ
Im C3 = πCF (3 − ρ) ,
4π β
1−β
t= , (3.12)
1+β
where CF denotes the colour factor which is equal to CF = (Nc2 − 1)/2Nc . In the equation
above we have introduced a fictitious mass of the gluon λ which is needed to regularize the
infrared divergence. The ultraviolet divergences, which cancel in the expressions above,
R Λ2 4
can be regularized with the cut-off method ( d k) which is known from old textbooks
on QED. In this way one avoids the intricacies of the γ5 -matrix prescription [38] which is
characteristic of n-dimensional regularization. The mass renormalization is carried out in
the pole mass scheme (on-shell renormalization). However one can also choose the MS-
scheme and carry out the analysis of the radiative corrections using the running mass (see,
e.g., [15]). The contributions from the Born reaction and the one-loop corrections are given
by
(V1 V2 ),(0)
Wµν + Wµν (V1 V2 ),VIRT

Nc β (1 + γ5 /s)
= Tr (/
p + m)Γ V1 ,(1)
a,µ (/
p 1 − m) e
Γ V2 ,(1)
a,ν δ(1 − x), (3.13)
32π 2 2
with Γeµ = γ0 Γµ† γ0 . The soft gluon part of the partonic tensor is given by
(V1 V2 ),SOFT
Wµν

Nc β SOFT (1 + γ5 /s)
= S Tr (/
p + m)Γ V1 ,(0)
a,µ (/
p 1 − m)Γ V2 ,(0)
a,ν δ(1 − x),
16π 2 2
(3.14)
with
Zω
αs d 3 p2 m2 m2 2p · p1
S SOFT = − CF + − . (3.15)
8π 2 E2 (p1 · p2 )2 (p · p2 )2 (p1 · p2 )(p · p2 )
0
The above integral will be evaluated in the rest frame of the vector boson V where ω is
a cut off on the gluon energy E2 which is taken to be much smaller than the quark mass
m (i.e., ω m). This is the reason why like in the case of the contribution from the Born
reaction and the virtual corrections the soft gluon part is proportional to δ(1 − x). The
result is
2
αs 4ω 2−ρ
S SOFT = CF − 2 ln + 2 − ln(t) − 2Li2 (t) − 2Li2 (−t)
4π λ2 β
2
4ω
+ ln ln(t) − 2 ln(t) ln(1 − t) − 2 ln(t) ln(1 + t)
λ2

π2
+ ln2 (t) + . (3.16)
6
(V V ),V +S
Addition of the virtual (V ) and soft (S) gluon parts leads to the expressions Wi 1 2
in Eq. (A.3). Note that in the combination C V +S = C1 + S SOFT the gluon regulator mass λ
vanishes, i.e.,
2
V +S αs 2−ρ 4ω 2−ρ
C = CF − 2 + ln(t) ln − − 4Li2 (t)
4π β m2 β
− 2Li2 (−t) − 4 ln(t) ln(1 − t) − 2 ln(t) ln(1 + t)

3 π2 5 − 4ρ
+ ln2 (t) − −2− ln(t) . (3.17)
2 2 β
The hard gluon part of the partonic tensor can be calculated in a straightforward way and
the results are given in Eq. (A.4). Notice that the upper bound on the integral over x in Eq.
(V V ),HARD
(2.12) for Wµν1 2 is given by 1 − 2mω/Q2 so that the energy cut-off ω is cancelled
(V1 V2 ),V +S
between Wi and Wi(V1 V2 ),HARD . The results for Ti are given in Eqs. (B.1)–(B.18).
We tried to shorten these expression as much as possible by minimizing the number of
independent polylogarithms. Note that there is a difference in sign between T12,N in Eq.
(B.17) and the equivalent expression in Eq. (33) of [14]. However substitution of T9,N (Eq.
(B.16)) and T12,N (Eq. (B.17)) in the cross section of Eq. (2.51) leads to the same result as
presented in Eq. (16) in [4].
The computation of the order αs2 corrections for massive quarks is extremely tedious
so that it has not been performed yet. The reason is that the partonic structure functions
for timelike processes like e+ e− -collisions cannot be written as the imaginary part of an
amplitude. This is in contrast to deep inelastic scattering where q 2 is spacelike or the total
cross section σtot (e+ e− → hadrons). In the latter case the cross section can be written
as the imaginary part of the hadronic vacuum polarization function so that one is able to
apply advanced methods to compute these type of quantities (see, e.g., [36,37]). However
for massless quarks we can compute the second order corrections to the partonic structure
functions contributing to the longitudinal polarization. This is feasible using conventional
techniques as is shown in [22–26]. In the case of massless quarks the computation is

facilitated because one has the following relations
v2 a2 x vq a q
W1 q = W1 q = − W3 ,
2
vq2 aq2 x vq a q v a
W2 = W2 = − W4 − 2W5 q q ,
2
v a x v 2 ,a 2 v 2 ,a 2
W6 q q = − W7 q q + W9 q q , (3.18)
2
which follow from Eq. (2.13) by putting s = p/m. Since m = 0 the contributions coming
from the vector currents vq2 lead to the same answer as those originating from the axial-
vector currents given by aq2 . In this case the matrix element only contains γ -matrices
which anti-commute with the γ5 -matrix. It also explains why the components proportional
to vq2 and aq2 which show up in the unpolarized structure functions are the same as the
contributions multiplying vq aq appearing in the polarized structure functions. The relations
in Eq. (3.18) break down for m 6= 0 as will be discussed at the end of this section.
The structure functions Wi in Eq. (2.13) can be related to the fragmentation functions
Fih (x, Q2 ) (unpolarized see [22–25]) and gih (x, Q2 ) (polarized see [26]) defined for the
process e+ e− → h + “X”. Here h represents a hadron in the final state which originates
from a light (anti-)quark. Using the relations in Eqs. (2.22)–(2.32), the quantities Ti can be
expressed into the first moments of the non-singlet quark coefficient functions as follows
Z1
Q2
T1 = T2 = dxC1,q x, 2 , (3.19)
µ
0
Z1
1 Q2 Q2
T3 = T4 = dx C1,q x, 2 − C2,q x, 2 , (3.20)
2 µ µ
0
Z1
Q2
T5 = dxC3,q x, 2 , (3.21)
µ
0
Z1
1 Q2
T7,L1 = − dx1C5,q x, 2 , (3.22)
4 µ
0
Z1
1 Q2 Q2
T7,L2 = − dx 1C5,q x, 2 − 1C4,q x, 2 , (3.23)
4 µ µ
0
Z1
1 Q2
T10,L = T11,L = − dx1C1,q x, 2 . (3.24)
4 µ
0
Here µ represents the mass factorization scale as well as the renormalization scale.
However since all coefficient functions above are of the non-singlet type the dependence on
the factorization scale vanishes while taking the first moments so that Ti only depends on
the renormalization scale. Because of the relations in Eq. (3.18) the coefficient functions
above are not mutually independent. In [26], it was demonstrated that one could derive
the following relations between the polarized 1Ci,q (i = 1, 4, 5) and the unpolarized Ci,q
(i = 1, 2, 3) coefficient functions. They are given by
1C1,q = C3,q , 1C4,q = C2,q , 1C5,q = C1,q , (3.25)

which means that in the case of massless quarks all quantities Ti are determined by
three independent coefficient functions only. The order αs2 contributions originate from
the processes
S + g + g,
V →H +H
S+H +H
V →H +H S,
S + q + q,
V →H +H (3.26)
which also includes the two-loop corrections to reaction (3.4) and the one-loop corrections
to process (3.9). The corresponding coefficient functions have been computed in [22–26]
for massless quarks and their first moments are presented in [17]. If we also include the
contributions from the lower order processes in (3.4) and (3.9) (see also [29–32]) we obtain
for massless quarks up to order αs2
T1m=0 = T2m=0
2
αs (µ2 ) αs (µ2 ) 2 2 7 11 Q
=1+ CF [1] + CF + CA CF − ln
4π 4π 2 3 µ2
2
347 4 Q 62
+ − 44ζ (3) + nf CF Tf ln 2
− + 16ζ (3) , (3.27)
18 3 µ 9
T3m=0 = T4m=0
2
αs (µ2 ) αs (µ2 ) 2 2 22 Q 380
= CF [2] + CF (−5) + CA CF − ln +
4π 4π 3 µ2 9
2
8 Q 136
+ nf CF Tf ln 2
− , (3.28)
3 µ 9

αs (µ2 ) 2
T5m=0 = 1 + CA CF (−44ζ (3)) + nf CF Tf (16ζ (3)) , (3.29)
4π
1
T7,L
m=0
1
= − T1m=0 , (3.30)
4
1
T7,L
m=0
2
= − T3m=0 , (3.31)
2
1
T10,L
m=0
= T11,L
m=0
= − T5 , (3.32)
4
where we have chosen the MS-scheme for the coupling constant renormalization. In the
above equations the colour factors are given by CA = Nc and Tf = 1/2 (for CF see
Eq. (3.12)). Furthermore nf is the number of light flavours and the scale µ represents
the renormalization scale. Notice that the sum T1 + T3 is equal to Re+ e− which is defined
by σtot (e+ + e− → hadrons)/σtot (e+ + e− → µ+ + µ− ). Finally the results obtained for
the spin dependent quantities T7,L1 , T10,L , T11,L depend on the way the limit m → 0 is
taken before or after the integration over the momenta. This is revealed by a comparison of
the order αs corrections in Eqs. (3.30)–(3.32) with those obtained for the same quantities
in Eqs. (B.7), (B.10), (B.11). We find the following relation
(1),m→0 (0),m→0 (1),m=0 (1),m→0 (0),m→0 (1),m=0
T7,L1 + 2T7,L1 = T7,L1 , Ti,L + 2Ti,L = Ti,L ,
i = 10, 11. (3.33)
Therefore we expect that the same difference between the massless and massive quark
approach will happen in higher order. This difference originates from the property that
the γ5 -matrix commutes with the mass term in the trace (see, e.g., Eq. (3.13)) contrary to
the ordinary gamma-matrix with which it anti-commutes. Hence the relations between the
polarized and the unpolarized partonic structure functions in Eq. (3.18) will break down
for the subleading terms. The origin of this phenomenon is explained in [15]. 2 It does
not show up in the relations between the vq2 - and aq2 -parts of the unpolarized structure
functions. Therefore only the coefficient functions 1Ci (i = 1, 4, 5) in Eqs. (3.22)–(3.24)
will get an anomalous term while going from the massless to the massive quark approach.
It also turns out that this term cancels in the combination 1C5 − 1C4 so that T7,L2 in
Eq. (3.23) will be unaffected at least up to order αs . It is unlikely that the latter will also
hold in higher order of perturbation theory.
4. Results
In this section we will present the effect of the higher order QCD corrections to the
polarization of the quark in e+ e− -collisions. When both the incoming leptons as well as
the outgoing quark are unpolarized the differential cross section is given by

d σ̄ 1 X dσ dσ
= (λe+ , λe− , W ) + (λe+ , λe− , −W )
dΩ 4 dΩ dΩ
λe+ ,λe−
1 X dσU
= . (4.1)
2 dΩ
λe+ ,λe−
When the incoming leptons are unpolarized but the quark is polarized the asymmetry is
defined by

dσW 1 X dσ dσ
= (λe+ , λe− , W ) − (λe+ , λe− , −W )
dΩ 4 dΩ dΩ
λe+ ,λe−

1 X dσL dσT dσN
= WL + WT + WN . (4.2)
2 dΩ dΩ dΩ
λe+ ,λe−
2 A similar phenomenon has been observed in polarized deep inelastic lepton–hadron scattering for the structure
function 1g1 see [33,34].
In the case the electron is polarized the asymmetry becomes equal to

dσW (λe− ) 1 X dσ dσ
= (λe+ , λe− , W ) − (λe+ , λe− , −W )
dΩ 2 dΩ dΩ
λe +
X
dσL dσT dσN
= WL + WT + WN . (4.3)
dΩ dΩ dΩ
λe +
When the positron is polarized and the electron is unpolarized we have the relation
dσW (λe+ ) dσW (−λe− )
= . (4.4)
dΩ dΩ
The polarizations of the quark are defined by
P P
λe+ ,λe− dσk /dΩ λ + dσk /dΩ
Pk = Wk P , Pk (λe− ) = Wk P e ,
λ + ,λ − dσU /dΩ
e e λ + dσU /dΩ e
k = L, T , N. (4.5)
For the longitudinally polarized quark we choose the following spin vector
b = n̂ → WL = 1,
W WT = 0, WN = 0. (4.6)
For the transversely polarized quark the spin vector is chosen to be in the plane spanned
by the electron and the quark momenta
b = n̂ × (n̂ × q̂e ) → WL = 0,
−
W WT = − sin θ, WN = 0. (4.7)
|n̂ × (n̂ × q̂e− )|
For the quark polarisation which is directed normal to the plan spanned by the electron and
the quark momenta we make the choice
b = n̂ × q̂e → WL = 0,
−
W WT = 0, WN = −1. (4.8)
|n̂ × q̂e− |
Notice that Wb is chosen in the same way as in [14] but opposite the choice made in [4]. 3
With the definitions above one can infer that on the Born level we have the following
properties. In the case of longitudinally polarized quarks (WT = 0, WN = 0) one obtains
for WL = ±1
X dσ X dσU
(λe+ , λe− = WL , WL , cos θ = 1) = (cos θ = 1),
dΩ dΩ
λe + λe +
X dσ
(λe+ , λe− = −WL , WL , cos θ = 1) = 0,
dΩ
λe +
X dσ X dσU
(λe+ , λe− = −WL , WL , cos θ = −1) = (cos θ = −1),
dΩ dΩ
λe + λe +
S is detected in the final state we have Pk (H

3 In the case the anti-quark H S) = −Pk (H ) for k = L, T but
S) = PN (H ).
PN ( H
X dσ
(λe+ , λe− = WL , , WL , cos θ = −1) = 0. (4.9)
dΩ
λe +
The above relations hold for massive and massless quarks. From Eqs. (4.3) and (4.5) it
follows
PL (λe− = ±1, cos θ = ±1) = 1, PL (λe− = ±1, cos θ = ∓1) = −1. (4.10)
For the transverse polarized cross section we also have a relation which only holds at
threshold Q → 2m. It reads for θ = π/2 (WT = −1)
X dσ X dσU
(λe+ , λe− = WT , WT , cos θ = 0) → (cos θ = 0),
dΩ dΩ
λe + λe +
X dσ
(λe+ , λe− = −WT , WT , cos θ = 0) → 0 (4.11)
dΩ
λe +
from which follows
lim PT (λe− = ±1, cos θ = 0) = ∓1. (4.12)

Q→2m
When Q > 2m we get |PT | < 1. Far away from threshold the transverse polarization
becomes very small and tends to zero. All relations above will be modified by QCD
corrections as we will see below.
We will now discuss the effect of the higher order QCD corrections on the polarizations
of the quarks where we neglect any higher order effect coming from the electro-weak
sector. Our results are obtained by choosing the parameters given in [39]. The electro-
weak constants are: MZ = 91.187 GeV/c2 , ΓZ = 2.490 GeV/c2 and sin2 θW = 0.23116.
For the strong parameters we adopt ΛMS = 237 MeV/c at nf = 5 so that the two-
loop corrected running coupling constant equals αs (MZ2 ) = 0.119. Further we take the
renormalization scale µ = Q. Furthermore we only study up-quark, bottom- and top-quark
production for which the following masses are chosen mu = 0, mb = 4.5 GeV/c2 and
mt = 173.8 GeV/c2 . In our plots we will show the polarizations computed in different
(i)
orders of perturbation theory. Hence we follow the notation in Eq. (3.1) and define Pk
(k = L, T , N ) to be the order αs contribution to the polarization. Similarly we define Pk,i
i
to be the order αsi corrected polarization. Finally we only show figures where the electron
beam is polarized. In the case the positron is polarized and the electron beam is unpolarized
the figures for λe+ = ±1 are the same as those for λe− = ∓1.
In Fig. 1 we have plotted the Born (zeroth order αs ) and the higher order corrections to
the longitudinal polarization of the up-quark at Q = MZ . The computation is done in the
massless quark approach were the anomalous terms are absent. Since the QCD corrections
for unpolarized beams are very small we could only show the order αs2 corrected result
PL,2 . For this case we have plotted the ratios PL,1 /PL,0 and PL,2 /PL,1 in Fig. 2. From
the latter figure we infer that the order αs corrections do not exceed the 6 pro-mille level.
The order αs2 corrections are at most 2 pro-mille. In the case of polarized beams they
become larger. For cos θ = ±1 the order αs and order αs2 corrections are 8.5% and 2.5%,
Fig. 1. Longitudinal polarization PL of the up-quark up to second order in αs at Q = MZ for

polarized (λe− = ±1) and unpolarized (unpol) electrons. The positron beam is unpolarized. (a) Born
PL,0 (dashed line); (b) order αs corrected PL,1 (dashed dotted line); (c) order αs2 corrected PL,2
(solid line).
respectively. Similar features are shown in Fig. 3 by bottom production at the same CM
energy which are computed in the massive quark approach so that the anomalous terms are
implicitly present. Here the corrections for polarized beams are larger than those obtained
for the up-quark and they amount to 25% at cos θ = ±1 but the corrections for unpolarized
beams are so small that they could not be shown in the figure. In Fig. 4 we have studied the
validity of the massless quark approach (m = 0) and the massive quark approach (m → 0)
for PL(1) in the case of bottom production at Q = MZ with unpolarized lepton beams. To
that order we have plotted the ratios of the various approaches with respect to the exact
result computed for m = mb = 4.5 GeV/c2 . As expected the massless approach given
by PL(1) (m = 0) does not work very well (see the dotted curve in Fig. 4) but also the
massive approach PL(1) (m → 0) is rather bad in particular near cos θ = ±1. It turns out
that only for mb < 0.2 GeV/c2 the difference between PL(1) (m → 0) and PL(1) (m = 0.2)
is less than 9%. For a comparison we have also shown the ratio of the unpolarized cross
sections d 2 σ (1) (m → 0)/d 2 σ (1) (m = 4.5) where no anomalous terms are present so that
d 2 σ (1) (m → 0) = d 2 σ (1)(m = 0). In this case the massless quark approach works rather
well except near cos θ = ±1. This is the justification for neglecting the bottom mass in
the order αs2 contributions to the forward backward asymmetry and the shape parameter
in [17,27]. In Fig. 5 we also plotted the longitudinal polarization of the top-quark which
Fig. 2. Ratios RL of the higher order and lower order corrected longitudinal polarization of the
up-quark at Q = MZ for unpolarized electrons and positrons. (a) Ratio of first order corrected and
the Born contribution to the polarization PL,1 /PL,0 (dashed line); (b) Ratio of the second order
corrected and the first order corrected polarization PL,1 /PL,0 (solid line).
is produced at Q = 500 GeV. Like in the two previous cases the order αs corrections to
PL are extremely small even when the incoming leptons are polarized. Notice that for up-
quark and bottom-quark production at Q = MZ the ZZ interference term in Eq. (2.49)
dominates the longitudinal polarization. However at Q = 500 GeV it turns out that the γ γ
(Eq. (2.45)) and γ Z (Eq. (2.53)) interference terms become important too although they
partially cancel each other because the latter is negative.
For an analysis of the transverse PT and normal PN polarization we have to limit
ourselves to heavy quark production since these quantities vanish for massless quarks.
In Fig. 6 we have shown PT at Q = MZ for bottom quarks. The QCD corrections are
much larger than for PL . This also holds for unpolarized beams. On the other hand the
absolute values for PT are rather small which is mainly due to the fact that the bottom
quark is produced far away from threshold Q = MZ 2mb so that PT ∼ 2mb /MZ 1.
For top-quark production at Q = 500 GeV (see Fig. 7) the situation is different. This quark
is produced rather close to threshold and the ratio 2mt /Q ∼ 0.7 is large enough so that for
polarized beams the maximum value of |PT | is close to one. However due to the difference
Q − 2mt there is a shift from θ = π/2 to larger angles (here about θ = 2π/3) where PT
attains its maximum (λe− = −1) or its minimum (λe− = 1).
The normal polarization PN is presented for the bottom quark (Q = MZ ) in Fig. 8.
From this figure we infer that PN is about a factor of five smaller than PT in Fig. 6.
Furthermore the difference between the Born approximation PN,0 , which is wholly due to
(0) γZ
the quantity T13,N in d 2 σN (Eq. (2.55)), and the QCD corrected result PN,1 is much larger
than observed for PL and PT . Notice that the QCD corrections to PN are not determined by
Fig. 3. Longitudinal polarization PL of the bottom-quark up to first order in αs at Q = MZ for

PL,0 (dashed line); (b) order αs corrected PL,1 (solid line).
the Z-peak in Eq. (2.51) only but they also receive contributions from the γ Z interference
(1)
term in Eq. (2.55). Actually it is the structure function T13,N (see Eq. (B.18)) which is
mainly responsible for the difference in Fig. 8 between PN,0 and PN,1 around cos θ = 0.8
for λe− = 1. When one is off-resonance the contributions in Eqs. (2.52)–(2.55) which are
proportional to Im P γ Z ∼ MZ ΓZ are heavily suppressed. This is the reason why for top
production at Q = 500 GeV (see Fig. 9) the Born approximation to the normal polarization
(0) (1)
given by PN,0 is almost zero. Therefore in this case neither T13,N nor T13,N , both appearing
in Eq. (2.55), play any role anymore. Hence PN in Fig. 9, which is smaller than PT in Fig. 7
by a factor of forty, is completely dominated by the QCD contributions coming from the
(1) (1)
structure functions T9,N (Eq. (B.16)) and T12,N (Eq. (B.17)). This observation is important
because besides of these corrections above, PN can also receive contributions coming from
CP-violating terms in the neutral current sector which is a signal of new physics. Hence it
is important to compute the QCD corrections beyond order αs which is not an easy task
as we have discussed below Eq. (3.3). Finally we want to emphasize that for the top-quark
all interference terms are equally important for the determination of PT and PN . This
observation was already mentioned for PL above. This is because t t¯-production occurs far
above the Z-boson resonance so that the dominance of the Z-propagator disappears.
As far as possible we have also made a comparison with some results obtained in the
literature. First we agree with the results presented for the top-quark for polarized and
Fig. 4. Ratios RL of order αs contributions to various quantities for unpolarized electrons and
positrons for bottom production at Q = MZ . (a) Ratio of the longitudinal polarization in the massless
quark approach with mb = 0 and the exact longitudinal polarization mb = 4.5 (dotted line); (b) Ratio
of the unpolarized cross section d 2 σ/dΩ for mb → 0 and mb = 4.5 (dashed dotted line); (c) Ratio of
the longitudinal polarization in the massive quark approach with mb → 0 and the exact longitudinal
polarization for mb = 4.5 (dashed line).
unpolarized beams shown in Figs. 2, 9 in [4]. We also found agreement with the plots
presented for PL and PT in [13–15] where the contributions T6(1) (Eq. (B.6)) and T8,L (1)
(Eq. (B.9)) were not taken into account. However these contributions are negligible. On
the other hand we disagree with the normal polarization PN plotted for the bottom-quark
(Fig. 2b) and the top-quark (Fig. 3b) in [14]. This is due to the difference in minus sign
(1)
between T12,N and the equivalent expression in Eq. (33) of [14] which we have mentioned
(1) (1)
below Eq. (3.17). Note that the signs for T9,N (Eq. (B.16)) and T12,N (Eq. (B.17)) are the
same as the contributions appearing in Eq. (16) of [4].
Before finishing this section we would like to comment on some results which are
obtained in [16] and [6]. Here one has expressed W b in Eq. (2.14) into the following
orthonormal basis (see also [40])
b = sin ξ cos ψ n̂1 + sin ξ sin ψ n̂2 + cos ξ n̂3 ,
W (4.13)
with
n̂ × (n̂ × e3 ) n̂ × e3
n̂1 = , n̂2 = , n̂3 = −n̂,
|n̂ × (n̂ × e3 )| |n̂ × e3 |
Fig. 5. Same as in Fig. 3 but now for top-quark production at Q = 500 GeV.
e1 = (1, 0, 0), e2 = (0, 1, 0), e3 = (0, 0, 1). (4.14)

Notice that n̂ is the direction of the three-momentum of the observed quark in the CM frame
of the electron–positron pair whereas n̂3 = −n̂ points into the direction of the momentum
q = qe+ + qe− in the rest frame of the observed quark. On the basis ei (i = 1, 2, 3) the
vectors n̂i take the following form:
n̂1 = (cos θ cos φ, cos θ sin φ, − sin θ ),
n̂2 = (sin φ, − cos φ, 0),
n̂3 = −(sin θ cos φ, sin θ sin φ, cos θ ). (4.15)
Substituting Eq. (4.13) in Eq. (2.16) and using the parameterization for the quark
momentum in (2.15) the spin four-vector s in the CM frame of the incoming leptons has
the following components
Qαx
s0 = − cos ξ,
2m
Qx
s1 = − cos ξ sin θ cos φ + sin ξ cos ψ cos θ cos φ + sin ξ sin ψ sin φ,
2m

Qx
s2 = − cos ξ sin θ sin φ + sin ξ cos ψ cos θ sin φ − sin ξ sin ψ cos φ,
2m

Qx
s3 = − cos ξ cos θ − sin ξ cos ψ sin θ. (4.16)
2m
Fig. 6. Transverse polarization PT of the bottom-quark up to first order in αs at Q = MZ for

PT ,0 (dashed line); (b) order αs corrected PT ,1 (solid line).
Similarly, using Eq. (4.13) one finds that
b = − cos ξ,
WL = n̂ · W
WT = Wb 3 − n̂ · W
b cos θ = − sin ξ cos ψ sin θ,
WN = Wb 2 cos φ − Wb 1 sin φ = − sin ξ sin ψ. (4.17)
Comparing the equation above with our choices of the polarization vector in Eqs. (4.6)–
(4.8) we obtain ξ = 0, π, ψ = arbitrary (longitudinal), ξ = ±π/2, ψ = 0 (transverse), and
ξ = ±π/2, ψ = π/2 (normal) respectively. Notice that the same angles were also adopted
in [12–15]. A different choice was made made in [6,16] where the values of the angles
are taken at ψ = 0 and ξ = arbitrary. In this case only the longitudinal and transverse
polarization of the heavy quark can be studied since WN = 0. The reason for this choice
is that the authors in [6,16] did not include the contributions coming from the imaginary
parts of the virtual corrections and the Z-boson propagator so that the normal polarization
vanishes. Furthermore in [16] one has adopted three different bases for the spin vector
which are called the helicity, the beamline and the off-diagonal bases. The helicity bases is
defined by the condition
cos ξ = ±1. (4.18)

Fig. 7. The same as in Fig. 6 but now top-quark production at Q = 500 GeV.
which is equivalent to our choice for the longitudinal polarization of the quark. The
beamline basis is given by
cos θ + β p
cos ξ = , β = 1 − ρ. (4.19)
1 + β cos θ
This condition only applies to the Born and the soft plus virtual gluon contribution where
there is no hard gluon emission. Here the spin of the quark is polarized along the positron
momentum in the rest frame of the quark. The above equation is not valid when there is
hard gluon emission. In the later case one obtains
q
x cos θ + αx
cos ξ = , αx = x 2 − ρ, (4.20)
x + αx cos θ
which reproduces the Eq. (4.19) in the limit x → 1 (soft gluon region) where αx → β.
Since cos ξ in Eq. (4.20) is a function of the integration variable x in Eq. (2.12) the
choice above cannot be used in those expressions where the x integration is already done.
Therefore in the case of hard gluon radiation the integral over x is very complicated since
this variable appears in the numerator as well as denominator of the expression for cos ξ .
This will lead to a non-trivial dependence of the cross section on cos θ . Note that the choice
in Eq. (4.19) at the level of hard gluon emission is at variance with the interpretation that
the spin of the top quark is polarized along the positron momentum direction in the rest
frame of the quark. The third choice, called off-diagonal basis, corresponds to the case
when the like-spin configuration of the quark anti-quark pair vanishes identically. As for
Fig. 8. Normal polarization PN of the bottom-quark up to first order in αs at Q = MZ for

approximation (γ Z-interference term) PN,0 (dashed line); (b) order αs (QCD) corrected results PN,1
(solid line).
the beamline basis this choice leads to an x-dependence of cos ξ which is non-trivial so
that it can be only applied to the Born and soft plus virtual gluon contribution.
Summarizing the results obtained in this paper we have presented the complete first or-
der QCD corrections to polarized (heavy) quark production. The most of these corrections
were already calculated in [12–15]. We agree with these results except those obtained for
the normal polarization in [14]. Further we were able to compress the expressions as far
as possible by minimizing the number of polylogarithms. Moreover we also computed the
second order QCD corrections to the production of light quarks when the latter are longitu-
dinally polarized. Furthermore we discovered that the massless quark approach which was
so successful to compute the higher order corrections to unpolarized quantities in the kine-
matical regime Q m failed in the case of the longitudinal polarization. As we have seen
for bottom-quark production this was not only due to the appearance of anomalous terms
but also to the slow convergence to the zero mass limit. Our results reveal that the QCD
corrections to the longitudinal polarization are rather small for light as well as heavy quarks
except for very small or very large values of the scattering angle θ . These corrections be-
come more important for the transverse polarization whereas in the case of the normal
polarization they completely dominate the process. For this reason it is very important to
compute the order αs2 corrections to the normal polarization which is a difficult task.
Fig. 9. Normal polarization PN of the top-quark up to first order in αs at Q = 500 GeV for polarized
(λe− = ±1) and unpolarized (unpol) electrons. The positron beam is unpolarized. (a) λe− = −1
(dashed line); (b) unpolarized electrons (solid line); (c) λe− = 1 (dotted line). The lower curves
represent the Born approximation (γ Z-interference term) PN,0 whereas the upper curves stand for
the order αs (QCD) corrected results PN,1 .
Acknowledgements
V. Ravindran would like to thank J. Blümlein, H.S. Mani and S.D. Rindani for
discussions. The work of W.L. van Neerven was supported by the EC network ‘QCD and
Particle Structure’ under contract No. FMRX-CT98-0194.
Appendix A
Since the leptonic current is conserved for massless leptons, which implies qµ Lµν =
qν Lµν = 0, we present only those partonic tensors which contribute to the cross section
(see Eqs. (2.12), (2.13)). In the Born approximation (3.4) we obtain

1 1
W1(V1 V2 ),(0) = vqV1 vqV2 − β + aqV1 aqV2 − β 3 δ(1 − x),
4 4
(V1 V2 ),(0)
W2 = − vq vq + aq aq βδ(1 − x),
V1 V2 V1 V2
(V V ),(0) 1
W3 1 2 = vqV1 aqV2 + aqV1 vqV2 βδ(1 − x),
2
W4(V1 V2 ),(0) = 0,
1
W5(V1 V2 ),(0) = vqV1 aqV2 + aqV1 vqV2 βδ(1 − x),
2
(V1 V2 ),(0) 1
W6 = vqV1 aqV2 + aqV1 vqV2 βδ(1 − x),
2
(V1 V2 ),(0)
W7 = 0,
(V1 V2 ),(0)
W8 = − aqV1 aqV2 βδ(1 − x),
(V V ),(0) 1
W9 1 2 = vqV1 vqV2 + aqV1 aqV2 βδ(1 − x),
2
(V1 V2 ),(0) 1
W10 = vqV1 aqV2 − aqV1 vqV2 βδ(1 − x),
2
(V1 V2 ),(0)
W11 = 0, (A.1)
with V1 , V2 = γ , Z. For the order αs correction (3.6) it is convenient to decompose the

partonic structure functions Wi (i = 1–9) as follows
(V1 V2 ),(1) (V1 V2 ),V +S (V1 V2 ),HARD
Wi = Wi + Wi ,
(V V ),V +S (V V ),VIRT (V V ),SOFT
with Wi 1 2 = Wi 1 2 + Wi 1 2 . (A.2)
The soft plus virtual gluon contributions are given by

(V1 V2 ),V +S β V1 V2 V +S
W1 =− v v C + aqV1 aqV2 β 2 2Re C2 + C V +S δ(1 − x),
2 q q
(V1 V 2),V +S
W2 = −2β vqV1 vqV2 Re C2 + C V +S + aqV1 aqV2 2Re C2 + C V +S δ(1 − x),

W3(V1 V2 ),V +S = β vqV1 aqV2 + aqV1 vqV2 Re C2 + C V +S

− vqV1 aqV2 − aqV1 vqV2 iIm C2 δ(1 − x),
(V1 V2 ),V +S 2β V1 V2
W4 = vq aq + aqV1 vqV2 Re C2
ρ

+ vqV1 aqV2 − aqV1 vqV2 iIm C2 δ(1 − x),

(V1 V2 ),V +S 1 1
W5 = β vqV1 aqV2 + aqV1 vqV2 − Re C2 + 3Re C2 + 2C V +S
2 ρ

1
+ vqV1 aqV2 − aqV1 vqV2 − iIm C2 − iIm C2 δ(1 − x),
ρ
V V
W6(V1 V2 ),V +S = β vq 1 aq 2 + aq 1 vq 2 Re C2 + C V +S
V V

− vqV1 aqV2 − aqV1 vqV2 iIm C2 δ(1 − x),
β
W7(V1 V2 ),V +S = − vqV1 vqV2 Re C2 + aqV1 aqV2 Re C3 δ(1 − x),
ρ

(V1 V2 ),V +S 1 V +S
W8 = −β aq aq 4Re C2 − Re C3 + 2C
V1 V2
δ(1 − x),
ρ

(V1 V2 ),V +S β V1 V2 1
W9 =− vq vq Re C2 − Re C2 − 2C V +S
2 ρ

1
+ aqV1 aqV2 − 4Re C2 + Re C3 − 2C V +S δ(1 − x),
ρ

(V V2 ),V +S 1 1
W10 1 = β vqV1 aqV2 − aqV1 vqV2 − Re C2 + 3Re C2 + 2C V +S
2 ρ

1
+ vq aq + aq vq
V1 V2 V1 V2
− iIm C2 − iIm C2 δ(1 − x),
ρ
(V V2 ),V +S 2β V1 V2
W11 1 =− v v iIm C2 δ(1 − x). (A.3)
ρ q q
The hard gluon contributions are
(V1 V2 ),HARD
W1

1
= CF vqV1 vqV2 − 20ρ − 11ρ 2 − ρ 3
αx (4 − 4x + ρ)2 (1 − x)
+ 8x + 18ρx − 2ρ 2 x − 2ρ 3 x + 8x 2 + 58ρx 2 + 23ρ 2 x 2

+ ρ 3 x 2 − 44x 3 − 74ρx 3 − 8ρ 2 x 3 + 32x 4 + 18ρx 4 − 4x 5
ln(ξ )
+ 2 ρ − ρ 2 − 2ρ 2 x − 2x 2 + 7ρx 2 + ρ 2 x 2
4αx (1 − x)

− 2ρx 3 − 2x 4

1
+ CF aqV1 aqV2 − 20ρ + 25ρ 2 + 16ρ 3
αx (4 − 4x + ρ)2 (1 − x)
+ 2ρ 4 + 8x + 10ρx − 80ρ 2 x − 20ρ 3 x + 8x 2 + 50ρx 2

+ 55ρ 2 x 2 − 44x 3 − 34ρx 3 + 2ρ 2 x 3 + 32x 4 − 6ρx 4 − 4x 5

ln(ξ )
+ 2 ρ + 2ρ 3 − 8ρ 2 x − 2x 2 + 7ρx 2 + 2ρx 3 − 2x 4 ,
4αx (1 − x)
W2(V1 V2 ),HARD

4
= CF vqV1 vqV2 4ρ + 35ρ 2 + 17ρ 3 + 2ρ 4
αx3 (4 − 4x + ρ)2 (1 − x)
+ 24x − 82ρx − 110ρ 2x − 22ρ 3 x − 40x 2 + 186ρx 2 + 85ρ 2 x 2

+ ρ 3 x 2 + 4x 3 − 126ρx 3 − 8ρ 2 x 3 + 16x 4 + 18ρx 4 − 4x 5
ln(ξ )
− 4 ρ − ρ 2 − 2ρ 3 + 10ρ 2 x + 2x 2 − 13ρx 2 − ρ 2 x 2
αx (1 − x)

+ 2ρx + 2x
3 4

4
+ CF aq aq
V1 V2
4ρ + 31ρ 2 + 18ρ 3 + 2ρ 4
αx3 (4 − 4x + ρ)2 (1 − x)
+ 24x − 106ρx − 108ρ 2x − 24ρ 3 x − 40x 2 + 274ρx 2

+ 95ρ 2 x 2 + 2ρ 3 x 2 + 4x 3 − 242ρx 3 − 18ρ 2x 3 + 16x 4 + 82ρx 4

+ 2ρ 2 x 4 − 4x 5 − 12ρx 5
ln(ξ )
− 4 ρ − 2ρ 2 − 2ρ 3 + 12ρ 2 x + 2x 2 − 15ρx 2 − 2ρ 2 x 2
αx (1 − x)

+ 6ρx 3 + 2x 4 − 2ρx 4 ,
W3(V1 V2 ),HARD
1
= CF vqV1 aqV2 + aqV1 vqV2
2
2
× − 3 68ρ 2 + 35ρ 3
αx (4 − 4x + ρ)2 (1 − x)
+ 4ρ 4 − 128ρx − 208ρ 2 x − 45ρ 3 x + 48x 2 + 332ρx 2
+ 172ρ 2 x 2 + 5ρ 3 x 2 − 112x 3 − 304ρx 3 − 52ρ 2 x 3 − 3ρ 3 x 3

+ 104x 4 + 156ρx 4 + 24ρ 2 x 4 − 64x 5 − 56ρx 5 + 24x 6
ln(ξ )
− 4 3ρ 2 + 4ρ 3 − 2ρx − 21ρ 2 x + 18ρx 2 + 5ρ 2 x 2
2αx (1 − x)

− 4x − 2ρx − 3ρ x + 6ρx − 4x ,
3 3 2 3 4 5
(V1 V2 ),HARD
W4
1
2
8
× 7 64ρ 2 + 124ρ 3
αx (4 − 4x + ρ)2
+ 45ρ 4 + 4ρ 5 − 528ρ 2x − 404ρ 3x − 63ρ 4 x + 112ρx 2
+ 976ρ 2 x 2 + 319ρ 3 x 2 + 6ρ 4 x 2 + 96ρx 3 − 432ρ 2x 3 − 18ρ 3 x 3
− 176x 4 − 764ρx 4 − 158ρ 2 x 4 − 9ρ 3 x 4 + 432x 5 + 756ρx 5

+ 75ρ 2 x 5 − 336x 6 − 206ρx 6 − ρ 2 x 6 + 80x 7 + 6ρx 7
2 ln(ξ )
− 13ρ 2 + 4ρ 3 − 18ρx − 39ρ 2 x + 58ρx 2 + 10ρ 2 x 2
αx6

− 12x 3 − 21ρx 3 + 4x 4 + ρx 4 ,
(V1 V2 ),HARD
W5
1
2
2
× − 3 80ρ + 96ρ 2
αx (4 − 4x + ρ)2 (1 − x)
+ 37ρ 3 + 4ρ 4 − 96x − 440ρx − 272ρ 2x − 46ρ 3 x + 400x 2

+ 736ρx 2 + 193ρ 2 x 2 + ρ 3 x 2 − 568x 3 − 446ρx 3 − 14ρ 2 x 3

+ 320x 4 + 76ρx 4 + ρ 2 x 4 − 56x 5 − 6ρx 5
ln(ξ )
− 4 4ρ + 5ρ 2 + 4ρ 3 − 26ρx − 22ρ 2 x + 8x 2
2αx (1 − x)

+ 49ρx 2 + ρ 2 x 2 − 16x 3 − 8ρx 3 + ρx 4 ,
W6(V1 V2 ),HARD
1
2
8
× − 8 + 12ρ
αx (4 − 4x + ρ)2 (1 − x)

+ 8ρ 2 + ρ 3 + 12x − 31ρx − 9ρ 2 x − 2x 2 + 18ρx 2 + ρx 3 − 2x 4

2 ln(ξ )
+ 2 ρ 2 + x − 3ρx + x 3 ,
αx (1 − x)
W7(V1 V2 ),HARD

2
= CF vqV1 vqV2 − 64ρ − 36ρ 2 − 4ρ 3 + 96x
αx3 (4 − 4x + ρ)2

+ 172ρx + 39ρ 2 x − 200x 2 − 102ρx 2 + ρ 2 x 2 + 104x 3 − 6ρx 3

ln(ξ )
+ − 4ρ − 4ρ 2
+ 15ρx − 8x 2
+ ρx 2
2αx4

2
+CF aqV1 aqV2 3 − 64ρ − 16ρ 2 − ρ 3 + 96x
αx (4 − 4x + ρ)2
+ 132ρx + ρ 2 x − 3ρ 3 x − 200x 2 − 46ρx 2 + 19ρ 2 x 2

+ 120x 3 − 22ρx 3 − 16x 4

ln(ξ )
+ − 4ρ − ρ + 15ρx − 3ρ x − 8x + ρx ,
2 2 2 2
2αx4
W8(V1 V2 ),HARD

2
= CF aqV1 aqV2 − 52ρ + 31ρ 2 + 4ρ 3 − 56x
αx (4 − 4x + ρ)2 (1 − x)

− 130ρx − 34ρ 2 x + 112x 2 + 72ρx 2 − ρ 2 x 2 − 56x 3 + 6ρx 3

ln(ξ )
+ 2 ρ − 4ρ + 10ρx − 8x + ρx ,
2 2 2
2αx (1 − x)
W9(V1 V2 ),HARD

1
= CF vqV1 vqV2 − − 8 − 17ρ − 4ρ 2
αx (4 − 4x + ρ)(1 − x)

+ 22x + 18ρx − 4x 2 + 3ρx 2 − 10x 3

ln(ξ )
+ 2 ρ + 4ρ 2 − 4x − 10ρx + 8x 2 − 3ρx 2 + 4x 3
4αx (1 − x)

1
+CF aqV1 aqV2 32 − 4ρ + ρ 2 − 120x
αx (4 − 4x + ρ)2 (1 − x)
+ 62ρx + 30ρ 2 x + 4ρ 3 x + 120x 2 − 144ρx 2 − 35ρ 2 x 2

− 8x 3 + 86ρx 3 − 24x 4
ln(ξ )
+ 2 ρ − 4x − 2ρx + 4ρ 2 x + 8x 2
4αx (1 − x)

− 11ρx + 4x ,
2 3
(V V2 ),HARD
W10 1
1
= CF vqV1 aqV2 − aqV1 vqV2
2
2
× 76ρ + 33ρ 2
αx (4 − 4x + ρ)2 (1 − x)
+ 4ρ 3 − 72x − 190ρx − 38ρ 2 x + 144x 2 + 120ρx 2 + ρ 2 x 2

ln(ξ )
− 72x − 6ρx + 2
3 3
ρ + 4ρ − 14ρx + 8x + ρx ,
2 2 2
2αx (1 − x)
(V V2 ),HARD
W11 1 = 0. (A.4)
In the expressions above we have introduced the following notations
q
ρ − 2x − 2αx
αx = x 2 − ρ, ξ= . (A.5)
ρ − 2x + 2αx
Appendix B
In this appendix we present the order αs contributions to the structure functions Ti

Eqs. (2.21)–(2.32) computed in Section 3. The unpolarized parts are given by

(1) 1 √
T1 = CF ρ(1 + ρ)F1 + ρ(1 − 3ρ)F2 + 2 8 − 5ρ − ρ 2 F3 + 2βF4
2
1 1
+ β(2 + 13ρ) + 64 − 39ρ − 7ρ 2 Li2 (t)
2 2
1
+ − 48 + 36ρ − 5ρ ln(t) + 2 4 − 3ρ − ρ ln(t) ln(1 + t) ,
2 2
4
(B.1)

1 √
T2(1) = CF ρ(1 + 2ρ)F1 + ρ(1 − 4ρ)F2 + 2 8 − 13ρ + 2ρ 2 F3
2
1 1
+ 2β(1 − ρ)F4 + β(2 + ρ) + 64 − 103ρ + 18ρ 2 Li2 (2)
2 2
1
+ − 48 + 60ρ − 9ρ ln(t) + 2(4 − 7ρ) ln(t) ln(1 + t) ,
2
4
(B.2)

1 √
T3(1) = CF − ρ(1 + ρ)F1 + ρ(−1 + 3ρ)F2 + 2ρ(5 − ρ)F3
2
1
+ βρF4 + 2β(1 − ρ) + ρ(39 − 9ρ)Li2 (t) + ρ(−7 + 3ρ) ln(t)
2
+ 6ρ ln(t) ln(1 + t) , (B.3)

(1) 1 √
T4 = CF − ρ(1 + 2ρ)F1 + ρ(−1 + 4ρ)F2 + 2ρ(1 + 2ρ)F3
2
1 7
+ β 8 − 38ρ + 3ρ 2 + ρ(1 + 2ρ)Li2 (t)
4 2
1
+ ρ − 32 + 8ρ − 3ρ ln(t) + 2ρ(1 + 2ρ) ln(t) ln(1 + t) , (B.4)
2
8
(1) √
T5 = CF 4β(−2 + ρ)G1 + 2(4 − 5ρ)G2 − 4 ρ + 4ρ
√ √
+ 8(−1 + ρ) ln(1 + t) + 8 ln(1 + t − t) + 16(−1 + ρ) ln(1 − t)

+ 2(2 − 4ρ + 3βρ − 2β) ln(t) , (B.5)
(1)
Im T6 = CF [2πρβ]. (B.6)
The longitudinal polarized structure functions are equal to
√
(1) 1 ρ 1
T7,L = CF β(2 − ρ)G 1 − 8 + 2ρ + 3ρ 2
G 2 − (8 + 29ρ) + (2 + 35ρ)
1 4 8 8
1 √
+ 2(1 − ρ) ln(1 + t) + − 32 + 60ρ − 17ρ 2 ln(1 + t − t)
16
√ 1
+ 4(1 − ρ) ln(1 − t) + − 32 + 4ρ + 32β
32

+ 72ρβ + 17ρ 2 ln(t) ,
(B.7)
√
(1) ρ ρ 1
T7,L2 = CF (10 + 3ρ)G2 + (8 + 13ρ) − (2 + 19ρ)
2 2 2

