Calculation of Nonperturbative Terms in Open String Models: Julie D. Blum

Nuclear Physics B 634 (2002) 322
www.elsevier.com/locate/npe
Calculation of nonperturbative terms in open string

models
Julie D. Blum
Physics Department, University of Texas at Austin, Austin, TX 78712, USA
Received 19 December 2001; accepted 9 April 2002
Abstract
Nonperturbative corrections in type II string theory corresponding to Riemann surfaces with one
boundary are calculated in several noncompact geometries of desingularized orbifolds. One of these
models has a complicated phase structure which is explored. A general condition for integrality of
the numerical invariants is discussed. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
String theory has provided mathematicians with many interesting conjectural results
that would be in some cases significantly more difficult to obtain through traditional
techniques. Calculations that count the number of maps from Riemann surfaces into
CalabiYau manifolds are one example. Recently, these calculations have been extended
to Riemann surfaces with boundaries. The addition of boundaries for a generic type II
string theory reduces the supersymmetry to N = 1 in four dimensions, and the counting of
maps corresponds to holomorphic terms in the field theory. Such calculations are possibly
relevant to an extension of the Standard Model.
In the following we will calculate the nonperturbative terms in a type II string
theory generated by Riemann surfaces with one boundary, disk instantons. Most of the
techniques that will be employed here are discussed in [1,2], and other places. In honor
of human decency, no additional references will be mentioned. We will focus on three
noncompact models, the blowups of the Z2 Z2 , Z2 Z4 , and Z7 orbifolds of C3 . The
three models support the conjecture that numerical invariants related mathematically to the
Euler characteristic of the moduli space of open string instantons and physically to the
E-mail address: julie@zippy.ph.utexas.edu (J.D. Blum).
0550-3213/02/$ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 8 6 - 9
J.D. Blum / Nuclear Physics B 634 (2002) 322
counting of domain walls are integral in phases where Khler parameters have a geometric
interpretation. The second model has an intricate moduli space with many phases. We
explore the phase structure of this model and see how it reduces in various limits to simpler
models. The complications of this model necessitated an understanding of what conditions
ensure integrality of the numerical invariants. We explain why the geometric phases always
give integers and the conditions under which fractions are possible.
2. Z2 Z2
2.1. Toric geometry
The toric geometry for this model can be described by a linear sigma model (twodimensional abelian gauge theory with N = 2 supersymmetry). There are six chiral fields
carrying the following charges under three U (1) gauge fields.
l1 = (1, 0, 0, 1, 1, 1)
l2 = (0, 1, 0, 1, 1, 1)
l3 = (0, 0, 1, 1, 1, 1)
The CalabiYau condition requires
D-terms in the gauge theory.
j
li |xj |2 = ri .
j
j li
(2.1)
= 0. From the set of charges, one derives the
(2.2)
These equations can be solved leading to the toric diagram, shown in Fig. 1.
The diagram is a projection of three dimensions onto the plane, and the angles shown
are not meant to be accurate. Generically, the diagram represents a T3 fibration. It shrinks
Fig. 1. Toric diagram of Z2 Z2 blowup.
Fig. 2. Toric diagram of Z2 Z2 flopped phase: r3 r3 .
to a two-torus along planes xi = 0, to a circle along lines xi = xj = 0, and to a point

at the intersection of three lines. From the diagram one sees that there are three twospheres and six noncompact two-cycles. Since this diagram resembles the intersection of
three conifolds, one would expect other phases that replace a P1 by an S3 . One can easily
visualize a flopped phase as where r3 r3 in Fig. 2.
The equation for the Z2 Z2 singularity can be obtained from gauge invariant
combinations of chiral fields as w2 = xyz where w, x, y, and z are complex variables.
To show explicitly that there are three P1 s, one should solve the PicardFuchs
equations. These equations determine the Khler parameters for the P1 s corrected by
worldsheet
instantons or equivalently complex parameters in the local mirror geometry,

i.e., i , where is the holomorphic three-form and i is a three-cycle. For this model
the equations are

2

2
=0
1 (1 2 3 ) z1 1 (2 3 )
(2.3)
i
and cyclic permutations where i = zi zi and zi = eti with ti = ri + ii , the initial

parameter in the linear sigma model. One
Khler parameter. Here i is a FayetIliopoulos

takes linear combinations of the i normalized appropriately to obtain ti , the instanton
corrected Khler parameters. For the calculation here we need the inverse solutions which
turn out to be
q1 (1 + q2 q3 )
z1 =
(2.4)
(1 + q1 q2 )(1 + q1 q3 )
and permutations where qi = eti . Note that in finding unique solutions for the Picard
Fuchs equations, we have frequently had to change to a different basis of P1 s. The
simplicity of these solutions makes this model amenable to obtaining exact results without
great labor.
2.2. The mirror and open string amplitudes
The equation for
the mirror is readily derived from Re(yi ) = |xi |2 , the D-term
equations, and xz = i eyi where x and z are complex variables. Setting y5 = u, y6 = v,
and y4 = 0 to fix a constant solution of the D-term equations yields

xz = P (u, v) = 1 + eu + ev + et1 +u+v + et2 +vu + et3 +uv .
(2.5)
A noncompact, supersymmetric Lagrangian three-cycle in the original manifold is

determined by three additional constraints which in this case take the form
|x5 |2 |x4 |2 = c1 ,
|x6 |2 |x4 |2 = c2 ,

Arg(xi ) = 0 and/or .
(2.6)
Let us write the above in the form

(1)j
lp |xj |2 |3-cycle = cp
(2.7)
and

l (2)i Im log(xi |3-cycle ) = 0 and/or
(2.8)
(1)i (2)i
(1)i
for later use. Here
= 0 makes the cycle Lagrangian and
l
= 0 is
i lp l
i (2)ii p
= 0 implies
necessary for supersymmetry of the Lagrangian cycle. Additionally, i lj l
that (2.8) is gauge invariant.If the cycle does not
intersect a compact or noncompact
two-cycle in the base, both i Arg(xi ) = 0 and i Arg(xi ) = are needed to give a
composite cycle without boundary. For this case any worldsheet disks that intersect a Dbrane wrapped on the two cycles will be oppositely oriented with respect to the two cycles
and
If the cycle intersects the toric base, one can choose either

not make a contribution.
Arg(x
)
=
0
or
Arg(x
i
i ) = to get a cycle without boundary, and there generally
i
i
will be a nonvanishing contribution from disks wrapping part of a P1 and intersecting the
cycle in a circle. Allowing the three-cycle to end on the P1 where x6 = x4 = 0 (which
will be denoted as phase I), the classical limit in the mirror corresponds to v = i ,
Re u = c1 r2 2 , and xz = 0. More generally, one can choose the mirror twocycle of the Lagrangian three-cycle to be parametrized by z with x = P (u, v) = 0, a
Riemann surface which is the moduli space of this cycle. The coordinates u and v can
be considered as transverse coordinates to a two-cycle inside a CalabiYau manifold.
The disk amplitude Fg=0,h=1 (g is the genus, h is the number of boundaries) can
be determined classically in the mirror as u F0,1 = v where classically v = 0 and u
parametrizes the area of a disk. In the original manifold, these disks can be interpreted
as domain walls (e.g., fourbranes wrapping a disk) ending on a sixbrane wrapped on the
three-cycle. The tension of these domain walls is corrected by an amount u that must be
added to the classical area of a disk. The amplitude F0,1 takes the form

1
mi n
eukn
Nk,m
qi
,
F0,1 =
(2.9)
n2
k,n,m
where k, n, mi are integers, u is the instanton corrected domain wall tension, and
Nk,m
counts the number of domain walls wrapping the two-cycle parametrized by i mi ti
with a boundary wrapping the S1 k times. The assumption is that one counts isolated
domain walls and that the Nk,m
should be integers. If one has a continuous family of
domain walls, fractions may be possible. In section three we will determine a criterion
for obtaining integers and present an argument for integrality in those cases. Under
mirror symmetry a domain wall fourbrane becomes a domain
wall fivebrane with tension

1
determined classically by u = u u where u = 2i
Cu u dv and v v + 2i around
the one-cycle on the Riemann surface Cu . This tension corresponds to the difference in the
superpotential F0,1 as one crosses the domain wall. In the original manifold, the change in
Imv corresponded to the change in Wilson line as one crossed the domain wall.
There is an ambiguity in F0,1 due to the possibility of redefining the disk coordinate
u u + nv where n is an integer since v = 0 classically. One requires n to be an integer
so that eu is invariant under u u + 2i. This ambiguity can sometimes be related to
moving the Lagrangian cycle to a different phase along the toric base, and the amplitude
F0,1 is not invariant. If the toric base is modeled by type IIB fivebranes, the ambiguity
corresponds to an SL(2, Z) transformation of type IIB.
Proceeding with the calculation, the Riemann surface P (u, v) = 0 looks like Fig. 3
where the legs extend to infinity.
There are nine phases, but the three compact P1 s and the six noncompact two-cycles
of the original manifold are related by symmetries reducing the number of inequivalent
phases to two (Fig. 4).
For instance, the three inner phases are exchanged by exchanging z3 z2 z1 along
with v u u, and phase II and phase III are exchanged under z1 z3 , v v.
Phase II corresponds to c1 = c2 + r2 , c2 0.
In phase I one obtains

v=i ln 2 1 + z1 eu + z2 eu + ln 1 + eu

2

+ 1 + eu 4z3 (eu + z1 e2u + z2 ) .
(2.10)
Fig. 3. Riemann surface P (u, v) = 0 in mirror of Z2 Z2 blowup.
Fig. 4. Phases for noncompact three-cycle in Z2 Z2 blowup.
Extracting the piece of v that is independent of eu gives v, and one has

t1 t1 t3 t3
+ i.
2
2
To get u, we exchange t3 and t2 so
v =
(2.11)
t1 t1 t2 t2
+ i.
(2.12)
2
2
In the above equation for v, we must substitute, v = v + v, u = u + u, and zi (qi ).
Note that not all of the classical symmetries of the superpotential are preserved by the
corrections, and one cannot determine the correction uniquely by symmetry. The result is

v = ln 1 q1 eu 1 q2 eu

(1)m+b+c
u =
m,n,a,b,c,d,e
q1a+d+e q2m+nabc+d q3m+nc+e

(2n + m 1)!eu(2a+b+c)
.
n!a!(c e)!e!(b d)!(m c)!d!(n a b)!
(2.13)

The indicates that we omit terms independent of eu . Clearly the first term of v has
the required form. The lowest order terms of the second summation can be examined by
hand or calculated on a computer using Mathematica, and one does obtain Nk,m
that are
integers after integrating and comparing with (2.9). One can show explicitly that all N1,m
are integers. One finds

N1,0,m,m =
m

(1)n+m (n + m 1)!m
n=1
N1,1,m1,m =
m1

n=0
n!2 (m n)!
(1)n+m (n + m 1)!
,
n!2 (m n 1)!
Table 1
m3 = 5, k = 5
m2
m1 = 0
0
1
2
3
4
5
5
126
756
1764
1764
635
14
350
2100
4900
4900
1764
14
350
2100
4900
4900
1764
6
150
900
2100
2100
756
1
25
150
350
350
126
0
1
6
14
14
5
Table 2
m3 = 5, k = 6
m2
m1 = 0
0
1
2
3
4
5
42
630
2940
5880
5292
1764
126
1890
8820
17640
15876
5292
140
2100
9800
19600
17640
5880
70
1050
4900
9800
8820
2940
15
225
1050
2100
1890
630
1
15
70
140
126
42
Table 3
m3 = 5, k = 7
m2
m1 = 0
0
1
2
3
4
5
198
2310
9240
16632
13860
4356
630
7350
29400
52920
44100
13860
756
8820
35280
63504
52920
16632
420
4900
19600
35280
29400
9240
105
1225
4900
8820
7350
2310
9
105
420
756
630
198
and
N1,0,m1,m = N1,1,m,m =
m

(1)n+m (n + m 1)!
n=1
n!(n 1)!(m n)!
We have verified that all Nk,m1 ,m2 ,m3 are integers for k 10, m1 1, m2 1, and m3 10.
Rather than present this data which is very cumbersome, we give Tables 13 with k and
m3 set to specific values.
The diagonal symmetry Nk,m1 x,m2 y,m1 +m2 = Nk,m1 +y,m2 +x,m1 +m2 is generated by
d b d in (2.13). The similar symmetry
Nk,m1 x,m1 +m3 k,m3 y = Nk,m1 +y,m1 +m3 k,m3 +x
is generated by e c e.
In phase II we do the coordinate transformation u u = u v + t2 , v v = v.
Correspondingly, u = u v + t2 = t3 2 t3 + t2 2 t2 and v = v. Phase II is almost

equivalent to phase I. One obtains phase II from (2.13) by ignoring the first log term,
10
exchanging q2 q3 and u v.
The first term is similar to the inner phase of a conifold
so it looks like the S3 symmetry relating the inner and two outer phases is broken by the
finite P1 s.
One can extract amplitudes in the flopped phase of Fig. 2 by taking q3 q3 = 1/q3 ,
q1 q1 q3 , q2 q2 q3 , and exp u q3 exp u in (2.13). As a check on our results, the
conifold in the two inequivalent phases is retrieved in the limit q1 , q3 0. Replacing
the P1 by an S3 in this limit via a conifold transition also yields the same result as the
calculation of the expectation value of a Wilson line in the ChernSimons theory on S3 . It
would be interesting to extend such calculations to the case of multiple S3 s.
3. Z2 Z4
3.1. Toric geometry
Let us move on to the Z2 Z4 case. Here we increase the complexity of the calculation,
but the results reduce precisely to the Z2 Z2 case in a particular limit. We start with the
following set of charges under six U (1)s for nine fields
l1 = (1, 1, 0, 0, 0, 1, 1, 0, 0),
l2 = (0, 1, 1, 0, 0, 1, 1, 0, 0),
l3 = (0, 1, 0, 0, 0, 1, 1, 0, 1),
l4 = (0, 0, 0, 0, 0, 1, 2, 1, 0),
l5 = (0, 0, 0, 1, 0, 0, 1, 1, 1),
l6 = (0, 0, 0, 1, 1, 0, 1, 1, 0),
(3.1)
j
yielding j li |xj |2 = ri . Solving these equations leads to the toric diagram (Fig. 5) in one
particular region of moduli space where r7 = r2 +r3 r5 > 0, r8 = 2r2 r3 +r4 +r5 > 0,
and all of the ri are large. The four-cycle represented by the hexagon is a P2 blown up in
succession at three points (one obtains inequivalent four-cycles depending on how one
does this). It is also equivalent to the F2 Hirzebruch surface blown up at two points. By
taking r4 and r8 to infinity, the above diagram and the theory reduce to two decoupled
Z2 Z2 cases. There are possible flop transitions to other geometric phases for r1 , r2 ,
r3 , r5 , r6 , and r7 but not for r4 and r8 . Shrinking r8 to a negative value removes a P1 as
|x3 | > 0 everywhere, and we enter a nongeometric phase where a Khler parameter loses its
correspondence to a geometric P1 . The result of flopping r7 is shown in Fig. 6. Two more
flops (r4 r2 r2 r4 , r3 r3 ) and taking all ri to infinity with r4 r3 > 0 and finite
reduces the diagram to the blowup of a Z3 orbifold (see Fig. 7). The flops r2 r2 and
r7 r7 generate a P1 P1 , and one can take external Khler parameters to infinity to
obtain this model. The equations for the Z2 Z4 singularity are v 2 = yu and v 4 = wzu2 .
Before solving the PicardFuchs equations, one must choose a basis that depends both
on the open as well as closed string phase. A priori there are many inequivalent basis
choices, but in this model the requirement that the open string expansion converge in a
particular phase limits the choices. Requiring the expansions in the zi to converge in a
11
Fig. 5. Toric diagram of Z2 Z4 blowup.
Fig. 6. Toric diagram of Z2 Z4 flopped phase: r7 r7 .
neighborhood of the origin yields a unique solution for the particular phase. By uniqueness
the solution solves the equations in any basis obtained from the original basis by a linear
transformation in which the transformation matrix has no negative entries. The solution
may not be unique in any particular basis. Also, even when the expansions are convergent
in a particular basis, there may be a correction which is nonperturbative with respect
to that basis and necessary for integrality. We have found a unique basis for this phase in
which the above problem does not occur. We need to define r9 = r6 r7 for this basis. The
12
Fig. 7. Toric diagram of Z2 Z4 flopped phase: r7 r7 , r4 r2 r2 r4 , r3 r3 .
solutions are as follows:

q1 eN+T
,
1 + q1 q3
z1 =
z2 = q2 (1 + q1 q3 )eN+MT
q3 eNT
z3 =
,
1 + q1 q3
z7 = q7 (1 + q2 q3 q9 )eP +MT ,
z8 = q8 eN+P 2M
q9 eM+2T
z9 =
,
(1 + q2 q3 q9 )2
= (t1 + t2 + 2t7 + 2t8 + t9 )2 + instanton corrections,
(3.2)
where is a solution corresponding to a four-cycle,

T=
(1)n+p+r+m (n + m + p r 2q 1)!z1r z2 z3m z7n z8s z9

m!q!r!(m p r + s)!(s n)!(p m r)!
m,n,p,q,r,s
1
,
(n 2q)!(p 2s + n q)!
N=
(1)s (2r s 1)!zr zr z2n zs zn

1 2 7 8 9
,
n!r!(s
2n)!(r
+
n 2s)!
r,n,s
M=
(1)n+r (2s n r 1)!zr zr z2n zs zn

1 2 7 8 9
,
r!n!(s
2r)!(s
2n)!
r,n,s
P=
(1)s (2n s 1)!zr zr z2n zs zn

1 2 7 8 9
n!r!(s
2r)!(r
+
n 2s)!
r,n,s
and
13
and the zi s must be found as a function of the qi s perturbatively.

Examining Fig. 5 one sees that there is a reflection symmetry about a line through the
equator of P1 (r4 ) and P1 (r8 ). The above solutions do not reflect this symmetry because any
choice of basis necessarily breaks this symmetry. In this case the selection of q3 breaks the
symmetry. We had previously chosen the basis with q6 instead of q9 and found that the open
string expansion did not give integers without the term (1 + q2 q3 q9 ) = (1 + q2 q3 q6 /q7 ),
but the perturbative solution, t6 , of the PicardFuchs equations does not converge if we
include this term. The pieces of the solution involving negative powers solve the Picard
Fuchs equations by themselves so there are ambiguities of the solution in general. Clearly,
the nonperturbative pieces are essential for integrality of the numerical invariants. We
also note that this model is the only one treated so far where the above ambiguity involving
negative powers occurs. In different phases of the theory we need to resolve the Picard
Fuchs equations.
Up to total order fourteen in the qi s (linear order in q9 ) we find that the coefficients
in the expansions of the inverse mirror map are integral. The expansions in this phase are
q1 q3
(z1 z3 = (1+q
)
q )2
1 3
z1 = q1 + q1 q2 q3 + q1 q2 q3 q8 + q1 q2 q7 q8 2q1q2 q3 q7 q8 + ,
z2 = q2 + q1 q2 q3 + q2 q8 + q1 q2 q3 q8 + ,
z7 = q7 q2 q3 q7 + q7 q8 + q1 q2 q7 q8 2q2 q3 q7 q8 + q2 q3 q7 q9
+ q2 q3 q7 q8 q9 + ,
z8 = q8 + q1 q2 q8 + ,
z9 = q9 2q2 q3 q9 q8 q9 + q1 q2 q8 q9 2q2q7 q8 q9
+ 4q2 q3 q7 q8 q9 2q1q2 q3 q7 q8 q9 + .
(3.3)
In the limit that z7 = z8 = z9 = 0,

= 1 + q1 q2 ,
= 1 + q2 q3 , and the solutions
are precisely those of the Z2 Z2 case. The limit z1 = z2 = z31 = z7 = z8 = z9 = 0
while z2 z3 z7 z8 is finite yields the solution for the blowup of the Z3 orbifold. Taking
z1 = z21 = z3 = z71 = z8 = z9 = 0 with z2 z3 and z2 z7 z8 finite yields the P1 P1 case.
eN
eT
3.2. The mirror and open string calculations

Putting y2 = u, y6 = v, and y7 = 0 the equation for the mirror is
xz = P (u, v) = 1 + eu + ev + eu+vt1 + euvt2 + evut3
+ evt2 t7 t8 + e2v+u2t2t8 + e3v+u3t22t7 2t8 t9 .
(3.4)
This choice is sensible for open string phases on the four-cycle. Notice again that this
equation reproduces the Z2 Z2 case when z7 = z8 = z9 = 0. To obtain the standard
version of the Z3 case, take v = v u t3 and the previously discussed limit. The standard
version of P1 P1 results from u = u v t2 and the above limit. The Riemann surface
P (u, v) = 0 can be visualized Fig. 8.
A noncompact, supersymmetric three-cycle intersecting the toric
base of the original
manifold is determined by |x2 |2 |x7 |2 = c1 , |x6|2 |x7 |2 = c2 , and i Arg(xi ) = 0. There
14
Fig. 8. Riemann surface P (u, v) = 0 in mirror of Z2 Z4 blowup.
Fig. 9. Phases for noncompact three-cycle in Z2 Z4 blowup.
are sixteen phases for the Lagrangian three-cycle with symmetries relating the phases in
the lower half of Fig. 9 to those in the upper half. For instance, the phase with a three-cycle
intersecting P1 (r7 ) is equivalent to phase I under the obvious permutation of the ri s and
v v. Also, phases II and III are equivalent. There are, accordingly, nine inequivalent
phases.
Five of these phases can be parametrized by u = 0 after a change of coordinates while
the rest correspond to v = 0. The latter yield quartic equations which can be solved and
expanded, but the amount of computer time required seems prohibitively large. Fortunately,
the u = 0 phases are quadratic. These are still quite complicated because the expansion
involves seven variables. We will, thus, restrict ourselves to two of the u phases and only
test the integer hypothesis at low order in the expansion. Phase I corresponds to c1 = 0 and
15
0 < c2 < r2 . In phase II c1 = r3 and c2 0. In phase I we calculate the zero mode piece
of u to be
t1 t1 t3 t3
(3.5)
= i T + N.
2
2
We can determine v by expanding an adjacent phase in which u v = 0, and we find
u = i +
t1 t1 3(t2 t2 ) t7 t7 t8 t8 t9 t9
4
4
2
2
4
= i T + ln(1 + q1 q3 ).
v = i +
(3.6)
Solving (3.4) for u and substituting instanton corrected variables, we can write the
expansion for u in this phase.
(1)q (n 1)!eNn eMr eP q
u =
p!q!r!(n p q r)!
n,p,q,r

p 3n3p2rq 2(nprq) 2n2p2rq nprq
v(4p+2r+q3n)
q1 q2
q7
q8
q9
e

(1)b+d (2n + m 1)!

n!a!b!c!d!e!f !(n a b c d)!(p f )!(m p e)!
a,b,c,d,e,f,m,n,p

3n3a3b2cd+p+f
q1a+e q2
eT (2n+m) eMc eNb eP d ev(4a+3b+2c+d+m2p2n)

n+e+f 2(nabcd)+p 2n2a2b2cd+p nabcd+f
q3
q7
q8
q9
(3.7)
.
One can easily verify that the above expansion reduces to (2.13) in the limit q7 = q8 =
q9 = 0. Also, the limit q1 = q9 = 0 yields the F2 blown up at two points. One can also
show that the inner phase of the Z3 model with ambiguity n = 1 is achieved in the limit
q1 = q2 = q31 = q7 = q8 = q9 = 0 with q2 q3 q7 q8 finite and v v + t3 . In choosing this
basis we have required that the expansion be convergent in a neighborhood of the origin
in both open and closed string
In phase I this requirement entails that negative
variables.
n
winding terms of the form ( i qi i )emv have n2 m. In an earlier calculation we chose a
basis not meeting this last requirement and many terms had fractional invariants. Note that
the basis with q6 instead of q9 does meet this latter requirement but still has fractions due
to the nonperturbative piece. Up to linear order in q9 , quadratic order in q7 , cubic order in
q1 , q3 , and q8 , quartic order in q2 and sixth order in ev , all of the numerical invariants in
this phase are integral. We present in Table 4 terms corresponding to the homology classes
that failed to be integral in the badly chosen basis.
3.3. Numerical invariants
What is the meaning of the numerical invariants given in Table 4 and why do we
anticipate that the numerical invariants of this phase are integers? We consider the moduli
space of maps from a Riemann surface of genus zero with one boundary into the Calabi
Yau such that the relative homology class of the image is labeled by the wrapping number
16
Table 4
Numerical invariants
k
m1
m2
m3
m7
m8
m9
Nk,m
m1
m2
m3
m7
m8
m9
Nk,m
2
2
2
3
3
3
3
4
4
4
4
6
6
6
6
1
0
0
0
1
1
3
1
3
0
0
1
3
0
0
3
4
4
3
4
4
3
3
3
4
4
3
3
4
3
2
2
2
3
3
3
3
2
2
2
2
2
2
2
3
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
3
2
2
3
2
2
2
2
3
2
2
2
3
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
64
36
36
621
2214
2214
72
128
0
56
56
224
0
84
936
2
2
2
2
3
3
3
3
4
4
4
6
6
6
6
1
2
0
0
0
1
0
1
0
0
0
3
1
0
1
3
4
3
4
3
4
3
4
4
6
6
3
4
4
4
0
0
1
2
0
0
0
0
0
2
2
3
3
2
3
2
2
2
2
2
2
2
2
2
4
4
2
2
2
2
2
2
2
2
2
2
2
2
2
4
5
2
2
3
3
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
0
0
1
4
0
0
0
0
0
0
0
1026
10656
84
10656
on each two-sphere of the basis and the winding number around a noncontractible circle
on the Lagrangian three-cycle. These maps should be holomorphic in the interior of the
disk. The moduli space of these maps is generally noncompact, and one must add in
extra maps that may be singular to define a compact space. There may be disconnected
components of the moduli space when there are homotopically inequivalent maps into
some relative homology class. One then defines a cohomology class analogous to the Euler
class, and the numerical invariant is obtained by integrating this class over the moduli
space. For the case of genus zero closed strings, one should fix three complex parameters
corresponding to SL(2, C) transformations of the complex plane while for disks one fixes
three real parameters corresponding to SL(2, R) transformations of the upper half plane. If
the dimension of the moduli space is zero after this fixing, the maps are isolated and the
numerical invariants can be interpreted as counting curves. Otherwise, the integral over the
moduli space could give fractions when there are orbifold singularities in the moduli space.
Of course, there are many technicalities needed to make the above discussion rigorous.
A first principles calculation from the nonlinear sigma model point of view as described
above is generally difficult. In this paper our determination of the invariants has been
facilitated by an equivalent calculation on the local mirror. The drawback is that one does
not have a direct argument that the invariants should be integers. Another approach is to
start with the boundary linear sigma model that flows to the conformally invariant nonlinear
sigma model at low energies. The relevant correlation functions that yield the numerical
invariants are in a topological sector of the theory. The corresponding correlators in a
topologically twisted version of the linear sigma model are scale invariant so the two
calculations should be equivalent. The correlators are intersection forms
on the moduli
1
space of classical solutions with a given instanton number mi = 2
Fi and winding

1
where
F
is
the
gauge
field
strength
which
is integrated over
number kp = 2
p
i
the two-dimensional worldsheet with boundary and p is a boundary field which

couples to the theta angle. It would be interesting to calculate correlators in the boundary
17
linear sigma model corresponding to the numerical invariants. This correspondence

between the two theories has been verified in several examples for closed worldsheets.
Our aim here is to discuss the structure of the boundary linear sigma model moduli space
in order to determine a criterion for integrality of the intersection forms. Assuming the
correspondence is valid, we can apply this criterion to the nonlinear sigma model. Note that
the moduli spaces of the two theories are different, but the correlators of this topological
sector are expected to be the same.
In the linear sigma model the moduli space can be described as a space of holomorphic
sections xi of a line bundle over the upper half complex plane H of degree d i =

(1)i
i
i
i
j (mj lj + kp lp ) (cf. (2.7)). One usually calls this bundle O(d ). If d 0, one can
d i
write a section as xi = n=0 xin zn where z is a coordinate on H and the xin must be
chosen so that solutions of xi = 0 all lie in H . If d i < 0, there is no holomorphic section
and xi = 0. Note that we have tried to indicate O(0) deformations of two-spheres on the
toric diagram by drawing adjacent lines parallel. The moduli space takes the form
Mm,
k
=
(Xm,
k
Im,
k
)
,
(3.8)
G
where Xm,
k
is the space of xin such that xi has zeroes in H . The maps xi are singular
whenever for some z, the values of xi are impossible. We can readily see what is meant
by impossible by examining Fig. 5. If we denote Di = {(x1 , x2 , . . . , x9 )|xi = 0}, then
I = {(x1 , x2 , . . . , x9 ) kl=1 Dl | kl=1 Dl M = } O where M is the toric manifold
corresponding to Fig. 5. Then Im,
k
is the set of xin such that for all z, (x1 , x2 , . . .) I . The
moduli space is compactified by including maps where some points on the worldsheet are
mapped into I . In general the moduli space could still be noncompact, but noncompact
O(0) directions are generally cut off by the disk. For k
= 0 and m
= 0, the moduli
space is the Lagrangian three-cycle. Here O is determined by the disk constraints. For
instance, in phase I of our case, O = {(x1 , x2 , . . . , x9 ) D3 D6 (D7 (M D2 ))
(D2 (M D7 ))}. We mod out the set of allowed xin by complex gauge transformations

l i l (1)i (2)i
G : xim j,p (jj pp l )xim where j C , p R+ , and U (1)(cf. (2.7), (2.8).
Writing xi = r0 (z r1 ) (z rn ), we see that this makes sense because we require
r1 , . . . , rn H but r0 C and
r0 = 0 if xi
/ 0. The rescalings and rotations preserve the
zeroes of xi , and the condition i Arg(xi ) = 0 is redundant at the boundary of the disk on
the toric base so 1 for disks stuck at the base. Note that all disks in the geometric phase
are stuck at the base. We have shown that the moduli space is well defined. It is now easy
to see that I includes any regions that are fixed by G unless we shrink some of the twospheres to enter a nongeometric, orbifold phase. Since the moduli space of the geometric
phase is smooth, the intersection form must give integers. On the other hand, the moduli
space is frequently singular in nongeometric phases and fractions are possible. One may
expect fractional terms when the boundary of the disk is fixed by the orbifold.
In phase II our coordinate transformation is u u = u t3 , v v = v. We obtain
u = i + t21 + t23 = i + ln(1 + q3 ) and v as in (3.6). This phase reproduces precisely
the outer phase (the three cycle intersects a noncompact two-cycle) of the Z3 case in the Z2
reflection of the limit discussed previously. This phase also reduces to the outer phase of
the blown up F2 in the limit that q1 = q9 = 0. All of the corresponding terms are integral.
18
In this phase v > 0 and all terms have k < 0. Calculating to the same order as in phase I,
we find that all terms are integral.
We have also calculated disk instantons in the nongeometric phase obtained by flopping
P1 (r6 ) and then flopping P1 (r6 + r7 ) at x4 = x7 = 0. In this phase x4 > 0 so the flopped
P1 is nongeometric. The calculation gives half-integer invariants for terms of the form
q24 q32 q8 e2nv . (r8 = r8 r2 + r4 + r6 , n Z+ ) These results are puzzling. Although there
is a Z2 orbifold singularity on the geometric P1 (r8 ) in this phase, the presence of the
disk prevents the moduli space from being an orbifold. We believe the resolution of this
paradox is that there is a nonperturbative contribution corresponding to P (3.2). The
relative homology classes in question are present in
Table 4. Depending on the choice of
coordinates, one can have square root branch cuts ( q8 ) in the nongeometric phase but one
can always find coordinates without branch cuts. The correction P is an expansion in the
flopped Khler parameter for which we need the analytic continuation to this phase. There
are other terms in this phase in which the moduli space does contain orbifold singularities
/ 0), and we anticipate fractional invariants. We have not pursued this
(x3 = x5 = 0, x2
calculation further.
We have in this model a flop transition that augments the four-cycle from an F2 blown
up at two points to one blown up at four points (Fig. 10).
We can flop r1 and r6 to obtain this phase. The conifold type transitions occur for the
part of the diagram that reduces to the Z2 Z2 case. As in that case one can substitute S3 s
for P1 s. Our amplitudes F0,1 should correspond to ChernSimons theory in a complicated
background.
Fig. 10. Flopped phase of Z2 Z4 blowup: r1 r1 , r6 r6 .
19
4. Z7
The Z7 case is interesting because there are three adjacent four-cycles. However, there
are no geometric flops unlike the previous examples. The charges of six chiral fields under
three U (1)s are
l1 = (0, 1, 3, 0, 1, 1),
l2 = (1, 2, 1, 0, 0, 0),
l3 = (2, 1, 0, 1, 0, 0).
(4.1)
j
Solving the equations j li |xj |2 = ri generates the toric diagram (Fig. 11) where r4 =
r1 + 3r2 and r5 = r1 + 3r2 + 5r3 .
The instanton corrected Khler parameters are given by
z1 = q1 e3M+P ,
z2 = q2 eM2P +Q ,
z3 = q3 e2Q+P ,
(4.2)
where
M=
(1)n+m (3n m 1)!zn zm zp

1 2 3
,
2 p!(m 2p)!(n 2m + p)!
n!
n,m,p
P=
(1)m+p (2n m p 1)!zp zn zm

1 2 3
,
2 (n 3p)!m!(n 2m)!
p!
n,m,p
Fig. 11. Toric diagram of Z7 blowup.
20
and
Q=
(1)m (2n m 1)!zp zm zn

1 2 3
.
2 (m 3p)!(n 2m + p)!
n!p!
n,m,p
These expressions for qi must be inverted to obtain zi (qi ). The expansions are
z1 = q1 + 6q12 + q1 q2 + 10q12q2 + 4q12q22 + q1 q2 q3 + 10q12q2 q3
+ 8q12 q22 q3 + 4q12q22 q32 + ,
z2 = q2 2q1 q2 + 5q12 q2 2q22 + 6q1q22 20q12q22 + q2 q3 2q1 q2 q3 + 5q12 q2 q3
3q22 q3 + 8q1q22 q3 26q12q22 q3 2q22 q32 + 6q1q22 q32 20q12 q22 q32 + ,
z3 = q3 + q2 q3 2q1q2 q3 + 5q12q2 q3 2q1 q22 q3 2q32 3q2 q32 + 6q1 q2 q32
15q12q2 q32 2q22 q32 + 12q1q22 q32 44q12q22 q32 + .
(4.3)
There are also three more solutions to the PicardFuchs equations corresponding to fourcycles. The Z3 and Z5 cases can be obtained in the appropriate limit. To obtain Z3 we set
z2 = z3 = 0.
The equation for the mirror can be written as
xz = P (u, v) = 1 + eu + ev + euvt1 + e2ut2 + e3u2t2t3 ,
(4.4)
where y2 = u, y5 = v, and y3 = 0. The Riemann surface that describes moduli of the threecycle is a genus three surface with legs extending to infinity (Fig. 12).
Clearly, we recover the O(3) P2 (Z3 blowup) when z2 = z3 = 0. Letting our
Lagrangian three-cycle be determined by |x2 |2 |x3 |2 = c1 , |x5 |2 |x3 |2 = c2 , and

i Arg(xi ) = 0; we find eight inequivalent phases taking into account the obviously
symmetric phases (Fig. 13) .
Again some of the phases are described by quadratic equations, whereas others require
a quartic equation. Phase I has c2 = 0 and c1 r21 while in phase II, c1 = c2 r2 2 . In
phase I
u = i +

1
t1 + 4t2 + 2t3 = i M + P
7
Fig. 12. Riemann surface P (u, v) = 0 in mirror of Z7 blowup.
(4.5)
21
Fig. 13. Phases for noncompact three-cycle in Z7 blowup.
Table 5
Phase I: m1 = 2, k = 3
m3
0
1
2
3
4
Table 6
Phase I: m1 = 2, k = 4
m2 = 0
9
0
0
0
0
26
26
0
0
0
49
76
49
0
0
76
144
144
76
0
110
232
284
236
118
Table 7
Phase I: m1 = 2, k = 5
m3
m2 = 0
0
1
2
3
4
15
0
0
0
0
m3
0
1
2
3
4
m2 = 0
12
0
0
0
0
36
36
0
0
0
72
112
72
0
0
116
224
224
116
0
172
372
460
376
180
Table 8
Phase I: m1 = 2, k = 6
48
48
0
0
0
99
156
99
0
0
166
324
324
166
0
250
552
684
556
258
m3
1
2
3
4
m2 = 0
0
0
0
0
62
0
0
0
208
132
0
0
444
444
226
0
772
960
776
355
and
v = i +

1
3t1 2t2 t3 = i M.
7
(4.6)
In phase II u is unchanged while v = v u = P . The results are shown in the

Tables 512.
There are nongeometric phases obtained by flopping the Khler parameters. One would
need to analytically continue the expansions (4.2) to calculate in these phases. The
boundary conditions on the disk prevent the open instanton moduli space from being
22
Table 9
Phase II: m1 = 2, k = 3
m3
0
1
2
3
4
Table 10
Phase II: m1 = 2, k = 4
m2 = 0
3
0
0
0
0
8
8
0
0
0
19
28
19
0
0
46
84
84
46
0
110
232
284
236
118
m3
Table 11
Phase II: m1 = 2, k = 5
m3
0
1
2
3
4
0
1
2
3
4
m2 = 0
4
0
0
0
0
12
12
0
0
0
32
48
32
0
0
76
144
144
76
0
172
372
460
376
180
Table 12
Phase II: m1 = 2, k = 6
m2 = 0
5
0
0
0
0
18
18
0
0
0
49
76
49
0
0
116
224
224
116
0
250
552
684
684
258
m3
0
1
2
3
4
m2 = 0
7
0
0
0
0
26
26
0
0
0
36
112
36
0
0
166
324
324
166
0
347
772
960
776
355
an orbifold even when the closed string instanton moduli space is an orbifold unless the
boundary of the disk is fixed by the orbifold. In this case fractional invariants are possible.
Acknowledgements
I wish to acknowledge J. Distler and A. Iqbal for beneficial discussions and R. McNees
for answering Mathematica related questions. This work was supported in part by NSF
grant PHY-0071512.
References
[1] M. Aganagic, A. Klemm, C. Vafa, hep-th/0105045.
[2] D. Morrison, M. Plesser, Nucl. Phys. B 440 (1995), hep-th/9412236.

Dielectric fundamental strings in matrix string

theory
D. Brecher a , B. Janssen a , Y. Lozano b
a Centre for Particle Theory, Department of Mathematical Sciences, University of Durham, South Road,
Durham DH1 3LE, United Kingdom

b Departamento de Fsica, Universidad de Oviedo, Avda. Calvo Sotelo 18, 33007 Oviedo, Spain
Received 11 April 2002; received in revised form 25 April 2002; accepted 25 April 2002
Abstract
Matrix string theory is equivalent to type IIA superstring theory in the light-cone gauge, together
with extra degrees of freedom representing D-brane states. It is therefore the appropriate framework
in which to study systems of multiple fundamental strings expanding into higher-dimensional Dbranes. Starting from Matrix theory in a weakly curved background, we construct the linear couplings
of closed string fields to type IIA Matrix strings. As a check, we show that at weak coupling the resulting action reproduces light-cone gauge string theory in a weakly curved background. Further dualities
give a type IIB Matrix string theory and a type IIA theory of Matrix strings with winding. We comment on the dielectric effect in each of these theories, giving some explicit solutions describing fundamental strings expanding into various Dp-branes. 2002 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 11.27.+d
Keywords: D-branes; M(atrix) theories
1. Introduction
The idea that a collection of branes can undergo an expansion into a single higherdimensional D-brane under the influence of a background RamondRamond (RR)
potential was first explored by Emparan [1]. Although his main concern was with the
description of N fundamental strings expanding into a Dp-brane, he also realised that
D(p 2)-branes could undergo such an expansion. Emparans analysis of this effect was
E-mail addresses: dominic.brecher@durham.ac.uk (D. Brecher), bert.janssen@durham.ac.uk (B. Janssen),
yolanda@string1.ciencias.uniovi.es (Y. Lozano).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 4 4 - 9
24
D. Brecher et al. / Nuclear Physics B 634 (2002) 2350
entirely at the level of the abelian theory relevant to the description of the single Dpbrane. Switching on a background RR (p + 2)-form field strength, he found solutions of
the combined BornInfeldChernSimons theory with topology M2 S p1 , where Mn
denotes an n-dimensional Minkowski space, and with N units of dissolved electric flux,
corresponding to the fundamental strings. Similar solutions, with topology Mp1 S 2 ,
and with N units of dissolved magnetic flux, corresponding to D(p 2)-branes, are easily
found.
It was some years later that a description from the point of view of the lowerdimensional D(p 2)-branes was provided [2]. This involved an analysis of certain
couplings in the non-abelian BornInfeldChernSimons action proposed by Taylor and
van Raamsdonk [3] and Myers [2]. From this perspective, the expansion is due to the fact
that the transverse coordinates of the D(p 2)-branes are matrix-valued. Myers original
analysis was of a collection of N D0-branes, in the presence of a background RR four2 , where
form field strength: they spontaneously expand into a D2-brane of topology RSNC
2
SNC denotes the non-commutative two-sphere. The D2-brane is uncharged with respect to
the four-form, but has a non-zero dipole moment. Hence the name of a dielectric D2brane. Configurations for arbitrary p, and more general configurations involving fuzzy
cosets, were described in [4].
These two descriptions of the dielectric effectabelian and non-abelianare, of
course, dual to one another. In the limit of large N , the non-commutative nature of the
fuzzy two-sphere is lost, it becomes a smooth manifold, and the two descriptions do indeed
agree.
In principle, all such dielectric branes should have a corresponding supergravity solution
describing the back-reaction of the brane on spacetime. Technical difficulties, however,
have limited the analysis of such solutions to that of N D4-branes [5,6] or N F-strings [6]
expanding into a D6-brane with topology M5 S 2 and M2 S 5 , respectively. In both cases,
there is a stable and an unstable radius of the dielectric sphere, the form of the effective
potential matching the abelian worldvolume analysis precisely [5,7].
There is thus much evidence, from both the worldvolume and supergravity perspectives,
that the dielectric effect is not limited to D-branes, but that dielectric F-strings also exist.
One can then pose the following question: can the expansion of F-strings into a Dp-brane
be described from the point of view of the strings themselves? It is the purpose of this
paper to answer precisely this question.1
Since, from the strings perspective, the dielectric effect should be due to matrix-valued
coordinates, we are led to a consideration of Matrix string theory [911]. Starting from the
Matrix theory action [12,13], one compactifies on a circle, reinterprets the resulting theory
as a (1+1)-dimensional super YangMills theory on the dual circle [14], and then performs
the so-called 9-11 flip, which implements a further S- and T-duality [11]. It is easy to see
that the result is a (1 + 1)-dimensional super YangMills theory describing N fundamental
strings in the type IIA theory. In the weak string coupling limit, one recovers N copies of
the GreenSchwarz action describing light-cone gauge string theory.
1 A recent attempt to answer this question from the perspective of Matrix string theory has already been
made [8], although the results seem somewhat opaque to the authors and there is little overlap with this work.
25
To discuss possible dielectric solutions of Matrix string theory, we need to know how the
RR potentials couple to the worldvolume of the (Matrix) string. Since the Matrix theory
action in a weakly curved background is known [15,16], one can in principle run through
the above chain of dualities to derive the Matrix string theory action in a corresponding
weakly curved background. Indeed, Schiappa has already considered this computation,
and has written down an obvious dielectric-like solution [17]. However, it is not at all
clear that this solution corresponds to a dielectric string, for reasons explained later in this
paper. Moreover, we find couplings to various components of the RR fields over and above
those found by Schiappa [17] and, for this reason, we consider the problem of deriving the
Matrix string theory action in a weakly curved background from scratch. We should further
note that one of us has also already derived couplings of the type IIA F-string to various
background RR potentials [18].
One might think that F-strings could expand into a Dp-brane with topology M2 S p1
for any value of p, but this is not the case. As mentioned above, one generically finds a
stable and an unstable spherical solution of the Dp-brane theory with dissolved electric
flux. Expansion into a D2-brane, however, is atypical: only an unstable solution exists [1].
Indeed, this observation is mirrored in the corresponding (albeit smeared) supergravity
solution [7]. One might suspect, therefore, that stable cylindrical D2-brane solutions
of the type IIA Matrix string theory do not exist, and we aim to address this issue
here.
A further T-duality takes us to a type IIB Matrix string theory which, as the S-dual of the
D-string theory, describes N F-strings in the static gauge. Just as D-strings can expand into
D3-branes, so can F-strings and the dielectric solution presented in [2] is equally applicable
here.
In the following section, we consider Matrix theory in a weakly curved background [15,16]. It is easier to consider the chain of dualities leading to Matrix string
theory in ten- rather than eleven-dimensional language, so we choose to work with the
D0-brane theory in a weakly curved background. This has been derived from the Matrix
theory results in [19]we show explicitly that it reproduces the lowest-order expansion
of the non-abelian D0-brane theory of Taylor and van Raamsdonk [3] and Myers [2]. Tduality, taking us to the D-string theory, is considered in Section 3, and the 9-11 flip in
Section 4. We show that the resulting linear action reproduces the light-cone gauge string
theory action in a weakly curved background, thus lending some weight to our results.
Indeed, we need the extra couplings, relative to the results of [17], for this to be the case.
A further T-duality taking us to the type IIB theory is considered in Section 5 and, although we cannot perform the S-duality rigorously, we argue that the result is equivalent
to the S-dual of the D-string theory. In Section 6, we apply once more a T-duality transformation, this time in a direction transverse to the IIB string, giving a theory of type IIA
F-strings with winding number rather than light-cone momentum. Some of the resulting
couplings of both the type IIA and the type IIB strings have already been considered by
one of us [18]. Dielectric solutions are considered in Section 7, where we give some explicit solutions describing the expansion of F-strings into Dp-branes for different p. We
conclude in Section 8.
26
2. Matrix theory
The bosonic sector of the Matrix theory action in a flat background is [13]
Sflat =
1
R

R 2 i j i j
1
X
X
,
X
,
X
dt Tr Dt Xi Dt Xi +
,
2
16 2 lP6
(2.1)
where lP denotes the eleven-dimensional Planck length, R is the radius of the eleventh
dimension and Dt Xi = t Xi + i[At , Xi ]. We will choose the gauge At = 0 throughout
this paper at the expense of losing
explicit gauge invariance. With the string coupling
set through the relation R = gs , this action is the non-relativistic limit of the nonabelian BornInfeld action describing N D0-branes, with mass quantized in units of 1/R:
the dimensional reduction to one dimension of ten-dimensional YangMills theory with
2 = g /(4 2 3/2 ). The connections are more subtle, however (see, e.g., [20]). After
gYM
s
all, when N is large, Matrix theory captures eleven-dimensional physics in the infinite
momentum frame [13]. For finite N , it is equivalent to a DLCQ or null compactification of
M-theory [2123], whereas the type IIA D0-brane theory comes about via a spacelike
compactification. The two actions are then related by an infinite boost in the eleventh
1/3
dimension. At any rate, if we take T0 = 1/R and R = gs , so that lP = gs
, then
we recover the YangMills description of D0-branes.
2.1. Linear couplings in Matrix theory
Matrix theory in an arbitrary background is understood only rather poorly (although
see [2426] for early work on this subject) and, in the above sense, this is related to the
question as to what is the form of the non-abelian BornInfeld theory in curved space.
Kabat and Taylor have derived the linear Matrix theory couplings to bosonic background
fields [15] and Taylor and van Raamsdonk have extended these calculations to include
fermionic backgrounds [16]. They have further derived the linear couplings of the D0brane [19] and Dp-brane [3] theories from Matrix theory. The results certainly seem to
agree with the linear order expansion of the combined non-abelian BornInfeldChern
Simons action proposed by Taylor and van Raamsdonk [3] and Myers [2], including the
form of the overall symmetrized trace due to Tseytlin [27]. More precise expressions for
the Matrix theory couplings, to all orders in both derivatives of the background fields and
the fermionic coordinates, can be had by dimensional reduction of the eleven-dimensional
supermembrane vertex operators constructed in [28,29].
In other words, one can write the Matrix theory action as
S = Sflat + Slinear ,
(2.2)
with Sflat given by (2.1), and where the linear action has the form [15]
Slinear =
1
R

dt STr

1
hAB T AB + AABC J ABC + AABCDEF MABCDEF , (2.3)
2
27
where A, B = 0, . . . , 10 and STr denotes the symmetrized trace.2 The eleven-dimensional

metric, hAB , 3-form potential, AABC , and its 6-form dual, AABCDEF , couple to
the energymomentum tensor, membrane current and 5-brane current, respectively.
The form of these currents has been worked out explicitly, and they match elevendimensional supergravity predictions [15]. There are also higher-order multipole couplings
to derivatives of the background fields, but we will not be concerned with them here. One
can relate this linear action to that relevant to the description of D0-branes, via the infinite
boost mentioned above. The D0-brane currents, denoted I , are related in this manner to the
eleven-dimensional currents T , J and M, as explained in [19].
The linear D0-brane action is then

1
1
...
Slinear =
dt STr h Ih + I + b Is + b1 ...6 I5 1 6
R
2
1 (5)
...
(3)
C
+ C(1) I0 + C
I2 +
I 1 5
60 1 ...5 4

1
1 ...7
C(7)
+
I
(2.4)
,
336 1 ...7 6
where , = 0, . . . , 9. The currents Ih , Is , I5 and Ip couple respectively to the metric, h ,
the NeveuSchwarzNeveuSchwarz (NSNS) 2-form potential b(2) , its 6-form Hodge
dual b (6) , and the RR (p + 1)-form potentials C (p+1) . The potentials C (5) and C (7) , the
Hodge duals of C (3) and C (1) , have been rescaled relative to [3,19]. Note that there should
also be couplings to C (9) , but these are not determined by the analysis of [3,19]. The
currents appearing in (2.4) are given in terms of the dimensional reduction of the Born
Infeld field strength,
F0i = F 0i = t Xi X i ,
Fij = F ij =
R
2lP3

i Xi , Xj ,
(2.5)
where i, j = 1, . . . , 9. Usually R is taken to be R = gs , and so one has 2 =

2lP3 /R. After the 9-11 flip of Section 4, however, the role of the ninth and the eleventh
coordinate are interchanged, and this relation is no longer true. For this reason we define
a general quantity 2lP3 /R, the D0-brane theory and its duals being recovered upon
taking = .
Substituting for (2.5), the NSNS currents are [19]:

2
1
1
1
1
I = 1 F 0i F 0i + F ij F ij
F F F F F F
2
4
8
4

1
1
1
2
= 1 2 [X, X]2 4 [X, X]4 [X, X]2
4
4
8

1
1
1
(2.6)
X 2 1 + X 2 2 [X, X]2 ,
2
4
4
2 We include both the overall factor of 1/R and the overall gauge trace in the action, rather than in the currents,
for convenience.
28
1 0i 0i 1 ij ij
1
1
F F + F F = 1 + X 2 2 [X, X]2 ,
(2.7)
2
4
2
4

1
1
Ih0i = F 0i 1 + F 0j F 0j + F j k F j k F ij F j k F 0k
2
4

1
1
1
(2.8)
i
2
2
= X 1 + X 2 [X, X] 2 Xi , Xj Xj , Xk X k ,
2
4

1
ij
Ih = F 0i F 0j + F ik F kj = X i X j 2 Xi , Xk Xk , Xj ,
(2.9)
1
i i j j
X ,X X ,
Is0i = F ij F 0j =
(2.10)
2
2

1
1
1
1
1
ij
Is = F ij 1 + F kl F kl F 0j F 0j + F 0i F 0k F kj F 0j F 0k F ki
2
4
2
2
2
1
+ F ik F kl F lj
2

i i j
1
1 2
2
=
1 X 2 [X, X]
X ,X
2
2
4

i i j k k j i k k
i
X X , X X X X , X X 3 Xi , Xk Xk , Xl Xl , Xj ,
2
2
(2.11)
where we have defined
Ih00 = 1 +
X 2 X i X i ,

[X, X]2 Xi , Xj Xi , Xj ,

[X, X]4 Xi , Xj Xj , Xk Xk , Xl Xl , Xi .
The RR currents are [19]
I00 = 1,
I0i
= F
(2.12)
0i
= X i ,
i i j
X ,X ,
=
6
(2.13)
1
= F ij
(2.14)
6
1
ij k
I2 = F 0i F j k + F 0j F ki + F 0k F ij
6
i i j k j k i k i j
(2.15)
=
X X ,X + X X ,X + X X ,X ,
6
1
0ij kl
I4
= F ij F kl + F ik F lj + F il F j k
2

1
= 2 Xi , Xj Xk , Xl + Xi , Xk Xl , Xj + Xi , Xl Xj , Xk ,
2
(2.16)
15 0[i j k lm]
15 [i j k l m]
ij klm
= F F F = 2 X X ,X X ,X ,
I4
(2.17)
2
2
0ij
I2
29
i [i j k l m n]
X ,X X ,X X ,X ,
(2.18)
3

7i
ij klmnp
I6
(2.19)
= 7F 0[i F j k F lm F np] = 3 X [i Xj , Xk Xl , Xm Xn , Xp] .
Up to a couple of caveats, we show below that the above currents agree with the
expansion, to linear order in the background fields, of the multiple D0-brane theory of
Taylor and van Raamsdonk [3] and Myers [2]. The first such caveat is that we must have
...
I5 1 6 = 0 if the two actions are to match [2]. In other words, D0-branes cannot expand
into NS5-branes. We will, therefore, drop this term from now on. The second is that the
ij
currents Ih00 and Ih derived from Matrix theory actually contain further terms [19], the
form of which seems less certain. Indeed, although the expressions (2.7) and (2.9) match
precisely the first-order expansion of the non-abelian BornInfeld action, as given in [2,3],
ij
we have not been able to match these extra terms in Ih00 and Ih . We will, for this reason,
ignore them.
0ij klmn
I6
= F [ij F kl F mn] =
2.2. Matrix theory vs. multiple D0-branes

We will now show that the above currents are recovered in the expansion to linear order
in the background fields of the multiple D0-brane theory.3 If we set the NSNS fields to
zero, the multiple D0-brane action is [2,3,27]
S = Sflat + SCS ,

1 X i (Q1 )ij X j det(Qij ) ,

Sflat = T0 dt STr

i(iX iX )/
(n)
SCS = T0 STr P e
C
,
where T0 = 1/R = 1/(gs ) and

i i j
Qij = ij +
X ,X .
The pull-back is defined in terms of gauge covariant derivatives, such that

P [] = 0 + Dt Xi i dt = 0 + X i i dt,
(2.20)
(2.21)
(2.22)
(2.23)
(2.24)
with obvious generalizations to forms of higher degree. The interior multiplication is

iX ...p = Xi i1 ...p ,
(2.25)
1
giving rise to the relevant commutators in the action. The ChernSimons action (2.22)
becomes

i
1
i
SCS = T0 STr P C (1) + (iX iX )C (3) 2 (iX iX )2 C (5) 3 (iX iX )3 C (7)
2
6

1
+
(iX iX )4 C (9) .
(2.26)
244
3 See also [30].
30
It is easy to see that the D0-brane RR currents, (2.12)(2.19), of the previous subsection,
can be written such that

C(1) I0 dt = P C (1) ,
(2.27)

i
(3)
I2 dt = P iX iX C (3) ,
C
(2.28)

1
1 (5)
...
C1 ...5 I4 1 5 dt = 2 P (iX iX )2 C (5) ,
(2.29)
60
2

i
1 (7)
...
C1 ...7 I6 1 7 dt = 3 P (iX iX )3 C (7) ,
(2.30)
336
6
which reproduce the ChernSimons action (2.26). Since we have set the NSNS 2-form to
zero, these linear couplings are actually exact. We should also note that one could derive
the correct Matrix theory coupling to C (9) in (2.4) by comparing with the ChernSimons
action (2.26).
The NSNS fields are somewhat harder to deal with. To linear order in the background
fields, the non-abelian BornInfeld action is
SNBI = SNBI ( = 0) Sflat ,
with Sflat as in (2.21). This gives the dilaton current:

1 X det Q,
I = 1 XQ
(2.31)
(2.32)
1 X = X i (Q1 )ij X j . As shown already in [2,27], expanding the current (2.32)

where XQ
to the relevant order, gives the Matrix theory result (2.6).
The action containing the couplings to the metric and NSNS 2-form is more
complicated:

i

SNBI ( = 0) = T0 dt STr P E00 + E0i Q1 k E kj Ej 0 det Qi j ,
(2.33)
where
i i k
X , X Ekj .
To linear order in the background fields, E = + h + b , so that

1
k i l m
i
i
k
Q j =Q k j + Q
X , X (hmj + bmj ) ,
l
Qi j = i j +
which gives, again to linear order,

1
i
1
i i l m
1
k
i
Q
X
(h
=
Q
,
X
+
b
)
Q
k
j
l
j,
mk
mk
(2.34)
(2.35)
(2.36)
so we have all the ingredients we need to derive the linear BornInfeld couplings to the
metric and NSNS 2-form potential. We find

1 X 1/2 ,
Ih00 = det Q 1 XQ
(2.37)

1 X 1/2 Q1 (ij ) X j ,
det Q 1 XQ

1
ij
1 X 1/2
det Q 1 XQ
Ih =
2

j
i
X i Q1 k X k X l Q1 lm Xm , Xi Q1 k X k
j

i
1
X Q1 k Xk , Xi + (i j ) ,
1 XQ
1
1 X 1/2 Q1 [ij ] X j ,
Is0i =
det Q 1 XQ
2

1
ij
1 X 1/2
det Q 1 XQ
Is =
4

j
i
X i Q1 k X k X l Q1 lm Xm , Xi Q1 k X k
1
j k i
i
1
1 XQ X Q
k X , X (i j ) .
Ih0i =
31
(2.38)
(2.39)
(2.40)
(2.41)
One can verify that the lowest order expansion of these currents reproduces the results
(2.7)(2.11).
3. T-duality: multiple D-strings

To construct the Matrix string theory of Dijkgraaf, Verlinde and Verlinde [11], one
compactifies Matrix theory on a circle in, say, the x 9 direction. Taylor has shown that
this is equivalent to a (1 + 1)-dimensional super YangMills theory on the dual circle [14].
Taking now i, j = 1, . . . , 8, and denoting the dual coordinate by x,
the worldvolume fields
transform as [14]

1
i i j
i i j
Fij = F ij =
(3.1)
X , X
X ,X ,
d x

2 R9

1
i 9 i
F9i = F 9i =
(3.2)
X , X
d x Dx Xi ,
9
2 R

1
0i
i
F0i = F = X
(3.3)
d x X i ,
9
2 R

1
F09 = F 09 = X 9
(3.4)
d x A x ,
9
2 R
9 = /R9 is the radius of the dual circle. Of course, this is just T-duality applied
where R
to Matrix theory. By construction, the multiple D0-brane action (2.20) considered in the
previous section is covariant under T-duality, so an application of T-duality to the D0-brane
action (2.4) derived from Matrix theory should reproduce the non-relativistic limit of the
multiple D-string action.
32
Certainly, the action (2.1) in a flat background becomes, with DXi Dx Xi and
A Ax ,

2
1
2
2 2
1
1
Sflat =
(3.5)
,
dt d x Tr X 2 2 DX2 +
A + 2 Xi , Xj
9
2
2
2
4
2R R
which, with = , is just the non-relativistic limit of the flat space D-string action; with
T-duality acting on the string coupling as

,
gs g s = gs
(3.6)
R9
the overall factor is T1 = 1/(gs ) as required. Turning to the linear action (2.4), we must
consider T-duality applied to both the worldvolume and background fields. As far as the
currents are concerned, we simply take the I s written in terms of the BornInfeld field
strength (2.5), and reinterpret the relevant components through the action of T-duality as
given by (3.1)(3.4) above [3]. This is a simple re-writing, the results being collected in
the appendix.
To linear order, the action of T-duality on the background fields is:
ha9 ba9,
(p)
h9a b9a ,
h99 h99 ,
1
h99 ,
2
(3.7)
(p1)
Ca1 ...ap1 9 Ca1 ...ap1 ,
(3.8)
where a, b = 0, . . . , 8, and all other fields are invariant. A simple application of these rules
gives

1
1
1
1
ab
a9
99
dt d x STr hab Ih 2ha9 Is h99 Ih + h99 I
Slinear =
9
2
2
2
2R R
(2)
(2)
(4)
ba9Iha9 + bab Isab + C (0) I09 + Ca9 I0a + 3Cab I2ab9 + Cabc9 I2abc
1 (4)
1 (6)
1 (6)
a ...a
a ...a 9
Ca1 ...a4 I4a1 ...a4 9 +
Ca1 ...a5 9 I4 1 5 +
Ca1 ...a6 I6 1 6
+
12
60
48

1 (8)
a1 ...a7
C
.
+
I
336 a1 ...a7 9 6
(3.9)
There should also be an extra coupling to C (8) , coming from the unknown Matrix theory
coupling to C (9) .
The resulting action should be equivalent to the linearized version of the D-string action.
With the results in the appendix, it is easy to see that the RR terms can indeed be written
as

i
SRR = T1 STr P C (0) F + C (2) + i(iX iX )C (2) F + (iX iX )C (4)
1
1
i
(iX iX )2 C (4) F 2 (iX iX )2 C (6) 2 (iX iX )3 C (6) F
2
2
6

1
i
(iX iX )4 C (8) F ,
3 (iX iX )3 C (8) +
(3.10)
6
243
33
We should note that only half of the terms necessary to form the pullback
where F09 = A.
(8)
of C
are present in the linear action (3.9). The missing terms come from the C (9)
coupling in the D0-brane action (2.4), as mentioned above. At any rate, this is just the
expansion of the ChernSimons action

i(iX iX )/
(n)
F
e
,
SCS = T1 STr P e
(3.11)
C
as expected. It is computationally more involved to check the NSNS couplings in (3.9)
against the non-abelian BornInfeld theory of D-strings, and we will not do this here.
4. Matrix string theory: multiple IIA F-strings

4.1. The 9-11 flip
Having constructed the (1 + 1)-dimensional theory of the D-string, we are now in a
position to perform the so-called 9-11 flip, a rotation
x 9 x 11 ,
x 11 x 9,
(4.1)
which will give us the type IIA Matrix string theory action in a linear background.
Whereas Schiappa [17] considers the action of the 9-11 flip on the currents, and leaves
the background fields invariant, we take the view here that it is the currents which are
invariant under the 9-11 flip. After all, in the flat space case, the 9-11 flip does not
change the worldvolume fields [11]. Moreover, Schiappa has argued that the
currents are
in fact invariant under the S- and T-dualities [17]. We simply take R9 = gs , so that
9 . We then define the dimensionless worldsheet coordinates

9 = /gs and lP = gs4/3 R
R
=
x
,

R9
R
t,

(4.2)
and perform the rescalings Xi Xi and gs gs /(2). The flat space action (3.5)
becomes the Matrix string action of [11]:

1
gs2 2
1
1 2 1
2
2
A + 2 [X, X] ,
Sflat =
(4.3)
d d Tr X DX +
2
2
2
2
4gs
where < < and 0 < 2 . Weakly coupled string theory at gs = 0 is recovered
in the IR limit, so is described by strongly coupled (1 + 1)-dimensional YangMills. The
conformal field theory which describes this IR limit is a sigma model on an orbifold
target space [11]. The matrix-valued coordinates must commute in this limit, so can be
i corresponding to the positions of
simultaneously diagonalised, the eigenvalues x1i , . . . , xN
the N strings. Then the action (4.3) reduces to a sum of GreenSchwarz actions for lightcone gauge string theory:
Sflat =
1
2

d d
N

1
n=1
x n2

1 2
xn .
2
(4.4)
34
The couplings to linear background fields considered herein should thus tell us something
about light-cone gauge string theory in weakly curved backgrounds.
The action of the 9-11 flip on the background fields is easy to derive. Consider, for
example, the type IIB metric fluctuation ha9 . Its T-dual on the type IIA side is ba9 , or in
eleven-dimensional language Aa9 11. Under the 9-11 flip Aa9 11 Aa11 9 = ba9 . In
other words, ha9 ba9 under the 9-11 flip, which is just (linearized) S-duality followed
by (linearized) T-duality in the x 9 direction. Arguing in a similar manner, one finds that the
linear background fields transform in the following way under the 9-11 flip:

1
1
hab hab ab h99 , ha9 ba9,
h99 + h99 ,
2
2
1
(3)
(1)
+ h99 ,
bab Cab9 ,
ba9 Ca ,
2
(2)
(2)
Cab
bab ,
Ca9
ha9 ,
C (0) C9(1) ,
Ca(4)
Ca(5)
,
1 ...a4
1 ...a4 9
(4)
(3)
Cabc9
Cabc
,
Ca1 ...a5 9 ba1 ...a5 9 ,
Ca1 ...a8 Ca1 ...a8 9 , Ca1 ...a7 9 Ca1 ...a7 .
(6)
(8)
Ca(6)
Na(7)
,
1 ...a6
1 ...a6 9
(9)
(8)
(7)
(4.5)
(7)
Na1 ...a6 9
Here,
is the field that couples minimally to a type IIA KaluzaKlein monopole
whose Taub-NUT direction is along x 9 . It is easy to verify that the above transformations
are precisely what one would find by performing a linearized S-duality (in the string frame)
h h ,
C
(0)
(0)
b2 C (2) ,
C (2) b2 , C (6) b6 ,
,
,
(4.6)
C
(8)
(8)
(4.7)
x9
followed by a linearized T-duality in the

direction, as in (3.7) and (3.8), plus the
linearized T-duality rules for the field b6 [31]:
(7)
ba1 ...a6 Na1 ...a6 9 ,
ba1 ...a5 9 ba1 ...a5 9 .
(4.8)
Performing the transformations (4.5) on the linear action (3.9), and substituting for
(4.2), one finds

1
1
1

IIA
Slinear
hab ab h99 Ihab + 2ba9Isa9
=
d d 2 STr
2
2
2
R

1
1
1
3
(3)
(1)
+
+ h99 Ih99
h99 I + Ca(1) Iha9 Cab9 Isab C9 I09
2
2
2
2
1 (5)
1
a ...a
(3) abc
Ca1 ...a4 9 I4a1 ...a4 9 +
ha9 I0a + 3bab I2ab9 + Cabc
I2 +
ba1 ...a5 9 I4 1 5
12
60

1 (7)
1
a ...a 9
a1 ...a7
Na1 ...a6 9 I6 1 6
Ca(7)
+
I
(4.9)
.
48
336 1 ...a7 6
Writing the currents in
the appendix in terms of the dimensionless quantities and , and
after rescaling Xi Xi and gs gs /(2), this is the action describing Matrix string
theory in a weakly curved background. There is no need to write out the new couplings
DX/, [X, X]/ and A appears
in full, suffice it to say that each term of the form X,
35
multiplied by a factor of R/ , each [X, X] term appears multiplied by a factor of 1/gs ,

and each A term appears with a factor of gs . We should note that the couplings derived
by Schiappa [17] are as above, but without those couplings to fields with a component in
the x 9 direction. As we will see below, these latter are necessary to match with light-cone
gauge string theory.
4.2. Light-cone gauge string theory in a general background
To examine the action (4.9), let us consider the weakly coupled string theory, gs = 0.
Just as in the flat case, the BornInfeld field strength drops out entirely, and one is forced
onto the space of commuting matrices as above. Moreover, the couplings to all Ramond
Ramond fields vanish, as one might expect, and we find
IIA
Slinear
1
=
2

N

1
1
h00 h09 + h99
d d
2
R2 2
n=1

(h0i h9i )xni + (b0i b9i )xni + (h00 h99 ) xn2 + xn2
R
4
1 i j
j
j
i
i
+ hij xn xn xn xn + bij xn xn + b09 x n xn
2

R
1
1 i 2
h0i xni xn2 + xn2 + bi9
xn xn + xn2 + x ni xn xn
+
2
2

2
R
3
1 2
1
2
2 2
+
(4.10)
h99 (xn xn ) + xn + xn
.
4
8
4
+
If we now define the (somewhat non-standard) light-cone coordinates
1
X = X0 X9 ,
2
(4.11)
then we have
IIA
Slinear
1
=
2

d d

h++ +
2 h+i xni + b+i xni
2
R
R
N

n=1
1
1
j
j
h+ xn2 + xn2 + hij xni xn xni xn

2
2

R
j
+ bij xni xn + b+ xn xn + O
.

(4.12)
We wish to compare this action to light-cone gauge string theory in a weakly curved
background. To this end, consider

1
S =
d d x x G (x) + 7 x x B (x) , (4.13)

4
36
where , denote the worldsheet coordinates and , is the worldsheet metric and
its determinant. We are always free to take the worldsheet to be flat, in which case

1
S =
d d x x G (x) + x x G (x) 2x x B (x) . (4.14)

4
Since the worldsheet energymomentum tensor vanishes, this must be supplemented with
the constraint

x x x x G = 0.
(4.15)
The light-cone gauge is defined by taking x + (, ) = . Since we interested only in
linear backgrounds, we set G = + h , where + = 1 and ij = ij , and
B = b . As usual in a light-cone treatment of gravity [32], we can further use spacetime
diffeomorphisms to set G = 0 = Gi . The string action is then

1
1
d
d
2x + x 2 x 2 + h++ + h+i x i + h+ x
S=

2
2
1 i j
+ hij x x x i x j + b+ x + b+i x i + bij x i x j
2

1 2
x + x 2 x i x x x i .
+ bi
(4.16)
2
At first sight, the linear couplings to the background fields do not seem to match those
in the action (4.12). However, the constraint (4.15) can be solved for x giving, again to
linear order in h ,
1
1
1
x = x 2 + x 2 + h++ + x i h+i + x i x j + x i x j hij + x 2 + x 2 h+ ,
2
2
2
2
(4.17)
x = x x + x i h+i + x i x j hij + x xh+ .
(4.18)
To linear order, then, we can replace x and x in the string action with (x 2 + x 2 )/2
and x x, respectively. In that case, we find exact agreement with the Matrix string theory
result (4.12). In other words, we have succeeded in reproducing the correct form of the
light-cone gauge type IIA string action in a weakly curved background.
4.3. RamondRamond couplings
Turning to the RR couplings in the action (4.9), let us consider the 3-form couplings,
which could potentially give rise to D2-brane solutions. If we set the BornInfeld field to
zero, we find

(3) i (3)
i
C + X Ci ,
d d STr
SC (3) =
(4.19)
2gs
R +
1 (3) j i
(3)
X ,X ,
= C+ij
C+
(4.20)
2

1
(3) j
Ci(3) = Cij(3)k Xk , Xj + C+j
(4.21)
X , Xi .
2
37
The couplings given in [17] differ considerably from the ones given here, since in [17]
(3)
only the C0ij
and Cij(3)k terms were considered. Yet from our analysis it is clear that the
other terms not only contribute, but are necessary in order to be able to write the C (3)
couplings in the light-cone gauge.
A nice check of the 3-form couplings (4.19) is against the eleven-dimensional
supermembrane theory. The connection between the light-cone treatment of the elevendimensional supermembrane in flat space and Matrix theory is well known [33]. One
can further argue that Matrix string theory is found by compactifying the light-cone
supermembrane theory on a circle, the off-diagonal elements of the Matrix string theory
fields being related to the infinite tower of KaluzaKlein modes along the compact
direction [34]. Given further that the light-cone supermembrane theory in an arbitrary
supergravity background is known [3537], one could in principle derive Matrix string
theory in an arbitrary background using these techniques. Indeed, the Lagrangian density
for the light-cone supermembrane in a curved (but not entirely general) background has
been derived in [36] and it is easy to see that the methods of [34] would generate some of
the couplings in (4.19) of our Matrix string theory action. More specifically, the couplings
(3)
(3)
(3)
to C+ij and Cij k are reproduced although, since the gauge C+j = 0 was chosen in [36],
the coupling to these components of C (3) cannot be checked. Of course, neither can the
couplings to the remaining RR fields, since it is not known how or, indeed, if their elevendimensional counterparts couple to the supermembrane.
It can be seen from (4.9) that as regards the RR 5-form potential, only the terms of the
(5)
(5)
form Ca1 ...a4 9 contribute. It is easy to check that the terms involving Ca1 ...a5 would couple
to the current I5 , which was previously set to zero in order to match the action (2.4) with
the action of the multiple D0-brane theory of [2,3]. Setting the BornInfeld vector to zero,
the remaining 5-form RR field couplings can be written as:

R
(5)
i
(5)
(5)
C+ + X i C+i
,
X i Ci
SC (5) =
(4.22)
d d STr
4gs
R

(5)
C
= Xk , Xj DXi Cij(5)k .
(4.23)
Similarly, the C (7) couplings only have contributions involving terms of the form Ca(7)
,
1 ...a7
couple
to
currents
for
which
the
explicit
expression
is
not
since terms of the form Ca(7)
1 ...a6 9
known, corresponding to N (7) couplings in the D0-brane action (2.4). The Ca(7)
terms
1 ...a7
can be written as

R2
1 (7)
1 (7) i (7)
i
, (4.24)
C
C
+
+
X
C
d
d
STr
SC (7) =
i
R
R
96gs3
2
2
with

C(7) = Xn , Xm Xl , Xk Xj , Xi Cij(7)klmn .
(4.25)
Let us for the sake of completeness also consider the couplings to b (6) and N (7) . As
for C (5) , only the terms with a 9-component appear in the action. The other components
38
couple to currents for which the explicit expression is not known. The b (6) couplings can
be written as:

R
(6)
1
i (6)
i (6)
Sb (6) =
(4.26)
d d STr
b + X b+i X bi ,
16gs2
R +
where

(6)
b
= Xl , Xk Xj , Xi bij(6)kl ,
and the N (7) couplings are given by

(7)
1 R2
i (7)
i (7)
N
SN (7) =
X
N
+
X
N
d
d
STr
+
+i
i ,
R
16gs2
(4.27)
(4.28)
where

(7)
N
= Xm , Xl Xk , Xj DXi Nij(7)klm .
(4.29)
Note that the

couplings of the different fields occur at a different order of the expansion
parameter R/ .
5. T-duality once more: multiple IIB F-strings

To describe fundamental strings in the type IIB theory, we perform another T-duality in
the x 9 direction, as in (3.7) and (3.8). As before, we assume that the worldvolume fields
(the currents) do not change, so the flat action (4.3) is unchanged. The linear action (4.9)
becomes

1
1
1

(2) a9
IIB
Slinear
=
Ih + ( h99 )Ih99
d d 2 STr (hab ab )Ihab + Ca9
2
R
2
2
1
(2) ab
(4) abc
( + h99 )I Cab
Is 2ha9 Isa9 + ba9 I0a C (0) I09 + Cabc9
I2
2
1 (4)
1
1
C
+ 3bab I2ab9 +
I a1 ...a4 9 +
ba ...a 9 I a1 ...a5 +
ba ...a I a1 ...a6 9
12 a1 ...a4 4
60 1 5 4
48 1 6 6

1 (8)
a ...a
Ca1 ...a7 9 I6 1 7 ,
(5.1)
336
which should describe strongly coupled Matrix strings in the IIB theory, with the currents
as
appendix, up to the coordinate transformations (4.2), and the rescalings Xi
in the
i

X and gs gs /(2).
Note that precisely the same action is obtained if one applies the S-duality rules (4.6)
and (4.7) to the D1-brane action (3.9). Although this might be expected from the S-duality
connection between D- and F-strings in the type IIB theory, it is not clear a priori how
such an S-duality should be done directly, since we are dealing with non-abelian fields; as
in YangMills theory, we cannot rigorously perform a worldvolume duality transformation
to show the S-duality equivalence between the two.
39
It is worth pointing out however that the S-duality transformation of the NSNS 2-form:
b(2) C (2) ,
C (2) b(2),
(5.2)
implies that the dynamics will now be governed by open D-strings, given that the invariant
2-form field strength that will couple in the worldvolume is constructed with C (2) instead
of b(2) . Therefore the BornInfeld field A should transform into a new worldvolume vector
field A whose abelian component forms a gauge invariant field strength with the RR 2form. The right S-duality transformation rule should be then
A A ,
A A,
(5.3)
in order to match (5.2). This cannot, however, be seen explicitly at the level of the linearized
actions that we have considered.
The action above can be rewritten in a more convenient form, filling in the expressions
for the currents as given in Appendix A (upon the rescaling). In particular, for the Chern
Simons action we have:

1
i
(2) gs (0)
IIB
SCS =
C F +
(iX iX )C (4)
b +
d d STr P
2
R2
R
gs R
1
R
+ i(iX iX )b(2) F 2 (iX iX )2 b (6)
(iX iX )2 C (4) F
2gs
2gs

iR
iR 2
+
(5.4)
(iX iX )3 C (8) 2 (iX iX )3 b (6) F .
6gs
6gs3
Again we note that only half of the terms necessary to form the pullback of C (8) are present
in the linear action. Some of the above couplings have been given before in [18].
6. T-duality along a transverse direction: multiple IIA strings with winding number
Let us now also describe type IIA fundamental strings with winding number. The IIA
strings that are described by Matrix string theory carry momentum p+ . This is the charge
which is related by the 9-11 flip to the number of D-particles. T-duality in the x 9 direction
gives IIB strings with winding number, which we have just checked are S-dual to multiple
D-strings. Strongly coupled IIA F-strings with winding number can then be obtained from
IIB strings by performing a T-duality transformation in a direction transverse to the IIB
strings. These strings with winding number have interesting dielectric properties that we
will discuss in the next section.
Calling z the T-duality direction and a = (0, i), where now i = 1, . . . , 7, the linear action
that is obtained from (5.1) is given by:

1
1
1

IIA
Slinear
=
d d 2 STr
hab + ab hzz Ihab baz Ihaz
2
R
2
2

1 1
1
1
(3) az
hzz + Ihzz + Ca9z
h99 + hzz Ih99 C9(1) Ihz9
Ih
2 2
2
2
40

1
1
(3)
hzz + h99 I Cabz Isab 2Ca(1) Isaz 2ha9Isa9 2bz9Isz9
2
2
(5)
+ ba9I0a + hz9 I0z Cz(1) I09 + Cabc9z I2abc 3Cab9I2abz + 3bab I2ab9
1 (7)
1
a ...a
N
ba ...a z9 I a1 ...a4 z
6haz I2az9 +
I 1 5+
60 a1 ...a5 9z 4
12 1 4 4
1 (5)
1 (3)
1 (9)
C
C
+
I a1 ...a4 9 + Cabc I4abcz9
I a1 ...a7
12 a1 ...a4 z 4
3
336 a1 ...a7 9z 6

1 (7)
1 (7)
1
Ca1 ...a6 9 I6a1 ...a6 z +
Na1 ...a6 z I6a1 ...a6 9 + ba1 ...a5 z I6a1 ...a5 z9 .
+
48
48
8
(6.1)
The direction z in which the T-duality is performed appears as an isometry direction in the
transverse space of the strings. We denote the corresponding Killing vector as
k = z ,
k = z + hz .
(6.2)
In a manner similar to the KaluzaKlein monopole, the non-abelian strings do not see this
special direction, the embedding scalar Xz is not a degree of freedom of the strings, but is
transformed under T-duality into a world volume scalar [31]. This worldvolume scalar
forms an invariant field strength with baz (see [18] for the details), and can, therefore,
be associated to fundamental strings wrapped around the isometry direction z, which
themselves end on the Matrix strings.
The action (6.1) can be written in a covariant way as a gauged sigma model,
where gauge-covariant derivatives D X are used to gauge away the embedding scalar
corresponding to the isometry direction [38]:
D X = D X k D X k ,
(6.3)
with = , . These gauge-covariant derivatives reduce to the standard covariant

derivatives D X for = z and are zero for = z. The pull-backs that appear in the action
of the F-strings with winding are constructed from these gauge covariant derivatives. For
example,

P b (2) = b DX DX dt dx

R
R
R2
i
i
i
j
= b09 + b0i DX + bi9 X + bij X DX dt dx.

(6.4)

Filling in the expressions for the currents as given in the appendix (upon the rescaling)
we can write the ChernSimons action as:

(1)
1
(2)
gs
(1)
ik C F + 2 b
k D
Slinear =
d d STr P
2
R
R
R
i
i
i[X,] C (3) + (iX iX )C (3) D
+ i(i[X,] )k (1) F +
gs R
gs
(3)
R
i
(iX iX )ik C (5)
+ (iX iX ) i[X,] C F
gs R
gs
R
1
+
(iX iX )2 ik C (5) F + 2 i[X,] (iX iX )ik b (6)

gs
2gs
41
R
R2
i
(iX iX )2 ik b (6) F
(iX iX )2 ik b (6) D +
[X,]

2

2g
2gs
s
1
iR 2
(iX iX )2 ik N (7) 2 (iX iX )3 ik N (7) F
2
2gs
6gs
iR 2
R
3 i[X,] (iX iX )3 C (7) +
(iX iX )3 ik C (9)
3
2gs
6gs

,
(6.5)
where we have introduced the following types of interior multiplication:

(ik )1 ...p = k 1 ...p = z1 ...p ,

i[X,] ... = Xi , i1 ...p .
1
(6.6)
Again only half of the terms to form the pullback of C (7) and C (9) are present in the linear
action (6.1). Some of the couplings in (6.5) have been derived earlier in [18], applying Tand S-duality relations to the actions of D-branes. We will see in the next section that the
presence of the isometry direction will enable us to find dielectric solutions of F-strings
expanding into Dp-branes with p even.
7. Dielectric solutions
It is well-known in the case of D-branes that some of the linear couplings in (2.4)
and (3.9) give rise to stable, dielectric configurations. In particular, it was shown [2,4]
that a set of coinciding Dp-branes in the presence of a (p + 4)-form RR field strength
2 . This configuration is stable
will expand into a non-commutative or fuzzy two-sphere, SNC
2 . This dielectric
and can be identified with a fuzzy D(p + 2)-brane of topology Mp+1 SNC
D(p + 2)-brane has no net (monopole) charge with respect to the (p + 4)-form RR field
strength, but it does have a dipole moment.
In this section we will comment on various solutions describing F-strings expanding
into D-branes.
7.1. Dielectric D-strings
It is instructive to review the case of N D-strings expanding into a dielectric D3-brane
in our notation. The action for N coinciding D-strings is given by the action (3.5) for
a flat background plus the action (3.9) for the linear couplings, which in the static case
(X i = DXi = A = 0) gives rise to the following potential:

1
i k j i (5)
2
X X X F09ij k ,
VD1 = STr 2 [X, X] +
(7.1)
4
3
where, in order to obtain the couplings to the five-form field strength, we performed a
partial integration and a non-abelian Taylor expansion [2,39]
(4)
(4)
(4)
C
(X) = C
(0) + k C
(0)Xk + .
(7.2)
42
The equation of motion

i j j i k j (5)
X , X F09ij k = 0,
X ,X ,X +
2
is clearly satisfied for the following choice of F (5) and Xi :
(7.3)
(5)
F09ij k = f ij k ,

Xi = f i , where i , j = 2iij k k ,
(7.4)
4
where i, j, k = 1, 2, 3. In other words, three of the transverse coordinates of the D-string
form a N N matrix representation of SU(2). The D-strings have expanded into a fuzzy
two-sphere with radius
1 i i
2 2 2
Tr X X =
f N 1 ,
(7.5)
N
16
2 , with no net D3-brane charge, but
representing a (fuzzy) D3-brane of topology M2 SNC
(5)
with a dipole moment with respect to F . The potential (7.1) for the solution (7.4) gives
R2
2 4 2
f N N 1 ,
(7.6)
384
which is clearly negative. The dielectric D3-brane thus has a lower energy than the original
configuration of N coinciding D-strings, and the latter will spontaneously decay into the
former.
VD1 =
7.2. Dielectric F-strings in type IIB

It is clear that for the type IIB F-strings, a similar effect will occur. If we compare
the linear couplings (5.4) of the type IIB F-strings with those of the D-strings (3.10),
we notice that, although the former now has a monopole charge with respect to b(2) , the
dielectric coupling to C (4) is identical to that for the D-strings. Hence the action (5.4) for N
coinciding type IIB F-strings gives rise to the same type of potential:

1
i
(5)
IIB
= STr 2 [X, X]2 +
Xk Xj Xi F09ij k ,
VF1
(7.7)
3gs
4gs
where now we performed a non-abelian Taylor expansion
R
(4)
(4)
(4)
C (X) = C (0) + k C (0)Xk + .

Clearly, the potential (7.7) has a solution of the same type as (7.4):
(5)
(7.8)
F09ij k = f ij k ,

gs
Xi = f i , where i , j = 2iij k k .
(7.9)
4
The interpretation now is that the N F-strings have expanded to form a fuzzy D3-brane
2 , the S-dual of the D1D3 configuration described above. Again,
with topology M2 SNC
this is a stable configuration with radius and energy as in (7.5) and (7.6) respectively.
43
Similarly, the (iX iX )2 b (6) couplings in (5.4) indicate that F-strings can expand into an
NS5-brane with quadrupole moments, this being the S-dual of a D1-D5 configuration. A
solution of N D-strings expanding into a D5-brane was given in [4], using the gamma
matrices of SO(5). This solution however turned out to be unstable, and we expect to find
a similar unstable solution here.
7.3. Matrix string theory: dielectric F-strings with momentum in type IIA
The case of type IIA Matrix string theory is more subtle. A solution, claimed to
describe N F-strings expanding into a spherical D2-brane, has been given in [17]. There,
the C (3) -couplings were considered in the static gauge = t, = x 9 , rather then in the
light cone gauge as in Section 4. The potential is then

1
i
IIA
2
k j i (4)
VF1 = STr 2 [X, X] +
(7.10)
X X X F0ij k ,
4gs
3gs
(4)
which clearly has a minimum of the form (7.9) with F0ij k = f ij k . However, the question
arises as to whether this solution really corresponds to a string expanding into a D2-brane,
given that the worldvolume of the D2-brane ((t, x i ) in Cartesian coordinates) does not
contain the world sheet direction, x 9 , of the string. How a collection of strings could give
rise to such a solution, where no trace of the spatial component of the worldvolume of
the strings is found, is unclear. Moreover, the dipole coupling (iX iX )C (3) in the Chern
Simons action, that would describe such a situation, cannot appear as a pull-back to a
two-dimensional worldvolume. Nor is it possible to construct from the action (4.19) a RR
four-form field strength coupling of the form F09ij or F+ij , which would include the
world sheet coordinate of the string in the dielectric brane.
Let us take static gauge X9 = in the linear action (4.9) so that we can better identify
the dielectric couplings that appear in the ChernSimons part of this action. In that case,
one immediately realizes that the couplings cannot be written as a pull-back to a twodimensional worldvolume. Moreover, the different terms can be organized as pull-backs
to a one-dimensional worldvolume which has X9 as an isometric direction. It turns out
that the resulting action describes multiple pp-waves carrying momentum along the X9
direction, as in [40]. Pp-waves carrying momentum in a compactified direction are, of
course, T-dual to fundamental strings wound around this direction [41,42]. There is a
sense, then, in which waves can be considered as fundamental strings carrying momentum
(see [43] for a discussion of this point). We have found that the action that describes type
IIA Matrix strings coupled to background fields reduces to that describing multiple waves
when one goes to the static gauge, which is in agreement with the fact that Matrix strings
carry momentum along the compactified direction.
The interpretation in terms of waves sheds some light on the dielectric solution for
IIA Matrix strings constructed in [17]. Recall that the description of multiple waves
is in terms of a gauged sigma model in which the direction of propagation of the
waves is isometric [40]. Indeed, in this action one finds a coupling (iX iX )C (3) to the
worldvolume [18], but now the pull-backs include gauge-covariant derivatives (see the
discussion about gauged sigma-models in Section 6) which gauge away X9 as a world
44
volume scalar. The resulting dielectric D2-brane is therefore constrained to move in the
space transverse to the isometric direction.
In summary, we believe that the solution
(4)
F0ij
k = f ij k ,

gs
(7.11)
4
presented in [17] corresponds to a set of N gravitational waves expanding into a spherical
D2-brane, which again is a stable configuration with radius and energy as in (7.5)
and (7.6).4
We have argued in the introduction that we do not expect to find stable solutions
describing F-strings expanding into a cylindrical D2-brane. Since we cannot find a coupling
of the F-strings to either F09ij or F+ij , which would give rise to such a dielectric brane,
this expectation is confirmed. Such a configuration simply does not exist in the type IIA
Matrix string theory. Although we find a possible cylindrical D2-brane solution below, it
is unstable. Whether it corresponds to the unstable abelian D2-brane solution is unclear,
however.
7.4. Dielectric F-strings with winding in type IIA

Coinciding F-strings with winding can expand into a spherical D4-brane in the presence
of an RR 6-form field strength. We can see this by analysing the C (5) couplings in the
action (6.1), which can be rewritten as the following coupling to the RR six-form field
strength:

i
(6)
IIA
d d STr Xk Xj Xi F09zij k .
SF1
(7.12)
3gs
It is then clear that a stable solution of the potential

1
i
(6)
IIA
VF1
(7.13)
= STr 2 [X, X]2 +
Xk Xj Xi F09zij
k ,
3gs
4gs
is given by
(6)
F09zij
k = f ij k ,

gs
(7.14)
4
and corresponds to N F-strings expanding into a D4-brane, which is itself wrapped around
the isometry direction z. We believe this solution corresponds to the (smeared) supergravity
solution of an F-string expanding into a D4-brane, the T-dual of the fully localized F1D6
solution given in [6,7].
Coinciding F-strings with winding also have couplings to the RR 3-form field
potential, which seems to suggest an expansion into a cylindrical D2-brane. Yet these C (3) couplings have an unusual form, due to the presence of the world volume scalar field . In
4 A similar type of solution, for a magnetic 4-form field strength, has been constructed in [44].
the static case the potential is given by

1
1
IIA
VF1
= STr 2 [X, X]2 2 [X, ]2
4gs
2gs

j i
j (4)
i i j
i
X , X + X , X + , X X F09ij .
6gs
45
(7.15)
It is easy to check that the following ansatz

(4)
F09ij = f ij ,
i j
X , X = 0,

Xi , = igs f ij Xj ,
(7.16)
where i, j = 1, 2, is a solution of the equations of motion

(4)

= 0,
Xi , Xj , Xj + Xi , , igs Xj , F09ij

i
(4)
= 0.
Xi , , Xi gs Xi , Xj F09ij
2
(7.17)
The solution (7.16) is the algebra of the symmetry group of a cylinder. A similar solution
was recently given in [45] using D0-brane matrix degrees of freedom.5 This brane has no
net monopole charge with respect to the F (4) field strength but carries a dipole moment in
the XZ and YZ planes.
From (7.16), the worldvolume of the D2-brane is (t, x 9 , x i ) in Cartesian coordinates,
which suggests a cylindrical worldvolume. However, the proper interpretation is not
entirely clear since the solution explicitly involves the worldvolume scalar field . It is
possible that this solution is dual to the unstable solution of the abelian D2-brane theory,
since it turns out to correspond to a maximum of the potential and is therefore unstable.
The only stable radius for the cylinder is
R2 =
1 i i
Tr X X = 0,
N
(7.18)
as might be expected from the analysis of [6,7]. There is another remark we would like to
make: an explicit representation of the cylindrical algebra (7.16) is [45]
ab = gs f ab b,
aab = eit a1,b ,
X1 = R a + a ,
aab
= eit a+1,b ,
X2 = iR a a
(7.19)
(7.20)
where a, b = 1, . . . , N and for some arbitrary phase , is only possible for the infinitedimensional case. There are no finite-dimensional representations of (7.16), which makes
the solution rather unphysical, requiring an infinite number of F-strings for the effect to
take place.
5 This can be done by adding a small number of D0-branes to the F1D2-brane system and taking then the
D0-brane density to zero.
46
8. Discussion
The dielectric effect of Myers [2] is not limited to D-branes, but should be observable
in a theory of fundamental strings too. To describe such dielectric F-strings, we are led
to a consideration of Matrix string theory, which captures type IIA superstring theory
in the light-cone gauge, together with extra degrees of freedom representing D-brane
states.
We have shown explicitly that the Matrix theory action in a weakly curved background
reproduces the lowest order expansion of the combined non-abelian BornInfeldChern
Simons of Taylor and van Raamsdonk [3] and Myers [2]. Using T-duality and the 9-11
flip, we have derived the Matrix string theory action in a weakly curved background and
shown that, in the weak string coupling limit, the result reproduces light-cone gauge string
theory. We have not been able to find explicit solutions describing dielectric Matrix strings
expanding into a D2-brane in this theory, and have argued that this is to be expected. We
have further argued that the solution found in [17] (in which the static, as opposed to the
light-cone, gauge was considered) in some sense describes dielectric particles or pp-waves,
carrying momentum.
We have derived a, strongly coupled, type IIB Matrix string theory, S-dual to the Dstring theory, and a, strongly coupled, type IIA theory of Matrix strings with winding.
As regards the former, the standard D1D3 dielectric solution of the D-string theory
equally well describes a collection of F-strings expanding into a D3-brane and other,
albeit unstable, solutions of the D-string theory carry through in a similar way. The
theory of Matrix strings with winding exhibits an isometry direction, in a manner similar
to the theory of the KaluzaKlein monopole, and one can find dielectric D4-brane
solutions which are smeared over this direction. Finally, we have mentioned a cylindrical
solution, the interpretation of which is somewhat unclear since it involves the worldvolume
scalar .
On a different note, one might expect to find solutions of the type IIA Matrix string
theory corresponding to the cylindrical supertubes of [46,47]. As far as both the D2brane theory and the supergravity solutions are concerned, these involve a cylindrical D2brane with dissolved F-string and D0-brane charge. The configuration is supported from
collapse by an angular momentum in the circle direction. It should be clear that possible
solutions of Matrix string theory corresponding to such supertubes would then involve
couplings to both C (3) and C (1) . As far as the D0-branes are concerned, however, the D2brane worldvolume is a non-commutative cylinder [48,49]. Then, since the F-strings lie
in the direction along the cylinder, the worldsheet direction would also have to be noncommutativeand this cannot occur in Matrix string theory.
The proper determination of both the Matrix and the Matrix string theories in an
arbitrary background remains an open problem. Of course, they could be determined
by duality transformations of multiple D-brane theories in an arbitrary background, but
the correct form of the latter is also only poorly understood; and so are the duality
transformations for these non-abelian actions. We are tempted to think that one might make
progress via a comprehensive treatment of light-cone gauge supermembrane theory in a
general background.
47
Acknowledgements
We wish to thank Roberto Emparan for collaborations during the initial stages of this
project. D.B. and B.J. would further like to thank Robbert Dijkgraaf, Clifford Johnson, Paul
Saffin and Douglas Smith for useful discussions. D.B. is supported in part by the EPSRC
grant GR/N34840/01.
Appendix A. D-string currents

T-duality applied to the D0-brane currents is explained in the text. We give the final
results here for completeness. With i, j = 1, . . . , 8, we have
I00 = 1,
I i = X i ,
0
I09
= A,
(A.1)
(A.2)
(A.3)

i
0ij
I2 =
(A.4)
Xi , Xj ,
6
I209i =
(A.5)
DXi ,
6

i
ij k
I2 =
(A.6)
X i Xj , Xk + X j Xk , Xi + X k Xi , Xj ,
6

i j i
9ij
I2 =
(A.7)
i A X , X X DXj + X j DXi ,
6

1
0ij kl
= 2 Xi , Xj Xk , Xl + Xi , Xk Xl , Xj + Xi , Xl Xj , Xk ,
I4
2
(A.8)
j k
k i
i j
i
09ij k
i
j
k
= 2 DX X , X + DX X , X + DX X , X ,
I4
(A.9)
2

15
ij klm
= 2 X [i Xj , Xk Xl , Xm] ,
I4
(A.10)
2

3
9ij kl
I4
(A.11)
= 2 A X[i , Xj Xk , Xl] + 4i X [i Xj , Xk DXl] ,
2

i
0ij klmn
I6
(A.12)
= 3 X[i , Xj Xk , Xl Xm , Xn] ,

09ij klm
I6
(A.13)
= 3 DX[i Xj , Xk Xl , Xm] ,

7i
ij klmnp
I6
(A.14)
= 3 X [i Xj , Xk Xl , Xm Xn , Xp] ,

9ij klmn
= 3 i A X[i , Xj Xk , Xl Xm , Xn] 6X [i Xj , Xk Xl , Xm DXn] .
I6
(A.15)
48
In terms of pullbacks of background fields, we have (for = and a, b, = 0, . . . , 8)

C (0) I09 dt dx = P C (0) F,
(A.16)

(2) a

i
(2) ab9
Ca9 I0 + 3Cab
(A.17)
dt dx = P C (2) + P (iX iX )C (2) F,
I2

1 (4)
(4)
C
Cabc9 I2abc +
I a1 ...a4 9 dt dx
12 a1 ...a4 4

1
i
P (iX iX )2 C (4) F,
= P (iX iX )C (4)
(A.18)
2

1 (6)
1 (6)
a ...a
a ...a 9
C
C
I 1 5+
I 1 6 dt dx
60 a1 ...a5 9 4
48 a1 ...a6 6

i
i
= 2 P (iX iX )2 C (6) 2 P (iX iX )3 C (6) F,
(A.19)
2

6
1 (8)
C
I a1 ...a7 + dt dx
336 a1 ...a7 9 6

i
1
= 3 P (iX iX )3 C (8) +
(A.20)
P (iX iX )4 C (8) F,
3
6
24
giving the action (3.10) in the text. The dots stand for the unknown coupling to C (9) which,
of course, could be determined by demanding that it can be rewritten in terms of the
pullback. The dilaton current is

1 2
1 2
2
2 2
1
2
2
I = 1 X 1 + X + 2 DX +
A 2 [X, X]
2
4
4
4
8
2
2
1

4 [X, X]4 2 X DX
8
2

1
2
2
2 2
1
+ 2 DX2 1 X 2 2 DX2 +
A 2 [X, X]2
4
4
2
4
8
2

4 DXi Xi , Xj Xj , Xk DXk
2

1 2
2 2
2
2 2
1
2
2
A 1 + X 2 DX +
A 2 [X, X]
2
4
4
4
8

1
+ 2 X i Xi , Xj Xj , Xk X k
2

1
2
2 2
1
1 2
2
2
2
A 2 [X, X]
2 [X, X] 1 X + 2 DX
4
4
4
4
8

i2
2 ADXi Xi , Xj X j ,
(A.21)
and the remaining NSNS currents are

Ih00 = 1 +
1 2
2
2 2
1
X + 2 DX2 +
A 2 [X, X]2 ,
2
2
2
4
(A.22)
49

1
2
1
Ih0i = X i 1 + X 2 + A 2 + 2 DX2 2 [X, X]2
2
2
2
4

i i j 2
2 X , X ADXj i Xj , Xk X k ,
(A.23)

i
1
2 2
2
1
Ih09 = A 1 + X 2 +
A 2 DX2 2 [X, X]2 + 2 DXi Xi , Xj X j ,
2
2
2
4
(A.24)
2

1
ij
Ih = X i X j 2 DXi DXj 2 Xi , Xk Xk , Xj ,
(A.25)

i
Ih9i = A X i 2 Xi , Xj DXj ,
(A.26)
2
Ih99 = 2 A 2 2 DX2 ,
(A.27)

2
1
i i j j

0i
i
Is =
(A.28)
X ,X X ,
ADX +
2

X DX,
Is09 =
(A.29)
2

i i j
2
1
1 2 2 2
ij
2
2
Is =
X ,X
A + 2 DX 2 [X, X]
1 X
2
2
2
2
4
2
2

1
[i
i
X [i Xj ] , Xk X k +
AX DXj ] + 3 DX[i Xj ] , Xk DXk
i i k k l l j
3 X ,X X ,X X ,X ,
(A.30)
2

2 2
2
1
1
Is9i =
DXi 1 X 2 +
A 2 DX2 2 [X, X]2
2
2
2
2
4

i i j j
i
A X ,X X +
X DX X 3 Xi , Xj Xj , Xk DXk . (A.31)
2
2
2
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
R. Emparan. Phys. Lett. B 423 (1998) 71, hep-th/9711106.

R.C. Myers, JHEP 9912 (1999) 022, hep-th/9910053.
W. Taylor IV, M. van Raamsdonk, Nucl. Phys. B 573 (2000) 703, hep-th/9910052.
S.P. Trivedi, S. Vaidya, JHEP 0009 (2000) 041, hep-th/0007011.
M.S. Costa, C.A.R. Herdeiro, L. Cornalba, Nucl. Phys. B 619 (2001) 155, hep-th/0105023.
R. Emparan, Nucl. Phys. B 610 (2001) 169, hep-th/0105062.
D. Brecher, P. Saffin, Nucl. Phys. B 613 (2001) 218, hep-th/0106206.
P.J. Silva, Matrix string theory and the Myers effect, hep-th/0111121.
L. Motl, Proposals on nonperturbative superstring interactions, hep-th/9701025.
T. Banks, N. Seiberg, Nucl. Phys. B 497 (1997) 41, hep-th/9702187.
R. Dijkgraaf, E. Verlinde, H. Verlinde, Nucl. Phys. B 500 (1997) 43, hep-th/9703030.
50
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
M. Claudson, M.B. Halpern, Nucl. Phys. B 250 (1985) 689.

T. Banks, W. Fischler, S. Shenker, L. Susskind, Phys. Rev. D 55 (1997) 5112, hep-th/9610043.
W. Taylor IV, Phys. Lett. B 394 (1997) 283, hep-th/9611042.
D. Kabat, W. Taylor IV, Phys. Lett. B 426 (1998) 297, hep-th/9712185.
W. Taylor IV, M. van Raamsdonk, JHEP 9904 (1999) 013, hep-th/9812239.
R. Schiappa, Nucl. Phys. B 608 (2001) 3, hep-th/0005145.
Y. Lozano, Phys. Rev. D 64 (2001) 106011, hep-th/0012137.
W. Taylor IV, M. van Raamsdonk, Nucl. Phys. B 558 (1999) 63, hep-th/9904095.
J. Polchinksi, Prog. Theor. Phys. Suppl. 134 (1999) 158, hep-th/9903165.
L. Susskind, Another Conjecture about M(atrix) theory, hep-th/9704080.
N. Seiberg, Phys. Rev. Lett. 79 (1997) 3577, hep-th/9710009.
A. Sen, Adv. Theor. Math. Phys. 2 (1998) 51, hep-th/9709220.
M.R. Douglas, Nucl. Phys. (Proc. Suppl.) 68 (1998) 381, hep-th/9707228.
M.R. Douglas, A. Kato, H. Ooguri, Adv. Theor. Math. Phys. 1 (1998) 237, hep-th/9708012.
M.R. Douglas, H. Ooguri, Phys. Lett. B 425 (1998) 71, hep-th/9710178.
A.A. Tseytlin, Nucl. Phys. B 501 (1997) 41, hep-th/9701125.
A. Dasgupta, H. Nicolai, J. Plefka, JHEP 0005 (2000) 007, hep-th/0003280.
J. Plefka, Int. J. Mod. Phys. A 16 (2001) 660, hep-th/0009193.
Y. Hyakutake, Expanded Strings in the Background of NS5-branes via a M2-brane, a D2-brane and D0branes, hep-th/0112073.
E. Eyras, B. Janssen, Y. Lozano, Nucl. Phys B 531 (1998) 275, hep-th/9806169.
M. Goroff, J.H. Schwarz, Phys. Lett. B 127 (1983) 61.
B. de Wit, J. Hoppe, H. Nicolai, Nucl. Phys. B 305 (1988) 545.
Y. Sekino, T. Yoneya, Nucl. Phys. B 619 (2001) 22, hep-th/0108176.
E. Bergshoeff, E. Sezgin, P.K. Townsend, Phys. Lett. B 189 (1987) 75.
B. de Wit, K. Peeters, J.C. Plefka, Supermembranes and supermatrix models, in: I. Antoniadis, L.E. Ibaez,
J.W.F. Valle (Eds.), Valencia 1997, Beyond the Standard Model: From Theory to Experiment, World
Scientific, 1998, p. 433, hep-th/9712082.
B. de Wit, K. Peeters, J. Plefka, Nucl. Phys. B 532 (1998) 99, hep-th/9803209.
E. Bergshoeff, B. Janssen, T. Ortn, Phys. Lett. B 410 (1997) 131, hep-th/9706117.
M.R. Garousi, R.C. Myers, Nucl. Phys. B 542 (1999) 73, hep-th/9809100.
E. Bergshoeff, P.K. Townsend, Nucl. Phys. B 490 (1997) 145, hep-th/9611173.
E. Bergshoeff, I. Entrop, R. Kallosh, Phys. Rev. D 49 (1994) 6663, hep-th/9401025.
B. Janssen, Fortsch. Phys. 47 (1999) 201.
C.M. Hull, Nucl. Phys. B 509 (1998) 216, hep-th/9705162.
S.R. Das, S.P. Trivedi, S. Vaidya, JHEP 0010 (2000) 037, hep-th/0008203.
I. Bena, The polarization of F1 strings into D2 branes: Aut Caesar aut nihil, hep-th/0111156.
D. Mateos, P.K. Townsend, Phys. Rev. Lett. 87 (2001) 011602, hep-th/0103030.
R. Emparan, D. Mateos, P.K. Townsend, JHEP 0107 (2001) 011, hep-th/0106012.
D. Bak, K. Lee, Phys. Lett. B 509 (2001) 168, hep-th/0103148.
D. Bak, S. Kim, Junctions of supersymmetric tubes, hep-th/0108207.

Non-supersymmetric orbifolds
Anamara Font a , Alexis Hernndez b
a Departamento de Fsica, Facultad de Ciencias, Universidad Central de Venezuela,
A.P. 20513, Caracas 1020-A, Venezuela

b Departamento de Fsica, Universidad Simn Bolvar, A.P. 89000, Caracas 1080-A, Venezuela
Received 27 February 2002; accepted 23 April 2002
Abstract
We study compact non-supersymmetric ZN orbifolds in various dimensions. We compute the
spectrum of several tachyonic type II and heterotic examples and partially classify tachyon-free
heterotic models. We also discuss the relation to compactification on K3 and CalabiYau manifolds.
2002 Elsevier Science B.V. All rights reserved.
1. Introduction
Orbifold compactifications provide a large class of exactly solvable string vacua in
which the number of supersymmetries is reduced [1]. In this paper we will discuss type II
and heterotic orbifolds in which supersymmetry is completely broken by the twisted
boundary conditions. This generalizes the HarveyDixon construction [2] of the tendimensional non-supersymmetric strings that can be found in various other ways [3].
To our knowledge, the study of non-supersymmetric compact orbifolds has not been
carried out systematically until now. Compactifications of the d = 10 non-supersymmetric
theories were considered some time ago [4]. Non-supersymmetric closed string models
in lower dimensions have been devised using asymmetric constructions [5], the Green
Schwarz formulation [6], ScherkSchwarz compactifications [7] and compactifications on
magnetic backgrounds [8]. The question of vanishing of the cosmological constant in nonsupersymmetric strings has been addressed as well [9].
In the non-supersymmetric orbifolds that we discuss, supersymmetry is broken at the
string scale. Thus, potential phenomenological applications in d = 4 will face the hierarchy
problem. Now, an approach to solve this problem in a similar scenario has actually
E-mail addresses: afont@fisica.ciens.ucv.ve (A. Font), ahernand@fis.usb.ve (A. Hernndez).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 3 6 - X
52
A. Font, A. Hernndez / Nuclear Physics B 634 (2002) 5170
been proposed [10] and other mechanisms could be conceived in the future. At any rate,
since experimental evidence indicates that supersymmetry is broken, non-supersymmetric
strings should be explored. Moreover, these theories have interesting features that are
worth studying in more depth. Notably, such theories typically have tachyons and are thus
unstable. It is then necessary to investigate processes of vacuum stabilization. Recently,
condensation of closed type II tachyons has been analyzed in the case of non-compact
orbifolds [11]. The fate of tachyons in compact orbifolds has however not been considered
in detail. The case of heterotic tachyons has only been discussed in d = 10 [12]. In
this paper we will present several tachyonic examples in d = 8, 6, 4, that could serve as
starting points to study type II and heterotic tachyon condensation in compact orbifolds.
Furthermore, tachyons could be of relevance in electroweak symmetry breaking [13] and
in inflation [14]. Yet another interesting issue is that of duality of string theories without
supersymmetry. We have found several non-tachyonic heterotic models whose strong
coupling duals could be described as in the SO16 SO16 case [15].
This paper deals with type II and heterotic non-supersymmetric compactifications on
T 10d /ZN for d = 8, 6, 4 and N 6. Our main results consist of concrete examples. In
particular, we have determined all tachyon-free heterotic models for N 5. As a check
on the massless spectra we have verified factorization of anomalies. We have also found
relations to compactifications of the d = 10 non-supersymmetric strings on smooth K3 and
CalabiYau manifolds.
This paper is organized as follows. In Section 2 we introduce the basic notation and
determine the allowed ZN , N 6, twists that break supersymmetry. In Sections 3 and 4
we describe type II and heterotic orbifolds. In Section 5 we present our final comments.
For the sake of self-containedness we include Appendix A with a review of tools needed
to obtain the tachyonic and massless spectrum in compact orbifolds.
2. Generalities
We focus on T 10d /ZN orbifolds although the analysis can be easily extended to
ZN ZM actions and discrete torsion can also be included. In our standard notation the ZN
generator, denoted , acts diagonally on internal complex coordinates as Yi = e2ivi Yi .
Since N = 1, the vi are of the form k/N . The vi are restricted by the condition that
must act crystallographically on the torus lattice. In particular this requires that the number
of fixed points det(1 ) be an integer. All such twists have been classified in Ref. [16].
Spacetime supersymmetry imposes the extra condition
v1 v2 v3 = 0 mod 2.
Modular invariance gives the additional constraint

vi = 0 mod 2.
N
(1)
(2)
Notice that (2) also ensures that is of order N acting on the world-sheet fermionic degrees
of freedom.
53
Table 1
Twist vectors for non-supersymmetric ZN actions, N 6
ZN
(v1 , v2 , v3 )
ZN
(v1 , v2 , v3 )
Z2
(0,0,1)
Z6
(0,0, 31 )
Z3
(0,0, 32 )
Z4
(0, 0, 12 )
(0, 14 , 34 )
( 12 , 12 , 12 )
Z5
(0, 15 , 35 )
(0, 16 , 12 )
(0, 16 , 56 )
(0, 13 , 23 )
( 16 , 16 , 23 )
( 13 , 13 , 13 )
( 12 , 12 , 13 )
All crystallographic twists satisfying (1) and (2) were given in Ref. [1]. If we relax (1)
there will be no gravitini left in the untwisted sector so that supersymmetry is generically
broken. In this case we can still use the results of Ref. [16] to find the allowed (v1 , v2 , v3 ).
The results for N 6 are shown in Table 1. For greater N there are of course other
solutions. Notice that in some Z2M examples, the generator leaves the holomorphic
two or three form invariant so that the resulting orbifold is either a singular K3 or a
singular CalabiYau. Hence, we will be able to describe a class of non-supersymmetric
compactifications in the orbifold limit of these manifolds.
It is also necessary to specify the torus lattice that allows a given action. We mostly
consider products of two-dimensional sublattices. More precisely, for order two and order
four rotations we take the SO4 root lattice whereas for order three and order six rotations
we take the SU 3 root lattice. The Z5 action is realized as the Coxeter rotation on the SU 5
root lattice.
Several of the twist vectors in Table 1 have been considered previously. The Z2 orbifold
was originally analyzed in the context of non-supersymmetric 10-dimensional strings [2].
The Z5 has been considered in Ref. [17] in relation to type IIB orbifolds of AdS5 S 5 .
The Z4 with eigenvalues ( 12 , 12 , 12 ) has been used in [18] and [19] to discuss type 0B
orbifolds and in [20] to construct non-supersymmetric tachyon-free orientifolds. The Z4
with (0, 0, 12 ) has also been used to build a type IIB orbifold [19]. More recently, some of
these generators have been considered in the case of non-compact orbifolds [11].
3. Type II
Symmetric type II orbifolds with the non-supersymmetric twists in Table 1 will have
tachyons in some twisted sector. Notice that in some of the Z2M examples, the Mth
twisted sector is actually untwisted in the sense that M acts trivially in spacetime. In
such cases the M sector always includes a tachyon that could be referred to as untwisted.
Since in these examples M = (1)FS , where FS is the spacetime fermion number, there
are altogether no spacetime fermions in the orbifold spectrum. In particular, the Z2 twist
(0, 0, 1) leads to the ten-dimensional type 0 closed strings. Below we will mostly describe
examples with spacetime fermions.
54
3.1. d = 8
In this case the little group SO8 is broken to SO6 SO2 , with 8v = (6, 0) + (1, 1) +
(1, 1) and 8s = (4, 12 ) + (4, 12 ), where the second entry gives the U1 SO2 charge.
As an example let us take the T 2 /Z3 with v = (0, 0, 0, 23 ). Using the results shown
in Appendix A, we see that the massless states in the untwisted sector include the metric,
dilaton and antisymmetric tensor. In the type IIB case, the remaining tachyonic and
massless matter is given by
0 : 1 + 1 + 15 + 4 + 4,
: 3(1 ) + 3(1 + 1 + 15 + 4 + 4),
(3)
where states are labelled by their SO6 representations and 1 denotes a tachyon. The three
tachyons (one per each fixed point) in the sector have m2R = m2L = 13 . The massless
states in the sector all have oscillators acting on the twisted vacuum. To the above we
must add the antiparticles.
In type IIA, in the untwisted sector there appears instead
(1 + 4) (1 + 4) = 1 + 6 + 10 + 4 + 4
(4)
and in the twisted sector the same with multiplicity three.

3.2. d = 6
In this case we can classify the massless states according to the little group SO4 . We
use SO4 SU 2 SU 2 so that the vector ( 1, 0 ) becomes the ( 12 , 12 ) representation, the
spinor ( 12 , 12 ) becomes ( 12 , 0) and ( 12 , 12 ) is (0, 12 ). In the untwisted sector there always
appear massless states in the SU 2 SU 2 representations (1, 1) + (0, 0) + (1, 0) + (0, 1)
that correspond to the metric, dilaton and antisymmetric tensor. The remaining tachyonic
and massless spectrum depends on the particular ZN .
As a first example we take the T 4 /Z4 with v = (0, 0, 14 , 34 ). In the and 3 sector
there appear tachyons, denoted (0, 0) , with m2R = m2L = 14 . There are no tachyons in
the 2 sector since 2v is supersymmetric. However, the states in this sector do not fill
supersymmetric multiplets. In type IIB, the tachyonic and massless matter includes

0:
14(0, 0) + 2(0, 1) + 4(1, 0) + 8 0, 12 ,

, 3 : 8(0, 0) + 40(0, 0) + 8(1, 0) + 32 12 , 0 ,

(5)
2:
50(0, 0) + 10(0, 1) + 24 0, 1 .
2
Notice that this massless spectrum is anomaly-free since the number of (0, 1) and (1, 0)
tensors is the same and the number of (0, 12 ) and ( 12 , 0) fermions is the same. We recall that
in the supersymmetric T 4 /Z4 , the massless spectrum consists of a gravity multiplet and 21
tensor multiplets, exactly as in type IIB compactification on a smooth K3. In the case at
hand, the massless states in the sector seem to fill tensor multiplets but the states differ
in the twisted oscillators acting on them. In type IIA, the tachyonic and massless matter
includes instead

+ 4 0, 12 + 4 12 , 0 ,

, 3 : 8(0, 0) + 32(0, 0) + 8 12 , 12 + 16 12 , 0 + 16 0, 12 ,

2:
40(0, 0) + 10 12 , 12 + 12 0, 12 + 12 12 , 0 .
0:
8(0, 0) + 6
1
1
2, 2
55
(6)
In the supersymmetric orbifold the massless spectrum has one gravity multiplet and 20
vector multiplets as in a K3 compactification.
Acting on the internal coordinates, the non-supersymmetric Z4 has the same action
as the supersymmetric one with v = (0, 0, 14 , 14 ). We may thus suspect that the nonsupersymmetric orbifold is related to a non-supersymmetric compactification on K3. The
same observation applies to the T 4 /Z6 with v = (0, 0, 16 , 56 ).
Yet another example that can be related to a K3 compactification is the T 4 /Z6 with
v = (0, 0, 13 , 23 ). In this case there are no fermions in the spectrum. Besides the untwisted
tachyon, there are 18 twisted tachyons. The massless matter in the type IIB (type IIA)
orbifold includes 128 (80) singlets and 24 antisymmetric tensors (48 vectors). Now, this Z6
is of the form Z3 Z2 where the Z3 has the supersymmetric twist v = (0, 0, 13 , 13 ) and
the Z2 has the (1)FS twist v = (0, 0, 0, 1). Hence, we expect this orbifold to correspond
to a K3 compactification of type 0 strings. Indeed, counting the zero modes in K3 of the
massless type 0 fields we find the same massless matter as in the orbifold. However, it is
not clear how the 18 twisted tachyons would appear in type 0 on K3 or disappear in the
orbifold upon blowing up the singularities. In supersymmetric type II T 4 /ZN orbifolds the
massless spectrum coincides directly with the K3 spectrum.
Finally let us discuss the T 4 /Z5 that has no supersymmetric analog. In type IIB the
tachyonic and massless matter is given by

8(0, 0) + 2(0, 1) + 2(1, 0) + 4 0, 12 + 4 12 , 0 ,
0:

, 4 : 10(0, 0) + 20(0, 0) + 10(1, 0) + 20 12 , 0 ,

(7)
2 , 3 : 10(0, 0) + 20(0, 0) + 10(0, 1) + 20 0, 1 .
2
In type IIA we instead find

4(0, 0) + 4 12 , 0 + 4 0, 12 + 4 12 , 12 ,
0:

, 4 : 10(0, 0) + 10(0, 0) + 10 12 , 12 + 10 12 , 0 + 10 0, 12 ,

2 , 3 : 10(0, 0) + 10(0, 0) + 10 12 , 12 + 10 12 , 0 + 10 0, 12 .
(8)
3.3. d = 4
Now the massless states are classified by the little group SO2 , i.e., by helicity . For
a L R state, = r p where r can be read from the first component of the SO8
weight r, and likewise for p . In the untwisted sector, as usual, we find the metric, dilaton
and antisymmetric tensor and now also a graviphoton. The remaining matter depends on
the ZN .
As an example, we take the Z6 with v = (0, 16 , 16 , 23 ). In the sectors 5 and there appear
tachyons with m2R = m2L = 16 and multiplicity three (one per each fixed point). There are
56
no tachyons in other twisted sectors since 2v and 3v are supersymmetric. However, the
matter in these sectors do not fill supersymmetric multiplets. Let us now consider type IIB
more specifically. In the 3 sector there further appear 5 massless vectors. We recall that in
the supersymmetric case with v = (0, 16 , 16 , 13 ) there are 5 vector multiplets in this sector.
The remaining massless states, labelled by helicity, are

0 : 11(0) + 8 12 ,

5 : 15(0) + 12 12 ,

2 : 30(0) + 24 12 ,

(9)
3 : 18(0) + 20 12 .
In all the d = 4 spectra we do not include the antiparticles. In type IIA we find extra
massless vectors, five in the untwisted sector, 3 in the + 5 , 15 in 2 + 4 and 6 in 3 .
The remaining massless states are

0 : 6(0) + 4 12 + 4 12 ,

5 : 12(0) + 6 12 + 6 12 ,

2 : 15(0) + 12 12 + 12 12 ,

(10)
3 : 17(0) + 10 12 + 10 12 .
This T 6 /Z6 orbifold can be seen as a singular CalabiYau with h11 = 29 and h12 = 5.
Besides the gravity multiplet and the dilaton hypermultiplet, the type IIA supersymmetric
compactification has then 29 vector multiplets and 5 hypermultiplets.
We now discuss the T 6 /Z6 with v = (0, 13 , 13 , 13 ) so that Z6 is of the form Z3 Z2
where the Z3 has the supersymmetric twist v = (0, 13 , 13 , 23 ) and the Z2 has the (1)FS
twist v = (0, 0, 0, 1). Hence, we expect this orbifold to be related to type 0 compactification
on the singular T 6 /Z3 CalabiYau with h11 = 36 and h12 = 0. In this orbifold there are
no twisted tachyons, only the untwisted tachyon that appears in the 3 sector. Besides the
graviton, the massless spectrum in the type IIB (IIA) orbifold includes 222 (78) singlets
and 2 (74) vectors. On the other hand, counting zero modes on a CY of the type 0B
massless fields gives (6 + 6h11 + 2h12 ) singlets and (2 + 2h12 ) vectors. For type 0A there
are instead (6 + 6h12 + 2h11 ) singlets and (2 + 2h11 ) vectors, as one might guess from
mirror symmetry. Thus, the full spectrum of this non-supersymmetric T 6 /Z6 coincides
with the type 0 compactification.
4. Heterotic
To study heterotic strings we need to describe the embedding in the gauge degrees of
freedom. We use the bosonic formulation and realize the embedding by a shift vector V
such that NV , where is either the E8 E
8 or the Spin(32)/Z2 lattice. Modular
invariance of the partition function imposes N I VI = 0 mod 2 but this is trivially
satisfied. Modular invariance further requires

N V 2 v 2 = 0 mod 2
(11)
57
Table 2
Z2 embeddings and groups
E8 E8
Spin(32)/Z2
2V
7 8
2, 0 0
2 6 2 6
1 ,0 1 ,0
7 7
2, 0 2, 0
8 8
1 ,0
4 12
1 ,0
15
2, 0
1 16
2
Gauge group
SO16 E8
E7 SU 2 E7 SU 2
SO16 O16
SO16 SO16
SO24 SO8
SO32
U16
For instance, in Z2 we can take V = 0 that leads back to the original heterotic string [2].
Indeed, in some cases supersymmetry can be recovered because missing partners appear in
the extra twisted sectors. More generally, tachyons can be eliminated from the spectrum by
level matching. The prime example is the non-supersymmetric non-tachyonic SO16 SO16
heterotic string in ten dimensions.
In the E8 E8 case, to determine the corresponding shift vectors V , and the
corresponding unbroken gauge group, we use the method of deleting nodes in the extended
Dynkin diagram [21]. For Spin(32)/Z2 we use a similar procedure. All Z2 embeddings
were given in Ref. [2] and are conveniently collected in Table 2 (the notation, say 1n ,
means that the entry 1 is repeated n times). The number of allowed V s grows rapidly
with N so that we will refrain from giving complete lists. In the following we will instead
describe some selected examples briefly.
We have searched systematically for tachyon-free models for the ZN , with N 5. By
analyzing the mass formula for left movers (43) we find that for some, typically larger
V 2 , the putative tachyons cannot have m2R = m2L and hence disappear by level matching.
Tachyons can also be eliminated by the generalized orbifold projection described in the
appendix. Concrete examples are described below.
In the following we will also provide some tachyonic examples. In particular we will
consider the standard embedding V = (v1 , v2 , v3 , 0, . . . , 0) that always leads to models
with tachyons.
4.1. d = 8
We consider first the T 2 /Z3 with v 2 = 49 . In Table 3 we display the five E8 E8 allowed
shifts. There are six embeddings in Spin(32)/Z2 that lead to gauge group of the form
U3k+1 SO306k and that have 3V of type (13k+1 , 0153k ) or (13k , 2, 0153k ).
As an example we consider the standard embedding. In E8 E8 the resulting gauge
group is SO14 U1 E8 and the matter spectrum, including tachyons, is

1
0 : (1 + 4) 14, 1 + 64s ,
+ (1, 0) ,
2
2 2

1
2
: 3 1 14,
+ 1,
3 2
3 2

1
1
4
2
(12)
+ 3(1 + 4) 14,
+ 64s ,
+ 1,
+ 2 1,
.
3 2
6 2
3 2
3 2
58
Table 3
Z3 embeddings and groups in E8 E8
3V
7 8
2, 0 0
2 6 2 6
1 ,0 1 ,0

2, 12 , 06 2, 07

1
7
2 6
2 5, 1 1 , 0
1 5, 17 1 5, 17
2
2
Gauge group
SO14 U1 E8
E7 U1 E7 U1
E6 SU 3 SO14 U1
SU 9 E7 U1
SU 9 SU 9
Here (1 + 4) indicates scalar plus fermion and (1 ) tachyon. Together with the SO14
representations we also give the U1 charge Q. In Spin(32)/Z2 the gauge group is
SO30 U1 and the spectrum is

0 : (1 + 4) 30, 1 + (1, 0) ,
2

1
2
: 3(1 ) 30,
+ 1,
3 2
3 2

1
4
2
(13)
+ 3(1 + 4) 30,
+ 1,
+ 2 1,
.
3 2
3 2
3 2
In all the d = 8 heterotic spectra such as the above antiparticles are not included. Given
the results in Appendix A, it is straightforward to compute the spectrum for all V s. We
find that there is only one tachyon-free model with gauge group SU 9 SU 9 and massless
matter given by

0 : (1 + 4) (1, 1) + (84, 1) + (1, 84) ,
(14)
: 3(1 + 4)(9, 9).
For the T 2 /Z4 with v 2 = 14 there are 14 embeddings in E8 E8 and 17 in Spin(32)/Z2 .
There are only two tachyon-free models in which potential tachyons are eliminated by the
generalized orbifold projection. One model, with V = 18 (7, 17 ) 18 (7, 17), has group
(SU 8 SU 2 )2 and massless matter given by

0 : (1) (70, 1, 1, 1) + (1, 1, 70, 1) + 2(1, 1, 1, 1)

+ (4) (28, 2, 1, 1) + (1, 1, 28, 2) ,
:
2:
4(4)(8, 1, 8, 1),

(4) (28, 1, 1, 2) + (1, 2, 28, 1) .
(15)
The other non-tachyonic model has V = 14 (112 , 04 ), group SU 12 SO8 U1 and the
following massless matter

1
,
0 : (1) 66, 1, 1 + 66, 1, 1 + 2(1, 1, 0) + (4) 12, 8v ,
6
6
2 6

1
3
3
: 4(4) 66, 1,
+ 1, 8s ,
+ 2 1, 1,
,
2 6
2 6
2 6

1
2
(16)
: (4) 12, 8c , .
2 6
Notice that in the above Tr Q = 0.
59
We have verified anomaly factorization in all models. To compute the anomaly

polynomial we just need the contribution of a 4 fermion in a representation Ri of the
gauge group. This is given by
A4 =
1
1
1
1 2 2
Tri F 5
tr R 2 Tri F 3 +
tr R 4 Tri F +
tr R Tri F.
5
12
240
192
(17)
We see then that factorization requires cancellation of the irreducible tr F 5 for all nonAbelian factors. Moreover it must be that Tr Q = 0 and 12 Tr Q5 = 5 Tr Q3 for all U1
factors. Summing the contribution of all massless fermions in each model we find that the
anomaly polynomial factorizes as

1
Trall F3
A=
(18)
va tr Fa2 tr R 2 .
12
a
Here a runs over all gauge factors and tr Fan stands for trace in the fundamental
representation, whereas Trall F3 stands for trace of all gauge factors over all massless
fermions. The coefficients va depend only on the group. Using the conventions of Ref. [23]
1
for SU N , U1 , SON , E6 , E7 and E8 respectively. To prove
they are va = 2, 1, 1, 13 , 16 , 30
Eq. (18) we use the results of Ref. [23] together with identities such as
Traij F 5 = (N 16) tr F 5 + 10 tr F 2 tr F 3 ,
1
Traijk F 5 = (N 6)(N 27) tr F 5 + 10(N 6) tr F 2 tr F 3
2
for antisymmetric SU N representations with N 5.
(19)
4.2. d = 6
We study first the T 4 /Z4 . The generator with v = (0, 0, 14 , 34 ) has the same V s that
appear in the supersymmetric v = (0, 0, 14 , 14 ). There are 12 in E8 E8 and 14 in
Spin(32)/Z2 . In the latter case the resulting gauge groups can be of the form Um SO2n
SO322m2n , with m + 4n 2 = 0 mod 8. These arise from 4V with vector structure of
type (1m , 2n , 016mn ). It is also possible to obtain gauge group U152k U2k+1 that
result from 4V without vector structure of type 12 (1152k , (3)2k+1).
For the standard embedding in E8 E8 the resulting gauge group is SO12 SU 2
U1 E8 and the charged m2 0 matter is

0:
12, 2, 12 + 32c , 1, 12 + 2(1, 1, 0) + c.c.
2(0, 0) + 0, 12

+ 2 12 , 0 (32s , 2, 0) + (1, 1, 1) + (1, 1 1) ,

, 3 : 4(0, 0) 12, 1, 14 + 2 1, 2, 14 + c.c. + 4 2(0, 0) + 12 , 0

32s , 1, 14 + 1, 2, 34 + 2 12, 1, 14 + 3 1, 2, 14 + c.c. ,

1

2:
0, 2 10(32c , 1, 0) + 6(12, 2, 0) + 32 1, 1, 12 + 32 1, 1, 12

+ 2(0, 0) 10(12, 2, 0) + 6(32c , 1, 0) + 32 1, 1, 12 + 32 1, 1, 12 . (20)
60
In the SO32 heterotic the standard embedding has gauge group SO28 SU 2 U1 and the
following tachyonic plus massless matter

0 : 2(0, 0) + 0, 12
28, 2, 12 + 2(1, 1, 0) + c.c.

+ 2 12 , 0 (1, 1, 1) + (1, 1 1) ,

: 4(0, 0) 28, 1, 14 + 2 1, 2, 14 + c.c.

+ 4 2(0, 0) + 12 , 0 2 28, 1, 14 + 3 1, 2, 14 + 1, 2, 34 + c.c. ,

2 : 0, 12 6(28, 2, 0) + 32 1, 1, 12 + 32 1, 1, 12

(21)
+ 2(0, 0) 10(28, 2, 0) + 32 1, 1, 12 + 32 1, 1, 12 .
In this Z4 we find several non-tachyonic examples. In both heterotic strings there are
models with group SO12 U2 SO16 . There is also a model with group SO10 SO10 U6
and another with SO10 SU 4 SU 8 SU 2 and massless matter

2(0, 0) + 0, 12 (16, 4, 1, 1) + (1, 1, 28, 2) + 2(1, 1, 1, 1) + c.c.
0:

+ 2 12 , 0 (10, 6, 1, 1) + (1, 1, 70, 1) ,

, 3 : 4 2(0, 0) + 12 , 0 (1, 4, 8, 1) + c.c. ,

1
2:
0, 2 10(10, 1, 1, 2) + 6(1, 6, 1, 2)

(22)
+ 2(0, 0) 6(10, 1, 1, 2) + 10(1, 6, 1, 2) .
This model has V = 14 (3, 13 , 04 ) 18 (7, 17).
We next discuss the T 4 /Z6 with v = (0, 0, 13 , 23 ). There is a large number of allowed
embeddings, a few of which with 3V 2 = 23 mod 2 and 3V give back supersymmetric
models. Although we have not searched systematically for tachyon-free models we have
found a particular non-tachyonic example with group SO12 SU 2 SO16 U1 that can
be obtained in both heterotic strings. For instance, in the SO32 heterotic with embedding
V = 16 (36 , 12 , 08 ) the resulting spectrum is

0:
2(0, 0) 12, 2, 1, 12 + (1, 1, 1, 1) + 2(1, 1, 1, 0) + c.c.

+ 0, 12 1, 2, 16, 12 + c.c. + 2 12 , 0 (12, 1, 16, 0),

, 5 : 9 12 , 0 32s , 1, 1, 16 + c.c. ,

2 , 4 : 18(0, 0) 12, 2, 1, 16 + 2 1, 1, 1, 23 + 5 1, 1, 1, 13 + c.c.

+ 9 0, 12 1, 2, 16, 16 + c.c. ,

3:
2 0, 12 (32c , 2, 1, 0) + (1, 1, 128, 0)

(23)
+ 12 , 0 32s , 1, 1, 12 + 32s , 1, 1, 12 .
N
As in d = 10, the E8 E8 spectrum follows from (23) upon exchanging the k and 2 k
fermions. As explained in Section 3.2, in this case Z6 = Z3 Z2 where the Z3 respects
61
supersymmetry and Z2 is equivalent to (1)FS . Thus, this orbifold is arguably related to

K3 compactification of the 10-dimensional non-supersymmetric heterotic theories. In fact,
we claim that the model (23) can also be obtained from compactification of the tachyonfree SO16 SO16 theory on K3.
To support our claim we first observe that
SO16 SO12 SU 2 SU 2
(24)
Thus, one SO16 can be broken to SO12 SU 2 by an SU 2 background with instanton number
k = 24. This is the standard embedding in the SO16 SO16 theory whose bosonic fields
are (g , B , ) and A in the adjoint (120, 1) + (1, 120). The fermionic fields are 8s
spinors in the (16, 16) representation and 8c spinors in (128, 1) + (1, 128). The number
of zero modes of the massless d = 10 fermions that transform in a representation R of
G10 = SO16 can be determined from an index theorem [24]. One first decomposes R under
G10 G H as

(Ri , Si ),
R=
(25)
i
where Ri and Si denote representations of the unbroken group G and the bundle group H .
In the case at hand, G = SO12 SU 2 and H = SU 2 but many other examples can be
worked out. The basic formula for the number of massless d = 6 fermions transforming in
the representation Ri is
ni = kT (Si ) dim Si ,
(26)
where T (Si ) is TrSi

For SU 2 , T (2) = and T (3) = 2. We use conventions such
that positive (negative) ni corresponds to (0, 12 ) (( 12 , 0)) spinors. Also, recall that in
the supersymmetric case, (26) gives actually the number of hypermultiplets which have
2(0, 12 ). Hence there is an extra factor of two to be taken into account. To count the zero
modes of the gauge field presumably we can use supersymmetry as a calculational tool and
suppose that there exist gauginos in the adjoint. In the supersymmetric case, gaugino zero
modes transforming as (0, 12 ) spinors belong to hypermultiplets in which the scalars arise
from the gauge field. Hence, the number of scalar zero modes is also given by (26) taking
into account a multiplicity of four since hypermultiplets have four scalars. Recall further
that zero modes of the metric and antisymmetric tensor give rise to 80 neutral scalars. We
will need the following decompositions
T 2.
1
2
16 = (12, 1, 1) + (1, 2, 2),

128 = (32c , 2, 1) + (32s , 1, 2),
120 = (66, 1, 1) + (1, 3, 1) + (1, 1, 3) + (12, 2, 2).
(27)
Then, (26) with k = 24 implies that from the (16, 16) fermions there arise 2 ( 12 , 0)
fermions transforming as (12, 1, 16) and 20 (0, 12 ) fermions transforming as (1, 2, 16).
This agrees with (23) ignoring the U1 charges, i.e., assuming that the U1 is broken.
Results for the number of fermions in the (32c , 2, 1) and (32s , 1, 1) representations
also agree in this way. To obtain the massless scalars, we also use (26) as explained
above. Then, from the decomposition of the 120 we find 40 scalars transforming as
62
(12, 2, 1), same as in the orbifold, and 180 singlets. Adding the 80 moduli gives 260
singlets as compared to 264 in the orbifold. Presumably four scalars, corresponding to
one hypermultiplet, disappear when the U1 is higgsed away as in the supersymmetric
compactification.
We might as well consider compactification of the d = 10 tachyonic models on
K3. For example, we can start with the SO16 E8 theory whose matter consists of
tachyons transforming as (16, 1) and massless fermions 8s transforming as (128, 1) plus
8c transforming as ( 128, 1) [2]. The standard embedding is obtained by embedding an SU 2
instanton bundle with k = 24 in SO16 so that the unbroken group is SO12 SU 2 E8 . Using
Eqs. (27) and (26) one can easily compute the resulting massless content. Now, we can
compare the results with those of the standard embedding in the T 4 /Z6 with Z6 = Z3 Z2 .
In this orbifold the gauge group is SO12 SU 2 U1 E8 and the massless states match
those in the K3 compactification neglecting U1 charges. As expected, in the orbifold
there is an untwisted tachyon transforming as (12, 1, 0) but there appear furthermore the
following twisted tachyons

9 12, 1, 13 + 2 1, 2, 16 + c.c. .
(28)
The states transforming as (1, 2, 16 ) are of blowing-up type, i.e., they have left-moving
oscillators acting on the twisted vacuum.
We finally consider the T 4 /Z5 orbifold that has no supersymmetric analog. Given
2
v = 25 we find 18 distinct embeddings in E8 E8 . In the SO32 heterotic there are
instead 20 possibilities that lead to gauge groups of the form SO3210n U5n and
SO302j 4n Un+j Un+1 , j = 1, 6, 11. There are groups that appear in both heterotic
strings, namely SO12 U5 U5 , SO14 U7 U2 , U8 U8 and SO10 U3 U8 . In
this orbifold the shift V = 0 is allowed. Hence, there are non-supersymmetric models in
six dimensions with group E8 E8 and SO32 whose tachyonic and massless spectrum
is
4(0, 0) + 2(0, 12 ) + 2( 12 , 0),

, 4 : 30(0, 0) + 50(0, 0) + 50 12 , 0 ,

2 , 3 : 30(0, 0) + 50(0, 0) + 50 0, 12 .
0:
(29)
All states are neutral. In this Z5 orbifold there is only one tachyon-free model with
V = 15 (4, 14 , 03 ) 15 (4, 14, 03 ), group SU 45 and the following massless spectrum

(0, 0) + 0, 12 (5, 10, 1, 1) + (1, 1, 5, 10) + (1, 1, 1, 1) + c.c.

+ (0, 0) + 12 , 0 (10, 5, 1, 1) + (1, 1, 10, 5) + c.c. ,

, 4 :
5 (0, 0) + 12 , 0 (5, 1, 5, 1) + c.c. ,

2 , 3 : 5 (0, 0) + 0, 12 (1, 5, 1, 5) + c.c. .
0:
(30)
As a final example, we give the spectrum for the model with standard embedding in
E8 E8 . The gauge group is SO12 U12 E8 and the tachyonic plus massless matter
turns out to be
63

(0, 0) + 0, 12
12, 12 , 12 + 32c , 0, 12 + (1, 1, 0) + (1, 0, 0) + c.c.

+ (0, 0) + 12 , 0

12, 12 , 12 + 32s , 12 , 0 + (1, 0, 1) + (1, 0, 0) + c.c. ,

1
3
+ 1, 35 , 15 + 2 1, 25 , 15 + c.c.
5(0, 0) 12, 10
, 10
, 4 :

1 1
1
3
+ 5 (0, 0) + 12 , 0
32s , 10
, 5 + 12, 10
, 10

+ 1, 25 , 45 + 2 1, 35 , 15 + 3 1, 25 , 15 + c.c. ,

3
1
2 , 3 : 5(0, 0) 12, 10
+ 1, 15 , 35 + 2 1, 15 , 25 + c.c.
, 10

1
3
1
+ 5 (0, 0) + 0, 12
, 10
32c , 15 , 10
+ 12, 10

+ 1, 45 , 25 + 2 1, 15 , 35 + 3 1, 15 , 25 + c.c. .
(31)
0:
All states are neutral under E8 .

We have checked anomaly factorization in all models. The starting point is the
contribution to the anomaly polynomial due to a ( 12 , 0) fermion in a representation Ri
of the gauge group. This is

dim Ri
5 2 2
1
4
A 1 =
(32)
tr R + tr R
+ Tri F 4 tr R 2 Tri F 2 .
( 2 ,0)
240
4
4
For (0, 12 ) fermions there is an overall minus sign. Cancellation of the irreducible tr R 4
requires equal number of (0, 12 ) and ( 12 , 0) fermions which is a first easy check on the
spectrum. Cancellation of the irreducible tr F 4 term can also be simply checked using the
results of Ref. [23] or Ref. [25]. In fact, we have found that in all models the full anomaly
polynomial factorizes as

2
2
2
za tr Fa
va tr Fa tr R .
A=
(33)
a
The coefficients za depend on the spectrum of massless fermions and can be computed
using the results in Ref. [23] or in Ref. [25] that employs slightly different conventions
such that va = 1 for all groups.
4.3. d = 4
In the T 6 /Z4 with v = (0, 12 , 12 , 12 ) we have looked systematically for tachyon-free
models. We find that tachyons are eliminated by the orbifold projection in examples with
groups U6 U10 , U2 U14 and SU 8 SU 2 E6 U2 . There are two more non-tachyonic
examples in this orbifold that have groups U8 SO6 SO10 and SU 8 U12 SO14 and
can be constructed in both heterotic strings.
For other T 6 /ZN we have not performed a systematic search of non-tachyonic models,
although we have some examples. For instance, in the Z6 with v = (0, 16 , 16 , 23 ) there are
models with group SO16 SO10 SU 2 U12 in both heterotic strings.
64
In the Z6 with v = (0, 13 , 13 , 13 ) there are many allowed embeddings, including a few
with 3V 2 = even and 3V that actually produce supersymmetric models. As explained
in Section 3.3, this orbifold can be related to compactifications of the d = 10 nonsupersymmetric theories on the singular T 6 /Z3 CalabiYau with h11 = 36 and h12 = 0.
Thus, we expect to find non-tachyonic models arising from the SO16 SO16 theory. Indeed,
in both heterotic strings we obtain tachyon-free examples with group SO16 U6 SO4 ,
SO14 U1 SO12 U2 and SO16 SO10 SU 3 U1 . In the latter the massless spectrum
turns out to be

0 : 3(0) 1, 10, 3, 1 + 1, 1, 3, 2 + 3(1, 1, 1, 0) + 12
6
6

1
3
3
(128, 1, 1, 0) + 3 1, 16, 3,
+ 1, 16, 1,
+ 1, 16, 1,
,
2 6
2 6
2 6

1
: 27 2 16, 1, 1, 1 ,
6

1
2 : 27(0) 1, 10, 1, 1 + 3(1, 1, 3, 0) + 1, 1, 1, 2 + 27 12 1, 16, 1,
,
6
6
2 6

1
1
(34)
.
3:
2 (16, 10, 1, 0) + 3 16, 1, 3,
6
= (1, 07 ) 16 (3, 13, 04 ). In the
This is obtained in E8 E8 with V

SO(32) heterotic with
1 8 5 3
V = 6 (0 , 3 , 1 ) we obtain the same states.
The model in Eq. (34) should correspond to the standard embedding in the SO16 SO16
on the T 6 /Z3 CalabiYau orbifold. To support this claim we count the number of zero
modes of the d = 10 fields as in the supersymmetric case. We start by decomposing
representations under
SO16 SO10 SU 3 U1 .
(35)
The second SO16 is broken by a background in SU 3 equal to the spin connection.