ρ √ ρ
+ (−24 + 7ρ) ln(1 + t − t) + (24 − 52β − 7ρ) ln(t) , (B.8)
4 8

(1) π
Im T8,L = CF − ρβ , (B.9)
2
√
(1) 5ρ ρ 1 β β
T10,L = CF F1 − (4 + ρ)F2 − (8 + ρ)F3 − F4 +
8 4 2 2 2
1 1
− (64 + 3ρ)Li2 (t) + (12 − 3ρ) ln(t)
8 4
1
− (4 + 3ρ) ln(t) ln(1 + t) , (B.10)
2

(1) ρ √ 1 β
T11,L = CF (5 − ρ)F1 − ρF2 + − 8 + 7ρ − 3ρ 2 F3 + (−1 + ρ)F4
8 2 2
β 1
+ (1 + 3ρ) + − 64 + 61ρ − 25ρ 2 Li2 (t)
2 8
1 1
+ 12 − 9ρ + ρ 2 ln(t) + − 4 + ρ − ρ 2 ln(t) ln(1 + t) .
4 2
(B.11)
The transverse polarized structure functions are

√ √
(1) √ ρ ρ ρ
T7,T = CF ρβ(2 − ρ)G1 − (16 + 7ρ)G2 + (48 + 17ρ) − (62 + 3ρ)
4 8 8
√
√ ρ √
+ 2 ρ(1 − ρ) ln(1 + t) + 40 + 2ρ − 3ρ 2 ln(1 + t − t)
16
√
√ √ ρ
+ 4 ρ(1 − ρ) ln(1 − t) + − 104 + 62ρ + 168β + 16ρβ
32

+ 3ρ 2 ln(t) . (B.12)

(1) π√
Im T8,T = CF − ρβ(1 + ρ) , (B.13)
4
√ √
(1) ρ 5ρ ρ 1√
T10,T = CF (4 + ρ)F1 − F2 + (−20 + 7ρ)F3 − ρβF4
16 8 4 2
√ √
3√ ρ ρ
− ρβ + (−156 + 57ρ)Li2 (t) + (26 − 13ρ) ln(t)
4 16 8
√
ρ
+ (−12 + 3ρ) ln(t) ln(1 + t) , (B.14)
4
√ √
(1) ρ ρ √ 1√ 7 ρ
T11,T = CF F1 + (−5 + ρ)F2 − ρ F3 + ρβ(14 − 3ρ) − Li2 (t)
4 8 8 4
√
ρ √
+ 20 − 12ρ + 3ρ 2 ln(t) − ρ ln(t) ln(1 + t) . (B.15)
16
The normal polarized structure functions are

(1) π√
T9,N = CF − ρ (1 − ρ) , (B.16)
4

(1) π√
T12,N = CF − ρ β(1 + ρ) , (B.17)
4
√ √
(1) √ ρ ρ
Im T13,N = CF ρβ(ρ − 2)G1 + (8 − 13ρ)G2 + (20 + 9ρ)
4 8
ρ √
− (3ρ + 26) − 2 ρ (1 − ρ) ln(1 + t)
8
√
ρ √ √ √
+ 24 − 2ρ − 3ρ 2 ln(1 + t − t ) − 4 ρ(1 − ρ) ln(1 − t)
16
√
ρ
+ 40 − 62ρ − 8β + 48ρβ + 3ρ 2 ln(t) ,
32
(B.18)
where Li2 (x) is defined in [19–21]. Furthermore the functions Fi (i = 1–4) and Gi
(i = 1, 2) appearing in the expressions above are given by
1
F1 = Li2 t 3 + 4ζ (2) + ln2 (t) + 3 ln(t) ln 1 + t + t 2 ,
2

F2 = Li2 −t 3/2 − Li2 t 3/2 + Li2 −t 1/2 − Li2 t 1/2 + 3ζ (2)
√ √ 3 √
+ 2 ln(t) ln(1 + t) − 2 ln(t) ln(1 − t) + ln(t) ln(1 + t − t)
2
3 √
− ln(t) ln(1 + t + t),
2
F3 = Li2 (−t) + ln(t) ln(1 − t),
F4 = 6 ln(t) − 8 ln(1 − t) − 4 ln(1 + t),

G1 = Li2 −t 3/2 − 3Li2 −t 1/2 − 4Li2 t 1/2 − Li2 (−t)
1 1
− ζ (2) − ln2 (t),
2 8
√
t 1 1 1 1
G2 = Li2 − ln2 (t) − ln(t) ln(1 + t) + ln2 (1 + t) − ζ (2), (B.19)
1+t 8 2 2 2
where the variable t is defined in (3.12). Further we are interested in the values taken by
Ti when m → 0. In the unpolarized case they become
(1),m→0 (1),m→0 (1),m→0 (1),m→0
T1 = T2 = 1, T3 = T4 = 2,
T5(1),m→0 = 0, T6(1),m→0 = 0, (B.20)
whereas the longitudinal structure functions tend to the limits
(1),m→0 1 (1),m→0 (1),m→0
T7,L1 = , T7,L2 = −1, T8,L = 0,
4
(1),m→0 (1),m→0 1
T10,L = T11,L = . (B.21)
2
All transverse and normal polarized quantities become zero in the limit m → 0.
References
[1] P.M. Zerwas (Ed.), e+ e- Collisions at TeV Energies: The Physics Potential, DESY, Hamburg,
Germany, 1996, p. 550 (DESY 96-123D).
[2] S.D. Rindani, M.M. Tung, Phys. Lett. B 424 (1998) 125.
[3] N.S. Craigie, Phys. Rept. 99 (1983) 69.
[4] J.H. Kühn, A. Reiter, P.M. Zerwas, Nucl. Phys. B 272 (1986) 560.
[5] M. Anselmino, P. Kroll, B. Pire, Phys. Lett. B 167 (1986) 113.
[6] S. Parke, Y. Shadmi, Phys. Lett. B 387 (1996) 199.
[7] J.B. Stav, H.A. Olsen, Z. Phys. C 57 (1993) 519.
[8] J.B. Stav, H.A. Olsen, Phys. Rev. D 50 (1994) 6775.
[12] J.G. Körner, A. Pilaftsis, M.M. Tung, Z. Phys. C 63 (1994) 509.
[13] S. Groote, J.G. Körner, M.M. Tung, Z. Phys. C 70 (1996) 281.
[14] S. Groote, J.G. Körner, Z. Phys. C 72 (1996) 255.
[15] S. Groote, J.G. Körner, Z. Phys. C 74 (1997) 615.
[16] J. Kodaira, T. Nasuno, S. Parke, Phys. Rev. D 59 (1999) 014023.
[17] V. Ravindran, W.L. van Neerven, Phys. Lett. B 445 (1998) 214.
[18] V. Ravindran, W.L. van Neerven, Phys. Lett. B 445 (1998) 206.
[19] L. Lewin, Polylogarithms and Associated Functions, North-Holland, Amsterdam, 1983.
[20] R. Barbieri, J.A. Mignaco, E. Remiddi, Nuovo Cimento A 11 (1972) 824.
[21] A. Devoto, D.W. Duke, Riv. Nuovo Cimento 6 (7) (1984) 1.
[22] P.J. Rijken, W.L. van Neerven, Phys. Lett. B 386 (1996) 422.
[23] P.J. Rijken, W.L. van Neerven, Phys. Lett. B 392 (1997) 207.
[24] P.J. Rijken, W.L. van Neerven, Nucl. Phys. B 487 (1998) 233.
[25] W.L. van Neerven, Acta Phys. Polon. B 29 (1998) 2573.
[26] P.J. Rijken, W.L. van Neerven, Nucl. Phys. B 523 (1998) 245.
[27] S. Catani, M.H. Seymour, JHEP 9907 (1999) 023.
[28] B. Falk, L.M. Sehgal, Phys. Lett. B 325 (1994) 509.
[29] D. de Florian, R. Sassot, Nucl. Phys. B 488 (1997) 367.
[30] V. Ravindran, Phys. Lett. B 398 (1997) 169.
[31] V. Ravindran, Nucl. Phys. B 490 (1997) 272.
[32] M. Stratmann, W. Vogelsang, Nucl. Phys. B 496 (1997) 41.
[33] R. Mertig, W.L. van Neerven, Z. Phys. C 60 (1993) 489.
[34] R. Mertig, W.L. van Neerven, Z. Phys. C 65 (1995) 360, Erratum.
[35] O.V. Teryaev, O.L. Veretin, hep-ph/9602362.
[36] K.G. Chetyrkin, J.H. Kühn, M. Steinhauser, Phys. Lett. B 371 (1996) 93.
[37] K.G. Chetyrkin, J.H. Kühn, M. Steinhauser, Nucl. Phys. B 482 (1996) 213.
[38] W.L. van Neerven, Acta Phys. Polon. B 29 (1998) 1175.
[39] C. Caso et al., Review of Particle Physics, Eur. Phys. J. C 3 (1998) 19, 24, 87.
[40] R. Hagedorn, Relativistic Kinematics, Benjamin, New York, 1964.
Asymmetric orientifolds, brane supersymmetry

breaking and non-BPS branes
Carlo Angelantonj a,b,∗ , Ralph Blumenhagen c , Matthias R. Gaberdiel d,1
a Laboratoire de Physique Théorique de l’École Normale Superiéure 24, rue Lhomond,
75231 Paris Cedex 05, France 2
b Centre de Physique Théorique, École Polytechnique, F-91128 Palaiseau, France 3
c Humboldt-Universität zu Berlin, Institut für Physik, Invalidenstrasse 110, 10115 Berlin, Germany
d Department of Applied Mathematics and Theoretical Physics, Centre for Mathematical Sciences,
Wilberforce Road, Cambridge CB3 0WA, UK
Abstract
A new class of six-dimensional asymmetric orientifolds is considered where the orientifold oper-
ation is combined with T-duality. The models are supersymmetric in the bulk, but the cancellation
of the tadpoles requires the introduction of brane configurations that break supersymmetry. These
can be described by D7-brane–antibrane pairs, non-BPS D8-branes or D9-brane–antibrane pairs.
The transition between these different configurations and their stability is analysed in detail.  2000
1. Introduction
In the course of the last two years various attempts at constructing non-supersymmetric
open string models have been undertaken [1–21]. These models always describe consistent
string compactifications, but the stability of the resulting theories is often difficult to
establish. In particular, the theories often develop tachyonic modes in certain regions of
the moduli space. This indicates an instability of the system to decay into another, quite
possibly supersymmetric, configuration.
E-mail addresses: angelant@lpt.ens.fr (C. Angelantonj), blumenha@physik.hu-berlin.de (R. Blumenhagen),
m.r.gaberdiel@damtp.cam.ac.uk (M.R. Gaberdiel).
1 Address after 1 October 2000: Department of Mathematics, King’s College London, Strand, London WC2R
2LS, UK.
2 Unité mixte du CNRS et de l’ENS, UMR 8549.
3 Unité mixte du CNRS et de l’EP, UMR 7644.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 1 8 - 6
546 C. Angelantonj et al. / Nuclear Physics B 589 (2000) 545–576
Non-supersymmetric tachyon-free models in various space–time dimensions have been

constructed using a variety of different approaches. In one approach one starts with the
ten-dimensional (non-supersymmetric) tachyonic type 0B string theory and performs a
suitable orientifold projection that removes the closed-string tachyon [2,3]. The various
projections correspond to inequivalent choices of the Klein-bottle amplitude consistent
with the constraints of [22–25]. Tadpole cancellation then requires the introduction of
D-branes, whose open-string spectrum is also non-supersymmetric; in the resulting theory
supersymmetry is therefore broken in the bulk as well as on the branes. It is also possible
to consider compactifications of these models [7,9,12,16], where the orientifold projection
is combined with an action of the space–time group.
An alternative way to break supersymmetry is via Scherk–Schwarz compactifications
[26–28]. In the simplest case of a circle compactification, higher-dimensional fields are
allowed to be periodic up to an R-symmetry transformation. Modular invariance suggests
a suitable extension of this mechanism to closed strings [29–34] and then, via orientifolds,
to open strings [4–6] as well. As a result supersymmetry is broken both in the bulk and
on the branes, 4 and different cosmological constants are generated both in the bulk and
on various branes. An interesting variant of Scherk–Schwarz compactifications involving
asymmetric Z2 orbifold projections [35–37] leads to a non-supersymmetric spectrum with
Fermi–Bose degeneracy at all mass levels. Non-abelian gauge symmetries can be generated
via orientifolds [8,10]; this leads to models where supersymmetry is preserved on the
branes (at lowest order), but is absent in the bulk.
In a different approach, named in [13] brane supersymmetry breaking, supersymmetry
is broken only in the unoriented open-string sector where different orientifold planes and
D-branes are suitably combined. In the simplest ten-dimensional case [11] the orientifold
projection involves O− planes with positive tension and positive RR charge instead of
the more familiar O+ with negative tension and negative RR charge. The projection on
the closed string spectrum is insensitive to the particular orientifold plane involved and
therefore yields the standard closed sector of the type I string, but the cancellation of the
massless RR tadpoles now requires the introduction of antibranes with negative charge and
positive tension. Thus, the orientifold projection for the open-string bosons is reverted
and leads to a USp(32) gauge group while the spinors are still in the antisymmetric
representation, consistently with anomaly cancellation. As a result, a positive cosmological
constant is generated on the branes thus reflecting the impossibility to cancel the NSNS
tadpole. In this construction the choice of the types of orientifold planes is optional and
two different open-string spectra (the supersymmetric SO(32) and the non-supersymmetric
USp(32)) can both be consistently tied to a single closed-string spectrum. In lower-
dimensional models, however, brane supersymmetry breaking is often demanded by the
consistency of the construction [13] and represents a natural solution [19] to old problems
in four-dimensional open-string model building [38–41]. Further non-supersymmetric
4 Actually, in the M-theory construction, supersymmetry is still present at the massless level, although it is
broken for the massive modes.
C. Angelantonj et al. / Nuclear Physics B 589 (2000) 545–576 547
deformations affecting only the open-string sector involve a background magnetic field
[42,43].
Finally, in all (supersymmetric and non-supersymmetric) orientifold models one has
the additional option to add pairs of branes and antibranes consistently with tadpole
cancellations [17–19,21,44]. As a result the bulk is not affected, while supersymmetry
(if present) is broken on the branes where a cosmological constant is generated. Using
similar deformations, quasi-realistic theories with three generations in the standard model,
left-right symmetric extensions of the standard model or Pati–Salam gauge groups have
been constructed [17,18,21]. Depending on the concrete model, the string scale can be in
the TeV range, or at an intermediate scale; the latter is typical for models featuring gravity
mediated supersymmetry breaking.
For models containing parallel branes and antibranes of the same dimension, the stability
of the resulting configuration is problematic. In particular, the system develops a tachyon
(and thus becomes unstable) if the branes and antibranes come close together. In order to
obtain a stable configuration it is therefore necessary to remove the moduli that describe
the relative distance between the brane and the antibrane. This can partially be achieved
by considering so-called fractional D-branes which are trapped at the fixed points of
some orbifold. However, the theory typically still contains bulk moduli that describe the
separation between the different fixed points, and it is therefore necessary to remove those
moduli as well. In all examples that have been studied so far, this could only be achieved
dynamically, and the details were out of reach of concrete computations.
In this paper we study a six-dimensional orientifold, where we combine the world-sheet
parity reversal with T-duality, or rather T-duality together with the symmetric reflection
of two directions. 5 (Similar models were also considered in [20].) As we shall show, the
Klein-bottle amplitude only leads to a twisted RR sector tadpole in our case. This is of
significance since it implies that the tadpoles cannot be canceled by any supersymmetric
brane configuration: every BPS brane carries untwisted RR charge, but since the entire
D-brane configuration must have vanishing untwisted RR charge, it will necessarily also
involve antibranes and therefore break supersymmetry. Thus one should expect that the
resulting non-supersymmetric theory (for a suitable brane configuration) is stable, and this
is indeed what we shall find.
In fact we shall find different brane configurations that cancel the tadpoles (as well as the
six-dimensional gravitational anomaly), and we shall be able to understand how they can
decay into one another. The simplest solution consists of fractional D7-brane–antibrane
pairs where the D7-branes and the anti-D7-branes are localized at different fixed points of
the underlying orbifold. 6 If the underlying torus is an orthogonal torus at the SU(2)4 point
without any B-field, the ground state of the string between the brane–antibrane pairs is
either massless or massive. Most of the bulk moduli that describe the shape of the torus are
removed by the orientifold projection (T-duality is only a symmetry for a specific class of
torii). However, certain shear deformations remain, and if the torus is deformed in this way,
5 These two descriptions are T-dual to one another.

6 The orientifold group is Z , and it contains a Z orbifold.
4 2
an open-string tachyon develops in the string between some D7-brane–antibrane pair. This
tachyon indicates the instability of the brane–antibrane system to decay into a non-BPS
D8-brane with magnetic flux. The resulting configuration of non-BPS D8-branes can be
described in detail and it also cancels the tadpoles (as well as the irreducible gravitational
and gauge anomalies); as far as we are aware, this is the first time that non-BPS D-branes
have naturally appeared in the tadpole cancellation of an orientifold model.
The configuration of non-BPS D8-branes still contains massless scalars in the open
string spectrum, and these can indeed become tachyonic if another bulk modulus is turned
on. The system then decays into a configuration involving D9-branes and anti-D9-branes
where both branes and antibranes carry magnetic flux of appropriate type. This can also be
constructed in detail, and indeed cancels the tadpoles as well as the irreducible anomalies.
The configuration is stable in a certain domain of the moduli space. However, if the torus
is tilted sufficiently, yet another tachyonic mode appears in the open string spectrum, and
the system decays into a configuration of diagonal D7-brane–antibrane pairs; this final
configuration appears to be stable.
The paper is organized as follows. We begin in Section 2 by describing the model and
the Klein-bottle amplitude. Sections 3–6 deal with various brane configurations that cancel
the RR tadpoles. In Section 7 we discuss the stability of the different configurations, and
their deformations into one another. Finally, Section 8 contains some conclusions. We have
included an appendix where the more technical material referring to the construction of
boundary and crosscap states is discussed.
2. The definition of the model and the Klein-bottle amplitude
The model that we shall discuss in this paper is the asymmetric orientifold of type IIB
string theory compactified on a 4-torus, whose coordinates are labeled by x6 , . . . , x9 . The
orientifold group is Z4 , and it is generated by ΩΘ4 , where Ω denotes the standard type
IIB orientifold, and Θ4 describes T-duality of the 4-torus. For the orthogonal self-dual
SU(2)4 torus with vanishing internal B-field, T-duality is equivalent to the asymmetric Z2
operation I4L . In the following we shall discuss mostly this case, and we shall then use the
notation I4L for T-duality.
The orientifold group contains a Z2 orbifold subgroup that is generated by I4 ; thus the
theory is equivalently described as a Z2 orientifold of type IIB on K3. The ΩΘ4 symmetry
fixes the metric of the T 4 completely, whereas the six independent values of the internal
B-field are free parameters.
It is actually more convenient to consider the theory that is obtained from the above after
T-duality in the x7 and x9 directions, say. If we denote by R the (symmetric) reflection in
these two coordinates, the theory in question is then described by
type IIB on SU(2)4

. (2.1)
(1 + I4 ) + ΩΘ4 R(1 + I4 )
Under this T-duality, some of the B-field moduli of the original model are mapped
to geometric moduli of (2.1). In fact, looking at the massless modes that survive the
projection, one finds that infinitesimally the following deformations are allowed
   
0 δg67 0 δg69 0 0 δb68 0
 δg67 0 δg78 0   0 0 0 δb79 
δG =   0
, B = .
δg78 0 δg89   −δb68 0 0 0 
δg69 0 δg89 0 0 −δb79 0 0
(2.2)
For simplicity we shall mainly consider the moduli that preserve the decomposition of
T 4 as T 4 = T 2 × T 2 , where the first T 2 has coordinates x6 , x7 , and the second has
coordinates x8 , x9 . (These moduli correspond to g67 and g89 .) Let us determine to which
global deformation of the torus these moduli correspond to. The action of the various
symmetries on the complex structure U and the Kähler structure T of T 2 is given as
Ω : (U, T ) 7→ (U, −TS),

S, −TS),
R : (U, T ) 7→ (−U (2.3)
Θ4 : (U, T ) 7→ (−1/U, −1/T ),
so that the combined action is
S, −1/T ).
ΩRΘ4 : (U, T ) 7→ (1/U (2.4)
Thus ΩRΘ4 leaves a given T 2 invariant provided that T = i and |U |2 = 1; this gives
indeed rise to a one-dimensional moduli space.
Every two-torus can be described as the quotient of the complex plane, T 2 = C/Λ,
where Λ is a lattice. The tori that have a complex structure U satisfying |U | = 1 are
characterized by the property that Λ is generated by the basis vectors
1 κ
e1 = , e2 = + iR, (2.5)
R R
where κ 2 + R 4 = 1. (More precisely U is a phase, U = exp(iφ), and φ is the angle between
e1 and e2 .) Furthermore, T = i implies that the B-field vanishes. The torus is shown in
Fig. 1.
We can write the left- and right-moving momenta as
1
pL = √ [U m1 − m2 − TS(n1 + U n2 )],
i U2 T 2
1
pR = √ [U m1 − m2 − T (n1 + U n2 )], (2.6)
i U2 T 2
and we can thus directly determine the action of ΩRΘ4 on them; this leads to

pL → −iU p̄R ,
ΩRΘ4 : (2.7)
pR → iU p̄L .
Note, that the relation (ΩRΘ4 )2 = I4 holds in general, even for U 6= i.
Fig. 1. The shape of the torus.
2.1. The Klein-bottle amplitude and the crosscap state
First we study the Klein-bottle amplitude for the model (2.1) defined on the SU(2)4
torus. We want to determine the correct crosscap states from the Klein-bottle amplitude,
which can be directly computed in the loop channel; this is given by
Z∞
dt ΩRI4L + ΩRI4R −2πt (L0 +L̄0 )
K = 8C Tr1,I4 PGSO e , (2.8)
t4 4
0
where C = V6 /(8π 2 α 0 )3 and the momentum integration over the non-compact directions
has already been performed. In the untwisted sector we get
f34 f44 − f44 f34 − f24 f04 + f04 f24
Tr1 (. . .) = (2.9)
f14 f24
with argument q = exp(−2πt). The various f -functions are defined by
√ 1 Y ∞

f0 (q) = 2 q 12 1 − q 2n = 0,
n=0
∞
Y
1
f1 (q) = q 12 1 − q 2n ,
n=1
√ 1 Y ∞

f2 (q) = 2 q 12 1 + q 2n , (2.10)
n=1
∞
Y
f3 (q) = q − 24
1
1 + q 2n−1 ,
n=1
Y∞

f4 (q) = q − 24
1
1 − q 2n−1 ,
n=1
where we have used the notation of [45]. In the I4 twisted sector the trace vanishes, as the
action of T-duality I4L on the sixteen fixed points is given by the traceless matrix [46,47]
O4
1 1 1
M= √ . (2.11)
2 1 −1
i=1
After a modular transformation to the tree channel, writing t = 1/(4l) and using the
transformation rules of the fi functions,
√
f0 e−π/ l = f0 e−πl , f1 e−π/ l = lf1 e−πl ,

f2 e−π/ l = f4 e−πl , f3 e−π/ l = f3 e−πl , (2.12)
we get
Z∞
f34 f24 − f44 f04 − f24 f34 + f04 f44
e=C
K dl 128 (2.13)
f14 f44
0
with argument q = exp(−2πl). It is immediate from this expression that (2.13) only
contains a twisted sector tadpole. This is in contrast to ordinary Z2 -orientifolds, where
only the untwisted sector propagates between the two crosscap states. The fact that in our
case only the twisted sector, g = I4 , is allowed to flow between the two crosscap states
is a consequence of the relation g = (ΩRI4L )2 = (ΩRI4R )2 . The relevant crosscap states,
|ΩRI4L,R i are characterized by the equation

Xµ (σ, 0) − RI4L,R Xµ (σ + π, 0) ΩRI4L,R = 0 (2.14)
together with a similar condition for the world-sheet fermions; the solution to these
equations is constructed in the appendix.
Under I4L (or I4R ) the two different crosscap states, |ΩRI4L i and |ΩRI4R i, are mapped
into one another. This follows from the identity

I4L ΩRI4L,R = ΩRI4R,L I4L . (2.15)
Thus the physical crosscap state is the sum of |ΩRI4L i and |ΩRI4R i. Also, since I4L
(or I4R ) maps the sixteen fixed points non-trivially into one another, the crosscap state
must involve coherent states in different twisted sectors; the most symmetric solution
involves then all sixteen fixed points equally, and this is what we shall consider in the
following. Finally, the crosscap state is constrained by the condition that the overlap with
itself reproduces the tree level Klein-bottle amplitude (2.13). This requires that it involves
components both from the twisted NSNS and the twisted RR sector, and that we have
hCL |e−lHcl |CR i = 0, (2.16)
where |CL i and |CR i denote the total crosscap states. A solution to all of these constraints
is given by
|CL i = (|L1 i + |L3 i) ⊗ (|L1 i + |L3 i) + (|L1 i + |L3 i) ⊗ (|L4 i − |L2 i)
+ (|L4 i − |L2 i) ⊗ (|L1 i + |L3 i) + (|L4 i − |L2 i) ⊗ (|L4 i − |L2 i),
|CR i = (|R1 i + |R2 i) ⊗ (|R1 i + |R2 i) + (|R1 i + |R2 i) ⊗ (|R4 i − |R3 i)

+ (|R4i − |R3 i) ⊗ (|R1 i + |R2 i) + (|R4 i − |R3 i) ⊗ (|R4 i − |R3 i). (2.17)
Here

|Li i ⊗ |Lj i = ΩRI4L NSNS,T (ij ) + ΩRI4L RR,T (ij ) , (2.18)
where |ΩRI4L i is defined in the appendix, and the twisted sector denoted by T (ij ) is
localized at the fixed point Ti of the T 2 with coordinates x6 , x7 , and at the fixed point Tj
of the T 2 with coordinates x8 , x9 ; on each T 2 we denote the different fixed points by

1 1 1 1
T1 = (0, 0), T2 = , 0 , T3 = 0, , T4 = , . (2.19)
2 2 2 2
The notation for |Ri i ⊗ |Rj i is analogous.
Both |CL i and |CR i can be thought to consist of four parallel O7-planes that ‘stretch’
between four fixed points each (and fill the uncompactified space). For example (|L1 i +
|L3 i) ⊗ (|L1 i + |L3 i) defines an O7-plane that is localized at x6 = x8 = 0 and ‘stretches’
between the four fixed points with x7 = 0, 1/2 and x9 = 0, 1/2; similar statements also
hold for the other terms in (2.17).
The N = (0, 1) supersymmetric massless spectrum consists of the supergravity
multiplet, 11 tensor multiplets and 10 hypermultiplets. There exist various configuration
of D-branes that cancel the twisted sector tadpoles from the Klein-bottle and therefore
cancel the anomalies from the closed string sector. These configurations will be discussed
in turn.
3. The D7-brane–antibrane configuration
The simplest configuration of branes that cancels the tadpoles consists of D7-branes and
antibranes that are arranged in the same way as the O7-planes. The branes in question are
so-called ‘fractional’ branes whose boundary states have components in the untwisted as
well as the twisted sectors. In particular, this implies that the branes are stuck at the fixed
planes. Since the Klein-bottle amplitude does not have any untwisted RR tadpoles, we have
to introduce D7-branes and D7-antibranes (D7) in pairs.
The boundary states of the D7-branes are schematically described by (see for example
[48] for an introduction into these matters)
|D7i = (|U, NSi + |U, Ri) + (|T, NSi + |T, Ri), (3.1)
where U denotes the untwisted sector and T the twisted sector. The normalization factors
for the untwisted and twisted sector parts have to be determined by loop channel-tree
channel equivalence. Under the action of ΩRI4L , a D7-brane is mapped to a D f7-brane
that is orthogonal to the former. In particular this implies that the tree exchange between a
D7-brane and a D f7-brane vanishes. The boundary states for the antibranes, D7, are of the
form
|D7i = (|U, NSi − |U, Ri) − (|T, NSi − |T, Ri), (3.2)
and thus the open string that stretches between a D7- and a D7-brane has the opposite GSO
projection. (It also has the opposite I4 projection.) In order to achieve a local cancellation
of the tadpoles the charge assignment for the various twisted sector ground states must be
chosen as for the O7-planes in (2.17).
We also have to guarantee that the total configuration is invariant under I4L,R , i.e., under
T-duality. Under this operation, spatial distances are exchanged with Wilson lines which
in turn are related to the relative signs of the twisted (RR) charges at different fixed points.
A consistent choice for the D7-branes and D7-branes is described in Table 1 (see also
Figs. 2 and 3). (The last column in Table 1 describes the action on the Chan–Paton factors
for the term in the open string with the insertion of I4 ; this can be determined from the
given boundary states by world-sheet duality.)
Given the branes described in Table 1, it is straightforward to compute the annulus
amplitude in the tree channel, and the result is
Z∞ 8
f3 − f48 2 2

Ã = C dl N 22
8
2
Θ0,1 + Θ1,1
f1 NSNS,U
0

f28 − f08
−8 2
Θ0,1 2
Θ1,1
f18 RR,U

f34 f24 − f44 f04 − f24 f34 + f0 f44
+8
f14 f44 NSNS−RR,T
4 4
f3 f4 − f44 f34 f44 f24 − f04 f34
+8 −8 , (3.3)
f14 f24 NSNS,U f14 f34 NSNS,T
where the argument is q = exp(−2πl), and we have used the standard definition for
X 2k(m+j/2k)2
Θj,k q 2 = q . (3.4)
m∈Z
Table 1
D7-branes for model I
Brane Location Wilson line Twisted sector (γI4 )
D71 x6 = 0, x8 = 0 θ7 = 0, θ9 = 0 (T1 + T3 )(T1 + T3 ) I

D72 x6 = 0, x8 = 12 θ7 = 0, θ9 = 12 (T1 + T3 )(T4 − T2 ) I
D73 x6 = 12 , x8 = 0 θ7 = 12 , θ9 = 0 (T4 − T2 )(T1 + T3 ) I
D74 x6 = 12 , x8 = 12 θ7 = 12 , θ9 = 12 (T4 − T2 )(T4 − T2 ) I
f71
D x7 = 0, x9 = 0 θ6 = 0, θ8 = 0 (T1 + T2 )(T1 + T2 ) I
f
D7 x7 = 0, x9 = 12 θ6 = 0, θ8 = 12 (T1 + T2 )(T4 − T3 ) I
2
f
D7 x7 = 12 , x9 = 0 θ6 = 12 , θ8 = 0 (T4 − T3 )(T1 + T2 ) I
3
f74
D x7 = 12 , x9 = 12 θ6 = 12 , θ8 = 12 (T4 − T3 )(T4 − T3 ) I
Fig. 2. The location of the D7-branes.
Fig. 3. The location of the D7-branes.
In the above, we have assumed that we have the same number, N, of each of the D-branes.
We should note that the massless untwisted RR-tadpole vanishes and that the fourth term in
(3.3) does not give rise to any massless tadpole, either. Upon world-sheet duality, writing
t = 1/(2l), the annulus amplitude becomes
Z∞
dt 2 f38 − f48 − f28 + f08
A=C N 4
Θ0,1 + Θ1,1
4
t4 f18
0
4 4
f38 + f48 − f28 − f08 f f − f44 f34 − f24 f04 + f0 f24
+ 2
2 Θ0,1 2
Θ1,1 +4 3 4
f18 f14 f24

f34 f24 − f24 f34 f24 f44 − f04 f34
+4 −4 , (3.5)
f14 f44 f14 f34
where the argument is q = exp(−πt). Here we have used that the Θj,1 functions transform
as
r
τ
Θ0,1 (e−2π/τ ) = Θ0,1 e−2πτ + Θ1,1 e−2πτ ,
2
r
τ
Θ1,1 (e−2π/τ ) = Θ0,1 e−2πτ − Θ1,1 e−2πτ . (3.6)
2
The first term in (3.5) arises from open strings stretched between parallel D7-branes or
parallel D7-branes, without the insertion of I4 in the trace. The second term comes from
open strings stretched between parallel D7- and D7-branes without I4 insertion. The third
term is due to parallel D-branes with I4 insertion in the trace, the fourth term is from
orthogonal D-branes without I4 insertion and finally the fifth term comes from orthogonal
branes with I4 insertion.
The Möbius strip amplitude is determined, in the tree channel, by the overlap between
the total crosscap state and the different D7-brane states. The different contributions
include for example
Z∞ Z∞
f34 f24 − f44 f04 − f24 f34 + f04 f44
dlhΩRI4L |e−lHcl |D71,4i = −C dl 4N ,
f14 f44
0 0
Z∞
dlhΩRI4R |e−lHcl |D71,4i = 0, (3.7)
0
where the argument is q = i exp(−2πl). Adding all these terms together, we find that the
contribution of the (twisted) RR sector is given by
Z∞ 4 4
f f − f04 f44 f24 f44 − f04 f34
Mf = C dl 32N 2 3 + . (3.8)
f14 f44 f14 f34 RR,T
0
Note that (3.8) is indeed real, as under complex conjugation one has
(f3 (iq))∗ = e− 24 f4 (iq), (f4 (iq))∗ = e− 24 f3 (iq).
πi πi
(3.9)
In order to determine the massless spectrum of the open strings we have to determine
the Möbius strip amplitude in loop channel for the different combinations separately, and
we find that
Z∞
dt ΩRI4L + ΩRI4R −2πt L0
Tr PGSO e
t 4 11̃,44̃ 4
0
Z∞
N dt f34 f24 − f44 f04 − f24 f34 + f04 f44 f44 f24 − f34 f04 − f24 f44 + f04 f34
= + ,
2 t4 f14 f44 f14 f34
0
(3.10)
with argument q = i exp(−πt), so that for D7-branes both the NS and the R sector are
symmetrized. For the two pairs of D7-branes we obtain
Z∞
dt ΩRI4L + ΩRI4R −2πt L0
Tr PGSO e
t 4 22̃,33̃ 4
0
Z∞
N dt f44 f04 − f34 f24 − f24 f34 + f04 f44 f34 f04 − f44 f24 − f24 f44 + f04 f34
= + ,
2 t4 f14 f44 f14 f34
0
(3.11)
so that the NS sector is now antisymmetrized whereas the R sector is still symmetrized.
The change of sign between M and M f is due to
f3 → f4 , f4 → e−iπ/4 f3 (3.12)
under the P = T ST 2 S transformation (that relates the tree and the loop channel Möbius
strip amplitude).
All three tree-level diagrams do not have a massless untwisted RR tadpole, and therefore
only the twisted (massless) RR tadpole needs to be canceled. It follows from (2.13), (3.3)
and (3.8) that this requires
8N2 − 64N + 128 = 8(N − 4)2 = 0, (3.13)
so that we need four D-branes of each kind. On the other hand, we cannot cancel
one untwisted and one twisted NSNS tadpole; this will therefore lead to a shift in the
background via the Fischler–Susskind mechanism [49,50].
The massless spectrum in the closed string sector consists of the N = (0, 1) supergravity
multiplet in addition to 11 tensor multiplets and 10 hypermultiplets, and the massless
spectrum arising from the different open strings is listed in Table 2.
It is worth mentioning that the sector of open strings starting and ending on the same
D-brane does not contain any scalar moduli. This implies that the D7-branes are not
allowed to move off the fixed points. This is basically a consequence of the fact that all
four fractional branes carry the same twisted RR charge; pairs of fractional branes can
move off a fixed point, but only if they carry the opposite twisted charge.
The distance between the different branes and antibranes is such that the ground state
‘tachyon’ is either massless or massive; in particular, the open string spectrum therefore
does not contain any actual tachyons. Since the branes and antibranes are fixed to lie on
the fixed points, their distance is determined in terms of the radii of the underlying torus.
However, these radii are not moduli any more since the theory is only well-defined for a
self-dual torus, and the configuration is (at least at this level) stable.
The non-supersymmetric spectrum of Table 2 also contains N+ = 256 massless fermions
of (2, 1) chirality, and N− = 144 massless fermions of (1, 2) chirality, giving rise to
1N = N+ − N− = 112; this is precisely what is needed to cancel the non-factorizable
gravitational anomaly. Moreover, the configuration of Table 2 is also free of irreducible
gauge anomalies, consistently with the RR tadpole cancellation [51,52].
Table 2
Massless open string spectrum for model I (orthogonal D7-branes)
Spin States
(2, 2) U(4)×U(4)×U(4)×U(4)
(2, 1) 2 × [(adj, 1, 1, 1) + (1, adj, 1, 1) + (1, 1, adj, 1) + (1, 1, 1, adj)]
(1, 1) 2 × [(4, 4, 1, 1) + (4, 4, 1, 1) + (4, 1, 4, 1) + (4, 1, 4, 1)]+

2 × [(1, 4, 1, 4) + (1, 4, 1, 4) + (1, 1, 4, 4) + (1, 1, 4, 4)]
(1, 2) [(10 + 10, 1, 1, 1) + (1, 10 + 10, 1, 1) + (1, 1, 10 + 10, 1) + (1, 1, 1, 10 + 10)]
(1, 1) 2 × [(10 + 10, 1, 1, 1) + (1, 6 + 6, 1, 1) + (1, 1, 6 + 6, 1) + (1, 1, 1, 10 + 10)]
(2, 1) [(4, 4, 1, 1) + (4, 4, 1, 1) + (4, 1, 4, 1) + (4, 1, 4, 1)]+

[(1, 4, 1, 4) + (1, 4, 1, 4) + (1, 1, 4, 4) + (1, 1, 4, 4)]
(1, 2) [(4, 1, 1, 4) + (4, 1, 1, 4) + (1, 4, 4, 1) + (1, 4, 4, 1)]
(1, 1) 2 × [(4, 1, 1, 4) + (4, 1, 1, 4) + (1, 4, 4, 1) + (1, 4, 4, 1)]
The arrangement of D7- and D7-branes listed in Table 1 is not the only possible
configuration of parallel D7-branes and antibranes that cancels the RR tadpole. In
fact, we can exchange the roles of some of the branes and antibranes, and consider
instead the brane configuration described in Table 3. (This configuration differs from that
described in Table 1 by the property that D72 is now a brane, whereas D74 is now an
antibrane.)
This modification only changes the annulus amplitude which, in the tree channel, now
becomes
Z∞ 8
f3 − f48
e
A = C dlN 2 2 2
Θ0,1 2 2
+ Θ1,1
8
f1 NSNS,U
0

f28 − f08
−4 2
Θ0,1 + Θ1,1
2
Θ0,1 Θ1,1
f18 RR,U
4 4
− f4 f04 − f24 f34 + f0 f44
f3 f2 4
+8
f14 f44 NSNS−RR,T
4 4
f f −f f 4 4
+ 8 3 4 4 44 3 . (3.14)
f1 f2 NSNS,U
In particular, the contribution of the massless RR sector states in (3.14) is unmodified
compared to (3.3), and therefore N = 4 still cancels the RR tadpoles. The massless
spectrum of this model is described in Table 4.
Again, the massless fermion spectrum cancels the non-factorizable gravitational
anomaly of the closed string spectrum, and ensure the vanishing of irreducible gauge
anomalies as well.
Table 3
D7-branes for model II
Brane Location Wilson line Twisted sector (γI4 )
D71 x6 = 0, x8 = 0 θ7 = 0, θ9 = 0 (T1 + T3 )(T1 + T3 ) I

D72 x6 = 0, x8 = 12 θ7 = 0, θ9 = 12 (T1 + T3 )(T4 − T2 ) −I
D73 x6 = 12 , x8 = 0 θ7 = 12 , θ9 = 0 (T4 − T2 )(T1 + T3 ) I
D74 x6 = 12 , x8 = 12 θ7 = 12 , θ9 = 12 (T4 − T2 )(T4 − T2 ) −I
f71
D x7 = 0, x9 = 0 θ6 = 0, θ8 = 0 (T1 + T2 )(T1 + T2 ) I
f72
D x7 = 0, x9 = 12 θ6 = 0, θ8 = 12 (T1 + T2 )(T4 − T3 ) −I
f
D7 x7 = 12 , x9 = 0 θ6 = 12 , θ8 = 0 (T4 − T3 )(T1 + T2 ) I
3
f
D7 x7 = 12 , x9 = 12 θ6 = 12 , θ8 = 12 (T4 − T3 )(T4 − T3 ) −I
4
Table 4
Massless open string spectrum for model II (orthogonal D7-branes)
Sector Spin States
ii (2, 2) U(4) × U(4) × U(4) × U(4)

(2, 1) 2 × [(adj, 1, 1, 1) + (1, adj, 1, 1) + (1, 1, adj, 1) + (1, 1, 1, adj)]
i(i + 2) (1, 1) 2 × [(4, 1, 4, 1) + (4, 1, 4, 1) + (1, 4, 1, 4) + (1, 4, 1, 4)]
i ĩ (1, 2) [(10 + 10, 1, 1, 1) + (1, 10 + 10, 1, 1) + (1, 1, 10 + 10, 1) + (1, 1, 1, 10 + 10)]

(1, 1) 2 × [(10 + 10, 1, 1, 1) + (1, 10 + 10, 1, 1) + (1, 1, 6 + 6, 1) + (1, 1, 1, 6 + 6)]
i (i^
+ 2) (2, 1) [(4, 1, 4, 1) + (4, 1, 4, 1) + (1, 4, 1, 4) + (1, 4, 1, 4)]
i (i^
+ 3) (1, 1) 2 × [(4, 1, 1, 4) + (4, 1, 1, 4) + (1, 4, 4, 1) + (1, 4, 4, 1)]
The two D7-brane models are actually closely related (see also [14] for a similar
construction). Suppose we add to the second theory four bulk D7-brane–antibrane pairs
that are parallel to the x6 − x8 plane, together with their images under ΩRI4L , i.e., together
with another four D7-brane–antibrane pairs in the bulk that are parallel to the x7 − x9
plane. We can then consider moving the antibranes to the fixed planes at x6 = 0, x8 =
1/2 and x7 = 0, x9 = 1/2, respectively, while moving the branes to the fixed planes at
x6 = x8 = 1/2 and x7 = x9 = 1/2, respectively. A bulk brane carries twice the untwisted
RR charge of a fractional brane, and therefore precisely changes the sign of the untwisted
RR charge (if it is of opposite sign). Thus the above operation transforms model II into
model I.
4. Non-BPS D8-branes
Both of the above models contain massless scalars in the i(i + 2) sector. This sector
consists of open strings that stretch between a D7-brane and a parallel anti-D7-brane,
for example between D71 and D73 . The ‘tachyonic’ ground state from the NS sector is
therefore invariant under the GSO-projection; its zero momentum and winding component
is removed by the orbifold projection, but states with non-trivial winding and momentum
survive. For the actual configuration that we are considering (where the brane and the
antibrane are separated by a finite distance along the x6 direction and where they carry a
relative Wilson line in the x7 direction), the lowest lying physical state is in fact massless.
This suggests that the corresponding scalars describe a marginal transformation along
which the D7–anti-D7-brane system can be deformed into a non-BPS D8-brane that fills
the space between the two D7-branes [53–56] (see also [48,57,58] for a review of these
matters).
Actually, the relevant non-BPS D8-brane is not a conventional non-BPS D-brane since
the charge distribution at the end-points can not be described in terms of constant Wilson
lines. (This is basically a consequence of the fact that the two D7-branes, D71 and D73 ,
out of which the D8-brane forms do not have the same Wilson line in the x7 direction.)
In fact, the non-BPS D8-brane into which the system decays carries a non-trivial magnetic
flux. One way to see this is to observe that the charge distribution requires that the Wilson
line in the x7 direction depends nontrivially on x6 , i.e.,
1
A6 = , A7 = x6 , A9 = 0, (4.1)
2
and therefore that the magnetic flux F67 = ∂6 A7 − ∂7 A6 = 1. Alternatively, the correct
charge distribution can be described by the superposition of a conventional non-BPS D8-
brane (where all eight twisted RR charges are +) together with a non-BPS D6-brane
that stretches along x9 (and is localised at x6 = 12 , x7 = x8 = 0), both of whose twisted
RR charges are −. Since the magnitude of the twisted RR charge at the end of a non-
BPS D6-brane is twice that of a non-BPS D8-brane, the total charge of the configuration
agrees then with that of the D7-brane–antibrane configuration. On the other hand, the open
string between the D6-brane and the D8-brane contains a tachyon, and the system decays
(presumably) into a non-BPS D8-brane with magnetic flux.
It is not difficult to describe the non-BPS D8-brane with magnetic flux in terms
of boundary states. Let us consider the configuration that is relevant to the previous
discussion: it is localised at x8 = 0, and has magnetic flux F67 = 1. The boundary
conditions for the internal directions are then
∂σ X6 + ∂τ X7 = 0,
∂σ X7 − ∂τ X6 = 0,
∂σ X8 = 0,
∂τ X9 = 0, (4.2)
and the exponential of the bosonic oscillators is of the form
X
1 6 7
|Bi = exp − α−n α̃−n − α−n
7 6
α̃−n − α−n
8 8
α̃−n + α−n
9 9
α̃−n |k, ωi (4.3)
n
n
and similarly for the fermions. Under ΩRI4L this is mapped to