Neglecting U1 charges to simplify and already including the unbroken SO16 we have
(1, 120) = (1, 45, 1) + (1, 1, 8) + (1, 1, 3) + (1, 1, 3) + (1, 10, 3) + (1, 10, 3),
(16, 16) = (16, 10, 1) + (16, 1, 3) + (16, 1, 3 ),
(1, 128) = (1, 16, 3) + (1, 16, 3) + (1, 16, 1) + (1, 16, 1),
(128, 1) = (128, 1, 1).
(36)
Since the (16, 10, 1) and the (128, 1, 1) are inert under SU 3 , there is only one zero
mode for each that transforms respectively as (16, 10) and (128, 1) under SO16 SO10 .
This agrees with the orbifold result. From (16, 1, 3) + (16, 1, 3 ) there follow instead
h11 + h12 zero modes transforming as (16, 1). Now, h11 = 36 and h12 = 0 so that the total
number coincides with the orbifold spectrum in Eq. (34) if we count SU 3 dimensionality
as multiplicity, i.e., assuming that SU 3 is broken. In fact, the 81 scalars transforming
as (1, 1, 3, 0) in the 2 sector are precisely analogous to the blowing-up modes in the
supersymmetric orbifold. Next, recalling that in d = 10 the (1, 128) has opposite chirality,
we find h11 + 1 zero modes transforming as (1, 16), and h12 + 1 transforming as (1, 16 ).
Notice that the number of families is |h12 h11 |. Again, when SU 3 is broken, this counting
agrees with the orbifold result. Finally we can also deduce the number of charged scalars
arising from the zero modes of the gauge field. From the adjoint decomposition we obtain
65
h11 + h12 states in the (1, 10) representation. In the orbifold there are indeed 36 such
scalars.
We can as well consider CY compactification of the tachyonic SO16 E8 theory.
The standard embedding leads to group SO10 U1 E8 . Using (36) one can easily
compute the number of charged zero modes that depend on h11 and h12 . One can then
compare for instance to the standard embedding in the non-supersymmetric T 6 /Z6 with
v = (0, 13 , 13 , 13 ) which has group SO10 U1 SU 3 E8 . The charged massless states agree
assuming that SU 3 is broken upon repairing the orbifold singularities. Among the orbifold
massless states there are indeed blowing-up scalars. In the orbifold there is only one
untwisted charged tachyon, transforming in the 10 of SO10 , that in the CY compactification
presumably arises from the untwisted d = 10 tachyon transforming as 16 of SO16 .
To verify anomaly factorization we need the contribution to the anomaly polynomial
due to a Weyl fermion in a representation Ri of the gauge group. This is
A 1 = Tri F 3
2
1
tr R 2 Tri F.
8
(37)
Cancellation of the irreducible tr F 3 term amounts to cancellation of non-Abelian cubic

anomalies. We find that in all models the anomaly properly factorizes as

1
2
2
Tr Fb
va tr Fa tr R ,
A=
(38)
8
a
b
where b runs only over the U1 factors with Tr Qb = 0. This factorization requires several
non-trivial relations among the U1 charges such as Tr Qb = 8 Tr Q3b . It can be shown that
there is at most one combination of U1 s that is anomalous.
5. Final comments
In this work we have used compact orbifolds to construct tachyonic and non-tachyonic
non-supersymmetric models in various dimensions. Many more examples can be built.
For instance, one can use ZN ZM orbifolds including discrete torsion [26]. Another
interesting extension is to add Wilson lines to further break the gauge group [27].
Connections to non-perturbative orbifolds [28] and orbifolds of M-theory [25,29] can be
explored as well.
We have found that some ZN non-supersymmetric orbifolds can be related to K3 and
CalabiYau (CY) compactifications of the type 0 and non-supersymmetric heterotic strings
in d = 10. Indeed, there are tachyon-free orbifolds that match compactifications of the nontachyonic SO16 SO16 string. Since orbifolds are well defined string vacua and stable in the
absence of tachyons, it then seems that even without supersymmetry compactification on
K3 and CY manifolds gives consistent vacua at least at tree level. The standard embedding
in K3 compactification that we discussed appears to be one example of a whole class of
d = 6 non-tachyonic models whose massless spectrum can be easily obtained and verified
to satisfy anomaly factorization using basic results. We have also analyzed the standard
embedding in CY and compared in a particular case to the corresponding orbifold.
66
We might also ask what happens when a d = 10 tachyonic theory is naively

compactified on K3 or a CY and how this relates to a corresponding orbifold in which the
spectrum is exactly known. Concretely, in d = 6 we can compare to the orbifold T 4 /Z6
with v = (0, 0, 13 , 23 ) and in d = 4 to T 6 /Z6 with v = (0, 13 , 13 , 13 ). For type 0 on K3 we
found that the massless states agree but there is a mismatch in the number of twisted
tachyons. For type 0 in CY, results fit those of type II on the orbifold in which there is
only one untwisted tachyon as in type 0. In the heterotic case we studied compactification
of the tachyonic SO16 E8 theory with standard embedding. In d = 6, the orbifold has
extra twisted charged tachyons that resemble blowing-up modes. In d = 4, CY and orbifold
results match.
Further orbifold examples associated to compactification of non-supersymmetric
d = 10 theories on K3 and CY can be worked out by taking the orbifold point group to be
ZN Z2 where the ZN is of supersymmetric type and Z2 is (1)FS . In some orbifolds such
as the T 4 /Z4 with v = (0, 0, 14 , 34 ) or the T 6 /Z6 with v = (0, 16 , 16 , 23 ), there should also be
a relation to K3 or CY compactifications in which supersymmetry is broken. However, the
interpretation is not as simple as in the ZN Z2 case.
To conclude we want to stress that there are many other non-supersymmetric orbifolds,
such as T 4 /Z5 , that are not of K3 or CY type. From the point of view of the orbifold
construction all modular invariant choices of twist and embedding are on the same footing
and can be equally analyzed. In particular, numerous examples can be built and used to
study tachyon condensation. It would be specially interesting to examine heterotic models.
Acknowledgements
A.F. thanks Dieter Lst and Ralph Blumenhagen for conversations, Stefan Theisen for
helpful discussions and observations on the manuscript, and Luis Ibez and Fernando
Quevedo for comments on the manuscript. A.F. is also grateful to the Institute of Physics of
Humboldt University, Berlin, for hospitality and the Alexander von Humboldt Foundation
for support while part of this work was carried out. This work is partially supported by a
CDCH-UCV grant No. 03.214.2001 and a Conicit grant No. G-2001000712.
Appendix A. Orbifold spectrum
To find the spectrum for each model, we follow the analysis in the supersymmetric
case [22]. There are N sectors twisted by k , k = 0, 1, . . . , N 1. Each particle state is
created by a product of left and right vertex operators L R. At a generic point in the torus
moduli space, the tachyonic and massless right movers follow from
1
1
(r + kv)2 + Ek .
(39)
2
2
In this formula NR is the occupation number of the right-moving oscillators and Ek is the
twisted oscillator contribution to the zero point energy given by
m2R = NR +
Ek =
3

k
i=1
vi (1 kvi ).
(40)
67
When kvi > 1 we must substitute kvi (kvi 1) in (40). The vector r is an SO8 weight.
When r belongs to the scalar or vector class, r takes the form (n0 , n1 , n2 , n3 ), with ni
integer, this is the NeveuSchwarz (NS) sector. When r belongs to an spinorial class it
+ 12 , n2 + 12 , n3 + 12 ), this is the Ramond (R) sector. The GSO
takes the form (n0 + 12 , n1
projection turns out to be ra = odd. For example, in the untwisted sector, the condition
m2R = 0 implies r 2 = 1 and the possible solutions are

1 1 1 1
r = ( 1, 0, 0, 0 ) = 8v ,
(41)
r = , , ,
= 8s ,
2 2 2 2
where underlining means permutations. The vector v in (39) is (0, v1 , v2 , v3 ), with the vi
given in Table 1. For simplicity we are setting = 2 everywhere.
The mass formula for left-movers depends on the type of string. For type II we have
1
1
(p + kv)2 + Ek ,
(42)
2
2

where p is an SO8 weight as well. In type IIB the GSO
projection is also pi = odd in
both NS and R sectors. In type IIA one has instead
pi = even in the R sector. In the
untwisted sector the spinor weights are then those of 8c .
For the heterotic string in the lattice formulation the left mass formula is instead
m2L = NL +
1
(P + kV )2 + Ek 1,
(43)
2
where P belongs to the lattice . The vectors (P + kV ) determine the representation of the
state under the gauge group. NL is the occupation number for all left-moving oscillators.
States in a k -twisted sector with given m2R = m2L are characterized by (NR , r; NL , p) in
type II or by (NR , r; NL , P ) in the heterotic string. The degeneracy of these states follows
from the generalized orbifold projection given by
m2L = NL +

k 1 N1
k , 3 (k, 3).
D =
N
(44)
3=0
In the heterotic string the phase is

(k, 3)

1
= exp 2i (r + kv) 3v (P + kV ) 3V + k3 V 2 v 2 + 3(L + R ) .
2
(45)
In type II the phase is analogous with p, v instead of P , V . The factors R,L appear only
in the case of states with oscillation number due to the internal (complex) coordinates Yi .
The phase e2i indicates how the oscillator is rotated by . In the sector, for example,
i , i
the right-moving oscillators are of the form v
1vi , . . . with R = vi and oscillators
i
i
i
1+vi , 2+vi , . . . with R = vi . Similarly, for left-movers there can be oscillators
i
i
i
i
, 2+v
, . . . with L = vi .
vi , 1vi , . . . with L = vi and oscillators 1+v
i
i
n
m
In (44) ( , ) is a numerical factor that counts the fixed point multiplicity. More
concretely, (1, 3 ) = 1, so that in the untwisted sector D(1) projects out precisely the
68
states non-invariant under . In twisted sectors ( k , 3 ) is the number of simultaneous

fixed points of k and 3 .
As an example, let us work out the type IIB string on T 2 /Z3 with v = (0, 0, 0, 23 ). In
the untwisted sector there are no tachyons and the massless states are as follows. With
r v = p v = 0 we have
r
( 1, 0, 0, 0)
p
( 1, 0, 0, 0).
(46)
The first three entries in r, p correspond to non-compact coordinates and give the Lorentz
representations according to the little group SO6 . The NSNS states in (46) are thus
6 6 = 20 + 15 + 1. These are the metric, the antisymmetric tensor and the dilaton. With
r v = p v = 23 mod int we have
r
1
(0, 0, 0, 1)
1
1
1
2,2,2,2
1 1 1 1
2, 2, 2,2
1
(0, 0, 0, 1)
1
1
1
2,2,2,2
1 1 1 1
2, 2, 2,2 .

(47)
The four spinor weights form the 4 of SO6 . Then, for instance when we combine right and
left we obtain 1 + 15 in the RR sector. With r v = p v = 13 mod int the solutions are as
in (47) with opposite sign. In the -twisted sector the solutions with m2R 0 are
NR
m2R
13
1
1
2,2,2,2
1 1 1 1
2, 2, 2,2
(0, 0, 0, 1)
1
3
r
1
(0, 0, 0, 1)

1
(48)
The solutions with m2L 0 are similar. The solutions in the 2 -twisted sector are as in (48)
with opposite sign and will lead therefore to the antiparticles of the states in the sector.
In type IIA we have to take the ps arising instead from 8c . This implies, for example,
to change the p spinor weights in (47) to the 4 of SO6 .
References
[1] L. Dixon, J. Harvey, C. Vafa, E. Witten, Strings on orbifolds, Nucl. Phys. B 261 (1985) 678;
L. Dixon, J. Harvey, C. Vafa, E. Witten, Strings on orbifolds II, Nucl. Phys. B 274 (1986) 285.
[2] L. Dixon, J. Harvey, String theories in ten dimensions without spacetime supersymmetry, Nucl. Phys.
B 274 (1986) 93.
[3] L. Alvarez-Gaume, P. Ginsparg, G.W. Moore, C. Vafa, An O(16) O(16) heterotic string, Phys. Lett. B 171
(1986) 155;
N. Seiberg, E. Witten, Spin structures in string theory, Nucl. Phys. B 276 (1986) 272;
H. Kawai, D.C. Lewellen, S.H. Tye, Classification of closed fermionic string models, Phys. Rev. D 34 (1986)
3794;
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
69
W. Lerche, D. Lst, A.N. Schellekens, Ten-dimensional heterotic strings from Niemeier lattices, Phys. Lett.
B 181 (1986) 71.
V.P. Nair, A.D. Shapere, A. Strominger, F. Wilczek, Compactification of the twisted heterotic string, Nucl.
Phys. B 287 (1987) 402;
K.S. Narain, M.H. Sarmadi, Removing tachyons by compactification, Phys. Lett. B 184 (1987) 165;
P. Ginsparg, C. Vafa, Toroidal compactification of nonsupersymmetric heterotic strings, Nucl. Phys. B 289
(1987) 414;
H. Itoyama, T.R. Taylor, Supersymmetry restoration in the compactified O(16) O(16) heterotic string
theory, Phys. Lett. B 186 (1987) 129;
D. Lst, Compactification of the O(16) O(16) heterotic string theory, Phys. Lett. B 178 (1986) 174;
Y. Kwon, The anomaly consideration of O(16) O(16) heterotic string theory, Mod. Phys. Lett. A 2 (1987)
531;
T. Sasada, Asymmetric orbifold models of nonsupersymmetric heterotic strings, Prog. Theor. Phys. 95
(1996) 249, hep-th/9508098.
H. Kawai, D.C. Lewellen, S.H. Tye, Four-dimensional type II strings and their extensions: type-III strings,
Phys. Lett. B 191 (1987) 63;
T.R. Taylor, Model building on asymmetric Z3 orbifolds: nonsupersymmetric models, Nucl. Phys. B 303
(1988) 543;
A. Font, L.E. Ibez, F. Quevedo, A. Sierra, Twisted N = 2 coset models, discrete torsion and asymmetric
heterotic string compactifications, Nucl. Phys. B 337 (1990) 119.
A.H. Chamseddine, J.P. Derendinger, M. Quirs, Nonsupersymmetric four-dimensional strings, Nucl. Phys.
B 311 (1988) 140.
R. Rohm, Spontaneous supersymmetry breaking in supersymmetric string theories, Nucl. Phys. B 237 (1984)
553;
S. Ferrara, C. Kounnas, M. Porrati, Superstring solutions with spontaneously broken four-dimensional
supersymmetry, Nucl. Phys. B 304 (1988) 500;
C. Kounnas, M. Porrati, Spontaneous supersymmetry breaking in string theory, Nucl. Phys. B 310 (1988)
355;
I. Antoniadis, A possible new dimension at a few TeV, Phys. Lett. B 246 (1990) 377;
I. Antoniadis, C. Muoz, M. Quirs, Dynamical supersymmetry breaking with a large internal dimension,
Nucl. Phys. B 397 (1993) 515, hep-ph/9211309;
K. Benakli, Perturbative supersymmetry breaking in orbifolds with Wilson line backgrounds, Phys. Lett.
B 386 (1996) 106, hep-th/9509115;
C.A. Scrucca, M. Serone, A novel class of string models with ScherkSchwarz supersymmetry breaking,
JHEP 0110 (2001) 017, hep-th/0107159.
J.G. Russo, A.A. Tseytlin, Heterotic strings in uniform magnetic field, Nucl. Phys. B 454 (1995) 164, hepth/9506071;
J.G. Russo, A.A. Tseytlin, Magnetic flux tube models in superstring theory, Nucl. Phys. B 461 (1996) 131,
hep-th/9508068.
G.W. Moore, AtkinLehner symmetry, Nucl. Phys. B 293 (1987) 139;
K.R. Dienes, New string partition functions with vanishing cosmological constant, Phys. Rev. Lett. 65 (1990)
1979;
S. Kachru, J. Kumar, E. Silverstein, Vacuum energy cancellation in a non-supersymmetric string, Phys. Rev.
D 59 (1999) 106004, hep-th/9807076;
J.A. Harvey, String duality and non-supersymmetric strings, Phys. Rev. D 59 (1999) 026002, hepth/9807213;
G. Shiu, S.H. Tye, BoseFermi degeneracy and duality in non-supersymmetric strings, Nucl. Phys. B 542
(1999) 45, hep-th/9808095;
D.M. Ghilencea, H.P. Nilles, S. Stieberger, Divergences in KaluzaKlein models and their string
regularization, hep-th/0108183.
K.R. Dienes, Solving the hierarchy problem without supersymmetry or extra dimensions: an alternative
approach, Nucl. Phys. B 611 (2001) 146, hep-ph/0104274.
A. Adams, J. Polchinski, E. Silverstein, Dont panic! Closed string tachyons in ALE spacetimes, hepth/0108075;
70
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
J.A. Harvey, D. Kutasov, E.J. Martinec, G. Moore, Localized tachyons and RG flows, hep-th/0111154;
A. Dabholkar, C. Vafa, tt geometry and closed string tachyon potential, hep-th/0111155;
J.R. David, M. Gutperle, M. Headrick, S. Minwalla, Closed string tachyon condensation on twisted circles,
hep-th/0111212;
S. Nam, S.J. Sin, Condensation of localized tachyons and spacetime supersymmetry, hep-th/0201132.
T. Suyama, Closed string tachyons in non-supersymmetric heterotic theories, JHEP 0108 (2001) 037, hepth/0106079;
T. Suyama, Melvin background in heterotic theories, Nucl. Phys. B 621 (2002) 235, hep-th/0107116;
T. Suyama, Charged tachyons and gauge symmetry breaking, hep-th/0112101.
R. Blumenhagen, L. Goerlich, B. Kors, D. Lst, Noncommutative compactifications of type I strings on tori
with magnetic background flux, JHEP 0010 (2000) 006, hep-th/0007024;
G. Aldazabal, S. Franco, L.E. Ibez, R. Rabadn, A.M. Uranga, Intersecting brane worlds, JHEP 0102
(2001) 047, hep-ph/0011132.
C.P. Burgess, M. Majumdar, D. Nolte, F. Quevedo, G. Rajesh, R.J. Zhang, The inflationary braneantibrane
universe, JHEP 0107 (2001) 047, hep-th/0105204.
J.D. Blum, K.R. Dienes, Duality without supersymmetry: the case of the SO(16) SO(16) string, Phys.
Lett. B 414 (1997) 260, hep-th/9707148;
J.D. Blum, K.R. Dienes, Strong/weak coupling duality relations for non-supersymmetric string theories,
Nucl. Phys. B 516 (1998) 83, hep-th/9707160.
J. Erler, A. Klemm, Comment on the generation number in orbifold compactifications, Commun. Math.
Phys. 153 (1993) 579, hep-th/9207111.
S. Kachru, E. Silverstein, 4d conformal theories and strings on orbifolds, Phys. Rev. Lett. 80 (1998) 4855,
hep-th/980218.
R. Blumenhagen, A. Font, D. Lst, Non-supersymmetric gauge theories from D-branes in type 0 string
theory, Nucl. Phys. B 560 (1999) 66, hep-th/9906101.
I.R. Klebanov, N.A. Nekrasov, S.L. Shatashvili, An orbifold of type 0B strings and non-supersymmetric
gauge theories, Nucl. Phys. B 591 (2000) 26, hep-th/9909109.
R. Blumenhagen, A. Kumar, A note on orientifolds and dualities of type 0B string theory, Phys. Lett. B 464
(1999) 46, hep-th/9906234.
T.J. Hollowood, R.G. Myhill, The 112 breakings of E8 , Int. J. Mod. Phys. A 3 (1988) 899;
L.J. Dixon, Some world-sheet properties of superstring compactifications, on orbifolds and otherwise,
Lectures given at the 1987 ICTP Summer Workshop in High Energy Physics and Cosmology, Trieste, Italy,
June 29 August 7, 1987.
L.E. Ibez, J. Mas, H.P. Nilles, F. Quevedo, Heterotic strings in symmetric and asymmetric orbifold
backgrounds, Nucl. Phys. B 301 (1988) 157;
A. Font, L.E. Ibez, F. Quevedo, A. Sierra, The construction of realistic four-dimensional strings through
orbifolds, Nucl. Phys. B 331 (1990) 421.
J. Erler, Anomaly cancellation in six-dimensions, J. Math. Phys. 35 (1994) 1819, hep-th/9304104.
M.B. Green, J.H. Schwarz, P.C. West, Anomaly free chiral theories in six-dimensions, Nucl. Phys. B 254
(1985) 327.
V. Kaplunovsky, J. Sonnenschein, S. Theisen, S. Yankielowicz, On the duality between perturbative heterotic
orbifolds and M-theory on T 4 /ZN , Nucl. Phys. B 590 (2000) 123, hep-th/9912144;
E. Gorbatov, V.S. Kaplunovsky, J. Sonnenschein, S. Theisen, S. Yankielowicz, On heterotic orbifolds,
M-theory and type I brane engineering, hep-th/0108135.
A. Font, L.E. Ibez, F. Quevedo, ZN ZM orbifolds and discrete torsion, Phys. Lett. B 217 (1989) 272.
L.E. Ibez, H.P. Nilles, F. Quevedo, Orbifolds and Wilson lines, Phys. Lett. B 187 (1987) 25.
G. Aldazabal, A. Font, L.E. Ibez, A.M. Uranga, G. Violero, Non-perturbative heterotic D = 6, 4, N = 1
orbifold vacua, Nucl. Phys. B 519 (1998) 239, hep-th/9706158.
M. Faux, D. Lst, B.A. Ovrut, Intersecting orbifold planes and local anomaly cancellation in M-theory,
Nucl. Phys. B 554 (1999) 437, hep-th/9903028;
M. Faux, D. Lst, B.A. Ovrut, Local anomaly cancellation, M-theory orbifolds and phase-transitions, Nucl.
Phys. B 589 (2000) 269, hep-th/0005251.

Orbifold matrix model

Hajime Aoki a , Satoshi Iso b , Takao Suyama b
a Department of Physics, Saga University, Saga 840-8502, Japan
b High Energy Accelerator Research Organisation (KEK), Tsukuba, Ibaraki 305-0801, Japan
Received 9 April 2002; accepted 12 April 2002
Abstract
We study a matrix model describing type IIB superstring in orbifold backgrounds. We particularly
consider a C3 /Z3 orbifold model whose six-dimensional transverse space is orbifolded by Z3
discrete symmetry. This model is chiral and has d = 4 N = 1 supersymmetry of YangMills type as
well as an inhomogeneous supersymmetry specific to matrix models. We calculate one-loop effective
action around some backgrounds, and the result can be interpreted as interactions mediated by
massless particles in IIB supergravity in orbifold background, if the background is in the Higgs
branch. If the background is in the Coulomb branch, the dynamics is effectively governed by the
reduced model of d = 4 super YangMills theory, which might be interpreted as exchange of massless
particles in the twisted sector. But the perturbative calculation cannot reproduce the supergravity
result. We also show that this model with a large Higgs vacuum expectation value becomes IIB
(IKKT) matrix model. 2002 Published by Elsevier Science B.V.
1. Introduction
A large N reduced model has been proposed as a nonperturbative formulation of type
IIB superstring theory [1]. It is defined by the action:1

1

1
1
I [AI , ] .
S = 2 Tr [AI , AJ ] AI , AJ +
(1.1)
g
4
2
It is a large N reduced model [2] of ten-dimensional super YangMills theory. Here is a
ten-dimensional MajoranaWeyl spinor field, and AI and are N N hermitian matrices.
It is formulated in a manifestly covariant way which enables us to study the nonperturbative
E-mail addresses: haoki@cc.saga-u.ac.jp (H. Aoki), satoshi.iso@kek.jp (S. Iso), tsuyama@post.kek.jp
(T. Suyama).
1 In this paper, we take a metric convention
I J = diag(1, 1, . . . , 1).
0550-3213/02/$ see front matter 2002 Published by Elsevier Science B.V.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 9 5 - X
72
H. Aoki et al. / Nuclear Physics B 634 (2002) 7189
issues of superstring theory. In fact we can, in principle, predict the dimensionality of

spacetime, the gauge group and the matter contents by solving this model [3].
Among several issues on the matrix formulation of superstring, one is to answer which
model should be regarded as the most fundamental one. This question might be related to
the universality in the lattice gauge theory. In this case, the most important symmetry we
have to keep is the gauge symmetry and we know that various models with the gauge
invariance are in the same universality class. In the case of matrix models, maximal
supersymmetry may play the same role but it is not yet certain. Hence, it is meaningful
to consider various matrix models, investigate their dynamics and make connections
between them. If matrix models can describe the dynamics of spacetime, we should be
able to generate various spacetime backgrounds from the same model, and in this respect
universality in matrix models will be related to the background independence.
Another related issue is to construct a phenomenologically interesting model with
chiral fermions in d = 4. In string theory, there are several mechanisms to generate
four-dimensional chiral fermions including compactification with nonzero index of Dirac
operator, D-branes wrapped on intersecting cycles etc. [4]. In matrix models, we may
similarly be able to consider corresponding mechanisms but no investigations have been
performed yet.
In this paper we study a matrix model in C3 /Z3 orbifold background [5,6]. This model
can be obtained from IIB matrix model by imposing some constrains. In particular we
consider a model with four-dimensional N = 1 YangMills type supersymmetry. In matrix
models, there is also an inhomogeneous supersymmetry as well, and the model has totally
d = 4 N = 2 supersymmetries. This model is chiral since the chiral multiplets are in bifundamental representation of the gauge group. There are two types of classical solutions,
corresponding to the Higgs branch and the Coulomb branch, respectively. The Higgs
branch solution forms a triplet under Z3 transformation and there are three images in C3
plane whose positions in d = 4 directions are the same. On the other hand, the Coulomb
branch solution is restricted on the orbifold fixed point and cannot move into C3 direction.
The Coulomb branch solution is continuously connected to the Higgs branch solution at
the orbifold singularity. When the Higgs branch solution comes on the orbifold singularity,
the three images are liberated from each other and they can move independently into d = 4
directions. As BPS solutions, the former represents ordinary D-branes with three mirror
images by Z3 , and when they lie on the singularity they can move freely into the fourdimensional direction as fractional branes.
We calculate one-loop effective action of the orbifold matrix model to clarify its
correspondence to the interaction in type IIB supergravity in the orbifold background [7].
Firstly we show that we can reproduce the correct supergravity interaction for classical
solutions in the Higgs branch in orbifold background. Secondly the dynamics of fractional
branes in the Coulomb branch is shown to be governed by the reduced model of d = 4
super YangMills theory instead of d = 10 YangMills. In string theory, fractional branes
[8] have a charge of RR fields in the twisted sector and the interaction between them should
have four-dimensional behaviour in the long distance. In this respect, it is qualitatively
consistent with the string picture. But perturbatively the reduced model of d = 4 super
YangMills theory does not reproduce the correct interactions mediated by massless fields
in the twisted sector. (The induced interaction is inversely proportional to the fourth
73
power of the distance instead of the expected quadratic power.) This discrepancy will
be a limitation of perturbative calculations and it is not certain whether incorporation of
nonperturbative effects can help it.
This model has smaller supersymmetries compared with IIB matrix model and it is
not apparent if it is in the same universality class as IIB matrix model. But if it is, the
orbifold matrix model must be obtained from IIB matrix model through condensation of
some fields and so must be the reverse. These relations will also shed light on the issue how
to realize four-dimensional chiral fermions from IIB matrix model. In this paper, we show
that the orbifold matrix model in the Higgs phase with a large expectation value becomes
IIB matrix model.
The organisation of the paper is as follows. In Section 2, we give a brief review of
IIB matrix model. In Section 3, we construct orbifold matrix models, and particularly
consider a model with N = 1 supersymmetry of YangMills type in d = 4. In Section 4,
we analyse this orbifold matrix model. First we list several classical solutions with or
without noncommutativity of backgrounds. Then we calculate one-loop effective action
around these solutions and show that the result is partly consistent with the string theory. In
Section 5, we show that this model with a large Higgs vacuum expectation value becomes
IIB matrix model. Section 6 is devoted to conclusions and discussions.
2. Brief review of IIB matrix model

Action of the IIB matrix model is defined in (1.1). This model has N = 2 tendimensional supersymmetry if we regard the distribution of eigenvalues as spacetime.
Since this matrix model action can be obtained by reducing the original ten-dimensional
spacetime to a single point, the action has the remnant of the supersymmetry of super
YangMills type:
i
(1) AI = i I .
(1) = [AI , AJ ] I J ,
(2.1)
2
In addition to this homogeneous symmetry, there is an inhomogeneous fermionic symmetry
(2) = ,
(2)AI = 0.
If we take a linear combination of

(1) = (1) + (2) ,
(2.2)
(1)
(2)
and
as

(2) = i (1) (2) ,
we obtain the N = 2 supersymmetry algebra:

(i) (j )

(i) (j )

(j )
(j )
(i) = 0,
(i) AI = 2i I ij .
(2.3)
(2.4)
The N = 2 supersymmetry is a crucial element of superstring theory. It imposes strong

constraints on the spectra of particles. Furthermore it determines the structure of the
interactions uniquely in the light-cone string field theory [9]. The IIB matrix model is a
nonperturbative formulation which possesses such a symmetry. Therefore, it has a very
good chance to capture the universality class of IIB superstring theory. These symmetry
considerations force us to interpret the eigenvalues of AI as the spacetime coordinates.
74
The simplest classical solutions are those where all bosonic matrices are commutable
and simultaneously diagonalisable. Since the distributions of the eigenvalues determine the
extent and the dimensionality of spacetime, the structure of spacetime can be dynamically
determined by the theory. Indeed, spacetime exits as a single bunch and no single
eigenvalue can escape from the rest [10]. Numerical simulations and analysis with
various approximations have indicated that the Lorentz symmetry might be spontaneously
broken [10,11].
The next simplest solutions are those with nonzero but c-number commutator [AI , AJ ]
= iI J 1NN . These solutions correspond to BPS-saturated backgrounds. Indeed, the
solutions are invariant under transformations if we set equal to 12 I J I J in the N = 2
supersymmetry (2.1) and (2.2). By expanding the matrix around this classical solution, we
can obtain a supersymmetric noncommutative gauge theory [12].
Finally, in this section, we comment on the calculation of the one-loop effective action
in IIB matrix model and its interpretation in type IIB supergravity. In [1], it was shown
that when the bosonic matrices are expanded around backgrounds having a block-diagonal
form

d(1) + X(1)
A = X =
(2.5)
,
d(2) + X(2)
the one-loop effective action becomes
W (12) =
1
4(d (1) d (2) )8

(1) (1) (1) (1)

(1) (1) (1) (1)
4n2 Tr f
f f f 8n2 Tr f
f f f
(1) (1) (1) (1)
(1) (1) (1) (1)
+ 2n2 Tr f f f f + n2 Tr f f f f
(1) (2) (2) (2)

(1) (2) (2) (2)
Tr f f f + 12 Tr f
Tr f f f + (1 2)
48 Tr f
(1) (1) (2) (2)
(1) (1) (2) (2)
Tr f f
+ 6 Tr f
f
f Tr f f
48 Tr f
9

+ O 1/ d (1) d (2) .
(2.6)
Here we keep terms which vanish for finite N . (These terms play an important role in
calculating interactions between objects with charges of antisymmetric fields such as Dbranes.) Observing the tensor structures, we find the exchanges of massless particles in IIB
supergravity corresponding to graviton, scalars and antisymmetric fields. In this paper, we
perform similar calculations in the matrix model in orbifold backgrounds.
3. C3 /Z3 orbifold matrix model

3.1. Z3 orbifolding
We now construct matrix models in C3 /Z3 orbifold background [6]. The IIB matrix
model consists of ten hermitian bosonic matrices AI (I = 0, . . . , 9) and 16 components
75
d = 10 MajoranaWeyl fermionic matrices. We take complex coordinates for the sixdimensional transverse space and write them as Bi (i = 1, 2, 3),
B1 = A4 + iA5 ,
B2 = A6 + iA7 ,
B3 = A8 + iA9.
(3.1)
We characterise the action of Z3 group by three integers (a1 , a2 , a3 ). The Z3 transformation

is defined to act on these complex coordinates as Bi ai Bi . Different choices of these
integers lead to different models with a different number of unbroken supersymmetries in
d = 4.
From the D-instanton point of view, since there are three images of each D-instanton,
we have to extend the size of the matrices three times to represent these three mirror
images. The Z3 group rotates these three images, too. We write hermitian matrices as
tensor products of 3 3 matrices on which Z3 acts and the rest. The 3 3 matrices can be
expanded in terms of t Hooft matrices,
1
0 0 1
,
U =
(3.2)
V = 1 0 0,
0 1 0
where U V = V U . In a representation where the coordinates of each mirror image of Dinstantons is diagonalized, the Z3 transformation interchanges these diagonal blocks, and is
expressed as M V MV , where M is AI or . While this representation is geometrically
clear, for later convenience, we take another representation where the Z3 transformation is
given as M U MU .
We now impose Z3 invariance conditions on matrices. The Z3 condition for the gauge
field (A , = 0, . . . , 3) is
A = U A U .
(3.3)
The off-diagonal blocks are projected out and only the diagonal blocks A(i,i) , i = 1, 2, 3,
of 3 3 matrices survive after this Z3 projection. If the size of each block matrix is N N ,
these gauge fields are associated with U (N)3 symmetry. The Z3 invariance condition for
the complex fields Bi is given by
Bi = ai U Bi U
(3.4)
and is written as ai + i j = 0 ( mod 3) for (i, j ) blocks.

Fermionic matrices are similarly projected out by this Z3 transformation. A convenient
way to pick up Z3 invariant components out of the original d = 10 MajoranaWeyl
fermionic matrix is to write it in lower-dimensional Weyl-basis. Since d = 10 chirality
operator = 019 is written as a product of lower-dimensional chirality operators
as = (i 0123)(i 45 )(i 67 )(i 89 ), a positive
chirality fermion in d = 10 can be

decomposed as a sum of the fermions = 3i=0 (L,i + R,i ) in Table 1. Since the d = 10
charge conjugation interchanges left and right as (L,i )c = ( c )R,i , a d = 10 Majorana
Weyl fermion contains four independent complex degrees of freedom L,i , and the
right-handed fermions are related to the left as R,i = (L,i )c . In the following we write
L,i without index L, and R,i as its charge conjugation. Hence, the fermionic 3N 3N
76
Table 1
i 0123
i 45
i 67
i 89
L,0
(a1 + a2 + a3 )/2
+
+
+
+
L,1
L,2
1
2
(a1 a2 a3 )/2
(a1 + a2 a3 )/2
+
+
L,3
R,0
3
(0 )c
(a1 a2 + a3 )/2
(a1 a2 a3 )/2
R,1
(1 )c
(a1 + a2 + a3 )/2
+
+
+
+
R,2
R,3
(2 )c
(3 )c
(a1 a2 + a3 )/2
(a1 + a2 a3 )/2
Majorana
bi
matrix is decomposed as
=
3

A A c A
i + i
T ,
(3.5)
i=0
where T A s are generators of 3N 3N hermitian matrices. Since the Z3 transformation

acts on each complex plane C in C3 , Z3 invariant condition depends on how each fermion
transforms under rotations in each C plane. Hence, according to the eigenvalues of i 45 ,
i 67 and i 89 , each of the fermionic matrices is subject to the following Z3 invariance
condition:
i = bi U i U ,
(3.6)
where the charges bi s are calculated from ai s as in Table 1. Invariant blocks of 3 3

matrices satisfy bi + i j = 0 ( mod 3).
An action of the orbifold matrix model is given from the IIB matrix model action (1.1)
by restricting the matrices by the Z3 constraints (3.3), (3.4) and (3.6). First we rewrite the
IIB matrix model action (1.1) in terms of the d = 4 gauge field and the complex coordinate
fields. The bosonic part of IIB matrix model action becomes

3

1
2
Sb = 2 Tr [A , A ] + 2
[A , Bi ] A , Bi
4g
i=1

3

1

+
(3.7)
Bi , Bj Bi , Bj + Bi , Bj Bi , Bj
.
2
i,j =1
The Tr is a trace over 3N 3N matrices. The bosonic part of the orbifold matrix model
action is given by (3.7) where matrices are constrained by the conditions (3.3), (3.4). The
potential term for Bi field can be written as
2

1

2
Tr
Bi , Bi + 2
|[Bi , Bj ]|

2
i
i,j
(3.8)
77
and each term in this expression comes from the integration of D term and F term,
respectively, in the superspace formalism. Comparing to the N N IIB matrix model
(1.1), the 3N 3N orbifold matrix model has three times as many as matrices.
The fermionic part of the IIB matrix model action (1.1) is rewritten in terms of the
four-dimensional fermions in Table 1 as
3
3

1
i [A , i ] + 2
(i )c (i) Bi , 0
Sf = 2 Tr
2g
i=0
i=1

3

(j )
c
+
(3.9)
|ij k |(i ) [Bj , k ] + h.c. ,
i,j,k=1
where we have defined complex gamma matrices

1
1
(1) 4 i 5 ,
(3.10)
(1) 4 + i 5 ,
2
2
and similarly for (2) and (3) . The fermionic part of the orbifold matrix model is given
by (3.9) where fermionic matrices are restricted by the condition (3.6).
By Z3 orbifolding, the original symmetry of IIB matrix model with size 3N 3N
is generally reduced from SO(9, 1) U (3N) to SO(3, 1) U (N)3 . In some models
with special charge assignments for ai s, under the unbroken U (N)3 gauge symmetry,
some of the Z3 invariant fermion fields transform as bi-fundamental representations, and
the resulting gauge theory becomes a so-called quiver gauge theory. Furthermore, there
remains another symmetry which is a subgroup of the transverse SO(6) symmetry. A model
in the next subsection has unbroken U (3) symmetry, which rotate three chiral superfields.
As for supersymmetry, we decompose the supersymmetry parameter and of (2.1)
and (2.2) into lower-dimensional Weyl basis in Table 1. If some of these supersymmetry
parameters are invariant under the Z3 transformation, they transform Z3 invariant fields
among themselves. Therefore, the number of zero charges of bi (i = 0, . . . , 3) is the
number of unbroken supersymmetries. While in matrix models, we have inhomogeneous
supersymmetries (2.2) as well as the YangMills type (2.1), in this paper, we will name the
model by the number of YangMills type supersymmetry.
3.2. N = 1 model
We particularly consider a model with Z3 charge assignment ai = 2 for i = 1, 2, 3, and
accordingly b0 = 0, bi = 2 (i = 1, 2, 3). The Z3 invariance conditions (3.3), (3.4), (3.6)
can be satisfied if we restrict the matrices to the following form
(1)
0
0
Bi(2)
0
0
A
Bi = Bi(3)
A(2)
0 ,
0
0 ,
A = 0
Bi
= 0
(2)
Bi
0
(3)
Bi
0
0
A(3)
Bi(1) ,
0
(1)
Bi
78
(1)
0 = 0
0
(0 )c =
)c(1)
(0
0
0
0
(0 )c(2)
0
(i )c(3)
0
0
i = i(3)
0(3)
0
(i ) = 0
(i )c(2)
0 ,
(2)
0
(1)
i(2)
0 ,
0
0
0 ,
(0 )c(3)
0
(i )c(1) .
0
(3.11)
Inserting these expansions into (3.7) and (3.9), we can obtain the explicit form of the action
for the orbifold matrix model.
As we will see at the end of this subsection, this model has N = 1 YangMills type
supersymmetry. A and 0 fields form a vector multiplet while Bi and i fields make
three chiral multiplets. The left- and right-handed fermions in the chiral multiplets (i
and (i )c ) are in different bifundamental representations of U (N)3 and hence the model
is chiral in d = 4. Hence, this model is phenomenologically attractive, though it is also
interesting to consider other models with different charge assignments, symmetries, and
chirality.
We can also write the Z3 invariant fields as

A =
(3.12)
Aa U a ,
Bi =
Bia U a V ,
a=0,1,2
0 =
0a
a=0,1,2
U ,
a
i =
a=0,1,2

ia U a V ,
(3.13)
a=0,1,2
where U and V are the t Hooft matrices (3.2). Since A s are hermitian, (Aa ) = Aa
.
Charge conjugated and hermitian conjugated fields are expanded as
a

Bi =
Bi V U a ,
a=0,1,2

(0 ) =
c
a=0,1,2
i =
0a
c
U a ,

ia V U a ,
(i )c =
c

ia V U a ,
a=0,1,2
c

ia U a V .
(i )c =
(3.14)
a=0,1,2
These expansions are more useful than the block representation (3.11) when we investigate
this model in the next section.
Since only b0 vanishes, there is one d = 4 YangMills type supersymmetry in this
model. Inhomogeneous fermionic transformations (2.2) also make the orbifold matrix
model action invariant only if the supersymmetry parameter is the 0 type. The Yang
Mills type supersymmetry of this orbifold matrix model is given by
(1) 0 =

i
[A , A ] + Bi , Bi 0 ,
2
(3.15)

i
|ij k | Bj , Bk (j ) (k) 0 + 2[A , Bi ] (i) (0 )c ,
2
(1) A = i0 0 + i(0 )c (0 )c ,
(1) Bi = 2i(0 )c (i) i .
(1) i =
79
(3.16)
(3.17)
(3.18)
The other supersymmetry is

(2) 0 = 0 .
(3.19)
Here 0 and 0 are supersymmetry parameters restricted to the 0 type component in Table 1. Combining them similarly to (2.3), they become d = 4 N = 2 spacetime supersymmetry whose commutator gives translation of A ( = 0, . . . , 3). In matrix models, bosonic
fields are interpreted as spacetime coordinates and, hence, the above supersymmetry can
be identified with the N = 2 four-dimensional spacetime supersymmetry. In this paper,
however, we call this model N = 1 orbifold matrix model, counting only the number of
YangMills type supersymmetries.
4. Investigations of N = 1 orbifold matrix model
4.1. Classical solutions
We now investigate the orbifold matrix model, especially N = 1 type model.
Equations of motion can be obtained from the equations of motion of IIB matrix model
[AI , [AI , AJ ]] = 0 by restricting AI to the Z3 invariant parts. This is because the
projected-out fields by Z3 symmetry always enter in the action with another projected-out
field.
We first search for commutable solutions [AI , AJ ] = 0. This equation can be written in
terms of the Z3 invariant components in the block representation (3.11) as
(i) (i)
A , A = 0,
(4.1)
(2)
A(1)
Bi
(2)
(3)
(2)
Bi A(3)
= A Bi
(3)
(1)
(3)
Bi A(1)
= A Bi
(1)
Bi A(2)
= 0,
(1) (3)
(1) (3)
(2) (1)
(2) (1)
(3) (2)
(3) (2)
Bi Bj Bj Bi = Bi Bj Bj Bi = Bi Bj Bj Bi = 0,
Bi(1) Bj(1) Bj(2)Bi(2) = Bi(2) Bj(2) Bj(3) Bi(3) = Bi(3) Bj(3) Bj(1) Bi(1)
(4.2)
(4.3)
= 0. (4.4)
There are two types of solutions to these equations. The first type has generally nonvanishing vev of Bi and corresponds to the Higgs branch of super YangMills theory. In
()
this case, A must be independent of the block index () and has the form
,
A(cl)
(4.5)
X
=
X
where X are N N diagonal matrices. Eq. (4.2) can be satisfied if Bi() is a diagonal
matrix. To satisfy the last two Eqs. (4.3) and (4.4), we further assume that Bi() is
80
independent of the block index () and has the form
(2)(cl)
0
0
Bi
0
(3)(cl)

(cl)
Bi = Bi
0
0 = Zi
(1)(cl)
0
0
0
B
i
0
0
Zi
Zi
0 ,
0
(4.6)
()(cl)
can be put
where Zi is an N N diagonal matrix. More generally, phases of Bi
()(cl)
i
differently as Bi
= e Zi (where is an N N diagonal matrix). However, since
they transform under U (1)3N subgroup of the U (N)3 symmetry as
+ (+1 1 ),
(4.7)
3
their relative phases can be gauged away. Since the sum i=1 i is invariant under these
rotation, the common phase of Zi cannot be gauged away generally. We can only rotate the
common phase of Zi by 2/3, which corresponds to changing the basis of 3 3 matrices.
Writing the above solution (4.5) and (4.6) in terms of U, V expansion, we have
Bi(cl) = Zi V .
A(cl)
= X 133 ,
(4.8)
The second type of solution forms a singlet under Z3 rotation in C3 plane, and Bi has
(cl)
vanishing vev. In this case, A can be an arbitrary diagonal matrix
(1)
X
2
a
(2)
=
A(cl)
(4.9)
=
X
X U a .
X(3)
a=0
This type of solutions corresponds to the Coulomb branch in the super YangMills. These
solutions are restricted on the orbifold fixed point and cannot move into C3 direction. This
branch is connected with the Higgs branch solutions at Zi = 0 and X(1) = X(2) = X(3) .
When the Higgs branch solutions come to the orbifold fixed point, three images with the
same four-dimensional coordinates are liberated, and become possible to move into d = 4
direction independently from each other.
We then consider solutions with nonvanishing commutators [AI , AJ ] = 0. The first type
(4.8) with the following commutators satisfies the equations of motion [AI , [AI , AJ ]] = 0:
[X , X ] = iC 1NN ,
[Zi , Zj ] = Eij 1NN ,

Zi , Zj = Dij 1NN ,
[X , Zi ] = Fi 1NN ,
(4.10)
where C , Dij , Eij and Fi are c-number coefficients. If Eij = Fi = 0, this solution
becomes half BPS-saturated and preserves half of the supersymmetries. This solution
corresponds to an ordinary D-brane.
The second type (4.9) is a classical solution if the commutator is proportional to a unit
matrix
(i) (i)
(i)
X , X = iC
(4.11)
1NN .
81
(i)
This solution is also half BPS-saturated if the coefficient C is independent of the block
index (i). An example is
X0(i) = p,
X1(i) = q,
X2(i) = x2(i) 1NN ,
X3(i) = x3(i) 1NN ,
(4.12)
where [p,
q]
= i. It describes three fractional D-strings whose x2 and x3 coordinates can
be different. They are constrained in the zi = 0 plane.
4.2. Noncommutative orbifold field theory
Expanding our orbifold matrix model around the classical solutions (4.10), we can
obtain noncommutative field theory in the orbifold background. Here we set Zi = 0, for
simplicity. We write the classical solution X as x to emphasise the noncommutativity of
coordinates. Matrices are expanded around this noncommutative background as
x + A (1)
A =
(4.13)
x + A (2)
.
x + A (3)
All other matrices do not have classical background and are considered as fluctuating
fields. All the fluctuating fields are written in terms of noncommutative plane waves in
four-dimensions as

A a =
(4.14)
A a (k) exp(ik x ).
k
By inserting these expansions and adopting the same technique to derive noncommutative
field theory from IIB matrix model [12], the orbifold matrix model becomes noncommutative orbifold field theory with N = 1 supersymmetry in d = 4.2
If the background is replaced by a tensor product x 1nn , the resulting noncommutative field theory acquires U (n)3 gauge symmetry. This model contains d = 4 vector
multiplet (Aa and 0a ) in an adjoint representation of U (n)3 and three chiral multiplets
(Bia and ia ) in bi-fundamental representations. It is a chiral four-dimensional theory. We
may consider the number of chiral multiplets, three, as the number of generations (species)
[14]. The global symmetry U (3), which is a subgroup of SO(6) part in SO(9, 1), becomes
the horizontal symmetry among generations. Phenomenological analysis of this model will
be left for future work (see also [13]).
We then consider classical solutions (4.12). In this case, if the positions of three
fractional D-strings are largely separated, chiral multiplets Bi and i acquire large masses
2 In these cases, x always enters in the action in a form of commutators. But if the background is also
extended to the C3 direction and Bi is expanded around noncommutative background z i , the same technique
cannot be applied to obtain noncommutative field theory, because we are faced with a problem that the operator
z i enters into the action in a different form from commutators. Typically it is like (zi A a a A a z i ) where
the fluctuating fields are expanded in terms of the U, V expansion. Hence, if we expand the fluctuating fields
similarly, we cannot obtain an ordinary kinetic term. But by deforming the expansion basis of matrices from the
ordinary noncommutative plane waves, we may be able to obtain well behaved field theory. A simple example of
This satisfies aM qMa = wM where [a, a ] = 1.

deformed plane wave is M = exp(w a )q a a exp(w a).
82
proportional to the distances, and only the vector multiplet survives in low energy. The
low energy fields contain the same field contents as the d = 4 super YangMills gauge
theory. Expanding around the D-string solution (4.12), the effective theory becomes d = 2
noncommutative gauge theory obtained from the reduced matrix model of d = 4 SYM
theory. This is consistent at least qualitatively with the fact that fractional branes have
charges of RR fields in twisted sector, which propagate in d = 4 subspace of d = 10.
4.3. Effective action and interaction between diagonal blocks
We now calculate the one-loop effective action of orbifold matrix model between two
diagonal blocks by the background field method. To fix gauge invariance in perturbative
calculation, we add the gauge fixing term

1
1 cl I 2 cl I,cl
Sgf = 2 Tr
(4.15)
AI , A + AI , b A , c ,
2
g
where b and c are ghost fields and Acl
I includes background configurations of bosonic
matrices. When we expand
A = Acl
+ A ,
i ,
Bi = Bicl + B
(4.16)
the quadratic part of the quantum fluctuation in the bosonic action becomes

I,cl J
cl cl I J
1

A
+
2
AI , AJ A , A
,
A
,
A
Sb,2 = 2 Tr Acl
J
I
2g

,cl cl

cl
1
2

+ Bi , A B cl , A
= 2 Tr Acl
,B
, A + A , Bi A
i
i
2g
1 cl cl cl cl
Bi , Bj Bi , Bj + Bi , Bj Bi , Bj
+
2
cl cl cl cl cl cl

+ A , B
i
+ 2 A , A A , A + A , Bi A , B
A , B
i
i
1 cl cl cl cl cl cl
+
Bi , Bj + Bi , Bj
Bi , Bj .
Bi , Bj Bi , Bj + Bi , Zj
2
(4.17)
We first consider the background of the first-type classical solution (4.8) with arbitrary
X and Zi . Bosonic matrices are expanded as
A = X 133 + A a U a ,
a U a V .
Bi = Zi V + B
i
(4.18)
Inserting these expansions into the action (4.17) and taking the trace over 3 3 matrices
constructed by U and V , we have

3
ia X , B
a

Sb,2 = 2 Tr X , A a X , A ,a + X , B
i
2g

a
+ Zi A a a A a Zi a Z A a
A Z
i

1 a
ja Zi a Z B
a a
Zi Bj a B
+
i j Bj Zi
2

a a B
a Zi a Z B
a a
+ Zi B
j
j
i j Bj Zi

a A ,a
a B
+ 2 [X , X ] A ,a , A ,a + [X , Zi ] a A ,a B
i
i

a a B
a A ,a
+ X , Zi A ,a B
i
i
a, a,

1
B
a, B

a,
+ [Zi , Zj ]a B
B
i
j
j
i
2
a a

B

a a
+ Zi , Zj a B
i j Bj Bi
a, a

B
ja B
j B
a, .
+ Zi , Zj B
i
i
83
(4.19)
Here Tr is a trace over N N after taking the 3 3 part. Similarly the quadratic part in
quantum fluctuation of the fermionic matrices becomes
3
3

a c (i) a

3

Sf,2 = 2 Tr
ia X , ia + 2
i Zi 0 a 0a Zi
2g
i=0
i=1

3

a c (j ) a

Zj ka a ka Zj + h.c.
|ij k | i
+
(4.20)
i,j,k=1
The last term survives due to the special property for the Z3 orbifold matrix model,
V 3 = 133 .
We are interested in the induced interactions between backgrounds that are separated
distantly and hence assume that X and Zi are of 2 2 block form

X(1)
Zi(1)
,
Zi =
.
X =
(4.21)
X(2)
Zi(2)

The difference between the c-number parts of two blocks, ( (x(1) x(2) )2 +

2 1/2 is assumed to be much larger than the fundamental scale g 1/2 of
i |zi(1) zi(2) | )
the matrix model. In order to calculate interactions between these two blocks, we need to
a in the off-diagonal blocks;
integrate over the fluctuations A a , B
i
A a =

A a(12)
A a(21)
ia =
B

a
B
i(21)
a
B
i(12)
(4.22)
In the quadratic actions (4.19), (4.20), a = 0 sector is always decoupled from a = 1, 2

sector. This a = 0 part is the same as the quadratic part of the IIB matrix model action
with a gauge fixing term around the background A = X , Bi = Zi . Hence, integration
over a = 0 sector gives the same contribution as the blockblock interactions in IIB matrix
model (2.6), which can be interpreted as exchanges of massless particles in d = 10 type
IIB supergravity. The typical form is
((x(1) x(2)
)2

1
tr(f(1) f(1) ) tr(f (2) f (2)) + .
2
4
+ |zi(1) zi(2) | )
(4.23)
The contributions from the other sectors a = 1, 2 can be evaluated similarly. First expand
further the actions (4.19), (4.20) in terms of block fields (4.21), (4.22). The bosonic part
84
becomes (we drop the coefficient 3/2g 2 )

,a
Tr X(1) A a(12) A a(12)X(2) X(2) A ,a
(21) A(21) X(1)

a a

a X(2) X B

a B
+ X(1) B
i(12)
i(12)
(2) i(12) Bi(12) X(1)

a
+ Zi(1)A a(12) a A a(12)Zi(2) a Zi(2)
A a
(21) A(21) Zi(1)
a a a
1
a a

a

Zi(1) B
+
j (12) Bj (12) Zi(2) Zi(2) Bj (12) Bj (12) Zi(1)
2
a

a

a a

a
a Z
B
Z
Z
B
+ Zi(1) B
i(2)
j (21)
j (21)
j (21) i(1)
i(2) j (21)

,a ,a
,a
,a
+ 2 [X(1) , X(1) ] A (12)A (21) A (12) A (21)
a ,a

,a a

+ [X(1) , Zi(1) ] a A (12) B
i(12) Bi(21) A(21)

,a a

a A ,a
A B
a B
+ X(1) , Z
i(1)
(12) i(21)
i(12)
(21)
a a a a
1

B
B
+ [Zi(1), Zj (1)]a B
i(21)
j (12) Bj (21)
i(12)
2

a B
a a
a
+ Zi(1) , Zj(1) a B
i(21) j (12) Bj (21) Bi(12)

a a
a
a

+ Zi(1) , Zj(1) B
i(21) Bj (21) Bj (12) Bi(12)

+ (1) (2), (12) (21) .
(4.24)
The fermionic part becomes

3

a
a
a
Tr
i(21)
X(2) i(21)
i(21)
X(1)
i=0
+2
3

a c (i)

a
a
i(12) Zi(2) 0(21)
a 0(21)
Zi(1)
i=1
3

a c (j ) a

a
a
|ij k | i(12)
a k(21)
Zj (1)
Zj (2) k(21)
i,j,k=1

+ (1) (2), (12) (21) + h.c.
(4.25)
Then we can classify the fluctuating fields into three classes:

1. Fields with a = 0,
1 , B
2 , 1 , 2 ,
2. A 1(12) (= (A 2(21)) ), B
i(12)
i(21)
i(12)
i(21)
2 , B
1 , 2 , 1 .
(= (A 1
) ), B
3. A 2
(12)
(21)
i(12)
i(21)
i(12)
i(21)
The quadratic actions are decoupled into three parts, each of which contains only the fields
in each class. Furthermore the quadratic action for each class becomes the same as the
85
a = 0 case (class 1) when we absorb the Z3 phase a into Zi(2) and redefine fields as
1 B
1 ,
B
i(12)
i(12)
Zi(2) Zi(2),
1
1
i(12)
i(12)
(i = 1, 2, 3),
(4.26)
for class 2 fields, and

2
2
i(12)
i(12)
2 B
B
,
2 Zi(2) Zi(2) ,
2
2
2 i(12)
i(12)
(i = 1, 2, 3).
(4.27)
for class 3 fields. Therefore, if the background is in the Higgs branch (4.18), we obtain the
same blockblock interactions as in IIB matrix model, except for the replacement of the
typical interaction form (4.23) by
2

1
+ |zi(1) a zi(2) |2 )4
((x(1) x(2)

tr(f(1) f(1)) tr(f (2) f (2) ) + .
a=0
)2
(4.28)
The a = 0 terms can be interpreted as interactions between the first block and the Z3
transformed mirror images of the second block. The tensor structure is the same as the
IIB result (2.6) and each term can be interpreted as an interaction mediated by massless
particles of type IIB supergravity in the untwisted sector in the orbifold background.
In the remainder of this subsection, we consider an effective action between Coulomb
branch solutions or fractional branes in the same twisted sector. In matrix models, each
twisted sector corresponds to each diagonal block matrices in (4.9). In string theory,
fractional branes have charges of massless fields in the twisted sector, and the induced
interaction is expected to have d = 4 behaviour instead of d = 10 behaviour. In the
Coulomb branch, since the three fractional branes (4.12) can move freely into fourdimensional spacetime, we can put them distantly from each other. Then the chiral
multiplets including Bi and i become massive (mass is proportional to the distances)
and are decoupled in the low energy. Only the vector multiplet of A and 0 survives. The
low energy theory now becomes three decoupled reduced models of d = 4 super Yang
Mills gauge theory. The effective interaction between fractional branes in the same twisted
sector is therefore governed by this d = 4 reduced model. This behaviour is qualitatively
consistent with the supergravity point of view. But in the perturbative calculation the
effective interaction is inversely proportional to fourth power of the distances (instead of
quadratic) and inconsistent with the expected power of the supergravity calculation. It is
not certain if we can overcome this difficulty nonperturbatively.
5. Recovery of IIB matrix model

As we mentioned in the introduction, it is meaningful to make connection between
IIB matrix model and the orbifold matrix model. If we can dynamically generate the
orbifold matrix model from IIB matrix model, we could realize four-dimensional chiral
fermion in IIB matrix model and we might also obtain some notion of universality class
and background independence in matrix models. In this section we report their connection
86
in the opposite direction that IIB matrix model can be obtained from the orbifold matrix
model.
In the Higgs phase (4.8) with
A(cl)
= X 133 = 0,
(cl)
Bi
= Zi V = zi 1NN V ,
(5.1)
where zi are complex numbers and |zi |2 g, a = 0 sector acquires a large mass and
only a = 0 sector survives. The resulting model becomes the IIB matrix model, and
ten-dimensional Poincare invariance and N = 2 supersymmetry of IIB matrix model are
recovered. This result is quite natural because in the large zi limit, the orbifold fixed point
will become invisible, and fields connecting different mirror sectors will have large mass
proportional to the vev |zi |.
We expand the matrices around the classical solution (5.1) as
A = A a U a ,
a U a V ,
Bi = zi 1NN V + B
i
0 = 0a U a ,
i = ia U a V ,
(5.2)
and insert them into the orbifold matrix model action (3.7). We then obtain the quadratic
term,

2
3 a

1 Tr 2|zi |2 A a A a
Sb,2 = 2
4g
a=0,1

1
a B
ja + 2zi zj B
a B
ja zi zj B
a B
a zi zj B
ia B
a
+ 4|zi |2 B
j
j
i
i
j
2

3

2
3 a
a B
ja
1 z12 Tr A a A a
+
= 2
B
j
2g
a=0,1
j =2

1 a a a a
B1 + Bj
.
+ B
+ B1
(5.3)
4 1
In the second line, we take z1 real and z2 = z3 = 0 by using the U (3) SO(6) invariance
of the orbifold matrix model. Similarly, (3.9) gives the quadratic fermionic term

c

c

3 a

1 z1 1a (1) 0a a 3a (1) 2a + h.c. (5.4)
Sf,2 = 2
g
a=0,1
All of the a = 0 modes are massless, while a = 0 modes acquire large masses of the order
a=1 (B
a=1 ) . This is a NambuGoldstone mode associated
of z1 except one mode, B
1
1
with the symmetry breaking U (N)3 U (N) induced by the vev (5.1). Indeed, the vacuum
transforms under U (N)3 as

z1 1NN V , i a U a = iz1 a 1 a U a V ,
(5.5)
and B1a = iz1 (a 1) a satisfies the relation (B1a ) = B1a . As we will show in the
remainder of this subsection, this NambuGoldstone mode disappears from the spectrum
by the Higgs mechanism.
87
To discuss general structure of the action, we call the classical solution Z, massless
modes in a = 0 sector L, the above NambuGoldstone mode in a = 0 sector G, and
massive modes in a = 0 sector M, in general. Then the triple and the quartic parts of
the quantum fluctuation in the action have the form, in general:

Sb,3 = ZLLL, ZLGG, ZGGG, ZLLM, ZLGM, ZGGM, O M 2 ,
(5.6)

Sb,4 = {LLLL, LLGG, LGGG, GGGG, O(M)},
(5.7)

Sf,3 = {LLL, O(M)}.
(5.8)
The LLLL terms in (5.7) and LLL terms in (5.8) are exactly same as the IIB matrix model
action. The ZLLL terms in (5.6) vanish because Z and L commute. The ZLLM terms in
(5.6) vanish after taking a trace of 3 3 blocks because only M is in a = 0 sector while Z
and L are in a = 0 sector.
Therefore, if we take unitary gauge, setting G = 0, and integrate out the massive
mode M, then we obtain the IIB matrix model action. If the ZLLM terms were
nonvanishing, these terms would induce new LLLL terms, apart form the IIB matrix model
action.
We can also see this Higgs mechanism in other gauges than the unitary gauge. In the
large z limit, the broken symmetry U (N)3 /U (N) becomes the translational symmetry of
the NambuGoldstone mode G. Therefore, the action, (5.6) and (5.7), should not depend
on G, after we integrate out M. Indeed, ZLGG, ZGGG terms in (5.6) vanish because

2
a

+B
a B1 , B a ,
1 a Tr B
Tr B1 , B1 6z1
(5.9)
1
1
1
a=0,1
and this kind of terms must include M. Furthermore, after M is integrated out, the
ZLGM, ZGGM terms in (5.6) induce LLGG, LGGG, GGGG terms, which cancel
those terms in (5.7). This cancellation is easy to see in the ZZMM, ZGGM, GGGG
terms:

1
3
1
Sb 2 |1 |2 Tr 2z12 M1 M1 z1 M1 G2 + M1 G2 + G2 G2
4g
4
2

3
1
1
=
(5.10)
|1 |2 2 Tr z1 M1 G2 z1 M1 G2 ,
4g 2
2 2
2 2
where
1 B
1 + B
1
1
B
B
1
(5.11)
M1 = 1 1 .
1 ,
2
2
In this way, in the large zi limit, the NambuGoldstone mode disappears from the
spectrum, only a = 0 sector survive, and the IIB matrix model is recovered from the
orbifold matrix model .
Finally, in this section, we will shortly discuss the vacuum structure of the orbifold
matrix model. We have shown that the orbifold matrix model is reduced to IIB matrix
model in the Higgs branch with a large scalar vev. Is this vacuum dynamically stable? In
field theory, this vacuum preserves half of supersymmetry and it is perturbatively stable.
In the case of matrix model, we can discuss more about the stability of the vacuum. As
G=
88
studied in [10], eigenvalues of bosonic matrices in IIB matrix model tend to gather and
no single eigenvalue can escape from the others. The dynamics of the eigenvalues in
orbifold matrix model will also share the same property and then three mirror images
of each eigenvalue will be bounded around the fixed point. If this is true, the vacuum
with large vev in the Higgs branch is not realized dynamically and IIB matrix model is
not recovered dynamically from the orbifold matrix model. Instead the nonperturbative
vacuum of the orbifold matrix model will be such that the orbifold fixed point cannot be
neglected and have phenomenologically more interesting properties. On the other hand,
it is more interesting if we can show that the orbifold matrix model can be dynamically
generated from IIB matrix model. We want to come back to this problem.
6. Conclusions and discussions

In this paper, we have studied matrix models in orbifold background, in particular,
C3 /Z3 orbifold with N = 1 four-dimensional supersymmetry. We find two types of
classical solutions, the Higgs branch solution and the Coulomb branch solution. When
the solutions have noncommutativity, they correspond to ordinary D-branes and fractional
branes, respectively. We then calculate the one-loop effective action around these solutions
and show that in the former case we can correctly reproduce the supergravity result in
the orbifold background. In the latter case, although we can obtain qualitatively consistent
picture with the string theory, perturbative calculation does not reproduce the supergravity
result. Finally, we show that this model includes IIB matrix model in a special point of
the perturbative moduli in a sense that the orbifold matrix model is reduced to IIB matrix
model when the scalar fields Bi have large expectation values in the Higgs branch.
There are several issues that are not studied in this paper. One of them is whether
the orbifold matrix model is equivalent to the IIB matrix model. As we discussed in the
introduction, various matrix models might be related to each other. Since IIB matrix model
is obtained from the orbifold matrix model, we now want to show the reverse. That is to
say, it is desirable if we can obtain the orbifold matrix model from IIB matrix model by
some mechanism such as condensations of the fields. When we simply consider vacuum
expectation values of scalar fields Bi in IIB matrix model of size 3N 3N , we cannot
obtain four-dimensional chiral theory. We have to violate parity and it is not yet known how
we cab twist the background to make the lower-dimensional theory chiral. Perhaps we have
to extend the size of IIB matrix model and consider parity violating condensation outside
of the 3N 3N matrix. These are under investigations together with other mechanisms to
generate four-dimensional chiral fermions from IIB matrix model dynamically.
Acknowledgements
The work (H. Aoki and S. Iso) is supported in part by the Grant-in-Aid for Scientific
Research from the Ministry of Education, Science and Culture of Japan. We thank N. Ishibashi for informing us of his thesis and J. Nishimura for discussions on dynamics of
orbifold matrix models.
89
References
[1] N. Ishibashi, H. Kawai, Y. Kitazawa, A. Tsuchiya, A large-N reduced model as superstring, Nucl. Phys.
B 498 (1997) 467, hep-th/9612115.
[2] T. Eguchi, H. Kawai, Phys. Rev. Lett. 48 (1982) 1063.
[3] For review, see H. Aoki, S. Iso, H. Kawai, Y. Kitazawa, A. Tsuchiya, T. Tada, Prog. Theor. Phys. Suppl. 134
(1999) 47, hep-th/9908038.
[4] Several mechanisms to generate chiral fermions are listed in the paper, A.M. Uranga, hep-th/0201221.
[5] Field theories in this orbifold background was studied in the paper, S. Kachru, E. Silverstein, Phys. Rev.
Lett. 80 (1998) 4855, hep-th/9802183.
[6] There are several papers which studied on orbifolding matrix models. Itoyama and Tokura constructed USp
matrix model by orbifolding IIB matrix model, H. Itoyama, A. Tokura, Prog. Theor. Phys. 99 (1998) 129,
hep-th/9708123;
H. Itoyama, A. Tokura, Phys. Rev. D 58 (1998) 026002, hep-th/9801084;
M(atrix) theory on S1 /Z2 was studied by N. Kim, S.J. Rey, Nucl. Phys. B 504 (1997) 189, hep-th/9701139;
T. Banks, L. Motl, JHEP 9712 (1997) 003, hep-th/9703218;
M(atrix) theory on C3 /Z3 was studied by A. Miyake, A. Sugamoto, hep-th/0203223.
[7] Discrepancy in higher order terms between matrix theory calculation and supergravity result was reported
in the paper, M.R. Douglas, H. Ooguri, Phys. Lett. B 425 (1998) 71, hep-th/9710178.
[8] D.E. Diaconescu, M.R. Douglas, J. Gomis, JHEP 9802 (1998) 013, hep-th/9712230;
M.R. Douglas, B.R. Greene, D.R. Morrison, Nucl. Phys. B 506 (1997) 84, hep-th/9704151;
M.R. Douglas, JHEP 9707 (1997) 004, hep-th/9612126.
[9] M. Fukuma, H. Kawai, Y. Kitazawa, A. Tsuchiya, String field theory from IIB matrix model, Nucl. Phys.
B 510 (1998) 158, hep-th/9705128.
[10] H. Aoki, S. Iso, H. Kawai, Y. Kitazawa, T. Tada, Prog. Theor. Phys. 99 (1998) 713, hep-th/9802085.
[11] J. Nishimura, F. Sugino, hep-th/0111102;
J. Ambjorn, K.N. Anagnostopoulos, W. Bietenholz, F. Hofheinz, J. Nishimura, hep-th/0104260;
J. Nishimura, G. Vernizzi, JHEP 0004 (2000) 015, hep-th/0003223.
[12] H. Aoki, N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, T. Tada, Nucl. Phys. B 565 (2000) 176, hep-th/9908141;
N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, Nucl. Phys. B 573 (2000) 573, hep-th/9910004;
S. Iso, H. Kawai, Y. Kitazawa, Nucl. Phys. B 576 (2000) 375, hep-th/0001027;
M. Li, Strings from IIB matrices, hep-th/961222.
[13] P.H. Frampton, hep-th/0108011;
G. Aldazabal, L.E. Ibanez, F. Quevedo, A.M. Uranga, JHEP 0008 (2000) 002, hep-th/0005067.
[14] A realistic model with three generations from type I open string in C3 /Z3 orbifold background was
constructed by Ishibashi in his thesis, 1988, unpublished.

Radiative ScherkSchwarz supersymmetry

breaking
G.V. Gersdorff a , M. Quirs a , A. Riotto b
a Instituto de Estructura de la Materia (CSIC), Serrano 123, E-28006 Madrid, Spain
b INFN, sezione di Padova, Via Marzolo 8, I-35131 Padua, Italy
Received 9 April 2002; accepted 18 April 2002
Abstract
We analyze the ScherkSchwarz (SS) supersymmetry breaking in brane-world five-dimensional
theories compactified on the orbifold S 1 /Z2 . The SS breaking parameter is undetermined at the
tree-level (no-scale supergravity) and can be interpreted as the Hosotani vacuum expectation value
corresponding to the U (1)R group in five-dimensional N = 2 (ungauged) supergravity. We show that
the SS breaking parameter is fixed at the loop level to either 0 or 1/2 depending on the matter content
propagating in the bulk but in a rather model-independent way. Supersymmetry breaking is therefore
fixed through a radiative ScherkSchwarz mechanism. We also show that the two discrete values of
the SS parameter, as well as the supersymmetry breaking shift in the spectrum of the bulk fields, are
altered in the presence of a brane-localized supersymmetry breaking arising from some hidden sector
dynamics. The interplay between the SS and the brane localized breaking is studied in detail. 2002
Elsevier Science B.V. All rights reserved.
PACS: 11.30.Pb; 11.30.Qc; 04.50.+h
1. Introduction
If supersymmetry plays an important role in constructing consistent high-energy
physics theories, a relevant issue is to explain how supersymmetry is spontaneously
broken in the low-energy world. Recent ideas on extra dimensions and the speculation
that our visible universe coincides with a four-dimensional brane living in the bulk
of the extra dimensionsthe so-called brane-world scenarioshave given rise to new
Work supported in part by CICYT, Spain, under contract FPA2001-1806, and by EU under contracts HPRNCT-2000-00152, HPRN-CT-2000-00148 and HPRN-CT-2000-00149.
E-mail address: mariano@makoki.iem.csic.es (M. Quirs).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 1 6 - 4
G.V. Gersdorff et al. / Nuclear Physics B 634 (2002) 90104
91
appealing possibilities regarding how to realize supersymmetry breaking [19]. The extradimensional framework is particularly interesting because it provides a new geometrical
perspective in understanding some of the problems of conventional four-dimensional
theories.
In a recent paper [10] five-dimensional (5D) supersymmetric theories compactified
on the orbifold S 1 /Z2 were considered and the ScherkSchwarz (SS) supersymmetry
breaking [11] was interpreted as the Hosotani breaking [12] of the local SU(2)R symmetry
present in off-shell N = 2 supergravity. In particular, it was shown that SU(2)R is gauged
by auxiliary fields A M , M = , 5 whose extra-dimensional components background
V51 + iV52 that make part of the F -component of the radion multipletprovide different
bulk masses to fermions and bosons in N = 1 supermultiplets and then trigger Scherk
Schwarz supersymmetry breaking [10,13]. The vacuum expectation value (VEV) of the
field V51 + iV52 is, however, not fixed at the tree level, a reminiscent situation of no-scale
supergravity in four dimensions [14], which means that the scale at which supersymmetry
is broken is classically undetermined.
In this paper we want to go one step further and introduce one-loop radiative corrections
to dynamically determine the ScherkSchwarz supersymmetry breaking parameter. Our
goal is to show that the ScherkSchwarz supersymmetry breaking parameter can be led
by radiative corrections to two discrete values, either 0 or 1/2, corresponding to unbroken
or broken supersymmetry. The actual value of depends upon the matter content of the
theory in the bulk. In order to do the analysis it is convenient to deal with only physical
degrees of freedom, integrate out the auxiliary fields and work with on-shell N = 2
supergravity. This will be done in Section 2 where the on-shell version of the mechanism
in Ref. [10] will be presented. The one-loop effective potential will be computed in
Section 3 where the ScherkSchwarz breaking will be dynamically determined. Nonlocal and spontaneous supersymmetry breaking can also be induced by brane-localized
dynamics giving rise to effective supersymmetry breaking superpotentials [15,16]. The
interplay between this mechanism and that induced by the V51 + iV52 VEV is studied in
Section 4 that gives rise to a variety of possibilities for the supersymmetry breaking shift
in the spectrum of the bulk fields depending on the actual value of the supersymmetry
breaking superpotential. Finally we draw our conclusions in Section 5.
2. Bulk supersymmetry breaking

In this section we want to introduce the ScherkSchwarz mechanism for supersymmetry
breaking in the context of on-shell D = 5 N = 2 supergravity, the on-shell version of the
mechanism presented in Ref. [10]. D = 5 N = 2 supergravity compactified on the Z2
orbifold has been recently analyzed in Refs. [17,18] and given an off-shell formulation in
Refs. [19,20]. The starting point is the minimal supergravity multiplet in five dimensions
a and the graviphoton B as
containing the graviton gMN , the SU(2)R doublet gravitino M
M
propagating fields and, as auxiliary fields, the SU(2)R gauge fields VM , an antisymmetric
tensor v AB , an SU(2)R triplet t, a real scalar C and an SU(2)R doublet spinor a . The
orbifold parity assignments are displayed in Table 1.
92
Table 1
Parity assignment of the minimal supergravity multiplet in five dimensions
Field
Even
Odd
gMN
M
BM
VM
g , g55
1 , 2
L
5L
B5
V3 , V51,2
v 5
t 1,2
C
L1
g5
2 , 1
L
5L
B
V1,2 , V53
v
t3
v AB
t
C
L2
Table 2
Parity assignment of the tensor multiplet
Field
Even
Odd
Y
BMNP
N
Y 1, Y 2
Y3
B5
B
N
1
L
2
L
However, the action based on the minimal multiplet is not physical, as was observed
in [20]. A simple way of realizing this fact is that it contains 40B + 40F degrees of
freedom while the minimal D = 5 off-shell supergravity must contain at least 48B + 48F
degrees of freedom [21]. Another way of realizing that the minimal multiplet by itself does
not describe the physical on-shell action is the variation of the action with respect to C
which leads to the unphysical equation of motion det gMN = 0. Therefore, introduction of
an additional (compensator) 8B + 8F multiplet is required. This compensator will serve
to (partially) fix the SU(2)R local symmetry in the on-shell theory. Depending on the
choice of the additional supermultiplet off-shell supergravity looks different although the
physical on-shell versions are the same for all cases. Different choices have been done
in Ref. [20]: a nonlinear multiplet (version I), a hypermultiplet (version II) and a tensor
multiplet (version III). Here we will adopt the latter formulation of off-shell supergravity
and introduce an additional tensor supermultiplet. In this case it is not possible to fix the
whole SU(2)R symmetry by means of the compensator field Y but a residual U (1)R gauged
by an auxiliary field remains. However, this residual degree of freedom can be fixed by
means of the equations of motion of the compensating multiplet.
The D = 5 tensor multiplet contains an SU(2)R bosonic triplet Y , an antisymmetric
three-form potential BMNP , a real scalar N and an SU(2)R doublet spinor a . The orbifold
parity assignments of the tensor multiplet are displayed in Table 2. The SU(2)R gauge
fixing is done by the compensator field Y as

0
Y = e 1 ,
0
u
(2.1)
93
where the scalar field u has been introduced. This gauge leaves the U (1)R subgroup
generated by 2 unbroken. Now, if we take as the Lagrangian
Lgrav = Lminimal + Ltensor,
(2.2)
the terms proportional to C in Lgrav are (1

Variation with respect to C now
yields the equation of motion u = 0, which shows that the previously mentioned problem
of Lminimal is solved. The VEV u = 0 will be fixed hereafter.
The auxiliary fields that are relevant for supersymmetry breaking are V51 , V52 , t 1 and t 2 ,
the components of the F -term of the radion multiplet

2
h55 + iB5 , 5L
(2.3)
, V51 + iV52 + 4i t 1 + it 2 .
eu )C.
The relevant terms in Lgrav containing V51 , V52 , t 1 and t 2 are

2
2
i
1
2
N BP QR + V51 12 t 1
Lgrav = P P MN DM N MNP QR VM
2
12
2
48 t 2 12Nt 2 N 2 ,
(2.4)
where DM is the covariant derivative with respect to local Lorentz and local U (1)R
transformations
2
DM = DM + i 2 VM
(2.5)
and DM the covariant derivative with respect to local Lorentz transformations.

The variation of (2.4) with respect to auxiliary fields provides the on-shell supergravity
Lagrangian. In general, auxiliary fields X have couplings in Laux , whose variation provides
the vacuum expectation value (VEV) X0 , and couplings to matter (propagating) spinor
+ aB X2 |A|2 . If we are only interested in mass terms
and scalar A fields as Lint = aF X
we can replace X X0 into Lint since the difference when solving the field equations from
Laux and Laux + Lint are quartic terms in matter fields. In particular, the field equations for
the auxiliary fields V51 , N , t 1 and t 2 yield
1

V5 = N
= t 1 = t 2 = 0,
(2.6)
while the field equation for the three-form tensor field BMNP gives
2
[M VN]
= 0.
(2.7)
The obvious solution to the latter equation is

2
VM
= M C,
(2.8)
i.e., a pure gauge that is non-trivial in a compact space. Since C is odd, the simplest choice
x5
R
leads to the background
C =
(2.9)
V2 = 0
(2.10)
94
and
(2.11)
R
that breaks supersymmetry through the coupling (2.5) in (2.4) as in Ref. [10].
Using the coupling of V52 to the gravitino field in (2.4) one obtains the gravitino mass
eigenvalues for the KaluzaKlein modes in the orbifold compactification S 1 /Z2
V52 =
n+
,
(2.12)
R
where n is an integer number. Supersymmetry breaking is also manifest in the spectrum
of bulk vector multiplets and scalar hypermultiplets. The action for the super Yang
Mills fields and hypermultiplets have been computed in Ref. [20] by embedding the
corresponding D = 5 N = 2 supermultiplets into linear multiplets. The relevant couplings
of the SU(2)R doublets, gauginos and hyperscalars A, with the SU(2)R gauge bosons
VM is provided by the covariant derivative in the interaction Lagrangian,
(n)
m3/2 =
i M
M A
2 ,
Lmatter =
DM +
D
2
where
M = DM + i VM
D
(2.13)
(2.14)
is the original covariant derivative with respect to the whole SU(2)R gauge group. After
M DM
using the field equations for the auxiliary fields V51 = V53 = 0 we obtain that D
and the mass eigenvalues for gauginos and hyperscalars can be computed as in Ref. [10]
and yield
(n)
(n)
m1/2 = m0 =
n+
.
R
(2.15)
All the mass eigenvalues depend on the field V52 in the same way. Therefore, the mass
of zero modes for gauginos, gravitinos and hyperscalars is m(0) = V52 which is then
the scale of supersymmetry breaking. This scale remains undetermined at the tree level,
a behaviour typical of ScherkSchwarz breaking and reminiscent of no-scale supergravity
models [14]. This flatness will be lifted when loop corrections will be accounted, thus
leading to the possibility of breaking radiatively supersymmetry through the Scherk
Schwarz mechanism. This is the subject of the next section.
3. Radiative determination of supersymmetry breaking

The one-loop effective potential can be most easily computed using 5D functional
techniques. In particular, for hyperscalars it is given by

2NH
d 5p 2
ln p + p52 ,
V0 =
(3.1)
5
2
(2)
95
where NH is the number of hypermultiplets, the factor of 2 comes from the degrees of
freedom of a complex scalar and

dp5 /2
p5
is understood with p5 = m(n)

0 . For gauginos it is

2NV
d 5p 2
V1/2 =
ln p + p52 ,
2
(2)5
(3.2)
where NV is the number of vector multiplets, the factor of 2 comes from the degrees of
freedom of a Majorana gaugino and p5 = m(n)
1/2 . Both in (3.1) and (3.2) we have already
taken into account a global factor of 1/2 for the orbifold S 1 /Z2 compactification that
removes half of the degrees of freedom corresponding to the circle.
The contribution of the gravitino is easily obtained using the functional techniques of
Ref. [22]. In particular, one can see that in the 5D harmonic gauge N Na = 0 the one loop
effective potential for the 5D gravitino can be easily worked out. It gives,

4
d 5p 2
ln p + p52 ,
V3/2 =
(3.3)
2 (2)5
where a 5D gravitino has eight degrees of freedom on-shell, the factor 4 appears from the
orbifold action and p5 = m(n)
3/2 .
Using now the mass eigenvalues for gravitinos (2.12) and gauginos and hyperscalars
(2.15), the one-loop effective potential in the presence of the background V52 = /R for
a system of NH hyperscalars and NV vector multiplets propagating in the bulk of the fifth
dimension can be computed as [7]

3(2 + NV NH ) 2i
Li5 e
+ h.c. ,
64 6 R 4
where Lin (x) are the polylogarithm functions
Veff () =
Lin (x) =
k

x
k=1
kn
(3.4)
Using the power expansion of Li5 (e2i ) + h.c. around = 0,

Li5 e2i + h.c. = 2 (5) 4 2 (3)2 + ,
(3.5)
we can see that the potential (3.4) has a maximum (minimum) at

=0
(3.6)
for NH < 2 + NV (NH > 2 + NV ). This means that for NH > 2 + NV the global minimum
of the potential is at = 0 and there is no induced supersymmetry breaking. On the other
hand, for NH < 2 + NV the global minimum is at
= 1/2,
(3.7)
96
supersymmetry is broken radiatively and the KK spectrum of bulk gauginos and

hyperscalars is shifted from n/R to (n + 1/2)/R. We have dubbed this phenomenon
radiative ScherkSchwarz supersymmetry breaking. Notice that our findings are quite
model-independent since they depend only upon the total number of vector multiplets and
hypermultiplets in the bulk.
Some comments are now in order. The first one refers to radius stabilization. As one can
see from (3.4) the one-loop potential does not stabilize the value of R, related to the radion
field. In fact the behaviour is runaway: in particular, for NH < 2 + NV , we obtain at the
global minimum of the effective potential the asymptotic behaviour R 0. However, we
expect the radion to be stabilized by some other mechanism that does not affect the present
results. Massive fields in the bulk seem to be required for radion stabilization [23].
The second comment concerns the appearance of a non-vanishing vacuum energy at the
minimum of the effective potential of order 1/R 4 . This fact is a generic feature of many
models with flat potential, or zero cosmological constant, at the tree-level. We expect on
general grounds that the non-vanishing energy will disturb our original assumption of flat
spacetime. We are not computing the back reaction on the spacetime metrics but expect
its modification to be suppressed as 1/(M5 R)4 , where M5 is the Planck scale of the higherdimensional theory.
The last comment refers to the quality of the supersymmetry breaking solution = 1/2
we have found. Since all the supersymmetric spectrum depends on the actual value of
[7] the value = 1/2 is very constraining for electroweak symmetry breaking in the
different considered models. In the model presented in Ref. [7], where all matter fields
were localized on the branes while Higgs and vector multiplets live in the bulk, the case
= 1/2 is phenomenologically problematic because the theory does not allow for an
H1 H2 Higgs mixing and therefore the VEV of H1 is zero. The same happens for the
alternative model [24] where gauge multiplets and SU(2)L singlets live and the bulk and
SU(2)L doublets are localized on the brane. This problem is however avoided in models
with a single Higgs hypermultiplet, as those based on S 1 /Z2 Z2 [25], although these
models develop a FayetIliopoulos quadratic divergence [26].
We have checked as well that in models with NH < 2 + NV the stability of the minimum
at = 1/2 is not lifted by two-loop corrections. Then it would be desirable to introduce an
extra mechanism that would lead to an effective potential with a minimum at = 0, 1/2.
Since the orbifold S 1 /Z2 has branes at the fixed points, a hidden sector at one of the fixed
points (say at x 5 = 0) can break spontaneously supersymmetry by some brane-localized
dynamics, as, e.g., gaugino condensation. In this case there will be an interplay between
the mechanism that we have discussed up to now and the brane dynamics. This will be the
object of study in the next section.
4. Brane-assisted supersymmetry breaking

The couplings of the minimal and tensor multiplet to the branes have been analyzed in
Ref. [20]. Using the gauge (2.1) they lead to the total Lagrangian,

L = Lgrav + W x 5 Lbrane ,
(4.1)
97
where W is a constant superpotential left out when some brane fields in the hidden sector
are integrated out and
1
Lbrane = 2N 2V51 12t 2 + 2 + ,
2
(4.2)
where we have fixed the VEV of the auxiliary fields that do not play any role in the mass
spectrum of SU(2)R doublets.
We are assuming the hidden sector in the brane located at x 5 = 0 and therefore the
observable brane (where any localized matter fields should be located) at x 5 = R.1 Notice
that the gauge fixing condition (2.1) breaks SU(2)R in the bulk to the U (1)R generated
by 2 , and to nothing in the branes since only the auxiliary gauge field A3 is even (see
Table 1). The bulk U (1)R will drop out by elimination of the auxiliary field V52 from the
BMNP field equation (2.8).
However, the presence of the brane Lagrangian (4.2) will provide a brane mass
contribution to the 5D gravitino and will alter the field equations of V51 , t 1 and t 2 with
respect to the bulk solution (2.6). This in turn will modify the spectrum of mass eigenvalues
of gauginos and hyperscalars as we will see in this section.
The field equation for the auxiliary fields V51 , N , t 1 and t 2 from the Lagrangian (4.1) in
the presence of the brane term (4.2) lead to
1 2
t = t = 0,

N
= W x 5 ,

V51 = W x 5 .
(4.3)
After elimination of the auxiliary fields V51 , N , t 1 and t 2 by their field equations (4.3)
there appear terms of the form 2 (x 5 ). We have checked that these singular terms cancel
leading to a zero cosmological constant at the tree-level. This latter property is expected to
be spoiled by radiative corrections as we have discussed in the previous section.
The presence of the gravitino brane mass term in (4.2) as well as the non-trivial
solution for the field V51 in (4.3) will alter the mass spectrum for gravitinos, gauginos
and hyperscalars, whose mass eigenvalues will departure from the calculated values in
(2.12) and (2.15), and need to be recalculated. The KaluzaKlein spectrum of bulk vector
multiplets, hyperscalars and gravitinos in the presence of the auxiliary fields V52 = /R
and V51 = W (x 5 ) can be computed in different ways. One can either directly solve for
the corresponding equations of motion in D = 5 or diagonalize the corresponding infinite
mass matrices. In this paper, we adopt another strategy, i.e., the Dyson resummation.
4.1. Gauginos
We can expand gauginos in modes according to the parity action

x , x 5 = i 5 3 x , x 5 ,
1 Of course the choice of hidden and observable branes is absolutely arbitrary and can be exchanged by
a coordinate redefinition.
98
i.e., 1L even and 2L odd, as

1
5

1
L
(n) cos nx /R
=
.
L
sin nx 5 /R
2L
R n=
(4.4)
The mass term for gauginos from (4.1), taking into account the classical VEV of V51 in
(4.3), can be written as

5 = i 5 5 + i 1 V 1 + i 2 V 2 .
M1/2 = i 5 D
(4.5)
5
5
The mode expansion (4.4) leads to
1 n + (n) (n)
W (n) 5 (m)
Lm1/2 =
,
+
2 n
R
R n,m
(4.6)
where (n) is defined as a four-component Majorana spinor. The appearance of 5 in the

second term of (4.6) comes from the fact that the (brane) contribution of V51 in the covariant
5 to the gaugino mass is imaginary with respect to that of V 2 .
derivative D
5
The (infinite) mass matrix for modes (n) arising from (4.6) is a very complicated one.
A simple way of diagonalizing it is by considering the first term of (4.6) in the unperturbed
propagator

i
n
m=
p R ),
n+ nm = iR cot(/
p
/
R
m,n
m,n
where p
/ = p , and the second term as a perturbation
W 5
R
that can be introduced in the propagator by a Dyson resummation.
We will then compute the fermion propagator in the bulk but with end points on the
brane x 5 = 0,
5

(n) (m)
p , x = 0 p , x 5 = 0 =
p
p
m = i
n,m
+
i
tan( p
/ R ).
R cot( p
/ R )
(4.7)
tan( p
/ R ) W 5
The mass eigenvalues are then computed as the poles of the propagator in the particle rest
i.e.,
frame, p = (m, 0),
0

det tan m R W 5 ,
(4.8)
leading to
(n)
m1/2
n + 1/2
,
=
R

1
1/2 = arctan
W 2 + tan2
.
1 + W 2 tan2
(4.9)
For W = 0 this reduces to the usual ScherkSchwarz shift 1/2 = while for = 0 one
finds 1/2 = 1/ arctan W .
99
4.2. Gravitinos
Gravitinos can be expanded in modes, according with the parities of Table 1, as
1

5
L
1
(n) cos nx /R
(4.10)
L
=
.
2
sin nx 5 /R
L
R n=
The mass terms for gravitinos from the Lagrangian (4.1) including brane terms in Lbrane
can be written as,
1 n + (n) (n)
W (n) (m)
Lm3/2 =
(4.11)
+
,
2 n
R
R n,m
where (n) is defined as a four-component Majorana spinor.
Diagonalization of the gravitino mass matrix can be done using techniques similar to
those used in the previous subsection for the gauginos. The main difference being that there
is not any 5 factor in the second term of (4.11) and hence no 5 factor in the mass insertion
between modes (n) and (m) . The Dyson resummation of the perturbation would lead to
the resummed propagator with poles (in the particle rest frame) defined as solution of the
equation,

det tan m 0 R W ,
(4.12)
leading to
(n)
m3/2 =
n + 3/2
,
R
3/2 = +
1
arctan W.
(4.13)
4.3. Hyperscalars
The action of the Z2 parity on hyperscalars is defined as

A x , x 5 = 3 A x , x 5 ,
i.e., A1 is even and A2 odd. We can expand hyperscalars in modes as

1
5

1
A
(n) cos nx /R
=
.
A
sin nx 5 /R
A2
R n=
(4.14)
The mass term for hyperscalars from (4.1), taking into account the classical VEV of V51 in
(4.3), can be written as
5 .
5 D
M20 = D
The mode expansion (4.14) leads to the mass Lagrangian,

n + 2

W m n (n)
W2
+i
L0 =
A (n) A(n)
A (m) (0)
A .
R
R
R2
n
m,n
(4.15)
(4.16)
100
Again, as happened for gauginos and gravitinos, the infinite mass matrix for the modes
A(n) arising from (4.16) is very complicated and the simplest method for diagonalizing it
is by considering the first term of (4.16) as part of the propagator

i
n-------- m=
--------
n+ 2 nm
2
m,n
m,n p
R

R
cot(pR ) + cot(pR + )
= D (p) = i
(4.17)
2p
and the second term as the perturbations
W2
,
R
W m+
n- ---
-----m=
,
R R
W n+
,
n - - - - - - - - - m = +
R R
that can be included in the propagator by a Dyson resummation. Notice the appearance of
the interaction proportional to
1
(0) =
1,
R n
n - - - - - - - - m = i(0)
a common feature in S 1 /Z2 compactifications [5]. This factor is necessary for the
consistency of the theory, as we will see next.
We will then compute the propagator in the bulk but with end points on the brane x 5 = 0,

(n) (m)
A
A p , x 5 = 0 A p , x 5 = 0 =
p A
p
n,m
- - - - - - - + - - - -- - - - + - - -
- - - - -- - - - + - - -
- - - - -- - - - +
D (p)
(4.18)
.
1 W 2 cot(pR ) cot(pR + )
The mass eigenvalues are then computed as the poles of the propagator in the particle
i.e., the solutions of the equation
rest frame, p = (m, 0),
tan(mR ) tan(mR + ) W 2 = 0.
(4.19)
Eq. (4.19) can be readily solved as

n + 0
(4.20)
, 0 = 1/2 .
R
where 1/2 was defined in (4.9). The fact that mass eigenvalues for gauginos and
hyperscalars are the same is a consequence of the fact that both matter fields behave
M while
identically with respect to the initial SU(2)R , i.e., with the covariant derivative D
the gravitino is acted by SU(2)R by means of the covariant derivative DM as a consequence
of the gauge fixing in the tensor multiplet. Notice that the presence of the factor (0) was
required to get this identity.
m(n)
0 =
101
4.4. The one-loop effective potential

The previous calculation of the mass shifts for KK modes, 3/2,1/2,0(, W ) shows that
in the two extreme cases so far in the literature W = 0, i.e., no brane effects, and = 0,
i.e., no consideration of the ScherkSchwarz breaking,2 the KK mass spectrum for all bulk
fields is the same. However, unlike the case of W , that is supposed to arise from some brane
dynamics in the hidden sector and so it is a free parameter for our analysis, the parameter
should be found from one-loop corrections. The output will in general lead to distinct
mass spectra that should be obtained upon minimization of the effective potential.
As we have seen above, the presence of the brane induced supersymmetry breaking
(4.2) modified the mass eigenvalues for gravitinos (4.13), gauginos (4.9) and hyperscalars
(4.20). These in turn modify the one-loop effective potential computed in Section 3 that
results now in
i2 (,W )

3
3/2
Li5 e
+ h.c.
Veff () =
6
4
32 R

3(NV NH ) i21/2 (,W )
Li5 e
+ h.c. .
+
(4.21)
64 6 R 4
We expect that the brane effects modify continuously the location of the minimum of the
effective potential away from its minima for W = 0 at = 0 or = 1/2. We have studied
numerically two extreme cases.
In Fig. 1 we have studied the case where only the gravitational and gauge sector are
living in the bulk, while matter and Higgs fields are localized in the observable brane and
Fig. 1. Plot of the minimum of the effective potential min (dotted line) and the corresponding lightest KK mode
mass for the gravitino (dashed line) and gaugino/hyperscalar (solid line) in units of 1/R for the case of only
Standard Model gauge bosons propagating in the bulk: NV = 12, NH = 0.
2 Observe nevertheless that considering the ScherkSchwarz parameter is not optional since it corresponds to
a flat direction of the corresponding tree-level potential.
102
Fig. 2. The same as in Fig. 1 for the case of all Standard Model fields propagating in the bulk: NV = 12, NH = 49.
thus do not participate in the one-loop effective potential. Of course the supersymmetric
partners of localized matter will receive radiative masses from the bulk supersymmetry
breaking [7]. In particular, we have considered the supersymmetric Standard Model gauge
sector in the bulk with NV = 12 and NH = 0. The result is shown in the plot where different
quantities are shown vs the parameter W . The dotted line is the value of the Scherk
Schwarz parameter at the minimum min . We can see that it is equal to 1/2 for W = 0 and
goes smoothly to zero in the limit W . Dashed and solid lines correspond to the values
of the gravitino and gaugino/hyperscalar lightest KK modes. Fig. 2 corresponds to the other
extreme case where all gauge sector and matter (including two Higgs hypermultiplets)
propagate in the bulk. For the case of the supersymmetric Standard Model NV = 12 and
NH = 3 15 + 4 = 49. We can see that the minimum starts at min = 0 at W = 0 and goes
to min = 1/2 in the limit W . In this limit the masses of lightest KK modes can be
much smaller than 1/R.
5. Conclusions
In this paper we have considered a supersymmetric five-dimensional brane-world
scenario where the fifth dimension is compactified on S 1 /Z2 . In this set-up, the Scherk
Schwarz supersymmetry breaking can be interpreted as the Hosotani breaking of the
SU(2)R local symmetry present in off-shell N = 2 supergravity. We have shown that the
ScherkSchwarz supersymmetry breaking parameter is undertemined at the tree-level, but
can acquire two discrete values after loop-corrections from supersymmetric bulk fields are
introduced. In this way, supersymmetry may get broken radiatively through the Scherk
Schwarz mechanism. We have also computed the contribution to the soft supersymmetry
breaking masses of bulk fields in the case in which a source of supersymmetry breaking
appears localized on the hidden brane. In such a case, the ScherkSchwarz supersymmetry
breaking parameter is lifted away from the discrete values 0 and 1/2 and the spectrum
103
of the KK modes of the bulk fields assumes a variety of possibilities depending upon the
strength of supersymmetry breaking on the hidden brane.
Acknowledgements
A.R. thanks the High-Energy Theory group of CSIC where part of this work was
done for the kind hospitality. M.Q. thanks the High-Energy Theory group of University
of Padova where this work was completed for the kind hospitality. One of us (A.R.) thanks
F. Feruglio for useful discussions. The work of G.G. was supported by the DAAD.
References
[1] I. Antoniadis, Phys. Lett. B 246 (1990) 377;
I. Antoniadis, C. Muoz, M. Quirs, Nucl. Phys. B 397 (1993) 515;
I. Antoniadis, K. Benakli, Phys. Lett. B 326 (1994) 69;
I. Antoniadis, K. Benakli, M. Quirs, Phys. Lett. B 331 (1994) 313;
I. Antoniadis, M. Quirs, Phys. Lett. B 392 (1997) 61.
[2] P. Horava, E. Witten, Nucl. Phys. B 460 (1996) 506, hep-th/9510209;
P. Horava, E. Witten, Nucl. Phys. B 475 (1996) 94, hep-th/9603142;
P. Horava, Phys. Rev. D 54 (1996) 7561, hep-th/9608019.
[3] I. Antoniadis, M. Quiros, Nucl. Phys. B 505 (1997) 109, hep-th/9705037;
I. Antoniadis, M. Quiros, Phys. Lett. B 416 (1998) 327, hep-th/9707208;
I. Antoniadis, M. Quiros, Nucl. Phys. Proc. Suppl. 62 (1998) 312, hep-th/9709023.
[4] H.P. Nilles, M. Olechowski, M. Yamaguchi, Phys. Lett. B 415 (1997) 24, hep-th/9707143;
H.P. Nilles, M. Olechowski, M. Yamaguchi, Nucl. Phys. B 530 (1998) 43, hep-th/9801030;
H.P. Nilles, hep-ph/0004064.
[5] E.A. Mirabelli, M.E. Peskin, Phys. Rev. D 58 (1998) 065002, hep-th/9712214.
[6] L. Randall, R. Sundrum, Nucl. Phys. B 557 (1999) 79, hep-th/9810155.
[7] A. Pomarol, M. Quirs, Phys. Lett. B 438 (1998) 255;
I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quiros, Nucl. Phys. B 544 (1999) 503, hep-ph/9810410;
A. Delgado, A. Pomarol, M. Quiros, Phys. Rev. D 60 (1999) 095008, hep-ph/9812489.
[8] D.E. Kaplan, G.D. Kribs, M. Schmaltz, Phys. Rev. D 62 (2000) 035010, hep-ph/9911293;
Z. Chacko, M.A. Luty, A.E. Nelson, E. Ponton, JHEP 0001 (2000) 003, hep-ph/9911323.
[9] T. Gherghetta, A. Pomarol, Nucl. Phys. B 602 (2001) 3, hep-ph/0012378.
[10] G. von Gersdorff, M. Quiros, Phys. Rev. D 65 (2002) 064016, hep-th/0110132.
[11] J. Scherk, J.H. Schwarz, Phys. Lett. B 82 (1979) 60;
J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61;
E. Cremmer, J. Scherk, J.H. Schwarz, Phys. Lett. B 84 (1979) 83.
[12] Y. Hosotani, Phys. Lett. B 126 (1983) 309;
Y. Hosotani, Phys. Lett. B 129 (1983) 193;
Y. Hosotani, Ann. Phys. 190 (1989) 233.
[13] Z. Chacko, M.A. Luty, JHEP 0105 (2001) 067;
T. Kobayashi, K. Yoshioka, Phys. Rev. Lett. 85 (2000) 5527;
D. Mart, A. Pomarol, Phys. Rev. D 64 (2001) 105025, hep-th/0106256;
D.E. Kaplan, N. Weiner, hep-ph/0108001.
[14] E. Cremmer, S. Ferrara, C. Kounnas, D.V. Nanopoulos, Phys. Lett. B 133 (1983) 61;
A.B. Lahanas, D.V. Nanopoulos, Phys. Rep. 145 (1987) 1.
[15] J.A. Bagger, F. Feruglio, F. Zwirner, Phys. Rev. Lett. 88 (2002) 101601, hep-th/0107128;
J. Bagger, F. Feruglio, F. Zwirner, JHEP 0202 (2002) 010, hep-th/0108010.
104
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
T. Gherghetta, A. Riotto, Nucl. Phys. B 623 (2002) 97, hep-th/0110022.

E. Bergshoeff, R. Kallosh, A. Van Proeyen, JHEP 0010 (2000) 033, hep-th/0007044.
R. Altendorfer, J. Bagger, D. Nemeschansky, Phys. Rev. D 63 (2001) 125025, hep-th/0003117.
T. Kugo, K. Ohashi, Prog. Theor. Phys. 104 (2000) 835, hep-ph/0006231;
T. Kugo, K. Ohashi, Prog. Theor. Phys. 105 (2001) 323, hep-ph/0010288;
T. Fujita, T. Kugo, K. Ohashi, hep-th/0106051;
T. Kugo, K. Ohashi, hep-th/0203276.
M. Zucker, Nucl. Phys. B 570 (2000) 267, hep-th/9907082;
M. Zucker, JHEP 0008 (2000) 016, hep-th/9909144;
M. Zucker, Phys. Rev. D 64 (2001) 024024, hep-th/0009083;
M. Zucker, Off-Shell Supergravity in Five Dimensions and Supersymmetric Brane World Scenarios, Ph.D.
Thesis, Bonn University, BONN-IR-2000-10, August 2000.
M.F. Sohnius, P.C. West, Nucl. Phys. B 216 (1983) 100.
R. Barbieri, S. Cecotti, Z. Phys. C 17 (1983) 183;
L. Baulieu, A. Georges, S. Ouvry, Nucl. Phys. B 273 (1986) 366;
M.K. Gaillard, V. Jain, Phys. Rev. D 49 (1994) 1951, hep-th/9308090.
E. Ponton, E. Poppitz, JHEP 0106 (2001) 019, hep-ph/0105021.
A. Delgado, M. Quiros, Nucl. Phys. B 607 (2001) 99, hep-ph/0103058;
A. Delgado, G. von Gersdorff, P. John, M. Quiros, Phys. Lett. B 517 (2001) 445, hep-ph/0104112;
A. Delgado, G.V. Gersdorff, M. Quiros, Nucl. Phys. B 613 (2001) 49, hep-ph/0107233;
M. Quiros, J. Phys. G 27 (2001) 2497.
R. Barbieri, L.J. Hall, Y. Nomura, Phys. Rev. D 63 (2001) 105007;
N. Arkani-Hamed, L.J. Hall, Y. Nomura, D. Smith, N. Weiner, Nucl. Phys. B 605 (2001) 81;
R. Barbieri, L.J. Hall, Y. Nomura, Nucl. Phys. B 624 (2002) 63, hep-th/0107004;
R. Barbieri, L.J. Hall, Y. Nomura, hep-ph/0110102.
D.M. Ghilencea, S. Groot Nibbelink, H.P. Nilles, Nucl. Phys. B 619 (2001) 385, hep-th/0108184.

Bd B d mixing and the Bd J /Ks asymmetry

in general SUSY models
D. Becirevic a , M. Ciuchini b , E. Franco a , V. Gimnez c ,
G. Martinelli a , A. Masiero d , M. Papinutto e , J. Reyes a ,
L. Silvestrini a
a Dip. di Fisica, Univ. La Sapienza and INFN, Sezione di Roma, P.le A. Moro 2, I-00185 Rome, Italy
b Dip. di Fisica, Univ. di Roma Tre and INFN, Sezione di Roma III, Via della Vasca Navale 84,
I-00146 Roma, Italy

c Dep. de Fsica Terica and IFIC, Univ. de Valncia, Dr. Moliner 50, E-46100, Burjassot, Valncia, Spain
d Dip. di Fisica G. Galilei, Univ. di Padova and INFN, Sezione di Padova, Via Marzolo 8,
I-35121 Padua, Italy

e Dip. di Fisica, Univ. di Pisa and INFN, Sezione di Pisa, via Buonarroti 2, I-56100 Pisa, Italy
Abstract
We present a next-to-leading order determination of the gluino-mediated SUSY contributions
to Bd B d mixing and to the CP asymmetry aJ /Ks in the framework of the mass-insertion
approximation. Using hadronic matrix elements recently computed on the lattice, we obtain improved
constraints on the squark-mass splittings. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
With the advent of B factories, B physics is playing a key role in testing the Standard
Model (SM) picture of flavour and CP violation and in probing virtual effects from new
physics at low energies. In particular, measurements of md , the mass difference in
the Bd B d system, and of aJ /Ks , the time-dependent CP asymmetry in Bd J /Ks
decays, can be used to put stringent constraints on new-physics contributions to B = 2
E-mail address: luca.silvestrini@roma1.infn.it (L. Silvestrini).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 9 1 - 2
106
D. Becirevic et al. / Nuclear Physics B 634 (2002) 105119
processes. In terms of the matrix elements of the effective B = 2 Hamiltonian we have

B=2
md = 2 Abs B d |Heff
(1)
|Bd ,
aJ /Ks = sin 2eff sin md t,
(2)
where

B=2
2eff = Arg B d |Heff
|Bd .
(3)
Within the SM the value of sin 2eff = sin 2 can be connected to the CP violation angle
of the Unitarity Triangle (UT)
sin 2 =
2 sin (1 cos )
,
1 + 2 2 cos
(4)

where = 2 + 2 in the Wolfenstein parametrisation [1]. Assuming that there are
no new physics effects in Bd B d and K 0 K 0 mixing, sin 2 is well determined by the
standard UT analysis. For example, Ref. [2] quotes
sin 2 = 0.698 0.066.
(5)
The first direct measurements of aJ /Ks at the B factories suggested a rather low value of
the asymmetry [3] which, if confirmed, would have been a hint of new physics [46]. The
most recent world average [7]
sin 2 = 0.79 0.10,
(6)
however, is in very good agreement with Eq. (5) and therefore does not favour extra
contributions. In spite of this, it is worth investigating, in neutral B meson systems, signals
of new physics which may emerge in the near future thanks to decreasing experimental
errors and theoretical uncertainties. On the one hand, for B = 2 transitions, where the
SM seems to agree with the data, it is important to improve the theoretical accuracy
for constraining new physics contributions to box diagrams. On the other, more accurate
predictions of mixing help to uncover signals of new physics in B decay asymmetries, for
example B K or B K. These processes are dominated by loop contributions and
one measures simultaneously the effect of mixing and CP violation in decay amplitudes,
from both SM and new physics virtual particles. Finally, of particular interest is the Bs B s
mixing for which the SM amplitude is real and the phase entirely originates from new
physics effects. This may give rise to CP violating effects which would be absent in the SM.
In this work, we aim at testing new-physics effects in B = 2 transitions using the most
recent experimental and theoretical results: the new data from B factories and the very
recent lattice determinations of the relevant matrix elements, which allow the computation
of md and of aJ /Ks at the Next-to-Leading Order (NLO), apart from some additional
model-dependent O(s (MW )) terms. First of all, we present model-independent formulae
for B = 2 (Bd and Bs ) transitions which include all possible new-physics contributions.
These formulae are obtained by considering the most general effective Hamiltonian for
B=2 . We write down the expression for m ( m ) and for in
B = 2 processes, Heff
d
s
eff
terms of the Wilson coefficients at the electroweak scale, including NLO QCD corrections
in the running from MW to = mb [8,9] and the matrix elements very recently computed
107
in lattice QCD [10], in the same RI-MOM renormalisation scheme. These expressions can
be readily used to compute md and eff in any extension of the SM.
As a concrete example of physics beyond the SM, we then consider the Minimal
Supersymmetric Standard Model (MSSM) with arbitrary soft breaking terms and we
study model-independent constraints on the SUSY parameter space coming from md
and aJ /Ks . We closely follow the analysis of Ref. [11], where upper bounds on new
sources of flavour and CP violation were obtained from the study of K 0 K 0 mixing. For
earlier analyses, either at tree level or with LO evolution, using nave Vacuum Insertion
Approximation (VIA) B parameters, see Refs. [12] and [13], respectively. Ref. [11] and
the present work can be considered as first steps toward a full NLO model-independent
analysis of flavour and CP violation beyond the SM, with special focus on its most
studied extension, namely the general MSSM. We stress that the inclusion of hadronic
matrix elements from lattice QCD, for operators renormalised consistently with the Wilson
coefficients at the NLO, is highly important to reduce the uncertainties in the study of
flavour and CP violation. Indeed, while matrix elements for operators appearing in the
SM have been studied for a long time, it was only recently that similar results have been
obtained for operators that only come from physics beyond the SM [10,14]. More work
is still to be done along these lines. For example, the large uncertainties in the SUSY
constraints from the electric dipole moment of the neutron could be drastically reduced by
a lattice computation of the relevant matrix elements [15].
The paper is organised as follows. In the next section we introduce the most general
B=2 with the inclusion of the NLO QCD corrections in the evolution from the scale
Heff
of new physics down to low energy. We provide the expression of the Wilson coefficients
at the low hadronic scale ( mb ) as a function of the Wilson coefficients and of s at the
new physics scale. Section 3 deals with the evaluation of the hadronic matrix elements of
B=2 . Here we replace the traditional values of the B
the local operators appearing in Heff
parameters in the VIA with the values recently obtained from a lattice computation in the
RI-MOM scheme [10]. We then turn to SUSY and present in Section 4 the expression
B=2 . As long as we deal with general squark
of gluino-mediated contributions to Heff
mass matrices, the inclusion of the gluino-mediated FCNC alone is sufficient to get the
correct bulk of the SUSY contribution. In specific models corresponding to particularly
restricted squark mass matrices, however, there are regions of the SUSY parameter space
where charginos, stops and/or charged Higgs are relatively light and their exchange gives
contributions even larger than gluinos to FCNC [16]. This happens more easily in the
case of Bd B d mixing, rather than for K 0 K 0 , because the relevant mixing angles and/or
Yukawa couplings are larger in this case. Unfortunately, the experimental information and
the theoretical accuracy are presently insufficient to attempt a simultaneous analysis which
includes a spanning of the squark mass matrices and of the stop/chargino mass values. For
this reason, in this first study which uses the new measurements of sin 2, NLO corrections
and lattice B parameters, we include gluino-mediated FCNC only. Our quantitative results
are presented in Section 5 and summarised in Tables 2 and 3. We perform our computations
in the mass insertion approximation [17]. We choose the super-CKM basis for the fermion
and sfermion states, where all the couplings of these particles to neutral gauginos are
flavour diagonal, while genuine SUSY FC effects are exhibited by the non-diagonal entries
of the sfermion mass matrices. Denoting by 2 the off-diagonal terms in the sfermion mass
108
matrices (i.e., the mass terms relating sfermions of the same electric charge, but different
flavour), the sfermion propagators can be expanded as a series in terms of = 2 /m
2,
2,
where m
is the average sfermion mass. As long as 2 is significantly smaller than m
we can just take the first term of this expansion and then the experimental information
concerning FCNC and CP violating phenomena translates into upper bounds on the s [11,
12]. Tables 2 and 3 contain new constraints on the parameters . With respect to previous
analyses, we find that the inclusion of the direct measurements of sin 2 allows imposing
quite stringent constraints on Im 13 [18]. On the theoretical side, as already observed
in [11], the QCD evolution from large scales and the use of the lattice B parameters induces
sizable changes in the limits on the values of the s. The major effect of our improved
computation is felt by leftright operators.
2. Effective Hamiltonian for B = 2 processes beyond the SM
The most general effective Hamiltonian for B = 2 processes beyond the SM has the
following form:
B=2
Heff
=
5
Ci Qi +
i=1
3
i Q
i ,
C
(7)
i=1
where

Q1 = dL bL
dL bL ,

Q2 = dR bL
dR bL ,

Q3 = dR bL dR bL
,

Q4 = dR bL
dL bR ,

Q5 = dR bL dL bR
,
(8)
1,2,3 are obtained from the Q1,2,3 by the exchange L R. Here

and the operators Q
qR,L = PR,L q, with PR,L = (1 5 )/2, and and are colour indexes. In the case of
the Bs system one has simply to replace the d with the s quark in the operators appearing
in Eqs. (8).
F =2 has been computed
The NLO anomalous dimension matrix for the most general Heff
in [8] and the results have been confirmed in [9]. We use the Regularisation-Independent
anomalous dimension matrix in the Landau gauge (also known as RI-MOM), since we
will make use of matrix elements computed in lattice QCD with the same choice of
renormalisation scheme.
A full NLO computation would also require the O(s ) corrections to the matching
conditions which determine the Wilson coefficients, see Eqs. (15). These corrections are
model-dependent and have been computed only in few specific cases [19,20]. One might
argue that, being of order s (MS ) (here and in the following MS represents the scale of
new physics), these contributions should be small, as suggested by the cases of the SM
and of the two Higgs doublet model [19]. This statement, however, can only be confirmed
109
Table 1
Input quantities used in the phenomenological analysis. mdMS and mbMS are the MS masses whereas mt is the
pole mass
Constant
Value
Constant
Value
mBd
5.279 GeV
fBd
200 30 MeV
mMS
d (2 GeV)
7 MeV
mbMS (mMS
b )
4.23 GeV
mt
174 5 GeV
s (MZ )
0.119
|Vcb |
(40.7 1.9) 103
|Vub |
(3.61 0.46) 103
by explicit computations in specific models. Due to the absence of O(s ) corrections to

F =2 are affected by a residual scheme dependence, which would be
the matching, our Heff
cancelled by the terms of order s (MS ) in the Wilson coefficients Ci (MS ). These terms
are model dependent.
The Ci (MS ) are obtained by integrating out all new particles simultaneously at the scale
MS . We then have to evolve the coefficients down to the hadronic scale = mb = 4.6 GeV
(mb mb ( = mb ) is the RI-MOM mass), which is the renormalisation scale of the
operators used in Ref. [10]. By coincidence, mb has the same numerical value as the pole
pole
mass, mb = 4.6 GeV, extracted at the NLO from the MS mass mbMS (mMS
b ) = 4.23 GeV,
see Table 1. For consistency, we have to evolve the Wilson coefficients at the same
renormalisation scale. The SM contribution can be computed independently and evolved
from MW to using the well-known NLO QCD corrections [21].
We give here an analytic formula for the expression of the Wilson coefficients at the
scale = mb as a function of the initial conditions at the SUSY scale Ci (MS ) and of
s (MS ). This formula has been obtained by using, for the SM parameters, the values in
Table 1. For MS > mt we obtain
pole (r,s)
(r,s) ai
=
b i + ci
Cs (MS ),
Cr mb
(9)
i
where, in the evolution of the coefficients from MS , we have chosen MS = (Mg + Mq )/2
and = s (MS )/s (mt ). The magic numbers are given below:
ai = (0.286, 0.692, 0.787, 1.143, 0.143),
bi(11) = (0.865, 0, 0, 0, 0),
(22)
bi = (0, 1.879, 0.012, 0, 0),
(23)
bi = (0, 0.493, 0.18, 0, 0),
(32)
bi = (0, 0.044, 0.035, 0, 0),
bi(33) = (0, 0.011, 0.54, 0, 0),
bi(44) = (0, 0, 0, 2.87, 0),
(45)
bi = (0, 0, 0, 0.961, 0.22),
(54)
bi = (0, 0, 0, 0.09, 0),
bi(55) = (0, 0, 0, 0.029, 0.863),
ci(11) = (0.017, 0, 0, 0, 0),

(22)
ci = (0, 0.18, 0.003, 0, 0),
(23)
ci = (0, 0.014, 0.008, 0, 0),
(32)
ci = (0, 0.005, 0.012, 0, 0),
ci(33) = (0, 0.000, 0.028, 0, 0),
ci(44) = (0, 0, 0, 0.48, 0.005),
(45)
ci = (0, 0, 0, 0.25, 0.006),
(54)
ci = (0, 0, 0, 0.013, 0.016),
ci(55) = (0, 0, 0, 0.007, 0.019),
(10)
110
and we have only written the non-vanishing entries. The magic numbers for the evolution
13 are the same as the ones for the evolution of C13 . Formulae (9) and (10)
of C
pole
are combined with the B parameters evaluated at mb = 4.6 GeV in the MOM-RI
1
scheme [10], given in Eqs. (13). In this way it is possible to compute the contribution
to md,s and eff at the NLO in QCD for any model of new physics in which the new
contributions with respect to the SM originate from extra heavy particles. One just has to
plug in the expression for the Ci evaluated at the large energy scale MS in his favourite
model. When the O(s ) corrections to the Ci (MS ) are available, one can obtain a full
NLO, regularisation-independent result; in the cases where this corrections have not been
computed yet, the results contain a residual systematic uncertainty of order s (MS ). We
note that, due to the presence of large entries in the NLO anomalous dimension matrix,
a systematic uncertainty is present in the QCD evolution from the SUSY scale to the
hadronic one. This uncertainty stems from different ways of writing the NLO evolution
matrix, which differ by terms of O(s2 ), and has been taken into account in the numerical
analysis as uncertainty in the numerical value of the magic numbers. For this reason we
have presented only the significant figures of the magic numbers, which have an uncertainty
(44)
of one unit on the last digit. The only exception is c4 for which the uncertainty is 3 units.
3. Hadronic matrix elements
The matrix elements of the operators Qi between neutral B mesons in the VIA are given
by:
1
B d |Q1 |Bd VIA = mBd fB2d ,
3
2

5
m Bd
mBd fB2d ,
B d |Q2 |Bd VIA =
24 mb + md

2
m Bd
1
mBd fB2d ,
B d |Q3 |Bd VIA =
24 mb + md
2

1
1
m Bd
B d |Q4 |Bd VIA =
+
mBd fB2d ,
24 4 mb + md

2
1
m Bd
1
+
mBd fB2d ,
B d |Q5 |Bd VIA =
8 12 mb + md
(11)
where mBd is the mass of the Bd meson and mb and md are the masses of the b and d
quarks, respectively. Note that the normalisation of the operators used in this paper differs
by a factor of 2mBd from the one used in Ref. [10]. Here and in the following, the same
13 ,
expressions of the B parameters of the operators Q13 are valid for the operators Q
since strong interactions preserve parity. In the case of the Bs system one has simply to
replace the d with the s quark in the operators and expressions appearing in Eqs. (11).
1 Magic numbers in the NDR scheme can be found in Ref. [22].
111
In the case of the renormalised operators, we define the B parameters as follows:

1 ()|Bd = 1 mBd fB2 B1 (),
B d |Q
d
3

2
5
m Bd

mBd fB2d B2 (),
B d |Q2 ()|Bd =
24 mb () + md ()

2
m Bd
3 ()|Bd = 1
mBd fB2d B3 (),
B d |Q
24 mb () + md ()

2
1
m Bd

B d |Q4 ()|Bd =
mBd fB2d B4 (),
4 mb () + md ()

2
1
m Bd

B d |Q5 ()|Bd =
mBd fB2d B5 (),
12 mb () + md ()
(12)
i () (or simply Q
i ) denotes the operators renormalised at the scale .
where the notation Q
For consistency with the calculation of the B parameters in Ref. [10], the quark masses,
evolved at the scale , must be computed in the Landau RI-MOM scheme [23,24]. With the
numbers in Table 1, this corresponds to mb = mb (mb ) = 4.6 GeV and md (mb ) = 5.4 MeV
(RI-MOM).2
A few words of explanation are necessary at this point. The B parameter of the
matrix element B d |Q1 |Bd has been extensively studied on the lattice due to its
phenomenological relevance [25,26], and used in many UT analyses [2,27]. For the other
operators, instead, all the studies beyond the SM have used the VIA B parameters, which
in some cases, as will be shown below, is a crude approximation. The numerical results for
the B parameters Bi () refer to the RI-MOM scheme. We have used the same definition
of the B parameters as in Ref. [10] from which we have taken the values. In our numerical
study, we have used
B1 (mb ) = 0.87(4)+5
4 ,
B2 (mb ) = 0.82(3)(4),
B3 (mb ) = 1.02(6)(9),
B4 (mb ) = 1.16(3)+5
7 ,
B5 (mb ) = 1.91(4)+22
7 .
(13)
We added a conservative 10% systematic error to the errors quoted above, based on
previous experience on discretisation, extrapolation and quenching effects. In the cases
where a comparison has been possible, the quenching effects for all the B parameters
computed so far have always been found much smaller than all other uncertainties.
4. Effective Hamiltonian for B = 2 processes in SUSY
B=2
In this section, we describe the computation of gluino-mediated contributions to Heff
at the NLO in QCD.
2 The precise value of m is not very important for our analysis.

d
112
Several cases may be considered: (i) mq mg , (ii) mq mg and (iii) mq mg , where

mq is the average squark mass and mg is the gluino mass. Case (ii), in which mq mg ,
cannot be realised in the framework considered here, in which the soft SUSY breaking
terms are introduced at the Planck scale. This is due to the fact that the evolution from
the Planck to the electroweak scale forbids such a mass hierarchy. In fact, neglecting the
effects of Yukawa couplings and A-terms, one obtains in the down sector the following
approximate expression for the ratio x = m2g /m2q at the electroweak scale, in terms of the
value x0 at the Planck scale [28]:
x
9x0
9
.
1 + 7x0
7
(14)
Even if one starts at the super-large scale with an extreme hierarchy between squark and
gluino masses (x0 1), at the electroweak scale the two masses will be of the same order.
For this reason, we will not consider the case mq mg in our analysis. Case (iii), in which
mq mg , can be realised in some special models, such as effective supersymmetry [30]
or models with a light gluino [31]. The NLO corrections in this case have been computed
in Ref. [32]. In this first study, however, we do not consider these cases and present only
results for the case mq mg .
The mass insertion approximation, which we have adopted in our analysis, presents
the major advantage that one does not need the full diagonalisation of the sfermion mass
matrices to perform a test of FCNC in the SUSY model under consideration. It is enough to
compute ratios of the off-diagonal over the diagonal entries of the sfermion mass matrices
and compare the results with the general bounds on the s which we provide here from
mBd and aJ /Ks . This formulation of the mass insertion approximation in terms of the
parameters is particularly suitable for model-independent analyses, but involves two
further assumptions: the smallness of the off-diagonal mass terms with respect to the
diagonal ones, and the degeneracy of the diagonal mass terms in the super-CKM basis.
The latter assumption, related to the use of the average squark mass m,
is well justified in
our case, since, on quite general grounds, one does not expect a sizable non-degeneracy of
the down-type squarks. In SUSY GUT it may happen that the sbottom mass is smaller than
the d and s squark masses. Also in this case, namely for non-degenerate diagonal mass
terms, however, it is possible to define a generalised mass insertion approximation [29]
and apply the methods exposed in this paper.
There exist four different mass-insertions connecting flavours d and b along a
d ) , ( d )
d
d
sfermion propagator: (13
LL
13 RR , (13 )LR and (13 )RL . The indexes L and R refer
to the helicity of the fermion partners. The amplitude for B = 2 transitions in the full
theory at the LO is given by the computation of the diagrams in Fig. 1. At the lowest order
in QCD, in the basis of Eq. (8), the Wilson coefficients are given by [12]:
C1 =
d 2
s2
24 x f6 (x) + 66 f6 (x) 13
,
LL
216m2q
C2 =
d 2
s2
204 x f6 (x) 13
,
RL
2
216mq
113
Fig. 1. Feynman diagrams for B = 2 transitions, with h, k, l, m = {L, R}.
C3 =
d 2
s2
36 x f6 (x) 13
,
RL
2
216mq
d d
s2
504 x f6 (x) 72 f6 (x) 13
,
LL 13 RR
2
216mq
d d
132 f6(x) 13
,
LR 13 RL
C4 =
d d
s2
24 x f6 (x) + 120 f6(x) 13
,
LL 13 RR
2
216mq
d d
,
180 f6(x) 13
LR 13 RL
C5 =
2
d 2
1 = s
,
C
24 x f6 (x) + 66 f6 (x) 13
RR
2
216mq
2

2 = s 204 x f6 (x) d 2 ,
C
13 LR
2
216mq
3 =
C
d 2
s2
36 x f6 (x) 13
,
LR
2
216mq
(15)
where x = m2g /m2q and the functions f6 (x) and f6 (x) are given by:
6(1 + 3x) ln x + x 3 9x 2 9x + 17
,
6(x 1)5
6x(1 + x) ln x x 3 9x 2 + 9x + 1
f6 (x) =
.
3(x 1)5
f6 (x) =
(16)
In the absence of O(s ) corrections to the matching, we interpret the Ci given above as
coefficients computed at the large energy scale MS mq mg , i.e., Ci Ci (MS ).
114
5. Numerical analysis of B = 2 processes

In this section, we illustrate the procedure followed in our analysis and give the main
B=2 |B which
results. md and sin 2 are defined in terms of the matrix element B d |Heff
d
we schematically write as
B=2
|Bd = Re ASM + i Im ASM
B d |Heff
d 2
d 2
+ ASUSY Re 13
+ i ASUSY Im 13
,
AB
AB
(17)
where ASM is the Standard Model contribution which depends on the CKM matrix
d
)AB denotes the generic effective
parameters, in particular on the CP violation phase . (13
coupling appearing in Eq. (15). In the following, for simplicity, the constraints are obtained
d )
by imposing that the sum of the SUSY contributions proportional to a single (13
AB
parameter and the SM contribution does not exceed by more than 1 the experimental
value of md and sin 2. This is justified (a posteriori) by noting that the constraints on
d
different (13
)AB parameters exhibit a hierarchical structure, and therefore interference
effects between different contributions would require a large amount of fine tuning. We
d
d
d
d
)LL only; (ii) (13
)LL = (13
)RR ; (iii) (13
)LR
consider the following four cases: (i) (13
d
d
only; (iv) (13 )LR = (13 )RL . For this reason, in all the cases considered here, the amplitude
d 2
in Eq. (17) only depends on a single (13
)AB .
The procedure used in our numerical analysis is the following:
we scan over all possible values of between 0 and 2 since, in the presence of
SUSY effects, this parameter is no longer constrained by the UT analysis. In particular,
the case of in the second quadrant has recently been studied in connection with
K [33]. Furthermore, since 5K may be subject to large SUSY contributions
d )
due to the (12

AB couplings, values of > 180 are allowed;
d
d
)AB between 1 and 1;
for a fixed value of , we scan over Re(13 )AB and Im(13
we let the other experimental and theoretical parameters, in particular the B
parameters, vary within the ranges given in Table 1 and in Eq. (13);
d )
d
in the ( , Re(13
AB , Im(13 )AB ) space, we select the region where the predicted
values of md and sin 2 lie within 1 from the measured values.
As can be seen from Eq. (17), the amplitude only depends on the square of the
d )2 . Therefore, the selected values of ( d )
coupling, (13
AB
13 AB have a twofold ambiguity
d
d
d
d
(Re(13 )AB Re(13
)AB and Im(13
)AB Im(13
)AB ) as will be evident from our
results.
We are ready to discuss the physics results. In Figs. 23, for mq = mg = 500 GeV, we
d
d
show the allowed regions in the ( , Re(13
)AB , Im(13
)AB ) space and the corresponding
d
d
projections on the (Re(13 )AB , Im(13 )AB ) plane. For the readers convenience, we have
coloured regions corresponding to in the four quadrants with different colours. An
alternative way of presenting our results is given in Fig. 4 where the allowed values of
d
d
)AB are plotted as a function of Arg(13
)AB .
Abs(13
d )
d
We give the maximum allowed values of | Re(13
AB | and | Im(13 )AB | in Table 2 for
x = 0.25, 1 and 4. We also give results for mq = 250 GeV and 1000 GeV in Table 3. We
115
d ) , Im( d ) ) space with ( d )

d
d
Fig. 2. Allowed regions in the ( , Re(13
LL
13 LL
13 LL only (left) and (13 )LL = (13 )RR
d ) Im( d )
(right). The two lower plots are the corresponding projections in the Re(13
plane.
Different
LL
13 LL
colours denote values of belonging to different quadrants.
d )
d
d
d
d
Fig. 3. Allowed regions in the ( , Re(13
LR , Im(13 )LR ) space with (13 )LR only (left) and (13 )LR = (13 )RL
d
d
(right). The two lower plots are the corresponding projections in the Re(13 )LR Im(13 )LR plane. Different
colours denote values of belonging to different quadrants.
d )
find that the constraints on the (13
AB parameters are less effective by a factor from 10
d
to 100 with respect to those extracted on (12
)AB from S = 2 transitions. The reason for
this difference is twofold. On the one hand, one naturally expects constraints of the order of
sin3 c for Bd B d mixing and sin5 c for K 0 K 0 mixing, dictated by the top contribution
116
d )
d
Fig. 4. Allowed values of Abs(13
LR as a function of Arg(13 )AB in the four considered cases. Different colours
denote values of belonging to different quadrants.
in the SM amplitude. On the other, in the K 0 K 0 case, LR contributions are chirally

enhanced, so that limits coming from chirality-changing operators are more effective.
d
d
Obviously, specific models of flavour, where (12
)AB and (13
)AB are correlated, can
constrain the squark-mass splittings more severely. In these cases one could even attempt
the inclusion of chargino contributions to the mixing [18,34]. Our general formulae for
the effective Hamiltonian and the magic numbers remain valid, and the phenomenological
analysis can be carried out along the same lines.
We now discuss the difference between our results and those obtained with tree level or
LO Wilson coefficients and VIA matrix elements. In this case, with our definition of the B
parameters given in Eq. (12), with = (mb /mBd )2 = 0.76, we have used
B1VIA = 1.0,
B2VIA = 1.0,
B3VIA = 1.0,
B4VIA = 1.13,
B5VIA = 2.14.
(18)
With NLO corrections and lattice B parameters, the constraints are generically looser
than using LO and VIA B parameters. The theoretical uncertainties of the NLO constraints
are, however, much smaller. In the LL = RR case, the NLO bounds are even more stringent
than the LO ones, in spite of the uncertainties on the B parameters. The most striking effect
is obtained in the LR = RL case for x = 4. Here, the strong cancellation taking place at the
d )
LO is exacerbated by NLO corrections, jeopardising any possibility to constrain (13
LR
in this case. Clearly, given the cancellation, this result should be interpreted with care and
verified by calculations in any given model.
117
Table 2
d)
d
Maximum allowed values for | Re(ij
AB | and | Im(ij )AB |, with A, B = (L, R), for an average squark mass
mq = 500 GeV and for different values of x = m2g /m2q . We give the results in the following cases: (i) with
the tree level Wilson coefficients, namely without evolution from MS to mb , and VIA B parameters, denoted
by TREE; (ii) with LO evolution and VIA B parameters, denoted by LO; (iii) with NLO evolution and lattice
B parameters, denoted by NLO. The missing entries correspond to cases in which no constraint was found for
d)
|(ij
AB | < 0.9
x
TREE
0.25
1.0
4.0
4.9 102
1.1 101
6.0 101
0.25
1.0
4.0
1.1 101
2.6 101
0.25
1.0
4.0
0.25
1.0
4.0
2.6 101
LO
d ) |
| Re(13
LL
5.4 102
NLO
TREE
6.2 102
1.4 101
7.0 101
3.1 102
3.4 102
4.7 102
1.2 101
1.3 101
1.3 102
2.9 101
3.4 101
2.0 102
1.2 101
6.7 101
d ) |
| Im(13
LL
2.8 101
d ) |
| Re(13
LR
3.0 101
1.5 102
3.4 102
2.7 102
3.0 102
3.8 102
5.3 102
4.1 102
4.5 102
1.2 101
3.9 102
7.6 102
8.7 102
1.2 101
3.0 102
d ) |
| Im(13
LR
6.0 102
6.6 102
9.2 102
3.3 102
6.6 102
7.4 102
1.0 101
8.3 102
1.5 102
3.6 102
2.7 101
LO
d )
| Re(13
LL=RR |
2.0 102
2.2 102
3.0 102
d )
| Im(13
LL=RR |
8.0 103
9.0 103
1.3 102
d )
| Re(13
LR=RL |
2.7 102
5.4 102
2.5 101
d )
| Im(13
LR=RL |
9.0 103
2.4 102
5.7 101
NLO
1.9 102
2.1 102
2.8 102
8.0 103
9.0 103
1.2 102
2.6 102
5.2 102
9.0 103
2.3 102
With the future measurements of ms at proton colliders, and hopefully of the

corresponding CP violating effects in Bs mixing and decays, it will be soon possible to
perform a similar analysis for the Bs system. Work along this line is in progress [35].
6. Conclusions
In this work we have provided an improved determination of the gluino-mediated
SUSY contributions to Bd B d mixing and to the CP asymmetry aJ /Ks in the framework
of the mass-insertion method. The improvement consists in introducing the NLO QCD
B=2 [8,9] and in replacing the VIA matrix elements with their recent
corrections to Heff
lattice computation [10]. As a glimpse at Table 2 readily reveals, these improvements affect
B=2 that one considers.
previous results in a different way, according to the operators of Heff
The effect is particularly large for leftright operators.
B=2 at the
We have provided an analytic formula for the most general low-energy Heff
NLO, in terms of the Wilson coefficients at the high energy scale. This formula can be
readily used to compute md and sin 2 in any extension of the SM with new heavy
particles.
118
Table 3
d)
d
Maximum allowed values for | Re(ij
AB | and | Im(ij )AB |, with A, B = (L, R), for an average squark mass
mq = 250 GeV or 1000 GeV and for different values of x = m2g /m2q . We give the results with NLO evolution
and lattice B parameters. The missing entries correspond to cases in which no constraint was found for
d)
|(ij
AB | < 0.9
x
mq = 250 GeV
0.25
1.0
4.0
6.5 102
3.2 101
mq = 1000 GeV
d
| Re(13 )LL |
1.3 101
3.1 101
d ) |
| Im(13
LL
0.25
1.0
4.0
1.3 101
1.5 101
d )
| Re(13
LR |
0.25
1.0
4.0
1.5 102
2.2 102
3.3 102
4.8 102
6.2 102
6.8 102
9.4 102
d )
| Im(13
LR |
0.25
1.0
4.0
2.8 101
5.0 101
1.4 101
1.5 101
2.2 101
mq = 250 GeV
mq = 1000 GeV
d
| Re(13 )LL=RR |
3.8 102
1.0 102
1.4 102
4.2 102
5.9 102
d )
| Im(13
LL=RR |
4.0 103
6.0 103
1.6 102
1.8 102
2.5 102
d )
| Re(13
LR=RL |
2.5 102
5.3 102
1.1 101
d )
| Im(13
LR=RL |
9.0 103
2.1 102
4.5 102
FCNC and CP violating phenomena (in particular in B physics) are promising candidates for some indirect SUSY signal before LHC, and are in many ways complementary to
direct SUSY searches. From this point of view the theoretical effort to improve as much as
possible our precision on FCNC computations in a SUSY model-independent framework
is certainly worth and, hopefully, rewarding.
Acknowledgements
We thank Fabrizio Parodi for information about the world average of the direct
measurements of sin 2 and for discussions. This work has been partially supported by
the European Communitys Human potential programme Hadron Phenomenology from
Lattice QCD contract number HPRN-CT-2000-00145 and by the RTN European project
Across the Energy Frontier contract number HPRN-CT-2000-0148.
References
[1] L. Wolfenstein, Phys. Rev. Lett. 51 (1983) 1945.
[2] M. Ciuchini et al., JHEP 0107 (2001) 013, hep-ph/0012308.
[3] T. Affolder et al., CDF Collaboration, Phys. Rev. D 61 (2000) 072005, hep-ex/9909003;
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
119
D.G. Hitlin, BABAR Collaboration, hep-ex/0011024;

H. Aihara, BELLE Collaboration, hep-ex/0010008.
A.J. Buras, R. Buras, Phys. Lett. B 501 (2001) 223, hep-ph/0008273.
G. Eyal, Y. Nir, G. Perez, JHEP 0008 (2000) 028, hep-ph/0008009.
A. Masiero, M. Piai, O. Vives, Phys. Rev. D 64 (2001) 055008, hep-ph/0012096.
This average has been taken from http://parodi.home.cern.ch/parodi/ckm/ckm.html and it is based on the
following references: Opal Collaboration, Eur. Phys. J. C 5 (1998) 379;
Aleph Collaboration, Phys. Lett. B 492 (2000) 259;
CDF Collaboration, Phys. Rev. Lett. 81 (1998) 5513;
Babar Collaboration, Phys. Rev. Lett. 86 (2001) 2515;
Belle Collaboration, Phys. Rev. Lett. 87 (2001) 091802.
M. Ciuchini, E. Franco, V. Lubicz, G. Martinelli, I. Scimemi, L. Silvestrini, Nucl. Phys. B 523 (1998) 501,
hep-ph/9711402.
A.J. Buras, M. Misiak, J. Urban, Nucl. Phys. B 586 (2000) 397, hep-ph/0005183.
D. Becirevic, V. Gimenez, G. Martinelli, M. Papinutto, J. Reyes, hep-lat/0110091;
See also hep-lat/0110117.
M. Ciuchini et al., JHEP 9810 (1998) 008, hep-ph/9808328.
F. Gabbiani, A. Masiero, Nucl. Phys. B 322 (1989) 235;
J.S. Hagelin, S. Kelley, T. Tanaka, Nucl. Phys. B 415 (1994) 293;
E. Gabrielli, A. Masiero, L. Silvestrini, Phys. Lett. B 374 (1996) 80, hep-ph/9509379;
F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321, hep-ph/9604387.
J.A. Bagger, K.T. Matchev, R.J. Zhang, Phys. Lett. B 412 (1997) 77, hep-ph/9707225.
L. Conti et al., Nucl. Phys. Proc. Suppl. 73 (1999) 315, hep-lat/9809162;
A. Donini, V. Gimenez, L. Giusti, G. Martinelli, Phys. Lett. B 470 (1999) 233, hep-lat/9910017.
G. Martinelli, Plenary talk at LATTICE 2001, The XIX International Symposium on Lattice Field Theory,
August 1924, 2001, Berlin, hep-lat/0112011.
M. Misiak, S. Pokorski, J. Rosiek, Adv. Ser. Direct. High Energy Phys. 15 (1998) 795, hep-ph/9703442.
L.J. Hall, V.A. Kostelecky, S. Raby, Nucl. Phys. B 267 (1986) 415.
C.K. Chua, W.S. Hou, hep-ph/0110106.
J. Urban, F. Krauss, U. Jentschura, G. Soff, Nucl. Phys. B 523 (1998) 40, hep-ph/9710245.
T.-F. Feng et al., Phys. Rev. D 63 (2001) 015013, hep-ph/0008029.
A.J. Buras, M. Jamin, P.H. Weisz, Nucl. Phys. B 347 (1990) 491.
A.J. Buras, S. Jager, J. Urban, Nucl. Phys. B 605 (2001) 600, hep-ph/0102316.
G. Martinelli, C. Pittori, C.T. Sachrajda, M. Testa, A. Vladikas, Nucl. Phys. B 445 (1995) 81.
A. Donini, V. Gimenez, G. Martinelli, M. Talevi, A. Vladikas, Eur. Phys. J. C 10 (1999) 121, heplat/9902030.
C.W. Bernard, Nucl. Phys. Proc. Suppl. 94 (2001) 159, hep-lat/0011064.
S. Ryan, Plenary talk at LATTICE 2001, The XIX International Symposium on Lattice Field Theory, August
1924, 2001, Berlin, hep-lat/0111010.
A. Hocker, H. Lacker, S. Laplace, F. Le Diberder, Eur. Phys. J. C 21 (2001) 225, hep-ph/0104062.
D. Choudhury, F. Eberlein, A. Konig, J. Louis, S. Pokorski, Phys. Lett. B 342 (1995) 180, hep-ph/9408275.
A.J. Buras, A. Romanino, L. Silvestrini, Nucl. Phys. B 520 (1998) 3, hep-ph/9712398.
A.G. Cohen, D.B. Kaplan, A.E. Nelson, Phys. Lett. B 388 (1996) 588, hep-ph/9607394.
G.R. Farrar, Phys. Rev. Lett. 53 (1984) 1029.
R. Contino, I. Scimemi, Eur. Phys. J. C 10 (1999) 347, hep-ph/9809437.
G. DAmbrosio, G. Isidori, hep-ph/0112135.
A. Masiero, M. Piai, A. Romanino, L. Silvestrini, Phys. Rev. D 64 (2001) 075005, hep-ph/0104101.
D. Becirevic et al., in preparation.