X
1 6 7
e = exp
|Bi α−n α̃−n − α−n
7 6
α̃−n − α−n
8 8
α̃−n + α−n
9 9
α̃−n |k, ωi. (4.4)
n
n
The last boundary state describes a non-BPS D8-brane that is localised at x9 = 0, and that
has magnetic flux F67 = −1. This is indeed the appropriate non-BPS D8-brane into which
f can decay.
f71 and D7
the combination of D 3
Schematically speaking, the boundary state of the whole non-BPS D8-brane has the
form
|D8i = (|U, NSi + |T, Ri), (4.5)
where again the normalization is fixed by world-sheet duality to the loop channel
Z∞
dt 1 + (−1)F I4 −2πt L0
A=C Tr e . (4.6)
t4 4
0
In order to cancel the twisted sector tadpoles we need four such non-BPS D8-branes with
parameters as shown in Table 5, where again, D8 f i is the image of D8i under ΩRI L .
4
As was explained in some detail in [59], when dealing with branes with background
gauge fields two issues need special attention. Firstly, the zero-mode spectrum of the open
strings between two parallel D-branes with background gauge flux changes to
|r + U s|2 T2
M2 = , (4.7)
U2 |n + T m|2
where in our case U = T = i, n = m = 1 and n = −m = 1, respectively. Thus compared
to a D8-brane without magnetic flux one gets an extra factor of one-half for the zero mode
spectrum
r 2 + s2
M2 = , (4.8)
2
Table 5
Non-BPS D8-branes
Brane Location Wilson line Twisted sector F67
D81 x8 = 0 θ9 = 0 (T1 + T3 )(T1 + T3 ) + (T4 − T2 )(T1 + T3 ) 1

D82 x8 = 12 θ9 = 12 (T1 + T3 )(T4 − T2 ) + (T4 − T2 )(T4 − T2 ) 1
f1
D8 x9 = 0 θ8 = 0 (T1 + T2 )(T1 + T2 ) + (T4 − T3 )(T1 + T2 ) −1
f2
D8 x9 = 12 θ8 = 12 (T1 + T2 )(T4 − T3 ) + (T4 − T3 )(T4 − T3 ) −1
√
leading to an extra factor of 2 in the normalization of the D8-brane boundary states. 7 In
the overlap of two non-BPS D8-branes with F = 1 and F = −1 (for which the winding and
momentum sum is absent), one therefore obtains an extra multiplicity of two, implying that
in loop channel every state in this open-string sector is two-fold degenerate. Both effects
can easily be seen in the T-dual picture involving branes at angles. A D-brane with non-
trivial gauge flux is generally mapped to a brane wrapping around rational cycles of the T 2
different from the two fundamental ones; apparently, this changes the zero mode spectrum.
The second effect is due multiple intersection points of D-branes intersecting at angles
[60–63] (see also [64]). In our case the T-dual D-branes stretch along the main and the
off-diagonal of the T-dual torus. Therefore it is evident that we really get an extra factor
of two for open strings between a non-BPS brane with F = 1 and a non-BPS brane with
F = −1.
Taking these two effects into account, the annulus amplitude becomes in tree channel
Z∞ X 2
f38 − f48
e= C
A dl N 2
e −2πlm2
f18 NSNS,U m∈Z
0
X 2 X 2
e−πln (−1)n e−πln
2 2
× +
n∈Z n∈Z
4 4
f24 f34 − f0 f4
4 4 f3 f4 − f44 f34
−8 +8 . (4.9)
f14 f44 RR,T f14 f24
The first two terms arise from the overlap of the boundary states for two branes with
identical magnetic fields, and the last term is the untwisted part of the overlap of the
boundary states for two branes with opposite magnetic fields. Since the twisted sector
ground states for two such branes are orthogonal to each other, the twisted sector
contribution vanishes. Upon a modular transformation this becomes in loop channel
Z∞ X 2
dt 2 f38 − f28 −πt m2
A=C N e
t4 f18
0 m∈Z
X 2 X 2
12
e−2πt n e−2πt (n+ 2
2
× +
n∈Z n∈Z
4 4
f44 f34− f0 f2
4 4 f3 f2 − f24 f34
−4 +4 , (4.10)
f14 f24 f14 f44
where we have used the Poisson resummation formula,
X R X −π(nR)2/ l
e−πl(m/R) = √
2
e . (4.11)
m∈Z
l n∈Z
7 Using the Poisson resummation formula, the factor of one-half in the zero mode spectrum leads to an extra
factor
√ of two for the normalisation of the tree-channel zero mode contribution; this requires an additional factor
of 2 for the boundary state.
Note, that in (4.10) the contribution for two branes with opposite magnetic fields and
(−1)F I4 insertion vanishes identically. This implies that (−1)F I4 acts with opposite signs
on the two-fold degenerate ground states in this sector. Using the crosscap states defined
in the appendix we can determine the tree-channel Möbius amplitude
Z∞
f24 f32 f3 f4 − f04 f42 f4 f3 f24 f42 f4 f3 − f04 f32 f3 f4
f= C
M dl 32N + (4.12)
f14 f42 f3 f4 f14 f32 f4 f3
0
with argument q = i exp(−2πl). The tadpole cancellation condition is therefore as before

in (3.13), i.e., we need N = 4 non-BPS D8-branes of each kind. In loop channel, the Möbius
amplitude is then
Z∞
i π2 f2 f4 f4 f3 − f0 f3 f3 f4 −i π2 f2 f3 f3 f4 − f0 f4 f4 f3
4 2 4 2 4 2 4 2
dt
M=C 2N e + e .
t4 f14 f32 f4 f3 f14 f42 f3 f4
0
(4.13)
Taking into account that (−1)F I4 acts with opposite signs on the two-fold degenerate
ground states of the open string between a brane with F = 1 and one with F = −1, and
that at the massless level the loop channel Möbius amplitude (4.13) vanishes we derive the
massless open string spectrum presented in Table 6.
The spectrum of massless fermions cancels again the anomaly; in order for this to work,
it is important that the extra multiplicities arising from double intersection points in the
D8–D8 f sector is taken into account. This provides an independent confirmation of our
claim that the non-BPS D8-branes carry magnetic flux.
Table 6
Massless open string spectrum on non-BPS D8-branes
Sector Spin States
ii (2, 2) U(4) × U(4)

(2, 1) 4 × [(adj, 1) + (1, adj)]
i(i + 1) (1, 1) 2 × [(4, 4) + (4, 4)]
i ĩ (1, 2) [(10 + 10, 1) + (1, 10 + 10)]

(2, 1) [(6 + 6, 1) + (1, 6 + 6)]
(1, 1) 2 × [(10 + 10, 1) + (1, 10 + 10) + (6 + 6, 1) + (1, 6 + 6)]
i (i^
+ 1) (2, 1) [(4, 4) + (4, 4)]
(1, 2) [(4, 4) + (4, 4)]
(1, 1) 4 × [(4, 4) + (4, 4)]
5. The configuration with D9–D9-branes
The spectrum in Table 6 contains massless scalars in the (12) sector. They arise from
open strings that stretch between the non-BPS D8-branes localised at x8 = 0, and those
that are localised at x8 = 12 . These massless scalars describe the marginal deformation
along which the non-BPS D8-branes can be deformed into a D9–D9-brane pair.
The D9–D9-branes into which the system can decay have to carry magnetic flux in both
the x6 − x7 and the x8 − x9 directions in order to reproduce again the correct twisted RR
sector charges. The corresponding boundary states are then characterised by the equations
∂σ X6 + F67 ∂τ X7 = 0,
∂σ X7 − F67 ∂τ X6 = 0,
∂σ X8 + F89 ∂τ X9 = 0,
∂σ X9 − F89 ∂τ X8 = 0. (5.1)
For F67 = F89 = +1, the exponential of the bosonic oscillators is then of the form
X
1 6 7
|Bi = exp − α−n α̃−n − α−n α̃−n + α−n α̃−n − α−n α̃−n |k, ωi
7 6 8 9 9 8
(5.2)
n
n
and similarly for the fermions. Under the action of ΩRI4L this boundary state is mapped
to
X
1 6 7
e = exp
|Bi α−n α̃−n − α−n
7 6
α̃−n + α−n
8 9
α̃−n − α−n
9 8
α̃−n |k, ωi. (5.3)
n
n
This corresponds then to a D9-brane state with magnetic flux F67 = F89 = −1
Again schematically, the boundary states of the D9-brane and the D9-brane have the
form
|D9i = (|U, NSi + |U, Ri) + (|T, NSi + |T, Ri),
|D9i = (|U, NSi − |U, Ri) − (|T, NSi − |T, Ri), (5.4)
where the normalisations are determined by world-sheet duality. The open string loop
amplitude of the open string stretched between the D9- and the D9-brane is then
Z∞
dt 1 1 − (−1)F 1 − I4 −2πt L0
A=C Tr e . (5.5)
t4 2 2 2
0
In particular, since the I4 projection appears now with the opposite sign, the ground state
tachyon is removed. Under ΩRI4L the two boundary states (5.4) are now mapped to
f = (|U, NSi − |U, Ri) − (|T, NSi − |T, Ri)
|D9i
f = (|U, NSi + |U, Ri) + (|T, NSi + |T, Ri).
|D9i (5.6)
f
f is an antibrane and D9 a brane (since ΩR maps a D9-brane into an
In particular, D9
antibrane and vice versa). Thus we are led to consider a configuration of D9-branes and
antibranes as described in Table 7.
Table 7
D9–D9-branes
Brane Twisted sector F67 F89
D9 (T1 + T3 )(T1 + T3 ) + (T4 − T2 )(T1 + T3 ) 1 1

+(T1 + T3 )(T4 − T2 ) + (T4 − T2 )(T4 − T2 )
D9 (T1 + T3 )(T1 + T3 ) + (T4 − T2 )(T1 + T3 ) 1 1
+(T1 + T3 )(T4 − T2 ) + (T4 − T2 )(T4 − T2 )
f
D9 (T1 + T2 )(T1 + T2 ) + (T4 − T3 )(T1 + T2 ) −1 −1
+(T1 + T2 )(T4 − T3 ) + (T4 − T3 )(T4 − T3 )
f
D9 (T1 + T2 )(T1 + T2 ) + (T4 − T3 )(T1 + T2 ) −1 −1
+(T1 + T2 )(T4 − T3 ) + (T4 − T3 )(T4 − T3 )
Table 8
Massless open string spectrum on D9–D9-branes.
Sector Spin States
ii (2, 2) U(4) × U(4)

(2, 1) 2 × [(adj, 1) + (1, adj)]
i(i + 1) (1, 1) 8 × [(4, 4) + (4, 4)]

(2, 1) 2 × [(4, 4) + (4, 4)]
i ĩ (2, 1) 2 × [(6 + 6, 1) + (1, 6 + 6)]

(1, 1) 2 × [(10 + 10, 1) + (1, 10 + 10) + (6 + 6, 1) + (1, 6 + 6)]
i (i^
+ 1) (1, 2) 2 × [(4, 4) + (4, 4)]
(1, 1) 4 × [(4, 4) + (4, 4)]
The annulus amplitude in tree channel then becomes

Z∞ 8 X 2 2 −2κrs 2
f −f8 −2πl r +s
√
e= C
A dlN2 2 3 8 4 e 1−κ 2
f1 NSNS,U r,s∈Z
0

f24 f34 − f04 f44 f34 f44 − f44 f34 − f24 f04 + f04 f24
−8 +8 ,
f14 f44 RR,T f14 f24
(5.7)
where we have introduced, for later convenience, κ = κ67 = κ89 to denote the tilt of the
x6 − x7 and the x8 − x9 torus. Using the crosscap states defined in the appendix we can
also determine the tree-channel Möbius amplitude 8
8 For κ 6= 0, the crosscap states are modified in the obvious way; this does not effect the Klein-bottle amplitude
however.
Z∞
f24 f32 f42 − f04 f42 f32
f= C
M dl 64N , (5.8)
f14 f32 f42
0
and thus read off the tadpole cancellation condition. This turns out to be the same as (3.13),
independent of κ, and we thus have to choose N = 4. The massless spectrum depends on
the other hand on κ (as we shall discuss below); for κ = 0 it is shown in Table 8. As before,
the massless fermions cancel the non-factorizable gravitational anomaly.
6. The configuration with diagonal D7-branes
In the previous sections we have described a number of different tadpole cancelling

configurations that are related to each other by standard deformations. There exists one
other interesting brane distribution that is not so obviously related to these configurations,
but that is relevant for the stability analysis of the theory. This configuration consists of
D7-branes and antibranes that stretch diagonally across the tori. If we denote by y1 and y2
the coordinate along the main diagonal of the two T 2 s, this brane configuration is described
in Table 9 (see also Fig. 4).
Table 9
Diagonal D7–D7-branes
Brane Location Wilson line Twisted sector
D7 y1 = y2 = 0 θ1 = θ2 = 0 (T1 + T4 )(T1 + T4 )
D7 y1 = y2 = 12 θ1 = θ2 = 12 (T3 − T2 )(T3 − T2 )
Fig. 4. The configuration with diagonal D7-branes.

From the point of view of the boundary states, it is actually not surprising that this
configuration also cancels the tadpoles. The total crosscap state |Ci = |CL i + |CR i only
has twisted RR charge at eight of the sixteen corners, since the contributions from |CL i and
|CR i cancel at the other eight fixed points. (This can be directly seen from (2.17).) Where
the contributions add, the charge is twice as large as before, and we therefore expect that
we shall need eight D7-brane–antibrane pairs of the above type. This is indeed correct: the
tree channel annulus amplitude of the above brane system is
Z∞ 8
1 1−κ f3 − f48
e= C
A dl N 2
2 1+κ f18 NSNS,U
0
X q 4 X q 4
−2πl 1−κ 2
1+κ r −2πl 1−κ 2
1+κ r
× e + (−1)r e
r∈Z r∈Z

1 1−κ − f0 8 f28
−
2 1+κ f18 RR,U
X q 4 X q 4
−2πl 1−κ
1+κ r
2
r −2πl
1−κ 2
1+κ r
× e − (−1) e
r∈Z r∈Z
4 4
f3 f2 − f44 f04 − f24 f34 + f04 f44
+2 , (6.1)
f14 f44 NSNS−RR,T
where we have again performed the calculation for general tilt parameter κ. On the other
hand, the contribution to the twisted RR tadpole of the tree channel Möbius amplitude is
Z∞ 2 0 2 !
ϑ 1/2 ϑ 1/4
Mf = C dl 32N 0
(6.2)
0 2
η 6ϑ
0 1/4 RR,T
with
α ∞
ϑ α2 Y
2 − 24 1 + q n− 2 +α e2πiβ
β 1 1
(q) = e2πiαβ q
η
n=1

× 1 + q n− 2 −α e−2πiβ
1
(6.3)
and
∞
Y
1
η(q) = q 24 1 − qn . (6.4)
n=1
Thus the tadpole cancellation condition becomes
2N2 − 32N + 128 = 2(N − 8)2 = 0. (6.5)
This requires indeed N = 8 D7-branes and D7-branes. For κ = 0, the massless open string
spectrum that follows from this configuration is given in Table 10.
In this case there are 112 fermions of spin (2, 1), and this cancels indeed the non-
factorizable gravitational anomaly. There also exists the configuration for which the D7-
Table 10
Massless open string spectrum on diagonal D7–D7-branes
Sector Spin States
ii (2, 2) Sp(8) × SO(8)

(2, 1) 2 × [(28, 1) + (1, 28)]
ii (1, 1) 8 × (8, 8)
branes and antibranes stretch along the off-diagonal; the analysis of this case is identical to
the above.
7. Regimes of stability
In the previous sections we have discussed a number of different tadpole cancelling

configurations, all of which are free of tachyons at the point in moduli space where the
torus is an orthogonal SU(2)4 torus. In this section we want to explore which of these
configurations is stable at a more general point in moduli space; we shall only consider a
one-parameter subspace of the six-dimensional moduli space of tori (that was described in
Section 2), but it is clear that at least the essential arguments and observations will hold
more generally.
Let us consider the deformation of the torus where we tilt both the (67) and the (89)
torus, and let us, for simplicity, assume that κ = κ67 = κ89 . As we increase κ, the distance
between a brane at xi = 0, and an antibrane at xi = 1/2 is reduced; in particular, if the
‘tachyonic’ open string between these two branes is massless for κ = 0, it will become
tachyonic for κ 6= 0. This is precisely what happens for the (13) string of the original
orthogonal D7–D7-brane system, and therefore the marginal perturbation along which
this system can be deformed into a system of non-BPS D8-branes becomes relevant. The
argument also applies to the i(i + 1) sector of the non-BPS D8-brane system, and the
marginal deformation of this system into the D9–D9-brane system becomes also relevant.
Thus, for κ 6= 0, either of the configurations described in Sections 3 and 4 decays into the
D9–D9-brane configuration described in Section 5
D7–D7 (Section 3) −→ non-BPS–D8 (Section 4)

−→ D9–D9 (Section 5). (7.1)
On the other hand, from the point of view of the D9–D9-system, the massless scalars in
the i(i + 1) sector of Table 8 (that describe the marginal deformation back to the non-BPS
D8-brane system) become massive for κ 6= 0. Indeed, the mass formula for the KK states
on each T 2 is given by
T2 |r − U s|2 1
M2 = −
|n + T m|2 U2 2
r 2 + s 2 − 2κrs 1
= √ − . (7.2)
2 1 − κ2 2
For κ = 0, the massless states in the i(i + 1) sector correspond to states with r = ±1, s = 0
or r = 0, s = ±1. If κ 6= 0, these states become indeed massive. (This also follows directly
from the loop amplitude
Z∞ X 2 2 −2κrs 2
dt 2 f38 − f28 +s
−πt r √
A=C N e 1−κ 2
t4 f18
0 r,s∈Z

f44 f34 − f04 f24 f34 f24 − f44 f04 − f24 f34 + f04 f44
−4 +4 , (7.3)
f14 f24 f14 f44
that can be obtained by a modular transformation from the tree channel amplitude (5.7).
For 0 < κ < 3/5, the open string spectrum of the D9–D9-brane system is tachyon free,
and this should imply that the configuration is indeed stable. At κ = 3/5, however, the
states with (r, s) = ±(1, 1) become massless, and for 3/5 < κ < 1, in fact tachyonic.
This implies that the stable configuration in this domain is described by another system.
Intuitively, these tachyons arise because for κ > 3/5, the D9-branes are stretched too much
along the main diagonal direction of the torus, and it becomes energetically preferable
to decay into two non-BPS D8-branes that stretch along the off-diagonal direction. The
corresponding brane configuration can be constructed, but it is also always unstable to
decay into the diagonal D7-brane system that we described in Section 6. Thus we find
that the stable configuration for 3/5 < κ < 1 is described by diagonal D7-branes and
antibranes!
Conversely, we can analyze the stability of the diagonal D7-brane system for all values
of κ. The mass formula for KK and winding states for the off-diagonal branes on each T2
is given by
r
r 2 + s2 1 + κ 1
M2 = − . (7.4)
2 1−κ 2
For κ > 0 the system is therefore non-tachyonic and the massless states in the ii sector
become massive for κ > 0. For κ 6 0 the open string between the D7- and the D7-brane
becomes tachyonic and the system decays into a D7 and a D7 along the off-diagonal.
It is worth mentioning that the system does not develop a marginal deformation (let alone
a relevant deformation) for κ = 3/5; this suggests that the configuration of diagonal D7-
branes and antibranes is actually stable for all values of κ, and that the D9-brane–antibrane
system is only metastable (for |κ| < 3/5).
Finally, we have to address the question of what the effect of the dynamically generated
potential for κ is. In the D9-brane configuration, the NSNS tadpoles do not depend on κ
(as follows from (5.7)), and the first κ dependent contribution to the potential arises at one
loop. Since there exist NSNS tadpoles, we are not really sitting in a string theory vacuum
and the background fields get modified by the Fischler–Susskind mechanism. Nevertheless,
since the κ dependence of Λ1-loop only arises via the Kaluza–Klein and winding modes,
Fig. 5. Λ1-loop (κ) − Λ1-loop (0) for the D9-brane system.
we are confident that qualitative features of the κ dependence can be reliably extracted
from the one-loop partition functions computed above. In particular we compute
Λ1-loop (κ) − Λ1-loop(0) = −A(κ) + A(0), (7.5)
where in fact the contributions from the torus, the Klein-bottle and the Möbius strip vanish
and only the first term in (7.3) contributes. Numerically evaluating (7.5 yields the curve
depicted in Fig. 5.
It follows from this result that to one-loop, κ = 0 is a stable minimum. It is separated by
a finite potential barrier from the configuration of diagonal D7-branes (into which it can
decay at κ = 3/5).
For the configuration of diagonal D7-branes, the NSNS tadpole depends on κ, thus
giving rise to a tree level contribution to the potential (see Fig. 6)
r
−Φ 1−κ
V (Φ, κ) ∼ e c N + V1-loop + · · · . (7.6)
1+κ
Fig. 6. The tree level potential for κ for the configuration with diagonal D7-branes.
The potential is minimized in the singular limit κ = 1; this simply expresses the fact that
the tension of the D7-branes pulls the two sides of each T 2 together. However, the point
κ = 1 is infinitely far away in moduli space and does not represent an actual decay mode.
Moreover, it might also happen that higher loop or non-perturbative contributions to the
potential stabilize κ at a finite value 0 6 κ < 1. The actual form of these contributions is
however beyond computational control.
8. Conclusion
In this paper we have constructed a new kind of asymmetric orientifold in six dimensions
which is supersymmetric in the bulk and non-supersymmetric on the branes. Tadpole
cancellation naively led to the introduction of pairs of fractional D7- and D7-branes,
which were localized on different fixed points, thus preventing the development of
tachyons. However, by turning on some of the closed string moduli the configuration
of D7–D7-branes became unstable and via non-BPS D8-branes eventually decayed into
pairs of D9–D9-branes with magnetic flux. For this configuration we computed the one-
loop cosmological constant and found that the system is stabilized on the SU(2)4 torus.
However, this configuration is only metastable, and it is separated by a finite energy barrier
from the stable system consisting of diagonal D7-branes and antibranes.
One of the lessons of this analysis is that models that are tachyon-free at first sight (such
as the configuration described in Section 3) can in fact be highly unstable, and may well
decay into more exotic brane configurations. In fact, in order to get control over the stable
configurations, it is important to analyze the various closed string moduli in some detail.
As far as we are aware, the present paper is the first example where this has been done to
any degree.
Most of the configurations that we found were at criticality in the sense that some
of the tachyonic modes were precisely massless. Non-supersymmetric configurations at
criticality sometimes lead to precise Bose–Fermi degeneracy in the open string spectrum
[65]. However, for the class of configurations that we considered, here, this did not occur;
it would be interesting to find a model where Bose–Fermi degeneracy is realized in the
supersymmetry breaking open string sector.
Finally, if we compactify the six-dimensional model on a further T 2 and T-dualise this
torus, we obtain a four-dimensional model, for which the latter torus can be made large,
thus leading to some brane world scenario. It would therefore be interesting to generalize
our approach directly to four-dimensional models.
Acknowledgements
C.A. is supported by the “Marie Curie” fellowiship HPMF-CT-1999-00256, in part

by the EEC-IHP “Superstring Theory” contract HPRN-CT-2000-00122, in part by
the EEC contract ERBFMRX-CT96-0090, in part by the EEC contract ERBFMRX-
CT96-0045 and in part by the INTAS project 991590. R.B. is supported in part by
the EEC contract ERBFMRX-CT96-0045, and thanks the Erwin Schrödinger Institute
for Mathematical Physics, where part of this work has been performed. M.R.G. is
supported by a Royal society University Research Fellowship, and acknowledges partial
support from the PPARC SPG programme, “String Theory and Realistic Field Theory”,
PPA/G/S/1998/00613.
We would like to thank Gary Shiu for collaboration during an early stage of this work.
We also thank Massimo Bianchi, Dieter Lüst, José F. Morales and Augusto Sagnotti for
helpful discussions.
Appendix A. The crosscap states
Let us recall from (2.14) that the crosscap states |ΩRI4L,R i satisfy the equation

Xµ (σ, 0) − RI4L,R Xµ (σ + π, 0) ΩRI4L,R = 0, (A.1)
where µ = 2, . . . , 9. In terms of the oscillator modes, this can be rewritten as
m

αrm ± m e−iπr α̃−r ΩRI4L,R = 0, for m ∈ {6, 7, 8, 9},
m

αrm ∓ m eiπr α̃−r ΩRI4L,R = 0, for m ∈ {6, 7, 8, 9}, (A.2)
µ
αnµ + (−1)n α̃−n ΩRI4L,R = 0, for µ ∈ {2, 3, 4, 5},
where

+1, for m = 6, 8,
m = (A.3)
−1, for m = 7, 9.
The upper sign in (A.2) corresponds to |ΩRI4L i, whereas the lower sign refers to |ΩRI4R i.
The first two conditions in (A.2) are compatible only for r ∈ Z + 1/2, confirming our
general observation that the crosscap state is a coherent state in the I4 twisted sector. The
crosscap conditions (A.2) also give rise to a relation among the zero-modes which, as
expected, can only be solved trivially. The conditions that arise for the fermionic modes
are similar

m

ψrm ± im ηe−iπr ψ̃−r ΩRI4L,R , η = 0, for m ∈ {6, 7, 8, 9},
µ
ψ µ + iηe−iπr ψ̃ ΩRI L,R , η = 0, for µ ∈ {2, 3, 4, 5},
r −r 4 (A.4)
where, as usual, η = ±1 gives rise to the different spin structures. The solution to these
equations is given by

ΩRI L,R , η
4
X
5 X
(−1)n µ µ
X X e±iπr
= M exp − α−n α̃−n + m m
α−r α̃−r
n r
µ=2 n∈Z m∈{6,8} r∈Z+
1
2
X X e∓iπr
+ m m
α−r α̃−r
r
m∈{7,9} r∈Z+ 1
2
"
X
5 X X X
e−iπr ψ−r ψ̃−r ∓ e−iπr ψ−r
µ µ
+ iη − m m
ψ̃−r
µ=2 r m∈{6,8} r
#!
X X
± e−iπr ψ−r
m m
ψ̃−r |TL,R , ηi, (A.5)
m∈{7,9} r
where the overall normalization M is determined by world-sheet duality. The moding of

µ µ
the fermions ψr and ψ̃r depends on whether we are considering the twisted NSNS or the
twisted RR sector of the theory.
The ground states in the twisted sectors are constrained by the conditions that arise
from the fermionic zero modes in (A.4). In the twisted RR sector, the theory has only
fermionic zero modes in the directions unaffected by the orientifold, and therefore the
standard argument applies (see for example [45]). On the other hand, there are fermionic
zero modes for ψ0m with m = 6, 7, 8, 9 in the twisted NSNS sector, and they give rise to
the conditions
m
ψ+ |TL , +i = 0, for m = 6, 8,
m
ψ− |TL , +i = 0, for m = 7, 9 (A.6)
and similarly for TR ,

m
ψ− |TR , +i = 0, for m = 6, 8,
m
ψ+ |TR , +i = 0, for m = 7, 9. (A.7)
Here we have defined
1
m
ψ± = √ ψ0m ± i ψ̃0m . (A.8)
2
For the IIB orbifold under consideration, the GSO-projection in the twisted NSNS sector
is given by
1 e
1 + (−1)F 1 + (−1)F , (A.9)
4
e
where the two operators (−1)F and (−1)F are defined by
Y
9 √ Y
9

(−1)F = 2 ψ0m = m
ψ+ + ψ−
m
,
m=6 m=6
e Y
9 √ m Y 9

(−1)F = 2 ψ̃0 = m
ψ+ − ψ−
m
. (A.10)
m=6 m=6
Thus if we define
|TL , −i = ψ− ψ+ ψ− ψ+ |TL , +i,
6 7 8 9
(A.11)
from which it follows that
|TL , +i = ψ+ ψ− ψ+ ψ− |TL , −i,
6 7 8 9
(A.12)
we have that
(−1)F |TL , ±i = |TL , ∓i,
e
(−1)F |TL , ±i = |TL , ∓i, (A.13)
and therefore
(|TL , +i + |TL , −i) (A.14)
is a GSO-invariant state. Since the GSO-operators act in the standard way on the oscillator
exponential, this implies that

ΩRI L ≡ ΩRI L , + + ΩRI L , − (A.15)
4 4 4
is GSO-invariant. The analysis for ΩRI4R is identical since the comparison of (A.6) with
(A.7) implies that we can define |TR , +i = |TL , −i and |TR , −i = |TL , +i. This analysis
applies separately for each twisted sector of the theory.
The actual crosscap states also have to be invariant under the orientifold projection
ΩRI4L (which generates the whole orientifold group). Since these crosscap states are
effectively O7-planes, the invariance under ΩR is familiar. In order to understand this
more explicitly (compare [66] for a similar analysis), we recall that Ω acts on the fermionic
modes as
Ωψrm Ω −1 = ψ̃rm ,
Ω ψ̃rm Ω −1 = −ψrm , (A.16)
so that
m −1
Ωψ+ Ω = −iψ+
m
,
m −1
Ω ψ̃− Ω = +iψ−
m
. (A.17)
If we denote by |C9, ηi the ground state that satisfies (A.4) without m , then Ω is defined
to satisfy Ω|C9, ηi = |C9, ηi. Since |TL , +i = ψ− ψ− |C9, +i, it then follows that |TL , ηi
7 9
has eigenvalue −1 under the action of Ω, i.e., Ω|TL , ηi = −|TL , ηi. On the other hand, R
acts on the ground states as
7 9 9
R = 4ψ07 ψ̃07 ψ09 ψ̃09 = − ψ+
7
+ ψ−
7
ψ+ − ψ− 7
ψ+ + ψ−9
ψ+ − ψ− 9
, (A.18)
and thus R|TL , ηi = −|TL , ηi. This implies that ΩR leaves |TL , ηi invariant. Again, the
action on the oscillator states is trivial, and therefore also |ΩRI4L i is invariant under ΩR.
The same argument obviously also applies to |ΩRI4R i.
References
[1] M. Bianchi, A. Sagnotti, On the systematics of open string theories, Phys. Lett. B 247 (1990)
517.
[2] A. Sagnotti, Some properties of open string theories, hep-th/9509080.
[3] A. Sagnotti, Surprises in open string perturbation theory, hep-th/9702093.
[4] I. Antoniadis, E. Dudas, A. Sagnotti, Supersymmetry breaking, open strings and M-theory,
Nucl. Phys. B 544 (1999) 469, hep-th/9807011.
[5] I. Antoniadis, G. D’Appollonio, E. Dudas, A. Sagnotti, Partial breaking of supersymmetry, open
strings and M-theory, Nucl. Phys. B 553 (1999) 133, hep-th/9812118.
[6] I. Antoniadis, G. D’Appollonio, E. Dudas, A. Sagnotti, Open descendants of Z2 × Z2 freely
acting orbifolds, Nucl. Phys. B 565 (2000) 123, hep-th/9907184.
[7] C. Angelantonj, Non-tachyonic open descendants of the 0B string theory, Phys. Lett. B 444
(1998) 309, hep-th/9810214.
[8] R. Blumenhagen, L. Görlich, Orientifolds of non-supersymmetric asymmetric orbifolds, Nucl.
Phys. B 551 (1999) 601, hep-th/9812158.
[9] R. Blumenhagen, A. Font, D. Lüst, Tachyon free orientifolds of type 0B strings in various
dimensions, Nucl. Phys. B 558 (1999) 159, hep-th/9904069.
[10] C. Angelantonj, I. Antoniadis, K. Förger, Non-supersymmetric type I strings with zero vacuum
energy, Nucl. Phys. B 555 (1999) 116, hep-th/9904092.
[11] S. Sugimoto, Anomaly cancellations in the type I D9–anti-D9-system and the USp(32) string
theory, Prog. Theor. Phys. 102 (1999) 685, hep-th/9905159.
[12] R. Blumenhagen, A. Kumar, A note on orientifolds and dualities of type 0B string theory, Phys.
Lett. B 464 (1999) 46, hep-th/9906234.
[13] I. Antoniadis, E. Dudas, A. Sagnotti, Brane supersymmetry breaking, Phys. Lett. B 464 (1999)
38, hep-th/9908023.
[14] G. Aldazabal, A.M. Uranga, Tachyon-free non-supersymmetric type IIB orientifolds via brane–
antibrane systems, JHEP 9910 (1999) 024, hep-th/9908072.
[15] C. Angelantonj, Comments on open-string orbifolds with a non-vanishing Bab , Nucl. Phys.
B 566 (2000) 126, hep-th/9908064.
[16] K. Förger, On non-tachyonic ZN × ZM orientifolds of type 0B string theory, Phys. Lett. B 469
(1999) 113, hep-th/9909010.
[17] G. Aldazabal, L.E. Ibanez, F. Quevedo, Standard-like models with broken supersymmetry from
type I string vacua, JHEP 0001 (2000) 031, hep-th/9909172.
[18] G. Aldazabal, L.E. Ibanez, F. Quevedo, A D-brane alternative to the MSSM, hep-ph/0001083.
[19] C. Angelantonj, I. Antoniadis, G. D’Appollonio, E. Dudas, A. Sagnotti, Type I vacua with brane
supersymmetry breaking, hep-th/9911081, to appear in Nucl. Phys. B.
[20] M. Bianchi, J.F. Morales, G. Pradisi, Discrete torsion in non-geometric orbifolds and their open
string descendants, Nucl. Phys. B 573 (2000) 314, hep-th/9910228.
[21] G. Aldazabal, L.E. Ibanez, F. Quevedo, A.M. Uranga, D-branes at singularities: a bottom-up
approach to the string embedding of the Standard Model, hep-th/0005067.
[22] D. Fioravanti, G. Pradisi, A. Sagnotti, Sewing constraints and non-orientable open strings, Phys.
Lett. B 321 (1994) 349, hep-th/9311183.
[23] G. Pradisi, A. Sagnotti, Ya.S. Stanev, Planar duality in SU(2) WZW models, Phys. Lett. B 354
(1995) 279, hep-th/9503207.
[24] G. Pradisi, A. Sagnotti, Ya.S. Stanev, The open descendants of non-diagonal SU(2) WZW
models, Phys. Lett. B 356 (1995) 230, hep-th/9506014.
[25] G. Pradisi, A. Sagnotti, Ya.S. Stanev, Completeness conditions for boundary operators in 2D
conformal field theory, Phys. Lett. B 381 (1996) 97, hep-th/9603097.
[26] J. Scherk, J.H. Schwarz, Spontaneous breaking of supersymmetry through dimensional
reduction, Phys. Lett. B 82 (1979) 60.
[27] J. Scherk, J.H. Schwarz, How to get masses from extra dimensions, Nucl. Phys. B 153 (1979)
61.
[28] E. Cremmer, J. Scherk, J.H. Schwarz, Spontaneously broken N = 8 supergravity, Phys. Lett.
B 84 (1979) 83.
[29] R. Rohm, Spontaneous supersymmetry breaking in supersymmetric string theories, Nucl. Phys.
B 237 (1984) 553.
[30] C. Kounnas, M. Porrati, Spontaneous supersymmetry breaking in string theory, Nucl. Phys.
B 310 (1988) 355.
[31] S. Ferrara, C. Kounnas, M. Porrati, F. Zwirner, Superstrings with spontaneously broken
supersymmetry and their effective theories, Nucl. Phys. B 318 (1989) 75.
[32] C. Kounnas, B. Rostand, Coordinate dependent compactifications and discrete symmetries,
Nucl. Phys. B 341 (1990) 641.
[33] I. Antoniadis, C. Kounnas, Superstring phase transition at high temperature, Phys. Lett. B 261
(1991) 369.
[34] E. Kiritsis, C. Kounnas, Perturbative and non-perturbative partial supersymmetry breaking:
N = 4 → N = 2 → N = 1, Nucl. Phys. B 503 (1997) 117, hep-th/9703059.
[35] S. Kachru, E. Silverstein, 4d conformal theories and strings on orbifolds, Phys. Rev. Lett. 80
(1998) 4855, hep-th/9802183.
[36] S. Kachru, J. Kumar, E. Silverstein, Vacuum energy cancellation in a non-supersymmetric
string, Phys. Rev. D 59 (1999) 106004, hep-th/9807076.
[37] J.A. Harvey, String duality and non-supersymmetric strings, Phys. Rev. D 59 (1999) 026002,
hep-th/9807213.
[38] A. Sagnotti, Anomaly cancellations and open-string theories, hep-th/9302099.
[39] G. Zwart, Four-dimensional N = 1 ZN × ZM orientifolds, Nucl. Phys. B 526 (1998) 378, hep-
th/9708040.
[40] Z. Kakushadze, G. Shiu, S.-H.H. Tye, Type IIB orientifolds, F-theory, type I strings on orbifolds
and type I — heterotic duality, Nucl. Phys. B 533 (1998) 25, hep-th/9804092.
[41] G. Aldazabal, A. Font, L.E. Ibanez, G. Violero, D = 4, N = 1 type IIB orientifolds, Nucl. Phys.
B 536 (1998) 29, hep-th/9804026.
[42] C. Bachas, A way to break supersymmetry, hep-th/9503030.
[43] M. Bianchi, Ya.S. Stanev, Open strings on the Neveu–Schwarz penta-brane, Nucl. Phys. B 523
(1998) 193, hep-th/9711069.
[44] G. Aldazabal, A.M. Uranga, Tachyon-free non-supersymmetric type IIB orientifolds via brane–
antibrane systems, JHEP 9910 (1999) 024, hep-th/9908072.
[45] J. Polchinski, Y. Cai, Consistency of open superstring theories, Nucl. Phys. B 296 (1988) 91.
[46] J. Lauer, J. Mas, H.P. Nilles, Duality and the role of nonperturbative effects on the world-sheet,
Phys. Lett. B 226 (1989) 251.
[47] E.J. Chun, J. Lauer, J. Mas, H.P. Nilles, Duality and Landau–Ginzburg models, Phys. Lett.
B 233 (1989) 141.
[48] M.R. Gaberdiel, Lectures on non-BPS Dirichlet branes, hep-th/0005029.
[49] W. Fischler, L. Susskind, Dilaton tadpoles, string condensates and scale invariance, Phys. Lett.
B 171 (1986) 383.
[50] W. Fischler, L. Susskind, Dilaton tadpoles, string condensates and scale invariance. 2, Phys.
Lett. B 173 (1986) 262.
[51] G. Aldazabal, D. Badagnani, L.E. Ibanez, A.M. Uranga, Tadpole versus anomaly cancellation
in D = 4, 6 compact IIB orientifolds, JHEP 9906 (1999) 031, hep-th/9904071.
[52] M. Bianchi, J.F. Morales, Tadpoles and anomalies, JHEP 0003 (2000) 030, hep-th/0002149.
[53] A. Sen, Stable non-BPS bound states of BPS D-branes, JHEP 9808 (1998) 010, hep-th/9805019.
[54] O. Bergman, M.R. Gaberdiel, Stable non-BPS D-particles, Phys. Lett. B 441 (1998) 133, hep-
th/9806155.
[55] A. Sen, Tachyon condensation on the brane–antibrane system, JHEP 9808 (1998) 012, hep-
th/9805170.
[56] M.R. Gaberdiel, B. Stefański, Dirichlet branes on orbifolds, Nucl. Phys. B 578 (2000) 58, hep-
th/9910109.
[57] A. Sen, Non-BPS states and branes in string theory, hep-th/9904207.
[58] A. Lerda, R. Russo, Stable non-BPS states in string theory: a pedagogical review, hep-
th/9905006.
[59] R. Blumenhagen, L. Görlich, B. Körs, D. Lüst, Asymmetric orbifolds, noncommutative
geometry and type I vacua, hep-th/0003024.
[60] R. Blumenhagen, L. Görlich, B. Körs, Supersymmetric orientifolds in 6D with D-branes at
angles, Nucl. Phys. B 569 (2000) 209, hep-th/9908130.
[61] R. Blumenhagen, L. Görlich, B. Körs, A new class of supersymmetric orientifolds with
D-branes at angles, hep-th/0002146.
[62] C. Angelantonj, R. Blumenhagen, Discrete deformations in type I vacua, Phys. Lett. B 473
(2000) 86, hep-th/9911190.
[63] R. Blumenhagen, L. Görlich, B. Körs, Supersymmetric 4D orientifolds of type IIA with D6-
branes at angles, JHEP 0001 (2000) 040, hep-th/9912204.
[64] G. Pradisi, Type I vacua from diagonal Z3 -orbifolds, hep-th/9912218.
[65] M.R. Gaberdiel, A. Sen, Non-supersymmetric D-brane configurations with Bose–Fermi
degenerate open string spectrum, JHEP 9911 (1999) 008, hep-th/9908060.
[66] O. Bergman, M.R. Gaberdiel, A non-supersymmetric open string theory and S-duality, Nucl.
Phys. B 499 (1997) 183, hep-th/9701137.
Physics potential at a neutrino factory: can we

benefit from more than just detecting muons?
A. Bueno, M. Campanelli, A. Rubbia ∗
Institut für Teilchenphysik, ETHZ, CH-8093 Zürich, Switzerland
Abstract
In order to fully address the oscillation processes at a neutrino factory, a detector should
be capable of identifying and measuring all three charged lepton flavors produced in charged
current interactions and of measuring their charges to discriminate the incoming neutrino helicity.
This is an experimentally challenging task, given the required detector mass for long-baseline
experiments. We address the benefit of a high-granularity, excellent-calorimetry non-magnetized
target-detector, which provides a background-free identification of electron neutrino charged current
and a kinematical selection of tau neutrino charged current interactions. We assume that charge
discrimination is only available for muons reaching an external magnetized-Fe spectrometer. This
allows the clean classification of events into electron, right-sign muon, wrong-sign muon and no-
lepton categories. In addition, high granularity permits a clean detection of quasi-elastic events,
which by detecting the final state proton, provide a selection of the neutrino electron helicity without
the need of an electron charge measurement. From quantitative analyses of neutrino oscillation
scenarios, we conclude that in many cases the discovery sensitivities and the measurements of
the oscillation parameters are dominated by the ability to measure the muon charge. However, we
identify cases where identification of electron and tau samples contributes significantly.  2000
PACS: 16.60.Pq; 13.15.+g

Keywords: Neutrinos; Oscillations; Matter effects; Neutrino factory; CP violation
1. Introduction
The firmly established disappearance of muon neutrinos of cosmic ray origin [1,2]
strongly points toward the existence of neutrino oscillations [3–7]. The first generation
long baseline (LBL) experiments — K2K [8,9], MINOS [10,11], OPERA [12] and
E-mail addresses: antonio.bueno@cern.ch (A. Bueno), mario.campanelli@cern.ch (M. Campanelli),
andre.rubbia@cern.ch (A. Rubbia).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 3 9 - 3
578 A. Bueno et al. / Nuclear Physics B 589 (2000) 577–608
ICANOE [13–15] — will give a conclusive and unambiguous signature of the oscillation
mechanism and will provide the first precise measurements of the parameters governing
the oscillation mechanism. MiniBOONE [16] and the LBL programs will test the LSND
signal [17–20].
A neutrino “factory” [21,22] is based on the decay of muons circulating in a storage
ring. Neutrino factories raised the interest of the physics community, since they appear
natural follow-ups to the current experimental LBL program and could open the way to
future muon colliders.
As many studies have shown [23–27], the physics potential of such facilities is indeed
very vast. An entry-level neutrino factory could test the LSND signal in a background free
environment [24]. More importantly, a neutrino factory source would be of sufficiently
high intensity to perform very long baseline (transcontinental) experiments. It could also
bring the neutrino sector into the realm of precision measurements.
The neutrino oscillation phenomenology may be complicated and involve a combination
of transitions to νe , νµ and ντ . It is quite evident that future neutrino factories will
provide ideal conditions for the neutrino oscillation physics [23,25–34]. The neutrino
flavor phenomenology could be completely explored: a precise measurement of the mass
difference and mixing matrix elements is achievable, a test of the unitarity of the mixing
matrix can be performed, a direct detection of Earth matter effects is feasible [28–30] and
CP-violation effects could be studied on the leptonic sector [31–34].
The combination of data from atmospheric neutrino and first generation LBL exper-
iments will provide some preliminary information on the possible sub-leading electron
mixing [13–15]. A neutrino factory can largely improve the sensitivity on this mixing an-
gle.
Neutrino sources from muon decays provide clear advantages over neutrino beams
from pion decays. The exact neutrino helicity composition is a fundamental tool to study
neutrino oscillations. It can be easily selected, since µ+ → e+ νe ν̄µ and µ− → e− ν̄e νµ can
be separately obtained.
At a neutrino factory, one could independently study the following flavor transitions:
µ− → e− ν̄e νµ
→ νe → e− appearance (1)
→ νµ disappearance, same sign muons (2)
−
→ ντ → τ appearance, high energy nu’s (3)
→ ν̄e disappearance (4)
+
→ ν̄µ → µ appearance, wrong sign muons (5)
+
→ ν̄τ → τ appearance, high energy nu’s (6)
plus 6 other charge conjugate processes initiated from µ+
decays.
The other main advantages over traditional pion beams are (1) the beam is free of
systematics and the composition is well known, therefore ideal for disappearance studies;
(2) the two neutrinos in the beam have opposite helicities, therefore one can envisage
oscillation appearance searches without intrinsic beam backgrounds (3) the muon energy
A. Bueno et al. / Nuclear Physics B 589 (2000) 577–608 579
is monochromatic and in principle adaptable (4) muon storage rings allow for multiple
baselines, and hence a complete exploration of the L/E domains of oscillations (5) because
very high intensity will be needed for muon colliders, very intense muon sources will
produce very intense neutrino sources, at least a factor 100 more intense than existing high
energy facilities.
While physics motivations are well understood, it is not yet clear which design of
detector would best allow to take full advantage of the neutrino factory beams.
We think that, in order to fully explore the neutrino oscillation processes, the detector
should be capable of:
1. Measuring and identifying all three lepton flavors: electron, muon and tau;
2. Measuring the sign of the lepton charge;
3. Separating between charged and neutral current interactions.
Experimentally, it is a very challenging task to build detectors with (1) mass scales of
the order of tens of kilotons required for long-baseline experiments, (2) with sufficient
granularity to cleanly identify electron and tau leptons and (3) which measure the charge
of these leptons.
Various solutions have been explored recently. One based on nuclear emulsions and
magnetized iron has been discussed in [35]. The main challenge there is to reach the
required mass. Large magnetized calorimeters have been discussed in [36]. Such high-
density detectors, while “easily” conceived as massive objects, have intrinsically very
coarse granularity and only allow the clean measurement of muons. They certainly do not
have sufficient power to adequately identify and measure electron or tau charged current
states.
In this paper, we are motivated by the recent progress made in the direction of the
design of the multikiloton ICANOE detector [13–15]: accordingly, we consider a 10 kton
(fiducial) high granularity low density liquid argon imaging target, complemented with a
high-acceptance external muon spectrometer.
Thanks to its extremely high granularity target and its excellent calorimetric properties,
this design provides the clean identification and measurement of all three neutrino flavors:
electron, muons and taus. However, only the sign of the muons reaching the muon
spectrometer can be determined. 1
The aim of this paper is to understand the potentials of a non-magnetized high-
granularity target detector which, compared to traditional high density iron calorimeters,
brings the measurement of electrons and taus.
We study the physics potentials of such a detector configuration for three-family neutrino
mixing and for three different baselines (732, 2900 and 7400 km).
In Section 2 we summarize three-family neutrino mixing framework, including the
treatment of propagation through matter.
In Section 3, we explain in details how the event distributions and rates are obtained
from a detailed simulation of neutrino interactions and detector effects.
1 The possibility of the measurement of the electron charge will be addressed in a future work.
In Section 4, we construct for given oscillation scenarios, event variable distributions for
various event classes. All events can be subdivided into the electron, the same sign muons,
the wrong sign muons and the no lepton samples.
In addition, in Section 5 we discuss the possibility to further discriminate final states
between νe and νµ origins from ντ by means of kinematical analysis of the events.
We also address in Section 6 the possibility to tag quasi-elastic events which provide
indirect neutrino helicity discrimination however at a large statistical price.
Section 7 is devoted to describing our fits of the oscillation parameters. The fits are
expected to give back the input reference oscillation parameters and are used to estimate
the precision with which we can estimate these parameters.
Section 8 presents the results for the important case in which the oscillation effects can
be approximated by one mass scale.
Since the information about the oscillation parameter is redundantly available in the
visible energy distributions of the various event classes, we address in Section 9 the
question of the consistency between the different observed oscillations processes. The
ability to treat the appearance of electron or tau neutrinos gives good over-constraints on
the mixing matrix.
Finally, in Section 10, we analyze the three-family scenario including possible CP-
violation.
2. Three-family neutrino oscillation framework
2.1. Mixing matrix parameterization
We consider neutrino oscillations in a three-family scenario: the flavor eigenstates να

(α = e, µ, τ ) are related to the mass eigenstates νi0 (i = 1, 2, 3) by the mixing matrix U
να = Uαi νi0 , (7)
and we parameterize it as:
 
c12 c13 s12 c13 s13 e−iδ
U =  −s12 c23 − c12 s13 s23 eiδ c12 c23 − s12 s13 s23 eiδ c13 s23  (8)
s12 s23 − c12 s13 c23 eiδ −c12 s23 − s12 s13 c23 eiδ c13 c23
with sij = sin θij and cij = cos θij . We confine without loss of generality the mixing angle
θ13 to values in the interval [0, π/4], and θ12 , θ23 , δ to the interval [0, π/2]. We present
results in terms of sin2 2θ13 , sin2 θ23 and sin2 θ12 , all running in the interval [0, 1]. The
reason for this choice can be for example seen in Appendix A, where we recall oscillation
probabilities in the one mass scale approximation. The oscillation probabilities for νe →
νµ and νe → ντ depend on sin2 2θ13 and on the sin and cosine of θ23 . The angle θ23 must
span the interval [0, π/2], however, the θ13 can vary within interval [0, π/4]. Note that we
consider small θ13 angles, so the cos4 θ13 dependence in νµ → ντ is very mild.
For δ = 0 (i.e., U is real), the general expression for the three-family neutrino oscillation
probability is:
X
P (να → νβ ; E, L) = P (ν̄α → ν̄β ; E, L) = δαβ − 4 Jαβj k sin2 (∆j k ), (9)
j >k
where in natural units ∆j k ≡ 1m2j k L/(4E) = (m2j − m2k )L/(4E), E is the neutrino energy,
L is the neutrino path-length, and the Jarlskog term is Jαβj k = Uβj Uβk Uαj Uαk .
We naturally assign the mass difference squared 1m212 to explain the solar neutrino
deficit and the mass difference squared 1m232 to describe the atmospheric neutrino
anomaly. We will take as reference value 1m232 = 3.5 × 10−3 eV2 . To cover possible
ranges of this value, we will also consider two other values 1m232 = 5 × 10−3 eV2 and
1m232 = 7 × 10−3 eV2 . We will always assume maximal (2–3)-mixing sin2 2θ23 = 1. In
case of the solar neutrino deficit solution, the values for 1m212 and mixing sin2 2θ12 are not
uniquely defined by experiments. We will limit ourself to the LMA–MSW solution with
parameters 1m212 = 1 × 10−4 eV2 and hypothesize a maximal (1–2)-mixing sin2 θ12 = 0.5.
In this paper, we do not consider the LSND result which would force us to include more
states with new parameters beyond three-family mixing.
This choice implies that we will always work in a situation where |1m221| < |1m232 | ≈
|1m231|. In first approximation, the oscillation phenomena governed by the two mass
differences decouple and the effects produced by 1m212 are small at high energy for the
considered baselines. In the first part of this paper, we will neglect 1m212 effects and
work in the so-called “one mass scale approximation” [37]. In a second phase, we will
be concerned with CP-violation effects and will have to include 1m212 .
For simplicity, we will take m1 < m2 < m3 which implies 1m232 > 0. We recall that
neutrino oscillations through matter can be used to distinguish 1m232 > 0 from 1m232 < 0.
In vacuum, we will express the oscillation probability as a function of the seven
following parameters: (a) the three mixing angles θ12 , θ13 , θ23 ; (b) the two mass differences
squared 1m212 , 1m232 ; (c) the baseline L; (d) the neutrino energy E.
2.2. Matter effects
Since we will consider very long distances between neutrino production and detection,
this will only be possible in practice for neutrinos traveling inside the Earth. In this case,
the neutrino oscillation probabilities will be modified by an additional diagram due to
the interaction of electron neutrino with the electrons in the matter [38–40]. One can
maintain the neutrino oscillation formalism derived in vacuum but define effective masses
and mixing angles valid in matter. For example, the effective masses will result from the
diagonalization of the Hamiltonian:
 2   
m1 0 0 D 0 0
U  0 m22 0  U † +  0 0 0  , (10)
0 0 m23 0 0 0
where

√ ρ E
D = 2 2 GF ne E = 7.56 × 10−5 eV2 . (11)
g cm−3 GeV
Here, ne is the electron density and ρ the matter density. For antineutrinos, we must replace
D by −D. For ρ(g cm−3 )E(GeV) ≈ 40, the effective mass parameter D is of the order of
the mass splitting that is derived from the atmospheric neutrino anomaly. We then expect
matter effects to be important.
Within the two-family mixing scheme, the modification of the flavor transition in matter
is taken into account by the mixing angle in matter θm , which is:
sin2 2θ
sin2 2θm (D) = 2 . (12)
sin2 2θ + D
1m2
− cos 2θ
For neutrinos, a resonance condition will be met when D ' 1m2 cos 2θ and the oscillation
amplitude will reach a maximum. The resonant neutrino energy E res is
1.32 × 104 cos 2θ 1m2 (eV2 )
E res (GeV) ≈ . (13)
ρ(g/cm3 )
Rather than two-family mixing, we have adopted throughout this study three-family
framework. In this context, we use the analytic expressions for the matter mixing angles
and mass eigenvalues calculated in [43,44] (see Appendix A).
The mass eigenvalues in matter M1 , M2 and M3 are:
√
A 1p 2 3p 2 p
M1 = m1 + −
2 2
A − 3B S − A − 3B 1 − S 2 , (14)
3 3 3
√
A 1p 2 3p 2 p
M22 = m21 + − A − 3B S + A − 3B 1 − S 2 , (15)
3 3 3
A 2p 2
M32 = m21 + + A − 3B S, (16)
3 3
where A, B and S are given in Appendix A. For the mixing angles in matter the analytical
expressions read:
−(M24 − αM22 + β)1M31
2
m
sin2 θ12 = 2 (M 4 − αM 2 + β) − 1M 2 (M 4 − αM 2 + β)
, (17)
1M32 1 1 31 2 2
M34 − αM32 + β
m
sin2 θ13 = 2 1M 2
, (18)
1M31 32
2 + F 2 c 2 + 2GF c s c
G2 s23 23 23 23 δ
m
sin2 θ23 = , (19)
G +F
2 2
where α, β, G and F are found in Appendix A.

To illustrate matter effects in three-neutrino mixing framework, we show in Fig. 1 the
values of the mixing angles in matter, plotted as a function of D/1m232 , or equivalently of
ρ × E. The parameter values in vacuum correspond to our reference values for atmospheric
and LMA–MSW solar experiments. The resonant behavior of θ13 m is clearly visible. It gives
maximum oscillation at a neutrino energy of about 12 GeV for a density of 3.7 g cm−3 .
m
There is a similar resonant behavior for θ12 but it occurs at low energy since it is driven by
m m
1m21 . For D > 1m32 , the angles sin 2θ12 and sin2 θ23
2 2 2
tend to rise slightly, because the
2
non-vanishing 1m21 splitting removes the degeneracy between muon and tau flavors.
√
Fig. 1. Evolution of mixing angles in matter as a function of D(= 2 2 GF Ne Eν )/1m232 . The
reference vacuum parameters are given in the figure.
Our results have been computed assuming a constant density along the whole neutrino
path, and equal to the mean density, obtained integrating over the earth profile [41]. This
approximation yields, as shown in, e.g., Ref. [45], similar results to those obtained by
numerical integration using the actual Earth’s density profile.
The oscillation probability through matter will be a function of eight parameters: (a)
the vacuum three mixing angles θ12 ,θ13 ,θ23 ; (b) the vacuum two mass differences squared
1m212 , 1m223 ; (c) the average earth density ρ; (d) the baseline L; (e) the neutrino energy E.
3. Choice of baseline and event rates
The exact parameters of a neutrino factory are not yet completely fixed and realistic sce-
narios are in the process to be defined [22]. However, based on [22], we can assume that
the muons in the storage ring have an energy Eµ = 30 GeV and that after one year of op-
eration, the factory should deliver about 1020 “useful” muons decays of both polarities in
the straight section pointing towards the far detector location. We base our ultimate reach
on 1021 “useful” muons decays. Even an integrated intensity of 1022 might be eventually
reachable.
We compute the fluxes assuming unpolarized muons and disregarding muon beams
divergences within the storage ring. We integrate the expected event rates using a neutrino–
Table 1
Expected events rates for a 10 kton (fiducial) detector in case no oscillations occur for 1020 muon
decays. Ntot is the total number of events and Nqe is the number of quasi-elastic events
Event rates for various baselines
L = 732 km L = 2900 km L = 7400 km

Ntot Nqe Ntot Nqe Ntot Nqe
νµ CC 226000 9040 14400 576 2270 90

µ− νµ NC 67300 – 4120 – 680 –
20
10 decays ν̄e CC 87100 3480 5530 220 875 35
ν̄e NC 30200 – 1990 – 300 –
ν̄µ CC 101000 4040 6380 255 1000 40

µ+ ν̄µ NC 35300 – 2240 – 350 –
20
10 decays νe CC 197000 7880 12900 516 1980 80
νe NC 57900 – 3670 – 580 –
nucleon Monte-Carlo generator [49]. The total charged current (CC) cross section is
technically subdivided into three parts: the exclusive quasi-elastic scattering channel σQE
and the inelastic cross section σinelasic which includes all other processes except charm
production which is included separately.
Table 1 summarizes the expected rates for the 10 kton fiducial mass and 1020 muon
decays (expected 1 year of operation). Ntot is the total number of events and Nqe is the
number of quasi-elastic events.
Even though our study is site non-specific, the chosen baselines could correspond to the
distances between the Laboratori Nazionale del Gran Sasso (LNGS) and neutrino factories
at (1) CERN (L = 732 km, hρEarth i = 2.8 g/cm3 ), (2) Canary Islands (L = 2900 km,
hρEarth i = 3.2 g/cm3 ) and (3) Fermilab (L = 7400 km, hρEarth i = 3.7 g/cm3 ).
4. The four main classes of events
Muon identification, charge and momentum measurement provide discrimination

between νµ and ν̄µ charged current (CC) events. Good νe CC versus ν NC discrimination
relies on the fine granularity of the target. Finally, the identification of ντ CC events
requires a precise measurement of all final state particles.
It is natural to classify the events in four classes [23]. We illustrate them for the case of
µ− stored in the ring.
1. Right sign muons (rsµ): the leading muon has the same charge as those circulating
inside the ring. Their origin is from
(a) non-oscillated νµ CC;
(b) νµ → ντ CC, τ − → µ− decays;
(c) hadron decays in neutral currents.
2. Wrong sign muons (wsµ): the leading muon has opposite charge to those circulating
inside the ring. Opposite-sign leading muons can only be produced by neutrino
oscillations, since there is no component in the beam that could account for them.
(a) ν̄e → ν̄µ oscillations;
(b) ν̄e → ν̄τ oscillations, τ + → µ+ decays;
(c) hadron decays in neutral currents.
3. Electrons (e): events with a prompt electron and no primary muon identified. Events
with leading electron or positron are produced by the charged-current interactions of
the following neutrinos:
(a) non-oscillated ν̄e neutrinos;
(b) νµ → νe oscillations;
(c) ν̄e → ν̄τ or νµ → ντ oscillations with τ → e decays.
4. No lepton (0`): events corresponding to NC interactions or ντ CC events followed
by a hadronic decay of the tau lepton. Events with no leading electrons or muons will
be used to study the νµ → ντ oscillations. These events can be produced in
(a) neutral current processes;
(b) ν̄e → ν̄τ or νµ → ντ oscillations with τ → hadrons decays.
The last two classes can only be cleanly studied in a fine granularity detector.
The most effective way to fit the oscillation parameters is to study the visible energy
distribution of the four classes of events defined above, since assuming the unoscillated
spectra are known, they contain direct information on the oscillation probabilities.
Of course, for electron or muon charged current events, the visible energy reconstructs
the incoming neutrino energy. In the case of neutral currents or the charged current of
tau neutrinos, the visible energy is less than the neutrino energy because of undetected
neutrinos in the final state. The information is in this case degraded but can still be used.
Our analyses are performed on samples of fully generated Monte-Carlo [49] events,
which include proper kinematics of the events, full hadronization of the recoiling jet and
proper exclusive polarized tau decays when relevant. 2 Nuclear effects, which are taken
into account by the FLUKA model [50], are included as they are important for a proper
estimation of the tau kinematical identification.
The detector response is included in our analyses using a fast simulation which
parameterizes the momentum and angular resolution of the√emerging particles, using
√ the following values: electromagnetic shower 3%/ E ⊕ 1%, hadronic shower
essentially
≈ 20%/ E ⊕ 5%, and magnetic muon momentum measurement 20%.
Hadron decay background can be quite large in a low density target and could be
quite dangerous. Fortunately, it can be easily suppressed by a cut on the muon candidate
momentum, Pµ > 2 GeV, which reduces it to a tolerable level. Fig. 2 illustrates the relative
background expected for νµ and ν̄µ NC and CC processes as a function of the muon
momentum. After the cut, the expected contamination for νµ CC events is at the level
of 10−5 . Real charged current events maintain an efficiency above 95%.
2 Our simulation has been bench-marked on the comparison of the kinematic features of the lepton and hadronic
jet of real neutrino data accumulated in the NOMAD experiment (see, e.g., [13–15]).
Fig. 2. Fractional background due to the decay of charged mesons as a function of the measured
muon momentum.
Fig. 3. Visible energy spectrum for electron Fig. 4. Same as Fig. 3 for right-sign muon
events: νe CC (dashed line), ντ and ν̄τ (dotted sample: νµ CC (dashed line), ντ and ν̄τ
line) and ν̄e CC (dot-dashed). The solid his- (dotted line) and meson decay background
togram shows the sum of all contributions. (dot-dashed). The solid histogram shows the
sum of all contributions.
The visible energy is computed as the modulus of the vector sum of the momenta of each
visible particle in the event. Figs. 3, 4, 5 and 6 show the reconstructed visible energy at the
baseline L = 7400 km normalized to 1020 µ’s for each event class for a specific oscillation
scenario with 1m232 = 3.5 × 10−3 eV2 , sin2 θ23 = 0.5 and sin2 2θ13 = 0.05. The different
contributions including backgrounds for each event class have been evidenced in the plots.
For example, in Fig. 4, the different processes that contribute to the right-sign muon class
are unoscillated muons, taus and background events.
Fig. 5. Same as Fig. 4 for wrong sign muon Fig. 6. Same as Fig. 4 for the no-lepton sample:
sample: νµ CC (dashed line), ντ and ν̄τ ν NC (dashed line), ντ and ν̄τ (dotted line).
(dotted line) and meson decay background The solid histogram shows the sum of all
(dot-dashed). The solid histogram shows the contributions.
sum of all contributions.
5. Further classification based on kinematical analysis
An efficient identification of ντ induced charged current events requires a precise

measurement of all final state particles. Excellent calorimetry allows to take full advantage
of the special kinematic features of ντ events. We independently search for the leptonic
and hadronic tau decay modes.
For the τ → lνν decay mode, the main background comes from νl CC. To enhance the
separation between τ and background events, we demand the event missing PT to be larger
than 0.6 GeV and the transverse momentum of the lepton candidate, PTl , to be smaller than
0.5 GeV. This set of cuts is referred to as “loose cuts” and it will be used to perform a check
on appearance/disappearance consistency (see Subsection 9). In Table 2, we show that the
overall τ efficiency for “loose cuts” is 40% for a CC background level of ∼ 5 × 10−3 .
Fig. 7 shows the energy spectra for the four event classes after application of these cuts,
for ντ CC and other types of events. No energy cut has been applied, but the fact of using
the energy spectra in the fit also exploits the difference in energy spectra of ντ CC events.
A set of “tight cuts” is also applied, aiming at having one expected background event at
the farthest location in case 1020 “useful” muons decays are delivered. As we can see from
Table 2, the overall tau efficiency in this case amounts up to 20%.
For hadronic decays the most important source of background corresponds to NC events.
If we demand a PTmiss smaller than 1 GeV and a transverse momentum of the hadron
candidate with respect to the total event momentum, QT , larger than 0.5 GeV (“loose
cuts”) only 2% of the initial background survives for an overall tau efficiency of 30%. If
we require only one NC background event survivor at L = 7400 km (“tight cuts”), the
signal efficiency drops to 6% due to the stringent QT requirement imposed.
Table 2
ντ appearance search for the leptonic and hadronic decay modes of the tau lepton. Overall signal
efficiencies and fractional remaining backgrounds are quoted. The case labelled as “Tight cuts”
correspond to the situation where one event background is expected, for 1020 muon decays, at the
farthest location (L = 7400 km)
νµ → ντ appearance search
Cuts τ →l CC background Cuts τ →h NC background
Initial 100% 100% Initial 100% 100%
Loose cuts
PTl < 0.5 GeV 50% 14% PTmiss < 1 GeV 72% 40%
PTmiss > 0.6 GeV 40% 0.5% QT > 0.5 GeV 30% 2%
Tight cuts
PTl < 0.5 GeV 50% 14% PTmiss < 1 GeV 72% 40%
PTmiss > 1 GeV 20% 0.08% QT > 1 GeV 6% 0.07%
6. Quasi-elastic final states
The quasi-elastic process, while rare, is a clean process that allows to separate neutrino
from antineutrino events, in principle for all neutrino flavors, since ν` + n → `− + p and
ν̄` + p → `+ + n. The recoil proton is easily identifiable within the high-granularity target.
This channel is particularly interesting to study oscillation in the electron channel.
Starting from negative muons circulating in the storage ring, we look for exclusive
electron–proton final states. These provide “background-free” oscillation signals:
µ− → e− ν̄e νµ
→ νe + n → e− + p (20)
− −
→ ντ + n → τ + p → e νν + p (21)
since ν̄e + p → e+ + n.
The selection of quasi-elastic events is the only way to identify the helicity of neutrino
electrons in absence of measurement of the electron charge.
7. Oscillation parameters fitting
Given the adopted parameterization of the mixing matrix, we have a priori a total of 7
free parameters, which can be represented by the vector:

PE = 1m221, 1m232 , sin2 θ12 , sin2 2θ13, sin2 θ23 , δ, ρ . (22)
The values of the parameters governing the oscillations are extracted from a global fit
of the visible energy distributions obtained for each event class. The fit is performed with
Fig. 7. Visible energy spectrum for the four event classes after application of loose kinematic cuts
(see text). The presence of τ events in the third class (wrong-sign muons) is an indication of the
process νe → ντ .
the MINUIT [42] package, and is expected to get back the same values of the parameters,
starting from the reference distributions.
At each iteration, a different set of parameters is probed, and with the same procedure
used to get the reference histograms.
For a given polarity λ of the muons in the storage ring, we compute χ 2 ’s of the difference
between the binned oscillated spectra, which will be function of the parameters, and the
reference histograms. We define a χ 2 for each of the four classes of events, i.e., the
electrons (e), the right-sign muon (rsµ), the wrong sign muons (wsµ) and the no lepton
class (0`):
2
χλ,all = χλ,e
2
+ χλ,rsµ
2
+ χλ,wsµ
2
+ χλ,0`
2
, (23)
where

X Nic (PE ) − Nic (PEref ) 2
2
χλ,c E
P = . (24)
2
σi,c
i
The sum runs over 25 equally spaced energy bins (Ei , i = 1, . . . , 25). Here, Nic (PEref ) is
the expectation for the reference values and Nic (PE ) is the “data” obtained for a given set
of the oscillation parameters PE . σi,c contains both statistical and systematic contributions
to the total error. In order to improve the statistical treatment of bins with low statistics, a
bin that possesses less than 40 events is assumed to be Poisson distributed and therefore
its contribution is computed as 2(Nic (PE ) − Nic (PEref )) + 2Ni ln(Nic (PE )/Nic (PEref )) (see
Ref. [51]).
The systematic error takes into account the uncertainties in the knowledge of the beam,
neutrino cross sections and selection efficiencies and we assume it amounts up to 2%
uncorrelated from bin to bin.
We assume that a neutrino factory will operate with alternate runs of opposite muon
polarities, therefore eight energy distributions can be fitted simultaneously:
2
χall = χ+,all
2
+ χ−,all
2
. (25)
The values of the fitted parameters PE are obtained minimizing the χ 2 (PE ).
It is in practice not always possible to fit all the free parameters, since for some
parameter-space regions, the oscillation effects at the chosen baselines and energies can
be negligible. In particular, this can be the case for 1m221 and sin2 θ12 which drive the solar
oscillations. For values 1m221 10−4 eV2 , we are insensitive to the “solar” sector.
We therefore adopted successive fitting procedures with an increasing number of free
parameters.
At first, we can simplify the three-family oscillation picture if the oscillations produced
by 1m221 can be neglected at the considered baselines and energies. In this case, only the
mass difference squared 1m232 and the two mixing angles θ13 and θ23 are relevant. The
oscillation probabilities are given in Appendix A. For example, for νe → νµ oscillations,
it is:
2 m 2
P (νe → νµ , E, L) = sin2 2θ13m
sin θ23 ∆32 , (26)
where ∆232 = sin2 ((M32 − M22 )L/(4E)). The fit has in this case 4 free parameters, which
can be represented by the vector:

PE1ms = 1m232, sin2 2θ13 , sin2 θ23, ρ . (27)
The δ phase has disappeared, since within this approximation it becomes unphysical. The
results are presented in Section 8, assuming a 1m232 parameter varying in the range of
values favoured by current atmospheric data (1m232 = 3.5, 5, 7 × 10−3 eV2 ), a maximal
(2–3)-mixing sin2 θ23 = 0.5 and a θ13 value compatible with CHOOZ results [46,47] and
recent fits to data [48] (sin2 2θ13 = 0.05).
We return to the general three-family scenario in Section 10 where we consider
sensitivity to the CP violation.
In order to compute the precision of the determination of the parameters, we consider
two methods: (1) One-dimensional “scan” of a given parameter; the other variables are left
2 + 1,
free and minimized at each step; the 1, 2, 3 sigmas are given by, respectively, χmin
+4 and +9. (2) Two-dimensional “scan” of a two-parameter plane; the other variables are
left free and minimized at each point in the plane; the 68%, 90%, 99% C.L. are given by,
2 + 2.4, +4.6 and +9.2.
respectively, χmin
8. Results for one mass scale approximation
8.1. Case of two-family mixing: θ13 = 0
The detection of the dip in the energy distribution of the right-sign muon sample
dominates the precision on the measurement of the mixing angle θ23 and the mass
difference 1m232 . This is true provided that the beam energy and baseline are chosen
in such a way that the νµ disappearance maximum is visible in the oscillated spectrum.
Table 3 summarizes the expected accuracies. With 1020 muon decays of each polarity,
precisions of 1–2% are expected in the determination of 1m232 while for the precision on
sin2 θ23 is around 10%, in agreement with results quoted in [26,27].
We compare in Fig. 8 how the precision on the mass difference and mixing angle changes
when experimental resolutions and backgrounds are disregarded. We see that our fits are
barely affected and only for the cases where the dip is not seen, the instrumental effects and
backgrounds spoil at the level of a few per cent the accuracy on the oscillation parameters.
8.2. Case of three-family mixing: θ13 6= 0
In order to obtain the best sensitivity on the mixing angle θ13 , the search for wrong sign
muons is ideal, since it will be a direct signature for νe → νµ oscillations.
For this kind of study, since we are dealing with the smallest number of signal events,
the sensitivity does strongly depend on the ability of rejecting background. Fig. 9 shows for
L = 7400 km, the sensitivity on θ13 for two different muon normalizations (1020 and 1021
muon decays of each polarity). For each pair of values (1m232, θ13 ), the fit was performed
leaving θ23 free.
Table 3
Precision in the measurement of the mixing angle assuming two family mixing for three possible
mass differences and two very large baselines. In all the cases the precision obtained in the
measurement of 1m232 is 1%
Two family mixing

δ(sin2 θ23 ) for θ23 = 45◦ , θ13 = 0
Only right-sign muons All classes

2
χ±,rsµ 2
χ±,all
1m232 (eV2 ) L = 7400 km L = 2900 km L = 7400 km L = 2900 km
7 × 10−3 22% 8% 20% 8%

5 × 10−3 12% 12% 11% 10%
3.5 × 10−3 10% 18% 10% 16%
Fig. 8. Two-dimensional 90% C.L. contours for 1m232 and sin2 θ23 for three different 1m232 values.
Inner contours: no background and perfect muon resolution; outer contours: backgrounds and muon
resolution included.
Fig. 9. Sensitivity on θ13 .
In obtaining the previous plot, we did not on purpose apply strong background cuts,
since we believe that very high rejection powers obtained on paper may not stand the proof
of real experimental conditions, with non-Gaussian behaviours, tails of distributions, etc.
To also give the maximum of sensitivity that can be obtained in principle, we also
illustrated in Fig. 9 the effect that a background free environment would have in the
expected sensitivity. At 90% C.L., we obtain sin2 2θ13 < 10−3 –10−4 depending on the
number of muon decays. This represents two orders of magnitude improvement with
respect to quoted sensitivities at CNGS. In case we consider 1021 muon decays and
backgrounds are reduced to a negligible level, the obtained sensitivity is consistent with
the one quoted in [36].
For a background-free environment gives the best sensitivity, the amount of information
added by other event classes (i.e., the electrons) is negligible.
8.2.1. Determination of θ13

The measurement of θ13 can profit from long baselines, since matter effects will enhance
the oscillation signal. In presence of backgrounds, it is more favorable to enhance neutrinos
signal even at the cost of the suppression of the antineutrino oscillations.
Table 4 summarizes the expected precision on the measurement of the oscillation
parameters.
In this case, since for the chosen value of sin2 2θ13 = 0.05 the number of signal events
is quite large, there is not any more a strong need for a background-free environment.
Therefore, the inclusion of other event classes, like the electrons, can help to constrain the
oscillation parameters.
Table 4
Precision on the measurement of the oscillation parameters
Three-family mixing
sin2 θ23 = 0.5, sin2 2θ13 = 0.05
All classes Only muons

2
χall 2 + χ2
χrsµ wsµ
L = 2900 km L = 7400 km L = 2900 km L = 7400 km
1m232 = 3.5 × 10−3 eV2

δ(1m232 ) 1.4% 0.9% 1.4% 0.9%
δ(sin2 θ23 ) 14% 8% 16% 9%
δ(sin2 2θ13 ) 15% 10% 17% 15%
1m232 = 5 × 10−3 eV2
δ(1m232 ) 0.4% 0.8% 0.4% 0.8%
δ(sin2 θ23 ) 11% 8% 10% 12%
δ(sin2 2θ13 ) 11% 9% 14% 16%
1m232 = 7 × 10−3 eV2
δ(1m232 ) 0.4% 0.6% 0.4% 0.6%
δ(sin2 θ23 ) 7% 8% 8% 18%
δ(sin2 2θ13 ) 8% 6% 9% 20%
Table 5
Expected number of electron type events for 1021 µ− decays. The three contributions to the total
number of electron events are shown separately. Rates have been computed assuming oscillations
with 1m232 = 3.5 × 10−3 eV2 , sin2 θ23 = 0.5 and sin2 2θ13 = 0.05
νe appearance search with quasi-elastic electron class:

events for 1021 µ− decays
ν̄e CC νµ → νe CC νµ → ντ CC, τ → e
Baseline Total Elastic Total Elastic Total Elastic
L = 732 km 860000 43000 2090 84 3990 110

L = 2900 km 54300 2700 1720 70 3300 90
L = 7400 km 8300 410 960 40 1450 40
8.3. Sensitivity to θ13 with quasi-elastic events
An exclusive way of detecting the effects of a non-vanishing θ13 is through the appear-
ance of wrong sign electrons. Although, with the assumed detector configuration, there is
no ability to directly measure the charge of the leading electrons, there is the possibility of
disentangling final state electrons from positrons through the use of quasi-elastic events.
We look for electron–proton final states. For example, in the target of ICANOE, a proton
can be resolved if its kinetic energy is larger than about 100 MeV corresponding to a range
of more than 2 cm.
Table 5 shows the expected rates contributing to the electron class before any cut is
applied. The background is twofold: (a) quasi-elastic ντ CC events followed by τ → e,
(b) ν̄e CC with an extra proton from nuclear origin.
This last background can be estimated from data themselves studying the reaction
ν̄µ p → µ+ n. Demanding a back to back electron–proton event topology with a proton
kinetic energy in excess of 100 MeV, we estimate the expected background to be less than
one event for an overall signal efficiency of 50%.
The quasi-elastic channel provides, in a “background free” environment, 20 to 40 gold-
plated events depending on the selected baseline. It is a very clean channel but is limited
by statistics.
8.4. Fit of the average Earth density parameter
In matter, the amplitude of neutrino oscillation goes through a maximum for an energy
given by Eq. (13). Since θ13 is small, the MSW resonance peak is only a function of
ρ and 1m232 , This can be seen in Fig. 10. Since the mass difference is constrained by
the disappearance of right-sign muons, ρ is well-determined by the energy distribution of
wrong sign muons.
We extract the density from the fit, leaving it as a free parameter, as well as 1m232 ,
θ23 and θ13 . The precision on the determination of ρ depends on the baseline, as shown
Fig. 10. Variation of the MSW resonance peak for wrong sign muons as a function of Earth’s density.
The plot is normalized to 1021 µ+ decays.
Table 6
Precision on the determination of ρ, from a global fit where also 1m232 , θ23 and
θ13 are left as free parameters. This result has been obtained for 2 × 1021 muon
decays
Distance (km) Density (g/cm3 ) Relative error (%)
732 2.8 18
2900 3.2 10
7400 3.7 2
in Table 6, obtained considering 2 × 1021 muon decays. For the longest baseline, where
matter effect are large, a precision as good as 2% can be obtained.
The influence of ρ in our fits is addressed in Fig. 11. We see that for L = 7400 km and
three different muon normalizations, the fact that ρ is either considered as a free parameter
or fixed during the fit does not influence the accuracy in the determination of the mixing
angles.
9. Over-constraining the oscillation parameters
The information about the oscillation parameter is redundantly available in the visible
energy distributions of the various event classes. This allows us to address the question of
the consistency between the different observed oscillations processes.
Fig. 11. 68% C.L. two-dimensional contours for sin2 2θ13 and sin2 θ23 : influence of ρ in the
determination of the mixing angles for three different muon normalizations and L = 7400 km. In
the upper plot ρ is fixed during the fit, while in the lower one is taken as a free parameter.
Given the high statistical accuracy of the measurement, the consistency can be tested
with good accuracy. In the three active neutrino mixing scheme, this implies that
P
y=e,µ,τ P (νx → νy ) should be equal to one for x = e, µ, τ and the same holds for
antineutrinos.
Other models can predict different values (i.e., oscillations to sterile neutrinos exist, the
sum would be smaller than one).
Let us concentrate on the oscillations into τ neutrinos. In the case of negative muons in
the ring, they can be originated from νµ → ντ or ν̄e → ν̄τ oscillations. The latter case is
particularly interesting, since coupling a neutrino factory with a detector with τ identifi-
cation capabilities is probably the only way to identify and measure such a process. The
ν̄e → ν̄τ can be revealed experimentally from the presence of τ candidates in the wrong-
sign muon sample, due to the opposite helicity of electron and muon neutrinos in the beam.
To have a quantitative estimation of the consistency check of the various oscillation
modes to τ neutrino, we assign two global normalization factors, α and β to the oscillation
probabilities P (νµ → ντ ) and P (νe → ντ ). If no new phenomena occur, these parameters
should be exactly one. From a global fit to the visible energy distributions it is possible to
extract the values of these parameters, and the precision obtainable on their measurement.
To select τ events from the background, still retaining a high efficiency on the signal,
we apply a set of loose kinematic cuts (see Table 2).
Background levels of the order of one event can be reached applying tighter cuts, but in
these cases the statistics is too small and the results obtained are slightly worse.
Since the τ lepton decays into muons, electrons and hadrons, we expect that a fit to
all event classes would result in a remarkable improvement on the precision for the α
Fig. 12. Precision on the determination of the α parameter when the fit is performed using: all event
classes (solid line), right-sign muons + electrons (dashed line) and right-sign muons only (dotted
line).
parameter. Fig. 12 shows how α determination improves as the different event classes are
included in the fit.
Table 7 shows the expected precisions in the determination of α and β. We observe that
for 1020 decays, α is better determined (accuracy around 5%) for L = 2900 km, however,
for 1021 muons, the accuracy is about 1% regardless of the baseline and the mass difference
and therefore νµ oscillations into a sterile neutrino can be largely ruled out.
The accuracy on β, and therefore the first experimental evidence for νe → ντ is much
worst, given the smaller statistics available, since this oscillation probability is smaller than
the corresponding νµ → ντ by a factor sin2 2θ13 , taken to be in this case 0.05. We observe
that a somewhat better determination exists at L = 7400 km since this oscillation mode
is largely influenced by matter effects given its dependence on θ13 . We conclude that a
precision at the level of a few per cent on the observation of νe → ντ oscillations would
require O(1022) useful muon decays.
10. Results for the general three-family scenario and CP violation
Let us consider now a more complex scenario. In this case, the value of the mass
difference 1m212 is not any more negligible, and is actually assumed to be 10−4 eV2 ,
Table 7
Precision on the determination of the parameters α and β that quantify, respectively, the amount of
νµ → ντ and νe → ντ present in the data. In the three neutrino framework, the reference values are
α, β = 1
Appearance/disappearance test
Baseline 1m232 (×10−3 eV2 ) 1020 µ± 1021 µ± 1022 µ±
Precision on α ⇒ α × P (νµ → ντ )
7400 km 3.5 5.5% 2% 0.6%
5 6% 2% 0.6%
7 11% 3% 1%
2900 km 3.5 4% 2% 0.6%
5 3% 1% 0.4%
7 2.5% 1% 0.4%
Precision on β ⇒ β × P (νe → ντ )
7400 km 3.5 60% 20% 7%
5 35% 10% 5%
7 25% 7% 2%
2900 km 3.5 75% 25% 9%
5 25% 15% 5%
7 30% 10% 4%
one of the highest values compatible with the large mixing angle MSW solution for solar
neutrinos [41].
The oscillation does not depend any more on only three parameters, but all four
independent angles of the mixing matrix and the two mass differences become important.
In the most general case, the phase δ can be different from zero, producing a complex
mixing matrix, and thus generating CP violation.
We recall that for neutrinos propagating in vacuum, the oscillation probability after a
distance L for neutrinos can be expressed as:
P (να → νβ ; E, L) = PCP even (α, β; E, L) + PCP odd (α, β; E, L) (28)
and for antineutrinos:
P (ν̄α → ν̄β ; E, L) = PCP even (α, β; E, L) − PCP odd (α, β; E, L). (29)
As an illustration, the probability for νµ to νe conversion in matter, assuming three
family mixing and CP violation, is given by:

PCP odd (µ, e) = cos θ13
m
sin δ m sin 2θ12
m m
sin 2θ13 m
sin 2θ23
2 L 2 L 2 L
1M12 1M13 1M23
× sin sin sin , (30)
4E 4E 4E
where the CP phase in matter is:

−iδm (G2 e−iδ − F 2 eiδ )s23 c23 + GF c23
2 − s2
e =q
23
. (31)
2 + F 2 c 2 + 2GF c s c
23 23 δ G c23 + F s23 − 2GF c23 s23 cδ
G2 s23 2 2 2 2
23
Here, G and F are given in Appendix A.

In case of degenerate mass differences, |1m221| ≈ |1m232|, the δ phase is significantly
modified from its original value in vacuum for δ > π/4 radians, while in the case |1m232| >
|1m221| the CP phase is almost unaffected by the presence of matter.
In addition, neutrino propagation in a dense medium makes more difficult to experimen-
tally extract a genuine CP violation signal, since the asymmetrical behavior of matter, with
respect to neutrinos and antineutrinos, induces fake CP violation effects.
10.1. CP statistical significance
To evaluate the sensitivity to this kind of measurement as a function of the oscillation

parameters and of the selected baseline, we compare the case where CP violation is
maximal (δ = π/2) and the case of no CP violation (δ = 0). Fig. 13 shows the sensitivity
N(δ = π/2, Ei ) − N(δ = 0, Ei )
SCP (Ei ) ≡ √ , (32)
N(δ = 0, Ei )
i.e., the difference between the two extreme cases, divided by the statistical error (being N
the number of events in each energy bin, for a given value of δ). We can see that for the
baseline of 732 km, this quantity is positive for almost the full energy range, so there is no
real shape variation in the spectrum, and the CP effect is similar to what would be obtained
with a change in the angle θ13 . On the other hand, for larger baselines this curve crosses
the zero, and the CP violation produces a visible deformation of the energy spectrum.
10.2. Fitting of the δ parameter
To evaluate the precision reachable on the measurement of these three oscillations

parameters, we perform fits assuming the following reference values:
1m232 = 3.5, 5, 7 × 10−3 eV2 ,
1m212 = 1 × 10−4 eV2 ,
sin2 θ23 = 0.5, θ23 = 45◦,
sin2 2θ13 = 0.05, θ13 = 6.5◦ ,
sin2 θ12 = 0.5,
δ > 0. (33)
Fig. 14 shows for the baseline L = 2900 km, the expected visible energy spectra for
electron and wrong sing muon class for the two cases δ = 0 and δ = π/2 for a 1m232 =
3.5 × 10−3 eV2 , and their difference in terms of number of events. As expected, in absolute
values, the CP violation affects the two classes in a similar way, but the effect is more
visible and significant for the wrong-sign muons, since the total number of events is
smaller.
νe → νµ
Fig. 13. Statistical significance of CP violation effects for three possible 1m232 values mass at the
three considered baselines. N(δ = π/2) is the number of expected events assuming maximal CP
violation in vacuum and N(δ = 0) corresponds to the number of expected events in absence of CP
violation, for 2 × 1021 muons.
Fig. 14 is normalized to 1021 µ+ decays, since this is the minimum amount of decays
required to produce a statistically significant observation of CP violation.
The sensitivity to fitting δ depends on the actual value of 1m232 . In what follows, we
restrict the discussion to 1m232 = 3.5 × 10−3 eV2 . For the largest baseline L = 7400
km, CP violation effects are less detectable even in case δ = π/2, since matter effects
increase the number of wrong sing muons and the statistics is reduced due to the distance.
At L = 732 km, the effect of a non-vanishing δ is more striking thanks to the higher event
rates and the smaller neutrino path in matter. However, this effect and the one produced
by a change of θ13 are difficult to disentangle with a single measurement (see top plot in
Fig. 15). The correlations between δ and θ13 prevent a precise determination of any of them.
Nonetheless, a measurement at L = 7400 km which provides an accurate determination of
Electrons
Wrong-sign muons
Fig. 14. Energy spectra and differences between no CP violation (δ = 0) or maximal CP violation
(δ = π/2). The two upper plots refer to the electron class, the lower ones to the wrong-sign muon
class. The plots on the left show the energy spectra for the two cases, the ones on the right the
different in number of events as a function of the energy. The helicity of muons circulating in the
ring has been chosen in such a way to enhance the effect, i.e., negative muons have been used for the
upper plots, and positive ones for the lower plots.
θ13 , can be combined with data collected at L = 732 km to produce a precise determination
of δ.
The other possibility to test CP violation is to perform a single measurement at L =
2900 km. As shown in Fig. 14, at this distance the effect of δ 6= 0 is twofold: not only the
event rate is modified but also the spectral shape. This last effect cannot be produced by
a change on θ13 . Fig. 15 shows that at this distance the correlation between δ and θ13 has
diminished and therefore a better determination of the parameters can be achieved with a
single measurement.
Fig. 15. Correlation between θ13 and CP phase δ for two different baselines and 2 × 1021 decays.
Matter effects introduce asymmetries between neutrinos and antineutrinos, which can
be misinterpreted as CP violating effects. However, for a baseline of L = 2900 km, matter
effects are less important than for the longest baseline. The “true” CP effect determined by
δ and the matter effects driven by the density ρ can be measured at the same time, even if a
priori correlated, since their energy dependence can be different. This is shown in Fig. 16,
where the result of a simultaneous fit to the average matter density ρ and the phase δ is
presented for the two cases where ρ is left as a free parameter or it is known beforehand
with a 3% accuracy.
In case no CP violation is observed, the allowed δ values at 90% C.L. as a function of
1m212 are shown in Fig. 17 for two different muon normalizations. On the other hand, if a
significant effect is detected, the precision achievable on the measurement of the CP phase
δ is shown in Fig. 18. We fit δ using all event classes, leaving the five parameters governing
the oscillation free. Assuming a reference value of 90◦ for δ and 1021 muon decays of each
polarity, we get: 90 ± 15◦ . Therefore a precision around 20% is expected. Finally, we note
that the change on the expected precision for θ13 is negligible when 1m212 and θ12 are
included in the fit.
Fig. 16. Correlation between the average matter density ρ and the CP phase δ for L = 2900 km. In
the lower plot we leave Earth’s mean density as a free parameter in the fit. In the upper plot we assume
that density is known within 3%. We see that the presence of matter does not spoil the possibility of
performing a measurement of δ.
10.3. Use of quasi-elastic events
Quasi-elastic events can also be useful to spot the presence of CP violation. Table 8
shows the expected number of QE electron events after kinematic cuts (a back to back
electron–proton final state with proton kinetic energy in excess of 100 MeV). Three
different assumptions for the total number of muon decays and the baseline of 2900 km
have been assumed. In case CP is conserved, we expect 35 (96) quasi-elastic electron
events for 1m232 = 3.5(7) × 10−3 eV2 and 1021 muon decays of each polarity. In case
δ = π/2, we expect 26 and 85 events for the two mass differences considered. The effect
is at the one sigma level. To obtain a statistically conclusive signal for CP violation would
require more than 1021 muon decays.
11. Conclusions
In this paper, we tried to understand in deeper detail the capabilities of an experiment

at the Neutrino Factory, with the aim of exploiting as much as possible the possibility of
Fig. 17. 90% C.L. sensitivity on the CP phase δ as a function of 1m212 for two different
normalizations: solid (dashed) line corresponds to 1021 (1020 ) muons decays of each polarity.
Table 8
Expected QE electron events (Nele ) at a baseline of 2900 km in case there is no CP violation and
in case CP violation phase in vacuum is maximal. The last column shows, in number of sigmas, the
statistical significance expected for the QE electron sample in case CP is violated in the lepton sector
CP violation with quasi-elastic events
L = 2900 km Nele Nele Stat.