7D bosonic higher spin gauge theory:

symmetry algebra and linearized constraints
E. Sezgin a , P. Sundell b
a Center for Theoretical Physics, Texas A&M University, College Station, TX 77843, USA
b Department for Theoretical Physics, Uppsala University, Box 803, S-75108, Uppsala, Sweden
Received 11 March 2002; accepted 16 April 2002
Abstract
We construct the minimal bosonic higher spin extension of the 7D AdS algebra SO(6, 2), which we
call hs(8 ). The generators, which have spin s = 1, 3, 5, . . . , are realized as monomials in Grassmann
even spinor oscillators. Irreducibility, in the form of tracelessness, is achieved by modding out an
infinite-dimensional ideal containing the traces. In this a key role is played by the tree bilinear traces
which form an SU(2)K algebra. We show that gauging of hs(8 ) yields a spectrum of physical fields
with spin s = 0, 2, 4, . . . which make up a UIR of hs(8 ) isomorphic to the symmetric tensor product
of two 6D scalar doubletons. The scalar doubleton is the unique SU(2)K invariant 6D doubleton.
The spin s 2 sector comes from an hs(8 )-valued one-form which also contains the auxiliary
gauge fields required for writing the curvature constraints in covariant form. The physical spin s = 0
field arises in a separate zero-form in a quasi-adjoint representation of hs(8 ). This zero-form
also contains the spin s 2 Weyl tensors, i.e., the curvatures which are non-vanishing on-shell. We
suggest that the hs(8 ) gauge theory describes the minimal bosonic, massless truncation of M-theory
on AdS7 S 4 in an unbroken phase where the holographic dual is given by N free (2, 0) tensor
multiplets for large N. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
A higher spin (HS) gauge theory is a general covariant field theory with an additional
infinite set of local symmetries. The full set of symmetries is based on a rigid HS algebra,
which typically extends the spacetime (super) AdS group. The first step in the construction
of a HS gauge theory is therefore to identify this algebra and find its representations.
Extensive results along these lines already exist in the literature for D 5 [18]. In this
E-mail address: sezgin@physics.tamu.edu (E. Sezgin).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 9 9 - 7
E. Sezgin, P. Sundell / Nuclear Physics B 634 (2002) 120140
121
paper we construct the minimal bosonic HS algebra in D = 7 and its linearized massless
field equations. Altogether these results show that the HS theories in diverse dimensions
have many features in common, which points to some underlying universal principle.
As will be discussed in more detail in [9], HS gauge theories in diverse dimensions that
include massive higher spin fields, are the anti-holographic duals of free conformal field
theories in the limit of large number of free fields [6,1012]. HS field equations do not
seem to admit any truncation to ordinary (super)gravity, because lower spin fields appear
as sources of HS fields. This is consistent with the fact that the stress-energy tensor of a
free CFT does not form a closed OPE algebra, as has been illustrated in the case of the 3D
supersingleton [13] (though the conserved charges of course form the finite-dimensional
conformal group which closes). Thus, if one wishes to give an anti-holographic description
of a renormalization group flow in the vicinity of a (free) CFT with conserved HS currents
one should use HS gauge theory instead of ordinary (super)gravity. In particular, the HS
theory discussed in this paper is the minimal bosonic truncation of a supersymmetric HS
theory which is proposed in [9] to contain solutions describing the flow from the strongly
coupled AN1 fixed point [14] to the free theory of N (2, 0) tensors.
Our construction of hs(8 ) is given in terms of a basic Grassmann even oscillator y
which is a Dirac spinor of SO(6, 1). The generators of hs(8 ) are traceless monomials in the
oscillators. The trace is taken with the symmetric charge conjugation matrix. Importantly,
the tracelessness condition is imposed as a coset condition, namely, hs(8 ) = L/I, where
L is a certain Lie algebra which contains both traceless generators and generators with
non-zero trace, and I L is an ideal containing all the generators with non-zero trace.
In fact, the above construction is similar to the one of hs(2, 2) in D = 5 [6,7], which is
the minimal bosonic HS extension of SO(4, 2). The ideal I occurring in that case contains
the trace K yy.
The spectrum of the 5D theory is given by the product of two scalar
doubletons, which have vanishing K charge. This yields massless fields in D = 5 with
spin s = 0, 2, 4, . . . . The spin s = 0 field and the on-shell curvatures (Weyl tensors)
are collectively described in HS theory by a master field which is a zero-form. As a
consequence of modding out I the master field is dressed up by K 2 -expansions which
are crucial for getting the correct AdS lowest weights [6,7].
K yy and K+ y y,
In D = 7, the ideal I is generated by the traces, K3 yy,

which form an SU(2)K algebra (note that the charge conjugation matrix is symmetric in
D = 7 whereas it is anti-symmetric in D = 5). Massless fields in AdS7 arise in tensor
products of two 6D conformal tensors known as doubletons [15].1 The spectrum of the
theory is given by the product of two scalar doubletons. The scalar doubleton is the only
one with vanishing SU(2)K charge (in the (2, 0) superextension of the theory this scalar is
one of the five scalars in the tensor multiplet). As in five dimensions, this yields massless
fields of spin s = 0, 2, 4, . . . . We will show that the gauging of hs(8 ) in D = 7 yields
a matching set of gauge fields. In particular, we verify that for each spin s the algebra
generators are in one-to-one correspondence with the previously known set of physical
1 In the N = 2 superextension of the 7D theory the superdoubleton squares belong to an isolated series
[16,17]. On the other hand, the massless fields of the D = 5, N = 4 higher spin theory [8], which arise in
the square of a certain superdoubleton, saturate the bound of a continuous series [16,17]. Thus the 5D theory can
be higgsed in a continuous fashion [9].
122
and auxiliary spin s gauge fields given in terms of a particular set of spin s Lorentz tensors
[18,19]. Moreover, modding out the ideal I the 7D master scalar field is dressed up with
K 2 -expansions, where now K 2 = KI KI . It is gratifying that these expansions indeed lead
to the correct critical mass-terms in the linearized equations for and the Weyl tensors
such that the AdS energies assume the correct massless values.
This paper is organized as follows. In Section 2 we describe the doubleton representations of the AdS7 group and compute explicitly the massless 7D field content from the
product of two scalar doubletons. In Section 3 we construct the bosonic HS algebra hs(8 ).
In Section 4 we gauge hs(8 ) by introducing a master gauge field in the adjoint representation and a master scalar field in a quasi-adjoint representation of hs(8 ), and in particular
we compute the K 2 -dressing. In Section 5 we write the linearized curvature constraints and
show that they yield the correct spectrum of massless fields. Our results are summarized
and further discussed in Section 6. Appendix A contains the details of the calculation of the
mass terms occurring in the linearized field equations. Appendix B contains the harmonic
analysis on AdS7 S 4 needed in the computation of the AdS energies.
2. SO(6, 2) representation theory

Physical representations of SO(6, 2) can be obtained from a set of Grassmann even
oscillator y ( = 1, . . . , 8), which is a SO(6, 1) Dirac spinor, obeying the oscillator
algebra2
y y y y = 2C ,
(2.1)
where denotes the product of the oscillator algebra. We also define a Weyl ordered
product as follows:
y y = y ya = y y C .
(2.2)
The Weyl ordered product extends in a straightforward fashion to contraction rules between
arbitrary polynomials of the oscillators similar to those used in expanding products of Dirac
matrices. For example,
(y y ) (y y ) = y y y y + C y y C y y C C .
(2.3)
The SO(6, 2) generators are given by (A = a, 8)

i
i
a8 = a ,
ab = ab ,
2
2
[MAB , MCD ] = iBC MAD + 3 more,
(MAB ) = MAB .
MAB = 12 y AB y = 12 y
AB y,
(2.4)
(2.5)
The spin group Spin(6, 2) is a non-compact version of SO(8; C), which we denote by
SO(8 ). The 7D spin group Spin(6, 1) is a subgroup of SO(8 ).
2 We use mostly positive signature and Dirac matrices ( a ) obeying { a , b } = 2ab (a =
0, 1, 2, 3, 4, 5, 7). The charge conjugation matrix C is symmetric and real and ( a C) are anti-symmetric.
The Dirac conjugate of y is defined by y = (y i 0 C) .
123
The maximal compact subgroup SO(6, 2) is given by L0 = SO(6)U (1)E where SO(6)
is the group of spatial rotations and U (1)E is generated by the AdS energy E = M08 .
There is a three-grading SO(6, 2) = L L0 L+ , where L contain non-compact
energy-lowering and energy-raising operators M such that [E, M ] = M . The threegrading also requires [L+ , L+ ] = [L , L ] = 0, [L+ , L ] = L0 and [L , L0 ] = L . The
reality conditions are (L0 ) = L0 and (L+ ) = L . Positive energy representations of
SO(6, 2) are weight spaces D(E0 ; n1 , n2 , n3 ) labeled by the lowest energy E0 and SO(6)
highest weight labels (n1 , n2 , n3 ) obeying n1 n2 |n3 |. In case of integer spin these
determine an SO(6) Young tableaux with n1 boxes in the first row, n2 boxes in the second
row and |n3 | boxes in the third row. If n3 = 0 the Young tableaux can be self-dual or
anti-self-dual, which corresponds to n3 > 0 and n3 < 0, respectively. The SO(6) Young
tableaux can be converted into an SU(4) Young tableaux by contracting it with SO(6)
Dirac matrices. A column with N boxes (N = 1, 2, 3) stacked on top of each other is
contracted with the rank N Dirac matrix. Their symmetry properties are as follows: ( R )ij
(R = 1, . . . , 6, i, j = 1, . . . , 4) is anti-symmetric, ( RST )ij and ( RST )ij are symmetric
In the case of
and with definite self-duality properties, and ( RS )i j belongs to 4 4.
half-integer spin all three highest weight labels are half-integers determining a -traceless
tensor-spinor whose tensor structure is determined as above by the highest weight labels
(n1 12 , n2 12 , n3 ((n3 ) 12 ), where ((n3 ) is the sign of n3 . For example, (1, 0, 0)
denotes the (real) 6-plet, (1, 1, 1) denotes the self-dual and anti-self-dual (complex) 10plets, ( 12 , 12 , 12 ) denotes the chiral SU(4) spinors and ( 32 , 12 , 12 ) has the (1, 0, 0) tensor
structure and denotes a chiral and -traceless vector-spinor.
Our next aim is to describe the UIRs of SO(6, 2) which arise in the oscillator
construction [15]. A convenient choice of Dirac matrices is

0 1
i 0
0
C=
,
=
(2.6)
.
1 0
0 i
The Dirac spinor oscillator y obeying (2.1) then split into two sets of U (4)-covariant
oscillators (i = 1, . . . , 4):

y = 2 a i , bj ,
(2.7)
(bi ) = bi ,
(ai ) = a i ,

j
j
j
j
a i , a = i ,
(2.8)
b i , b = i .
The U (4)-covariant oscillators have a unitary representation in the Fock space built on the
vacuum state |0 defined by
ai |0 = bi |0 = 0.
(2.9)
The identification (2.7) implies that the Fock space is a unitary representation space of
SO(6, 2). Other unitary representations of SO(6, 2) can then be obtained by considering
tensor products of several copies of the Fock space.
The Fock space and its tensor products decompose into physical UIRs of SO(6, 2).
The three-grading L L0 L+ takes the following form in the (single) Fock space
representation:
L+ :
Lij = a [i b j ] ,
(2.10)
124
L0 :
L :
Lij = a i aj + bi bj + 4ji ,
(2.11)
Lij = a[i bj ] .
(2.12)
Lij ,
is bounded from below and contains a

The energy operator, which is the trace-part of
constant contribution when it is written in normal-ordered form:

1
1
1
i
E = y 0 y = y y = a i ai + bi bi + 2 = (Na + Nb ) + 2.
(2.13)
4
4
2
2
Since L0 = U (4) SU(4) U (1)E where SU(4) is the diagonal sum of SU(4)a and
SU(4)b , it is possible to form lowest weight states (lws) carrying the same L0 weight
by exchanging a i and bi oscillators. To be more precise, the lws carry an extra label of
the SU(2)K group under which (a i , bi ), i = 1, . . . , 4, transform as doublets. In fact, the
SU(2)K is generated by
i
K+ = y y = ibi ai ,
4
i
K = yy = ia i bi ,
4
1
1
1
= y y + 2 = (Nb Na ).
K3 = yy
4
4
2
For computational purposes it is convenient to write (I = 1, 2, 3)
(2.14)
(2.15)
(2.16)
1
KI = y A (I )AB y B ,
(2.17)
yA = y B BA = (y , y ),
8
where (I )AB are symmetric Pauli matrices tensored with the SO(8 ) charge conjugation
matrix and yA obeys

0
C
.
AB =
yA yB = yA yB + AB ,
(2.18)
C
0
By construction [SO(6, 2), SU(2)K ] = 0. Thus the Fock space and its tensor products
decompose into (2j + 1)-plets of identical SO(6, 2) weight spaces, which we denote by
D (j ) (E0 ; n1 , n2 , n3 ).
The Fock space F of a single set of oscillators decomposes into SO(6, 2) SU(2)K
weight spaces, known as doubletons, as follows [15]

D (s) (E0 = s + 2; s, s, s).
F=
(2.19)
s=0, 21 ,1, 32 ,...
The lws with K3 -eigenvalue m is given by

a (i1 a i2s2m bi2s2m+1 bi2s ) |0.
(2.20)
The doubletons describe off-shell conformal tensors in six dimensions3 with spin s and
scaling dimension E0 . In particular, the SU(2)K -singlet describes a scalar field with scaling
dimension 2.
3 The compact SO(6) labels are related to the non-compact SO(5, 1) labels, i.e., the six-dimensional spin
labels, and the energy E0 to the six-dimensional scaling dimension by a non-unitary rotation [20].
125
The P -fold tensor product F P of the oscillator Fock space F can be described by
adding a flavor-index (r, s = 1, . . . , N ):

j
j
bi (r), bj (s) = rs i .
ai (r), a j (s) = rs i ,
(2.21)
The representation of SO(6, 2) SU(2)K on the tensor product is given by

Lij =
Lij (r),
Li j =
Li j (r),
Lij =
Lij (r),
r
KI =
(2.22)
KI (r).
(2.23)
In particular, the total energy operator is given by

E = 2P +

1
Na (r) + Nb (r) .
2 r
(2.24)
Thus the tensor product decomposes into weight spaces with lowest energy E0 s + 2P
(the lws with E0 > s + 2P are obtained by anti-symmetrizing oscillators carrying different
flavor indices). In particular, P = 2 yields massless fields in AdS7 [15].4
3. The massless spectrum of the theory

As discussed in the Introduction, the minimal bosonic 7D HS theory is conjectured to
be the anti-holographic dual of the 6d theory of N free scalar fields i , i.e., N copies
of the scalar doubleton, in the limit of large N [9,12]. The composite operators of this
theory couple to (non-normalizable) bulk modes of the HS theory. In particular, the bilinear
operators are the spin s = 0 operator i i and a set of conserved, symmetric and traceless
tensors of spin s = 2, 4, 6, . . . [24]. These tensors are in one-to-one correspondence with
the massless representations in the symmetric tensor product of two 6D scalar doubletons.
The anti-symmetric tensor product contains massless representations which correspond to
descendants. The massless spectrum of our 7D theory is, therefore, given by

S = D(2; 0, 0, 0) D(2; 0, 0, 0) S .
(3.1)
In order to decompose D(2; 0, 0, 0) D(2; 0, 0, 0) under SO(6, 2) we compute the
lws with energy E0 = 4 + s for s = 0, 1, 2, . . . . To this end, we expand a general state
| D(2; 0, 0, 0) D(2; 0, 0, 0) with that energy as follows:
s
s

()

| =
() i
=
1 j1 ,...,i j ;k1 l1 ,...,ks ls
=0
=0
Li1 j1 (1) Li j (1)Lk1 l1 (2) Lks ls (2)|0.
(3.2)
4 These fields also satisfy the masslessness criteria defined in [21]. In [21] certain UIRs of SO(6, 2) that cannot
be obtained by multiplying any number of 6D doubletons. However, the unitarity bounds of Osp(8 |4) seem to
exclude this possibility [22,23].
126
The quantities Li1 j1 (1) Li j (1) and Lk1 l1 (2) Lks ls (2) are irreducible under
SU(4). Their SO(6) highest weight labels are (, 0, 0) and (s , 0, 0), respectively. Thus
the SU(4) tensors (0) and (n) are irreducible, and given by the SU(4) Young tableaux
with two rows of length s. This SU(4) irrep, which we shall denote by Rs , has spin given
by s and can be converted to a real, symmetric rank s SO(6) tensor by contracting it with
s (anti-symmetric) SO(6) Dirac matrices. Its SO(6) highest weight labels are (s, 0, 0). The
remaining () , 0 < < s, are reducible and decompose into Rs plus a set of various
other irreps, {R () } say (where each irrep occur once and only once), which we write as

() = () (Rs ) +
(3.3)
() R () , 0 < < s.
R ()
The state | is a lws provided that

s1

()

= 0,
Lij | Lij (1) + Lij (2) |
(3.4)
=0
where | () contains factors of L+ (1) and s 1 factors of L+ (2). Thus

()

= 0,
= 0, . . . , s 1.
| ()
()
(3.5)
(+1) .
The state
is a linear combination of contributions from
and
From
| (0) = 0 and the fact that the contributions from the various irreps are linearly
independent it follows that

(1) R (1) = 0.
(1) (Rs ) = s 2 (0) ,
(3.6)
From | (1) = 0 it then follows that
2

(s 1)2 (1)
s
(Rs ) =
(Rs ) =
(3.7)
(0) ,
(2) R (2) = 0.
4
2
Iterating this procedure we find that there is precisely one lws with energy E0 = 4 + s,
which belongs to the SU(4) irrep Rs described above. Collecting the above results, we find
the following explicit expression for this state:
2
s

s
(1)
Li1 j1 (1) Li j (1)Li+1 j+1 (2) Lis js (2)|0,
|E0 ; s, 0, 0 =
=0
(3.8)
where separate symmetrization of the i and j indices is assumed. The lws with even spin
belong to the symmetric tensor product and those with odd spin to the anti-symmetric
product. In summary:

D(E0 = s + 4; s, 0, 0),
D(2; 0, 0, 0) D(2; 0, 0, 0) S =
(3.9)
(2)
D(2; 0, 0, 0) D(2; 0, 0, 0)

A
s=0,2,4,...
D(E0 = s + 4; s, 0, 0).
(3.10)
s=1,3,5,...
We have thus computed the massless spectrum of our minimal bosonic HS theory, which
hence consists of massless states with spin s = 0, 2, 4, . . . and vanishing SU(2)K charge.
127
4. The higher spin algebra hs(8 )

Our next task is to determine the HS symmetry algebra, hs(8 ). From the boundary point
of view, hs(8 ) is the algebra of charges of the set of conserved currents built from the s 2
symmetric traceless tensors [24]. These rigid symmetries of the free CFT induce local
symmetries in the anti-holographic bulk theory, including general covariance. However,
instead of determining hs(8 ) from the current algebra we construct it directly in terms
of the oscillators. We do this by making use of the properties of the spectrum derived in
the previous section, and the knowledge of which gauge fields are required for writing the
covariant constraints for a massless field of given spin s 2 in a linearization around AdS7
[18,19]. In fact, the canonical set of spin s gauge fields which contains physical as well as
auxiliary fields is in one-to-one correspondence with the above mentioned set of conserved
currents of that spin [24].
Thus, the methodology we adopt is to impose constraints on general oscillator
expansions. Essentially we impose three types of constraints: (1) we project out all
monomials except those which are of degree 44 + 2, where 4 = 0, 1, 2, . . . is a level index;
(2) we impose neutrality under SU(2)K ; and (3) we mod out an ideal which contains all
the traces.5
We begin by taking A to be the space of arbitrary polynomials f (y, y)
in the oscillators,
which is an associative algebra with a -product defined by the Weyl ordering (2.2). An
element of A can therefore be expanded in terms of Weyl ordered monomials in the basic
oscillator y and its Dirac conjugate y with complex coefficients which are multispinors.
We shall use the following normalization convention (which we give here for a single
monomial):
1 1
y y m y 1 y n f1 ...m ,1 ...n .
m!n!
We next define a linear anti-automorphism of A as follows:

f (y, y)
= f (iy, i y),
f (y, y)
=
(f1 f2 ) = (f2 ) (f1 ).
(4.1)
(4.2)
(4.3)
A linear anti-automorphisms of an associative algebra can be used to define a Lie

subalgebra. In our case we define the following Lie subalgebra of A:

L = P A : (P ) = P = P , [KI , P ] = 0 ,
(4.4)
where KI are the SU(2)K generators defined in (2.17). The Lie bracket of L is
[P1 , P2 ] = P1 P2 P2 P1 .
(4.5)
The generators of L have expansions with multispinor coefficients which in general have
non-zero trace parts. In order to impose tracelessness we define a new ordering of elements
in A which amounts to factoring out the trace parts explicitly [6]. An element P L is thus
5 The last step leads to an algebra which cannot be represented on doubletons that carry non-zero SU(2)
K
charge.
128
expanded as:
P=
I1 ...In
P(n)
(y, y)
KI1 KIn ,
(4.6)
n=0
I1 ...In
(y, y)
has an expansion in terms of traceless, Weyl ordered multispinors and
where P(n)
I1 ...In
the SU(2)K indices I1 . . . In are symmetric. The conditions on P(n)
imply that
in (I1
( P(n) I2 ...In )K .
2 IK
The expansion (4.6) leads to the following decomposition of L:
I1 ...In
I1 ...In
= KI P(n)
KI P(n)
L = L(0) L(1) L(2) = L(0) + I,
(4.7)
I = L(1) L(2) ,
(4.8)
where L(n) represents the nth term in (4.6). If we were to change the prescription for
ordering the SU(2)K generators in (4.6) this would only affect the multispinors in I. Thus
the traceless multispinors in L(0) are uniquely defined by (4.6).
The space L(0) of traceless generators decomposes into levels labeled by 4 = 0, 1, 2, . . .
consisting of elements of the form
P=
1
P ... , ... y 1 y n y 1 y n ,
(n!)2 1 n 1 n
n = 24 + 1,
(4.9)
where the multispinor P1 ...n ,1 ...n belongs to an (irreducible) Young tableaux of

Spin(6, 2) SO(8 ) with two rows of equal length n. In order to see this we first note
that from the condition
[K3 , P ] = 0,
(4.10)
it follows that P must have an equal number of y and y oscillators. The conditions
[K , P ] = 0
(4.11)
then set to zero the SO(8 ) irreps in P described by Young tableaux which have more
boxes in the first row than in the second row.
The reality condition P = P implies that the real dimension of (4.9) equals that of
the corresponding SO(8) Young tableaux. The Young tableaux is reducible under the 7D
spin group Spin(6, 1) SO(8 ). It decomposes into irreps labeled by (tensorial) SO(6, 1)
Young tableaux with two rows of which the first one has n boxes and the second one m
boxes, where 0 m n. The counting works out simply because SO(8 ) decomposes
under Spin(6, 1) in the same way as SO(6, 2) decomposes under SO(6, 1). The conversion
from the SO(8 ) irrep to the set of SO(6, 1) irreps can be explicitly done by contracting the
SO(8 ) Young tableaux with m second rank Dirac matrices ( ab ) and n first rank Dirac
matrices ( c ) , where the SO(6, 1) indices belong to the SO(6, 1) Young tableaux. We
express the resulting generators as:
1
(m,nm)
P
y 1 y n y 1 y n
(n!)2 1 ...n ,1 ...n
am c1 ...cnm b1
= Pba11 ...
Ma1 Mabmm Pc1 Pcnm ,
... bm
n = 24 + 1,
0 m n.
(4.12)
129
Upon gauging, these generators lead to the canonical set of gauge fields with spin s =
24 + 2 (4 = 0, 1, 2, . . .) as discussed earlier. In particular, the zeroth level consists of the
SO(6, 2) generators Pa and Mab corresponding to the gravitational gauge fields.
The space I defined in (4.8) forms an ideal
[L, I ] I.
(4.13)
Moreover, there is a redundancy such that content of L(0) is reproduced in the SU(2)K
trace part of L(n) for n = 2, 4, . . . . For example, the generators P I J KI KJ give rise
to one copy of L(0) , and so on. We are therefore led to defining the HS algebra by

hs 8 = L/I.
(4.14)
The elements of hs(8 ) are thus equivalence classes [P ] of elements in L defined by

[P ] = P L | P P I .
(4.15)
The Lie bracket of [P1 ] and [P2 ] is given by

[P1 ], [P2 ] = [P1 , P2 ] .
In order to examine the representation theory of the HS algebra
following grading of hs(8 ):
(n)

hs 8 =
L ,
E, L(n) = nL(n) ,

nZ
L(n) , L(m)
= L(m+n) ,
(n)
L
= L(n) .
(4.16)
hs(8 )
we observe the
(4.17)
(4.18)
Moreover, from (4.12) it follows that hs(8 ) is a subalgebra of the universal enveloping
algebra Env(SO(6, 2)):

hs 8 = Env(SO(6, 2))/R,
(4.19)
where R is an ideal generated by various polynomials in Env(SO(6, 2)) which vanish by
Fierz identities that arise when the single oscillator realization of SO(6, 2) is used. By a
choice of ordering we can thus write
p q r
L
: p r = n;
L(n) = L+ L0

p + q + r = 24 + 1, 4 = 0, 1, 2, . . . /R,
(4.20)
where L L0 L+ is the three-grading of SO(6, 2), 4 is the level index and we are using
the non-commutative -product. From (4.20) it follows that L(0) L0 + (L0 )3 + when
acting on an SO(6, 2) lowest weight state. A physical representation of hs(8 ) is thus a
0 ; m1 , m2 , m3 ) based on an SO(6, 2) lowest weight state
lowest weight representation D(E
|E0 ; m1 , m2 , m3 obeying the additional conditions
L(n) | = 0,
n = 1, 2, 3, . . . .
(4.21)
Since the single oscillator vacuum state |0 obeys (4.21), the scalar doubleton, which is the
fundamental SO(6, 2) UIR, is also the fundamental UIR of hs(8 )
0, 0, 0) = D(2; 0, 0, 0).
D(2;
130
Tensor products of the scalar doubleton also form hs(8 ) representations. If P hs(8 )
N is given by
then the
representation of P on the N -fold tensor product (D(2; 0, 0, 0))
P= N
P
,
where
P
acts
on
the
rth
factor
in
the
tensor
product.
Thus,
in particular,
r
r=1 r
the massless spectrum S given in (3.1) is an hs(8 ) representation. The SO(6, 2) lowest
weight state |4; 0, 0, 0 = |0 S is an hs(8 ) lowest weight. There are no other hs(8 )
lws in S. To see this we first note that any hs(8 ) lws must also be an SO(6, 2) lws. It
thus suffices to show that for each lws in (3.8) with s = 2, 4, 6, . . . there exists at least one
energy-lowering operator in hs(8 ) which does not annihilate this state. To exhibit such an
operator we first write (3.8) in the following schematic form (s = 2, 4, 6, . . .):
s
s + s1 +
+ L1
L2 + + L+
|0.
|s + 2; s, 0, 0 = L+
(4.22)
1
2
Acting on |s + 2; s, 0, 0 with the element L+ (L )s L(s+1) (the level of this state is
given by 4 = s/2) yields
+ s

s
s
L L
|s + 2; s, 0, 0 = L+
|
+ L+
1 L1
2 L2
+
+
= L1 + L2 |0 = 0.
(4.23)
It follows that the massless spectrum S defined in (3.1) is an irreducible hs(8 ) multiplet:
0, 0, 0).
S = D(4;
(4.24)
5. Gauging hs(8 )
In order to gauge hs(8 ) we introduce an hs(8 )-valued one-form [A] and a zero-form
obeying the conditions:
(A) = A = A,
() = = (),
[KI , A] = 0,
KI = KI = 0.
(5.1)
(5.2)
Here is the anti-automorphism defined in (4.2), (4.3) and is an automorphism acting

on SU(2)K -invariant elements f A as follows:

f (m,n) = (1)n f (m,n) ,
(5.3)
(f1 f2 ) = (f1 ) (f2 ),
where m and n are related to the SO(6, 1) highest weight labels as:
m1 = m + n,
m2 = m,
m3 = 0.
(5.4)
Here we have used the fact that any SU(2)K invariant element f A can be expanded
in terms of such SO(6, 1) irreps as explained in the previous section; see the analysis
following (4.9). In order to show that is an automorphism, we make use of the fact
that if f1 , f2 A are SU(2)K invariant then also f1 f2 is SU(2)K invariant. Thus f1 f2
can also be expanded in terms of SO(6, 1) Young tableaux with m3 = 0. Using the basic
rules for SO(6, 1) tensor products one can then verify that is an automorphism.
The conditions on defines a representation of hs(8 ) which we call quasi-adjoint. It
is essential to introduce this representation to accommodate the physical scalar field and
the spin s 2 Weyl tensors as well as all the derivatives of these fields. The curvature and
131
covariant derivative are defined by

F[A] = [FA ] = [dA + A A], D[A] = d + A (A),

[(] = ( (().
[(] [A] = d( + [A, (] ,
(5.5)
(5.6)
The conditions on implies that D[A] and [(] are also quasi-adjoint elements and
that they do not depend on the choice of A and (. The role of the automorphism is to
distinguish between the translations-like and rotations-like generators in the HS algebra;
for example,6
(Pa ) = Pa ,
(Mab ) = Mab .
(5.7)
Hence, if denotes an SO(3, 2) valued connection, we find that D = +

dx {P , } , where is the Lorentz covariant derivative. As will be shown in the next
section, this means that the whole one-form D can be constrained without having the
consequences of setting to constant.
We next solve the conditions on . By subtracting the two KI -conditions in (5.2) we
obtain [KI , ] = 0. Thus we can choose to expand as follows:
=

n=0
(n) f(n) =

1 I1 ...In
(n) (y, y)K
I1 KIn f(n) K 2 ,
n!
(5.8)
n=0
I1 ...In
where (n)
(y, y)
consists of traceless multispinors which are also traceless in their
internal SU(2)K indices, i.e.,
in
I1 ...In
I1 ...In
KI (n)
(5.9)
= KI (n)
(I(IK1 (n) I2 ...In )K ,
2
J KI ...I
J K (n) 1 n2 = 0, n > 1.
(5.10)
The quantities f(n) (K 2 ) are analytical functions of K 2 = KI KI , which we determine

below. Note that had we chosen to use -products of SU(2)K generators in (5.8), instead
of classical products, the KI -condition would only admit the trivial solution = 0.
It thus remains to impose KI = 0. After some algebra, where it is convenient to use
(2.17) and (2.18), we find

(n)
f(n) ,
KI ((n) f(n) ) = KI (n) D[f(n) ] 4 |(n) | + 4
(5.11)
KI
where D is the second order differential operator

1
1
(z) zf(n)
(z),
D f(n) (z) = f(n) (z) 7 + |(n) | f(n)
(5.12)
8
4
6 For D = 4, 5, the map can also be defined as a linear transformation of the oscillators. This does not seem
to be possible in D = 7, since (a C) and (ab C) have the same symmetry. It is worth noting that there
is an alternative way of implementing the map by extending the oscillator algebra A with an inner Kleinian
= A (A) has the map (f (y, y;
) = f (y, y;
),
operator = (1)2K3 . The associative algebra A
ab = Mab . More generally, hs(8 ) is generated by
a = Pa and M
and the AdS generators can be taken to be P
to obey (5.2) and imposing the curvature constraints
m+ P (m,24+1m) (m = 0, . . . , 24+1) for = 1. Taking
(6.1), (6.2), which are now expansions in , one finds that = 0 gives rise to an entire set of auxiliary gauge
fields. Thus the physical field content remains the same in this framework.
132
and |(n) | denotes the number of irreducible spinor indices in the multispinor defining
(n) :
y A A I1 ...In (y, y)
= |(n) | I1 ...In (y, y).
From KI = 0 it

2

f(n) K 2 = F |(0) |; K ,
0,
(5.13)
follows that7
n = 0,
n > 0,
where the function F (w; z) is given by

(4z)n 12 (w + 7)
F (w; z) =

.
n! 12 (w + 7 + 2n)
(5.14)
(5.15)
n=0
From [KI , (0) ] = 0, and following the analysis below (4.9), it follows that the y and
y expansion of (0) (y, y)
gives rise to traceless multispinors (0)1 ...s ,1 ...s that belong
to spin s SO(8 ) Young tableaux that have two rows of equal length s. Taking into account
also the condition () = () we find (from now on we drop the subscript (0))

(m,n;0) (y, y)F
2s; K 2
(y, y)
=
(5.16)
m = 0,2,4,...
n = 0,1,2,...
(m,n;k) (y, y),
(5.17)
m = 0,2,4,... k=0
n = 0,1,2,...
yields a traceless multispinor (m,n;0)

where the y and y expansion of (m,n;0) (y, y)
1 ...s ,1 ...s
which is equivalent to a spin s SO(6, 1) Young tableaux with s = m + n boxes in the first
row and m boxes in the second row. The multispinor (m,n;k)
(k = 1, 2, 3, . . .)
1 ...s+k ,1 ...s+k
which by definition contain k traces times an SO(6, 1) Young tableaux with s = m + n
boxes in the first row and m boxes in the second row, is given by (5.15) and (5.17). For
example,
= 0,
(m,n;1)
1 ...s+1 ,1 ...s+1
(m,n;2)
1 ...s+2 ,1 ...s+2 =
(s + 2)2 (s + 1)2 (m,n;0)

1 ...s ,1 ...s
2(7 + 2s)
(Cs+1 s+1 Cs+2 s+2 Cs+2 s+2 Cs+1 s+2 ),
(5.18)
where separate symmetrization in the and indices is assumed.

The spin s = 0 sector of is a single real scalar field with the following K 2 -dressing:

8
1 + K2 + ,
(5.19)
7
7 A solution to K = K = 0 can also be constructed formally as = , where is the projector
I
I
of the single oscillator Fock space (2.19) onto the spin s = 0 doubleton, which is the SU(2)K -singlet.
133
where the coefficient 8/7 is read off from (5.15). The spin s = 1, i.e., || = 2, sector
contains an SO(6, 1) vector a . In the spin s = 2 sector, i.e., for || = 4 we find a
symmetric traceless tensor ab and a traceless tensor Cab,cd = Ccd,ab obeying C[ab,c]d = 0.
These tensors match the on-shell second order derivatives of a scalar and a graviton,
respectively. This pattern extends to higher spins such that the tensorial content of is
isomorphic to the derivatives of a spin s = 0 field and a set of spin s = 2, 4, 6, . . . Weyl
tensors (obeying KleinGordon equations and Bianchi identities for s 2).
6. The linearized curvature constraints

In analogy with the 5D case [6], we propose the following linearized curvature
constraints (s = 2, 4, 6, . . .):
= ea eb (ab ) (s,0;0)
Flin
1 ...n ,1 ...n ,
1 ...s1 ,1 ...s1
(6.1)
D = 0,
(6.2)
where is the AdS background

1 ab
a
= i P ea + M ab ,
2
d + = 0,
(6.3)
and F lin is the linearized curvature:

F lin = dA + A + A .
(6.4)
These constraints are invariant under the linearized form of the gauge transformations
(5.6):
( A = d( + [, (],
( = ( (().
(6.5)
= 0. The integrability of
Consistency of the constraints requires integrability, i.e.,
the scalar constraint (6.2) follows immediately from the flatness of . The integrability of
the curvature constraint (6.1) is equivalent to the Bianchi identity dF lin + [, F lin] = 0,
which takes the following form in components:
d2
lin
+
[ F],
1 ...s1 ,1 ...s1
(s 1)2
lin
([ )1 F]
2 ...s1 ,1 ...s1
2

lin
([ )1 F]
1 ...s1 ,2 ...s1 = 0, (6.6)
where separate symmetrization in and indices is assumed and is the Lorentz

covariant derivative. In order to verify this we use the component form of the scalar
constraint (6.2) which reads:
1
n2
1 ...n ,...n + ( ) 1 ...n ,...n + ( )1 1 2 ...n ,2 ...n = 0,
2
2
(6.7)
where again separate symmetrization in and indices is assumed. Inserting the curvature
constraint (6.1) into the Bianchi identity (6.6) and making use of (6.7) we find that (6.6)
134
decomposes into two irreducible parts. In the notation defined in (4.12) these are given by
an (s, 0) part and an (s 1, 1) part (by construction these equations have no trace parts).
These are satisfied due to the following Fierz identities (s 2):
(s,0;0)
(a[b ) (cd] ) (1 ...(s2 ,1 ...s2 = 0,
(6.8)
(s1,1;0)
([a ) (bc] )
(1 ...(s2 ,1 ...s2 = 0.
(6.9)
In order to verify these it is important to use the symmetries of (s,0;0) and (s1,1;0) as
implied by the properties of their SO(6, 1) Young tableaux. To implement these properties
it is convenient to use expansions similar to (A.4).
The scalar constraint (6.2) implies that the independent fields in are given by
(s,0;0) (s = 0, 2, 4, . . .) and that the remaining components are derivatives of these fields;
schematically (s,n;0) n (s,0;0). The constraint also implies the following Klein
Gordon equations (the derivation of the mass-term for arbitrary s is given in Appendix A):

2
m2 = 8 2s.
m2 (s,0;0) = 0,
(6.10)
Using the harmonic analysis in Appendix B, we find the following lowest energy of
(s,0;0):
E0 = s + 4.
(6.11)
As discussed in Section 2, this is the correct value for a massless spin s field. Thus
the independent field content of the quasi-adjoint representation is isomorphic to the
spectrum S in (3.1). We remark that the spectrum S is a massless hs(8 ) multiplet. The
global hs(8 ) transformations on are realized in terms of gauge transformations (5.6)
with rigid parameters ( obeying the Killing equation
D ( = 0.
(6.12)
K 2 -expansion
found in the previous section play a crucial role in obtaining the

The
correct critical mass-value in (6.10). Let us demonstrate this in the case of spin s = 0. We
examine the leading equations in the spin s = 0 sector of (6.2):
1
= ( ) , ,
(6.13)
2
1
1
, = ( ) , ( ) .
(6.14)
2
2
It follows that

1
( ) , + ( ) .
=
(6.15)
4
The second term on the right-hand side contributes to m2 by 14. The multispinor in the
first term can be decomposed into SO(6, 1) irreps as follows:
(2,0;0)
(1,1;0)
(0,2;0)
(0,0;2)
, = ,
+ , + , + , .
(6.16)
The contributions to m2 from the first three terms vanish, due to the Fierz identity

(m,n;0)
( ) ,
m + n = 2.
(6.17)
= 0,
135
(0,0;2)
The SO(6, 1) singlet , yields a non-zero contribution to m2 given by

1
21
( )
(2C C + 2C C 4C C ) = 6.
4
74
As a result we find that the spin s = 0 field obeys

m2 = 0,
m2 = 14 + 6 = 8,
(6.18)
(6.19)
which leads to the lowest weight energy E0 = 4.

The curvature constraint (6.1), when written in SO(6, 1) tensorial basis, is of the
canonical form [18,19]. Thus the gauge fields consist of (s = 2, 4, 6, . . .)
Generalized vielbeins:
A(0,s1)
1 ...s1 ,1 ...s1 ,
Auxiliary gauge fields:
A1 ...s1 ,1 ...s1 , . . . , A 1 ...s1 ,1 ...s1 .
(1,s2)
(6.20)
(s1,0)
(6.21)
The auxiliary gauge fields can be solved for in terms of derivatives of the generalized
vielbeins. The linearized field equations, which are second order equations, are obtained
(1,s2)
(0,s1)
by solving for A
from F
= 0 and substituting into a certain projection of the
(1,s2)
. Upon fixing a Lorentz-like gauge, one finds that the physical fields
constraint on F
carry the representations D(s + 4; s, 0, 0).
In summary, as a result of the constraints (6.1) and (6.2) the independent set of fields in
the theory are

, A a , A abc , A abcde , . . . ,
(6.22)
which obey second order field equations for massless fields carrying the representations in
the spectrum S given in (3.1) and (3.9).
7. Conclusions
We have constructed a minimal bosonic HS extension of the AdS7 group SO(6, 2),
which we call hs(8 ), consisting of generators of spin s = 1, 3, 5, . . . . We have realized
this symmetry in a 7D HS gauge theory with massless fields of spin s = 0, 2, 4, . . . , which
are given in (6.22). The s 2 fields are contained in a hs(8 ) valued one-form and the
spin s = 0 field, the Weyl tensors and their derivatives are assembled in a zero-form
which transforms in a quasi-adjoint representation of hs(8 ). The spectrum of physical
fields form a UIR of hs(8 ) which is isomorphic to the symmetric product of two scalar
doubletons. The gauge fields (6.20), (6.21) and the spin s = 0 field correspond to the full
set of conserved bilinear currents [24] and the quadratic mass term i i , respectively,
of a 6D theory of free scalar fields i (i = 1, . . . , N ) which carry the scalar doubleton
representation.
The internal SU(2)K algebra given in (2.17) plays a key role for embedding hs(8 )
and its quasi-adjoint representation in an associative oscillator algebra. Irreducibility is
achieved by imposing the SU(2)K invariance conditions on the master fields as in (5.1),
(5.2), choosing the ordering prescriptions (4.6) and (5.8) and modding out the ideal I
defined in (4.8). The SU(2)K invariance implies that a 6D realization of hs(8 ) in terms of
136
doubletons must necessarily be given in terms of the scalar doubleton (which is part of the
tensor multiplet in the case of (2, 0) supersymmetry).
The structure of the above 7D HS theory suggests that it describes the massless sector
of a hs(8 ) gauge theory which includes massive fields and whose holographic dual is the
6D scalar field theory in the limit of large N [9,12]. The scalar doubleton theory is the
minimal bosonic truncation of the theory of N free (2, 0) tensor multiplets. We expect
that the latter theory has an anti-holographic dual for large N which is a 7D gauge theory
based on a superextension of hs(8 ) with N = 2 supersymmetry. Once we have computed
the interactions in the 7D theory, these ideas can be tested explicitly by comparing an 6D
n-point functions to the corresponding amplitude of the bulk theory. Since both sides are
weakly coupled when N is large [9,12] this provides an explicit example of a directly
verifiable AdS/CFT correspondence, similar to the one proposed in [6,10,11] for D = 5.
The results obtained this far on HS gauge theories in diverse dimensions point to
underlying universal features. In particular:
They are gauge theories of HS algebras that are infinite-dimensional extensions of the
finite-dimensional AdS group. These algebras are based on oscillator realizations, or
equivalently AdS group enveloping algebras modded by certain ideals.
The massless sector of a HS gauge theory is a UIR of the HS algebra given by
the symmetric tensor product of two ultra-short multiplets, known as singletons or
doubletons, which decomposes into an infinite tower of AdS supermultiplets where
the first level is the supergravity multiplet.
Their massless field content is given by an adjoint gauge field A and a quasi-adjoint
zero-form .
Their background independent field equations follow from a universal set of curvature
constraints.
These properties and other arguments which will be presented elsewhere [9] suggest that
HS gauge theories have holographic duals given by various free, large N conformal field
theories and that finite N corrections are encoded into the bulk theory in a universal
background independent quantization scheme. This would describe an unbroken phase
of type IIB string/M-theory.
Clearly much remains to be done to develop HS gauge theories further. The supersymmetric extension of the linearized 7D HS gauge theory presented here, and the linearized
6D HS gauge theory based on the HS extension of the AdS6 superalgebra F4 should be
straightforward. As for the interactions, they are known fully in D = 4 [1] and some cubic couplings have been computed in the bosonic 5D HS theory [7]. Those in D = 4 are
given in a closed form but considerable amount of work remains to be done to exhibit their
structure explicitly. The construction of the full interactions in D > 4 HS gauge theories
also remains an open problem though we do no expect any fundamental obstacle in achieving this. In testing ideas of higher spin AdS/CFT correspondence, it is also important to
incorporate the massive HS multiplets which arise in higher than second order tensor product of singletons/doubletons [15]. The coupling of massless and massive HS multiplets is
therefore an important open problem.
137
Another open problem is to understand the role played by the U (1)K and SU(2)K
charged higher spin doubletons in d = 4 and d = 6, respectively. They may ultimately
be necessary in the description of an unbroken phase of M-theory. In any event, it seems
likely that their inclusion will lead to a HS conformal field theory in the boundary and
corresponding generalized HS gauge theory in the bulk based on a HS algebra whose
maximal finite-dimensional subalgebra is the symplectic extension of the AdS algebra [26].
Acknowledgements
P.S. is thankful to U. Danielsson and F. Kristiansen for discussions. This research project
has been supported in part by NSF Grant PHY-0070964.
Appendix A. Calculation of mass terms

In order to derive the linearized field equation obeyed by the spin s Weyl tensor
(s,0;0)
1 ...s ,1 ...s we start from the following components of the scalar constraint (6.7):
1
s2
1 ...s ,1 ...s = ( ) 1 ...s ,1 ...s ( )1 1 2 ...s ,2 ...s ,
2
2
1
1 ...s+1 ,1 ...s+1 = ( )( (1 ...s+1 ,1 ...s+1
2
(s + 1)2
(A.1)
( )1 1 2 ...s+1 ,2 ...s+1 .
2
Here and in the remainder of this section we assume separate symmetrization of and
indices. Combining the two equations and restricting to the (s, 0; 0) sector we find:

1
(s,0;0)
(s,0;0)
(s,0;0)
= ( ) ( ) (s),(s)
+ s( ) 1 (s),(s1)
2 (s),(s)
4

1
(s,0;0)
(s,0;2)
+ s( )1 (s1),(s) + ( )( ((s),(s)
4
(s,0;0)
m2 (s),(s),
(A.2)
where (s) = 1 . . . s and idem . The contribution to m2 from the first three terms on
the right-hand side is readily found to be
7
14 s.
2
(A.3)
The calculation of the contribution to m2 from the last term in (A.2) is more elaborate.
We need to use (5.18) and break up the overlapping symmetrizations. In doing so it is
convenient to go over to the SO(6, 1) tensorial basis using

(s,0;0)
...bs
(s),(s)
(A.4)
= a1 b1 as bs ab11...a
,
s
1 1
s s
138
as
where ba11 ...
... bs belongs to the SO(6, 1) Young tableaux with two rows of length s. Making
also use of the Fierz identities

a
(2,0;0)
bd
= (2),(2)
,
b c d ac
(A.5)
1 1
2 2
a
c
(2,0;0)
b d ab cd = 4(2),(2)
(A.6)
,
1 2
1 2
where the right-hand sides are defined as in (A.4) and extra indices on have been
suppressed, we find that the contribution to m2 from the last term in (A.2) is given by

2
3
1
24s + 180s + 336 = 6 + s.
8(7 + 2s)
2
(A.7)
Adding the two contributions in (A.3) and (A.7) we find

m2 = 8 2s.
(A.8)
Appendix B. Harmonic analysis

To determine the SO(6, 2) content of the spectrum, we shall follow the technique used
in [25] which is based on the analytic continuation of AdS7 to S 7 , and consequently
the group SO(6, 2) to SO(8). The Casimir eigenvalues for an SO(6, 2) representation
D(E0 ; n1 , n2 , n3 ), where n1 n2 |n3 | are SO(6) highest weight labels, and an SO(8)
representation with highest weight labels 41 42 43 |44 |, are given by

C2 SO(6, 2) = E0 (E0 6) + n1 (n1 + 4) + n2 (n2 + 2) + n23 ,

C2 SO(8) = 41 (41 + 6) + 42 (42 + 4) + 43 (43 + 2) + 424 .
(B.1)
The continuation from AdS7 to S 7 requires the identification:

41 = E0 ,
42,3,4 = n1,2,3 .
2 AdS 2 S 7 ,
7
(B.2)
A tensor T(m1 m2 m3 ) on S 7 in an irrep R of SO(7) with highest weight labels m1 m2

m3 0 can be expanded as

(41 42 43 44 ) 1
T(m1 m2 m3 ) (x) =
(B.3)
Tp(41 42 43 44 ) D(m
Lx ,
1 m2 m3 ),p
(41 42 43 44 )
p
where Lx is a coset representative of a point x S 7 and (41 42 43 44 ) label all SO(8)

representations satisfying the embedding condition
41 m1 42 m2 43 m3 |44 |.
(B.4)
The Laplacian acting on T yields

(4 4 4 4 )

(4 4 4 4 )
2 S 7 D(m11 m2 23m34),p = C2 SO(8) C2 SO(7) D(m11 m2 23m34),p ,
(B.5)
where the SO(7) Casimir is given by

C2 SO(7) = m1 (m1 + 5) + m2 (m2 + 3) + m3 (m3 + 1).
(B.6)
139
(m,n)
Using the notation introduced in (5.17), the SO(7) highest weight labels of 1 ...2s
(m + n = s) are
m1 = m + n,
m2 = m,
m3 = 0.
(B.7)
The embedding condition (B.4) implies

41 s,
42 = s,
43 = 0, 1, . . . , s,
44 = 0.
(B.8)
The on-shell conditions on (s,0;0), which follow from the (s, 0; 0) part of (A.1), remove
the SO(6, 2) irreps with 43 = 0, 1, . . . , s 1. Recalling (A.8) we thus find that the lowest
(s,0;0)
weight energy of (s),(s) is given by

E0 = 3 + 9 + s(s + 5) + s(s + 3) s(s + 4) 8 2s = 4 + s.
(B.9)
(s,0;0)
Thus the physical field content in (s),(s) is the massless spin s field carrying the
representation D(s + 4; s, 0, 0).
References
[1] M.A. Vasiliev, Consistent equations for interacting gauge fields of all spins in 3 + 1 dimensions, Phys. Lett.
B 243 (1990) 378.
[2] S.E. Konstein, M.A. Vasiliev, Extended higher-spin superalgebras and their massless representations, Nucl.
Phys. B 331 (1990) 475.
[3] E. Sezgin, P. Sundell, Higher spin N = 8 supergravity, JHEP 9811 (1998) 016, hep-th/9805125.
[4] E. Sezgin, P. Sundell, Higher spin N = 8 supergravity in AdS4 , hep-th/9903020.
[5] M.A. Vasiliev, Higher spin gauge theories: star-product and AdS space, hep-th/9910096.
[6] E. Sezgin, P. Sundell, Doubletons and 5D higher spin gauge theory, hep-th/0105001.
[7] M.A. Vasiliev, Cubic interactions of bosonic higher spin gauge fields in AdS5 , hep-th/0106200.
[8] E. Sezgin, P. Sundell, Towards massless higher spin extension of D = 5, N = 8 gauged supergravity,
JHEP 0109 (2001) 025, hep-th/0107186.
[9] E. Sezgin, P. Sundell, to appear.
[10] E. Witten, private communications.
[11] B. Sundborg, Stringy gravity, interacting tensionless strings and massless higher spins, hep-th/0103247.
[12] E. Witten, Talk given at J.H. Schwarz 60th Birthday Conference, Cal Tech, November 23, 2001.
[13] E. Bergshoeff, E. Sezgin, Y. Tanii, Stress-tensor commutators and Schwinger terms in singleton theories,
Int. J. Mod. Phys. A 5 (1990) 3599.
[14] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity,
Phys. Rep. 323 (2000) 183, hep-th/9905111.
[15] M. Gunaydin, S. Takemae, Unitary supermultiplets of OSp(8 |4) and the AdS7 /CFT 6 duality, Nucl. Phys.
B 578 (2000) 405, hep-th/9910110.
[16] B. Eden, S. Ferrara, E. Sokatchev, (2, 0) superconformal OPEs in D = 6, selection rules and nonrenormalization theorems, JHEP 0111 (2001) 020, hep-th/0107084.
[17] S. Ferrara, E. Sokatchev, Universal properties of superconformal OPEs for 1/2 BPS operators in 3 D 6,
hep-th/0110174.
[18] V.E. Lopatin, M.A. Vasiliev, Free massless bosonic fields of arbitrary spin in d-dimensional de Sitter space,
Mod. Phys. Lett. A 3 (1987) 257.
[19] R.R. Metsaev, Arbitrary spin massless bosonic fields in d-dimensional anti-de Sitter space, hep-th/9810231.
[20] S. Fernando, M. Gunaydin, S. Takemae, Supercoherent states of OSp(8 |2N ), conformal superfields and
the AdS7 /CFT 6 duality, hep-th/0106161.
[21] S. Ferrara, C. Fronsdal, Conformal fields in higher dimensions, hep-th/0006009.
140
[22] S. Minwalla, Restrictions imposed by superconformal invariance on quantum field theories, Adv. Theor.
Math. Phys. 2 (1998) 781, hep-th/9712074.
[23] S. Ferrara, E. Sokatchev, Superconformal interpretation of BPS states in AdS geometries, Int. J. Theor.
Phys. 40 (2001) 935, hep-th/0005151.
[24] S.E. Konstein, M.A. Vasiliev, V.N. Zaikin, Conformal higher spin currents in any dimension and AdS/CFT
correspondence, JHEP 0012 (2000) 018, hep-th/0010239.
[25] E. Sezgin, The spectrum of the eleven-dimensional supergravity compactified on the round seven sphere,
Phys. Lett. B 138 (1984) 57.
[26] M.A. Vasiliev, Conformal higher spin symmetries of 4d massless supermultiplets and osp(L, 2M) invariant
equations in generalized (super)space, hep-th/0106149.

Domain wall from gauged d = 4, N = 8

supergravity: part I
Changhyun Ahn, Kyungsung Woo
Department of Physics, Kyungpook National University, Taegu 702-701, South Korea
Received 4 September 2001; accepted 18 April 2002
Abstract
By studying already known extrema of nonsemisimple InonuWigner contraction CSO(p, q)+
and noncompact SO(p, q)+ (p + q = 8) gauged N = 8 supergravity in 4 dimensions developed
by Hull some time ago, one expects there exists nontrivial flow in the 3-dimensional boundary
field theory. We find that these gaugings provide first-order domain-wall solutions from direct
extremization of energy density.
We also consider the most general CSO(p, q, r)+ with p + q + r = 8 gauging of N = 8
supergravity by two successive SL(8, R) transformations of the de WitNicolai theory, that is,
compact SO(8) gauged supergravity. The theory found earlier has local SU(8) CSO(p, q, r)+
gauge symmetry as well as local N = 8 supersymmetry. The gauge group CSO(p, q, r)+ is
spontaneously reduced to its maximal compact subgroup SO(p)+ SO(q)+ U (1)+r(r1)/2 . The
T -tensor we obtain describes a two-parameter family of gauged N = 8 supergravity from which
one can construct A1 and A2 tensors. The effective nontrivial scalar potential can be written as the
difference of positive definite terms. We examine the scalar potential for critical points at which the
expectation value of the scalar field is SO(p)+ SO(q)+ SO(r)+ invariant. It turns out that there
is no new extra critical point. However, we do have flow equations and domain-wall solutions for the
scalar fields are the gradient flow equations of the superpotential that is one of the eigenvalues of A1
tensor. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
One of the interesting issues in recent research is the domain wall/quantum field theory
(DW/QFT) correspondence initiated by [1] between supergravity, in the near horizon
region of the corresponding supergravity brane solution, compactified on domain wall
E-mail addresses: ahn@knu.ac.kr (Ch. Ahn), a0418008@rose0.knu.ac.kr (K. Woo).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 1 3 - 9
142
Ch. Ahn, K. Woo / Nuclear Physics B 634 (2002) 141191
spacetimes that are locally isometric to anti-de-Sitter (AdS) space but different from it
globally and quantum (nonconformal) field theories describing the internal dynamics of
branes and living on the boundary of such spacetimes. DW/QFT correspondence was
motivated by the fact that the AdS metric in horospherical coordinates is a special case
of the domain wall metric [2,3]. R-symmetry of the supersymmetric QFT on the boundary
of domain worldvolume should match with the gauge group of the corresponding gauged
supergravity. Compact gaugings are not the only ones for extended supergravities but there
exists a rich structure of noncompact and nonsemisimple gaugings (note that the unitarity
property is preserved since in all extrema of scalar potential, noncompact gauge symmetry
is reduced to some residual compact subgroup). Such a theory plays a fundamental role
in the description of the DW/QFT correspondence as the maximally compact gauged
supergravity has played in the AdS/conformal field theory (CFT) duality [46] that is a
correspondence between certain compact gauged supergravities and certain conformal field
theories. It would be interesting to identify the appropriate noncompact and nonsemisimple
gauged supergravities corresponding to each choice of brane configuration.
One of the questions we addressed was whether the maximal supergravity theories
with noncompact gauge groups can be obtained from higher-dimensional theory. N = 8
gauged supergravity theories have been constructed in 4 dimensions with gauge groups
SO(p, 8 p) where p = 0, 1, 2, 3, and 4 or with nonsemisimple contractions of these
gauge groups [714]. In 7 dimensions, N = 4 gauged supergravity theories have been
constructed with gauge group SO(p, 5 p) with p = 0, 1, 2 [15]. In five dimensions
there exist gauged N = 8 supergravity theories with gauge groups SO(p, 6 p) with
p = 0, 1, 2, 3 or SU(3, 1) [16]. Although odd-dimensional gauged supergravity theories
do not appear to allow gaugings of nonsemisimple contractions, some researchers have
attempted to resolve the difficulties in five dimensions [17,18]. It has been shown that
the SO(p, q) gaugings and their nonsemisimple contractions can be obtained from the
appropriate higher-dimensional supergravity theories. The spheres used to compactify the
SO(p) gaugings are replaced by hyperboloids for the noncompact SO(p, q) gaugings and
generalized cylinders for the nonsemisimple contractions [14].
Since embedding or consistent truncation of gauged supergravity is known to characterize S7 compactification of eleven-dimensional supergravity,1 we also are interested in
the domain-wall solution in four-dimensional gauged supergravity. In [28], a renormalization group flow from N = 8, SO(8) invariant UV fixed point to N = 2, SU(3) U (1)
invariant IR fixed point was found by studying the de WitNicolai potential which is invariant in the SU(3) U (1) group. For this interpretation it was crucial to know the form
of superpotential that was encoded in the structure of the T -tensor of the theory. Very re1 By generalizing compactification vacuum ansatz to the nonlinear level, solutions of eleven-dimensional
supergravity were obtained directly from the scalar and pseudo-scalar expectation values at various critical points
of the N = 8 supergravity potential [19]. They reproduced all known KaluzaKlein solutions of the elevendimensional supergravity: round S7 [20], SO(7) -invariant, parallelized S7 [21], SO(7)+ -invariant vacuum
[22], SU(4) -invariant vacuum [23], and a supersymmetric one with G2 invariance. Among them, round S7 and G2 -invariant vacua are stable, while SO(7) -invariant ones are known to be unstable [24]. In [25,26] threedimensional conformal field theories were classified by using AdS/CFT correspondence. In particular, researchers
[27] have studied the SU(3) U (1) critical point, from the point of higher-dimensional analysis, which does not
belong to the classification [19] but is a supersymmetric critical point of four-dimensional gauged supergravity.
143
cently, the lift to M-theory of the solution described in [28] was constructed [27] (see also
[29]). Moreover, it was natural and illuminating to ask whether one can construct the most
general superpotential for critical points in four-dimensional N = 8 gauged supergravity:
(1) SU(3)-invariant sectors, (2) SO(5)-invariant sectors and (3) (SO(3) SO(3))-invariant
sector [30]. In order to find and study BPS domain-wall solutions by minimization of the
energy functional, one has to reorganize it into complete squares. Then one should expect
that the scalar potential takes the form of squares of physical quantities. One important
feature of the de WitNicolai d = 4, N = 8 supergravity is that the scalar potential can
be written as the difference of two positive square terms. Together with kinetic terms this
implies that one may construct the energy functional in terms of complete squares.
The other gaugings of N = 8 supergravity could be obtained in the same way the SO(8)
gauging. One could proceed in the same way as the de WitNicolai theory by changing the
supersymmetry transformations and adding them to the Lagrangian. Contrary to N = 4
supergravity in four or seven dimensions, as a result of the complicated nonlinear tensorial
structure, it is necessary to prove that the modified A1 and A2 tensors satisfy a number
of rather involved and lengthy quantities as in [31], to demonstrate the supersymmetry of
the theory. However, in [714], an indirect and simple method which uses some results
known in the de WitNicolai theory was found to generate different gaugings from SO(8)
compact-gauged supergravity theory in such a way that one obtains the full nonlinear
structure automatically and is guaranteed gauge invariance and supersymmetry. The first
step was to construct a real, self-dual anti-symmetric (SO(p)+ SO(q)+ )-invariant fourform tensor using both the generator of SL(8, R) and SO(8) matrices. Next, it was
necessary to describe the projectors that project the SO(8) Lie algebra onto its subalgebras
in terms of a four-form tensor in order to provide a convenient way to deal explicitly with
the SL(8, R) transformation. Then exhaustive manipulations of the invariance of four-form
tensor were crucial for the existence of those gaugings and finiteness of coupling constantdependent, covariant derivative terms as we take the infinity limit of some real parameter.
Then we possess an explicit form for the T -tensor in terms of the standard parametrization
of the scalar coset space.
In this paper, in Section 2, we analyze known vacua of four-dimensional N = 8
noncompact and nonsemisimple gauged supergravity developed by Hull [714]. We claim
no originality for most of the results presented in Sections 2.12.3, although our derivations
are hopefully illuminating. What we will do is to find out
a superpotential from given T -tensor for known gauged supergravity theory and
BPS domain-wall solutions from an energy functional written in terms of complete
squares.
In Section 3, we will consider most general gaugings CSO(p, q, r)+ initiated by Hull
where p + q + r = 8 by using two successive SL(8, R) transformations on the compact
gauged supergravity. What we will do is to find
a T -tensor in SU(8)-basis,
a superpotential and a scalar potential from our findings of T -tensor and
domain-wall solutions.
144
In Section 4, we present our main results. In appendices, we present some details which
are necessary for the calculations in Sections 2 and 3.
2. Domain wall from SO(p)+ SO(q)+ sectors of N = 8 supergravity
Let us consider an ungauged supergravity theory with N local Majorana supersymmetries, 4 N 8 given by CremmerJulia theory [32] who constructed it by dimensionally
reducing 11-dimensional supergravity. Recall that since a Majorana spinor in four dimensions has four real components, the total number of supercharges for the maximal N = 8
theory becomes 32. Note that there is no scalar field in the graviton multiplet for N < 4.
If the maximum spin is to be two, the number N cannot be larger than 8. The scalar fields
lie in a coset space G/H where G is some noncompact group and H its maximal compact
subgroup. Group H is a local symmetry of the whole action while group G is a global
symmetry of the equations of motion only (not the action) because it acts on the spin-1
fields through duality transformations. However, there exists a noncompact subgroup L
of G, which is a global (rigid) symmetry of the action. One can gauge subgroup K of
the global symmetry group L of the action where the dimension of K cannot exceed the
number of vector fields in the model. To gauge the theory, one adds minimal YangMills
couplings for K both to the Lagrangian L0 , which is the Lagrangian of the ungauged theory, and to the supersymmetry transformation rules of the ungauged theory with the vector
fields of the theory acting as gauge connections. One should add coupling constant dependent terms to both the action and supersymmetry transformation laws in such a way that
local supersymmetry is restored and gauge invariance is maintained. Then one obtains a
theory with Lagrangian L = L0 + Lg where Lg consists of minimal gauge couplings with
coupling constant g, fermionic bilinear terms proportional to g, and a scalar potential proportional to g 2 . The minimal couplings and scalar potential break the symmetry G of the
equations of motion and the symmetry L of the action down to K while leaving the local
symmetry H unchanged. Then the gauge theory has both H K local gauge symmetry
and N -extended local supersymmetry.
The ungauged N = 8 supergravity (in this paper, we restrict ourselves to N = 8
theory) has a symmetry G H = E7 global SU(8)local of the equations of motion where
the 28 vectors correspond to a global Abelian symmetry between particles. Motivated
by the fact that realistic theories of fundamental interactions are based on local, nonAbelian symmetries, de Wit and Nicolai [31,33] gauged the subgroup K = SO(8) of (the
L = SL(8, R) subgroup of ) E7 that is a global symmetry of L0 , and obtained a theory
with a local K H = SO(8) SU(8) symmetry. The gauge group K L is a local
symmetry: L L under K while the remainder L\K of the noncompact group L is a
global symmetry of L0 but not of Lg : L L
= L0 + L
g under L\K. In other words,
acting with L\K changes the gauge covariantizations, fermion bilinear terms, and scalar
potential in Lg while keeping the L0 unaffected. This is an invertible field-redefinition for
finite value of t which appears in (1) that leads to an equivalent theory, invariant under the
local supersymmetries and local gauge symmetry.
The contraction procedure involves a sequence of change of basis transformations
depending on the parameter [34]. Although the transformation becomes singular in the
145
zero limit of a parameter, the Lie bracket exists and is well defined in this singular
limit. The original and contracted algebras are not isomorphic. Note that nonsingular
changes of bases can never lead to new algebras because under such a transformation
the new structure constant tensor possesses exactly as much information as the original.
Let us consider a sequence of nonsingular elements E( ) of L with real parameter
and E(1) = 1, and an identity transformation, whose limit point E(0) is singular and not
in L. As long as E( ) remains nonsingular ( = 0), the structure constants have the usual
tensor properties. Acting on the Lagrangian with E( ) yields a sequence of Lagrangian:
L L
( ) = L0 + L
g ( ). If one also rescales the coupling constant g to an -dependent
one through g g
( ) for some choices of the sequence E( ) in L, the limit of L
g ( )
as 0 ( L
g (0)) exists and is well defined (the new structure constants characterize
a Lie algebra) so that L
(0) = L0 + L
g (0) gives the Lagrangian for a gauge-invariant
supersymmetric theory. The gauge group corresponding to L
(0) is not K = SO(8) itself
but an InonuWigner [35] contraction of K denoted by CSO(p, q)+ with p + q = 8
[10,11]. A new (different from the de WitNicolai compact gauged supergravity theory)
gauging, inequivalent to the original one, is obtained by a singular, noninvertible field
redefinition. One can also continue the Lagrangian L
( ) to negative values of . In this
case, L
(1) is the Lagrangian for another gauging and the gauge group is noncompact
SO(p, q)+ with p + q = 8 [10,11].
In Section 2.1, starting with the action of L = SL(8, R) element on the de WitNicolai
theory, we present a superpotential which is an eigenvalue of A1 tensor, that is, a partially
contracted T -tensor. In Section 2.2, with explicit -dependence on the T -tensor, one
recovers more general scalar potetential which will be reduced to the one in Section 2.1
when we put = 0 and obtains a more general superpotential. In Section 2.3, we present
other cases, CSO(p, q)+ and SO(p, q)+ gaugings where p = 6, 5, 4, 3, 2, 1 and q = 8 p
and review their critical points in a scalar potential. In Section 2.4, we construct a domainwall solution from an energy functional. Finally in Section 2.5, as an aside, we will
concentrate on the construction of a scalar potential for the vacuum expectation value
given in terms of real, anti-self-dual (not self-dual), totally anti-symmetric tensor. The
parametrization for this singlet-space is invariant under the SO(p) SO(q) where
p + q = 8.
2.1. Superpotential in CSO(7, 1)+ = ISO(7)+ gauging [7]
Following the procedure we have introduced, the action of the noncompact part of
SL(8, R), L\K, on the theory can be used to other gauged N = 8 supergravity. Let us
consider the acting with the L = SL(8, R) E7(+7) element

0
tX+I J KL
E(t) = exp
(1)
,
tXI+J KL
0
on the de WitNicolai theory where t is a real parameter proportional to ln where
was introduced before and X+I J KL is some real and self-dual totally anti-symmetric tensor
that satisfies
I J KLMNP Q +MNP Q
+

X
.
X+I J KL = X
I J KL =
24
146
Since E(t) is in the real SL(8, R) subgroup of E7(+7) , the ungauged CremmerJulia action
L0 remains unchanged but the g-dependent part Lg is modified nontrivially (changes
the minimal couplings and rotates the A1 and A2 tensors into one another). This gives
one-parameter family of Lagrangian related to the de WitNicolai theory (t = 0 where
E(0) = 1, identity transformation, or equivalently = 1 and E( = 1) = 1) by the
SL(8, R) field-redefinition given by E(t). For all finite values of t, this yields a theory
which is equivalent to the de WitNicolai theory by field-redefinition. However, a different
gauging might be found in the limit t (equivalent to 0) if it exists. For many
choices of the four-form X+I J KL , the limit does not exist. The simplest and special choice
for which this limit exists [7] is2
X+I J KL = Y I J KL +
I J KLMNP Q MNP Q

Y
,
24
(2)
where
Y I J KL =

1 I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
1234 + 1256
.
+ 1278
+ 1375
+ 1368
+ 1458
+ 1467
2
I J KL has 1 when I, J, K and L

Here = +1 for SO(7)+ -invariant X+I J KL and MNP
Q
form an even permutation of M, N, P , Q and 1 when they form an odd permutation of
M, N, P , Q and vanishes. We will come to the case in which = 1 later in Section 2.5
which holds for SO(7) -invariant XI J KL . The four-form tensor X+I J KL is closely
related to the torsion parallelizing seven-sphere S7 [21,24,3638] and invariant under the
SO(7)+ -subgroup of SO(8). Turning on the vacuum expectation value proportional to
X+I J KL in the de WitNicolai theory gives rise to spontaneous symmetry breaking of
SO(8) into SO(7)+ . Regarded as 28 28 matrices, X+I J KL has 21 eigenvalues of 1 and
7 eigenvalues of +3. Introducing the projector P+ into the 21-dimensional eigenspace (P+
projects the generators of SO(8) onto those of SO(7)+ while P projects the generators of
SO(8) into the remainder SO(8)\SO(7)+ ),3 they are given in terms of X+I J KL

3 IJ
1
P+I J KL =
KL X+I J KL ,
4
3
and4
IJ
PI J KL = KL
P+I J KL =

1 KL
+ X+I J KL .
4 IJ
Therefore, one has

X+I J KL = P+I J KL + 3PI J KL.
(3)
2 We emphasize that the way we have chosen X+I J KL here is different from the one in [30] in the sense that
in [30] the SU(2) matrix of SU(8) appears in the last 2 2 block diagonal. However, in this paper, we take it as
the first 2 2 block diagonal matrix. The nonzero component of X+I J KL is either 1/2 or 1/2 as in [7].
3 Note that although the subscript minus sign in P is nothing to do with the anti-self-dual part SO(7) of
SO(8), we will follow the same notation as in the previous literature [7]. In Section 2.5, we take those projectors
as P1 and P2 .

4 I J is defined as I J = 1 I J I J J I + J I = 1 I J I J .
KL
KL
K L
L K
L K
L K
2 K L
2!2! K L
147
One can easily check that the projectors have the following properties5 which will be used
throughout this paper
P2 = P ,
P P = 0.
Here the product P2 is that of 28 28 matrices, (P2 )I J KL = PI J MN PMNKL . The

28 SO(8) generators I J are projected onto a 21-dimensional subspace by P+ , I+J =
P+I J KL KL and this subspace is the Lie algebra for the SO(7)+ -subgroup of SO(8); in
other words, the subgroup stabilizes a right-handed side SO(8) spinor (see Appendix B).
Similarly, the remaining 7 generators are generated by IJ = PI J KL KL .
The change of the minimal couplings under supersymmetry gives a net change of the
action under an infinitesimal local supersymmetry that can be parametrized by a T -tensor
[7]. An expression for the T -tensor, Ti
j kl can be obtained by realizing that a variation of
AIJ leads to a variation of the SU(8)-connection Bi j

Ti
j kl = u kl I J + v klI J MI J KL uim KM u j m LM vimKM v j mLM

+ NI J KLMN vimKL u j m MN uim KL v j mMN ,
(4)
where MI J KL and NI J KLMN are defined in terms of projectors
1
MI J KL = P+I J KL + PI J KL ,
2

1 I J [K L] P QMN
KLMN
= P [P Q] P
P+P QMN .
NI J
2
The supersymmetry of the theory is restored by adding L
g to the ungauged action L0 and
to the supersymmetry transformation rules where A1 , A2 tensors, that appear in L
g , have
a functional dependence on the scalar field but with T
tensor. That is, for example,
A
1 ij =
4
ij m
,
T
21 m
4
A
2l ij k = Tl
[ij k] .
3
(5)
The parametrization for the SO(7)+ -singlet space6 that is an invariant subspace under a
particular SO(7)+ subgroup of SO(8) becomes
I J KL = 4 2 sXI+J KL ,
where s is a real scalar field.
Therefore, 56-beins V(x) can be written as a 56 56 matrix whose elements are the
function of scalar s by exponentiating the vacuum expectation value I J KL . On the other
hand, 28-beins uij KL and vij KL are elements of this V(x). One can explicitly construct
28-beins uij KL and vij KL in terms of scalar s and they are given in Appendix E (E.1).
Now the complete expression for A
1 and A
2 tensors are given in terms of s using (4) and
(5). It turns out from (5) that A
1 tensor has a single real eigenvalue, z1 with degeneracies

5 In terms of X+I J KL , we have the following relation, I J 1 X+I J KL KL + X+I J KL = 0.
3
KL
IJ
6 The 35-dimensional fourth rank self-dual anti-symmetric tensor representation of SO(8) splits into the
SO(7)+ representation 35 27 + 7 + 1 where the singlet 1 is nothing but SO(7)+ -invariant tensor X+I J KL .
148
8 and has the following form

A
1 ij = diag(z1 , z1 , z1 , z1 , z1 , z1 , z1 , z1 ),
7
z1 = es .
8
(6)
Similarly, A
2 tensor can be obtained from the triple product of uij KL and vij KL fields, that
is, from (5). They can be written as
1
A
2i j kl = es X+ij kl .
4
(7)
Finally, the scalar potential together with new A

1 and A
2 tensors can be written, by
combining all the components of A
1 , A
2 tensors, as [7,9]

2
2
1
3
35
V7,1, =0 = g 2 A
1 ij A
2i j kl = g 2 e2s
(8)
4
24
8
which implies that there is no SO(7)+ -invariant critical point of potential by differentiating
this scalar potential with respect to a field s. The eigenvalue z1 provides a superpotential
which will be analyzed in detail in Sections 2.3, 2.4. The scalar potential can be written as

2
V7,1, =0 = g 2 (s z1 )2 6z12 = g 2 4(s z1 )2 6z12 ,
7
where s = 14 s. The theory [7] constitutes a gauging of the 28-dimensional, noncompact

I SO(7)+ symmetry of the CremmerJulia action L0 . The theory has N = 8 local
supersymmetry and H K =0,p=7,q=1 = SU(8) I SO(7)+ local gauge symmetry
where I SO(7)+ is the isometry group of Euclidean 7 space, R7 . In the symmetric
gauge the diagonal SO(7)+ subgroup is manifest. The noncompact gauged N = 8
supergravity theories can be obtained by compactification of 11-dimensional supergravity
on hyperboloids of a constant negative curvature. The contracted version corresponds to a
limit in which the hyperboloid degenerates to an infinite cylinder [14]. Thus, the ISO(7)+
theory corresponds to a compactification on the cylinder S6 R1 that can be replaced by
S6 S1 because the near-horizon limit of the D2-brane is different from that of M2-brane
[1]. As near-horizon limits of the k torus T k reduction of the M2-brane, one expects that
the corresponding theory is CSO(8 k, k)+ gauged N = 8, d = 4 supergravity.
2.2. Superpotential in noncompact SO(7, 1)+ gauging [8]
A suitable one-parameter family (that cannot be more than 28-dimensional) where is
a real parameter of 28-dimensional subgroups of L = SL(8, R) each parametrized by some
real anti-symmetric generator, is generated with nonzero . The commutation relations of
the generators are given in the previous subsection and the only difference is that there
exists another nonzero commutator. Then the 21 + generate SO(7)+ group in which the
seven linearly independent transforms as a 7 representation. When > 0, the algebra
is that of SO(8) and normalization is obtained when = 1. When < 0, one obtains
SO(7, 1)+ , the normalization being obtained when = 1. When = 0, it gives I SO(7)+
as done previously. Then the one parameter family of gauged N = 8 supergravities can be
described by inserting -dependent terms where the T -tensor is given by [8]

Ti
j kl ( ) = Ti j kl (1 ) u kl I J + v klI J

1
jm
PI J KL uim KM u LM vimKM v j mLM
2

+ NI J KLMN vimKL u j m MN uim KL v j mMN .
149
(9)
When = 1, one gets the de WitNicolai model with SU(8) SO(8) gauge symmetry.
When = 0, one has SU(8) I SO(7)+ gauge symmetry. Moreover, when = 1,
a different theory with SU(8) SO(7, 1)+ gauge symmetry was obtained. All the < 0
theories are equivalent to = 1 model related to the SL(8, R) transformation and all
the > 0 theories are equivalent to = 1 de WitNicolai theory related to the SL(8, R)
transformation. Moreover, the = 0 theory was obtained by limiting either > 0 or
< 0 models, under which SO(8) or SO(7, 1)+ is transformed from an InonuWigner
contraction to ISO(7)+ . As we have done before, we can describe 28-beins in terms of s.
It turns out that the A
1 tensor has a single eigenvalue z1 with a multiplicity of 8 which will
provide a superpotential of a scalar potential and has the following form generalizing (6)
A
1 ij = diag(z1 , z1 , z1 , z1 , z1 , z1 , z1 , z1 ),
z1 =

1 s
7e + e7s .
8
(10)
Additionally, we can construct an A

2 tensor generalizing (7) which is the combination of
the triple product of 28 beins, as given in
A
2i j kl =

1 s
e e7s X+ij kl .
4
(11)
Therefore, the scalar potential generalizing (8) in the SO(7)+ -invariant direction by
summing all the components of A
1 and A
2 tensors and counting the degeneracies correctly
is given by [8,9]

2
2

1
3
1
V7,1, = g 2 A
1 ij A
2i j kl = g 2 35e2s 14 e6s + 2 e14s .
4
24
8
This can be written as a superpotential: V7,1, = g 2 [ 27 (s z1 )2 6z12 ] = g 2 [4(s z1 )2 6z12 ]
where s = 14 s. It is easily determined that there are no SO(7)+ -invariant critical

points. The theory [8] has N = 8 local supersymmetry and H K =1,p=7,q=1 =
SU(8) SO(7, 1)+ local gauge symmetry. SO(7, 1)+ gauge symmetry is broken down
to its compact subgroup.
2.3. Superpotential in other CSO(p, q)+ and SO(p, q)+ gaugings [10,11]
Starting from the SO(8) gauging, the ISO(7)+ and SO(7, 1)+ gaugings were obtained
by exploiting the transformations generated by the SO(7)+ -invariant fourth-rank antisymmetric tensor. Now if one uses the (SO(p)+ SO(8p)+ )-invariant fourth-rank tensor
to generate transformations, one expects an SO(p, 8 p)+ gauging and a gauging of a
certain contraction of SO(p, 8 p)+ about its compact subgroup SO(p)+ [10,11]. Let us
150
consider the (SO(p)+ SO(q)+ )-invariant generator of SL(8, R),

0
1pp
Xab =
0
1qq
with
p + q = 0,
p + q = 8,
where 1pp is the p p identity matrix. The embedding of this SL(8, R) in E7 is such that
Xab corresponds to the 56 56 E7 generator which is a noncompact (SO(p)+ SO(q)+ )invariant element of the SL(8, R) subalgebra of E7

0
X+I J KL
,
XI+J KL
0
where the real, self-dual totally anti-symmetric (SO(p)+ SO(q)+ )-invariant four-form
tensor XI+J KL can be written in terms of a symmetric, trace-free, 8 8 matrix with SO(8)
right-handed spinor indices, Xab using SO(8) matrices (see Appendix B)
1
XI+J KL = (I J KL)ab Xab ,
8
(12)
where I J KL = [I J K L] and an arbitrary SO(8) generator LI J acts in the righthanded spinor representation by (LI J I J )ab . When p = 7 and q = 1, this expression of
(12) through matrix coincides exactly with the one in (2). We also present (12) explicitly
in Appendix A for various p and q.
Regarded as a 28 28 matrix, X+I J KL has eigenvalues , and = ( + )/2
with degeneracies d , d and d , respectively. Let it be recalled that SO(7)+ -invariant
four-form tensor has eigenvalues of 1 and +3. The eigenvalues and eigenspaces of the
(SO(p)+ SO(q)+ )-invariant tensor are summarized in Table 1, including the case of
(p, q) = (7, 1). By introducing projectors as done in previous cases, P , P and P onto
corresponding eigenspaces, we have a 28 28 matrix equation that generalizes (3) to
arbitrary p and q
X+I J KL = PI J KL + PI J KL + PI J KL .
Projector P (P ) projects the SO(8) Lie algebra onto its SO(p)+ (SO(q)+ ) subalgebra
while P does onto the remainder SO(8)/(SO(p)+ SO(q)+ ).
Then the -dependent T -tensor [10,11] has a much more complicated expression that
generalizes (9)

Ti
j kl ( ) = Ti j kl (1 ) u kl I J + v klI J

1
jm
PI J KL + PI J KL uim KM u LM vimKM v j mLM
2

I J RS KLMN
vimKL u j m MN + uim KL v j mMN ,
ZRS
+ P
(13)
151
Table 1
Eigenvalues and eigenspaces of the (SO(p)+ SO(q)+ )-invariant tensor, X+ where |X+ |2 = d ||2 + d ||2 +
d | |2 . We have taken this table from [10]. In [39], they displayed the signature of the KillingCartan form by
writing the numbers n+ , n and n0 of its positive, negative and zero eigenvalues. Here we identify d + d with
n+ and d with n
p
= ( + )/2
d = p(p 1)/2
d = q(q 1)/2
d = pq
|X+ |2
7
6
5
4
3
2
1
1
2
3
4
5
6
7
1
1
1
1
1
1
1
7
3
5/3
1
3/5
1/3
1/7
3
1
1/3
0
1/5
1/3
3/7
21
15
10
6
3
1
0
0
1
3
6
10
15
21
7
12
15
16
15
12
7
84
36
20
12
36/5
4
12/7
where we introduce the new quantity ZIMN

J KL
ZIMN
J KL =

1
(P P )I J MP PNP KL PI J MP (P P )NP KL .
2
(14)
The 28-beins uij KL and vij KL are given in Appendix E and the projectors PI J KL
( = , , ) are given in Appendix F. This T
tensor [10] defines new A
1 and A
2 tensors.
These models will have N = 8 local supersymmetry and local SU(8) K,p,q invariance.
The gauge groups are
SO(7, 1)+ ,
SO(6, 2)+ ,
SO(5, 3)+
and SO(4, 4)+ ,
when = 1 (t = i/( )). When = 0 (t = ) there exist the inhomogeneous

groups
CSO(7, 1)+ = ISO(7)+ ,
CSO(3, 5)+ ,
CSO(6, 2)+ ,
CSO(2, 6)+
CSO(5, 3)+ ,
CSO(4, 4)+ ,
and CSO(1, 7)+ .
Any other choice of > 0 ( < 0) gives a model equivalent to the SO(8) (SO(p, q)+ )
gauging by field-redefinition. The gauge symmetry K,p,q is broken down to its maximal
compact subgroup or some subgroup thereof. There are three inequivalent distinct
gaugings. From the expression (13) one gets a single eigenvalue z1 with degeneracies 8
which has the following form
A
1 ij = diag(z1 , z1 , z1 , z1 , z1 , z1 , z1 , z1 ),
z1 =
1 s
p s
pe + q e q
8
(15)
which include all the cases p, q and and generalize (10). Similarly, one can construct A
2
generalizing (11)
A
2i j kl =
q s
p s
e e q X+ij kl .
4
(16)
152
Finally the K,p,q -invariant scalar potential as a function of p, q, and s by counting the
degeneracies correctly can be written as7

2
2
3
1
V = g 2 A
1 ij A
2i j kl
4
24
2

2
1 s
1 q s
pq s
pq s +ij kl
2 3
8
pe + q e
e e
= g
X
4 8
24 4
with
+ 2 1
X = p(p 1)||2 + 1 q(q 1)||2 + pq| |2 .
2
2
The potentials Vp,q, for the K,p,q gauging are given by [11]

1
V7,1, = g 2 35e2s 14 e6s + 2 e14s ,
8

V6,2, = 3g 2 e2s + e2s ,

3
V5,3, = g 2 5e2s + 10 e2s/3 + 2 e10s/3 ,
8

V4,4, = g 2 e2s + 4 + 2 e2s ,

3
V3,5, = g 2 e2s + 10 e2s/5 + 5 2 e6s/5 ,
8

V2,6, = 3g 2 e2s/3 + e2s/3 ,

1
V1,7, = g 2 e2s 14 e6s/7 35 2 e2s/7 .
(17)
8
Of course, the potential V7,1, is identical to the one in Sections 2.1 and 2.2 and is obtained
by putting p = 7 and q = 1 into the general expression of a scalar potential. Note that
for = 1, the potentials for the SO(p, q)+ gauging and the SO(q, p)+ gauging coincide
with each other due to the fact that the potential Vp,q, can be obtained from Vq,p, by
rescaling s ps/q. However, this is not true for = 0 because Vp,q, =0 = Vq,p, =0 .
From the above effective nontrivial scalar potential one expects that the superpotential
W maybe encoded in either the A
1 or A
2 tensors. It turns out that the eigenvalue of the
A
1 tensor z1 provides a superpotential and one can check that the scalar potential can be
written in terms of a superpotential as follows, observed newly in this paper

q
1
1
s
p s
p s
Wp,q ( ; s) = z1 = pes + q e q = pe 2p + q e 2q ,
8
8

2q
2
2
2
s Wp,q ( ; s ) 6Wp,q ( ; s )
Vp,q ( ; s) = g
p

2
(18)
= g 2 4 s Wp,q ( ; s) 6Wp,q ( ; s )2 ,
7 It is known [12] that for finite real t, the T -tensor can be obtained from the old one, de WitNicolai T -tensor
by replacing V with VE(t)1 and scaling by a factor of et : Ti
j kl (V) = et Ti j kl (VE(t)1 ). This can be used
to give a simple calculation of the potential in the (SO(p)+ SO(q)+ )-invariant direction in the space of scalar
field.
153
where s = 2p/q s. The scalar potential has critical points at (1) critical points of
superpotential and at (2) points for which superpotential satisfies some differential
equation. By differentiating W with respect to field s, one finds that there are no
critical points of superpotential corresponding to supersymmetric critical points except
the trivial critical point which has N = 8 supersymmetry and whose cosmological
constant = 6g 2 for which W = 1. The other critical points of scalar potential yield
nonsupersymmetric vacua that may or may not be stable. The superpotential has the
following values at the various critical points.
SO(8) case: N = 8. By differentiating the scalar potential with respect to real scalar
field s, there exists a solution of s = 0 when = 1 for all possible values of p and q. This
is nothing but de WitNicolais SO(8)-invariant critical point and vacuum, which is fully
2 2
supersymmetric
(because in this case, s W |s=0 = 0 implying that V = 6g W . In other
words, |W | = V /6g 2 . All the eight eigenvalues of the A
1 tensor give rise to the number
of supersymmetries) and hence are stable. All the scalar potential Vp,q, becomes 6g 2
when s = 0 for = 1.
SO(7)+ SO(1)+ case: N = 0. This is exactly the SO(7)+ -invariant critical point of
the SO(8) theory. As in Table 2, it has no supersymmetry and is unstable.
SO(5)+ SO(3)+ case: N = 0. In this case, the value of the scalar potential gives a
positive cosmological constant where the eigenvalue of the A
1 tensor is 12 33/8 and
the A
2 tensor has a value of 2 35/8X+ij kl . It is known to be unstable.
SO(4)+ SO(4)+ case: N = 0. At this critical point, the value of the scalar potential
gives a positive cosmological constant where the A
1 tensor vanishes and the A
2 tensor has
a value of 4X+ij kl . It is known to be unstable. The positivity of the cosmological constant
from the analysis of 11-dimensional field equations for SO(5, 3)+ and SO(4, 4)+ theories
was confirmed in [14].
SO(2)+ U (1)+15 case: N = 0. When = 0, the potential vanishes implying that
for any value of s, there exists a zero cosmological constant critical point. In addition,
the potential is also flat in the SO(2)+ SO(6)+ -invariant direction. Nonetheless, global
Table 2
Summary of various critical points [11] in the context of superpotential observed in this paper first: gauge
symmetry, supersymmetry, vacuum expectation value of field, superpotential and cosmological constants. For
SO(3)+ SO(5)+ case, one can check it by the change of variable of SO(5)+ SO(3)+ case, s 3s/5 that
corresponding potential of SO(3)+ SO(5)+ is obtained while by change of variable, s s/7, the potential
of SO(1)+ SO(7)+ can be found from SO(7)+ SO(1)+ case. Although the corresponding superpotential of
these two cases may be different from the original ones, the scalar potentials are the same
Gauge symmetry
q =8p
SO(8)
SO(7)+ SO(1)+
8
0
any
7
any
1
1
1
SO(5)+ SO(3)+
SO(4)+ SO(4)+
SO(3)+ SO(5)+
SO(2)+ U (1)+15
SO(1)+ SO(7)+
0
18 ln 5
0
0
0
0
0
5
4
3
2
1
3
4
5
6
7
1
1
1
0
1
38 ln 3
0
5 ln 3
8
any
7 ln 5
8
W
1
3 51/8
2
12 33/8
1 33/8
2
es /4
3 51/8
2
V
6g 2
2 53/4 g 2
2 31/4 g 2
2g 2
2 31/4 g 2
0
2 53/4 g 2
154
SO(6)+ symmetry remains unbroken by the vacuum. In this case, the eigenvalue of the A
1
tensor is equal to es /4 and the A
2 tensor is 3es X+ij kl .
2.4. Domain wall in CSO(p, q)+ and SO(p, q)+ gaugings [10,11]
Let us begin with the resulting Lagrangian of the scalar-gravity sector by explicitly
determining the scalar kinetic terms appearing in the action in terms of s. The scalar kinetic
1
|A ij kl |2 where the generalized g-dependent A ij kl can be obtained
term is 96

A ij kl = 2 2 u ij I J v klI J v ij I J u kl I J

1
+ 4 2(1 )gAI J PI J KL + PI J KL
2

ij
klLM
ij KM kl
+ v
u LM
u KM v

KLMN ij
u KL u kl MN v ij KL v klMN .
+ PI J RS ZRS
(19)
By taking the product of A I J KL and its complex conjugation and taking into account
the multiplicity with vanishing AI J , we arrive at the following expression for (p, 8 p)
where p = 7, 6, 5, 4, 3, 2, 1
2
1
A I J KL = (7, 3, 5/3, 1, 3/5, 1/3, 1/7) s s.
96
Let us define a new variable s , in order to have usual canonical kinetic terms, normalized
by 1/2, as

2p
s.
s =
q
Therefore, the resulting Lagrangian of scalar-gravity sector takes the form:

1
1
4
d x g R s s Vp,q ( ; s ) ,
2
2
(20)
together with (17) where s replaced by s . Having established the holographic duals of both
supergravity critical points, and examined small perturbations around the corresponding
fixed point field theories, one can proceed the supergravity description. The supergravity
scalar whose vacuum expectation value leads to the new critical point tells us what relevant
operators in the dual field theory would drive a flow to the fixed point in the IR. To construct
the kink corresponding to the supergravity description of the nonconformal (in special case:
RG) flow from one scale to two other connecting critical points in d = 3 field theories, the
form of a 3d Poincar-invariant metric breaking the full conformal group invariance takes
the form [40]
ds 2 = e2A(r) dx dx + e2B(r) dr 2 ,
= (, +, +),
(21)
characteristic of spacetime with a domain wall where r is the coordinate transverse to

the wall (can be interpreted as an energy scale) and A(r) is the scale factor in the fourdimensional metric.
155
Our interest in domain wall spacetimes arises from their relevance to dual field
theories. The distance from horizon U = corresponds to long distance in the bulk (UV
in the dual field theory) and U = 0 (near horizon corresponds to short distances in the
bulk (IR in the dual field theory). We are looking for interpolating solutions. We will
show how supergravity can provide a description of the entire flow from the maximal
supersymmetric UV theory to the IR fixed point. With the above ansatz (21) the equations
of motion for the scalars and the metric from (20) read
3
1
1
r2 A r Ar B + (r A)2 + (r s )2 + e2B Vp,q, = 0,
2
4
2
r2 s + 3r Ar s r Br s e2B s Vp,q, = 0.
(22)
By substituting the domain-wall ansatz (21) into the Lagrangian (20), the energy-density
E[A, s ] [41], with the integration by parts on the term of r2 A, per unit area transverse to
r-direction is given by

E[A, s ] =
dr e
3A+B

3e2B 2(r A)2 + r2 A r Ar B

1 2B
2
e
(r s ) Vp,q, (s ) .
2
We are looking for a nontrivial configuration along r-direction in order to find out the
first-order differential equations satisfying the domain-wall, let us rewrite and reorganize
the energy-density by complete squares plus others due to usual squaring-procedure as
follows:
1
E[A, s ] =
2

2
dr e3A+B 6 eB r A + 2 gWp,q ( ; s )

2
+ eB r s 2 2 gs Wp,q ( ; s )

+ 12 2 geB Wp,q ( ; s )r A + 4 2 geB r Wp,q ( ; s) ,

where superpotential Wp,q ( ; s ) is givenby (18). Then one can easily check the last two
terms in the above can be combined as 4 2 gr (e3A Wp,q ( ; s )). Therefore, one arrives at
1
2

2
dr e3A+B 6 eB r A + 2 gWp,q ( ; s )

2

+ eB r s 2 2 gs Wp,q ( ; s ) + 2 2 g e3A Wp,q ( ; s ) .

Finally, we find BPS bound, inequality of the energy-density

E[A, s ] 2 2 g e3A() Wp,q ( ; s)() e3A() Wp,q ( ; s )() .
(23)
156
Then E[A, s ] is extremized by the following so-called BPS domain-wall solutions. The
first-order differential equations for the scalar field are the gradient flow equations of a
superpotential defined on a restricted slice of the scalar manifold and simply related to the
potential of gauged supergravity on this slice. The equations describing the flow are then
r s = 2 2 eB gs Wp,q ( ; s ),
r A = 2 eB gWp,q ( ; s ).
(24)
There exists a supersymmetry [41,42] of the background with a nonvanishing metric
and a single scalar field, for each spinor satisfying the Killing spinor condition. The
background satisfying (24) preserve half the supersymmetry. It is straightforward to verify
that any solutions s (r), A(r) of (24) satisfy the gravitational and scalar equations of motion
given by the second-order differential equations (22). Embedding or consistent truncation
means that the flow is entirely determined by the equations of motion of supergravity in
four dimensions and any solution of the truncated theory can be lifted to a solution of
untruncated theory [43]. Using (24), the monotonicity [44] of r A which is related to the
local potential energy of the kink leads to r2 A 0 when B is constant. Note that the value
of superpotential at either end of a kink may be thought of as determining the topological
sector. The analytic solutions of (24) for (p, q) = (4, 4) when B is a constant become
(e 2 g(cr) 1)
,
s(r) = 2 log
(e 2 g(cr) + 1)

A(r) = (1 + 2 g)c + log 2 sinh 2 g(r c) ,
where c is some constant. For other values of (p, q), the analytic solutions exist only for
= 0.
2.5. SO(7) invariant sector from SO(8) gauging
The four-form tensor8 XI J KL is invariant under the SO(7) subgroup of SO(8).
Turning on the vacuum expectation value proportional to XI J KL in the de WitNicolai
theory gives rise to spontaneous symmetry breaking SO(8) into SO(7) . Regarded as a
28 28 matrix, XI J KL has 21 eigenvalues of 1 and 7 eigenvalues of 3. Introducing
the projector P1 onto the 21-dimensional eigenspace (P1 projects the generators of SO(8)
onto those of SO(7) while P2 projects the generators of SO(8) onto the remainder
SO(8)\SO(7) ), they are given in terms of XI J KL

3 IJ
1
KL + XI J KL ,
P1I J KL =
4
3

1
IJ
I J KL
P2I J KL = KL
.
P1I J KL = IKL
J X
4
8 The SL(8, R) does act on the vector potential and is generated by the SO(8) and self-dual part. The remainder
of E7 including the anti-self-dual part does not act on the vector potentials but does act on the field strengths.
Therefore, contrary to the self-dual case we have discussed in previous sections, the anti-self-dual case does not
act on the vector potential. We thank C.M. Hull for pointing out this to us.
157
One can easily check that they satisfy

P12 = P1 ,
P22 = P2 ,
P1 P2 = P2 P1 = 0.
The 28 SO(8) generators I J are projected onto a 21-dimensional subspace by P1 ,

I1J = P1I J KL KL and this subspace is the Lie algebra for the SO(7) subgroup of SO(8);
in other words, the subgroup stabilizes a left-handed SO(8) spinor (see Appendix B).
The remaining 7 generators are I2J = P2I J KL KL . The usual commutation relations for
SO(8) are given in terms of I1J and I2J .
Viewed as a 28 28 matrix, XI J KL has eigenvalues , and = ( + )/2 with
degeneracies d , d and d , respectively (for the explicit construction of XI J KL see
Appendix A). The eigenvalues and eigenspaces of the (SO(p) SO(q) )-invariant tensor
are summarized similarly. By introducing projectors as done in previous cases, P , P
and P onto corresponding eigenspaces, we have a 28 28 matrix equation to arbitrary
p and q. The parametrization for the (SO(p) SO(q) )-singlet space that is invariant
subspace under a particular SO(p) SO(q) subgroup of SO(8) becomes
I J KL = 4 2 isXIJ KL,
where s is a real scalar field. Note the presence of imaginary number i. As in the
previous consideration, the A
1 tensor we obtained is a single complex eigenvalue with
degeneracies 8
A
1 ij = diag(z1 , z1 , z1 , z1 , z1 , z1 , z1 , z1 ),
p

1
1
p s
s
z1 = (1 + i) pes + qe q + (1 i) pes + qe q .
(25)
16
16
For the A
2 tensor we get
ps

ps

p
A
2i j kl = (1 + i) e q es + (1 i) e q es Xij kl .
8
Therefore, we are now ready to calculate the full expression of a scalar potential and it
turns out

5

1
V7,1 = g 2 e14s 1 + e4s 1 5e4s + e8s ,
16

V6,2 = 3g 2 e2s 1 + e4s ,

5
3
V5,3 = g 2 e10s/3 1 + e4s/3 ,
16

V4,4 = g 2 4 + e2s + e2s ,

5
3
V3,5 = g 2 e2s 1 + e4s/5 ,
16

V2,6 = 3g 2 e2s/3 1 + e4s/3 ,

5

1
V1,7 = g 2 e2s 1 + e4s/7 1 5e4s/7 + e8s/7 .
16
Note that the potential Vp,q can be obtained from Vq,p by rescaling s ps/q. The
eigenvalue of the A
1 tensor z1 provides a superpotential and one can check that the scalar
158
Table 3
Summary of critical points in the context of superpotential: symmetry group, supersymmetry, vacuum expectation
values of field, superpotential, and cosmological constants. For either case, it is exactly SO(7) -invariant critical
point of the SO(8) theory. It has no supersymmetry and is unstable
Gauge symmetry
q =8p
SO(8)
any
any
SO(7) SO(1)
SO(1) SO(7)
1 ln 1 (1 + 5)
2
2
7 ln 1 (1 + 5)
2
2
2
6g
353/4
8
353/4
8
2585 g 2
258 5 g 2
potential can be written in terms of superpotential:

Wp,q (s) = |z1 |,

2
2
2 2q
2
s Wp,q (s) 6Wp,q (s) = g 2 4 s Wp,q (s) 6Wp,q (s)2 ,
Vp,q (s) = g
p
where s = 2p/q s and z1 is given by (25). The kinetic terms are equivalent to the previous
cases. In this case, there are no such first-order differential equations for either a flow
between SO(8) fixed point and SO(7) SO(1) fixed point or a flow between SO(8)
and SO(1) SO(7) , contrary to the previous SO(p)+ SO(q)+ embedding case. The
superpotential has the following values at the two critical points (Table 3).
3. Domain wall from SO(p)+ SO(q)+ SO(r)+ sectors of N = 8 supergravity
Let us consider a sequence of nonsingular elements E( ) of L = SL(8, R) with real
parameter and E(1) = 1, identity transformation, whose limit point E(0) is singular and
not in L. As long as E( ) remains nonsingular ( = 0), the structure constants have
the usual tensor properties. Acting on the Lagrangian with E( ) yields a sequence of
Lagrangian: L L
( ) = L0 + L
g ( ). If one also rescales the coupling constant g by
a -dependent one through g g
( ) for some choices of the sequence E( ) in L, a limit
of L
g ( ) exists and is well defined. One can continue the Lagrangian L
( ) to negative
values of . In this case, L
(1) is the Lagrangian for different gauging and the gauge
group is noncompact SO(p, q + r)+ with p + q + r = 8 which will be discussed in
Section 3.2. One continues to consider a sequence of nonsingular elements F ( ) of L with
real parameter and F (1) = 1, identity transformation, whose limit point F (0) is singular
and not in L. As long as F ( ) remains nonsingular ( = 0), the structure constants have
the usual tensor properties. Acting on the Lagrangian L
with F ( ) yields a sequence of
Lagrangian: L
L
( ) = L0 + Lg
(, ). If one also rescales the coupling constant g
by a -dependent one through g

g
( ) for some choices of the sequence F ( ) in L,

the limit of Lg
( ) exists as 0 so that L
( = 0) = L0 + Lg
( = 0) gives the
Lagrangian. The gauge group corresponding to L
( = 0, = 1) is an InonuWigner
contraction of K,,p,q,r denoted by CSO(p, q, r)+ with p + q + r = 8 [11].
In Section 3.1, we start with the most general gaugings which generalize previous
considerations by introducing two parameters, and . The gauging denoted by
CSO(p, q, r)+ preserves a metric with p positive eigenvalues, q negative eigenvalues and
159
r zero eigenvalues. In Section 3.2, by analyzing two successive SL(8, R) transformations

(repeating twice) in the context of SO(p, q + r)+ and SO(p + q, r)+ gaugings, we discover
a T
tensor which depends on these two parameters, and . As done in previous sections,
the A1 and A2 tensors can be easily determined by realizing that 56-beins are product of
each 56-bein for each parametrization of the singlet-space. In Section 3.3 it turns out that
one has a scalar potential which can be written as a superpotential in very simple form
and in Section 3.4, we find the domain-wall solutions. In Section 3.5, by starting with
SO(p)+ SO(q)+ SO(r)+ invariant generator of SL(8, R) directly, one can construct
the projectors corresponding to this invariant four-form tensor, which we will compare
with the approach given in Sections 3.23.4.
3.1. Nonsemisimple and noncompact gaugings [11]
It is possible to gauge the 28-dimensional subgroup K,,p,q,r of L = SL(8, R) whose
algebra
[ab , cd ], = ad bc ac bd bd ac + bc ad ,

0
0
1pp
,
p + q + r = 8,
ab =
0
1qq
0
0
0
1rr
where a, b = 1, . . . , 8 and ab = ba .
When (, ) = (1, 1), this leads to the algebra of SO(8) and the de WitNicolai
gauging is recovered. When (, ) = (1, 0) it will give CSO(p + q, r)+ algebra which was
discussed in the previous section and the maximal compact subgroup is SO(p + q)+
U (1)+r(r1)/2. Moreover, when (, ) = (1, 1), one gets SO(p + q, r)+ algebra which
was already considered and the maximal compact subgroup is SO(p + q)+ SO(r)+ .
When (, ) = (1, 1), it will give noncompact SO(p, q + r)+ gauging whose
maximal compact subgroup is SO(p)+ SO(q + r)+ . When (, ) = (1, 0), it gives
a certain nonsemisimple algebra of the InonuWigner contraction of SO(8) about its
SO(p, q)+ subgroup, denoted by CSO(p, q, r)+ [11]. The maximal compact subgroup
is SO(p)+ SO(q)+ U (1)+r(r1)/2. Note that CSO(p, q, 1)+ = ISO(p, q)+ , the
inhomogeneous group. For (, ) = (1, 1), one gets SO(p + r, q)+ algebra.
When = 0, it gives InonuWigner contraction CSO(p, q + r)+ which was already
considered.
The CSO(p, q, r)+ gauging initiated by Hull which preserves a metric with p positive
eigenvalues, q negative eigenvalues and r zero eigenvalues can be obtained by group
contractions of SO(8) as follows. One decomposes each SO(8) generator into the
part () in the SO(p)+ subalgebra, the part () in the SO(q)+ subalgebra, the part
( ) in the SO(r)+ subalgebra, and the remainders () , () , and () where =
() + () + ( ) + () + () + () . See also the discussion around in (32). One
performs the rescaling as

() + (() + ( ) + () ) + (() + ()).
160
The rescaled algebra can be represented as

[(), () ] () ,
[() , () ] () ,
[(), () ] () ,
[() , () ] () ,
[() , () ] () ,
[() , () ] () ,
[( ) , ( ) ] ( ) ,
[( ) , () ] () ,
[( ), () ] () ,
[() , ()] () + () ,
[() , ()] () ,
[(), () ] () ,
[() , () ] () + ( ) ,
[() , () ] () ,
[(), () ] () + ( ) ,
with other commutators vanishing. By taking the contraction, 0, the SO(r)+ subgroup
generated by ( ) collapses to an Abelian group U (1)+r(r1)/2 and the maximal compact
subgroup of CSO(p, q, r)+ is SO(p)+ SO(q)+ U (1)+r(r1)/2. The generators ( )
commute all the generators except appearing on the right-hand sides of [(), () ]
and [() , () ]. The methods described in previous section will be used to obtain a
CSO(p, q, r)+ gaugings.
3.2. T -tensor in CSO(p, q, r)+ gaugings [11]
The CSO(p, q, r)+ gaugings can be obtained by acting on the SO(p, q + r)+ gauging
first. Some idea in this direction was already given in the paper of [14]. Let us consider
the SO(p)+ SO(q + r)+ invariant generator of SL(8, R) we have discussed in previous
section,

0
1pp
Xab =
(26)
0
1(q+r)(q+r)
with
p + (q + r) = 0,
p + q + r = 8.
(27)
Regarded as a 28 28 matrix, real, self-dual totally anti-symmetric (SO(p)+ SO(q +

r)+ )-invariant four-form tensor Xt+I J KL has eigenvalues , and = ( + )/2
with degeneracies d , d and d , respectively. The eigenvalues and eigenspaces of the
(SO(p)+ SO(q + r)+ )-invariant tensor are summarized in Table 1. By introducing
projectors, P,t , P,t and P ,t onto corresponding eigenspaces, we have a 28 28 matrix
equation. Projector P,t (P,t ) projects the SO(8) Lie algebra onto its SO(p)+ (SO(q +
r)+ ) subalgebra while P ,t does onto the remainder SO(8)/(SO(p)+ SO(q + r)+ ). Note
that q over there is replaced by q +r here. The projectors can be constructed from Xt+I J KL .
The combination gAIJ in the minimal couplings will be finite as t if g is rescaled to
g(t) = get
for constant which we have chosen as 1 so that
J

J
J
+ e()t AI()
+ e( )t AI(
g(t)AIJ (t) = g AI()
)

J
J
J
+ AI()
+ AI(
= g AI()
) ,
(28)
161
where = e()t as before. One finds that on taking the limit t ( 0) one obtains
a gauging with gauge group contraction of SO(8) about its SO(p)+ subgroup. If, instead,
one analytically continues to t = i/( ), one obtains a gauging of SO(p, q + r)+ .
Let us consider the additional, second SL(8, R) transformation using the (SO(p + q)+
SO(r)+ )-invariant generator of SL(8, R),

0
1(p+q)(p+q)
Xab =
(29)
0

1rr
with

(p + q) +
r = 0,
p + q + r = 8.
(30)
Regarded as a 28 28 matrix, real, self-dual totally anti-symmetric (SO(p + q)+

SO(r)+ )-invariant four-form tensor Xs+I J KL has eigenvalues
,
and
= (
+
)/2
with degeneracies d
, d
and d
, respectively. The eigenvalues and eigenspaces of the
(SO(p + q)+ SO(r)+ )-invariant tensor are summarized in Table 1. By introducing
projectors, P
,s , P
,s and P
,s onto corresponding eigenspaces, we have a 28 28
matrix equation. Projector P
,s (P
,s ) projects the SO(8) Lie algebra onto its SO(p + q)+
(SO(r)+ ) subalgebra while P
,s does onto the remainder SO(8)/(SO(p + q)+ SO(r)+ ).
Note that p over there is replaced by p + q here. The projectors can be constructed
from Xs+I J KL similarly. The combination gAIJ in the minimal couplings will be finite
as s if g is rescaled to
g(s) = ge s
for some constant
(taken as 1) so that by acting [exp(sXs+ )]I J KL on the right-hand
side of (28)

g(s, t)AIJ (s, t) = g PI

J,sKL + e( )s PI
J,sKL + e( )s PI
J,sKL

()t KL
AKL
A() + e( )t AKL
() + e
( )
IJ

J
IJ
J
J
= g AI(
A(
) + AI(
AI(
) + A(
) +
) +
)

IJ
+ A(
) ,
where = e( )s as before. Here we used the fact that

P
,s P,t = P
,s P ,t = P
,s P,t = 0
(31)
which can be shown by the explicit expression of projectors given in Appendix F and we
J
denote the simplified notations for AI(
) , where = , , , = , , as follows:
J
I J MN AMN ,
AI(
) (P
,s P,t )
J
I J MN AMN ,
AI(
) (P
,s P,t )
J
I J MN AMN ,
AI(
) (P
,s P ,t )
J
I J MN AMN ,
AI(
) (P
,s P,t )
J
I J MN AMN ,
AI(
) (P
,s P,t )
J
I J MN AMN .
AI(
) (P
,s P ,t )
162
KLMN , as a single projector.

Now we can think of the product of these projectors, PI
J,sKL P,t
Therefore, let us define them, to satisfy the usual property of projectors, as
P
,s P,t P ,
P
,s P,t P ,
P
,s P ,t P ,
P
,s P,t P ,
P
,s P,t P ,
P
,s P ,t P .
(32)
We will see that = ( + )/2, = ( + )/2 and = ( + )/2 and and are related
to s in (27) and (30). Projector P (P ) [P ] projects the SO(8) Lie algebra onto its
SO(p)+ (SO(q)+ ) [SO(r)+ ] subalgebra while P (P ) [P ] projects onto the remainder

SO(8)
SO(8)
SO(8)
SO(p)+ SO(q)+
SO(p)+ SO(r)+
SO(q)+ SO(r)+
which will be discussed in Section 3.5. One obtains these projectors explicitly from
the relation (32) where the projectors in (SO(p)+ SO(q)+ )-invariant sector are given
in Appendix F. In terms of these new projectors, one can write the combination
g(s, t)AIJ (s, t) as
IJ IJ
IJ
J
J
J
AI()
(33)
+ AI()
+ AI(
) + A() + A() + A() .
By expanding g(t, s)D(A , , ) with respect to both t and s, there exist many terms
that seem to diverge as t or s . However, by exploiting some identities of the
generators given in Appendix D, it implies that those divergent terms vanish identically
and therefore a limit of t or s exists.
By simplifying the expressions appearing in g(t, s)D(A , , ), one gets, for example,
the first 28 28 block diagonal terms given by
A() + P A() P + P A() P + P A()P + P A() P + P A() P

+ P A() P + A() + P A() P + P A()P + P A() P + P A() P

+ P A() P + P A() P

+ A( ) + P A() P + P A() P + P A() P + P A() P

+ P A() P + P A() P ,
(34)
where we used the properties between projectors and vector fields:
P A() P = P A() P = P A() P = 0.
One can prove that (34) becomes the one we have considered for SO(p, q + r)+ gauging
when = 1 by combining -dependent terms with -dependent terms9 and removing the
projectors P
,s (
=
,
,
) with (32) under the extensive manipulation of properties of
projectors. On the other hand, when = 1, it becomes the one in SO(p + q, r)+ gauging by
9 When = 1, (34) becomes,

A() + P,t A( ) P ,t + P ,t A( ) P,t + A() + P,t A( ) P ,t + P ,t A( ) P,t .
163
combining the , -independent terms with -dependent terms and removing the projectors
P,t ( = , , ). In this case, we can write it similarly.10
Collecting all other terms by simplifying other three 28 28 blocks we get
g(t, s)D(A , , ) = gD(A )

MN
MN
MN
MN
A() + 12 A() + A() ,
Z()I
J KLA() Z()I J KLA()
(1 )g

MN
MN
MN
MN
Z()I
A() + 12 A() + A()
J KLA() Z()I J KLA(),

MN
MN
MN
MN
Z()I
A( ) + 12 A() + A() ,
J KLA() Z()I J KLA()
(1 )g

MN
MN
MN
MN
Z()I
A( ) + 12 A() + A()
J KLA() Z()I J KLA() ,
MN
where Z(
)I J KL are quadratic forms of projectors
MN
Z()I
J KL =
1
(P P )I J MP PNP KL PI J MP (P P )NP KL
2

PI J MP PNP KL PI J MP PNP KL ,
(35)
MN
and Z()I
J KL can be written by performing the change of the above indices in (35) as
MN
, , , , and Z()I
J KL which can be expressed by changing
the indices in (35) as , , , , . Then our SU(8) T
tensor
encoding the structure of the scalar sector of the N = 8 supergravity can be read off and
one arrives at the final complicated expression:
Ti
j kl (, )

= Ti j kl (1 ) u kl I J + v klI J

KM j m

1
u LM vimKM v j mLM
uim
PI J KL + PI J KL + PI J KL
2

I J RS KLMN

KLMN
+ P
vimKL u j m MN + uim KL v j mMN
Z()RS PI J RS Z()RS

(1 ) u kl I J + v klI J

KM j m

1
uim
u LM vimKM v j mLM
PI J KL + PI J KL + PI J KL
2

I J RS KLMN
KLMN
vimKL u j m MN + uim KL v j mMN .
Z()RS PI J RS Z()RS
+ P
(36)
10 When = 1, (34) becomes

A(
) + P
,s A(
) P
,s + P
,s A(
) P
,s + A(
) + P
,s A(
) P
,s + P
,s A(
) P
,s .
164
Let us examine the structure of T

-tensor.11 When = 1, it consists of -independent
part plus -dependent part. One can prove that by plugging P ( = , , , , , )
into the product of P
,s (
=
,
,
) and P,t ( = , , ). According to (32),
the expressions of projectors proportional to 1 are identical to those in (13) for
(SO(p + q)+ SO(r)+ )-invariant sector. On the other hand, when = 1, the above
(36) will consist of -independent part plus -dependent part. By substituting P ( =
, , , , , ) back into P
,s (
=
,
,
) and P,t ( = , , ). According to
(32), the expressions of projectors proportional to 1 are the same as those in (13)
for (SO(p)+ SO(q + r)+ )-invariant sector. Therefore, the expressions of projectors
proportional to 1 in (36) are the difference between the one in (p, q + r) and the
one in (p + q, r). One can easily see that the expressions of projectors proportional to
1 in (36) are the one in (p + q, r). This implies that one can use the projectors in
(36) for those in (SO(p)+ SO(q)+ )-invariant sector. Alternatively, one can exploit those
projectors from (48) directly.
When (, ) = (1, 1), this leads to the algebra of SO(8) and one obtains de Wit
Nicolai gauging with SU(8) SO(8) gauge symmetry. When (, ) = (1, 0) one has
CSO(p + q, r)+ algebra with SU(8) CSO(p + q, r)+ gauge symmetry. Moreover, when
(, ) = (1, 1), one gets SO(p + q, r)+ algebra with SU(8) SO(p + q, r)+ gauge
symmetry. When (, ) = (1, 1), it will give noncompact SO(p, q + r)+ gauging with
SU(8) SO(p, q + r)+ gauge symmetry. When (, ) = (1, 0), it yields a nontrivial
nonsemisimple algebra of the InonuWigner contraction of SO(8) about its SO(p, q)+
subgroup, denoted by CSO(p, q, r)+ with SU(8) CSO(p, q, r)+ gauge symmetry. For
(, ) = (1, 1), one gets SO(p + r, q)+ algebra with gauge symmetry SU(8) SO(p +
r, q)+ . Finally, when = 0, it gives InonuWigner contraction CSO(p, q + r)+ with gauge
symmetry SU(8) CSO(p, q + r)+ . The gauge group will be spontaneously broken to its
maximal compact subgroup.
3.3. Superpotential and scalar potential in CSO(p, q, r)+ gaugings [11]
The parametrization for the (SO(p)+ SO(q)+ SO(r)+ )-singlet space that is
invariant subspace under a particular SO(p)+ SO(q)+ SO(r)+ subgroup of SO(8)
becomes

I J KL = 4 2 mXI+J KL,s + nXI+J KL,t ,
where m, n are two real scalar fields. The two scalar fields parametrize an (SO(p)+
SO(q)+ SO(r)+ )-invariant subspace of the full scalar manifold E7(7)/SU(8). The 56beins V can be written as a 56 56 matrix by exponentiating the vacuum expectation value
I J KL . One can construct 28-beins uij KL and vij KL in terms of scalars m, n explicitly,
KL and v
which can be given in terms of the products of uij, KL
ij KL,s , as
t , vij KL,t , uij, s
11 It is more convenient to use the SL(8, R) basis in order to compare operators in the dual CFT. This is
nothing but a triality rotation of the SU(8) basis we have considered in which the two representations of SO(8)
are converted to each other using gamma matrices. In an SL(8, R) basis, the expression of T -tensor was already
given as Eq. (21) in [11].
165
given in Appendix E.12 Now the full expression for A

1 and A
2 tensors are given in terms
of m, n using (5) and (36) with T
tensor.

1
I J P Q
0

2 2
V(x) = exp
1

MNKL
0
2 2

1
0
I J P Q,t
2 2
= exp
1

MNKL,t
0
2 2

1
0

I J P Q,s
2 2
exp
1

MNKL,s
0
2 2

IJ

I
J
I
J
uij, s
uij, t
vij KL,t
vij KL,s
vij KL
uij
=
,
=
(37)
v klI J u kl KL
vtklI J u kl KL,t
vsklI J u kl KL,s
where I J KL,s = 4 2 mXI+J KL,s and I J KL,t = 4 2 nXI+J KL,t , which commute each
other. It turns out from (36) that the A
1 tensor has a single real eigenvalue, z1 , with
degeneracies 8 that has the following form
A
1 ij = diag(z1 , z1 , z1 , z1 , z1 , z1 , z1 , z1 ),
p
1
m p n
p+q m (q+r)
n
.
z1 = pem+n + qe (q+r) + re r
(38)
8
It is now straightforward to verify that this yields (15) for (p + q, r) gauging when = 1
and n = 0. However, when = 1 and m = 0 it becomes (15) for (p, q + r) gauging. In
particular the superpotential, W , for the flow is found as one of the eigenvalues of the this
symmetric tensor. Additionally, we can construct the A
2 tensor from (36) which are the
combinations of the triple product of 28-beins as given in

p+q +ij kl
1
1
p n +ij kl
p n
A
2i j kl = (q + r)em en e q+r Xt
+ r e q+r em e r m Xs
4
4
= em (A2,t )i j kl + e
p
q+r
n
(A2,s )i j kl ,
(39)
SO(p)+
r)+
where (A2,t )i
is the same as the one (16) for
SO(q +
sector and

+ij kl
+ij kl
+
+
j
kl
for SO(p + q) SO(r) . Moreover Xt
is =,, P,t while Xs
(A2,s )i
j kl
12 One can express u KL and v

IJ
I J KL in terms of sum of product of 4 4 matrices as follows:

uI J KL = diag u1,t u1,s + v1,t v1,s , u2,t u2,s + v2,t v2,s , u3,t u3,s + v3,t v3,s ,

u4,t u4,s + v4,t v4,s , u5,t u5,s + v5,t v5,s , u6,t u6,s + v6,t v6,s , u7,t u7,s + v7,t v7,s ,

vI J KL = diag u1,t v1,s + v1,t u 1,s , u2,t v2,s + v2,t u 2,s , u3,t v3,s + v3,t u 3,s ,

u4,t v4,s + v4,t u 4,s , u5,t v5,s + v5,t u 5,s , u6,t v6,s + v6,t u 6,s , u7,t v7,s + v7,t u 7,s ,
and uI J, KL
where each ui,t and ui,s corresponds to seven 4 4 block diagonal matrices for uI J, KL
s respectively
t
as in Appendix E and vi,t and vi,s for vI J KL,t and vI J KL,s , respectively. Their complex conjugations hold
similarly.
166

is
=
,
,

P
,s .13 Finally the K,,p,q,r -invariant scalar potential as a function of
p, q, r, , and m, n by combining all the components can be written as

1
j kl 2
2 3
ij 2
Vp,q,r (, ; m, n) = g
A
A
4 1
24 2i
2pn
1 2(p+q)m
q+r
r
=
e
1536

V1 + V2 + V3 + V4 2 + V5 2 + V6 2 2 ,
where we introduce intermediate functions Vi s as the coefficients of the and
16m
+ 16n
V1 = e r q+r p p2 (q + r) + (2 + q + r)(q + r)2

+ 2p 72 + q 2 r + r 2 + q(1 + 2r) ,

pq 144 + p2 + q 2 + 2q(1 + r) 2r + r 2 + 2p(1 + q + r) ,
8m

+ 8n
V3 = 2e r q+r pr 144 + p2 + q 2 + 2q(1 + r) 2r + r 2 + 2p(1 + q + r) ,

16m
V4 = e r q p3 + q 2 r + (2 + r)r 2 + p2 (2 + 2q + 3r) + 2q 72 r + r 2

+ p q 2 + r(4 + 3r) + q(2 + 4r) ,

8m
V5 = 2e r qr 144 + p2 + q 2 + 2q(1 + r) 2r + r 2 + 2p(1 + q + r) ,
V6 = r p3 + q 3 + 2q 2(1 + r) 144r + q(2 + r)r + p2 (2 + 3q + 2r)

+ p 3q 2 + 4q(1 + r) + (2 + r)r .
V2 = 2e
16m
8n
r + q+r
13 One can prove A

and A
can be obtained by analytic continuation. The T
tensor we obtained is
1

2
Ti
j kl E(n) F (m), , . By considering only the SL(8, R) transformation by , this can be reduced to

et Ti
j kl E(t + n)1 F (m), 0, . Moreover, this becomes e(tn) Ti
j kl E(t)1 F (m), e()n , .

Now we arrive at the following intermediate expression: en Ti
j kl 1 F (m), e()n , . Next we apply
SL(8, R) transformation by . Then by doing a similar procedure we arrive at the final expression:

Ti
j kl E(n) F (m), , = e m en Ti
j kl 1, e()n , e( )m .
At the origin, I J KL = 0, V = 1, the T
tensor is from (36)
kl 3
3
ij kl
1 (1 )a11 (1 )a21 ij
(1 )a11 Xt
2
2
3
ij kl
(1 )a21 Xs .
(40)
2

Finally we possess all the information of Ti
j kl E(n) F (m), , because by transforming e()n ,

e( )m as in (40) we get Ti
j kl 1, e()n , e( )m . From this, one can obtain A
1 tensor which
4 T
j kl E(n) F (m), , . It turns out that it coincides with the one in (38). We used the numerical
is 21
i
Ti
j kl (1, , ) =
p
p
p+q
values:
= 1 = , = q+r
,
= p+q
r and a1 = q+r + 1, a2 = r + 1. Additionally, we have checked

4
j
kl
[j
kl]
= 3 Ti
E(n) F (m), , is identical to the one in (39).
that A2,i
167
By looking at the form of scalar potential, it is easy to see that Vr,q,p ( = 1, =

r
1; m, n) can be obtained from Vp,q,r = 1, = 1; p+q
n, q+r
p m . Under the
change of real fields, they are equivalent to each other. Moreover, the potential Vr,q,p ( =

r
1, = 1; m, n) can be obtained from Vp,q,r = 1, = 1; p+q
n, q+r
p m . On this
basis, the kinetic terms are not the usual ones but there exists a cross term, m n which
makes it difficult to find first-order differential equations for domain-wall solutions. Now
we have to change the basis for which one has usual kinetic terms. We calculated all
the quantities for 21 possible cases of CSO(p, q, r)+ gaugings and summarized them in
Appendix G: kinetic terms in terms of old fields,14 change of variables, superpotential,
and scalar potential as new fields. From the results in Appendix G, one can describe a
superpotential and scalar potential in terms of new real scalar fields m
and n that are related
to the old fields m and n as follows:
(q + r) pq(p + q)
2r(p + q)
r pq(p + q)
m
n,
n=
m.
m=
4q(p + q)
2(p + q)
4pq
Then in terms of new fields the superpotential can be written as

q
p
r
r
1
2 p(p+q)
m+
2 q(p+q)
m
2(p+q) n
2(p+q) n
Wp,q,r (, ; m,
n)
= pe
+ qe
8

p+q
n
+ re 2r ,
(41)
and the supergravity potential is then given by

2

2

n)
= g 2 4 m Wp,q,r (, ; m,
n)
+ 4 n Wp,q,r (, ; m,
n)
Vp,q,r (, ; m,

6Wp,q,r (, ; m,
n)
2 .
(42)
Note that the coefficients, 4 and 4, in the first and second terms in the above are a simple
generalization of (18) for two scalar fields.
The superpotential has the following values at the various critical points in Table 4.
14 One can generalize the kinetic terms (19) of (SO(p)+ SO(q)+ )-invariant sector to write down

A ij kl = 2 2 u ij I J v klI J v ij I J u kl I J

1
+ 4 2(1 )gAI J PI J KL + PI J KL + PI J KL
2

u ij KM v klLM + v ij KM u kl LM

KLMN P I J RS Z KLMN u ij
kl
ij KL v klMN
+ PI J RS Z()RS
KL u MN v
()RS

1
+ 4 2(1 )gAI J PI J KL + PI J KL + PI J KL
2
ij

klLM
ij
KM
kl
u KM v
+ v
u LM

KLMN P I J RS Z KLMN u ij
kl
ij KL v klMN .
u
+ PI J RS Z()RS
KL
MN
()RS
Of course, in this case we put AIJ to zero because we are interested in the scalar plus gravity parts of the
Lagrangian.
168
Table 4
Summary of various critical points in the context of superpotential: supersymmetry, vacuum expectation values
of fields, superpotential and cosmological constants. There is no SO(p)+ SO(q)+ U (1)+r(r1)/2 critical
point of potential for = 1 and = 0. The nontrivial CSO(p, q, r)+ gauging in this section does not provide
any new extra critical points
N
any
any
any
6g 2
0
0
1
1
3
4
5
8
1
2
3
8
1
4
1
8
0
1
ln 5
78 ln 5
ln 5
34
58
12
38
14
ln 5
ln 5
ln 5
ln 5
any
5
8
38
5
8
38
38
38
ln 3
1
2
21
1
2
21
21
21
5
8
5
8
5
8
5
8
38
38
1
2
14
3
8
18
1
4
1
8
ln 5
ln 3
ln 5
ln 3
ln 5
ln 3
ln 5
ln 3
ln 3
ln 3
ln 3
ln 3
ln 3
ln 3
1
2
1
2
1
2
1
2
21
21
em /4
33/8
2 31/4 g 2
33/8
33/8
33/8
33/8
0
2g 2
33/8
2 31/4 g 2
33/8
33/8
33/8
33/8
33/8
2g 2
78 ln 3
21 33/8
ln 3
7
8
34
3
4
58
12
1
2
21
1
2
21
21
ln 3
ln 3
ln 3
ln 3
ln 3
ln 3
ln 3
ln 3
33/8
ln 3
ln 3
2 53/4 g 2
ln 5
ln 3
3 51/8
2
2 31/4 g 2
33/8
33/8
33/8
33/8
33/8
(continued on next page)
169
Table 4 (continued)
N
2g 2
1, 0, 1
any
any
em+n /4
The first row corresponds to the maximal supersymmetric case of de WitNicolais

SO(8)-invariant trivial critical point.
The second row corresponds to the SO(7)+ -invariant critical point of the SO(8)
theory that is equivalent to the second one in Table 2. We find Vr,q,p ( = 1, = 1; m, n) =
r
n, q+r
Vp,q,r ( = 1, = 1; p+q
p m). This means that by a change of variables,
the solutions of (p, q, r) = (4, 1, 3), (5, 1, 2), (6, 1, 1) are obtained from (p, q, r) =
(3, 1, 4), (2, 1, 5), (1, 1, 6), respectively.
The third row is the gauging of CSO(p + q, r)+ = CSO(2, 6)+ that corresponds to
the sixth in Table 2.
The fourth row implies SO(p + q, r)+ = SO(5, 3)+ that corresponds to negative
superpotential, being equivalent to the third one in Table 2, or SO(p + q, r)+ = SO(3, 5)+
that corresponds to positive superpotential, being equal to the fifth in Table 2. In each case,
the potentials are the same, although the superpotentials are different at the critical points.
The fifth row implies SO(p + q, r)+ = SO(4, 4)+ , being equivalent to the fourth one
in Table 2.
For the sixth row one has SO(p, q + r)+ = SO(3, 5)+ gauging with positive
superpotential and SO(p, q + r)+ = SO(5, 3)+ with negative superpotential. According
to the symmetry between the potential, one can see that all the critical points in this row
can be obtained from those in fourth row: Vr,q,p ( = 1, = 1; m, n) = Vp,q,r ( = 1, =
r
1; p+q
n, q+r
p m). This implies that within a given class with the same scalar potential,
these solutions are coming from the same critical point in the fourth row but are viewed
along different directions in scalar space.
For the seventh row, we have SO(p, q + r)+ = SO(4, 4)+ gauging. Additionally, in
this case, all the critical points are similarly obtained from those in the fifth row. These
solutions are coming from the same critical point in the fifth row but are viewed along
different directions in scalar space.
For the eighth row, one has either SO(p + r, q)+ = SO(5, 3)+ gaugings with negative
superpotential or SO(p + r, q)+ = SO(3, 5)+ with positive superpotential.
In the ninth row SO(p + r, q)+ = SO(4, 4)+ gauging.
Finally the last row corresponds to CSO(p, q + r)+ = CSO(2, 6)+ gauging.
170
For the eighth and ninth rows we have the following symmetry in the potential:
r
n, q+r
Vr,q,p ( = 1, = 1; m, n) = Vp,q,r ( = 1, = 1; p+q
p m). In other
words, the solutions corresponding to (p, q, r) = (2, 5, 1), (3, 3, 2), (4, 3, 1) can be
obtained from (p, q, r) = (1, 5, 2), (2, 3, 3), (1, 3, 4) respectively in the eighth row.
Similarly, in the ninth row the solution of (p, q, r) = (1, 4, 3) is related to (p, q, r) =
(3, 4, 1).
3.4. Domain wall in CSO(p, q, r)+ gaugings [11]
The resulting Lagrangian of scalar-gravity sector takes

1
1
1
d 4 x g R m
m
n
n Vp,q,r (, ; m,
n)
.
2
2
2
(43)
With the ansatz (21) the equations of motion for the scalars and metric read
3
1
1
1
r2 A r Ar B + (r A)2 + (r m)
2 + (r n)
2 + e2B Vp,q,r (, ; m,
n)
= 0,
2
4
4
2
r2 m
+ 3r Ar m
r Br m
e2B m Vp,q,r (, ; m,
n)
= 0,
n)
= 0.
r2 n + 3r Ar n r Br n e2B n Vp,q,r (, ; m,
(44)
By plugging the domain-wall ansatz (21) into the Lagrangian (43), the energy-density per
unit area transverse to r-direction with the integration by parts on the term of r2 A can be
expressed similarly and after rewriting and recombining the energy-density by summation
of complete squares plus other terms, one gets
1
E[A, m,
n]
=
2

2
dr e3A+B 6 eB r A + 2 gWp,q,r (, ; m,
n)

2
+ eB r m
2 2 gm Wp,q,r (, ; m,
n)

2
+ eB r n 2 2 gn Wp,q,r (, ; m,
n)
+ 12 2 geB Wp,q,r (, ; m,
n)
rA

B
n)
.
+ 4 2 ge r Wp,q,r (, ; m,
n))
By recognizing that the last two terms can be written as 4 2 gr (e3A Wp,q,r (, ; m,
we arrive at

2
1
dr e3A+B 6 eB r A + 2 gWp,q,r (, ; m,
n)

2
+ eB r m
2 2 gm Wp,q,r (, ; m,
n)

2
+ eB r n 2 2 gn Wp,q,r (, ; m,
n)

3A

+ 2 2 g e Wp,q,r (, ; m,
n)
.

171
Therefore, one finds the energy-density bound

n)()
e3A()
E[A, m,
n]
2 2 g e3A() Wp,q,r (, ; m,

Wp,q,r (, ; m,
n)()
.
This E[A, m,
n]
is extremized by domain-wall solutions and the first-order differential
equations for the scalar fields one finds are the gradient flow equations of the superpotential
(41):
r m
= 2 2 eB gm Wp,q,r (, ; m,
n),
B
r n = 2 2 e gn Wp,q,r (, ; m,
n),
B
r A = 2 e gWp,q,r (, ; m,
(45)
n).
It is easy to check whether solutions m(r),
n(r)
and A(r) of (45) satisfy the gravitational

and scalar equations of motion in the second-order differential equations (44). The analytic
solutions of (45) for = 0 when B is a constant become

2r
2
c1 + 2 log g(8q+pr)x+c
2q p+q
4 2(p+q)
,

m(x)
=
q
p(p+q) (8q + pr)

r
2
8q p+q
c1 2pr log g(8q+pr)x+c
4 2(p+q)

n(x)
=
,
r
(p+q) (8q + pr)

2
p(p + q) log g(8q+pr)x+c

4 2(p+q)
,
A(x) = c1 +
8q + pr
where we change variable r into x in order not to confuse it with the integer r and c1
and c2 are constant. One expects to have nontrivial analytic solutions for nonzero for
particular (p, q, r) and , as in Section 2. In particular, for CSO(1, 1, 6)+ gauging where
we fix = 1 and = 0, we have

7gx + c2
2
m(x)
,
3 c1 + 2 log tan
7
4 2

7gx + c2
2
4c1 + 6 log tan
,
n(x)
=
14
4 2

7gx + c2
1
7gx + c2
1
7c1 tan
sin
+ 1 log
.
A(x) =
7
2
4 2
2 2
Similarly, one also has an analytic solution of CSO(3, 3, 2)+ gauging where = 1 and
= 0.
3.5. CSO(p, q, r)+ gaugings from SO(8) gaugings
Thus far, the values of p, q and r are greater than or equal to 1. If we allow those
values to have zero, then one can classify them as follows: (1) CSO(p, 0, 0)+ = SO(p)+ ,
172
(2) CSO(p, q, 0)+ = SO(p, q)+ , (3) CSO(p, 0, r)+ = CSO(p, r)+ , and (4) CSO(p, q, r)+ .
In this section, we take a different route from previous the case. Some idea in this direction
was already given in the paper of [14]. Let us consider the (SO(p)+ SO(q)+ SO(r)+ )invariant generator of SL(8, R),
0
0
1pp
1qq
0
Xab = 0
(46)
0
0
1rr
with
p + q + r = 0,
p + q + r = 8,
where 1pp is p p identity matrix. The embedding of this SL(8, R) in E7 is such that
Xab corresponds to the 56 56 E7 generator which is a noncompact (SO(p)+ SO(q)+
SO(r)+ )-invariant element of the SL(8, R) subalgebra of E7

0
X+I J KL
,
XI+J KL
0
where the real, self-dual totally anti-symmetric (SO(p)+ SO(q)+ SO(r)+ )-invariant
four-form tensor XI+J KL can be written in terms of a symmetric, trace-free, 8 8 matrix
with SO(8) right-handed spinor indices, Xab using SO(8) matrices (see Appendix B)
1
XI+J KL = (I J KL)ab Xab ,
(47)
8
where I J KL = [I J K L] and an arbitrary SO(8) generator LI J acts in the righthanded spinor representation by (LI J I J )ab . One can show that X+I J KL (47) can be
decomposed into Xt+I J KL and Xs+I J KL :
X+I J KL = Xt+I J KL + Xs+I J KL ,
where the real, self-dual totally anti-symmetric (SO(p)+ SO(q + r)+ )-invariant fourform tensor Xt+I J KL was expressed in the previous subsection as matrices with (26)
and (SO(p + q)+ SO(r)+ )-invariant four-form tensor Xs+I J KL with (29). Moreover,
and in (46) consist of t that was defined as (26) and (27) (we replace by t ) and s
as (29) and (30). We also replace
with s . Therefore we have
= t + s ,
= s
p
t .
q +r
Regarded as a 28 28 matrix, X+I J KL has eigenvalues , , , = ( + )/2, =

( + )/2, = ( + )/2 with degeneracies d , d , d , d , d and d , respectively.
The eigenvalues and eigenspaces of the (SO(p)+ SO(q)+ SO(r)+ )-invariant tensor
are summarized in Table 5. By introducing projectors, P , P , P , P , P and P onto
corresponding eigenspaces, we have a 28 28 matrix equation

PI J KL .
X+I J KL =
=,, ,,,
173
Table 5
Eigenvalues
and eigenspaces of the SO(p)+ SO(q)+ SO(r)+ invariant tensor, X+ where |X+ |2 =

2
=,, ,,, d | | . The degeneracies are given in d = p(p 1)/2, d = q(q 1)/2, d =
r(r 1)/2, d = pq, d = pr and d = qr. In [39], they displayed the signature of the KillingCartan form
by writing the numbers n+ , n and n0 of its positive, negative and zero eigenvalues. Here we identify d + d
with n+ , d with n and d + d + d with n0
p
|X+ |2
1
1
2
1
2
3
1
2
3
4
1
2
3
4
5
1
2
3
4
5
6
1
2
1
3
2
1
4
3
2
1
5
4
3
2
1
6
5
4
3
2
1
6
5
5
4
4
4
3
3
3
3
2
2
2
2
2
1
1
1
1
1
1
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
6/7
6/7
2/3
6/7
2/3
2/5
6/7
2/3
2/5
0
6/7
2/3
2/5
0
2/3
6/7
2/3
2/5
0
2/3
2
10/21
26/35
14/15
8/7
2
8/5
38/21
2
34/15
8/3
22/7
10/3
18/5
4
14/3
50/7
22/3
38/5
8
26/3
10
10/7
10/7
4/3
10/7
4/3
6/5
10/7
4/3
6/5
1
10/7
4/3
6/5
1
2/3
10/7
4/3
6/5
1
2/3
0
16/21
22/35
8/15
3/7
0
1/5
2/21
0
2/15
1/3
4/7
2/3
4/5
1
4/3
18/7
8/3
14/5
3
10/3
4
4/21
2/35
2/15
1/7
2/3
3/5
10/21
2/3
14/15
4/3
8/7
4/3
8/5
2
8/3
22/7
10/3
18/5
4
14/3
6
0
0
1
0
1
3
0
1
3
6
0
1
3
6
10
0
1
3
6
10
15
0
1
0
3
1
0
6
3
1
0
10
6
3
1
0
15
10
6
3
1
0
15
10
10
6
6
6
3
3
3
3
1
1
1
1
1
0
0
0
0
0
0
1
2
2
3
4
3
4
6
6
4
5
8
9
8
5
6
10
12
12
10
6
6
5
10
4
8
12
3
6
9
12
2
4
6
8
10
1
2
3
4
5
6
6
10
5
12
8
4
12
9
6
3
10
8
6
4
2
6
5
4
3
2
1
64/7
432/35
96/5
120/7
32
168/5
176/7
32
208/5
56
288/7
48
288/5
72
96
624/7
96
528/5
120
144
192
Projector P (P ) [P ] projects the SO(8) Lie algebra onto its SO(p)+ (SO(q)+ ) [SO(r)+ ]

SO(8)
subalgebra while P (P ) [P ] projects onto the remainder SO(p)SO(8)
+ SO(q)+
SO(p)+ SO(r)+

SO(8)
. The projectors can be constructed from X+I J KL ,
SO(q)+ SO(r)+
P =

=

MN

1
I J X+I J MN ,
)
for = , , , , ,
(48)
where ,
= , , , , , .
(49)
and it is easily checked that they satisfy

P2 = P ,
P P
= 0
( =
),
Then using the relation obtained by the properties of projectors above

I J KL
KLMN
exp sXs+
exp tXt+

=
e s PI
J KL
e t PKLMN

=
,
,
=,,
one gets

I J KL
KLMN MN

A
exp tXt+
g(s, t)AIJ (s, t) get e s exp sXs+
174
= get e s

=
,
,
e s PI
J,sKL
KLMN MN
e t P,t
A
=,,
J
I J KL AKL for = , , , , , .
which will be the same as (33) together with AI(
) P
In this section, the main difference with the previous section is that we started with
projectors directly constructed from (SO(p)+ SO(q)+ SO(r)+ )-invariant four-form
tensor. Of course, these projectors are very complicated expressions because they are fifth
power of X+I J KL or IKL
J given in (48). In the previous section, according to (32), we
identified the product of projectors in SO(p, q + r)+ and SO(p + q, r)+ with a single
projector (48) in this section.
4. Conclusion
In summary:
The main results in Section 3 is described by (24). There are BPS domain-wall
solutions interpolating between a maximally supersymmetric SO(8) critical point and
various nonsupersymmetric ones.
The analytic solution is available for only p = q = 4 with general . For = 0, we
also have solutions for general (p, q). That is, for SO(4, 4)+ and CSO(p, 8p)+ gaugings,
where p = 1, . . . , 7, there exist analytic solutions. Among these gaugings, only SO(4, 4)+
and CSO(2, 6)+ cases contain critical points according to Table 2. Note that the presence
of domain-wall solutions do not have any critical points.
In Section 3, the crucial part is to obtain a T -tensor as found in (36). Although it is
rather complicated and involved, all the components of a T -tensor can be obtained from
the information on both the projectors and 28-beins established by (SO(p)+ SO(q)+
SO(r)+ )-singlet space. The only nontrivial gauging comparable to that in Section 2
corresponds to both = 1 and = 0 that gives rise to CSO(p, q, r)+ gauging. The other
values of and gave us the previous gaugings discussed in Section 2.
Finally, we arrived at (41) and (42) that is a general expression, our new findings, for
two scalar fields as the one (18) for one scalar field.
Moreover, similar domain-wall solutions are described by (45). Although the scalar
potential for this case looks different from the case of SO(p)+ SO(q)+ , the structure
of the critical points are reduced to those in (SO(p)+ SO(q)+ )-invariant sector. We
emphasize that although CSO(p, q, r)+ gaugings (in this case = 1 and = 0) do not
have any critical points analyzed in Section 3.3, they do have domain wall solutions and
even possess analytic solutions for particular (p, q, r) combinations.
Recently, 11-dimensional embedding [27,45] of supersymmetric vacua of compactgauged supergravity was found. For solutions with varying scalars (due to the r-dependence
of vacuum expectation values), the ansatz for the field strength was more complicated. In
this direction it was crucial to know about the 11-dimensional analog of superpotential,
so-called geometric superpotential, in order to achieve the M-theory lift of the RG flow.
Provided that the r-dependence of the vevs is controlled by the RG flow equations, an exact
175
solution to the 11-dimensional EinsteinMaxwell equations was obtained. As mentioned

in the introduction, the 11-dimensional origin of SO(p, q)+ and CSO(p, q)+ gaugings was
found in [14] for constant scalars. In this paper, we describe explicit r-dependence on the
vevs by domain-wall solutions. It is natural to ask whether 11-dimensional embedding of
various vacua we have considered of noncompact and nonsemisimple gauged supergravity can be obtained. In a recent paper [46], the metric on the 7-dimensional internal space
and domain wall in 11-dimensions was found. However, they did not provide an ansatz
for an 11-dimensional three-form gauge field. It would be interesting to study the geometric superpotential, 11-dimensional analog of superpotential we have obtained. We expect
that the nontrivial r-dependence of vevs makes EinsteinMaxwell equations consistent not
only at the critical points but also along the supersymmetric RG flow connecting two critical points.
In [38], all critical points of the scalar potential of the N = 8 supergravity with SO(8)
gauge symmetry that break the local SO(8) down to a solution with symmetry that is at least
some specified subgroup of SO(8) were found. One considers only those scalars which are
singlets of that subgroup and searches critical points of the potential restricted to be a
function only of the singlets. Schurrs lemma tells us that any critical point of restricted
potential will be a critical point of the original complete scalar potential. Then the problem
of finding critical points of the potential is reduced to the simpler one of finding critical
points of the restricted potential which is a singlet sector. In this paper, we applied similar
techniques to the noncompact and nonsemisimple gauged supergravities and the subgroup
is to be SO(p)+ SO(q)+ for the SO(p, q)+ gaugings and CSO(p, q)+ gaugings while
that will be SO(p)+ SO(q)+ SO(r)+ for the CSO(p, q, r)+ gaugings.
In [38], the specified subgroup H was taken to be SU(3) for SO(8) gauged supergravity.
One can think of the H subgroup as a compact subgroup of SO(p, q)+ gauged model
because this is necessary to the validity of Schurrs lemma. When 56-beins commute the
SL(8, R) transformation E(t), it is rather easy to calculate the scalar potential. However,
it may happen that for the noncommutativity of 56-beins V and E(t), it will be rather
complicated to find the scalar potential because of the presence of additional Baker
Hausdorff terms appearing in the calculations of exponentials of matrices. According to
[13], it was found that no G2 -invariant critical points exist for SO(7, 1)+ gauging, no
SU(3)-invariant critical points for SO(6, 2)+ gauging and a SO(5)-invariant critical point
with positive cosmological constant, and no supersymmetry for SO(5, 3)+ gauging. It
would be interesting to investigate whether there exist any critical points of the potential
restricted to the H -singlet sector for the most general CSO(p, q, r)+ gaugings we have
considered in this paper. Here group H is a compact subgroup of this model.
Acknowledgements
This research was supported by Kyungpook National University Research Fund, 2000
and grant 2000-1-11200-001-3 from the Basic Research Program of the Korea Science &
Engineering Foundation. CA thanks Max-Planck-Institut fr Gravitationsphysik, AlbertEinstein-Institut where part of this work was undertaken and thanks B. de Wit, C.M. Hull
and H. Nicolai for discussions.
176
Appendix A. Four-form (anti)self-dual tensors in 28 28 matrices

Let us consider the SO(p) SO(q) invariant generator of SL(8, R),

0
1pp
Xa b =
with p + q = 0, p + q = 8,
0
1qq
where 1pp is a p p identity matrix. The embedding of this SL(8, R) in E7 is such that
Xa b corresponds to the 56 56 E7 generator with XI J KL

0
XI J KL
,
XIJ KL
0
where the real, anti-self-dual totally anti-symmetric tensor XI J KL is given by the
following form through the matrix
1
XIJ KL = (I J KL)a b Xa b ,

(A.1)
8
where I J KL = [I J K L] as in Section 2.5 and an arbitrary SO(8) generator LI J
acts in the left-handed spinor representation (see Appendix B for this representation) by
(LI J I J )a b . When p = 7 and q = 1, one can see that this expression of (A.1) through
I J KL
matrix coincides exactly with the one in Section 2.5 or X7,1
presented below explicitly.
We have seen real (anti)self-dual tensors in the SU(8)-basis through matrices in (12)
and (A.1). Now one can express them as the following forms which will be a useful and
illuminating description, viewed as a 28 28 matrix representation, after doing the
matrix algebra
I J KL
I J KL
MNP Q
= Yp,q
+ I J KLMNP Q Yp,q
,
Xp,q
24
where self-duality + corresponds to = 1 and anti-self-duality corresponds to
I J KL tensors are given for each p and q in
= 1 and Yp,q

1 I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
Y7,1
1234 + 1256
,
=
+ 1278
+ 1375
+ 1368
+ 1458
+ 1467
2

1 I J KL
I J KL
I J KL
I J KL
Y6,2
=
+ 1278
,
1234 + 1256
2

1
I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
31234
,
Y5,3
=
+ 1256
+ 1278
+ 1537
+ 1368
+ 15
+
48
1647
6
1 I J KL
I J KL
Y4,4
= 1234
,
2

1
I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
Y3,5
31234
,
=
+ 1526
+ 1278
+ 1357
+ 1368
+ 1458
+ 1647
10

1 I J KL
I J KL
I J KL
I J KL
Y2,6
=
+ 1278
,
1234 + 1526
6

1
I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
I J KL
1234
.
Y1,7
=
+ 1526
+ 1278
+ 1357
+ 1368
+ 1548
+ 1467
14
177
I J KL
Actually the case of X5,3
can be identified with SO(5) -singlets among six scalars
[47] when restricted to equal real parameters (I J KL depends on only three real parameters
because of SO(3) rotation).
Appendix B. SO(8) matrices and its representations

The 28 SO(8) generators are denoted by MN where M, N = 1, 2, . . . , 8 and they can
be decomposed into MN = (mn , m1 ). Here mn = nm where m, n = 2, 3, . . . , 8
are the 21 generators of SO(7). Then the 8 8 SO(7) gamma matrices satisfy {m , n } =
2mn and the generators act on the 8-dimensional spinor representation of SO(7) by
1 mn
4 mn . Then the 16 16 SO(8) gamma matrices have the following form, MN =
diag((MN )ab , (MN )a b ) where

MN = MN = mn ,
M, N = 2, 3, . . . , 8,
M1 = M1 = m
and a, b are right-handed spinor indices and a,

b are left-handed spinors. The SO(8) has
three different eight-dimensional representations: the vector representation 8v generated
by MN , the right-handed spinor representation 8s generated by 14 mn mn , and the lefthanded spinor representation 8c generated by 14 mn mn . This induces three inequivalent
SO(7) subgroups of SO(8). That is, the stability group of the vector, SO(7) is generated by
MN , M, N = 2, 3, . . . , 8, the stabilizer of a right-handed spinor, SO(7)+ is generated by
MN MN , and the stabilizer of a left-handed spinor, SO(7) is generated by MN MN .
The SO(7)+ -singlet under the branching rule of 35-dimensional fourth rank self-dual
anti-symmetric tensor representation of SO(8) into SO(7)+ corresponds to the SO(7)+ invariant tensor X+I J KL given in Section 2.3. Moreover, we present explicit realizations
of matrices we are using here as follows [32,48]:

0
,
3

0 1
5
=
,
1 0

0 2
7
=
,
2 0
2 =
3
0

3 =
2
0
0
3

0
8
=
1
=
6

0
,
2

3
,
0

1
,
0

4=
1
0
0
1
where i s and i s are given in terms of usual 2 2 Pauli matrices i s

=
1

=
1
0
1
0
i 2
1
0

,

i 2
,
0
2

0
3
i
3
,
=
,
=
0
0 i 2

0 1
i 2
0
2
3
,
=
.
=
1 0
0
i 2
2
0
3
178
Appendix C. Some identities between invariant generators and projectors in

SO(p)+ SO(q)+ sectors
J
, the invariance of X+I J KL under the SO(p)+
For any SO(p)+ SO(q)+ generator I()
implies

()
()
0
0
E(t)1
E(t) =
,
0
()
0
()
which is equivalent to
[P , () ] = [P , () ] = [P , () ] = 0.
Similarly, the invariance of X+I J KL under the SO(q)+ implies

()
()
0
0
E(t)1
E(t) =
,
0
()
0
()
which will lead to vanishing of commutators between P,, and ()
[P , () ] = [P , () ] = [P , () ] = 0.
One gets the following identities
P () P = P () P = P () P = P () P = 0,
P ()P = P () P = P () P = P () P = 0,
P () P = P () P = P () P = P () P = 0.
Moreover, one gets for the
SO(8)/(SO(p)+
SO(q)+ )
(C.1)
generator ( )
P ( ) P = P ( ) P = P ( ) P = P ( ) P = P ( ) P = 0.
(C.2)
With 1 = P + P + P , the combinations of (C.2) will give us

P ( ) P = P ( ) ,
P ( ) P = ( ) P ,
P ( ) P = P ( ) ,
P ( ) P = ( ) P .
(C.3)
By combining the first (second) and third (fourth) relations of (C.3) respectively and using
(C.2) it is easily checked that
(P + P )( ) = ( ) P ,
( ) (P + P ) = P ( ) .
Appendix D. Some identities between invariant generators and projectors in

SO(p)+ SO(q)+ SO(r)+ sectors
J , the invariance of X +I J KL under
For any SO(p)+ SO(q)+ SO(r)+ generator I()
+
the SO(p) implies

()
()
0
0
E(t)1
E(t) =
,
0
()
0
()
179
which is equivalent to
[P , () ] = 0
for = , , , , , .
Similarly, the invariance of X+I J KL under the SO(q)+ implies

0
0
()
1 ()
E(t)
E(t) =
,
0
()
0
()
which will lead to
[P , () ] = 0 for = , , , , , .
Similarly, the invariance of X+I J KL under the SO(r)+ implies

0
( )
0
1 ( )
E(t)
E(t) =
,
0
( )
0
( )
which will lead to
[P , ( ) ] = 0
for = , , , , , .
Using the relations (49), one gets the following identities

P ()P
= P () P
= P ( ) P
= 0,
for ,
= , , , , , , =
.
J
Moreover, one gets for the SO(8)/(SO(p)+ SO(q)+ ) generator I()
P () P = 0,
= , , , , ,
P () P = 0,
= , , , , ,
P ()P = 0,
= , , , , ,
P () P = 0,
= , , , ,
P () P = 0,
= , , , , ,
= , , , , .
P () P = 0,

With 1 = =,, ,,, P , the combinations of (D.1) will give us
P () P = P () ,
P () P = P () ,
P () P = P () ,
P () P = () P ,
P () P = () P ,
P () P = P () ,
P () P = P () .
(D.1)
(D.2)
SO(8)/(SO(p)+
SO(r)+ )
P () P = 0,
= , , , , ,
P () P = 0,
= , , , , ,
P ()P = 0,
= , , , , ,
P () P = 0,
= , , , , ,
P () P = 0,
= , , , ,
P () P = 0,
= , , , , .
generator
J
I()
(D.3)
180
With 1 =
=,, ,,, P ,
the combinations of (D.3) will give us
P () P = P (),
P ()P = P () ,
P () P = P () ,
P () P = P (),
P () P = ()P ,
P ()P = () P ,
P () P = P () .
(D.4)
SO(8)/(SO(q)+
SO(r)+ )
P () P = 0,
= , , , , ,
P () P = 0,
= , , , , ,
P () P = 0,
= , , , , ,
P () P = 0,
= , , , , ,
P () P = 0,
= , , , , ,
generator
J
I()
= , , , .
P () P = 0,

With 1 = =,, ,,, P , the combinations of (D.5) will give us
P () P = P () ,
P () P = P () ,
P () P = P () ,
P () P = P () ,
P () P = P () ,
P () P = () P ,
P () P = () P .
(D.5)
(D.6)
Using (D.2), (D.4) and (D.6) it is easily checked that

(P + P )() = () P ,
() (P + P ) = P () ,
(P + P )() = () P ,
() (P + P ) = P (),
(P + P )() = () P ,
() (P + P ) = P () ,
P () = ()P ,
P () = () P ,
P () = () P ,
P () = ()P ,
P () = () P ,
P () = () P ,
P () = () P ,
P () = () P ,
P () = () P .
Appendix E. 28-beins uI J KL and v I J KL for each invariant sector

The 28-beins uI J KL and vI J KL fields can be obtained by exponentiating the vacuum
expectation values I J KL . The nonzero components of those have the following seven
4 4 block diagonal matrices respectively
uI J KL = diag(u1 , u2 , u3 , u4 , u5 , u6 , u7 ),
vI J KL = diag(v1 , v2 , v3 , v4 , v5 , v6 , v7 ).
Each Hermitian submatrix is a 4 4 matrix and we denote anti-symmetric indices
explicitly for convenience. For simplicity, we make an empty space corresponding to
lower triangle elements. We also denote + = 1 (self-dual), = i (anti-self-dual) and
= 1 corresponding to self-dual case or 1 anti-self-dual case. We write down here each
Hermitian matrices.
(SO(7) SO(1) )-invariant sectors:
[12] [34] [56] [78]

B B B
[12] A
u1 = [34]
A
B
B ,
[56]
A
B
[78]
A
[13] [24] [57] [68]

[13] A B B B
u2 = [24]
A
B
B ,
[57]
A
B
[68]
A
[14] [23] [58] [67]

B B B
[14] A
u3 = [23]
A
B
B ,
[58]
A
B
[67]
A
[15] [26] [37] [48]

[15] A B B B
A
B
B ,
u4 = [26]
[37]
A
B
[48]
A
[16] [25] [38] [47]

B B B
[16] A
u5 = [25]
A
B
B ,
[38]
A
B
[47]
A
[17] [28] [35] [46]

[17] A B B B
u6 = [28]
A
B
B ,
[35]
A
B
[46]
A
[18] [27] [36] [45]

B B B
[18] A
u7 = [27]
A
B
B ,
[36]
A
B
[45]
A
[12] [34] [56] [78]

G G G
[12] F
v1 = [34]
F
G
G ,
[56]
F
G
[78]
F
181
182
[13] [24]
G
[13] F
v2 = [24]
F
[57]
[68]
[14] [23]
G
[14] F
v3 = [23]
F
[58]
[67]
[15] [26]
[15]
F
G
F
v4 = [26]
[37]
[48]
[16] [25]
G
[16] F
v5 = [25]
F
[38]
[47]
[17] [28]
G
[17] F
v6 = [28]
F
[35]
[46]
[18] [27]
G
[18] F
v7 = [27]
F
[36]
[45]
[57] [68]
G G
G
G ,
F
G
F
[58] [67]
G G
G
G ,
F
G
F
[37] [48]
G G
G
G ,
F
G
F
[38] [47]
G G
G
G ,
F
G
G
F
[35] [46]
G G
G
G ,
F
G
F
[36] [45]
G G
G
G ,
F
G
F
(E.1)
where
A = cosh3 s,
B = cosh s sinh2 s,
F = sinh3 s,
G = sinh s cosh2 s.
From now on, we do not include the index pairs into the 4 4 matrices ui and vi , for
simplicity. For example, when we write u2 = u3 below, this implies that although the
indices they possess are different, the corresponding matrix elements are identical.
u1 =
B
A
B
B
A
B
B
,
B
A
u2 = C144 = u3 = u4 = u5 = u6 = u7 ,
v1 =
v2 =
G
F
0 H
0
183
G G
G
G
,
F
G
F
0 0
0 0
= v3 = v4 = v5 = v6 = v7 ,
0 H
0
where
A = cosh3 s,
B = cosh s sinh2 s,
C = cosh s,
= sinh3 s,
G = sinh s cosh2 s,
H = sinh s.
A B C C
A B C C
A
C
C
A
C
C
u2 =
u1 =
,
,
A
B
A
B
A
A
A B C C
D E E E
A
C
C
D
E
E
u3 =
u4 =
,
,
A
B
D
E
A
D
D E E E
D E E E
D
E
E
D
E
E
u5 =
u6 =
,
,
D
E
D
E
D
D
F G H H
D E E E
F
H
H
D
E
E
v1 =
u7 =
,
,
F
G
D
E
F
D
F G H H
F G H H
F
H
H
F
H
H
v2 =
v3 =
,
,
F
G
F
G
F
F
I J J J
I J J J
I
J
J
I
J
J
v5 =
v4 =
,
,
I
J
I
J
I
I
I J J J
I J J J
I
J
J
I
J
J
v6 =
v7 =
,
,
I
J
I
J
I
I
where

2s
s
s
A = 1 + 2 cosh
cosh3
,
B = cosh(s) sinh2
,
3
3
3
184

s
s
s
C = 2 cosh
+ cosh(s) sinh2
,
D = cosh3
,
3
3
3

2s
s
s
s
sinh2
,
F = 1 + 2 cosh
sinh3
,
E = cosh
3
3
3
3

1
s
5s
s
G = cosh2
sinh(s),
H=
sinh
sinh
,
3
4
3
3

s
1
s
I = sinh3
,
J=
sinh
+ sinh(s) .
3
4
3
u1 = A144 = u2 = u3 ,
u4 = u5 = u6 = u7 = 144 ,
0 B 0 0
0 0 0
v1 =
v4 = v5 = v6 = v7 = 0,
= v2 = v3 ,
0 B
0
where
A = cosh s,
B = sinh s.
(SO(3)
SO(5) )-invariant sectors:
A B C C
A B C
A
C
C
A
C
u1 =
u2 =
,
A
B
A
A
A B C C
D E E
A
C
C
D
E
u3 =
u4 =
,
A
B
D
A
D E E E
D E E
D
E
E
D
E
u5 =
u6 =
,
D
E
D
D
F G H
D E E E
F
H
D
E
E
u7 =
v1 =
,
F
D
E
D
F G H H
F G
F
H
H
F
v3 =
v2 =
,
F
G
F
I J J J
I J
I
J
J
I
v4 =
v5 =
,
I
J
I
C
C
,
B
A
E
E
,
E
D
E
E
,
E
D
H
H
,
G
F
H H
H
H
,
F
G
F
J J
J
J
,
I
J
I
v6 =
J
I
J
J
I
J
J
,
J
I
v7 =
J
I
185
J
J
I
J
J
,
J
I
where

3s
s
s
2s
cosh3
,
B = cosh
sinh2
,
A = 1 + 2 cosh
5
5
5
5

1
s
3 s
C=
cosh(s) cosh
,
D = cosh
,
4
5
5

s
s
2s
s
sinh2
,
F = 1 + 2 cosh
sinh3
,
E = cosh
5
5
5
5

3s
1
s
2 s
sinh
,
H=
sinh(s) sinh
,
G = cosh
5
5
4
5

s
s
s
I = sinh3
,
J = cosh2
sinh
.
5
5
5
All these functions of s can be obtained from those in SO(5) SO(3) by replacing s
with 3s/5 and using the properties of hyperbolic functions. For example, each C that seems
to look different is the same by a simple change of variable.
B
B
u1 =
u2 = C144 = u3 = u4 = u5 = u6 = u7 ,
,
B
A
F G G G
F
G
G
v1 =
,
F
G
F
0 H 0 0
0 0 0
v2 =
= v3 = v4 = v5 = v6 = v7 ,
0 H
0
A
B
A
B
B
A
where

s
,
A = cosh3
3

s
F = sinh3
,
3

s
s
B = cosh
sinh2
,
3
3

s
s
G = sinh
cosh2
,
3
3

s
C = cosh
,
3

s
H = sinh
.
3
with s/3.
186
B
B
u1 =
= u3 = u4 ,
B
A
A B B B
A
B
B
u2 =
= u6 ,
A
B
A
A B B B
A
B
B
u5 =
,
A
B
A
A B B B
A
B
B
u7 =
,
A
B
A
F G G G
F
G
G
v1 =
= v3 = v4 ,
F
G
F
F G G G
F
G
G
v2 =
= v6 ,
F
G
F
F G G G
F
G
G
v5 =
,
F
G
F
F G G G
F
G
G
v7 =
,
F
G
F
A
B
A
B
B
A
where

s
A = cosh
,
7

s
,
F = sinh3
7
3

s
2 s
B = cosh
sinh
,
7
7

s
s
G = sinh
cosh2
.
7
7
with s/7.
187
Appendix F. Projectors of SO(p)+ SO(q)+ sectors in 28 28 matrices

I J KL ( = , , ) of (SO(p)+ SO(q)+ )-invariant sectors can be
The projectors P,p,q
I J KL and P I J KL only because P I J KL can be obtained from
obtained explicitly. We list P,p,q
,p,q
,p,q
J KL = 1 P I J KL P I J KL :
those PI,p,q
,p,q
,p,q
I J KL
P,7,1
= diag(F1 , F2 , F1 , F3 , F4 , F2 , F1 ),
I J KL
= 0,
P,7,1
I J KL
P,6,2
= diag(F1 , F9 , F10 , F9 , F10 , F9 , F10 ),
I J KL
P,6,2
= diag(F5 , 0, 0, 0, 0, 0, 0),
I J KL
P,5,3
= diag(F10 , F9 , F10 , F8 , F7 , F5 , F6 ),
I J KL
= diag(F5 , F6 , F8 , 0, 0, 0, 0),
P,5,3
I J KL
P,4,4
= diag(F10 , F9 , F10 , 0, 0, 0, 0),
I J KL
P,4,4
= diag(F9 , F10 , F9 , 0, 0, 0, 0),
I J KL
P,3,5
= diag(F7 , F8 , F6 , 0, 0, 0, 0),
I J KL
= diag(F9 , F10 , F9 , F8 , F6 , F7 , F5 ),
P,3,5
I J KL
= diag(F7 , 0, 0, 0, 0, 0, 0),
P,2,6
I J KL
= diag(F3 , F10 , F9 , F9 , F10 , F10 , F9 ),
P,2,6
I J KL
P,1,7
= 0,
I J KL
P,1,7
= diag(F3 , F4 , F3 , F3 , F1 , F4 , F2 ),
where the 4 4 block diagonal matrices Fi s are
3 1 1 1
3
1
1
1 1 3 1 1
1 1
3 1
F2 =
F1 =
,
1 1 3
8 1 1 3 1
8
1 1 1 3
1 1
1
3
1 1 1
3 1 1
1 1
1 1 3
3
1 1
1
F3 =
F4 =
,
3
1
1
1
3
8 1 1
8
1 1 1
3
1
1 1
1 1 1 1
1 1 1 1
1 1 1 1 1
1 1 1
1 1
F5 =
F6 =
,
,
1 1
8 1 1 1 1
8 1 1
1 1 1 1
1 1 1 1
1 1 1 1
1
1 1
1 1 1 1 1
1 1
1 1
F8 =
F7 =
,
1 1 1 1
8
8 1 1 1
1 1 1 1
1 1 1
1
1
,
1
3
1
1
,
1
3
1
1
,
1
1
188
1
1 1
F9 =
4 0
0
1
1
0
0
0
0
1
1
0
0
,
1
1
1 1
1 1 1
F10 =
0
0
4
0
0
0
0
0
0
.
1 1
1 1
Appendix G. Kinetic terms, superpotential and potential in

SO(p)+ SO(q)+ SO(r)+ sectors
We list here (1) the kinetic terms in terms of original variables, m and n, (2) new
variables, m
and n in order to have usual canonical expression of kinetic terms, (3)
superpotential in terms of new fields, and (4) scalar potential in SO(p)+ SO(q)+
SO(r)+ sectors. In all cases, the scalar potential can be expressed in terms of superpotential
as (42). In this revised version, we list only six cases due to space limitations and refer to
the original version in the hep-th archive for remaining cases.
SO(1, 7)+ SO(2, 6)+ CSO(1, 1, 6)+ :
1
2
1
K1,1,6 (m, n) = m m m n n n,
3
7
7
7 2
6
3 2
m
n,
n=
m,
m=
4
2
4
6 n
1 2 2 m+
2 3 6 n
2
W1,1,6 (, ; m,
n)
= e
,
e2 2 m + + 6e 2 m+
8

1 22 m
6 n 4 2 m
2 3 6 n
V1,1,6(, ; m,
e
n)
= e
2e2 2 n 12e3 2 m+
+ 2
8
2 3 6 n 2
4 3 6 n 2 2
12e 2 m+
24e2 2 m+
.
There exists a SO(7)+ -invariant critical point of SO(8) theory for = 1 and = 1 and a
(SO(2)+ SO(6)+ )-invariant critical point for = 1 and = 0.
SO(1, 7)+ SO(3, 5)+ CSO(1, 2, 5)+ :
3
2
1
K1,2,5 (m, n) = m m m n n n,
5
7
7
5 6
7 6
5 6
m
n,
n=
m,
m=
24
6
8
30 n
6
1 2 6 m+
6m
m+
4 1530 n
6
3
W1,2,5 (, ; m,
n)
= e
+ 2 + 5e
,
e
8
56
30
2 6
1 6 m
4 1530 n
V1,2,5(, ; m,
n)
= e 3 3 n e 3 m 4e 3 n 10e 6 m+

4 30
6
8 1530 n 2 2
20e 15 n 2 15e 3 m+
.
There exists an (SO(3)+ SO(5)+ )-invariant critical point for = 1 and = 1.
189
SO(1, 7)+ SO(4, 4)+ CSO(1, 3, 4)+ :

2
1
K1,3,4 (m, n) = m m m n n n,
7
7
7 3
3
2
m
n,
n=
m,
m=
6
2
6
2 n
4 3
3
1 2 3 m+3
2 n
6
W1,3,4 (, ; m,
e 3 m + 3 + 4e 3 m+
n)
= e
,
8
2 3
2 3
1
23 m
2 n
2 n
2 n 2
V1,3,4(, ; m,
n)
= e
6e 3 m
8e 3 m e 3 m
3
24e 3 m 2 8e 2 n 2 2 .
There exists an (SO(4)+ SO(4)+ )-invariant critical point for = 1 and = 1, and an
SO(1, 7)+ SO(5, 3)+ CSO(1, 4, 3)+ :
5
2
1
K1,4,3 (m, n) = m m m n n n,
3
7
7
7 5
30
3 5
m
n,
n=
m,
m=
40
10
8
30 n
5
1 2 5 m+
4 1530 n
10
W1,4,3 (, ; m,
e 5 m + 4 + 3e 5 m+
n)
= e
,
8
30 n
6 5
1 2 5 m+
4 1530 n
5
e2 5 m 8e 5 m 6e 5 m+
V1,4,3(, ; m,
n)
= e
8 2
8

5
2 5
4 1530 n 2
8 1530 n 2 2
24e 5 m+
3e 5 m+
.
There exists an (SO(5)+ SO(3)+ )-invariant critical point for = 1 and = 1, and a
SO(1, 7)+ SO(6, 2)+ CSO(1, 5, 2)+ :
2
1
K1,5,2 (m, n) = 3 m m m n n n,
7
7
7 30
30
6
m
n,
n=
m,
m=
60
6
20

30
2 30
30
1
66 n
2 3 6 n
W1,5,2 (, ; m,
e 5 m + 5 + 2e 15 m+
n)
= e 15 m
,
8
2 30
7 30
1 2 30 6 n
4 30 m
2 3 6 n
3
e 5 + 10e 5 m + 4e 15 m+
V1,5,2(, ; m,
n)
= e 15 m

30
2 3 6 n 2
.
+ 15 2 + 20e 15 m+
There exists an (SO(3)+ SO(5)+ )-invariant critical point for = 1 and = 1.
SO(1, 7)+ SO(7, 1)+ CSO(1, 6, 1)+ :
2
1
K1,6,1 (m, n) = 7 m m m n n n,
7
7
190
7 42
42
14
m
n,
n=
m,
m=
168
14
24
42
42
42
1
1414 n
3 n)
W1,6,1 (, ; m,
e 3 m + 6 + e 21 (m+4
n)
= e 21 m
,
8
42
14
42
4 42
1 2 42 m
3 n)
V1,6,1(, ; m,
n)
= e 21 7 n e2 3 m 12e 3 m 2e 21 (2m+
42
2 42
3 n)
2
3 n)
2 2
24 2 12e 21 (m+4
+ e 21 (m+4
.
There are no critical points, in this case.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
H.J. Boonstra, K. Skenderis, P.K. Townsend, JHEP 9901 (1999) 003, hep-th/9807137.
H. Lu, C.N. Pope, P.K. Townsend, Phys. Lett. B 391 (1997) 39, hep-th/9607164.
M. Cvetic, S. Griffies, S.-J. Rey, Nucl. Phys. 381 (1992) 301, hep-th/9201007.
J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
C.M. Hull, Phys. Rev. D 30 (1984) 760.
C.M. Hull, Phys. Lett. B 142 (1984) 39.
C.M. Hull, Physica D 15 (1985) 230.
C.M. Hull, Phys. Lett. B 148 (1984) 297.
C.M. Hull, Class. Quantum Grav. 2 (1985) 343.
C.M. Hull, N.P. Warner, Nucl. Phys. B 253 (1985) 650.
C.M. Hull, N.P. Warner, Nucl. Phys. B 253 (1985) 675.
C.M. Hull, N.P. Warner, Class. Quantum Grav. 5 (1988) 1517.
M. Pernici, K. Pilch, P. van Nieuwenhuizen, N.P. Warner, Nucl. Phys. B 249 (1985) 381.
M. Gunaydin, L.J. Romans, N.P. Warner, Phys. Lett. B 154 (1985) 268;
M. Gunaydin, L.J. Romans, N.P. Warner, Phys. Lett. B 164 (1985) 309;
M. Gunaydin, L.J. Romans, N.P. Warner, Nucl. Phys. B 272 (1986) 598.
L. Andrianopoli, F. Cordaro, P. Fre, L. Gualtieri, Class. Quantum Grav. 18 (2001) 395, hep-th/0009048.
L. Andrianopoli, F. Cordaro, P. Fre, L. Gualtieri, Fortschr. Phys. 49 (2001) 511, hep-th/0012203.
B. de Wit, H. Nicolai, N.P. Warner, Nucl. Phys. B 255 (1985) 29.
M.J. Duff, C.N. Pope, KaluzaKlein supergravity and the seven sphere, in: S. Ferrara, J.G. Taylor, P. van
Nieuwenhuizen (Eds.), Supersymmetry and Supergravity 82, World Scientific, Singapore, 1983.
F. Englert, Phys. Lett. B 119 (1982) 339.
B. de Wit, H. Nicolai, Phys. Lett. B 148 (1984) 60.
C.N. Pope, N.P. Warner, Phys. Lett. B 150 (1985) 352.
B. de Wit, H. Nicolai, Nucl. Phys. B 231 (1984) 506.
C. Ahn, S.-J. Rey, Nucl. Phys. B 565 (2000) 210, hep-th/9908110.
C. Ahn, S.-J. Rey, Nucl. Phys. B 572 (2000) 188, hep-th/9911199.
R. Corrado, K. Pilch, N.P. Warner, hep-th/0107220.
C. Ahn, J. Paeng, Nucl. Phys. B 595 (2001) 119, hep-th/0008065.
C.V. Johnson, K.J. Lovis, D.C. Page, JHEP 0110 (2001) 014, hep-th/0107261.
C. Ahn, K. Woo, Nucl. Phys. B 599 (2001) 83, hep-th/0011121.
E. Cremmer, B. Julia, Nucl. Phys. B 159 (1979) 141.
B. de Wit, H. Nicolai, Phys. Lett. B 108 (1982) 285.
R. Gilmore, Lie Groups, Lie Algebras, and Some of Their Applications, Wiley, 1974.
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
191
E. Inonu, E.P. Wigner, Proc. Natl. Acad. Sci. USA 39 (1953) 510.
L. Castellani, N.P. Warner, Phys. Lett. B 130 (1983) 47.
F. Englert, M. Rooman, P. Spindel, Phys. Lett. B 130 (1983) 50.
N.P. Warner, Phys. Lett. B 128 (1983) 169.
F. Cordaro, P. Fre, L. Gualtieri, P. Termonia, M. Trigiante, Nucl. Phys. B 532 (1998) 245, hep-th/9804056.
H. Lu, C.N. Pope, E. Sezgin, Phys. Lett. B 371 (1996) 46, hep-th/9511203.
K. Skenderis, P.K. Townsend, Phys. Lett. B 468 (1999) 46, hep-th/9909070.
C.M. Hull, JHEP 0111 (2001) 061, hep-th/0110048.
L. Girardello, M. Petrini, M. Porrati, A. Zaffaroni, JHEP 9812 (1998) 022, hep-th/9810126.
C. Ahn, T. Itoh, hep-th/0112010, to appear in Nucl. Phys. B.
G.W. Gibbons, C.M. Hull, hep-th/0111072.
L.J. Romans, Phys. Lett. B 131 (1983) 83.
F. Gliozzi, J. Scherk, D. Olive, Nucl. Phys. B 122 (1977) 253.

Classification and one loop renormalization of

dimension six and eight operators in
quantum gluodynamics
J.A. Gracey
Theoretical Physics Division, Department of Mathematical Sciences, University of Liverpool, Peach Street,
Liverpool, L69 7ZF, United Kingdom
Abstract
We determine the complete set of independent dimension six and eight Lorentz scalar operators
in YangMills theory for an arbitrary colour group. The anomalous dimension mixing matrix is
determined at one loop. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
Quantum chromodynamics, (QCD), is widely accepted as the quantum field theory
which describes the strong interactions of the nuclear constituents. Indeed at large energies
the theory behaves as if it were virtually a free field theory allowing one to apply perturbative techniques to describe high energy parton interactions. However, a full understanding
of the strong interactions at lower energies scales is still sought. For example, the generation of quark masses and quark confinement are not fully understood and are believed to
be intimately related to the infrared properties of QCD. One approach to understand such
phenomena is to use effective field theories or models which have similar properties to the
original QCD Lagrangian. One such model is the NambuJona-Lasinio model [1], which
involves four quark interaction terms. From the point of view of standard renormalization
theory such terms are not renormalizable in four spacetime dimensions. This is readily
apparent from a simple-dimensional analysis since the canonical dimension of such interactions is six and therefore their coupling must incorporate a dimensionful scale to have a
Lagrangian of canonical dimension four. An alternative point of view of these operators is
E-mail address: jag@amtp.liv.ac.uk (J.A. Gracey).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 3 4 - 6
J.A. Gracey / Nuclear Physics B 634 (2002) 192208
193
that at large energies where the perturbative approximation is valid, the coupling of these
operators is driven to zero and they are known as irrelevant operators. However, it could
be the case that in the approach to the infrared the coupling, or the anomalous dimension
of the operator, gains a large correction to alter the canonical dimension in such a way
that they become relevant in the infrared rgime. Therefore, they would then be regarded
as sensible and important operators for understanding the phenomenology at such scales.
Whilst such operators and models or effective field theories have received wide attention
the actual connection of these models with the original QCD Lagrangian has yet to be fully
established in detail. Since one can analyse the structure of the gauge invariant composite
operators of any quantum field theory from the point of view of renormalization theory,
it is the purpose of this article to consider the operators of dimension six which one can
be built in YangMills theories. The aim is to determine the basis set of dimension six
operators and then to compute their anomalous dimensions at one loop.
There are various reasons for such a study aside from those already stated. First, we will
concentrate on YangMills theories with an arbitrary colour group since previous analyses
of this problem, we believe, have been incomplete. Therefore, it is appropriate to focus
on the gluonic sector of QCD before returning to the full theory in a later article. For
instance, the earlier work of [2] only considered dimension six operators which did not
involve covariant derivatives of the gluon field strength. We will demonstrate that there is
an extra independent operator which was omitted from that analysis. Second, whilst the
work of [3,4] considered such field strength covariant derivatives, the operators were only
considered at zero momentum. To fully treat composite operators and the determination
of their anomalous dimensions one must renormalize them with a non-zero momentum
flowing through them. As is well known doing otherwise can lead to erroneous anomalous
dimensions. (See, for example, [57].) Moreover, we will show that one cannot readily
drop operators which are total spacetime derivatives of lower-dimensional operators but
which are overall dimension six. These are crucial to preserving identities similar to the
Bianchi identities which are valid in the classical theory and which must be preserved in
the quantum theory. This is another reason for concentrating on YangMills theory since
these technical issues can become over-complicated in QCD. Further, previous calculations
[2,8], were only concerned with specific unitary colour groups. We take a more general
line here by analysing YangMills theory with a general Lie group which will allow
one, for instance, to understand the properties of dimension six operators in the large Nc
limit. Another reason for this is that the construction of the basis set of operators will not
need to appeal to the group tensor properties of a particular Lie group. For instance, the
totally symmetric Casimir, d abc , only exists in the group SU(Nc ) for Nc 3 and does
not always have a counterpart in other classical or exceptional Lie groups. Also, it is
possible to consider various group representations in the arbitrary case. Other motivations
for considering dimension six operators come from calculations such as [9,10]. In [3,4,
9], for example, dimension six operators have been shown to be important in hadronic
scattering. In [10] the structure of the mixing matrix of anomalous dimensions plays a
crucial role in the large order perturbative behaviour of physical quantities such as the
hadronic decay of the or e+ e annihilation into hadrons. Essentially the operator or
combination or operators which has the dominant eigenanomalous dimension drives the
structure of the perturbative series at large powers of the strong coupling constant and
194
is related to the Borel properties of the series. For other (gluonic) correlation functions
the dominant eigenoperator could be different and therefore it is important to have the
anomalous dimensions for the full set of dimension six operators. Finally, although we
have focused on the motivation for considering dimension six operators, we will also study
dimension eight operators at the same level. Again previous analyses in our view have not
been fully complete and therefore it is important to establish the full picture.
The paper is organised as follows. In Section 2, we discuss at length the background requirements and results for constructing the basis set of dimension six operators. This is repeated in Section 3 for the case of dimension eight operators before discussing the one loop
renormalization of all operators in Section 4. Concluding remarks are given in Section 5.
2. Classification of dimension six operators

To systematically classify dimension six and eight gauge invariant operators it is
appropriate to choose a notation where this important property is manifest. As we are
dealing with YangMills operators it seems appropriate therefore to choose a group valued
field strength, G , and gauge potential, A , where
G = Ga T a ,
A = Aa T a
(2.1)
and T a are the usual colour group generators obeying the Lie algebra of the colour group G
a b
T , T = if abc T c
(2.2)
with structure constants f abc which satisfy the usual Jacobi identity
f abe f cde + f ace f dbe + f ade f bce = 0.
(2.3)
Moreover, the covariant derivative of a group valued object X, satisfies

D X = X + ig[A , X],
(2.4)
where g is the coupling constant. Consequently, one has

[D , D ]X = ig[G , X],
(2.5)
where the field strength is defined as

G = A A + igA A .
(2.6)
Given the geometric nature of G it satisfies the Bianchi identity

D G + D G + D G = 0
(2.7)
which is a basic symmetry property of the YangMills field. In this notation, the gauge
transformations of the following entities are
i
A U A U + ( U )U
g
G U G U ,
D G U D G U ,
(2.8)
195
where U is a group valued x-dependent unitary matrix with

U U = U U = 1.
(2.9)
Hence, to construct gauge invariant operators whether Lorentz scalar or otherwise one
need only consider colour group traces of objects which transform covariantly under (2.8).
For YangMills theories such objects would be the field strength itself and any number of
covariant derivatives of it. Objects with a gauge potential clearly cannot lead to a gauge
invariant operator. To illustrate how one systematically constructs a set of linearly independent operators of a specific dimension, which satisfy the criteria of the renormalization
theorems, we consider dimension six operators in detail first. Since operators can be related
to each other by operators which involve the equation of motion, then we note at the outset
that in quantum gluodynamics, the equation of motion is
D G = 0.
(2.10)
Thus if an operator can be related by the symmetries of the theory to another operator
plus one which includes the object D G then they are regarded as being dependent [7].
The method we have followed is first to write down all possible structures involving the
objects G , D and which are of dimension six, Lorentz scalars and which are gauge
invariant by the above construction. With this basic set it is straightforward to decorate
all the slots for the Lorentz indices in all possible ways, though initially dropping those
operators involving D G . The ordinary derivative, , is allowed in this construction
as it has dimension one but due to the imposition of gauge invariance it can only appear
outside a trace of operators and hence it occurs in operators with total derivatives. We will
discuss their treatment later, as they will be important, and concentrate for the moment on
non-total derivative operators.
Clearly to have a colour singlet operator one must have at least two field strengths in
the trace which for dimension six operators implies only one trace operation is allowed.
Therefore, the three allowed structures are
Tr(G.. G.. G.. ),
Tr(D. G.. D. G.. ),
Tr(G.. D. D. G.. ),
(2.11)
where the dots indicate the location of the Lorentz indices and the covariant derivative
in any string acts only on the object immediately to the right. Although one can simply
enumerate all the cases, making use of the symmetry properties reduces the amount of
work. For instance, a common object in the structures is
D1 Dn G1 2 .
(2.12)
However, in certain cases the following index pattern will be present

D1 D Dn G2 ,
D1 D D Dn G1 2 ,
(2.13)
where we have indicated the explicit contractions. (There could be additional contractions
among the sets {1 , . . . , n } and {1 , 2 } but we focus on those illustrated explicitly.) For
the first example, applying (2.5) recursively to move the covariant derivative to the right
yields operators with a higher number of legs together with the operator
D1 Dn D G2
(2.14)
196
which has the same number of legs as the original operator. We define the number of legs
on an operator as the lowest number of gluon fields in any term when the field strength and
the covariant derivatives are written in terms of A . Clearly, (2.14) is an operator which
vanishes on an equation of motion. Therefore, the operator we began with can be written
in terms of higher leg operators plus an operator which is ignored from the point of view of
establishing linear independence [7]. From focusing on this particular structure embedded
within an operator an algorithm to determine the basis set of operators follows naturally. In
other words using symmetries we write as far as possible the lower leg operators in terms
of higher leg ones and equation of motion operators. These then become dependent and can
be dropped from the original set of potential operators. For the second operator of (2.13),
it can be written in terms of the first operator by commuting the covariant derivative D
closest to the field strength to obtain higher leg operators plus
D1 D Dn D G1 2 .
(2.15)
Using (2.7) this is related to

D1 D Dn D 1 G2 + D1 D Dn D 2 G1
(2.16)
which is of the form of the first operator of (2.13). Hence any operator involving factors of
the form of (2.13) where two Lorentz indices are contracted in a string, are not independent
and depend on higher leg operators. This is a general result which is not limited to
dimension six operators. From (2.11) it is easy to see that no member of the last set of
operators is independent. For the remaining two possibilities in (2.11) the lemma implies
there can be no index contraction in D. G.. leaving the cases

Tr D G D G ,
Tr D G D G .
Tr D G D G ,
(2.17)
The last two are related by the antisymmetry of G and either can be written using the
Bianchi identity as proportional to the first which therefore means

O62 = Tr D G D G
(2.18)
remains as the only two leg independent dimension six operator. For the remaining
structure of (2.11) there is only one way of slotting the Lorentz indices non-trivially giving
the independent three leg operator

O61 = Tr G G G .
(2.19)
By taking the trace explicitly this is related to
f abc Ga Gb Gc
(2.20)
Tr(T (a T b T c) )
where the term involving the symmetric tensor

vanishes by symmetry.
All that remains are the operators involving a total derivative acting on the trace of
two field strength operators. One might expect that such operators are unimportant as they
would not contribute when inserted in diagrams at zero momentum. However, it will turn
out that they are crucial for ensuring consistency in the one loop renormalization and preserving identities which follow from the Bianchi identity or symmetries. Indeed total derivative operators are known to be important in QCD. For example, the renormalization of the
197
= G and is
where G
axial vector anomaly involves the operator G G
is the
the usual totally antisymmetric rank four pseudotensor. As is well known G G
total derivative of the ChernSimons current. To compute the anomalous dimension of the
singlet axial current correctly one must ensure the axial anomaly equation is satisfied as
an operator equation quantum mechanically. This requires the renormalization of the total
. (See, for example, [11,12].) Therefore, in the context of our
derivative operator G G
dimension six and eight classification we also consider such operators, though only those
which are Lorentz scalar and not pseudoscalar. In addition to the earlier lemmas, we now
introduce new results which relate various operators by a total derivative operator. For example, using the distributivity property of D acting on the product of typical objects X
and Y which involve products of G and its covariant derivatives, we have
Tr(XD Y ) = Tr(XY ) Tr(D XY )
(2.21)
which implies
1
Tr(XD X) = Tr(XX).
2
(2.22)
In the first of these results it is tempting to omit the first term on the right. However, from
, can have an anomalous dimension when
the renormalization theorems it, like G G
renormalized at non-zero momentum. Moreover, the same relation provides us with the
strategy for classifying total derivative operators. In addition to beginning with these operators with the lowest number of legs, one considers those with the lowest number of
external derivatives and rearranges them to produce ones with a higher number. Two basic
structures for dimension six emerge, which are
. Tr(G.. D. G.. ),
. . Tr(G.. G.. ).
(2.23)
Ignoring operators which involve the equation of motion, this gives the following candidate
operators,

Tr G D G ,

Tr G G .

Tr G D G ,

Tr G G ,
(2.24)
The first operator is related to the third by (2.22) and is thus not independent. Likewise the
second operator is related to the first through the Bianchi identity. Finally, using the results
(2.21), (2.22) and the Bianchi identity, we have

1

Tr G D G = Tr G G + Tr G G
4
(2.25)
leaving the last operator, say, as the only independent dimension six one from the set of
total derivative operators, (2.24), where only the results (2.21) and (2.22) were applied.
It remains to check what relations emerge when these latter results are applied to the set
of operators which do not initially involve a total derivative. Therefore, if we consider
198

Tr D G D G and apply (2.21) and (2.22) then we find,

Tr D G D G

1
= Tr G G + 2 Tr G D D G 4ig Tr G G G . (2.26)
2
Hence, overall we are left with only two independent dimension six operators.
3. Classification of dimension eight operators

The procedure to classify operators of dimension eight follows that for the dimension
six case and rather than reproduce similar arguments we will concentrate on the essential
differences. First, with the higher dimension it is clear that more structures akin to (2.11)
are possible. Moreover, one has to consider operators built out of more than one colour
group trace. Considering the two leg operators first, like the dimension six case there is
only one such operator,

O821 = Tr (D D G ) D D G
(3.1)
which is a natural generalization of (2.18). All other two leg operators either involve an
equation of motion operator or can be written as (3.1) and higher leg operators. For three
leg operators there is a similar reduction in the number of possibilities though one is left
with

O831 = Tr D G D G G ,

O832 = Tr D G D G G
(3.2)
as the two independent operators. As there is now a trace over three group generators
it might be expected that the Feynman rule of each operator will involve d abc . For the
latter operator the antisymmetry of G ensures that only f abc emerges. Whilst using the
Bianchi identity on the first operator produces

1
Tr D G D G G
2
(3.3)
which likewise only involves f abc . For operators of the form Tr(D. D. G.. G.. G.. ) it might
be expected that our lemma could still allow for several independent operators. In other
words if there are no contracted indices on the field strength with two covariant derivatives
then the lemma is not applicable. This leaves the four cases

Tr D D G G G ,
Tr D D G G G ,

Tr D D G G G .
Tr D D G G G ,
(3.4)
Clearly the second and fourth operators of this set are each related to four leg operators. For
the remaining two using the Bianchi identity allows one to rewrite each as either the second
or fourth operator. Similar arguments systematically applied to the other possible structures
leave (3.2) as the only three leg dimension eight operators. Finally, for the four leg
operators which therefore involve four field strength factors and no covariant derivatives,
199
one has the possibility of a double group trace structure. Systematically enumerating the
allowed Lorentz structures simply leaves the eight basic operators

O841 = Tr G G G G ,
O842 = Tr G G G G ,

O843 = Tr G G G G ,
O844 = Tr G G G G ,

O846 = Tr G G Tr G G ,
O845 = Tr G G Tr G G ,

O847 = Tr(G G ) Tr G G ,
O848 = Tr(G G ) Tr G G . (3.5)
The remaining operators which must be considered now involve those which are total
derivatives. As before if we focus on those operators which have at least one external
derivative acting on the trace of the field strengths then the possible candidates for
independence are, after relating possible operators within the same structures by, for
example, the Bianchi identity,

Tr G G ,
Tr G G ,

Tr D G D G ,
Tr G G G ,

Tr G G G .
Tr G G D G ,
(3.6)
Of this set the last three can be related to operators which do not involve total derivatives
and therefore they are dependent and excluded from the basis. For example, if one
considers O831 and applies (2.21) we have

Tr D G D G G

= Tr D D G G G Tr D G G D G

Tr G G D G Tr G D G G .
(3.7)
The first term is related to operators which have already been shown to be independent by
a route not involving (2.21) or (2.22) whilst the second and fourth terms are equation of
motion operators. Repeating the same manipulations on O831 but integrating by parts with
the other covariant derivative first, one can relate the fourth operator of (3.7) to the fifth
and therefore neither are independent. Next by considering O832 we have

Tr D G D G G
1

1
= Tr G G G Tr G D D G G ,
(3.8)
6
2
where (2.21) has been applied twice. Therefore, to summarize the three independent
dimension eight total derivative operators are

O822t = Tr G G ,
O821t = Tr G G ,

O823t = Tr D G D G .
(3.9)
Whilst this completes the classification of the operators necessary for renormalizing
all dimension six and eight YangMills operators there are several points which still need
to be addressed. First, each of the operators we have produced only represents one in a
tower of such gauge invariant colour singlet operators. For instance, one can introduce
200
tensor products of group generators into each trace without spoiling gauge invariance or
altering the dimension of the operator. The former property follows from the fact that
under the gauge transformations an operator will contain products of U and U where U
is a group element. If group generators are present the cancellation of U factors via (2.9)
is obstructed. However, it is possible to show (infinitesimally) that
U T aU U T aU = T a T a
(3.10)
for all U . Moreover,

f abc U T a U U T b U U T a U = f abc T a T b T c .
(3.11)
Hence, one can in principle produce an infinite set of additional operators from the
basic set we have constructed. Examples include Tr(T a T b G T a G T b G ) and
Tr(T a D G T a D G ). However, for the latter one can easily reduce this to

CF 12 CA Tr T a D G T a D G
(3.12)
using T a T a = CF , where CF is the usual rank two Casimir and CA is its value in the
adjoint representation. For the former the manipulation of the group generators would lead
to other group Casimirs, [13]. Although this increases the number of possible operators
to consider for an arbitrary gauge group, we will not classify them but regard them as
derivable from the base set. It is worth noting that in the explicit renormalization each
additional operator will in fact be generated. This is because inserting them in the Greens
function group generators are introduced at each vertex through the use of the Lie algebra,
(2.2). However, the main reason we do not need to consider such operator generalizations
is that at least for the classical Lie groups the tensor product T a T a can, in principle, be
decomposed by a group identity. For instance, in SU(Nc )

1
1
a a
TI J TKL = I L J K
(3.13)
I J KL .
2
Nc
The other classical Lie groups have similar relations which allows one to rewrite operators
with generator strings in terms of the base set of operators. However, the relations are
not necessarily related in a group invariant way. In other words one would have to
consider each set of (classical) Lie groups separately. As we are interested in performing
a calculation without reference to particular groups we will allow for the possibility of
these new operators being generated at each loop order. Therefore, in this context we need
to define extra dimension eight operators since these will be generated in our one loop
renormalization. They are

O842a = Tr G G T a G G T a ,

O844a = Tr G G T a G G T a
(3.14)
and their relation to (3.5) is readily determined for G = SU(Nc ) using (3.13), for example.
One final consideration needs to be addressed and that is the classification of the
dimension six and eight operators which vanish on the equation of motion. This is for an
important reason. We will renormalize the operators by inserting them in a Greens function
201
where the external legs are multiplied by the physical polarization vectors. However, not
only will operators which we have classified be generated but operators which vanish on
the equation of motion. Not all such operators have a zero Feynman rule when the external
legs are put on-shell and therefore they can occur with a non-zero pole with respect to the
regularization. Hence, one has to classify such operators and produce a basis of them. Once
one has determined the set of operators which a specific operator mixes into, the explicit
mixing matrix of anomalous dimensions of physical operators is determined by excluding
the equation of motion operators. Given this technical point we have also classified all
dimension six and eight operators which involve the equation of motion (2.10). In this
construction we proceed as before writing down all possible Lorentz structures and using
the symmetries to remove dependent operators. However, unlike previously we insist that
the structures all contain a factor of the form D G . This means that for dimension six
there are only two leg operators and dimension eight at most three leg operators. As the
procedure then parallels our previous construction we merely write down the operators
which form the basis for this sector. For dimension six we have

Tr G D D G
Tr D G D G ,
(3.15)
and for dimension eight the following form the basis

Tr G D G D G ,
Tr G G D D G ,

Tr D D G D D G ,
Tr D G D D D G ,

Tr D G D D D G ,
Tr D D G D D G ,

Tr G D D D D G .
(3.16)
It is worth noting that whilst these are the full set, some but not all vanish for all legs when
put in a Greens function where the external legs are on-shell which therefore reduces the
number one has to consider for an operator to mix into.
4. One loop renormalization

Having established the sets of independent dimension six and eight operators which
forms the basis, we can now determine their anomalous dimensions. This is achieved
by respecting the standard renormalization theorems, [57]. In essence each operator is
inserted in a Greens function with the same number of external legs as the operator itself.
For gauge invariant operators this translates into the lowest number of legs in the operator
since the covariant derivative and gluon field strength involve terms with various numbers
of legs. When the operator in inserted it has a non-zero external momentum flowing into
it. If it is inserted at zero momentum one cannot readily resolve the resulting set of
operators it mixes into straightforwardly. Indeed if one considers the simple example of
the renormalization of (Ga )2 as discussed in [7], an incorrect value for the -function
would be obtained if one took the naive values for the renormalization which emerged.
More detailed discussion of this issue has been given in [1416]. Therefore, we will insert
each operator with a non-zero momentum which means that whilst each Greens function
has n gluon legs external, it is in fact an (n + 1)-point function due to the extra external
202
Fig. 1. Gluonic Greens function with operator insertion.
momentum. For operators which are a total derivative this property is important. The
general structure of the Greens function is illustrated in Fig. 1. In addition to inserting
at non-zero momentum we exclude the possibility of gauge variant operators emerging
in the mixing matrix by ensuring the external gluons are on-shell. Thus for the external
gluon A a (p) in momentum space, we multiply the Greens function by (p), where
are the spin-1 physical polarization vectors, and set p p = 0 and (p)p = 0 for
each such gluon. Consequently, when each operator is inserted the resulting output will
involve a large number of terms involving different combinations of the factors i j , pi j
(i = j ) and pi pj (i = j ) where 1 i,j n. To resolve these into the operator basis we
have computed the Feynman rules for each independent operator with the same number
of gluon legs as the original operator and multiplied by (p) subjecting them to the
same restrictions as above. The full set with arbitrary coefficients is compared with the
operator output and the parameters fixed to cancel off all the one loop divergences. Given
that YangMills is renormalizable and that we have a basis set of operators then there is
no redundancy or overconstraint in the computation of the parameters which are therefore
uniquely determined.
To find the pole structure of each Greens function we have chosen to compute using
dimensional regularization where the spacetime dimension is d = 4 2. The ultraviolet
infinities will emerge as simple poles in at one loop. However, since we are working with
Greens functions which are at least 3-point, (in the case of a two leg operator insertion),
we cannot use the M INCER algorithm, [17], which applies only to 2-point propagator type
integrals. Moreover, if one uses massless propagators to compute, say, 4-point or higher
Greens functions then there is a danger of obtaining spurious infrared infinities which in
dimensional regularization are inseparable from the ultraviolet ones we seek. Instead we
are forced to infrared regularize our one loop integrals by introducing a mass m in the
gluon propagator which acts as an infrared cutoff. Recently, a similar approach has been
used in [18,19] to systematically compute analogous Greens functions for dimension five
operators in QCD. Therefore, we use for our gluon propagator,

k k
1
(1
,
(4.1)
(k 2 + m2 )
(k 2 + m2 )
where m appears naturally as an infrared regularization. Moreover, we use a covariant
gauge fixing with parameter . To ensure that our renormalization procedure with such
203
a (gauge symmetry violating) propagator is valid, we have checked the full one loop
renormalization of YangMills using (4.1) and a mass independent renormalization scheme
[20,21]. This is important since the operators we are interested in are composite and
therefore each field present in the operator will be renormalized requiring the wave function
renormalization. For example,

= ZA ZDGDG Tr D G D G
Tr D Go D G
(4.2)
o
1/2
where Ao = ZA A and the subscript, o , denotes the bare quantity. In this expression ZA
is the usual gauge dependent gluon wave function renormalization and ZDGDG is related
to the gauge independent anomalous dimension of the particular operator which we seek,
if we ignore operator mixing for the moment. Therefore, by computing with a non-zero
we have a check which is that the operator renormalization constants which emerge
must be gauge independent. Further, by first renormalizing YangMills at one loop, this
allows us to check that the Feynman rules we use are consistent. This is important since
given the nature of the operators we are considering, whose Feynman rules can involve
over four thousand terms,1 we have used a symbolic manipulation approach. The Feynman
diagrams are generated with Q GRAF, [22], and converted into a format recognizable by the
language F ORM, [23]. We have written a programme to convert the Q GRAF output into a
typical Feynman integral with propagators and vertices substituted. The group theory for
each graph is performed before the integrals are evaluated. This is achieved by reducing
each diagram to the corresponding vacuum bubble graph by systematically rewriting the
propagators using the result, [18,19],
1
(2kp p2 )
1
= 2
+ 2
.
(4.3)
2
2
2
((k p) + m ) (k + m ) (k + m2 )((k p)2 + m2 )
This relation is exact but for mass regularized propagators all terms bar the first involve
an external momentum in the numerator. If one continually repeats the substitution
with the termination rule that O(p4 ) terms, say, can be dropped due to YangMills
renormalizability, then the procedure will stop leaving only one loop massive vacuum
bubbles which are easily calculated due to

(n 12 d) 2 1 dn
1
=
(4.4)
m 2
1
(k 2 + m2 )n
d
2
(4)
(n)
k

for any n > 0 where k = d d k/(2)d . We have implemented the above algorithm in
F ORM, [23]. Finally, for operators which mix under the renormalization there will be extra
terms in relations similar to (4.2). In general we have
Ooi = Zij Oj
(4.5)
which gives the mixing matrix of anomalous dimensions

ij (a) =
ln Zij .
(4.6)
1 This number represents the number of terms in the four and five legs parts of the operator O
844 . At one loop
the six leg part of this operator is not required for the anomalous dimension.
204
Having summarized our method and the general formalism, we now record our results.
First, for the two independent dimension six operators there is no mixing and we find that
61 (a) =

CA
a + O a2 ,
2
62 (a) =

11
CA a + O a 2
3
(4.7)
where a = g 2 /(16 2 ) and the subscript on the anomalous dimension corresponds to the
appropriate operator. The one loop expression for 61 (a) has been computed previously in
[2,8] and we note that we get consistency with both calculations which were performed in
the background field gauge. For completeness we note that in both instances the dimension
six operator which was considered were multiplied by powers of a. Including the respective
contributions from the -function allows one to compare the final expressions for the
anomalous dimension of each operator. The anomalous dimension for O62 is new and
is the same as the one loop YangMills -function, [24,25].
For the dimension eight operators the full one loop mixing matrix of anomalous
dimensions partitions into blocks defined by the number of legs on the operator. Therefore,
we list only the entries in each of the blocks. First, for the single two leg operator

11
CA a + O a 2 .
3
For three legs, we find
821,821(a) =
(4.8)

7
1
831,831(a) = CA a + O a 2 ,
831,832(a) = CA a + O a 2 ,
6
3

1
832,832(a) = CA a + O a 2 ,
832,831(a) = O a 2 ,
(4.9)
2
so that this sub-matrix is triangular at this order. For the four leg dimension eight operators
the mixing matrix is further divided into sectors defined by the number of colour traces in
the original operator. Thus, for the single trace operators we have

25
1
CF 7CA a + O a 2 ,
841,842(a) = CA a + O a 2 ,
841,841(a) =
3
3
2
841,843(a) = (16CF 11CA )a + O a ,

52
841,844(a) =
CF 11CA a + O a 2 ,
3

23
2
841,841a (a) = CF a + O a 2 ,
841,842a (a) = CF a + O a 2 ,
3
3
2

52
841,843a (a) = 16CF a + O a ,
841,844a (a) = CF a + O a 2 ,
3

50
4
CF 8CA a + O a 2 ,
842,841(a) =
842,842(a) = CA a + O a 2 ,
3
3

2
56
38
CF CA a + O a ,
842,843(a) =
3
3

842,844(a) = (16CF 10CA )a + O a 2 ,
205

2
842,842a (a) = 16CF a + O a 2 ,
842,841a (a) = CF a + O a 2 ,
3

56
842,843a (a) = CF a + O a 2 ,
842,844a (a) = 16CF a + O a 2 ,
3

47
843,842(a) = O a 2 ,
843,841(a) = 12CF CA a + O a 2 ,
6

34
CF 6CA a + O a 2 ,
843,843(a) =
3

14
CF 2CA a + O a 2 ,
843,844(a) =
3

23
13
843,841a (a) = CF a + O a 2 ,
843,842a (a) = CF a + O a 2 ,
3
3
2

34
14
843,843a (a) = CF a + O a ,
843,844a (a) = CF a + O a 2 ,
3

3
2

19
844,841(a) = 6CF CA a + O a ,
844,842(a) = CA a + O a 2 ,
6

14
11
844,843(a) =
CF CA a + O a 2 ,
3
3

10
13
CF CA a + O a 2 ,
844,844(a) =
3
3
2

5
23
844,841a (a) = CF a + O a ,
844,842a (a) = CF a + O a 2 ,
3
3
2

14
10
844,843a (a) = CF a + O a ,
(4.10)
844,844a (a) = CF a + O a 2 .
3
3
To simplify the expressions for the anomalous dimensions of the double colour trace
operators we have introduced the intermediate operators
O8410 = f abe f cde Ga Gb Gc Gd ,
O8411 = f abe f cde Ga Gb Gc Gd
(4.11)
which are not independent as they are related by

O842 = O841 12 T (R)O8411 ,
O844 = O843 + 12 T (R)O8410
where Tr(T a T b ) = T (R) ab . Therefore, we have

22
CA a + O a 2 ,
3

845,846(a) = O a 2 ,
845,847(a) = O a 2 ,

845,848(a) = O a 2 ,
845,8410(a) = 28CA a + O a 2 ,

11
846,845(a) = CA a + O a 2 ,
845,8411(a) = 2CA a + O a 2 ,
6

846,846(a) = O a 2 ,
846,847(a) = O a 2 ,
845,845(a) =
(4.12)
206

846,848(a) = O a 2 ,

4
846,8410(a) = CA a + O a 2 ,
3
2
11
846,8411(a) = CA a + O a ,
3

1
28
847,845(a) = CA a + O a 2 ,
847,846(a) = CA a + O a 2 ,
3
3

2
847,848(a) = 8CA a + O a 2 ,
847,847(a) = CA a + O a 2 ,
3

34
25
847,8410(a) = CA a + O a 2 ,
847,8411(a) = CA a + O a 2 ,
3
3
2

1
14
848,846(a) = CA a + O a 2 ,
848,845(a) = CA a + O a ,
6
3

10
848,847(a) = 4CA a + O a 2 ,
848,848(a) = CA a + O a 2 ,
3
2

2
11
848,8411(a) = CA a + O a 2 .
848,8410(a) = CA a + O a ,
3
3
Finally, for the total derivative operators we have
(4.13)

11
CA a + O a 2 .
(4.14)
3
Having completed the full one loop renormalization we are now in a position to examine
some of the results used in constructing the initial set of independent operators. For
instance, the operators which involve a total derivative were related via the identities (2.21)
and (2.22). By considering O62 these imply,

Tr G D D G = Tr G D G Tr D G D G
(4.15)
821t,821t (a) = 822t,822t (a) = 823t,823t (a) =
which was used in [3] but with the first term on the right omitted. However, one can
compute the one loop renormalization of each operator in this result following the
procedures we have discussed previously. In particular, each operator is inserted in a gluon
2-point function with a non-zero momentum flowing through the operator. It turns out that
the total derivative operator of (4.15) has an anomalous dimension which is the same as
O62 whilst their is no renormalization of the other operator. This is consistent since it is
related to operators which vanish on the equation of motion or which involve higher legs
and so has no two leg projection. Therefore, it would appear that omitting total derivative
operators in calculations could lead to erroneous results.
5. Discussion
We conclude with various remarks. First, we have computed the one loop anomalous
dimensions of a set of linearly independent gauge invariant dimension six and eight
operators for an arbitrary Lie group. Whilst we have reproduced results that had been
derived previously we believe our calculation is more comprehensive since the systematic
classification of all operators has been performed for the first time and operators which
207
are total derivatives have been considered. Their effect cannot be neglected since operators
are renormalized at non-zero momentum. Further, it transpires that there is an additional
operator of dimension six which appears to have been omitted from phenomenological
considerations. Second, our study has laid the foundation for extending the work in various
directions. For instance, with the basis of independent physical operators it ought to be
possible to renormalize these operators at two loops in YangMills theory. Also, given
the fact that several dimension eight operators at one loop have the same anomalous
dimensions it would be interesting to see if the degeneracy is lifted at this order. Moreover,
since there are more independent operators in the basis than would previously appear to
have been considered it would be worthwhile to extend both the dimension six and eight
bases to QCD when quark fields are included. Although at one loop previous analyses
would seem to suggest a lack of mixing between the quark and gluon sectors the extension
to two loops would also be worth pursuing since examination of the Feynman diagrams
which are generated at two loops suggest that there will be mixing. In a related context
the operators we have focused on have all been Lorentz scalars. Given the recent interest
in CP -violation and the role dimension six operators play in probing physics beyond the
standard model, [26], it will be important to repeat the one loop calculations for Lorentz
pseudo-scalar operators. Indeed it would be interesting to see if an independent analogue
of O62 exists. We hope to return to these issues in future work.
Acknowledgements
The author thanks Dr. D.B. Ali and Prof. L. Dixon for useful discussions and the latter
for pointing out Ref. [9].
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
Y. Nambu, G. Jona-Lasinio, Phys. Rev. 122 (1961) 345.

A.Yu. Morozov, Sov. J. Nucl. Phys. 40 (1984) 505.
E.H. Simmons, Phys. Lett. B 226 (1989) 132.
E.H. Simmons, Phys. Lett. B 246 (1990) 471.
S.D. Joglekar, B.W. Lee, Ann. Phys. 97 (1976) 160.
S.D. Joglekar, Ann. Phys. 100 (1976) 395;
S.D. Joglekar, Ann. Phys. 108 (1977) 233;
S.D. Joglekar, Ann. Phys. 109 (1977) 210.
J.C. Collins, Renormalization, Cambridge Univ. Press, 1984.
S. Narison, R. Tarrach, Phys. Lett. B 125 (1983) 217.
L.J. Dixon, Y. Shadmi, Nucl. Phys. B 423 (1994) 3;
L.J. Dixon, Y. Shadmi, Nucl. Phys. B 452 (1995) 724.
M. Beneke, V.M. Braun, N. Kivel, Phys. Lett. B 404 (1997) 315.
J. Kodaira, Nucl. Phys. B 165 (1980) 129.
S.A. Larin, Phys. Lett. B 303 (1993) 113.
T. van Ritbergen, J.A.M. Vermaseren, S.A. Larin, Phys. Lett. B 400 (1997) 327;
T. van Ritbergen, A.N. Schellekens, J.A.M. Vermaseren, Int. J. Mod. Phys. A 14 (1999) 41.
R. Hamberg, W.L. van Neerven, Nucl. Phys. B 379 (1992) 143.
W. Furmanski, R. Petronzio, Phys. Lett. B 97 (1980) 437;
E.G. Floratos, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 152 (1979) 493.
208
[16] J.C. Collins, R.J. Scalise, Phys. Rev. D 50 (1994) 4117;

B.W. Harris, J. Smith, Phys. Rev. D 51 (1995) 4550.
[17] S.G. Gorishny, S.A. Larin, L.R. Surguladze, F.K. Tkachov, Comput. Phys. Commun. 55 (1989) 381;
S.A. Larin, F.V. Tkachov, J.A.M. Vermaseren, The Form version of Mincer, NIKHEF-H-91-18.
[18] M. Misiak, M. Mnz, Phys. Lett. B 344 (1995) 308.
[19] K.G. Chetyrkin, M. Misiak, M. Mnz, Nucl. Phys. B 518 (1998) 473.
[20] D.J. Gross, F.J. Wilczek, Phys. Rev. Lett. 30 (1973) 1343.
[21] H.D. Politzer, Phys. Rev. Lett. 30 (1973) 1346.
[22] P. Nogueira, J. Comput. Phys. 105 (1993) 279.
[23] J.A.M. Vermaseren, F ORM version 2.3, CAN Amsterdam, 1992.
[24] H. Kluberg-Stern, J.B. Zuber, Phys. Rev. D 12 (1975) 467.
[25] J.C. Collins, A. Duncan, S.D. Joglekar, Phys. Rev. D 16 (1977) 438.
[26] S. Weinberg, Phys. Rev. Lett. 63 (1989) 2333.

Brane splitting via quantum tunneling

Selena Ng, Malcolm Perry
Department of Applied Mathematics and Theoretical Physics, Centre for Mathematical Sciences,
Wilberforce Road, Cambridge CB3 0WA, United Kingdom
Received 19 November 2001; received in revised form 25 February 2002; accepted 15 April 2002
Abstract
We study the two-centred AdS7 S 4 solution of eleven-dimensional supergravity using the
Euclidean path-integral approach, and find that it can be interpreted as an instanton, signalling
the splitting of the throat of the M5-brane. The instanton is interpreted as indicating a coherent
superposition of the quantum states corresponding to classically distinct solutions. This is a surprising
result since it leads, through the AdS/CFT correspondence, to contradictory implications for the dual
(2, 0) superconformal field theory on the M5-brane. We also argue that similar instantons should exist
for other branes in ten- and eleven-dimensional supergravity. The counterterm subtraction technique
for gravitational instantons, which arose from the AdS/CFT correspondence, is examined in terms
of its applicability to our results. Connections are also made to the work of Maldacena et al on antide Sitter fragmentation. 2002 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Sq; 04.70.Dy
1. Introduction
Several years ago, Brill [1] considered an instanton connecting AdS2 S 2 to a geometry
containing two or more AdS2 S 2 centres. Since AdS2 S 2 is the near-horizon geometry of
the extremal ReissnerNordstrm black hole, Brill argued that the instanton describes the
semi-classical splitting of the AdS2 S 2 throat into two or more throats. It is well known
that the extremal ReissnerNordstrm black hole, and independently its throat AdS2 S 2 ,
can be considered as supersymmetric solutions of four-dimensional supergravity. In ten
and eleven dimensions there exist analogous supersymmetric solutions of the supergravity
action with anti-de Sitter near-horizon geometry, such as the D3 (AdS5 S 5 ), M2 (AdS4
E-mail addresses: s.k.l.ng@damtp.cam.ac.uk (S. Ng), m.j.perry@damtp.cam.ac.uk (M. Perry).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 4 6 - 2
210
S. Ng, M. Perry / Nuclear Physics B 634 (2002) 209229
S 7 ) and M5 (AdS7 S 4 ) branes. The natural question which then arises is whether the
fragmentation process also occurs here in the context of string or M-theory.
One might argue that fragmentation is forbidden on the grounds that each of
these AdSp+2 S Dp2 spacetimes is a supersymmetric solution of supergravity in
D dimensions, and thus stable. However, rather than indicating an instability, the existence
of such an instanton indicates a quantum superposition of states. One expects that each
classical solution of string or M theory is an approximation to the corresponding quantum
state. Here we will consider the family of classical solutions with n AdSp+2 S Dp2
centres1 (for given p and D), which can be thought of as analogous to the classical vacua
of YangMills theory labelled by winding numbers n. In four-dimensional YangMills
theory, the existence of instantons connecting two vacua with two given winding numbers
is interpreted as meaning that the true quantum vacuum, often referred to as the theta
vacuum, is a linear superposition of the quantum states associated to each of the classical
vacua. By analogy, we should interpret the instanton as indicating that the quantum state
|n associated to the n-centred AdSp+2 S Dp2 geometry is not necessarily an exact
eigenstate of the M-theory Hamiltonian. Instead we expect an exact eigenstate to be
a coherent superposition of all the states {|n }.
However, a second and more serious objection to higher-dimensional AdS fragmentation is due to the AdS/CFT correspondence [35]. According to this duality, AdS2 is dual
to quantum mechanics, and fields in quantum mechanics do not have well-defined vacuum
expectation values, that is, they can fluctuate. Hence one would expect that the wavefunction describing the quantum state should be constructed as a superposition of each of the
different classical vacua, which is in agreement with the existence of Brills instanton. On
the other hand, higher-dimensional AdS spaces are dual to quantum field theories. Scalar
fields in field theories can, in contrast to quantum mechanics, be given fixed vacuum expectation values which are classical and constant. Since it costs an infinite action in field
theory to change such vacuum expectation values dynamically, superposition of different
exact vacua in the moduli space of the theory is forbidden. Translating to the AdS side, this
would seem to imply that higher-dimensional analogues of the Brill instanton should not
be allowed, if the AdS/CFT correspondence holds true.
Having established the above objections, it is somewhat surprising that we do identify
in this paper, analogously to the four-dimensional case, an instanton which interpolates
between the near-horizon AdS7 S 4 geometry of the M5-brane and the near-horizon2
two-centred AdS7 S 4 geometry of the two-centred M5-brane. We postpone discussion on
the resolution of this puzzle to the final section.
The standard procedure for identifying instantons, which we follow, is to use the
Euclidean path integral approach to quantum gravity [6]. We will be interested in
contributions to the zero-temperature vacuum amplitude, given by the path integral

Z = D[]eI []
(1.1)
1 We will henceforth refer to these as n-centred AdS

Dp2 , although note that the geometry for
p+2 S
n = 1 is not strictly a direct product, only approaching one-centred AdSp+2 S Dp2 asymptotically.
2 We mean this in the sense of [2] where the limit L 0 is taken, and L is the Planck length (LD2 =
p
p
p
GD ).
211
over all fields which are real on the Euclidean section and with boundary conditions
appropriate to the zero-temperature vacuum, and where I is the Euclidean action. In the
semi-classical approximation, these contributions correspond to gravitational instantons,
that is, nonsingular and geodesically complete solutions of the Euclidean equations of
motion with finite action.
The Euclidean action is defined, for a metric gab and (p + 1)-form gauge potential Ap+1
on a D-dimensional manifold M with boundary, as

1
1
2
Fp+2
I = 2
dDx g R
2 (p + 2)!
D
M
2
2
(1.2)
d D1 x K h I0 [h].
D
M
R is the Ricci scalar of the metric, Fp+2 = dAp+1 is the (p + 2)-form field strength, hab is
2 = 16G where
the induced metric on the boundary, K is the extrinsic curvature, and D
D
GD is the D-dimensional gravitational constant. The surface integral is necessary in order
to obtain an action which depends only on first derivatives of the metric and not normal
derivatives to the boundary [7,8], and ensures that the variational principle is well-defined.
Finally, I0 [h] is another surface term which depends only on the induced metric hab , and
whose purpose will become clearer below.
Since the volume of anti-de Sitter space is infinite, the action (1.2) evaluated on an
asymptotically AdSp+2 S Dp2 space generally diverges. A common solution to this
problem is to first regularise the action by imposing cut-offs, and then subtract the action
I0 of a background which acts as a zero-point and contains the same infrared divergences
[9,10]. This zero-point action must be written as a surface integral and corresponds exactly
to the term I0 [h] in (1.2) above. However, this background subtraction technique is not in
general well-defined. If for a particular geometry there exists a reference background with
the same intrinsic metric on the boundary, the process works. Unfortunately, one cannot
always find such a reference background. This is indeed the case for the multi-centred
AdSp+2 S Dp2 spacetimes.
The counterterm approach [1113], which was motivated by the AdS/CFT correspondence, provides an alternative and well-defined procedure. Since the partition function of
the string theory is conjectured to be equal to the generating functional of the dual conformal field theory, one can remove the divergences in the supergravity action by adding
local counterterms on the boundary, giving a manifestly-finite, renormalised on-shell action. These local counterterms are proportional to surface integrals of the induced metric,
and hence also take the form of the term I0 [h] in (1.2). As a side-remark, a consequence
of this is that any interpolating, nonsingular, asymptotically AdS solution of the Euclidean
field equations will have finite action, and can be interpreted as a gravitational instanton.
Much of the literature on using counterterm subtraction in Euclidean quantum gravity
has focused on gauged supergravities obtained by reduction on a sphere of the ten- or
eleven-dimensional theory. However, the multi-centred AdSp+2 S Dp2 solutions are
warped geometries, due to the choice of direction connecting the two centres. This means
that, as far as we are aware, the D-dimensional theory cannot be consistently reduced to
212
a (p + 2)-dimensional gauged supergravity, and thus the existing counterterms in [14] for
asymptotically AdS spaces cannot be faithfully applied. An alternative would be to attempt
to formulate the counterterms directly in the higher dimensions, and recent progress in this
area has been made in [15,16]. This will be discussed further in Section 4.
In the next section we will review the work of Brill on AdS2 S 2 , and in Section 3 we
present the AdS7 S 4 instanton for the M5-brane using background subtraction. We will
also compare our results with the claim in a paper by Maldacena et al. [2] that brane
creation by the field strength in supersymmetric AdS spaces occurs only for AdS2 . In
Section 4 we examine the limitations of the background subtraction procedure, and discuss
how these might be overcome by suitable counterterm regularisation. Section 5 contains
preliminary results for the M2- and D3-branes, and a more conclusive result for the D1/D5
system, including a short discussion on correctly accounting for the contribution from selfdual field strengths to the instanton action. We also briefly speculate on nonconformal
D-branes and the NS5-brane. In the concluding section, we comment that our result for the
M5-brane implies the superposition of distinct quantum vacua in the dual six-dimensional
N = (2, 0) superconformal field theory, in contradiction to the cluster decomposition
principle, and speculate on its resolution. Note that all quantities, unless otherwise stated,
will be presented in Euclidean signature.
2. The Brill instanton

For a one-form potential and metric in four dimensions, the supergravity action (1.2)
reduces to

1
1
2
I = 2 d 4x g R
(2.1)
F22 2
d 3 x K h I0 [h].
2 2!
4
4
M
One family of solutions to the equations of motion of this action is the metric and two-form
field strength

ds 2 = H 2 dt 2 + H 2 ds 2 E3 ,
(2.2)
F2 = 2 !3 dH,
(2.3)
where H is a harmonic function on E3 and !3 is the Hodge dual also on E3 . One recovers
(one-centred) AdS2 S 2 from the metric (2.2) by choosing
H=
b
,
|x|
(2.4)
where b is a constant and x represents Cartesian coordinates on E3 . The apparent

singularity at x = 0 is a nonsingular horizon of the metric (2.2). In fact, the metric
represents the near-horizon limit of the extremal ReissnerNordstrm black hole with
charge and mass b. Since the gravitational attraction between extremally charged black
holes is balanced by the electric repulsion, another choice is the multi-centred harmonic
213
function
H=
n

k=1
bk
.
|x x k |
(2.5)
This corresponds to n black holes each of charge (of equal sign) bk centred at x = x k .
Geometrically one has a spacetime which tends asymptotically to AdS2 S 2 with radius
bk whenever one of these centres is approached, that is, as x x k , and to AdS2 S 2
with radius b1 + + bn as the outer boundary |x| is approached. This n-centred
AdS2 S 2 spacetime thus interpolates between the n+1 AdS2 S 2 spaces. In the following
we will restrict ourselves to the two-centred case n = 2, however, it is straightforward to
extend the analysis to general values of n.
Let us now, following Brill [1], compute the Euclidean action (2.1) for the twocentred AdS2 S 2 geometry. As mentioned in the introduction, one expects the result
to be infrared-divergent since AdS is noncompact. We will thus background subtract the
reference action of one-centred AdS2 S 2 in order to obtain a finite, meaningful result.
The metric equation of motion demands that the Ricci scalar vanishes in the action
(2.1), reducing the bulk term to an integral of the two-form field strength. The sourcefree equation of motion d F2 = 0 for the field strength implies that one can write
for a one-form potential A.
So the bulk term can be converted into
2 F2 = d A,
F

4 1 2
(2.6)
d x g F2 = F2 F2 = F2 d A =
F2 A + C,
2!
M
since dF2 = 0 by the Bianchi identity. Note that although A is only specified up to a gauge
transformation, the surface integral above is obviously gauge-invariant. Here C represents
Since we are only
finite contributions to the action resulting from the nonsmoothness of A.
interested in the finiteness of the action rather than its specific value, we will not explicitly
evaluate these contributions here.
In spherical coordinates on E3 such that

ds 2 E3 = dr 2 + r 2 d 2 + sin2 d 2
(2.7)
the two-centred harmonic function takes the form
b1 b2
H=
+ ,
r1
r2
where
r12 = r 2 + a 2 + 2ar cos ,
r22 = r 2 + a 2 2ar cos .
(2.8)
(2.9)
The two-form field strength, its dual and associated one-form potential are given by
F2 = 2r H r 2 sin d d + 2 H sin d dr,
2 = 2H 2r H dt dr + 2H 2 H dt d,
F
A = 2H 1 dt.
Since the boundary of AdS lies at r = in these coordinates, we need to regularise the
action by introducing an upper cutoff R on the radial integration, and an upper cutoff
214
V1 on the temporal integration, then taking the limit R followed by V1 after

performing background subtraction. The unit outward normal to the regularised boundary
surface r = R is n = H 1 r , giving the extrinsic curvature term

K h n h = r 2 H 1 r H + 2r sin .
(2.10)
Thus the regularised action can be written in these coordinates as

1
C
2
4 I (R) =
F2 A 2
d 3 x K h 42 I0 [h] +
2
2
M
M

C
d 3 x r + r 2 H 1 r H sin + .
= 4
2
(2.11)
r=R
Note that this vanishes identically for the one-centred geometry, making the background
subtraction trivial. Substituting in the value of the two-centred harmonic function (2.8) and
expanding in powers of 1/R about 1/R = 0, we obtain the result

C
4 I (R) = 16V1 R R + O R 1 + ,
2
which is finite in the limit R !
There are several points to note here. Firstly, although we are only interested in whether
or not the action is finite and not in its exact value, we should point out that, as noted
by Brill [1], the finite contributions result in the action giving the expected Bekenstein
Hawking entropy S = A/4, where A is the nonzero horizon area of the Reissner
Nordstrm black hole. The contribution to the path integral is then roughly exp(2I /h )
which is infinitely suppressed, as expected, in the classical limit h 0.
Secondly, note the particular interpretation of the Brill instanton. In Lorentzian
signature, the solution is static in time and inhomogeneous in space. In the Euclidean
continuation, however, one can interpret the solution as being static in space and
inhomogeneous in time (with a suitable redefinition of the coordinates as in [1]), since
each spacetime direction now carries the same sign. The instanton thus describes being
in an initial state (the one-centred universe) in one asymptotic region, and rolling to the
final state (the two-centred universe) asymptotically. Both the initial and final states, being
supersymmetric, are degenerate minima of the action of the theorydiffering from the
so-called bounce instanton solutions in which the solutions are saddlepoints and not
genuine minimaand thus the quantum system is expected to be a superposition of these
two states.
Thirdly, we have taken the magnetic solution (2.10) for the field strength F2 . We could
2 . This leads to a real instanton since in addition to the
also have chosen its electric dual F
Wick rotation tL tE = itL , one analytically continues the electric charge QL QE =
iQL , thus ensuring that fields which are real in Lorentzian space correspond to real fields
on the Euclidean section [6].
Finally, note that the finite action is independent of the distribution of the charges, since
the action vanishes independently of the choice of harmonic functionthis agrees, by
taking the limit |b1 | |b2 |, with the result of Maldacena et al. [2] where a one-centred
AdS2 S 2 universe was shown to fragment into a macroscopic universe and a microscopic
215
brane. This can also be viewed as brane creation by the two-form field strength F2 with
charge proportional to b1 .
3. The M5 throat instanton

As mentioned in the introduction, one might expect that a similar instanton exists in
the context of ten- or eleven-dimensional supergravity. It turns out that the simplest case
to consider is the throat of the M5-brane in eleven-dimensional supergravity, since it is
similarly nonsingular and couples magnetically to the three-form potential A3 in the theory.
The relevant bosonic action is

1
2
11
2
F + F4 F4 A3
11 I = d x g R
2 4! 4
M
M
10
2
2 d x K h 11 I0 [h].
(3.1)
M
We are interested in the solution

ds 2 = H 1/3 ds 2 E6 + H 2/3 ds 2 E5 ,
F4 = !5 dH.
(3.2)
(3.3)
is now a harmonic function on E5
is the Hodge dual on E5 . The usual M5-brane
H
and !5
solution is obtained by taking H = 1 + b3 /|x|3 where x are Cartesian coordinates on E5 ,
b is a constant, and the apparent singularity at x = 0 again corresponds to a horizon. If one
discards the requirement of asymptotic flatness, then one can go directly to the throat or
near-horizon geometry AdS7 S 4 by setting
H=
b3
,
|x|3
(3.4)
and similarly for the n-centred geometry. The constant b3 is now proportional to the charge
or the number of M5-branes located at x = 0, and determines the radius l of AdS7 by
l = 2b.
We would like to evaluate the supergravity action for the two-centred AdS7 S 4
geometry analogously to the previous section. Choosing spherical coordinates on E5 ,

ds 2 E5 = dr 2 + r 2 d 2 + sin2 d32 ,
(3.5)
where d32 is the metric on the unit 3-sphere, we can write the two-centred harmonic
function as
H=
b13
r13
b23
r23
with r1 , r2 defined as in (2.9). The metric equation of motion sets

R=
1
F 2.
6 4! 4
(3.6)
216
As in the previous section, the bulk term can be converted into a surface integral. Noting
that

1 4 1
r H r H + 4r 3 sin3 3 ,
K h=
3

F4 = r H r 4 sin3 d vol S 3 + H r 2 sin3 dr vol S 3 ,

7 F4 = d A 6 = H 2 r H dr vol E6 + H 2 H d vol E6 ,
F

A 6 = H 1 vol E6 ,
where 3 = 3 , the volume of the unit three-sphere, and that the ChernSimons term
vanishes for the above solution, we can rewrite the regularised action as

1
2
2
I (R) =
F4 A 6 2 d 10 x K h 11
I0 [h]
11
3
M
M

2
d 10 x r 4 H 1 r H + 8r 3 4 11
I0 [h].
=
(3.7)
r=R
Introducing further upper cutoffs on the brane worldvolume directions by compactifying

them on a six-torus with volume V6 , we obtain

9
2
2
11
I (R) = V6 4 5R 3 + 23 a 2 R + O R 1 11
I0 [h],
5
where
k (b1 , b2 ) =
b1k b2k
b1k + b2k
(3.8)
describes the charge distribution of the configuration. This is clearly divergent in the limit
R , as we expect. In order to overcome this divergence, we will perform background
subtraction, choosing as our zero-point the one-centred action
2
I0 (R) = 5V6 4 R 3 .
11
(3.9)
The background-subtracted action is then

9
2
11
I (R) = 23 V6 4 a 2 R + O R 1 ,
5
which vanishes in the limit R followed by V6 only in the case that the charges
are equally distributed, that is, 3 = 0, and otherwise diverges linearly in R (and V6 ).
How should we interpret this result? Analogously to Brills instanton, with a suitable
choice of coordinates, one can interpret the instanton as tunneling between a universe
with two equally-charged centres and a universe with the same total charge, again at zero
temperature. As we shall see below, the asymptotic boundary of two-centred AdS7 S 4 in
the case b13 = b23 matches exactly to the boundary of one-centred AdS7 S 4 with charge
b3 = b13 + b23 , and thus the background subtraction is well-defined. This is not the case
when b13 = b23 , where the background subtraction calculation must be taken with caveats.
217
This means that although the above calculation suggests that no instanton exists which
interpolates between a two-centred spacetime with b13 = b23 and the one-centred geometry,
we cannot conclusively assert this without applying a better-defined procedure such as
the use of counterterms. Nevertheless, note that the result from background subtraction is
consistent with the analysis of [2] in which the special case |b1 | |b2 | is considered. Here
they find that fragmentation of a supersymmetric AdS universe into one macroscopic and
one microscopic part, or the creation of a brane with charge given by b1 , is allowed only
for AdS2 , but not for higher-dimensional AdS spaces.
Note that, as opposed to the Brill instanton, in the case of the M5-brane the finite
contribution vanishes since it is proportional to the horizon area, which vanishes for the
M5-brane. Hence the zero action in this case correctly coincides with the zero entropy
associated to the M5-brane.
4. The need for counterterms

Let us examine in more detail how closely the boundary geometry of two-centred
AdSp+2 S Dp2 matches that of the one-centred space, so that we can determine in
which cases background subtraction can be faithfully applied. We will analyse this briefly
for AdS2 S 2 and AdS7 S 4 , then outline the counterterm technique and discuss its
applicability to the M5 throat instanton action.
4.1. Boundary matching in background subtraction
For AdS2 S 2 , the induced metric on the boundary is
ds 2 = H 2 dt 2 + H 2 R 2 d22,
(4.1)
in the limit R , and where H in the above equation is understood to be a function

of the regulatory constant R rather than the coordinate r. The S 2 radius for two-centred
AdS2 S 2

b1
b2
HR =
+
R
R1 R2
tends exactly to the one-centred S 2 radius b at the boundary, provided that charge
conservation holds, that is, b1 + b2 = b. Note that R1 , R2 are defined as r1 , r2 in (2.9)
with r replaced by R. However, the scale factor of the worldvolume coordinates

H 2 = (b1 + b2 )2 R 2 + 21 aR cos + a 2 1 3 1 21 cos2

+ O R 1
(4.2)
does not match asymptotically, where 1 is defined as in (3.8). The leading order term
agrees with the one-centred value of R 2 /b2 (with charge conservation), but one is left with
nonvanishing terms for any value of 1 , that is, regardless of charge distribution! Despite
this nonmatching, however, the results in Section 2 suggest that background subtraction
218
still reproduces the correct result. We shall see in a few paragraphs that this is indeed the
case.
Let us now turn to AdS7 S 4 . Here the boundary metric is

ds 2 = H 1/3 ds 2 E6 + H 2/3 R 2 d42
(4.3)
in the limit R . The S 4 radius for two-centred space matches that for the one-centred
space in the boundary limit.3 The worldvolume scale factor
1/3

[R + 3 a cos ] + O R 1
H 1/3 = b13 + b23
(4.4)
again does not in general match asymptotically, except if 3 = 0, that is, when b13 = b23 ! For
unequal distributions, however, one is left with a finite term which renders the background
subtraction procedure untrustworthy. So although the results of Section 3 imply that no
instanton exists for a two-centred spacetime with b13 = b23 , we cannot state this definitively
without applying a better-defined procedure. Such a procedure is, as mentioned in the
introduction, counterterm subtraction.
4.2. Counterterms and the M5 throat instanton
Counterterm subtraction has been employed successfully to calculate gravitational
actions and thermodynamics of black holes such as the TaubNUTAdS, TaubBolt
AdS [13], and KerrAdS [17], for which the appropriate reference background was either
unknown or ambiguous. In each of these cases, background subtraction provided a good
approximation to the results obtained using counterterms, in some cases coinciding.
The idea behind this technique is as follows. The AdS/CFT correspondence equates
the partition functions of the two theories. In the low energy limit, one can thus use
the supergravity action in string theory to calculate the generating functional of the
conformal field theory on the boundary. It turns out that the divergences which appear
in the gravitational action are local integrals on the boundary [5], and thus can be removed
by subtracting local surface counterterms. The form of these counterterms has been
developed in [11,12]. In particular, the counterterms to the purely gravitational action are
(for 0 p 5) [13]
2
d p+1 x h
Ict = 2
p+2
M

p
lR
l3
p+1 2
ab
+
+
R
R
R
+ Ilog , (4.5)
ab
l
2(p 1) 2(p 3)(p 1)2
4p
where M is the asymptotically AdSp+2 space, R is the Ricci scalar of the induced metric
h on the boundary M, and Ilog is a logarithmically divergent term corresponding to the
conformal anomaly [11] which only contributes in dimensions for which p is odd. Notice
that no counterterms are necessary for p = 0, which corresponds to the action for the
3 In fact, one can show that the S Dp2 radius for the two-centred AdS
Dp2 always tends to the
p+2 S
corresponding one-centred value in the boundary limit.
219
Brill instanton. The action above should be read such that for p = 1 only the first term is
included, for p = 2, 3 only the first two terms are included, and for p = 4, 5 all three terms
are included. In addition, for odd values of p there is a contribution from the logarithmic
term Ilog . Recall that l is the radius of AdS.
For the AdS7 S 4 instanton, the relevant counterterm action is

bR b 3
2
5
3 2
6
ab
+
+
Ict (AdS7 ) = 2
d x h
+ Ilog ,(4.6)
Rab R R
2b
4
8
10
7
AdS7
since l = 2b for the AdS7 S 4 metric (3.2). As a first check we compute the counterterms
for the one-centred AdS7 S 4 action. Since the boundary of AdS7 is Ricci-flat, we have
that Ilog vanishes [11] and

5
V6
2
=
Ict (AdS7 ) = 2
d 6x h
5R 3 .
2b
7
b4 72
r=R
2 = 2 b 4 , giving
To uplift to eleven dimensions we simply use that 11
4
7

V6 4
Ict AdS7 S 4 = 2 5R 3 ,
11
which is exactly the value of the one-centred AdS7 S 4 action obtained in (3.9), as we
expect.
The counterterms presented in (4.6) arise from including a source for the stressenergy tensor in the dual conformal field theory, that is, they only cancel divergences
corresponding to the purely gravitational action. However, the supergravity action (3.1)
also contains matter coupled to gravity in the form of the field strength F4 . One might
wonder if additional counterterms are necessary. To check this, one should first determine
which seven-dimensional fields correspond to the eleven-dimensional F4 . This can be
done since the KaluzaKlein truncation of eleven-dimensional supergravity to maximally
gauged supergravity in seven dimensions with gauge group SO(5) has been shown to
be consistent in [18]. One should then apply the counterterm criterion for the sevendimensional scalar fields and other matter, which was derived in [19,20] by turning on
a source for each dual operator in the conformal field theory and following the same
approach as for pure gravity. Without explicitly undertaking this procedure, however, the
fact that the desired matching is obtained by adding the gravitational counterterm alone
suggests that further counterterms should not be required.
We would like to compute counterterms for the two-centred action. However, this
is not straightforward. The reason is that the counterterms presented in [1113] have
been formulated for asymptotically AdS spacetimes. In the case that the solution of
interest takes the form (asymptotically AdSp+2 ) S Dp2 , one can in general reduce the
D-dimensional theory to (p + 2)-dimensional gauged supergravity, and find the analogous
(asymptotically AdSp+2 ) solution. The counterterms which arise from the dual conformal
theory living on the (p + 1)-dimensional boundary can then be applied. Unfortunately the
two-centred (and in general n-centred) AdSp+2 S Dp2 spacetime is, as mentioned in the
introduction, a warped geometry. Rather than being a direct product of an asymptotically
220
AdS space with a sphere, it takes the form of asymptotically (AdSp+2 S Dp2 ). Certain
asymptotically (AdSp+2 S Dp2 ) geometries, such as some continuous distributions of
p-branes, can be consistently reduced to a solution of a (p + 2)-dimensional supergravity
[21,22]. Indeed, the counterterm renormalisation technique was recently applied to such
a solution in [23]. However, it is not clear to us that a discrete distribution of p-branes,
as the n-centred AdSp+2 S Dp2 geometries describe, finds an analogue in a reduced
(p + 2)-dimensional supergravity,4 and thus it would appear that one cannot apply the
counterterms, as originally formulated, to this class of solutions. An alternative which
is available to us is to attempt to formulate the counterterms directly in the elevendimensional theory, as was carried out in [15] for the PolchinskiStrassler solutions in
type IIB supergravity, and in [16] for AdS3 S 3 and AdS5 S 5 . We will not consider this
here but leave it as a further direction to pursue.
5. Other brane throat instantons?

In the introduction to this paper we mentioned that we expect the result for the
M5-brane to hold analogously for other supersymmetric branes in ten- and elevendimensional supergravity. Here we will present preliminary results for the M2- and
D3-branes using background subtraction, and a more concrete result for the D1/D5 system
with a discussion on counterterms and subtleties in dealing with self-dual field strengths.
We finish by briefly commenting on other nonconformal D-branes and the NS5-brane.
5.1. The M2- and D3-branes
Like the M5-brane, the M2- and D3-branes are nonsingular and have anti-de Sitter nearhorizon geometriesAdS4 S 7 and AdS5 S 5 , respectively. However, the results achieved
using background subtraction have been less conclusive. The M2 throat instanton action
contains a divergent term of order R 4 , while the action of the D3 throat instanton contains
a divergent term of order R 2 .
Despite this, we would argue that these results cannot provide the complete picture, and
that if we are able to apply a suitable counterterm technique as mentioned above, we should
recover results analogous to that for the M5-brane. To motivate this, let us analyse whether
the boundary geometries match.
The boundary metric of the two-centred AdS4 S 7 geometry corresponding to the twocentred M2-brane is, in the limit R ,

ds 2 = H 2/3 ds 2 E3 + H 1/3 R 2 d72 ,
(5.1)
where
H=
b16
R16
b26
R26
4 This is because one would need to turn on an infinite number of chiral fields [24] to accommodate the
candidate solution in a reduced supergravity.
221
The S 7 radius can easily be shown to match to the one-centred value, but the scale factor
for the worldvolume coordinates is
4 4

R + 46aR 3 cos + O R 2 ,
H 2/3 = b16 + b26
(5.2)
where 6 is defined as in (3.8).
For the two-centred AdS5 S 5 geometry corresponding to the two-centred D3-brane,
we have

ds 2 = H 1/2 ds 2 E4 + H 1/2 R 2 d52 ,
(5.3)
where
H=
b14
R14
b24
R24
The worldvolume scale factor is then

1/2 2
H 1/2 = b14 + b24
R + 24 aR cos + O(1) ,
(5.4)
with 4 defined as in (3.8).

A pattern should now emerge: if we consider (4.4), (5.2) and (5.4) ((4.2) is an
exception), we see that the higher the dimension of the sphere, the higher the power
of R at which the boundary geometries do not match5 R 0 for the four-sphere, R for
the five-sphere, and R 3 for the seven-sphere. This corresponds to one power of R less
than the highest order divergent term left in each of the on-shell instanton actions after
background subtraction! We would therefore argue that one should not trust the results
obtained from background subtraction except for the M5-brane case with two equal centres
(and, of course, the Brill instanton), and that one should expect an appropriate counterterm
procedure to show that analogous instantons exist for the M2- and D3-branes, at least in
the cases with two equal centres.
5.2. The D1/D5 system and self-dual field strengths
The D1/D5 system of type IIB supergravity is yet another with anti-de Sitter nearhorizon geometry, in this case, AdS3 S 3 E4 . The action for this system is

1
2
10
2
2
10 I = d x g R
(5.5)
F3 2 d 9 x K h 10
I0 [h].
2 3!
M
We are interested in the solution6

ds 2 = H 1 ds 2 E2 + H ds 2 E4(1) + ds 2 E4(2) ,
(5.6)
5 The following is meant for arbitrary values of ; choosing equal charge distribution = 0 results in
k
k
reducing the power of R by one, hence resulting in exact matching at the boundary for the AdS7 S 4 geometry
with two equal centres.
6 Since the three-form field strength is self-dual and thus carries both magnetic and electric components,
it will necessarily become complex in the continuation to Euclidean space. The issue of how to interpret this
remains unresolved. This comment also applies to the self-dual five-form field strength in the D3-brane solution
above.
222
F3 = dA2 + !6 dA2 ,

A2 = iH 1 vol E2 .
(5.7)
(5.8)
The subscripts on the two E4 spaces above are simply for clarity. Here E2 is the space
tangent to both branes, E4(1) is transverse to both branes, E4(2) is transverse to the D1-branes
but tangent to the D5-branes, and !6 refers to the Hodge dual on the space E2 E4(1) . We
will choose spherical coordinates on E4(1)

ds 2 E4(1) = dr 2 + r 2 d32 ,
(5.9)
and compactify E4(2) onto a four-torus T 4 of volume V4 . The choice of harmonic function
b2
(5.10)
r2
corresponds to the special nondilatonic case in which the D1- and D5-brane charges are
equal and proportional to b2 , giving the metric (5.6) the geometry AdS3 S 3 T 4 with
b the radius of both AdS3 and S 3 . We have chosen to study this particular case since the
resulting geometry for the two-centred choice
H=
H=
b12
r12
b22
r22
(5.11)
is a direct product of asymptotically (AdS3 S 3 ) with T 4 . This means that we can

essentially disregard the four-torus and work in six dimensions, with a geometry analogous
to the previous cases we have studied above. As before, b1 and b2 represent the charges of
the two centres, and r1 , r2 are defined as in (2.9).
As we did in Section 2 for the Brill instanton and Section 3 for the M5 throat instanton,
let us now calculate the two-centred AdS3 S 3 T 4 Euclidean action using background
subtraction. The bulk action vanishes by virtue of the equations of motion (for R) and
self-duality of F3 .7 Since

1 3 1
2
r H r H + 3r 3
K h=
2
where 3 = 3 , the unit three-sphere volume, we obtain

2
2
I = 2
d 9 x K h 10
I0 [h]
10
r=R

2
= V2 V4 3 4R 2 + 22 a 2 + O R 1 10
I0 [h],
where 2 is defined as in (3.8) and the coordinates on E2 have been compactified onto
a two-torus T 2 of volume V2 , which will be taken to infinity after background subtraction.
7 In the Euclidean continuation, the standard self-duality formula becomes F = i F , hence leading to the
3
3
problem outlined in the previous footnote.
223
Choosing the one-centred action

2
10
I0 (R) = 4V2 V4 3 R 2
(5.12)
as the zero-point, we obtain the background-subtracted action

2
10
I (R) = V2 V4 3 22 a 2 + O R 1 .
Similarly to the M5-brane throat, the action only vanishes in the limit R followed
by V2 , V4 if 2 vanishes, that is, in the case of equal charges. So, using background
subtraction, there exists a zero-temperature instanton describing the splitting of one D1/D5
throat into two equally charged D1/D5 throats. Note that the background subtraction
result for splitting into two throats of unequal charge distribution (2 = 0) again does
not contradict the result of [2], nor the more detailed analysis of [25].
How trustworthy is background subtraction in this case? The worldvolume coordinate
scale factor for the two-centred geometry is
1 2

R + 22 aR cos + 1 4 cos2 1 22 a 2
H 1 = b12 + b22

+ O R 1 ,
(5.13)
and does not match the one-centred value R 2 /b2 for any value of 2 . However, we would
argue, as for the Brill instanton (4.2), that the result obtained by background subtraction for
the equal-centred case is correct. Perhaps it is the low dimension of the transverse sphere,
and hence the low exponent of the highest divergent term in the action, which results in the
correct result via background subtraction.
As expounded in the previous section, the only rigorous way of calculating the action
is to use counterterm subtraction, rather than background subtraction. AdS3 corresponds to
p = 1 in the action (4.5), giving the only relevant counterterms

h
2
2
+ Ilog .
Ict (AdS3 ) = 2
(5.14)
d x
b
3
AdS3
2 =
For the one-centred AdS3 S 3 T 4 spacetime, Ilog again vanishes and, using that 10
2
2
3 V2 V4 b 3 , we have

V2 V4 3 2
Ict AdS3 S 3 T 4 =
2R .
2
10
(5.15)
This is half the value of the one-centred AdS3 S 3 T 4 action (5.12) calculated above!
How do we account for this discrepancy?
In general, varying a given action involves integrating by parts to obtain a bulk term and
a boundary term. Forcing the bulk term to vanish determines the equations of motion for
the system, and the vanishing of the boundary term determines the boundary conditions
which the fields must satisfy for that particular action.
However, it can happen that there is a solution which obeys these equations of motion
derived from the action but which does not obey the specified boundary conditions. In
these cases, to compute the correct action one needs to add boundary terms to the original
224
action such that the variation of this modified action still reproduces the same equations of
motion, but now determines boundary conditions which coincide with the given solution.
In our particular case we should consider the variation of the F 2 term in the bulk action:

F F = 2 d(A) F = 2 A d F + 2 A F.
(5.16)
M
The choice of sign in the bulk term above corresponds to the rank of A. If the boundary term
does not vanish when evaluated on a particular solution, then this suggests that to obtain
the correct value of the on-shell action, one should really add to the action the boundary
term

2 A F.
(5.17)
M
Of course, in the case of AdS3 S 3 there are some subtleties due to the lack of a covariant
action incorporating the self-dual F3 , however, it turns out that

1
2 A2 F3 = 2V2 V4 3 R 2
(5.18)
2
M
is precisely the right term to add to I0 (5.12) in order to equate it with Ict (5.15).8
As for counterterms in the two-centred case, recent progress in [16] seems to indicate
that the counterterm renormalised actions for both the two-centred AdS3 S 3 and AdS5
S 5 geometries do indeed vanish.
5.3. Nonconformal D-branes and the NS5-brane
Although the near-horizon geometry of the supersymmetric Dp-brane (p = 3) solutions
of type II supergravity is only conformal to AdSp+2 S 8p (or E(6,1) S 3 for p = 5), one
can still apply the same analysis above by working in what is called the dual frame [26].
This is defined as the frame in which the metric and the dual (8 p)-form field strength
couple to the dilaton in the same way, and it is in this frame that all Dp-branes (except
for p = 5) have AdSp+2 S 8p as their near-horizon geometry. It has also been argued
[26] that the dual frame is holographic in the sense that taking the decoupling limit of the
Dp-brane solution leads directly to a supergravity description, so it would be convenient to
compute quantities in this frame in order to easily make statements about the corresponding
conformal field theory.
One final brane for which it might be interesting to perform the same analysis is
the supersymmetric NS5-brane in type IIA/B supergravity. Its near-horizon geometry is
no longer anti-de Sitter, rather it is S 3 R R5,1 , and the string theory in the NS5brane background is conjectured [27] to be dual to the NS5-brane worldvolume theory.
This worldvolume theory, often referred to as a little string theory, turns out to be in
8 Although it is not clear to us why this should be so, taking F 2 = |dA2 | + | ! dA2 | rather than zero also
6
3
2
2
gives precisely the right contribution to agree with the counterterm result.
225
an appropriate limit a six-dimensional superconformal field theory with N = (2, 0) for

type IIA,9 and N = (1, 1) for type IIB. One can as before construct supersymmetric multicentred near-horizon geometries, and compute the appropriate Euclidean action evaluated
on these solutions.
We expect that similar results will follow for the branes discussed in this subsection, but
leave this for future work.
6. Speculations
In his original paper [1], Brill conjectured the existence of an instanton which describes
the fragmentation of the complete extremal ReissnerNordstrm wormhole and agrees
with the AdS2 S 2 instanton in the interior throat region. Unfortunately this has not
yet been realised. Nevertheless, we would similarly conjecture that an instanton which
describes the splitting of the M5-brane with two equal centres, and not just its two-centred
AdS7 S 4 throat, also exists.
If this is true, then drawing from the analogy we made in the introduction, an
interpretation would be that the exact eigenstate of the full M-theory Hamiltonian, which
we label naively as corresponding to the M5-brane, is in fact a coherent superposition
of the quantum states |n associated to each of the n-centred M5-branes which reside
in the same charge sector of the supersymmetry multiplet, so that charge conservation
holds. Intuitively, one can motivate this since the BPS property of the branes means that
there are no forces involved in separating them, and thus the two-centred or in general
n-centred geometries with n separate stacks of branes are equally stable configurations
with the same total energy. The question which remains from our analysis is whether the
M5-brane eigenstate also superposes states which correspond to a non-uniform, or
unequal, distribution of branes in the stacks, as in the case of the extremal Reissner
Nordstrm black hole.
What does all this imply for the dual conformal field theory? The AdS/CFT correspondence tells us that M-theory on AdS7 S 4 is dual to a six-dimensional N = (2, 0) superconformal field theory [3], about which little is known. Since, by the results of Section 5,
we expect a similar interpretation to apply to the AdS5 S 5 throat of the D3-brane, we will
discuss, in the following paragraph, the correspondence in terms of AdS5 S 5 and N = 4
super YangMills, which is much better understood. Nevertheless, one should read the
comments below as applying to any pair of anti-de Sitter space and conformal field theory
related by the correspondence, in particular, to AdS7 S 4 and the (2, 0) superconformal
field theory.
Now string theory on AdS5 S 5 with one centre is dual to N = 4 YangMills at
the superconformal point, where the vacuum expectation values of all the scalar fields
vanish. Similarly, string theory on AdS5 S 5 with more than one centre (for clarity we
will discuss the case of two centres) is dual to the conformal field theory at a point in its
9 This is the same (2, 0) theory dual to M-theory on AdS S 4 , see [28] and references therein for further
7
details. We thank T. Dasgupta for clarifying this point.
226
Fig. 1. On the left-hand side is the CFT picture, with the infrared fixed points shown, one of which is
superconformal (SC). The dual AdS picture is shown on the right-hand side. The dashed and dashed/dotted arrows
represent RG flow from the ultraviolet (UV) to the infrared (IR) for the one-centred and two-centred AdS S
spaces, respectively, while the plain arrow represents the nonzero transition amplitude between the one-centred
and two-centred states in both pictures.
moduli space where some of the scalar fields have acquired a nonzero vacuum expectation
value, hence breaking the gauge group in a manner described in [24]. That the metric
on two-centred AdS5 S 5 interpolates between one-centred AdS5 S 5 at r and
a two-centred geometry at r 0, corresponds to the renormalisation group (RG) flow from
N = 4 YangMills at the superconformal fixed point in the ultraviolet to N = 4 Yang
Mills at some fixed point in the infrared.10
However, this is not what we believe the instanton on the supergravity side implies for
the conformal field theory. As touched on in the introduction, the fact that the transition
amplitude between the one-centred and two-centred AdS5 S 5 spaces is nonvanishing
implies that the true vacuum of N = 4 YangMills is actually a superposition of vacua in
which the scalar fields have different vacuum expectation values. This is depicted in Fig. 1.
On the other hand, one expects the vacua above to be exact vacua of the full quantum
N = 4 YangMills theory, and hence their superposition to be forbidden by the cluster
decomposition principle. How can we understand this contradiction?
One should keep in mind that the result presented in this paper was calculated in
semi-classical supergravity, and that while this serves as a low-energy approximation to
M-theory, it may be that M-theory finds some way of resolving the contradiction by
suppressing the tunneling, so that the effective contribution of such instantons to the
amplitude vanishes. Since the full M-theory is not yet known, it is possible that some
as-yet-unknown symmetry not present in the low-energy supergravity description may
10 For one-centred AdS S 5 the flow is trivial.
5
227
impose a new superselection rule, preventing the tunneling process. It would be very
interesting to investigate just how such a suppression might occur. Since the contradiction
only arises in applying the AdS/CFT duality, one could also question if there is a subtlety
in its application here to supergravity. However, we are inclined to believe that the
correspondence holds true, and that the tunneling will be suppressed in the full M-theory.
Yet another possibility is that the tunneling is instead suppressed even at the supergravity
level by the presence of fermionic zero modes.11 This issue is presently under investigation
and we hope to report on it soon, for both the Brill instanton and the M5-throat instanton.
Finally, we would like to comment on the implications of our results for the issue of
black hole entropy. Some years ago, it was claimed [2931] using semiclassical topological
arguments, that all extremal black holes, regardless of horizon area, should have zero
entropy, and that the BekensteinHawking area law applies only to nonextremal black
holes. Since the Hawking temperature of extremal black holes vanishes, the zero entropy
claim is supported by the third law of thermodynamics in its strongest form. While the
extremal M5-, M2- and D3-branes are in accordance with both the area law and the zero
entropy claim due to their vanishing horizon areas, the extremal ReissnerNordstrm black
hole forces a contradiction.
On the other hand, string theory appears to favour the area law, and the thesis that
extremal black holes should be obtained as limits of their nonextremal counterparts. The
counting of the microscopic quantum degrees of freedom corresponding to the black hole
entropy was pioneered by Strominger and Vafa in [32]. When applied to the Reissner
Nordstrm black hole, the string analysis results in the expected nonzero Bekenstein
Hawking entropy [33]. There was an attempt to resolve this apparent contradiction between
the macroscopic and microscopic counting of the entropy in [34].
As mentioned above, the extremal M5-, M2- and D3-branes each have vanishing
horizon area and hence vanishing entropy. This agrees with the fact that each of the
above branes is BPS (supersymmetric), and thus there should be no quantum corrections
to the semiclassical entropy as given by the BekensteinHawking area law. Furthermore,
an object with vanishing entropy should truly be fundamental, in the sense that it has
no constituent degrees of freedom. The extremal M5-, M2- and D3-branes can indeed be
considered to fall into this category.
Acknowledgements
S.N. would like to thank M. Bianchi, G.W. Gibbons, H.S. Reall, K. Skenderis,
M.M. Taylor-Robinson and P.K. Townsend for interesting discussions. In particular, S.N. is
grateful to D. Mateos for many valuable comments, many more enlightening conversations,
and for a careful reading of the manuscript. Finally, we would like to thank a referee for
useful and insightful comments. The work of S.N. was supported by Churchill College
Cambridge and an ORS Award.
11 We thank M. Bianchi for pointing this out, and for discussions regarding this.
228
References
[1] D.R. Brill, Splitting of an extremal ReissnerNordstrm throat via quantum tunneling, Phys. Rev. D 46
(1992) 1560, hep-th/9202037.
[2] J. Maldacena, J. Michelson, A. Strominger, anti-de Sitter fragmentation, J. High Energy Phys. 02 (1999)
011, hep-th/9812073.
[3] J.M. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math.
Phys. 2 (1998) 231, hep-th/9711200.
[4] S. Gubser, I. Klebanov, A. Polyakov, Gauge theory correlators from noncritical 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] S.W. Hawking, in: S.W. Hawking, W. Israel (Eds.), General Relativity, an Einstein Centenary Survey,
Cambridge Univ. Press, 1979.
[7] J.W. York Jr., Role of conformal three-geometry in the dynamics of gravitation, Phys. Rev. Lett. 28 (1972)
1082.
[8] G.W. Gibbons, S.W. Hawking, Action integrals and partition functions in quantum gravity, Phys. Rev. D 15
(1977) 2752.
[9] S.W. Hawking, D.N. Page, Thermodynamics of black holes in anti-de Sitter space, Commun. Math. Phys. 87
(1983) 577.
[10] E. Witten, Anti-de Sitter space, thermal phase transition, and confinement in gauge theories, Adv. Theor.
Math. Phys. 2 (1998) 505, hep-th/9803131.
[11] M. Henningson, K. Skenderis, The holographic Weyl anomaly, J. High Energy Phys. 07 (1998) 023, hepth/9806087.
[12] V. Balasubramanian, P. Kraus, A stress tensor for anti-de Sitter gravity, Commun. Math. Phys. 208 (1999)
413, hep-th/9902121.
[13] R. Emparan, C.V. Johnson, R.C. Myers, Surface terms as counterterms in the AdS/CFT correspondence,
Phys. Rev. D 60 (1999) 104001, hep-th/9903238.
[14] K. Skenderis, Asymptotically anti-de Sitter spacetimes and their stress energy tensor, Int. J. Mod. Phys. A 16
(2001) 740, hep-th/0010138.
[15] M. Taylor-Robinson, Anomalies, counterterms and the N = 0 PolchinskiStrassler solutions, hepth/0103162.
[16] M. Taylor-Robinson, Higher-dimensional formulation of counterterms, hep-th/0110142.
[17] A.M. Awad, C.V. Johnson, Holographic stress tensors for KerrAdS black holes, Phys. Rev. D 61 (2000)
084025, hep-th/9910040;
A.M. Awad, C.V. Johnson, Higher dimensional KerrAdS black holes and the AdS/CFT correspondence,
Phys. Rev. D 62 (2000) 125010, hep-th/0008211.
[18] H. Nastase, D. Vaman, P. van Nieuwenhuizen, Consistent nonlinear KK reduction of 11d supergravity on
AdS7 S 4 and self-duality in odd dimensions, Phys. Lett. B 469 (1999) 96, hep-th/9905075.
[19] S. de Haro, K. Skenderis, S.N. Solodukhin, Holographic reconstruction of spacetime and renormalization in
the AdS/CFT correspondence, Commun. Math. Phys. 217 (2001) 595, hep-th/0002230.
[20] M. Taylor-Robinson, More on counterterms in the gravitational action and anomalies, hep-th/0002125.
[21] D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, Continuous distributions of D3-branes and gauged
supergravity, J. High Energy Phys. 07 (2000) 038, hep-th/9906194.
[22] M. Cvetic, S.S. Gubser, H. L, C.N. Pope, Symmetric potentials of gauged supergravities in diverse
dimensions and coulomb branch of gauge theories, Phys. Rev. D 62 (2000) 086003, hep-th/9909121.
[23] M. Bianchi, D.Z. Freedman, K. Skenderis, How to go with an RG flow, hep-th/0105276.
[24] I.R. Klebanov, E. Witten, AdS/CFT correspondence and symmetry breaking, Nucl. Phys. B 556 (1999) 89,
hep-th/9905104.
[25] N. Seiberg, E. Witten, The D1/D5 system and singular CFT, J. High Energy Phys. 04 (1999) 017, hepth/9903224.
[26] H.J. Boonstra, K. Skenderis, P.K. Townsend, The domain-wall/QFT correspondence, J. High Energy
Phys. 01 (1999) 003, hep-th/9807137.
[27] O. Aharony, M. Berkooz, D. Kutasov, N. Seiberg, Linear dilatons, NS5-branes and holography, J. High
Energy Phys. 10 (1998) 004, hep-th/9808149.
229
[28] O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity,
Phys. Rep. 323 (2000) 183, hep-th/9905111.
[29] G.W. Gibbons, R.E. Kallosh, Topology, entropy and Witten index of dilaton black holes, Phys. Rev. D 51
(1995) 2839, hep-th/9407118.
[30] S.W. Hawking, G.T. Horowitz, S.F. Ross, Entropy, area and black hole pairs, Phys. Rev. D 51 (1995) 4302,
hep-th/9409013.
[31] C. Teitelboim, Action and entropy of extreme and non-extreme black holes, Phys. Rev. D 51 (1995) 4315,
hep-th/9410103.
[32] A. Strominger, C. Vafa, Microscopic origin of the BekensteinHawking entropy, Phys. Lett. B 379 (1996)
99, hep-th/9601029.
[33] J.M. Maldacena, A. Strominger, Statistical entropy of four-dimensional extremal black holes, Phys. Rev.
Lett. 77 (1996) 428, hep-th/9603060.
[34] S. Das, A. Dasgupta, P. Ramadevi, Can extremal black holes have nonzero entropy?, Mod. Phys. Lett. A 12
(1997) 3067, hep-th/9608162.

Muon anomalous magnetic moment in a calculable

model with one extra dimension
Giacomo Cacciapaglia a , Marco Cirelli a , Giampaolo Cristadoro b
a Scuola Normale Superiore and INFN, Piazza dei Cavalieri 7, I-56126 Pisa, Italy
b Enrico Fermi Department of Physics, University of Pisa, via Buonarroti 2, I-56126 Pisa, Italy
Abstract
In the framework of a recently proposed extension of the Standard Model, with N = 1
supersymmetry in 5 dimensions, compactified on R1 /Z2 Z2 , we compute the muon anomalous
magnetic moment at one loop to order (MW R)2 , where R is the compactification radius. We find the
corrections to be small with respect to the SM pure weak contribution and not capable of explaining
the present discrepancy between theory and experiment for any sensible value of R. 2002 Elsevier
Science B.V. All rights reserved.
PACS: 11.10.Kk; 12.60.Jv; 13.40.Em
1. Introduction
Models with extra (space) dimensions often fail to produce quantitative predictions
for physical observables and, as a consequence, can hardly be ruled out or confirmed
by present energy experiments. On the other hand, in [1] a model was proposed where
calculability is achieved for a number of physical quantities. In fact, in spite of the
non-renormalizability of 5-dimensional YangMills theory, they are finite and cut-offindependent, owing to the underlying supersymmetric structure. The insensitivity to
ultraviolet physics of such observables is one of the main goals of the model. The
branching ratio of B Xs [2] and the Higgs production via gluon fusion [3] have already
been computed, while in the present paper we deal with the muon anomalous magnetic
moment a .
E-mail addresses: cacciapa@cibs.sns.it (G. Cacciapaglia), mcirelli@cibs.sns.it (M. Cirelli),
cristado@cibs.sns.it (G. Cristadoro).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 3 7 - 1
G. Cacciapaglia et al. / Nuclear Physics B 634 (2002) 230246
231
The measurement of a is one of the most stringent tests for new physics scenarios,
thanks to its current impressive precision. Moreover, the recent results of E821 experiment
at BNL [4] have highlighted a discrepancy with the present Standard Model prediction and
have, therefore, made even more interesting the comparison with reliable predictions made
by new models. In particular, attempts have been made in order to study the effect of extra
space dimensions [5].
Let us briefly review the present status of the SM prediction [6]. The QED contribution
has been computed up to order 5 and it gives the core of the experimental value:
aQED = (11 658 470.0 0.3) 1010. The purely weak terms are also known with a small
uncertainty, up to two loops: aweak = (15.1 0.4) 1010 . On the contrary, hadronic
loop corrections provide relevant contributions but also carry the greatest theoretical
uncertainty [7]. In fact, since it is quite hard to deal with low energy QCD, experimental
inputs are also needed. Light-by-light scattering yields albl = (8.6 3.2) 1010
and hadronic vacuum polarization a = (684.9 6.4) 1010 . Summing up all these
contributions and uncertainties, the SM prediction is:
hvp
aSM = (11 659 162.0 7.5) 1010.

On the other hand, the present world average for the experimental value (including BNL
results) is [6]:
a = (11 659 202.3 15.1) 1010 .
exp
This leads to a discrepancy a bit larger than 2 :

a aSM = (40.3 16.9) 1010.
exp
(1)
Actually, this result is not well established. On the theoretical side, recent results [8,9]
claim that the light-by-light contribution has opposite sign, thus reducing the discrepancy
to about 1 . At the same time, the error on the experimental value should be improved by
further results at BNL.
In this work we compute the corrections a to the anomalous magnetic moment due
to the new fields in the theory, up to order (MW R)2 , where R is the compactification radius
of the 5th dimension.
2. The model
In this section we briefly summarize the main features of the model [1] we are working
with. It is a 5D theory with N = 1 supersymmetry, compactified on R1 /Z2 Z2 , and
the gauge group is the Standard Model one: SU(3) SU(2) U (1). The fields content
is made up of the vector hypermultiplets and a matter hypermultiplet for each SM matter
field Q, U, D, L, E, H , all living in the bulk.
From a 4D point of view, a matter hypermultiplet splits into a pair of chiral supermultiplets with conjugate quantum numbers (property that we indicate with superscript c),
while a vector hypermultiplet splits into a vector and a chiral supermultiplet. Under the
orbifolding, global supersymmetry is completely broken la ScherkSchwarz, leaving the
232
Table 1
Gauge, matter and Higgs fields content of the theory with their orbifolding properties
A , M , H
(+, +)
cos 2n
Ry
, M , H
(+, )
cos 2n1
R y
c
c
, H
, M
(, +)
sin 2n1
R y
c c
, H
, M
(, )
sin 2n
Ry
2n , n = 0, 1, 2, 3, . . .
R
2n1 , n = 1, 2, 3, . . .
R
2n1 , n = 1, 2, 3, . . .
R
2n , n = 1, 2, 3, . . .
R
SM fields as the only zero modes. Anyhow, restricted local supersymmetric transformations still hold. The most general Lagrangian, according to this set of symmetries (gauge
group, orbifold parities and local supersymmetry), is:
L(x, y) = L5 + (y)L4 + (y R/2) L4 ,
(2)
where L5 is N = 1 supersymmetric in 5D whereas L4 and L4 are 4D Lagrangians invariant

under different N = 1 supersymmetries valid on the fixed points of the two parities. To be
consistent with Ref. [1], we will write all hypermultiplets in terms of supermultiplets of
the 4-dimensional N = 1 supersymmetry located on the brane y = 0. So, the gauge vector
and matter hypermultiplets are:

1

(W , ) , = ( + iW5 ) ,
2

c
.
(X , X ) Xc , X
Under the two parities all the fields have definite transformation properties, given in Table 1
together with the corresponding eigenfunctions and the spectrum of every tower of Kaluza
Klein states.
As pointed out in Ref. [10], a FayetIliopoulos (FI) term D can be located on the two
branes, thus introducing a free parameter in the potential. For our purposes, the only effect
of this term is that the value of 1/R, which is correlated to the Higgs mass, can increase up
to about a TeV [11]. In the particular case that the FI term is radiatively generated, with a
cut-off at a scale 5/R, its effect is small enough and well inside the errors on 1/R, that
mainly come from the z factors. Thus, in the following we still refer to the prediction of
Ref. [1]:
1
= 370 70 GeV.
R
In [12] it is shown that in the model under consideration the hypercharge current
has an anomaly, which is localized on the two branes. A relevant feature is that, from
a 4-dimensional point of view, the integrated anomaly vanishes. Gauge invariance can
however be recovered with a suitable modification of the theory. The overall consistency of
the model modified by the addition of a ChernSimons term has been checked in [13] and
in [14]. Such modifications, anyhow, do not affect the calculation presented in this paper,
since anomalous contributions only enter at higher order.
233
The terms in the 5D action, that are relevant for the following calculation, are:

1
S = d 5 x Tr FMN F MN Tr i D i D
4
+ iX D X + iXc D Xc

M c
c
D X
+ (DM X ) DM X + DM X

c
c
+ 2 g(5) X X X X + h.c.

c
+ 2 g(5) Xc X + X
X + Xc X + h.c.

c
c
c
c
+ y R L E H
0 + L H 0 E + E H 0 L
2

c
c
c
c
E H + + H + E + E H +
,
(3)
, ,
and contains
where the fermionic fields are Weyl spinors, the gauge fields WM
the group generators, e.g., a T a (the generators being normalized according to
Tr(T a T b ) = ab ) and the index X can run over E, L, H . The gauge coupling g is intended
to be the UY (1) coupling g or the SU L (2) coupling g according to the gauge fields
involved.
From now on, for the four gauge hypermultiplets we use the notation:
QED
Zs
W s
(A , , c , ),
(Z , z , z c , z ),

W , w , w c , w .
We remind that, also in this case, the subscribed c only means that the second spinors have
conjugate quantum numbers.
2.1. Mass eigenstates
The brane interactions introduce mass mixing among the states of muon and smuon KK
towers: since it will be useful for our calculation to work with the mass eigenstates, we
proceed to determine eigenvalues and eigenvectors from the very beginning (see, e.g., the
discussion in [2]). For the fermions, the mass eigenvalues turn out to be:
m0 = m
3
,
R
m
n =m
2n
R
(n = 1, 2, . . .),
corresponding to mass eigenvectors whose Weyl spinor components we denote by (0 , c0 )

c
and (
n , n ). Thus m is the mass of the ordinary muon. For the scalars:
m
n =
(2n 1)
m
R
(n = 1, 2, . . .),
corresponding to eigenvectors n and nc .
234
The interaction eigenstates1 are expressed in terms of the mass eigenstates (at order 3)
by
1
+
n n ,
2n
n

1
3
3
+
+ 1
+3
L,l =
nl n n ,
1+
2(2n) l
2(2n) l
2
n

3
1
3
+
c
+
1+
3
1
ln n + n
E,l =
2(2n) l
2(2n) l
2
n
30 ,
2l

3
1
3
+
L,l =
1+
l + 1
l
2(2n 1)
2(2n 1)
2

+3
Jln n+ n ,
L,0 = 0 + 3
(4)
(5)
(6)
(7)

3
1
3
c
E,l
l+ + 1 +
l
= 1
2(2n 1)
2(2n 1)
2

+
+3
Jnl n + n ,
(8)
with
ln =
2(2l)
(2l)2 (2n)2
and Jln =
2(2l 1)
,
(2l 1)2 (2n 1)2
and zero for n = l. Expressions which are identical to (4), (5), (6) hold for (E,0 , E,n ,
c ) if ( , ) are replaced by (c , c ); for ( c , ) it is sufficient to replace
L,n
0
E
n
n
n
L
0
with nc in Eqs. (7) and (8).
2.2. Couplings to the mass eigenstates
Starting from Eq. (3), one has to integrate out the 5th dimension y and rewrite
the interaction vertices in terms of the mass eigenstates (4)(8). Taking into account
momentum conservation along the 5th direction, p5 , the bulk interactions give rise to
several trilinear vertices. The muon masses and Higgs vertices, on the other hand, being
located on the brane, violate p5 -conservation and are proportional to 3. As we are interested
in O(R 2 ) contributions, only graphs where the internal lines carry the same index n, in
terms of the mass eigenstates, have to be taken into account. This simplifies the calculation
of the relevant vertices. As an explicit example, the interaction terms between KK photons
1 Note that in this paper the brane Lagrangians L and L are reversed respect to Ref. [1]. For more details,
4
4
see Appendix A in [3].
II
IV
235
III
Fig. 1. General patterns of graphs contributing to the (g 2).
and the various components in the muon towers, in the mass eigenstate basis, are:

1
3
3
eL,0 L,n An = e 0
+ 1+
An ,
1
2(2n) n
2(2n) n
2

1
e c
3
c
0 +
E,2n
E,n
An = e
n n A n .
2(2n)
2
2
In this particular case, summing up the two contributions, the couplings of order 3 vanish,
so that the net result is:

1
e 0 +
(9)
n + n A n .
2
This allows to recognize e/ 2 as the correct coupling with a KK photon.

On the other hand, the interactions of the zero mode
gauge bosons are still diagonal
in n and have a universal coupling, e.g., e = e(5) / 2R for the

photon. Note that this
relation holds for all gauge couplings, so in the following g = g(5) / 2R is the 4D gauge
coupling, while s = sin W and c = cos W are functions of the usual SM Weinberg angle.
3. Calculation of a
The effective Lagrangian term which fixes the notation for the anomalous magnetic
moment a = (g 2)/2 is:

ie
a
F .
(10)
2m
In the Feynmant Hooft gauge ( = 1, see below), one can identify the five general
types of diagrams in Fig. 1, each giving a contribution a whose explicit expressions are
listed in Appendix A.
L=
236
As in the following we will distinguish between diagrams with chirality flip on the
external or internal fermionic line, in Eqs. (51)(54) it is easy to trace back the external
cross contribution and the internal cross one (the latter proportional to mf /m). On the
other hand, in graph V the chirality flip is provided by the only scalar interaction with the
fermion.
3.1. QED
In this section we consider all the diagrams involving QED hypermultiplet fields. The
one loop Standard Model contribution [15] simply comes from graph IV with a massless
photon in the loop, with g+ = g = e:
.
a =
(11)
2
3.1.1. KK vectors
The interaction Lagrangian for KK photons has been displayed in (9) and the graphs are
of type IV. As a general rule, dealing with the diagrams involving KK fields, it is useful to
separate the external and internal cross contributions. This is motivated by the fact that in
the latter case it is necessary to use couplings and loop masses at order 3 while in the former
O(1) is enough. The following sketched diagrams are drawn in terms of Weyl fermions and
the emitted zero-mode photon is implicitly attached to any internal charged line. The states
in the loop are KK modes of the field indicated, always carrying the same index n for the
reasons explained above. The resulting a , for a given internal KK state, are:

32
e2

dx 4x(1 x)2 8x(1 x) , (12)
2
2
16 (2n)
e2
32
16 2 (2n)2

dx 8x(1 x) 16x(1 x)2 ,
(13)
and the trivial integration over the Feynman parameter x is left explicit in order to discuss
supersymmetric cancellations (see below).
3.1.2. Scalars
The gauge scalar field A5 mixes mode by mode with the massive vector A . Including
also the mass term for the gauge scalar , the mass Lagrangian, for each mode n, is:

2

1
2n
2n
1 2n 2 2
A5

Lmass = A
(14)
A A5 ,
R
2 R
2
R
where the R gauge fixing term is there to cancel the mixing terms. It is convenient to
work in the = 1 gauge: we obtain a photon propagator with no longitudinal modes and
237
two scalars and A5 (with equal mass) that can be combined in the complex scalar .
The interaction terms are then:
c

(15)
2 e E E + Lc L + h.c. ,
so that at the relevant order in 3, only an external cross graph is possible. Equivalently, in
terms of the real scalar fields it happens that the internal cross diagram with an A5 field is
cancelled out by the same diagram with a . The external cross diagram is of type II so
that the resulting a is

32
e2
2
16 (2n)2

dx 4x(1 x)2 .
(16)
3.1.3. Fermions
The gauge fermions and c are mixed by the kinetic term along the 5th direction.
They are involved in the following interaction terms

3
1
n n+ +
n + n+ ,
2 e0 n
2n 1
2

1 +
3 +
c
n + n +
n , ,
+ 2 e0 c,n
(17)
2n 1 n
2
and in similar expressions (replacing with c ) for the interactions of c0 with n and 0
with c,n . From graph II, one gets
e2
32

16 2 (2n 1)2
e2
32
2
16 (2n 1)2

dx 4x(1 x)2 2, (18)

dx 8x(1 x)2 2.
(19)
238
3.1.4. Total QED contribution

First of all, in order to test the computation, we consider the supersymmetric limit in
which both fermions and bosons have the same parities and therefore the same bulk mass.
In this case, we find that the total contributions of the KK modes vanishes, as expected. Of
course, the appearance of zero modes for the matter and gauge partners also cancels the
SM result.
Then, we collect all the QED contributions and we obtain:
aQED =
2
e2
2
(mR)
.
16 2
18
(20)
3.2. Neutral weak sector

In this section we consider all the diagrams involving Z hypermultiplet fields and
neutral Higgs. We define

1
1
2
gL =
+ s g,
c
2
gR =
s2
g.
c
The one-loop SM result [16] comes from graph IV with a zero mode Z boson and couplings
g = gL , g+ = gR :
aZ =
g 2 m2 4s 4 2s 2 1
.
48 2 MZ2
c2
(21)
The discussion for the KK modes mimics the QED case, thus in the following we only
point out the differences.
3.2.1. KK vectors
The interaction Lagrangian consists in
1
1

c
0 jL+ +
c0 jR+ c+
n + j L n Z n +
n + jR n Zn + h.c.
2
2
(22)
where
3
(gL + gR ),
2(2n)
3
(gL + gR ),
jL = gL
2(2n)
jL+ = gL +
3
(gR + gL ),
2(2n)
3
(gR + gL ).
jR = gR +
2(2n)
jR+ = gR
(23)
The two a then amount to

32
1
dx gL2 + gR2
2
2
16 (2n)

2x(1 x)2 4x(1 x) ,
(24)
239

1
2
dx gL gR + (gL + gR )
2

8x(1 x) gL gR 16x(1 x)2 .
(25)
32
1

16 2 (2n)2
3.2.2. Scalars
With respect to the QED case (14), the Lagrangian for scalar fields is a bit more
complicated, involving also mixing terms with H 0 , the imaginary part of H 0 . With the
gauge fixing piece already included, the Lagrangian for each mode n is:

2

2n
1
1 2n
2
Lmass = Z
Z5 MZ H 0
H 0 MZ Z5 MZ2 H
0
R
2 R
4

2

1 2n 2 2
1
2n
Z
Z5 MZ H 0
Z
.
2 R
2
R
(26)
Vector-scalar mixings are again cancelled and, for = 1, there are no off-diagonal terms
H 0 Z5 , so the three real scalars Z , Z5 , H 0 are the mass eigenstates. The first two
combine in the complex field Z in terms of which, at the relevant order in 3, only the
external cross graph is possible:
32
1

16 2 (2n)2

dx gL2 + gR2 2x(1 x)2 .
(27)
3.2.3. Fermions
c , for each mode n, is:
The mixing matrix for the four neutralinos z , z c , H 0 , H
0

Lmass = (H 0 , z )
2n1
R
MZ
MZ
2n1
R
c
H
0
z c

.
They are involved in the following interaction terms:

1
3
+
+
2 0 z n gL n n + (gL gR )
+ n
2(2n 1) n
2

1
+ 2 c0 z n gL gR and c
2

c c 1

3
+ 2 0 z n gR n + n+ (gR gL )
n n+
2(2n 1)
2

1
+ 2 0 z nc gL gR and c
2

c
3
1
1
+ 2 0 n + n+ + c0 nc + nc+ H,n
.
v
2
2
(28)
(29)
240
The external cross graphs are readily computed:
1
32

dx
16 2 (2n 1)2

gL2 + gR2 2x(1 x)2 2,
(30)
and it is possible to write three internal cross diagrams, using also the off-diagonal
propagator in Eq. (28):

32
1

dx
16 2 (2n 1)2

gL2 gR2 2x(1 x) + gL gR 8x(1 x)2 ,

32
1

dx
16 2 (2n 1)2

gR2 gL2 2x(1 x) + gL gR 8x(1 x)2 ,
32
1

16 2 (2n 1)2
(31)
(32)
g
dx (gL + gR )4x(1 x).
c
(33)
3.2.4. Total neutral weak contribution

Again, as in case of QED, one can check that in the supersymmetric limit, discussed
above, the contribution of the KK modes vanishes. Moreover, in the limit gL = gR = e,
the QED results are recovered.
Then, we sum up all the results of the neutral weak sector and obtain:

s 2 2s 4 2
g2
Z
2 1 1
+
.
(mR) 2
a =
(34)
16 2
c 4
3
3 12
3.3. Charged weak sector
In this section we consider all the diagrams involving W hypermultiplets fields and
charged Higgs. The SM result [17] comes from
graph III, where the loop contains a zero
mode W boson and a neutrino and g+ = g/ 2, g = 0:
aW =
5g 2 m2
.
2
96 2 MW
(35)
3.3.1. KK vectors
The interaction terms involving the KK vectors are quite simple
g 3 c
g
Wn 0 ,n
Wn 0 ,n
2
2 2 2n
and the external and internal cross diagrams are readily computed using graph III:

32
g2

dx (1 x)2 (x 3/2),
16 2 (2n)2
32
g2

16 2 (2n)2
3
dx (1 x)2 .
2
241
(36)
(37)
(38)
There also are diagrams of type V. The only nonvanishing one has a Higgs scalar, since
the contributions of W5 and W cancel out:
32
g2
16 2 (2n)2

dx(1 x)2 .
(39)
3.3.2. Scalars
The gauge fixing procedure works as in the Zs case: the interaction Lagrangian terms
only allow for an external cross contribution:

g c + L + Lc + h.c.
32
g2

16 2 (2n)2

dx x(1 x)2 .
3.3.3. Fermions
c
The charginos w , w c , H + , H
+ mix for each n according to:
c
2n1

2MW
2n 1 + +
H +
R
w c w .
Lmass = H + , w c
2n1
w
R
2MW
R
They enter the following interaction terms
3
g,n
c w c ,
w n+ 0 g
2(2n 1) ,n n 0
3
c
+ c
g,n
,n w c,n
w c,n
0 + g
0 ,
2(2n 1)
(40)
(41)
242
2m c c
+ c .
v 0 H ,n ,n
The external cross diagrams are of the usual form
g2
32

dx x(1 x)2 2,
16 2 (2n 1)2
(42)
(43)
and there are three internal cross diagrams with propagators (w + ) w c+ , (w c ) w

c :
and (w c ) H
+

32
g2
2
16 (2n 1)2
g2
32

16 2 (2n 1)2
32
g2

16 2 (2n 1)2
dx(1 x)2 ,
(44)

dx(1 x)2 ,
(45)
dx4(1 x)2 .
(46)
3.3.4. Total charged weak contribution

Again in this case, we correctly find that in the supersymmetric limit the sum of the KK
states contributions vanishes.
We then gather all the contributions of the charged weak sector into the result:
aW =
g2
7 2
.
(mR)2
2
16
6 12
(47)
3.4. Result to leading order in (MW R)2

Collecting all partial results (20), (34) and (47) we have:
aKK =
g 2 m2 11 18s 2
(MW R)2 .
2
192 MW
12c2
(48)
243
Numerically, for 1/R = 370 70 GeV, it is

weak
10
aKK = 0.07+0.06
= 1.1+0.6
,
0.02 a
0.3 10
which means that, for any sensible value of R (see [11]), aKK is well inside the
uncertainties that affect the SM prediction.
4. MW R corrections
In the previous section, we have neglected the ElectroWeak Symmetry Breaking in
the masses of the KK modes for the gauge fields, since this was enough to get Eq. (48).
However, in this way we are neglecting contributions that in principle can be quite relevant,
since the dimensionless expansion parameters MW R or MZ R can be as large as 1/3. In
this chapter we estimate these contributions. They are recovered if the full gauge fields
masses are kept in the denominators of Eqs. (51)(55), so the corrections to our previous
results can be expressed in a simple way. Of course, these corrections only apply to the
pure weak sector.
To begin with, let us focus on the neutral weak sector. First of all, what we have
to do is to study the exact spectrum: defining the dimensionless parameter = MZ R,
for every n there is a vector boson Z whose full mass squared is (2n)2 + 2 , in unit
of 1/R 2 . The scalars fields masses are determined from Eq. (26): the nth KK mode
of z has mass-squared (2n)2 + 2 , while, for the real field H 0 , it has mass-squared
(2n)2 + 32 2 . For the neutralinos, diagonalization of Eq. (28) shows that there are two
degenerate states with mass-squared (2n 1)2 + 2 . The rest of the computation proceeds
as
in 1the previous section. To get the modified results, it is sufficient to substitute the sum
in Eqs. (24), (25), (27), (30)(33) with:
(2n)2
N2
1
or
,
N2 + 2
(N 2 + 2 )2
where N = 2n for bosons and N = 2n 1 for fermions. The first is the one to be used in all
external cross diagrams, while for internal cross ones the formulae are a bit more involved.
To be specific, in the case of the vectors, Eq. (25) becomes:

1 2
1
1
2
(g
8x(1 x)
3
g
+
+
g
)
dx
g
L R
L
R
2
16 2
(2n)2 + 2

(2n)2
2
(49)
gL gR 16x(1 x)
.
((2n)2 + 2 )2
With a bit of algebra, all contributions can be recast in a common form and the net result
is:
aZ =

1 m2
(MW R)2 gL2 + gR2 f1 ( ) + gL gR f2 ( ) ,
2
2
16 MW
(50)
where the functions f1 and f2 are defined in Appendix B. This replaces Eq. (34), with all
corrections to any order in (MW R)2 included. Numerically, for 1/R = 370 70 GeV, the
244
result is only about 3% smaller than Eq. (34). In the case of the charged weak sector, the
calculation is very similar, so we expect the correction to be not so far from a few percent.
5. Conclusion
We have performed the calculation at one loop of the corrections to muon anomalous
magnetic moment coming from the presence of one extra dimension in the model in
Ref. [1], to first order in (mR)2 and in (MW R)2 . We have also shown that the complete
calculation in (MW R)2 yields small deviations, so to our purposes the simple analytic
form given in Eq. (48) is a good approximation. The corrections are relatively small, for
any sensible value of R, at the 10% level of the pure weak contribution in the SM, and well
inside the uncertainties of the hadronic contribution.
Acknowledgements
We would like to thank Riccardo Barbieri and Riccardo Rattazzi for useful discussions
and suggestions. We are also grateful to Gilberto Colangelo for informations about the
present status of SM (g 2) . This work was supported by the EC under the RTN contract
HPRN-CT-2000-00148.
Appendix A
In this appendix we list the contributions to a from the five graphs2 in Fig. 1:
aI
Qs 2
=
m
16 2
1
dx
h2+ + h2
mf
2x(1 x)2 +
h+ h 2x(1 x)
2
m
aII
Qf 2
=
m
16 2
xm2f
1
dx
1
,
+ (1 x)m2s x(1 x)m2
h2+ + h2
mf
2x(1 x)2 +
h+ h 2(1 x)2
2
m
(51)

xm2s
1
,
+ (1 x)m2f x(1 x)m2
2 Similar results can also be found in Refs. [18].
(52)
Qv 2
=
m
16 2
aIII
1
dx
2 + g2
g+
mf
2(1 x)2 (2x 3) + g+ g

6(1 x)2
2
m
245
aIV
Qf 2
=
m
16 2
aV
Qv
=
m
16 2
1
dx
0
xm2f
1
,
+ (1 x)m2v x(1 x)m2
2 + g2

g+
4x(1 x)2 8x(1 x)

2

mf
g+ g 8x(1 x)
+
m
1
,
2
xmv + (1 x)m2f x(1 x)m2
1
dx (g+ h+ + g h )(1 x)2
0
(53)
(54)
1
,
xm2f + (1 x)m2v x(1 x)m2
(55)
where Qs,f,v are the electric charges of the particles emitting the zero-mode photon and
ms,f,v are the masses in the loop. Note that in Eq. (55) we assume ms = mv . The couplings
are defined by:

s f h+ P + + h P ,
e s + W A ,
v f g+ P + + g P ,

e g (k p+ ) + g (p+ p ) + g (p k) W+ W A ,
where P are projectors on the two Weyl components of the Dirac fields.
Appendix B
In this appendix we list the functions to be inserted in Eq. (50):
1
f1 = 8

0

y y 2 coth
y + 2y
dy 2 4 y 4 y 2 tanh
2
2

4
1
= 2+
48 5 5 Li2 e Li2 e
6
8
3
3

+ 336 4 4 Li3 e Li3 e

+ 1440 3 3 Li4 e Li4 e

+ 4320 2 2 Li5 e Li5 e

+ 8640 Li6 e Li6 e + 8640 Li7 e Li7 e

42 4 4 (3) + 17145(7) ,
246
1
f2 = 8

y y 2 coth
y + 2y
dy 8 y 2 2 3y 2 2 2 y 2 tanh
2
2

4
4
= 2 6 8 8 5 5 Li2 e Li2 e

+ 136 4 4 Li3 e Li3 e

+ 960 3 3 Li4 e Li4 e

+ 3840 2 2 Li5 e Li5 e

+ 8640 Li6 e Li6 e + 8640 Li7 e Li7 e

+ 28 4 4 (3) 930 2 2 (5) + 17145(7) ,
where (n) is the Riemann zeta function, not to be confused here with our variable
= MZ R. In the integrals, the hyperbolic tangent is produced by the sum over the
fermionic KK states, while the cotangent and the 2y pieces by the bosonic states.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
R. Barbieri, L.J. Hall, Y. Nomura, Phys. Rev. D 63 (2001) 105007, hep-ph/0011311.