(δ = 0) (δ = π/2) significance
1m232 = 3.5 × 10−3 eV2 1021 µ± 35 26 1.5 σ

23 = 0.5, sin 2θ23 = 0.05
sin2 θ 2 5 × 1021 µ± 175 130 3.4 σ
1m212 = 10−4 eV2 , sin2 θ12 = 0.5 1022 µ± 350 260 4.8 σ
1m232 = 7 × 10−3 eV2 1021 µ± 96 85 1.1 σ

23 = 0.5, sin 2θ23 = 0.05
sin2 θ 2 5 × 1021 µ± 480 425 2.5 σ
1m212 = 10−4 eV2 , sin2 θ12 = 0.5 1022 µ± 960 850 3.6 σ
studying several neutrino transitions at the same time. For this reason, we have considered
as a baseline detector a large Liquid Argon TPC with external muon identifier, the most
versatile design proposed so far for large neutrino experiments.
Fig. 18. Measurement of the CP-violating phase δ for 2 × 1021 µ decays and L = 2900 km. The two
mass differences and the three mixing angles have been left free during the fit. The reference value is
δ = π/2. The three curves are obtained using respectively in the fit all classes, only the muon classes,
or only electrons and NC-like events.
Assuming an oscillation scenario favoured by present experimental results, the leading

oscillation would be between the second and third neutrino family, which are maximally
mixed. Therefore, a very precise determination of the parameters governing this transition,
θ23 and 1m223 is essential also for the understanding of all other processes. This is mainly
achieved using the information coming from the νµ disappearance (right-sign muon class),
provided that the baseline and beam energy are chosen in such a way that the first
oscillation maximum is visible as a dip in the oscillated spectrum.
The maximal sensitivity to θ13 is achieved for very small background levels, since we
are looking in this case for small signals; most of the information is coming from the clean
wrong-sign muon class, and from quasi-elastic events.
On the other hand, if its value is not too small, for a measurement of θ13 , the
signal/background ratio could be not so crucial, and also the other event classes can
contribute to this measurement.
Like for a B-Factory, a ν-Factory should have among its aims the overconstraining of
the oscillation pattern, in order to look for unexpected new physics effects. This can be
achieved in global fits of the parameters, where the unitarity of the mixing matrix is not
strictly assumed. Using a detector able to identify the τ lepton production via kinematic
means, it is possible to verify the unitarity in νµ → ντ and νe → ντ transitions. For this
latter, the possibility of a kinematical τ identification for wrong-sign muon events could
allow for the first time a clear identification of this type of oscillations.
The study of CP violation in the lepton system is a very fascinating subject, and
probably the most ambitious goal of this kind of machines. It will only be possible for
high beam intensities, and if the parameters governing the solar neutrino deficit are in the
region usually indicated as large mixing angle MSW solution. Matter effect can mimic
CP violation; however, a multiparameter fit at the right baseline can allow a simultaneous
determination of matter and CP-violating parameters.
Also for CP violation measurements most of the information would come from the
wrong-sign muon class, but since in this case the electron class would also be affected,
the study of these events (and of the very clean quasi-elastic interactions) can provide
essential cross-checks for these delicate measurements.
Appendix A. Oscillations in matter
In Ref. [43], the authors compute analytic expressions for the mass eigenvalues, mixing
angles and CP-violation phase in matter, assuming the mixing matrix U is parametrized
“à la CKM”. Since several missprints were observed, we reproduce here the corrected
expressions:
For neutrinos, the mass eigenvalues in matter M1 , M2 and M3 are:
√
A 1p 2 3p 2 p
M12 = m21 + − A − 3B S − A − 3B 1 − S 2 , (A.1)
3 3 3
√
A 1p 2 3p 2 p
M22 = m21 + − A − 3B S + A − 3B 1 − S 2 , (A.2)
3 3 3
A 2p 2
M32 = m21 + + A − 3B S, (A.3)
3 3
where
A = 1m221 + 1m231 + D, (A.4)

B = 1m2211m231 + D 1m231 c13
2
+ 1m221 c13
2 2
c12 + s13
2
, (A.5)
C = D1m221 1m231 c13
2 2
c12 , (A.6)
√
D = 2 2 GF Ne E for neutrinos, (A.7)
3
1 2A − 9AB + 27C
S = cos arccos p , (A.8)
3 2 (A2 − 3B)3
and cij = cos θij and sij = sin θij . The mixing angles and CP phase in matter are:

− M24 − αM22 + β 1M31 2
sin θ12 =
2 m
, (A.9)
1M32 M1 − αM1 + β − 1M31 M2 − αM22 + β
2 4 2 2 4
M34 − αM32 + β
m
sin2 θ13 = 2 1M 2
, (A.10)
1M31 32
2 + F 2 c 2 + 2GF c s c
G2 s23 23 23 δ
m
sin2 θ23 = 23
, (A.11)
G2 + F 2

−iδm (G2 e−iδ − F 2 eiδ )s23 c23 + GF c23
2 − s2
e =q
23
, (A.12)
2 + F 2 c 2 + 2GF c s c
23 23 δ G c23 + F s23 − 2GF c23 s23 cδ
G2 s23 2 2 2 2
23
where

α = m23 c13
2
+ m22 c13
2 2
c12 + s13
2
+ m21 c13 s12 + s13
2 2 2
, (A.13)

β = m3 c13 m2 c12 + m1 s12 + m2 m1 s13 ,
2 2 2 2 2 2 2 2 2
(A.14)
2
G = 1m231 M32 − m21 − 1m221 − 1m221 M32 − m21 − 1m231 s12 c13 s13 , (A.15)

F = M3 − m1 − 1m31 1m21c12 s12 c13 .
2 2 2 2
(A.16)
For antineutrinos, we must replace D by −D.
In the case of one-mass scale approximation, the oscillation probabilities are:
2
P (νe → νe , E, L) = 1 − sin2 2θ13
m
∆32 ,
2 m 2
P (νe → νµ , E, L) = sin 2θ13 sin θ23 ∆32 ,
2 m
2 m 2
P (νe → ντ , E, L) = sin2 2θ13
m
cos θ23 ∆32 ,
2 m 2 m 2
P (νµ → νµ , E, L) = 1 − 4 cos2 θ13m
sin θ23 1 − cos2 θ13 m
sin θ23 ∆32 ,
2 m 2
P (νµ → ντ , E, L) = cos θ13 sin 2θ23 ∆32 ,
4 m
(A.17)

where 132 = sin M3 − M2 L/(4E) .
2 2 2 2
References
[1] Y. Fukuda et al., Super-Kamiokande Collaboration, Phys. Lett. B 433 (1998) 9.

[2] C.W. Walter, Invited talk, on behalf of the Super-Kamiokande Collaboration, at International
Europhysics Conference on High-Energy Physics (EPS-HEP 99), Tampere, Finland, July 15–
21, 1999.
[3] B. Pontecorvo, J. Exp. Theor. Phys. 33 (1957) 549; Sov. Phys. JETP 6 (1958) 429.
[4] B. Pontecorvo, J. Exp. Theor. Phys. 34 (1958) 247; Sov. Phys. JETP 7 (1958) 172.
[5] Z. Maki, M. Nakagawa, S. Sakata, Prog. Theor. Phys. 28 (1962) 870.
[6] B. Pontecorvo, J. Exp. Theor. Phys. 53 (1967) 1717.
[7] V. Gribov, B. Pontecorvo, Phys. Lett. B 28 (1969) 493.
[8] E362 Proposal for a long baseline neutrino oscillation experiment, using KEK-PS and Super-
Kamiokande, February 1995.
[9] H.W. Sobel, in: Proceedings of Eighth International Workshop on Neutrino Telescopes, Venice,
Vol. 1, 1999, p. 351.
[10] E. Ables et al., MINOS Collaboration, P-875: A long baseline neutrino oscillation experiment
at Fermilab, FERMILAB-PROPOSAL-P-875.
[11] The MINOS detectors technical design report, NuMI-L-337, October 1998.
[12] K. Kodama et al., OPERA Collaboration, OPERA: A long baseline ντ appearance experiment
in the CNGS beam from CERN to Gran Sasso, CERN/SPSC 99-20; SPSC/M635; LNGS-LOI
19/99.
[13] F. Arneodo et al., ICARUS and NOE Collaboration, ICANOE: Imaging and calorimetric neu-
trino oscillation experiment, LNGS-P21/99; INFN/AE-99-17; CERN/SPSC 99-25; SPSC/P314.
[14] F. Cavanna et al., ICANOE Collaboration, ICANOE: Answers to questions and remarks
concerning the ICANOE project, LNGS-P21/99-ADD2; CERN/SPSC 99-40; SPSC/P314
Add 2.
[15] A. Rubbia, ICARUS and NOE collaborations, hep-ex/0001052. Updated information can be
found at http://pcnometh4.cern.ch.
[16] E. Church et al., A proposal for an experiment to measure muon-neutrino → electron-neutrino
oscillations and muon-neutrino disappearance at the Fermilab Booster: BooNE, FERMILAB-P-
0898. Updated information on the BOONE proposal can be found at http://www.neutrino.lanl.
gov/BooNE/.
[17] C. Athanassopoulos et al., LSND Collaboration, Phys. Rev. C 54 (1996) 2685.
[18] C. Athanassopoulos et al., LSND Collaboration, Phys. Rev. Lett. 77 (1996) 3082.
[21] S. Geer, Phys. Rev. D 57 (1998) 6989.
[22] Information on the neutrino factory studies and mu collider collaboration at BNL can be found
at http://www.cap.bnl.gov/mumu/. Information on the neutrino factory studies at FNAL can
be found at http://www.fnal.gov/projects/muon\_collider/. Information on the neutrino factory
studies at CERN can be found at http://muonstoragerings.cern.ch/Welcome.html/
[23] A. Bueno, M. Campanelli, A. Rubbia, hep-ph/9808485; ETHZ-IPP-98-05.
[24] A. Bueno, M. Campanelli, A. Rubbia, hep-ph/9809252; CERN-EP/98-140.
[25] A. de Rújula, M.B. Gavela, P. Hernández, Nucl. Phys. B 547 (1999) 21.
[26] V. Barger, S. Geer, R. Raja, K. Whisnant, hep-ph/9911524.
[27] V. Barger, S. Geer, R. Raja, K. Whisnant, hep-ph/0003184.
[28] A. Bueno, M. Campanelli, A. Rubbia, hep-ph/9905240; Nucl. Phys. B, in press.
[29] V. Barger, S. Geer, K. Whisnant, Phys. Rev. D 61 (2000) 053004.
[30] M. Freund, M. Lindner, S.T. Petcov, A. Romanimo, hep-ph/9912457.
[31] A. Donini, M.B. Gavela, P. Hernández, S. Rigolin, hep-ph/9909254.
[32] M. Koike, J. Sato, hep-ph/9909469.
[33] G. Barenboim, R. Scheck, hep-ph/0001208.
[34] H. Fritzsch, Z. Xing, hep-ph/0002248.
[35] D.A. Harris, A. Para, Neutrino oscillation appearance experiment using nuclear emulsion and
magnetized iron, hep-ex/0001035.
[36] A. Cervera et al., Golden measurements at a neutrino factory, hep-ph/0002108.
[37] P. Lipari, hep-ph/9903481.
[38] L. Wolfenstein, Phys. Rev. D 17 (1978) 2369.
[39] L. Wolfenstein, Phys. Rev. D 20 (1979) 2634.
[40] S.P. Mikheyev, A.Yu. Smirnov, Sov. J. Nucl. Phys. 42 (1986) 913.
[41] J.N. Bahcall, P.I. Krastev, A.Y. Smirnov, Phys. Rev. D 58 (1998) 096016; hep-ph/9807216, and
references therein.
[42] F. James, MINUIT manual, CERN program libraries, p. 43.
[43] H.W. Zaglauer, K.H. Schwarzer, Z. Phys. C 40 (1988) 273.
[44] V. Barger, K. Whisnant, S. Pakvasa, R.J.N. Phillips, Phys. Rev. D 22 (1980) 2718.
[45] I. Mocioiu, R. Shrock, hep-ph/0002149.
[46] M. Apollonio et al., CHOOZ Collaboration, Phys. Lett. B 466 (1999) 415.
[47] M. Apollonio et al., CHOOZ Collaboration, Phys. Lett. B 420 (1998) 397.
[48] G.L. Fogli, E. Lisi, A. Marrone, G. Scioscia, Phys. Rev. D 59 (1999) 033001.
[49] A. Ferrari, A. Rubbia, NUX: A neutrino event generator, ICARUS internal note, in preparation.
[50] A. Ferrari, P.R. Sala, ATL-PHYS-97-113, ATLAS internal note, Trieste; in: A. Gandini,
G. Reffo (Eds.), Proc. of the Workshop on Nuclear Reaction Data and Nuclear Reactors Physics,
Design and Safety, ICTP, Miramare-Trieste, Italy, April 15–May 17, 1996, Vol. 2, World
Scientific, 1998, p. 424.
[51] S. Baker, R. Cousins, Nucl. Instrum. Methods 221 (1984) 437.
Twist-three effects in two-photon processes

A.V. Belitsky a,b,∗ , D. Müller b
a C.N. Yang Institute for Theoretical Physics, State University of New York at Stony Brook,
Stony Brook, NY 11794-3840, USA
b Institut für Theoretische Physik, Universität Regensburg, D-93040 Regensburg, Germany
Abstract
We give a general treatment of twist-three effects in two-photon reactions. We address the
issue of the gauge invariance of the Compton amplitude in generalized Bjorken kinematics and
relations of twist-three ‘transverse’ skewed parton distributions to twist-two ones and interaction
dependent three-particle correlation functions. Finally, we discuss leading order evolution of twist-
three functions and their impact on the deeply virtual Compton scattering.  2000 Elsevier Science
B.V. All rights reserved.
PACS: 11.10.Hi; 12.38.Bx; 13.60.Fz

Keywords: Two-photon processes; Deeply virtual Compton scattering; Skewed parton distribution; Twist-three
contributions
1. Introduction
Exclusive two-photon reactions involving two hadrons (scattering or production),

like deeply virtual Compton scattering (DVCS) [1–3], γ ∗ γ → hh with small invariant
mass of the hadron system [1,4], etc., are of special interest since they involve new
nonperturbative characteristics which generalize conventional parton densities and/or
distribution amplitudes. The leading twist factorization gives the amplitude of these
processes in terms of a perturbatively calculable coefficient function and a skewed parton
distribution (for DVCS) [3,5,6] or a generalized distribution amplitude (for γ ∗ γ → hh) [7]
which are responsible for soft physics. However, the twist-two analysis of the Compton
scattering at non-zero t-channel momentum transfer, 1 6= 0, is inadequate since it
violates
p explicitly the gauge invariance of the amplitude due to approximation involved,
1⊥ / −q 2 1. This calls for a consistent treatment of the effects suppressed in 1⊥ ,
∗ Corresponding author: C.N. Yang Institute for Theoretical Physics, State University of New York at Stony
Brook, Stony Brook, NY 11794-3840, USA; Tel.: +1 631 632 4171; Fax: +1 631 632 7954.
E-mail address: belitsky@insti.physics.sunysb.edu (A.V. Belitsky).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 4 2 - 3
612 A.V. Belitsky, D. Müller / Nuclear Physics B 589 (2000) 611–630
i.e., twist expansion. 1 As a first step one takes the contributions linear in 1⊥ which
are of twist-three. Obviously, then the amplitude will be gauge invariant to the twist-four
accuracy. Repeating the steps one improves the amplitude accordingly.
Let us demonstrate the peculiarities of the off-forward kinematics on a simple example
of a free Dirac fermion theory. As we will see, even and odd parity structures ‘talk’ to each
other in the case at hand, when the operators with total derivatives within the context of
the operators product expansion are relevant. This will be the source of restoration of the
electromagnetic gauge invariance of the two-photon amplitude defined by a chronological
product T {jµ (x)jν (y)} of currents jµ (x) = ψ̄γµ ψ. The leading light-cone singularity
(x − y)2 → 0 arises from the hand-bag diagram and reads
T {jµ (x)jν (y)} = i ψ̄(x)γµ S
/(x − y)γν ψ(y) + i ψ̄(y)γν S/(y − x)γµψ(x), (1)
where S/(x) = 2π1 2 xx/4 is the free quark propagator. Taking into account the equation of
motion ∂/ ψ = 0 and ∂/ S/(x) = −iδ(x) it is a simple task to show that the hand-bag diagram
respect current conservation. After performing the decomposition of the Dirac structure in
Eq. (1) it reduces to 2
T {jµ (x)jν (y)} = Sµν;ρσ iSρ (x − y){ψ̄(x)γσ ψ(y) − (x ↔ y)}
− iµνρσ iSρ (x − y){ψ̄(x)γσ γ5 ψ(y) + (x ↔ y)}, (2)
where Sµν;ρσ = gµρ gνσ + gµσ gνρ − gµν gρσ . One finds that current conservation does
not separately occur in both terms. Employing the density matrix |P1 ihP2 | for free
fields, we find, of course, that the amplitude hP2 |T {jµ (x)jν (y)}|P1 i respects current
conservation, however, there appears a cancellation between both matrix elements of
different parity on the r.h.s. in Eq. (2). So we realize that the current conservation is
not manifest in the decomposition (2). This arises due to operators with total derivatives
in twist decomposition of ψ̄(x)γµ (1, γ5 )ψ(y), see later Eq. (28). However, once this
decomposition is performed the gauge invariance is restored automatically. As compared
to this simple example the only complication in QCD to leading order emerges due to the
presence of the interaction dependent three-particle contributions. This does not present
any difficulty and will be solved in the next section.
Our consequent presentation is organized as follows. The next section is devoted to a
detailed description of the twist-three formalism for the generalized Compton amplitude
and as a result the restoration of the gauge invariance sketched above. In Section 3
we address the issue of a twist decomposition of light-ray (and local) operators in a
situation when total derivatives are relevant. Then we turn in Section 4 to the two
cases of matrix elements of operators sandwiched between spin-0 and 12 hadrons and
derive relations for twist-three two-particle ‘transverse’ functions in terms of twist-two
(Wandzura–Wilczek contribution) and three-particle correlation functions. In Section 5 we
comment on the phenomenological consequences for different asymmetries measurable in
the DVCS process, and then in Section 6 point out that the evolution of twist-three skewed
1 Here we imply kinematical definition of twist.

2 We use the conventions for Dirac and Lorentz tensors from Itzykson and Zuber [8].
A.V. Belitsky, D. Müller / Nuclear Physics B 589 (2000) 611–630 613
distributions is known in leading logarithmic approximation from analogous studies for

forward kinematics. Finally, we give our conclusions.
2. Generalized two-photon amplitude
The amplitude of scattering of two virtual photons with momenta q1 and q2 on a hadron
target, with incoming (outgoing) momentum P1 (P2 ), is given by a Fourier transform of a
correlator of electromagnetic currents,
Z
Tµν = i d 4 x eiq·x hP2 |T {jµ (x/2)jν (−x/2)}|P1i. (3)
One introduces the vectors q = 12 (q1 + q2 ) and P = P1 + P2 , 1 = P2 − P1 = q1 − q2 to

describe the amplitude. They can be used to construct a pair of the light-cone vectors nµ
and n?µ , such that n2 = n?2 = 0 and n · n? = 1, as follows
p
2ξ 1 − 1 − 4(ξ δ)2
nµ = − p qµ − p Pµ ,
q 2 1 − 4(ξ δ)2 2q 2δ 2 1 − 4(ξ δ)2
p
ξ δ2 1 + 1 − 4(ξ δ)2
nµ = p
?
qµ + p Pµ , (4)
1 − 4(ξ δ)2 4 1 − 4(ξ δ)2
where δ 2 ≡ (M 2 − 12 /4)/q 2 . The transverse metric and antisymmetric tensor are
⊥ ≡g ⊥
µν − nµ nν − nν nµ , µν ≡ µν−+ . Here and in the following we use
defined as, gµν ? ?
the conventions for the generalized Bjorken variable ξ = −q 2/P · q and skewedness
η = 1 · q/P · q which are scaling variables of (3). Neglecting the corrections O(δ 2 ) we
have then to twist-four accuracy
ξ
nµ = − (2qµ + ξ Pµ ), n?µ = 12 Pµ , 1µ = 2η n?µ + 1⊥
µ. (5)
q2
This approximation will be used throughout in our analysis.
In the generalized Bjorken kinematics −q 2 → ∞, P · q → ∞, 1 · q → ∞, 12 = finite,
ξ and η being fixed the contribution to (3) comes from the diagrams in Fig. 1
Tµν = Tµν
2
+ Tµν
3
, (6)
where Tµν 2 comes from hand-bag diagram (left) while T 3 from antiquark–gluon–quark
µν
one (right). Explicit calculation, done in the light cone gauge B+ = 0, gives
Z
1
Tµν = − 2 dx C (+) (x, ξ ) Sµν;ρ− V Kρ (x, η) + 2iµνρσ qσ AOρ (x, η)
2
2q
2
q
− C (−) (x, ξ ) Sµν;ρ+ V Oρ (x, η)
ξ

− iµνρ− 2x AOρ (x, η) − AKρ (x, η) , (7)
for the first one, and

Fig. 1. Two- and three-leg coefficient functions (plus crossed diagrams) for DVCS process which
form a gauge invariant amplitude to twist-three accuracy.
Z Z
1 dτ
3
Tµν =− dx
2q 2 τ − i0

× C (+) (x, ξ )Sµν;ρ− V Sρ− (x, x − τ, η) + V Sρ+ (x + τ, x, η)

+ C (−) (x, ξ )Sµρ;ν− V Sρ− (x, x − τ, η) − V Sρ+ (x + τ, x, η) , (8)
for the second. Here Sµν;ρσ tensor defined in the introduction and tree coefficient functions
are
C (±) (x, ξ ) = (1 − x/ξ − i0)−1 ± (1 + x/ξ − i0)−1 . (9)
We use here the following conventions for generalized functions
Z
dκ iκx
I
Oρ (x, η) = e hP2 |IOρ κ2 , − κ2 |P1 i,
2π
Z
dκ iκx
I
Kρ (x, η) = e hP2 |IKρ κ2 , − κ2 |P1 i,
2π
Z
I ± dκ1 dκ2 i κ1 (x1 +x2 )+iκ2 (x1 −x2 ) κ1
Sρ (x1 , x2 , η) = e2 hP2 |ISρ± κ1
2 , κ2 , − 2 |P1 i, (10)
2π 2π
where the two-quark operators are defined by
I
Oρ (κ, −κ) = ψ̄(−κn)IΓρ ψ(κn),
↔
⊥
I
Kρ (κ, −κ) = ψ̄(−κn) IΓ+ i ∂ ρ ψ(κn), (11)
with IΓ= {γρ , γρ γ5 } for I = V , A and antiquark–gluon–quark ones read

V ± e+ρ (κ2 n) ψ(κ1 n),
Sρ (κ1 , κ2 , κ3 ) = ig ψ̄ (κ3 n) γ+ G+ρ (κ2 n) ± iγ+ γ5 G

A ± e+ρ (κ2 n) ψ(κ1 n).
Sρ (κ1 , κ2 , κ3 ) = ig ψ̄ (κ3 n) γ+ γ5 G+ρ (κ2 n) ± iγ+ G (12)
The latter two are related to each other by the ‘duality’ equation
⊥ A ±
iρσ Sσ = ±V Sρ± . (13)
By means of the quark equation of motion, D/ ψ = 0, we can obtain the following relation
between the correlation functions introduced so far
∂ V ⊥ ⊥
O (κ, −κ) − iρσ ∂+ AOσ⊥ (κ, −κ) + i V Kρ (κ, −κ) + iρσ
⊥ ⊥A
∂σ O+ (κ, −κ)
∂κ Z ρ

+ dλ w(λ − κ)V Sρ− (κ, λ, −κ) + w(λ + κ)V Sρ+ (κ, λ, −κ) = 0,
⊥ ∂ A ⊥
∂+ V Oρ⊥ (κ, −κ) − iρσ ⊥ A
Oσ (κ, −κ) + ρσ Kσ (κ, −κ) − ∂ρ⊥ V O+ (κ, −κ)
Z ∂κ

+ dλ w(λ − κ)V Sρ− (κ, λ, −κ) − w(λ + κ)V Sρ+ (κ, λ, −κ) = 0, (14)
where w(κ) = −θ (κ) for the ML prescription on the infrared pole in the gluon propagator.
Here ∂ stands for the total derivative. In terms of the momentum fraction space functions
introduced in Eqs. (10) these relations read
2x V Oρ⊥ (x, η) + 2ηiρσ
⊥ A ⊥
Oσ (x, η) − iρσ⊥
1⊥ V ⊥
σ O+ (x, η) − Kρ (x, η)
A
Z
dτ V −
− Sρ (x, x − τ, η) + V Sρ+ (x + τ, x, η) = 0,
τ − i0
2xiρσ Oσ (x, η) + 2ηV Oρ⊥ (x, η) − 1⊥
⊥ A ⊥ ⊥ A ⊥
ρ O+ (x, η) − iρσ Kσ (x, η)
V
Z
dτ V −
+ Sρ (x, x − τ, η) − V Sρ+ (x + τ, x, η) = 0. (15)
τ − i0
So that expressing K in terms of other correlation functions in Eq. (7) we get the result
Z
1 (1) (−)
Tµν = − 2 dx Tµν C (x, ξ )V O+ (x, η) + Tµν (2) (+)
C (x, ξ )AO+ (x, η)
q

+ Tµν;ρ C (−) (x, ξ )V Oρ⊥ (x, η) + Tµν;ρ C (+) (x, ξ )AOρ⊥ (x, η) ,
(3) (4)
(16)
where the support properties of distributions restrict the integration range of the variable x
within the interval [−1, 1]. Here the Lorentz tensors are

Tµν
(1)
= n?µ qν + ξ n?ν − 12 1⊥ ? ? 1 ⊥
ν + nν qµ + ξ nµ + 2 1µ − q · n gµν ,
?
⊥

Tµν
(2)
= iµν−σ qσ − 2i ρσ 1⊥ σ nµ gρν + nν gρµ ,
? ?
(3)
Tµν;ρ = qν + (2ξ − η)n?ν gµρ + qµ + (2ξ + η)n?µ gνρ ,
(4) ⊥

Tµν;ρ = iµνρσ qσ − iηρσ n?µ gσ ν + n?ν gσ µ . (17)
These expressions are obviously target independent. In the case of scalar target our result
reduces to the one obtained in Ref. [9].
It is an easy exercise to check that, after accounting for twist-three corrections, gauge
(i) (i)
invariance is fulfilled up to O(12⊥ ), i.e., twist-four effects: q1ν Tµν = q2µ Tµν = O(12⊥ ),
q 2
where we have used Sudakov decomposition for the photon momenta q1µ = − 2ξ nµ −
2
(ξ − η)n?µ + 12 1⊥ q 1 ⊥
µ and q2µ = − 2ξ nµ − (ξ + η)nµ − 2 1µ . Interesting to note that the last
?
(i)
two structures Tµν;ρ are gauge invariant provided we simultaneously contract them both
(i)
with incoming and outgoing photons q1ν q2µ Tµν;ρ = 0. Namely, e.g., for the last tensor we
(4)
get q1ν Tµν;ρ e⊥
= − 12 ξ Pµν Pν 1 (4) e⊥
ρ and q2µ Tµν;ρ = − 2 ξ Pµ Pµν 1ρ , where we use throughout
1
the convention 1 e⊥ ⊥
ρ ≡ iρσ 1ρ and introduced a projector Pµν = gµν − q1µ q2ν /q1 · q2
which gives ξ Pµ Pµν = qν + 12 (2ξ − η)Pν and ξ Pµν Pν = qµ + 12 (2ξ + η)Pµ up to terms of
order O(1⊥ ) which we drop since they would exceed the accuracy we work to. A similar
(3)
result we get for Tµν;ρ e⊥ being replaced by 1⊥ . Using the definitions of the photon
with 1
and hadron momenta in terms of light-cone vectors, consequent simple algebra leads to the
following structures in photon and hadron momenta

q2 q1µ q2ν ξ P · q2 P · q1
Tµν =
(1)
gµν − + Pµ − q1µ Pν − q2ν ,
2ξ q1 · q2 2 q1 · q2 q1 · q2

Pµ q2θ Pν q1λ
Tµν
(2)
= 2i θλρσ Pρ qσ gµθ − gνλ − ,
P · q2 P · q1

(3) q1µ q2ρ P · q1
Tµν;ρ = ξ gµρ − Pν − q2ν
q1 · q2 q1 · q2

q1ρ q2ν P · q2
+ ξ gνρ − Pµ − q1µ ,
q1 · q2 q1 · q2

(4) Pµ q2θ Pν q1λ
Tµν;ρ = iθλρσ qσ gµθ − gνλ − , (18)
P · q2 P · q1
where a missing twist-four part is restored minimally according to previous discussion.
All structures have a well defined forward limit. It will be shown in Section 5 that these
terms are equivalent to those used in our previous studies [10,11]. In subsequent sections
we derive relations between the ‘transverse’ generalized distributions introduced here and
conventional twist-two ones as well as three-particle correlation functions.
3. Twist decomposition
In this section we present a decomposition of the two-quark operators into separate twist
components to twist-three accuracy. Similar analyses have been done in Refs. [12–15].
However, an essential new ingredient of our study is the treatment of operators with total
derivatives. Since the group-theoretical notion of twist as the dimension minus spin of an
operator is well defined for local operators, our strategy will be thus: first, an expansion of
a light cone operator in infinite Taylor series in local ones, second, an extraction of definite
twist components and as a final step the resummation of the result back into nonlocal
form. An alternative approach directly based in terms of light-ray operators is described in
Ref. [13]. In the following we will not take care of trace terms, proportional to nρ , since
they only contribute at twist-four level.
The Taylor expansion of Eq. (11) in terms of local operators simply reads
∞
X (−iκ)j
Oρ (κ, −κ) = nµ1 · · · nµj Oρ;µ1 ...µj ,
j!
j =0
↔ ↔
with Oρ;µ1 ...µj = ψ̄Γρ i Dµ1 · · · i Dµj ψ, (19)
↔ → ←
where Dµ = D µ − D µ and Dµ = ∂µ − igBµ . To have a well-defined decomposition into
twist-two and three contributions we extract tensors Rρ;µ1 ...µj with (j + 1) indices corre-
ρ µ1 µ2 . . . µn µ1 µ2 . . . µn
sponding to the two Young tables and .
ρ
This can easily be done with the result
Oρ (κ, −κ) = R2ρ (κ, −κ) + R3ρ (κ, −κ)

∞
X
(−iκ)j 2j
= nµ1 · · · nµj R2ρ;µ1 ...µj + R3ρ;µ1 ...µj , (20)
j! j +1
j =0
with
↔ ↔
R2ρ;µ1 ...µj = S ψ̄Γρ i Dµ1 · · · i Dµj ψ,
ρµ1 ...µj
↔ ↔
R3ρ;µ1 ...µj = S A S ψ̄Γρ i Dµ1 · · · i Dµj ψ, (21)
µ1 ...µj ρµ1 µ1 ...µj
where Sµ1 ...µj is a symmetrization (and trace subtraction) operation of j indices with
weight 1/j ! and antisymmetrization being defined by Aµ1 µ2 tµ1 µ2 = 12 (tµ1 µ2 − tµ2 µ1 ). In
terms of light-ray operators the twist-two part reads
Z1 (
R2ρ (κ, −κ) = du ψ̄(−uκn)Γρ ψ(κn)
0
Zu
κ → ←
+ dτ ψ̄(−uκn)Γ+ ∂ ρ (uκn)− ∂ ρ (−uκn)
2
−u
)

− 2igBρ (τ κn) ψ(uκn) (22)
in the light-cone gauge B+ = 0. If we would work in a covariant gauge this would result,
apart from restoration of the gauge-link factors, to the substitution in the square brackets
→ ←
∂ρ (uκn)− ∂ ρ (−uκn) − 2igBρ (τ κn)
→ ←
→D ρ (uκn)− D ρ (−uκn) + 2iτ κgG+ρ (τ κn). (23)
One can easily project onto the twist-two part from the operator Oρ by contraction with
the vector nρ , namely, R2+ = O+ .
The conversion of the twist-three part in terms of three-particle operators is more tricky
as compared to the forward case due to relevance of operators involving total derivatives.
From the Eqs. (20) and (22) one can easily derive, however, the following equation, cf.
[12],
Z1 (
V
Oρ (κ, −κ) = V R2ρ (κ, −κ) + κ du iu +ρµν ∂µ AOν (uκ, −uκ)
0
Zu
κ
+ dτ (u − τ )V Sρ+ (uκ, τ κ, −uκ)
2
−u
)

− (u + τ )V Sρ− (uκ, τ κ, −uκ) , (24)
→ ←
with ∂µ = ∂ µ + ∂ µ being the total derivative, and similar expression for AOρ which is
obtained upon substitution Γρ → Γρ γ5 . These two equations set an infinite iteration in
total derivatives. We can represent these relations in a schematic form

V
O = V R2 + J ∗ ∂ AO + V S , A
O = AR2 + J ∗ ∂ V O + AS ,
with J being an integral operator. A formal solution to this system of equations reads
−1 V 2
V
O = 1 − J 2∂ 2 ∗ R + ∂J ∗ AR2 + J ∗ (J ∂ + 1) ∗ V S ,
and similar for AO. So that IO − IR2 is a twist-three part of IO. The series can be summed
up into total translations in the light cone formalism. Let us demonstrate first the effect of
total derivatives in the basis of local operators. The moments of Eq. (24) look like
j
V
Oρ;j = V R2ρ;j + i+ρµν i∂µ AOν;j −1
j +1
j −1
2 X + −
− (j − k)V Sρ;j,k − k V Sρ;j,k , (25)
j +1
k=1
where obviously Oρ;j = nµ1 . . . nµj Oρ;µ1 ...µj and we have introduced three-particle local
operators
↔ ↔
V ±
Sρ;j,k = ig i j −2 ψ̄ ∂ k−1
+
e+ρ ∂ j+−k−1 ψ,
γ+ G+ρ ± iγ+ γ5 G (26)
and the analogous one for odd parity, i.e., V → A and γ+ → γ+ γ5 . The solution for R3
reads
j −1
1 X
V
R3ρ;j = (j − l)(i∂+ )l σl+1 i+ρµν i∂µ AR2ν;j −l−1
2j
l=0

− σl (i∂ρ nσ − gρσ i∂+ )V R2σ ;j −l−1
j −2 j −l−1
1X X + l V −

− (i∂+ )l (j − k − l)V Sρ;j −l,k − (−1) k Sρ;j −l,k , (27)
j
l=0 k=1
with σl = 2 [1 − (−1) ].
1 l Now we can resum this equation back into light-ray operators, or
this can directly be obtained from Eq. (25), with the result
Z1 (
κ
V
R3ρ (κ, −κ) = du ui+ρµν ∂µ AR2ν κ, (ū − u)κ + AR2ν (u − ū)κ, −κ
2
0

− u(∂ρ nσ − gρσ ∂+ ) V R2σ κ, (ū − u)κ − V R2σ (u − ū)κ, −κ
Zu

+ κ dτ (u − τ )V Sρ+ κ, (τ + ū)κ, (ū − u)κ
−u
)

− (u + τ )V Sρ− (u − ū)κ, (τ − ū)κ, −κ , (28)
and same for V ↔ A. Contrary to the forward scattering there appeared contributions
from different parity operators to a given twist-three one and the ‘center-of-mass’ of two-
and three-particle operators gets shifted by a total translation exp (±i ūκ∂). This equation
gives a relation between skewed parton distributions of different ‘twists’ when sandwiched
between hadronic states, which we discuss in the next section.
4. Twist-three skewed parton distributions
In this section we define twist-three skewed parton distributions and their relation to
twist-two ones. Before we deal with spin- 21 functions, we consider first a more simple
case of spinless target. For generality, we deal with the incoming and outgoing hadrons of
different masses P12 = M12 6= P22 = M22 .
4.1. Spin-0 target
It is instructive to start the analysis with the expectation values of local operators. Since
for spin-0 hadron we can not form a ‘twist-two’ axial-vector, only the vector twist-two
operator develops non-zero reduced matrix elements, which are given by

hP2 |V R2ρ;µ1 ...µj |P1 i = S Pρ · · · Pµj Bj +1,j +1 + 1ρ Pµ1 · · · Pµj Bj +1,j + · · ·
ρµ1 ...µj

+ 1ρ · · · 1µj Bj +1,0 ,
hP2 |AR2ρ;µ1 ...µj |P1 i = 0. (29)
The reduced matrix elements Bj k are defined as moments of a skewed parton distribution
B(x, η):
Z1
1 dk
Bj,j −k = dx x j −1 B(x, η)|η=0 , where 0 6 k 6 j, 1 6 j. (30)
k! dηk
−1
For the analysis of the moments of skewed parton distributions we have to project both
sides of (29) with the light-cone vectors nµ1 · · · nµj . Since
nµ1 · · · nµj S 1ρ · · · 1µk Pµk+1 · · · Pµj
ρµ1 ...µj
1 j +1−i j −k
= k1ρ 1k−1
+ P+ + (j + 1 − k)Pρ 1k+ P+
j +1
1 j
= k1ρ ηk−1 + (j + 1 − k)Pρ ηk P+ , (31)
j +1
the coefficients in front of the two monomials are generated by derivatives of Bj +1 (η) w.r.t.
the skewedness parameter, namely,

j η d j 1 d
hP2 | Rρ;j |P1 i = Pρ P+ 1 −
V 2
Bj +1 (η) + 1ρ P+ Bj +1 (η),
j + 1 dη j + 1 dη
j +1
hP2 |V R2ρ;j |P1 inρ = P+ Bj +1 (η), (32)
with
j +1
X Z1
Bj +1 (η) = η Bj +1,j +1−k =
k
dx x j B(x, η).
k=0 −1
The moments Bj +1 (η) can be taken from the + component of the operator R+ . As
a consequence of symmetrization the first equation contains a Wandzura–Wilczek term
proportional to 1⊥ρ = 1ρ − ηPρ , which effectively enters as a twist-three contribution to
the scattering amplitude. The matrix element of the light-ray operator can be obtained in a
straightforward manner by resummation:
hP2 |V R2ρ (κ, −κ)|P1 i
Z1 Z1 !
d
= dx e−iκP+ x Pρ B(x, η) + 1⊥
ρ dy W2 (x, y) B(x, η) , (33)
dη
−1 −1
where the kernel reads W2 (x, y) = θ (x)θ (y − x)/y + θ (−x)θ (x − y)/(−y).
Next we define the reduced matrix elements of the antiquark–gluon–quark operators.
Since these operators are partially antisymmetrized, we have obviously two vectors 1⊥ ρ
and the dual ones 1e⊥ρ = i ⊥ 1⊥ , introduced above, in our disposal. Thus, the general
ρσ σ
decomposition of reduced matrix elements reads
±
hP2 |V Sρ;j,k |P1 i = 1⊥ ±
ρ Sj,k P+ ,
j ±
hP2 |ASρ;j,k e⊥
|P1 i = 1 e± j
ρ Rj,k P+ . (34)
The duality relation (13) reduces immediately the number of independent contributions
to two instead of four, namely, R e± = S ± . It turns out convenient to work in a mixed
j,k j,k
representation for the skewed parton distributions. We introduce a representation that
depends on the position of the gluon field and a Fourier conjugate variable with respect
to a distance between both quark fields:
V ± ⊥Z
Sρ (κ, uκ, −κ) 1ρ
hP2 | |P1 i = P+
2
dx e−iκxP+ S ± (x, u, η). (35)
AS ± (κ, uκ, −κ) e⊥
1
ρ ρ
The moments with respect to the momentum fraction x are given by a polynomial of order
j − 2 in the variable u:
Z1
Sj± (u, η) ≡ dx x j −2 S ± (x, u, η)
−1
j −1
X
j −2 1 + u k−1 1 − u j −k−1 ±
= Sj,k (η), (36)
k−1 2 2
k=1
where Sj,k (η) are polynomials in η of order j defined in Eq. (34). As a simple consequence
of our definition we have
Z1 X j −1
1+u ± k
du Sj (u, η) = 2 S ± (η),
2 j (j − 1) j,k
−1 k=1
Z1 X j −kj −1
1−u ±
du Sj (u, η) = 2 S ± (η). (37)
2 j (j − 1) j,k
−1 k=1
Finally, to find the matrix element of the operator Oρ;j = R2ρ;j + j2j
+1 Rρ;j , it remains to
3
insert our findings (32) and (37) into the solution (27):

−j 1⊥ρ
P+ hP2 | Oρ;j |P1 i = Pρ +
V
Bj +1 (η)
η
( j )
X σj +1−k d k+1
⊥ j −k
+ 1ρ (−η) − Bk+1 (η)
j +1 dη η
k=0
X
j Z1
k!(−η)j −k 1−u +
− 1⊥
ρ du Sk (u, η)
(j + 1)(k − 2)! 2
k=2 −1

j −k 1 + u −
− (−1) Sk (u, η) . (38)
2
Now we consider the axial-vector case, which is handled in the same way. The main
difference is that there is no twist-two part for spinless target. However, a non-vanishing
twist-two part of the vector operator induces now a Wandzura–Wilczek type relation due
to the epsilon tensor in Eq. (27) and provides
X
j
−j σj −k d k+1
e⊥
P+ hP2 |AOρ;j |P1 i = 1 ρ (−η) j −k
− Bk+1 (η)
j +1 dη η
k=0
X
j Z1
k!(−η)j −k 1−u +
e⊥
−1 ρ du Rk (u, η)
(j + 1)(k − 2)! 2
k=2 −1

1+u −
− (−1)j −k Rk (u, η) . (39)
2
The final step is a summation of local operators, see Eq. (20), which leads to the
expectation values of the light-ray operators in terms of Fourier transform of skewed parton
distributions in parity even and odd cases
hP2 |V Oρ (κ, −κ)|P1 i

Z1 ( Z1 → ←
−iκP+ x dy x y d y d
= dx e 1⊥
ρ W+ , − B(y, η)
|η| η η dη η dy
−1 −1

1
+ Pρ + 1⊥ B(x, η)
η ρ
Z1 Z1
⊥ dy 1 − u 00 x y +
− 1ρ du W , S (y, u, η)
|η| 2 η η
−1 −1
)
1 + u 00 x y −
+ W − , − S (y, u, η) , (40)
2 η η
hP2 |AOρ (κ, −κ)|P1i
Z1 ( Z1 → ←
dy x y d y d
= dx e −iκP+ x e⊥
1 ρ W− , − B(y, η)
|η| η η dη η dy
−1 −1
Z1 Z1
dy 1 − u 00 x y
e⊥
−1 ρ W du , R + (y, u, η)
|η| 2 η η
−1 −1
)
1 + u 00 x y −
+ W − , − R (y, u, η) . (41)
2 η η
Here W 00 (± xη , ± yη ) ≡ d2
dy 2
W (± xη , ± yη )
and the W kernels read

Θ(x, y) 1+x y−x
W (x, y) = , with Θ(x, y) = sign(1 + y)θ θ ,
1+y 1+y 1+y
1
W± (x, y) = W (x, y) ± W (−x, −y) . (42)
2
In restoration of non-local form one uses the following result for Mellin moments of
W -kernels
Z1 Xj
dx j x y (−1)j −k j −k k
x W , = η y ,
|η| η η j +1
−1 k=0
Z1 Xj
dx j x y (−1)j −k ± 1 j −k k
x W± , = η y . (43)
|η| η η 2(j + 1)
−1 k=0
Note that the piece η1 1⊥

ρ B(x, η) in Eq. (40) is cancelled by a term arising from the
convolution with the W+ kernel.
4.2. Spin- 21 target
Now we are in a position to discuss a spin- 21 target. It is convenient to express the

expectation value of local operators in terms of spinor bilinears
(b, e S(P2 , S2 )(1, γ5 )U (P1 , S1 ),

b) = U

hρ , e
hρ = US(P2 , S2 )γρ (1, γ5 )U (P1 , S1 ),
(tρσ ,e S(P2 , S2 )iσρσ (1, γ5 )U (P1 , S1 ),
tρσ ) = U (44)
Obviously, the dual tensor form factor e tρσ is obtained from tρσ by contraction with the
-tensor and can, therefore, be eliminated. Furthermore, equation of motion shows that in
each parity sector we have the relation between the structures (44)
Pρ b = M+ hρ − tρσ 1σ , 1ρ b = M− hρ − tρσ P σ ,
1ρ e
b = M+ehρ − etρσ P σ , Pρ e
b = M−e
hρ − e
tρσ 1σ , (45)
where M± = M2 ± M1 . As demonstrated above for a scalar target, the symmetrization
provides us a Wandzura–Wilczek term proportional to 1⊥ . To take advantage of the
analysis already performed in the preceding section and for a symmetrical handling of even
and odd parity sectors, our basis is spanned by the (pseudo) scalar and the (axial) vector
bilinears. At the end, we express the result in terms of conventional ones, introduced by Ji
[2],
1σ e
1ρ
e
hρ , eρ = tρσ , hρ , eeρ = b. (46)
M+ M+
Obviously, for the (pseudo) scalar bilinears we can take the results deduced for scalar
target. The only new structure for the spin- 21 target is proportional to the hρ , ehρ .
In parallel to Section 4.1, the local matrix elements read

hP2 |V R2ρ;µ1 ...µj |P1 i = S hρ Pµ1 . . . Pµj Aj +1,j +1 + · · · + 1µ1 . . . 1µj Aj +1,0
ρµ1 ...µj
b
···,
+ (47)
M+
where the ellipses (here and later on) stand for the r.h.s. of Eq. (29) (and corresponding
equations from the preceding subsection). Analogous relation holds for the parity
odd case. The momentsR 1 of the skewed parton distribution are defined analogously to
Eq. (30), Aj +1 (η) = −1 dx x j A(x, η). The difference arises only from the symmetrization
procedure
nµ1 . . . nµj S hρ 1µ1 . . . 1µk Pµk+1 . . . Pµj
ρµ1 ...µj
j −1
P+
= hρ ηk P+ + (j − k)Pρ h+ ηk + k1ρ h+ ηk−1 , (48)
j +1
and provides now a new structure
j −1
P+ ⊥ d
hP2 | Rρ;j |P1 i =
V 2
(j + 1)Pρ h+ + hρ − Pρ h+ + 1ρ h+ Aj +1 (η)
j +1 dη
b
+ ···,
M+
j b j +1
hP2 |V R2ρ;j |P1 inρ = h+ P+ Aj +1 (η) + P Bj +1 (η). (49)
M+ +
Next we resum the local result and get
Z1
h+ b
hP2 | V
R2ρ (κ, −κ)|P1 i = dx e−iκP+ x Pρ A(x, η) + B(x, η)
P+ M+
−1
Z1
⊥ h+ d b d
+ dy W2 (x, y) 1ρ A(y, η) + B(y, η)
P+ dη M+ dη
−1

h+
+ hρ − Pρ A(y, η) . (50)
P+
Projecting this expression with vector nρ we obtain conventional definitions for twist-two
skewed parton distributions. The basis used presently can be easily expressed in terms of
Ji’s parametrization as follows, for Dirac structures
e M+ M+
b= e
e+ , b= (h+ − e+ ), (51)
1+ P+
and skewed parton distributions
e= H
A e, e = ηE,
B e A = H + E, B = −E. (52)
Pρ Pρ
For the following, it is useful to note as well that M+ (hρ − P+ h+ ) = (tρσ
− P+ t+σ )1σ .
For the time being we introduce three-particle skewed parton distributions without an
explicit spinor bilinear decomposition,
V ± Z ±
Sρ (κ, uκ, −κ) −iκxP+
Sρ (x, u, η)
hP2 | |P1 i = P+ dx
2
e . (53)
AS ± (κ, uκ, −κ) Rρ± (x, u, η)
ρ
A discussion of the parametrization will be given below.
Due to length of the consequent formulas we give below only results for the vector case
since the axial one follows from the former by substitutions. Combining the Eq. (49) with
the two-particle part of Eq. (27) we get for the h-part of the local operators V Oρ
1−j
P+ hP2 |V Oρ;j |P1 i