R. Barbieri, G. Cacciapaglia, A. Romito, Nucl. Phys. B 627 (2002) 95, hep-ph/0107148.
G. Cacciapaglia, M. Cirelli, G. Cristadoro, Phys. Lett. B 531 (2002) 105, hep-ph/0111287.
H.N. Brown et al., Muon g 2 Collaboration, Phys. Rev. Lett. 86 (2001) 22272231, hep-ex/0102017;
A. Czarnecki, W.J. Marciano, Phys. Rev. D 64 (2001) 013014, hep-ph/0102122.
P. Nath, M. Yamaguchi, Phys. Rev. D 60 (1999) 116006, hep-ph/9903298;
R. Casadio, A. Gruppuso, G. Venturi, Phys. Lett. B 495 (2000) 378, hep-th/0010065;
S.C. Park, H.S. Song, Phys. Lett. B 506 (2001) 99, hep-ph/0103072;
K. Agashe, N.G. Deshpande, G.-H. Wu, Phys. Lett. B 511 (2001) 85, hep-ph/0103235;
T. Appelquist, B.A. Dobrescu, Phys. Lett. B 516 (2001) 85, hep-ph/0106140.
J. Prades, Invited talk at Kaon 2001, 613 June 2001, Pisa, Italy, hep-ph/0108192.
For a recent review: W.J. Marciano, B.L. Roberts, hep-ph/0105056.
M. Knecht, A. Nyffeler, Phys. Rev. D 65 (2002) 073034, hep-ph/0111058;
M. Knecht, A. Nyffeler, M. Perrottet, E. de Rafael, Phys. Rev. Lett. 88 (2002) 071802, hep-ph/0111059.
I. Blokland, A. Czarnecki, K. Melnikov, Phys. Rev. Lett. 88 (2002) 071803, hep-ph/0112117.
D.M. Ghilencea, S. Groot Nibbelink, H.P. Nilles, Nucl. Phys. B 619 (2001) 385, hep-th/0108184.
R. Barbieri, L.J. Hall, Y. Nomura, hep-ph/0110102.
C.A. Scrucca, M. Serone, L. Silvestrini, F. Zwirner, Phys. Lett. B 525 (2002) 169, hep-th/0110073.
L. Pilo, A. Riotto, hep-th/0202144.
R. Barbieri, R. Contino, P. Creminelli, R. Rattazzi, C.A. Scrucca, hep-th/0203039.
J. Schwinger, Phys. Rev. 73 (1948) 416417;
J. Schwinger, Phys. Rev. 76 (1949) 790817.
G. Altarelli, N. Cabibbo, L. Maiani, Phys. Lett. B 40 (1972) 415.
S.J. Brodsky, J.D. Sullivan, Phys. Rev. 156 (1967) 1644.
I. Vendramin, Nuovo Cimento A 101 (1989) 731;
K.R. Lynch, Phys. Rev. D 65 (2002) 053006, hep-ph/0108080, hep-ph/0108081.

Next-to-leading order QCD corrections to

differential distributions of Higgs boson production
in hadronhadron collisions
V. Ravindran a , J. Smith b,1 , W.L. van Neerven c,2
a Harish-Chandra Research Institute, Chhatnag Road, Jhunsi, Allahabad, 211019, India
b C.N. Yang Institute for Theoretical Physics, State University of New York at Stony Brook, NY 11794-3840, USA
c Instituut-Lorentz, University of Leiden, PO Box 9506, 2300 RA Leiden, The Netherlands
Received 28 January 2002; accepted 23 April 2002
Abstract
We present the full next-to-leading order corrected differential distributions d 2 /(dpT dy),
d/dpT and d/dy for the semi-inclusive process p + p H + X . Here X denotes the inclusive
hadronic state and pT and y are the transverse momentum and rapidity of the Higgs-boson H ,
respectively. All QCD partonic subprocesses have been included. The computation is carried out
in the limit that the top-quark mass mt which is a very good approximation as long as mH ,
pT < 200 GeV. Our calculations reveal that the dominant subprocess is given by g +g H + X but
the reaction g + q(q)
H + X is not negligible. Another feature is that the K-factor representing
the ratio between the next-to-leading order and leading order differential distributions is large. It
varies from 1.4 to 1.7 depending on the kinematic region and choice of parton densities. We show
that a reliable determination of the differential cross sections requires good knowledge of the gluon
density in the region where x < 103 . Further we study whether the differential distributions are
dominated at large transverse momentum by soft-plus-virtual gluon contributions. This is of interest
for the resummation of large corrections which occur near the boundary of phase space. We also
compare our results with those previously reported in the literature. 2002 Elsevier Science B.V.
All rights reserved.
PACS: 13.85.-t; 14.80.Bm
E-mail address: neerven@lorentz.leidenuniv.nl (W.L. van Neerven).

1 Partially supported by the National Science Foundation grant PHY-0098527.
2 Work supported by the EC network QCD and Particle Structure under contract No. FMRX-CT98-0194.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 3 3 - 4
248
V. Ravindran et al. / Nuclear Physics B 634 (2002) 247290
1. Introduction
The Higgs boson, which is the corner stone of the standard model, is the only particle
which has not been discovered yet. Its discovery or its absence will shed light on the
mechanism how particles acquire mass as well as answer questions about supersymmetric
extensions of the standard model or about compositeness of the existing particles and the
Higgs boson. The LEP experiments [1] give a lower mass limit of about mH 114 GeV/c2
and fits to the data using precision calculations in the electro-weak sector of the standard
model indicate an upper limit mH < 200 GeV/c2 with 95% confidence level. After the
end of the LEP program the search for the Higgs will be continued at hadron colliders
in particular at the TEVATRON and the LHC. If the Higgs mass is in the above range
the principal production mechanisms at hadron colliders are gluongluon fusion g + g
H + X or W + W -fusion appearing in the reaction q + q q + q + H + X . In these
processes X denotes an inclusive hadronic state. For mH < 135 GeV/c2 the process
q + q V H + V + X with V = , Z, W has also to be taken into consideration. The
Higgs boson will be observed via its decay products among which the channels H b + b
and H + are the most prominent ones although H + + should also be
considered. However the large backgrounds make the observation of these decays very
difficult and it will take a lot of experimental and theoretical effort to detect the Higgs
boson provided it is there.
In this paper we concentrate on Higgs production channels where the lowest-order
reaction proceeds via the gluongluon fusion mechanism. In the standard model the Higgs
boson couples to the gluons via heavy quark loops among which the top-quark loop is the
most prominent since the coupling of the Higgs to a fermion loop is proportional to the
mass of the fermion (for a review see [2]). In lowest order (LO) the gluongluon fusion
process g + g H , represented by the top-quark triangle graph, was computed in [3].
The next-to-leading (NLO) processes given by gluon bremsstrahlung g + g g + H and
g + q(q)
q(q)
+ H were presented in [46] from which one can derive the transverse
momentum (pT ) and rapidity (y) distributions of the Higgs boson. The total integrated
cross section, which also involves the computation of the QCD corrections to the topquark loop, has been calculated in [7]. This calculation is rather cumbersome since it
involves the computation of two-loop triangular graphs with massive quarks. Recently also
the next-to-next-to-leading (NNLO) processes involving all two-to-three parton processes
have been computed in [8] using the helicity method which means that the matrix elements
are presented in four dimensions. From the experience gained from the NLO corrections it
is clear that it will be very difficult to obtain the NNLO one-particle inclusive distributions
from these calculations let alone the total cross sections.
Fortunately one can simplify the calculations if one takes the large top-quark mass limit
mt . In this case the Feynman graphs are obtained from an effective Lagrangian
describing the direct coupling of the Higgs boson to the gluons. The LO and NLO
contributions to the total cross section in this approximation were computed in [9] and
they found that the error introduced by taking the mt limit is less than about 5%
provided mH 2mt . A similar investigation was done for the differential distributions of
the processes g + g g + H and g + q(q)
q(q)
+ H in [10]. Here the approximation
is valid as long as mH and pT are smaller than mt . This is corroborated by the recent
249
calculations in [8] which show that for mH = 120 GeV/c2 the approximation is valid for
jets with transverse momenta smaller than 200 GeV/c. The NNLO matrix elements using
the effective Lagrangian were computed in [11,12] albeit in four dimensions. The oneloop corrections to the two-to-two parton subprocesses were presented in [13], where the
computation of the loop integrals was performed in n-dimensions but the matrix element
was still presented in four dimensions. The results in [1113] were used to compute the
transverse momentum and rapidity distributions of the Higgs boson up to NLO [14]. The
effective Lagrangian method was also applied to obtain the NNLO total cross section by
the calculation of the two-loop corrections to the Higgs-gluongluon vertex in [15], the
soft-plus-virtual gluon corrections in [16] and the computation of the two to three body
processes in [17].
In this paper we present the full NLO computation of the double differential
distributions d 2 /(dpT dy) for Higgs boson production in hadronhadron collisions using
the gluongluon fusion mechanism in the mt approximation. Here we have included
all partonic subprocesses. A similar calculation has been performed in [14] but our
approach differs from it in various aspects. First our calculation is purely analytical and
follows the calculation carried out for the DrellYan process describing vector boson
production in hadronhadron collisions (see [18]). The approach in [14] was mainly
numerical and based on the methods explained in [19]. Moreover it used the two-tothree particle matrix elements in [11,12] and the two-to-two particle matrix elements
including virtual corrections in [13] which were all presented in four dimensions. In our
calculation the matrix elements as well as the loop integrals and phase space integrals
are computed in n dimensions. The advantage of the analytical approach is that one gets
more insight into the structure of the radiative corrections. This is particularly important
for the large corrections, due to soft gluon radiation and collinear fermion pair production,
which arise near the boundary of phase space, where the pT of the Higgs boson gets large.
Resummation of this type of corrections has been carried out for the total cross section
in [20]. Resummation of small pT contributions due to the Sudakov effect has been done
in [21]. In view of the experimental problems to observe the Higgs boson, a recalculation
of all the NLO corrections is necessary to be sure that the theoretical predictions are
correct. We find that the contribution due to the (anti)quarkgluon subprocess is substantial
in particular at large transverse momentum and that the important region of x in the
gluon density is x < 103 . We also investigate the region of applicability of the softplus-virtual (S+V) approximation for the calculation of the differential cross sections.
Finally we mention that another paper has just appeared on the NLO corrections to the
g + g > H + g channel, using the helicity framework [22].
Our paper will be organized as follows. In Section 2 we give an outline of the kinematics and present the Born contributions. In Section 3 we present the virtual contributions.
In Section 4 the gluon bremsstrahlung corrections to the Born reactions are computed and
the soft gluon cross sections are explicitly shown. In Section 5 we show how the mass factorization is carried out. In Section 6 we give differential cross sections for protonproton
collisions at the LHC and make comparisons with results obtained earlier in the literature.
Some particular expressions are given in Appendices A and B. Other formulae, in particular those for the two-to-three body cross sections, are too long to be published. They are
available upon request as files written in the algebraic manipulation program FORM [23].
250
2. Lowest order contributions to Higgs production

In the large top-quark mass limit the Feynman rules (see, e.g., [12]) can be derived from
the following effective Lagrangian density
1
(2.1)
O(x) = Ga (x)Ga, (x),
4
where (x) represents the Higgs field and G is an effective coupling constant given by

s (2r ) 2
2 2r
2
2 2
2
G =4 2
(2.2)
GF F ( )C s r , 2 .
4
mt
Leff = G(x)O(x) with
In the expression above m and mt denote the masses of the Higgs boson and the top
quark, respectively. The running coupling is given by s (2r ) where r denotes the
renormalization scale and GF is the Fermi constant. Further C is the coefficient function
which originates from the QCD corrections to the top-quark triangle graph describing the
process H g + g in the limit mt . On the Born level the width of this decay process
is given by
(H g + g) =
(N 2 1)m3 GF 2 2 2 2
s r F ( ),
128 3 2
(2.3)
where N denotes the number of colours. Further the function F ( ) occurring in Eqs. (2.2),
(2.3) is defined by
F ( ) = 1 + (1 )f ( ),
4m2t
,
m2
1
f ( ) = arcsin2
for 1,
2

1 1
1
+ i
f ( ) = ln
4
1+ 1
for < 1.
(2.4)
In the large mt -limit F ( ) behaves as

2
.
(2.5)
3
The coupling G in Eq. (2.2) is presented for general mt on the Born level only whereas C is
computed in higher order for mt . In order to keep some part of the top-quark massdependence we take for G the expression in Eq. (2.2). This is an approximation because the
gluons which couple to the Higgs boson via the top-quark loop in the partonic processes
describing Higgs-production are often virtual. The virtual-gluon momentum dependence
is neither described by F ( ) in Eq. (2.4) nor by C. For on-mass-shell gluons the latter
quantity has been computed in the large top-quark mass limit up to order s in [7,9,11] and
up to s2 in [20,24]. In second order it reads

2 2r
s(5) (2r )
(11)
C s r , 2 = 1 +
4
mt
lim F ( ) =
s(5) (2r )
+
4
2
251

2777
67 8 2r
2r
+ 19 ln 2 + nf + ln 2 .
18
6
3 mt
mt
(2.6)
Here nf denotes the number of light flavours and s(5) is presented in a five flavour number
scheme.
In this paper we study the semi-inclusive reaction with one Higgs-boson H in the final
state. It will be denoted by
H1 (P1 ) + H2 (P2 ) H (p5) + X ,
(2.7)
where H1 and H2 denote the incoming hadrons and X represents an inclusive hadronic
final state. In lowest order the partonic reactions contributing to Eq. (2.7) are denoted by
a(p1 ) + b(p2 ) c(p3 ) + H (p5 ),
p1 + p2 + p3 + p5 = 0,
a, b, c = q, q,
g,
(2.8)
and the partonic cross sections in n-dimensions are given by

d n p5 + 2 + 2
1
d n p3
(1)
abc H = Kab G2 g 2 4n
p3 p5 m2
n1
n1
2s
(2)
(2)
(1)
2

M
(1 ) .
abc H
(2.9)
M (1)
Here
denotes the amplitude of the process and the strong coupling constant is given
by g with g 2 = 4s . The scale originates from the fact that the coupling constant
acquires a dimension in n dimensions. The quantity Kab represents the spin and colour
1
if one integrates over m
average over the initial states including the statistical factor m!
identical particles in the final state. Further we have assumed that the Higgs boson is mainly
produced on-mass-shell. However one can also use the narrow width approximation. This
can be achieved by replacing

m
1
,
p52 m2
(2.10)
(p52 m2 )2 + m2 2
where is the total width of the Higgs boson which is dominated by the decay H g + g.
Choosing the C.M. frame of the incoming partons we have the following parametrization
1
p1 =
P12 (1, 0, . . . , 0, 1),
2
1
p2 =
P12 (1, 0, . . . , 0, 1),
2
P12 m2
p3 =
(1, . . . , sin 1 , cos 1 ),
2 P12

1
p5 =
(2.11)
P12 + m2 , . . . , P12 m2 sin 1 , P12 m2 cos 1 ,
2 P12
with
Pij = (pi + pj )2 ,
t u
cos 1 =
.
s m2
P12 = s,
P15 = t,
P25 = u,
s + t + u = m2 ,
(2.12)
252
From Eq. (2.9) we infer that

s
2d
2 (1)
abc H
dt du

s (2r ) tu /2
S
= Kab G
s + t + u m2
(1 + /2) 4
2 s
(1)
2
, with n = 4 + .
M
2
abc H
The spherical factor S is defined by

S = exp (E ln 4) .
2
(2.13)
(2.14)
On the Born level we have the following subprocesses

g + g g + H,

(1)
2

4
4
4
8
2
M
= N N2 1 1
1+
s + t + u + m + 2m stu ,
ggg H
stu
2
1
1
,
Kgg =
4(1 + /2)2 (N 2 1)2
q + q g + H,

(1)
2

M
= CA CF 1 1 + t 2 + u2 + tu ,
q qg
H
s
2
1 1
,
Kq q =
4 N2
q(q)
+ g q(q)
+ H,

(1)
2

M
= CA CF 1 1 + s 2 + t 2 st ,
qgq H
u
2
1
1
,
Kqg =
4(1 + /2) N(N 2 1)
(2.15)
(2.16)
(2.17)
(2.18)
(2.19)
(2.20)
(2.21)
(2.22)
(2.23)
with
CA = N,
CF =
N2 1
.
2N
(2.24)
(1)
2
The n-dimensional matrix element |Mggg
H | is proportional to n 2 provided m = 0
so that it vanishes in two dimensions. The same feature also appears in the lowest order
matrix element for the decay H g +g derived from the effective Lagrangian in Eq. (2.1).
Notice that the factor n 2 = 2(1 + /2) also shows up in the spin average quantities Kab .
3. One-loop corrections to the 2 2-body reactions

The one-loop corrections to the gluongluon subprocess in Eq. (2.15) entails the computation of forty-two graphs. Fourteen graphs lead to independent expressions whereas the
remaining ones can be obtained via crossing from the s-to t-channel or from the s- to
u-channel. The Feynman integrals show ultraviolet, infrared and collinear singularities
253
which will be regularized using n-dimensional regularization. Hence the matrix element
and the Feynman integrals have to be computed in n dimensions for which we used the
algebraic manipulation program FORM (version 3.0) [23]. Since the Feynman integrals
also contain the integration momentum in the numerator we have to apply tensorial reduction. Here we followed the procedure in [25] which was extended to n-dimensions in [26].
The scalar integrals for the two-, three- and four-point functions can be found in [27]. The
one-loop correction to the gluongluon differential distribution (2.15) can be written as
s2
VIRT
d 2 ggg
H
dt du

s (2 ) 2
1
= s + t + u m2 S2 G2
2
4
(N 1)2

6
s
t
u
1
N 2 + 2 ln 2 4 ln 2 4 ln 2 + 6

2
t
u
sm
+ Li2
+ Li2
+ Li2
s
m2
m2
t
u
t u
t
s
u
s
3 ln 2 ln 2 + ln 2 ln 2 + ln 2 ln 2 ln2 2 ln2 2

2

2
2
1
1
t m
um
m t
1 2 m2 u
1
ln2
+ ln2
ln
+
ln2
2
t
2
u
2
2
m2
m2

2
9 (1)
s
t
u 11
Mggg H
2 ln 2 + 4 ln 2 + 4 ln 2 + (2)
2
2

1
(1)
2 ,
MBggg
+ (N nf )
(3.1)
H
4
2 2 2

m2

MB (1)
(3.2)
stu + m2 (st + su + tu) .
ggg H = N N 1
3
stu
Here the pole terms 1/k , k = 1, 2 (see Eq. (2.13)) represent the three types of divergences
mentioned above and Li2 (x) denotes the dilogarithmic function defined in [28]. The virtual corrections to the quarkantiquark reaction in Eq. (2.18) require the computation of
fourteen graphs. The procedure is the same as outlined above Eq. (3.1) and yields
s2
d 2 qVIRT
qg
H
dt du

2 2

2 2 2 s ( )
= s + t + u m S G
4

1 t 1 u 5
1
2
+ ln
2 nf
+ ln
N
3 3 2 3 2
9

2
s
t
u 11 1
+ CA 2 + 2 ln 2 2 ln 2 2 ln 2
3

t u
s m2
t
s
u
s
ln 2 ln 2 + ln 2 ln 2 + ln 2 ln 2
+ Li2
s
254

38
11 t
11 u 7
t
2 u
ln
ln
(2)
+
ln
2
2
6
2
6
2
2
9

1
t
u
t
u
4
+ Li2
+ CF 2 + 2 ln 2 2 ln 2 + 3
+
Li
2
2
m
m2

t u 1 2 t m2
1 2 u m2
1 2 m2 t
2 ln 2 ln 2 ln
ln
+ ln
2
t
2
u
2
m2

2

(1)
2
m u
1
3 t 3 u

+ ln2
+ ln 2 + ln 2 + 9(2) 4 Mq qg
H
2
2
2
2
m

2
1
(1)
,
+ (CA CF )
(3.3)
MBq qg
H
2

2
MB (1)

(3.4)
q qg
H = CA CF (t u).
ln2
The square of the matrix element for the virtual corrections to the quarkgluon subprocess
in Eq. (2.21) can be obtained from the one calculated for the quarkantiquark subprocess
in Eq. (2.18) via crossing from the s-channel to the u-channel. This yields
s2
VIRT
d 2 qgq
H
dt du

s (2 ) 2
1
= s + t + u m2 S2 G2
4
N(N 2 1)

8
1 t 2 u 1
s
2
+ ln 2 + ln 2 ln 2
nf
3 3
3
3
9

t
8 1
t
s
2
u
t u
+ Li2
+ CA 2 + 2 ln 2
ln 2 ln 2 + ln 2 ln 2
3
m2

1 2 m2 u
u
s
1 2 u m2
2 t
2 u
+ ln
ln 2 ln 2 ln 2 + ln 2 ln
2
u
2
m2

s
50
5 t
11 u 11
9
ln 2 +
ln 2 + (2) +
ln 2
6
3
2
9

t
u
s
1
4
+ CF 2 + 2 ln 2 4 ln 2 + 2 ln 2 + 5

2
sm
t
t u
2 ln 2 ln 2
+ Li2
+ Li2
s
m2

1 2 m2 t
u
s
1 2 t m2
2 u
+ ln
+ 2 ln 2 ln 2 2 ln 2 ln
2
t
2
m2

2
s
5 t
u 5
13 (1)
+ ln 2 + 5 ln 2 ln 2 + (2)
Mqgq H
2
2
2

1
2
(1)
+ (CA CF )
(3.5)
MBqgq H ,
2

2
MB (1)

(3.6)
qgq H = CA CF (s + t).
255
Notice that the MB (1) in Eqs. (3.2), (3.4) and (3.6) are presented in four dimensions and
higher-order terms in are ignored. This is sufficient because the factors with which they
are multiplied are finite in the limit 0. Our results agree with those presented in [13].
(1)
Notice that in the latter paper the Born amplitudes Mabc
H are presented in four dimensions because they are constructed using the helicity method. We have computed them in
n dimensions. Apparently this does not affect the factors that multiply the Born matrix
elements and which contain all the pole terms in 1/k .
4. Gluon bremsstrahlung and other 2 3-body processes

In this section we give an outline of the computation of the two-to-three parton processes
which show up in NLO. The calculation proceeds in an analogous way as in the case
for heavy flavour production presented in [29]. The processes under consideration will be
denoted by
a(p1 ) + b(p2 ) c(p3 ) + d(p4 ) + H (p5 ),
a, b, c, d = q, q,
g,
pi2 = 0,
i = 14,
5

pi = 0,
i=1
2
p52 = m .
(4.1)
The cross section corresponding to the reaction above can be expressed in n dimensions as
follows

1 2 4
d n p3
d n p4
d n p5
(2)
abcd H = Kab G g
2s
(2)n1
(2)n1
(2)n1

+ 2 + 2 + 2
2
n (n)
p3 p4 p5 m (2) (p1 + p2 + p3 + p4 + p5 )
2
(2)

Mabcd
(4.2)
H (1 , 2 ) ,
where Kab is defined below Eq. (2.9). Choosing the C.M. frame of the outgoing partons 3
and 4 we get the following parametrization in n dimensions
p1 = 1 (1, 0, . . . , 0, 0, 1),

p2 = 2 , 0, . . . , 0, |p5 | sin , |p5 | cos 1 ,
p3 = 4 (1, 0, . . . , sin 1 sin 2 , sin 1 cos 2 , cos 1 ),
p4 = 4 (1, 0, . . . , sin 1 sin 2 , sin 1 cos 2 , cos 1 ),

p5 = 5 , 0, . . . , 0, |p5 | sin , |p5 | cos ,
(4.3)
where the energies i are given by

1 =
P12 + P15 m2
,
2 P34
5 =
P15 + P25
,
2 P34
2 =
cos =
P12 + P25 m2
,
2 P34
P15 m2 + 21 5
.
21 |p5 |
3 = 4 =
1
P34 ,
2
(4.4)
256
The invariants are denoted by

Pij = (pi + pj )2 ,
P12 = s,
P15 = t,
P25 = u,
P34 = s4 = s + t + u m .
2
(4.5)
From | cos | 1 and s4 = P34 0 we derive the boundary conditions

sm2
m2 u s + t,
(4.6)
m2 m2 t s.
m2 t
In the subsequent part of this paper it is more convenient to choose s4 as integration variable
instead of u. In this case the integration boundaries are
t (m2 s t)
,
m2 m2 t s.
(4.7)
m2 t
From the above kinematics the differential cross section in Eq. (4.2) becomes equal to

2 (2)
2 2 tu m2 s /2 s /2
1
S2
4
4
abcd H
2d
2 s ( )
s
= Kab
G
dt du
2
(1 + )
4
2 s
2
(2)
2

Mabcd
(4.8)
H ,
0 s4
(2)
2
where |Mabcd
H | is the second order matrix element integrated over the polar angle 1
and the azimuthal angle 2 . It is given by
(2)
M
abcd H
2
=

d1 (sin 1 )
0

1+
(2)
dn1 M
2
(2)
d2 (sin 2 ) Mabcd H (1 , 2 )
abcd H (1 , 2 )
2
.
(4.9)
As in the calculation of the lowest order matrix elements in Eqs. (2.16), (2.19), (2.22) the
next-to-leading order expressions |M (2)(1 , 2 )|2 have to be computed in n dimensions
because we use n-dimensional regularization for the collinear and infrared singularities
which arise from the integration over the momenta of the final state partons. We use the
algebraic manipulation program FORM (version 3.0) as in the earlier computation of the
virtual corrections. Further we choose the Feynman gauge for the gluon propagators and
the sum over the physical polarizations of the external gluons is given by

? (p, )? (p, ) = P (l, p), with l P = P l = 0, l 2 = 0, (4.10)
=L,R
with
P = g +
l p + l p
,
lp
(4.11)
where l is an arbitrary lightlike vector. There are three types of matrix elements depending
on the number of gluons and (anti)quarks appearing in the initial and final state. They will
be denoted by ggggH , ggqqH and qqqqH . Notice that the quark q can also represent
257
an antiquark q depending on the type of process. The Feynman graphs can be found
in [12]. The matrix element for ggggH involves the computation of twenty-six graphs
whereas eight graphs contribute to ggqqH . In the case of qqqqH one has to distinguish
between identical and non-identical quarks. For identical quarks we have two Feynman
diagrams whereas the non-identical quarks are represented by one graph only. In the case of
ggggH the amplitude is denoted by M (2) and the matrix element squared is obtained
by contraction over the Lorentz indices as follows
(2)

M (1 , 2 )2 = P P P P M (2) M (2) .

(4.12)
We checked that the dependence on l pi disappears for arbitrary l. However, this

was only possible by writing the expression in Eq. (4.12) over a common denominator
and expressing the result into five independent kinematical invariants. It turns out that
Eq. (4.12) has the same property as the square of the Born matrix element in Eq. (2.16)
namely that it is proportional to n 2 provided m = 0. In the case of ggqqH the amplitude
(2)
is given by M
. Here we checked that the dependence on l pi of the matrix element
squared given by

(2)
M (1 , 2 )2 = P P M (2) M (2),
(4.13)
already disappears after a simple partial fractioning. We made a comparison with the results in [12] which are presented in four dimensions only because they are computed using
the helicity method. After corrections for some misprints3 the expressions obtained from
Eqs. (4.12) and (4.13) are in agreement with those presented in Appendix A of [12]. The
procedure to bring the matrix element over a common denominator, as discussed below
Eq. (4.12), is not suitable for partial fractioning (see below) because it leads to high powers in the angular-dependent kinematical variables, which show up in the numerator and
denominator. To avoid this complication we proceed in a different way. Since the l pi
terms are linear-independent for i = 14 they have to cancel among themselves. This we
checked by deleting them and comparing the truncated matrix element squared with the
one obtained by bringing all terms over a common denominator, which is manifestly independent on l pi . The difference between these two expressions turned out to be zero.
Moreover the total power of the angular-dependent kinematical invariants appearing in the
numerator and denominator of the truncated matrix element does not exceed four so that
it becomes amenable for partial fractioning. After this procedure the integration over the
angles is performed in the following way. The expression |M (2)(1 , 2 )|2 depends on the
variables Pij defined in Eq. (4.5) among which five are linearly independent. Because of the
parametrization in Eq. (4.3) the variables P12 , P15 , P25 , P34 are independent of the angles
1 , 2 whereas the remaining ones P13 , P23 , P14 , P24 , P35 and P45 are angular-dependent.
3 We found additional misprints in [12]. The S

34 in the denominators of the two terms in (A.12) should
read S24 . In (A.16) one has instead of n13 = S24 n12 (2 4) and n23 = S23 n13 (3 4) the expressions
n13 = S24 n12 (1 4) and n23 = n13 (3 4).
258
One can distinguish two types of integrals:

l
l
J Pijk , Pm5
= dn1 Pijk Pm5
, i, j = 14, m = 3, 4,
with 4 k 2, 0 l 2,

l
l
J Pijk , Pmn
,
= dn1 Pijk Pmn
(4.14)
i, j, m, n = 5,
with 3 k 2, 3 l 2.
(4.15)
The basic integrals of the first class are given by

k l
J P13
, P45 = 2k (P25 P34 )k Ik,l (A, B, C),
k l
J P13
, P35 = 2k (P25 P34 )k Ik,l (A, B, C),
with

Ik,l (A, B, C) =
(4.16)
dn1 (1 cos 1 )k (A + B cos 1 + C sin 1 cos 2 )l .
The basic integrals of the second class are represented by

k l
J P13
, P24 = 2k+l (P25 P34 )k (P15 P34 )l Ik,l (),
k l
J P13 , P23 = 2k+l (P25 P34 )k (P15 P34 )l Ik,l ( ),
with

Ik,l () =
(4.17)
(4.18)
dn1 (1 cos 1 )k (1 cos 1 cos sin 1 cos 2 sin )l
(n/2 1 k) (n/2 1 l) (n 3)
= 21kl
2 (n/2 1) (n 2 k l)

n
2
F1,2 k, l, 1, cos
,
2
2
(4.19)
where F1,2 (a, b, c; x) denotes the hypergeometric function which can be found in [30].
The other integrals can be derived by interchanging the inclusive momenta p3 and p4 or
interchanging the parametrization for p1 and p2 . The angular integrals above are presented
in Appendix C of [29] for the case 2 k 2. Because the square of the matrix element
for Higgs production contains two extra powers of Pij in the numerator the previous
computation has to be extended to cover the cases k = 3, 4. These extra powers can
be attributed to the momentum-dependence in the effective Higgs gluongluon coupling
described by the Lagrangian in Eq. (2.1). The colour decompositions of the integrated
expressions read as follows
g + g g + g + H,
(2)
2
2
(2)
2
2
M

,
gggg H = N N 1 Mgggg H
(4.20)
g + g qi + qi + H,

(2)
2
2

2
M
= nf CA CF CA M (2)
+ CF M (2)

,
ggqi qi H
ggq q H
ggq q H
(4.22)
(4.21)
(4.23)
259
q + q g + g + H,

(2)
2
2

2
M

(2)
+ CF M (2)

,
q qgg
H = CA CF CA Mq qgg
H
q qgg
(4.24)
q + g q + g + H,

(2)
2
2

2
M

(2)
+ CF M (2)

.
qgqg H = CA CF CA Mqgqg H
qgqg H
(4.26)
(4.25)
(4.27)
The colour and spin average factors Kgg , Kq q and Kqg are given in Eqs. (2.17), (2.20) and
(2.23), respectively. In NLO we encounter some new subprocesses. The first one is given
by the reaction in Eq. (4.22) where a sum over all light flavours qi with i = 1, . . . , nf is
understood. The next one is quarkquark scattering (non-identical and identical quarks)
represented by
q1 = q2 ,
q1 + q2 q1 + q2 + H,
(2)
2
(2)
2
M

,
q1 q2 q1 q2 H = CA CF Mq1 q2 q1 q2 H
(4.28)
(4.29)
q + q q + q + H,
(2)
2
(2)
2

2
M

+ CF M (2)
,
qqqq H = CA CF Mq1 q2 q1 q2 H
qqqq H
A
(4.30)
(4.31)
which have the colour and spin average factors

1 1
.
8 N2
The second subprocess is quarkantiquark scattering
Kq1 q2 = Kqq =
(4.32)
q1 + q2 q1 + q2 + H,
q1 = q2 ,
(2)
2
(2)
2
M

,
q1 q 2 q1 q2 H = CA CF Mq1 q2 q1 q2 H
(4.33)
qi = q,
q + q qi + qi + H,
(2)
2

2
M
= (nf 1)CA CF M (2)
,
q qq
q qq
a q a H
i q i H
(4.35)
(4.34)
qa = q,
(4.36)
q + q q + q + H,

(2)
2
2

2
M

(2)
+ M (2)

q qq
q H = CA CF Mq1 q2 q1 q2 H
q qq
H
a a
(4.37)

2
,
+ CF Mq(2)
qq
q H
F
(4.38)
and the colour and spin average factor reads

1 1
.
(4.39)
4 N2
Further we have summed over all quark flavours qi in the final state provided they are
not equal to the flavour of the quark q in the initial state (see Eq. (4.35)). Notice that the
colour decomposition given above also holds before the squares of the matrix elements
are integrated over the angles. Some of the expressions in Eqs. (4.20)(4.38) are singular
in the limit s4 0, where s4 is defined in Eq. (4.7), which is due to soft gluon radiation
or soft (collinear) fermion pair production. This will lead to infrared singularities when
the differential cross section in Eq. (4.8) is convoluted with the parton densities even after
Kq1 q2 = Kq q =
260
mass factorization is carried out. The convolution involves an integration over s4 so that
one has to split up the partonic cross sections into hard and soft gluonic parts as follows
(2)
HARD
SOFT
d 2 abcd
d 2 abcd
d 2 abcd
H
H
H
= s2
(s4 > ) + s 2
(s4 ),
dt du
dt du
dt du
where is chosen in such a way that it satisfies the inequalities
s2
s,
t,
u.
(4.40)
(4.41)
The soft gluon contribution to the differential cross section is defined by

2 SOFT
2 2 tu /2
1
S2
abcd H
2d
2 s ( )
= Kab
G
s
dt du
2
(1 + )
4
2 s
/2

(2)
2

s4
M

s + t + u m2
ds4
abcd H .
2
(4.42)
0
(2)
Only the singular part of |Mabcd H |2 , which behaves as 1/s4 , contributes to the above
integral whereas the non-singular terms vanish in the limit s4 0. This singular part is
SOFT
2
called |Mabcd
H | and behaves in two different ways:
SOFT 2
1

,
(I) Mabcd
(4.43)
H
s4
and
/2

SOFT 2
1 ss4
(II) Mabcd
(4.44)
.
H
s4 tu
SOFT for the various subprocesses are presented in Appendix A. Adding the latter
The d 2 ab
to the virtual contributions presented in Section 3 leads to the soft-plus-virtual differential
cross sections defined by
S+V
SOFT
VIRT
d 2 ab
d 2 ab
d 2 ab
= s2
+ s2
.
(4.45)
dt du
dt du
dt du
The infrared singularies cancel in this expression so that the double pole terms 1/2 ,
occurring in both the soft and the virtual contributions, all vanish.
s2
5. Renormalization and mass factorization of the partonic differential cross sections

The virtual cross sections in Section 3 have two types of ultraviolet divergences.
One is removed by renormalization of the operator O(x) occurring in Eq. (2.1) and the
other by renormalization of the strong coupling constant. The renormalization constant
corresponding to the former is

1
2

2
s 2
s
4 2 2
=1+
+
0 +
+
ZO = 1 + (s )
1
4
4
2 0
(1)
(2)
= 1 + ZO
+ ZO
+ ,
(5.1)
261
up to order s2 [31], where 0 and 1 are the lowest order coefficients of the beta-function
given by
2
s
s
0
1 + ,
4
4
11
2
34
10
1 = CA2 2nf CF nf CA .
0 = CA nf ,
3
3
3
3
(s ) =
(5.2)
In lowest order ZO is the same as the renormalization constant for the strong coupling
constant which reads

2
4 2 1
s 2
s
0 +
Zs = 1 +
+
1 +
4
4
2 0
= 1 + Z(1)
+ Z(2)
+ .
s
s
(5.3)
Since the operator insertion always appears twice in the differential cross sections the latter
2.
have to be multiplied by ZO
After renormalization one has still to perform mass factorization to remove the
remaining collinear divergences. This is achieved by the formula
s2
d 2 abH + X
(s, t, u, )
dt du
1
1
d 2 ef H + X
dx1
dx2
=
ea (x1 , )f b (x2 , )s 2
(s , t, u)
x1
x2
d t d u
0
ea f b s2
d 2 ef H + X
,
d t d u
(5.4)
where an implicit summation over e, f is understood, with

s = x1 x2 s,

t = x1 t m2 + m2 ,

u = x2 u m2 + m2 .
(5.5)
In Eq. (5.4) d 2 represents the singular cross section which still contains the collinear
divergences indicated by . These divergences are removed by the kernels ab so that the
finite cross sections d 2 are left. Both therefore depend on the mass factorization scale .
Since the mass factorization has to be carried out for the semi-inclusive cross section up to
NLO the kernels ab have to be corrected up to order s only. They are denoted by
(1)
ab (x, ) = ab (1 x) + ab (x, ).
(5.6)
In Section 4 the cross sections were split up into hard gluon (H ) (Eq. (4.40)) and softplus-virtual gluon parts (S+V) (Eq. (4.45)). Therefore one proceeds with the kernels in the
same way so that the mass factorization can be carried out for both parts separately
(1)
HARD,(1)
S+V,(1)
ab
= ab
+ (1 x)ab
.
(5.7)
262
The various kernels are expressed into the splitting functions Pab , depending on the partons
a, b, which represent the residues of the collinear singularities
s 1 (0)
S P (x),
4 qq
s 1 (0)
(1)
S P (x),
= g(1)
gq
q =
4 gq
1 (0)
s
(1)
(1)
S Pqg
= qg
(x),
qg
=
4 2
s 1 (0)
(1)
S P (x).
=
gg
(5.8)
4 gg
The splitting functions are given by

3
1
(0)
(x) = 4CF 2
1 x + (1 x) ,
Pqq
1x +
2

2
1 + (1 x)
(0)
Pgq
,
(x) = 4CF
x

(0)
Pqg
(x) = 8Tf x 2 + (1 x)2 ,

1
11
4
1
(0)
2
Pgg (x) = 8CA
+ 2 + x x + (1 x) nf (1 x).
1x + x
12
3
(5.9)
The collinearly finite cross sections can be derived from Eq. (5.4) by using the above
kernels. In LO they are equal to the Born results derived in Section 2. In NLO the finite
order s2 contributions can be written as
(1)
= q(1)
qq
q =
g + g g + g + H,
s
(2)
d 2 gggg
H
dt du
(2)
(1)
(1)
d 2 gggg H
d 2 ggg H
d 2 ggg H
(1)
(1)
gg
s 2
s 2
= s2
gg
,
dt du
d t d u
d t d u
(5.10)
g + g q + q + H,
s2
(2)
d 2 ggq
q H
dt du
(2)
(1)
(1)
d 2 ggq
d 2 qgq
d 2 qg
q H
H
q H
(1)
(1)
2qg
s2
s2
= s2
2qg
,
dt du
d t d u
d t d u
(5.11)
q + q g + g + H,
s2
d 2 q(2)
qgg
H
dt du
(1)
d 2 q(2)
d 2 q(1)
d 2 qqg
qgg
H
qg
H
2
(1)
2
(1)
2
qq s
=s
qq s
,
dt du
d t d u
d t d u
(5.12)
263
q + g q + g + H,
s
(2)
d 2 qgqg
H
dt du
=s
(2)
d 2 qgqg
H
dt du
(1)
gg
s 2
(1)
qq
s
d t d u
(1)
s 2
gq
(1)
s
qg
(1)
d 2 ggg
H
d t d u
(1)
d 2 q qg
H
2
d t d u
(5.13)
q + q q + q + H,
(2)
d 2 qqqq H
2
dt du
(2)
(1)
(1)
d 2 qqqq H
d 2 qgq H
d 2 qgq H
(1)
(1)
gq
s2
s2
= s2
gq
,
dt du
d t d u
d t d u
q1 + q2 q1 + q2 + H,
s
d t d u
(1)
d 2 qgq H
2
q1 + q2 q1 + q2 + H,
s
(1)
d 2 qgq
H
(5.14)
q + q q + q + H,
d 2 q(2)
qq
q H
2
dt du
(1)
(1)
d 2 q(2)
d 2 qgq
d 2 qgq
qq
q H
H
H
(1)
2
gq
= s2
(5.15)
s2
g(1)
.
q
dt du
d t d u
d t d u
Notice that the finite parts of these cross sections can become negative due to the above
HARD
subtractions. The expressions for the finite hard gluon cross sections d 2 abcd
H are too
long to be published. They exist as FORM [23] files and they are available on request. Here
we can only show those parts which behave like 1/s4 since they are of interest later on. In
the soft gluon limit s4 0 they read
lim s 2
s4 0
HARD
d 2 gggg
H

2
lim s 2
s4 0
lim s 2
dt du
s (2 )
4
2
HARD
d 2 ggq
q H
dt du

2
11 (1)
tu
s4
N
1
Mggg H ,
3 ln 2 ln 2
2
2
(N 1) s4
s 12

= G2
s (2 )
4
2

2
nf
1 1 (1)
Mggg H ,
2
2
(N 1) s4 6
(5.16)
(5.17)
d 2 qHARD
qgg
s4 0
dt du

s (2 ) 2 1 1
s4
tu
11
C
ln
= G2
+
ln
A
4
N 2 s4
2
2 s 12

2
(1)
s4
tu

M
+ CF 4 ln 2 2 ln 2
q qg
H ,
(5.18)
264
lim s 2
HARD
d 2 qgqg
H
s4 0
dt du

2 2
1
1
tu
3
s4
s4
2 s ( )
= G
CA 2 ln 2 ln 2 + CF ln 2
4
4
N(N 2 1) s4
(1)
2
Mqgq H ,
(5.19)
lim s 2
d 2 qHARD
qq
q H
s4 0
dt du

= G2
s (2 )
4
2

2
nf 1 1 (1)

Mq qg
H .
2
N s4 6
(5.20)
Notice that the expressions above satisfy a supersymmetric relation. It turns out that the
expressions for gg X + H in Eqs. (5.16) plus (5.17), q q X + H in Eqs. (5.18)
plus (5.20) and qg X + H in Eq. (5.19) become equal for a N = 1 supersymmetry
where CA = CF = nf = N .
After renormalization of the operator O(x) and the coupling constant using ZO and Zs
in Eqs. (5.1) and (5.3) respectively the ultraviolet divergences vanish in the soft-plus-virtual
S+V
cross section d 2 ab
in Eq. (4.45). The remaining collinear divergences are removed via
S+V
in Eqs. (5.7)(5.9). The combined effect of these three
mass factorization using ab
operations leads to the following expressions
s2
S+V
d 2 ggg
H
dt du
= s2
S+V
d 2 ggg
H
dt du
(1)

d 2 ggg H
(1)
S+V,(1) 2
+ Z(1)
,
s
+
2Z
2
gg
O
s
dt du
(5.21)
s2
d 2 qS+V
qg
H
dt du
= s2
d 2 qS+V
qg
H
dt du
(1)

d 2 q qg
H
(1)
S+V,(1) 2
+ Z(1)
,
+
2Z
2
s
qq
O
s
dt du
(5.22)
S+V
d 2 qgq
H
dt du
=s
S+V
d 2 qgq
H
dt du
Z(1)
s
(1)
+ ZO s 2
(1)
d 2 qgq H
dt du
(1)
d 2 q qg
H
S+V,(1)
S+V,(1) 2
qq
.
s
s
gg
dt du
dt du
The results for the finite soft-plus-virtual differential cross sections read
(1)
d 2 ggg H
2
s2
S+V
d 2 ggg
H
dt du

2 s (2 ) 2
1
1
5
2
ln
nf
= s + t + u m G
4
6 2 18
(N 2 1)2

t
3 2
tu
11
ln 2 + ln 2
ln 2 + Li2
+N
2
12
m2

t
u
s m2
s
u
s
ln 2 ln 2 ln 2 ln 2
+ Li2
+ Li2
2
s
m
(5.23)

t m2
1 2 t 1 2 u 1 2 s
1
ln 2 + ln 2 + ln 2 ln2
2
2
2
2
t

2

2

2
1 2 m t
1 2 m u
1 2 um
+ ln
+ ln
ln
2
u
2
2
m2
m2

2
2
67 (1)
1
(1)
Mggg H + (N nf )
MBggg H ,
+ 3 (2) +
36
4
265
s2
s2
(5.24)
d 2 qS+V
qg
H
dt du

s
s (2 ) 2 1
1
1
5
ln
ln
= s + t + u m2 G2
n
+
f
4
6 2 3 2 6
N2

1
s m2
tu
11
ln 2 + Li2
+ CA ln2 2 + ln 2
2
s
s 12

s
1 2 t 1 2 u 1 2 s
11
73
ln 2 5(2) +
ln 2 ln 2 + ln 2
2
2
2
6
12

tu
t
u
+ CF 2 ln2 2 2 ln 2 ln 2 + Li2
+ Li2
2
s
m
m2
t
t
u
s
u
s
ln 2 ln 2 ln 2 ln 2 + ln2 2 + ln2 2

2
2
1 2 t m
1 2 um
1 2 m2 t
ln
ln
+ ln
2
t
2
u
2
m2

2

(1)
2
1
s
m u
3

+ ln2
+ ln 2 + 8(2) 4 Mq qg
H
2
2
m
2

2
1
,
MBq(1)
+ (CA CF )
(5.25)
qg
H
2
S+V
d 2 qgq
H
dt du

1
s (2 ) 2
= s + t + u m2 G2
4
N(N 2 1)

1 u 5
u
tu
ln 2
+ CA ln2 2 ln 2 ln 2 + Li2
nf
3
9
s
m2

s
1 2 u m2
u
1 2 m2 u
2 u
ln 2 ln 2 + ln 2 ln
+ ln
2
u
2
m2

11 u
38
ln 2 + 4 (2) +
6
9

1 2
t
3
ln 2 ln 2 + Li2
+ CF
2
4
m2
266

t
s m2
t 1
u 1
s
s
1
ln 2 ln 2 + ln2 2 ln2 2 + ln2 2
s
2
2
2

2
1 2 t m2
9 (1)
3 u
1 2 m2 t
Mqgq H
ln
+ ln 2 (2)
+ ln
2
2
t
2
m
2
4

1
2
(1)
.
MBqgq
+ (CA CF )
(5.26)
H
2
+ Li2
(1)
The expressions multiplying the Born matrix elements |Mabc H |2 satisfy a supersymmetry relation. If we choose a N = 1 supersymmetry so that the quarks are put in the
adjoint representation, i.e., CA = CF = nf = N then all expressions are equal except for
the rational constants which are equal to 19/12 in Eq. (5.24), 15/12 in Eq. (5.25) and
17/12 in Eq. (5.26), respectively. These differences originate from the fact that we use
n-dimensional regularization rather than n-dimensional reduction. The result for the latter
case can be obtained via a finite renormalization. It turns out that for n-dimensional reduction the rational numbers become equal to 7/4 for all three reactions which provides
us with a strong check on our calculations. Finally we want to comment on the scale .
In the computation of the radiative corrections we have assumed that the renormalization
scale r is equal to the mass factorization scale . If one wants to distinguish between
both scales one has to substitute

2
s (2r )
0 ln r2 ,
s 2 = s 2r 1 +
(5.27)
4
in all finite expressions.

6. Differential distributions for the process p + p H + X
In this section we will present the differential cross sections for Higgs-boson production
in protonproton collisions at the LHC. Here we study the dependence of the cross sections
on input parameters like the QCD scale , the renormalization/factorization scale
and the dependence on the chosen set of parton densities. We also study which region
in the gluon density is the most important. Furthermore we make a comparison with
similar results presented in previous papers. Finally we also study at which values of
the kinematical variables the soft-plus-virtual (S+V) gluons start to dominate the cross
sections. This will give an indication about the validity of the S+V gluon approximation
which is of importance when one wants to improve the perturbation series via resummation
techniques leading to a better prediction for the differential cross sections.
The hadronic cross section d is obtained from the partonic cross section dab as
follows
S
2d
2 H1 H2
dT dU
S, T , U, m
1
a,b=q,g x1,min
dx1
x1
1
x2,min

dx2 H1
fa x1 , 2
x2

d 2 ab

s, t, u, m2 , 2 .
fbH2 x2 , 2 s 2
dt du
(6.1)
267
In analogy to Eq. (2.12) the hadronic kinematical variables are defined by

S = (P1 + P2 )2 ,
T = (P1 + p5 )2 ,
U = (P2 + p5 )2 ,
(6.2)
where P1 and P2 denote the momenta of hadrons H1 and H2 , respectively (see Eq. (2.7)).
In the case parton p1 emerges from hadron H1 (P1 ) and parton p2 emerges from hadron
H2 (P2 ) we can establish the following relations
p1 = x1 P1 ,
s = x1 x2 S,
x1,min =
p2 = x2 P2 ,

t = x 1 T m2 + m2 ,
U
,
S + T m2
x2,min =

u = x 2 U m2 + m2 ,
x1 (T m2 ) m2
.
x 1 S + U m2
(6.3)
When the matrix element behaves as 1/s4 , which occurs in the case of gluon bremsstrahlung or collinear fermion pair emission, it is more convenient to choose the integration
variable s4 instead of x2 in order to get better numerical stability. The cross section
becomes
S2

d 2 H1 H2
S, T , U, m2
dT dU
s4,max
1
dx1
ds4
=
x1
s4 x1 (T m2 ) m2
a,b=q,g x1,min

d 2 ab

faH1 x1 , 2 fbH2 x2 (s4 ), 2 s 2
s, t, s4 , m2 , 2 .
dt du
The function x2 (s4 ) can be derived from

s4 = s + t + u m2 = x1 x2 S + x1 T m2 + x2 U m2 + m2 ,
(6.4)
(6.5)
and it reads
x2 (s4 ) =
s4 x1 (T m2 ) m2
,
x 1 S + U m2

s4,max = x1 S + T m2 + U.
(6.6)
When parton p1 emerges from hadron H2 (P2 ) and parton p2 emerges from H1 (P1 ) one
obtains the same expressions as in Eqs. (6.3)(6.6) except that T and U are interchanged.
This result has to be added to Eq. (6.4). When the partonic cross section is symmetric under
t u one can also use the representation in Eq. (6.4) without adding the result where
T and U are interchanged provided one makes the replacement faH1 fbH2 faH1 fbH2 +
faH2 fbH1 .
The expression in Eq. (6.4) simplifies when the partonic cross sections correspond to
the Born reactions in Eq. (2.13) or the S+V contributions in Eqs. (5.24)(5.26). In this case
d 2 (s4 ) and the integral in Eq. (6.4) becomes one-dimensional. For two-to-three body
processes with no soft gluons or collinear fermion pairs in the final state the integral is twodimensional as given in Eq. (6.4). If the processes are of the gluon bremsstrahlung type,
or contain collinear fermion pairs in the final state we have to split the two-to-three body
partonic cross section in Eq. (4.40) into two parts, i.e., a hard gluon part with s4 > and
268
a soft gluon part with s4 . The terms containing the cut off parameter cancel between
the S+V cross sections originating from Eqs. (5.24)(5.26) and the corresponding terms
in the hard gluon hadronic cross section in Eq. (6.4), namely,
1
s
4,max
dx1
x1
x1,min

ds4
faH1 x1 , 2 fbH2 x2 (s4 ), 2
2
2
s4 x1 (T m ) m
2 HARD
2 d ab

(6.7)
s, t, s4 , m2 , 2 .
dt du
In order to achieve the cancellation analytically we rewrite the expression above as follows
s4,max
1

dx1
ds4
faH1 x1 , 2 fbH2 x2 (s4 ), 2
2
2
x1
s4 x1 (T m ) m
s
x1,min
2 HARD
ab
1
+
x1,min
s, t, s4 , m ,
dx1
x1
s
4,max
s
4,max

ds4
faH1 x1 , 2
2
2
dt du
x1 (T m ) m

2 HARD

d
H2
ab
fb x2 (0), 2 lim s 2
s, t, s4 , m2 , 2
s4 0
dt du
s
2d

ds4
faH1 x1 , 2 fbH2 x2 (0), 2
2
2
x1 (T m ) m
2 HARD
2 d ab

s, t, s4 , m2 , 2 .
(6.8)
dt du
The expression between the square brackets is integrable in s4 so we can put = 0 in
it. One can perform the integration over s4 analytically in the second part by using the
HARD in the limit s 0. The lni /2 terms, which arise from
simple expressions d 2 ab
4
the integration over s4 , then cancel those appearing in the S+V cross sections.
In practice one is not interested in the Higgs boson differential cross sections depending
on T and U but rather in the rapidity y and the transverse momentum pT distributions.
Neglecting the masses of the incoming hadrons we have the following relations

2
2
2
T = m S pT + m cosh y + S pT2 + m2 sinh y,

U = m2 S pT2 + m2 cosh y S pT2 + m2 sinh y,
(6.9)
lim s
s4 0
so that the cross section becomes

2 H1 H2

d 2 H1 H2
2
2
2d
,
y,
m
S
=
S
S, T , U, m2 .
S,
p
T
2
dT dU
dpT dy
(6.10)
The kinematical boundaries are

m2 S T 0,
S T + m2 U
Sm2
+ m2 ,
T m2
(6.11)
269
from which one can derive

S
S
1
1
ln 2 y ln 2 ,
2 m
2 m
2 )2
(S
+
m
pT2 ,max =
m2 ,
4 S cosh2 y
0 pT2 pT2 ,max ,

with
(6.12)
or
(S m2 )2
p T2 ,max ,
0 pT2
4S
4S(pT2 + m2 )
1 1 + 1 sq
,
sq =
.
ymax = ln
2 1 1 sq
(S + m2 )2
ymax y ymax ,
with
(6.13)
Since the cross section diverges for pT 0 we cannot perform the integral over this
kinematical variable down to zero. However we can perform the integral over the rapidity
and obtain the transverse momentum distribution

d H1 H2
S, pT2 , m2 =
dpT
ymax

dy
ymax

d 2 H1 H2
S, pT2 , y, m2 ,
dpT dy
(6.14)
with ymax given in Eq. (6.13). There is an alternative way to obtain the distribution above.
This is shown in Appendix B. We checked that both procedures lead to the same numerical
result. In the case we plot the rapidity distribution we have to impose a cut pT ,min on the
transverse momentum integration, i.e.,

d H1 H2
S, y, m2 =
dy
p
T ,max
dpT
pT ,min

d 2 H1 H2
S, pT2 , y, m2 ,
dpT dy
(6.15)
with pT ,max given in Eq. (6.12). Actually the differential cross section dies off so fast that it
is sufficient to put pT ,max = 8 pT ,min . Finally we define what we mean by leading order
(LO) and next-to-leading order (NLO). In LO the differential cross section is defined by
d 2 (1)

d 2 LO
S, pT2 , y, m2 =
S, pT2 , y, m2 ,
dpT dy
dpT dy
(6.16)
and the gluongluon-Higgs coupling is given by G in Eq. (2.2) with C = 1 (see Eq. (2.6)).
We also adopt the leading logarithmic representation for the running coupling and the
parton densities. The NLO corrected differential cross section reads

d 2 NLO
S, pT2 , y, m2
dpT dy

(5) 2 2 (1)
d 2 (2)

s ( )
d
S, pT2 , y, m2 +
S, pT2 , y, m2 , (6.17)
= 1 + 22
4
dpT dy
dpT dy
where the LO contribution in this formula is now multiplied with C 2 = 1 + 22 s /4 .
Furthermore the running coupling and parton densities are represented in next-to-leading
order for which we have chosen the MS-scheme. In this way one obtains a result which
270
Table 1
Various parton density sets with the values for the QCD scale and the running coupling s
MRST98 (LO, lo05a.dat)
LO
5 = 130.5 MeV
sLO (MZ ) = 0.125
MRST98 (NLO, ft08a.dat)
NLO
= 220 MeV
5
=
181 MeV
LO
5
NLO
= 202 MeV
5
LO
=
131 MeV
5
NLO
= 173 MeV
5
= 220 MeV
NLO
5
sNLO (MZ ) = 0.1175
CTEQ4 (LO, cteq4l.tbl)

CTEQ4 (NLO, cteq4m.tbl)
GRV98 (LO, grv98lo.grid)
GRV98 (NLO, grvnlm.grid)
MRST99 (NLO, cor01.dat)
sLO (MZ ) = 0.132

sNLO (MZ ) = 0.116
sLO (MZ ) = 0.125
sNLO (MZ ) = 0.114
sNLO (MZ ) = 0.1175
is consistently corrected up to NLO. In our computations the number of light flavours is

taken to be nf = 5 which holds for the running coupling, the partonic cross sections and the
number of quark flavour densities. Further we have chosen for our plots the parton densities
obtained from the sets MRST98 [32] CTEQ4 [33], GRV98 [34] and MRST99 [35] (see
Table 1). Notice that the GRV sets do not contain charm and bottom quark densities. For
simplicity the factorization scale is set equal to the renormalization scale r . For our
plots we take 2 = m2 + pT2 unless mentioned otherwise. Here we want to emphasize
that the magnitudes of the cross sections are extremely sensitive to the choice of the
renormalization scale because the effective coupling constant G s (r ), which implies
that d LO s3 and d NLO s4 . However the slopes of the differential distributions are
less sensitive to the scale choice if they are only plotted over a limited range.
For the computation of the Higgs-gluongluon effective coupling constant G given in
Eq. (2.2) we choose the top-quark mass mt = 173.4 GeV/c2 and the Fermi constant GF =
1.16639 GeV2 = 4541.68 pb. In this paper we will only
study Higgs boson production in
protonproton collisions at the center of mass energy S = 14 TeV characteristic of the
LHC. Since the hadrons H1 and H2 are now identical the y differential cross sections are
symmetric.
The effect of the NLO corrections to Higgs-boson production were already studied
earlier by the authors in [14]. In order to compare with their results we present LO and NLO
differential cross sections in pT , integrated over y, for m = 120 GeV/c2 and 2 = m2 +pT2
in Fig. 1(a) and 1(b), respectively. The MRST98 parton densities [32] were used for these
plots. We note that the NLO results from the q(q)g
and qq channels are negative at
small pT so we have plotted their absolute values multiplied by 100. It is clear that the
gg subprocess dominates but the q(q)g-subprocess
is also important. The corresponding

results for the y distributions integrated over the pT region between pT ,min = 30 GeV/c
and pT ,max = 240 GeV/c are given in Fig. 2(a) and 2(b), respectively. The latter value
was chosen because the cross section above 240 GeV/c is extremely small and can be
neglected. Here the scale is 2 = m2 + pT2 ,min . Again we see that the gg-subprocess
dominates. Using the recent MRST99 set [35] we present the dependence of the NLO
pT - and y-distributions on the Higgs mass in Fig. 3 and 4, respectively.
Next we plot the scale-dependence of the above differential cross sections in Figs. 5
and 6. For the pT -distributions we have chosen the scale factors = 20 , = 0 and =
0 /2 with 20 = m2 + pT2 . In the case of the y-distributions we adopted 20 = m2 + pT2 ,min
where pT ,min = 30 GeV/c (see Eq. (6.15)). Furthermore we have again adopted the
271
(a)
(b)
Fig. 1. (a) The differential cross section d/dpT integrated over the whole rapidity range (see Eq. (6.14)) with
m = 120 GeV/c2 and 2 = m2 + pT2 . The LO plots are presented for the subprocesses gg (long-dashed line),
q(q)g
(dot-dashed line) and 100 (q q)
(dotted line) using the parton density set MRST98(lo05a.dat). (b) Same
as (a) in NLO except for 100 abs(q q)
(dotted line) and the additional subprocess 100 abs(qq) (short-dashed
line) using the parton density set MRST98(ft08a.dat).
272
(a)
(b)
Fig. 2. (a) The differential cross section d/dy for pT ,min = 30 GeV/c and pT ,max = 240 GeV/c (see Eq. (6.15))
with m = 120 GeV/c2 and 2 = m2 + pT2 ,min . The LO plots are presented for the subprocesses gg (long-dashed
line), q(q)g
(dot-dashed line) 100 q(q)
(dotted line) using the parton density set MRST98(lo05a.dat). (b) Same
as (a) in NLO with the additional subprocess 100 qq (short-dashed line) using the parton density set
MRST98(ft08a.dat).
273
Fig. 3. The mass-dependence of d NLO /dpT (Eq. (6.14)) using the set MRST99 with 2 = m2 + pT2 for Higgs
masses m = 120 GeV/c2 (solid line), m = 160 GeV/c2 (dashed line) and m = 200 GeV/c2 (dot-dashed line).
Fig. 4. The mass-dependence of d NLO /dy for pT ,min = 30 GeV/c (Eq. (6.15)) using the set MRST99 with
2 = m2 + pT2 ,min . The notation is the same as in Fig. 3.
274
(a)
(b)
Fig. 5. (a) The scale-dependence of d LO /dpT integrated over the whole rapidity range (see Eq. (6.14)) with
m = 120 GeV/c2 , 20 = m2 + pT2 using the MRST98 parton density sets. The results are shown for = 0 /2
(dashed line), = 0 (solid line) and = 2 0 (dot-dashed line). The upper three curves are the NLO results.
(b) The quantity N (pT , /0 ) (see Eq. (6.18)) plotted in the range 0.1 < /0 < 10 at fixed values of pT
with m = 120 GeV/c2 and 20 = m2 + pT2 using the MRST98 parton density sets. The results are shown for
pT = 30 GeV/c (solid line), pT = 70 GeV/c (dashed line), pT = 100 GeV/c (dot-dashed line). The upper three
curves on the left-hand side are the LO results whereas the lower three curves refer to NLO.
275
Fig. 6. The scale-dependences of d LO /dy and d NLO /dy integrated over the region pT ,min = 30 GeV/c and
pT ,max = 240 GeV/c (see Eq. (6.15)). The notation and parameters are as in Fig. 5.
MRST98 parton density set in [32] since it contains both LO and NLO versions. In the case
of the pT -distribution in Fig. 5(a) one observes a small reduction in the scale dependence
while going from LO to NLO. This reduction becomes more visible when we plot the
quantity

d (pT , )/dpT
N pT ,
(6.18)
=
0
d (pT , 0 )/dpT
in the range 0.1 < /0 < 10 at fixed values of pT = 30, 70 and 100 GeV/c. The upper
set of curves at small /0 are for LO and the lower set are for NLO. Notice that the
NLO plots at 70 and 100 are extremely close to each other and it is hard to distinguish
between them. Further one sees that the slopes of the LO curves are larger that the slopes
of the NLO curves. This is an indication that there is better stability in NLO, which was
expected. However there is no sign of a flattening or an optimum in either of these curves.
This implies that one will have to calculate the differential cross sections in NNLO to find
a better stability under scale variations. In the case of the y-distributions we adopted the
scale 20 = m2 + pT2 ,min where pT ,min = 30 GeV/c (see Eq. (6.15)). As shown in Fig. 6
there is hardly any reduction in the scale-dependence for the y-distributions between the
LO and NLO curves.
Besides the dependence on the factorization and renormalization scales there are two
other uncertainties which affect the predictive power of the theoretical cross sections. The
first one concerns the rate of convergence of the perturbation series which is indicated by
276
the K-factor defined by

d NLO
.
(6.19)
d LO
Another uncertainty is the dependence of the cross section on the specific choice of parton
densities, which can be expressed by the factors
K=
d CTEQ
d GRV
GRV
,
R
=
.
(6.20)
d MRST
d MRST
The quantities in Eqs. (6.19) and (6.20) are plotted for the pT distributions in Fig. 7(a)
and (b), respectively, where we have chosen the same parameters and scales ( = 0 )
as in Figs. 5(a), (b) and 6. Depending on the parton density set the K-factors shown in
Fig. 7(a) are pretty large and vary from 1.4 at pT = 30 GeV/c to 1.7 at pT = 150 GeV/c.
In Fig. 7(b) we show the dependences of the ratios defined in Eq. (6.20) on the parton
density sets. From this figure we infer that both the GRV98 and the CTEQ4 densities
lead to larger cross sections than those computed from MRST98. The differences between
the various NLO cross sections is maximal at small pT and diminishes at large pT .
There is a uniform fall-off in the R values as a function of pT . In the case of the ydistributions in Fig. 8(a) we observe the same pattern. Again the CTEQ4 parton densities
yield the smallest K-factor, while there is not much difference between the results for the
MRST98 and GRV98 sets. Notice that there are small differences between Fig. 7(a), (b)
and the corresponding Fig. 2(a) in [14]. We almost agree with the K-factors but find a
stronger dependence of the R-factors with respect to pT . The same applies to Fig. 8(a), (b)
compared with the corresponding Fig. 2(b) in [14]. Although the general pattern of the
curves is similar there are small differences. In order to track down these differences the
authors of [14] informed us that they did not use a fixed scale at 20 = pT2 ,min + m2 but
a variable scale 20 = pT2 + m2 when integrating over pT to calculate the y-distributions.
We have also run our programs with the latter scale, which yields different y-distributions.
The y-cross sections become smaller due to the smaller coupling constant but the R-ratios
hardly change with respect to the ones given in Fig. 8(b). The discrepancy between our
results and those in [14] becomes more conspicuous if we look at the absolute values of
the differential cross section d/dpT . To compare we have read off the central values
of the bins from their Fig. 1(a) and, after changing bin sizes, have replotted their values
versus ours in Fig. 9(a) for the MRST98 set and the same input parameters. We only show
the case where the scale is = 0 . Although the results are similar they are not identical.
Even taking into account the small errors in reading their bin values (which are the cause
of the oscillations in the FGK curves) the slopes of their results are slightly steeper than
ours. To show the differences more clearly we have also plotted the ratio between our
curves and theirs in Fig. 9(b). From this figure one infers that at small pT our cross section
is smaller whereas at large pT it becomes larger than the cross section obtained in [14].
This observation also holds for the LO cross section which is already known for a very
long time. Therefore to be sure we also made a comparison with the LO pT -distribution in
Fig. 6 of [5] by reading off their values. Choosing the same input parameters and parton
density set we completely agree with the results in [5]. The authors of [14] have informed
us that the binned results shown in Fig. 1(a) of their paper have been smoothed (with
R CTEQ =
277
(a)
(b)
Fig. 7. (a) The K-factors in Eq. (6.19) for d/dpT integrated over the whole rapidity range (see Eq. (6.14)) with
m = 120 GeV/c2 and 2 = m2 + pT2 for the MRST98 sets (solid line), the GRV98 sets (dot-dashed line), and the
CTEQ98 sets (dashed line). (b) The ratios R in Eq. (6.20) for the same differential cross sections as in (a), R GRV
in LO (solid line), R GRV in NLO (dot-dashed line), R CTEQ in LO (dashed line) and R CTEQ in NLO (dotted
line).
278
(a)
(b)
Fig. 8. (a) The K-factors in Eq. (6.19) for d/dy integrated over the region pT ,min = 30 GeV/c and
pT ,max = 240 GeV/c (see Eq. (6.15)) with m = 120 GeV/c2 and 2 = m2 + pT2 ,min . The parton density sets
and notation are as in Fig. 7(a). (b) The ratios R in Eq. (6.20) for the same differential cross sections as in (a).
The notation is the same as in Fig. 7(b).
279
(a)
(b)
Fig. 9. (a) Comparison of d/dpT in LO and NLO obtained from our calculation (RSN) versus the one in Fig. 1(a)
in [14] (FGK). (b) Same as in (a) but now for the ratios in LO and NLO.
280
a corresponding rebinning) in order to make the results clearer. This is often necessary in
an exclusive calculation where the other variables are integrated over numerically. In an
inclusive calculation such as ours there is no need for any rebinning as these integrations
have been done analytically. We conclude that the rebinning is the probable cause of the
small differences between our results and those in [14]. Note that we have checked several
of the figures in [22] using the same parton densities and parameters and find excellent
agreement. This yields an independent check on the validity of our NLO results in the gg
channel.
We now analyse why the MRST98, GRV98 and CTEQ4 parton densities yield different
results for the differential cross sections. This can be mainly attributed to the small
x-behaviour of the gluon density because gluongluon fusion is the dominant production
mechanism. To investigate the small x-behaviour we consider the product of the gluon flux
with the corresponding partonic cross section

dgg

xS, pT2 , 2 .
Fgg xS, pT2 , 2 = gg x, 2
(6.21)
dpT
In this case we have removed the factor G2 s /4 from the LO (Born) cross section
and the factor G2 (s /4)2 from the NLO contribution to the partonic cross section.
This was done to suppress the dependence on the strong coupling constant s so that the
differences between the several Fgg can be only attributed to the various parton density
(1)
sets. In Fig. 10(a) we have plotted the LO partonic cross section dgg
/dpT versus log10 x
3
for different values for pT . For x < 10 this cross section rises steeply whereas it flattens
out for x > 103 . The latter behaviour is changed when we look at the NLO partonic cross
(2)
section dgg /dpT plotted versus log10 x in Fig. 10(b). Here the rise at small x becomes
even steeper whereas for x > 103 the flatness shown by the Born cross section is replaced
by a steep decrease when x 1 except at high pT . Notice that the NLO partonic cross
section becomes negative (here for x 8.0 102 ) due to mass factorization as explained
below Eq. (5.15).
To show the effect of the gluon flux we have plotted Fgg (xS, pt2 , m2 ) as a function of
log10 x in Fig. 11 for pT = 100 GeV/c using three parton density sets in Table 1. From
Fig. 11 we infer that the GRV98 set contains the steepest gluon density whereas the gluon
densities from the other sets are about equal. From this observation one can understand
the relative ordering of the plots in Figs. 7(a) and (b). Another feature is that the hadronic
cross section, which is represented by the integral over x of Fgg , receives its main support
from the small x-region. Using the representation for the hadronic cross section d/dpT
as given in Eq. (B.7) we have computed this quantity where now all subprocesses are
included. Further we have chosen two different integration regions. The first one is the
full range xmin x xmax with xmax = 1 (see Eq. (B.8) and the second range is given by
xmax = 5 xmin . In Fig. 12 we have shown the pT -distributions due to the two different
integration regions. They do not differ by more than 10% which shows that the whole
differential cross section is dominated by the small x-region x < 5 xmin . Hence apart
from the different coupling constants the difference between the cross sections is due
to the different gluon densities. Future HERA data will have to provide us with unique
gluon densities before we can make more accurate predictions for the Higgs differential
distributions. Note that we have chosen the axes in Fig. 12 to coincide with those in
281
(a)
(b)
(1)
Fig. 10. (a) The partonic differential cross section dgg /dpT (Born) with m = 120 GeV/c2 and 2 = m2 + pT2
for pT = 200 GeV/c (solid line), pT = 150 GeV/c (long-dashed line), pT = 100 GeV/c (dot-dashed line),
pT = 75 GeV/c (short-dashed line) and pT = 50 GeV/c (dotted line). (b) Same as in (a) but now for the NLO
(2)
differential cross section dgg /dpT .
282
Fig. 11. The flux multiplied by the partonic cross section represented by the quantity Fgg (xS, pT2 , m2 ) in
Eq. (6.21) with pT = 100 GeV/c, m = 120 GeV/c2 and 2 = m2 + pT2 plotted as a function of log10 x for
MRST98 (solid lines), GRV98 (dot-dashed lines) and CTEQ4 (dashed lines). The upper curves are for NLO and
the lower ones for LO.
Fig. 1(b) so that one can see the pT -dependence of the K-factor for these two densities
by overlaying the plots.
After having studied the small x-region we now investigate the large x-region of the
partonic cross sections where S+V gluons and collinear quark antiquark pairs dominate
the radiative corrections. The S+V gluon part of the partonic cross section is obtained
by omitting the hard contributions which are regular at s4 = 0 so that we only keep the
singular parts presented in Eqs. (5.16)(5.20). Furthermore we include the S+V partonic
cross sections in Eqs. (5.24)(5.26). These two contributions constitute the S+V gluon
approximation. To study its validity we compute in NLO the ratio
d S+V
(6.22)
,
d EXACT
for the pT and y distributions which are given in Figs. 13 and 14, respectively. Here
we use the MRST99 parton density set. One expects that the approximation becomes
better at larger transverse momenta where pT = p T ,max in Eq. (6.13) at the boundary
of phase space, which leads to x = 1 using Eq. (B.8). However in Fig. 13 the highest
value of pT , given by pT = 150 GeV/c, is still very small with respect to pT ,max
S/2 = 7 103 GeV/c. Therefore it is rather fortuitous that the approximation works
so well for pT > 100 GeV/c where one obtains R S+V < 1.2. The boundary of phase
R S+V =
283
Fig. 12. The NLO hadronic cross section d/dpT obtained from Eq. (B.7) (all subprocesses included) with
m = 120 GeV/c2 and 2 = m2 + pT2 . The parton density set is GRV98 with xmax = 1 (solid line) and with
xmax = 5xmin (dashed line).
space is also approached when the Higgs mass increases. This

can also be inferred from
Eq. (B.8) where at fixed pT , xmin 1 when m2 S 2 S pT . Therefore we have
plotted R S+V for various Higgs boson masses in Fig. 13. This figure reveals that the
largest Higgs mass leads to the worst approximation contrary to our expectations which
means that the kinematics is not in the large x-region. In Fig. 14 we plot the ratio R S+V
for the y-distributions for the same mass range as in Fig. 13 where we have chosen the
cut pT,min = 30 GeV/c in Eq. (6.15). Here the approximation is less good than for the
pT distributions in Fig. 12 and like in the latter figure it becomes better when the mass
of the Higgs gets smaller. At the lowest value chosen for the mass, i.e., m = 120 GeV/c2
we observe that R S+V 1.4. Furthermore we see no variation in R S+V with respect to
the values of y. The approximation becomes better, see Fig. 15, if pT ,min is chosen to be
larger than 30 GeV/c so that it becomes closer to the values given in Fig. 13. Finally the
figures discussed above show that the S+V gluon approximation overestimates the exact
NLO result. However this overestimate becomes smaller when the transverse momentum
gets larger. In particular for pT > 200 GeV/c the S+V approximation is good enough
so that resummation techniques can be used to give a better estimate of Higgs boson
production corrected up to all orders in perturbation theory. This statement will also hold
when Higgs boson production is described according to the standard model approach
where the boson is coupled via top-quark loops to the gluons without taking mt
as we did above. Although the pT distributions will change for the exact and S+V gluon
284
Fig. 13. The ratio R S+V in Eq. (6.22) for the pT distributions using the set MRST99 with 2 = m2 + pT2 ,min
and various Higgs masses given by m = 120 GeV/c2 (solid line), m = 160 GeV/c2 (dashed line) and
m = 200 GeV/c2 (dot-dashed line).
Fig. 14. The ratio R S+V in Eq. (6.22) for the y distributions using the MRST99 set with 2 = m2 + pT2 ,min for
pT ,min = 30 GeV/c and the same Higgs masses as in Fig. 13.
285
Fig. 15. The ratio R S+V in Eq. (6.22) for the y distributions using the MRST99 set with 2 = m2 + pT2 ,min and
m = 120 GeV/c2 , for various values of pT ,min , namely pT ,min = 100 GeV/c (solid line), pT ,min = 200 GeV/c
(dashed line), pT ,min = 250 GeV/c (dot-dashed line) and pT ,min = 300 GeV/c (dotted line).
approximation at large pT , the ratio R S+V in Fig. 13 will be less affected by the large top
quark mass approach because of cancellations between the numerator and denominator in
Eq. (6.22). Notice that the S+V gluon approach in the case of the exact cross section can
be obtained from Eqs. (5.16)(5.20) and (5.24)(5.26) by replacing the approximate Born
(1)
matrix elements Mij k H , obtained from the effective Lagrangian in Eq. (2.1), by the exact
expressions presented in [46].
Summarizing the above we have calculated the NLO corrections to the differential
cross section for Higgs boson production in the large top quark mass approach which
can be obtained from an effective Lagrangian. The calculation was carried out in standard
n-dimensional regularization where the MS scheme was chosen for renormalization and
mass factorization. We have presented some of the NLO results in the region where the
effective Lagrangian should be trustworthy. It turns out that the gluongluon subprocess
dominates the hadronic cross section but the (anti)quarkgluon subprocess is certainly not
negligible. For the transverse momentum distributions there is a small reduction in scaledependence while going from the LO to the NLO cross section which is not observed for
the rapidity distributions. Further there is still a large uncertainty in our predictions because
the K-factor varies from approximately 1.4 to 1.7 depending on the parton density set. Also
the dependence on the running coupling and the parton density set is appreciable. The latter
is mainly due to the small x behaviour of the various gluon densities since both the partonic
cross sections and the gluon densities increase very steeply at decreasing x. Finally we have
286
shown that the S+V approximation is quite reasonable provided pT ,min > 100 GeV/c in
spite of the fact that x is still too small to belong to the large x-region. This means that this
approximation can be used to resum the large corrections due to S+V gluons in order to
obtain a better estimate of the all order corrected cross section.
Acknowledgements
We thank S. Dawson, B. Field and R. Kauffman for their assistance in tracking down
the misprints in [12]. We also thank D. de Florian, M. Grazzini and Z. Kunszt for their
comments on the first version of this manuscript.
Appendix A
Here we list the soft contributions to the partonic cross sections as defined in Eq. (4.42)
which were not explicitly given in Section 4. For the various processes we obtain
s2
s2
s2
SOFT
d 2 gggg
H
dt du

2 2

N
2 2 2 s ( )
= s + t + u m S G
2
4
(N 1)2

47 1
tu
tu
+ ln 2 ln 2 + ln2 2
2 + 4 ln 2 + 2 ln 2
s
6
s 2

2
59
tu
5
295 (1)
1 2 tu
35
ln
ln
(2) +
+ ln 2
Mggg H ,
2
s 12 2 12 2 s 2
36
(A.1)
SOFT
d 2 ggq
q H
dt du

2 2

1
2 2 2 s ( )
nf
= s + t + u m S G
2
4
(N 1)2

2
1
tu
11 (1)
1
1
+ ln 2 + ln 2
Mggg H ,
3 6
6 s 18
d 2 qSOFT
qgg
H
dt du

s (2 ) 2 1
= s + t + u m2 S2 G2
4
N2

11 1
tu
tu
2
+ ln 2 ln 2 ln2 2
CA 2 + 2 ln 2
s
6
s 2

tu
67
1 2 tu
11
11
3
ln
ln
+ ln 2
(2) +
2
36
s 12 2 12 2 s 2
(A.2)

+ CF
s2
s2

(1)
2
1
4
2
M

+
2
ln
+
4
ln
(2)
q qg
H ,
2
2
2
287
(A.3)
SOFT
d 2 qgqg
H
dt du

1
s (2 ) 2
= s + t + u m2 S2 G2
4
N(N 2 1)

1
2
1
2
CA 2 + 2 ln 2 1
+ ln 2 ln 2 (2) +
2
2

tu
tu
7 1
+ ln 2 ln 2
+ CF 2 + 2 ln 2 + 2 ln 2
2
s

2
7
7 (1)
1 2
1 2 tu
7
tu
Mqgq H ,
+ ln 2 + ln 2 ln 2 ln 2 2(2) +
2
2
4 s
2
s 4
(A.4)
d 2 qSOFT
qq
q H
dt du

s (2 ) 2
1
= s + t + u m2 S2 G2
nf 2
4
N

2

1
tu
1
1
5 (1)

Mq qg
+ ln 2 + ln 2
H .
3 6
6 s 18
(A.5)
Appendix B
There is an alternative way to obtain the transverse momentum distribution as presented
in Eq. (6.14). First one computes the partonic cross section

dab
s, pT2 , m2 , 2 =
2
dpT
s
4,max
ds4
0

d 2 ab
s, pT2 , s4 , m2 , 2 ,
2
dpT ds4
(B.1)
with

d 2 ab
s
s, pT2 , s4 , m2 , 2 =
2
dpT ds4
(s + m2 s4 )2 4s(pT2 + m2 )
2

d 2 ab

d ab
s, t, u, m2 , 2 +
s, u, t, m2 , 2 ,
dt du
dt du
and

s4,max = s + m2 2 s pT2 + m2 .
(B.2)
(B.3)
288
The reason is that if one changes the variables t and u into s4 and pT2 one has two
possibilities

2

1
t = s4 + m2 s + s + m2 s4 4s pT2 + m2 t1 ,
2

2

2
1
2
2
2
u1 ,
u = s4 + m s s + m s4 4s pT + m
(B.4)
2
or

2

2
1
2
2
2
t = s4 + m s s + m s4 4s pT + m
t2 = u1 ,
2

2

1
u = s4 + m2 s + s + m2 s4 4s pT2 + m2 u2 = t1 ,
2
(B.5)
which follows from the substitution

(s + m2 s4 )2 4s(pT2 + m2 )
sinh y =
,

2 s pT2 + m2
s + m2 s4
cosh y =
,
2 s pT2 + m2
(B.6)
in Eq. (B.2). Hence one has to compute the sum d (t1 , u1 ) + d (t2 , u2 ) which is equal to
d (t1 , u1 ) + d (u1 , t1 ). Notice that when the partonic cross section is symmetric in t and
u one can replace the sums above by 2d (t1 , u1 ). This does not apply to the quarkgluon
subprocess because here the cross section is asymmetric in t and u. The hadronic cross
section is now obtained from

d H1 H2
S, pT2 , m2 =
dpT
xmax
a,b=q,g xmin
dab

H1 H2
dx ab
x, 2
xS, pT2 , m2 , 2 ,
dpT
(B.7)
with
xmin =

m2 + 2pT2 + 2 pT2 (pT2 + m2 )
S
xmax = 1,
(B.8)
and ab denotes the partonic flux defined by

H1 H2
ab
x, 2 =
1
1
dx1

dx2 (x x1 x2 )faH1 x1 , 2 fbH2 x2 , 2 .
(B.9)
The lower boundary xmin follows from the inequality in Eq. (6.13) where s = xS. We have
checked that expressions Eqs. (6.14) and (B.7) lead to the same result.
289
References
[1] R. Barate et al., ALEPH Collaboration, Phys. Lett. B 495 (2000) 1;
M. Acciarri et al., L3 Collaboration, Phys. Lett. B 495 (2000) 18;
P. Abreu et al., DELPHI Collaboration, Phys. Lett. B 499 (2001) 23;
G. Abbiendi et al., OPAL Collaboration, Phys. Lett. B 499 (2001) 38.
[2] J.F. Gunion, H.E. Haber, G.L. Kane, S. Dawson, The Higgs Hunters Guide, AddisonWesley, Reading,
MA, 1990;
J.F. Gunion, H.E. Haber, G.L. Kane, S. Dawson, hep-ph/9302272, Erratum.
[3] F. Wilczek, Phys. Rev. Lett. 39 (1977) 1304;
H. Georgi, S. Glashow, M. Machacek, D. Nanopoulos, Phys. Rev. Lett. 40 (1978) 692;
J. Ellis, M. Gaillard, D. Nanopoulos, C. Sachrajda, Phys. Lett. B 83 (1979) 339;
T. Rizzo, Phys. Rev. D 22 (1980) 178.
[4] I. Hinchliffe, S.F. Novaes, Phys. Rev. D 38 (1988) 3475.
[5] R.K. Ellis, I. Hinchliffe, M. Soldate, J.J. van der Bij, Nucl. Phys. B 297 (1988) 221.
[6] R.P. Kauffman, Phys. Rev. D 44 (1991) 1415;
R.P. Kauffman, Phys. Rev. D 45 (1992) 1512.
[7] D. Graudenz, M. Spira, P. Zerwas, Phys. Rev. Lett. 70 (1993) 1372;
M. Spira, A. Djouadi, D. Graudenz, P. Zerwas, Nucl. Phys. B 453 (1995) 17.
[8] V. Del Duca, W. Kilgore, C. Oleari, C. Schmidt, D. Zeppenfeld, Phys. Rev. Lett. 87 (2001) 122001;
V. Del Duca, W. Kilgore, C. Oleari, C. Schmidt, D. Zeppenfeld, Nucl. Phys. B 616 (2001) 367.
[9] S. Dawson, Nucl. Phys. B 359 (1991) 283;
A. Djouadi, M. Spira, P. Zerwas, Phys. Lett. B 264 (1991) 440.
[10] U. Baur, E. Glover, Nucl. Phys. B 339 (1990) 38.
[11] S. Dawson, R.P. Kauffman, Phys. Rev. Lett. 68 (1992) 2273.
[12] R.P. Kauffman, S.V. Desai, D. Risal, Phys. Rev. D 55 (1997) 4005;
R.P. Kauffman, S.V. Desai, D. Risal, Phys. Rev. D 58 (1998) 119901, Erratum.
[13] C.R. Schmidt, Phys. Lett. B 413 (1997) 391.
[14] D. de Florian, M. Grazzini, Z. Kunszt, Phys. Rev. Lett. 82 (1999) 5209.
[15] R.V. Harlander, Phys. Lett. B 492 (2000) 74.
[16] R.V. Harlander, W.B. Kilgore, Phys. Rev. D 64 (2001) 013015;
S. Catani, D. de Florian, M. Grazzini, JHEP 0105 (2001) 025.
[17] R.V. Harlander, W.B. Kilgore, hep-ph/0201206.
[18] B. Arnold, M.H. Reno, Nucl. Phys. B 319 (1989) 37;
B. Arnold, M.H. Reno, Nucl. Phys. B 330 (1990) 284, Erratum;
R.J. Gonsalves, J. Pawlowski, C.-F. Wai, Phys. Rev. D 40 (1989) 2245.
[19] S. Frixione, Z. Kunszt, A. Signer, Nucl. Phys. B 467 (1996) 399;
S. Frixione, Nucl. Phys. B 507 (1997) 295.
[20] M. Krmer, E. Laenen, M. Spira, Nucl. Phys. B 511 (1998) 523.
[21] S. Catani, E. DEmilio, L. Trentadue, Phys. Lett. B 211 (1988) 335;
C.-P. Yuan, Phys. Lett. B 283 (1992) 395;
C. Balazs, C.-P. Yuan, Phys. Rev. D 59 (1999) 114007;
C. Balazs, C.-P. Yuan, Phys. Rev. D 63 (2001) 059902, Erratum;
C. Balazs, C.-P. Yuan, Phys. Lett. B 478 (2000) 192;
C. Balazs, J. Huston, I. Puljak, Phys. Rev. D 63 (2001) 014021;
D. de Florian, M. Grazzini, Phys. Rev. Lett. 85 (2000) 4678;
D. de Florian, M. Grazzini, Nucl. Phys. B 616 (2001) 247.
[22] C.J. Glosser, hep-ph/0201054.
[23] FORM by J.A.M. Vermaseren, version 3.0 available from http://www.nikhef.nl/form.
[24] K.G. Chetyrkin, B.A. Kniehl, M. Steinhauser, Phys. Rev. Lett. 79 (1997) 353.
[25] G. Passarino, M. Veltman, Nucl. Phys. B 160 (1979) 151.
[26] W. Beenakker, PhD thesis, Universiteit Leiden, 1989.
[27] T. Matsuura, S.C. van der Marck, W.L. van Neerven, Nucl. Phys. B 319 (1989) 570.
[28] L. Lewin, Polylogarithms and Associated Functions, North-Holland, Amsterdam, 1983.
290
[29] W. Beenakker, H. Kuijf, W.L. van Neerven, J. Smith, Phys. Rev. D 40 (1989) 54.
[30] A. Erdely (Ed.), Higher Transcendental Functions, Bateman manuscript, Vol. 1, McGrawHill, New York,
1953;
M. Abramowitz, J.A. Stegun, Handbook of Mathematical Functions, National Bureau of Standards, 1964.
[31] H. Kluberg-Stern, J.B. Zuber, Phys. Rev. D 12 (1975) 467;
V.P. Spiridonov, K.G. Chetyrkin, Yad. Fiz. 47 (1988) 818.
[32] A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Eur. Phys. J. C 4 (1998) 463.
[33] H. Lai et al., Phys. Rev. D 55 (1997) 1280.
[34] M. Glck, E. Reya, A. Vogt, Eur. Phys. J. C 5 (1998) 461.
[35] A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Eur. Phys. J. C 14 (2000) 133.

Conformal invariant saturation

H. Navelet, R. Peschanski
CEA, Service de Physique Theorique, CE-Saclay, F-91191 Gif-sur-Yvette Cedex, France
Abstract
We show that, in oniumonium scattering at (very) high energy, a transition to saturation happens
due to quantum fluctuations of QCD dipoles. This transition starts when the order 2 correction
of the dipole loop is compensated by its faster energy evolution, leading to a negative interference
with the tree level amplitude. After a derivation of the one-loop dipole contribution using conformal
invariance of the elastic 4-gluon amplitude in high energy QCD, we obtain an exact expression
of the saturation line in the plane (Y , L) where Y is the total rapidity and L, the logarithm of
the onium scale ratio. It shows universal features implying the BalitskyiFadinKuraevLipatov
(BFKL) evolution kernel and the square of the QCD triple Pomeron vertex. For large L, only the
higher BFKL Eigenvalue contributes, leading to a saturation depending on leading log perturbative
QCD characteristics. For initial onium scales of same order, however, it involves an unlimited
summation over all conformal BFKL Eigenstates. In all cases, conformal invariance is preserved for
the saturation mechanism based on dipole loops. 2002 Elsevier Science B.V. All rights reserved.
PACS: 12.38.-t; 13.60.-r
1. Saturation from QCD dipoles

A regime of QCD at small coupling constant where the density of partons begins to
saturate has since long been the subject of many studies [118]. In one class of models,
one expects the effect due to the high energy multiplication of partons due to the QCD
dynamics. Another way which was proposed to investigate high density partonic effects
is the consideration of collisions at high energy on heavy nuclei. Indeed, the number of
partons is supposed to be large in the background of the collision providing boundary
conditions favorable for the saturation mechanism to happen. In the present paper, we will
focus on the first scheme, i.e., the saturation mechanism due to the energy evolution of the
E-mail address: pesh@spht.saclay.cea.fr (R. Peschanski).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 1 4 - 0
292
H. Navelet, R. Peschanski / Nuclear Physics B 634 (2002) 291308
parton density in a purely perturbative QCD framework. More precisely, we will consider
even perturbative QCD boundary conditions, in order to select the features of saturation
which could be fully calculated from the QCD Lagrangian.
On the experimental ground, saturation has not been clearly identified in the evolution
of structure functions in the HERA range, even if some evidence for partial saturation
has been found by analyzing vector meson diffractive production in impact parameter
space [19]. The problem of identifying saturation effects in actual or future experiments
is very stimulating, since the increase of the energy range is at reach at Tevatron and
later at LHC. Even, some lessons will be available from experiments with heavy nuclei at
RICH. The goal of our purely perturbative calculations, even if they appear to be far from
the experimental realizations, is to give an example of saturation which could be fully
computed, without referring to a non-perturbative ansatz. In practice, later on, it will be
possible, using factorization properties, to combine our calculations with some reasonable
non-perturbative wave functions to get predictions for realistic processes, like hard hadron
hadron interactions at Tevatron and LHC.
Our idea is to start from the most elementary hard processes, namely oniumonium
scattering at (very) high energies. The reaction can be described [4,5] by the wavefunctions of the onia in terms of q q colour singlet dipoles at the time of the (hard)
interaction, which then interact through double gluon exchanges. This description, which
corresponds to the tree level dipole interactions, is equivalent to the usual Balitsky, Fadin,
Kuraev, Lipatov (BFKL) formalism [20] for the same process, as can be proven [21] using
the conformal invariant symmetry of the BFKL evolution equation. When energy increases,
the wave functions has an ever increasing density of colour dipoles and the tree level dipole
interaction is expected to be modified by unitarity corrections which are already present in
the perturbative regime. The first order correction is the one-loop dipole contribution which
was first considered in Ref. [22] as a formal calculation of next-to-next order contributions
in the coupling constant. In the present paper we show that, despite its higher order, this
one-loop unitarity correction gives rise to the onset of saturation in oniumonium scattering
and thus allows one to give a complete perturbative QCD description of the saturation
critical line. This saturation mechanism is to be distinguished from the one which would
come from the interaction of a onium (e.g., a collection of independent dipoles) with a large
nucleus (see, e.g., [8,1113,15,17,18]), which, in some cases, can be associated with a nonlinear branching process without dipole recombination. In the case we consider, the nonlinearities due to the high density of partons correspond to the recombination of QCD
dipoles.
We write the oniumonium elastic cross-section
1
d
|FQ |2 ,
=
(1)
dQ2 4
where Q is the transfer momentum. The amplitude reads

FQ = i
2
1
d d

dz dz (, z) , z fQ , Y ,
(2)
where the onium is defined as a heavy qq state with q (respectively q) having the fraction
z (respectively 1 z) of the longitudinal momentum in an infinite momentum frame. It
293
is defined by its wave-function (, z) over a basis of dipole states of transverse size

0 1 , where 0 , 1 are the positions of the dipoles constituents in transverse
coordinate space. fQ (, |Y ) is the dipoledipole elastic amplitude for a full rapidity
interval Y and a momentum transfer Q, the calculation of which at one dipole-loop order
and saturation properties are the subject of the present paper.
In fact the physical origin of the saturation mechanism that we want to investigate is the
following: it is known [4] that the wave function of an Onium in the infinite momentum
frame develop a large number of dipole states as a function of the rapidity (which plays
the rle of time in a cascading mechanism). However, the larger is the dipole number,
the smaller is their size and thus they do not contribute to saturation. This can be proven
rigorously by the fact that the dipoledipole cross-section add together to give back the
BFKL cross-section [21]. It is interesting to note the use of conformal invariance to get
this exact bootstrap property between an s-channel and t-channel description of the
same amplitude.
However, now if one computes the dipoledipole interaction at one-loop level, the
system is allowed to take into account quantum fluctuations where virtual dipole pairs may
be created during the time of interaction. Moreover, all sizes of dipoles can contribute in the
loop, which goes beyond the limitations of the tree-level mechanism. This is expected from
perturbative unitarity arguments [4] which expect the damping of the energy dependence of
the amplitude by saturation effects due to multi-loop dipole contributions. This is precisely
the goal of our work to put this expectation on a quantitative level.
Through unitarity arguments, the QCD saturation phase corresponding to the mechanism we consider requires an estimate of all dipole loop contributions, at least in the high
energy approximation. In the present paper we will concentrate on the first step, i.e., the
transition to saturation which is expected to be due to the interference between the tree and
1-loop level. We expect that the techniques which we were led to consider in this work,
relying notably on the conformal invariance properties, will be useful to investigate this
intringuing phase of matter.
Our main results are the following:
(i) Up to one-loop QCD dipole level, the dipoledipole amplitude reads

fQ , Y
(0)
(1)
fQ + fQ

h
2
h
Q
EQ () d(h) e(h)Y
= dh E
4

4 2(1/2)Y

g3P h, 12 , 12 2

h
2 e
h

E ()
1 3 dh E
Q
Q
(h) 2 1 .
4
2 Y
2
In Eq. (3), the symbolic notation

dh

n=
(3)
294
corresponds to the integration over the quantum numbers associated to the continuous
unitary irreducible representations of SL(2, C), namely,
1n
1+n
(4)
,
h = i +
, n N , R.
2
2
Note that for odd n there is no contribution to the elastic onium amplitude (see second
reference of [21]). The SL(2, C) Eigenvalue of the BFKL kernel is
h = i +
(h) =

Nc
Nc
(h)
2 (1) (h) 1 h ,
(5)
h () are the SL(2, C) Eigenfunctions [23]. g (h, h , h ) is

where (log ) , and EQ
a b
3P
the known QCD triple Pomeron coupling function [24,25], whose definition is given in
h
Appendix A, together with EQ
() and the other ingredients (such as the functions a(h),
d(h)) of the SL(2, C)-invariant formalism.
The symmetry of the tree level amplitude remains valid at one-loop level, leading to
a conformal invariant saturation scheme.
(ii) The evaluation of formula (3) when the condition

L log 1
is fulfilled can be done in the forward direction (Q = 0) in terms of the leading

Eigenfunction and Eigenvalue of the BFKL kernel, namely, with n = 0. By contrast, for
similar dipole sizes (L = O(1)) infinitely many Eigenfunctions and Eigenvalues contribute
to the loop expression, leading to a quite different structure.
The QCD critical saturation line is derived in the L 1 regime and is given by:

c)
(h ) (h
log L,
c + 5 log Y
Y
(6)
2L hc h = Y

1
2
2
(1/2)Y and Y
c O{log(1/ 2 )} is a large constant (formula (16) in the text).
where Y
The transition to saturation occurs when the first member of (6) exceeds the second one.
means that the rapidity (i.e., energy) dependence of the
Note that using the variable Y
amplitude is quantified by reference to the BFKL intercept 1 + (1/2). Hence the overall
, L dependence is determined by universal numbers, independent of the coupling
Y
constant.
In formula (6), there appears two critical values of the effective anomalous dimensions
hc for the tree level and h for the one-loop contribution. They are both defined as solution
of implicit equations
2L
(hc )
=
,

(1/2)
Y

h = 2(1/2).
(7)
hc is a moving saddle point in the tree-level integrand of (3) while h 0.183 is the fixed
location of a double pole singularity in the one-loop integrand of (3).
The plan of the paper is the following. The next Section 2 is devoted to the derivation
of formula (3) and Section 3 to that of formula (6). A discussion of the resulting saturation
line and an outlook are proposed in Section 4. The SL(2, C)-invariant formalism which we
295
use is presented in Appendix A and some technical parts of the derivation of the one-loop
amplitude are present in Appendices B and C.
2. The one-loop dipole amplitude

(1)
At one-loop level, the amplitude fQ has a simple and physically appealing formulation [4] in the transverse coordinate space, see Fig. 1. We write

1
fQ(1) , Y
(8)
d 2 r eiQ T (1) 0 1 ; 0 1 Y ,
2
with impact parameter r. The dipoledipole amplitude in coordinate space reads

T (1) 0 1 ; 0 1 Y = y + y
y
y
2
2
2
2
d 2 a0 d 2 a1 d 2 b0 d 2 b1 d a0 d a1 d b0 d b1
|a b |2
|a b |2
0
0

n2 0 1 ; a0 a1 , b0 b1 |y y,
y n 2 0 1 ; a 0 a 1 , b 0 b 1 |y y , y
T (a0 a1 , a a )T(b0 b1 , b b ),
d y
d y
(9)
Fig. 1. One-loop dipole amplitude in transverse coordinate space. The transverse coordinates i of the incident
and of the two interacting dipoles are simply denoted by their respective indices i. The coordinates 2, 2 refer to
the integration points, see Appendix B.
296
where 0 , 1 are the transverse coordinates of one of the initially colliding dipoles
(respectively, 0 , 1 for the second one), while a0 a1 and b0 b1 , are the two interacting
dipoles emerging from the dipole 0 1 after evolution in rapidity (respectively, {i}
{i } , for the second one).
The overall minus sign comes from the well-known AGK rules [26] relating the
amplitudes to the discontinuity here calculated in the c.o.m. frame of the dipoledipole
reaction. The tree level interaction amplitudes T are given in Appendix A. In formula (9),
y)
stands for the probability of producing two dipoles after
n2 (0 1 ; a0 a1 , b0 b1 |y y,
a mixed rapidity evolution [4,22], namely, with a rapidity y y with one-Pomeron type of
evolution and a rapidity y with two-Pomeron type of evolution, see Fig. 2. Further on, one
has to integrate over y,
y . The solution for n2 is given in Appendix B. The integration over
intermediate variables yields a drastic simplification due to the appearance of quite a few
-functions. This tedious but straightforward derivation is given in detail in Appendix C.
Starting with the general formula (C.7) obtained in Appendix C, we first evaluate the
high energy behaviour of the integral over ha,b the dominant contribution of which, as
well-known from the ordinary BFKL analysis, is obtained for na,b = 0. The integrals are
dominated by the usual BFKL saddle-points at ha,b = 1/2; (1/2) = 4 log 2. Taking into
account in the saddle-point derivation the logarithmic corrections due to the zeroes in the
prefactors at ha,b = 1/2, one gets:
8

1

2
4 e2( 2 )Y
(1)

fQ , Y

3
2 2
12 Y

g3P h, 12 , 12 2
h
h

dh EQ EQ ()
(10)
.
(h) 2 12
Formula (10) is then reported in the full amplitude (3).
Fig. 2. One-loop dipole amplitude in rapidity space, the initial dipoles 0 1 , 0 1 evolve in the rapidity range
y y , y y with one-Pomeron type of evolution and then with two-Pomeron type of evolution over the range
y , see Appendix C, the resulting
y,
y . Note that y + y Y , is the total rapidity range. After integration over y,
amplitude depends on the total range Y .
297
A few comments on our result (10) are in order:

(i) Conformal invariance is explicitly realized, due to the decomposition on the
h ( )E h (). This property being shared by the
combination of SL(2, C) Eigenvectors E
Q
Q
tree-level amplitude, see (A.1), is thus valid for the whole amplitude (3) including oneloop level. Hence, saturation which is expected from the compensation between tree level
and one-loop due to the different energy dependence is genuinely a conformal invariant
phenomenon. We conjecture that this will be true to all orders, since only tensorial SL(2, C)
properties play a rle in the symmetry properties of the amplitude. Thus the saturation
regime based on dipole loop contribution, if mathematically convergent, is expected to
respect the conformal symmetry.
(ii) The key difference between the tree-level and one-loop amplitude properties lies in
the difference in the analytic singularities of the integrands. While the tree level term in
(3) is dominated by a moving saddle-point at high energy, the one loop amplitude (10) has
double poles at
(h) = 2(1/2) = 8 log 2
which depend only on the anomalous dimensions h and thus are independent of the energy.
This very peculiar property means that the one-loop contribution may depend (and thus
can give information) on BFKL properties which are not reached in the conventional
analysis. In particular, all higher conformal spins may contribute, contrary to the tree level
contribution. Also, contributions for values of h not limited to the interval [0+ , 12 ] can give
rise to poles. This may be interpreted as unlimited contributions of higher twist terms of
the BFKL solution.
(iii) Apart the factor 4 , which is the natural strength scale of the one-loop amplitude,
there is a dynamical factor connected to the square of the function g3P (h, 12 , 12 ). This
function is the QCD analogue of the triple Pomeron coupling. However, it is not the value
g3P ( 12 , 12 , 12 ) which is here relevant, as was the case in the QCD dipole analysis of the
triple Pomeron coupling in [25], since h = 12 is not solution of the double pole equation.
One could also note that the square of g3P appears, contrary to saturation equations on an
extended target, where the dependence is linear [11].
3. Saturation critical line

As discussed in Section 1, saturation is expected to take place when the one-loop
amplitude, with coupling of order 4 , becomes comparable in strength with the tree
level amplitude of order 2 . Formula (3) is completely explicit and thus can serve for
analyzing saturation depending on transverse momentum or, equivalently using the Fourier
transforms (A.4), on impact parameter. For sake of simplicity we will concentrate on
the forward contribution at Q = 0, that is (through unitarity) for the total cross-section
integrated over impact parameter. In this limit, the SL(2, C) Eigenvectors simplify [23] to
give
h1/2 h1/2

h E h ()
.
E
(11)
Q
Q

298
Consequently, we may infer that the key order parameter for the discussion of saturation
will be the scale invariant ratio of incident dipole sizes / , in conformity with the
conformal invariance of the amplitude. This will in turn be determined by the ratio of
the characteristic scales of the incoming systems which are encoded in their corresponding
QCD dipole wave functions, see (2).
The general solution of (3) is easily obtained by a summation over the double pole
contributions verifying (hi ) = 2(1/2), whose location are characterized by an infinite
number of values hi = (ni , i ). One finds a formal double series

(1)
fQ=0 , Y
6 4 2 1 Y
2
e 2
1 3
2
2 Y

g3P h, 1 , 1 2 hi h i
log g3P h, 12 , 12
2
2

L+
, (12)
(h )

h
i
h=hi
n ,
i
where, for sake of simplicity, we used the notation L log |/ |.

For incident dipoles of comparable size, L O(1), we may note that the one loop result
then depends on an infinite sum over (ni , i ) which is due to the analytic properties of the
function (h). It even raises the question of the convergence of the double series (12).
On a physical ground, since the anomalous dimensions hi have no obvious limitation,
this situation can be interpreted as due to a dependence of the one loop amplitude on
all higher twist contributions, including those with non-zero conformal spins ni . This is
a quite strange example where higher twist calculations are required to all orders in a purely
perturbative framework, in striking analogy with actual non-perturbative situations.
Leaving this regime for further work, we will here focus on the case where only the
first pole with h = (i , ni = 0) contributes, i.e., where the equivalent of a leading twist
contribution is sufficient. This approximation is realized in a region where L 1 (or L 1
by symmetry).
The dominant double pole in (12) such that (h ) = 2( 12 ) = 8 log 2 is h (n = 0,
0.183). Consequently, factorizing the tree level coupling, the resulting one loop
amplitude can be written

2
1
(1)

L Y1 3 log Y + log L ,
exp 2Y + 2 h
fQ=0 L, Y
(13)
4
2
(1/2)Y , and
where we define Y
2 1 3
12
2

(h )

,
Y1 = log

16 2 g3P h , 12 , 12 12
(14)
which is a priori large, being in the perturbative domain 1 and constant up to 1/L
correction terms.
, L) where the tree-level
We define the saturation critical line as the line in the (Y
contribution equals the one loop one. It is a simple extension to the conformal invariant
configuration of the usual one [1]. Beyond this line, saturation is expected to take place.
299
We now make use of the well-known saddle-point approximation of the tree level
amplitude

2
(0)
exp (hc ) Y
+ 2(hc 1/2)L
fQ=0 L, Y
4
(1/2)

2 1
1
2

1/2 log Y , (15)
log 2

hc (1 hc )2
12
where the tree level saddle point hc is given by the well-known implicit equation given
in (7). Inserting the results (13) and (15) in (3) with a redefinition of the constant

2( 1 )
1
2

1 + log
c = Y
,
Y
(16)
h2c (1 hc )2
( 12 )
we get Eq. (6) announced in Section 1 for the saturation critical line.
4. Discussion, outlook
Let us describe general features of the critical line, when is very small but otherwise
not specified. Following expression (6) (see Fig. 3), it is straightforward to realize that there
/2L > (1/2)/ (h ) 0.09. Indeed
is an absolute bound when L, Y 1 given by Y
below this bound, the compensation can no more take place between tree level and one
loop contributions. It is interesting to note that the bound implies that a renormalization
group evolution, with L moving to large values while keeping Y fixed, cannot lead from
a weak density to a saturation region, as expected from the general theoretical picture of
the QCD dynamics of partons.
One can delimit three typical regimes in the saturation transition, depending upon the
behaviour of the tree level amplitude:
: in this region, saturation is obtained in the region
(I) 0 L0 L1 Y

0 + 0.63L,
> Y
0 + 2L 1 h Y
Y
2
0 is a large effective rapidity scale incorporating the logarithmic prefactors.
where Y
: it is well-known that the BFKL behaviour discussed in (I)
(II) L0 L1 L2 < Y
acquires an extra diffusion term (of the gluon transverse momenta along the BFKL ladder
contributions) in the intermediate region. Within the saddle-point approximation it leads to
the following saturation curve:

12 2L2
1

h 1
,
Y > Y0 + 2L
(17)

2
2 Y
where the last term is due to diffusion. Quite interestingly, saturation is favoured by
diffusion. Indeed, formula (17) approximately interpolates between the behaviour (I)
vs. L.
and (II) by a smooth decrease of the limiting slope of Y
300
, L. The saturation region

Fig. 3. Critical Saturation line. The critical saturation line is represented in the plane Y
is under the curve. The region beyond the validity of leading twist approximation is below a value 1 L0 . The
regions (I)(III) correspond to different regimes of compensation between tree and one-loop dipole contributions
(see text).
< L: near the bound 2L/Y

< ( 1 )/ (h ) 10. In this region, one can
(III) L2 Y
2
use the relations

1 2L
1
hc
;
(hc ) .

2 Y
hc
0 one obtains, within
Using this and again introducing the effective large rapidity scale Y

a good approximation at large (Y , L), the constraint

2
0
8L
Y

.
Y>
(18)
+
2
12
/2L > ( 1 )/ (h ).
It is easy to check that (18) satisfies the bound Y
2
We have not yet discussed the rle of the logarithmic prefactors. Using our explicit
expressions, it will be easy to incorporate them in a more complete study which we
leave for the future. They may play a phenomenological rle. Indeed, it is important
of (6) is due to the double zeroes appearing in the
to note that the large factor 52 log Y
loop couplings to the intermediate dipoles, see (C.7). This has the important effect of
a strong energy dependent damping on the effective triple Pomeron couplings, which could
otherwise be large [25]. Another possible important rle would be, by mere extrapolation to
higher dipole loops, to establish a useful hierarchy between these different orders. Indeed
the coupling strength expected at order n, namely, ( 2 exp ( 12 )Y )n cannot provide
301
a hierarchical structure in the saturation regime. We are thus conjecturing a series of critical
saturation lines corresponding to the onset of compensation between a growing number of
higher loop order contributions. The study of this regime, necessary to understand the full
saturation behaviour in the conformal invariant set-up will be pursued in the near future.
In fact, our work suggests a strategy to attack the full problem of saturation from dipole
loops. It seems possible to analyze the n-loop dipole diagrams in the same way as we have
shown here for the 2-loop case. It is clear that many diagrams can contribute, depending
on the multiple rapidities where new dipoles can branch. However, we may conjecture
that the integration over these intermediate rapidity intervals would dominantly lead to the
QCD multi-Pomeron exchange and the corresponding 1 n dipole vertices which were
analyzed in Ref. [28]. Then, it will be possible to explore the conformal invariant saturation
regime it self and not only the transition to it.
Another point merits some care, when the onium typical sizes are of the same order.
A quite interesting regime seems to set in when the sum (12) over all the double poles in
the one-loop integrals becomes relevant. While being still at small coupling constant, this
regime shares with the non-perturbative one the features of a summation over higher twist
contributions. It would be nice to solve this case in detail.
Finally, the step between our calculations and actual experimental possibilities, which
is not done in the present work, could be done either through factorization properties or
by the consideration of hard subprocesses with very high energies like MuellerNavelet
jets at LHC or collisions at a future linear accelerator. Relevant studies are certainly
deserved in the future.
Acknowledgements
We acknowledge fruitful discussions with Yuri Kovchegov, Heribert Weigert and
Edmond Iancu.
Appendix A. SL(2, C)-invariant formalism

Using the SL(2, C)-invariant formalism, the solution of the BFKL equation obtained at
leading log level of perturbative QCD [20] reads
2
(0)
fQ , Y =
4

h E h () d(h) e(h)Y ,
dh E
Q
Q
(A.1)
where the SU(3) coupling is fixed, small but unspecified, in the leading log approximation we consider.
In Eq. (A.1), the symbolic notation

dh

n=
302
corresponds to the integration over the quantum numbers associated to the continuous
unitary irreducible representations of SL(2, C), namely,
h = i +
1n
,
2
1+n
h = i +
,
2
n N,
R.
(A.2)
h
EQ
() and (h) are, respectively, the SL(2, C) Eigenfunctions and Eigenvalues of the
BFKL kernel [23]. The Eigenvalues read
(h) =

Nc
Nc
(h)
2 (1) (h) 1 h ,
where (log ) . The SL(2, C) Eigenvectors are defined1 by

2 2
h
2 iQb h
EQ () =
d r e E r ,r +
,
||b(h)
2
2
with

||b(h)
h
()
d 2 r eiQr EQ
Eh r , r +
2
2
8 4

h
h
= ()hh 2
,
r 2 /4
r 2 2 /4
(A.3)
(A.4)
(A.5)
r is the 2d impact-parameter, and introducing useful notations

where h = 1 h,
1
,
( 2 + (n 1)2 /4)( 2 + (n + 1)2 /4)

1+|n|
3 4h+h1 i + 2
(i + |n|/2)
b(h) =
,

1+|n|
1/2 h i +
(i + |n|/2)
d(h) =
|b(h)|2
4 /2
a(h) =
= 2
.
2
2
+ n2 /4
(A.6)
We will also make use of the formalism in the full coordinate space, the 4-dipole
amplitude T (0) being defined at tree level of QCD dipole contributions by

1
(0)
(0)

T
(A.7)
0 , 1 ; 0 , 1 Y
d 2 Q eirQ fQ , Y ,
2
where the impact parameter is r = 12 (0 + 1 0 1 ).
Inserting (A.1) into (A.7), one finds

2

n2
T (0) 0 , 1 ; 0 , 1 Y =
dh Gh 0 , 1 ; 0 , 1 2 +
d(h) e(h)Y ,
4
4
(A.8)
1 Note that an analytic expression of the Eigenvectors E h () in the mixed representation has been
Q
provided [21] in terms of a combination of products of two Bessel functions. For simplicity, we did not include
the impact factors [23,27]. Phenomenologically, the leading contribution to the amplitude (A.1) comes from the
n = 0 component which corresponds to the BFKL Pomeron.
303
where, using the SL(2, C) Eigenvectors in coordinate space (A.5), the Green function Gh
in coordinate space [21,23] can be written as

h

h (0 , 1 )E h (0 , 1 )
G 0 , 1 ; 0 , 1 d 2 E

|b(h)|2
h E h ().
=
(A.9)
d 2 Q eibQ E
Q
Q
4
We will also define for convenience the instantaneous (i.e., Y = 0) amplitude

T 0 , 1 ; 0 , 1 T (0) 0 , 1 ; 0 , 1 Y = 0 .
(A.10)
Due to completeness and frame independence of the formalism, many formulae can be
simplified using these instantaneous amplitudes between dipoles.
The single multiplicity distribution of dipoles n1 after an evolution characterized by the
rapidity Y is simply related [21] to the previous expressions (e.g., (A.9)), namely,

1
n1 0 , 1 ; 0 , 1 Y = 2 2 dh Gh 0 , 1 ; 0 , 1 2 + n2 /4 e(h)Y . (A.11)
| |
Let us finally note a useful completeness formula [21] valid at tree level:

T (0) 0 1 ; 0 1 Y = y + ya + y
2
2
2
1
d a0 d 2 a1 d a0 d a1
2
2 (2)4
|a |2
|a |2
(0)
n1 (0 1 ; a0 a1 |y)n 1 0 1 ; a0 a1 y T (a0 a1 , a0 a1 |ya ),
(A.12)
where 0 , 1 are the transverse coordinates of one initial dipole (respectively 0 1 for the
second one), a0 , a1 are those of the interacting dipole emerging from the dipole 0 ,
1 after evolution in rapidity y (respectively, a , a and y for the second one), which
0
1
interaction is described by the amplitude T (a0 , a1 ; a0 , a1 |ya ) as in (A.8).
Eq. (A.12) has the physical interpretation that, at tree dipole level, the multiplication
of dipoles due to rapidity evolution does not lead to saturation, but to the same tree level
amplitude in a kind of bootstrap property.
Appendix B. The two-dipole density distribution n2

The two-dipole density distribution n2 results from the solution of a specific evolution
equation which consists of an extension of the mixed evolution of the one formulated in
Refs. [4,24,31], namely:
y)
n2 (0 1 ; a0 a1 , b0 b1 |y y,

Nc
dh dha dhb
e1 (yy)
=
d
1
1 (h)
2 2
|a b |2

y
e
h,h ,h
n a b (0 1 ; a0 a1 , b0 b1 ),
d
(ha ) + (hb ) 2
(B.1)
304
where, by convention, di di /2 and

h,ha ,hb
n2
(0 1 ; a0 a1 , b0 b1 )

1
h,ha ,hb
=
d 2 d 2 d 2 R
,,
a(h)a(ha )a(hb )
E ha (a0 , a1 )E hb (b0 , b1 )E h (0 , 1 ),
(B.2)
with
h,h ,h
a b
R,,
d 2 r0 d 2 r1 d 2 r2 h
E (r0 , r1 )E ha (r0 , r2 )E hb (r1 , r2 ),
|r01 r02 r12 |2
(B.3)
where = 0 1 , a = a0 a1 , b = b0 b1 .
Conformal invariance implies the well-known tensorial reduction [29]

a b
R,,
[ ]hha hb [ ]ha hb h [ ]hb ha h [ ]hha hb [ ]ha hb h
h,h ,h
[ ]hb ha h g3P (h, ha , hb ),
(B.4)
where g3P (h, ha , hb ) happens to be [25] the QCD triple Pomeron coupling as obtained in
the dipole formulation, namely:

ha
hb
d 2 r0 d 2 r1 d 2 r2
r12
h r02
g3P (h, ha , hb ) =
[r01 ]
[r01 ]h
2
|r01 r02 r12 |
r0 r2
(1 r1 )(1 r2 )
h b
h a

r12
r02
.
(B.5)
r0 r2
(1 r1 )(1 r2 )
Considering the Fourier transforms (A.4) for the SL(2, C) Eigenvectors Eqh in the mixed
representation, one writes
h,ha ,hb
n2
(0 1 ; a0 a1 , b0 b1 )
= g3P (h, ha , hb )
b(h)b(ha )b(hb )
|a b |
27 10
d 2 qa d 2 qb d 2 Q (2)(Q + qa + qb )

v
w
h
()Eqhaa (a )Eqhbb (b )ei(qa ra +qb rb +Qr) d 2 v d 2 w ei((qa qb ) 2 +Q 2 )
EQ

1+hha +hb
1+h+ha hb
[v]1hha hb (w v)/2
(w + v)/2
{a.h.},
(B.6)
where = v, + = w, 2ra = a0 + a1 , 2rb = b0 + 1 and r is the overall
impact parameter. The notation {a.h.} indicates the anti-holomorphic part of the bracketed
term in the integrand for which the integration variables are complex conjugates and the
exponents hi are replaced by h i . Equivalently, the distribution n2 for the lower vertex is
given by the same equation (B.6) by using prime indices.
305
Appendix C. The general one-loop amplitude

Implementing equations (A.8) and (B.6) in the master equation (3) and first integrating
over impact parameters, one gets:

(2)8 qa qa qb qb qa qa qb qb .
Then, integrating over the intermediate dipole sizes, and using the orthogonality [23]
relation
2

1
d h
qh () = (2) h, h + (2) h, 1 h ei(h) ,
(C.1)
Eq ()E
2
2
4
||
where ei(h) Eqh ()/Eq1h () is just a , q-independent phase factor given in Ref. [23],
one finds

(2)8 24 (2) ha , ha (2) hb , hb (2) ha , 1 ha (2) hb , 1 hb ,
where (2) (h, h ) nn ( ). The integration over qa qb finally gives (2)2 (v v ).
Plugging in formulae (8), (9) the result obtained in formula (B.6) and integrating over
-functions, one gets:

(1)
fQ , Y = y + y

4 2 Nc 2
d(ha ) d(hb )
dha dhb
=
8
2 2
|b(ha )|2 |b(hb )|2

g3P (h, ha , hb ) g3P (h , ha , hb ) h
h HQ h, h
EQ ()E
dh dh
Q

b(1 h)
b(h )
Yy
y
d y
d y
0
0

d d1 d d1

1 (yy)
ey+
((ha ) + (hb ) )(1 (h))
e y +1 (y y )
,
((ha ) + (hb ) )(1 (h ))
(C.2)
where, after the change of variables w = v(1 2t), w = v(1 2t ),

HQ h, h

= d 2 v d 2 t d 2 t eiQv(t t ) v hh 1 t 1+h+hb ha (1 t)1+hhb +ha t ha hb h
h +h h

{a.h.}
1 t a b
(C.3)
After integration over v one obtains [30]

hh h

h

h h

2
2

i 2 (nn )

HQ h, h =
d 2t d 2t t t
e
hh

Q
Q
h +h h
1+h+h ha

b
t
(1 t)1+hhb +ha t ha hb h 1 t a b
{a.h.}
(C.4)
306
where, by definition, (z) (z)/ (1 z ).

The remaining integral can be proven2 to reduce to
hh h
h

2
2
2
HQ h, h =

Q
Q
(2h 1)(2h 1)

n+na +nb
(h + ha hb ) 1 h ha + hb
h h + ()

1 h h .
(C.5)
The integral over h can now easily be performed. The remaining poles at h = h and
h = 1 h give twice the same contribution due to the over completeness relation of the
h generators [23]. One finally obtains:
EQ

(1)
fQ , Y

n+na +nb
g3P (h, ha , hb ) 2
4 4 Nc 2
d(ha ) d(hb ) (1)

=
dh
dh
dh
a
b

2
1
8
b(ha )b(hb )b(h)
2
| 2 h|2
h

h

E ()E
Q
d y
d y
0
Yy
y
0

d d1 d d1

1 (yy)
ey+
((ha ) + (hb ) )(1 (h))
e y +1 (y y )
.
((ha ) + (hb ) )(1 (h))
(C.6)
This one-loop amplitude preserves the global conformal invariance of the tree-level
BFKL 4-gluon amplitude, since the only scale-dependence on Q is present in the
h ()E
h ( ).
conformal Eigenvectors | |EQ
Q
It was noted in Ref. [22] that the integration over rapidity variables yields two different
contributions depending on the sign of the quantity (ha ) + (hb ) (h). Indeed for
(ha ) + (hb ) < (h), the relevant poles are situated at = 1 = = 1 = (h),
leading to a contribution associated with the single Pomeron dependence e(h)Y . In the
opposite case, namely (ha ) + (hb ) > (h), the relevant poles are situated at = 1 =
= 1 = (ha ) + (hb ). The resulting amplitude

fQ(1) , Y

h
1
4
h
Q
8
dh E
EQ ()
2
2 2
2

( 12 ha )( 12 hb )
dha dhb
ha (1 ha )hb (1 hb )
2

g3P (h, ha , hb )
,
e((ha )+(hb ))Y
(C.7)
(h) (h ) (h )
a
2 One shows using general methods [30] that H (h, h ) is a combination of functions whose coefficients
Q
can then be computed directly on formula (C.4).
307
is the one we are interested in for the saturation problem since it possesses the QCD
double Pomeron energy behaviour e((ha )+(hb ))Y > e(h)Y . This is precisely the energy
behaviour which will compensate for the higher order 2 in the coupling constant. Notice
that the expression depends only on the sum Y = y + y , as it should do from longitudinal
boost invariance.
References
[1] L.V. Gribov, E.M. Levin, M.G. Ryskin, Phys. Rep. 100 (1983) 1.
[2] A.H. Mueller, J. Qiu, Nucl. Phys. B 268 (1986) 427.
[3] N.N. Nikolaev, B.G. Zakharov, Z. Phys. C 49 (1991) 607;
N.N. Nikolaev, B.G. Zakharov, Z. Phys. C 53 (1992) 331;
N.N. Nikolaev, B.G. Zakharov, Z. Phys. C 64 (1994) 651;
N.N. Nikolaev, B.G. Zakharov, J. Exp. Theor. Phys. 78 (1994) 598.
[4] A.H. Mueller, Nucl. Phys. B 415 (1994) 373;
A.H. Mueller, B. Patel, Nucl. Phys. B 425 (1994) 471;
A.H. Mueller, Nucl. Phys. B 437 (1995) 107.
[5] A.H. Mueller, Nucl. Phys. B 335 (1990) 115;
Yu.A. Kovchegov, A.H. Mueller, S. Wallon, Nucl. Phys. B 507 (1997) 367;
A.H. Mueller, Eur. Phys. J. A 1 (1998) 19;
A.H. Mueller, Nucl. Phys. A 654 (1999) 370;
A.H. Mueller, Nucl. Phys. B 558 (1999) 285.
[6] J.C. Collins, J. Kwiecinski, Nucl. Phys. B 335 (1990) 89;
J. Bartels, G.A. Schuler, J. Blmlein, Z. Phys. C 50 (1991) 91;
J. Bartels, G.A. Schuler, J. Blmlein, Nucl. Phys. Proc. Suppl. 18 (1991) 147.
[7] J. Bartels, E.M. Levin, Nucl. Phys. B 387 (1992) 617;
J. Bartels, Phys. Lett. B 298 (1993) 204;
J. Bartels, Z. Phys. C 60 (1993) 471;
J. Bartels, Z. Phys. C 62 (1994) 425;
J. Bartels, M. Wsthoff, Z. Phys. C 66 (1995) 157;
J. Bartels, C. Ewertz, JHEP 9909 (1999) 026.
[8] L. McLerran, R. Venugopalan, Phys. Rev. D 49 (1994) 2233;
L. McLerran, R. Venugopalan, Phys. Rev. D 49 (1994) 3352;
L. McLerran, R. Venugopalan, Phys. Rev. D 50 (1994) 2225;
A. Kovner, L. McLerran, H. Weigert, Phys. Rev. D 52 (1995) 6231;
A. Kovner, L. McLerran, H. Weigert, Phys. Rev. D 52 (1995) 3809;
R. Venugopalan, Acta Phys. Pol. B 30 (1999) 3731;
E. Iancu, L. McLerran, Phys. Lett. B 510 (2001) 145;
L. McLerran, The color glass condensate and small x physics: 4 lectures, hep-ph/0104285;
E. Iancu, A. Leonidov, L. McLerran, Nucl. Phys. A 692 (2001) 583;
E. Ferreiro, E. Iancu, A. Leonidov, L. McLerran, Nucl. Phys. A 703 (2002) 489;
A. Capella et al., Phys. Rev. D 63 (2001) 054010.
[9] G.P. Salam, Nucl. Phys. B 449 (1995) 589;
G.P. Salam, Nucl. Phys. B 461 (1996) 512;
G.P. Salam, Comput. Phys. Commun. 105 (1997) 62;
G.P. Salam, A.H. Mueller, Nucl. Phys. B 475 (1996) 293.
[10] E. Gotsman, E.M. Levin, U. Maor, Nucl. Phys. B 464 (1996) 251;
E. Gotsman, E.M. Levin, U. Maor, Nucl. Phys. B 493 (1997) 354;
E. Gotsman, E.M. Levin, U. Maor, Phys. Lett. B 245 (1996) 369;
E. Gotsman, E.M. Levin, U. Maor, Eur. Phys. J. C 5 (1998) 303;
E. Gotsman, E.M. Levin, U. Maor, E. Naftali, Nucl. Phys. B 539 (1999) 535;
308
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
A.L. Ayala Filho, M.B. Gay Ducati, E.M. Levin, Nucl. Phys. B 493 (1997) 305;
A.L. Ayala Filho, M.B. Gay Ducati, E.M. Levin, Nucl. Phys. B 551 (1998) 355;
A.L. Ayala Filho, M.B. Gay Ducati, E.M. Levin, Eur. Phys. J. C 8 (1999) 115.
I. Balitsky, Nucl. Phys. B 463 (1996) 99;
High-Energy QCD and Wilson Lines, To be published in the Boris Ioffe Festschrift, in: M. Shifman (Ed.),
At the Frontier of Particle Physics/Handbook of QCD, World Scientific, Singapore, 2001, hep-ph/0101042.
J. Jalilian-Marian, A. Kovner, L. McLerran, H. Weigert, Phys. Rev. D 55 (1997) 5414;
J. Jalilian-Marian, A. Kovner, H. Weigert, Phys. Rev. D 59 (1999) 014014;
J. Jalilian-Marian, A. Kovner, H. Weigert, Phys. Rev. D 59 (1999) 099903, Erratum;
A. Kovner, J. Guilherme Milhano, H. Weigert, Phys. Rev. D 62 (2000) 114005.
H. Weigert, Unitarity at small Bjorken x, hep-ph/0004044.
M.A. Braun, Eur. Phys. J. C 16 (2000) 337, hep-ph/0101070.
Yu.A. Kovchegov, L. McLerran, Phys. Rev. D 60 (1999) 054025;
Yu.A. Kovchegov, L. McLerran, Phys. Rev. D 62 (2000) 019901, Erratum;
Yu.A. Kovchegov, E.M. Levin, Phys. Rev. D 60 (1999) 034008;
Yu.A. Kovchegov, Phys. Rev. D 59 (1999) 014017;
Yu.A. Kovchegov, Phys. Rev. D 61 (2000) 074018.
E.M. Levin, K. Tuchin, Nucl. Phys. B 537 (2000) 99;
E.M. Levin, K. Tuchin, Nucl. Phys. A 691 (2001) 779;
E.M. Levin, K. Tuchin, Nucl. Phys. A 693 (2001) 787.
K. Golec-Biernat, M. Wsthoff, Phys. Rev. D 59 (1998) 014017.
N. Timneanu, J. Kwiecinski, L. Motyka, Eur. Phys. J. C 23 (2002) 513.
S. Munier, A.M. Stasto, A.H. Mueller, Nucl. Phys. B 603 (2001) 427.
L.N. Lipatov, Sov. J. Nucl. Phys. 23 (1976) 642;
V.S. Fadin, E.A. Kuraev, L.N. Lipatov, Phys. Lett. B 60 (1975) 50;
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP 44 (1976) 45;
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP 45 (1977) 199;
I.I. Balitsky, L.N. Lipatov, Sov. J. Nucl. Phys. 28 (1978) 822.
H. Navelet, R. Peschanski, Nucl. Phys. B 507 (1997) 353;
H. Navelet, S. Wallon, Nucl. Phys. B 522 (1998) 237.
H. Navelet, R. Peschanski, Phys. Rev. Lett. 82 (1999) 1370.
L.N. Lipatov, Zh. Eksp. Teor. Fiz. 90 (1986) 1536, [Eng. trans. Sov. Phys. JETP 63 (1986) 904];
L.N. Lipatov, Phys. Rep. 286 (1997) 131.
R. Peschanski, Phys. Lett. B 409 (1997) 491.
A. Bialas, H. Navelet, R. Peschanski, Phys. Rev. D 57 (1998) 6585;
G.P. Korchemsky, Nucl. Phys. B 550 (1999) 397.
V.A. Abramovsky, V.N. Gribov, O.V. Kancheli, Sov. J. Nucl. Phys. 18 (1974) 308;
J. Bartels, M.G. Ryskin, Z. Phys. C 76 (1997) 241.
H. Navelet, R. Peschanski, Nucl. Phys. B 515 (1998) 269.
R.A. Janik, R. Peschanski, Nucl. Phys. B 549 (1999) 280.
A. Polyakov, Pisma Zh. Eksp. Teor. Fiz. 12 (1970) 538.
J.S. Geronimo, H. Navelet, math-ph/0003019, to appear in J. Math. Phys.
A. Bialas, R. Peschanski, Phys. Lett. B 355 (1995) 301.

Numerical evaluation of the general massive

2-loop sunrise self-mass master integrals
from differential equations
M. Caffo a,b , H. Czyz c,1 , E. Remiddi a,b
a INFN, Sezione di Bologna, I-40126 Bologna, Italy
b Dipartimento di Fisica, Universit di Bologna, I-40126 Bologna, Italy
c Institute of Physics, University of Silesia, PL-40007 Katowice, Poland
Abstract
The system of 4 differential equations in the external invariant satisfied by the 4 master integrals
of the general massive 2-loop sunrise self-mass diagram is solved by the RungeKutta method in
the complex plane. The method, whose features are discussed in details, offers a reliable and robust
approach to the direct and precise numerical evaluation of Feynman graph integrals. 2002 Elsevier
Science B.V. All rights reserved.
PACS: 11.10.-z; 11.10.Kk; 11.15.Bt; 12.20.Ds; 12.38.Bx
1. Introduction
High precision measurements in high energy physics (or more in general in the
determination of particle properties) require more and more precise calculations of multiloop Feynman diagrams to have sufficiently precise theoretical predictions to compare
with.
The nowadays widely accepted procedure of expressing radiative correction calculations in terms of a limited number of master integrals (MI) [1] reduces the problem to the
careful determination of these quantities. The method has also the advantage that, with
E-mail addresses: caffo@bo.infn.it (M. Caffo), czyz@us.edu.pl (H. Czyz), remiddi@bo.infn.it
(E. Remiddi).
1 Work supported in part by the European Communitys Human Potential Programme under contract HPRNCT-2000-00149 Physics at Colliders.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 1 5 - 2
310
M. Caffo et al. / Nuclear Physics B 634 (2002) 309325
Fig. 1. The general massive 2-loop sunrise self-mass diagram.
a correct bookkeeping of the recurrence relations arising from integration by parts identities, the MI of a given problem can be reused in more complicated calculations.
The analytical calculation of MI, in terms of the usual polylogarithms and their
generalizations, is in general possible only when the number of different scales (internal
masses and external momenta or Mandelstam variables) is small, like in QCD calculations,
where all masses are set to zero or in the QED cases, where only the electron mass is
different from zero, or when the external variables are fixed to particular values (zero or
mass shell condition). Another possibility of big help in analytic calculations is sometimes
offered by the exploitation of particular simplifying conditions, like the smallness of some
ratios of the parameters allowing the corresponding expansion.
In the general massive case, relevant in the electroweak theory, the number of
parameters prevents from obtaining results in the usual analytic form already in the case of
the 2-loop sunrise self-mass diagram shown in Fig. 1.
This diagram has indeed a long history of investigation and its MI were even recognized
to be expressible in closed form as a combination of four Lauricella functions, a special
class of generalized hypergeometric series [2] (and earlier references therein). The method
provides efficient multiple series expansions for the regions of small |p2 |, i.e., |p2 | <
max(m2i ), and of large |p2 |, i.e., |p2 | > (m1 + m2 + m3 )2 , but some problems arise in
the intermediate region.
Great efforts were therefore devoted to investigate the properties in the special points
(i.e., p2 = 0, , pseudothresholds and threshold). The analytical expansions of the MI
at 0 and are given in [2] and [3]; the values at the pseudothresholds and threshold in [4];
the analytical expansions at the pseudothresholds are in [5]; a semi-analytical expansion
at the threshold is in [6] and also in the configuration space technique in [7], while the
complete analytical expansions at the threshold are presented in [8].
For numerical evaluation purposes, it is possible to cast the general massive self-mass
diagram as a double integral representation and in the particular case of the sunrise diagram
in a single integral representation [2,9] (and earlier references therein). The configuration
space technique is also exploited in the numerical approach [2,10]. In a recent approach
rearrangements of the integrand, driven by the BernsteinTkachov theorem, are introduced
to improve numerical convergence [11]. A different and interesting method is the use of
the recurrence relations as difference equations to numerically evaluate the MI [12].
In the present paper we exploit the numerical evaluation of the four MI related to the
general massive 2-loop sunrise self-mass diagram [13], using the differential equations
in p2 , obtained in [3] and the RungeKutta method [14] to solve them for complex values
of p2 . The interest is not limited to provide with a fast routine precise numerical values
for all the four MI in the general massive case and for all the values of p2 , but extends
311
to the investigation of the reliability of the method, as it can be easily extended to the
numerical evaluation of other less studied diagrams.
In Section 2 the master differential equations are recalled and the analytic properties
of the MI are reviewed. Section 3 contains a description of the method used to solve the
system of differential equations and to determine the accuracy. In Section 4 the control
tests and the comparisons with other values reported in the literature are presented. Finally,
in Section 5 our conclusions on the application of the method to present and further work
are presented.
2. Analytical properties and behaviours of the MI

We use here the following definition of the four MI related to the general massive 2-loop
sunrise self-mass diagram in n continuous dimensions and with fully Euclidean variables

82n
Fj n, m21 , m22 , m23 , p2 =
((2)n2 )2

1
d n k1 d n k2 2
,
2
2
2
(j
)
(j
1
2
(k1 + m1 )
(k2 + m2 ) ) ((p k1 k2 )2 + m23 )3 (j )
j = 0, 1, 2, 3
(1)
and i = 1, 2, 3; i (0) = 1, for j = 0; i (j ) = 1, for j = i; i (j ) = 2, for j = i.
Wherever necessary to avoid ambiguities, the usual imaginary displacements m2i
2
mi i, where is an infinitesimal positive number, are understood.
At variance from [3], were the mass scale was given the value = 1, here we choose
= m1 + m2 + m3 ,
(2)
which comes out to be the appropriate mass scale parameter for the numerical discussion.
The expansion of the MI around n = 4 has the form [3]

Fj n, m21 , m22 , m23 , p2

1
1
(2) 2
(1) 2
= C 2 (n)
Fj
m1 , m22 , m23 , p2 +
m1 , m22 , m23 , p2
Fj
2
(n 4)
(n 4)

+ Fj(0) m21 , m22 , m23 , p2 + O(n 4) ,
(3)
where the coefficient C(n)

(4n)
n
C(n) = 2
,
3
2
(4)
not expanded, can be replaced by its value C(4) = 1, at n = 4, when multiplying a function
regular in (n 4). The coefficients of the poles in (n 4) of F0 (n, m21 , m22 , m23 , p2 ) are
known to be [3]

1
F0(2) m21 , m22 , m23 , p2 = m21 + m22 + m23 ,
8
312

1 p2 3 2

+ m1 + m22 + m23
F0(1) m21 , m22 , m23 , p2 =
8 4
2

2
2
2
m3
m1
m2
2
2
2
m1 log
+ m2 log
+ m3 log
,
2
2
2
(5)
while those for
Fi (n, m21 , m22 , m23 , p2 ),
i = 1, 2, 3, are

1
Fi(2) m21 , m22 , m23 , p2 = ,
8
2

mi
1
1
(1) 2
2
2
2
Fi
m1 , m2 , m3 , p = + log
.
16 8
2
(6)
From now on we deal only with the finite parts of the MI (i.e., we subtract the
n 4 poles) and we do not write anymore, for short, the arguments of the functions
Fj(0) (m21 , m22 , m23 , p2 ) Fj(0) , unless we need to refer explicitly to them.
The differential equations satisfied by the finite part of the MI expansions, given in [3],
can be written as
p2
(0)
F = F0(0) + m21 F1(0) + m22 F2(0) + m23 F3(0) + T0 ,
p2 0
3

(0)
p2 D m21 , m22 , m23 , p2

F
=
Mi,j Fj(0) + Ti ,
p2 i
i = 1, 2, 3,
(7)
j =0
where the explicit form of the functions T0 , Ti (polynomials of p2 and m2i , and logarithms
of m2i /2 ) and Mi,j (polynomials of p2 and m2i ) can be found in [3]. The function
D(m21 , m22 , m23 , p2 ) is defined by

2
2
2

2
2
2
2
D m21 , m22 , m23 , p2 = p2 pth
p pps1
p pps2
p pps3
,
2
pth
= (m1 + m2 + m3 )2 ,
2
pps1
= (m1 + m2 m3 )2 ,
2
pps2
= (m1 m2 + m3 )2 ,
2
= (m1 m2 m3 )2 ,
pps3
(8)
2 and at the three pseudothresholds p2 , p2 , p2 . Indeed,

and vanishes at the threshold pth
ps1 ps2 ps3
the values of p2 at which the coefficients of the derivatives in Eq. (7) vanish, together with
p2 = , are the singular points of the differential equations, which we will call special
points because special care is required in the numerical computation of the MI.
The differential equations (7) allow a class of solutions wider than just the functions
Eq. (1), but the initial conditions at p2 = 0 imposed by Eq. (1) identify uniquely the
solutions (actually the regularity at p2 = 0, even without the explicit knowledge of the
functions in that point, is enough to fix the solutions). Once the initial condition at p2 = 0 is
2 , p2 = p2 , p2 = p2 are also
fixed, one finds that the pseudo-threshold values p2 = pps1
ps2
ps3
313
2 is a branch point [8] (in agreement,

regular points [5], while the threshold value p2 = pth
of course, with standard textbook results [15]).
It is convenient to use reduced masses and reduced external invariant
mi,r
mi
,
m1 + m2 + m3
pr2
p2
,
(m1 + m2 + m3 )2
(9)
(0)
together with a dimensionless version of F0 , defined by

(0)
F0
;
(10)
(m1 + m2 + m3 )2
as the other master integrals are already dimensionless, the values of all the functions are
2 =
now pure numbers. In terms of the new variables pr2 , mi,r the threshold is located at pth,r
2
2
pth (mi,r ) = 1 and the pseudo-thresholds are in pps1,r
= pps1 (mi,r ), pps2,r
= pps2 (mi,r ),
2
pps3,r = pps3 (mi,r ). We also recall here that according to the Euclidean definition our p2
is positive for space-like p, and negative for time-like p.
Typical plots of the real and imaginary parts of the MI for two sets of masses are shown
in Figs. 25. The plots are obtained by means of the FORTRAN program described in
(0)
F0,r
(0)
(0)
Fig. 2. Plots of Re F0,r (labeled as (0)) and Re Fi

m3 = 4 and = m1 + m2 + m3 .
(labeled as (i)) as a function of pr2 for m1 = 2, m2 = 1,
314
(0)
(0)
Fig. 3. Plots of Re F0,r (labeled as (0)) and Re Fi

m3 = 200 and = m1 + m2 + m3 .
the next two sections and each consists of 6000 points per function calculated with the
relative precision of 106 .
The behavior of the functions depends of course on the values of the masses, however
some of their properties are quite general, as it appears from the following discussion. The
(0)
real part of F0,r as seen in Figs. 2 and 3 has one local minimum and one local maximum
2
for finite pr and goes to for pr2 . We do not have analytically the exact position
of the local extrema, but as they appear to lie outside the region 1 pr2 1 they can be
(0)
found approximately with the help of the asymptotic expansion (large pr2 ) for F0,r , which
is [3]
(0)
F0,r =

3

13
pr2
1
log2 pr2
m2i,r
log pr2
+
32
4
32
i=1
3

1
log pr2
m2i,r log m2i,r + .
16
i=1
(11)
(0)
(0)
Fig. 4. Plots of Im F0,r (labeled as (0)) and Im Fi

m3 = 4 and = m1 + m2 + m3 .
315
(0)
The asymptotic behavior of Re F0,r is obvious from Eq. (11), while for the positions
of the maximum and minimum one finds approximately from the first term only, which
2
2
is also the leading, pr,max
= 9.5 and pr,min
= 9.5, independent from the mass values.
The approximate values of the function at those points, again from the first term only,
(0) 2
(0) 2
are Re F0,r (pr,max
) = 0.3 and Re F0,r (pr,min
) = 0.3, also independent from the mass
values. Taking into account the asymptotic behavior one expects at least three zeros of the
function, provided that in nonasymptotic region there are no additional extrema (which is
(0)
actually the case). The second derivative of Re F0,r at threshold is infinite [8], but it does
(0)
not change sign at that point, even if the position of the flex point of Re F0,r
is not far from
the threshold.
The other MI go one into the other by the exchange of the values of the related masses.
From the expression of their expansion for large pr2
Fi(0) =

1
1
log2 pr2 +
log pr2 log m2i,r + 1 + ,
32
16
(12)
we see that their real parts all go to in both asymptotic regions pr2 , however
the position of the maximum cannot be obtained just from the first terms of the asymptotic
316
(0)
(0)
Fig. 5. Plots of Im F0,r (labeled as (0)) and Im Fi

m3 = 200 and = m1 + m2 + m3 .
expansion, as it is positioned in the region of small pr2 . The analytic expansions at threshold
(0)
[8] show that the derivatives of the Re Fi are infinite, but they do not change sign exactly
at that point.
The imaginary parts of all the functions plotted in Figs. 4 and 5 exhibit no complicated
structure and their asymptotic behaviors can be simply deduced from Eq. (11) and Eq. (12).
Observe that, due to our definition of pr2 , the proper analytic continuation for time-like p,
hence negative pr2 = |pr2 |, is obtained by giving a positive infinitesimal imaginary part to
pr2 , so that log(pr2 ) log(pr2 i) = log |pr2 | i (at variance with the default option
of FORTRAN compilers).
3. The numerical method

For the numerical solution of the system of differential equations we use the fourthorder RungeKutta method [14]. The method starts from the known values of the solutions
in an initial point, then it calculates the values of the solutions in a nearby point at distance
with an expansion in based on the differential equations, omitting terms of order 5 .
317
Repeating the procedure along a path of length L in N steps, so that the step is of
length = L/N , a relative error of approximately N5 = L5 /N 4 is accumulated, and
the requested accuracy is obtained by a suitable choice of L and N . This method is known
for its robustness, and indeed it works quite well in our case allowing us to obtain a relative
accuracy of 1010 1012 (the FORTRAN program is written in double precision) within
reasonable CPU time (see discussion at end of this section). More sophisticated methods
exist and could be implemented, but the simplicity of the used one has the advantage of a
better control on the accuracy.
To obtain the four MI related to the general massive 2-loop sunrise self-mass diagram
we use the system of four linear differential equations given in Eq. (7). For the necessary
initial conditions we use the values of the MI in the special points, where the differential
equations simplify, allowing the analytic calculation of the MI; but as also the coefficients
of the derivatives vanish there, to start the numerical evaluation from that points the values
of the first derivatives must be provided as well. We use for that purpose the analytic values
presented in [3,5,8]. Starting from pr2 = 0, numerical instabilities arise when approaching
any of the pseudo-thresholds; therefore, to obtain the value of the MI in the proximity
2
2
2
2
= 1 or of the pseudo-thresholds pps1,r
, pps2,r
, pps3,r
, we take
of the threshold pth,r
the threshold or pseudo-thresholds themselves as the starting points of the numerical
evaluation. Further, for very large values of pr2 it is more convenient to start from the
value x = 1/pr2 = 0, so that the asymptotic expansion at large pr2 is also needed.
We use the RungeKutta method in the complex plain of pr2 ; the initial condition
guarantees as in the real case the uniqueness of the solution, provided that we do not
2 = 1) to
cross the cut [16], which extends in the present case from the threshold (pth,r
along the real axis. As already remarked, due to the use of Euclidean variables in
Eq. (1), the proper sign of the imaginary part of the solutions is obtained with a path laying
in the lower half complex plane of pr2 . Using pr2 as a complex variable has the advantage
that the initial conditions at pr2 = 0 can be used to obtain the solutions everywhere, with a
path which does not approach too much the pseudo-thresholds and threshold. This feature
is relevant, as the method reaches much faster the required accuracy, when starting from
pr2 = 0, rather than from anyone of the other special pointseven if the result is of course
independent of the chosen path.
The program is organized as an independent subroutine, whose arguments are the input
values of the masses mi , of p2 (which is real) and for the required accuracies, and the
output values of the MI with their errors. Actually, the input accuracies refer separately
to the real part, the imaginary part and the absolute value of the functions. However, the
accuracy of the imaginary part is controlled by the program only when not vanishing, i.e.,
for pr2 < 1, but its value is in any case an indication of the precision of the result. The
program starts usually from the initial conditions at pr2 = 0. However, if |pr2 | > 900, it
starts from initial conditions at x 1/pr2 = 0, while if |pr2 ps2 | < 0.001, where ps2 is the
threshold or any of the pseudo-thresholds, it starts from the initial conditions at ps2 .
When starting from pr2 = 0, the fourth-order RungeKutta method is applied to the
system of equations (7) and the required final value of pr2 is reached following the
rectangular path in the complex plane: (0, 0), (0, 0.1), (pr2 , 0.1), (pr2 , 0). When
pointing to time-like values pr2 < 0, this path has the merit of avoiding the numerically
troublesome points of the threshold and pseudo-thresholds. The same path is also used for
318
reaching space-like values pr2 > 0, although no special points occur along on the real axis,
as it turns out that the requested precision is usually reached much faster along a complex
path than along the real one.
The fourth-order RungeKutta for the system of equations (7) is also used when the
2
2
2
, pps2,r
, pps3,r
, along a similar
starting point of pr2 is one of the pseudo-thresholds pps1,r
2
2
2
2
rectangular path: (pps,r , 0), (pps,r , 0.1), (pr , 0.1), (pr , 0).
When starting from pr2 = 1 (the threshold) the same system of equations cannot be
used, as the first derivatives of the master integrals Fi , i = 1, 2, 3, are infinite at that point.
th , F th through the definitions
Instead, we introduce new functions F0,r
i
2
th
m1,r m2,r m3,r xth
log(xth ) + F0,r
32
m1,r m2,r m3,r

(0)
Fi =
xth log(xth ) + Fith , i = 1, 2, 3,
16
mi,r
(0)
F0,r =
(13)
where xth = pr2 + (m1,r + m2,r + m3,r )2 = pr2 + 1, we generate algebraically the system of
differential equations which they satisfy, and then we solve numerically that new system
within the program. We do not report here the new system of equations as one can easily
(0)
obtain it from the Eq. (7) using Eq. (13). The function F0,r does not have the same
problems of the Fi(0) , i = 1, 2, 3, but the subtraction is performed anyway to simplify
the equations. The required value of pr2 is then reached along the triangular path in the
complex plane (1, 0), ((pr2 1)/2, 0.01), (pr2 , 0).
In the asymptotic region |pr2 | > 900 we perform the change of variables pr2 x
1/pr2 , then we subtract from the MI the terms not vanishing at x = 0 (the original MI are
indeed divergent at x 0) and finally we write a system of equations for the subtracted
as and F as , i = 1, 2, 3 defined as
MI F0,r
i
(0)
F0,r

3
3
2 13

1
pr2
1
log pr
+
log2 pr2
log pr2
=
m2i,r
m2i,r log m2i,r
32
4
32
16
i=1
1
32
i=1
i=1

as
m2i,r 5 6 log m2i,r + log2 m2i,r + F0,r

1
1
log2 pr2 +
log pr2 log m2i,r + 1
Fi(0) =
32
16
2

1
+
1 4 log mi,r + log2 m2i,r + Fias .
(14)
32
as , F as can be obtained in a simple way
Again the new system of equations for F0,r
i
substituting Eq. (14) into Eq. (7). The numerical solution is then obtained in the variable x
along the complex triangular path (0, 0), (x/2, 0.01), (x, 0).
The errors assigned to the final results of the RungeKutta method are estimated by
comparing them to the results obtained with a number of steps 10 times smaller then for
the final results. The difference of the two results is taken as the estimate of the absolute
error. To account for
the cumulated rounding error, we estimate the relative error in a N
step calculation as N 1015 (as the program works in double precision), and then take
319
the cumulated rounding error as the relative error times the value of the result. We finally
take the sum of the absolute error and the cumulated rounding error as an indication of the
error in the result.
The initial number of steps Ni for |pr2 | < 1 is taken to be Ni = 2/min(accuracies)
(where min(accuracies) is the smallest of the accuracies required in calling the routine)
and Ni = 2|pr2 |/min(accuracies) for |pr2 | 1, but is set to Ni = 20 if the number comes
out smaller then 20. If the required precision is not reached the number of steps is increased
by a factor 4, the system is solved once more and the procedure for estimating the error is
repeated. For a high required accuracy it might happen that the estimated error grows when
the number of steps is increased (because of an accumulation of the rounding errors, etc.).
In that case the program gives out the best result (i.e., the one with the smallest error). It
may also happen, in the case the accuracy obtained in a given step is almost equal to the
required one, that in the next step the accuracy obtained is much higher then the required
one.
Typical running times on PC with Intel Pentium III (1 GHz) CPU are the following: for
a required accuracy of 107 a fraction of a second for |pr2 | 0.2, 2 seconds for |pr2 | 2
and 8 seconds for |pr2 | 30; for a required accuracy of 1011 20 seconds for |pr2 | 0.2,
3.5 minutes for |pr2 | 2 and 59 minutes for |pr2 | 30.
The program is available from authors upon request and we report in Tables 13 a few
results, which can serve as a benchmark. The reported results were all obtained asking the
accuracies to be 1011 . In Table 3 the values at pr2 = 0, at threshold (pr2 = 1) and at three
2
2
2
, pps2,r
, pps3,r
) are calculated from the known analytical
pseudo-thresholds (pr2 = pps1,r
results [35,8] incorporated into the program, therefore no error is indicated (the imaginary
parts vanish for those values of pr2 ).
Table 1
(0)
(0)
The benchmark values of F0,r and F1 for masses m1 = 1, m2 = 9, m3 = 200 and = m1 + m2 + m3
pr2
(0)
Re F0,r
1000
113.349786296(3)
30
0.230629539580(2)
15
0.044413679180(8)
1.5
0.2536003902785(5)
(0)
Im F0,r
96.922241476(2)
2.31604333072(1)
0.964565210432(6)
0.0032295545873(1)
(0)
Re F1
0.4811355157(1)
2.86686026427(1)
3.47247970401(2)
5.03180993338(1)
(0)
Im F1
3.255217395(1)
2.46075767262(1)
2.23096593908(2)
0.564979990606(8)
Table 2
(0)
(0)
The benchmark values of F2 and F3 for masses m1 = 1, m2 = 9, m3 = 200 and = m1 + m2 + m3
pr2
(0)
Re F2
1000
1.4689531571(1)
30
0.880149099634(4)
15
1.26943734837(1)
1.5
2.066516140875(4)
(0)
(0)
(0)
Im F2
Re F3
Im F3
2.393163661(1)
1.624323639603(8)
1.42094925579(1)
0.239828506599(7)
0.81238274708(3)
0.1097269441157(5)
0.195266696001(4)
0.2185459121886(5)
1.17931522036(6)
0.496397637240(4)
0.366097572324(2)
0.0195952353591(3)
320
Table 3
(0)
(0)
(0)
(0)
The benchmark values of F0,r , F1 , F2 and F3 for masses m1 = 1, m2 = 9, m3 = 200 and =
m1 + m2 + m3
pr2
(0)
F0,r
(0)
F1
1
0.279454902855371
4.83093725524177
0.99
0.2798928396415(5)
4.80520291678(1)
2
0.280281633048667
4.78848044412341
pps3,r
0.9
0.2836780811878(5)
4.68669935479(1)
2
pps2,r
0.286233415451605
4.62888926299134
0.825 0.2866587055221(5)
4.620058473554(8)
2
0.286906928933491
4.61498769104262
pps1,r
0.8
0.2876221116285(5)
4.60069487253(1)
0.1
0.3101507246241(3)
4.261426874519(4)
0
0.312816604092084
4.22788075922252
1.0
0.3340476235037(6)
3.970691522343(6)
30.0
0.323213333716(3)
2.33632892937(2)
1000
115.539777092(3)
0.8050335305(2)
(0)
(0)
F2
F3
1.89253423642110
1.884686031133(5)
1.87869903011051
1.836380342548(8)
1.80974413459626
1.805561498336(6)
1.80314761916276
1.796297920597(6)
1.619837556072(1)
1.60134292154365
1.455289136414(2)
0.425519391740(4)
1.7888035846(2)
0.185424043556224
0.1846128537367(4)
0.183940826371101
0.1786103121374(3)
0.174898795219002
0.1742958493924(4)
0.173945546345814
0.1729424293568(4)
0.1432933391084(2)
0.139821925866842
0.1103513887593(2)
0.178170159306(2)
1.11980894482(3)
4. Tests and comparisons

Several checks were done in the past [4,5,8] to verify that the analytical expansions of
the master integrals in the special points, used within the numerical program, satisfy the
differential equations and agree with the results existing in the literature.
A remarkable feature of the extension of the RK-method to the complex plane is that
it provides some natural self-consistency checks of the algorithm implementation. Starting
from a special point and moving to a chosen value of p2 with different paths in the p2
complex plane, the values obtained for the master integrals should agree inside the errors
of the method discussed previously. One has, however, to remember that paths chosen in
the upper and lower half-plane, respect to the real axis, give opposite sign to the imaginary
part of the master integrals for time-like values of the external invariant above threshold,
pr2 < 1. An even more complete test is to reach the same value of p2 starting from
different special points, hence following different paths, and compare the values of the
master integrals at p2 obtained along the various paths. If the values coincide inside
the assigned errors the consistency between the differential equations, the expansions in
the special points used as initial values, the implementation of the RK-method and the
algorithm for estimating the errors are cross-tested in a rather effective way.
We have performed several of the mentioned checks in the different regions of p2
obtaining the requested agreement.
The only published precise numerical results for the general massive case (all different
nonzero mass values) are presented in [2,11], in the form of a combination of the general
massive case with massless cases, to cancel the pole singularities in (n 4). In our notation
that combination is

(0)
(0)
T123N p2 , m21 , m22 , m23 = 16 +F0 m21 , m22 , m23 , p2 F0 m21 , 0, m23 , p2

F0(0) 0, m22 , m23 , p2 + F0(0) 0, 0, m23 , p2 ,
(15)
321
which has also the property of being independent of ; the overall factor (16) accounts
for the different definition of the master integral in Eq. (1) and our p2 corresponds to (p2 )
in [2] and to s in [11].
(0)
To obtain the values for F0 (0, 0, m23, p2 ) we use the analytic formula presented in [3],
while for the values of F0(0) (m21 , m22 , m23 , p2 ), F0(0) (m21 , 0, m23 , p2 ) and F0(0) (0, m22 , m23 , p2 )
we use the present program. Although the value zero for the masses is not allowed, we
have checked that the limit can be in practice reached numerically. Comparing the results
obtained for the mass values from 106 to 109 , we can estimate the error coming from
having a mass not exactly zero, by taking the difference between the results obtained with
the two smallest values used for the mass to be set to zero. As the error due to zero mass
limit is sometimes comparable with the error due to the RK-method, we sum the two
errors for each of the considered functions. The final error of T123N is the sum of the errors
assigned by the algorithm to each of the contributing functions in Eq. (15). The larger
errors (or less efficiency in calculations) come from the zero mass contributions, for which
an approach based entirely on analytical expressions is in preparation [17]. Furthermore,
the choice of equal values for two or even all the three masses, reduces the number of
the independent equations in the system of differential equations, generating potential
numerical problems, although less serious than those for the zero mass.
In [2] the values for T123N (p2 , m21 , m22 , m23 ) are presented, for different sets of the
masses m1 , m2 , m3 , and for the two regions of small |p2 | < (m1 + m2 + m3 )2 and large
|p2 | > (m1 + m2 + m3 )2 in Tables 1 and 2 of [2], respectively.
Table 4
Comparison for small |p 2 | < (m1 + m2 + m3 )2 . In each box the first entry is the value of the multiple series of
Table 1 of [2], the second entry is our result (the error in the last digits is enclosed in parenthesis). Our value of
p 2 corresponds to p 2 in [2]
m1
m2
m3
p2
T123N
10
10
20
10
25
10
30
10
49
15
50
20
100
20
150
25
150
25
200
7.3129877443
7.3129877442(26)
4.1493850173
4.1493850171(18)
2.3353847298
2.3353847298(14)
0.8117674738
0.8117674738(15)
0.3167501084
0.3167501085(24)
7.9471022759
7.9471022760(33)
6.0171476156
6.0171476159(59)
6.3903568683
6.3903568686(73)
14.5339444977
14.5339444982(87)
15.0523063012
15.0523063010(95)
p2
9
20
25
30
49
50
100
150
150
200
T123N
6.93244055931
6.93244055924(50)
3.63591843327
3.63591843320(95)
1.9428452190
1.9428452191(10)
0.6306847352
0.6306847353(12)
0.1950338472
0.1950338472(21)
6.8270303849
6.8270303852(19)
4.9485063889
4.9485063897(42)
4.7506023184
4.7506023184(66)
1.21816923644
1.21816923648(68)
11.8790613597
11.8790613597(83)
322
Table 5
Comparison for small |p 2 | < (m1 + m2 + m3 )2 and for m1 = 10, m2 = 20, m3 = 100. In each box the first entry
is the multiple series value of [2], the second entry is our result (the error on the last digits is in parenthesis) the
third entry is from the numerical integration in Table 7 of [11]. Our value of p2 corresponds to p 2 in [2] and to
s in [11]
p2
1
p2
T123N
70.6856984
70.6856977(39)
70.686011
70.6798310
70.6798305(21)
70.680106
25
25
T123N
70.75620346
70.75620352(11)
70.756299
70.609519049
70.609519051(97)
70.609231
p2
81
81
T123N
70.92141286
70.92141291(70)
70.921481
70.446146678
70.446146655(35)
70.446044
Table 6
Comparison for p 2 large (|p 2 | > (m1 + m2 + m3 )2 ) and negative. In each box first entry is the multiple series
value of Table 2 of [2], the second entry is our result (the error on the last digits is in parenthesis). Our p2
corresponds to p 2 in [2]
m1
m2
m3
p2
Re T123N
Im T123N
80
100
100
150
150
150
150
200
200
250
0.587432001
0.587431990(41)
1.28284949
1.28284943(16)
0.3286481685
0.3286481687(16)
1.26795173
1.26795173(21)
1.5662482672
1.5662482670(14)
0.9865824304
0.9865824312(19)
4.7638745(12)
4.7638748416(32)
1.6960823345
1.6960823350(17)
1.86355967
1.86355979(34)
2.64395201
2.64395222(31)
11.262835755
11.262835744(24)
20.899600723
20.899600741(10)
11.84587606309
11.84587606302(60)
26.491194705
26.491194694(41)
11.9689438774
11.9689438778(12)
16.10213970663
16.10213970687(77)
29.601246304(26)
29.60124631204(68)
6.02417248918
6.02417248906(81)
20.39852237
20.39852240(12)
27.60904430
27.60904439(18)
We repeat in Table 4 for the same values of the masses and p2 the results of Table 1
of [2] for the multiple series (first entry), pushed to a large number of terms in some cases,
and our results (second entry). The results are in excellent agreement.
Also in Table 7 of [11] the values for the same combination T123N (there called Sc ) are
presented for small s, equal to our p2 , |p2 | < (m1 + m2 + m3 )2 and for m1 = 10, m2 = 20,
m3 = 100. They are repeated here in Table 5, where in each box the first entry comes from
the multiple series of [2] with a large number of terms, the second entry is the present
result, the third entry is from the numerical integration of [11]. Again, we have excellent
323
Table 7
Comparison for p 2 large (|p 2 | > (m1 + m2 + m3 )2 ) and positive. In each box first entry is the multiple series
value of Table 2 of [2], the second entry is our result (the error on the last digits is in parenthesis). Our p2
corresponds to p 2 in [2]
m1
m2
m3
p2
T123N
m1
m2
m3
p2
50
100
100
120
150
150
150
200
200
3.728125558
3.728125610(46)
8.79126989
8.79127000(11)
7.0830520665
7.0830520661(28)
11.361931056
11.361931121(31)
3.3018636831
3.3018636830(21)
250
T123N
7.1836810855
7.1836810854(54)
12.430997190
12.430997250(30)
10.85647158
10.85647158(26)
9.64611359
9.64611360(29)
13.57188440
13.57188463(21)
agreement with the multiple series of [2], and the accuracy of the numerical integration of
[11] is within a few ppm, inside the relative 105 precision declared there.
In Table 6 we report the results of Table 2 of [2] for the combination T123N , for large
and negative p2 (i.e., |p2 | > (m1 + m2 + m3 )2 , the value of p2 here corresponds to p2
in [2]), so that this time there is also an imaginary part. In each box the first entry comes
from the multiple series of [2], with a large number of terms, the second entry is the present
result. Again, we have excellent agreement with the multiple series of [2] in most of the
cases, in few cases there is a deviation of two times the assigned error, that we attribute
to our procedure of approaching the zero mass. In the seventh box we assign an error also
to the multiple series, because, although each sum is taken up to 70 terms, the results are
not yet stable. The assigned error is the difference with the sums taken up to 60 terms. We
attribute the difficulty to the chosen value of p2 = 150, which is too near to the threshold
value (3 + 4 + 5)2 = 144.
In Table 7 we report the results of Table 2 of [2] for the combination T123N , for large
and positive p2 (i.e., |p2 | > (m1 + m2 + m3 )2 ; our p2 corresponds to p2 in [2]). In each
box the first entry comes from the multiple series of [2], with a large number of terms, the
second entry is the present result. Also here we have excellent agreement with the multiple
series of [2] in most of the cases, in few cases there is a deviation about two times the
assigned error, that we attribute to our procedure of approaching the zero mass.
5. Conclusions
We propose to solve numerically, by means of the RungeKutta method extended to
the complex plane, the system of the differential equations satisfied by the MI related
to the diagrams, which due to the large number of occurring parameters cannot be
calculated analytically.
We apply the method to the study of the simplest nontrivial diagram, the general massive
2-loop sunrise self-mass, which is already exhibiting a number of intriguing analytic
properties. We obtain, for all the allowed values of the parameters and of the external
324
invariant p2 , very accurate values of the MI within reasonable CPU time, in good
agreement with the results already present in the literature.
The method can be naturally extended to the other self-mass diagrams of the same
order, like the 2-loop 4-propagator self-mass diagram for which the differential equation is
already known [18].
The extension to higher order self-mass diagrams will only increase linearly the number
of the MI and of the differential equations in the system, while the growing of the number
of parameters is not a problem at all. Also the extension to diagrams with three or more
external legs, which means multi variable cases, can be easily envisaged.
The true difficulty of the method is the need of the initial conditions for starting the
numerical solution of the differential equations; clearly the initial conditions have to be
provided by an independent method. Of special values are, in this respect, the special points
(such as 0, , thresholds and pseudothresholds) where an analytic calculation is easier and
sometimes possible, as in the case discussed in this paper. When the special points are used,
also the first derivative of the MI have to be provided as an independent input to the Runge
Kutta approach, but that is analytically a relatively simpler task, amounting to an iteration
of the expansions provided by the differential equations.
Acknowledgements
We thank Sandro Rambaldi for his invaluable advise on the use of RungeKutta method
to solve differential equations.
One of us (H.C.) is grateful to the Bologna Section of INFN and to the Department of
Physics of the Bologna University for support and kind hospitality.
Appendix A. Corrections to some analytic formulae of [5,8]

We report here for completeness the correct form of the formulae, which are wrongly
reported in our previous publications [5,8] and are used here to obtain the numerical values
of the MI at pseudo-thresholds and threshold.
In Section 5 of [5] there are three misprints. In Eq. (41) the factor 1/16 in front of the
integral should be missing. In the first line of Eq. (44) log(y) should read log(yS ) and in
Eq. (47) there is a missing factor 4 in front of I2 .
In [8] there is one misprint in Eq. (45): the second line from the end should be of the
opposite sign ( + ). In Eq. (31) of [8] the expression for I3 (m1 , m2 , m3 ) should be
symmetric in all the masses, so it becomes
I3 (m1 , m2 , m3 ) +
I3 (m1 , m1 , m2 )
I3 (m2 , m1 , m1 ),
I3 (m1 , m2 , m3 ) =
with

I3 (m1 , m2 , m3 )

3
3

log m
log m
1
m1
m2
+
= m1 m2 dm3
m3 (m1 + m2 + m3 ) m3 + m1 m3 + m2
(A.1)
325

m1 + m2
log(t t1 ) log(t t2 )
= i log
4m1

m1 + m2
log(t + t2 ) log(t + t1 )
+ log
4m2

+ log(t t1 ) 2 log(1 t1 ) log(t1 ) log(t t2 ) 2 log(1 t2 ) log(t2 )

log(t + t1 ) 2 log(1 + t1 ) log(t1 )

+ log(t + t2 ) 2 log(1 + t2 ) log(t2 )

t t1
t t2
t t1
t t2
2 Li2
+ 2 Li2
+ Li2
Li2
1 t1
1 t2
t1
t2

t + t1
t + t2
t + t1
t + t2
+ 2 Li2
(A.2)
2 Li2
Li2
+ Li2
,
1 + t1
1 + t2
t1
t2
where the expressions of the logarithms account now properly for their imaginary part.
Consequently, Eq. (42) of [8] should be replaced by

1
5
1
K
.
= +
log(2) + Cl2
b
(A.3)
32
32 32
8
2
References
[1] F.V. Tkachov, Phys. Lett. B 100 (1981) 65;
K.G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 (1981) 159.
[2] F.A. Berends, M. Bhm, M. Buza, R. Scharf, Z. Phys. C 63 (1994) 227.
[3] M. Caffo, H. Czyz, S. Laporta, E. Remiddi, Nuovo Cimento A 111 (1998) 365, hep-th/9805118.
[4] F.A. Berends, A.I. Davydychev, N.I. Ussyukina, Phys. Lett. B 426 (1998) 95, hep-ph/9712209.
[5] M. Caffo, H. Czyz, E. Remiddi, Nucl. Phys. B 581 (2000) 274, hep-ph/9912501, see also Appendix A of the
present article.
[6] A.I. Davydychev, V.A. Smirnov, Nucl. Phys. B 554 (1999) 391, hep-ph/9903328.
[7] S. Groote, A.A. Pivovarov, Nucl. Phys. B 580 (2000) 459, hep-ph/0003115.
[8] M. Caffo, H. Czyz, E. Remiddi, Nucl. Phys. B 611 (2001) 503, hep-ph/0103014, see also Appendix A of the
present article.
[9] P. Post, J.B. Tausk, Mod. Phys. Lett. A 11 (1996) 2115, hep-ph/9604270.
[10] S. Groote, J.G. Krner, A.A. Pivovarov, Eur. Phys. J. C 11 (1999) 279, hep-ph/9903412;
S. Groote, J.G. Krner, A.A. Pivovarov, Eur. Phys. J. B 542 (1999) 515, hep-ph/9806402.
[11] G. Passarino, Nucl. Phys. B (2001) 257, hep-ph/0108252.
[12] S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087, hep-ph/0102033;
S. Laporta, Phys. Lett. B 504 (2001) 188, hep-ph/0102032.
[13] O.V. Tarasov, Nucl. Phys. B 502 (1997) 455, hep-ph/9703319.
[14] W.H. Press, S.A. Teukolsky, W.T. Vetterling, B.P. Flannery, Numerical Recipes in FORTRAN. The Art of
Scientific Computing, Cambridge Univ. Press, 1994.
[15] C. Itzykson, J.B. Zuber, Quantum Field Theory, McGrawHill, New York, 1980, Sections 63, p. 301, and
references therein.
[16] E.L. Ince, Ordinary Differential Equations, Dover, New York, 1956.
[17] H. Czyz, A. Grzelinska, M. Zabawa, in preparation.
[18] M. Caffo, H. Czyz, S. Laporta, E. Remiddi, Acta Phys. Pol. B 29 (1998) 2627, hep-th/9807119.

Quantum mechanics on noncommutative

Riemann surfaces
Bogdan Morariu a , Alexios P. Polychronakos b,1
a Department of Physics, Rockefeller University, New York, NY 10021, USA
b Physics Department, University of Ioannina, 45110 Ioannina, Greece
Abstract
We study the quantum mechanics of a charged particle on a constant curvature noncommutative
Riemann surface in the presence of a constant magnetic field. We formulate the problem by
considering quantum mechanics on the noncommutative AdS2 covering space and gauging a discrete
symmetry group which defines a genus-g surface. Although there is no magnetic field quantization
on the covering space, a quantization condition is required in order to have single-valued states on
the Riemann surface. For noncommutative AdS2 and sub-critical values of the magnetic field the
spectrum has a discrete Landau level part as well as a continuum, while for over-critical values we
obtain a purely noncommutative phase consisting entirely of Landau levels. 2002 Elsevier Science
B.V. All rights reserved.
1. Introduction
Noncommutative quantum field theories have been studied very intensely over the
last few years especially because of their relation to M-theory compactifications [1] and
string theory in nontrivial backgrounds [24]. They are interesting because they preserve
some of the nonlocal properties inherent in string theory. For example, T-duality is a
manifest symmetry [5,6]. (For recent reviews of noncommutative gauge theory see [7].)
Recently, noncommutative ChernSimons was shown to give an alternative description of
the fractional quantum Hall effect [812].
E-mail addresses: morariu@summit.rockefeller.edu (B. Morariu), poly@teorfys.uu.se
(A.P. Polychronakos).
1 On leave from Theoretical Physics Department, Uppsala University, Sweden.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 9 8 - 5
B. Morariu, A.P. Polychronakos / Nuclear Physics B 634 (2002) 326338
327
At low enough energies the single-particle sector becomes relevant and thus it is enough
to consider noncommutative quantum mechanics. (For early studies of noncommutativity
in quantum mechanics see [1318].) In particular, one can consider the quantum mechanics
of a charged particle moving on a two-dimensional noncommutative surface in the presence
of a constant magnetic field. The problem on the plane and the sphere has been considered
in [1921], on the noncommutative torus in [22] and on noncommutative AdS2 in [23].
In this paper we generalize this to higher genus noncommutative Riemann surfaces. In
Section 2 we review noncommutative U (1) gauge theory on AdS2 . We study the quantum
mechanics of a charged particle on noncommutative Euclidean AdS2 in a constant magnetic
field in Section 3. This problem has also been considered in [23]; however, because only
representations of the Lie algebra sl(2, R) which integrate to representations of the group
SL(2, R) were used, a quantization of the magnetic field resulted. Such a quantization is
certainly not observed in the commutative limit, since the topology of AdS2 is trivial. We
show that more general representations are allowed such that the magnetic field is not
quantized.
We also discuss the energy spectrum. Unlike the usual Landau levels on the plane, for
commutative AdS2 the Hamiltonian has both a discrete spectrum and a continuum. Semiclassically this can be understood as follows: on a plane, for any finite energy the classical
orbits are closed and single valuedness of the wave function phase around the orbit leads
to a quantization of the energy. On AdS2 , if the energy is above a threshold, we have open
trajectories and no quantization of the energy. The spectrum for noncommutative AdS2 is
similar, except that for a magnetic field above a critical value Bcrit = 1/ all motion is
bounded and there is only a discrete spectrum.
In Section 4 we construct quantum mechanics on a noncommutative Riemann surface
by modding AdS2 by a discrete subgroup of SO(2, 1) which defines the cycles of
a genus-g surface. Gauging of this discrete subgroup is just the requirement that the
Hilbert space is projected to states that transform trivially under the action of the subgroup,
which corresponds to invariance of (scalar) wavefunctions around the cycles, up to
gauge transformations and vacuum angles. We show that this gauging requires a certain
quantization condition for the magnetic field and demonstrate that in the commutative
limit this condition reduces to the standard Dirac quantization of the flux. The Landau
level spectrum for a noncommutative Riemann surface is the same as that of AdS2 but
with finite degeneracy. One also expects a discrete spectrum above the threshold, but little
is known about this even in the commutative case. A partial list of studies of the same
problem on a commutative Riemann surface is [24,25]. The concept of a noncommutative
Riemann surface was also discussed in [26].
Finally, in the last section we briefly discuss some open issues for future investigation.
2. Gauge theory on the noncommutative AdS2

In this section we discuss U (1) gauge theory on the noncommutative AdS2 . We
follow closely the treatment of the noncommutative sphere in [21]. Field theory on the
noncommutative sphere was introduced in [27] and studied rather extensively in [28]. First
328
consider the Lie algebra
[xi , xj ] = i ij k xk ,
(1)
r
where and r are real parameters which we take to be positive, 123 = 1 and indexes are
raised and lowered with the metric = diag(1, 1, 1). The rescaled generators Ri = r xi
satisfy the sl(2, R) relations
[Ri , Rj ] = i ij k Rk ,
(2)
with the quadratic Casimir

R 2 = R12 + R22 R32 .
(3)
Let us briefly describe the unitary representations of sl(2, R). These representations are
infinite-dimensional since the metric is of indefinite signature. Usually, in the mathematical
literature [29] one finds the description of the representations of the Lie algebra which can
be integrated to true representations of the groups SL(2, R) or SO(2, 1). While somewhat
less familiar than the unitary representations of su(2), they can nevertheless be obtained
exactly in the same way. One starts with an arbitrary R3 eigenstate |m of unit norm and
obtain other states in the representation by applying R = R1 iR2 . Using the fact that
Ri are Hermitian, one can calculate the norm of these states and require it to be positive.
After this analysis [3032], one obtains representations which are of the following types:
Principal discrete series. These representations act on the Hilbert space

Dj = |j ; m; m = j, j 1, j 2, . . . .
(4)
The state |j ; m has R3 = m, and the state |j ; j has the highest weight in Dj while
|j ; j has the lowest weight in Dj+ . The Casimir equals R 2 = j (1 j ) where j is an
arbitrary positive real number.
Principal continuous series. These representations act on the Hilbert space

Cj = |j, ; m; m = , 1, 2, . . .
(5)
labeled by two continuous parameters j and . The Casimir is given by R 2 = j (1 j )
for j = 1/2 + is where s is real and positive. The parameter is real and can be chosen
to satisfy [0, 1). The states have R3 = m.
Complementary continuous series. These representations act on the Hilbert space

Ej = |j, ; m; m = , 1, 2, . . . .
(6)
The parameter is real and can be chosen to satisfy [0, 1) while j is real in the
interval j (1/2, 1) and must satisfy j (1 j ) > (1 ).
Identity representation. This is the trivial one-dimensional representation.
The representations in the discrete series form a discrete set only if we require them to
integrate to representations of either the group SL(2, R) or SO(2, 1). Then, j must be an
integer or half integer for SL(2, R), while for SO(2, 1) it must be an integer. In general,
329
a unitary representation of a semi-simple Lie algebra is also a unitary representation of the

of all the groups G with the given algebra. Since such a group
universal covering group G

to obtain representations
G has the form G = G/
where is a discrete subgroup of G,
Equivalently, a necessary and
of G we must restrict to -invariant representations of G.
sufficient condition for a representation of a semi-simple Lie algebra to integrate to a
representation of the Lie group G, is to be a good representation of a maximal compact
subgroup of G. Regarded as a Riemannian manifold (with the metric given by the Killing
metric), the universal covering group of SL(2, R) or SO(2, 1) is in fact the familiar AdS3 of
unit radius and nonperiodic time.2 It has the topology D R, where D denotes a disk. We
can obtain SL(2, R) by identifying time with period 4 and SO(2, 1) by identifying time
with period 2 . Both groups have the topology D S 1 . This leads to the quantization of
j described above.
Noncommutative AdS2 of radius r is defined as the matrix algebra generated by xi in
the Dj+ irreducible unitary representation where the Casimir satisfies
x 2 = x12 + x22 x32 = r 2 ,
and x3 is positive definite. We must take j > 1 so that
is given by
=
(7)
x2
be negative. Then the parameter
r2
.
j (j 1)
(8)
For states with x1 , x2 0, x3 r, (1) reduces to the planar noncommutativity relation

[x1 , x2 ] = i and thus is identified as the noncommutativity parameter. Note that for
fixed r, since j can vary continuously, there is no quantization of .
In the operator approach, scalar fields on noncommutative AdS2 space are defined as
arbitrary operators on the Hilbert space and thus can be identified with arbitrary elements
of the algebra . We can implement the infinitesimal action of sl(2, R) on the generators
of the noncommutative AdS2 as [Ri , xj ] = i ij k xk . Since this action is a derivation, we can
define it also on an arbitrary element of the algebra as
Li () = [Ri , ].
(9)
We can then define the derivative operators i = ri Ri on , which satisfy

1
[i , j ] ij k k = 0.
(10)
r
We now formulate gauge theory on the noncommutative AdS2 . The covariant derivative
operators can be defined as a perturbation of the derivative operators
Di = i + iAi .
(11)
Under gauge transformations, which are just time-dependent infinite-dimensional unitary

matrices U , the covariant derivative operators transform as
Di = U Di U 1 .
2 For an illuminating discussion of SL(2, R) and its covering group see [33].
(12)
330
It is convenient to also introduce covariant coordinates [34]

Xi = i Di = xi Ai ,
(13)
parametrizing a noncommutative two-dimensional membrane. The requirement that there

be only two independent components of the gauge field on AdS2 is equivalent to the
requirement that there be no transversal excitations of the membrane. So the Xi satisfy
the hyperboloid condition X2 = r 2 , or, equivalently,
2
r
j (j 1)
2
=
.
D =
(14)
r2
This can be rewritten as
x i Ai + Ai x i A2 = 0.
(15)
In the commutative limit 0, (15) is just the condition that Ai is tangent to the
hyperboloid.
We can define a gauge covariant field strength as
1
iFij = [Di , Dj ] ij k Dk .
(16)
r
Notice that Fij = 0 for vanishing Ai or any other gauge equivalent configuration. For a
commutative time we also introduce D0 = 0 + iA0 and define
iF0i = [D0 , Di ].
(17)
Since the integral on AdS2 is just = 2 Tr() the Maxwell action takes the form

1
S = 2 dt 2 Tr F F .
(18)
4g
3. Quantum mechanics and spectrum on noncommutative AdS2

In this section we discuss the quantum mechanics of a charged particle in a constant
magnetic field on a noncommutative AdS2 .
The magnetic field, defined as Bi = 12 i j k Fj k , takes the form
1
iBi = i j k Dj Dk + Di .
(19)
r
To have a uniform magnetic field we will take Bi proportional to the gauge-covariant
coordinate Xi
B
i B
Di ,
Bi = Xi =
r
r
and this together with Eq. (19) implies
[Di , Dj ] =
1 B k
ij Dk ,
r
(20)
(21)
331
which, up to a rescaling of Di , are just the sl(2, R) relations. Thus we have

1 B
Ki ,
(22)
r
where Ki satisfy the algebra (2). Since Di still have to satisfy (14), we take the
representation of Ki to be irreducible and of the form Ds with s > 1. We will show shortly
that the choice of Ds+ or Ds depends on the value of B. By a gauge transformation we can
bring the Ki in the standard form where K3 is diagonal. The relation (14) implies that s
must satisfy
Di = i
(1 B)2 =
j (j 1)
.
s(s 1)
(23)
Since neither j nor s are quantized when considering unitary representations of the Lie
algebra, the relation (23) does not imply any quantization of B as was assumed in [23].
This result is compatible with the commutative limit where B is not quantized, since AdS2
has a trivial topology.
For a charged field , with the gauge transformation = U , we define the covariant
derivative action as
Di () = Di i .
(24)
On the right-hand side, Di represents an element of the algebra while on the left-hand side
it denotes an action on . We can also write this as
1
iDi () = ( Ki Ri ),
(25)
r
where = 1 B.
Note that is a matrix multiplied on the left by Ds representation matrices and on
the right by Dj+ representation matrices. It is more convenient to have both of these
multiplications described as actions on the left. Since the generators are Hermitian,
transposition is equivalent to complex conjugation and this takes Dj+ into Dj . Concretely,
to the matrix nm we associate the state
| =
mn |s + m
s | j nj ,
(26)
n,m=0
and then the relation (25) can be written as

iDi | =
1
(j )
(s)
Ri
|,
+ Ri
r
(27)
where Ri
(Ri ) denote operators acting on states of the Ds (Dj ) representations. In
this notation, the action of the generators Ji of the sl(2, R), representing the infinitesimal
symmetry of AdS2 , takes the form
(s)
(j )
|.
+ Ri
Ji | = Ri
(28)
(s)
(j )
In particular, J3 can be identified with angular momentum around the origin.
332
The equation of motion for can be obtained from an action of the Schrdinger type

1

i
S = dt 2 Tr i + Di () D () .
(29)
2
Then the Hamiltonian is given by H = 12 D 2 , and with a little bit of algebra it can be
rewritten as

2 2
Br
H = 2 J2 +
(30)
.
2r
The spectrum and eigenstates of the Hamiltonian are trivially related to those of J 2 , and
thus they are given by pure representation theory. They can be obtained from the following
tensor product decompositions
Ds
Dj
Ds+j
+m ,
(31)
m=0
Ds+
Dj
D+n
nI
C1/2+is
,
(32)
s=0
where I = {n Z; 1/2 < + n |s j |}. In (32) the + sign is taken for s > j and
= |s j | mod(1) [0, 1). Note that in (32), the r.h.s. contains representations from both
the discrete and the continuous series, and that the discrete series start at k 12 . The dots
in (32) stand for complementary series representations. However, in the expansion of a
normalizable state in terms of energy eigenstates the complementary series representations
have zero measure [33], thus they do not contribute to the spectrum.
To choose between Ds+ and Ds we require that the Hamiltonian (30) be bounded from
below. For B < 1/ , since is positive we choose Ds+ . By (32) there is only a finite number
of discrete series representations and because of the second term in (30) the Hamiltonian
is positive definite. The spectrum consists of a finite set of discrete Landau level energies

2 2
Br
, n I,
En = 2 ( + n)(1 n) +
(33)
2r
above which there is a continuous spectrum, starting at the threshold energy

Ethres =
B2r 2
.
+
8r 2
2
(34)
For B > 1/ , since is negative, we have a Hamiltonian bounded from below if we

choose Ds . In this case there is only a discrete energy spectrum given by

2 2
Br
, n = 0, . . . , .
En = 2 (j + s + n)(1 j s n) +
(35)
2r
This phase is a purely noncommutative one.
We can check that, in the limit r 2 with constant , the above spectrum reproduces
the Landau levels on the noncommutative plane found in [20]. In that limit the continuous
333
Fig. 1. Continuous spectrum and Landau levels.
spectrum is pushed to infinity. For the discrete levels we have, up to O(r 2 ) corrections,

r2
1
r2 1
1
+ ,
s=
+ ,
En = n +
j=
(36)
|B|,
2
| | 2
2
in agreement with the planar result. The density of states agrees as well. This gives an
independent justification for the choice of Ds for the representation of the covariant
derivatives in the case B > 1/ , since the system maps to the correct over-critical planar
phase. At the commutative AdS2 limit, 0 for constant r, we recover the standard results
[24].
The form of the spectrum we obtained is depicted in Fig. 1: for small positive values
of the magnetic field (|s j | < 12 ) the spectrum is entirely continuous, with a threshold
as in (34). For positive B corresponding to s j = 12 , a single Landau level peels
from the bottom of the continuum. For s j = 32 a second level peels, and so on. As
B Bcrit = 1/ , an infinity of Landau levels has formed, while the continuum is pushed
to infinity. Above Bcrit no more Landau levels are formed and there is no continuum.
Similarly, for negative values of B, Landau levels peel from the continuum at points at
which s j equals negative half-integers. Since s > 1, there is a lowest such point, for
s j = [j 32 ] 12 corresponding to some B. , at which the last Landau level peels.
For B < B. no more Landau levels form. We see that, for negative B, there is a maximum
2
number of Landau levels Nmax = [j 32 ]. For large r or small , Nmax r . The entirely
discrete spectrum above Bcrit = 1/ and the existence of Nmax are purely noncommutative
effects.
4. Noncommutative Riemann surfaces

In this section we will formulate quantum mechanics on a noncommutative Riemann
surface by gauging a discrete symmetry group of the action (29). To set the stage, we first
review how to obtain a commutative Riemann surface endowed with a constant curvature
metric by modding out the upper half-plane (or the mass hyperboloid) by the action of a
Fuchsian group [35].
334
Consider a Riemann surface of genus g on which we have chosen a canonical

homology basis with generators ai , bi , i = 1, . . . , g, i.e., the intersection numbers are given
by
ai aj = 0,
ai bj = ij ,
bi bj = 0.
(37)
Let us pick a representative in the homology class of each generator which also goes
through a fixed point P on . Then, ai and bi can be interpreted as generators of the
fundamental group 1 () based at P of the surface . As such, they satisfy
g

ai bi ai1 bi1 = 1.
(38)
i=1
To understand Eq. (38), take the above homology generators passing through P to be
geodesics and then cut along them. The resulting surface, called the cut Riemann surface
c , is a 4g-gon and the product on the l.h.s. of (38) is just the boundary cycle. This is
obviously contractable to a point.
The group of isometries of the mass hyperboloid x 2 = r 2 is SO(2, 1). Group elements
of SO(2, 1) acting without a fixed point are called hyperbolic (they are called elliptic if they
have a finite fixed point and parabolic if the fixed point is at infinity). Consider a discrete
subgroup of SO(2, 1) isomorphic to the fundamental group 1 () and containing only
hyperbolic elements. Then must be generated by gai and gbi satisfying
g

gai gbi ga1
= 1.
gb1
i
i
(39)
i=1
All the nondegenerate Riemann surfaces of genus g can be obtained by modding out
the mass hyperboloid by the action of such a group . One can chose a covering of
the hyperboloid such that each fundamental region is isomorphic to the cut Riemann
surface c .
The action (29) is invariant under the infinitesimal sl(2, R) transformations (28). These
transformations, actually, involve both space translations and gauge transformations. They
commute with the Hamiltonian and correspond to the magnetic translations of the particle.
They are, thus, the appropriate transformations to be used in order to reduce the Hilbert
space to the one of the genus-g Riemann surface. As we will see, if j and s are chosen
appropriately, one can integrate the infinitesimal action of these generators and represent
the group on the set of states.
We therefore define quantum mechanics on the noncommutative Riemann surface as the
system obtained by gauging the group , in analogy to the commutative case. Since this
group is discrete this just means that we must project onto the subspace of gauge invariant
states. More generally, we can require invariance up to a phase (vacuum angle)
U (g )V 1 (g ) = ei ,
= 1, . . . , 2g.
(40)
In the above, the index runs over the ai and bi cycles, while U and V denote the
Ds and Dj+ representations of g . For s = j the set of s satisfying (40) form the
algebra of functions on the noncommutative Riemann surface. For s = j the set of s
335
satisfying (40) define a projective module which is the noncommutative analogue of the
set of sections of a vector bundle.
Using (40) repeatedly we obtain the consistency condition
g

g

1
1
gai gbi ga1
V 1
gai gbi ga1
= .
gbi
gbi
U
(41)
i
i
i=1
i=1
As we will now show, Eq. (41) implies a quantization of s j .

For j and s integers, Ds and Dj+ are also representations of SO(2, 1) and the
relation (39) implies

g

g

1 1
1 1
gai gbi gai gbi
=V
gai gbi gai gbi
= 1,
U
(42)
i=1
i=1
thus the consistency condition (41) is satisfied trivially. However, for j and s real positive,
since the representations U and V are only representations of the universal covering group
1) we only have
SO(2,
g

g

1 1
is
1 1
U
(43)
gai gbi gai gbi
=e
gai gbi gai gbi
= eij ,
,
V
i=1
i=1
as we will explain shortly. In this case, the consistency condition (41) is satisfied if the two
phases in (43) are equal. The origin of the above phases is as follows: since all the g are
hyperbolic, they can be written as exponentials of elements in the Lie algebra. Using the
exponential map, g can also be understood as group elements in the universal covering
1). The product on the l.h.s. of (39) with the multiplication performed in
group SO(2,
the universal covering group does not necessarily give the identity but some element of
1) which projects to the identity of SO(2, 1). By looking at the form of the R3 (or
SO(2,
K3 ) generators one can see that such an element is represented by a phase.
Let us associate to each g a curve in SO(2, 1) denoted g (t) representing a portion
of a one-dimensional subgroup passing
through g such that g (0) is the identity and
g
g (1) = g . Then to the product i=1 (gai gbi ga1
gb1
) we associate a curve of length 4g
i
i
by translating and joining the curves g (t) in the obvious way: for t [0, 1) the curve is
given by ga1 (t); for t [1, 2) the curve is given by ga1 (1)gb1 (t 1); and so on. Due to the
relation (39) this must be a closed curve in SO(2, 1). However, the curve winds 2(g 1)
1).
around the noncontractable S 1 cycle of SO(2, 1) and thus it is an open curve in SO(2,
Before we calculate the winding in our problem, let us describe one way of obtaining it
for an arbitrary closed curve g(t) in SO(2, 1). Fix a reference point P on the hyperboloid
and a reference tangent vector at P . The curve g(t)P is a closed curve on the hyperboloid.
The action of g(t) on the reference vector gives a periodic vector field around the curve
g(t)P . The winding is just the number of times the vector spins around itself as it goes
once around the curve and is a topological invariant.
In our problem the curve g(t)P is just the boundary of c , and the reference vector is
parallel transported around the boundary of c . Under parallel transport on a hyperboloid
of radius r around a closed loop enclosing an area A, a vector is rotated by an angle
= A/r 2 . Since the scalar curvature is given by R = 2/r 2 , using the GaussBonnet
336
theorem one can find the area of the surface to be A = 4(g 1)r 2 . Thus under parallel
transport around a vector rotates an angle = 4(g 1). Since we have eiR3 = eij ,
the phase is given by eij = e4i(g1)j . The group defined by the relation (39) is only
represented projectively

g

1 1
gai gbi gai gbi
= e4i(g1)j .
V
(44)
i=1
Projective representations of were also considered in [26] where they were obtained with
the help of a gauge field on the Poincar plane. Here we see that projective representations
naturally occur if j is not an integer or half-integer. Finally, the consistency condition (41)
implies the quantization
n
,
s j =
(45)
2(g 1)
where n is an arbitrary integer.
From experience with the noncommutative sphere and torus we know that a more
B(1 B)1 . This would be the strength
relevant quantity is a rescaled magnetic field B
of the SeibergWitten mapped commutative gauge field in the planar case. From Eq. (23)
we obtain

= 1 s(s 1) j (j 1) .
B
(46)
r2
Since j is fixed for a given r and by relation (8), and s j is quantized as in (45) we
can only take discrete values. Note however that, unlike the commutative case,
see that B
are not equally spaced.
the values of B
As a check, consider the commutative limit, obtained by taking j and s to infinity while
keeping r and B finite (we must choose Ds+ ). In this limit we have B = r12 (s j ), thus
we must keep s j finite. Using this, we obtain the following integral quantization for the
flux
AB = 2n.
(47)
This is the expected Dirac quantization (or integrality of the first Chern number).
5. Concluding remarks
We have formulated the problem of a charged particle on a noncommutative genus-g
Riemann surface and found the condition required for the existence of scalar wavefunctions. The spectrum of the particle, on the other hand, has not been fully identified. To
achieve this, we would need to identify the physical states which satisfy the genus-g condition (40). This is, in principle, a purely group-theoretic problem. We expect the degeneracy of each discrete Landau level to become finite, and also the continuous spectrum to
be fragmented into discrete nondegenerate states. Besides the Landau levels, there should
be additional states below the threshold corresponding to linear combinations of complementary series states. It is interesting to understand how the continuous spectrum for the
337
noncommutative AdS2 emerges from the discrete spectrum of the noncommutative Riemann surface as we take the genus g and thus the area to infinity. In particular, note that
only principal series states contribute to the continuum. As we take the area to infinity the
density of states increases, but to obtain a continuum the density must scale as the square
root of the area. Presumably, this is what happens for the principal but not for the complementary continuous representation states. Carrying out this calculation and identifying the
full spectrum and degeneracies is a very interesting open issue.
The Dirac-like quantization condition for the strength of the magnetic field was derived
by demanding invariance of the wavefunction under magnetic translations around the
cycles of the noncommutative Riemann surface. It should be stressed that, as in the
noncommutative torus case and unlike the sphere, this is not a requirement for consistency
of the problem. In fact, we could have promoted the wavefunction into a multicomponent
vector by tensoring it with an N -dimensional vector space VN and demand invariance under
combined magnetic translations and U (N) transformations, which would have resulted in
an N -fold decrease in the unit of quantization in (45). This corresponds to overlapping
N copies of the fundamental domain of the Riemann surface.
In the toroidal case [22], the problem can be analyzed entirely in the canonical
framework by defining physical coordinate and momentum variables which are welldefined on the torus. The representation theory of the algebra of these observables
reproduces the above extended wavefunctions. In the genus-g case there is no immediately
obvious complete set of such observables. Formulating and analyzing the noncommutative
Riemann problem in terms of such canonical observables is an interesting open problem.
Finally, we should remark that, although here we have only considered AdS2 =
SL(2, R)/U (1), it is obvious that the construction can be generalized to G/H where G
is a real semi-simple Lie group and H is its maximal compact subgroup. Application of
this technique to physically relevant situations, such as the noncommutative gravity setting
of [36], would be an interesting possibility. Moreover, the methods developed in this paper
could also be applied to the study of D-branes on AdS3 , see [37] and references therein.
Acknowledgements
We would like to thank K. Bering for discussions and C. Bachas for pointing out the
connection to D-branes. This work was supported in part by the US Department of Energy
under Contract Number DE-FG02-91ER40651-TASK B.
References
[1] A. Connes, M.R. Douglas, A. Schwarz, JHEP 9802 (1998) 003, hep-th/9711162.
[2] C. Chu, P. Ho, Nucl. Phys. B 550 (1999) 151, hep-th/9812219;
C. Chu, P. Ho, Nucl. Phys. B 568 (2000) 447, hep-th/9906192.
[3] V. Schomerus, JHEP 9906 (1999) 030, hep-th/9903205.
[4] N. Seiberg, E. Witten, JHEP 9909 (1999) 032, hep-th/9908142.
[5] A. Schwarz, Nucl. Phys. B 534 (1998) 720, hep-th/9805034.
[6] D. Brace, B. Morariu, B. Zumino, Nucl. Phys. B 545 (1999) 192, hep-th/9810099;
338
D. Brace, B. Morariu, B. Zumino, Nucl. Phys. B 549 (1999) 181, hep-th/9811213.

[7] J.A. Harvey, hep-th/0102076;
M.R. Douglas, N.A. Nekrasov, hep-th/0106048;
R.J. Szabo, hep-th/0109162.
[8] L. Susskind, hep-th/0101029.
[9] A.P. Polychronakos, JHEP 0104 (2001) 011, hep-th/0103013;
A.P. Polychronakos, JHEP 0106 (2001) 070, hep-th/0106011.
[10] S. Hellerman, M. Van Raamsdonk, JHEP 0110 (2001) 039, hep-th/0103179.
[11] D. Karabali, B. Sakita, hep-th/0106016.
[12] B. Morariu, A.P. Polychronakos, JHEP 0107 (2001) 006, hep-th/0106072.
[13] R. Jackiw, hep-th/0110057;
G. Dunne, R. Jackiw, C. Trugenberger, Phys. Rev. D 41 (1990) 661.
[14] G.G. Athanasiu, E.G. Floratos, S. Nicolis, J. Phys. A 29 (1996) 6737, hep-th/9509098;
E.G. Floratos, S. Nicolis, hep-th/0006006.
[15] J. Lukierski, P.C. Stichel, W.J. Zakrzewski, Ann. Phys. 260 (1997) 224, hep-th/9612017.
[16] D. Bigatti, L. Susskind, Phys. Rev. D 62 (2000) 066004, hep-th/9908056.
[17] C. Duval, P.A. Horvathy, Phys. Lett. B 479 (2000) 284, hep-th/0002233.
[18] J. Gamboa, M. Loewe, J.C. Rojas, Phys. Rev. D 64 (2001) 067901, hep-th/0010220.
[19] V.P. Nair, Phys. Lett. B 505 (2001) 249, hep-th/0008027.
[20] V.P. Nair, A.P. Polychronakos, Phys. Lett. B 505 (2001) 267, hep-th/0011172.
[21] D. Karabali, V.P. Nair, A.P. Polychronakos, hep-th/0111249.
[22] B. Morariu, A.P. Polychronakos, Nucl. Phys. B 610 (2001) 531, hep-th/0102157.
[23] R. Iengo, R. Ramachandran, hep-th/0111200.
[24] A. Comtet, Ann. Phys. 173 (1987) 185.
[25] M. Guztwiller, Chaos in Classical and Quantum Mechanics, Springer, Berlin, 1990;
M. Antoine, A. Comtet, S. Ouvry, J. Phys. A 23 (1990) 3699;
J. Bolte, F. Steiner, J. Phys. A 24 (1991) 3817;
J.E. Avron, M. Klein, A. Pnueli, Phys. Rev. Lett. 69 (1992) 128;
R. Iengo, D.P. Li, Nucl. Phys. B 413 (1994) 735, hep-th/9307011.
[26] G. Bertoldi, J.M. Isidro, M. Matone, P. Pasti, Phys. Lett. B 484 (2000) 323, hep-th/0003200.
[27] J. Madore, Class. Quantum Grav. 9 (1992) 69.
[28] U. Carow-Watamura, S. Watamura, Commun. Math. Phys. 183 (1997) 365, hep-th/9605003;
U. Carow-Watamura, S. Watamura, Commun. Math. Phys. 212 (2000) 395, hep-th/9801195;
U. Carow-Watamura, S. Watamura, Int. J. Mod. Phys. A 13 (1998) 3235, q-alg/9710034;
C. Klimcik, Commun. Math. Phys. 199 (1998) 257, hep-th/9710153;
J. Gratus, q-alg/9710014;
S. Iso, Y. Kimura, K. Tanaka, K. Wakatsuki, Nucl. Phys. B 604 (2001) 121, hep-th/0101102.
[29] V. Bargmann, Ann. Math. 48 (1947) 568.
[30] L.J. Dixon, M.E. Peskin, J. Lykken, Nucl. Phys. B 325 (1989) 329.
[31] M.S. Plyushchay, J. Math. Phys. 34 (1993) 3954.
[32] J. Maldacena, H. Ooguri, J. Math. Phys. 42 (2001) 2929, hep-th/0001053.
[33] R. Carter, G. Segal, I.G. Macdonald, Lectures on Lie Groups and Lie Algebras, Cambridge Univ. Press,
1995.
[34] J. Madore, S. Schraml, P. Schupp, J. Wess, Eur. Phys. J. C 16 (2000) 161, hep-th/0001203.
[35] H.M. Farkas, I. Kra, Riemann Surfaces, Springer, Berlin, 1980.
[36] V.P. Nair, hep-th/0112114.
[37] C. Bachas, M. Petropoulos, JHEP 0102 (2001) 025, hep-th/0012234.

The Hamiltonian dynamics of bounded spacetime

and black hole entropy: the canonical method
Mu-In Park
Department of Physics, City College of the CUNY, New York, NY 10031, USA
Abstract
From first principles, I present a concrete realization of Carlips idea on black hole entropy
from conformal field theory on the horizon in any dimension. The new formulation is free of the
inconsistencies encountered in Carlips formulation. By considering a correct gravity action, of which
the variational principle is well defined at the horizon, I derive a correct classical Virasoro generator
for the surface deformations at the horizon through the canonical method. The existence of classical
Virasoro algebra is crucial in obtaining an operator Virasoro algebra, through canonical quantization,
which produces the right central charge and conformal weight A+ /h G for the semiclassical black
hole entropy. The coefficient of proportionality depends on the choice of ground state, which has to be
put in by hand to obtain the correct numerical factor 1/4 of the BekensteinHawking (BH) entropy.
The appropriate ground state is different for rotating and non-rotating black holes but otherwise
it has a universality for a wide variety of black holes. As a byproduct of my results, I am led to
conjecture that non-commutativity of taking the limit to go to the horizon and computing variation
is proportional to the Hamiltonian and momentum constraints. It is shown that almost all the known
uncharged black hole solutions satisfy the conditions for the universal entropy formula. 2002
Elsevier Science B.V. All rights reserved.
PACS: 04.70.Dy; 11.25.Hf; 04.60.Ds
Keywords: Black hole entropy; Virasoro algebra; Canonical quantization
1. Introduction
In the recent six years, there have been several outstanding approaches toward a
statistical mechanical computation of the BekensteinHawking (BH) entropy [14]. (See
E-mail address: muinpark@yahoo.com (M.-I. Park).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 9 2 - 4
340
M.-I. Park / Nuclear Physics B 634 (2002) 339369
also Ref. [5] for an earlier work.) But there is no complete and consistent understanding
about the statistical origin of black hole entropy so far. Each approach assumes a specific
model in a certain regime but a universal mechanism, which can be applied to any kind
of black holes, had been unclear until the recent seminal work of Carlip [6]. According
to Carlip, the symmetries of a horizon, which is treated as a boundary, is a universal
mechanism for black hole entropy. This generalizes Stromingers approach [4], in which
the statistical entropy of the BTZ black hole [7] is computed from the classical Virasoro
algebra at the boundary at infinity [8,9]. By looking at surface deformations [8,10,11]
of the rt plane that leave the horizon fixed, he has shown that the symmetry algebra
contains a classical Virasoro algebra independently of spatial dimensions when appropriate
boundary conditions are chosen. With the aid of Cardys formula for the asymptotic states,
the BH entropy was derived as the leading term of a steepest descent approximation.
The relevance of the rt plane at the horizon to black hole entropy, which resembles the
Euclidean gravity formulation in a radial slicing [12], was a key observation: this made
it possible to elevate Stromingers idea to an arbitrary black hole in higher dimensions
without requiring any microscopic model for quantum states.
There are two big differences between Carlips approach and Stromingers, apart
from the difference in spacetime dimension. The first is that Carlips Virasoro algebra is
computed at the horizon, in contrast to Stromingers, which is computed at the boundary
at infinity where there is no horizon. Carlips approach can encode the details of the
metric at the horizon unlike the Stromingers. The second is that only one copy of the
Virasoro algebra which lives at the horizon is involved, in contrast to the two copies
of the Virasoro algebra which live at the boundary at infinity. Similar results have been
subsequently derived in other frameworks as dimensionally reduced gravity and covariant
phase space methods [13,14].
Unfortunately it has been known that Carlips formulation is not complete [15,16] and
there has been no complete resolution so far [17], as far as I know, though his idea has
received wide interest. This paper addresses a resolution of the problem. This provides a
concrete realization of Carlips idea from first principles.
In Section 2, a new canonical Hamiltonian which satisfies the usual variational principle
in the presence of the boundary is derived from an action principle for Einstein gravity
in any dimension. Full (bulk + boundary) diffeomorphism (Diff ) generator is derived
immediately from a known theorem in gravity theory. Carlips Hamiltonian and Diff
generator are also derived from an action principle with a different choice of boundary
action term, but the variational principle is not well defined in general.
In Section 3, a general black hole metric in any dimension is introduced with the suitable
metric fall-off conditions at the horizon. Fall-off-preserving conditions are derived for Diff
symmetry, which restricts the sub-leading terms as well as the leading terms for the metric
and the Diff parameters.
In Section 4, a grand canonical ensemble, in which the horizon temperature and
angular velocity as well as the horizon itself are fixed, is introduced. As an immediate
consequence of the ensemble and the fall-off conditions, the new Hamiltonian satisfies the
usual variational principle quite well at the black hole horizon, which is differentiable in
the usual terminology.
341
In Section 5, other consequences of the grand canonical ensemble to the Diff at the
horizon are investigated. First, the new Diff generator satisfies all the conditions for a
(differentiable) variational principle. Second, Carlips two main assumptions are derived
which leads to a Virasoro-like algebra from the surface deformation algebra at the horizon:
from the condition of fixed horizon and its temperature, I derive the equation which
expresses radial Diff parameter r in terms of the (time or angular) derivatives of temporal
or angular Diff parameter t , ; from the condition of fixed horizon angular velocity,
I derive the equation of zero angular surface deformations = 0, which reduces the
deformation algebra to the rt plane in the surface deformation space. Now, with the
help of explicit spacetime dependence on t , which is inspired by the null surface at the
horizon as well as Carlips two derived equations, I show that the surface deformation
algebra becomes a Virasoro algebra with a classical central extension. In deriving this
result, I find, after a tedious computation, a peculiar property that first taking the limit to
go to the horizon and then computing variation do not commute by the amount of the
Hamiltonian constraint. This leads me to conjecture that the non-commutativity of the two
limiting process is proportional to the Hamiltonian and momentum constraints.
By restricting to the case of only one independent rotation, I obtain the usual momentum
space Virasoro algebra with the classical central charge c and the conformal weight
which are proportional to the horizon area A+ .
In Section 6, a non-rotating black hole is analyzed with slight modifications of the
formulas for a rotating black hole. The main difference is the appearance of an additional
factor 1/2 in the formula compared to the rotating horizon. There is no natural
connection between the temporal derivative and the angular derivative of Diff parameters .
But one can still introduce an arbitrary velocity parameter, which is not related the
horizons rotation, to obtain a Virasoro algebra. Its c and have additional factor 2
compared to the rotating case.
In Section 7, the canonical quantization of the Virasoro algebra is considered. I find that
the existence of the classical Virasoro algebra is crucial in obtaining an operator Virasoro
algebra which produces the right central charge and conformal weight A+ /h G for the
semiclassical blackhole entropy. Quantum corrections of operator ordering have negligible
contribution O(1/ h ) for a large semiclassical BH entropy. In order to obtain the correct
numerical factor 1/4 of the BH entropy, the minimum conformal weight min , which fixes
the ground state of a black hole configuration, of the classical Virasoro algebra is assumed
differently for rotating and non-rotating black holes: for the rotating one, it is assumed
that min = (32G)1 3A+ /T , which has an explicit dependence on /T with the
inverse Hawking temperature and a temporal-period T . For the non-rotating one, a , T
independent ground state min = 0 can be assumed. The arbitrary /T dependences in
c and exactly cancel each other through effective ones ceff and eff such as the final
entropy has no /T dependence. The appropriate ground state is different for the rotating
and the non-rotating black holes but otherwise it has a universality for a wide variety of
black holes.
In Section 8, several applications are considered and it is found that almost all the known
uncharged solutions satisfy a universal statistical entropy formula, which is the same as
the BH entropy. The theory is extended to include the cosmological constant term and
it is found that the BTZ solution and the rotating de Sitter space have the universal BH
342
entropy also. This computation is contrary to the other recent computations at the spatial
infinity where there is no horizon and which, therefore, may not be relevant to a black
hole. Especially for the rotating de Sitter solution in n = 3 (Kerr-dS3), no complex number
appears at any intermediate step in contrast to the other computations at spatial infinity,
which is hidden inside the horizon.
I conclude with remarks on several remaining questions which are under investigation.
I shall adopt unit in which c = 1.
2. The variational principle and surface deformations for bounded spacetime:

a general treatment
2.1. The variational principle for bounded spacetime in general
Let me start with the spacetime split of the EinsteinHilbert action on a n-dimensional
manifold M, accompanied by the extrinsic curvature terms on the boundary M [18,19]
S = SEH + S M ,
where1
1
SEH =
16G
(2.1)

(16G)2
1
ab
2
pab p
p
d x N h R+
h
n2
n
1
S M =
8G

+ 2t ( h K) + 2a h KN a h hab b N ,

d
n1
t
tf
1
h K
+
dt d n2 x na Na K.
8G
t
i
ti
(2.2)
(2.3)
Here, the boundaries are M = tf ti B with the spacelike boundaries tf and

ti at the final and initial times, respectively, and the intersection B of an arbitrary
timelike boundary with a time slice t .2 N and N a (a = 1, 2, . . . , n 1) are the lapse
and shift functions, respectively. hab is the induced metric on t and h is its determinant.
R ab and K ab are the intrinsic and extrinsic curvature tensors of the hypersurface t ,
ab
ab
ab
and R = h
ab R ab and Kab = hab K are their curvature scalars, respectively. pab =
1
h(K Kh ) is the canonical momentum conjugate to hab and p = p hab .
(16G)
1

g =
N 2 + Na N a
Na
hac hcb = ab ,
Nb
hab

,
g =
N 2
N 2 N a
N2 N b
ab
h N 2 N a N b

,
N a = hab Nb .
I follow the convention of Wald [18] with an exception of (2.6).

2 Here, I assume that the spacelike hypersurface intersects orthogonally the timelike boundary B for each t.
t
See Hawking and Hunters paper [20] for a generalization to a non-orthogonal intersection.
343
na is the unit normal (hab na nb = 1) to the boundary B on a constant time slice t , and
ab = hab na nb and are the induced metric on the boundary B and its determinant,
respectively. Then the first and the second boundary terms of S M cancel the first and the
second total derivatives terms of SEH , respectively, which are proportional to K. The firstorder total derivative action S is closely related to the so-called gammagamma action
which eliminates all the second derivatives of g ( = 0, 1, . . . , n), but the advantage of
the action S is that this can be written in a manifestly covariant form and moreover the
first-order time derivatives of N, N a are removed from the start.
For a general hypersurface t , the variation of S becomes

1
1
(n)
G g
&+
d ,
S =
(2.4)
16G
16G
M
where is the pull-back of

(n1)
& r r r (n 1)& + 16GP
= r
(2.5)
on M[21]. Here, the unit normal vector r of M is normalized as g r r = 1 (the

upper sign for a timelike boundary and the lower sign for a spacelike boundary) and the
induced volume element and the metric on M are given by
(n1)
&1 2 ...n1 = r (n) &1 2 ...n1 ,
= r r + g .
(2.6)
Here, P = ( )(n1) & is the canonical momentum (n 1)-form conjugate

to with the extrinsic curvature = r . It is clear that, if the induced metric
and the unit normal r are fixed on M, the equation of motion for g is the usual
Einstein equation G = 0 at the boundary as well as in the bulk. On the other hand, since
there is no physical evidence of modification of the Einstein equation at the boundary in
any case, I use the guiding principle, for the general treatment of bounded systems, that
the equation of motion is the usual Einstein equation for any kinds of variations. The
boundary condition, which specifies what quantities are fixed at the boundary, changes as
the boundary term S M changes. But it will be shown that my choice of S M is the right
one which is relevant to the black hole horizon when the horizon is treated as a boundary.
However, in this case I do not require the fixed induced metric or the fixed unit normal
at the horizon but rather some appropriate fall-off conditions for them. I am going to treat
this problem within the Hamiltonian formulation, which fits my purpose well; in general,
the required boundary condition in the action formulation is different from that of the
Hamiltonian formulation.
The canonical Hamiltonian becomes3 [6,22]

H N, N a = d n1 x NHt + N a Ha
3 This Hamiltonian form was first studied by Brown, Martinez and York [22] within the context of
thermodynamic partition function. But the physical content is different to that of this paper because of different
boundary conditions. See also [23] for another literature on the Hamiltonian.
344

1
d n2 x 16GN a pr a + na Da N
8G
B

H N, N a + HB N, N a ,
+
(2.7)
where H [N, N a ] and HB [N, N a ] are the bulk and boundary terms on and B,
respectively. Ht and Ha are the bulk Hamiltonian and momentum constraints

16G
1
h
R+
p2 ,
Ht =
(2.8)
pab pab
Ha = 2Db pb a ,
16G
n2
h
where Da denotes the covariant derivative with respect to the spatial metric hab . The
variation in H [N, N a ] due to arbitrary variations in hab , na , N, N a becomes

H N, N a = H N, N a + HB N, N a

1
= bulk terms +
d n2 x nr r N + nr r N
8G
B

1 r
a
r
+ n N Dr + 16GN pa
(2.9)
2
with

H N, N a = bulk terms +
1 r
n N Dr
2
B

1 r
r
a
n Dr N 16GN pa ,
2

1
a
n2
HB N, N =
d
x nr r N + nr r N
8G
1
8G
d n2 x
(2.10)

1 r
r
r
a
a
+ n Dr N + 16GN pa + 16GN pa ,
2
(2.11)
where I have chosen a coordinate system of na = (nr , 0, . . . , 0).4
The bulk terms are the usual variation terms for pab , hab , which produce the bulk
equations of motion [18] as well as the Hamiltonian and momentum constraints
Ht 0,
Ha 0.
(2.12)
The additional variations at the boundary could affect the bulk equation of motions in
general. Let me first consider the first term in the boundary terms of (2.9). By using, from
the definition of nc ,
1
nc = hab hcb na
2
(2.13)
4 Greek letters from the beginning of the alphabet are the boundary indices which do not include radial
coordinate r whence Greek letters from the middle of the alphabet are spacetime indices [6].
the first term becomes

1
d n2 x hab hrb na r N .
16G
345
(2.14)
So, this term would produce an additional term in the bulk equation of motion of the form

pab (x), H N, N a
boundary =

1
(2.15)
(r rB ) na hbr r N
16G
unless the quantity in the bracket ( ) vanishes on the boundary B. Here rB is the radius of the
boundary B. Similarly, the last boundary term in (2.9) produces an additional contribution
Ha
boundary = (r rB )2pr a
(2.16)
to the momentum constraints Ha in general. Contrary to these two contributions, which
are proportional to (r rB ), the second and third terms produce a highly singular
term r (r rB ), which produces the divergence of (0) even when I compute the

commutation relations between the integrated quantities {H [N, N a ], H [N , N a ]}, which
should be related to the measurable things. In order to avoid this problem, I need to assume
the boundary conditions which restrict the radial derivatives of the variations N and
on the boundary B as
r
(2.17)
n r N
B = 0,
r
N n Dr
= 0
(2.18)
B
but N and can be arbitrary otherwise. One may also take, instead of (2.17), (2.18),
N|B = 0 and |B = 0 but this could be too strong a condition depending on the
property of B. Hence I keep (2.17) and (2.18) for a general treatment.
Now, in summary, about the contributions of the boundary terms in (2.9), there are two
possible contributions (2.15) and (2.16) to the corresponding bulk ones from the first and
last terms in (2.9), respectively. The two badly-behaving terms are removed by restricting
r N and r as in (2.17) and (2.18). In other words, the Hamiltonian (2.7) produces
the Einstein equation on the boundary B as well as in the bulk when the contributions of
(2.15) and (2.16) vanish, and the boundary conditions (2.17) and (2.18) are satisfied on B
(gab is fixed on the boundaries at infinity tf ,ti as usual such that there is no contributions
from tf ,ti ).
For completeness, let me remark on another interesting type of boundary C where there
is no added boundary action S M such as when the total action is just the EinsteinHilbert
action S = SEH . In this case, in (2.4) becomes

2
2
16G P + 1
|C =
16GP
n2
n2

(n1)
+ r
& r r r (n1) &.
(2.19)
This, in general, contains P term as well as and r terms. Hence, for the
boundary C, one must fix P as well as and r on it, which may in turn over-specify
the boundary degrees of freedom. The canonical Hamiltonian becomes, for a timelike
346
boundary C,

H N, N a =

d n1 x NHt + N a Ha
1
+
8G

d
C
n2
a
n Na p + 16GN a pr a
h

a
+ n Da N .
16G
x
2n
(2.20)
This has the same form that Carlip has postulated, which can
be seen by expressing the
first p term in terms of extrinsic curvature K = 16G((2 n) h )1 p [6]. But it has a
drawback that the variational principle is not well defined in general (see Appendix A for
details; see also Ref. [15]), which would be related to the over-specification of boundary
data in (2.19). So, I will not consider this boundary C anymore hereafter and concentrate
only the boundary B.
2.2. Diff generators
Now, let me consider the generators of the spacetime Diff
x = ,
g = g + g + g .
(2.21)
The generators may be obtained directly from the usual Noether procedure [2427].
But there is a well-known theorem which identifies the generators from a well-defined
Hamiltonian [11]: if H [N, N a ] is a canonical Hamiltonian of the gravity theory which does
not have boundary term in the variations H [N, N a ] with the lapse and shift functions N
and N a , the Diff generator L[ ] of

pab = pab , L[ ]
hab = hab , L[ ] ,
(2.22)
is given by substituting N, N a in the Hamiltonian H [N, N a ] with the so-called surface
deformation parameters [8,10,11]
a = a + t N a ,
(2.23)
respectively, giving

L[ ] = H N, N a (N,N a )( t , a ) .
(2.24)
t = N t ,
The Diff of gt is instead given by the basic formula (2.21) essentially due to the absence
of the canonical momentum conjugate to gt . Hence the generator becomes (I denote
L[ ] = H [ ])

H [ ] = d n1 x H + J [ ],
(2.25)
where
J [ ] =
1
8G

d n2 x 16G a pr a + na Da t
347
(2.26)
for the boundary B, I assume the boundary conditions for the well-defined variations
H [N, N a ] without boundary terms.5 The usual Diff at the boundary as well as in the
bulk is generated by the cancellation of the boundary Diff from the bulk part in (2.25)
and another boundary Diff from the boundary generator J [ ]. Another interesting effect,
which is crucial to my analysis of black hole entropy, of J [ ] is that it may produce the
classical central extension in the symmetry algebra in general [8,28].
Since I am interested in the symmetry algebra of H [ ] on the physical subspace where
the Hamiltonian and momentum constraints of (2.12) are imposed, I must compute the
Dirac bracket in general [26]. The Dirac bracket algebra would show some interesting
effects of the boundary but this is very complicated in my case. Rather, in this paper, I
themselves without
will use an effective method which gives the Dirac bracket of H []s
6
tedious computations [8,28]

H [ ], H []
J [ ], J []
= J [ ]
(2.27)
from the definition of the Dirac bracket. Here, the Dirac bracket on the right-hand side has
the form

J []
J [],
(2.28)
= J { , }
SD + K[ , ]
in general, where K[ , ]
is a possible central term and { , }
SD is the Lie bracket for the
algebra of surface deformations [8,10,11]:
{ , }
tSD = a a t a a t ,

{ , }
aSD = b b a b b a + g ab t b t t b t .
(2.29)
3. Fall-off conditions at a black hole horizon

Now, let me consider a general black-hole-like metric in n spacetime dimensions [30]
with the BoyerLindquist coordinates (t, r, x )

2

ds 2 = N 2 dt 2 + f 2 dr + N r dt + dx + N dt dx + N dt ,
(3.1)
where the lapse function N vanishes at the horizon and behaves as follows near the outermost horizon r = r+

N 2 = h x (r r+ ) + O(r r+ )2 ,
2
= na a N
r ,
+
(3.2)
5 Of course, this theorem will be modified if I allow the boundary terms in the variation principle [29].
6 This can be equivalently written as {J [ ], J []}
= ( J [ ] J [])/2
in order that the antisymmetry
under is manifest. This can be realized through an explicit form of the Dirac bracket also.
348
where is the inverse Hawking temperature, which is constant on the horizon r+ , and na
is the unit normal to the horizon boundary r = r+ on a constant time slice t .
The suitable fall-off conditions near the horizon r+ are
h 1
N + O(1),
4

t N r r g = O(N)g ,
Nr = O N2 ,
f=
= O(1),
N = O(1),
(3.3)
(3.4)
(3.5)
D N + D N = O(N).
(3.6)
The conditions in (3.4) are the perturbations of the stationary black hole of N r = 0; I
shall keep N r in this and the next sections for generality; but N r = 0 shall be considered
in Section 5.3 by computing the Virasoro algebra for a stationary horizon. (3.6) is the
condition of constant angular velocity = N . However, by comparing with (3.5) one
can see that (3.6) gives a non-trivial restriction if one considers D O(1). To make
it more explicit, let me consider a decomposition
N = f + K ,
(3.7)
where f = O(1), D f + D f = 0, and K vanishes at the horizon r+ . Then, it is easy

to see that (3.6) is satisfied if

r O N 1 ,
(3.8)
K O(N).
(Note that r N = 2f 1 .) Therefore, I require the form (3.7), with the condition (3.8),
for consistency with (3.3)(3.6). These are the basic setup of the fall-off conditions at the
horizon r+ .
As another requirement of consistency, the Diff symmetry (2.21) should not accidently
violate these fall-off conditions. The suitable conditions for this requirement are7

t O(1),
O(1),
r O N2 ,
(3.9)

r t O N 2 ,
(3.10)
r O(1),
2
.
r N O(1),
(3.11)
r O N
This shows that the fall-off-condition-preserving Diff requires the suitable behaviors of
the sub-leading terms of Diff parameters from (3.10) as well as the behavior of the leading
terms (3.9). Moreover, it depends also upon the behavior of the sub-leading term of the
metric from (3.11). Then, it is straightforward to compute that

Kr = O N 1 ,
K = O(1),
K = O N 1 (3.12)
Krr = O N 3 ,
and

pr r = O N 1 ,

p = O N 2 ,
p r = O(1),

p = O N 2 .

pr = O N 2 ,
(3.13)
7 Notice that the hat of in (3.10) is not mistyping. In the unhatted expression, it becomes =
r
N r t + O(1), which is important when I prove the condition (2.18) in Diff (5.1).
349
Note that there is no inequality relation in contrast to the fall-off preserving conditions
(3.8)(3.11).
4. Grand canonical ensemble and the variational principle at the black hole horizon
I have considered the variational principle for a bounded spacetime in general in the
two preceding sections. Now, let me consider the variational principle at the black hole
horizon r+ when the horizon is treated as a boundary with the fall-off boundary conditions
as described in the last section. But, for discussing the statistical mechanical property of
a black hole, I must specify the type of ensemble first of all. For this purpose, I take
a grand canonical ensemble, in which the horizon r+ and its temperature and angular
velocity are fixed [22]. Full elaboration on the impact of this ensemble will be given in the
next section, but here I only mention a direct consequence for the variational principle:
when the horizon r+ is treated as the boundary B, the boundary condition (2.17) is
automatic since
N
r+ = 0
(4.1)
in order that the boundary remains at the horizon r = r+ , which is the solution of
N + N = 0, and also in order that the temperature , which is the coefficient of N + N ,
is unchanged. However, the condition (2.18) is not automatic and I must impose this for
consistency.
Moreover, the fixed angular velocity N in the ensemble insures that there is no
boundary contributions to the momentum constraint H (2.16), which is the coefficient of
N .
One can then see that, using the fall-off conditions in (3.13), the only non-vanishing
1
term in (2.16) is Hr |boundary = O(N
n1). Butr this is not harmful because its contribution to
the bulk Hamiltonian vanishes: d
x N Hr O(N). On the other hand, the boundary
contributions of (2.15) vanish also since

prr (x), H N, N a
boundary O N 2 ,
(4.2)
and all other components in (2.15) vanish trivially from my choice of coordinates and the
metric (3.1). Therefore, the new Hamiltonian (2.7) admits the variational principle even at
the black hole horizon such that the usual bulk equations of motion and the bulk constraints
(2.12) can be applied to the horizon as well as outside of the horizon.
Moreover, I note that since the first term in the boundary part of the Hamiltonian (2.7) is
N a pr a = N pr + O(N) O(1), the Hamiltonian at the black hole horizon, which will
be the true Hamiltonian on the physical subspace, reduces to

H
r N, N a =
+
1
8G

r+

d n2 x 16GN pr + nr Dr N O(1).
(4.3)
350
5. Diff and Virasoro algebra at the horizon

In Section 2.2, the general Diff generators and their symmetry algebras have been
considered for an arbitrary boundary B. But, now let me restrict to the horizon in particular,
where the general Diff reduces to (3.9)(3.11). But, to this end, I need more detailed
knowledge on the Diff in the grand canonical ensemble for the black hole.
5.1. Diff at the horizon: boundary conditions
First, I note that the boundary condition (2.18) is automatic
N nr Dr O(N)
by using the condition (3.8) and the fact that

x = O(N).
(5.1)
(5.2)
Moreover, the general requirement (4.1) of the fixed horizon and temperature for the
grand canonical ensemble in the variational principle at the black hole horizon should be
applied to Diff transformation also. Now let me elaborate in some details what the further
implications of this for Diff . From the fact that g t t = 1/N 2 and g t t = 2 t t , one finds

4Nf t r
N + r + 2N 2 t N a a t + N 2 + O N 3

O N2 .
N 2 =
(5.3)
Hence, an arbitrary Diff transformation produces change in the inverse temperature8 1

by (2)1 nr r N|r+ = (4f N)1 r N 2 |r+ unless the condition (4.1) is satisfied for
the Diff transformation also, i.e.,
N 2
r = 0.
(5.4)
+
In other words, the grand canonical ensemble is satisfied if

t NN r
N 3
N
2f t
r
t
=
r
t N +

g t t ,
2f
2f
N
4f
(5.5)
which allows to express r in terms of t .9 Hence, another boundary condition (2.17) is

also automatic in the grand canonical ensemble.
Therefore, the Diff generator (2.25), for the boundary B as the horizon, is well-defined
without the additional contributions from the boundary when (5.5) as well as all the other
8 Notice that another term nr N from (3.2) should be dropped since this gives the temperature at the lifted
r
position r+ (nr + nr ) not the horizon r+ nr .

9 Notice that here, there is no perturbation on the black hole radius r since N 2 = O(N 2 ) (r r ); this
+
+
can be more directly seen from x r |r+ = r |r+ = 0 due to (2.21) and (3.9). But this does not imply the fixed

horizon area A+ = r+ d n2 x , i.e., fixed entropy, which might cause the ensemble to be absurd: rather, one

has A+ = r+ d n2 x = O(1) according to the solution (5.2).
351
conditions (2.17) and (2.18) for the metric and the conditions for the Diff parameters are
satisfied (all the other boundary contributions vanish as the Hamiltonian did (Section 4)).
The existence of the well-defined Diff generator implies, according to the Noether theorem
[2527], that the Diff symmetry is not broken even with the boundary.
Furthermore, with the well-defined generator H [ ] (2.25) all the information on the
Diff of the bulk part d n1 x H is encoded into the boundary (horizon r+ ) part J [ ]
when it is treated on the physical subspace of H 0; this is another aspect of the t Hooft
holography principle at the black hole [33].
5.2. Diff at the horizon: no angular surface deformations
In the last subsection I have used the condition of fixed temperature for a grand
canonical ensemble. Now, let me consider the condition of fixed angular velocity, which
is due to a chemical potential for the black hole system. Since N , I must compute
N to explore the rotational property of a grand canonical ensemble. On the other hand,
since gt = N , I can compute N (from N = gt N ) as
N = t N + N + t N N + O(N).
(5.6)
Then, one finds that the condition of fixed angular velocity N , i.e., N |r+ = 0 of
a grand canonical ensemble is satisfied when I restrict the surface deformation space of
(2.23) to the rt plane:
= 0;
(5.7)
of course, this does not mean the rt plane in the space of spacetime Diff . Furthermore,
the solution is unique in the grand canonical ensemble.10
Now, Eq. (5.7), together with Eq. (5.5), has been fixed from the conditions of the grand
canonical ensemble on the black hole; these were the two main assumptions in the Carlips
formulation. So, is this the end of story of the grand canonical ensemble? To answer this, let
me elaborate what the condition (5.7) further implies. From the definition (2.21), Eq. (5.7),
i.e., = N t implies
x = N t
(5.8)
and so,

t
x
(no sum)
(5.9)
with = N . This does not show any definite information about arbitrary spacetime
variations. But, let me introduce one assumption about t inspired by these equations (5.8)
and (5.9) that t is at rest on the horizon.
10 The question of angular Diff with = 0 has been raised by Carlip in his talk [6]. I can rule out this in a
grand canonical ensemble. But it is unclear whether this can be ruled out in other ensembles also.
352
This assumption determines the spacetime dependence of t as11

1
f r
t
t
= t 1 N r +
x ,x ,
N

(5.10)
where t (1 f N 1 N r )r = constant is a radial, outgoing null geodesic for a generic

Kerr black hole with the ReggeWheeler tortoise coordinate

r =
(5.11)
log(r r+ ) + O(r r+ )
4
near the horizon r+ ; an appropriate integration constant for dr/dr = N/f is chosen here.
The angular dependence on x reflects the assumption that t is at rest on the horizon
which is rotating with velocity ; however, notice that x dependence in t is arbitrary
in general. Now then, one can show that

f
f
r t =
(5.12)
1 N r t t ,
N
N
t t = t
(no sum),
(5.13)
which makes it possible to express all the derivatives in terms of one-dimensional

derivatives t or . Here, (5.13) shows that t respects the symmetry (5.8) exactly.
Moreover, (5.12) is consistent with the boundary condition (3.10). One can also show that
another non-vanishing Diff parameter satisfies the same equations as (5.12) and (5.13).
5.3. The Virasoro algebra at a stationary horizon
So far my computation was valid for any N r , which is O(N 2 ). But, now let me focus
on a time slice with N r = 0 such that a stationary black hole is considered. Then, the
Diff of J [ ] with respect to the surface deformations becomes

1
2 J [1 ] =
d n2 x 2 nr r 1t + nr r 2 1t
8G
r+
1
+ nr r 1t 2 + 16G2 1 a par + 16G1 a 2 par .
2
(5.14)
It is straightforward to check that the second to the fourth terms vanish based on the
relations for the metric and the Diff parameters. The only non-vanishing contributions
are the fifth and the first terms.
The fifth term. This term needs a tedious computation of {pab , H [ ]} from (2.22). But
the final result is very simple (the derivation is sketched in Appendix B): this becomes,
11 For N r = 0, a more generalized form t = t (t 1 r + ( )1 x , x ) may considered by
relaxing the assumption that t is at rest on the horizon. But one can show that taking different , , and T /
corresponds to taking different ground state.
from = 0,

2 d n2 x 1 r 2 pr r = d n2 x 1 r 2 t Ht + O(N),
r+
353
(5.15)
r+
where

16G
1
Ht =
(5.16)
p p
p p + O N 2
n2
h
is the Hamiltonian constraint evaluated near the horizon r+ . This result shows a very
peculiar property since the Diff of pr r does not preserve the fall-off behavior pr r =
O(N 1 ) near the horizon, but rather becomes as pr r = O(N 2 ) such that the lefthand side of (5.15) is O(1), which does not vanish in general. Notice that this is
sharply contrary to the process of first taking the limit to go to the horizon in the
corresponding term 2 r+ d n2 1 r pr r = O(N) = 0 in J [ ] and then computing the
functional differentiation 0 = 0; these two processes do not commute in general.12
However, the result (5.15) shows the interesting property that the surviving terms are
nothing but the constraints of the system. Hence, the problematic situation of noncommutativity of the two limiting processes of computing the variations and the Poisson
bracket is avoided by the genuine constraints of the system. This implies that there is
no non-commutativity problems automatically when one consider the Dirac bracket, in
which constraints can be implemented consistently through variations. Moreover, since the
usual ReggeTeitelboim approach [28], which finds the appropriate J [ ] such that J [ ]
cancels the boundary part of the variations of the bulk symmetry generator Hbulk[ ], uses
the limiting process first and then the variating, this approach might not be always true. The
only resolution is that the difference of the two processes is proportional to the constraints
as in my case. But since there is no other fundamental constraints besides the Hamiltonian
and momentum constraints in my black hole system, I am led to a conjecture, for any black
hole system, that non-commutativity of taking the limit to go to the horizon and computing
variation is proportional to the Hamiltonian and momentum constraints.
I cannot generally prove this conjecture since it is not clear whether the fall-off
conditions that I have studied in this paper are the most general for the Virasoro algebra
to exist, but this should be the case for the consistency of functional differentiation;
furthermore, it seems that this conjecture may be extended to any gauge theories,
which have the Gauss law constraints as the fundamental constraints, if there are some
appropriate boundaries where the variational principles are well-defined [26].
The first term. The first term in (5.14) is the main term. From (2.13), one has
1 r
n grr .
2f 2
On the other hand, from (2.21) one obtains
nr =
(5.17)

f 2
t N t t + O N 1 ,
grr = 2f 2 t N t +
(5.18)
12 This has been first pointed out by Carlip in the interpretation of a result of Ho and I [15,16].
354
where I have used r = (N/2f )(t N ) t from (5.5), t t = N t from

(5.13) with N r = 0, the rotational symmetry N N 2 = 0, and the condition (3.11). Now
then, the first term in (5.14) becomes

nr
1
d n1 x 2 r 1 t 2 grr
16G
f
r+
1
=
16G
n1

t
2
8N
t
t
2 .
x 1 N N N 2 +
(5.19)
r+
Here, I have taken a safe integration by parts for [ part does not contribute] coordinate
due to the rotational symmetry N = N = 0.

By summarizing the computation, 2 J [1 ] of (5.14) is reduced to

1
2 J [1 ] =
d n2 x 2 nr r 1t + 16Gf 2 1r 2 prr + O(N)
8G
r+

4N
1
n2
t
t
t
N N N 2
2
d
x 1
=
8G
r+

+ d n2 x 1 r 2 t Ht + O(N).
(5.20)
r+
On the other hand, notice that

1
J {1 , 2 }SD =
d n2 x na Da {1 , 2 }tSD + 16G{1 , 2 }aSD par
8G
r+
1
8G
d n2 x

r 4f N
1t 2t (1 2) + O(N).
n
(5.21)
r+
Here I have used the fact that

{1 , 2 }tSD = 2NN 1t 2t (1 2),

{1 , 2 }SD = O N 2 ,
{1 , 2 }rSD = O N 2 ,
{1 , 2 }aSD pr a = O(N) (5.22)
from (5.5), (5.12), (5.13), and = 0. Then (5.20) can be written as follows:

2 J [[1 ]] = J {1 , 2 }SD + K {1 , 2 }SD

1
d n2 x Nf 1 {1 , 2 }tSD Ht + O(N),
+
4
(5.23)
r+
where

K {1 , 2 }SD =
1
8G

d n2 x
r+

t
4N
N N N 2t +
2t
1
1
8G

d n2 x

t
4
t 2t .
1 t t t 2t
355
(5.24)
r+
From the relation (2.27), one further has a Virasoro-type algebra

J [1 ], J [2 ] = 2 J [1 ]

= J {1 , 2 }SD + K {1 , 2 }SD

1
+
d n2 x Nf 1 {1 , 2 }tSD Ht
4
(5.25)
r+
with a central term K[{1 , 2 }SD ]. Notice that, the Virasoro algebra has been generalized to
higher dimensions depending on the number of independent rotations [30]: for uncharged
black holes, the number of independent rotations is given by [(n 1)/2], which is the
number of Casimir invariants of the SO(n 1) rotation group. ([(n 1)/2] denotes the
integer part of (n 1)/2.) Moreover, since the relation (2.27) is valid with the Dirac
bracket, where the constraints are imposed consistently, the last constraint term in (5.25)
has no effect on the central term K[{1 , 2 }aSD ].
Now, in order to obtain a more familiar momentum-space Virasoro algebra, I adopt the
Fourier expansion of (5.10) as usual:

T
2in
t

1
=
(5.26)
n exp
J [ ] =
n Jn
t r N r ,
+
4 n
T
n
with n = n . The normalization of (5.26) is chosen such that one can obtain the standard
factor i(mn)Jm+n in the Virasoro algebra below. Contrary to the lower-dimensional case
(n 3), the four- and higher-dimensional n have x dependences in general, in which the
momentum-space representation is different from the usual one; in this case the central
term is expressed as an integral not just a number. But, for our purpose, I consider the
case where n has no other coordinate-dependence. Then one obtains the familiar classical
Virasoro algebra in the momentum space
{Jm , Jn } = i(m n)Jm+n iKm+n ,
(5.27)
where
A+ T
Jp =
p,0 ,
16G

2
A+
T
2
m m
m,n
Km+n =
16G
(5.28)
and A+ is the area of the horizon r+ . The standard form of the central term is also obtained
by a constant shift on J0

2
A+
T
1
32G T

2
T
A+
1+
=
32G T
J0 J0 = J0 +
(5.29)
356
leading to
{Jm , Jn } = i(m n)Jm+n i

c 2
m m 1 m,n
12
(5.30)
with a central charge

3A+
.
4G T
Notice the sign change of the second term in (5.29) from the shift.
c=
(5.31)
6. The Virasoro algebra for a non-rotating horizon

So far, I have considered the Virasoro algebra for the surface deformation algebra on a
black hole horizon when there is at least one non-zero rotation, i.e., Kerr black hole. But
non-rotating (N = 0) black hole, which has now a spherical symmetry, can be analyzed
similarly with some modifications in the formulas for the rotating horizon.
The main differences in the basic relations are as follows (note N r = 0). First, Eq. (5.5)
becomes
1 N
t t (non-rotating horizon),
2 f
N
t t (rotating horizon).
r =
f
r =
(6.1)
The second difference is that the condition of fixed angular velocity N |r+ = 0 in a
grand canonical ensemble produces

= r, x (non-rotating horizon),
= N t
(rotating horizon)
(6.2)
from (5.6); even without rotation, the angular Diff does not vanish, in contrast to a nonrotating limit of a rotating horizon; but now there is no connection between and t .
However, the basic fall-off conditions (3.3)(3.6) and their preserving conditions (3.8)
(3.11) are the same as for the rotating horizon; except that some of the conditions for the
non-rotating case are milder than those of the rotating case, but in this case also, the nonrotating case can be obtained as a limit of the rotating case:
r O(1) (non-rotating horizon),
r + N r t O(1) (rotating horizon)
(6.3)
is one example.
The remaining analysis on the Virasoro algebra is straightforward and it is easily found
that the coordinate-space Virasoro algebra has the form of (5.25) with

1 1
J [ ] =
d n2 x na Da t ,
2 8G
r+
1 1

K {1 , 2 }SD =
2 8G

d n2 x
r+
357

4
t 1t 2t + t 1t t t 2t

2t 1t t 2t 2t t 1t 2t ,

1
(6.4)
2Nt 1t 2t (1 2) .
2
The overall factor 1/2 is the result of the same factor in (6.1). But, due to the absence of
the relation (5.13), this is not truly a Virasoro algebra because the derivative t cannot be
integrated by parts and so the manifest 1 2 antisymmetry is not attained in contrast to
the generic definition (2.27). Hence, in order that J [ ] be a symmetry generator as (2.27),
I am required to connect t t with t for some angular coordinates x as follows:13
{1 , 2 }tSD =
t t = v t .
(6.5)
v ,
But now, the speed

which is arbitrary, has no connection to the horizons rotation. The
last two terms cancel from the integration by parts and the central term (6.4) becomes the
same as (5.24) with the differences in the overall factor 1/2 and with v instead of N :

1 1
4v
a
n2
t
t
t
v v v 2 +
2
K {1 , 2 }SD =
d
x 1
2 8G
r+

1 1
4
n2
t
t
t
=
(6.6)
t 2 .
d
x 1 t t t 2
2 8G
r+
Then, for the Diff t which lives at the horizon

1
t
t
= t r + x
v
the momentum-space Virasoro algebra has the standard form (5.30) with

2
T
3A+
A+
1+
.
,
c=2
J0 = 2
32G T
4G T
(6.7)
(6.8)
The change of factor 1/2 in (6.4) to 2 in (6.8) comes from the normalization of the Fourier
expansion of t

T
2in
t
1
t r v
=2
n exp
(6.9)
4 n
T
in order to obtain the correct standard form factor i(m n)Jm+n in (5.27). Notice that
furthermore, since should be periodic for 2 rotation, v behaves as 2/T .
13 On the other hand, with no angular dependence, the orthogonality disappear, and the whole algebra contains
the time-dependent factor exp{2 i(m + n)(t r )/T } as well as other unwanted terms which are proportional to
(m2 n2 ); this is the same situation as in two-dimensional gravity [34].
358
7. Canonical quantization and black hole entropy

The computation of black hole entropy from the canonical quantization of the classical
Virasoro algebra is rather straightforward at this stage.
When one uses the standard canonical quantization rule
[Lm , Ln ] = i h {Jm , Jn },
Jm+n Lm+n ,
(7.1)
where Lm is a quantum operator, the classical Dirac bracket algebra (5.30) becomes an
operator algebra

h c 2
m m 1 m,n .
12
By considering the transformation
[Lm , Ln ] = h(m
n)Lm+n +
(7.2)
Lm h (:
Lm : + ha
m,0),
(7.3)
(7.2) becomes the standard operator Virasoro algebra for the normal ordered operator :
Lm :
(a is some number)

n :] = (m n):L
m+n : + ctot m m2 1 m,n
m :, :L
[:L
(7.4)
12
with
c
ctot = + cquant .
(7.5)
h
(See also Ref. [35] for some related discussions.) Here, cquant , which is O(1), is the
quantum effect due to operator reordering.
With the Virasoro algebra of :
Lm : in the standard form, which is defined on the plane,
one can use Cardys formula for the asymptotic states [1,14,36,37]

1
ctot

,
(ctot 24min)
log () 2
(7.6)
6
24
is the eigenvalue, called conformal weight, of :
where
L0 : for a black hole quantum state
min is its minimum value. When this is expressed in terms of the classical
|black hole and
eigenvalue J0 and the central charge c through
h a
(7.7)
h
from (7.3) and (7.5), one obtains

h cquant
2 1
c
2
2
log ()
+
24
h
a
c 24min + hc
h
quant
6
24
24
h

2
ceff eff /6 + O(h )
=
(7.8)
h
with
c
ceff = c 24min,
(7.9)
eff = .
24
=
359
This result shows explicitly how the classical Virasoro generator and central charge can
give the correct order of the semiclassical BH entropy (c = k = 1)
SBH
A+
4h G
(7.10)
since c A+ /G and A+ /G; details on the numerical factor 1/4 of the BH entropy
depends on min , which has to beput in by hand, and the quantum correction due
to reordering gives negligible O(1/ h ) effect to the entropy when one considers the
macroscopic ensemble of ceff eff 1.
Now, let me compute the black hole entropy explicitly. For rotating horizons, (7.8)
produces the entropy

2
A+ 2
A+ T
S = log ()
(7.11)
min
h
16G
8G
with
eff =
A+ T
.
32G
(7.12)
This gives the BH entropy (7.10) if one takes

min =
3A+
32G T
(7.13)
such as
3A+
.
(7.14)
G T
Note that min has an explicit /T dependence. But from the fact that this value is outside
of the classical spectrum of (16G)1 A+ , one can expect that the ground state of this
black hole may be described by another class of black holes; or this might be relevant to
the non-commutative spacetime near the ground state black hole which, probably, is very
light [38].
On the other hand, for non-rotating horizons, there is an additional factor 2 in (6.8)
and this has a remarkable consequence for the entropy computation: the entropy from the
Cardy formula gives

2
A+ T
A+ 2
S
(7.15)
min
h
8G
4G
ceff =
with
eff =
A+ T
.
16G
(7.16)
This gives the BH entropy (7.10) with a T -independent ground state

min = 0
(7.17)
360
such as
ceff =
3A+
.
2G T
(7.18)
This means that for an arbitrary choice of T and hence v the entropy is uniquely defined.
Moreover, the arbitrary T dependences in c and exactly cancel each other and one
obtains the T -independent correct entropy.14
The appropriate ground state is different for the rotating and the non-rotating black
holes, but otherwise it has a universality for a wide variety of other black holes: the fall-off
conditions described in Section 3 involve only weak assumptions about the black hole, and
once min is fixed, say for an isolated, uncharged Kerr black hole, its value is determined
also for a large number of other black holes carrying electric or magnetic charge (details
will appear elsewhere [41]).
8. Applications
I have shown that the statistical entropies of rotating and non-rotating stationary
(N r = 0) black holes through the Cardy formula have a universal form if the fall-off
conditions and several fall-off-preserving conditions are satisfied. So, the problem of the
entropy computation is reduced to verification of the conditions. My considerations are
general enough that almost all the known solutions satisfy the required conditions. In this
section I briefly discuss some of these. I shall adopt different units of G depending on the
usual conventions for the solutions in the literatures.
8.1. The Kerr and Schwarzschild black holes
The Kerr black hole solution, with one-rotation, in the 4 and higher dimensions is given,
in the standard form (3.1) [30,42] as

2
2
2
a
2 2
ds 2 = 2 dt 2 + 2 sin2 d n5 2 dt +
dr
+ 2 d 2 + r 2 cos2 d n4 ,
(8.1)
where
2 = r 2 + a 2 cos2 ,
= r 2 + a2

2 = 2 r 2 + a 2 + n5 a 2 sin2
r
,
r n5
(8.2)
14 This was claimed even in the rotating case by Carlip [6] due to the additionally introduced factor 2, which
is unclear in the present context. But his claim is exactly realized in the case of non-rotating case or in the ceff
and eff even when there is rotation. The connection to the usual T = relation in the path integral formulation
[39,40] needs further studies.
361
and n4 is the line element on a unit n 4 sphere; and a are the mass and the angular
momentum parameters, respectively. One can easily check that all the fall-off, fall-offpreserving, and differentiability conditions are satisfied in the n = 4 non-extremal Kerrsolution such that it has the universal statistical entropy (7.8) and gives the BH entropy
2 + a 2 ) = 8Mr , and
(7.10) with A+ = 4(r+
+
3 M 2 (M + M 2 a 2 )2
min =
(8.3)
.
T
M 2 a2
Here, M and a are the ADM mass and
the angular momentum per unit mass of the black
hole, respectively, and r = MG M 2 a 2 (G = 1). The higher-dimensional solutions
have exactly the same results if only the one-rotation solution (8.1) is concerned.
Moreover, since the Schwarzschild solution can be obtained as a non-rotating limit of
the Kerr solution, it is a trivial matter to check that the Schwarzschild solution also satisfies
all the fall-off and other related conditions. So, in this case also the universal entropy form
can be applied. Note that, the ground state of the non-rotating black hole is contained in
the full spectrum of a=0 and obtained as M 0 limit, which is nothing but the flat
spacetime: a=0 = ((8M 3 /T ) + (MT /8)) 0.
The lower ground state for the rotating black hole compared to the non-rotating one may
be understood qualitatively as follows: let me consider a black hole with a tiny angular
velocity, which is almost static. Then, increase the angular speed adiabatically slowly
such as no other physical properties of the black hole are changed. But in order that the
horizon is not naked by this adiabatic process, the mass of the black hole should also be
simultaneously increased by M 2 a 2 . In order to accommodate this increase in mass,
the vacuum becomes lower without changing the identity of the black hole. But a complete
understanding still needs to be discovered.
8.2. With a cosmological constant
My analysis can be also generalized to include the cosmological constant (CC) term
with only a small modification in the formulas of the preceding sections.
The CC term
1
d n x N h(2)
SCC =
(8.4)
16G
added to the action (2.2) changes the Hamiltonian constraints as

16G
h
1
t
(R 2) +
p2 .
H =
(8.5)
pab pab
16G
n2
h
But since this additional term does not generate any surface term in the variations, almost
all the results of the boundary conditions and the surface deformations in the preceding
sections are unchanged. The only exception is the computation of (5.15), which produces
the constraints term in the Virasoro algebra (5.25). But in this case again, there is no effect
of at the horizon. From
h t
t
t
= H=0 ,
H = H=0 +
8G
362
h rr
rr
h = p=0
,
(8.6)
16G
rr is again the constraint term,
one finds that the variations of 2 J [1 ] involving p
which is the same as (5.15).
However, if the CC generates its own horizon which is not due to the black hole, it
can be treated as another independent application of my original method. The interesting
examples are the BTZ solution in n = 3 ( < 0) and rotating de Sitter space solutions
( > 0).
rr
p
rr
= p=0
8.2.1. The BTZ black hole

The BTZ black hole solution [7] in n = 3 ( < 0) is similar to the Kerr solution.
But I consider this example since there are some points which are worthy of studying
in comparison with the other entropy computations in different contexts.
The BTZ solution is given by the standard form (3.1) with (G = 1/8)
f 2 = N 2 =
2 )(r 2 r 2 )
(r 2 r+
,
r2
N r = 0,
r+ r
(8.7)
,
N = r+ r ,
= r 2 .
r2
All the subsidiary conditions are satisfied the same way as in the Kerr-solution and so, the
universal entropy formula (7.8) is also applicable here with A+ = 2r+ and
N =
2
3l 2 r+
(8.8)
(J = 0),
min = 0 (J = 0).
2 r2
T r+

Here, r = (l/ 2) M M 2 (J / l)2 . The ground state of non-rotating solution
is obtained as M 0 : limM0 J =0 = (2)1 |M|T = 0. This is contrary to
Stromingers entropy computation, in which the Virasoro algebra is an algebra for an
asymptotic infinity, not for the horizon and the ground state gives M = 1. On the other
hand, the factor 2 in the non-rotating case (6.8) corresponds to the two copies of the
Virasoro algebra for the isometry group at spatial infinity [4,8,9].
min =
8.2.2. The rotating de Sitter space: n = 3 (Kerr-dS3 )

The de Sitter space of > 0 is peculiar in that it has its own horizon, called
cosmological horizon, without black holes [4347]. Moreover, remarkably this space can
have a rotating horizon with no lower bound of the mass, which is contrary to the Kerr
and the BTZ solutions. The n = 3 solution was studied first in Ref. [46] and more recently
studied in other contexts in Refs. [48,49]. The rotating de Sitter solution without black
holes exists also for n 4 and can be simply obtained from a zero-mass-black-hole limit
in the usual Kerrde-Sitter black hole solution. Moreover, the computation in my method
is interesting because the previous computations in the n = 3 case which have used the
isometry at spatial infinity, which is hidden inside the horizon, produce complex (and
imaginary c for the ChernSimons formulation) when rotations are involved [46,48,49]
always. But strangely enough, the final result of the entropy is real valued and identical
to the usual BH entropy [43]. So, it is important to check whether the complex number
363
disappears in all the intermediate steps or not when I treat the correct boundary r+ not the
suspicious boundary at infinity. In this subsection, I first consider n = 3 solution and then
consider n 4 in the next subsection.
The rotating de Sitter in n = 3, = l 2 > 0 is given by the standard metric (3.1) with
(G = 1/8)
2
r
J2
2
2
f =N =M
+ 2,
N r = 0,
l
4r
J
J
= r 2 ,
N = ,
N = 2 ,
(8.9)
2
2r
where M and J are the mass and angular momentum parameters
of the solution,

respectively. The solution has only one horizon at r+ = l/ 2 M + M 2 + (J / l)2 .
Notice that there is no constraint on M 2 bounded by J 2 for a horizon to exit: even a
negative value of M is allowed when J = 0 such as the mass spectrum (ranging from
to ) is continuous and so there is no mass gap in contrast to the BTZ solution; for J = 0
case, there is no horizon for M < 0.
The fall-off forms of the metric are the same as the BTZ solution and so the entropy
also has the universal from as (7.8) with A+ = 2r+ and
min =
3
T
(J = 0),
min = 0 (J = 0).
(8.10)
The static de Sitter ground state solution is obtained as M 0 : limM0 J =0 =

(2)1 T l 2 |M| = 0. Note that there is no complex number at any step of computation.
8.2.3. The rotating de Sitter solution: n 4 (massless Kerr-dSn )
The rotating de Sitter solution for n 4 is obtained from the M = 0 reduction of the
Kerrde-Sitter solution [43,50]. In the standard form (3.1) with (n = 4 for simplicity; but
it is a trivial matter to generalize to higher dimensions), the solution is given by (G = 1)
N2 =
r
,
2
(r + a 2 ) 3
N =
f2 =
1 a(r 2 + a 2 ) 2
sin ,
3
2
2
,
r
=
N r = 0,
N = 0,
(r 2 + a 2) 2
sin ,
N =
=
1 a
,
3
(8.11)
with
a 2
,
2 = r 2 + a 2 cos2 ,
=1+
3

r 2
a 2 2
r = r 2 + a 2 1
(8.12)
cos ,
,
= 1 +
3
3
which has the cosmological horizon at r+ = 3/.

This is very similar to the Kerr-solution and so one can easily check that all the
conditions of the Kerr-solution are still satisfied. So the universal entropy (7.8) can be
364
2 and
applied in this case, with A+ = 4r+
3
3r+
(a = 0),
min = 0 (a = 0).
(8.13)
4T
2 /2T ) +
The ground state of non-rotating
de Sitter space is obtained a=0 = r+ ((r+
(T /8)) 0 as r+ = 3/ 0, which corresponds to giving an infinite moment of

inertia to the space.
min =
I have shown that almost all the known solutions with the horizon have the universal
statistical entropy. This is identical to the BekensteinHawking entropy when the
appropriate ground states are chosen and the higher order quantum corrections of operator
orderings are neglected. Here the existence of the classical Virasoro algebra at the horizon
was crucial to the results. The remaining questions are as follows.
1. How can we understand extremal black holes in my method? Can this method explain
the discrepancy between the gravity side [51]15 and the string theory side [2], which
claim different entropies for the extremal black hole?
2. Can my method be generalized to non-stationary metrics such that expanding or
collapsing horizons can be treated [53]?
3. Can gauge and matter fields be introduced without perturbing the universal statistical
entropy formula? Can this study give another proof of the no-hair conjecture?
Some of the questions are being studied and will appear elsewhere [41].
Finally, computing the symplectic structure on the constraint surface H 0 with a
horizon boundary through the Dirac bracket method or the symplectic reduction will be an
outstanding challenge. The higher order quantum corrections of black hole entropy can be
computed by quantizing the classical symplectic structure.
Acknowledgements
I would like to thank Dimitra Karabali, Parameswaran Nair, Peter Orland, Alexios
Polychronakos, Bunji Sakita and Stuart Samuel for warm hospitality while I have been
staying at City College. I would like to thank Steve Carlip and Gungwon Kang for the
useful discussions in the very beginning of this work and Yun Soo Myung for inviting me
to Inje University, where some of the early ideas were initiated. I thank to Roman Jackiw
for introducing the interesting world of classical Virasoro algebra while I stayed at MIT
and continuous hospitality after then. This work was supported in part by postdoctoral
fellowships program from Korea Science and Engineering Foundation (KOSEF) and in
part by a CUNY Collaborative Incentive Research Grant.
15 There is a debate on the gravity side also. See Ref. [52].
365
Appendix A. Analysis on the boundary C

The variation in H [N, N a ] due to arbitrary variations in hab , na , N, N a is

H N, N a

16G r
1
n2
n Nr p
d
x
= bulk terms +
8G
2n h
C

a b r
16G
16G
r
Nr p
n n n Nr phab
+ n r N +
2n h
2(2 n) h

16G
16G
na p +
hra nr N a p .
+ N a 16Gpa r +
2n h
2n h
(A.1)
Here, I have also assumed the boundary conditions (2.17) and (2.18) in order to remove
the problematic terms which persist also on C as in (2.9).
The boundary contributions to the bulk equation of motion are

2
r
hab (x), H N, N a
boundary = (r rC )
(A.2)
n Nr hab ,
2n h

ab
p (x), H N, N a
boundary

2
nr Nr pab
= (r rC )
h 2n

(a b) r 1 a b r
2
1 (a b)r
r N n n n n Nr n h Nr p
+
2n h
2
2

(a b)r
n h r N .
(A.3)
16G
The boundary contribution to the momentum constraints is

r
2

n hra p + 2pr a .
H a boundary = (r rC )
2n h
(A.4)
In order to achieve the extremality under the variations with the boundary contributions,
I need the boundary conditions on C such that all the terms of hab , pab , N a vanish which
would be impossible in general. This can be easily checked within our black hole setup,
which shows non-vanishing boundary contributions, {pr (x), H [N, N a ]}|boundary O(1),
{p (x), H [N, N a ]}|boundary O(1), {hrr (x), H [N, N a ]}|boundary O(1), H r |boundary
O(N 2 ).
Finally, I note that there is a special choice of the slicing, called maximal slicing [28,
31,32], p = 0, i.e., p = 0 (but pr = O(1) needs not be zero), in which the Hamiltonian
H [N, N a ] is well defined, i.e., there is no boundary contribution, when the coordinate
t a
system Nr = 0 is used. But, even in this case, the Diff generator
1 H [ ,r ] is still ill
t
a
defined: {hrr (x), H [ , ]}|boundary = (r r+ )((2 n) h) 2 nr , hrr O(1)
[15].
366
Appendix B. Computing 2

r+
d n2 1 r 2 pr r
In this appendix, I compute 2

r+
d n2 1 r 2 pr r of (5.15). From (2.22), one has

pab = pab , H [ ]

h
h a b t
1
t
ab
ab
D D hab Dc D c t
=
R h R +
16G
2
16G

t
8G ab
32G t
1
1
cd
2
b ac
ab
p cp
p
p p
+ h
pcd p
n2
n2
h
h
c ab
a bc
b ac
+ Dc p Dc p Dc p .
(B.1)
In order to determine what is involved in the computation of (5.15), let me expand the
integrand of (5.15) as
1r 2 pr r = 1r f 2 2 prr + 1r 2 hr pr + 1r 2 hrr prr
(B.2)
by writing pr r = hbr pbr . [Here and after, I do not restrict to the case of N r = 0 for
generality, though I consider = 0 to avoid unnecessary complication.] The second and
the third terms are definitely O(N 2 ) and O(N), respectively. Naively, the first term is
O(N) if the fall-off condition (3.13) is preserved by Diff . But this is not a trivial matter.
Let me compute this first term in detail.
With the help of (B.1), the first term of (B.2) becomes
1r f 2 2 prr = A + B + C + D,

h
1 rr
2 r t
rr
R h R ,
A = f 1 2
16G
2

h
B = f 2 1r
D r D r 2t hrr Dc D c 2t ,
16G

8Gf 2 1r 2t rr
1
C=
p2
h pcd pcd
n2
h

t
2
r
32Gf 1 2 r rc
1
prr p ,
p cp
n2
h
c rr
r rc

2 r
2 r
D = f 1 Dc p f 1 Dc 2 p f 2 1r Dc 2r prc .
2
(B.3)
Term B reduces to
2
B = (16G)1f 3 1r hrr r 2t r O(N).

In order to compute the term A, one needs to compute the curvature tensor.

Rrr = r r
+ rrr r
+ O N 2 = O N 3 ,

r
r
R = r
+
rrr + O(1) = O N 1 ,

R = hrr Rrr + h R = O N 1 .
(B.4)
(B.5)
From this result, it is easy to see that

f 3 r t rr 2
1
A=
1 2 h
Rrr hrr R O(N).
16G
2
367
(B.6)
Term D reduces to

D = f 2 1r Dc 2c prr + 2f 2 1r D 2r pr .
(B.7)
The first term of this equation becomes

r r
f 2 1r ( h )1 r h 2r prr + f 2 1r 2r r prr + 2rrr prr + 2r
O(N).
p
(B.8)
On the other hand, the second term of (B.7) becomes

2f 2 1r hrr h D 2r pr

r r
a r
a r
= 2f 2 1r hrr h 2r pr + r
(B.9)
2 pr r
2 pa
2 pra .
Now, from the computation of each term inside the bracket [ ],

r r
2 pr O(1),
r
2r pr O(1),

a r
1
a r
,
r 2 pa O N
2 pra O N 1 ,
(B.10)
(B.9) becomes

2f 2 1r hrr h D 2r pr O(N).
(B.11)
This result is contrary to the naive expectation D such that this is O(N 3 ): this
a , makes O(N 3 ) = O(N).
implies that r , which is hidden in D through r
r
C is the most important term in (B.3).
The first term of C reduces to

8G1r 2t
1
p p
p p + O N 3 O(1).
(B.12)
n2
h
The second term reduces

32G1r 2t
1
pr r p + O N 2 O(N).
n2
h
(B.13)
(Notice that the n = 2 is meaningless in this formula, a separate consideration is required

for that case.)
Therefore, C becomes

8G1r 2t
1
p p + O(N) O(1).
C=
(B.14)
p p
n2
h
On the other hand, since the Hamiltonian constraint Ht becomes, near the horizon,

16G
1
p p + O N 2 O N 3
p p
Ht =
(B.15)
n2
h
368
from (B.5) and (B.12), C is nothing but

1r 2t
Ht + O(N) O(1).
2
In summary, one has

2 d n2 1 r 2 pr r = 2 d n2 (A + B + C + D)
C=
r+
(B.16)
r+
d n2 1r 2t Ht + O(N),
(B.17)
r+
A, B, D O(N).
(B.18)
Notice that I have not used any of the constraint equations Ht 0, Ha 0.
References
[1] S. Carlip, Phys. Rev. D 51 (1995) 632;
S. Carlip, Phys. Rev. D 55 (1997) 878.
[2] A. Strominger, C. Vafa, Phys. Lett. B 379 (1996) 99.
[3] A. Ashtekar, J. Baez, A. Corichi, K. Krasnov, Phys. Rev. Lett. 80 (1998) 904.
[4] A. Strominger, J. High Energy Phys. 02 (1998) 009.
[5] C. Teitelboim, Phys. Rev. D 53 (1996) 2870.
[6] S. Carlip, Phys. Rev. Lett. 82 (1999) 2828;
See also: S. Carlip, Talk at the ITP program on the Classical and Quantum Physics of Strong Gravitational
Fields, Santa Barbara, 1999, gr-qc/9912118.
[7] M. Baados, C. Teitelboim, J. Zanelli, Phys. Rev. Lett. 72 (1994) 957.
[8] J.D. Brown, M. Henneaux, Commun. Math. Phys. 104 (1986) 207.
[9] V. Balasubramanian, P. Kraus, Commun. Math. Phys. 208 (1999) 413;
H. Terashima, Phys. Lett. B 499 (2001) 229;
G. Barnich, F. Brandt, hep-th/0111246.
[10] C. Teitelboim, Ann. Phys. 79 (1973) 542.
[11] J. Guven, M.P. Ryan, Phys. Rev. D 45 (1992) 3559.
[12] J.D. Brown, G.L. Comer, E.A. Martinez, J. Melmed, B.E. Whiting, J.W. York Jr., Class. Quantum Grav. 7
(1990) 1433.
[13] S.N. Solodukhin, Phys. Lett. B 454 (1999) 213.
[14] S. Carlip, Class. Quantum Grav. 16 (1999) 3327;
J. Jing, M.-L. Yan, Phys. Rev. D 63 (2001) 024003.
[15] M.-I. Park, J. Ho, Phys. Rev. Lett. 83 (1999) 5595.
[16] S. Carlip, Phys. Rev. Lett. 83 (1999) 5596.
[17] O. Dreyer, A. Ghosh, J. Wisniewski, Class. Quantum Grav. 18 (2001) 1929;
J.-I. Koga, Phys. Rev. D 64 (2001) 124012.
[18] R. Arnowitt, S. Deser, C.W. Misner, in: L. Witten (Ed.), Gravitation, Wiley, New York, 1962;
B.S. Dewitt, Phys. Rev. 160 (1967) 1113;
R.M. Wald, General Relativity, Univ. Chicago, Chicago, 1984.
[19] J.W. York, Phys. Rev. Lett. 28 (1972) 1082;
J.W. York, Found. Phys. 16 (1986) 249.
[20] S.W. Hawking, C.J. Hunter, Class. Quantum Grav. 13 (1996) 2735.
[21] G.A. Burnett, R.M. Wald, Proc. R. Soc. London A 430 (1990) 57.
[22] J.D. Brown, E.A. Martinez, J.W. York, Phys. Rev. Lett. 66 (1991) 2281.
369
[23] V.O. Soloviev, Phys. Rev. D 61 (2000) 027502.

[24] R. Jackiw, in: S.B. Treiman, R. Jackiw, B. Zumino, E. Witten (Eds.), Current Algebra and Anomalies, World
Scientific, Singapore, 1985.
[25] P. Oh, M.-I. Park, Mod. Phys. Lett. A 14 (1998) 231.
[26] M.-I. Park, Nucl. Phys. B 544 (1999) 377.
[27] P. Oh, Phys. Lett. B 464 (1999) 19.
[28] T. Regge, C. Teitelboim, Ann. Phys. (N.Y.) 88 (1974) 286;
R. Benguria, P. Cordero, C. Teitelboim, Nucl. Phys. B 122 (1977) 61.
[29] For example, see: T. Lee, hep-th/9806113.
[30] R.C. Myers, M.J. Perry, Ann. Phys. (N.Y.) 172 (1986) 304.
[31] P.A.M. Dirac, Phys. Rev. 114 (1959) 924.
[32] R. Arnowitt, S. Deser, C.W. Misner, J. Math. Phys. 1 (1960) 434.
[33] G. t Hooft, gr-qc/9310026;
L. Susskind, J. Math. Phys. 36 (1995) 6377.
[34] M.-I. Park, J.H. Yee, Phys. Rev. D 61 (2000) 088501;
M. Cadoni, S. Mignemi, Phys. Rev. D 59 (1999) 081501.
[35] S. Carlip, Class. Quantum Grav. 17 (2000) 4175.
[36] J.A. Cardy, Nucl. Phys. B 270 (1986) 186.
[37] S. Carlip, Class. Quantum Grav. 15 (1998) 3609.
[38] M.-I. Park, hep-th/0201002.
[39] G.W. Gibbons, S.W. Hawking, Phys. Rev. D 15 (1977) 2752.
[40] J.D. Brown, J.W. York, Phys. Rev. D 47 (1993) 1420;
J.D. Brown, Phys. Rev. D 52 (1995) 7011.
[41] M.-I. Park, in preparation.
[42] S. Chandrasekhar, The Mathematical Theory of Black Holes, Oxford Univ. Press, New York, 1983.
[43] G.W. Gibbons, S.W. Hawking, Phys. Rev. D 15 (1977) 2738.
[44] S. Deser, R. Jackiw, Ann. Phys. 153 (1984) 405.
[45] J. Maldacena, A. Strominger, JHEP 9802 (1998) 014;
M. Baados, T. Brotz, M.E. Ortiz, Phys. Rev. D 59 (1999) 046002;
W.T. Kim, Phys. Rev. D 59 (1999) 047503;
F.-L. Lin, Y.-S. Wu, Phys. Lett. B 453 (1999) 222.
[46] M.-I. Park, Phys. Rev. B 440 (1998) 275.
[47] A. Strominger, JHEP 0110 (2001) 034;
M. Spradin, A. Strominger, A. Volovich, hep-th/0110007.
[48] V. Balasubramanian, J. de Boer, D. Minic, hep-th/0110108.
[49] B. Carneiro da Cunha, hep-th/0110169.
[50] S.W. Hawking, C.J. Hunter, M.M. Taylor Robinson, Phys. Rev. D 59 (1999) 064005.
[51] S.W. Hawking, G.T. Horowitz, S.F. Ross, Phys. Rev. 51 (1995) 4302;
C. Teitelboim, Phys. Rev. D 51 (1995) 4315;
G.W. Gibbons, R.E. Kallosh, Phys. Rev. D 51 (1995) 2839.
[52] O.B. Zaslavskii, Phys. Rev. Lett. 76 (1996) 2211.
[53] R. Brunstein, Phys. Rev. Lett. 86 (2001) 576;
E. Verlinde, hep-th/0008140;
A. Strominger, hep-th/0110087.