1⊥ X
j
σj +1−k d k+1
(−η)j −k 1⊥
ρ
= Pρ + h+ Aj +1 + h
ρ + −
η j +1 dη η
k=0
Xj
σj −k j −k e ⊥e d k+1
+ (hρ P+ − Pρ h+ ) Ak+1 + (−η) 1ρ h+ −
j +1 dη η
k=0

⊥ e

+ iρσ hσ P+ − Pσ eh+ A ek+1 + · · · . (54)
Transforming it finally to the non-local form, and adding the b-term and the missing three-
particle piece, one finds
hP2 |V Oρ (−κ, κ)|P1 i
Z1
−iκP+ x 1 ⊥
= dx e Pρ + 1ρ D(x, η)
η
−1
Z1 → ←
dy x y ⊥ d y d h+
+ W+ , 1ρ − D(y, η) + hρ − Pρ A(y, η)
|η| η η dη η dy P+
−1
→ ← e
x y e ⊥ d y d e ⊥ e h+ e
+ W− , 1ρ − D(y, η) + iρσ hσ − Pσ A(y, η)
η η dη η dy P+
Z1
1 − u 00 x y +
− du W , S (y, u, η)
2 η η ρ
−1

1 + u 00 x y −
+ W − , − Sρ (y, u, η) , (55)
2 η η
where we have introduced a shorthand notation for the combinations
h+ b
D(x, η) = A(x, η) + B(x, η),
P+ M+
e e
e η) = h+ A(y,
D(y, e η) + b e
B(y, η). (56)
P+ M+
The derivatives w.r.t. η in Eq. (55) acts only on skewed parton distributions but not on
spinor bilinears. The axial-vector case is deduced by the following trivial substitutions:
b ↔e b, hρ ↔ e e A↔A
hρ , B ↔ B, e and Sρ± → Rρ± . Conventional form is obtained by means
of substitutions (51), (52). Eq. (55) is our final result. Its transverse part gives an expression
for the ‘transverse’ twist-three skewed functions in terms of ‘known’ leading twist and
three particle functions.
Let us address briefly the parametrization of genuine twist-three part of the correlation
functions. An analysis shows that there exist four independent twist-three spinor bilinears,
which we can choose to be
b h+ eb e
h+
1⊥
ρ , 1⊥
ρ ; 1⊥
ρ , 1⊥
ρ ; (57)
M+ P+ M+ P+
while the dual ones read in the same sequence
b h+ eb e
h+
e⊥
1 ρ , e⊥
1 ρ ; e⊥
1 ρ , e⊥
1 ρ . (58)
M+ P+ M+ P+
t
Note that a fifth candidate, M+ P+ρ+ , and trace terms proportional to M+
2 n enter at twist-
ρ
four level. One can immediately see that bilinears used in Eq. (55) can be spanned by this
basis via relations
h+ η h+ 1 e
h+
hρ − Pρ =− 1⊥
ρ + e⊥
1 ρ
P+ 1−η 2 P+ 1 − η 2 P+
and
e
h+ η e
⊥ h+ 1 h
e
hρ − Pρ =− 1 ρ + e⊥ + .
1
P+ 1 − η2 P+ 1 − η2 ρ P+
The parametrization of correlation functions S and R then reads
b ± h+ ± e
b e± e
h+ e±
Sρ± (x, u, η) = 1⊥
ρ S1 + 1⊥
ρ
e⊥
S2 + 1 ρ
e⊥
S1 + 1 ρ S ,
M+ P+ M+ P+ 2
e
b ± e
h+ ± b e± h+ e±
Rρ± (x, u, η) = 1⊥
ρ R1 + 1⊥
ρ
e⊥
R2 + 1 ρ
e⊥
R1 + 1 ρ R , (59)
M+ P+ M+ P+ 2
where duality (13) requires eSi± = Ri± , Re± = S ± . To reexpress the parity even and odd
i i
sector only in terms of vector and axial-vector form vectors, respectively, one can equally
use an alternative basis, where the independent elements are
b h+ eb e
h+
1⊥
ρ , hρ − Pρ , 1⊥
ρ , e
hρ − Pρ , (60)
M+ P+ M+ P+
while the dual ones read in the same sequence
b h+ − e
1ρ e hρ 1+ e
b 1ρ h+ − hρ 1+
e⊥
1 ρ , , e⊥
1 ρ , . (61)
M+ P+ M+ P+
5. Power suppression of twist-three effects in DVCS
Now we point out the phenomenological consequences of our analysis for the DVCS
process, eN → e0 N 0 γ , with η = −ξ . Since DVCS interferes with the Bethe–Heitler
process, the possibility exists to get a direct access to the skewed parton distributions via
the interference term. In leading twist-two approximation of the DVCS hadronic tensor the
amplitude squares for unpolarized or longitudinal polarized nucleon behave as:

1 1 12min 1/2
|TDVCS | ∝ 2 ,
2
TBH TDVCS ∝
?
p 1− ,
Q −12 Q2 12
1
|TBH |2 ∝ 2 , (62)
1
where 12min = −4M 2 ξ 2 /(1 − ξ 2 ) follows from the kinematical boundary on which 1⊥
vanish. Consequently, the interference terms and so also the charge and spin asymmetries
vanish (this is not the case for transversely polarized target). Since the leading 1⊥
dependence could also arise from the twist-three contributions to the DVCS amplitude,
one may argue that those enter in the interference term without power suppression.
Consequently, the whole twist-two analyses would be spoiled. This possible complication
has been already studied in the past [11], unfortunately, it was not clearly emphasized that
twist-three contributions are power suppressed.
Our aim is to show that this is indeed the case, while the complete interference term will
be published elsewhere [16]. In our analysis [11] we used the following parametrization
of the hadronic scattering amplitude that arises from the operator product expansion in the
free fermion theory, where current conservation was restored in calculations by means of
projection operators:

q · V1 V2 ρ
Tµν (q, P , 1) = −Pµσ gσ τ Pτ ν + Pµσ Pσ Pρν + Pµρ Pσ Pσ ν
P ·q P ·q
A1 ρ A2 ρ
− Pµσ iσ τ qρ Pτ ν − Pµσ iσ τ Pρ Pτ ν , (63)
P ·q P ·q
where Pµν has been introduced in Section 2. Comparing with the results given in Eqs. (16),
(18), we find that the form factors are related to each other in the following manner
Z
dx (−)
V1 µ = C (x, ξ )V Oµ (x, η),
ξ
Z
dx (+)
A1 µ = C (x, ξ )AOµ (x, η), A2 µ = 0,
ξ
Z
dx 1 Pµ
V2 µ = ξ C (−) (x, ξ ) V Oµ (x, η) − q · V O (x, η)
ξ 2P ·q

i µν1q A
+ C (+) (x, ξ ) Oν (x, η) . (64)
2 P ·q
A straightforward calculation and power counting show that in the case of DVCS
kinematics, where η = −ξ , the contributions of V2 and A2 are power suppressed in
comparison to the known twist-two result:
∗
TBH TDVCS

2 − 2y + y 2 ξ 1 q·J † q · A†1
=− kσ − qσ Jσ − 1σ 2 q · V1 + ikq1J
1−y 12 q 4 y q q2

λ(2 − y)y ξ 1 σ q ·J † q · V1†
− k σ
− q Jσ − 1 σ q · A + ikq1J .
1 − y 12 q 4 y q2 1
q2
(65)
Here Jµ is the electromagnetic current, λ and kµ denote the polarization and momentum of
the incoming electron. We employed a simple rule that qµ , kµ − y1 qµ , and Pµ , 1µ , when
contracted with an electromagnetic current or spinor bilinears, give terms of order Q2 , Q1 ,
and Q0 , respectively. Note that in agreement with this counting our results for unpolarized
nucleon target coincide with Ref. [17], where the contribution of V2 has been taken into
account. Let us remark, that also in the case of a scalar target the same situation holds true,
as it already has been shown in [9].
Since, the twist-three contributions enter as 1/Q power suppressed term to all single spin
and charge asymmetries, the comparison of models for twist-2 skewed parton distributions
with experimental data will be contaminated in an expected manner. Moreover, we
emphasize again that the twist-3 functions are completely known in terms of twist-two
ones in absence of the gluonic contributions as it is the case in the naive parton model. It
is an interesting non-perturbative problem to estimate the size of the gluonic contributions
in comparison to the Wandzura–Wilczek term. Assuming that the dynamical twist-three
effects encoded in the three-particle operators are small, which presumably happens in
view of recent experimental data [18,19] and new lattice results [20,21] for the forward
kinematics, one can use the Wandzura–Wilczek part of the relation (55) as a model for
the ‘transverse’ twist-three skewed parton distributions, hP2 |IOµ⊥ |P1 i. This will allow to
give a numerical estimate of the power suppressed contributions to asymmetries for the
kinematics of present experiments [16].
6. Logarithmic scaling violation
Let us add a final remark on the logarithmic scaling violation of twist-three functions.
Obviously the Wandzura–Wilczek part evolves via the familiar twist-two generalized
exclusive evolution equation known to two-loop accuracy nowadays [22,23]. The three-
particle piece does not require a new study as well, since it was elaborated in great detail
recently in the context of twist-three functions measured in inclusive reactions (forward
limit restrictions was not made in those studies). Namely, the operators S alluded to above
fall into a class of the so-called quasi-partonic ones [24] and thus at leading order in the
coupling constant their evolution kernel is given by a sum of conventional skewed twist-two
ones with appropriate quantum numbers in subchannels: K q̄gq = Kq̄g + Kgq + Kq̄q . The
issue of its diagonalization (see a recent review [25]) has been addressed in detail recently
[26–30] within the context of integrable open spin chain models which arise in multicolour
limit. 3 Namely, for Nc → ∞ the kernel Kq̄q can be neglected and, e.g., N1c Kq̄g = ψ(Jˆq̄g +
3
2 ) + ψ(Jˆq̄g − 3 ) − 2ψ(1) and 1 Kgq = ψ(Jˆq̄g + 1 ) + ψ(Jˆq̄g − 1 ) − 2ψ(1) for S + , and
2 Nc 2 2
with 32 and 12 being interchanged for S − . Due to conformal invariance at tree level, the
kernels depend on the quadratic Casimir operator of the collinear conformal group SO(2, 1)
2
or, in other words, on the conformal spin in the subchannel only, Jˆ = Jˆ(Jˆ − 1). It turns
out that antiquark–gluon–quark system admits an extra integral of motion QS and is thus
completely integrable. The result of the analysis allows to find the eigenfunctions Ψ and
eigenvalues E of the system and thus solve the evolution equation,

X αs (Q20 ) Nc E{α} /β0

S x1 , x2 , x3 |Q2 = Ψ{α} (x1 , x2 , x3 ) S{α} (Q2
0 ) ,
αs (Q2 )
{α}
X
3
xi = η.
i=1
Here β0 = 43 TF Nf − 11 3 CA is the QCD beta function and a set of quantum numbers {α}
parametrizes solutions and can be chosen as eigenvalues of the conformal spin J and
the charge QS . Namely, the lowest anomalous dimension, E0 (J ), is known exactly [31],
while the rest of the spectrum, Eq (J ), can be described with a high accuracy using WKB
approximation [26–30] with the results
1
E0 (J ) = ψ(J + 3) + ψ(J + 4) − 2ψ(1) − ,
2

3 3
EQ (J ) = 2 ln J − 4ψ(1) + 2 Re ψ + iηS − ,
2 2
p
where ηS is related to the conserved charge by the relation ηS ≡ 12 2QS /J 2 − 3 and obeys
a WKB quantization condition which, once solved, gives quantized values of EQ .
3 The latter gives a good approximation since the corrections to the leading result are suppressed in 1/N 2 .
c
For practical purposes, one may also generalize the polynomial reconstruction method
as it was applied for the forward kinematics in Ref. [32–34] or use a brute force numerical
integration to solve the evolution equations.
7. Conclusions
In this paper we have studied twist-three effects for the two-photon amplitudes in the
generalized Bjorken kinematics. We have demonstrated explicitly the restoration of the
electromagnetic gauge invariance for a Compton-type amplitude to twist-four accuracy.
We have given as well the exact Lorentz structure of the amplitude motivated by our results
at leading order in the coupling constant. Obviously, beyond leading order, e.g., different
(1)
Lorentz structures in Tµν will be multiplied by independent functions (see Ref. [10]),
analogues of F1 and F2 in the forward scattering. Our analysis demonstrates a necessity of
introduction of twist-three two-particle generalized distributions. The latter can be related
by means of QCD equation of motion and Lorentz invariance to the familiar twist-two
ones and interaction dependent antiquark–gluon–quark correlation functions. The former
ones represent a generalization of the Wandzura–Wilczek type relation, while the last ones
give dynamical twist-three contributions. With the assumption of smallness of dynamical
contributions the Wandzura–Wilczek part with parametrization of leading twist skewed
parton distributions can serve as a model for ‘transverse’ functions. Contributions of these
functions to diverse asymmetries [11] will be considered elsewhere [16].
We thank A. Kirchner for a collaboration at an early stage of the work. A.B. would
like to thank A. Schäfer for the hospitality extended to him at the Institut für Theoretische
Physik, Universität Regensburg. This work was supported by Alexander von Humboldt
Foundation, in part by National Science Foundation, under grant PHY9722101 (A.B.) and
by DFG and BMBF (D.M.).
References
[1] D. Müller, D. Robaschik, B. Geyer, F.-M. Dittes, J. Hořejši, Fortschr. Phys. 42 (1994) 101.
[2] X. Ji, Phys. Rev. D 55 (1997) 7114.
[3] A.V. Radyushkin, Phys. Rev. D 56 (1997) 5524.
[4] M. Diehl, T. Gousset, B. Pire, O.V. Teryaev, Phys. Rev. Lett. 81 (1998) 1782.
[5] J.C. Collins, A. Freund, Phys. Rev. D 59 (1999) 074009.
[6] X. Ji, J. Osborne, Phys. Rev. D 58 (1998) 094018.
[7] A. Freund, Phys. Rev. D 61 (2000) 074010.
[8] C. Itzykson, J. Zuber, Quantum Field Theory, McGraw-Hill, New York, 1980.
[9] I.V. Anikin, B. Pire, O.V. Teryaev, On the gauge invariance of the DVCS amplitude, hep-
ph/0003203.
[10] A.V. Belitsky, A. Schäfer, Nucl. Phys. B 527 (1998) 235.
[11] A.V. Belitsky, D. Müller, L. Niedermeier, A. Schäfer, Leading twist asymmetries in deeply
virtual Compton scattering, hep-ph/0004059.
[12] I.I. Balitsky, V.M. Braun, Nucl. Phys. B 311 (1989) 541.
[13] P. Ball, V.M. Braun, Y. Koike, K. Tanaka, Nucl. Phys. B 529 (1998) 323.
[14] B. Geyer, M. Lazar, D. Robaschik, Nucl. Phys. B 559 (1999) 339.

[15] B. Geyer, M. Lazar, Twist decomposition of nonlocal light cone operators. 2. General tensors
of 2nd rank, hep-th/0003080.
[16] A.V. Belitsky, A. Kirchner, D. Müller, A. Schäfer, in preparation.
[17] M. Diehl, T. Gousset, B. Pire, J.P. Ralston, Phys. Lett. B 411 (1997) 193.
[18] E143 Collaboration, K. Abe et al., Phys. Rev. Lett. 76 (1996) 587.
[19] E155 Collaboration, P.L. Anthony et al., Phys. Lett. B 458 (1999) 529.
[20] M. Göckeler, P. Rakow, private communication.
[21] M. Göckeler, R. Horsley, W. Kürzinger, H. Oerlich, P. Rakow, G. Schierholz, The polarized
structure function g2 : a lattice study revisited, hep-ph/9909253; in: Polarized Protons at High
Energies — Accelator Challanges and Physics Opportunities.
[23] A.V. Belitsky, A. Freund, D. Müller, Nucl. Phys. B 574 (2000) 347.
[24] A.P. Bukhvostov, G.V. Frolov, L.N. Lipatov, E.A. Kuraev, Nucl. Phys. B 258 (1985) 601.
[25] A.V. Belitsky, Integrability of twist-three evolution equations in QCD, hep-ph/0007013.
[26] V.M. Braun, S.E. Derkachov, A.N. Manashov, Phys. Rev. Lett. 81 (1998) 2020.
[27] A.V. Belitsky, Phys. Lett. B 453 (1999) 59.
[28] A.V. Belitsky, Nucl. Phys. B 558 (1999) 259.
[29] A.V. Belitsky, Nucl. Phys. B 574 (2000) 407.
[30] S.E. Derkachov, G.P. Korchemsky, A.N. Manashov, Nucl. Phys. B 566 (2000) 203.
[31] A. Ali, V.M. Braun, G. Hiller, Phys. Lett. B 266 (1991) 117.
[32] B. Geyer, D. Müller, D. Robaschik, The evolution of the nonsinglet twist-3 parton distribution
function, hep-ph/9611452.
[34] D. Müller, Phys. Lett. B 407 (1997) 314.
Global obstructions to gauge-invariance

in chiral gauge theory on the lattice
David H. Adams
Math. dept. and Centre for the Subatomic Structure of Matter, University of Adelaide,
Adelaide, S.A. 5005, Australia
Received 25 April 2000; accepted 15 August 2000
Abstract
It is shown that certain global obstructions to gauge-invariance in chiral gauge theory, described
in the continuum by Alvarez-Gaumé and Ginsparg, are exactly reproduced on the lattice in the
Overlap formulation at small non-zero lattice spacing (i.e., close to the classical continuum limit).
As a consequence, the continuum anomaly cancellation condition dR abc = 0 is seen to be a necessary
(although not necessarily sufficient) condition for anomaly cancellation on the lattice in the Overlap
formulation.  2000 Elsevier Science B.V. All rights reserved.
PACS: 11.15.Ha; 11.30.Rd; 02.40.-k
1. Introduction
The Overlap formalism [1–4] provides a potential solution to the important problem of
constructing chiral gauge theories nonperturbatively on the lattice. Gauge anomalies are
a central issue in this context: one would like to show that a gauge-invariant formulation
of chiral gauge theories on the lattice is possible when the usual (continuum) anomaly
cancellation conditions are satisfied. Conversely, when these conditions are not satisfied,
one would like to see the continuum anomalies emerge in the lattice formulation. In
particular, an interesting test for a lattice formulation of chiral gauge theory is whether
it can capture the global obstructions to gauge-invariance of the continuum theory, which
reflect the topological structure of the determinant line bundle over the gauge orbit space
[5–8].
In this paper we show that the overlap reproduces a basic class of such obstructions,
described in the continuum by Alvarez-Gaumé and Ginsparg [5]. 1 This is a further
E-mail address: dadams@maths.adelaide.edu.au (D.H. Adams).

1 The possibility that this could happen had been previously mentioned in [4].
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 2 4 - 1
634 D.H. Adams / Nuclear Physics B 589 (2000) 633–656
demonstration of the ability of the overlap to reproduce topological features of the

continuum theory. 2 As a consequence we will see that the usual (i.e., continuum) local
anomaly cancellation condition ((1.4) below) is a necessary condition for local anomaly-
free chiral gauge theory on the lattice in the overlap formulation.
Global obstructions to the vanishing of local gauge anomalies in the overlap formulation
were previously considered by Neuberger in [9]. An analogue of the the geometric
perspective on the continuum chiral determinant as a section in a determinant line bundle
was described for the overlap, and a class of global obstructions (which are naturally
described in this setting) was explicitly constructed for the abelian theory. These were
seen to vanish precisely when the fermion content of the theory satisfies
X
eα3 = 0, (1.1)
α
where the eα ’s label the irreducible U (1) representations of the fermion species. This
is precisely the condition for cancellation of local gauge anomalies in the continuum
theory. Thus, in the abelian case, the continuum anomaly cancellation condition (1.1) is
a necessary condition for gauge-invariance of the lattice chiral determinant in the overlap
formulation. 3 Our result is a nonabelian variant of this. 4
In the continuum theory, the global obstructions of Alvarez-Gaumé and Ginsparg arise
as follows. Take spacetime to be the Euclidean 4-torus T 4 (the choice of dimension 4 is for
concreteness; everything generalises to the T 2n case for arbitrary n), gauge group SU(N),
and consider a family φθ of gauge transformations parameterised by θ ∈ S 1 . If the fermion
is in the fundamental representation then each φθ is a map from T 4 to SU(N), and the
family of these corresponds to a map Φ : T 5 → SU(N) , Φ(θ, x) = φθ (x). The action of
φθ on a gauge field A determines a family {Aθ }θ∈S 1 . The winding number of the phase of
the chiral determinant around this circle-family of gauge fields is an obstruction to gauge-
invariance of the chiral determinant (since if the determinant is gauge-invariant then it is
constant around the family {Aθ } and the winding number vanishes). In [5] it was shown
that this winding number equals the degree of the map Φ. Thus the obstruction is non-
vanishing precisely when there exist maps Φ : T 5 → SU(N) with non-vanishing degree
(which happens, e.g., when N = 3). 5
In the general case where the fermion content is specified by some arbitrary (typically
reducible) representation R of SU(N), the preceding generalises as follows. Instead of the
degree of Φ , which is given by an expression of the form d abc habc where
2 The parity-invariant overlap formulation of vector gauge theory in odd dimensions reproduces the global
gauge anomaly of the continuum theory [10,11], while for chiral gauge theory in even dimensions it has been
shown to reproduce Witten’s global anomaly [12,13].
3 Herbert Neuberger has pointed out to me that these obstructions also arise for nonabelian gauge groups with
U (1) subgroups, and that in four dimensions there are no additional restrictions in the nonabelian case, i.e., if all
U (1) subgroups are free from anomalies then so is the nonabelian group.
4 Although the class of obstructions that we consider is different: the ones in [9] involve a torus in the orbit
space of lattice gauge fields whereas ours involve a 2-sphere.
5 In [5] the spacetime was S 4 rather than T 4 , and a condition φ ≡ 1 was imposed, which allows Φ to be
0
viewed as a map from S 5 to SU(N ). There is no essential difference with the present case though, since there is
an isomorphism between the homotopy equivalence classes of Map(S k , SU(N )) and Map(T k , SU(N )).
D.H. Adams / Nuclear Physics B 589 (2000) 633–656 635

d abc = 2 tr (T a T b + T b T a )T c (1.2)
and the T a ’s are the generators of SU(N), the obstruction is given by dRabc habc where dRabc
is given by (1.2) with T a replaced by R(T a ) etc. Using the well-known fact that there is a
relation of the form 6
dRabc = c(R)d abc (1.3)
we see that the obstruction in the general case is c(R) times the degree of Φ. Thus in
the case where Map(T 5 , SU(N)) contains maps with non-vanishing degree a necessary
condition for gauge-invariance of the chiral determinant is c(R) = 0 , or
dRabc = 0. (1.4)
Of course, this is just the usual (necessary and sufficient) condition for anomaly
cancellation in the continuum theory (the non-abelian analogue of (1.1)).
In this paper we consider a lattice version of the preceding obstructions in the overlap
formulation, and show that they reduce to the continuum obstructions in the classical
continuum limit. Since the lattice and continuum obstructions are both specified by
integers, it follows that the lattice obstruction is exactly equal to the continuum one at
small non-zero lattice spacing (i.e., close to the classical continuum limit). Our approach
is similar to the recent analytic work of Bär and Campos on the lattice version of
Witten’s global anomaly [13]. When combined with the preceding observation (1.3),
our result implies that (1.4) is a necessary condition for gauge-invariance of the lattice
chiral determinant in the overlap formulation, at least in the case where Map(T 5 , SU(N))
contains maps with non-vanishing degree. 7
It should be emphasised that global obstructions, and hence the results of this paper,
are independent of the choice of phase in the overlap chiral determinant. In contrast,
the consistent gauge anomaly for the overlap chiral determinant does depend on the
phase choice. The consistent anomaly in the overlap formulation, and its classical
continuum limit, has been previously studied in a number of works [14–19], although
these have all involved some form of approximation (e.g., linearisation of the overlap)
and/or assumptions (e.g., weak field, slowly varying field). No such approximations or
assumptions are made in this paper.
The key question which these results lead on to (but which we do not pursue in this
paper) is whether (1.4) is a sufficient condition for existence of a local anomaly-free lattice
chiral gauge theory at non-zero lattice spacing. This is currently a topic of major interest
6 The existence of a relation of this form can be seen as follows. Since the representation ring of SU(N ) is
generated by the fundamental representation and its complex conjugate, it suffices to show (1.3) in the case
where R is a tensor product of copies of the fundamental representation. Then R(T a ) = T a ⊗ 1 ⊗ · · · ⊗ 1
+ 1 ⊗ T a ⊗ 1 ⊗ · · · ⊗ 1 + · · · + 1 ⊗ · · · ⊗ 1 ⊗ T a etc, and it follows that (R(T a )R(T b ) + R(T b )R(T a ))R(T c ) =
(T a T b + T b T a )T c ⊗ 1 ⊗ · · · ⊗ 1 + · · · + 1 ⊗ · · · ⊗ 1 ⊗ (T a T b + T b T a )T c + terms which have a single T a , T b ,
or T c in one of the tensor slots. Since tr(T a ) = 0 etc, it follows that the trace of these latter terms vanishes and
we get (1.3).
7 In higher dimensions there are cases where the maps all have vanishing degree yet the anomaly coefficient
d a1 ···an (the symmetrised trace of T a1 , . . . , T an ) is non-vanishing. E.g., in dimension 2n = 6 with gauge group
SU(3) one has π7 SU(3) = 0 and d a1 a2 a3 a4 6= 0 , cf. p. 472 of [5].
and activity [9,19–22]. To proceed with this in practice, a specific phase choice must be
made to begin with. (The standard choice is the so-called Brillouin–Wigner phase [4].) One
can then try to “improve” the starting phase choice in various ways to get a local anomaly-
free overlap when the cancellation condition holds. One practical approach is to average
along the gauge orbits (i.e., the FNN mechanism); see [9] and the references therein. 8
Another, more theoretical, approach is to reduce the problem of going from an arbitrary
starting phase choice to one for which the (local) anomalies vanish to that of solving a
system of finite-difference equations on the lattice [20,21]. 9 In fact the integrability of
these equations has been proved in the abelian case [20]. 10 There are strong indications
that the same can be done in the nonabelian case [19,21], although a complete proof of
this has not yet been given. Quite recently, a practically-oriented analytic prescription for
constructing anomaly-free non-compact chiral U (1) gauge theory on the lattice has been
given, starting from a adiabatic phase choice for the overlap [22].
In Section 2 we review the overlap construction of the chiral determinant. In Section 3
the lattice version of the global obstruction of Alvarez-Gaumé and Ginsparg is described,
and is shown to reduce to the continuum obstruction in the classical continuum limit.
The derivation of a key formula used to establish this is given separately in Section 4.
This formula ((3.18) below) is due to Lüscher [21], and our detailed explicit derivation in
Section 4 is intended to complement the rather brief argument in [21]. In Section 5 we
make some concluding remarks. Some details of our calculations are given in an appendix.
2. Overlap construction of the chiral determinant on the lattice
The spacetime is taken to be the Euclidean 4-torus T 4 with fixed edge length L. (Again,
the choice of dimension 4 is for concreteness and simplicity; the arguments and results
in the following generalise straightforwardly to the T 2n case for arbitrary n.) We consider
hyper-cubic lattices on T 4 with 2N sites along each edge and lattice spacing a = L/2N . 11
Given such a lattice, the space of lattice spinor fields ψ(x) is denoted C, and the space
of lattice gauge fields Uµ (x) is denoted U . The space C is finite-dimensional and comes
equipped with an inner product:
X
hψ1 , ψ2 i = a 4 ψ1 (x)∗ ψ2 (x), (2.1)
x
8 The viability of this approach has been a topic of debate in the literature [23–25], although there is a body of
evidence which is supportive of it — see, e.g., [26–30].
9 The formulation of Refs. [20,21] (a functional integral formulation based on a lattice Dirac operator satisfying
the Ginsparg–Wilson relation [31], which had been rediscovered outside of the overlap setting in the work of
Hasenfratz and collaborators [32–36]) is structurally identical to the overlap formulation after identifying the
chiral fermion measures in the functional integral with the many-body groundstates in the overlap. More on this
in Section 2, where the many-body groundstates are the “unit volume elements” in our terminology.
10 The argument in [20] relied on a result on the structure of the abelian axial anomaly [37], which has been
further elucidated in [38–42].
11 This N is of course not related to the N in SU(N ).
With suitable boundary conditions, the covariant forward and backward finite difference
operators a1 ∇µ± act on C by
∇µ+ ψ(x) = Uµ (x)ψ(x + aeµ ) − ψ(x), (2.2)

∇µ− ψ(x) = ψ(x) − Uµ (x − aeµ )−1 ψ(x − aeµ ), (2.3)
eµ denotes the unit vector in the positive µ-direction. We restrict to the case where Uµ (x)
and ψ(x) are periodic. This is the relevant case for considering the classical continuum
limit with topologically trivial gauge fields. Since the chiral determinant vanishes in the
topologically non-trivial case, this suffices for our purposes. Set ∇µ = 12 (∇µ+ + ∇µ− ); this
operator is anti-hermitian with respect to the inner product (2.1) since (∇µ± )∗ = −∇µ∓ . The
Wilson–Dirac operator is now given by

1 r 1
DWilson = ∇ / + a 2∆ , (2.4)
a 2 a
P
where ∇/ = µ γ µ ∇µ (the γ µ ’s are taken to be hermitian so ∇ / is anti-hermitian),
P P P
∆ = µ ∇µ− + ∇µ+ = µ (∇µ+ )∗ ∇µ+ = µ (∇µ− )∗ ∇µ− (hermitian, positive) and r > 0 is the
Wilson parameter. The hermitian operator

H (m) = γ5 (aDWilson − rm) = γ5 ∇ / + r 12 ∆ − m (2.5)
determines an orthogonal decomposition
(m,U ) (m,U )
C = C+ ⊕ C− , (2.6)
(m,U ) (m,U )
where C+ and C− are the subspaces spanned by the eigenvectors of H (m) with
positive and negative eigenvalues, respectively. (We are restricting to the m, U for which
(m,U )
H (m) has no zero-modes.) These subspaces are characterised by (m) = ±1 on C±
where
H (m)
(m) = p (2.7)
H (m)2
(the dependence on U has been suppressed). Noting that
|rm| H (0) − |m| γ5

1 m
(m) = q 2 → γ5 , for m → −∞ (2.8)
|rm| H (0) − |m| γ5
1 m
we see that in the m → −∞ limit (2.6) reduces to the usual chiral decomposition
C = C+ ⊕ C− (2.9)
independent of U . Set m = 1 (the canonical value; 0 < m < 2 would suffice) and let v±
(1,U )
and w± (U ) be unit volume elements 12 on C± and C± , respectively; these are unique up
12 A vectorspace V determines vectorspaces Λp V (p = 1, . . . , dim V ): the exterior algebra (= Grassmann

algebra) of V of degree p. An inner product in V induces an inner product in each Λp V . A “unit volume
element on V ” is an element v ∈ Λd V (d = dim V ) with |v| = 1. E.g., if v1 , . . . , vd is an orthonormal basis for
V then v1 ∧ · · · ∧ vd is a unit volume element. Since Λd V is 1-dimensional, a unit volume element is unique up
to ± if V is real, or up to a phase if V is complex.
to phase factors. Then the lattice versions of the right- and left-handed chiral determinants
in the overlap construction are, respectively,
hv+ , w+ (U )i (right-handed), (2.10)

hv− , w− (U )i (left-handed) (2.11)
(see [4] for background and motivation). The w± (U ) are required to depend smoothly on
U ; then the overlaps (2.10)–(2.11) are smooth in U . Note that a condition for non-vanishing
(1,U )
overlaps is dim C± = dim C± ≡ d. The overlaps are unique up to a phase factor, and their
norms are gauge-invariant (an easy consequence of the gauge-covariance of H (m), (m)).
Remark 2.1. The construction of the overlaps (2.10)–(2.11) requires that H (1) has no
zero-modes. This can be guaranteed by imposing the condition [43,44]
||1 − U (p)|| 6 0.04, ∀p (2.12)
on the lattice gauge field U , where U (p) is the product of the link variables around a
plaquette p. This condition is automatically satisfied in the classical continuum limit since
1 − U (px;µ,ν ) = a 2Fµν (x) + O(a 3 ).
We henceforth restrict U to be the space of lattice gauge fields satisfying (2.12).
Remark 2.2. The overlaps (2.10)–(2.11) are determined (up to a phase) solely by = (1).
The construction could be carried through given any hermitian operator with the property
2 = 1. The norms of the resulting overlaps would be gauge-invariant provided is gauge-
covariant.
Remark 2.3. The overlaps (2.10)–(2.11) can be written as 13

b + i ≡ det DL ,
( a2 )d hv+ , w+ i = hv+ , Dw (2.13)
b − i ≡ det DR ,
( 2 )d hv− , w− i = hv− , Dw (2.14)
a
where
1
D = (1 + γ5 ) (2.15)
a
and = (1) is given by (2.7). This follows easily from the facts that (1 + γ5 )w =
(1,U )
(1 ± γ5 )w for w ∈ C± and (1 ± γ5 )v = 2v for v ∈ C± . The relations (2.13)–(2.14)
show how the overlaps can be viewed as chiral determinants in an analogous way to the
continuum setting: Set γ̂5 = γ5 (1 − aD) = −, then γ̂52 = 1 and D γ̂5 = −γ5 D, which
b∓ := C±
implies that D maps C
(1,U )
to C± . Thus, modulo the factors (2/a)d , the right-handed
overlap can be viewed as a left-handed chiral determinant, and vice versa, as indicated in
b : Λp W → Λp V for all p,
13 We are using the fact that a linear operator D : W → V induces linear operators D
b 1 ∧ · · · ∧ wp ) = Dw1 ∧ · · · ∧ Dwp . Note that if W = V and d = dim V then D(w
defined by D(w b 1 ∧ · · · ∧ wd ) =
det D · w1 ∧ · · · ∧ wd .
(2.13)–(2.14). 14 In this form the overlap arises as the chiral determinant in Lüscher’s
formulation [20,21] after identifying the unit volume elements v± and w± (the many-
body groundstates in the overlap) with the chiral fermion measures. These observations
have been pointed out previously in [45–47]. As mentioned there, it is easy to see that an
operator D is of the form (2.15), with hermitian and 2 = 1, if and only if it satisfies the
the following two conditions:
Dγ5 + γ5 D = aDγ5 D, (Ginsparg–Wilson relation [31]), (2.16)
∗
D = γ5 Dγ5 , (γ5 -hermiticity). (2.17)
Also, clearly D is gauge-covariant if and only if is gauge-covariant. The operator (2.15),
with = (1) given by (2.7), is the Overlap Dirac operator introduced by Neuberger in
[48,49]. It should also be mentioned that the Ginsparg–Wilson relation was rediscovered
outside of the overlap setting in the work of Hasenfratz and collaborators — they
considered a different solution, the so-called perfect Dirac operator [32–36].
The nullspace of D is invariant under γ5 (this follows from the GW relation (2.16):
Dψ = 0 ⇒ D(γ5 ψ) = (aDγ5 D − γ5 D)ψ = 0) so index D ≡ Tr(γ5 |ker D ) is well-defined,
as was first noted in [36]. We only need to consider the lattice gauge fields U for which
(1,U )
dim C± = dim C± ≡ d, since the overlaps vanish otherwise. As noted in [21], this
corresponds to having index D = 0. Therefore, we henceforth take U to be the space of
lattice gauge fields satisfying (2.12) and index D = 0.
3. Global obstructions to gauge-invariance of the overlap
From now on we consider only the right-handed overlap hv+ , w+ (U )i (the situation
for the left-handed overlap is analogous). A lattice version of the obstructions considered
by Alvarez-Gaumé and Ginsparg is as follows. Let φθ be a family of lattice gauge
transformations parameterised by θ ∈ S 1 . We can assume that the fermion content is
specified by the fundamental representation of SU(N); it will be clear from what follows
that the general case is related to this case in the same way as in the continuum setting
discussed in the introduction. If U ∈ U is a lattice gauge field for which the overlap
hv+ , w+ (U )i is non-vanishing 15 then the action of φθ on U determines a map
S 1 → C − {0}, θ 7→ hv+ , w+ (φθ · U )i. (3.1)
Since |hv+ , w+ (U )i| is gauge-invariant, we have hv+ , w+ (φθ · U )i = eiα(θ) hv+ , w+ (U )i
for some phase α(θ ), and the map (3.1) has integer winding number W (Φ, U ) =
14 A careful consideration of the overlap prescription shows that the overlaps hv , w i really should be
± ±
multiplied by a factor (2/a)d as in (2.13)–(2.14). These factors are physically irrelevant though: they appear
both in the numerator and denominator in expressions for physical expectation values, and hence cancel out,
and they do not affect anomalies since these only have to do with the phase of the overlaps. Nevertheless, they
are relevant if one considers the chiral determinant on its own and wishes to use the lattice regularisation as an
alternative to, e.g., zeta-regularisation.
15 It can be seen from (2.13) that hv , w (U )i vanishes at the U for which D has zero-modes. Generically,
+ +
these are isolated points in U .
2π (α(1) − α(0)). Obviously, if the winding number is non-vanishing hv+ , w+ (U )i cannot

1
be gauge-invariant. To see that this is a genuine obstruction to gauge-invariance we note

e+ (U ) is another unit volume element
that it is independent of the choice of w+ (U ): if w
(1,U )
on C+ , smoothly varying with U , then w e+ (U ) = eiβ(U ) w+ (U ) where the phase factor
e iβ(U ) is smooth in U , and we have
e+ (φθ · U )i = eiα(θ)+iβ(φθ ·U ) hv+ , w+ (U )i.

hv+ , w
Assuming that {φθ · U }θ∈S 1 is a contractible circle in U (which is certainly true close to the
classical continuum limit), it follows that this has the same winding number as (3.1) since
since eiβ(U ) is a smooth, non-vanishing, globally defined function of U . Hence the winding
number W (Φ, U ) is an obstruction to gauge-invariance of the overlap, independent of the
choice of w+ .
Our main result is that this obstruction coincides with the continuum one at small non-
zero lattice spacing close to the classical continuum limit:
Theorem. If φθ is the restriction to the lattice of a family of continuum gauge

transformations (also denoted φθ ) and U is the lattice transcript of a topologically trivial
continuum gauge field, then there is an a0 > 0 (depending on the φθ ’s and U ) such that
W (Φ, U ) = deg(Φ), for all a < a0 , (3.2)

where deg(Φ) is the degree of the continuum map Φ : T 5 → SU(N) given by Φ(θ, x) =
φθ (x).
In light of the discussion in the introduction we conclude from this that, in the general
case where the fermion content is specified by a general representation R of SU(N),
a necessary condition for existence of a gauge-invariant construction of the overlap is
dRabc = 0. (3.3)
The remainder of the paper is concerned with the proof of the above theorem. We start
by expressing the obstruction as
Z1
1
W (Φ, U ) = d
dθ dθ loghv+ , w+ (φθ · U )i
2πi
0
Z
1
= d loghv+ , w+ i, (3.4)
2πi
S1
where S 1 denotes the circle {φθ · U }θ∈S 1 in U and d denotes the exterior derivative on
functions (or more generally, differential forms) on U . After noting that
dw+ = hw+ , dw+ iw+ + (dw+ )⊥ , (3.5)

where (dw+ )⊥ denotes the projection of dw+ onto the orthogonal complement of w+ in
Λd C, one finds [9]
hv+ , (dw+ )⊥ i
d loghv+ , w+ i = + hw+ , dw+ i. (3.6)
hv+ , w+ i
The first term on the right-hand side of (3.6) can be re-expressed as
hv+ , (dw+ )⊥ i
= Tr dDD −1 P+ , (3.7)
hv+ , w+ i
where P+ = 12 (1 + γ5 ) is the projection onto C+ . This is a straightforward consequence
of (2.13) and relations noted in [20,21]; for completeness we provide a derivation in the
appendix. Then

d loghw+ , w+ i = Tr dDD −1 P+ + hw+ , dw+ i. (3.8)
Set w+θ = w (φ · U ) and let D denote D with lattice gauge field φ · U . The gauge-
+ θ θ θ
θ −1
d
covariance of D gives dθ Dθ = [ dφ φ
dθ θ , Dθ ], leading to [20,21]

d dφθ −1 dφθ −1 −1
Tr Dθ Dθ−1 P+ = Tr φθ P+ − Tr Dθ φθ Dθ P+
dθ dθ dθ

dφθ −1
= Tr φ P+ − Pb−
dθ θ

1 dφθ −1 1 dφθ
= − a Tr φθ γ5 Dθ = − a Tr φθ−1 γ5 D1 , (3.9)
2 dθ 2 dθ
where we have used the fact that P+ D = D Pb− where Pb− = 12 (1 − γˆ5 ), γˆ5 = γ5 (1 − aD).
Substituting (3.8) into (3.4) and using (3.9) we get
Z

2πiW (Φ, U ) = Tr dDD −1 P+ + hw+ , dw+ i (3.10)
S1
Z1 θ
1 −1 dφθ θ dw+
= dθ − a Tr φθ γ5 D1 + w+ , . (3.11)
2 dθ dθ
0
We have derived this relation under the assumption that the overlap is non-vanishing for
U , or equivalently, that the Overlap Dirac operator D with lattice gauge field U has no
zero-modes. By construction W (Φ, U ) is clearly smooth, and therefore locally constant,
in such U . But it is ill-defined at the (generically isolated) points in U where the overlap
vanishes. One such point is the trivial field U = 1 (in this case the zero-momentum spinors
with definite chirality are zero-modes for D). However, the right-hand side of (3.11) is
clearly smooth in U for all U ∈ U (since D is smooth in the lattice gauge field when (2.12)
is satisfied [43]), and must therefore be a locally constant function of U for all U ∈ U .
In the continuum, any topologically trivial gauge field can be continuously deformed to
the trivial field. It follows that when the lattice spacing a is sufficiently small, the lattice
transcript U can be continuously deformed to the trivial lattice gauge field (using the lattice
transcript of the continuum path). Therefore, to prove the theorem it suffices to show that
there is an a0 > 0 such that
W (Φ) = deg(Φ), for all a < a0 , (3.12)
where W (Φ) denotes the right-hand side of (3.11) with trivial field U = 1. In this case D1
acts trivially in colour space, and since φθ−1 dφ θ
dθ is in the Lie algebra of SU(N) the trace
over colour indices in the first term in (3.11) vanishes, resulting in
Z1

θ d θ
2πiW (Φ) = dθ w+ , dθ w+
0
Z
= hw+ , dw+ i. (3.13)
S1
A calculation gives
dhw+ , dw+ i = hdw+ , dw+ i
= Tr(P dP dP ), (3.14)
where
P ≡ Pb− = 12 (1 − γˆ5 ) = 12 (1 + ) (3.15)
with = (1) given by (2.7). The last equality in (3.14) is derived in the appendix.
A simpler version of it (originating in [50]) was used in [9]; the same relation in a different
guise was subsequently noted in [20,21], and in more detail in [18]. Now, by Stokes
theorem, if B 2 is a disc in U with boundary S 1 it follows that
Z
2πiW (Φ) = Tr(P dP dP )
B2
Z
= Tr(P [∂θ P , ∂t P ]) dθ dt, (3.16)
B2
where (θ, t) are taken to be polar coordinates on a unit disc B 2 parameterising B 2 . We take
B 2 to be the lattice transcript of a disc-family of continuum gauge fields, also denoted B 2 ,
given by
A(θ,t ) = f (t)φθ dx φθ−1 , (θ, t) ∈ B 2 , (3.17)
where f (t) is an arbitrary smooth function equal to 1 in a neighbourhood of t = 1 and
vanishing in a neighbourhood of t = 0. The lattice transcript U (θ,t ) has the property
U (θ,1) = φθ · 1 so the boundary of B 2 is S 1 as required in (3.16). Note that (3.16) is
manifestly independent of the choice of w+ (U ), i.e., independent of the choice of phase in
the overlap.
A general formula for the classical continuum limit of the integrand in (3.16) has been
given by Lüscher in [21]: if U (s,t ) is the lattice transcript of a family A(s,t ) of continuum
gauge fields, and P = P (s,t ) is the corresponding family of projection operators (given by
(3.15)), then
Z
−1
lim Tr(P [∂s P , ∂t P ]) = d 4 x µνρσ d abc ∂s Aaµ (x)∂t Abν (x)Fρσ
c
(x). (3.18)
a→0 32π 2
T4
Using this, (3.12) follows easily from (3.16) as we will see below, thereby proving the
theorem. In [21] the locality, smoothness and symmetry properties of P were used to show
that the limit in (3.18) exists and is given by the integral over T 4 of a polynomial in the
gauge fields and its derivatives. However, the explicit form of this polynomial (i.e., the
integrand on the right-hand side of (3.18)) was not obvious, at least to the present author,
from the brief argument in [21]. Since (3.18) is an important formula in this context (it
was also a key ingredient in the arguments of Refs. [21] and [13]) we will give a detailed,
explicit derivation of it in Section 4. This is intended to complement the brief argument in
[21].
By (3.18), the classical continuum limit of (3.16) is
Z
−1
lim 2πiW (Φ) = dθ dt d 4 x µνρσ d abc ∂θ Aaµ (x)∂t Abν (x)Fρσ
c
(x) (3.19)
a→0 32π 2
B2 ×T 4
with A = A(θ,t ) given by (3.17). It remains to show that this is equal to 2πi deg(Φ). Then,
since W (Φ) is integer, we can conclude that W (Φ) = deg(Φ) for all lattice spacings a
smaller than some a0 > 0 and the theorem is proved. The right-hand side of (3.18) can be
shown to equal 2πi deg(Φ) by a direct calculation [13], but it is easier and perhaps more
illuminating to proceed indirectly as follows. We view the family A(θ,t ) as a gauge field on
B2 × T 4:
−1
µ (x)dx = f (t)φθ (x)∂µ φθ (x) dx
A(x, θ, t) = A(θ,t ) µ µ
(3.20)
e2 × T 4 by
and define another gauge field on B
e θ, s) = −f (s)∂θ φ −1 (x)φθ (x) dθ,
A(x, (3.21)
θ
where Be2 is another copy of the unit disc, with polar coordinates (θ, s). B 2 × T 4 and
e × T 4 can be glued together along their common boundary S 1 × T 4 to get the closed
B 2
e are related by a
manifold S 2 × T 4 . On the common boundary S 1 × T 4 the fields A and A
gauge transformation:
e = Φ −1 · A = Φ −1 AΦ + Φ −1 (dθ + dx )Φ,
A
where Φ : S 1 × T 4 → SU(N) is given by Φ(θ, x) = φθ (x). Therefore A and A e constitute
b
a gauge field A on an SU(N) bundle over S × T with topological charge − deg(Φ). The
2 4
topological charge is also given by the integral of the Chern character over S 2 × T 4 , thus
Z
i3 b3
− deg(Φ) = tr F
(2π)3 3!
S 2 ×T 4
" Z Z #
i3 e
= tr F −
3
tr F .
3
(3.22)
(2π)3 3!
B 2 ×T 4 e2 ×T 4
B
The second term vanishes: F e3 = 0 since Ae only involves the 1-form dθ . Regarding the
first term, from F = (dx + dθ + dt )A + A ∧ A we get
tr F 3 = tr(dθ Adt AF + dt Adθ AF + dθ AF dt A

+ dt AF dθ A + F dθ Adt A + F dt Adθ A)
abc
d
=3 dθ Aa dt Ab F c .
2
After substituting this for tr F 3 in (3.22) we see from (3.19) that lima→0 W (Φ) = deg(Φ)
as required.
4. A detailed derivation of Lüscher’s formula
In this section we give a detailed, explicit derivation of the formula (3.18). We begin by
noting that
1
Tr(P [∂s P , ∂t P ]) = Tr([∂s , ∂t ])
8
1
= Tr(∂s ∂t ). (4.1)
4
Setting H = H (1) we have
H γ5 X
=√ =√ , X = aγ5 H = ∇
/ +r 2∆ − 1
1
(4.2)
H 2 X∗ X
and a calculation using (4.1) gives
Tr(P [∂s P , ∂t P ])

1 γ5 X 1 1
= Tr √ ∂s (γ5 X) √ ∂t (γ5 X) √ (4.3)
4 X∗ X X∗ X X∗ X

1 1 1 1 √
+ Tr ∂s √ ∂t (γ5 X) + Tr ∂ (γ X)∂ X ∗X (4.4)
s 5 t
4 X∗ X 4 X∗ X

1 1 √ 1
− Tr ∂s √ X∗ Xγ5 X∂t √ . (4.5)
4 X X∗ X∗ X
We will see below that (4.4) ∼ O(a) and (4.5) = Symm + O(a) where Symm is
symmetric under interchange of ∂s and ∂t . Since Tr(P [∂s P , ∂t P ]) is antisymmetric under
this interchange it follows that

1 γ5 X 1 1
Tr(P [∂s P , ∂t P ]) = Tr √ ∂s (γ5 X) √ ∂t (γ5 X) √
8 X∗ X X∗ X X∗ X

γ5 X 1 1
− Tr √ ∂t (γ5 X) √ ∂s (γ5 X) √ + O(a). (4.6)
X∗ X X∗ X X∗ X
To prove these statements, and evaluate the a → 0 limit of (4.6), we use the fact that the
lattice transcript 16
16 Here A = A(s,t) and U = U (s,t) depend smoothly on the two parameters (s, t).
Z1 !