Canonical quantization of gravity

without frozen formalism
Giovanni Montani
ICRAInternational Center for Relativistic Astrophysics, Dipartimento di Fisica (G9),
Universit di Roma, La Sapienza, Piazzale Aldo Moro 5, 00185 Rome, Italy
Abstract
We write down a quantum gravity equation which generalizes the WheelerDeWitt one in view
of including a time dependence in the wave functional. The obtained equation provides a consistent
canonical quantization of the 3-geometries resulting from a gauge-fixing (3 + 1)-slicing of the
spacetime. Our leading idea relies on a criticism to the possibility that, in a quantum spacetime,
the notion of a (3 + 1)-slicing formalism (underlying the WheelerDeWitt approach) has yet a precise
physical meaning. As solution to this problem we propose of adding to the gravity-matter action the
so-called kinematical action (indeed in its reduced form, as implemented in the quantum regime),
and then we impose the new quantum constraints. As consequence of this revised approach, the
quantization procedure of the 3-geometries takes place in a fixed reference frame and the wave
functional acquires a time evolution along a one-parameter family of spatial hypersurfaces filling the
spacetime. We show how the states of the new quantum dynamics can be arranged into an Hilbert
space, whose associated inner product induces a conserved probability notion for the 3-geometries.
Finally, since the constraints we quantize violate the classical symmetries (i.e., the vanishing nature
of the super-Hamiltonian), then a key result is to find a (non-physical) restriction on the initial wave
functional phase, ensuring that general relativity outcomes when taking the appropriate classical
limit. However, we propose a physical interpretation of the kinematical variables which, based on
the analogy with the so-called Gaussian reference fluid, makes allowance even for such classical
symmetry violation. 2002 Elsevier Science B.V. All rights reserved.
PACS: 83.C
E-mail address: montani@icra.it (G. Montani).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 0 1 - 2
G. Montani / Nuclear Physics B 634 (2002) 370392
371
1. General statements
It is commonly believed that the canonical methods of quantization, in spite of their
successful predictions on many elementary particles phenomena (it stands up the great
agreement of quantum electrodynamics with experimental data), nevertheless they cannot
be extrapolated up to arbitrarily high energies (i.e., arbitrarily small distances), when
the granular nature of the fields and their interaction with the underlying space
time are expected to be important. Indeed this point of view is confirmed by the large
number of renormalization procedure are required to get satisfactory predictions already
on Minkowski space and moreover by the inconsistencies or ambiguities characterizing the
canonical quantization of matter field when referred to a curved spacetime [1].
The implementation of the canonical formalism to the gravitational field quantization,
leads to the so-called WheelerDeWitt equation (WDE) [2,3], consisting of a functional
approach in which the states of the theory are represented by wave functionals taken on the
3-geometries and, in view of the requirement of general covariance, they do not possess
any real time dependence. Due to its hyperbolic nature, the WDE is characterized by a large
number of unsatisfactory features [4] which strongly support the idea that is impossible any
straightforward extension to the gravitational phenomena of procedures well-tested only in
limited ranges of energies; however, in some contexts, like the very early cosmology [5,6]
(where a suitable internal time variable is provided by the universe volume) the WDE is not
a dummy theory and give interesting information about the origin of our classical universe,
see [7], which may be expected to remain qualitatively valid even for the outcoming of
a more consistent approach.
Over the last ten years the canonical quantum gravity found its best improvement in
a reformulation of the constraints problem in terms of the so-called Ashtekar variables,
leading to the loop quantum gravity theory [8,9]; this more recent approach overcomes
some of the WDE limits, like the problem of constructing an appropriate Hilbert space, but
under many aspects is yet a theory in progress.
The aim of this paper, more than to be the answer to the hard question whether canonical
quantum gravity has some predictive issue, concerns a criticism on a fundamental ansatz
on which is based the whole formalism: the ansatz we criticize is that, in a fully covariant
quantum theory has yet meaning to speak of an ArnowittDeserMisner (ADM) formalism
[3,10] to perform a (3 + 1)-slicing of the spacetime (as implicitly assumed in the WDE
approach). Indeed, the space or time character of a vector, in particular of the normal to
a 3-hypersurface, is sensitive to the metric field and cannot be definite precisely when the
gravitational field is in a quantum state (unless we consider simply a perturbation theory).
By other words we claim that the quantization procedure does not commutes with the
spacetime slicing operation; therefore, only two approaches appear to be acceptable and
self-consistent (see also the beginning of Section 4):
(i) The requirement of a covariant quantization of gravity is preserved, but any slicing of
the spacetime prevented.
(ii) The slicing representation of the spacetime is allowed, but any notion of full
covariance given up.
372
The first point of view was addressed in [11] and find a promising development in
the recent issue of the so-called spin foam formalism [12]. The analysis here presented
addresses the second point of view as leading statement and by using the discussion of the
canonical methods of quantization presented in [13], as a fundamental paradigm, we get a
reformulation of the WDE which overcomes many of its shortcomings.
In [13] is provided a wide discussion about the canonical methods of quantization as
referred to different systems (with a satisfactory guide to the previous literature on this
subject) in order to construct a quantum geometrodynamics on the base of its similarities
and differences with other theories (for a valuable and more recent review on the canonical
quantum gravity and the related problem of a physical time see [14]). Of particular interest
is to be regarded for our purposes the analysis concerning the quantum theory of matter
field on a fixed background, which shows the necessity of adding the so-called kinematical
action to achieve a satisfactory structure for the quantum constraints.
Our main statement is that the kinematical term may be retained even in the gravitational
case since in the ADM action the lapse function and the shift vector can be thought
as assigned (like in the matter field case) up to a restriction on the initial Cauchy
problem, i.e., the super-Hamiltonian and super-momentum constraints have to be satisfied
on the initial space-like hypersurface; indeed to fix the reference frame, i.e., the lapse
function and the shift vector, is equivalent to loss the super-Hamiltonian and the supermomentum constraints which (equivalent to the (0 )-components of the Einstein
equations), however, have an evolutive character, i.e., if satisfied on the initial hypersurface,
they are preserved by the dynamical evolution [13]. Thus from a classical point of view,
we add the kinematical action to the gravity-matter one, but we preserve the general
relativity framework simply by assigning a particular Cauchy problem. When passing to
the quantization procedure we get equations describing the evolution of a wave functional
no longer invariant under time displacements (the invariance under 3-diffeomorphisms is
yet retained), but therefore characterized by a dependence on the choice of the spatial
hypersurface, i.e., the time variable.
By taking the evolution of the wave functional along a one-parameter family of spatial
hypersurfaces, filling the spacetime, we show how the space of the solutions for the wave
equation can be turned into an Hilbert space and the quantum dynamics can be reduced
to a Schrdinger-like approach with an associated eigenvalue problem; it is important to
stress that the quantum evolution of the 3-geometries is, at the end, expressed directly in
terms of the parameter labeling the hypersurfaces (the so-called label time).
As last but fundamental achievement we find the quantum correspondence of the
restriction imposed classically on the Cauchy data to provide general relativity; indeed
we show that if the phase of the initial wave functional satisfies the usual HamiltonJacobi
equation, then the classical limit h 0 always reproduces the Einsteinian dynamics. This
feature of the model is crucial for its viability because ensures that, though violated on
quantum level, the right symmetries (i.e., the vanishing nature of the super-Hamiltonian)
are restored on the classical limit.
Since the approach here presented overlap in the formalism the so-called multi-time
formulation of canonical quantum gravity, as well as its smeared Schrdinger version [14,
15], it is worth stressing differences and similarities. Other important approaches based on
373
the same (so-called) embedding variables, and even referred to the path integral formalism,
can be found in [1618] (see also [19]).
In this well-known quantization scheme, the kinematical variables are identified with
non-physical degrees of freedom of the gravitational field, extracted by an ADM resolution
of the Hamiltonian constraints. Thus, unlike for our proposal, where the kinematical
variables are added by hand (as for any other field), in the multi-time approach, no
violation takes place of the classical symmetries. The main similarity is indeed only
the form of the quantum equations because the physical and dynamical meaning is
made significantly different in view of the different Hamiltonian densities (or smeared
Hamiltonians) respectively involved: in the present approach we have to do with the superHamiltonian, while in the multi-time formalism with the ADM reduced one.
More physical insight is obtained on the nature of the kinematical variables by making
a parallel between our model and the so-called Gaussian reference fluid [14,20]; we argue
that the formalism below developed corresponds to a generalization of the Gaussian case
to a generic reference fluid which can play the role of a physical clock. The physical
characterization of this fluid is provided by analysing the Hamiltonian equations involving
the kinematical variables; as a result we show that to this generic fluid can be associated
the energymomentum tensor of a dust.
In Section 2 we give a review of the quantum theory of the matter field in the spirit
of the discussion presented in [13], which constitutes the appropriate line of thinking for
the reformulation of the quantum geometrodynamics developed in Section 4 and following
a schematic derivation of the WDE approach presented in Section 3. By Section 5 we
compare our proposal with the Schrdinger multi-time formalism, devoting particular
attention to a minisuperspace model. In Section 6 we give a discussion on the physical
nature of our time variable, based on a generalization of the Gaussian fluid clock. Finally
Section 7 is devoted to brief concluding remarks and provides an application of the
proposed quantum gravity theory to a very simple model.
2. Quantum fields on curved background

We start by a brief review of the canonical quantization of a matter field on an assigned
spacetime background [1,13]. Let us consider a four-dimensional pseudo-Riemannian
manifold M4 , characterized by a metric tensor g (y ) (, , = 0, 1, 2, 3) (having
signature , +, . . . , +, +) and perform a (3 + 1)-slicing of the spacetime [3,10], by a
one-parameter family of spacelike hypersurfaces t3 : y = y (t, x i ) (i = 1, 2, 3), each of
which defined by a unique fixed value of t. The normal field to the family of hypersurfaces
n (y ) (n n = 1) and the three tangent vectors to this family ei i y (we adopted

the notation i ( ) ( )/x i ) form a reference basis in M4 .
Thus we may decompose the so-called deformation vector N t y (t ( ) ( )/t)
along the basis {n, ei }, i.e.,
N t y = Nn + N i ei ,
(1)
where N(t, x i ) and N i (t, x i ) are, respectively, called the lapse function and the shift vector.
By performing on the metric tensor the coordinates transformation y = y (t, x i ) and
374
using (1), the line element rewrites

ds 2 = g dy dy = N 2 dt 2 + hij dx i + N i dt dx j + N j dt ,
(2)
where hij g ei ej is the metric tensor induced on the hypersurfaces t3 . By the line
element (2) we get the expression of the contravariant normal vector in the new coordinates
(t, x i ), i.e., n = (1/N, N i /N); by the inverse components of the metric in these same
coordinates, i.e., g {1/N 2 N i /N 2 hij N i N j /N 2 }, we get the covariant vector field
n = (N, 0). These contravariant and covariant expressions for the normal field imply,
respectively, n ei = N i /N and n ei = 0.
Now, within the ADM formalism, we analyze the quantization of a self-interacting
scalar field (t, x i ) described by a potential term V () on a fixed gravitational background;
its dynamics is summarized by the action
S ( , ) =
3
t NH N i Hi
d x dt,
(3)
M4
where denotes the conjugate field to the scalar one and . the Hamiltonian terms H
and Hi read explicitly
1
1 ij
h h i j + h V (),
H 2 +
2
2 h
Hi i
(4)
being h det hij . This action should be varied with respect to and , but not N , N i and
hij since the metric background is assigned. However, it remains to specify the geometrical
meaning of the lapse function and the shift vector with respect to the spacetime slicing;
this aim is reached by adding to S the so-called kinematical action [13]

S p , y =

p t y N p d 3 x dt
(5)
M4
so getting, by (1) the full action for the system

S k S + S k

=
t + p t y N H + H k N i Hi + Hik d 3 x dt,
(6)
M4
H p n ,
k
Hik p ei .
(7)
In the above action n and hij are to be regarded as assigned functionals of y (t, x i ).
Adding the kinematical action does not affect the field equation for the scalar field,
while the variations with respect to p and y provide Eq. (1) and the evolution of the
kinematical momentum.
375
Finally, by varying, now even, with respect to N and N i we get the Hamiltonian
constraints
H = p n ,
Hi = p ei .
(8)
Clearly is to be assigned the following Cauchy problem assigned on a regular initial
hypersurface t30 , i.e., y (t0 , x i ) = y0 (x i )

t0 , x i = 0 x i ,
t0 , x i = 0 x i ,

y t0 , x i = y0 x i ,
(9)
p t0 , x i = p 0 x i .
At last, to complete the scheme of the field equations, we have also to specify the
lapse function, the shift vector and the metric tensor hij = hij (y ). This system can
be easily quantized in the canonical formalism by assuming the states of the system
be represented by a wave functional (y (x i ), (x i )) and implementing the canonical
= i h ( )/}. Then
variables {y , p , , } to operators {y , p = i h ( )/y , ,
the quantum dynamics is described by the equations

1 ij
2
= h
+
h h i j + h V () ,
i h n = H
y
2 h 2
= i h i .
i h ei = H
(10)
i
y
These equations have 5 3 degrees of freedom, corresponding to the values taken by

the four components of y and the scalar field in each point of a spatial hypersurface.
In (10) y plays the role of time variable, since it specifies the choice of a particular
hypersurface y = y (x i ).
In view of their parabolic nature, Eq. (10) have a space of solutions that, by an heuristic
procedure, can be turned into an Hilbert space whose inner product reads

1 | 2
1 2 D
= 0,
1 | 2
(11)
y
y =y (x i )
where 1 and 2 denote two generic solutions and D the Lebesgue measure defined on
the -function space. The above inner product induces the conserved functional probability
distribution ! | .
The semiclassical limit of this Eqs. (10) is obtained when taking h 0 and, by setting
the wave functional as

1
= exp i y ,
(12)
h
and then expanding in powers of h/i,
i.e.,
2
h
h
2 + .
= 0 + 1 +
(13)
i
i
By substituting (12) and (13) in Eqs. (10), up to the zero-order approximation, we find
the HamiltonJacobi equations
376

0 2 1 ij
1
=
+ h
h i j + V () ,
2
2 h
0
0
= i
ei
y
0
n
y
(14)
which lead to the identification 0 S k .
3. The WheelerDeWitt equation

Now we briefly recall how the WheelerDeWitt approach [2,13] faces the problem of
quantizing a coupled system consisting of gravity and a real scalar field, which implies also
the metric field now be a dynamical variable. The action describing this coupled system
reads

g

g
ij t hij + t N H g + H N i Hi + Hi
d 3 x dt,
S =
(15)
M4
where ij denotes the conjugate momenta to the three-dimensional metric tensor hij and
g
the gravitational super-Hamiltonian H g and supermomentum Hi takes the explicit form
16G
c3 3
ij kl
G
h R,
ij
kl
c3
16G
1
(hik hj l + hil hj k hij hk ),
2 h
Hg
Gij kl
g
Hi 2 3 j i .
(16)
In the above expressions 3R and 3 ( ) denote, respectively, the Ricci scalar and the
covariant derivative constructed by the 3-metric hij , while G is the Newton constant.
Since now N and N i are, in principle, dynamical variables, they have to be varied,
g
so leading to the constraints H g + H = 0 and Hi + Hi = 0 which are equivalent to the

(0)-components of the Einstein equations and, therefore, play the role of constraints for
the Cauchy data. It is just this restriction on the initial values problem, a peculiar difference
between the previous case, at fixed background, and the present one; in fact, now, on
the regular hypersurface t = t0 , the initial conditions {0 (x i ), 0 (x i ), hij 0 (x i ), ij 0 (x i )}
cannot be assigned freely, but they must verify on t30 the four relations {H g + H }t0 =
g
{Hi + Hi }t0 = 0.
Indeed behaving like Lagrange multipliers, the lapse function and the shift vector have
not a real dynamics and their specification corresponds to assign a particular slicing of
M4 , i.e., a system of reference.
In order to quantize this system we assume that its states be represented by a wave
functional ({hij }, ) (the notation {hij } means all the 3-geometries connected by
a 3-diffeomorphism) and implement the canonical variables to operators acting on this
wave functional (in particular, we set hij h ij , ij ij i h ( )/hij ). The quantum
dynamics of the system is then induced by imposing the operatorial translation of the
377
classical constraints, which leads, respectively, to the WheelerDeWitt equation and to

the supermomentum constraint one [2]:

g

g
+H
= 0,
= 0,
+H
H
H
(17)
i
i
which to be explicited requires a specific choice for the normal ordering of the operators.
Due to its hyperbolic nature this formulation of the quantum dynamics has some limiting
feature [4], which we summarize by the following four points:
(i) It does not exist any general procedure allowing to turn the space of the solutions
into an Hilbert one and so any appropriate general notion of functional probability
distribution is prevented.
(ii) The WDE does not contain any dependence on the variable t or on the function y ,
so loosing its evolutive character along the slicing t3 . Moreover individualizing an
internal variable which can play the role of time is an ambiguous procedure which
does not lead to a general prescription.
(iii) The natural semiclassical limit, provided by a wave functional of the form

({hij } )
{hij }, = exp i
(18)
h
leads, up to the zero order of approximation 0 ({hij }, ), to the right
HamiltonJacobi equation for the classical scalar tensor action S g , but, in the limit
corresponding to a fixed background N = N , N i = N i , hij = hij , the WDE does
not provide neither the HamiltonJacobi equation (14) expected associated to the
semiclassical limit ({hij }, ) = exp i{(0 ())/h } (h 0), neither the appropriate
quantum dynamics on a fixed background (10). In correspondence to the wave
functional = ({hij }, ) (), which instead provides
= 0,
H
= 0.
H
i
(19)
This discrepancy is due to the absence of a dependence on y in the action S g , which

cannot be clearly restored anyway.
(iv) At last we stress what is to be regarded as an intrinsic inconsistency of the
approach above presented: the WDE is based on the primitive notion of space-like
hypersurfaces, i.e., of a time-like normal field, which is in clear contradiction with the
random behavior of a quantum metric field [21]; indeed the space or time character of
a vector becomes a precise notions only in the limit of a perturbative quantum gravity
theory. This remarkable ambiguity leads us to infer that there is inconsistency between
the requirement of a wave equation (i.e., a wave functional) invariant, like the WDE
one, under space diffeomorphisms and time displacements on one hand, and, on the
other one, the (3 + 1)-slicing representation of the global manifold.
The existence of these shortcomings in the WDE approach, induces us to search for
a better reformulation of the quantization procedure which addresses the solution of
the above indicated four points as prescriptions to write down new dynamical quantum
constraints.
378
4. Reformulation of the quantum geometrodynamics

We start by observing that, within the framework of a functional approach, a covariant
quantization of the 4-metric field is equivalent to take the wave amplitude =
(g (x )); in the WDE approach, by adopting the ADM slicing of the spacetime, the
problem is restated in terms of the following replacement

g x N t, x l , N i t, x l , hij t, x l .
(20)
Then, since the lapse function N and the shift vector N i are cyclic variables, i.e., their
conjugate momenta pN and pN i vanish identically, we get, on a quantum level, the
following restrictions:
pN = 0,
pN i = 0
= 0,
N
= 0,
N i
(21)
by other words, the wave functional should be independent of N and N i . Finally,

the super-momentum constraint leads to the dependence of on the 3-geometries {hij }
(instead on a single 3-metric tensor hij ). The criticism to the WDE approach, developed
at the point (iv) of the previous section, concerns with the ill-defined nature of the
replacement (20). The content of this section is entirely devoted to reformulate the
quantum geometrodynamics, by preserving the (3 + 1)-representation of the spacetime,
but avoiding the ambiguity above outlined in the WDE approach.
g
Since, as well known [13], the Hamiltonian constraints H g +H = 0 and Hi +Hi = 0,

once satisfied on the initial hypersurface, are preserved by the remaining Hamilton
equations, then the above variational principle (15) would be equivalent to a new one
with assigned lapse function and shift vector (so loosing the Hamiltonian constraint), but
with the specification of a Cauchy problem satisfying these constraints. Therefore, we face
the question of introducing an external temporal dependence in the quantum dynamics
of gravity, by addressing, in close analogy to the approach discussed in Section 2, the
kinematical action (5) as till present even in the geometrodynamics framework. Therefore,
we adopt already on a classical level the full action

ij t hij + t + p t y N H g + H + p n
S gk =
M4

g

N i Hi + Hi + p ei d 3 x dt,
(22)
where n is to be regarded as a functional of y (t, x i ), but, differently from the case of a

fixed background, now hij (like ) are dynamical variables taken on t and x i . The variation
of this action with respect to the variables hjj ij provides the standard Einstein-scalar
dynamics, the variation with respect to y and p gives rise, respectively, to the relations
(1) and t p = p Nn /y + i (N i p ) and finally we have to vary with respect to N
and N i , so getting the constraints
H g + H = p n ,
Hi + Hi = p ei .
(23)
379
However, we observe that the equation t p = p Nn /y + i (N i p ) is a linear and

homogeneous first order differential system which ensures that if p is chosen to be zero
on the initial hypersurface t30 then it will remain zero during the whole evolution1 and
the considered theory reduces to ordinary general relativity. Thus we see how, as argued
above, the difference in addressing the presence of the kinematical action, respectively, to
the canonical case relies only on a different structure of the Cauchy problem.
The key point in our reformulation of the canonical quantization of gravity consists of
assuming (23) be the appropriate quantum constraints and individualizing the restriction to
be imposed on the initial wave functional to get p 0 when constructing its semiclassical
expansion.
In agreement with these considerations we require that the quantum states of the
system be represented by a wave functional = (y , hij , ) which, once performed
the canonical variables translation into the corresponding operators, satisfy the superHamiltonian and supermomentum constraints
i h n

g
+H
,
= H
i
i h ei

g
+H
.
= H
i
i
(24)
The second of the above equations exprimes the non-invariance of the wave functional
under space diffeomorphisms (i.e., x i = x i (x i )) and therefore, since we expect the
quantization procedure, even in a (3 + 1)-representation of spacetime, should yet preserve
the tensorial nature of the 3-geometry, then it is natural to require the left-hand side of this
equation vanish (on a classical level it corresponds to the restriction p n p ei =

0); thus we take, as describing the quantum geometrodynamics, the following system of
functional differential equations:
i h n

g
+H
,
= H
i

g + H
= 0,
H
i
i

= {hij }, y ,
(25)
where now the wave functional is taken again on the 3-geometries ({hij }) related by the
3-diffeomorphisms. These (4 3 ) equations, which correspond to a natural extension of
the WheelerDeWitt approach, have the fundamental feature that the first of them is now
a parabolic one and it is just this their property which allows to overcome some of the
above discussed limits of the WDE.
Though this set of equations provides a satisfactory description of the 3-geometries
quantum dynamics, nevertheless it turns out convenient and physically meaningful to take,
by (1), the wave functional evolution along a one-parameter family of spatial hypersurfaces
filling the universe.
By (1) the firsts of Eqs. (25) can be rewritten as follow (taking into account even the
second one of (25)):
i h

g
+H
.
t y = N H
(26)
1 We remark that if p vanishes on the initial hypersurface, then this equation implies that all its initial time
derivatives vanish.
380
Now this set of equations can be (heuristically) rewritten as a single one by integrating over
the hypersurfaces t3 , i.e.,

d 3 x .
g + H
i h t = i h
(27)
t y d 3 x = H
N H
y
t3
t3
The above equations (27) and (25) show how in the present approach the wave functional
is no longer invariant under infinitesimal displacements of the time variable.
is an Hermitian one, i.e., in the bracket Dirac
Let us now show that the operator H
notation, it verifies the relations
2 = H
1 | 2
1 | H
(28)
being 1 and 2 two generic solutions of Eqs. (27) and (25).

g the following normal ordering for its kinetic part
We start by choosing in H

( )
Gij kl ij kl h 2
Gij kl
.
hij
hkl
Hence we have
2
1 | H

DhD
Ft
g

+H
2 ,
d 3 x1 N H
(29)
(30)
t3
where Ft denotes the functional space ({hij }, ), as referred to the hypersurface t3 ,

DhD the associated Lebesgue measure and the complex conjugate wave functional,
which satisfies the Hermitian conjugate equation of (27) and (25) ones.
By observing that, as shown in [13] and easily checkable by repeating the analysis below
is an Hermitian operator, then it remains to be
performed for the gravitational terms, H
analyzed the term (the functional integrals on and h commute with each other and both
commute with the space integral):2

33

2
h
c
R
16G
h
2
2 .
Dh
1
(31)
Gij kl
c3
hij
hkl
16G
Ft
t3
Since Gij kl = Gklij , it is easy to check the relation

2
3

d x 1
Gij kl
hij
hkl
t3

=

1
1
2
d x
2 Gij kl
1 Gij kl
+
Gij kl
2
hij
hkl
hkl
hij
hkl
3
(32)
t3
2 The analysis of this section and of the following one has an heuristic character due to its functional approach
and it should be made rigorous by an appropriate discretization on a lattice.
381
and then, assuming (as done for the matter field [13]) the validity in the functional space
({hij }) of the functional Gauss theorem (the wave functional is required to vanish on the
boundary of this space):

(33)
Dh d 3 x
( ) = 0
hij
Ft
t3
is an Hermitian operator.
we conclude the proof that H
On the base of this results, by (27) we get
t 1 | 2 d 3 x t y 1 | 2
y
t3
= t 1 | 2 + 1 | t 2 =

i
2 = 0
H1 | 2 1 | H
h
(34)
and the generic character of the deformation vector allows us to write the fundamental
conservation law
1 | 2
= 0.
y
(35)
Thus we defined an inner product which turns the space of the solutions to Eqs. (27)
and (25) into an Hilbert space (so removing the shortcoming (i) of the WDE). Indeed now
it is possible to define the notion of a conserved functional probability distribution as:

! y , {hij }, ,
| = 1,
|
= 0.
y
(36)
Let us now reformulate the dynamics described by Eqs. (27) and (25) by means of the
eigenvalues problems for Eqs. (25). To this end we expand in the following functional
representation:

y , {hij },

i
=
(37)
D() {hij }, exp
d 3 x dy (n ) ,
h
Y
t3
where D denotes the Lebesgue measure in the functional space yt of the conjugate
function (x i ), a functional valued in this domain and n denotes the covariant normal
vector; indeed, once assigned n (y ), the field n can be written, in general, only formally
in a quantum spacetime.
Substituting this expansion into Eqs. (25) we get the eigenvalues problems

g
g
= ,
= 0.
+H
+H
(38)
H
H
i
i
g or H
,
Here (x i ) is not a 3-scalar, but it transforms, under 3-diffeomorphisms, like H
3
so ensuring that d x, as it should, be an invariant quantity.
382
Now we observe that, by (1), Eq. (37) rewrites

t

i
3

D() {hij }, exp
d x
dt t y (n ) ,
h
y , {hij }, =

{hij }, , t =

i
D() {hij }, exp
h
t
t0
t3
t
t0
dt

d 3 x (N)
(39)
(40)
t3
being t0 an assigned initial instant. To the same result we could arrive by choosing,
without any loss of generality, the coordinates system (t, x i ), i.e., y 0 t, y i x i ; indeed,
for this system, the spatial hypersurfaces have equation t = const, i.e., dy (dt, 0, 0, 0)
and we have n0 = N . By other words the wave functional (40) is to be interpreted directly
in terms of the time variable t, i.e., ({hij }, , t) and, in fact, it turns out solution of the
wave equation

{hij }, , t .
i h t {hij }, , t = H
(41)
The expansion (40) of the wave functional and the eigenvalues problems (38)
completely describe the quantum dynamics of the 3-geometries.
Now to conclude our analysis, it needs to recognize which restriction should be imposed
on the initial wave functional, say ({hij }, }, t0 ) = 0 ({hij }, ) (t = t0 defining the initial
hypersurface) to get in the classical limit the ordinary general relativity, i.e., the quantum
counterpart of the condition p0 = 0.
an Hermitian operator the same
To this aim we preliminary observe that, being H
g

remains valid for H (by identical proof) and the functionals are expected to be an
orthonormal basis, i.e., | = ( ) ( denoting the Dirac functional), on which
we can expand 0 ; in fact we get the functional relation () = 0 | .
As next step we express the wave functional as follows

exp i
(42)
h
being and , respectively, the modulus and the real phase (up to h)
of . By substituting
(42) into Eq. (41) we get for its real part an expression of the form:

N g
J +
H +H
d 3 x,
t =
(43)
NH
t3
J we denote the HamiltonJacobi operator; we stress how, in the above

where by H
equation, the terms relative to the functional contain, differently from the Hamilton
Jacobi one, h , as well as the imaginary part of (41) (which is an evolutive equation for ).
Now we observe that once assigned 0 ({hij }, , t0 ), Eq. (43) allows to calculate on
t30 (t ) |t =t0 and, by iteration, all higher order time derivatives; thus we can expand in
383
powers of t near enough to the initial hypersurface, i.e.:

=

1
(t )n t =t (t t0 )n .
0
n!
(44)
n=0
This expression permits to extend the solution toward the future by reassigning the Cauchy
data in t0 + ,t, ,t 1, and then iterating the procedure indefinitely. But if we require
that 0 satisfies the restriction
J 0 = 0
H
(45)
than we get
t |t =t0 = terms O(h ) + = 0 + terms O(h )(t t0 ) + .
(46)
This result ensures in the semiclassical limit h 0 the evolutive nature of the wave
functional phase, i.e., = 0 (the same is not true for the modulus which remains
an evolutive variable even in this limit). Thus an initial wave functional whose modulus
is a generic one, but whose phase satisfies the HamiltonJacobi equation (45) provides
a quantum evolution compatible with the classical limit of general relativity. However, we
have to note that to retain the invariance under the 3-diffeomorphisms of the quantization
procedure prevents, in general, the achievement of a correct limit for the quantum field
theory on a fixed background; indeed (25) provides, on a fixed background, the right
dynamics (10) only in those reference frames for which N i = 0.
We conclude our reformulation of the quantum geometrodynamics, by emphasizing
how, a restriction on the initial wave functional phase 0 (as the one required to get the
classical limit of general relativity), does not correspond to a real loss of physical degrees
of freedom on t30 , since the only meaningful information we can provide on the initial
quantum configuration of the system, consists of the functional probability distribution 0
(which, to get the classical limit, should be peaked around a specific solution of the Einstein
equations).
5. The multi-time and Schrdinger approach

In this section we provide a schematic formulation of the so-called multi-time approach
and of its smeared Schrdinger version, in view of a comparison with the proposal of
previous section.
The multi-time formalism is based on the idea that many gravitational degrees of
freedom appearing in the classical geometrodynamics have to be not quantized because
are not real physical ones; indeed we have to do with 10 3 variables, i.e., the values
of the functions N N i hij in each point of the hypersurface 3 , but it is well-known that
the gravitational field possesses only 4 3 physical degrees of freedom (in fact the
gravitational waves have, in each point of the space, only two independent polarizations
and satisfy second order equations). The first step is therefore to extract the real canonical
variables by the transformation

Hr P r , = 0, 1, 2, 3 r = 1, 2,
hij ij ,
(47)
384
where Hr , P r are the four real degrees of freedom, while play the role of embedding
variables.
In terms of this new set of canonical variables, the gravity-matter action (15) rewrites

g

t + P r t Hr + t N H g + H N i Hi + Hi
d 3 x dt,
S g =
M4
Hg
(48)
= H g ( ,
,Pr)
g
Hi
g
= Hi ( , , Hr , P r ).
where
and
, Hr
Now we provide an ADM reduction of the dynamical problem by solving the
Hamiltonian constraint for the momenta

+ h , Hr , P r , , = 0.
(49)
Hence the above action takes the reduced form

r

P t Hr + t h t d 3 x dt.
S g =
(50)
M4
Finally, the lapse function and the shift vector are fixed by the Hamiltonian equations
lost with the ADM reduction, as soon as, the functions t are assigned. A choice of
particular relevance is to set t = 0 which leads to

r

P t Hr + t h0 d 3 x dt.
S g =
(51)
M4
The canonical quantization of the model follows by replacing all the Poisson brackets
with the corresponding commutators; if we assume that the states of the quantum system
are represented by a wave functional = ( , Hr , ), then the evolution is described by
the equations
i h
= h ,

(52)
where h are the operator version of the classical Hamiltonian densities. In its smeared
formulation the multi-time approach reduces to the following Schrdinger equation
i h t = ,
= (t, Hr , ).
Here denote the quantum correspondence to the smeared Hamiltonian

h t d 3 x dt.
=
(53)
(54)
M4
Now, observing that the first of Eqs. (25) can be rewritten as follows
i h
g
+H
,
= n H
i
(55)
it exists a correspondence between the above multi-time approach and our proposal, viewed
by identifying the formulas (23)(54), (55)(58) and (28)(56). But the following two key
385
differences appear evident: (i) the embedding variables y are added by hand, while the
corresponding ones come from non-physical degrees of freedom; (ii) the Hamiltonians
H and (as well as their corresponding densities) describe very different dynamical
situations.
We show explicitly the parallel between these two approaches by their implementation
in a minisuperspace model: a Bianchi type IX Universe containing a self-interacting scalar
field. By using Misner variables (, + , ) [3] the classical action describing this system
reads:

S=
p + p+ + + p + p

cNe3 p2 + p2+ + p2 + p2 + V (, , ) dt, c = const, (56)
where () d( )/dt and the precise form of the potential term V is not relevant for our
discussion. For this model, since the Hamiltonian density is independent of the spatial
coordinates, then the multi-time approach and its smeared Schrdinger version overlap,
the same being true in our formalism.
In the spirit of our proposal the quantum dynamic of this model is described by the
equation

i h t = cNe3 h 2 2 2+ 2 2 + V , = (t, , , ),
(57)
to which it should be added the restriction that the initial wave function phase 0 =
0 (, , ) satisfies the HamiltonJacobi equation

( )2 + (+ )2 + ( )2 + ( )2 0 + V = 0.
(58)
In this scheme N(t) is an arbitrary function of the label time to be specified when fixing
a reference.
To set up the multi-time approach we have to preliminarily perform an ADM reduction
of the dynamics (56). By solving the Hamiltonian constraint obtained varying N , we find
the relation

p hADM = p2+ + p2 + p2 + V .
(59)
Therefore, action (56) rewrites as

S=
p+ + + p + p h
ADM dt.
(60)
Thus we see how plies the role of an embedding variable (indeed it is related to the
Universe volume), while are the real gravitational degrees of freedom (they describe
the Universe anisotropy). By one of the Hamiltonian equation lost in the ADM reduction
(i.e., when varying p in (56)), we get
= 2cNe3 p = 2cNe3 hADM .
(61)
Hence by setting = 1, we fix the lapse function as

N=
e3
.
2chADM
(62)
386
The quantum dynamics in the multi-time approach is summarized by the equation

i h = h 2 (2+ + 2 + 2 ) + V , = (, , ).
(63)
We stress that in this multi-time approach the variable , i.e., the volume of the
Universe, behaves as a time-coordinate and therefore the quantum dynamics cannot
avoid the Universe reaches the cosmological singularity ( ). On the other hand,
in the formalism we proposed, is on the same footing of the other variables and are
admissible stationary states for which it is distributed in probabilistic way.
This feature reflects a more general and fundamental difference existing between
the two approaches: the multi-time formalism violates the geometrical nature of the
gravitational field in view of real physical degrees of freedom, while the proposed quantum
dynamics implements this idea only up to the lapse function and the shift vector, but
preserves the geometrical origin of the 3-metric field.
6. Physical interpretation of the model

A fundamental question we have to answer is about the physical meaning of the
kinematical variables adopted in the present reformulation of the quantum dynamics.
Indeed in the multi-time approach the corresponding embedding variables have no physical
meaning since they do not represent any physical degree of freedom, but simply equivalent
ways to represent the same real dynamics in the spacetime. However, in our case, these
variables have been added by hand to the geometrodynamics and provide, in general,
a violation of the classical symmetries; more precisely, by restricting, as in (45), the
initial phase of the wave functional, we violate the symmetry of general relativity only
on a quantum level and they are restored on a classical limit, when h 0.
A viable point of view is to regard this feature of the dynamics as a prescription
of the nature; by other words we may argue that the only experimental knowledge at
our disposition concerns the classical dynamics only (which therefore is to be restored
anyway), but noting we know, in principle, about the quantum behavior of gravity (which,
in the proposed paradigm we prescribe to be summarized by the present equations).
Another, more physical, point of view relies on observing that any dynamical term
able to deform the dynamics of an assigned system with different issues for its evolution,
should have a precise physical meaning and can be identified with some kind of field. In
what follows, we address this last way of thinking and argue the kinematical term can be
interpreted as a matter fluid, in close analogy with the so-called Gaussian reference fluid
which was first proposed by K. Kuchar and his collaborators, see [14,20]. In this spirit the
restriction (45) is no longer a key requirement for the consistency of the theory, but the
validity of the idea requires that this material component be experimentally detected.
We start by schematically reviewing the Gaussian reference approach. In a system of
Gaussian coordinates {T , Xi } (i = 1, 2, 3), the line element of a generic gravitational field
takes the form

ds 2 = dT 2 + hij T , Xk dXi dXj ,
(64)
387
and therefore such a system is obtained from a generic one by imposing the covariant
Gaussian constraints
g T T = 1,
g T Xi = 0.
(65)
In the above formula we adopted the notation

being
generic coordinates
to which it corresponds the metric tensor g .
The idea now consists into add this constraint to the gravity-matter action (i.e., into
requiring the latter be restricted to Gaussian references) via four Lagrangian multipliers M
and M i ; thus we consider an additional action term of the form

M g T T 1 + M i g T Xi d 3 x dt.
S (G) =
(66)
/y ,
M4
A careful analysis of the whole variational problem, leads [20] to dynamical constraints
of the form
pT + hT = 0,
pXi + hx i = 0,
(67)
where pT and pXi denote the conjugate momenta, respectively, of the variables T
and Xi , while hT and hXi correspond to linear combinations of the super-Hamiltonian
and the super-momentum with coefficients depending on T , Xi and the 3-metric hij . The
linearity of the above constraints with respect to the added momenta, allows to perform
a satisfactory quantization of the model in the spirit of a Schrdinger-like formalism. The
physical interpretation of this proposal relies on regarding the added term as a kind of fluid
which interact with gravity. Though this point of view is affected by some shortcomings
(indeed this fluid has a quite peculiar dynamics), nevertheless it provides an appropriate
notion of reference fluid clock.
The link of our formalism with the above Gaussian reference fluid consists of
the observation that in Gaussian coordinates the spatial hypersurfaces have equation
T (y ) = const and therefore the quantity T describes the covariant normal field n ,
while the gradients Xi are provided via the reciprocal vectors ei as Xi = ei +
(N i /N)n . Hence it is natural to argue that our kinematical term corresponds to a generic
reference fluid, having an associated time-like vector n , for which the first of (65) is
generalized to the normalization condition g n n = 1; the second condition (65)
does not hold for a generic fluid since we have g n Xi = n [ei + (N i /N)n ] =
N i /N = 0. Our parallel with the Gaussian reference fluid becomes precise by stressing
how, in our case, the normalization condition for n should not be added as a constraint
(by some Lagrangian multiplier); indeed it is ensured by the relation (1), obtained when
varying the kinematical action with respect to p .
We recall that the dynamics of the generic fluid reference is described by the
Hamiltonian equations

t y = Nn + N i i y ,
(68)
t p = p Nn /y + i N i p .
Once assigned the vector n as a functional of y and the functions N(t, x i ), N i (t, x i ),
we can solve the first of these equations to get the kinematical variables y = y (t, x i ).
388
Hence by substituting this information into the second Hamiltonian equation and solving
it, we get the momentum p = p (t, x i ).
In order to outline the relation existing between this system of Hamiltonian equations
and the dynamics of a fluid, we multiply the second of (68) by n , so getting via the first
one

n t p + p Nn /y i N i p = t p n i N i p n = 0.
(69)
Hence by setting p n = and labeling with barred indices ,
,
. . . all the quantities
in the coordinates t, x i , we rewrite the above equation as follows

= 0 n = 0,
t p n i N i p n = N n
(70)
where /
h denotes a real 3-scalar and refers to the covariant 4-derivative. By the
covariance of this equation, we should have in general

n = n t = 0,
(71)
being t n n the energymomentum tensor of a dust with energy density and fourvelocity n = n (t, x i ). The real correspondence of this tensor with the kinematical term,
i.e., with the generic fluid reference, comes out when observing, first that the requirement
ei /y = 0 implies, on a classical level p ei = 0 and then how the kinematical
term p n , appearing on the right-hand side of the super-Hamiltonian constraint can

be rewritten in the expressive form

1
1
p n = = h = h t n n = 2 h t t y t y = 2 t0 0 = h t00
N
N
(72)
(of course, apart from t0 0 , all the other components t vanish identically).
Therefore, the super-Hamiltonian constraint acquires the familiar expression

H g + H = H g + H + h t00 = 0.
(73)
Thus we see how the generic reference fluid, having the energymomentum tensor of
a dust fluid, contributes to the full super-Hamiltonian by a quantity like the one due to
a spacetime dependent cosmological term, coinciding with its energy density.
In the light of these considerations, the requirement that p vanishes identically implies
that 0 and therefore such a reference fluid does not interact with gravity by its energy
momentum; by a purely phenomenological point of view, in this case, it behaves like a test
matter field, whose kinematics is fixed by (1). This identification with a reference fluid
allows us to achieve the fundamental result of upgrading the formal time y to a real
physical clock.
Finally we stress the following two points:
(i) The momentum equation is equivalent to the conservation law for the dust energy
momentum tensor and therefore
provides the behavior of . For a Gaussian system this
equation yields = c(x i )/ h c(x i ), being c a generic space function. This
behavior is just the one of a dust energy density, but it is worth noting that our analysis
does not oblige to be a positively defined quantity.
(ii) The classical limit of the super-Hamiltonian equation (25) provides an HamiltonJacobi
389
equation where the eigenvalue coincides (see (73)) with the function taken on a specific
hypersurface.
7. Concluding remarks and a simple model

As outcoming of our analysis, we get a reformulation of the canonical quantization of
gravity in which is removed the so-called frozen formalism typical of the WDE, i.e., the
wave functional becomes evolutive along a one-parameter family of spatial hypersurfaces
filling the spacetime and can be expressed in terms of a direct dependence on the
parameter t labeling the slicing (this result provides a solution to the shortcoming of the
WDE emphasized at the point (ii)); indeed the existence of an Hilbert space associated
to the solutions of the restated equation can be regarded as a consequence of the nonfrozen formalism here obtained (resulting into a parabolic nature of the super-Hamiltonian
quantum constraint). Instead of these successful results, the WDE shortcoming indicated
at the point (iii) is overcome only with respect to those reference frames where the
shift vector vanishes; the reasons for this incompleteness are to be regarded as a direct
consequence of retaining in the quantum geometrodynamics the invariance of the wave
functional under the 3-diffeomorphisms. Of course the relaxation of this restriction is
a possible issue of an extended theory; here we do not address this point because it implies
some non-trivial complications in constructing an Hilbert space (at least on our heuristic
level), but overall because it would correspond to leave a priori the notion of a covariant
3-geometrodynamics, without any strong physical motivation to pursue this way (indeed
we are forced to relax the time displacement invariance by the incompatibility of a quantum
spacetime and the (3 + 1)-slicing notion).
However, the main goal of our analysis is achieved by removing the shortcoming of the
WDE stated at the point (iv), i.e., now the quantization procedure takes place in a fixed
system (indeed only the lapse function should be specified, while the shift vector can
assume a generic value) and no ambiguity survives about the time-like character of the
normal field; by other words, in this new approach it is possible to quantize the 3-geometry
field on a fixed family of spatial hypersurfaces (corresponding to its evolution in the space
time), because this quantization scheme does not contradicts the strong assumption of
a (3 + 1)-slicing of the 4-dimensional manifold.
The discussion presented in Section 6, about the interpretation of the kinematical
variables as a generic reference fluid clock, has the very important merit to transform
a working formalism into a possible experimental issue; however, it should be supported
by further investigation on the physical consequences the clock fluid has when referred to
specific contexts: indeed the question about the appropriate definition of a reference frame
in a quantum spacetime is a really subtle one (see, for instance, [22,23]).
In spite of this available physical issue (ensured by the fluid interpretation), we
emphasize how to have found the (non-physical) restriction on the initial wave functional
phase (45), which ensures the classical limit coinciding with general relativity, is essential
for the consistency of the whole approach; in fact the physical meaning of this result
concerns the fundamental achievement of restoring, on a classical level, the invariance of
the theory under the time displacements, which is instead broken by the quantum dynamics.
390
We conclude by observing how the functional nature of all our approach implies it has
(like in the WDE) a heuristic value; but it appear rather reasonable that it can be made
rigorous when reformulated, in a discrete approach, as a theory on a suitable lattice.
Though is out of the aim of this paper to face this problem (to be regarded as a fundamental
subject of further investigations), nevertheless we here suggest that the best method to
reformulate the quantum dynamics on a discrete level, seems to be via the Regge calculus
[24,25] as applied to the 3-geometry field. The physical justification for a discrete approach
to quantum gravity relies on the expectation that the spacetime has a lattice structure (or
a granular morphology) on a Planckian scale.
At the end of this work we provide an application of the obtained theory to the
quantization of a very simple model, described by the following line element:
ds 2 = N(t)2 dt 2 r(t)4/3 ij dx i dx j ,
(74)
where ij denote the Krnecker matrix and the flat hypersurfaces t = const are taken to
have a closed topology, i.e., 0 x i < 2L (i = 1, 2, 3) (L = const is the radius of the
three cyrcles and has the dimensionality of a length). We allow r belongs to the positive
real axis 0 r < (since the metric is invariant under the exchange r r).
The ADM action describing the vacuum dynamics of this 1-dimensional model, reads
in the simple form

GN 2
2 L3 c3
p
pr r
r 2 dt =
dt,
S =
(75)
r
2G
2 2 L3 c3
where () t ( ) and pr denotes the conjugate momentum to the variable r. From
a classical point of view this model corresponds to the Euclidean 3-space (to which
is associated a generic time variable) since r pr = 0 r = r0 = const and the
corresponding HamiltonJacobi equation and solution read
2
d
= 0 = 0 = const.
(76)
dr
The quantum dynamics of the model is described by Eq. (27), which, in the present case
reduces to the very simple form:
N h 2 2
2 L3 c3
r (t, r),
.
(77)
2
G
Apart from the negative nature of its Hamiltonian, we see how this quantum gravity model
corresponds to the free non-relativistic particle; the general solution of the above equation
reads in the following wave-packet form:

t

p2
i
1
pr +
,
dp (p) exp
N t dt
(t, r) =
(78)
h
2
h
i h t (t, r) =
t0
where (p) is determined by the relation

1
(p) =
h

0

pr
dr 0 (r) exp i
,
h
0 (r) (t = 0, r).
(79)
391
Since (as well as 0 ) should verify the boundary conditions (r = 0) = 0 (in addition
to (r ) = 0), then should be antisymmetric in its argument, i.e., (p) = (p).
In terms of the expression (79) and by choosing the synchronous gauge N = c (t T ),
we may rewrite the wave function (78) as follows (taking into account the expression for
and setting T0 = 0):
(T , r) = e
i0 /h
1
2

dr 0 r

where lPl denotes the Planck length (lPl

(45), we set
0 = 0 ei0 /h ,

2

2 cT lPl
dx exp i x r r + x
,
2 2 L3
(80)

Gh /c3 ) and, in agreement with the restriction
(81)
being 0 a real function subjected only to the boundary conditions 0 (r = 0) = 0 and

0 (r ) = 0. At last, it is worth noting that, for any fixed T , takes place the (intriguing)
limit
lim
(cT lPl 2 /L3 )0
(T , r) = 0 (r),
(82)
where the initial probability distribution 02 can eventually be regarded as strongly peaked
around a fixed value r = r0 .
Acknowledgements
We are very grateful to Bryce DeWitt for his valuable comment on the topic faced by
this paper.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
N.D. Birrell, P.C.W. Davies, Quantum Fields in Curved Space, Cambridge Univ. Press, Cambridge, 1982.
B.S. DeWitt, Phys. Rev. 160 (1967) 1113.
C.W. Misner, K. Thorne, J.A. Wheeler, Gravitation, Freeman, San Francisco, 1973, Ch. 21, Ch. 43.
B.S. DeWitt, in: T. Piran (Ed.), Proc. eighth Marcell Grossmann meeting, Jerusalem, 2227 June 1997, p. 6.
E.W. Kolb, M.S. Turner, The Early Universe, AdisonWesley, Reading, 1990, p. 447.
J.B. Hartle, in: B. Iver et al. (Eds.), Highlights in Gravitation and Cosmology, Cambridge Univ. Press, 1988.
A.A. Kirillov, G. Montani, JETP Lett. 66 (1997) 7;
A.A. Kirillov, G. Montani, JETP Lett. 66 (1997) 475.
F. Mansouri, J.J. Scanio (Eds.), Quantum Gravity, University of Cincinnati, USA, 34 April 1992.
C. Rovelli, Loop quantum gravity, 1983, gr-qc/9701008.
R. Arnowitt, S. Deser, C.W. Misner, in: I. Witten, J. Wiley (Eds.), Gravitation: An Introduction to Current
Research, New York, 1962.
J.B. Hartle, S.W. Hawking, Phys. Rev. D 28 (1983) 12.
D. Oriti, Spacetime from algebra: spin foam models for non-perturbative quantum gravity, gr-qc/0106091,
to appear in Rep. Prog. Phys. (2001).
K. Kuchar, in: C.J. Isham et al. (Eds.), Quantum Gravity II, a second Oxford symposium, Clarendon, Oxford,
1981.
392
[14] C.J. Isham, Canonical quantum gravity and the problem of time, gr-qc/9201011.
[15] C.J. Isham, K.V. Kuchar, Ann. Phys. 164 (1985) 288;
C.J. Isham, K.V. Kuchar, Ann. Phys. 316 (1985) 288.
[16] J.J. Halliwell, J.B. Hartle, Phys. Rev. D 43 (1991) 1170.
[17] J.B. Hartle, in: A. Ashtekar, J. Stachel (Eds.), Conceptual Problems of Quantum Gravity, Birkhuser,
Boston, 1991.
[18] J.J. Halliwell, Phys. Rev. D 43 (1991) 2590.
[19] S. Tomonaga, Prog. Theor. Phys. 1 (1946) 27;
S. Tomonaga, in: J. Schwinger (Ed.), Selected Papers on Quantum Electrodynamics, Dover, 1958.
[20] K. Kuchar, C. Torre, Phys. Rev. D 43 (1991) 419.
[21] J. Zinn-Justin, Quantum Field Theory and Critical Phenomena, 3rd edn., Clarendon, Oxford, 1996, p. 228.
[22] C. Rovelli, Class. Quantum Grav. 8 (1991) 297.
[23] C. Rovelli, Phys. Rev. D 43 (1991) 442.
[24] T. Regge, Nuovo Cimento 19 (1961) 558.
[25] T. Regge, in: T. Piran (Ed.), Proc. eighth Marcell Grossmann meeting, Jerusalem, 2227 June 1997, p. 2.

Global analysis of Solar neutrino oscillation

evidence including SNO
and implications for Borexino
P. Aliani a , V. Antonelli a , M. Picariello a , E. Torrente-Lujan a,b,c
a Dipartimento di Fisica, Universit di Milano, and INFN Sezione di Milano, Via Celoria 16, Milano, Italy
b Departamento Fisica Teorica C-XI, Universidad Autonoma de Madrid, 28049 Madrid, Spain
c CERN TH Division, CH-1202 Geneve, Switzerland
Received 29 November 2001; accepted 23 April 2002
Abstract
An updated analysis of all available neutrino oscillation evidence in Solar experiments including
the latest SNO data is presented. Predictions for total rates and daynight asymmetry in Borexino
are calculated. Our analysis features the use of exhaustive computation of the neutrino oscillation
probabilities and of an improved statistical 2 treatment.
In the framework of two neutrino oscillations we conclude that the best fit to the data is obtained
2 /n = 0.81,
in the LMA region with parameters (m2 , tan2 ) = (5.4 105 eV2 , 0.38), (min
n = 38 degrees of freedom). Although less favored, solutions in the LOW and VAC regions are
still possible with a reasonable statistical significance. The best possible solution in the SMA region
gets a maximum statistical significance as low as 3%.
We study the implications of these results for the prospects of Borexino and the possibility of
discriminating between the different solutions. The expected normalized Borexino signal is 0.63
at the best fit LMA solution, where the DN asymmetry is negligible (approximately 103 ). In the
LOW region the signal is in the range 0.60.7 at 90% confidence level while the asymmetry is
120%. As a consequence, the combined Borexino measurements of the total event rate with an
error below 510% and daynight total rate asymmetry with a precision comparable to the one of
SuperKamiokande will have a strong chance of selecting or at least strongly favoring one of the Solar
neutrino solutions provided by present data. 2002 Published by Elsevier Science B.V.
PACS: 14.60.Pq; 26.65.+t
E-mail addresses: paul@lcm.mi.infn.it (P. Aliani), vito.antonelli@mi.infn.it (V. Antonelli),

marco.picariello@mi.infn.it (M. Picariello), torrente@cern.ch (E. Torrente-Lujan).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 3 0 - 9
394
P. Aliani et al. / Nuclear Physics B 634 (2002) 393409
1. Introduction
The Solar neutrino problem [110] has been defined, for example, as the difference
between the neutrino flux measured by a variety of Solar neutrino experiments and the
predictions of the Standard Solar Model (SSM). The explanation of this difference by the
neutrino oscillation hypothesis has been boosted by the SNO demonstration at the 3 level
of the appearance at the Earth of Sun-directed muon and tau neutrino beams [11] and the
first determination of the total 8 B neutrino flux generated by the Sun [12]. The agreement
of this flux with the expectations implies as a by-product the confirmation of the validity
of the SSM [1315].
At present the SNO experiment measures the 8 B Solar neutrinos via the reactions [16
19]: (1) Charged Current (CC): e + d 2p + e , (2) Elastic Scattering (ES): x + e
x + e . The first reaction is sensitive exclusively to electron neutrinos. The second,
the same as the one used in SuperKamiokande (SK), is instead sensitive, with different
efficiencies, to all flavors. The first results by SNO on Solar neutrinos (1169 neutrino events
collected during the first 240 days of data, Ref. [12]) confirm previous evidence from SK
and other experiments [1,3,4,20]. The Solar neutrino flux measured via the observed CC
reaction rate,
SNO
= 1.75 0.07 0.12 106 cm2 s1 ,
CC
is lower than the two existing ES reaction rate measurements: the more precise at SK
SK
ES
= 2.32 0.03 0.08 106 cm2 s1
and the one measured by SNO itself

SNO
ES
= 2.39 0.34 0.15 106 cm2 s1 .
SNO
SNO
SNO
SK
Comparison of the CC
and ES
and especially CC
and ES
can be considered as
the first direct evidence for Solar neutrino oscillations which cause the appearance of sundirected muon and tau neutrinos at the Earth. When considering only SNO results, it is
advantageous to use the experimental ratio of CC to ES measurements because of the
cancellation of systematic errors. The experimental value is
SNO
SNO
/ES
= 0.748 0.13
R exp = CC
which is 2 away from the no-oscillation expectation R exp = 1. Recall that the statistical
SNO SNO is clearly lower (1.4 ) [12]. If we use the ESsignificance of the difference CC
ES
SK result, the no-oscillation hypothesis is excluded at 3.3 [21,22,52,53]. Moreover the
allowed oscillations to only sterile neutrinos is excluded at the same level ( 3 ). Although
less favored, the oscillations to both active and sterile neutrinos are still allowed [2329].
Although of qualitative importance, the SNO results by themselves, due to their
large error bars, are still far from constraining the neutrino oscillation parameters in
a significant way. For this purpose we continue to need the data of all other Solar neutrino
experiments [1,2,4,30,31]. The aim of this work is to present an up-to-date analysis of
all available Solar neutrino evidence including the latest CC-SNO data in a two-neutrino
framework. The predictions for day and night signal rates in the near future real-time 7 Be
395
sensitive Borexino experiment [3235] are also calculated. Two important characteristics
of this paper are the use a thorough numerical computation of the neutrino oscillation
probabilities and the use of an improved statistical 2 analysis. The latter is explained in
detail in the next section. In summary we have followed the SK standard procedure as close
as possible allowing for a better treatment of correlated uncertainties.
As we show below, the best fit to the full data set including global rates and SK
energy spectrum is obtained in the Large Mixing Angle (LMA) region. Although less
favored, solutions in the Low mass (LOW) and Vacuum (VAC) regions are possible with
a reasonable statistical significance. A best fit solution in the Small Mixing Angle (SMA)
region is strongly disfavored.
This work is organized as follows. In Section 2 we present a summary of the methods
that are used: details of the computation of neutrino oscillation probabilities in the Sun
and in the Earth, scattering cross sections and calculation of the signal at the detectors.
We continue with a detailed account of the 2 statistical methods. In Section 3 we present
the results of different analyses: first considering total event rates only and then including
the SK energy spectrum information. In Section 4 we present our Borexino predictions for
the best fit solutions obtained in the previous sections. Finally in Section 5 we summarize
and draw conclusions.
2. Methods
2.1. Neutrino oscillations and expected signals
In general, our determination of neutrino oscillations in Solar and Earth matter and
of the expected signal in each experiment follows the standard methods found in the
literature [36]. Nonetheless, this work differs from previous ones in the treatment of the
computation of the neutrino transition probabilities. We completely solve numerically the
neutrino equations of evolution for all the oscillation parameter space. Although other
alternatives, as the use of semi-analytical expressions in portions of the (m2 , tan2 )
plane, could be previously justified in terms of economy of resources, at this stage a full
simulation is an affordable and better option. This is specially true in the calculation of the
smearing of the neutrino probability as a function of the neutrino production point when
introducing non-radial propagation. However no surprises are found: our numerical results
are in overall good agreement with the semi-analytical approach.
The survival probabilities for an electron neutrino, produced in the Sun, to arrive at
the Earth are calculated in three steps. The propagation from the production point to the
Suns surface is computed numerically in all the parameter range using the electron number
density ne given by the BPB2001 model [14] averaging over the production point. The
propagation in vacuum from the Suns surface to the Earth is computed analytically. The
averaging over the annual variation of the orbit is also exactly performed using simple
Bessel functions. To take the Earth matter effects into account, we adopt a spherical model
of the Earth density and chemical composition. In this model, the Earth is divided in
eleven radial density zones [37], in each of which a polynomial interpolation is used to
obtain the electron density. The gluing of the neutrino propagation in the three different
396
regions is performed exactly using an evolution operator formalism [36]. The final survival
probabilities are obtained from the corresponding (non-pure) density matrices built from
the evolution operators in each of these three regions. The night quantities are obtained
using appropriate weights which depend on the neutrino impact parameter and the sagitta
distance from neutrino trajectory to the Earths center, for each detectors geographical
location.
The expected signal in each detector is obtained by convoluting neutrino fluxes, oscillation probabilities, neutrino cross sections and detector energy response functions. We have
used neutrino-electron elastic cross sections which include radiative corrections [38,39].
Neutrino cross sections on deuterium needed for the computation of the CC-SNO measurements are taken from [40].
Detector effects are summarized by their respective response functions, obtained by
taking into account both the energy resolution and the detector efficiency. The resolution
function for CC-SNO is that given in [12]. We obtained the energy resolution function
for SK using the data presented in [4143]. The effective threshold efficiencies, which take
into account the live time for each experimental period, are incorporated into our simulation
program. They are obtained from [47].
2.2. The 2 calculations: definitions and procedures
The statistical significance of the neutrino oscillation hypothesis is tested with
a standard 2 method which we explain in some detail here.
In the most simple case, the analysis of global rates presented in Section 3.1, the
definition of the 2 function is the following:
T 1 th

2
glob
R R exp ,
= R th R exp 2
(1)
2 + 2 .
where 2 is the full covariance matrix made up of two terms, 2 = unc
cor
2
The diagonal matrix unc contains the theoretical, statistical and uncorrelated errors
2 contains the correlated systematic uncertainties. The R th,exp are vectors
while cor
containing the theoretical and experimental data normalized to the SSM expectations. The
length of these vectors is 3 or 4 depending on whether the CC-SNO experiment is included
or not:
th,exp
Ri
th,exp
= Si
/SiSSM ,
i = Cl, Ga, SK, (CC-SNO ).
The index i denotes the different Solar experiments: Chlorine (Cl), Gallium (Ga),
SuperKamiokande (SK) and Charged Current SNO (CC-SNO ).
The correlation matrices, whether including the SNO experiment or not, have been
computed using standard techniques [44,45].
To test a particular oscillation hypothesis (m2 , tan2 ) against the parameters of the
2
best fit, we perform a minimization of the glob
as a function of the oscillation parameters.
2
2
2
A point in parameter space (m , tan ) is allowed if the globally subtracted glob
2 (m2 , tan2 ) 2 < 2 (CL). Where 2 (90%, 95%, . . .) =
fulfills the condition glob
n
min
n=2
4.60, 5.99, . . . are the n = 2 degrees of freedom quantiles.
397
A more elaborated statistical analysis is necessary once one introduces the SK energy
spectrum into the analysis (Section 3.2).
The procedure assumed by the SK Collaboration, in order to present its own analysis

and obtain the tables of bin correlated errors, is that of using the summation 2 = i2
where each term is of the form [46,47]
exp
i fi (cor )Rith Ri ,
(2)
and where the expression

fi (cor )
1
1 + i,cor cor
(3)
is the response function for the correlated error in the ith-energy bin. The i,cor are bincorrelated uncertainties and is an overall flux normalization factor. The correlation
parameter cor is arbitrary and determined in the minimization process together with the
rest of the oscillation parameters.
SuperKamiokande calculation of tables for correlated errors (as Table 3 in Ref. [47]
and previous publications) and the 2 prescription given respectively by Eqs. (2) and (3)
are closely related. In this work we follow this definition of the 2 , however, for clarity
we write the expressions in a notation which renders the comparison with other published
works simple. It can be easily shown that the expression to be presented below leads to the
SK expressions. 1
Therefore, for the analysis of the full set of data including SK energy spectrum, we
2
2 . The first one
+ spec
consider a 2 function which is sum of two quantities 2 = glob
is given by Eq. (1) considering only Cl, Ga and CC-SNO total rates, while for the second
term we adopt the definition

t 2
th

2
2 1
spec
(4)
=
+ cor cor
R th R exp unc
R R exp + 2 + 2 .
d,n
A subindex i = d, n corresponding to separated day and night quantities is understood

for any R vector and matrix. We have introduced the flux normalization factor and
the correlation parameter cor . The complete variance matrix is not a constant quantity.
It is obtained by combining the statistical variances with systematic uncertainties and
dependent on this correlation parameter.
The corresponding 2 terms in Eq. (4) are
2 =
( th )2
,
2
2 =
th )2
(cor cor
.
2
The central value th = 1 corresponds to the BPB 2001 8 B flux. The uncertainty on
8
comes from the +19
14 % uncertainty in the B flux and we make an average of the twoside errors taking th = 1.00 0.17 [14]. The correlation parameter is assumed to
be constrained to vary in a Gaussian way with an error corresponding to and we take
th = 1.
cor
1 We have not included a term log det 2 ( ) log(1 + ) whose effect is negligible.
cor
cor
398
In this work we consider two cases. In the first case the flux normalization is taken
to be free (we allow for a very large in the 2 term). As a second possibility
we consider that both quantities and cor are constrained. Note that the unwanted
consequence of eliminating the last term 2 would be to obtain running-away solutions
for the optimal value of this parameter.
The 2 summation now contains 41 bins in total: 3 from the global rates (all the
experiments except SK) and 2 19 bins for the SK day and night spectrums. The full
correlation matrix is defined by blocks. The 3 3 block corresponding to the global rates
is defined as above. For each day and night spectrum the corresponding 19 19 block
correlation matrices are conservatively constructed assuming full correlation among energy
bins. 2 The components of the variance matrix are
2
cor ij = i,exp j,exp + i,cal j,cal ,
2
2
2
+ i,unc
,
unc ii = i,stat
where stat are the statistic errors and the quantities exp , cal , unc are, respectively,
the bin-correlated experimental, the spectrum calculation uncertainties and the binuncorrelated ones.
Notice that in the above sum we do not simultaneously include the global ratio for SK
and the partial spectrum bins. We ignore any correlation between the spectrum information
and the global rates of all experiments except SK itself. Finally, a remark is in order. In
this case the defined covariance matrix is dependent on one of the fitting parameters. One
does not, however, have to add any correction (i.e., log det terms) to the 2 expression
min . To test a particular oscillation hypothesis and obtain allowed
as long as we sit at cor
regions in parameter space we perform a minimization of the four-dimensional function
2 (m2 , tan2 , , cor ). The minimization of the expression with respect to and cor is
done analytically. The subsequent minimization in the (m2 , tan2 ) plane is numerical.
min , =
For cor = cor
min , a given point in the oscillation parameter space is allowed if
2 <
the globally subtracted quantity fulfills the condition 2 = 2 (m2 , tan2 ) min
2
2
n (CL). Where n=4 (90%, 95%, . . .) = 7.78, 9.4, . . . are the quantiles for four degrees of
freedom.
3. Results
We have used data on the total event rates measured at chlorine Homestake experiment,
at the gallium experiments SAGE [4,48], GNO [30] and GALLEX [31] and at the water and
heavy-water SK (live time 1258 days) [46] and SNO (ES, CC 240 days [12]) experiments
(see Table 1 for an explicit list of results and references).
For the purposes of this work it is enough to summarize all the gallium experiments in
one single quantity by taking the weighted average of their rates.
In addition, for SK we use the available information for the day and night energy
spectrums. For each of the day and night cases, this spectrum information contains 18 total
2 The introduction of the parameter
cor is equivalent to the relaxation of this condition during minimization.
399
Table 1
Summary of data used in this work. The expected signal (SSSM ) and observed ratios SData /SSSM with respect
to the BPB2001 model are reported. The SK and CC-SNO rates are in 106 cm2 s1 units. The Cl, SAGE and
GNOGALLEX measurements are in SNU units. In this work we use the combined results of SAGE and GNO
GALLEX: SGa /SSSM (Ga SAGE + GALLEX + GNO) = 0.579 0.050. The 8 B total flux is taken from the
6
2 s1
BPB2001 model [14]: (8 B) = 5.05(1+0.20
0.16 ) 10 cm
SSSM
SData /SSSM (1 )
2.32 0.03 0.08

2.39 0.34 0.15
2.56 0.16 0.16
67.2 7.0 3.2
74.1 6.7 3.5
0.451 0.011
0.347 0.029
0.332 0.056
0.521 0.067
0.600 0.067
Experiment [Ref.]
SK (1258d) [46]
CC-SNO (240d, 820 MeV) [12]
Cl [51]
SAGE [4,48]
GNO-GALLEX [30,31]
Table 2
Left: best fit oscillation parameters: m2 (eV2 ) and tan2 , from the analysis of global rates: Cl, Ga and SK.
The SNO measurement is not included (d.o.f. = 3 2). The significance level g.o.f. is in percentage. The
sampling error is 1% for m2 and tan2 ( ). Right: minimization residuals in 1 units for diverse experimentally
measured quantities: the global rates for the four experiments and SK global daynight asymmetry (ADN ). Only
the first three quantities appear in the 2 computation
m2
tan2
2 /n
m
SMA
VAC
7.8 106
8.4 1011
LMA
LOW
1.2 105
1.1 107
1.10 103
0.27
3.37
0.34
0.75
1.1
1.3
1.6
1.9
3.6
Solution
g.o.f.
29
26
21
17
5.7
SK
Cl
Ga
SNO
ADN
SK
0.5
1.9
1.7
0.0
2.0
0.8
1.4
1.4
1.3
1.9
0.2
0.8
0.9
0.2
1.0
2.5
0.2
0.3
0.7
2.5
1.3
1.3
1.3
1.1
0.3
recoil-electron energy bins of width 0.5 MeV in the range 5 to 14 MeV and an additional
bin spanning the remaining 1420 MeV range. The data and errors for individual energy
bins for SK spectrum has been obtained from Ref. [47]. The information from other SK
results as the global daynight asymmetry is already contained to a large extent in the
previous quantities and does not change the results to be presented further on [3,46].
3.1. Global rate analysis
The analysis of the global rates of the four experiments SK, Chlorine, Gallium and CCSNO is the simplest possibility but nonetheless it reveals important trends of the solutions
to the Solar neutrino problem. It illustrates how the CC-SNO data impose additional
constraints on the parameter space and in particular how the SMA solution loses its
statistical significance in favor of the LMA and LOW regions, becoming allowed only
at marginal 3 confidence levels.
We present results both with and without the inclusion of the CC-SNO global rate. In
the analysis of the global rates of the SK, Cl, Ga (CC-SNO ) experiments one has two
free parameters m2 and tan2 and 3 (4) experimental quantities, therefore, the effective
number of degrees of freedom (d.o.f.) is 1 (2).
On the left-hand side of Tables 2, 3 we present the best fit parameters or local minima
obtained from the minimization of the 2 function given in Eq. (1). Also shown are the
400
Table 3
Best fit oscillation parameters, m2 (eV2 ) and tan2 , and minimization residuals (see explanation in Table 2).
The analysis now includes the global rates for the four experiments: Cl, Ga, SK and CC-SNO (d.o.f. = 4 2)
Solution
m2
LMA
SMA
VAC
3.1 105
7.3 106
9.7 1011
LOW
3.2 107
tan2
2 /n
m
0.40
1.40 103
0.26
4.00
0.80
1.6
2.4
3.1
3.2
3.3
g.o.f.
21
8.8
4.6
4.1
3.5
SK
Cl
Ga
SNO
ADN
SK
1.4
1.0
3.8
4.1
2.7
0.9
1.5
2.3
2.5
1.2
1.5
0.4
1.1
1.3
3.1
1.3
0.9
1.5
1.9
1.7
1.6
1.3
1.3
1.3
0.9
2 per degree of freedom ( 2 /n) and the goodness of fit (g.o.f.) or significance
values of min
level of each point (definition of SL as in Ref. [49]). On the right-hand side of the same
tables we show the deviations (minimization residuals) from the expected values for all
the total event rates. We also include, for further reference, the deviations corresponding to
the SK day and night energy spectrums and global daynight asymmetry, although these
quantities are not incorporated in the 2 minimization. We first consider the case in which
the CC-SNO global rate is ignored. The absolute minimum is located at the SMA region
2 /n = 1.06 (d.o.f. n = 3 2),
(m2 , tan2 )SMA = (7.8 106 eV2 , 1.10 103), with min
the significance level for this point g.o.f. is 29%. Closely we find the twin minima situated
in the VAC region with parameters (m2 , tan2 )VAC = (8.4 1011 eV2 , 3.37(0.27)),
which receive a significance level only slightly smaller than the SMA minimum. The next
local minima are situated in the LMA region (m2 , tan2 )LMA = (1.23 105 eV2 , 0.34),
2 /n = 1.89 which corresponds to a still reasonable good fit g.o.f. is 17% and in
with min
2 /n = 3.62. For
the LOW region (m2 , tan2 )LOW = (1.11 107 eV2 , 0.75), with min
comparison, much-less significant results are obtained when sterile neutrino oscillations
are considered.
When one introduces the CC-SNO total event rate in the analysis the relative order
of the local minima changes. The absolute minimum is now located at the LMA region:
(m2 , tan2 )LMA = (3.12 105 eV2 , 0.40). The significance level of the absolute
2 /n = 1.61 (d.o.f. n = 4 2) which
minimum is clearly worse than in the previous case min
corresponds to g.o.f. = 21%.
2 /n = 2.40,
The significance level of the SMA region which follows is no better than min
2
2
SMA
6
g.o.f. is 8.8%, value which is obtained at (m , tan )
= (7.3 10 eV2 , 1.40
3
10 ). After that, we have the pair of minima at the VAC region (m2 , tan2 )VAC =
2 /n = 3.2 3.1. Finally, we find in the table
(9.7 1011 eV2 , 4.00(0.26)), with min
a minimum in the LOW closely situated to the one of the previous case.
In Fig. 1 we present graphically our results respectively before and after including the
global CC-SNO rate. In the plots one can see the regions which are allowed at 90, 95,
99 and 99.743% confidence levels. The main difference between the plots is an overall
reduction of the extent of the allowed area at any CL. The large contiguous areas at 3 and
4 which are present along tan2 1 in the first case are converted into well separated
patches after the inclusion of SNO results. The most serious consequence is however the
considerable reduction on the significance of the SMA region. From the analysis of the four
global rates a good part of this region becomes acceptable only at some marginal level. In
these and the next plots, the region above the line at high m2 is excluded at 99% CL
(a)
401
(b)
Fig. 1. Best fit solutions for the global rate analysis. Global rates from SK, Cl and Ga experiments are included.
(b) contains also the CC-SNO one. The black dots are the best fit points (the absolute minimum is located at the
SMA (a) or LMA (b) regions). The colored areas are the allowed regions at 90, 95, 99 and 99.7% CL relative to
the absolute minimum. The region above the solid line is excluded by CHOOZ results at 99% CL [50].
by the negative results of CHOOZ and Palo Verde [50]. Note that regions for very large
m2 103 eV 2 are excluded at the 23 level without need of the CHOOZ result once
we included the SNO data.
The reduction of the SMA region can be explained from the tables of deviation (right
part in Tables 2, 3). Before the introduction of the CC-SNO global rate in the computation
and 2 minimization the residuals for the CC-SNO data corresponding to the best fit are
very high (Table 2): 2.5 for the SMA and also for the LOW solution. The CC-SNO
residual values for remaining regions are however smaller. When the CC-SNO data is
incorporated into the fit the SMA and LOW regions become less favored. The effect is
much more important for the relatively small SMA region. For the LOW region the initial
area is large and nearby minima can be found to adjust for the CC-SNO data and still
provide reasonable quality fits.
3.2. Daynight SK spectra and global rates
In this section we present the results obtained when including the SK day and night
energy spectrum rates in addition to the total event rates of Cl, Ga experiments and CCSNO .
The definition of the 2 functions and minimization procedures used in the statistical
analysis are explained in detail in Section 2.2. The number of experimental data inputs is
2 19 + 3 = 41. One now has four free parameters, the oscillation parameters m2 and
tan2 , the 8 B neutrino flux normalization factor and the correlation parameter cor . Two
cases will be studied. In case A the flux normalization is considered a free parameter, the
402
Table 4
Best fit oscillation parameters, m2 (eV2 ) and tan2 , and minimization residuals (see explanation in Table 2).
The analysis now includes the global rates for three experiments Cl, Ga and CC-SNO , and SK day and night
energy spectra. There are four parameters: m2 , tan2 , and cor . We let the flux normalization vary freely,
so the d.o.f is = 41 4 (see the text). The sampling error due to the finite-grid is 1% for m2min and tan2 ( )
m2
tan2
2 /n
m
LMA
LOW
VAC
5.4 105
1.2 107
2.7 1010
SMA
7.3 106
0.36
0.75
2.70
0.32
1.30 103
0.8
0.8
0.8
0.9
1.1
Solution
g.o.f.
SK
Cl
Ga
SNO
ADN
SK
82
80
80
80
40
1.8
2.0
2.2
2.3
0.1
0.4
2.0
1.8
2.0
1.0
0.8
0.9
0.5
0.6
0.3
2.1
2.5
0.7
0.8
2.1
0.0
0.2
1.2
1.2
1.3
Table 5
Same as Table 4 except that here the flux normalization is constrained to vary with the SSM standard deviation
and the d.o.f is = 41 3 (see the text)
Solution
m2
5.4 105
LMA
LOW
VAC
7.5 108
8.9 1010
SMA
7.3 106
tan2
2 /n
m
0.38
0.84
1.73
0.48
1.30 103
0.8
0.9
0.9
0.9
1.1
g.o.f.
SK
Cl
Ga
SNO
ADN
SK
82
72
73
63
40
1.6
2.7
2.4
2.4
0.1
0.5
2.3
1.9
2.2
1.0
0.9
0.8
0.5
1.4
0.3
1.8
2.9
0.7
1.6
2.1
0.0
0.7
1.3
1.3
1.3
number of effective d.o.f. is then 41 4 = 37. In case B, the parameter is constrained

to vary around the BPB2001 central value with a standard deviation given by SSM. We
now have 38 d.o.f.
In Tables 4, 5 we present the best fit parameters or local minima obtained from the
minimization of the 2 function given in Eq. (4). As in the previous section, the values
2 per degree of freedom ( 2 /n) and the goodness of fit (g.o.f.) or significance level
of min
of all minima are shown. In the right part of the tables, the deviations from the expected
values or minimization residuals for a number of experimental quantities are listed.
The results for the free flux minimization or case A are presented first. As a summary
of numerical results, which appear in detail in Table 4 and Fig. 2(a)), the position
of the absolute minimum is located at the LMA region (m2 , tan2 )LMA = (5.4
2 /n = 0.82 (d.o.f. n = 37). The significance level for this point
105 eV2 , 0.36), with min
is noticeably larger (g.o.f. = 82%) than the levels obtained in global rates only analysis.
This is only in part an artifact of the statistical machinery. It is a satisfactory result
because it basically reflects the internal consistency of the data: although the number
of degrees of freedom have considerably increased, the 2 per d.o.f is still much the
same. The minimization value of the flux normalization is = 1.02 while the correlation
parameter is cor = 1.43. The next local minima are situated at the following positions:
2 /n = 0.85 and
the LOW region (m2 , tan2 )LOW = (1.2 107 eV2 , 0.75), with min
2
2
VAC
10
the two solutions at the VAC region (m , tan )
= (2.7 10
eV2 , 2.70(0.32)),
2
with min /n = 0.85. In this last region the best-fit flux normalization is down to =
0.56 while cor = 1.65. In case B, where the flux normalization is constrained to its
SSM value, the results are very similar and are shown in Table 5 and Fig. 2(b). The
(a)
403
(b)
Fig. 2. Global solutions for the analysis of Cl, Ga and CC-SNO experiments plus the daynight SK spectrum rates.
The black dots correspond to best fit points. The colored areas are the allowed regions at 90, 95, 99 and 99.7%
CL relative to the absolute minimum. (a) (case A of the text): the minimization with respect cor is performed
as before. The absolute normalization factor is allowed to vary freely. (b) (case B of the text): the absolute
normalization factor and correlation parameter cor are varied subject to constraints.
(a)
(b)
Fig. 3. Marginalized likelihood distributions for each of the oscillation parameters m2 (a), tan2 (b) appearing
in the 2 fit. The curves are in arbitrary units with normalization to the maximum height. The continuous lines
shows the distributions corresponding to Fig. 2(b). Dashed lines correspond to the case represented in Fig. 1(a).
position of the first minimum remains practically unchanged, and is located at the LMA
2 /n = 0.82 (d.o.f. n = 38)
region (m2 , tan2 )LMA = (5.4 105 eV2 , 0.38), with min
and g.o.f.= 82%. The next local minima are: the one situated in the LOW region
2 /n = 0.86, and the twin solutions
(m2 , tan2 )LOW = (7.5 108 eV2 , 0.84), with min
2 /n = 0.9.
2
2
VAC
10
at the VAC region (m , tan )
= (8.9 10
eV2 , 1.73(0.48)), with min
Finally, in both cases A and B we have a local minimum situated at the SMA region. These
minima are situated at m2 7.3 106 and tan2 1.3 103 with a considerably
poorer statistical significance.
404
Table 6
Borexino experiment expectations for the normalized total signal (S Bor ) and daynight asymmetry (ADN ) for all
of the best-fit solutions appearing in Table 5
Solution
m2 (eV2 )
tan2 ( )
S Bor
ADN
Bor (%)
LMA
LOW
VAC
5.4 105
7.5 108
8.9 1010
SMA
7.3 106
0.38
0.84
1.73
0.48
1.30 103
0.63
0.61
0.64
0.65
0.24
0.008
14.3
+0.02
0.07
0.02
From the Tables 25 and also from 2 landscape scans it can be seen that there are some
non-negligible regions in parameter space, not only the best fit points. We are therefore
2
justified in converting 2 into likelihood using the expression L = e /2 , and proceeding
to study the marginalized parameter constraints. This normalized marginal likelihood is
plotted in Fig. 3 for each of the oscillation parameters m2 and tan2 . We present
results corresponding to two cases: the global-rate analysis excluding CC-SNO and the
full analysis including SK energy spectrum. For tan2 we observe that the likelihood
function is concentrated in a region 0.2 < tan2 < 1 with a clear maximum at tan2 0.5
in sharp coincidence with previous results. The situation for m2 is less obvious although
the region at large mass differences is clearly favored. The half width of these curves can
be used as another estimate of the minimum error which can be assigned to the parameters
at the present and near future experimental situation.
4. Borexino implications
Borexino is a real-time detector for low energy (< 1 MeV ) spectroscopy. The
experiments goal is the direct measurement of the 7 Be Solar neutrino flux of all flavors
via neutrino-electron scattering in an ultra-pure scintillation liquid [3235].
As before, in order to estimate the Borexino signal one convolutes the Sun-neutrino
flux with a detector response function which includes the elastic scattering cross sections
e, e e, e , energy resolution and energy-dependent efficiency effects. Weassume
a Gaussian resolution function with an energy dependent width (Eexp ) = 0.048 Eexp +
0.01 MeV . At this stage we suppose unit efficiency over the nominal window of the
experiment 0.3 < E < 0.8 MeV and zero otherwise. The absolute 7 Be flux is taken from
BPB2001 [14] as for the other experiments. After one year of data taking, Borexino [54]
expects roughly 1520 103 Solar neutrino events with a nearly negligible statistical error
stat 0.8% whereas the error in discriminating signal from background is much larger
8% for the same period.
In Table 6 and Fig. 4 we present the expectations for the normalized daynight
signal S daynight at the Borexino experiment for all of the local minima solutions found
in the previous section (SK spectrum plus global rates with constrained flux): S Bor =
S DN (m2 , )/S0 where S0 is the expected signal in absence of oscillations and day
night signals are averaged. We also present the expected daynight asymmetry ADN
2(D N)/(D + N). As we will see below, the asymmetry on the daynight event rates is
405
Fig. 4. The signal at the Borexino experiment as a function of the oscillation parameters. The signal is normalized
to the no-oscillation case. Contours are drawn at S/S0 = 0.5, 0.6, 0.7 with full lines (respective from inside to
outside). Superimposed: the allowed regions from the global rate analysis including CC-SNO (see Fig. 2(b)) and
the 1 allowed regions (here enclosed by the dashed lines) from the 8 Be sensitive Cl experiment alone.
a valuable tool for distinguishing among the different oscillation solutions. In Fig. 4 we
present the expectations for the Borexino experiment as a function of the two-dimensional
oscillation parameters. The expected signal varies mildly in ample regions of the parameter
space. We observe that at the best fit solution, situated at the LMA region, the expected
Borexino normalized signal is S Bor = 0.63. In the whole LMA region (99.7% CL) the
signal varies between 0.5 and 0.7. If we restrict ourselves to the 90% CL allowed region
around the absolute minimum the signal is always within the range 0.60.7. The
asymmetry at the absolute minimum is negligible ADN < 105 , while in the LMA region
which surrounds it, it can reach the 1% level. A similar behavior appears in all the rest
of the parameter space except in the LOW region.
In the LOW region the signal varies between 0.60.7 (90% CL). The variation
increases to cover the range 0.50.8 at 99.7 CL. Here the daynight asymmetry is
expected to be at its maximum. From the table one observes values as high as ADN 30%.
In the VAC region the signal varies between 0.70.8. The variation of the expected
signal in the SMA region is much more abrupt. This prevents us from giving accurate
predictions for this case. The signal passes from a value S Bor 0.2 to S Bor 0.8 in a
very narrow region. As it is shown in Fig. 5, the combined Borexino measurements of
the total event rate with an error below 710% and daynight total rate asymmetry with
a precision comparable to that of SK will allows us to distinguish or at least to strongly
favor the Solar neutrino solutions provided by present data.
406
Fig. 5. Daynight asymmetry |ADN | versus normalized signal at Borexino corresponding to the 90, 95, 99,
99.7% CL allowed regions from the full analysis including global rates and SK spectra. The upper (lower) zones
correspond to the LOW and VAC (LMA) regions in Fig. 2(a). The stars are the expectations of the best fit solutions
in each of the regions.
5. Conclusions
We have analyzed experimental evidence from all the Solar neutrino data available
at this moment including the latest charged current SNO results. For the best solutions
which fit the present data, we have obtained and analyzed the expectations in Borexino. We
considered different combinations of non-redundant data to perform two kinds of analyses:
in the first one we only included total event rates for all the experiments, while in the
second we included global rates for Homestake, GNOGALLEXSAGE and CC-SNO
plus the day and night energy spectrum rates provided by SK. In this last case we studied
the possibility of free or constrained variations of the Boron Solar neutrino flux.
In the simplest case involving global rates only we have observed how the inclusion
of the CC-SNO data causes the strong decrease of the statistical significance of the SMA
solution.
In the most comprehensive case, global rates plus spectrum, the best fit was obtained in
2 /n =
the LMA region with parameters (m2 , tan2 )LMA = (5.4 105 eV2 , 0.38), (min
0.82, d.o.f. n = 38). Solutions in the LOW and VAC regions are still possible although
much less favored. The best possible solution in the SMA region gets a low statistical
significance.
We have analyzed in detail the expectations on the future experiment Borexino for
all the favored neutrino oscillation solutions. The expected Borexino normalized signal
is S Bor = 0.63 at the best-fit LMA solution while the daynight asymmetry is negligible
ADN < 105 . In the VAC region the signal is slightly higher 0.70.8 and the asymmetry
is still practically negligible. In the whole LMA region (99.7% CL) the signal varies
407
between 0.5 and 0.7. In the LOW region the signal is in the range 0.60.7 at 90% CL
while the asymmetry ADN 130%. We conclude that in the near future, after 23 years
of data taking, the combined Borexino measurements of the total event rate with an error
below 510% and daynight total rate asymmetry with a precision comparable to that of
SK will allows us to distinguish or at least to strongly favor the Solar neutrino solutions
provided by present data. Additional information and precision measurement should be
obtained also from Kamland and from future solar neutrino experiments [55].
Acknowledgements
It is a pleasure to thank R. Ferrari for many enlightening discussions and for his
encouraging support without which this work would have not been possible. We thank all
the Borexino group of Milano University and especially M. Giammarchi, E. Meroni and
B. Caccianiga for providing us essential information about the experiment and its signal
discrimination power. One of us (V.A.) would like to thank M. Pallavicini for interesting
and useful discussions about Borexino features and potentiality. We thank also all those
who sent us comments on early versions of this work. We acknowledge the financial
support of the Italian MURST, the Spanish CYCIT funding agencies and the CERN
Theoretical Division. The numerical calculations have been performed in the computer
farm of the Milano University theoretical group (FARMTH).
References
[1] R. Davis, Prog. Part. Nucl. Phys. 32 (1994) 13;
B.T. Cleveland et al., HOMESTAKE Collaboration, Nucl. Phys. (Proc. Suppl.) B 38 (1995) 47;
B.T. Cleveland et al., HOMESTAKE Collaboration, Astrophys. J. 496 (1998) 505526.
[2] Y. Fukuda et al., SuperKamiokande Collaboration, Phys. Rev. Lett. 82 (1999) 2430, hep-ex/9812011.
[3] Y. Fukuda et al., SuperKamiokande Collaboration, Phys. Rev. Lett. 82 (1999) 1810, hep-ex/9812009.
[4] J.N. Abdurashitov et al., SAGE Collaboration, Phys. Rev. Lett. 83 (23) (1999) 4686.
[5] P. Langacker, Talk given at 4th Intl. Conf. on Physics Beyond the Standard Model, Lake Tahoe, CA, 1318
December 1994, hep-ph/9503327, Published in Trieste HEP Cosmol. (1992), 0487522;
P. Langacker, Nucl. Phys. (Proc. Suppl.) B 77 (1999) 241, hep-ph/9811460;
P. Langacker, Talk given at 1th Int. Workshop on Weak Interactions and neutrinos (WIN99), Cape Town,
SA, 2430 January 1999, hep-ph/9905428.
[6] P. Ramond, Nucl. Phys. (Proc. Suppl.) B 77 (1999) 3, hep-ph/9809401.
[7] F. Wilczek, Nucl. Phys. (Proc. Suppl.) B 77 (1999) 511, hep-ph/9809509.
[8] S.M. Bilenky et al., Summary of the Europhysics neutrino Oscillation Workshop Amsterdam, The
Netherlands, 79 September 1998, hep-ph/9906251;
S.M. Bilenky, C. Giunti, C.W. Kim, Int. J. Mod. Phys. A 15 (2000) 625, hep-ph/9902462.
[9] E. Torrente-Lujan, hep-ph/9902339.
[10] P. Aliani, V. Antonelli, M. Picariello, E. Torrente-Lujan, hep-ph/0112101.
[11] G. Fiorentini, F.L. Villante, B. Ricci, hep-ph/0109275.
[12] Q.R. Ahmad et al., SNO Collaboration, Phys. Rev. Lett. 87 (2001) 071301, nucl-ex/0106015.
[13] S. Turck-Chieze, Nucl. Phys. Proc. Suppl. 91 (2001) 73;
E.G. Adelberger et al., Rev. Mod. Phys. 70 (1998) 1265, astro-ph/9805121;
A.S. Brun, S. Turck-Chieze, P. Morel, astro-ph/9806272.
[14] J.N. Bahcall, M.H. Pinsonneault, S. Basu, Astrophys. J. 555 (2001) 990, astro-ph/0010346.
408
[15] J.N. Bahcall, M.H. Pinsonneault, Rev. Mod. Phys. 67 (1995) 781.
[16] J.R. Klein, SNO Collaboration, in: Venice 1999, Neutrino telescopes, Vol. 1, 115125;
A.B. McDonald, SNO Collaboration, Nucl. Phys. Proc. Suppl. 77 (1999) 43.
[17] J. Boger et al., SNO Collaboration, Nucl. Instrum. Methods A 449 (2000) 172, nucl-ex/9910016.
[18] V. Barger, D. Marfatia, K. Whisnant, Phys. Lett. B 509 (2001) 19, hep-ph/0104166.
[19] J.N. Bahcall, P.I. Krastev, A.Y. Smirnov, JHEP 0105 (2001) 015, hep-ph/0103179.
[20] Y. Fukuda et al., SuperKamiokande Collaboration, Phys. Rev. Lett. 81 (1998) 1158;
Y. Fukuda et al., SuperKamiokande Collaboration, Phys. Rev. Lett. 81 (1998) 4279, hep-ex/9805021.
[21] A.W. Poon, SNO Collaboration, nucl-ex/0110005.
[22] M.V. Garzelli, C. Giunti, hep-ph/0111254.
[23] V. Barger, D. Marfatia, K. Whisnant, hep-ph/0106207.
[24] G.L. Fogli, E. Lisi, D. Montanino, A. Palazzo, Phys. Rev. D 64 (2001) 093007, hep-ph/0106247.
[25] P.I. Krastev, A.Y. Smirnov, hep-ph/0108177.
[26] J.N. Bahcall, M.C. Gonzalez-Garcia, C. Pea-Garay, JHEP 0108 (2001) 014, hep-ph/0106258;
J.N. Bahcall, M.C. Gonzalez-Garcia, C. Pea-Garay, hep-ph/0111150.
[27] A. Bandyopadhyay, S. Choubey, S. Goswami, K. Kar, hep-ph/0110307.
[28] S. Choubey, S. Goswami, D.P. Roy, hep-ph/0109017.
[29] A. Bandyopadhyay, S. Choubey, S. Goswami, K. Kar, Phys. Lett. B 519 (2001) 83, hep-ph/0106264.
[30] M. Altmann et al., GNO Collaboration, Phys. Lett. B 490 (2000) 1626.
[31] P. Anselmann et al., GALLEX Collaboration, Phys. Lett. B 285 (1992) 376;
W. Hampel et al., GALLEX Collaboration, Phys. Lett. B 388 (1996) 384;
T.A. Kirsten, Prog. Part. Nucl. Phys. 40 (1998) 8599;
M. Cribier, Nucl. Phys. (Proc. Suppl.) B 70 (1999) 284;
W. Hampel et al., GALLEX Collaboration, Phys. Lett. B 447 (1999) 127.
[32] T. Hagner, BOREXINO Collaboration, Part. Nucl. Lett. 104 (2001) 90.
[33] S.M. Bilenky, T. Lachenmaier, W. Potzel, F. von Feilitzsch, hep-ph/0109200.
[34] C. Arpesella et al., BOREXINO Collaboration, hep-ex/0109031.
[35] E. Meroni, Nucl. Phys. Proc. Suppl. 100 (2001) 42.
[36] E. Torrente-Lujan, Phys. Rev. D 59 (1999) 093006;
E. Torrente-Lujan, Phys. Rev. D 59 (1999) 073001;
E. Torrente-Lujan, Phys. Lett. B 441 (1998) 305;
V.B. Semikoz, E. Torrente-Lujan, Nucl. Phys. B 556 (1999) 353;
E. Torrente-Lujan, Phys. Lett. B 494 (2000) 255, hep-ph/9911458.
[37] I. Mocioiu, R. Shrock, Phys. Rev. D 62 (2000) 053017, hep-ph/0002149;
A. Dziewonski, in: D.E. James (Ed.), The Encyclopedia of Solid Earth Geophysics, Van Nostrand Reinhold,
New York, 1989.
[38] J. Bahcall, M. Kamionkowski, A. Sirlin, Phys. Rev. D 51 (1995) 6146.
[39] M. Passera, Phys. Rev. D 64 (2001) 113002, hep-ph/0011190.
[40] S. Nakamura, T. Sato, V. Gudkov, K. Kubodera, Phys. Rev. C 63 (2001) 034617, nucl-th/0009012.
[41] H. Ishino, Ph.D. Thesis, University of Tokio, 1999;
M. Nakahata et al., SK Collaboration, Nucl. Instrum. Methods 46 (1998) 301.
[42] M. Nakahata et al., SuperKamiokande Collaboration, Nucl. Instrum. Methods A 421 (1999) 113, hepex/9807027.
[43] N. Sakurai, Ph.D. Thesis, December 2000, Constraints of the neutrino oscillation parameters from 1117 day
observation of soloar neutrino day and night spectra in SuperKamiokande.
[44] G.L. Fogli, E. Lisi, Astropart. Phys. 3 (1995) 185.
[45] S. Goswami, D. Majumdar, A. Raychaudhuri, Phys. Rev. D 63 (2001) 013003, hep-ph/0003163.
[46] S. Fukuda et al., SuperKamiokande Collaboration, Phys. Rev. Lett. 86 (2001) 5656, hep-ex/0103033.
[47] S. Fukuda et al., SuperKamiokande Collaboration, Phys. Rev. Lett. 86 (2001) 5651, hep-ex/0103032.
[48] A.I. Abazov et al., SAGE Collaboration, Phys. Rev. Lett. 67 (1991) 3332;
D.N. Abdurashitov et al., SAGE Collaboration, Phys. Rev. Lett. 77 (1996) 4708;
J.N. Abdurashitov et al., SAGE Collaboration, Phys. Rev. C 60 (1999) 055801, astro-ph/9907131;
J.N. Abdurashitov et al., SAGE Collaboration, Phys. Rev. Lett. 83 (1999) 4686, astro-ph/9907113.
[49] D.E. Groom et al., Particle Data Group Collaboration, Eur. Phys. J. C 15 (2000) 1.
[50] M. Apollonio et al., CHOOZ Collaboration, Phys. Lett. B 466 (1999) 415, hep-ex/9907037;
M. Apollonio et al., Phys. Lett. B 420 (1998) 397;
F. Boehm et al., Phys. Rev. D 62 (2000) 072002, hep-ex/0003022.
[51] K. Lande, For the Homestake Collaboration, Nucl. Phys. (Proc. Suppl.) B 77 (1999) 1319.
[52] J.N. Bahcall, P.I. Krastev, E. Lisi, Phys. Rev. C 55 (1997) 494;
B. Fad, G.L. Fogli, E. Lisi, D. Montanino, Astropart. Phys. 10 (1999) 93.
[53] V. Berezinsky, hep-ph/0108166.
[54] M. Giammarchi, Borexino Collaboration, private communication.
[55] A. Strumia, F. Vissani, JHEP 0111 (2001) 048, hep-ph/0109172;
A. De Gouvea, C. Pea-Garay, Phys. Rev. D 64 (2001) 113011, hep-ph/0107186.
409

Erratum
Erratum to: Quartic mass corrections to Rhad

at O(s3)
[Nucl. Phys. B 586 (2000) 5672]
K.G. Chetyrkin a , R.V. Harlander b , J.H. Khn c
a Fakultt fr Physik, Albert-Ludwigs-Universitt Freiburg, D-79104 Freiburg, Germany
b TH Division, CERN, CH-1211 Geneva 23, Switzerland
c Institut fr Theoretische Teilchenphysik, Universitt Karlsruhe, D-76128 Karlsruhe, Germany
Received 3 May 2002
As shown in Ref. [1], a wrong value for the anomalous dimension 0 of the vacuum
energy E0 , defined by [2]
2
d
E 0 = 0 m4 ,
d2
(1)
which is an important ingredient in the calculation, had been adopted in Ref. [3].
The new value for the vacuum anomalous dimension leads to corrections in Eqs. (9), (10)
and in Table 2 of our original paper [3]. The numerical effect is too small to be visible in
any of the plots. The quantities that must be corrected are:
2 2 3
mQ
s
(v)
(v)
,
rQ,4 new = rQ,4 old + 24
s
2 2 3
mQ
s
(a)
(a)
rQ,4 new = rQ,4 old 24
(2)
.
s
In both cases, the effect on the coefficient of the (s /)3 (m2Q /s)2 term is around 0.5%.
Table 2 of Ref. [3] changes as follows:
Table 2
(v)
rQ for s = 10.5 GeV
s0
m0
m2
m4
1.0
0.0
0.0001477
s1
s2
s3
0.05482
0.003264
0.00002969
0.004581
0.001628
0.00001245
0.001898
0.000519
0.00002026
1.058
1.063
1.063
PII of original article: S0550-3213(00)00393-X.

E-mail address: robert.harlander@cern.ch (R.V. Harlander).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 5 3 - X
414
K.G. Chetyrkin et al. / Nuclear Physics B 634 (2002) 413414
References
[1] K.G. Chetyrkin, in preparation.
[2] V.P. Spiridonov, K.G. Chetyrkin, Yad. Fiz. 47 (1988) 818; Sov. J. Nucl. Phys. 47 (1988) 522;
K.G. Chetyrkin, J.H. Khn, Nucl. Phys. B 432 (1994) 337.
[3] K.G. Chetyrkin, R.V. Harlander, J.H. Khn, Nucl. Phys. B 586 (2000) 56.
Nuclear Physics B 634 [FS] (2002) 417432

HCIZ integral and 2D Toda lattice hierarchy

P. Zinn-Justin
Laboratoire de Physique Thorique et Modles Statistiques, Universit Paris-Sud, Btiment 100,
91405 Orsay Cedex, France
Received 8 April 2002; accepted 6 May 2002
Abstract
The expression of the large N Harish ChandraItzyksonZuber (HCIZ) integral in terms of
the moments of the two matrices is investigated using an auxiliary unitary two-matrix model, the
associated biorthogonal polynomials and integrable hierarchy. We find that the large N HCIZ integral
is governed by the dispersionless Toda lattice hierarchy and derive its string equation. We use this to
obtain various exact results on its expansion in powers of the moments. 2002 Published by Elsevier
Science B.V.
1. Introduction and definitions

The Harish ChandraItzyksonZuber integral [1,2] plays a key role in matrix models
[26], in particular in relation to (discretized) two-dimensional quantum gravity. It also
displays interesting connections [7,8] with free probability theory [9,10]. It is defined by
d eN tr AB ,
IN (A, B) =
(1.1)
U (N)
where A and B are N N matrices and d is the normalized Haar measure on U (N).
There are many ways to evaluate this integral [1,2,4,1113]; the result depends only on the
eigenvalues ai of A and bi of B, and is given by
IN (A, B) =
det(eNai bj )
.
(ai )(bj )
(1.2)

Here () denotes the Vandermonde determinant: (ai ) = i<j (ai aj ). In all equations,
constants that only depend on N will be ignored; they can be restored when required.
E-mail address: pzinn@ipno.in2p3.fr (P. Zinn-Justin).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 7 4 - 7
418
P. Zinn-Justin / Nuclear Physics B 634 [FS] (2002) 417432
For practical applications it is important to look at the large N limit of the HCIZ integral:
it corresponds to the planar limit of matrix models. One considers a sequence of N N
matrices A and B, whose spectral distributions converge to some fixed distributions as N
goes to infinity. It is known (and in fact, proved rigorously [14]) that the following quantity,
the large N free energy, is well-defined:
log IN (A, B)
N2
and depends on the spectral distributions of A and B only through their moments:
F [q , q ] = lim
(1.3)
tr q
tr
A ,
(1.4)
q = Bq , q 1.
N
N
We also define for future use the generating series of these moments, that is the resolvents
of A and B:

tr 1
= 1/a +
G(a) lim
(1.5a)
q /a q+1 ,
N N a A
q =
q1

lim tr 1 = 1/b +
G(b)
q /bq+1 .
N N b B
(1.5b)
q1
As a function of the q and the q , F has a power series expansion around 0. We shall
investigate here the determination of the coefficients of this expansion, as well as the more
general question of the evolution of F as one varies the q and q . We shall see that this
evolution is governed by an infinite set of partial differential equations, a particular scaling
of the Toda lattice hierarchy.
More precisely, we shall rewrite in Section 2 the HCIZ integral as a unitary twomatrix model, then show in Section 3 that the latter satisfies the Toda lattice hierarchy;
in Section 4, we shall take the large N limit which relates F to the dispersionless Toda
lattice hierarchy; finally, we shall apply this technology to the expansion of F in Section 5.
Section 6 is devoted to a summary of results.
2. HCIZ integral as a unitary two-matrix model

We shall first use a standard trick to rewrite the HCIZ integral as an integral over two
matrices. The derivation can be performed either by computing residues in the two-matrix
integral, or by character expansion; we choose the latter. We thus start with the following
identity, valid for any N N matrix M:

eN tr M =
(2.1)
cR dR R (M).
R0
Let us explain the symbols used. R stands for an analytic irrep of GL(N) with positive
highest weights, or the associated Young diagram; more explicitly, it can be described by
its shifted highest weights hi , 1 i N which form a strictly decreasing sequence (they
are related to the usual highest weights mi by hi = mi + N i). R (M) is the character
419
associated to the irrep R and the matrix M. cR and dR are coefficient of the expansion:
N |R|
cR = N
(2.2)
,
dR = (hi ),
i=1 hi !

where |R| = i hi N(N 1)/2 is the number of boxes of R. These explicit expressions
of the coefficients play no role in what follows since the trick we use is quite general.
Applying the identity (2.1) to the definition of the HCIZ integral (1.1) and using
orthogonality relations for matrix elements of irreps yields

cR R (A)R (B).
IN (A, B) =
(2.3)
R0
h
As a side remark, we note that using Weyls formula for characters (R (A) = det(ai j )/
(ai )), recombining the two resulting determinants into one single determinant and
performing the summation, one recovers Eq. (1.2).
We now want to compare expression (2.3) with the partition function of a matrix model
of two N N matrices U and V:

q
q
1 1
N [tq , tq ] =
(2.4)
dU dV e q1 tq tr U + q1 tq tr V +tr U V .
Here the tq are as yet unknown coefficients; they must however satisfy supq |tq |1/q < m <
< ; this renders the summation convergent provided that the
and supq |tq |1/q < m
spectral radii of U and V are less or equal to 1/m, 1/m.
Since the integrand is analytic in
U and V the exact contours of integration are irrelevant; for the sake of definiteness, we
2 .
choose U and V to be unitary up to a multiplicative constant: UU = m2 , VV = m
The measure of integration is simply the normalized Haar measures for mU and mV.
All
these details can be safely ignored if one remembers that in the integral (2.4) only the
contribution coming from poles at the origin should be picked up.
Note that we have defined (2.4) so that N does not appear anywhere explicitly in the
action. It is however convenient to rescale the matrices to restore the usual N dependence
of matrix models:

q/2 tr Uq +
q/2 tr Vq +N tr U1 V1
q1 tq N
dU dV e q1 tq N
.
N [tq , tq ] =
(2.5)
Next we apply one of the forms of the Cauchy identity:

q/2 tr Uq
e q1 tq N
=
sR tq N q/2 R (U),
(2.6)
R0
where sR is the Schur function associated to the Young diagram R.

Plugging twice (2.6) as well as the identity (2.1) into (2.5) results in:

sR1 tq N q/2 sR2 tq N q/2 cR dR
N [tq , tq ] =
R,R1 ,R2 0

dU dVR1 (U)R2 (V)R U1 V1 .
(2.7)
420
Again, orthogonality relations imply that

cR sR tq N q/2 sR tq N q/2 .
N [tq , tq ] =
(2.8)
R0
#rowsN
Note that we have written explicitly the constraint that the Young diagram R should have
no more than N rows, since contrary to prior expressions the summand in Eq. (2.8) is not
necessarily zero if the number of rows exceeds N .
Comparing Eqs. (2.8) and (2.3), we see that the two expressions are equal (up to a
numerical factor) iff
tq N q/2 =
1
tr Aq ,
q
tq N q/2 =
1
tr Bq
q
(2.9)
or in terms of the normalized moments introduced before

tq = N q/2+1
q
,
q
tq = N q/2+1
q
.
q
(2.10)
The constraint on the number of rows of R is automatically satisfied if the tq are of the
form (2.9). However, the expression (2.8) has the advantage that it is defined for arbitrary
(independent) tq at finite N , which is not the case of the original HCIZ integral (where there
are only N independent tq ). Also note that the condition of finiteness of supq1 |tq |1/q and
supq1 |tq |1/q is automatically satisfied at finite N if tq is of the form (2.9), and remains
true as N provided the spectra of A and B remain bounded.
3. Biorthogonal polynomials and Toda lattice hierarchy

We have found in the previous section that calculating the HCIZ integral is equivalent to
solving the matrix model given by Eq. (2.4). A convenient way to do so, at least formally, is
via biorthogonal polynomials. This is standard material, and is very similar to derivations
found in the literature [15]. N will then be identified with a tau function of the 2D Toda
lattice hierarchy.
3.1. Setup
We start with the following determinant form of the partition function (2.4):

du dv j i q1 tq uq +q1 tq v q +u1 v 1

u ve
N = det
,
2iu 2iv
0i,j N1
(3.1)
where once again the contours of integrations are sufficiently small circles around the
origin.
This suggests to introduce a non-degenerate bilinear form on the space of polynomials
by

q
du dv
q 1 1
p(u)q(v)e q1 tq u + q1 tq v +u v
q|p =
(3.2)
2iu 2iv
421
so that N = det MN , with MN = (mij ), 0 i, j N 1 and mij = v i |uj .

Next we define normalized biorthogonal polynomials qn (v) and pn (u) with respect to
the bilinear form above on C[v] C[u], that is qn |pm = nm . We write:
pn (u) =
n
pkn uk ,
qn (v) =
k=0
n
qkn v k .
(3.3)
k=0
There is an arbitrariness in the normalization which is fixed by assuming pnn = qnn h1

n .
Define upper triangular matrices PN = (pkn ) and QN = (qkn ), 0 n, k N 1; and
1
PN = QTN . Orthonormality of these polynomials is equivalent to writing
QTN MN PN = 1
(3.4)
or MN = PN PN1 , with PN1 upper triangular and PN lower triangular. In particular

2
N = det MN = N1
i=0 hi .
Next, introduce the semi-infinite matrices M = M , P = P , Q = Q , P = P and
the shift matrix Z: Zij = ij +1 . Set
U = P 1 ZP ,
V = Q1 ZQ,
(3.5)
U (respectively, V ) is the matrix of multiplication by u (respectively, v) in the basis (pn (u))

V T = P1 Z T P. By definition of the
of C[u] (respectively, (qn (v)) of C[v]). Also set U
mij , we have
M
= MZ q ,
tq
M
q
= Z T M.
tq
(3.6)
P 1 ,
We then easily find, using M = P
P 1

P
P
+ P1
= Uq,
tq
tq
P 1

P
P
q .
+ P1
=U
tq
tq
(3.7)
P
P
is upper triangular, whereas P1 t
is lower triangular, and they have opposite
P 1 t
q
q
diagonal elements (and similarly for derivatives with respect to tq ); so that if one defines
symbols ( )+ = ( )>0 + 12 ( )0 and ( ) = ( )<0 + 12 ( )0 for lower and upper
diagonal parts respectively (it is convenient to include one half of the diagonal part, though
this does not make them projections), we have

q
P
P

(3.8a)
= P U q ,
= P U
,
tq
tq
P q
P q
(3.8b)
= P U +,
=P U +
tq
tq
and finally

U
= U q +, U ,
tq

U
,
= U q +, U
tq

U q
= U , U ,
tq
q

U

.
= U
,U
tq
(3.9a)
(3.9b)
422
Eqs. (3.9) are equivalent to equations of ZakharovShabat type [16]

r
U +
tq
r
U ++
tq

q
U + + U q + , U r + = 0,
tr
q r
q

U U
, U + = 0,
tr
q r
r
q

U
U U
= 0,
, U
tq
tr
(3.10a)
(3.10b)
(3.10c)
in the following way:

which ensure compatibility of Eqs. (3.8). If we expand U and U
U = rZ +
uk Z Tk ,
(3.11a)
k=0
= ZTr +
U
Z k vk ,
(3.11b)
k=0
where r = diag(rn ), the uk = diag(uk;n ) and the vk = diag(vk;n ) are diagonal matrices,
then Eqs. (3.10) are the (semi-infinite) Toda lattice hierarchy of differential equations for
.
the coefficients of U and U
3.2. Examples
The simplest equation, the usual Toda lattice equation, which is the case q = r = 1 in
Eq. (3.10b), can be obtained directly from the determinant form (3.1) by applying Jacobis
2
determinant identity, which yields n+1 n1 = n t t n t1 n t n , or
1
2
n+1 n1
=
log n .
n2
t1 t1
(3.12)
The left-hand side is nothing but (hn / hn1 )2 = rn2 . This implies
2
2
rn+1
+ rn1
2rn2 =
2
log rn2
t1 t1
(3.13)
which is the Toda lattice equation in terms of the log rn2 . In the next section we shall see
how to use this equation in the large N limit.
Another equation one can derive (case q = 2, r = 3 of Eq. (3.10a)) is the usual
2
KadomtsevPetviashvili (KP) equation, which is given in terms of = 2 2 log N :
t1
3
4
= 0.
+
+
6
+
t1
t3
t1
t22
t13
(3.14)
423
3.3. String equation

So far we have not used the specific form of the interaction between our two matrices
in (2.4). This occurs when one tries to determine the string equation, that is an additional
relation which fixes a particular solution of the Toda hierarchy. Here it is very easy to derive
the string equation by arguments similar to those used in the standard two-matrix model.
We write that

q
du dv
d
q 1 1
0=
(3.15)
u
pj (u)qi (v)e q1 tq u + q1 tq v +u v
2iu 2iv du
expand and obtain

1 U 1 = 0,
qtq U q U
UD +
(3.16)
q1
where D is the derivative with respect to u in the basis pn (u), and U 1 is a particular
1 is a
left inverse of U whose matrix elements are defined by Eq. (3.15), and similarly U
1

particular right inverse of U . Using [D, U ] = 1 and U U = 1, we derive from Eq. (3.16)
the string equation:
1 1

= 1.
U ,U
(3.17)
Let us note that Eq. (3.17) is not the usual string equation of the hermitean two-matrix
model.
4. Large N limit
4.1. Classical/dispersionless limit
Next we want to consider the large N limit. As always in matrix models, this identifies
with a classical limit for the operators involved, the small parameter being h 1/N .
, Z become classical variables u, v,
Therefore in this limit, the various operators U , U
z, up to some rescalings which will be detailed later.
What happens to the Toda equation? The large N limit corresponds to the dispersionless
limit of the Toda hierarchy [17,18]. In this limit, operators of discrete difference in the
index n (e.g., Eq. (3.13)) can be replaced with a differential operator / with n/N .
Furthermore, the logarithm of the tau-function N is required to be of order N 2 (cf. our
Eq. (1.3)), with a smooth 1/N expansion,1 which leads to some simplifications in the
differential equations. We shall only write down a few examples of these dispersionless
equations; we refer the reader to, e.g., [17,18,20] for a complete and more rigorous
description of the dispersionless hierarchy. Let us simply mention that, as mentioned above,
1 The assumption of smoothness is less innocent than it seems; it is the equivalent of the single cut
hypothesis in matrix models (cf. a similar remark in [19] the context of 1D Toda). However, in formal power
series around tq = tq = 0 this hypothesis is certainly valid.
424
we are dealing with a classical limit, that is the commutators become of order h 1/N ,
and their leading term defines a Poisson bracket. Eqs. (3.9)(3.17) can be rewritten in the
large N limit, once rescaled, by simply replacing commutators with Poisson brackets. In
particular, the string Eq. (3.17) becomes:
1 1
=1
u ,v
(4.1)
(where now u1 and v 1 are ordinary inverses since in the large N limit all operators are
invertible). It is worth noting that the same string equation occurs in the interior Dirichlet
boundary problem [21]. Also, it is identical to the string equation of [22] in which the
radius has the unphysical value R = 1.
4.2. Scaling. Some examples
In order to give some examples, we need to give some additional details on the particular
scaling of the variables that is appropriate for our model. First, we have for our operators
, Z: (cf. the rescaling from Eq. (2.4) to Eq. (2.5)):
U, U
U N 1/2 u,
N 1/2 v,
U
Z z,
(4.2)
where u, v, z have a finite limit when N and are related by:

u = rz +
uk zk ,
k=0
v = rz1 +
(4.3a)
vk zk
(4.3b)
k=0
in terms of appropriately rescaled coefficients (in particular rN rN 1/2 ). We have

already derived the rescaling of the moments, cf. Eq. (2.10).
In what follows, we shall freely change from original to rescaled variables, using the
same letters by abuse of notation. In most circumstances this does not cause any confusion.
For example, let us see what happens to the KP Eq. (3.14). In the large N limit,
performing all the rescalings one term drops out and we find

2
+
2
=0
2 2 +
(4.4)
1
3
1
2
which is just the dispersionless KP equation (or KhokhlovZabolotskaia equation).
Let now see what happens to the Toda lattice equation (3.12) in the large N limit. Here,
it is important to notice that an extra ingredient is required if one considers equations in
which derivatives with respect to n appear. It is our scaling hypothesis, based on the fact
that we consider only the planar large N limit (first term in the 1/N expansion) and on
the scaling of the tq and tq . Considering Eq. (1.3), we write that:
2 F [nq/21 qt
n = c n e n
q/21 q t ]+O(1)
q ,n
q
(4.5)
The constant cn can be determined by setting tq = tq = 0 in Eq. (3.1); one easily finds

1
cn = ( n1
i=0 i!) .
425
More explicitly, if one introduces the dispersionless tau function F [, q , q ] =

limN ( N12 log n + 12 2 log N 34 ) with = n/N , then we have the scaling

1
3
F [, q , q ] = 2 log + 2 1 + 2 F q q/21 , q q/21 ,
2
4
where our old function F [q , q ] is recovered by setting = 1.
We now rewrite the Toda equation (3.12):
2
e 2
2
1 1
F,
(4.6)
(4.7)
which is just the dispersionless Toda equation (integrated twice); then apply the scaling
(4.6). The final result is the following partial differential equation for F :

q

exp 2F +
+ 1 q 2 q q F + q q F
2
q1
q
2
+1
r
2

+ 1 q r q r F + 2q r q r F + q r q r F = 1 1 F
(4.8)
q,r1
with q /q and q / q . This is in general a fairly complicated equation.

Application
Assume that all q and q except 1 and 1 are zero. This greatly simplifies Eq. (4.8).
Note furthermore that by obvious homogeneity property, F only depends on the product
x 1 1 . Rewriting Eq. (4.8) for F (x) leads to an ordinary differential equation:
e2F +9x
2 F
= F + xF
(4.9)
which can be solved: if = xdF /dx, then is the solution of the third degree equation
16 3 + 8 2 + (1 36x) + x(27x 1) = 0
(4.10)
that vanishes at x = 0. We have the following expansion:

=

n=0
x n+1
(3n)!2n
.
(n + 1)!(2n + 1)!
n
(4.11)
(3n)!2
Curiously, the number (n+1)!(2n+1)!
also appears in the context of the enumeration of planar
2
maps; it counts rooted maps with 2n nodes [23]. This suggests that a direct combinatorial
proof of this result might be possible. As a corollary, the HCIZ integral with 1 , 1 = 0 is
in the universality class of pure 2D quantum gravity: the singularity of (x) is of the form
= (x xc )3/2 + (with xc = 2/27, c = 1/12).
More generally, if one turns on a finite number of q and q , one can show using the
formalism developed in this section that F can be expressed in terms of the solution of an
2 The author would like to thank J.-B. Zuber for pointing this out to him.
426
algebraic equation. This is because the string equation implies that a = 1/u and b = 1/v
are polynomials in z1 , z:
a=
2+1
q zq ,
q=1
b=
2+1
q z q ,
(4.12)
q=1
where 2 = max{q | q = 0}, 2 = max{q | q = 0}, and as will be shown explicitly in the
sections below (cf. Eq. (4.18)), the knowledge of a and b allows to compute derivatives of
the free energy.3
For example, if only 1 , 2 , . . . , 2 , 1 are non-zero, by homogeneity one can set 1 = 1;
2
2
then, = 2 F2 2 4 , and r 2 = F are given by the degree 22 + 1 equation
1
1 + +
1 1
2

q=1
(1)q
(2q)!
q 2q+1
(q!)2
= 1 + 21 3 + 62 5 203 7 + = 0
(4.13)
with the initial condition = 1 for all q zero. One can check that if 2 3, = 2 4 does
satisfy the partial differential equation (4.4).
It is expected that by turning on an arbitrary number of q and q , one should be able to
recover the critical behavior (i.e., singularities of F and its derivatives) of all c < 1 minimal
models coupled to gravity.
4.3. Derivatives of F with respect to the tq
Our goal is now to express the first few derivatives of F with respect to the tq and the
tq . Let us define the differential operator
aq
(a) =
(4.14)
.
q tq
q1
In
language,
the abosonic
this isqthe positive modes of the full bosonic field (a) =
q

+
log
a
q1 q tq
q1 tq a . One could define similarly (b).
Our first task is to compute (a)F (see also [7,24] for a saddle point approach to this
question). We note that in terms of the matrix model (2.4), (a)F = limN tr log(1
aU), or
d
(a)F = u2 D(u) u,
da
(4.15)
1
is the resolvent of U, and with the identification u 1/a
where D(u) limN tr uU
(we recall that appropriate scaling as N is implied). Next we use the following
standard identity:
pN (u) = h1
N det(u U),
3 Note that since a, b have a rational parameterization, the curve (a, b) has genus zero.
(4.16)
427
where the average is again over N N matrices with the measure given by (2.4). This
expression of pN (u) can be proved by checking that qn |pN = nN using it. Considering
the logarithmic derivative with respect to u of Eq. (4.16) in the large N limit, we see that
D(u) is simply the operation of derivative with respect to u = 1/a in the basis of the pn (u).
This operator was already introduced in the discussion of the string equation; Eq. (3.16)
becomes, in the large N limit,

D(u) = u2 v 1 G(u) + u ,
(4.17)

where we recall that G(u) = u + q1 qtq uq+1 is the resolvent of the original matrix A.
Combining Eqs. (4.15) and (4.17) leads to
d
(4.18)
(a)F = b(a) G(a),
da
where b = 1/v, a = 1/u and u and v related by Eqs. (43), i.e., b(a) is obtained
by composing the series b(z) = 1/v(z) and the inverse series z(a) of a(z). Note that
d
(a) in
b(a) G(a) is simply the non-negative part of the Laurent expansion of b(a) = da
powers of a.
Note that to derive the identity (4.18), we had to use the string equation under the form
(3.16). In contrast, the second derivative (a1 )(a2 )F is universal, i.e., in our context
it means that it is true for any solution of the dispersionless Toda hierarchy regardless of
the string equation. Derivation of (a1 )(a2 )F requires two identities. We shall prove
the first one, which is an expression for z. We note that since Z is the shift operator, one
should have in the large N limit z = Pn+1 (u)/Pn (u), that is taking the logarithm and using
Eq. (4.16),
log z = log u log r
uq
q1
2
F,
q tq
(4.19)
where it is recalled that the dependence on of F is given by Eq. (4.6).

The second relation we use the dispersionless limit of the Fay identity [18], which we
shall not prove here and write under the form (see [25]):
(u1 u2 )e
u
q,r1
q r
u
2
2
1
q
r tq tr
= u1 e
q 2
q1 u1 tq
u2 e
q 2
q1 u2 tq
(4.20)
Combining Eqs. (4.19) and (4.20), using the variables a1 = 1/u1 , a2 = 1/u2 and the power
series inverse z(a) of a(z) leads to
(a1 )(a2 )F = log
z(a1) z(a2 )
+ log r
1/a1 1/a2
(4.21)
(cf. also a similar formula in [3]).

One can prove a formula for mixed derivatives: (a)(b)F
= log(1 z(b)/z(a)),
but it will not be needed here.
Formulae for higher numbers of derivatives become more and more complicated, (see
in particular residue formulae for three derivatives [20]), and we shall not write them
down here.
428
5. Expansion around 0 of the HCIZ integral

We now come to the main goal of this paper: to apply the equations above to the
expansion of the large N HCIZ free energy F [q , q ] in powers of q and q . We shall
consider the problem in an asymmetric way: we shall apply the operator (a) defined
in the previous section to the free energy, that is consider derivatives with respect to the
q , and then set q = 0 and see how the result depends on the q . The identities obtained
above involve the functions b(a) = 1/v(a = 1/u) and z(a); let us see how to compute
these functions when all tq are zero.
If we set tq = 0 in definition (3.2) of the bilinear form, we see that no positive powers
of u are generated, so that mij = v i |uj satisfies

mij =
0,
1
j ! sj i ,
j < i,
j i,
(5.1)

where we have defined sq by (v) exp( q1 tq v q ) = q0 sq v q ; they are Schur
functions corresponding to single row Young diagrams.
P 1 with P upper
Let us first use the fact that M is upper triangular. Since M
=P
n
is diagonal, i.e., qn (v) = n! v . The normalization
triangular and P lower triangular, P
hn = (n!)1/2 implies that rn = n1/2 , that is after rescaling r = 1. Finally, we conclude
from the definition (3.5) of V that all the coefficients vk in its expansion (or of its
, Eq. (3.11) are zero, or in the large N limit
transpose U
v = z1
(5.2)
(note that this is consistent with Eq. (4.12)). We have now reduced the problem to a single
function b(a) 1/v(a) = z(a); its calculation was performed in [7] using a saddle point
method, but it is particularly simple to rederive it in our framework. We start again from
Eq. (5.1) and notice that one can easily invert this matrix, which
of
yields the coefficients

P = M 1 P. If we introduce the eq defined by (v)1 exp( q1 tq v q ) = q0 eq v q
(they are (1)q times the Schur functions corresponding to one-column Young diagrams),
then expansion of the identity (v)(v)1 = 1 leads to

pij =
0,
i! ej i ,
j!
j < i,
j i.
(5.3)
We now define the matrix A = (aij ) by A = P 1 Z T P , which according to definitions (3.5) is a left inverse of U . We have

aik =

=
i!
k!

j,ij k1
1
sj i (j + 1)!ek1j
j!
k1i
i!
(q + i + 1)sq ek1iq .
k!
q=0
(5.4)
This time we recognize the expansion of (v)(v)1 =

we also have (v)(v)1 = q1 ntn v n1 , so that
n0 v
k < i + 1,
0,
k,
k
= i + 1,
aik =
i!
k! (k i 1)tki1 , k > i + 1.
n1
429
n
q=0 q sq enq ;
but
(5.5)
This is an exact expression at finite N . In the large N limit this can be simply rewritten,
performing the appropriate rescalings and remembering that z = b,
a=
1 q
+
.
b
bq+1
(5.6)
q1

We recognize the resolvent of the original matrix B (Eq. (1.5)): a(b) = G(b)
(again
d

this is consistent with the dual equation of (4.18), db
(b)F = a(b) G(b)).
Finally
b(a = 1/u) is the functional inverse of a(b); it is well-known that b(a) is nothing but the
generating series of free cumulants m
q of B [10] (cf. also the expression of connected
correlation functions in matrix models, as in Eq. (31) of [26]):
b(a) =
1
+
m
q a q1 ,
a
(5.7)
q=1
where m
q = q
q
2!

i,j1
i+j=q
i j +
q(q+1)
3!

i,j,k1
i+j+k=q
i j k + .
now apply the formulae obtained in previous section. We recall that (a)
We acan
q
q1 q tq . First we rewrite Eq. (4.18) for tq = 0:
1
d
(a)F = b(a) =
m
q a q1 .
da
a
(5.8)
q=1
This equation can be found in [7], but it can already be extracted from [2]; see also [8] for
a combinatorial proof.
The second derivatives are given by Eq. (4.21) with z(a) = b(a):
b(a1) b(a2 )
1/a1 1/a2

q1

k qk
= log 1
m
q
a1 a2
.
(a1 )(a2 )F = log
q=2
(5.9)
k=1
Since the expression is in fact known for all tq , one can differentiate once more and obtain
the expression for the third derivatives at tq = 0 with little extra work. First one computes
the expression of b(z) at first order in tq via Eq. (3.9b); one obtains after inversion

tq
cqk bk+1 + ,
tq 0,
rz = b +
(5.10)
q1
k0
430
where the coefficients cqk are best expressed using their generating function
aq
db(a)
cqk =
b(a)k2.
q
da
(5.11)
q1
Secondly one uses Eqs. (4.18) and (4.21) to deduce

(a2 )b(a1 ) =

log z(a1 ) z(a2 ) .

a1
(5.12)
Acting with (a3 ) on Eq. (4.21), setting tq = 0, using identities (5.10)(5.12) and
performing simple manipulations results in
(a1 )(a2 )(a3 )F =
db(a1)/da1
(b(a1) b(a2))(b(a1 ) b(a3))
db(a2)/da2
+
(b(a2) b(a1))(b(a2 ) b(a3))
db(a3)/da3
+
+ 1.
(b(a3) b(a1))(b(a3 ) b(a2))
(5.13)
One can go to higher derivatives by further differentiation, but most likely, the expressions
become more and more cumbersome.
6. Summary of results
In this paper we have investigated the expansion of the large N HCIZ integral free
energy in terms of powers of the moments of the matrices. We have shown that this free
energy is the dispersionless tau function for the dispersionless 2D Toda hierarchy of partial
differential equations, so that coefficients of the expansion satisfy an infinite set of nonlinear relations. Also, we have seen that the solution of the Toda hierarchy is selected by a
non-trivial string equation {u1 , v 1 } = 1. The following explicit results have been found:
First, one can compute some infinite series of coefficients of the expansion using
these differential equations. The example worked out in section 4 is the set of coefficients
of 1k 1k ; if x = 1 1 , then = xF /x was found to satisfy the equation 16 3 + 8 2 +
(1 36x) + x(27x 1) = 0, that is
1
4
176
(1 1 )5 + .
F = 1 1 + (1 1 )2 + (1 1 )3 + 6(1 1 )4 +
2
3
5
(6.1)
More generally, one can compute generating series of coefficients involving only a
finite number of distinct q and q as solutions of algebraic equations; a more complicated
example was given without proof, cf. Eq. (4.13). In principle, this allows to compute any
given coefficient required, although it does not give a closed expression for the whole
function of the infinite set of q and q .
Secondly, one can obtain all the coefficients (as functions of the q ) of the first
few orders of the expansion in products of q : we rewrite their generating functions here
431
(Eqs. (5.8), (5.9) and (5.13)):

1
d
(a)F = b(a) ,
(6.2.1)
da
a
b(a1) b(a2)
(a1 )(a2 )F = log
(6.2.2)
,
1/a1 1/a2
db(a1)/da1
(a1 )(a2 )(a3 )F =
(b(a1) b(a2 ))(b(a1) b(a3))
db(a2)/da2
+
(b(a2) b(a1 ))(b(a2) b(a3))
db(a3)/da3
+ 1,
(6.2.3)
+
(b(a3) b(a1 ))(b(a3) b(a2))

where in the rescaled variables, (a) = q1 a q q . The coefficients of this expansion

are functions of the q via the free cumulants m
q : b(a) = 1/a + q1 m
q a q1 . These can
be reexpanded themselves in powers of q , so that we find for example
F=
1
q q
q
q1
1
2
1
2
1
4
1
6
1
6
q r1 r2
q,r1 ,r2 1
r1 +r2 =q
q1 q2 r
r,q1 ,q2 1
q1 +q2 =r

q1 + q2 min(q1 , q2 , r1 , r2 ) + 1 q1 q2 r1 r2
q1 ,q2 ,r1 ,r2 1

q1 +q2 =r1 +r2
(q + 1)q r1 r2 r3
q,r1 ,r2 ,r3 1

q=r1 +r2 +r3
(r + 1)q1 q2 q3 r
r,q1 ,q2 ,q3 1

r=q1 +q2 +q3
(6.3)
with the appropriate symmetry of interchange of the q and q .
Acknowledgements
The author would like to thank P. Biane, B. Collins, B. Eynard and especially I. Kostov
and J.-B. Zuber for discussions.
432
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
H. Chandra, Amer. J. Math. 79 (1957) 87120.

C. Itzykson, J.-B. Zuber, J. Math. Phys. 21 (1980) 411.
J.M. Daul, V. Kazakov, I. Kostov, Nucl. Phys. B 409 (1993) 311, hep-th/9303093.
M.L. Mehta, Commun. Math. Phys. 79 (1981) 327.
V. Kazakov, Phys. Lett. A 119 (1986) 140;
D.V. Boulatov, V. Kazakov, Phys. Lett. B 186 (1987) 379.
P. Di Francesco, P. Ginsparg, J. Zinn-Justin, Phys. Rep. 254 (1995) 1133.
P. Zinn-Justin, Phys. Rev. E 59 (1999) 4884, math-ph/9810010.
B. Collins, Moments and cumulants of polynomial random variables on unitary groups, the ItzyksonZuber
integral and free probability, preprint.
D.V. Voiculescu, K.J. Dykema, A. Nica, Free Random Variables, American Mathematical Society, 1992.
R. Speicher, Math. Ann. 298 (1994) 611.
M. Stone, Nucl. Phys. B 314 (1989) 557.
D. Altschuler, C. Itzykson, Ann. Inst. Henri Poincar 54 (1991) 1.
S.L. Shatashvili, Commun. Math. Phys. 154 (1993) 421, hep-th/9209083.
A. Guionnet, O. Zeitouni, Large deviations asymptotics for spherical integrals, preprint.
M. Adler, P. van Moerbeke, Commun. Pure Appl. Math. 50 (1997) 241, hep-th/9706182.
K. Ueno, K. Takasaki, in: H. Morikawa, K. Okamoto (Eds.), Adv. Stud. Pure Math. 4, 1984, p. 1.
K. Takasaki, T. Takebe, Lett. Math. Phys. 28 (1993) 165176, hep-th/9301070.
K. Takasaki, T. Takebe, Rev. Math. Phys. 7 (1995) 743808, hep-th/9405096.
P. Zinn-Justin, Phys. Rev. E 62 (2000) 3411, math-ph/0005008.
I.M. Krichever, Funct. Anal. Appl. 22 (1989) 200213;
I.M. Krichever, Commun. Pure Appl. Math. 47 (1992) 437, hep-th/9205110.
A. Marshakov, P. Wiegmann, A. Zabrodin, Integrable structure of the Dirichlet boundary problem in two
dimensions, hep-th/0109048.
I. Kostov, String equation for string theory on a circle, hep-th/0107247.
W.T. Tutte, Canad. J. Math. 14 (1962) 402417.
P. Zinn-Justin, Commun. Math. Phys. 194 (1998) 631, cond-mat/9705044.
P. Wiegmann, A. Zabrodin, Conformal maps and integrable hierarchies, hep-th/9909147.
E. Brzin, C. Itzykson, G. Parisi, J.-B. Zuber, Commun. Math. Phys. 59 (1978) 35.

Multicharged dyonic integrable models

I. Cabrera-Carnero, J.F. Gomes, G.M. Sotkov, A.H. Zimerman
Instituto de Fsica Terica, IFT/UNESP, Rua Pamplona 145, 01405-900, So Paulo, SP, Brazil
Abstract
(1)
We introduce and study new integrable models (IMs) of An -nonabelian Toda type which admit
U (1) U (1) charged topological solitons. They correspond to the symmetry breaking SU(n + 1)
SU(2) SU(2) U (1)n2 and are conjectured to describe charged dyonic domain walls of N = 1
SU(n + 1) SUSY gauge theory in large n limit. It is shown that this family of relativistic IMs
corresponds to the first negative grade q = 1 member of a dyonic hierarchy of generalized cKP
type. The explicit relation between the 1-soliton solutions (and the conserved charges as well) of the
IMs of grades q = 1 and q = 2 is found. The properties of the IMs corresponding to more general
symmetry breaking SU(n + 1) SU(2)p U (1)np as well as IM with global SU(2) symmetries
are discussed. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
Topological solitons of two-dimensional N = 2 SUSY abelian affine An -Toda models
are known to describe certain domain walls (DW) solutions of four-dimensional N = 1
SUSY SU(n + 1) gauge theory [1,2]. According to Wittens arguments [3], for large
n this SUSY gauge theory should coincide with the ordinary (nonSUSY) QCD. It is
therefore reasonable to relate the DW of the (nonSUSY) SU(n + 1) YangMillsHiggs
type (YMH) model (for large n and maximal symmetry breaking SU(n + 1) U (1)n )
to the solitons of the corresponding nonsupersymmetric abelian affine An -Toda theory
for n . Together with these simple DW (with no internal structure), the superstring
description of 4-D gauge theories [2,4] provides examples of, say, U (1) (or U (1)l , l n)
charged DW of dyonic type or DW that requires nonmaximal breaking of SU(n + 1) to
SU(l) U (1)nl+1 [5], etc. Hence the properties of the abelian affine Toda solitons (and
E-mail addresses: cabrera@ift.unesp.br (I. Cabrera-Carnero), jfg@ift.unesp.br (J.F. Gomes),
sotkov@ift.unesp.br (G.M. Sotkov), zimerman@ift.unesp.br (A.H. Zimerman).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 4 7 - X
434
I. Cabrera-Carnero et al. / Nuclear Physics B 634 [FS] (2002) 433482
their N = 2 SUSY versions) are not sufficient to describe such dyonic DW carrying internal
chargesU (1)p or even nonabelian, say SU(2), etc. It is natural to expect appropriately
chosen nonabelian (NA) affine An -Toda models [6] (and their N = 2 supersymmetric
extension) to be the main tool in the description of DW with internal charges. Recently
the U (1) charged topological solitons of certain singular affine NA-Toda models [7,8] have
been interpreted as U (1) charged dyonic DW of 4-D SU(n + 1) gauge theory. It is expected
that such DW-2D soliton relation originates from the fact that specific symmetry reduction
(reflecting DW properties) of SU(n + 1) self-dual YM (SDYM) equations reproduces the
equations of motion of certain two-dimensional integrable models (2-D IMs). The simplest
representative of such dyonic IM (with p = 1, i.e., one U (1) symmetry) is given by the
Lagrangian [7]
( + ) 1
1
kij i j +
e
V
1
2
1 + 2 (n+1)
i=1
2n e
n1
p=1
Ln
(1.1)
with potential
n1

2 kij j
(1 +n1 )
2
1
V= 2
1 + e
n ,
e
+e
i=1
where kij = 2i,j i,j 1 i,j +1 , i, j = 1, . . . , n 1. It represents a pair of U (1) charged

fields , of LundRegge type [12] interacting with a set of neutral fields i of the An1
abelian affine Toda model. For imaginary coupling i0 its potential has n distinct
vacua and manifests discrete Z2 Zn and a global U (1) symmetry. As a consequence the
IM (1.1) admits both, U (1) charged and neutral, topological solitons. The semiclassical
spectrum of such dyonic solitons turns out to be quite similar to the one of charged DW
of N = 1 SU(n + 1) Supersymmetric YangMills (SYM) theory for nonmaximal breaking
SU(n + 1) SU(2) U (1)n1 .
Apart from the problem of exact identification of certain 2-D solitons as 4-D DWs [11]
an important question to be addressed is whether one can construct in this manner also
U (1)p , p 2, multicharged DWs and nonabelian (say SU(2)) DW as well. Which are
the 2-D IMs that admit topological solitons carrying U (1)p or SU(2) charges? What are
the main features of such solitons and how to derive their masses and charges? Should we
restrict ourselves in considering relativistic 2-D IMs of NA-Toda type only, or we also have
to include certain nonrelativistic IMs, as, for example, the nonlinear Schroedinger type
models? The present paper provides a partial answer to these questions. It is devoted to the
construction of both, relativistic and nonrelativistic dyonic IMs of global U (1) U (1),
(i.e., p = 2) symmetry and to the detailed study of their soliton spectra. We also introduce
and discuss its generalization to multicharged IMs with U (1)p (p = 1, . . . , n 1) and U (2)
global symmetries. The main purpose of the present research is the construction of soliton
(1)
solutions of the following An (p = 2) dyonic IM (and its nonrelativistic counterpart (5.8)):

n2
n
1
1
p=2
j+
1 1 e1
1 + 2
n n en2
kij i
Ln =
2
2(n 1)
i=1

n
n n en2
1 1 e1
+ 1 + 2
2(n 1)

( + )
2
p=2
1
n2
1 n 1 n + n 1 n 1 e
Vn
+
2(n 1)
435
(1.2)
with potential
p=2
Vn
n2

2 kij j
= 2
e
+ e(1 +n2 ) 1 + 2 n n en2
i=1

1 + 2 1 1 e1 n + 1 ,
where 0 = n1 = 0 and

2n
1 1 e1 + n n en2
2(n 1)
4 (n + 1)
1 1 n n e(1 +n2 ) .
+
4(n 1)
=1+
The Lagrangian (1.2) is invariant under global U (1) U (1) transformations: a =

2
2
ei a a , a = ei a a (a = 1, n), i = i (with charges Q1 and Qn ) as well as under
p=2
represents
the Z2 Zn1 discrete transformations. For imaginary coupling = i0 , Vn
(n 1) distinct zeros and now the IM (1.2) admits three different kinds of 1-soliton
solutions, namely:
neutral (of An2 -abelian Toda type) solitons [10];
U (1) charged 1-solitons (Q1 = 0, Qn = 0 and Qn = 0, Q1 = 0) of An1 (p = 1)
dyonic type [7,8];
new U (1) U (1) charged 1-solitons (Q1 = 0, Qn = 0, Qmag = 0) which appears as
bound states of the above (Q1 , 0) and (0, Qn ) 1-solitons.
From the Hamiltonian reduction point of view [6] the U (1) U (1) charged IM (1.2)
appears as a natural generalization of the U (1) dyonic model (1.1) including an extra pair
of charged fields n , n and a rather complicated kinetic term (involving quartic terms in
the denominator ) for a , a , a = 1, n. Hence, the methods developed in the construction
of solitons of the IM (1.1) [7,8] can be extended to the case of multicharged IM (1.2) with
certain modifications concerning the genuine (Q1 , Qn ) charged solitons. An important new
feature of the semiclassical spectra of these (Q1 , Qn ) 1-solitons (see Section 3.4):

4(n 1) (4j 02 jel )
,
Mjel ,j =
sin
4(n 1)
02
jel = 0, 1, . . . , j = 0, 1, . . . , (n 2),

4
el
2 jel
+ 0 ,
Qmag = 2 j ,
Q1 = 0
2
0

jel
2
+
Qel
(1.3)
0 , 0 R
n
0
2
436
2
el
is that their masses Mjel ,j depend on the difference of the U (1) charges Qel
1 Qn = 0 jel
only. The latter is quantized semiclassically (jel Z), while the sum of the charges
2
el
Qel
1 + Qn = 0 0 remains continuous (0 R). The new kind of dyonic effect that takes
place for the model (1.2) should be also mentioned. Together with the standard shift of Qel
a
by the magnetic charge Qmag (see Section 2.5):
02 a
j , a = 1, n,
2
which is induced by the topological term (2.42) (present in the IM (1.1) as well), one can
add to the Lagrangian (1.2) a new topological term

n
1
top
ln
.
ln
L =
n
1
4 2
el
Qel
a Qa
Its effect is another shift of the electric charges:

el
Qel
1 Q1
02
Q(1)
,
2 2
el
Qel
n Qn +
02
Q(n)
2 2
a
by the topological charges Q(a)
= 2 j .
Our dyonic IM (1.2) is not the only candidate to describe the U (1) U (1) charged
DWs. Answering the question about the nonrelativistic 2-D IMs admitting U (1) U (1)
charged topological solitons, we derive in Section 5, a new family of IMs of constrained
KadomtsevPetviashvili (cKP) type. They are represented by a system of integrable
second order differential equations (5.8) for the set of fields ul (x, t), ra (x, t), qa (x, t)
(l = 1, . . . , n 2; a = 1, n), which is invariant under global U (1) U (1) transformation
(a = const):
qa = eia qa ,
ra = eia ra ,
u l = ul .
Although the nonrelativistic IM (5.8) does not belong to the class of affine NA-Toda
IMs, it shares (by construction) the same algebraic structure underlining the relativistic
IM (1.2). This fact allows us to establish simple relations between the 1-soliton solutions
of both models, as well as between the conserved charges characterizing their soliton
spectra, similar to those found in [13] involving charges of sine-Gordon and mKdV and
its generalization to certain affine NA-Toda and cKP models in Ref. [14]. An interesting
feature of IM (5.8)
is that the fields ul (x, t) themselves serve as densities of the conserved
l = ul dx called fluxes. They appear to be the nonrelativistic counterparts of

charges Q
the electric and magnetic charges of the relativistic model (1.2) as one can see from the
following relation

1
el
el
l =
l 4j Qel
Q
(1.4)
1 + Qn + (n 1)Q1 .
2(n 1)
l , together with the particle number charge Q
0 given by
The fluxes Q

n2
1 Q
2(n 1) Q

Q0 =
sin 2(n 3) , = 2
(1.5)
437
determine the essential part of the 1-soliton spectrum of the nonrelativistic IM (5.10). An
important characteristic of the relation between the IMs (1.2) and (5.8) is that the masses
0 are proportional,
of the relativistic 1-solitons and the nonrelativistic particle number Q
0 = 1 M rel
Q
2 jel ,j
as one can verify by comparing Eqs. (1.3), (1.4) and (1.5). It is worthwhile to mention that
the existence of the nonrelativistic counterpart of the IM (1.2) is not a specific feature of
the particular IMs with U (1) U (1) symmetry we are considering. Applying the methods
presented in Section 5 one can construct an entire hierarchy of nonrelativistic IMs (of
higher order q = 2, 3, . . . , differential equations) for each abelian or nonabelian affine
Toda model, including the simplest case of U (1) dyonic IM (1.1).
This paper is organized as follows. Section 2 contains together with the path integral
derivation of the effective Lagrangian (1.2) (and its ungauged versions (2.12) and (2.15)),
the corresponding zero curvature representations as well as the proof of their classical
integrability. The construction of the new (Q1 , Qn )-charged topological 1-soliton solutions
of IM (1.2) is approached in two different manners. In Section 3 we present the explicit
1-soliton solutions obtained by applying the vacua Bcklund transformation, i.e., by
constructing and solving the corresponding soliton first order differential equations. The
derivation of the semiclassical 1-soliton spectrum is given in Section 3.4. Section 4 is
devoted to the vertex operator ( function) method. Following the standard Heisenberg
subalgebra constructions, the explicit form of the four different type (neutral and
charged) soliton vertex operators is obtained. For example, the (Q1 , Qn )-charged
1-soliton is represented by specific composite 4-vertex operator. The corresponding G0 -functions 0 , j , j = 1, . . . , n 2, a , a = 1, n, are realized as vacuum expectation
values of appropriate soliton vertex operators. It turns out that the cumbersome algebraic
manipulation to be performed in the derivation of the explicit form of the -functions
is simpler in the case of the nonrelativistic generalized constrained KP hierarchy models
constructed in Section 5. They have the same algebraic structure A(1)
n , Q, G0 , + (see
Section 2.1) as our relativistic dyonic model (1.2) but with the negative grade q = 1
constant element replaced by a constant element of grade q = 2. Another difference
lies in the zero grade parameters (i.e., the physical fields) of the nonrelativistic model.
They are directly related to the current (x g0 )g01 , while for the relativistic theory they
parametrize the group element g0 G0 . The precise statement presented in Section 5
is that the relativistic dyonic IM (1.2) appears as the first negative flow of the new
(1)
dyonic hierarchy An (p = 2, q) given by Eqs. (5.8). It is a generalization of the wellknown relation between the sine-Gordon and MKdV [19], LundRegge and Nonlinear
(1)
Schroedinger models [18], the generalized An (p = 1) dyonic models (1.1) and the
constrained cKP hierarchy [20] and their supersymmetric versions [21].
Section 6 contains preliminary discussion of the algebraic structure and the effective
actions of more general multicharged dyonic models, as, for example, the SL(3) generalization of the LundRegge model and SL(2)p U (1)np models (p = 1, 2, . . . , n 1).
438
2. Dyonic A(1)
n (p = 2) IM as gauged two-loop WZW model
2.1. Algebraic structure of the Hamiltonian reduction
The gauged G0 /G00 WZW model based on finite-dimensional Lie algebra G is given
by

= SWZW (g) k
1 +
SG/H (g, A, A)
d 2 x Tr A gg
2

1
1 + A0 A 0 ,
+ A g g + Ag Ag
(2.1)
where A = A + A0 , A = A + + A 0 , A G< , A + G> , A0 , A 0 G00 , g G and

G00 = exp(G00 ). As is well known all the conformal field theories: abelian and nonabelian
Toda models, parafermions [22,23], etc. can be represented as appropriate gauged WZW
models. The subgroups H< , H> G are generated by positive/negative grade subalgebras
G< and G> , respectively, according to the appropriately chosen grading operator Q,1
decomposing the Lie algebra G in graded subspaces,
n

2i H
si
, [Q, Gl ] = lGl , si Z,
i2
i=1

G=
Gl , [Gl , Gk ] Gl+k , l, k = 0, 1, . . . ,
Q=
(2.2)
where i denote the fundamental weights of G. The constant elements G1 of

(0)
2
grade 1 (say = ni=1
i Ei ) indicate those WZW currents that are taken to be
+
constants, i.e., Ji = i , Ji = i . In fact, Q is providing G with specific Gauss-like

decomposition G = H< G0 H> with the important difference that the zero grade subgroup
G0 (parametrized by physical fields, say i , a , a in Eq. (1.2)) is in general nonabelian.
Each conformal model is specified by the choice G, Q, .
an infinite-dimensional affine (KacMoody)
Similar construction takes place for G,
algebra [6]. They give rise to a larger class of (non-conformal) integrable models which
are in fact integrable deformations of the conformal models described above. The main
ingredient is now the so-called two-loop gauged WZW model (i.e., the model associated
[15] described formally by the same action (2.1) with an important
to an affine algebra G)
H
[8]. The
< , H> are infinite-dimensional (and so is g and A, A)
difference that all G,
zero grade subgroup G0 = G/H is however chosen to be finite-dimensional. The grading
operator Q is now given by
+
Q = hd
n

2i H
si
,
i2
i=1

d, Ea(n) = nEa(n) ,
(n)
d, Hi
(n)
= nHi , n Z, (2.3)
where h is a positive integer to be specified below. The major difference from the conformal
models is in the constant grade element that may include extra affine generators of the
1 See [6] and Appendix A of [23] for our algebraic notation.
1
2 The WZW currents in question have the from J = Tr(E g 1 g ) and J = Tr(E g
0 g0 ).
0
439
form E = 1 E . For instance, in the abelian An affine Toda models, where Q is

given by the principal gradation we have,
(1)
(0)
Q = (n + 1)d +
n

2i H
i=1
i2
=
n
(0)
(1)
Ei + E(1 ++n ) .
(2.4)
i=1
Since in this case the invariant subalgebra G00 G0 defined as [G00 , ] = 0 is trivial
(empty), these models have no continuous (local or global) symmetries. Their Lagrangians
are invariant under the discrete permutation group Sn (Weyl group of An ) [10].
On the other hand, the so-called homogeneous gradations
Q = d,
=
n
i h(1)
,
i
i=1
G00
(0)

= hi , i = 1, . . . , n = U (1)n
(2.5)
define a class of IMs of electric type with maximal number n of local U (1)n symmetries
and potential V = Tr(+ g0 g01 ) which has no nontrivial zeros. One can further gauge
fix axial or vector local symmetries by considering the G0 /G00 -gauged WZW model,
S(g0 , A0 , A 0 )

k
= SWZW (g0 )
Tr + g0 g01 d 2 x
2

k
0 g 1 + A 0 g 1 g0 A0 g0 A 0 g 1 + A0 A 0 d 2 x,
Tr A0 g
0
0
0
2
(2.6)
where the signs correspond to axial or vector gaugings, respectively, g0 = 0 g0 0 ,

( )1 and = 0 (z, z ) G 0 for axial and =
A 0 = A0 01 0 , A 0 = A 0
0 0
0
0
0
01 (z, z ) G00 for vector cases. Since G0 /U (1)l , 0 l n, is nonabelian and define
homogeneous space, the IMs of this type are known as homogeneous sine-Gordon models
[24]. They are generalization of the usual complex SG model (i.e., n = 1, known as
LundRegge [12]). Note that the corresponding effective action manifest global U (1)l
symmetries.
The two examples illustrate the extreme cases, the principal gradation, with no
continuous symmetries (G00 = ) but maximal number of topological charges and the
homogeneous gradation with maximal number of local (or global) U (1)l symmetries, but
no topological charges. An interesting class of IMs arises when the following intermediate
(dyonic type) gradation is considered
+
Q = hd

i=a1 ,a2 ,...,ap
2i H
,
i2
=
n

i=a1 ,a2 ,...,ap
(0)
Ei +
(1)
(1)
s Es , (2.7)
where a = (a1 , . . . , ap ) denote the corresponding set of omitted fundamental weights

s
a and simple roots a ; h = 1 + i=a1 ,a2 ,...,ap 2i
and s are appropriate composite
2
(1)
roots of An to specified below. The main feature of such dyonic IM is that (i) only part
of the Cartan subalgebra, say U (1)l , l p n 1, generate local (or global) Noether
440
symmetries and (ii) their potential for imaginary coupling constant and appropriate choice
of has (n l + 1) distinct zeros and Znl+1 as an asymptotic (discrete) symmetry group.
Hence they admit soliton solutions carrying both U (1)l electric and topological (magnetic)
Znl+1 charges. The simplest is the A(1)
n (l = p = 1) dyonic IM defined by
Q = nd +
n

2i H
i=2
i2
=
n
(0)
(1)
i Ei + E(2 ++n )
(2.8)
i=2
such that G0 = SL(2) U (1)n1 and G00 = {1 H }. It has been introduced and
systematically studied in Refs. [7,8]. The problem we address here concerns the
construction and the study of the properties of solitons for multicharged (p > 1) IMs.
Among the vast variety of families of p-multicharged dyonic IM we choose one simple
representative characterized by the form of the zero grade subgroup3 G0 = SL(2)p
U (1)np . Although the methods we are employing in the construction of the p-charged
dyonic models are universal (and valid for all 1 p n 1) we restrict ourselves for
simplicity to consider the case p = l = 2 only,
n1

n1

(0)
(1)
i H,
=
Ei + E(2 ++n1 ) ,
Q = (n 1)d +
(2.9)
i=2
i=2
and G00 = {1 H, n H }. The specific choice of si and i , namely, s = (0, 1, 1, . . . , 1, 0)

in Eq. (2.9) is not a strong restriction. Following the discussion of Section 2.7 of Ref. [7]
(concerning the Weyl families of An (p = 1) IMs) one can easily verify that any other
choices4 of s, preserving the structure of G0 , say s = (0, 1, 1, . . ., 1, 0, 1) is related to (2.9)
by certain Weyl transformation. The actions for both models (s and s ) are related by change
of variables similar to the one of An (p = 1) case presented in Ref. [7]. Our symmetric
choice (2.9), represents however certain advantages in the construction of U (1) U (1)
topological solitons in Sections 3 and 4.
2.2. Effective dyonic actions
(1)
Given the graded structure (2.9) of the affine group An , the elements of the zero grade
subgroup are parametrized as

n2

(0)
(0)
(0)
(0)
(0)
exp R1 1 H + Rn n H +
+ n E
i h
g0 = exp 1 E
1

exp 1 E(0)
+ n E(0)
n
1
i+1
i=1
(2.10)
3 Note that for given p, one has several inequivalent choices for s and , (1) that leads to different G , for
s
i
i
0
(1)
(2)
example, for p = 2, we have G0 = SL(2) SL(2) U (1)n2 and G0 = SL(3) U (1)n2 corresponding to
n1

Q1 = (n 1)d + i=2
i H and Q2 = (n 1)d + ni=3 i H , respectively.
4 Except the particular case s = (0, 0, 1, . . . , 1, 1) and Q = Q . Due to the fact that now = 0
2
1
2
(1 j = 0 for all other cases), this model is not Weyl equivalent to the original one, since it does not preserve
the structure of G0 .
441
and the corresponding positive/negative grade subgroups H< and H> contain all
nonphysical fields and are generated by the positive/negative grade G> (G< ) generators of
the A(1)
n affine algebra with respect to Q given in (2.7). As we have explained before, the
starting point in the derivation of the A(1)
n (p = 2) dyonic IM action is the gauged (H< , H>
invariant) G0 = G/H two-loop WZW model given by Eqs. (2.1) without the A0 A 0 term.
The result of the formal functional integration yields the following U (1) U (1)-ungauged
effective action

k
dz d z Tr + g0 g01 ,
SG0 (g0 ) = SWZW (g0 )
(2.11)
2
g0 G0 = SL(2) SL(2) U (1)n2 . Taking into account the parametrization (2.10), it
leads to the following action ( 2 = 2
k ):
1
j + 1 1 e(R1 1 ) + n n e(Rn n2 )
kij i
2
n2
LG0 =
i=1

1
1 + nRn R
n + R1 R
n + Rn R
1 V0 ,
nR1 R
2(n + 1)
(2.12)
where
n2

2 kij j
V0 = 2
e
+ e(1 +n2 ) 1 + 2 1 1 e(R1 1 )
i=1

2 (Rn n2 )
n+1 .
1 + e
n n
As we have mentioned, the fields j , j = 1, . . . , n 2, Ra , a , a , (a = 1, n) parametrize

the nonabelian zero grade subgroup G0 . The ungauged IM (2.12) represents an integrable
deformation of the G0 WZW model with the potential V0 . An important feature of (2.12)
is that it is invariant under chiral U (1) U (1) local gauge transformations wa (z), w a (z):
Ra = Ra + wa + w a ,
i = i ,

= ewa ,
= e w a
a
(2.13)
G00
reflecting the fact that G0 has two-dimensional -invariant subgroup

= {1 H ,
0
n H }, such that [G0 , ] = 0.
We further consider an intermediate A(1)
n (p = 2) dyonic IM with one local U (1) and one
global (axial) U (1) (generated by (1 + n ) H ) symmetry. It is obtained from Eqs. (2.6)
with A0 = a0 (1 + n ) H , A 0 = a 0 (1 + n ) H 5 and zero grade factor group element
f
g0 is taken in the form

n2

f
(0)
(0)
(0)
(0)

g = exp 1 E + n E exp R(1 n ) H +
i h
0
i+1
i=1
5 a (z, z ), a (z, z ) are arbitrary functions parametrizing the auxiliary fields.