Uµ (x) = T exp aAµ x + (1 − τ )aeµ dτ (4.7)
0
can be expanded in powers of a as
∞
X Z
Uµ (x) = an dτn · · · dτ1 Aµ (x, τn ) · · · Aµ (x, τ1 ), (4.8)
n=0 0<τ1 <···<τn <1
where Aµ (x, τ ) = Aµ (x + (1 − τ )aeµ ). Since Aµ (x) is a smooth function on the closed

manifold T 4 there is a finite K such that |Aµ (x)| < K for all x, µ. Then the norm of the
integral in the nth term of (4.8) is bounded by n!1 K n , so (4.8) is norm-convergent for all a.
The inverse Uµ (x)−1 also has an expansion in powers of a: using the fact that Uµ (x)−1 is
the parallel transport from x to x + aeµ specified by A we see that Uµ (x)−1 is given by
the right-hand side of (4.8) with Aµ (x, τ ) = Aµ (x + τ aeµ ).
P
Substituting these expansions in (2.2)–(2.3) gives an expansion ∇µ± = ∞ n ±
n=0 a (∇µ )n .
P∞ n
This in turn gives an expansion X = n=0 a Xn . It is not difficult to show that the
||(∇µ± )n ||’s and the ||Xn ||’s have a finite bound K 0 independent of a and n, so the
expansions are norm-convergent
√ when a is sufficiently small.
To expand 1/ X∗ X we note that
X∗ X = L + V , (4.9)
where
2
L = −∇ 2 + r 2 2∆ − m
1
(4.10)
V =V +V ,
(1) (2)
(4.11)
X
1 µ + − 1
V = rγ ∇µ ,
(1)
∇ν − ∇ν , V (2) = − γ µ , γ ν [∇µ , ∇ν ]. (4.12)
2 ν
4
P
Just as for X, the expansion (4.8) leads to expansions X∗ X = ∞ ∗
n=0 a (X X)n , L =
n
P∞ n P∞ n ∗
n=0 a Ln , V = n=0 a Vn where the ||(X X)n ||’s, ||Ln ||’s and ||Vn ||’s again have a
finite bound independent of a and n. Furthermore, explicit calculations show that
+ +
∇µ , ∇ν ψ(x) = a 2 Fµν (x) + O a 3 ψ x + aeµ + aeν , (4.13)
+ −
∇µ , ∇ν ψ(x) = a Fµν (x) + O a ψ x + aeµ − aeν ,
2 3
(4.14)
− +
∇µ , ∇ν ψ(x) = a 2 Fµν (x) + O a 3 ψ x − aeµ + aeν , (4.15)
− −
∇µ , ∇ν ψ(x) = a Fµν (x) + O a ψ x − aeµ − aeν .
2 3
(4.16)
It follows that V0 = V1 = 0, i.e., the expansion of V starts with the a 2 term, hence ||V || ∼
O(a 2). The leading term a 2 V2 is explicitly given (mod O(a 3)) by substituting (4.13)–
(4.16) into (4.11)–(4.12). We note from this that V2 = V2b T b where the T b ’s are the
generators of the Lie algebra of SU(N) and the V2b ’s are trivial in colour space.
From (4.8) we also get expansions of ∂s Uµ (x) and ∂s Uµ (x)−1 = Uµ (x)−1 ∂s Uµ (x) ×
P
Uµ (x)−1 in powers of a, leading to an expansion ∂s X = ∞ n
n=0 a (∂s X)n and expansions
∗
of ∂s (X X), ∂s L, and ∂s V . Note that these begin with the order a term, i.e., (∂s X)0 =
(∂s (X∗ X))0 = (∂s L)0 = (∂s V )0 = 0. For later use we also note the following: (1)
The lowest order term in the expansion of ∂s Uµ (x) (or ∂s Uµ (x)−1 ) is aAbµ (x)T b
(or −aAbµ (x)T b ). (2) Applying ∂s to (4.13)–(4.16) results in Fµν (x) → ∂s Fµν (x), so
(∂s V )0 = (∂s V )1 = 0, ||∂s V || ∼ O(a 2), and (∂s V )2 = (∂s V )b2 T b where the (∂s V )b2 ’s are
trivial in colour space.
Note that the γ µ ’s in (4.9) are all contained in V . The hermitian positive operator L
is trivial in Dirac indices and the lowest order term L0 in its expansion is diagonal with
respect to the plane wave basis {eiak·x }; the diagonal elements are
X X 2
L0 (ak) = sin2 (akµ ) + r 2 −1 + 1 − cos(akµ ) . (4.17)
µ µ
From this we see that there is a b > 0 independent of a such that L0 > 2b. Then, by√taking
a to be sufficiently small, we can achieve L > b and ||V || < 12 b, in which case 1/ X∗ X
can be expanded as follows:
Z∞
1 dσ 1
√ = ∗
X∗ X π X X + σ2
−∞
Z∞
dσ 1 1
=
π 1 + (L + σ 2 )−1 V L + σ2
−∞
Z∞ ∞
dσ X −1 k 1
= (−1)k L + σ 2 V . (4.18)
π L + σ2
−∞ k=0
For all p we have

p ∞ k
X k 1

2 −1 X 1 1 1 2
(−1) k
L + σ V < b = .
L + σ2 b 2 b + σ2 b + σ2
k=0 k=0
Since the integral of this over (−∞, ∞) is finite, the integral and sum in (4.18) can be
interchanged, resulting in a norm-convergent expansion in powers of V :
∞ Z∞
X
∗ −1/2 dσ −1 k −1
(X X) = (−1)k L + σ 2 V L + σ 2 . (4.19)
π
k=0 −∞
Since ||V || ∼ O(a 2) the kth term in the sum is ∼ O(a 2k ) and we conclude that
p Z∞
X
∗ −1/2 dσ −1 k −1
(X X) = (−1)k L + σ 2 V L + σ 2 + Rp+1 , (4.20)
π
k=0 −∞
where 1
a 2p
||Rp+1 || → 0 for a → 0. Similarly, we find
p Z∞
∗ −1/2
X dσ −1 k −1
∂s (X X) = (−1)k ∂s L + σ 2 V L + σ 2 + ∂s Rp+1 ,
π
k=0 −∞
(4.21)
where a12p ||∂s Rp+1 || → 0 for a → 0.

The bound ||∇µ± || 6 2 and triangle inequalities lead to an a-independent upper bound
L < b1 . Using this, the operator (L + σ 2 )−1 in (4.20) can be expanded as

1 1 1
= −L
L + σ2 b1 + σ 2 1 − bb1+σ 2
1
∞
X
2 −1
−m
= b1 + σ b1 + σ 2 (b1 − L)m . (4.22)
m=0
P
Substituting the expansion L = ∞ n
n=0 a Ln in (4.22), and then substituting in (4.20) the
P
resulting expansion of (L + σ 2 )−1 , along with the expansion V = ∞ a n V , we get
n
∗ −1/2
P∞ n ∗ −1/2 n=2
an expansion (X X) = n=0 a (X X)n . Similarly, after applying ∂s to (4.22)
and substituting the resulting expansion in (4.21), we get an expansion ∂s (X∗ X)−1/2 =
P∞ n ∗ −1/2 ) (note (∂ (X ∗ X)−1/2 ) = 0). These, together with the expansions
n=1 a (∂s (X X) n s
P 0
of X and ∂X, lead to expansions O = ∞ n=0 a On of the operators in (4.3)–(4.5). It can
n
e
be shown that there is a finite K independent of a and n such that ||On || < K e for these
17
operators. This technical result will be presented elsewhere [51]. This implies that the
aforementioned operator expansions are all norm-convergent when a is sufficiently small.
P
An immediate consequence is the following: in the resulting expressions ∞ n
n=0 a Tr(On )
P∞ n
for (4.3)–(4.5) the part n=5 a Tr(On ) vanishes in the a → 0 limit. To see this, let
{ψj }j =1,...,N be an arbitrary orthonormal basis for C; then

X n X
a Tr(On ) 6 a n |hψj , On ψj i|

n>5 n>5,j
X ∞
X
e
e = a4N K e a
6 anK a n = 4L4 K → 0,
1−a
n>5,j n=0
for a → 0, (4.23)
where we have used the fact that N ≡ dim C = 4(L/a)4 . This shows that we only need to
P
consider the a n terms with n 6 4 in the expansions ∞ n
n=0 a Tr(On ) of (4.3)–(4.5).
To proceed with the determination of (4.3)–(4.5) in the a → 0 limit, we note the
following: (i) Due to the presence of an odd number of γ5 ’s, terms in the expansions
involving a product of less that 4 γ µ ’s vanish. (ii) L, V (1) and V (2) are of order 0, 1
P
and 2, respectively, in the γ µ -matrices, cf. (4.12). (iii) In the expansion O = ∞ n=0 a On
n
±
of any operator O constructed from the ∇µ ’s, the term O0 is independent of the gauge
field, so (∂s O)0 = 0. Hence non-vanishing terms in such expansions are at least O(a) for
a → 0. (iv) As we have seen in (4.23), terms in the operator expansions which are of order
> 5 in a give vanishing contributions in the a → 0 limit.
17 It relies on the fact that A (x) is periodic. The general (i.e., topologically non-trivial) case is more
µ
complicated.
At this point we can derive the postulated formula (4.6). Consider the first trace in (4.4):
after substituting the expansion (4.19) for (X∗ X)−1/2 only the terms with at least two V ’s
are non-vanishing after taking the trace over spinor indices. Since ||V || ∼ O(a 2), such
terms are all of order > 4 in a. Since ∂t (γ5 X) ∼ O(a) it follows that the terms in the
expansion of ∂s (X∗ X)−1/2 ∂t (γ5 X) which are non-vanishing after taking the trace are all
of order > 5 in the a → 0 limit, so (4.4)∼ O(a) as claimed.
We now consider the trace (4.5). First, note from (4.20)–(4.22) that the lowest order term
involving V (or containing γ µ ’s) in the expansion of (X∗ X)−1/2 is
Z∞
dσ −1 −1
− L0 + σ 2 a 2 V2 L0 + σ 2 . (4.24)
π
−∞
Using (4.22) and the fact that [L0 , V ] ∼ O(a) we find that, modulo an O(a) term, this is
Z∞ Z∞
dσ −2 −3/2 dσ 1
−a 2
L0 + σ 2 V2 = −a 2 L0 V2
π π (1 + σ 2 )2
−∞ −∞
1 −3/2
= − a 2 L0 V2 . (4.25)
2
Similarly, the lowest order term containing γ µ ’s in the expansion ∂s (X∗ X)−1/2 is
−3/2
− 12 a 2 L0 (∂s V )2 . To simplify the notation in the following, we write ∂s On for (∂s O)n .
Now, using (i)–(iv) above, we find

Tr ∂s (X∗ X)−1/2 (X∗ X)1/2 γ5 X∂t (X∗ X)−1/2
−1/2 −1/2
= a 4 Tr ∂s (X∗ X)2 (X∗ X)0 γ5 X0 ∂t (X∗ X)2
1/2
+ O(a)
a4 −3/2 1/2 −3/2
= Tr L0 ∂s V2 L0 γ5 X0 L0 ∂t V2 + O(a), (4.26)
4
where we have also used (X∗ X)0 = L0 . We now supplement (i)–(iv) above with the
following observation (v): V2 and (∂V )2 commute with L0 modulo an O(a) term, and
commute with γ5 X0 modulo an O(a) term and a term of order 1 in the γ µ ’s. It follows that,
modulo an O(a) term, (4.26) is symmetric under interchange of ∂s and ∂t as claimed. This
proves the previously stated symmetry property of (4.5), thereby completing the derivation
of (4.6).
Turning now to the traces in (4.6), similar arguments to the preceding give

Tr γ5 X(X∗ X)−1/2 ∂s (γ5 X)(X∗ X)−1/2 ∂t (γ5 X)(X∗ X)−1/2
−1/2 −1/2 −1/2
= a 4 Tr γ5 X0 (X∗ X)2 ∂s (γ5 X)1 (X∗ X)0 ∂t (γ5 X)1 (X∗ X)0
−1/2 −1/2 −1/2
+a 4 Tr γ5 X0 (X∗ X)0 ∂s (γ5 X)1 (X∗ X)2 ∂t (γ5 X)1 (X∗ X)0
−1/2 −1/2 −1/2
+a 4 Tr γ5 X0 (X∗ X)0 ∂s (γ5 X)1 (X∗ X)0 ∂t (γ5 X)1 (X∗ X)2 + O(a)
1 −5/2
= − a 4 t cab + t acb + t abc Tr L0 γ5 X0 ∂s (γ5 X)a1 ∂t (γ5 X)b1 V2c + O(a),
2
(4.27)
where we have set t abc = tr(T a T b T c ). Modulo an O(a) term, (4.27) is antisymmetric
under interchange of ∂s (γ5 X)a1 and ∂t (γ5 X)b1 . To see this, note that ∂s (X∗ X)a1 =
γ5 X0 ∂s (γ5 X)a1 + ∂s (γ5 X)a1 γ5 X0 . Since (X∗ X)1 = L1 does not contain γ µ ’s, it follows that
γ5 X0 ∂s (γ5 X)a1 can be replaced by −∂s (γ5 X)a1 γ5 X0 in (4.27). The claimed antisymmetry
then follows from the cyclicity of the trace after using (v) above. Taking this into account
in (4.6), and noting 12 d abc = t cab + t cba = t acb + t bca = t abc + t bac , we get
3 4 abc −5/2
Tr(P [∂s P , ∂t P ]) = − a d Tr L0 γ5 X0 ∂s (γ5 X)a1 ∂t (γ5 X)b1 V2c + O(a).
32
(4.28)
We calculate

∂s (γ5 X)a1 ∂t (γ5 X)b1 = −γ µ (∂s ∇µ )a1 + 12 r(∂s ∆)a1 γ ν (∂t ∇ν )b1 + 12 r(∂t ∆)b1
eab + a term not involving γ µ ’s,
=V (4.29)
where
eab = V
V e(1)ab + V e(2)ab , (4.30)

e(1)ab = − 1 rγ µ (∂s ∇µ )a (∂t ∆)b − (∂s ∆)a (∂t ∇µ )b ,
V (4.31)
1 1 1 1
2

e(2)ab = − 1 γ µ , γ ν (∂s ∇µ )a (∂t ∇ν )b ,
V (4.32)
1 1
2
It follows from (4.28) and (4.29)–(4.32) that
−3 4 abc −5/2
Tr(P [∂s P , ∂t P ]) = a d Tr L0 γ5 X0 V eab V c + O(a)
2
32
−3 4 abc −5/2
= a d Tr L0 γ5 γ µ ∇µ (V e(1)ab V (2)c + V
e(2)ab V (1)c
32 2 2
1 e(2)ab (2)c
+r ∆V V2 + O(a). (4.33)
2
V (1)ab and V (2)ab can be determined as follows. Recalling ∂s Uµ (x) = a∂s Aµ (x) + O(a 2 )
we get

∂s ∇µ± 1 ψ(x) = ∂s Aµ (x) + O(a) ψ(x ± aeµ ) (4.34)
and calculations give
1
(∂s ∇µ )a1 (∂t ∇ν )b1 ψ(x) = ∂s Aaµ (x)∂t Abν (x) + O(a)
4
ψ(x + aeµ + aeν ) + ψ(x + aeµ − aeν )
× , (4.35)
+ψ(x − aeµ + aeν ) + ψ(x − aeµ − aeν )

(∂s ∇µ )a1 (∂t ∆)b1 − (∂s ∆)a1 (∂t ∇µ )b1 ψ(x)
X1
=− ∂s Aaµ (x)∂t Abν (x) − ∂s Aaν (x)∂t Abµ (x) + O(a)
ν
2

ψ(x + aeµ + aeν ) + ψ(x + aeµ − aeν )
× , (4.36)
+ψ(x − aeµ + aeν ) + ψ(x − aeµ − aeν )
which determine V e(1)ab and V

e(2)ab in (4.31)–(4.32). On the other hand, from (4.11)–(4.12)
(1)c (2)c
and (4.13)–(4.16) we find that V2 and V2 coincide up to O(a) with V e(1)ab and V
e(2)ab ,
respectively, after replacing ∂s Aµ (x)∂t Aν (x) with − 2 Fµν (x).
a b 1 c
Having determined the operators in (4.33) we can now evaluate the trace by first tracing
over spinor indices and then evaluating the remaining trace in the plane wave orthonormal
basis {φk (x)} for periodic scalar fields, given by
1 ik·x
φk (x) = e (4.37)
N
π
N = (2N)4 , kµ ∈ {−N, −N + 1, . . . , N − 1}. (4.38)
aN
The result is
I X
Tr(P [∂s P , ∂t P ]) = d abc 4
a µνρσ ∂s Aaµ (x)∂t Abν (x)Fρσ
c
(x), (4.39)
32π 2 x
where
X
I= a 4 ∆4 kIr (ak), (4.40)
k
(2π)4
∆4 k ≡ = the “volume per k” in (4.38),
a 4N
Q4 P P sin2 kµ
−3r ν=1 cos kν − 1 + µ (1 − cos kµ ) − µ cos kµ
Ir (k) = P P 2 5/2 . (4.41)
8π 2 sin2 kµ + r 2 −1 + (1 − cos kµ )
µ µ
−5/2
(The denominator in this expression comes from L0 in (4.33); we have used (4.17).)
Changing variables from k to ak in (4.40) leads to
Zπ
lim I = d 4 k Ir (k). (4.42)
a→0
−π
This integral was encountered in [52,53] in connection with the axial anomaly for fermions
with Overlap Dirac operator, and was found to equal 1 (independent of r). We can now take
the a → 0 limit in (4.39) and get the desired result:
Z
−1 abc
lim Tr(P [∂s P , ∂t P ]) = d d 4 xµνρσ ∂s Aaµ (x)∂t Abν (x)Fρσ
c
(x). (4.43)
a→0 32π 2
T4
In deriving this formula we have followed the approach used in [52,53] for calculating the
axial anomaly for the Overlap Dirac operator. Presumably the other approaches of Refs.
[54] and [55] could also be used to derive this formula.
Remark. The preceding also shows that

1 −1 abc
lim tr{[[∂s , ∂t ]](x, x)} = d µνρσ ∂s Aaµ (x)∂t Abν (x)Fρσ
c
(x)
a→0 8 32π 2
−1
= tr(F (x, s, t))3 . (4.44)
24
Here [· · ·](x, y) denotes the kernel of the operator [· · ·], F (x, s, t) is the curvature
of the gauge field A(x, s, t) = A(s,t ) µ
µ (x)dx in 6 dimensions, and the last coefficient
−1
arises as 2πi( (2πi)33! ) = 24π 2 . This result for the “topological field” q(x, s, t) =
1
1
8 tr{[[∂s , ∂t ]](x, x)} in 6 dimensions was used in [21] 18 to show the existence of a
w+ (U ) (or equivalently, the existence of a local gauge-invariant current jµ (x) satisfying
certain conditions) to all orders in an expansion in the lattice spacing a such that the
corresponding overlap hv+ , w+ i is gauge-invariant when dRabc = 0. The calculations in
this section leading to (4.44) above could therefore be useful as a starting point for finding
explicit expressions for the terms in jµ (x).
The main result of this paper is that the overlap formulation of chiral gauge theory on the
lattice reproduces the global obstructions to gauge-invariance discussed in the continuum
by Alvarez-Gaumé and Ginsparg [5]. We showed that the obstruction on the lattice reduces
to the continuum obstruction in the classical continuum limit. This, together with the fact
that the lattice obstruction is also an integer (winding number), implies that the lattice
obstruction coincides exactly with the continuum one for small non-zero lattice spacing
(i.e., close to the classical continuum limit). Thus the overlap formulation is seen to exactly
capture topological structure of the continuum theory in the nonabelian case, just as it
does in the abelian case considered previously in [9]. We mention again that, while we
have taken the spacetime to be the 4-dimensional, our arguments and results generalise
straightforwardly to Euclidean spacetime T 2n for arbitrary n.
It might be instructive to compare this with the situation for chiral Wilson fermions on
the lattice (where gauge-invariance is explicitly broken due to the Wilson term, and only
φθ ·U
restored in the a → 0 limit). In this case the consistent local anomaly dθd
log det(DWilson )+
has been shown to converge to the continuum anomaly in the classical continuum limit
[56–58], so the integral
Z1
1 d φθ ·U
WWilson (Φ) = dθ log det DWilson +
2πi dθ
0
converges to the continuum global obstruction deg(Φ) in this limit. However, for non-zero
lattice spacing the integral WWilson (Φ) is non-integer in general it does not have a winding
number interpretation since | det(DWilson
U
)+ | is not gauge-invariant. This is in contrast to
the overlap case where
Z1
1 d
W (Φ) = dθ loghv+ , w+ (φθ · U )i
2πi dθ
0
18 In [21] a more general q(x, s, t) was considered; however, to obtain the mentioned result it suffices to consider
the present q(x, s, t).
is an integer (winding number) since |hv+ , w+ (U )i| is gauge-invariant and hv+ ,

w+ (φθ ·U )i differs from hv, w+ (U )i only by a phase factor. Thus, in contrast to the overlap
case, the non-integer WWilson (Φ) is in general not equal to the integer-valued continuum
obstruction deg(Φ) at any non-zero lattice spacing; equality only occurs in the limit a → 0.
As a consequence of our main result, we found that the continuum anomaly cancellation
condition dRabc = 0 is a necessary condition for anomaly cancellation in the overlap on
the lattice (at least when Map(T 5 , SU(N)) contains maps with non-trivial degree). While
this is no surprise, our derivation is robust compared to other approaches: firstly, it is
independent of the choice of phase in the overlap (in contrast to the consistent local
anomaly which does depend on the phase choice), and secondly, no approximations, or
assumptions on the gauge field, have been used.
In the continuum argument of Ref. [5], the winding number obstruction W (Φ) is shown
to equal the index of a Dirac operator D in 6 = 2n + 2 dimensions. The index theorem is
then used to get index D = deg Φ. In this paper we have followed a different route: a lattice
version of the determinant line bundle approach of Refs. [6–8]. In fact, it is also possible
to give a lattice version of the original argument of Alvarez-Gaumé and Ginsparg in the
overlap setting, using a certain lattice Dirac operator in 2n + 2 dimensions (with the extra
2 dimensions being continuous) [59]. 19
The obstructions of Alvarez-Gaumé and Ginsparg are but one type of obstruction to
gauge-invariance of the chiral determinant. In general, the obstructions are manifestations
of non-trivial topological structure of the determinant line bundle over the orbit space
of gauge fields. This topic has been studied in detail in the continuum; see, e.g., [6–
8]. The results of [9] and the present paper suggest that in general the continuum
topological structure of the determinant line bundle can be reproduced on the lattice in
the overlap formulation (at least when the spacetime manifold is a 2n-dimensional torus).
The determinant line bundle comes equipped with a canonical U (1) connection, and the
difference Im Γ (A(1)) − Im Γ (A(0) ) for the effective action Γ (A) = log det(D+ A ) can be
expressed in terms of the parallel transport of this connection along a path joining A(0) to
A(1) [60]. This can in turn be expressed in terms of a spectral flow (η-invariant) of a Dirac
operator and a Chern–Simons term, both in 5 = 2n + 1 dimensions. Lattice versions of
these relations in the overlap setting have already been found [18,45].
Finally, it is interesting to note that the quantity Tr(P dP dP ) = 18 Tr( d d), which
in the present setting appears as the curvature of the overlap determinant line bundle (or
the ‘Berry curvature’ in the terminology of Ref. [9]), also arises as the curvature of a
determinant line bundle in canonical quantisation of the continuum theory. 20 See [61]
and the references therein. In that setting one considers a certain infinite-dimensional
Grassmannian manifold consisting of splittings V+ ⊕ V− of the Hilbert space of 1-particle
states; each splitting corresponds to an = PV+ − PV− . There is a canonical determinant
line bundle on this manifold, and its curvature turns out to be a renormalised version
19 In this case, the index density of the lattice D turns out to be the aforementioned topological field q(x, s, t)
in 6 = 2n + 2 dimensions which appeared in Ref. [21].
20 I thank Prof. J. Mickelsson for pointing this out to me.
of 18 Tr( d d). 21 It could be interesting to explore the apparent analogy between

this continuum formulation and the lattice overlap formulation. Recently, an obstruction
to canonical quantisation of the continuum theory on odd-dimensional spacetimes was
described in [63]. Instead of the ‘Berry curvature’ 2-cocycle, the obstruction there is given
in terms of a 3-cocycle known as the Dixmier–Douadly class. It could be interesting to see
if there is something analogous to this in the lattice overlap formulation.
Acknowledgements
This work was presented at the mini-workshop “New developments in lattice gauge
theory”, CSSM, Adelaide, 4–5 April 2000, and I thank Prof.’s Ting-Wai Chiu, Kazuo
Fujikawa, Urs Heller, Jouko Mickelsson and Tony Williams for interesting comments and
discussions at that time. I also thank Prof. Herbert Neuberger for correspondence on the
Overlap and, together with Prof.’s Martin Lüscher and Jun Nishimura, for feedback on
earlier versions of this paper. The author is supported by an ARC postdoctoral fellowship.
Appendix A. Derivation of (3.7): hv+ , (dw+ )⊥ i/hv+ , w+ i = Tr(dDD −1 P+ )
Let U (t) be a smooth curve in U . Using (2.13) we calculate

d d b +i
loghv+ , w+ i = loghv+ , Dw
dt dt
1
d b

= v+ , dt Dw+ + v+ , Db d w+ . (A.1)
b +i
hv+ , Dw dt
At t = 0 (and with our assumption that D has no zero-modes) we have

d b

d b b−1 b

v+ , dt Dw+ = v+ , dt D D Dw+ = v+ , Dw b + v+ , d D b b−1
dt D v+

b + d det D(t)D(0)−1
= v+ , Dw
dt t =0 C+

b + i Tr d DD −1 P+
= hv+ , Dw dt
d
so the first term on the right-hand side of (A.1) is Tr( dt DD −1 P+ ). Comparing (A.1) with
(3.6) we see that (3.7) holds iff

b d w+ = v+ , Dw
v+ , D b + w+ , d w+ . (A.2)
dt dt
(1,U (t ))
Choose an orthonormal basis w1 (t), . . . , wd (t) for C+ such that w+ = w1 ∧ · · · ∧ wd .
Then

Xd

b d w+ =
v+ , D b w1 ∧ · · · ∧
v+ , D d
∧ · · · ∧ wd .
dt dt wj (A.3)
j =1
21 In fact, a lattice regularisation of chiral gauge theory (different to the overlap) in which this quantity also
appears has been presented in Ref. [62]. Gauge-invariance appears to be problematic in this approach though: the
notion of gauge symmetry needs to be modified on the lattice in a way that involves non-local operators.
P (1,U (t ))
d
Substitute dt wj = dk=1 hwk , dt
d
wj iwk + ( dtd d
wj )⊥ in (A.3), where ( dt wj )⊥ ∈ C− .
b (1,U
Since D maps C± = C∓
)
to C∓ , the terms involving ( dt wj )⊥ give vanishing contribution,
d
and we get

Xd
b d w+ =
v+ , D b + ihwj , d wj i.
hv+ , Dw
dt dt
j =1
This equals the right-hand side of (A.2) as required, since

Xd

d
w+ , dt w+ = w+ , w1 ∧ · · · ∧ d
dt wj ∧ · · · ∧ wd
j =1
X
d

Xd

= hw+ , w+ i wj , dt
d
wj = d
wj , dt wj .
j =1 j =1
Appendix B. Derivation of (3.14)
It suffices to restrict to a surface in U with coordinates (s, t) and show

h∂s w+ , ∂t w+ i − h∂t w+ , ∂s w+ i = Tr(P ∂s P ∂t P ) − Tr(P ∂t P ∂s P ). (B.1)
(1,U (s,t ))
Let w1 (s, t), . . . , wd (s, t) be an orthonormal basis for C+ such that w+ =
w1 ∧ · · · ∧ wd . It is convenient to use bra-ket notation:
X
d X
d
P= |wk ihwk |, ∂s P = |∂s wk ihwk | + |wk ih∂s wk |.
k=1 k=1
Then
X
d
Tr(P ∂s P ∂t P ) = hwj |∂s P ∂t P |wj i
j =1
X
d
= hwj |(|∂s wk ihwk | + |wk ih∂s wk |)(|∂t wl ihwl | + |wl ih∂t wl |)|wj i
j,k,l=1
X X
= hwj |∂s wk ihwk |∂t wj i + hwj |∂s wk ih∂t wk |wj i
j,k j,k
X X
+ h∂s wj |∂t wj i + h∂s wj |wl ih∂t wl |wj i. (B.2)
j k,l
The first and fourth sums are clearly symmetric under ∂s ↔ ∂t . The second sum is likewise
symmetric under ∂s ↔ ∂t as is easily seen using hδwj |wk i + hwj |δwk i = δhwj |wk i = 0. It
follows that
X
Tr(P ∂s P ∂t P ) − Tr(P ∂t P ∂s P ) = h∂s wj |∂t wj i − h∂t wj |∂s wj i
j
and this is equal to the left-hand side of (B.1) as required.

References
[1] R. Narayanan, H. Neuberger, Phys. Lett. B 302 (1993) 62.

[2] R. Narayanan, H. Neuberger, Phys. Rev. Lett. 71 (1993) 3251.
[3] R. Narayanan, H. Neuberger, Nucl. Phys. B 412 (1994) 574.
[5] L. Alvarez-Gaumé, P. Ginsparg, Nucl. Phys. B 243 (1984) 449.
[6] M.F. Atiyah, I.M. Singer, Proc. Natl. Acad. Sci. USA 81 (1984) 2597.
[7] J.-M. Bismut, D.S. Freed, Commun. Math. Phys. 106 (1986) 159.
[8] J.-M. Bismut, D.S. Freed, Commun. Math. Phys. 107 (1986) 103.
[10] R. Narayanan, J. Nishimura, Nucl. Phys. B 508 (1997) 371.
[11] Y. Kikukawa, H. Neuberger, Nucl. Phys. B 513 (1998) 735.
[12] H. Neuberger, Phys. Lett. B 437 (1998) 117.
[13] O. Bär, I. Campos, hep-lat/0001025 (to appear in Nucl. Phys. B).
[14] S. Randjbar-Daemi, J. Strathdee, Phys. Lett. B 348 (1995) 543.
[15] S. Randjbar-Daemi, J. Strathdee, Nucl. Phys. B 443 (1995) 386.
[16] S. Randjbar-Daemi, J. Strathdee, Nucl. Phys. B 466 (1996) 335.
[17] S. Randjbar-Daemi, J. Strathdee, Phys. Lett. B 402 (1997) 134.
[19] H. Suzuki, hep-lat/0002009.
[20] M. Lüscher, Nucl. Phys. B 549 (1999) 295.
[22] H. Neuberger, hep-lat/0002032.
[23] M. Golterman, Y. Shamir, Nucl. Phys. Suppl. 47 (1996) 603.
[25] T. Izubuchi, J. Nishimura, JHEP 9910 (1999) 002.
[28] Y. Kikukawa, R. Narayanan, H. Neuberger, Phys. Lett. B 399 (1997) 105.
[29] Y. Kikukawa, R. Narayanan, H. Neuberger, Phys. Rev. D 57 (1998) 1233.
[30] Y. Kikukawa, Nucl. Phys. (Proc. Suppl.) 63 (1998) 587.
[31] P. Ginsparg, K.G. Wilson, Phys. Rev. D 25 (1982) 2649.
[32] P. Hasenfratz, F. Niedermayer, Nucl. Phys. B 414 (1994) 785.
[33] P. Hasenfratz, in: M. Asorey, A. Dobado (Eds.), in Non-perturbative quantum field physics
(Peñiscola, 1997), World Scientific, Singapore, 1998.
[34] P. Hasenfratz, Nucl. Phys. B (Proc. Suppl.) A-C 63 (1998) 53.
[35] P. Hasenfratz, Nucl. Phys. B 525 (1998) 401.
[36] P. Hasenfratz, V. Laliena, F. Niedermayer, Phys. Lett. B 427 (1998) 125.
[38] T. Fujiwara, H. Suzuki, K. Wu, Phys. Lett. B 463 (1999) 63.
[39] T. Fujiwara, H. Suzuki, K. Wu, Nucl. Phys. B 569 (2000) 643.
[40] T. Fujiwara, H. Suzuki, K. Wu, hep-lat/9910030.
[41] T. Fujiwara, H. Suzuki, K. Wu, hep-lat/0001029.
[42] H. Suzuki, hep-lat/99110009.
[43] P. Hernández, K. Jansen, M. Lüscher, Nucl. Phys. B 552 (1999) 363.
[45] T. Aoyama, Y. Kikukawa, hep-lat/9905003.

[50] J.E. Avron, R. Seiler, B. Simon, Phys. Rev. Lett. 51 (1983) 51.
[51] D.H. Adams, in preparation.
[52] D.H. Adams, hep-lat/9812003.
[54] Y. Kikukawa, A. Yamada, Phys. Lett. B 448 (1999) 265.
[55] K. Fujikawa, Nucl. Phys. B 546 (1999) 480.
[56] S. Aoki, Phys. Rev. D 35 (1986) 1435.
[57] A. Coste, C. Korthals Altes, O. Napoly, Nucl. Phys. B 289 (1987) 645.
[58] T. Jolicoeur, R. Lacaze, O. Napoly, Nucl. Phys. B 293 (1987) 215.
[59] D.H. Adams, hep-lat/9910036 v2.
[60] S. Della Pietra, V. Della Pietra, L. Alvarez-Gaumé, Commun. Math. Phys. 109 (1987) 691.
[61] J. Mickelsson, Commun. Math. Phys. 154 (1993) 403.
[62] J. Mickelsson, hep-lat/9602030.
[63] A.L. Carey, J. Mickelsson, hep-th/9912003.
Instantons, monopoles and the flux quantization

in the Faddeev–Niemi decomposition
Toyohiro Tsurumaru a,∗ , Izumi Tsutsui a , Akira Fujii b
a Institute of Particle and Nuclear Studies, High Energy Accelerator Research Organization (KEK), Tsukuba,
Ibaraki 305-0801, Japan
b Department of Physics, Graduate School of Science, Osaka University, Toyonaka, Osaka 560-0043, Japan
Abstract
We study how instantons arise in the low energy effective theory of the SU(2) Yang–Mills theory
in the context of the non-linear sigma model recently proposed by Faddeev and Niemi. We find a
simple relation between the instanton number ν and the charge m of the monopole that appears in the
effective theory. It is given by ν = mΦ/(2π), where Φ is the quantized flux associated with a U (1)
gauge field passing through the loop formed by the singularity of the monopole.  2000 Elsevier
Science B.V. All rights reserved.
PACS: 11.15.Tk; 11.27.+d; 12.39.Dc; 14.80.Hv

Keywords: Instanton; Monopole; Flux quantization; Skyrmion
1. Introduction
Yang–Mills (YM) theory is known to provide a basis for describing fundamental

interactions, most notably, the strong interaction. In the high energy regime the theory
admits perturbative studies thanks to the asymptotic freedom, whereas in the low energy
regime it defies a similar systematic analysis due to the strong coupling behavior. Recently,
Faddeev and Niemi made a suggestion [1] that the SU(2) YM theory LYM = 2g1 2 tr Fµν F µν
in a certain phase of the low energy regime may be described by a modified version of the
O(3) non-linear sigma model (NLSM) in four dimensions often referred to as the Skyrme–
Faddeev model,
1 1 2
LSF = 2 (∂µ n)2 + 2 abc na ∂µ nb ∂ν nc . (1.1)
2λ 4e
E-mail addresses: toyohiro.tsurumaru@kek.jp (T. Tsurumaru), izumi.tsutsui@kek.jp (I. Tsutsui),
fujii@het.phys.sci.osaka-u.ac.jp (A. Fujii).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 3 8 - 1
660 T. Tsurumaru et al. / Nuclear Physics B 589 (2000) 659–668
This model permits stable soliton solutions [2–6], which have been conjectured to be a
candidate for glueballs [1].
The key ingredient for establishing the connection between the SU(2) YM theory and
the Skyrme–Faddeev model is the so-called Faddeev–Niemi (FN) decomposition 1
A = Cn + (1 + σ ) dn × n + ρ dn. (1.2)
This is a new version of the Abelian projection which decomposes the YM field A into
a vector field C, a complex scalar φ = ρ + iσ and an iso-vector field n of unit length
P3
a=1 n n = 1. The vector field C is a gauge field associated with the U (1) gauge
a a
transformations with respect to n, that is, it transforms as C → C + dθ under the U (1)

subgroup U (x) = exp[θ na T a ] of SU(2) transformations for A. Thus, at on-shell, C has
two degrees of freedom when gauge-fixed, and together with n, ρ and σ it comprises the
six degrees of freedom of the on-shell SU(2) YM field. The original motivation for the
Abelian projection proposed by ’t Hooft and Polyakov [7–9] was to provide a qualitative
explanation for the color confinement by asserting that in the low energy regime the YM
field exhibits the dual Meissner effect in a condensate of magnetic monopoles. With regard
to this it is very important to study how instantons — a hallmark of the non-perturbative
aspect of the YM theory — can affect the magnetic monopole configurations in the
Skyrme–Faddeev model obtained under the FN decomposition (1.2). (The implication of
the FN decomposition for confinement and chiral symmetry breaking has been discussed
in Ref. [10].)
In this note we point out that, when instantons are present in the SU(2) YM theory
monopoles must necessarily appear in the low energy regime, i.e., in the Skyrme–Faddeev
model. The relation between instantons and monopoles has been discussed earlier by Christ
and Jackiw for physically static field configurations [11], where the instanton number was
shown to coincide with the monopole charge for SU(2). In the context of the original
Abelian projection, a similar relation has been found in [12,13] for generic configurations
under the Weyl gauge. More recently, it has been shown without reference to gauge fixing
that the instanton number is given, modulo the monopole charge, by the (generalized)
Hopf invariant associated with an adjoint scalar field [14]. Here we examine the problem
in the context of the FN decomposition, and argue that the relation between the instanton
number and the monopole charge is again given by a similar formula, involving this time
a quantized flux trapped by the monopole. Specifically, if the original YM gauge field that
admits the FN decomposition possesses a nonzero instanton number, the corresponding
vector field n must have singularity in closed circles, or “monopole loops”. We shall find
that the flux Φ associated with the U (1) gauge field C trapped by the monopole loop must
be quantized topologically, and that the instanton number ν is given by the product of the
monopole charge m and the topological integer of the flux Φ (see Fig. 1),
Φ
ν =m . (1.3)
2π
1 Notation: We denote su(2)-valued fields such as A = Aa T a = Aa dx µ T a often by iso-vectors as A =
µ
(A1 , A2 , A3 ). Here d is the spacetime exterior derivative, and T a := τ a /(2i), where τ a are the Pauli matrices, is
the su(2) basis we use.
T. Tsurumaru et al. / Nuclear Physics B 589 (2000) 659–668 661
Fig. 1. The singularity of n arises as a closed loop (monopole loop) in the spacetime for
non-vanishing instanton sectors. The instanton number ν is given by the monopole charge m times
the integer of the quantized flux Φ associated with the U (1) gauge field C passing through the surface
encircled by the monopole loop.
The rest of the paper is devoted to show this simple relation.
2. Instanton and singularity
Prior to the discussion on the relation (1.3), we first show that, under the FN
decomposition (1.2), the field n is necessarily singular if the original A possesses a
non-vanishing instanton number. Following the conventional procedure, we assign a
configuration of the gauge field on our spacetime diffeomorphic to S 4 by introducing a
set of local coordinate patches Vk ' D 4 for k = 1, 2. Let Ak be the gauge potential on the
patch Vk . On the overlap V1 ∩ V2 , the gauge fields are related by
A2 = U † A1 U + U † dU, (2.1)
with the transition function U (x) ∈ SU(2). The instanton number ν of the gauge field is
encoded in the transition function in its winding number on the boundary ∂V1 ' S 3 (or
equivalently on ∂V2 ),
Z Z
1 1 2 3
ν = − 2 tr(F ∧ F ) = π tr U † dU . (2.2)
8π 24
S4 ∂V1
If we now adopt the FN decomposition to each of the gauge fields on the patches, we
have
Ak = Ck nk + (1 + σk ) dnk × nk + ρk dnk , (2.3)
for k = 1, 2. We remark that the FN decomposition implies, in effect, a specific gauge fixing
and use of an on-shell condition for A (except for the part of the U (1) gauge field C), and
therefore, it is a nontrivial question if the decomposition (2.3) is actually available over the
entire patches we have introduced. In what follows we consider the class of configurations
for which this is possible, knowing that standard instanton configurations such as Witten’s
solution (see below) belong to this class. Further, for our effective picture of the Skyrme–
Faddeev model which has no local gauge symmetry, we need to regard the field n physical,
that is, it represents the low energy dynamical freedom of the YM theory irrespective of
the gauge chosen. In other words, we assume that given a configuration of the SU(2) gauge
field A with gauge equivalent configurations identified, the corresponding field n in the FN
decomposition can be uniquely determined up to a global O(3) transformation. This is a
crucial but necessary demand for the FN decomposition to work in practice, allowing us to
discuss the topological aspect of the YM theory in terms of the field n.
Upon this assumption we can perform with no loss of generality, a global O(3)
transformation so that we have n1 = n2 on the overlap V1 ∩ V2 . Thus the transition function
U defined on the overlap must be of the form,
X
U (x) = exp[θ (x)n(x)], n(x) := na (x)T a , (2.4)
a
with some function θ (x) on V1 ∩ V2 . Then it is easily seen that if n is regular everywhere
on ∂V1 , the winding number (2.2) of U (x) must be zero. Indeed, if we expand U (x) in
(2.4) in the form,
U (x) = α(x) + β(x)n(x), (2.5)
with
θ 1 θ
α(x) = cos = tr U, β(x) = sin = −2 tr(U, n), (2.6)
2 2 2
then α(x) and β(x) are also regular for regular n(x), because U (x) is given regularly
over the overlap. Since (α, β) may be regarded as a map from ∂V1 ' S 3 to S 1 , it can
be deformed continuously to the constant map (α, β) = (1, 0) on account of π3 (S 1 ) =
0. For regular n, this means that U (x) can be continuously deformed to the constant
map U (x) = 1, for which the winding number vanishes. It follows therefore that n must
necessarily be singular if the gauge field A has a non-vanishing instanton number. Note
that the regularity of U (x) demands that the parameter θ in (2.4) be such that the function
β in (2.5) converges to zero sufficiently fast when n approaches its singularity. Thus, for
singular n, the functions (α, β) are still given regularly, even though the deformation of
(α, β) mentioned above cannot be performed due to the constraint β = 0 at the singularity.
The singularity of n can be classified by the topological property π2 (S 2 ) = Z by
regarding n as a map from an arbitrary two-dimensional sphere S 2 in the spacetime to
the target S 2 . The topological integer can be provided by
Z
1
m= tr(n dn ∧ dn), (2.7)
2π
S2
which we call the “monopole charge”. In the four-dimensional spacetime we expect that the
singularity of n forms a one-dimensional closed loop [7–9], i.e., a monopole loop as shown
in Fig. 1. From the foregoing argument we find that, if ν 6= 0, then m 6= 0 or, equivalently,
if m = 0 then ν = 0.
At this point, it is instructive for us to look at explicit solutions of the YM instantons
that admit the FN decomposition. For this we take Witten’s ansatz [15] which is an axially
Fig. 2. The spacetime S 4 is covered with two local patches V1 , V2 ' D 4 . The monopole loop lies
across the overlap V1 ∩ V2 and hence can be cut into two pieces Γ1 and Γ2 at the two intersections
p and q with the boundary ∂V1 .
symmetric self-dual solution to the SU(2) YM equation with arbitrary instanton numbers.
With the coordinate (E
x , t) on a local patch, the ansatz assumes for the solution the most
general form which is invariant under SO(3) rotations
A0 x a ϕ2 + 1 ϕ1 xj xa
Aa0 = , Aaj = 2
j ak xk + 2 δj a r 2 − xj xa + A1 2 , (2.8)
r r r r
where ϕ, A0 , A1 are functions of the radius r = |E x | and t. Note that this is already
in the form of the FN decomposition (1.2) as seen by setting ϕ1 = ρ, ϕ2 = σ, Ci =
(xi /r)A1 , C0 = A0 and na (E x ) = x a /r [16], where now the monopole loop is located at
the origin of the space. If we let z = r + it and define ψ by Aµ = µν ∂ν ψ (µ, ν = 0, 1,
where 0 and 1 refers to t and r, respectively), self-dual solutions with instanton number
Q
ν = k − 1 are obtained by ψ = − ln[(1 − g ∗ g)/(2r)] with g(z) = ki=1 (ai − z)/(ai∗ + z)
using arbitrary complex constants ai satisfying Re ai > 0. Clearly, the instanton number
ν is not determined (and hence can be put even to zero) under the field na (E x , t) = x a /r
which is singular at the origin and has the monopole charge m = 1 (see (2.7)).
The above example shows that the instanton number ν is not determined by the
monopole charge alone, and that to account for the instanton number the configuration of
the U (1) gauge field C must also be considered. Indeed, for the ansatz (2.8) the instanton
number (2.2) reduces to
Z
1 1
ν = π d x ∂µ (ij µν ϕi Dν ϕj ) + µν Fµν ,
2
(2.9)
2 2
D2
where D 2 is a two-dimensional disc whose boundary is the monopole loop. We find that
the second term yields the flux of the gauge field C passing through the monopole loop,
and this will be seen the only contribution to the instanton number ν (i.e., the first term
given by the surface integral vanishes) on a general basis shortly.
3. Instanton number vs. monopole charge
The formula (2.9) suggests that the flux associated with C penetrating through the
surface encircled by the monopole loop contributes to the instanton number ν. We now
show that ν is in fact proportional to the flux in just the way (1.3) we mentioned in
the Introduction. We do this in two steps, first showing that the instanton number is
proportional to the monopole charge, and then arguing that the proportional factor is given
by the flux which is quantized.
To this end, recall that for nonzero ν the field n is singular on the boundary ∂V1 .
This implies that the monopole loop (which is the line of singularity of n) must intersect
with ∂V1 twice or more generally even times. Suppose for simplicity, that there are two
intersections on ∂V1 , which we denote as p and q. (When there are more than two
intersections we may choose a different set of coordinate patches so that the boundary
∂V1 intersects with the monopole loop only twice.) From (2.6) we find that in the overlap
∂V1 ∩ ∂V2 the function θ can be defined regularly. At the singular points p and q, we have
β = sin(θ/2) = 0 to ensure the regularity of U and hence
θ = 2π × integer, (3.1)
at both p and q.
Consider then the cylinder ∂V1 \ {p, q} ' S 2 × Ipq (without edges) obtained by
removing the two points p and q from ∂V1 ' S 3 , where Ipq is an interval from p to q
on the cylinder (see Fig. 3). Since U (x) is regular, the removal of the two points from
the domain of integration for the instanton number (2.2) does not alter the outcome. Thus,
instead of the domain ∂V1 ' S 3 we may use the cylinder to evaluate the instanton number
(2.2) as
Z Z
1 1
ν= tr(U dU ) =
† 3
(1 − cos θ )dθ tr(n dn ∧ dn), (3.2)
24π 2 4π 2
S 2 ×Ipq S 2 ×Ipq
where we have used

U † dU = (n dθ + sin θ dn) − [n, dn](1 − cos θ ). (3.3)
Fig. 3. The removal of the two points p and q from ∂V1 yields the cylinder ∂V1 \ {p, q} ' S 2 × Ipq .
Let us choose the coordinates of the cylinder S 2 × Ipq such that θ is constant on S 2 at
each point of Ipq . This choice of coordinates allows us to evaluate the integral (3.2) by two
separate integrals over θ and n as
Z Z
m m
ν= (1 − cos θ ) dθ = dθ, (3.4)
2π 2π
Ipq Ipq
where m is the monopole charge (2.7), and we have used (3.1) for the second equality. The
condition (3.1) also ensures that the r.h.s. of (3.4) gives an integer as required.
4. Flux quantization
It remains to show that the integer factor multiplying m in (3.4) is the flux Φ associated
with the U (1) gauge field C passing through the monopole loop. Let Γk , k = 1, 2, be the
contours obtained by cutting the monopole loop in half at p and q (see Fig. R 2). Then
R the
flux Φk penetrating the surface encircled by Γk and Ipq is given by Φk = ( Γk + Ipq )Ck .
Noting that C1 and C2 are related by the relation C2 = C1 + dθ on Ipq , we find that the
total flux is given by
Z Z Z
Φ = Φ1 + Φ2 = C1 + C2 + dθ. (4.1)
Γ1 Γ2 Ipq
We shall argue that the component of the U (1) field Ck along the monopole loop actually
vanishes at the monopole loop, and hence the contributions from the two contours Γ1 and
Γ2 in (4.1) disappear.
For this, we choose a local coordinate patch in V1 and parameterize it by (E x , t), where
the origin xE = 0E is taken to be at the monopole loop, and t is the local coordinate along the
monopole loop. With x̂ = xE /|E x | and r = |E
x |, we consider the limit
n0 (x̂, t) := lim n(r x̂, t). (4.2)
r→0
Here we suppose that n(x) has a certain limit dependent on the direction x̂. To specify
the direction, let us use the polar coordinates (r, ϑ, ϕ) and consider the unit vectors Eer , Eeϑ ,
Eeϕ associated with them. (Note that n has a direction-independent limit if and only if it is
regular, for which we have m = 0 and hence the relation (1.3) holds trivially.) For n with
m 6= 0, the radius of the sphere S 2 on which the monopole charge (2.7) is evaluated can be
taken as small as we wish without changing the value m. From this we recognize that the
derivative
E = Eer ∂r n + Eeϑ ∂ϑ n + Eeϕ ∂ϕ n,
∇n (4.3)
r r
diverges as 1/r on average, or more precisely, on an area of finite volume on the S 2 . Under
the existence of the limit (4.2) we obtain
|∂t n| |∂t n0 |
lim = lim = 0, (4.4)
r→0 ∇n
E r→0 ∇nE
qP
where we have used the notation |∇n| E = a 2
i,a (∂i n ) . On the other hand, from the
inequality
|∂ n|
(1 + σ )∂t n × n + ρ∂t n = (1 + σ )∇n
E × n + ρ ∇n E |∂t n| ,
E t 6 |A| (4.5)
E
∇n E
∇n
E is finite, we observe that the l.h.s. of (4.5) converges to zero. Hence it
and the fact that |A|
follows that
At → Ct n for r → 0. (4.6)
However, since At is smooth while n is singular at the origin r = 0, we must have
limr→0 At = 0 and
Ct → 0 for r → 0. (4.7)
Thus we see that the component Ct along the monopole loop vanishes identically at
the monopole loop. This in turn implies that the flux Φ in (4.1) is quantized as Φ =
2π × integer, where the integer is given by the difference in the integers (3.1) of the angle
θ at the singular points p and q. (The quantization condition takes a more familiar form
Φ = (2π h̄/g) × integer if the flux is evaluated for C/g with a properly rescaled g.) In
conclusion, we have shown that the relation (1.3) holds for the class of those n for which
the limit (4.2) exists. We note that Witten’s ansatz for instantons has such a limit, and
we expect that any physically interesting configurations will also have it, because violent
fluctuations will be smeared out in the low energy regime. For completeness, however, in
Appendix A we shall provide an outline of the argument for more generic configurations
for n.
Acknowledgement
The authors wish to thank A.J. Niemi for helpful discussions. This work is supported in
part by the Grant-in-Aid for Scientific Research (C) under Contract No. 11640301 provided
by the Ministry of Education, Science, Sports and Culture of Japan.
Appendix A
We here discuss the general case in which n may not have the limit (4.2). Let us
regard Aaµ as a four times three matrix and consider the quantity rank A which is the
number of linearly independent (column or, equivalently, row) vectors in the matrix Aaµ .
Choose a point on the monopole loop to which we assign the value t = 0. Consider then a
neighbourhood of the point with four-dimensional radius such that
p
rank A(x) > rank A(x = 0) for |x| = xµ x µ < . (A.1)
Since rank A(x = 0) = 0 means A(x = 0) = 0 and hence limr→0 Cµ = 0, we only need to
consider the cases where rank A > 1 at x = 0.
(i) Rank A(x = 0) = 3

In this case we have k µ (x) satisfying k µ Aµ = 0 given by
1 µαβγ
k µ := abc Aaα Abβ Acγ , (A.2)
3!
which is a non-vanishing smooth function for rank A = 3. As can be easily seen from (1.2),
we have (1 + σ )2 + ρ 2 6= 0 for r 6= 0, and hence k µ Aµ = 0 means that
k µ ∂µ n = 0. (A.3)
Thus k µ is a Killing vector for n, and hence it must point to the t-direction along the
monopole loop at x = 0. It then follows that At (x = 0) = 0 and, therefore, Ct (x = 0) = 0.
(ii) Rank A(x = 0) = 2 and 1
For Aµ with rank = 2, we can put Aµ and ∂µ n (which also has rank = 2 because n has
two independent freedoms) in the form
Aµ (x) = αµ a + βµ b + O(|x|), ∂µ n(x) = ξµ (x)e(x) + χµ (x)f(x), (A.4)
where αµ , βµ , a, b are constant vectors while other vectors ξµ , χµ , e, f are coordinate
dependent. Plugging these into the identity
(Aµ × n) × n = −(1 + σ )∂µ n × n + ρ∂µ n, (A.5)
which holds for r > 0, we see that the vectors ξµ (x) and χµ (x) are given by a linear
combination of αµ and βµ with coordinate dependent coefficients. We thus find that, for
|x| → 0, ∂µ n takes the form
∂µ n → αµ u(n) + βµ v(n), (A.6)
where u(n), v(n) are vectors orthogonal to n. Let γµ be a vector which is orthogonal to
both αµ and βµ and has no time-component. If we choose the z-direction of space to be
parallel to the vector γµ , then we get
|∂z n|
lim = 0. (A.7)
|x|→0 ∇n
E
This is, however, impossible because ∂i n must diverge uniformly for all i as r → 0 if n has
a non-vanishing monopole charge m 6= 0. Thus, we conclude that Aµ with rank = 2 cannot
arise at the monopole loop. The case Aµ with rank = 1 can also be denied by a similar
argument.
References
[1] L. Faddeev, A.J. Niemi, Phys. Rev. Lett. 82 (1999) 1624.

[2] L. Faddeev, A.J. Niemi, Nature 387 (1997) 58.
[3] J. Gladikowski, M. Hellmund, Phys. Rev. D 56 (1997) 5194.
[4] R. Battye, P. Sutcliffe, hep-th/9811077.
[5] R. Battye, P. Sutcliffe, Phys. Rev. Lett. 81 (1998) 4798.
[6] J. Hietarinta, P. Salo, Phys. Lett. B 451 (1999) 60.

[9] A. Polyakov, Nucl. Phys. B 120 (1977) 429.
[10] K. Konishi, K. Takenaga, IFUP-TH-61-99, hep-th/9911097.
[11] N. Christ, R. Jackiw, Phys. Lett. B 91 (1980) 228.
[12] H. Reinhardt, Nucl. Phys. B 503 (1997) 505.
[13] C. Ford, U.G. Mitreuter, T. Tok, A. Wipf, Ann. Phys. 269 (1998) 26.
[14] O. Jahn, J. Phys. A 33 (2000) 2997.
[15] E. Witten, Phys. Rev. Lett. 38 (1977) 121.
[16] L. Faddeev, A.J. Niemi, From Yang–Mills field to solitons and back again, hep-th/9901037.
CUMULATIVE AUTHOR INDEX B581–B589
Aarts, G. B587 (2000) 403 Berenstein, D. B589 (2000) 196

Abada, A. B585 (2000) 45 Berera, A. B585 (2000) 666
Abrikosov Jr., A.A. B586 (2000) 589 Berkovits, N. B587 (2000) 147
Accomando, E. B585 (2000) 124 Bertola, M. B581 (2000) 575
Achiman, Y. B584 (2000) 46 Bertola, M. B587 (2000) 619
Adams, D.H. B589 (2000) 633 Bertolini, M. B582 (2000) 393
Agashe, K. B588 (2000) 39 Bialas, P. B581 (2000) 477
Aglietti, U. B587 (2000) 363 Bialas, P. B583 (2000) 368
Ahn, C. B586 (2000) 611 Bianchi, M. B584 (2000) 216
Akemann, G. B583 (2000) 739 Bijnens, J. B585 (2000) 293
Aliev, T.M. B585 (2000) 275 Bilal, A. B582 (2000) 65
ALPHA Collaboration B588 (2000) 377 Binoth, T. B585 (2000) 741
Alvarez, O. B582 (2000) 139 Blanchard, P. B583 (2000) 368
Alvarez, O. B584 (2000) 659 Blanchard, Ph. B588 (2000) 229
Alvarez, O. B584 (2000) 682 Blumenhagen, R. B582 (2000) 44
Amico, L. B588 (2000) 531 Blumenhagen, R. B589 (2000) 545
Amorós, G. B585 (2000) 293 Blümlein, J. B581 (2000) 449
Anastasiou, C. B585 (2000) 763 Blümlein, J. B586 (2000) 349
Andreev, O. B583 (2000) 145 Bohm, A.R. B581 (2000) 91
Angelantonj, C. B589 (2000) 545 Bonini, G.F. B587 (2000) 403
Antoniadis, I. B583 (2000) 35 Bonini, M. B585 (2000) 253
Arhrib, A. B581 (2000) 34 Bonora, L. B589 (2000) 461
Arnaudon, D. B588 (2000) 638 Boudjema, F. B581 (2000) 3
Arnowitt, R. B585 (2000) 124 Braaten, E. B586 (2000) 427
Arutyunov, G. B586 (2000) 547 Brace, D. B588 (2000) 521
Asano, M. B588 (2000) 453 Brandt, F. B587 (2000) 543
Aschieri, P. B588 (2000) 521 Braun, V. B589 (2000) 381
Brignole, A. B582 (2000) 759
Bajnok, Z. B587 (2000) 585 Brignole, A. B587 (2000) 3
Balog, J. B583 (2000) 614 Brink, L. B586 (2000) 183
Bandos, I. B586 (2000) 315 Bronnikov, K.A. B584 (2000) 436
Bär, O. B581 (2000) 499 Bros, J. B581 (2000) 575
Barbieri, R. B585 (2000) 28 Bros, J. B587 (2000) 619
Barceló, C. B584 (2000) 415 Brower, R.C. B587 (2000) 249
Barr, S.M. B585 (2000) 79 Bruckmann, F. B584 (2000) 589
Barros, M. B584 (2000) 719 Buchbinder, I.L. B584 (2000) 615
Baulieu, L. B581 (2000) 604 Bueno, A. B589 (2000) 577
Bélanger, G. B581 (2000) 3 Buras, A.J. B586 (2000) 397
Belitsky, A.V. B589 (2000) 611
Bellucci, S. B587 (2000) 445 Cacciatori, S. B587 (2000) 277
Benakli, K. B583 (2000) 35 Caffo, M. B581 (2000) 274
Benaoum, H.B. B585 (2000) 554 Campanelli, M. B589 (2000) 577
Benedetti, R. B588 (2000) 436 Campbell, B.A. B581 (2000) 240
670 Nuclear Physics B 589 (2000) 669–675
Campos, I. B581 (2000) 499 Donato, F. B581 (2000) 3

Capdequi Peyranère, M. B581 (2000) 34 Dorn, H. B583 (2000) 145
Capella, A. B586 (2000) 382 Dorsner, I. B585 (2000) 79
Capitani, S. B582 (2000) 762 Dubin, A.Yu. B582 (2000) 677
Caravaglios, F. B589 (2000) 475 Dubin, A.Yu. B584 (2000) 749
Carena, M. B586 (2000) 92 Dutta, B. B585 (2000) 124
Carloni Calame, C.M. B584 (2000) 459
Carpentier, D. B588 (2000) 565 Eckern, U. B588 (2000) 531
Carter, G.W. B582 (2000) 571 Eden, B. B581 (2000) 523
Casas, J.A. B581 (2000) 61 Eguchi, T. B586 (2000) 331
Castellani, C. B583 (2000) 542 Einhorn, M.B. B582 (2000) 216
Castro Alvaredo, O.A. B581 (2000) 643 Ellis, J. B586 (2000) 92
Ceresole, A. B585 (2000) 143 Engelhardt, M. B585 (2000) 591
Chakrabarti, S.K. B582 (2000) 627 Engelhardt, M. B585 (2000) 614
Chakravarty, S. B587 (2000) 228 Enqvist, K. B582 (2000) 763
Chalmers, G. B585 (2000) 517 Epele, L.N. B583 (2000) 454
Chan, Ch.S. B581 (2000) 156 Erdmenger, J. B585 (2000) 517
Chayes, L. B588 (2000) 229 Erdmenger, J. B589 (2000) 3
Cheng, H.-C. B589 (2000) 249 Erickson, J.K. B582 (2000) 155
Chetyrkin, K.G. B583 (2000) 3 Erler, J. B586 (2000) 73
Chetyrkin, K.G. B586 (2000) 56 Erlich, J. B581 (2000) 309
Chiu, T.-W. B588 (2000) 400 Erlich, J. B584 (2000) 359
Chu, C.-S. B582 (2000) 65 Espinosa, J.R. B586 (2000) 3
Chu, C.-S. B585 (2000) 193 Evans, N. B581 (2000) 391
Ciafaloni, P. B589 (2000) 359 Eyras, E. B584 (2000) 251
Cirillo, E.N.M. B583 (2000) 584
Coleman, P. B586 (2000) 641 Fabrizio, M. B583 (2000) 542
Comelli, D. B589 (2000) 359 Faessler, A. B587 (2000) 25
Coste, A. B581 (2000) 679 Fanchiotti, H. B583 (2000) 454
Cox, J. B583 (2000) 331 Farhi, E. B585 (2000) 443
Creminelli, P. B585 (2000) 28 Faux, M. B589 (2000) 269
Csáki, C. B581 (2000) 309 Ferrández, A. B584 (2000) 719
Csáki, C. B584 (2000) 359 Ferreira, C.N. B581 (2000) 165
Cvetič, M. B584 (2000) 149 Ferreira, P.C. B589 (2000) 167
Cvetič, M. B586 (2000) 275 Feruglio, F. B582 (2000) 759
Cvetič, M. B586 (2000) 287 Finster, F. B584 (2000) 387
Czyż, H. B581 (2000) 274 Fortunato, S. B583 (2000) 368
Fosco, C.D. B582 (2000) 716
D’Hoker, E. B589 (2000) 38 Freedman, D.Z. B589 (2000) 3
Dall’Agata, G. B585 (2000) 143 Freund, M. B585 (2000) 105
Damgaard, P.H. B583 (2000) 347 Freund, M. B588 (2000) 101
Dasgupta, K. B587 (2000) 228 Fries, R.J. B582 (2000) 537
da Silva, A.J. B587 (2000) 299 Fries, R.J. B589 (2000) 381
Davidson, S. B587 (2000) 118 Fröhlich, J. B583 (2000) 381
de Azcárraga, J.A. B581 (2000) 743 Frolov, S. B586 (2000) 547
Delfino, G. B583 (2000) 597 Fuchs, J. B588 (2000) 110
Denef, F. B581 (2000) 135 Fujii, A. B589 (2000) 659
Denner, A. B587 (2000) 67 Fujikawa, K. B587 (2000) 419
Derendinger, J.-P. B582 (2000) 231 Fujikawa, K. B589 (2000) 487
Derkachov, S. B583 (2000) 691
Diakonov, D. B582 (2000) 571 Gaberdiel, M.R. B589 (2000) 545
Díaz, M.A. B583 (2000) 182 Gaillard, M.K. B588 (2000) 197
Dick, K. B588 (2000) 101 Gallot, L. B586 (2000) 206
Di Clemente, V. B581 (2000) 61 Gandolfo, D. B583 (2000) 368
Dittmaier, S. B587 (2000) 67 Gandolfo, D. B588 (2000) 229
Dobrescu, B.A. B589 (2000) 249 Gannon, T. B581 (2000) 679
Nuclear Physics B 589 (2000) 669–675 671
Ganor, O.J. B587 (2000) 228 Hirayama, T. B587 (2000) 207

Garcia-Garcia, A.M. B586 (2000) 668 Hisano, J. B584 (2000) 3
García Canal, C.A. B583 (2000) 454 Hoffmann, L. B589 (2000) 337
Garousi, M.R. B584 (2000) 284 Holland, K. B583 (2000) 331
Gates Jr., S.J. B584 (2000) 109 Hollik, W. B581 (2000) 34
Gavai, R.V. B586 (2000) 475 Hollowood, T.J. B581 (2000) 309
Geyer, B. B581 (2000) 341 Hollowood, T.J. B584 (2000) 359
Gherghetta, T. B586 (2000) 141 Hong, D.K. B582 (2000) 451
Ghoshal, D. B584 (2000) 300 Hormuzdiar, J. B581 (2000) 391
Girotti, H.O. B587 (2000) 299 Hosomichi, K. B589 (2000) 134
Gitman, D.M. B584 (2000) 615 Hou, B.-Y. B586 (2000) 711
Glover, E.W.N. B585 (2000) 763 Howe, P.S. B581 (2000) 523
Godbole, R. B581 (2000) 3 Howe, P.S. B587 (2000) 481
Gomes, M. B587 (2000) 299 Hsu, S.D.H. B581 (2000) 391
Gonnella, G. B583 (2000) 584 Huber, P. B585 (2000) 105
González-Sprinberg, G.A. B582 (2000) 3 Huber, P. B588 (2000) 101
Gorini, V. B581 (2000) 575 Huiszoon, L.R. B584 (2000) 705
Görlich, L. B582 (2000) 44 Hull, C.M. B583 (2000) 237
Gorsky, A. B584 (2000) 197
Govindarajan, T.R. B583 (2000) 291 Inoue, R. B581 (2000) 761
Graham, N. B585 (2000) 443 Intriligator, K. B581 (2000) 257
Grandi, N. B588 (2000) 508 Irvine, S.E. B584 (2000) 795
Grandjean, O. B583 (2000) 381 Ishibashi, M. B587 (2000) 419
Greene, B.R. B584 (2000) 480 Ishibashi, N. B583 (2000) 159
Gregori, A. B588 (2000) 178 Ishibashi, N. B588 (2000) 149
Grimus, W. B587 (2000) 45 Iso, S. B583 (2000) 159
Grojean, C. B584 (2000) 359 Ito, K. B586 (2000) 231
Groot Nibbelink, S. B588 (2000) 57 Ivanov, E. B587 (2000) 445
Grosse, H. B587 (2000) 568 Ivanov, N.Ya. B586 (2000) 382
Grzadkowski, B. B583 (2000) 49
Grzadkowski, B. B585 (2000) 3 Jacob, P. B587 (2000) 514
Guadagnini, E. B588 (2000) 436 Jaffe, R.L. B585 (2000) 443
Guan, X.-W. B583 (2000) 721 Jamin, M. B587 (2000) 331
Guimarães, M.E.X. B581 (2000) 165 Janik, R.A. B586 (2000) 163
Gukov, S. B584 (2000) 69 Jejjala, V. B589 (2000) 196
Gukov, S. B584 (2000) 109 Jurčo, B. B584 (2000) 784
Gunion, J.F. B583 (2000) 49
Guruswamy, S. B583 (2000) 475 Kaidalov, A.B. B586 (2000) 382
Kakushadze, Z. B589 (2000) 75
Harlander, R.V. B586 (2000) 56 Kamimura, K. B585 (2000) 219
Harmark, T. B585 (2000) 567 Kaminsky, K. B581 (2000) 240
Harris, M.G. B586 (2000) 518 Kanno, H. B586 (2000) 331
Harshman, N.L. B581 (2000) 91 Karakhanyan, D. B583 (2000) 691
Hart, A. B586 (2000) 443 Kaste, P. B582 (2000) 203
Hashimoto, K. B587 (2000) 207 Kausch, H.G. B583 (2000) 513
Hassan, S.F. B583 (2000) 431 Kawai, H. B583 (2000) 159
Heinrich, G. B585 (2000) 741 Kaya, A. B583 (2000) 411
Heinzl, T. B584 (2000) 589 Kaya, A. B587 (2000) 481
Heitger, J. B588 (2000) 377 Kazakov, V. B587 (2000) 645
Helayël-Neto, J.A. B581 (2000) 165 Kazama, Y. B584 (2000) 171
Heller, U.M. B583 (2000) 347 Ketov, S.V. B582 (2000) 95
Herdeiro, C.A.R. B582 (2000) 363 Ketov, S.V. B582 (2000) 119
Hikami, K. B581 (2000) 761 Keurentjes, A. B589 (2000) 440
Hikida, Y. B589 (2000) 134 Khuri, R.R. B588 (2000) 253
Hill, C.T. B589 (2000) 249 Kiem, Y. B586 (2000) 303
Hioki, Z. B585 (2000) 3 Kikukawa, Y. B584 (2000) 511
Kim, J.E. B582 (2000) 296 Losada, M. B587 (2000) 118

Kimura, Y. B581 (2000) 295 Low, I. B585 (2000) 395
Kinar, Y. B583 (2000) 76 Lü, H. B584 (2000) 149
Kirschner, R. B583 (2000) 691 Lü, H. B586 (2000) 275
Kiselev, V.V. B581 (2000) 432 Lucas, P. B584 (2000) 719
Kiselev, V.V. B585 (2000) 353 Ludwig, A.W.W. B583 (2000) 475
Kitazawa, N. B586 (2000) 261 Lunardini, C. B583 (2000) 260
Kitazawa, Y. B581 (2000) 295 Lunardini, C. B584 (2000) 459
Kitazawa, Y. B583 (2000) 159 Lüscher, M. B582 (2000) 762
Klein, B. B588 (2000) 483 Lüst, D. B582 (2000) 44
Klemm, D. B587 (2000) 277 Lüst, D. B589 (2000) 269
Kniehl, B.A. B582 (2000) 514 Lütken, C.A. B582 (2000) 203
Kogan, I.I. B584 (2000) 313
Kogan, I.I. B588 (2000) 213 Macfarlane, A.J. B581 (2000) 743
Kogan, I.I. B589 (2000) 167 Mahnke, N. B589 (2000) 381
Kogut, J.B. B582 (2000) 477 Mangano, M.L. B582 (2000) 759
Körs, B. B582 (2000) 44 Martins, M.J. B583 (2000) 721
Kosmas, T.S. B587 (2000) 25 Maru, N. B586 (2000) 261
Kotikov, A.V. B582 (2000) 19 Marucho, M. B583 (2000) 454
Kovacs, S. B584 (2000) 216 Mathieu, P. B587 (2000) 514
Kovalenko, S. B587 (2000) 25 Mathur, S.D. B587 (2000) 249
Kovalsky, A.E. B585 (2000) 353 Mathur, S.D. B589 (2000) 38
Kramer, G. B582 (2000) 514 Matusis, A. B589 (2000) 38
Krauth, W. B584 (2000) 641 McDonald, J. B582 (2000) 763
Krykhtin, V.A. B584 (2000) 615 Melnikov, V.N. B584 (2000) 436
Kühn, J.H. B586 (2000) 56 Merten, C. B584 (2000) 46
Kurosawa, K. B584 (2000) 3 Mesref, L. B589 (2000) 337
Kyae, B. B582 (2000) 296 Metsaev, R.R. B586 (2000) 183
Mihailescu, M. B587 (2000) 179
Laine, M. B582 (2000) 277 Mikhailov, A. B584 (2000) 545
Laine, M. B586 (2000) 443 Miramontes, J.L. B581 (2000) 643
Langacker, P. B586 (2000) 287 Misiak, M. B586 (2000) 397
Langmann, E. B587 (2000) 568 Modanese, G. B588 (2000) 419
Lazar, M. B581 (2000) 341 Moeller, N. B583 (2000) 105
LeClair, A. B583 (2000) 475 Montagna, G. B584 (2000) 459
Le Doussal, P. B588 (2000) 565 Montes, X. B582 (2000) 259
Lee, H.M. B582 (2000) 296 Morariu, B. B588 (2000) 521
Lee, J. B586 (2000) 427 Morel, A. B581 (2000) 477
Lee, J. B589 (2000) 119 Moschella, U. B581 (2000) 575
Lee, S. B586 (2000) 303 Moschella, U. B587 (2000) 619
Leigh, R.G. B589 (2000) 196 Moultaka, G. B581 (2000) 34
Lerche, W. B582 (2000) 203 Mouslopoulos, S. B584 (2000) 313
Lerda, A. B586 (2000) 206 Mukhopadhyay, B. B582 (2000) 627
Lesgourgues, J. B582 (2000) 593 Müller, B. B582 (2000) 537
Li, G.-L. B586 (2000) 711 Müller, D. B589 (2000) 611
Li, T. B582 (2000) 176 Muramatsu, T. B584 (2000) 171
Liivat, H. B588 (2000) 90
Likhoded, A.K. B585 (2000) 353 Naganuma, M. B586 (2000) 231
Lindner, M. B585 (2000) 105 Nakamura, A. B584 (2000) 528
Lindner, M. B588 (2000) 101 Nastase, H. B581 (2000) 179
Lipatov, L.N. B582 (2000) 19 Nastase, H. B583 (2000) 211
Logan, H.E. B586 (2000) 39 Natsuume, M. B588 (2000) 453
Loide, R.-K. B588 (2000) 90 Nelson, B. B588 (2000) 197
López, A. B582 (2000) 716 Nepomechie, R.I. B586 (2000) 611
Losada, M. B582 (2000) 277 Nersesyan, A.A. B583 (2000) 671
Losada, M. B585 (2000) 45 Niclasen, R. B583 (2000) 347
Nicrosini, O. B584 (2000) 459 Quirós, M. B581 (2000) 61

Niedermaier, M. B583 (2000) 614 Quirós, M. B583 (2000) 35
Niedermayer, F. B583 (2000) 614
Nielsen, H.B. B588 (2000) 281 Radu, E. B585 (2000) 637
Nierste, U. B586 (2000) 39 Rahimi Tabar, M.R. B588 (2000) 630
Nishimura, M. B588 (2000) 471 Rajesh, G. B587 (2000) 228
Nishino, H. B568 (2000) 491 Ramgoolam, S. B587 (2000) 179
NOMAD Collaboration B588 (2000) 3 Rasmussen, J. B582 (2000) 649
Nomura, Y. B584 (2000) 3 Rastelli, L. B589 (2000) 38
Nurmagambetov, A. B586 (2000) 315 Ravindran, V. B586 (2000) 349
Ravindran, V. B589 (2000) 507
Oda, H. B586 (2000) 231 Recknagel, A. B583 (2000) 381
Oeckl, R. B581 (2000) 559 Recknagel, A. B588 (2000) 552
Ohsawa, A. B581 (2000) 73 Reinhardt, H. B585 (2000) 591
Okada, N. B586 (2000) 261 Reisz, T. B581 (2000) 477
Okawa, Y. B584 (2000) 329 Remiddi, E. B581 (2000) 274
Okuyama, K. B588 (2000) 149 Restrepo, D.A. B583 (2000) 182
Oleari, C. B585 (2000) 763 Rétey, A. B583 (2000) 3
Oller, J.A. B587 (2000) 331 Ricciardi, G. B587 (2000) 363
Onishchenko, A.I. B581 (2000) 432 Rius, N. B587 (2000) 118
Osterloh, A. B588 (2000) 531 Rivelles, V.O. B587 (2000) 299
Robaschik, D. B581 (2000) 449
Ots, I. B588 (2000) 90
Roggenkamp, D. B588 (2000) 552
Ovrut, B.A. B589 (2000) 269
Rosier-Lees, S. B581 (2000) 3
Özpineci, A. B585 (2000) 275
Ross, G.G. B584 (2000) 313
Rossi, A. B587 (2000) 3
Pakman, R.L. B588 (2000) 508
Rossi, G. B584 (2000) 216
Palla, L. B587 (2000) 585
Roth, M. B587 (2000) 67
Panda, S. B584 (2000) 251
Rubbia, A. B589 (2000) 577
Papazoglou, A. B584 (2000) 313 Ruelle, P. B581 (2000) 679
Paschos, E.A. B588 (2000) 263 Rühl, W. B589 (2000) 337
Pasquali, L. B588 (2000) 263 Rummukainen, K. B583 (2000) 347
Patrascioiu, A. B583 (2000) 614 Russo, R. B582 (2000) 65
Paul, P.L. B581 (2000) 156 Russo, R. B585 (2000) 193
Pelizzola, A. B583 (2000) 584 Rychkov, V. B581 (2000) 116
Pépin, C. B586 (2000) 641
Pérez-Victoria, M. B589 (2000) 3 Saar, R. B588 (2000) 90
Pernici, M. B582 (2000) 733 Sabra, W.A. B587 (2000) 277
Pershin, V.D. B584 (2000) 615 Sadrzadeh, A. B586 (2000) 275
Peschanski, R. B586 (2000) 163 Saito, T. B584 (2000) 528
Petersson, B. B581 (2000) 477 Sakai, N. B586 (2000) 231
Petkou, A.C. B586 (2000) 547 Sakai, S. B584 (2000) 528
Petrov, K. B581 (2000) 477 Samuel, S. B585 (2000) 715
Philipsen, O. B586 (2000) 443 Santamaria, A. B582 (2000) 3
Piccinini, F. B584 (2000) 459 Santiago, J. B584 (2000) 313
Pich, A. B587 (2000) 331 Santoso, Y. B585 (2000) 124
Pickering, A. B581 (2000) 523 Satoh, Y. B588 (2000) 149
Pilaftsis, A. B586 (2000) 92 Satz, H. B583 (2000) 368
Pliszka, J. B583 (2000) 49 Sauser, R. B582 (2000) 231
Plyushchay, M.S. B589 (2000) 413 Savcı, M. B585 (2000) 275
Poghossian, R. B588 (2000) 638 Savvidy, K.G. B585 (2000) 567
Polyakov, A. B581 (2000) 116 Schaeffer, R. B581 (2000) 575
Pomarol, A. B586 (2000) 141 Schaeffer, R. B587 (2000) 619
Pope, C.N. B584 (2000) 149 Schäfer, A. B582 (2000) 537
Pope, C.N. B586 (2000) 275 Schalm, K. B584 (2000) 480
Pötter, B. B582 (2000) 514 Schaposnik, F.A. B582 (2000) 716
Schaposnik, F.A. B588 (2000) 508 Tan, C.-I. B587 (2000) 249
Scharnhorst, K. B581 (2000) 718 Tanaka, T. B582 (2000) 259
Schellekens, A.N. B584 (2000) 705 Taylor, W. B583 (2000) 105
Schlottmann, P. B586 (2000) 686 Taylor, W. B585 (2000) 171
Schmidhuber, C. B585 (2000) 385 Tekin, B. B588 (2000) 213
Schnabl, M. B589 (2000) 461 Tekin, B. B589 (2000) 167
Schomerus, V. B583 (2000) 381 Terashima, S. B584 (2000) 329
Schomerus, V. B588 (2000) 552 Tok, T. B584 (2000) 589
Schreiber, E. B583 (2000) 76 Tomasiello, A. B589 (2000) 461
Schubert, C. B585 (2000) 407 Toublan, D. B582 (2000) 477
Schubert, C. B585 (2000) 429 Tran, T.A. B586 (2000) 275
Schupp, P. B584 (2000) 784 Trentadue, L. B583 (2000) 307
Schvellinger, M. B588 (2000) 213 Tricarico, E. B585 (2000) 253
Schweigert, C. B588 (2000) 110 Trigiante, M. B582 (2000) 393
Schwetz, M. B581 (2000) 391 Trittmann, U. B587 (2000) 311
Schwetz, T. B587 (2000) 45 Troost, J. B581 (2000) 135
Sciuto, S. B585 (2000) 193 Tseytlin, A.A. B584 (2000) 233
Sedrakyan, A. B588 (2000) 638 Tsurumaru, T. B589 (2000) 659
Seiler, E. B583 (2000) 614 Tsutsui, I. B589 (2000) 659
Semenoff, G.W. B582 (2000) 155 Tsvelik, A.M. B586 (2000) 641
Sen, A. B584 (2000) 300
Sen, A. B587 (2000) 147 Uibo, H. B588 (2000) 90
Sevrin, A. B581 (2000) 135 Urban, J. B586 (2000) 397
Sezgin, E. B587 (2000) 481
Sheikh-Jabbari, M.M. B587 (2000) 195 Vafa, C. B584 (2000) 69
Sheikh-Jabbari, M.M. B589 (2000) 461 Vaidya, S. B583 (2000) 291
Shirman, Y. B581 (2000) 309 Vainshtein, A. B584 (2000) 197
Shiu, G. B584 (2000) 480 Valle, J.W.F. B583 (2000) 182
Shore, G.M. B581 (2000) 409 Vaman, D. B581 (2000) 179
Simón, J. B585 (2000) 219 Vaman, D. B583 (2000) 211
Smirnov, A.Yu. B583 (2000) 260 van der Bij, J.J. B585 (2000) 637
Smoller, J. B584 (2000) 387 van Holten, J.W. B588 (2000) 57
Sokatchev, E. B581 (2000) 523 van Neerven, W.L. B586 (2000) 349
Sommer, R. B582 (2000) 762 van Neerven, W.L. B588 (2000) 345
Sommer, R. B588 (2000) 377 van Neerven, W.L. B589 (2000) 507
Sonnenschein, J. B583 (2000) 76 van Nieuwenhuizen, P. B581 (2000) 179
Sonoda, H. B585 (2000) 725 Vasiliev, M.A. B586 (2000) 183
Sorba, P. B588 (2000) 638 Verbaarschot, J.J.M. B582 (2000) 477
Sorokin, D. B586 (2000) 315 Verbaarschot, J.J.M. B586 (2000) 668
Stanev, Y.S. B584 (2000) 216 Verbaarschot, J.J.M. B588 (2000) 483
Staudacher, M. B584 (2000) 641 Verbeni, M. B583 (2000) 307
Stefański jr., B. B589 (2000) 292 Vergados, J.D. B587 (2000) 25
Stein, E. B582 (2000) 537 Verlinde, H. B581 (2000) 156
Stein, E. B589 (2000) 381 Vernizzi, G. B583 (2000) 739
Stephanov, M.A. B582 (2000) 477 Vidal, J. B582 (2000) 3
Strigazzi, P. B586 (2000) 206 Visser, M. B584 (2000) 415
Strumia, A. B585 (2000) 28 Vogt, A. B588 (2000) 345
Sugawara, Y. B589 (2000) 134 Volkov, M.S. B582 (2000) 313
Sundell, P. B587 (2000) 481
Suneeta, V. B583 (2000) 291 Wackeroth, D. B587 (2000) 67
Suzuki, H. B585 (2000) 471 Wagner, C.E.M. B586 (2000) 92
Wágner, F. B587 (2000) 585
Takács, G. B587 (2000) 585 Walcher, J. B582 (2000) 203
Takanishi, Y. B588 (2000) 281 Walcher, J. B588 (2000) 110
Talavera, P. B585 (2000) 293 Walton, M.A. B584 (2000) 795
Tamada, M. B581 (2000) 73 Wang, Y.-J. B583 (2000) 671
Weigel, H. B585 (2000) 443 Youm, D. B589 (2000) 315

Weiss, N. B583 (2000) 76 Yu, J.Y. B588 (2000) 263
Weisz, P. B583 (2000) 614 Yue, R.-H. B586 (2000) 711
Wess, J. B584 (2000) 784 Yung, A. B584 (2000) 197
West, P.C. B581 (2000) 523
Wetterich, C. B587 (2000) 403
Wheater, J.F. B586 (2000) 518 Zanon, D. B587 (2000) 277
White, B.E. B581 (2000) 409 Zarembo, K. B582 (2000) 155
Wiedemann, U.A. B582 (2000) 409 Zayas, L.A.P. B582 (2000) 216
Wiedemann, U.A. B588 (2000) 303 Zee, A. B585 (2000) 395
Wipf, A. B582 (2000) 313 Zhang, R.-J. B586 (2000) 3
Wipf, A. B584 (2000) 589 Zhitnitsky, A. B582 (2000) 477
Witten, E. B584 (2000) 69 Zumino, B. B588 (2000) 521
Witten, E. B584 (2000) 109
Zvyagin, A.A. B586 (2000) 686
Wittig, H. B582 (2000) 762
Zwanziger, D. B581 (2000) 604
Wittig, H. B588 (2000) 377
Zwiebach, B. B587 (2000) 147
Yau, S.-T. B584 (2000) 387 Zwirner, F. B582 (2000) 759

Nucl - Phys.B v.589

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Nucl - Phys.B v.589

Caricato da

Copyright:

Formati disponibili

Nuclear Physics B 589 (2000) 3–37

Near-extremal correlators and vanishing

Received 11 April 2000; accepted 17 August 2000

E-mail addresses: dhoker@physics.ucla.edu (E. D’Hoker), jke@mitlns.mit.edu (J. Erdmenger),

2. The correlator hO4 O2 O2 O2 O2 i

2.1. hO4 O2 O2 O2 O2 i in SYM

Fig. 2. Graphs contributing to the correlator hO4 O2 O2 O2 O2 i at order g 2 .

2.2. hO4 O2 O2 O2 O2 i in AdS

Fig. 3. Witten diagrams contributing to E52 correlators in the AdS/CFT correspondence.

2 The normalization for k = 2 is in agreement with the discussion in [34].

Finally, there is a contact contribution given by diagram e of Fig. 3:

3. General near-extremal correlators in field theory

In this section we study near-extremal correlation functions in N = 4 SYM to order g 2

Fig. 6. Feynman graphs contributing to an E52 function at order g 2 .

B-term ∼ {a}{b}{d}{[a1, p]e}{e}

Fig. 7. Feynman graphs contributing to an E62 function at the free-field level.

En function, m 6 n − 2, is a sum of terms that factor into at least n − m − 1 pieces. The

Fig. 9. Relevant part of a general order-g 2 correction to a T graph.

4. General near-extremal correlators in AdS

5. Vanishing near-extremal supergravity couplings

are allowed by SU(4)R . We distinguish three possibilities:

We study the three cases in turn.

The main results of this paper are

5. The correspondence would be spoiled by non-factored supergravity contact diagrams

A.1. AdS integrals

A.2. Examples of five-point integrals

hO∆ (x)O∆1 (x1 )O∆2 (x2 )O∆3 (x3 )O∆4 (x4 )i

y0 −5 y0 4−∆ y0 ∆1 y0 δ = y0 δ+∆1 −∆−1

A.3. General AdS integrals

Fig. 13. Typical exchange diagram.

where Rb and Rb0 are regular in ε and

sub-extremal the region y ∼ x of the integral over y gives a vanishing contribution.

[1] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.

The operator product expansion of N = 4 SYM and

Received 18 April 2000; revised 3 August 2000; accepted 11 August 2000

1 Perturbative studies of 4-point functions in N = 4 SYM include [34–43].

Fig. 1. Scalar exchange in the ‘t-channel’.

5 A similar result has been obtained in [21] in a rather different formalism.

2. Supergravity exchanges versus OPE: power singularities

There is an intriguing relation [17,18] between supergravity exchange diagrams and

3. On the logarithmic behavior of conformal field theories

3.1. General analysis

11 These statements can be proved by the methods reviewed in Appendix A.

[Oij (x)]y ≡ Oi (x + y)Oj (x) − subtractions (i 6= j ). (3.14)

[Oα ]x = Vα(A) [O(A) ]x

Oi (x)Oj (0) ∼ singular terms + [Oα (0)]y + Mαβ log(x/y)[Oβ (0))]y ,

3.2. Logarithms in the supergravity picture

The AdS/CFT correspondence provides an interesting dual picture of the Yang–Mills

Fig. 3. Quartic graph.

z0∆ z0∆ z02∆

If we set x1 , x2 to zero the integral becomes

4. Dilaton–axion four point functions and anomalous dimensions

4.1. The use of U (1)Y symmetry and anomalous dimensions

The automorphism group U (1)Y of the conformal N = 4 supersymmetry algebra

SYM operator Desc SUGRA Dim Spin Y SU(4) Lowest reps

Ok ∼ tr Xk , k > 2 – hαα aαβγ δ k (0, 0) 0 (0, k, 0) 200 , 50, 105

E. D’Hoker et al. / Nuclear Physics B 589 (2000) 38–74

4.2. Comparison with large N SYM calculations

To tackle these issues it might be more convenient to consider a 4-point function of

It is a pleasure to thank Dan Freedman for very enjoyable collaboration in previous

Appendix A. Summary of supergravity 4-point functions

−Dσ Dσ Gµνµ0 ν 0 − Dµ Dν Gσ σ µ0 ν 0 + Dµ Dσ Gσ νµ0 ν 0

A.1. Explicit form of quartic graphs and series expansion

(a) t-channel limit: |x13 | |x12|, |x24 | |x12|, which corresponds to s, t → 0;