0
0
442

,
exp 1 E(0)
+ n E(0)
n
1
(2.14)
1
1
= 1 (R1 Rn ). In order to construct
where a = a e 4 (R1 +Rn ) , a = a e 4 (R1 +Rn ) and R
2
its action, we first consider the partition function for this intermediate An (p = 2) model,

ax

p=2
0) ,
(g
,
A
,
A
Zn = Dg0 DA0 D A 0 exp SG
0
0
0
/G
0
where the action
S ax 0
G0 /G0
of the axial gauge fixed model is given by (2.6) with upper signs.
Integrating over the corresponding auxiliary fields A0 , A 0 , we derive the effective action
representing this IM,
p=2
Ln,inter =
n2
(n 1) 1
j
kij i
R R +
(n + 1)
2
i=1

1
2

+
1 + n n e(n2 +R) 1 1 e(R1 )
0
4

2

+ 1 + 1 1 e(1 R) n n e(R+n2 )
4

( + )
2
1
n2
n 1 1 n + 1 n n 1 e
Vinter ,
4
(2.15)
where
n2

2 kij j

Vinter = 2
e
+ e(1 +n2 ) 1 + 2 1 1 e(R1 )
i=1

n2 )
2
(R+
1 + n n e
n+1 ,

and 0 = 1 + 4 ( 1 1 e(R1 ) + n n e(R+n2 ) ). Note that the denominator 0 in
(2.15) is quadratic in a , a , i.e., similar to An (p = 1) IM (1.1), but with an extra pair of
fields n , n and an additional chiral local U (1) gauge symmetry:
2
1 = ew 1 ,
n = ew n ,
+ w + w,
= R
1 = e w 1 ,
n = e w n ,
l = l .
Finally, in order to derive the action for the completely gauged IM (1.2) we substitute
A0 = a011 H + a0n n H , A 0 = a 01 1 H + a 0n n H in Eq. (2.6) with the upper signs
f
(axial gauging) and we take g0 in the form
n2

f
(0)
(0)
(0)
,
g0 = exp 1 E1 + n En exp
(2.16)
i hi+1 exp 1 E(0)
+ n E(0)
n
1
i=1
443
1
1
where a = a e 2 Ra , a = a e 2 Ra . Following the procedure described above (i.e.,
integrating over A0 , A 0 , etc.), we obtain the effective Lagrangian given by Eq. (1.2) (see
Ref. [20] for details).
It is worthwhile to mention that the conceptual difference between the completely
gauged IM (1.2) and the ungauged (2.12) and the intermediate models (2.15) is that the
action (1.2) has no local symmetries. Instead it has two global symmetries described by
transformations (2.32).
Similarly to the axial and vector A(1)
n (p = 1) dyonic IM [7,8], applying the method
developed in Ref. [17] one can derive the CPT-invariant vector model (1.2) starting from
Eqs. (2.6) with lower signs (). One can also consider the mixed case by axial gauging
one U (1) and vector gauging the other, but this topic is out of the purpose of the present
paper.
2.3. LeznovSaveliev equations and the zero curvature representation

The equations of motion for the models (1.2), (2.12) and (2.15) can be written in the
following matrix LeznovSaveliev form [6]

0 g 1 + , g0 g 1 = 0.
g
g01 g0 + , g01 + g0 = 0,
(2.17)
0
0
The local U (1) U (1) symmetries (2.13) reflect the fact that the traces of the above
H (0) = J H (0) = 0, where
equations with a H vanish, i.e., J
a
a

1
(0)
,
Ja H (0) = Tr g0 g0 a H

1
(0)
0 g a H
, a = 1, n,
Ja H (0) = Tr g
(2.18)
0
due to the G00 property that [a H, ] = 0, a = 1, n. The axial (vector) gauge fixing of
these U (1) U (1) symmetries is known to be equivalent to implementing the constraints
Ja H (0) = Ja H (0) = 0, i.e., to the system of equations

n+1
nR1 1 + 2
1 1 e1 + Rn = (n + 1)1 1 e1 ,
2n

n+1
R1 + nRn 1 + 2
n n en2 = (n + 1)n n en2 ,
2n

n
+
1
1 1 + 2
n = (n + 1)1
1 e1 ,
nR
1 1 e1 + R
2n

n en2 .
n 1 + 2 n + 1 n n en2 = (n + 1)n
R1 + nR
2n
The diagonalization of the above system leads to the definition of the auxiliary nonlocal
fields Ra , a = 1, n, which are going to play an important role in the construction of the
1-soliton solutions in the next sections,

n+1
(n 1)R1 = n1 1 1 + 2
n n en2 e1 n n en2 ,
2n

n
+1
(n 1)Rn = nn n 1 + 2
1 1 e1 en2 1 1 e1 ,
2n
444

1 1 + 2 n + 1 n n en2 e1 n
n en2 ,
1 = n1
(n 1)R
2n

n = nn
n 1 + 2 n + 1 1 1 e1 en2 1
1 e1 ,
(n 1)R
2n
(2.19)
where

n 2
=1+
1 1 e1 + n n en2 )
2(n 1)
(n + 1) 4
1 1 n n e(1 +n2 )
+
4(n 1)
j
and 1 j = n+1j
n+1 , n j = n+1 , j = 1, . . . , n 1.
The zero curvature representation
= 0,
[A, A]
A A
(2.20)
(1)
of the An (p
= 2) equations (2.17) is known to be the crucial ingredient in the proof of the

classical integrability of the model. As it is not difficult to check the pure gauge potentials
has the standard LeznovSaveliev form [6],
A, A,
A = g0 g01 ,
0 g 1
A = + + g
0
(2.21)
common for a large class of grades |q| = 1, (i.e., q( ) = 1, where [Q, ] = q( ) )
singular (or nonsingular) affine NA-Toda IM. As in the case of A(1)
n (p = 1) dyonic models
[25] one can complete the proof of the integrability by deriving the explicit form of the
classical R-matrix and the infinite set of conservation laws in involution. The calculation
is quite straightforward and standard for all affine abelian and NA Toda models.
2.4. Vacua structure and symmetries
The existence of soliton solutions for the An (p = 2) IM (1.2) is known to be intrinsically
connected to the following three related objects:
distinct zeros of the potential Vn = 12 (Tr(+ g0 g01 ) 2 (n 1));
set of nontrivial constant solutions of its equation of motion;
the existence of few different boundary conditions (i.e., admissible asymptotic values
at x = ) for the fields l , a , a , Ra .
p=2
The very origin of the nontrivial vacua structure is hidden within the discrete symmetries
of (1.2) and in the corresponding topological charges. In our case, for imaginary = i0 ,
the action (1.2) is invariant under the following discrete group of transformations

lN
2
, l = 1, 2, . . . , n 2,
l = l +
0 n 1
1
1 = w 2 (N+(n1)s1 ) 1 ,
n = w 2 (N(n1)sn ) n ,
1 = w 2 (N+(n1)s1 ) 1 ,
n = w 2 (N(n1)sn ) n ,
(2.22)
445
2 i
where w = e n1 and N is an arbitrary integer (defining different Zn1 vacua sectors) and
sa , sa are two pairs of even (or odd) integers (i.e., sa + sa = 2Sa , sa sa = 2L a and Sa ,
L a Z). The action (1.2) is also invariant under CP (but not CPT) transformation,
l = l ,
P x = x,
a = a ,
P = .
a = a ,
(2.23)
It is convenient to introduce the following new parametrization for the field variables a ,
a ,
1 i0 ( 1 1 1 )
e 2
sinh(0 R1 ),
0
1
1
n = ei0 ( 2 n2 n ) sinh(0 Rn ),
0
1 =
1 i0 ( 1 1 +1 )
e 2
sinh(0 R1 ),
0
1
1
n = ei0 ( 2 n2 +n ) sinh(0 Rn ).
0
1 =
As we shall see the angular variables a , a = 1, n, make transparent the origin of the
topological charges Qa . They also play an important role in the discussion of T-duality
transformations at the end of this subsection. The transformations (2.22) and (2.23) in this
parametrization acquires the form
a = a + L a ,
R a = Ra ,
0
2
R a = Ra ,
(2.24)
a = a +
L
0 a
with L a , L a arbitrary integers. Combining the a and a transformations we find

i
a
La
= a +
ln
,
La = 2L a L a .
a =
20
20
a
(2.25)
The conclusion is that together with the trivial vacua: l = 0, Ra = 0, a = 0 (i.e.,

N = La = 0) our model (1.2) possess an infinite set of distinct vacua given by

2
L
lN
(N)
l =
(2.26)
a(L) =
,
R(0)
a = 0,
0 n 1
20
(N)
defining its vacua lattice (l

(a) Zn1 :
(L)
, a ). Due to the Zn1 Z2 symmetry,
el wlN el ,
1
1 w 2 (N+(n1)s1 ) 1 ,
1
1 w 2 (N+(n1)s1 ) 1 ,
n w 2 (N(n1)sn ) n ,
1
n w 2 (N(n1)sn ) n ,
(b) Z2 :
a a ,
a a
we have a finite set of allowed boundary conditions (at x = ):

2 lN
L
a
,
Ra () = 0,
l(N) () =
a(L) () =
0 n 1
20
(2.27)
446
(i.e., a () = a () = 0, but a /a () = 0). One therefore expects the existence

of finite energy topological solitons, interpolating between different vacua (N , L()
a ). The
corresponding topological charges
j = N+ N = 0, 1, . . . , n 2 mod (n 1),
ja = L+
a La = 0, 1, . . . , mod 2
are represented by the following topological currents

2(n 1)
l ,
0
+
top
Ql =
Jl0 dx = lQmag ,
Jl =
Qmag =
and
,a
4
j
02

= 0 a ,
J0,a dx =
Qa =
(2.28)
ja
2
(2.29)
(for our U (1) U (1)-charged topological 1-solitons (3.14), (3.18), j = 1, ja = 0).

The transformations (2.22) and (2.23) that allowed us to determine the vacua lattice
(N) (L)
(1)
(l , a ) of our model, does not exhaust all the discrete symmetries of the An (p = 2)
dyonic IM (1.2). There exist two other Z2 discrete groups leaving the action (1.2) invariant,
1. Z2 -reflections
l = nl1 ,
l = 1, 2, . . . , n 2,
= 1 ,
= n ,
1 = n ,
2. composed Weyl transformation P :

g0 = g01 ,
(+ ) = ,
(Ea ) = E(a ) = Ea ,
a = 1, n,
n = 1 ;
(2.30)
2 = 1,
P = .
(2.31)
The Z2 Z2 symmetries (2.30) and (2.31) impose certain equivalence relations on the
vacua lattice (l(N) , a(L)) Zn1 Z2 . As a result the irreducible vacua lies in the coset
Zn1 Z2 /Z2 Z2 , i.e., (2.27) with the identifications (2.30) and (2.31). As we shall see
in Section 3 the Z2 symmetry (2.30) is crucial in the derivation of the first order soliton
equations (3.3) and (3.4). The Z2 transformation can be realized as certain compositions
of An2 -Weyl group transformations similar to the constructions of Section 2.6 of Ref. [7].
It is the analog of the Sn symmetries of the abelian An Toda model.
An important new feature of the dyonic IM (1.2) is that, together with the above discrete
symmetries, it manifest the following continuous symmetries
a = ei0 a a ,
2
a = ei0 a a ,
2
l = l
(2.32)
with a the arbitrary real parameters of the global U (1) U (1). Taking into account the
definition (2.19) of the nonlocal fields Ra , the corresponding (Noether) electric currents
447
(derived from (1.2)) can be written in the following simple form

20
20
(2.33)
Jn,el =
(nR1 + Rn ),
(R1 + nRn ).
n+1
n+1
It therefore follows that the topological soliton with nontrivial asymptotics of the fields Ra ,
i.e., Ra (+) = Ra (), will carry electric charges given by
,el
J1

Qel
1
J10,el dx =

20
n R1 () R1 () + Rn () Rn () ,
n+1
Jn0,el dx =

20
R1 () R1 () + n Rn () Rn () . (2.34)
n+1
Qel
n =
As it is shown in Section 3 together with the neutral (0, 0), we have (0, Qn ) and (Q1 , 0)
and (Q1 , Qn ) charged topological solitons.
It is worthwhile to mention the geometrical (target space -model) meaning of the
U (1) U (1)-symmetry as well as the T-duality O(2, 2|Z) transformations behind it.
Writing (1.2) in symbolic ( -model) form,

L = gMN L + bMN L M N V L ,
(2.35)
where L = {l , Ra , a } and noting that the U (1) U (1) transformation (2.32) acts as
a translations, a a + 02 a , we conclude that the target space geometry (described
by the metric gMN , antisymmetric tensor bMN and by the tachyon potential V ( L )) has
two isometries, i.e., gMN , bMN and V ( L ) are a -independent. This is an indication that
the target space geometry encoded in the kinetic part of (1.2) allows two equivalent T-dual
descriptions. As in the case of U (1) dyonic model (1.1) (see Refs. [7,17]) one can introduce
a ),
together with the axial model (1.2) its T-dual vector model described by Lvec (l , a , R
Lvec = Laxial +
dF
,
dt
(2.36)
where

1
dF
a a a ).
=
( a
dt
4(n + 1)
a=1,n
The T-duality transformations with the above generating function F (a , a ) are known to
, a )vector such that
act as the following canonical transformation (a , a )axial (
a
= x a ,
a = x a ,
(2.37)
and certain point transformations l = fl (l , Ra ) of the nonisometric coordinates (see

Refs. [7,17] for more details). The relation (2.36) between Lax and Lvec is a simple
consequence of the fact that both Hamiltonians are in fact equal, i.e., Hax = Hvec . An
important feature of the abelian T-duality (2.36) and (2.37) is that it maps the U (1) U (1)
(given by Eq. (2.34)) into the vector model analog Qa,vector
of the
charges Qel,axial
a
topological charges Qa,axial

(given by Eq. (2.29)). Since we have that (see Section 3 of
448
Ref. [17])

Ja,el (ax) = eaN L N = a ,

eaN L = gaN L + baN L , L = {l , Ra ; a }, a a ,
(2.38)
comparing with Eqs. (2.33) we conclude that the isometric coordinates a of the vector
model can be written as a linear combination of the axial model nonlocal fields Ra ,
2
2
(nR1 + Rn ),
(R1 + nRn )
n =
1 =
(2.39)
n+1
n+1
a,vec = dx x a , which makes the form (2.36) of the generating function F explicit.
and Q
The T-dual transformations of electric Qel,ax

and a -topological charges Qa,ax
are given
a
by the following interchange rule

a,vec el,vec
el,ax a,ax

a
Q
.
,Q
Qa , Q
2.5. Dyonic effect of the topological -terms

One of the main properties of the soliton spectrum of the dyonic IM (1.1) (with one
global U (1) symmetry) is known to be its intrinsic dyonic structure [7]. Similarly to
the electric and magnetic charges of dyons in 4-D SU(n + 1)YangMillsHiggs model
[26], the electric charge Qel (p = 1) of the U (1) charged topological solitons acquires
contributions from its topological (i.e., magnetic) charge Qmag (p = 1) (see Eq. (1.5) of
Ref. [7]):

p=1
p=1
2
j
(p
=
1)
=
+
j
,
Qax
0
el
el
2
4
Qmag (p = 1) = 2 jp=1 , jp=1 = 0, 1, . . . , (n 1),
(2.40)
0
where is an arbitrary real parameter. The origin of such dyonic effect comes from the
topological (total derivative) -term (an analog of the corresponding four-dimensional
topological -term [26])

n1

top
.
Lax (p = 1) =
(2.41)
k k ln
2
8
k=1
(1)
Since by construction the dyonic model An (p = 2) (1.2) with U (1) U (1) global
symmetry is an appropriate generalization of the model An (p = 1) (1.1) it is natural to
expect that similar dyonic effects to take place. In this case one can introduce the following
two kinds of topological terms:
1. Straightforward generalization of the U (1) term (2.41) to the U (1) U (1) case

n2
(a)
a
top
.
ln
L (p = 2) =
(2.42)
k
k
8 2
a
k=1 a=1,n
449
2. New kind of a mixed term

top
La (p

n
1

= 2) =
ln
ln
,
4 2
n
1
(2.43)
where k(a) , and are arbitrary real constants. Adding the -terms (2.42) and (2.43) to
the original Lagrangian (1.2), i.e., considering
top
top
Limpr (p = 2) = L(p = 2) + L + La
(2.44)
leads to the following improvements (shifts) of the corresponding electric currents

,el
J1,impr (p = 2)

n2
02 (n + 1) (1)
20
0 (n + 1)
n
=
nR1 + Rn
,
l l
ln
n+1
8 2
8 2
n
l=1
,el
Jn,impr (p
= 2)

n2
02 (n + 1) (n)
20
0 (n + 1)
1
. (2.45)
=
R1 + nRn
l l +
ln
n+1
1
8 2
8 2
l=1
The corresponding new improved electric charges take the following form

0,el
J1,impr
dx
= Qel
1

(1)
02
j
(1)
j +
,
2
2
0,el
Jn,impr
dx = Qel
n

02
j (n)
j (n)
,
2
2
(2.46)
1 n2 (a)
where (a) = n1
l=1 ll . For = 0, i.e., without the second topological term, one
recover the well-known simple effect that the electric charges acquires contribution from
the magnetic (j -topological) charge Qmag = 42 j . The new dyonic effect for = 0 is
0
specific for the U (1) U (1) case (1.2) (improved as Eq. (2.44)) and it consists in the fact
(a)
that the electric charges Qel
a,impr gets shifted by the topological charges j as well.
An interesting physical interpretation of the parametres la , as external constant
magnetic fields FMN ()
2
2
(a)
Fa ,l =
(2.47)
l ,
F1 ,2 =
4
4
(all other components vanish) comes from the following parallel with the open string (in
1
curved background gMN , bMN ) with two compactified directions say, X25 = 2
ln(1 /1 ),
X1 =
1
2
ln(n /n ) and boundary (ChanPaton) term included (as in Section 8.6 of [27]).
el
el
The string momenta P1 =
Qn and P25 = Q1 get contributions from the boundary (Wilson
Aa (X) dXa , Xa = a , Xi = {l , Ra , a }). It is important to note
line) term Sstrong = i
450
that for our 1-soliton solutions (3.17) the string coordinates

i
a
ln
a =
(X1 , X25 )
20
a
are indeed periodic (in worldsheet coordinate) with periods Ta =
w1 = sin ,
2
wa
wn = sin .
Hence the parallel with open string with two compactified dimensions is not accidental.
Following such string analogy, the topological -terms (2.42) and (2.43) can be rewritten
as Wilson line contributions for certain background gauge fields A (X )

a
Stop = d 2 z Ltop + Ltop

2
1
a
=
l l t a + (1 t n n t 1 )
dt
4
2
a=1,n

=
d a Aa (XM ),
(2.48)
where we have denoted by A a the following terms

2
n2
1
(1)
A 1 =
l l n ,
4
2
l=1

2
n2
1
(n)
A n =
l l + 1
4
2
(2.49)
l=1
and all other AM for M = a vanish. Therefore the magnetic fields FMN = M AN N AM
for our U (1) gauge potential (2.49) are given by Eq. (2.47).
3. Multicharged topological solitons

3.1. Soliton equations
The well-known relation between dressing and Bcklund transformation [28] provides
a simple derivation of 1-soliton equations for a large class of grade one affine NA-Toda
models. Arguments similar to those used for the A(1)
n (p = 1) dyonic IM [7], lead us to the
following compact form of 1st-order soliton differential equations (DE)

0 g 1 Y g0 Xg 1 , + = 0,
g01 g0 X g01 Y g0,vac , = 0,
(3.1)
g
0
0,vac
[X, ] = [Y, + ] = 0
(3.2)
imposed together with the image of the system (3.1) under Z2 discrete transformations
(2.30). The constant elements X(, ai ), Y (, bi ) of the universal enveloping algebra are
451
realized as
X = X01 I + X02 1 H + X03 n H +
n2
ak ( )k ,
k=1
Y = Y01 I + Y02 1 H + Y03 n H +
n2
bk (+ )k ,
k=1
where X0i (), Y0i (), ak () and bk () are arbitrary functions of the spectral parameter
to be determined and g0,vac = ei0 (c1 1 H +cn n H ) is the constant vacuum solution of
the LS Eqs. (2.17). The verification of the fact that second-order DE (2.17) are indeed
the integrability conditions for the first-order DE (3.1) (under conditions (3.2)) is quite
straightforward. We should mention that the consistency of Eqs. (3.1) with their Z2 image
requires certain algebraic relations (see Eqs. (3.4)) specific for the soliton equations far
f
a large class of affine NA-Toda models [7]. With the parametrization of g0 given by
Eqs. (2.16) at hand and the subsidiary constraints (2.19) for the nonlocal fields Ra we
derive the following explicit system of 1st-order DE,

p
p =
e
ep1 + 2 ( 1 1 n n )p,1 ,

p=
e p ep+1 2 ( 1 1 n n )p,n1 , p = 1, 2, . . . , n 1,
0 = n1 ,
1 = n ,

2
n1
2
n1
n1
1 + n n e
,
1 1 e
1 = 1 e
2

2
n = n e1 1 + 2 1 1 e1 n n e1 ,
2

2
1 = 1 e1 1 + 1 1 e1 ,
2

2
n1
n1
n = n e
1+
,
n n e
2

2
1 = 1 e1 1 + 1 1 e1 ,
2

2
n = n en1 1 + n n en1 ,
2

2
1 = 1 en1 1 + 2 n n en1 n n en1 ,
2

2
n = n e1 1 + 2 1 1 e1 n n e1 ,
(3.3)
2
where is the Bcklund transformation parameter,
=
b2 ec1 cn (Y01 Y02 )

b1 ec1 cn
=
=
= = eb .
a2
a1
X01 X02
452
Another consequence of Eqs. (3.1) is the following chain of algebraic relations,

e1 e2 = e2 e3 = = en2 en3
= en1 e1 + 2 ( n n 1 1 )
(3.4)
which should be satisfied together with the first-order DE (3.3). The new variables p , a
and a
p = p p1
1
R,
n+1
R = R1 Rn , 0 = n1 = 0,
n1
1 + 2 + + n1 =
R,
n+1
n = n e
1 = 1 e
R
2(n+1)
n = n e
R
2(n+1)
R
2(n+1)
1 = 1 e

p
k ,
p =
R+
n+1
p
k=1
R
2(n+1)
(3.5)
introduced in (3.3) have the advantage to simplify the construction of the solutions (1soliton) of Eq. (3.3). The nonlocal fields Ra , a = 1, n, satisfy the following set of 1st-order
DE,

(n n n 1 1 ),
(n 1)
( 1 1 n n n ),
Rn =
(n 1)
1 = (n 1 1 n n ),
R
(n 1)
n = (n n n 1 1 ),
R
(n 1)
R1 =
(3.6)
as a consequence of their defining Eqs. (2.19).

We next derive few consequences of the system (3.3), (3.4) and (3.6) representing the
following solitonic conservation laws,

ep + 1 ep+1 = 0,

+ 1 ep = 0, p = 1, 2, . . . , n 1,

1 ln aa = 0, a = 1, n,

+ 1 a a = 0, a = 1, n,

+ 1 (R1 Rn ) = 0,
(n1)

(n1)
ln 1 e 2(n+1) R = ln nn e 2(n+1) R , = , .
(3.7)
1
They play an important role in the construction of the solutions (1-soliton) of the system
(3.3) and (3.4).
453
3.2. Soliton solutions

The strategy we employ in solving the complicated 1st-order system (3.3) consists in:
1. to derive all first integrals of Eqs. (3.3);
2. to diagonalize the system of Eqs. (3.3) by using these 1st integrals and the solitonic
conservation laws (3.7), i.e., to separate the equations for the variables p , a , a , Ra ;
3. to solve each one of the corresponding individual 1st-order (one variable) equations.
The first integrals for the system (3.3) can be easily found by simple manipulations of
Eqs. (3.3), (3.7). The result is:
ep ep+1 = 2i sin = c0 , p = 1, . . . , n 1,
(n1)
(n1)
n 2(n+1)
1 2(n+1)
R
R
e
= ed2 ,
e
= ed1 , d1 + d2 = 2,
1
n
(3.8)
(3.9)
e+ n+1 (nR1 +Rn ) e+ n+1 (R1 +nRn ) = De ,
(3.10)
where c0 , d1 , d2 and D are arbitrary complex constants. We next substitute Eqs. (3.8)
and (3.4) in the first two equations of the system (3.3), thus obtaining the desired individual
equations for each p , = i0 ,

+ ei0 p = 1 e2i0 p 2i sin ei0 p ,
ei0 p = 0,
(3.11)
where we have introduced new variables ,

+ = x cosh(b) t sinh(b),
+ = sinh(b)t + cosh(b)x ,
= t cosh(b) x sinh(b),
= cosh(b)t + sinh(b)x ,
b = ln .
The solutions of Eqs. (3.11) have the following form

ei0 p = ei
Sp e2i+2+ cos() 1
.
Sp e2+ cos() + 1
(3.12)
Due to Eqs. (3.8) and (3.4) the integration constants Sp should satisfy the following
recurrence relations
Sp+1 = e2ii Sp ,
p = 1, 2, . . . , n 1,
and therefore we find

Sp = (1)p1 e2i(p1)S1 = (1)pn e2i(pn) X0 ,
where X0 is an arbitrary constant that determines the center of mass of the soliton. Finally,
taking into account the definitions (3.5) of the original fields p and R = R1 Rn in terms
454
of p we find 1-soliton solutions in the form:

(n1)
ef + (1)n1 e2i(n1) ef
,
(3.13)
ef + ef
ef + (1)ln+1 e2i(n1l) ef
, l = 1, 2, . . . , n 2,
l
n1l
+ ef ) n1 (ef + (1)n1 e2i(n1) ef ) n1
(3.14)
ei0 (n+1) R = ei(n1)( Sign())

ei0 l =
(ef
where f = + cos()+ 12 ln X0 . We next calculate R1 and Rn from Eqs. (3.10) and (3.13),
n0
i0 R1
Dei (n+1) R
(n1)
ei0 (n+1) R e2
= De
2iin( Sign()) (e
+ (1)n1 e2i(n1) ef ) n+1

1
(ef + ef ) n1 (A1 ef + A2 ef )
i0 Rn
Dei0 (n+1) R
(n1)
ei0 (n+1) R e2
1
De2i (ef + ef ) n1
1
(ef + (1)n1 e2i(n1) ef ) n+1 (A1 ef + A2 ef )
(3.15)
where A1 , A2 are given by

A1 = 1 ei(n1)( Sign()+2),
A2 = 1 ei(n1)( Sign()+2).
It remains to derive the solutions for a , a . The algebraic relation (3.4) together with
Eqs. (3.9), (3.13) and (3.14) allows us to calculate the product a a , a = 1, n,
1
02 1 1
N 2 (ef + ef ) n1 (ef + (1)n1 e2i(n1) ef )

=
(A1 ef + A2 ef )
(n2)
(n1)
(3.16)
where N 2 = (1 + e2i )((1)n1 e2i(n1) 1). For the ratio a /a we obtain the simple
equations

1
n
= + ln
= 0,
+ ln
1
n

1
n
= ln
= 2i sin()
ln
1
n
and therefore we find
1
= e2i( sin()+p1 ) ,
1
n
= e2i( sin()+pn ) ,
n
(3.17)
where p1 , pn are arbitrary constants. The explicit expression for a , a completes the final
form of our (Q1 , Qn )-charged topological 1-soliton solution
1 =
1 i( sin()+p1 )
e
1 1 ,
0
1 =
1 i( sin()+p1 )
e
1 1 ,
0
455
1
n =
N ed2 2 (n3)( Sign())ei( sin()+p1 ) (ef + (1)n1 e2i(n1) ef ) 2(n+1)

,
(n2)
1
(A ef + A ef ) 2 (ef + ef ) 2(n1)
1
N
n =
i
ed1 2 (n3)( Sign())ei( sin()+q1 ) (ef
1
+ (1)n1 e2i(n1) ef ) 2(n+1)

(n2)
(A1 ef + A2 ef ) 2 (ef + ef ) 2(n1)

(3.18)
i.e., the (Q1 , Qn ) 1-soliton of the dyonic model A(1)

n (p = 2) IM (1.2) is presented by the
set of functions (3.14), (3.16) and (3.18) of spacetime variables + and .
It is worthwhile to mention that both, the Lagrangian (1.2) and the 1st-order system
(3.3), (3.4) have two interesting limits:
(1)
1 = 1 = 0 (Rn = const) or n = n = 0 (R1 = const) leading to the An1 (p = 1)

dyonic IM (1.1) (see Ref. [7] for corresponding 1st-order system) and their 1-soliton
solutions [7,8] represent one charge (0, Qn ) or (Q1 , 0) topological solitons of our
An (p = 2) model (2.32);
(1)
1 = 1 = n = n = 0 gives rise to the An2 abelian affine Toda model and its
neutral (0, 0) 1-solitons are particular cases of our (Q1 , Qn )-solitons.
3.3. Soliton spectrum
One of the main properties of the soliton solutions is that they carry finite energy and
nontrivial topological charge (and electric charges Q1 , Qn in our case). Thus, it manifests
a
a
el
el
a particle-like spectrum, M, E, Qel
1 , Qn , Qmag , Q . The soliton charges Qa , Qmag , Q
as well as their energy E (and mass M) are known to depend on the boundary conditions,
i.e., asymptotics of fields l , a , a (at x ), and on the symmetries of the 1st-order
(BPS-like) equations (3.3). In fact, the explicit form of the soliton spectrum can be derived
without the knowledge of the exact 1-soliton solution (3.14)(3.18). The arguments are
quite similar to the abelian affine Toda [10] and A(1)
n (p = 1) dyonic IM [7] cases. One first
verify the following solitonic conservation law
= F + ,
F
n1

k
2
e
+ n n ,
F = 2
k=1
n1

k
+
2
F =
e
+ 1 1
2
(3.19)
k=1
(p=2)
and as a consequence the An (p = 2) potential Vn

(p=2)
Vn

1
F + F + .
2
(1.2) can be written in the form
456
Next step is to demonstrate (by using the 1-soliton Eqs. (3.3), (3.4) and (3.6)) that the
An (p = 2) stress-tensor components T00 = T + + T + 2V , T01 = T + T ,

1
1 1
2T + = kij i j +
n 1 1 + en2 1 n n
e
2

(1 +n2 )
2e
(1 n n 1 + n 1 1 n )
+
(3.20)
2(n 1)
where
and T = T + ( )
n
n
1 1 e1 ,
n n en2
n = 1 + 2
2(n 1)
2(n 1)
are total derivatives, i.e.,

T01 = x F + F + .
T00 = x F F + ,
1 = 1 + 2
Therefore the energy and the momentum of our (Q1 , Qn ) 1-soliton (3.14)(3.18) gets
contribution from boundary terms only,
+
+

E=
T00 dx = F F + ,
+
+

P=
T01 dx = F + F + . (3.21)
The last step is to calculate E, P and M = E 2 P 2 in terms of the asymptotics of the

fields l , a , a , Ra . The analysis of the classical vacua structure of the An (p = 2) IM
(1.2) (i.e., nontrivial constant solutions corresponding to multiple zeros of the potential
(p=2)
, with = i0 ) presented in Section 2.4 provides the complete list of the admissible
Vn
solitonic b.c.

2 lN
2 (a)
l () =
,
(a a )() = 0,
Ra () =
f ,
0 n 1
0
2 (1)
2
a
(n)
R() =
(3.22)
f f =
f ,
i ln
() = La ,
0
0
a
where N , La are arbitrary integers and fa are real numbers. Substituting the above
asymptotic values in Eq. (3.21) we derive the following mass formula,

n 1
4(n 1)
j (f+ f )
M=
,
sin

n1
n+1
02
j = N+ N = 0, 1, . . . , (n 2).
(3.23)
We can further simplify Eq. (3.23) by noting that according to Eqs. (2.28) and (2.34) we
have
n1
4
el
(f+ f ),
Qmag = 2 j .
Qel
(3.24)
1 Qn = 4
n+1
0
el
Therefore the 1-soliton masses are independent of Q and Qel
1 + Qn ,
2
el
0 Qmag Qel
4(n 1)
1 + Qn
M=
sin
.
4(n 1)
02
(3.25)
457
As in the case of the axial and vector An (p = 1) dyonic IM [7], one expects the 1-solitons
of the corresponding vector gauged An (p = 2) IM to exhibit mass spectrum with Qel
a
a . An important question to be answered concerns the relation between f(a) ,
replaced by Q
N , L introduced in (3.22) and the parameters that appear in the soliton solution (3.14)
(3.18), and further the question of whether we can have f+a = fa , N+ = N , etc. The
analysis is quite similar to the An (p = 1) case presented in Ref. [7], and is based on (a) the
soliton conservation laws in the form (3.8)(3.10) and (b) the direct evaluations of the
asymptotics at x of the fields l , a , a , Ra from Eqs. (3.14)(3.18). The result is
as follows,

l
2
l
Nl + K
,
l () =
0 n 1
i.e., j(a) = Qa = 0,

1
2
(n + 1)
f+ =
,
R(+) =
(n + 1) + 2S+
0
0
n1

2
(n + 1)
1
(n + 1) Sign()
2S
R() =
f =
0
0
n1
a () = a (+),
(3.26)
and from (3.8)(3.10) one gets further restrictions,

Nl = N
l
K
= 0,
(mod (n 1)),
S = N
(mod (n 1)),
N+ = N
as one can see from Eqs. (3.10) and (3.15). The R1 + Rn asymptotics are certain functions
of and , but they do not enter directly in the computation of the 1-soliton spectrum.
We should mention that the origin of the factor Sign() in the explicit form of the
soliton solutions (3.13), (3.15) and (3.18) as well as in the boundary values R() in
Eq. (3.26) comes from the consecutive application of Eq. (3.8) for x and t = t0
fixed. This leads to the following equation for f :

2
n1
sin
f N = sin .
n1 n+1
Its general solution turns out to depend on the sign of the angle . The same argument
takes place in the determination of the constants of integration Sp (see Eq. (3.12)) from
Eq. (3.8).
One might wonder whether massless 1-solitons exists. Indeed for specific values of
Qmag and Qel
a , when the following relation
el
02 Qmag Qel
1 + Qn = 4(n 1)s,
s Z,
(3.27)
takes place, we have M = 0 as one can see from Eq. (3.25). However by substituting
the explicit form of Qel
a in terms of f and , we conclude that Eq. (3.27) is satisfied for
= s = (s 12 ) only. Since cos(s ) = 0 leads to N(s ) = 0 and f (s ) = + cos(s )+
1
2 ln X0 = const and therefore the solutions for such values of = s coincide with the
trivial constant vacua solution. Hence, they do not describe 1-solitons. In conclusion, the
458
charges of the proper 1-soliton should satisfy the following selection rule,
el
02 Qmag Qel
1 + Qn = 4(n 1)s.
(3.28)
3.4. Semiclassical quantization

An important feature of the (Q1 , Qn ) charged topological 1-soliton we have constructed
is that at the rest frame it represents periodic particle-like motion (due to the dependence
2
of a , a ) with period = sin()
. Then, similarly to the SG breather [29], NLS and Lund
Regge 1-solitons [12] as well as U (1)-charged 1-solitons of the An (p = 1) IM, one can
apply the field theoretic analog of the BohrSommerfeld quantization rule

S
ax
+ E(v = 0) =
jel Z,
L
=
,
M
dx M M = 2jel ,
dt
Hax = M M Lax
(3.29)
in order to derive the semiclassical 1-soliton spectrum. Taking into account that a =
1 0,el
(see Eqs. (2.34), (2.38), (2.39)) and the following simple t-dependence of our
0 Ja
charged 1-soliton solution (3.14)(3.18) at rest frame ( = eb , cosh(b) = 1, sinh(b) = 0):
t l = t l = t (a a ) = t Ra = 0,
t ln
1
n
= t ln
= 2i sin()
1
n
we find that

dt

dx M M =

dt
dx a a =
2
(Q1 Qn ).
02
(3.30)
Therefore the BohrSommerfeld quantization rule implies

Q1 Qn = 02 jel ,
or equivalently

2 jel
+ 0 ,
Q1 = 0
2
Q1 = 02

jel
+ 0 ,
2
0 = 0 (, ),
(3.31)
where 0 is an arbitrary parameter, i.e., only the difference of the charges is quantized.
Their sum remains continuous. Taking into account the improvements (2.45), (2.46)
of the electric charges Qel
a introduced by topological -terms (2.42), (2.43) and the
corresponding topological shifts in the momenta a (i.e., of the electric currents Ja0,el
Eqs. (2.45)) we obtain the following (improved) charge quantization

2
(1)
impr
impr
02 jel = Q1 Qn = Q1 Qn 0 (1) (n) j +
j j(n)
2
2
and, therefore,
LQ = Q1 Qn = 02
459

( (1) (n) )
(1)
(n)
j +
j j
jel +
.
2
4
Since both 0 and are arbitrary reals we can choose such that 20 =
Then the charge spectrum takes the form,

2
1
(1)
(n)
Q1 = 0 jel + (1) j +
j + j
,
2
4

02
1 (n)
(1)
(n)
j + j
Qn =
jel + j
2
4
and the 1-soliton improved mass formula (3.25) becomes

4(n 1) (4j 02 jel )
.
sin
Mjel ,j =

4(n 1)
02
(3.32)
(n)
4 (j
j(1)).
(3.33)
(3.34)
The semiclassical version of the selection rule (3.28) that singles out the true charged
topological solitons takes now the form (02 = 2
k ):
jel =

4
j s(n 1) = 2k s(n 1) j ,
2
0
(3.35)
where jel Z, j = 0, 1, . . . , n 2 mod (n 1). Therefore for integer k, we have to

exclude certain jel according to Eq. (3.35), i.e., for fixed topological charge j , not all
electric charges jel are allowed. As a consequence the mass formula (3.34) makes sense
for this specific set of allowed (jel , j ) only.
Similarly to the An (p = 1) model (1.1) (see Sections 2 and 5 of Ref. [7]) due to
T-duality that maps electric charges of the axial model into the Qa -topological charges
of the vector model (and vice versa), the topological charges Qa (2.29) (Qa = 0 for our
1-soliton solution) also acquires improvements

a,impr
Q
(3.36)
= fa a j .
fa ( a ) is certain function of a determined by T-duality (see, for example, Eqs. (2.40) of
Ref. [7]). The complete semiclassical spectrum of our (Q1 , Qn )-charged 1-solitons (3.14)
(3.18) is represented by Eqs. (3.33)(3.36).
4. Soliton vertex operators

4.1. Dressing transformations
The first order soliton equations (3.3), (3.4) are by no means an effective tool for
deriving the explicit form (3.14)(3.18) of the multicharged topological 1-solitons of
the An (p = 2) dyonic IM. The systematic construction of the different species of N solitons and breathers however, requires more powerful group theoretical methods as the
460
(1)
An -vertex operators and the corresponding -functions [10]. An important advantage of

such method is that they make transparent the algebraic structure underlying the soliton
solutions, namely, the level one representation of the twisted An affine KacMoody algebra
[10,16]. The dressing transformation [28] are known to be the origin of all those methods
(including the vacuum Bcklund transformation (3.1) of Section 3). Let us consider two
0 , s = 1, 2, of Eqs. (2.17) written for the case of A(1)
arbitrary solutions Bs G
n extended
by d and the central term c, i.e.,
Bs = g0s es c+s d .
= A(B
s ) are related
The corresponding Lax (LS) connections (2.21) A(s) = A(Bs ), A(s)
by gauge (dressing) transformations ,+ = exp G<,> ,
A (2) = A (1)1 + ( )1 .
(4.1)
They leave invariant the equations of motion (2.17) as well as the auxiliar linear problem,
i.e., the pure gauge A defined in terms of the monodromy matrix T (Bs ),

A(Bs ) Ts (Bs ) = 0.
(4.2)
The consistency of Eqs. (4.1) and (4.2) imply the following relations
T2 = T1 ,
i.e., + T1 = T1 g (1) ,
(4.3)
is an arbitrary constant element of the corresponding affine group. Suppose

where g (1) G
T1 = T0 (Bvac ) is the vacuum solution,
1
= ,
Bvac Bvac
Avac = ,
1
vac Bvac
B
= 2 zc,
Avac = + + 2 zc,
(4.4)
and T0 = exp(z ) exp(z+ ) as one can easily check by using the fact that [+ , ] =
2 c. According to Eqs. (4.1) and (4.3), every solution T2 = T (B) can be obtained from the
vacuum configuration (4.4) by an appropriate gauge transformation . In fact, Eqs. (4.1)
with Avac and A vac as in Eq. (4.4) and
A(B) = B B 1 ,
A = + + BB
allows to derive as functionals of B, i.e., = (B). We next apply Eqs. (4.3),

1
1 + = Tvac g (1) Tvac
(4.5)
in order to obtain a nontrivial field configuration B in terms of g (1) G and certain highest
weight (h.w.) representation of the twisted algebra A(1)
n as we shall see in Section 4.2. The
first step consists in substituting Avac , A vac and A(B), A(B)

in Eq. (4.1) and then solving
it grade by grade remembering that may be decomposed in the form of infinite products
= et (0) et (1) ,
+ = ev(0)ev(1) ,
where t (i) and v(i), i = 1, 2, . . . , denote linear combinations of grade q = i generators.

For grade zero we find
t (0) = H (z),
2 z
zc
ev(0) = BeG(z)
461
where the arbitrary functions H (z), G(z) G00 and are fixed to zero due to the subsidiary
constraints (2.18), (2.19), i.e., H (z) = G(z) = 0. The equations for v(1), t (1) appears to
be of the form
1
BB
= [t (1), + ] + 2 zc.
B 1 B 2 z c = [v(1), ],
The next step is to consider certain matrix elements (taken for the h.w. representation |l !)
of Eq. (4.5). Since v(i)|l ! = 0 and "l |t (i) = 0, i > 0, we conclude that
"l |B|l !e
2 z
z
= "l |T0 g (1) T01 |l !.
(4.6)
Taking into account the explicit parametrization of the zero grade subgroup element B
(2.10) in terms of the physical fields, , i , a , a and choosing specific matrix elements
we derive their explicit spacetime dependence,
2 z
z
0 e
= "0 |T0 g (1) T01 |0 !,
nR1
Rn
2 z
z
R1
nRn
2 z
z
R1 e n+1 + n+1 +
Rn e n+1 + n+1 +
= "1 |T0 g (1) T01 |1 !,

= "n |T0 g (1) T01 |n !,
j e1 j R1 +n j Rn +j1 +
2 z
z
l e
= "j |T0 g (1) T01 |j !,
(1 l + 12 l,1 )R1 +( 21 l,n +n l )Rl +2 zz
= "l |T0 g
(1)
1
(0)
T01 E
|l !,
l
l = 1, n,
l e(1 l + 2 R1 l,1 )R1 +( 2 l,n +n l )Rn +
2 z
z
= "l |E(0)
T0 g (1) T01 |l !,
l
j = 2, . . . , n 1,
l = 1, n,
(4.7)
where for An we have 1 j = n+1j

n+1 , j = 1, 2, . . . , n 1. In order to make the
construction of the solution (4.7) complete it remains to specify the constant affine group
element g (1) . The constant element g (1) encodes the information (including topological
properties) about the N -soliton structure of Eqs. (2.17). We shall see in the next Section 4.2
that all 1-soliton solutions of the A(1)
n (p = 2) dyonic IM (1.2)the neutral (0, 0), U (1)charged, (Q1 , 0) and (0, Qn ), and the U (1) U (1)-charged (Q1 , Qn )are represented
(1)
by certain level one vertex operators (i.e., h.w. representations) of the twisted An algebra.
(1)
The twist being determined by the corresponding An (p = 2) graded structure Q, , G00 .
The multisoliton solutions and the breathers are associated to appropriate tensor products
of these basic vertices.
4.2. Vertex operators
Given A(1)
n affine algebra,
m n
Hi , Hj = cmm+n,0 i,j
m n
Hi , E = ()i Em+n ,
i, j = 1, . . . , rank G,
462
m+n
(, )E+
if + is a root,
m
m+n
E , E = H
(4.8)
+
cm
m+n,0 if + = 0,
0
otherwise
(m, n Z and c is its level to be taken to 1 in the vertex operator representations), provided
with the graded structure {Q, , G00 } as in Eqs. (2.9). It is important to mention that

a zn1 s1
the chiral two-loop KacMoody algebra spanned by J a (z; ) = n,sZ Jn,s
a
m
m
(where J0,m denote Hi and E ) has two independent central extensions:

a
Js1 (z1 ), Jsb2 (z2 ) = if abc Jsc1 +s2 (z1 )(z1 z2 )
k
c
+ i g ab (z1 z2 )s1 +s2 ,0 +
s1 g ab (z1 z2 )s1 +s2 ,0 .
2
2

n
The label k that appear as the coupling 2 = 2

k in the WZW action is related to the mode
expansion in the spacetime variables z, z . The central term c appearing in (4.8) is related to
the mode expansion in the spectral parameter . In the vertex operator construction below,
a
, representing the dressing symmetry takes
the only -part of the algebra spanned by J0,m
place.
Since form a Heisenberg subalgebra,
[+ , ] = 2 c
and we have to calculate the matrix elements ( -functions), say
"0 |ez ez + g (1) ez+ ez |0 !,
it is instructive to introduce a new basis for the A(1)
n , such that appear as part of its
Cartan affine subalgebra. Let us first observe that the following specific linear combination
of Hi(l) , E(k) [16],
a
ba+m(n1)
n1a
E(m)
i+1 +i+2 ++i+a
i=1
a
(m+1)
E(
,
i+1 +i+2 ++i+n1a )
i=1
a = 1, . . . , n 2,

(n 1)
0
(1 + n ) H (m) ,
bm(n1) =
2

(n + 1)
(1 n ) H (m)
bm =
2(n 1)
(4.9)
close new n-dimensional Heisenberg subalgebra,

b
a
= a + m(n 1) m,l a,b ,
, bb+l(n1)
ba+m(n1)
a

ba+m(n1) , bm = 0, a = 0, 1, . . . , n 1,
[bm , bl ] = mm+l,0 .
(4.10)
Therefore we can consider the generators

a = 0, 1, . . . , n 1, together with
bm (and their conjugate) as the generators of the affine Cartan subalgebra in the new basis
a
ba+m(n1)
,
463
(1)
for the twisted An .6 Next step is to complete this basis with the corresponding new step
operators e(l)
. Since we have by definition that

= b11
+ = b11 ,
commute with b00 and b0 , in the vanishing central charge case, they all commute and hence
share the same set of eigenvectors F ( ),

0

b0 , F ( ) = f00 ( )F ( ),
, F ( ) = f ( )F ( ),
[b0 , F ( )] = f0 ( )F ( ).
(4.11)
Following Ref. [16] we find four different types of eigenvectors

Fa,j ( ) =

(m)
c
+
m(n1)
hi+1
wa(ip)
a
(w 1)
mZ
n2
i=1
p=1
wbj (b+m(n1))
b=1 mZ
b
n2
n1b
wa(i1) E(m)
i+1 +i+2 ++i+b
i=1
a(i+n2b)
(m+1)
E(
i+1 +i+2 ++i+n1b )

,
(4.12)
i=1
2i
where j = 1, . . . , n 1, a = 1, . . . , n 2 and w = exp( (n1)
), together with
1,l ( ) =
F
n2
m(n1)
2
wpl p Em1 ++p+1 ,

mZ
1,l ( ) =
F
F ( ) =
p=0
n2
m(n1)
2
wpl p Emn ++np ,
mZ
l = 0, . . . , n 2,
p=0
m
(m+1)(n1)+1E(
.
1 ++n )
(4.13)
mZ
Their eigenvalues are obtained from

, Fa,j ( ) = wj wa 1 Fa,j ( ),

1,l ( ),
1,l ( ) = 1 wl F
,F

1,l ( ) = 1 wl F
1,l ( ),
,F

, F ( ) = 0.
(4.14)
6 Note that ba
a+m(n1) , a = 0, has no zero modes, i.e., only two of the generators of the new Cartan
0
, bm are untwisted.
subalgebra, namely bm(n1)
464
1,l , F
1,l , F are eigenvectors of all oscillators ba
In fact, Fa,j , F
a+m(n1) and bm , in
particular,
a1

ba1 +l(n1) , Fa2 ,j ( ) = wa1 j wa1 a2 1 a1 +l(n1) Fa2 ,j ( ),
[bl , Fa,j ( )] = 0.
(4.15)
We can therefore identify the step operators in the new basis as follows

d a+l(n1)
e(l)
=
Fa,j ( ).
j
2i
(1)
With the new basis at hand, we define the An affine group element g (1) as
g (1) = edF ( ),
1,l , F
1,l , F
F ( ) = Fa,j , F
(1)
or more generally as a product including all the An generators,

g (1) = ed1 F1 (1 ) ed2 F2 (2 ) edN FN (N ) .
(4.16)
The new basis introduced above drastically simplifies the calculation of the -functions
(4.7),

T0 g (1) T01 = exp d( )F ( ) = 1 + d( )F ( ),
(4.17)
where ( ) = exp(zf ( ) + z f + ( )). The last equality in (4.17) reflects the fact that
F 2 ( ) = 0 for A(1)
n due to the well-known properties of the short distance OPE (see
Ref. [16]).
An important question concerns the specific choice of the form (4.16) of g (1) that leads
(1)
to different species of neutral and charged solitons and breathers. As in the An (p = 1)
dyonic IM case [7,8] taking
ga(1) ( ) = edFa,n1 ( )
(4.18)
(1)
we reproduce the abelian affine An2 -Toda neutral 1-soliton solutions, since a =
a = 0. The corresponding U (1) charges vanish identically, i.e., Q1 = Qn = 0. Another
indication of the neutrality of such solution is that Fa,n1 ( ) has zero eigenvalue with
respect to the generators of the U (1) U (1) charges, namely, 1 H and n H .
The choice
(1)
g1 (1 , 2 ) = ed1 F1,1 (2 ) ed2 F1,1 (1 ) ,
gn(1) (1 , 2 ) = ed1 F1,1 (2 ) ed2 F1,1 (1 )
(4.19)
for 1 = eBi(+) , 2 = eB+i (and further conditions on d1 , d2 , d1 , d2 as in Ref. [8])

leads to nontrivial U (1) charged 1 (Q1 , 0) or n (0, Qn ) 1-soliton solutions. They
provide 1-soliton solutions with n = n = 0 or 1 = 1 = 0 coinciding with the
(1)
electrically charged 1-soliton solution of the An (p = 1) dyonic IM [7,8]. Similarly, the
3-vertex solution generated by
(1)
(1)
g1a (1 , 2 , 3 ) = g1 (1 , 2 )ga(1) (3 ),
(1)
gna
(1 , 2 , 3 ) = gn(1) (1 , 2 )ga(1) (3 )
(4.20)
465
for an appropriate choice of i , i = 1, 2, 3, and d, d1 , d2 , d1 , d2 (as in Ref. [8]) give rise to

U (1)-charged 1 - and n -breathers.
(1)
The simplest genuine An (p = 2) nontrivial (Q1 , Qn )-charged 1-soliton solution
(3.14)(3.18) correspond to

(1)
g1n
(1 , 2 , 3 , 4 ) = ed1 F1,1 (2 ) ed2 F1,1 (1 ) ed4 F1,1 (4 ) ed3 F1,1 (3 )
(4.21)
for 1 = eBi(+) , 2 = eB+i and d1 d2 = d3 d4 = d0 . The (Q1 , Qn )-charged breather

can be generated by taking
(1)
(1)
(1 , 2 , 3 , 4 , 5 ) = g1n
(1 , 2 , 3 , 4 )ga(1) (5 )
g1n
(4.22)
and special choice of l , l = 1, 2, 3, 4, 5, and the coefficients dl . The multisoliton solutions

are indeed represented by g (1) that includes products of the corresponding 1-soliton
(1)
(1)
vertices g1n and ga , say
g (1) =
N
!

g (1) 1s , 2s
s=1
again with specific relations among the parameters 1s , 2s , d1s , d2s , d1s , d2s .
5. Dyonic A(1)
n (p = 2, q) hierarchy
(1)
5.1. Constrained KP hierarchies of An (p = 2, q = 2) dyonic type

Our motivation to introduce A(1)
n (p = 2, q) dyonic hierarchy (of which q = 1
integrable model (1.2) is a member) and to study the relation between the 1-solitons of
q = 1 and say q = 2, dyonic integrable models is twofold:
(a) as we shall show in Section 5.3, the knowledge of (Q1 , Qn )-charged 1-soliton (3.14),
(3.18) of q = 1 model (1.2) allows to construct the corresponding 1-solitons of all q 2
models by simple change of variables (boththe field ones: ra (x, t), qa (x, t), ul (x, t)
a , a , l and the spacetime ones: z x, z t q );
(b) since the equations of motion of q 2 models are much simpler than the q = 1
ones (that follows from (1.2)), the Hirota method [10,30] (and the explicit derivation of
G0 -, -functions) works more effectively in say, q = 2 case than in the q = 1 one.
This suggests that in construction of multisolitons and breathers to first elaborate them
for q = 2 model and then to apply the above change of variables in order to find the
corresponding solutions of the q = 1 model. The origin of the A(1)
n (p = 2, q) dyonic
hierarchy is in the fact that keeping a part of q = 1 graded structure (2.9): Q, + , G00 and
q
replacing (of grade q = 1) by an arbitrary constant element Dq of Q-grade q 1
466
such that7

1
D1 =
Dq Ker(ad )
(5.1)
one can generate an infinite hierarchy of new integrable models. Their equations of motion
have the following compact zero curvature form [16,18]:
tq D0 x Dq [Dq , D0 ] = 0,
(5.2)
where
= x ) = + + (x g0 )g 1 ,
D0 = A(
0
Dq =
q
Dq(l) Gl .
Dq(l) ,
(5.3)
l=0
An important difference with q = 1 case is that now the physical fields ra (x, t),
qa (x, t) and ul (x, t) parametrize the zero grade algebra G0 (and not the zero grade group
element g0 G0 as in q = 1 case):
n2

(0)
(0)
+
q
E
ul h(0)
ra E
+
a a
l+1 .
a
A0 = (x g0 )g01 =
a=1,n
(5.4)
l=1
(l)
The remaining grade l = 0, 1, . . . , q 1 gauge potentials Dq introduce a set of

auxiliary fields Vk(0)(x, t), Vk(1) (x, t), etc. with k = 1, 2, . . . , n + 4, n 4:
Dq(0) =
n
(0) (0)
(0) (0)
(0) (0)
(0) (0)
Vi(0)hi + Vn+1
E1 + Vn+2
En + Vn+3
E1 + Vn+4
En ,
i=1
Dq(1) =
n2
(1) (0)
(1) (1)
Vi(1) E(0)
+ Vn1
E1 +2 + Vn(1) E(0)
+ Vn+1
E(2 ++n1 )
n1 +n
i+1
i=1
(1)
(0)
(1)
(0)
(1)
(0)
+ Vn+2 E(1 ++n1 ) + Vn+3 E(2 ++n ) + Vn+4 E(1 ++n ) ,
(5.5)
etc. Writing Eqs. (5.1) for each grade 0, 1, . . . , q + 1 we get the following system of
equations:
(q)

Dq , + = 0,
(q)
(q1)

(q)
x Dq + Dq , A0 + Dq
, + = 0,
(q2)

(q1)
(q1)
x Dq
+ Dq
, A0 + Dq
, + = 0,
..
.
(0)

(1)
(1)
x Dq + Dq , A0 + Dq , + = 0,

tq A0 x Dq(0) Dq(0) , A0 = 0.
(5.6)
7 The additional condition [D q , ] = 0 is imposed in order to obtain local equations of motion and allows to
q
relate the 1-solitons of q-model to the ones of q = 1 model.
467
The recipe in deriving the system of differential equations (of order q in x) for ra , qa , ul
(q)
consists in first determining Dq Ker(ad ) (i.e., starting from the first equation). For,
say, q = 2 we obtain (n 4)
D2(2)
= b2(2)
n3
E(0)
i+1 +i+2
i=1
2
(1)
E(
,
i+1 ++i+n3 )
(5.7)
i=1
(2)
a
where b2 is a particular case of ba+m(n1)
, a = 2, m = 0, given by Eq. (4.9). Substituting
(q1)
Eq. (5.7) in the second of the equations (5.6) we find a part of the auxiliary fields Vk
parametrizing Im(ad + ) in terms of ra , qa , ul . The third equation determines the remaining
(q1)
(q2)
lying in the Ker(ad + ) and those fields Vk
lying in the Im(ad(+ )), etc.
part of Vk
Applying this procedure for the case q = 2 we find the following system of equations
defining the A(1)
n (p = 2, q = 2) nonrelativistic dyonic integrable model:

n2
n2
2q1
1
2
t2 q1 x q1 +
x uk +
kkj uk uj + r1 q1 + rn qn = 0,
n1
2
k=1
k,j =1

n2

2r
1
2
t2 r1 + x r1
(1 n)x u1 +
x uk
n1
k=1

n2
1
+
2
k,j =1

n2
n2
2qn
1
2
t2 qn + x qn
x uk +
n1
2
k=1
k,j =1

n2

2rn
t2 rn x2 rn +
x uk
(1 n)x un2 +
n1
k=1

n2
1
+
2
k,j =1
(0)
t2 ul + x Vl
= 0,
l = 1, 2, . . . , n 2,
(5.8)
where
1
(x Ll + Ml ),
n1
l1
n2

Ll = 2(n l 1)
uk + (n 2l 1)ul 2l
uk ,
(0)
Vl
k=1
k=l+1
l

Ml = 2 (n l 1)r1 q1 lrn qn + (n l 1)
kij ui uj
i,j =1
468
n2
kij ui uj (n 2l 1)u2l .
i,j =l
We take the q = 2 A(1)

n (p = 2) integrable model defined by the above system of equations
as a representative of A(1)
n (p = 2) positive grade q 2 dyonic hierarchy to further study the
relation between its soliton solutions with the 1-solitons (3.14)(3.18) of q = 1 dyonic
model.
The simplest case n = 3 should be treated separately. Taking D2(s) (n = 3), s = 0, 1, 2,
in the form
D2(2) (n = 3) = (3 1 ) H (1),
(1)
(1)
+ r1 E
+ q3 E(0)
r3 E
,
D2(1) (n = 3) = q1 E(0)
1 +2
1 2
2 +3
2 3
(0)
D2(0) (n = 3) = (x q1 + q1 u)E(0)
(x q3 + q3 u)E(0)
(x r1 r1 u)E
1
1
3
1
(0)
(0)
+ (x r3 r3 u)E3 + (r1 q1 r3 q3 )h2 ,
2
(5.9)
we get the explicit form (5.15) of the equations of motion for the A(1)
2 (p = q = 2)
nonrelativistic IM.
5.2. Charged 1-solitons of q = 2 model
The construction of the soliton solutions for each member of An (p = 2, q) hierarchy
by vertex operator method is quite similar to the q = 1 case described in Section 4. The
main difference is in the form of the vacua (constant) solutions:
(q)
Dq(v) = Dq ,
D0(v) = + ,
(q)
Tq(v) = etq Dq ex+
(5.10)
(2)
(for q = 2 D2 is given by grade 2 constant element (5.7)) and in the choice of new set of
Tv g (1) Tv1 -matrix elements (different from (4.7) for q = 1 case) representing physical
fields ra , qa , ul . Note that the form of g (1) (say (4.18)(4.21)) remains unchanged since
(q)
[Dq , ] = 0, henceforth they share the same set of eigenvectors (4.12), (4.13). Following
the dressing procedure of Section 4.1 we find for q = 2 model:
(x B)B 1 = A0 = [t(1), + ],
ev(0) = B,
t(1) =
n2
(0)
(0)
(0)
(ul + x )E
r1 E
+ rn E
l+1
1 2
n1 n
l=1
(1)
(1)
(1)
(1)
+ q1 E1 ++n1 qn E2 ++n + x E2 ++n1 + W E1 ++n ,

(5.11)
(1)
where W (t, x) is an arbitrary function since E1 ++n Ker(ad + ); = 2 xt. Taking

into account the explicit form (5.11) of v(0), t(1) and (and the definition of the highest
weight states |j !, "j |) we consider appropriate matrix elements of Eqs. (4.5) in order to
469
extract the -functions representing ra , qa , ul :

0 e = "0 |Tv g (1) Tv
|0 !,
1,(n1) q1 e = "0 |E1 n1 Tv g (1) Tv1 |0 !,

(1)
2n qn e = "0 |E2 n Tv g (1) Tv1 |0 !,

(1)
12 r1 e(1 ) = "1 |E(0)

T g (1) Tv1 |1 !,
1 +2 v
n1,n rn e(n ) = "n |En1 +n Tv g (1) Tv1 |n !,
(0)
l e(l ) = "l |Tv g (1) Tv1 |l !,
l = 2, 3, . . . , n 1,
(5.12)
where
(1 ) = 21 R1 + 1 n Rn + ,
(n ) = 2n Rn + 1 n R1 + ,
(l ) = l1 + l 1 R1 + l n Rn + ,
l
n+1l
,
1 l =
.
l n =
n+1
n+1
l = 2, . . . , n 1,
(5.13)
(1)
Therefore the soliton solutions of An (p = 2, q = 2) IM (5.8) can be written in terms of

the above -functions as follows
12
n1,n
r1 = ,
rn =
,
1
n
1,(n1)
2n
q1 =
,
qn =
,
0
0
l+1
ul = x ln
(5.14)
, l = 1, 2, . . . , n 2.
0
It remains to calculate the matrix elements (5.12) for say, g (1) represented by 4-vertex
describing (Q1 , Qn )-charged 1-soliton. We shall present here the explicit form of this
(4-vertex) solution for the simplest case n = 3, q = 2 only, since the general case of
U (1) U (1)-charged 1-soliton of An (p = 2, q = 2) system (5.8) n = 3, 4, . . . will be
derived in Section 5.3 by applying the vacua Bcklund transformation method of Section 3.
For n = 3 the system (5.8) gets the following simple form (t = t2 ):
t q1 x2 q1 + q1 x u + q1 u2 + r1 q12 + r3 q1 q3 = 0,
t r1 + x2 r1 + r1 x u r1 u2 r12 q1 r1 r3 q3 = 0,
t q3 + x2 q3 q3 x u q3 u2 r3 q32 r1 q1 q3 = 0,
t r3 x2 r3 r3 x u + r3 u2 + r32 q3 + r3 r1 q1 = 0,
t u + x (r1 q1 ) x (r3 q3 ) = 0.
(5.15)
Its charged 1-solution (4-vertex) solution is obtained by simple algebraic manipulations

(involving A(1)
3 -algebra)
0 = 1 + (c12 b1 b2 + c34 b3 b4 )1 2 ,
2 = 1 + (d12 b1 b2 + d34b3 b4 )1 2 ,
470
1,n1 = b1 f1 1 ,
1,2 = b2 e2 2 ,
2n = b4 i4 2 ,
n1,n = b3 g3 1 ,
(5.16)
where
1 (1 ) = ex1 +t 1 ,
2 (2 ) = ex2t 2 ,
(5.17)
21 22
212 2
=
d
= d12 ,
c12 =
,
c
=
34
34
(1 2 )2 (1 + 2 )
(1 2 )2 (1 + 2 )
e2 = 22 = i4 .
f1 = 2 1 = g3 ,
(5.18)
The above solution is a particular case of the more general (4-parameter i ) 4-vertex (2soliton) solution, given by Eqs. (A.1)(A.6) of Appendix A. The restrictions
1 = 3 ,
1 = 2 = eB+i
4 = 2 ,
(5.19)
we have imposed on it, ensure that the solution (5.16)(5.19) represents finite (positive)
energy (Q1 , Qn )-charged topological 1-soliton, as we shall show in Section 5.4. One
can easily verify by substituting ra , qa , u (5.14) (written in terms of -functions (5.16)
or (A.2)) in the system (5.15) that they indeed satisfy Eqs. (5.15), i.e., that -function
method provides solutions of this system. It is important to note that for n = 3 and q > 2
arbitrary integer (i.e., A(1)
3 (p = 2, q) IM) the form of the corresponding 1-soliton solutions
(q)
is the same as in q = 2 case given by Eqs. (5.16), (5.18) with the only difference that 1
(q)
and 2 have been changed to:
1 ( ) = ex +tq ,
(q)
2 ( ) = ex tq .
(q)
(5.20)
(1)
The system of equations for A3 (p = 2, q = 2) model is indeed of order q in x (i.e., the

q
higher derivative is x ) and involves more products of derivatives terms then q = 2 system
(5.15).
5.3. 1-soliton relations: q = 2 vs q = 1 solitons
(1)
The common graded algebraic structure of q = 1 and q = 2 An (p = 2) integrable

models address the question whether one can construct q = 2 solitons in terms of
corresponding q = 1 ones. By comparing q = 1 and q = 2 dressing transformations
(4.1), the form of Tv g (1) Tv1 matrix elements
Tv (q)ebF ( )Tv1 (q) = ebq ( )F ( ),
z z +
q = 1, 2,
(2)
e ,
Tv (q = 2) = et2 B2 ex+ ,
Tv (q = 1) = e

z
(2) ( ) = exp t2 2 + x
(1) ( ) = exp + z ,
(5.21)
1,l , F
1,l , Fa, , F } do not depend on q, we observe that
and remembering that F ( ) = {F
the only difference between q = 1 and q = 2 -functions (4.7), (5.12) is in the factors
471
q ( ). Therefore, the simple change of variables

z
t2 2
z x,
(5.22)
transforms the q = 1 -functions in the corresponding q = 2 ones. The problem however

is that q = 2 fields ra (x, t), qa (x, t), ul (x, t) and q = 1 ones a (z, z ), a (z, z ), l (z, z )
are represented by different sets of G0 - -functions. This reflects the fact that by definition
q = 2 field variables parametrize the G0 -algebra
A0 = (x B)B 1 =
n2

(0)
(0)
+
q
E
ul h(0)
ra E
+
a
a
l+1 + x c
a
a=1,n
(5.23)
l=1
while the q = 1 fields are related to G0 -group element

n2

(0)
(0)
B = exp
a Ea exp
l hl+1 + 1 H R1 + n H Rn + c
a=1,n
exp
l=1
.
a E(0)
a
(5.24)
a=1,n
In order to make this relation explicit we introduce the q = 1 G0 WZW currents ra (z, z ),
qa (z, z ), u l (z, z ):
1
=
(B)B
n2

(0)
(0)
ra E
+
+
q
E
u l h(0)
a
l+1 + c
a
a
a=1,n
(5.25)
l=1
and realize them in terms of a (z, z ), a (z, z ), l (z, z ), Ra (z, z ):

#
"
nl
l+1
u l = l +
R1 +
Rn , l = 1, . . . , n 2,
n+1
n+1

1
1
R1 1
2
q1 = e
1 1 R1 ,
2

1
1
Rn n2
2
qn = e
n n Rn ,
2
"
#

1
1
2 R1
1 2
1
r1 = e
R1 ,
1 + 1 1 e
1 1 1 e 1 1 +
2
"
#

n
12 Rn
n2 2
n2
1 + n n e
rn = e
Rn . (5.26)
n n +
n n n2 e
2
For (Q1 , Qn )-1-solitons (i.e., 4-vertex solution) the complicated ra , qa expressions
considerably simplifies, due to 1st-order soliton equations (3.3):
1
1
1
1
1
1
1 e 2 R1 n+1 (R1 Rn ) ,
qn = n e 2 Rn + n+1 (R1 Rn ) ,
1
1
1
1
1
1
rn = n en2 2 Rn + n+1 (R1 Rn ) .
r1 = 1 e1 2 R1 + n+1 (R1 Rn ) ,
q1 =
(5.27)
472
Since by definition we have

z
2
qa (x, t) = qa z x, t ,

z
2
ra (x, t) = ra z x, t ,

z
2
ul (x, t) = u l z x, t
once the q = 1 one-soliton solution is known (3.14), (3.15), (3.18), then Eqs. (5.27)
(together with the change of variables (5.22)) determine the corresponding q = 2 onesoliton solution. The same is true for generic (say, multisoliton) solutions, but in this case
one has to use more complicated relations (5.26) between G0 -currents ra , qa , u l and G0 fields a , a , l , Ra . Let us consider the simplest case of n = 3 in order to demonstrate
the relation (5.27) between q = 2 and q = 1 one-solitons established above. We take
q = 2 (n = 3) 1-soliton in the form (5.14) with -functions given by Eqs. (5.16), (5.18),
(5.19). We next change the variables x, t2 to z, z (and that appears in ( ) according
to Eqs. (5.22)). Finally we apply Eqs. (5.27) replacing a , a , 1 = , Ra (a = 1, 3) by its
explicit form of q = 1 model one-solitons (3.13)(3.18). As a result we find the following
expressions of the q = 2 one-soliton parameters bi (i = 1, 2, 3, 4) in terms of the q = 1
ones D, d1 , d2 , 2 = d1 + d2 , , p1 :
b1 = b3 e d1 ,
b2 = b4 e d2 ,
D 2 Neip1 +
,
b1 =
2 (1 e2+2i )
D 2 Nei(p1 )+
b2 =
2 (1 e2+2i )
and the inverse relations:

b = b1 b2 b3 b4 ,
b = b1 b2 1 e2 ,
b+ + b
b+ + b i
,
D=
e ei ,
e2 =
b+ b
b+ b
e2ip1 =
(5.28)
b1 i
e
b2
(5.29)
and the appropriate choice of the soliton center of mass X0 . Similar relations between
the charged 1-solitons of q = 2 and q = 1 An (p = 1) dyonic models (with one U (1)
symmetry) have been established in our recent paper [20].
5.4. Conserved charges
As in the relativistic q = 1 A(1)
n (p = 2) model (1.2) (see Section 3.3) the soliton
spectrum of q = 2 A(1)
n (p = 2) IM (5.8) is determined by the first few conserved charges
(out of the infinite set of commuting charges) namely topological charges (fluxes) Ql ,
0 , momenta P, energy E. They all indeed depend on the asymptotic
particle density Q
a at x only. Since the boundary
values of G0 -group element fields: a , a , l , R
a , ( a () = a () = 0) and ra () = qa () = 0 define
values of l and R
l , as in the q = 1 case all the other charges appears
completely the topological charges Q
l and the soliton velocity . The problem we address in
to be certain functions of these Q
473
this section is (a) to calculate these charges for our 1-soliton solutions and (b) to find the
explicit relation between the conserved charges of q = 1 and q = 2 IMs. First step is to
derive the An (p = 2, q = 2) IM conserved currents. According to its equations of motion
(5.8) the fields ul represents the simplest conserved currents

l = dx ul .
t ul = x Vl(0) ,
Q
T (l) = ul ,
(5.30)
Taking into account Eqs. (5.8) once more one can easily verify that

1
0 = dx T0
Q
T0 = r1 q1 + rn qn + kij ui uj ,
(5.31)
2
defines the particle density conserved current. The momenta and energy densities have
more complicated form and for n = 3 model (see Eqs. (5.15)), they are given by:
Tx = r1 x q1 q1 x r1 r3 x q3 + q3 x r3 2u(r1 q1 r3 q3 ),

P = dx Tx ,
2
1
1
Tt = 2(x r1 x q1 + x r3 x q3 ) + (x u)2 + u2 + r1 q1 + r3 q3 + 2r1 q1 r3 q3
2
2
+ u(r1 x q1 q1 x r1 + r3 x q3 q3 x r3 ),

E = dx Tt .
(5.32)
The systematic way of deriving all the conserved currents (higher Hamiltonians) as
well as their representations as total x -derivatives consists in applying the dressing
e t (i) ).
transformations (4.1) to calculate certain traces of t (k) (remember = i=0
Starting from the vacua dressing:
1
)1
= + 1 + (
+ + BB
and writing it grade by grade (i.e., expanding in
(5.33)
k )
we obtain:
= [t(1), + ] = A0 ,
BB
(5.34)
1
[+ , t(2)] = [A0 , t(1)] + t(1),

(5.35)
2

1
[+ , t(3)] = [A0 , t(2)] [A0 , t(1)], t(1)
6
1
(5.36)
[t(1), t(1)] + t(2),
2
etc. We next multiply Eq. (5.35) by + and by taking trace we get:

1
T0 = tr A20 = x Tr + t(1)
(5.37)
2
which reproduces Eq. (5.31). Taking into account the explicit form (5.11) of t (1) we
further conclude that
n2

ul (n 1)x2 .
T0 = x
(5.38)
l=1
474
One can continue this procedure and to determine t(2) from (5.35), substituting it in
Eq. (5.36) and by taking trace with b2(2) from Eq. (5.7) (and Eq. (5.9) for n = 3 case) to
derive the explicit form of Tx , etc.
We next consider the problem of calculation of q = 2 model (5.8) conserved charges
l and Q
0 and their relation to q = 1 model (1.2) charges Qmag , Q1 , Qn and E, P , M.
Q
According to Eq. (5.34) we have

nl
l+1
ul = l +
(5.39)
R1 +
Rn , = x
n+1
n+1
l are determined by the asymptotic values of the
and therefore the corresponding charges Q
a
fields l , R
l () =
2l
N ,
(n 1)
a () = 2 f (a).
R
(5.40)
We have assumed that for t (and t2 ) fixed (say to zero) the x limits of the fields
a (x2 , t2 ) and l (z, z ), Ra (z, z ) do coincide. As a result we realize Q
l in terms
l (x2 , t2 ), R
of the q = 1 topological and electric charges Qmag , Q1 , Qn (2.28), (2.34) as follows
l =
Q
1
[l(4j Q1 + Qn ) + (n 1)Q1 ].
2(n 1)
(5.41)
0 and its relation with E, P , M of the q = 1 model, we represent

In order to calculate Q
derivative by applying 1st-order soliton equations (3.3) and ra ,
T0 (5.31) as total (x )
qa , ul a , a , l relations (5.26), (5.27):

n1

1
1 k
1
2
,
T0 = 2 Vrel =
(5.42)
e
+ n n =
x F
2
2 2
k=1
(p=2)
Vn
.
Note that the q = 2 coupling constant =

where Vrel is the potential
in (5.42) as a result of simple rescaling
(ra , qa , ul ) ( ra , qa , ul ).
appears
(5.43)
We next remind the form of the q = 1 model stress tensor (T00 , T01 ) derived in Section 3:
n1

1
k
2
T00 = +
x
e
+ n n ,
2
k=1
n1

1
k
2
x
e
+ n n ,
T01 = +
(5.44)
2
k=1
where we have used the relation

1
F + F + = (n 1)c0
(5.45)
475
(that follows from Eqs. (3.4) and F definitions). By comparing (5.42) to Eq. (5.44) we
conclude that:

1
1
P = T01 dx =
E = T00 dx = +
Q0 ,
Q0 ,

n1

0 = dx2 T0 = 1

Q
(5.46)
F
,
F
=
ek + 2 n n ,
dx
2
x
2
2
k=1
where
= eb

is related to the 1-soliton velocity of q = 1 model by

1 vrel 1/2
,
vrel = thb.
1 + vrel
0 is proportional to the q = 1 soliton mass M:

Therefore the q = 2 charge Q

1 1 + vrel 1/2
2(n 1) [02 Qmag Q1 + Qn ]

M=
Q0 =
sin
.
2 1 vrel
4(n 1)
02
0 -charge
The form (5.46) of Q
0 = F
|
Q

(5.47)
(5.48)
(5.49)
l only. We next observe that

suggests that it can be realized in terms of the ul -charges Q

1 Q
n2 = n 3 02 Qmag Q1 + Qn
Q
2(n 1)
as one can see from Eq. (5.41). Substituting it in Eq. (5.48) we find (n = 3)

1 Q
n2
Q
2(n
1)

0 =
Q
sin 2(n 3) .
(5.50)
(5.51)
For particular case n = 3 the analog of the above relation include also the additional
(1)
(1)
nonlocal charge Q1 = 4
(f+ f ):

0 (n = 3) = 4 sin Q1 + Q1 .
Q
(5.52)

2
4
l , Q
0 , P, E) of q = 2 model (5.8) requires the
The complete 1-soliton spectrum (Q
evaluation of the soliton momenta P and energy E (5.32) as well. The same procedure
0 calculation can be applied to the Tx
(1st-order soliton equations, etc.) we have used in Q
and Tt . We shall present the complete soliton spectrum of our q = 2 A(1)
n (p = 2) IMs in
future publication.
6. Dyonic IMs with nonabelian symmetries

We have established in Sections 1 and 2, that the specific choice of the A(1)
n
graded structure (Q, , G00 ) (2.9) giving rise to the dyonic A(1)
n (p = 2) IM (1.2) led
476
to topological solitons carrying internal (Noether) charges U (1) U (1). They appear
as natural generalization of the simplest A(1)
n (p = 1) dyonic IM (1.1) (see Refs. [7,
8]) and correspond to a particular member of a U (1)p -dyonic family of IMs A(1)
n (p),
p = 1, 2, . . . , n 1. Consider, for example, the NA-Toda models defined by the graded
structure similar to (2.7), i.e.,
Q = (n 2p)d +
p
n1
2i H +
i=1
j H
(6.1)
j =2p+1
and
=
n1
(0)
(1)
E
+ E(
.
i
2p +2p+1 ++n1 )
(6.2)
i=2p+1
Since the zero grade subalgebra G0 is given by

G0 = SL(2)p+1 U (1)np1 ,
G00 = U (1)p+1
(6.3)
G00 -gauged
the field content of the corresponding

IM consists in (p + 1)-pairs of charged
fields a , a , a = 1, 2, . . . , p + 1, and n p 1 neutral fields l , l = 1, 2, . . . , n p 1.
The effective action for these models can be derived using the methods described in
Section 2. The potential have n 2p 1 distinct zeros (for imaginary coupling) and the
IM admits U (1)p+1 -charged topological solitons.
(1)
Having constructed the U (1)p dyonic IM An (p), 1 p n 1, with abelian internal
symmetry, the question of constructing IM with nonabelian internal structure arises
naturally. It is clear that one has to choose Q and such that G00 G0 to be nonabelian.
For an appropriate choice of the potential (i.e., of and g0 , since V = Tr(+ g0 g01 )),
the model admits topological solitons carrying nonabelian charges (isospin). The simplest
(1)
case is to consider G00 = U (2) G0 = SL(3) U (1)n2 obtained by decomposing the An
algebra by the grading operator,
Q = (n 1)d +
n
i H,
=
i=3
n
(1)
E(0)
+ E(
,
i
3 +4 ++n )
(6.4)
i=3
(0)
where G00 = {E1 , 1 H, 3 H }. The model represents an SL(3)/SL(2) U (1) analog

(1)
of the complex sine-Gordon (SL(2)/U (1)) interacting with An2 abelian affine Toda
model. Its main ingredient is the new SL(3)/U (2) integrable model. It is defined by the
homogeneous gradation of A(1)
2 ,

(0)
Q = d,
= 2 H (1),
(6.5)
G00 = E
, h1 , h2 .
1
The corresponding effective action can be obtained by considering the A(1)
2 two-loop
gauged WZW model and applying the methods described in Section 2.1. We take Eqs. (2.1)
with A = A1 + A0 , A = A 1 + A 0 where
,
A1 = a1 E(1)
1
(1)
A 1 = a 1 E1 ,
A0 = a01 1 H + a02 (2 1 ) H,
A 0 = a 01 1 H + a 02 (2 1 ) H
(6.6)
477
and g0 = e1 E2 +2 E1 2 e1 E2 +2 E1 2 . We next perform the Gaussian functional

integrals in the definition of the partition function of the model,

SG/H = DBeSeff .
Z SL(3)/U (2) = DBDAD Ae
f
The effective action for the SL(3)/U (2)-model obtained in this way has the form

k
1 1 (1 + 1 1 + 2 2 ) +
2 2 (1 + 1 1 )
Seff =
dz d z 1
2

2 1 + 1 2
1 2 ) V ,
12 (1 2
(6.7)
2
3
2
where V = 2
k ( 3 + 1 1 + 2 2 ) and = (1 + 1 1 ) + 2 2 (1 + 4 1 1 ).
As in the complex sine-Gordon case, the form of the potential V is an indication that
the IM (6.7) can have only nontopological solitons carrying U (2) Noether charges. Hence
this model is not of dyonic type. The A(1)
n (U (2)), n > 3, IMs defined by grading (6.4)
representing the A(1)
abelian
affine
Toda
model interacting with the model (6.7) are
n2
however of dyonic type. Their potential is given by

n2
kij j
(1 +n2 )
1
Vn = 2
e
+e
+ e (1 1 + 2 2 ) n + 1
i=1
and for imaginary coupling i0 shows n 1 distinct zeros. Therefore as in A(1)

n (p)
dyonic models one expects to have topological solitons but now with U (2) internal degrees
of freedom. The explicit construction of such topological solitons with U (2) isospin can
be obtained by applying the methods of Sections 3 and 4.
(1)
The nonabelian dyonic models An (U (2)) described above represents the simplest
0
member of the vast family of dyonic models A(1)
n (As ), 1 s n 1, i.e., with G0 =
U (s + 1), defined by the graded structure
Qs = (n s)d +
n

i=s+2
i H,
=
n
(0)
(1)
Ei + E(s+2 ++n )
(6.8)
i=s+2
and G0 = SL(s + 2) U (1)ns1 , G00 = U (s + 1). The effective actions for these
nonabelian dyonic models can be derived by the methods of Section 2 and appear to be a
straightforward generalization of the simplest A(1)
2 (U (2)) action (6.7) we have constructed
above.
We should note that the discussion of different dyonic IMs introduced in the present
section is far from an effective recipe for their classification. Our purpose is to discuss
the properties of a vast family of dyonic IMs according to their symmetries: abelian
(i.e., U (1)p ), nonabelian, global or local, etc. The selection of the models was done by
first choosing the invariant subalgebra (i.e., symmetries) G00 G0 and next seeking for
appropriate grading operator Q and constant grade 1 elements (such that [G00 , ] =
[Q, G0 ] = 0) leading to consistent integrable models of dyonic type. It is clear that the
systematic classification program consists in (a) listing the admissible gradings for all the
affine algebras, (b) to define the constant elements of grade 1, , (c) to impose the
478
dyonic conditions, i.e., to separate and Q such that G00 is nontrivial and the potential
tr(+ g0 g01 ) has nontrivial distinct zeroes. Its realization is out of the scope of this paper.
Motivated by the problem of description of domain walls in 4-D SU(n + 1) SDYM we
(1)
have introduced and studied in the present paper a pair of An dyonic IMs (one relativistic
and one nonrelativistic) that admit U (1) U (1) charged topological solitons. The 1-soliton
solutions of both models have been explicitly constructed. Since the DWs tensions are
known to be proportional to soliton masses, the derivation of the semiclassical spectrum
(1.3) of these solitons should be considered as the main result of this paper. Although the
soliton mass formulae for U (1)- and U (1) U (1)-integrable models are quite similar,
the semiclassical quantization of the electric charges of the second model represents new
feature. Namely, only their difference Q1 Qn = 02 jel is quantized. Therefore one can
expect that the masses of the multicharged topological solitons of generic U (1)p , p 3,
dyonic IMs are simple generalization of the known particular cases p = 1, 2. The real
problem to be solved, however, is the semiclassical quantization of the corresponding
electric charges Qel
a , a = 1, 2, . . . , p. The methods to be used in the construction of these
multicharged solitons are indeed an appropriate extension of those presented in Sections 3
and 4. Whether and how these methods work in the case of nonabelian SU(2)-dyonic IM
(6.7) and what is the spectrum of the solitons carrying SU(2) isospin are interesting open
questions, which require further investigation.
Together with the dyonic IMs with global U (1)p symmetry we have derived in Section 2
integrable models with local symmetries: the so-called ungauged IM (2.12) with local
U (1) U (1) and the intermediate model (2.15) with one local and one global U (1)
symmetries. It is therefore interesting to know how their solitons or more general string-like
finite energy solutions looks like. By construction both models have the 1-soliton solutions
of the completely gauged IM (1.2) as particular solutions. Our preliminary analysis shows
that their soliton spectra is more complicated and also includes massless solitons and
solitonic strings. The complete discussion will be presented elsewhere.
The -model form of the Lagrangians (1.1), (1.2), (2.12) and (2.15) constructed in the
present paper suggests that they might have interesting string application. Although they
are not conformal invariant, it is natural to interpret their conformal affine versions [9,15]
(called CAT) as representing strings on curved background gMN (X), bMN (X) and V :

L = gMN (X) + bMN (X) XM XN V XL ,
(7.1)
where XL = (a , a , Ra , l , , ), L = 1, 2, . . . , D = n + 6 are the string coordinates

and gMN (X) is the target space metric, bMN (X)the antisymmetric tensor and V (X)
the tachyonic potential. For each nonconformal IM one introduces its CAT counterpart
by adding a pair of fields (, ), that restores conformal invariance, i.e., by extending
the (n + 4)-dimensional off-critical string to (n + 6)-critical one, for say n = 20. In the
479
particular case of ungauged model (2.12) we have

1
j + 1 1 e(R1 1 )
kij i
2
n2
LCAT
G0 =
i=1
+ n n e(Rn n2 ) +

1
1 + nRn R
n + R1 R
n + Rn R
1 V CAT , (7.2)
nR1 R
+
0
2(n + 1)
where
n2
2

ekij j + e(1 +n2 ) 1 + 2 1 1 e(R1 1 )

V0CAT = 2
i=1

n1
2 (Rn n2 )
.
1 + n n e
(n 1)e
Within this context the ungauged IM (2.12) (i.e., (7.2) with = 0 = const) describes the
off-critical (i.e., relevant deformation [4]) and renormalization group flow properties of
strings on AdS3 S3 Tn2 -target space, when Rn and l are taken imaginary and n , n
as complex conjugated to each other. The target space metric gMN (X) for gauge fixed IMs
(1.2) and (2.15) represents both horizons and singularity and therefore could be used in
the study of the off-critical behaviour of 4-D black-holes (and black-strings) with certain
additional flat directions Xl = l .
The exact quantum spectrum of 2-D solitons is known to be crucial in the description
of the strong coupling behaviour (and S-duality properties) of the corresponding quantum
integrable models [32]. The problem to be solved before any attempt for exact quantization
of the considered dyonic IMs is to answer the question: whether and for which 2-D
IMs the semiclassical spectrum survives the renormalization. The answer is known for
sine-Gordon (sG) and the abelian affine Toda models, where the only renormalization is
of the coupling constant and indeed the semiclassical spectrum is exact. The quantum
consistency of the LundRegge model [35], U (1)-dyonic IM (1.1) (see Section 7.2 of
Ref. [7]) and T-selfdual Fateevs IMs [32] requires that new counterterms to be added
to the original classical Lagrangians. Since the same counterterms are present in the
corresponding conformal quantum theories, i.e., the gauged G0 /G00 WZW models (2.1)
[33] one believes that they cannot change the soliton masses. But they do change the form
of the U (1) currents and hence they might shift the semiclassical values of the electric
charges. Recently the thermodynamic Bethe ansatz analysis [34] of the homogeneous
sine-Gordon models has confirmed the exactness of their semiclassical mass spectra. The
answer of the question whether such statement takes also place for multicharged dyonic
IMs requires the construction of their exact S-matrices, etc. It exists however a strong hint
that the n limit of the semiclassical (Q1 , Qn ) 1-soliton masses

el
Mjel ,j (n ) = 2 02 Qmag Qel
1 + Qn
0
is exact. The reason is that in the large n limit the counterterms become proportional to the
complicated a , a kinetic term and thus contribute to the 02 renormalization only.
480
Acknowledgements
We are grateful to L.A. Ferreira and H. Aratyn for discussions and Fapesp, Unesp and
CNPq for financial support.
Appendix A. Two-soliton solution

The generic 4-vertex solution of Eqs. (5.15) can be extracted from Eqs. (5.12), (5.14)
for n = 3 and the constant G0 group element

g (1) = eb2 F11 (2 ) eb1 F11 (1 ) eb4 F11 (4 ) eb3 F11 (3 ) ,
(A.1)
11 ( ), F
11 ( ) are given by Eqs. (4.13) (for n = 3). By calculating the matrix
where F
elements (5.12) we obtain the following explicit form of the G0 - -functions8 (and
therefore of ra , qa and u according to Eqs. (5.14)):
0 = 1 + b1 b2 c12 1 2 + b3 b4 c34 3 4 + c12341 2 3 4 b1 b2 b3 b4 ,
2 = 1 + b1 b2 d12 1 2 + b3 b4 d34 3 4 + d12341 2 3 4 b1 b2 b3 b4 ,
n1,n = (g3 + b1 b2 g123 1 2 )3 b3 ,
2n = (i4 + b1 b2 i124 1 2 )4 b4 ,
12 = (e2 + b3 b4 e2343 4 )2 b2 ,
1,n1 = (f1 + b3 b4 f134 3 4 )1 b1 ,
(A.2)
where
1 (1 ) = ex1 +t 1 ,
2 (2 ) = ex2t 2 ,
3 (3 ) = ex3 +t 3 ,
4 (4 ) = ex4t 4
2
2
2
2
(A.3)
and
c12 =
21 22
,
(1 2 )2 (1 + 2 )
c34 =
232 4
,
(3 4 )2 (3 + 4 )
d12 =
212 2
,
(1 2 )2 (1 + 2 )
d34 =
23 42
,
(3 4 )2 (3 + 4 )
c1234 =
41 22 32 4 (1 3 )(2 4 )
,
(1 + 2 )(1 2 )2 (2 3 )(1 4 )(3 + 4 )(3 4 )2
(A.4)
d1234 =
412 2 3 42 (1 3 )(2 4 )
,
(1 + 2 )(1 2 )2 (2 3 )(1 4 )(3 + 4 )(3 4 )2
(A.5)
8 We have determined such solution using the Mathematica program of Ref. [31].
e2 = 2 2 ,
f1 =
g3 =
i4 =
2 1 ,
2 3 ,
2 4 ,
2 2 2 3 42 (2 4 )
e234 =
,
(2 3 )(3 + 4 )(3 4 )2
2 2 1 32 4 (1 3 )
f134 =
,
(1 4 )(3 + 4 )(3 4 )2
2 2 12 2 3 (1 3 )
g123 =
,
(1 + 2 )(2 3 )(1 2 )2
2 21 22 4 (2 4 )
i124 =
.
(1 + 2 )(1 4 )(1 2 )2
481
(A.6)
It is worthwhile to note again that the 4-vertex solution for arbitrary q > 2 integer
(1)
A3 (p = 2, q) model have the form identical to the q = 2 one (A.2), (A.6) with i (i )
replaced by, say
1 ( ) = ex +tq , etc.
(q)
(A.7)
For the specific choice of i , namely

1 = eB1 i1 = 2 ,
3 = eB2 i2 = 4
the above four-vertex solution has real energy and momenta and thus represents charged
two-soliton solution, where Bi (vi = th(Bi )) are the two independent rapidities (velocities)
of the 2-soliton and i are related to the Q1 , Qn charges. As it was demonstrated in
Section 5.2, the limit when B2 = B1 , 2 = 1 leads to the charged 1-soliton solution (5.16).
References
[1] A. Kovner, M. Shiffman, A. Smilga, Phys. Rev. D 56 (1997) 79787989;
Also G. Dvali, M. Shiffman, Phys. Lett. B 396 (1997), hep-th/9612128;
A. Hanany, K. Hori, Nucl. Phys. B 513 (1998) 119, hep-th/9706089.
[2] E. Witten, Nucl. Phys. B 507 (1997) 658.
[3] E. Witten, Nucl. Phys. B 160 (1979) 461.
[4] J. Polchinski, M. Strassler, The string dual of a confining four-dimensional gauge theory, hep-th/0003136;
O. Ahrony, S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theory, string theory and gravity, Phys.
Rev. 323 (2000) 183, hep-th/9905111.
[5] J. Gauntlett, D. Tong, P. Townsend, Supersymmetric intersecting domain walls in massive hyperkahler sigma
models, Phys Rev. D 63 (2001) 085001, hep-th/0007124;
J. Gauntlett, R. Portugues, D. Tong, P. Townsend, D-brane solitons in supersymmetric models, Phys. Rev.
D 63 (2001) 085002, hep-th/0008221.
[6] A.N. Leznov, M.V. Saveliev, Group Theoretical Methods for Integration of Nonlinear Dynamical Systems,
Progress in Physics, Vol. 15, Birkhuser, Berlin, 1992;
L.A. Ferreira, J.L. Miramontes, J.S. Guillen, Nucl. Phys. B 449 (1995) 631.
[7] J.F. Gomes, E.P. Gueuvoghlanian, G.M. Sotkov, A.H. Zimerman, Electrically charged topological solitons,
Nucl. Phys. B 606 (2001) 441, hep-th/0007169.
[8] J.F. Gomes, E.P. Gueuvoghlanian, G.M. Sotkov, A.H. Zimerman, Dyonic integrable models, Nucl. Phys.
B 598 (2001) 615, hep-th/0011187.
[9] O. Babelon, L. Bonora, Phys. Lett. B 244 (1990) 220.
[10] H.C. Liao, D. Olive, N. Turok, Phys. Lett. B 298 (1993) 95;
D. Olive, N. Turok, J. Underwood, Nucl. Phys. B 409 (1993) 509;
482
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
D. Olive, N. Turok, J. Underwood, Nucl. Phys. B 401 (1993) 663;

T.J. Hollowood, Nucl. Phys. B 384 (1992) 523.
J.F. Gomes, E.P. Gueuvoghlanian, G.M. Sotkov, A.H. Zimerman, Self-dual domain walls, in preparation.
F. Lund, T. Regge, Phys. Rev. D 14 (1976) 1524;
F. Lund, Phys. Rev. Lett. 38 (1977) 1175;
B.S. Getmanov, JETP Lett. 25 (1977) 119;
N. Dorey, T.J. Hollowood, Nucl. Phys. B 440 (1995) 215.
A. Chodos, Phys. Rev. D 21 (1980) 2818.
J.L. Miramontes, Nucl. Phys. B 547 (1999) 623.
H. Aratyn, L.A. Ferreira, J.F. Gomes, A.H. Zimerman, Phys. Lett. B 254 (1991) 372;
L.A. Ferreira, J.F. Gomes, A. Schwimmer, A.H. Zimerman, Phys. Lett. B 274 (1992) 65.
H. Aratyn, L.A. Ferreira, J.F. Gomes, A.H. Zimerman, J. Phys. A 31 (1998) 9483;
H. Aratyn, L.A. Ferreira, J.F. Gomes, A.H. Zimerman, in: H. Aratyn et al. (Eds.), Supersymmetry and
Integrable Models, Lecture Notes in Physics, Vol. 502, Springer, Berlin, 1998, pp. 197210.
J.F. Gomes, E.P. Gueuvoghlanian, G.M. Sotkov, A.H. Zimerman, T-duality of axial and vector dyonic
integrable models, Ann. Phys. 289 (2001) 232, hep-th/0007116.
H. Aratyn, L.A. Ferreira, J.F. Gomes, A.H. Zimerman, J. Phys. A 33 (2000) L331.
D. Fioravanti, M. Stanishkov, Nucl. Phys. B 591 (2000) 685, hep-th/0005158.
I. Cabrera-Carnero, J.F. Gomes, E.P. Gueuvoghlanian, G.M. Sotkov, A.H. Zimerman, Non-Abelian Toda
Models and Constrained KP Hierarchies, to appear in the VII International Wigner Symposium, Maryland,
2001, hep-th/0109117.
E. Nissimov, S. Pacheva, Symmetries of supersymmetric integrable hierarchies of KP type, nlin.SI/0102010.
J. Balog, L. Feher, L. ORaifeartaigh, P. Forgacs, A. Wipf, Ann. Phys. 203 (1990) 76;
L. Feher , L. ORaifeartaigh, P. Ruelle, I. Tsutsui, A. Wipf, Phys. Rep. 222 (1992) 1.
J.F. Gomes, G.M. Sotkov, A.H. Zimerman, Parafermionic reductions of WZNW model, Ann. Phys. 274
(1999) 289362, hep-th/803234.
C.R. Fernandez-Pousa, M.V. Gallas, T.J. Hollowood, J.L. Miramontes, Nucl. Phys. B 484 (1997) 609;
C.R. Fernandez-Pousa, M.V. Gallas, T.J. Hollowood, J.L. Miramontes, Nucl. Phys. B 499 (1997) 673;
C.R. Fernandez-Pousa, J.L. Miramontes, Nucl. Phys. B 518 (1998) 745.
J.F. Gomes, E.P. Gueuvoghlanian, G.M. Sotkov, A.H. Zimerman, Classical integrability of non-abelian
affine Toda models, hep-th/0010257, to appear in the Proc. of the XXIII International Colloquium on Group
Theoretical Methods in Physics, Ed. G. Pogosyan et al., Dubna, 2000.
E. Witten, Phys. Lett. B 86 (1979) 283.
J. Polchinski, String Theory, Vol. I, Cambridge Univ. Press, 1998.
O. Babelon, D. Bernard, Int. J. Mod. Phys. A 8 (1993) 507.
S. Coleman, Phys. Rev. D 11 (1975) 2088.
H. Aratyn, C.P. Constantinides, L.A. Ferreira, J.F. Gomes, A.H. Zimerman, Nucl. Phys. B 406 (1993) 727.
A.G. Bueno, L.A. Ferreira, A.V. Razumov, Confinement and soliton solutions in the SL(3) Toda model
coupled to matter fields, hep-th/0105078.
V.A. Fateev, Nucl. Phys. B 479 (1996) 594.
A. Tseytlin, Nucl. Phys. B 399 (1993) 601;
A. Tseytlin, Nucl. Phys. B 411 (1994) 509.
O.A. Castro-Alvaredo, A. Fring, C. Korff, J.L. Miramontes, Nucl. Phys. B 575 (2000) 535;
O.A. Castro-Alvaredo, A. Fring, Nucl. Phys. B 604 (2001) 367.
H.J. de Vega, J.M. Maillet, Phys. Rev. D 28 (1983) 1441.

Large-N limit of the exactly solvable BCS model:

analytics versus numerics
J.M. Romn a , G. Sierra a , J. Dukelsky b
a Instituto de Fsica Terica, CSIC/UAM, Madrid, Spain
b Instituto de Estructura de la Materia, CSIC, Madrid, Spain
Abstract
We have studied the numerical solutions of Richardson equations of the BCS model in the limit of
large number of energy levels at half-filling, and compare them with the analytic results derived by
Gaudin and Richardson, which in turn leads to the standard BCS solution. We focus on the location
and density of the roots, the eigenvalues of the conserved quantities, and the scaling properties of the
total energy for the equally spaced and the two-level models. 2002 Elsevier Science B.V. All rights
reserved.
PACS: 74.20.Fg; 75.10.Jm; 71.10.Li; 73.21.La
1. Introduction
One of the most studied models in Condensed Matter Physics is the BCS model of
superconductivity, proposed by Bardeen, Cooper, and Schrieffer in 1957 [1]. The BCS
model is usually formulated in the grand canonical ensemble, which is appropriated to
systems with a macroscopic number of fermions. However, for small systems, such as
nuclei or ultrasmall metallic grains, one has to consider the canonical ensemble where the
BCS wave function is not adequate.
Fortunately enough, there is a simplified version of the BCS model which is exactly
solvable in the canonical ensemble. This occurs when the strength of the pairing interaction
between all the energy levels is the same. The exact solution of the reduced BCS model was
obtained by Richardson in 1963 [24], who also studied its consequences and properties
in a series of papers in the 60s and 70s [58]. The exactly solvable BCS model, in turn, is
E-mail address: sierra@sisifo.imaff.csic.es (G. Sierra).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 1 7 - 6
484
J.M. Romn et al. / Nuclear Physics B 634 [FS] (2002) 483510
closely related to the rational spin model of Gaudin, defined in terms of a set of commuting
Hamiltonians [9,10].
The integrability of the reduced BCS Hamiltonian was proved in 1997 by Cambiaggio,
Rivas, and Saraceno (CRS) [11], who constructed a set of conserved quantities in
involution, commuting with the BCS Hamiltonian and closely related to the rational
Gaudins Hamiltonians [9]. More recently Gaudins trigonometric model have been
generalized, to include a g-term la BCS, in Refs. [12,13]. There are also bosonic
pairing Hamiltonians satisfying the same type of equations as in the fermionic case [7,
14,15]. Generalizations of the SU(2) BCS model to other Lie groups [16] and supergroups
[17] have been worked out. Both, the BCS and Gaudins models are intimately linked
to conformal field theory [1820], ChernSimons theory [16], and Integrable Models in
Statistical Mechanics [2123] (see Ref. [24] for a short review on these topics). The exact
solution has also been used to study the effect of level statistics [25] and finite temperature
[26,27] on the ultrasmall metallic grains.
An important property of the exact solution is that the energy of the states and the
occupation numbers agree, to leading order in the number of particles, with the BCS theory
[1]. This result was obtained by Gaudin [10] and Richardson [8], using an electrostatic
analogy of the Richardsons equations, which have to be solved for finding the eigenstates
of the BCS Hamiltonian. More recent applications of this electrostatic model to nuclear
pairing can be found in Ref. [28].
In the limit of large number of levels N , Richardsons equations become integral
equations, which can be solved with techniques of complex analysis [10]. This is sufficient
to determine the location and density of roots, as well as the total energy of the ground
state to leading order in N . To study higher order corrections of the ground state energy
and the excitations, Richardson developed a multipole expansion of the fields appearing
in the electrostatic analogue model [8]. We shall follow Gaudins approach which is more
geometrical, but use also some of the results obtained by Richardson.
The aim of this paper is to make a revision and precise comparison between the
numerical solution and the analytic expressions for the location and density of roots of
the Richardsons equations, as proposed originally by Gaudin in his unpublished paper in
1968, which got finally published in his collected papers in 1995 [10]. The comparison
will be made in the limit of large-N for the two-level model and the equally spaced model.
These models have been extensively used in nuclear physics since they were introduced
by HogaasenFeldman [29] (two-level) and Richardson himself [6] (equally spaced). The
latter model has also been used to describe the physics of ultrasmall metallic grains
[3034] (for a review see Ref. [35]). In the large-N limit the equally spaced model is
superconducting for all values of attractive BCS couplings g, while the two-level model
displays a quantum phase transition between a superconducting state and a normal state as
a function of the BCS coupling constant.
The organization of the paper is as follows: in Sections 2 and 3 we give a primer of the
Richardsons exact solution of the BCS model, together with its integrability. In Section 4
we introduce the electrostatic analogue model of BCS. In Section 5 we review Gaudins
continuum limit of the Richardsons equations and their general solution, using complex
analysis. In Section 6 we derive the gap and chemical potential BCS equations under the
assumption that the roots of Richardson equations collapse into a single arc, and study
485
in detail the two-level and equally spaced models. In Section 7 we present our numerical
results and compare them with the analytic ones obtained in Section 6. We also study the
scaling properties of the roots and the ground state energy. In Section 8 we present our
conclusions and prospects. Finally, we describe in Appendix A the conformal mappings
associated to the equally spaced model.
2. Richardsons exact solution of the reduced BCS model

The reduced BCS model is defined by the Hamiltonian [1,30,35]
HBCS =

1
j cj cj G
cj + cj cj cj + ,
2

j, =
(1)
j,j
where cj, (respectively cj,

) is an electron annihilation (respectively creation) operator
in the time-reversed states |j, with energies j /2, and G is the BCS dimensionful
coupling constant. The sums in (1) run over a set of N doubly degenerate energy levels j /2
(j = 1, . . . , N). We adopt Gaudins notation, according to which j denotes the energy of
a pair occupying the level j [10]. We shall assume that the energy levels are all distinct,
i.e., i = j for i = j . The Hamiltonian (1) is a simplified version of the reduced BCS
Hamiltonian where all couplings have been set equal to a single one, namely G. Hereafter
we shall refer to (1) simply as the BCS Hamiltonian.
Richardson had long ago solved this model exactly for an arbitrary set of levels, j
[24]. To simplify matters, we shall assume that there are not singly occupied electronic
levels. As can be seen from (1), these levels decouple from the rest of the system; they
are said to be blocked, contributing only with their energy j /2 to the total energy E.
The above simplification implies that every energy level j is either empty (i.e., |vac), or
occupied by a pair of electrons (i.e., cj,+

cj,
|vac). Denote the total number of electrons
pairs by M. Then of course M N . The most studied case in the literature corresponds
to the half-filled situation, where the number of electrons, Ne = 2M, equals the number of
levels N [35]. In the absence of interaction (i.e., G = 0), all the pairs occupy the lowest
energy levels, forming a Fermi sea. The pairing interaction promotes the pairs to higher
energies and even for small values of M or G the levels are pair correlated [3133,35].
In order to describe Richardsons solution one defines the hard-core boson operators
bj = cj, cj,+ ,
bj = cj,+
cj,
,
Nj = bj bj ,
(2)
which satisfy the commutation relations,

bj , bj = j,j (1 2Nj ).
The Hamiltonian (1) can then be written as

j bj bj G
bj bj .
HBCS =
j
j,j
(3)
(4)
486
Richardson showed that the eigenstates of this Hamiltonian with M pairs have the
(unnormalized) product form [24]
|M =
M
B |vac,
B =
N

j =1
=1
1
b ,
j E j
(5)
where the parameters E ( = 1, . . . , M) are, in general, complex solutions of the M

coupled algebraic equations
1
1
=
G
j E
N
M
j =1
=1(=)
2
,
E E
= 1, . . . , M,
(6)
which play the role of Bethe ansatz equations for this problem [1824]. The energy of
these states is given by the sum of the auxiliary parameters E , i.e.,
E(M) =
M
E .
(7)
=1
The ground state of HBCS is given by the solution of Eqs. (6) which gives the lowest value
of E(M). The (normalized) states (5) can also be written as [5]

C
(j1 , . . . , jM )bj1 bjM |vac,
|M =
(8)
M! j ,...,j
1
where the sum excludes double occupancy of pair states and the wave function takes the
form
M
1
.
(9)
jk EP k
P k=1
The sum in (9) runs over all the permutations, P, of 1, . . . , M. The constant C in (8)
guarantees the normalization of the state [5] (i.e., M|M = 1).
(j1 , . . . , jM ) =
3. Integrability of the reduced BCS Hamiltonian

A well-known fact about the BCS Hamiltonian is that it is equivalent to that of a XY spin
model with long range couplings, and a position dependent magnetic field proportional
to j . To see this, let us represent the hard-core boson operators (2) in terms of the Pauli
matrices as follows,

1
bj = j ,
Nj = 1 jz ,
bj = j+ ,
(10)
2
in which case the Hamiltonian (4) becomes
1
j + G(N/2 M),
HBCS = HXY +
2
HXY =

j
j tj0

G +
T T + T T + ,
2
(11)
487
where the matrices

Ta =
N
tja
(a = 0, +, ),
j =1
satisfy the SU(2) algebra,

a b
T , T = fcab T c ,
with
1
tj0 = jz ,
2
with f++0 = f0 = 1,
tj+ = j+ ,
f0+ = 2,
tj = j ,
(12)
(13)
whose Casimir is given by

1 +
T T + T T + .
(14)
2
This spin representation of the BCS model is the appropriate one to study its
integrability, as it was shown by Cambiaggio, Rivas, and Saraceno (CRS), who constructed
a set of operators, Rj (j = 1, . . . , N) [11],
T T = T 0T 0 +
Ri = ti0 2g
N

ti tj
i j
(i = 1, . . . , ),
(15)
[HBCS , Ri ] = [Ri , Rj ] = 0 (i, j = 1, . . . , N).
(16)
j (=i)
satisfying the properties
Let us denote by j the eigenvalue of Rj acting on the state (8), i.e.,

Rj |MR = j |MR .
(17)
CRS, unaware of Richardsons solution, did not give an expression of j in their work.
This was found in reference [20] using CFT techniques,
M

N

1
1
1
1
,
i = + G
(18)
2
i E
2
i j
=1
j =1(=i)

which verifies i i = (M N/2) upon using Eqs. (6), so that it vanishes at half filling.
However CRS showed that HXY given in Eq. (11) can be expressed in terms of the
operators Ri as
2

3
HXY =
(19)
j Rj + G
Rj GN.
4
j
Then using the Eqs. (11), (18), and (6) one can show that E(M) is indeed given by Eq. (7).
(For the two-level system the last term in (19) must be replaced by 2G(N/4 + 1)N/4 ).
4. The GaudinRichardsons electrostatic model of BCS

Eqs. (6) admit a 2-dimensional electrostatic analogy, which will be very useful in the
study of the limit N [8,10]. Let us consider a set of N charges 1/2 fixed at the
488
positions j on the real axis, and a uniform field parallel to this axis with strength 1/2G.
The problem is to find the equilibrium positions of M charges +1 at positions E subject to
their mutual repulsion, the attraction with the 1/2 charges, and the action of the uniform
field. The 2D electrostatic potential is given by W + W , where W reads,

1
W {E }, {i } =
log(i E )
log(E E )
2
>
i,

1
1
1
log(i j ) +
E
i .
4
2G
2
i>j
(20)
It is easy to see that the Richardsons equations (6) arise as the stationary conditions
W/E = 0, while the conserved quantities (18) are proportional to the forces exerted
on the fixed charges, i.e.,
i = 2G
W
.
i
(21)
The total energy (7) is the center of gravity of the charges E , which must be located
symmetrically with respect to the real axis, i.e., if E is a solution to Eqs. (6), then E
must also be a solution. This condition is fulfilled in two cases: (i) either E is real, or
(ii) E and E form a complex conjugate pair. The formation of these complex pairs is
usually an indication of the superconducting properties of the ground state.
5. Continuum limit of Richardsons equations

In Ref. [10] Gaudin proposed a continuum limit of Eqs. (6), in order to obtain the BCS
formulas for the ground state energy and occupation number of levels. We shall follow
closely this reference. This continuum limit is defined by taking the number of levels N
going to infinity, while keeping fixed the following quantities:
g = GN,
M/N.
(22)
At the same time the pair energy levels i will be equivalent to a negative charge density
() located on an interval of the real axis. The total charge of this interval is given
by

N
() d = .
(23)
2
We shall suppose that is contained inside an interval (, ). For example, in the BCS
model with equally spaced levels, /2 is equal to the Debye energy.
The basic assumption made by Gaudin, which is supported by numerical results, is
that the solutions of Eqs. (6) organize themselves into arcs k (k = 1, . . . , K), which are
piece-wise differentiable and symmetric under reflection on the real axis. We shall call
the union of all these arcs, and r( ) the linear charge density of roots E in the complex
plane. Hence, the total number of pairs, M, and the total energy, E, are given by

r( ) |d | = M,
489
(24)

r( ) |d | = E.
(25)
The continuum limit of Eqs. (6) is

1
() d
r( ) |d |
P
= 0,

2G
(26)
which implies that the total electric field on every point of the arcs k is null. On the other
hand, the values of the conserved quantities (18) are given in the continuum by

1
r( ) |d |
( ) d
() = + G
(27)
P
, .
2
The formal solution of Eqs. (26) can be found as follows. First of all, let us orient each
arc k from the point ak to the point bk , and call Lk an anticlockwise path encircling k .
We look for an analytic field h( ) outside and the set , such that

1
h+ ( ) h ( ) d, ,
r( ) |d | =
(28)
2i
where h+ ( ) and h ( ) denote the limit values of h( ) to the right and left of . This
can be understood using the electrostatic equivalence, considering that the electric field
presents a discontinuity proportional to the superficial density of charge when crossing
such surface. Next, we define a function R( ), with cuts along the curves k , by the
equation
1/2
K

( ak )( bk )
,
R( ) =
(29)
k=1
and look for a solution which vanishes at the boundary points of , in the form

() d
.
h( ) = R( )
(30)
This field has to be constant at infinity, hence the first K 2 moments of must vanish,
i.e.,

3 () = 0, 0 3 < K 1.
(31)
This formula is empty for K = 1. The contour integral surrounding the charge density r( )
in (26) can be expressed as

1
1
r( ) |d |
h+ ( ) h ( )
h( )
=
d
=
d , C, (32)

2i

2i

490
where C is the region outside the curves in . Using Eqs. (29) and (30), one finds for
the principal value of (32)

r( ) |d |
d R( )
() d
1
=
P

2i

L

()R()
d + K1 () d, ,
=
(33)
where we have deformed the contour of integration L into two contours, one encircling the
interval (first term) and another one around the infinity (second term). We are assuming
that , and consequently L, do not cut the interval . The case where L intersects
actually arises in the equally spaced model, and will be discussed later on.
Plugging Eq. (33) into (26), we see that a solution is obtained provided the following
equations are satisfied:
()
,
() =
R()

1
()
d =
,
K1
R()
2G
(34)
(35)
which have to be supplemented by Eq. (31), which can be rewritten as

()
d = 0, 0 3 < K 1.
3
R()
(36)
The field h( ) gives also the density of charges r( ),

1
r( ) = h( ), ,
and its value is given by replacing (34) into (30), i.e.,

() d
h( ) = R( )
,
R()
(37)
(38)
whose value at infinity is 1/2G. Moreover, using Eq. (32) and performing the same
contour deformations that lead to Eq. (33), one can show that the conserved quantities (27)
are given by

G
( ) d
() = P GR()
(39)
=
h( + i0) + h( i0) , ,
R( )
2
so that () is the principal part of h() on the set . Finally the equations fixing the arcs
k are the equipotential curves of the total distribution, i.e.,

R
ak
h( ) d = 0,
k .
(40)
491
6. The BCS equations

In this section we shall show how the BCS equations describing the ground state of the
model can be derived from the formalism of Section 5 [10].
The basic assumption is that for the ground state all the roots E form a single arc, i.e.,
K = 1, leaving only two complex parameters a a1 and b b1 , which shall be denoted
as
a = 0 i,
b = 0 + i,
(41)
where 0 and > 0 are real parameters.

From Eq. (38) the electrostatic field is given by

()
d
2
2 1/2
.
h( ) = ( 0 ) +
[( 0 )2 + 2 ]1/2
(42)
And the conserved quantities () can be derived from Eq. (39). Eq. (35) yields the BCS
gap equation

() d
1
(43)
.

=
2
2
2G
( 0 ) +
Similarly, Eq. (24) becomes the chemical potential equation

1
0
M=
h( ) d =
1
() d,
2i
( 0 )2 + 2
L
(44)
while Eq. (25) gives the BCS expression for the ground state energy,

2
1
0
+
h( ) d =
E=
1
() d.
2i
4G
( 0 )2 + 2
L
(45)
Gaudins paper [10] contains a misprint in this equation since the term 4G
is quoted as
2
2G
.
Comparing these equations with the corresponding ones in the BCS theory, we deduce
the following relations between , 0 , and (), and BCS (BCS gap), (chemical
potential), and n(7) (single-particle energy density):

1
= 2BCS ,
(46)
.
0 = 2,
() = n
4 2
The factor 1/4, in the last term of (46), emerges from the normalization 1/2 of the charge
density (), times another factor 1/2, due to the fact that the separation between energy
pairs is twice the separation between single-particle energy levels.
To determine the equation of the curve we define the complex variable

z = ( 0 )2 + 2 = x + iy.
(47)
492
Using Eqs. (40) and (42) we deduce for ,

2

x ( 0 )2 + 2 + y 2
x
1
+ log
.
0 = d ()

2
( 0 )2 + 2 4
x + ( 0 )2 + 2 + y 2
(48)
Eqs. (43)(45) have been derived under the condition that does not cut the set , which
requires that (48) has no solutions with . In the equally spaced model the curve
may intersect , so this case has to be treated with some care (see below).
Finally, the mean occupation of the energy level j can be computed from the formula
Nj = E/j . In the large-N limit Nj can be shown to be given by [8,10]

1
0
N() =
(49)
1
, ,
2
( 0 )2 + 2
which agrees with the BCS result [1]. Next we shall apply the previous formalism to two
models.
6.1. The two-level model
Let us suppose that we have two energy levels 1 , which can accommodate at most
N/2 pairs each. The charge density is given by

N
(50)
( + 1 ) + ( 1 ) .
4
From Eq. (44) we deduce that the chemical potential vanishes, i.e., 0 = 0, while the gap
equation (43) gives
() =
12 + 2 = g 2 ,
g = GN.
(51)
This equation implies that g must be greater than 1 . The total energy of this solution
follows from (45),

N 2
1 + g 2 .
E=
(52)
4g
Integrating Eq. (42) we obtain the electric field, and accordingly the value of the conserved
quantities

N 2 + 2
1 2
h( ) =
(53)
21 + g 2 .
(1 ) =
2
2
2g 1
8g1

Similarly one finds (1 ) = (1 ), in agreement with the condition i i = 0 derived
in Section 3. The equation of the curve (48) becomes in this case
4x
(x g)2 + y 2
+ log
= 0,
g
(x + g)2 + y 2
(54)
or equivalently,
x 2 + y 2 + g2 =
2xg
,
th(2x/g)
(55)
493
Fig. 1. Plot in the -plane of the solutions of Eq. (76) for M = 100 and three values of g. The continuous lines
are the theoretical curves (55) (with the change of variables (47)) for g = 1.5, 1; and (63) for g = 0.5. All the
energies are in units of 1 .
where x and y are given by Eq. (47). In Gaudins paper [10] there is a misprint in Eqs. (52)
and (55).
From the numerical results shown in Fig. 1, when g > 1 , is an open arc between the
imaginary points i in the complex -plane. However when g = 1 , in the limit when
N , the arc closes at the origin, and for g < 1 it remains closed surrounding the point
1 .
To treat this case Gaudin considers a general situation where is a closed curve
surrounding a piece of and defines a function s( ) by the equation [10]
s( ) d = r( ) |d |.
(56)
Then Eq. (26) is solved by

i 1
() d
() d
+
,
s( ) =
2G
(57)

which can be verified using P d /( ) = i , when . For the two-level model
s( ) is given by

N 1
i 1
, .
s( ) =
(58)
2G
2 12 2
and Eqs. (24) and (25) yield

N
M = d s( ) = d 2() = ,
2
(59)
494

E=

d s( ) =
d 2() =
N
1 ,
2
(60)
Hence the state for g < 1 is a Fermi sea to leading order in N .

The conserved quantities are derived from an equation similar to (39), where h( ) is
replaced by s( ), i.e.,

G
s( + i0) + s( i0) ,
2

1
g
(1 ) =
,
1
2
41
() = i
,
(61)
verifying, as in the previous case, the symmetry (1 ) = (1 ).

To find the curve we integrate (58), and impose the result to be real accordingly to
(56). Imposing the imaginary part of Eq. (56) to vanish we obtain a differential equation
for the curve , which is a check of consistency.

1
x0 1
+ log
R
(62)
= 0, ,
g
2
+ 1
where x0 is an integration constant, which has a nonvanishing value, contrary to Gaudins
assumption. Eq. (62) can be written as
x 2 + y 2 + 12 =
2x1
,
th(2(x x0 )/g)
= x + iy.
(63)
To find x0 we have first to look for the point 0 where intersects the real axis. From
Fig. 1, the density of roots at this point vanishes, i.e., s(0 ) = 0, which implies

0 = 1 1 g/1 .
(64)
The value of x0 is found by imposing that 0 belongs to the curve (63), namely
x0 = 0
1 + 0
g
log
.
2
1 0
(65)
As g approaches 1 from below, both 0 and x0 go to zero, and the curve (63) coincides
with the curve (55) at the critical point g = 1 .
6.2. Equally spaced model
Let us consider now a model with N pair energy levels that are uniformly distributed in
the interval [, ], with a density () = 0
N
.
(66)
4
The distance between the single-particle energy levels, d, is given by d = /N = 2D /N ,
where D is the Debye energy. This is the model used in most of the studies on ultrasmall
superconducting grains [35]. At half-filling the number of pairs is M = N/2, which implies
that 0 = 0 (see Eq. (44)). On the other hand, the gap equation (43) yields the standard
0 =
495
result [1]
=
,
sinh(1/g)
g=
GN
,
(67)
where g is the usual dimensionless BCS coupling constant. Recall that is twice the BCS
gap BCS . The total energy of the BCS ground state can be derived from (45),

2
2
E=
(68)
1+
,
4d
while the energy of the Fermi sea state is given by

2 GN
.
(69)
4d
2
The condensation energy is defined as EC = E EFS , which in the weak coupling limit,
where , behaves as
EFS =
2BCS
,
(70)
2d
the well-known result.
Next, we shall study the shape of the arcs in this model. The electrostatic field h( ) is
given by

( )(2 + (2 + 2 )(2 + 2 )

h( ) = 0 log
(71)
.
( + )(2 + + (2 + 2 )(2 + 2 )
EC
To simplify matters, Gaudin considered the limit , where Eq. (71) becomes

2 + 2

h( ) = 0 log
,
| | .
+ 2 + 2
The conserved quantities () are given by

|( )(2 + (2 + 2 )(2 + 2 )|
g

() = log
4
|( + )(2 + + (2 + 2 )(2 + 2 )|
| 2 + 2 |
g
||
log
.
4
| + 2 + 2 |
(72)
(73)
In the low energy approximation the equation of the equipotential curves passing
through the points 0 i is given by

2 + 2

R log
(74)
+ 2 2 + 2 = 0.
+ 2 + 2
This curve cuts the real axis inside the interval [, ], which contradicts the assumptions
made so far. From numerical studies we can see that for weak couplings only a fraction
of energies E form complex conjugated pairs, which in the limit N organize
themselves into arcs, while the other energies remain real and located between the lower
496
energy levels (see the appendix for more details on this issue). Taking this into account and
using contour arguments, it can be shown that the BCS equations (43)(45) also hold in
cases where cuts the energy interval .
In the general case the curve is given by the equation

1 + 2 /2
2 + 2
R i arcsin
(75)
+ Argsinh
= 0.
1 2 /2
2 2
In the limit Eq. (75) turns into (74). We show in the appendix that for > the
arc does not touch the interval [, ] and all the roots E are complex. This happens
for g > g0 = 1.13459 (sinh 1/g0 = 1).
7. Numerical results
In this section we present the comparison between the analytic results derived in
Section 6, and the numerical ones obtained by solving directly the Richardsons equations
(6) for the models considered above.
7.1. Two-level model
In this model there are two energy levels, 1 , with a degeneracy N/2. Hence the
Richardsons equations (6) become
1
N
=
G
2
1
1
1 E
1 + E
M

=
2
,
E E
(76)
with M = N/2. The total number of solutions of this system of equations is given by the
N
combinatorial number CM
. The solution which corresponds to the ground state of the BCS
Hamiltonian is the one for which E (G) 1 , in the limit where G 0. For any
nonzero value of G, all the roots E form complex conjugate pairs surrounding the lower
energy 1 .
The reduced coupling constant for this case is taken as g = GN . Our numerical results
are presented in units of 1 , which is equivalent to set 1 = 1.
In Fig. 1 we plot the distribution of the M = 100 roots E , for three different values
of the coupling constant g = GN , together with the analytic curves derived in Section 6.
There is an optimal agreement between the numerical and analytic results in the three
regimes: normal (g < 1 ), critical (g = 1 ), and superconducting (g > 1 ). As we increase
the value of M up to 800 pairs, the fit of the numerical data to the analytical curves
improves, and in the limit M the discrete roots collapse into the continuous curves.
(M), whose real part lies closer to ,
In particular, the complex roots Emax (M) and Emax
0
approach their continuum limit value, namely,
lim Emax (M) = 0 + i,
(77)
497
Fig. 2. Plots of Im Emax (M) and 0 Re Emax (M) versus M , for the two-level model. 0 and are
given in Eqs. (78) and (79), while numerical values of are given in Table 1.
Table 1
Numerical and analytical values of the exponent appearing in Eq. (80). The numerical values are extracted from
the real and imaginary parts of Emax (M)
Analytical
Numerical ()
Numerical (0 )
where
g = 1.51
g = 1.01
g = 0.51
2/3
0.635
0.672
1/3
0.329
0.328
1/2
0.446
0.513
0,
g 1 ,
1 1 g/1 , g 1 ,

g 2 12 , g 1 ,
=
0 =
0,
g 1
(78)
(79)
in agreement with Eqs. (51) and (64).

Fig. 2 shows the following scaling behaviour Emax (M):
1
,
(80)
M
where the exponent depends on the regime of the system. The numerical and analytic
results of are given in Table 1.
The analytical value of can be derived as follows. For M sufficiently large we can
write
Emax (M) (0 + i)
Emax (M) = (0 + i) + ,
(81)
498
where is a small number. Using Eqs. (53), (56) and (58), the density of roots r(Emax )
behaves as
1/2, g > 1 ,
g = 1 ,
r(Emax ) AM( ) ,
(82)
= 2,
1,
g < 1 ,
where A is a parameter which only depends on g and 1 . The magnitude of can be
fixed by imposing the existence of at least one root in the interval (0 + i, 0 + i + ),
namely,
1 = r(Emax )| | AM| |+1
1
,
M 1/(1+)
(83)
which is the desired result (80) with = 1/(1 + ).

As we have seen in Fig. 1, the roots E lie on the analytic curves to a very good degree
of approximation. A further check of the analytic solution is to compare the density of roots
r( ), which is given by r(E) = |h(E)|/ for g 1 (see Eq. (37)), and by r(E) = |s(E)|
for g 1 (see Eq. (56)). In Fig. 3 we plot the fraction of roots per unit length r(E)/M as
a function of the arc length s [1, 1], where the s = 1 corresponds to the end points of
the arc . The agreement is extremely accurate for M = 800.
The value of the conserved quantities obtained in Eq. (53) also agree with the numerical
ones in the large-N limit, in the three different regimes.
Finally, it is interesting to study the behaviour of the energy as a function of the number
of pairs M. The leading term is given by Eq. (52) for g 1 and by Eq. (60) for g 1 .
In Fig. 4 we plot e e(M) versus 1/M, where e(M) = E(M)/M is the energy per pair.
Fig. 3. Density of roots r(E)/M as a function of the arc length. The continuous curves give the analytic result
while the step like curves are the numerical results obtained for M = 800 and three values of g. The total length
of the curve has been divided into 50 parts. The height of each part is the fraction of roots E per unit length E.
499
Fig. 4. Plot of e e(M) versus 1/M for the two-level-model and different values of g. The continuous lines
are given by Eqs. (84) for g = 1 , and Eq. (85) for g = 1 .
The linear behaviour of the numerical data means that the next leading correction to E(M)
is a constant term, E1 , whose value is given by the analytic expression

E1
g 12 /M, g 1 ,
= e e(M) =
(84)
M
(1 |0 |)/M, g 1 ,
where and 0 are given by Eqs. (51) and (64), respectively. The formula for g > 1
has been derived by Richardson in Ref. [8], studying the next-to-leading order of the
electrostatic model. The formula for g < 1 is an educated guess which needs a proof.
At g = 1 , the numerical data show a large deviation, for small M, from the formula (84),
namely 1 /M. There are theoretical reasons to believe that the scaling behaviour of E1 /M
at g = 1 , is given by the following formula (see Fig. 4):

E1
A
1
(85)
= 1
4/3 , , A = 0.62258,
M
M M
which would imply that the next-to-next leading correction of the total energy goes as
M 1/3 , rather than M 1 . In fact, the formula given by Richardson [8] for the O(M 1 )
energy correction, which is given by E2 = g(3g )(g )/(4N2 ), breaks down at
g = 1 where the gap vanishes. The derivation of Eq. (85) shall be given elsewhere.
7.2. Equally spaced model
This model is defined by N nondegenerate energy levels j = d(2j N 1), j =
1, . . . , N , where d = /N is the single-particle level spacing. The first numerical study
of this model, using the exact solution, was done by Richardson in Ref. [6], where he
introduced new set of variables, related to the roots E , in order to handle the singularities
arising at particular values of the coupling constant g (see below). This work was revised
recently in connection to ultrasmall metallic grains in [36,37].
500
Fig. 5. Evolution of the real and imaginary parts of E (g), in units of d = /N , for the equally spaced model
with M = N/2 = 8, as a function of the coupling constant g. For convenience the energy levels are chosen in this
figure as j = 2j . This figure confirms numerically the behaviour found in Ref. [6].
The ground state of the system, at half-filling M = N/2, is the one for which
limg0 E (g) ( = 1, . . . , N/2).
The reduced coupling constant is given by g = G/d = GN/ in this section. And all
our numerical results are presented in units of , which is equivalent to set = 1.
In Fig. 5 we plot the real and imaginary parts of E , for a system with M = N/2 = 8
pairs. For small values of g, all the pair energies are real and below the corresponding
energy levels at g = 0. At a certain value g1 , the two closest pairs to the Fermi energy,
namely EM and EM1 , coincide at M1 and then, for larger values of g > g1 they form
a complex conjugate pair, while the rest of the pair energies remain real. At higher critical
values of g the same phenomena happens for the other energies, until all of them become
complex. At the critical values of g the Richardson equations (6) are singular, but the
singularities can be resolved by the change of variables introduced in [6].
In Fig. 6 we show the distribution of roots for M = 100 pairs, together with the
theoretical curves obtained in Section 6, for three different values of g. For g = 0.5 and
1 there is a fraction of roots which become complex, falling into the arcs described by
Eq. (75), while the real roots extend from the cutting point of the arc with the real axis
down to the energy . For g = 1.5 all the roots are complex and fall into the theoretical
curve (75). The numerical data are consistent with the value g0 = 1.13459 above which all
the energies become complex.
As in the two-level case, the root Emax (M) closest to 0 + i satisfy the scaling law (80)
with an exponent = 2/3. This result can be proved along the same lines as for the two-
501
Fig. 6. Plot of the roots E for the equally spaced model in the complex -plane. The discrete symbols denote
the numerical values for M = 100. The continuous lines are given by Eq. (75). All the energies are in units of .
Table 2
Numerical and analytical values of the exponent appearing in Eq. (80) for the equally spaced model. The
notation is the same as in Table 1
Analytical
Numerical ()
Numerical (0 )
g = 1.5
g=1
g = 0.5
2/3
0.646
0.658
2/3
0.645
0.659
2/3
0.642
0.661
level case. In Table 2 we give the comparison between the theoretical and numerical values
of , and in Fig. 7 we show the scaling behaviour for three values of g.
In Fig. 8 we give the fraction of roots per unit length r(E)/M for M = 800 and three
values of g. Again, the numerical data fit accurately the theoretical result r(E)/M =
|h(E)|/ , where h(E) is given by Eq. (71).
A further check is presented in Fig. 9, where the distribution of the conserved quantities
(i ) for M = 20 is presented, together with their continuous limit curves given in Eq. (73).
A very good agreement is achieved even for such a small number of pairs.
We have finally studied the condensation energy EC /M as a function of M, obtaining
the behaviour EC (M) = e M + E1 , where E1 (g) is plotted in Fig. 10. An analytic
expression for E1 (g) is not known [8]. In order to understand the numerical results we
have made two fits. The first one is based on the result obtained for the two-level case,
namely,
E1 (g) = g
,
2
(86)
502
Fig. 7. Plots of Im Emax (M) and 0 Re Emax (M) versus M 2/3 , for the equally spaced model.
Fig. 8. Density profile of the complex roots for the equally spaced model with M = 800. Notations are the same
as in Fig. 3.
where is given by Eq. (67). Fig. 10 shows that (86) is a good fit for large values of g,
where all the roots are complex, forming a single arc. For g g0 = 1.13459 there is a
fraction Mcomplex /M of complex roots, and a fraction Mreal /M of real roots. This suggests
503
Fig. 9. Conserved quantities in the region in the equally spaced model for M = 20 and different values of g.
The continuous lines are given by Eq. (73).
Fig. 10. Plot of E1 (g) versus g for the equally spaced level model. The fits of the numerical data are given by
Eqs. (86) and (87) in units of .
the following mixed fit:

Mcomplex
Mreal
E1 (g)
= g
+ (A + Bg)g
,
2
M
M
A = 0.6735,
B = 0.1729,
(87)
which describes pretty well E1 (g) in the whole range of gs. In the weak coupling regime
there have been proposed several formulas for the finite size corrections of the condensation
504
energy EC using the DMRG method [33] and the exact solution [34]. In the latter reference
it is shown that for couplings g > g = 1/ ln N , the condensation energy EC can be splitted
in two contributions, one due to the complex energy pairs (the BCS term) and another one
due to the real roots (perturbative term). It would be interesting to clarify the relation
between formula (87) and Eq. (24) in Ref. [34].
8. Conclusions and prospects

In this paper we have reviewed and completed Gaudins continuum limit of the exactly
solvable BCS model, and compared its predictions with the numerical solutions describing
the ground state of the two-level and the equally spaced models. We have confirmed
previously known results [6,25,36,37], and obtained new ones, which we list below.
Analytic and numerical determination of the curves , where the roots E of the
Richardson equations lie, together with their density along . For the equally spaced
model the curves are given for arbitrary values of the coupling constant g, and for
the two-level model in the normal regime, i.e., g < 1 , the analytic curves differ
from those in [10].
Analytic expression of the eigenvalues j of the conserved quantities found by
Cambiaggio, Rivas, and Saraceno [11], which fit accurately the numerical data, even
for a few levels.
Study of the scaling properties of the ground state energy E(M) with the number
of pairs M which, in the large M limit, behaves generically as E(M) = e M +
E1 + O(1/M). For the two-level model, in the superconducting regime, we find an
agreement with the analytic result for E1 [8], while in the normal regime we have
guessed an analytic expression for E1 , which fits perfectly the numerical data. In
the critical case there are large deviations in E(M), which can be explained by a
term O(M 1/3). In the equally spaced model the condensation energy also receives a
constant contribution, E1 , which is fitted with a phenomenological formula which
combines the effects of the complex and real roots.
Scaling behaviour of the roots Emax (M) which lie closest to the end points of the
curves . This scaling can be characterized by a critical exponent , whose analytic
values fit well the numerical ones. In the superconducting regimes (i.e., g > 0 for
the equally spaced-model and g > 1 for the two-level model) we find = 2/3, while
in the normal (critical) regimes of the two-level model we find = 1/2 (1/3). In this
manner the exponent characterizes the nature of the ground state. An interesting
question is wether shows up in physical observables.
Generalization of Gaudins conformal mapping for the equally spaced model using
elliptic functions, which degenerate into trigonometric ones in the limit . This
suggest the existence of a rich mathematical structure.
Finally, we shall mention some problems which are worth to investigate along the lines
suggested by the present work.
505
Study of the excited states and the finite size effects in a more geometrical way,
completing the work initiated by Richardson in [8].
Quantum dots at finite temperature is another difficult problem, which has been treated
combining several techniques depending on the temperature regime [26,27,38]. Is it
possible to find a thermodynamic Bethe ansatz for this system?
Study of solutions of Richardson equations with several arcs, i.e., K > 1 in the notation
of Section 5. For the standard BCS model they must describe very high excited states
formed by separate condensates in interaction. This case may be relevant to systems
such as arrays of superconducting grains or quantum dots. From a mathematical point
of view, the cases with K > 1 seem to be related to the theory of hyperelliptic curves
and higher genus Riemann surfaces, which may shed some light on this physical
problem.
Generalization of Gaudins equations to Richardson models based on arbitrary Lie
groups G [16]. The standard case corresponds to the choice G = SU(2). The
electrostatic model for generic G has charges labelled by vectors of dimension rank G.
It is naturally to expect that the roots Ea (a = 1, . . . , rank G) should fall into several
arcs labelled by the index a = 1, . . . , rank G.
Acknowledgements
We thank Joo Lopes dos Santos, Rodolfo Cuerno, Miguel Angel Martn-Delgado, and
Javier Rodrguez-Laguna for computational help, and Didina Serban for Ref. [10]. This
work has been supported by the Spanish grants BFM2000-1320-C02-01 (J.M.R. and G.S.),
and BFM2000-1320-C02-02 (J.D.).
Appendix A. Conformal mapping for the equally spaced system

In this appendix we study the analytic structure underlying the shapes of the curves
defined by Eqs. (74) and (75).
A.1. Case
In this limit it is convenient to make a conformal mapping [10] from the upper half-plane
Im > 0, with a cut in the segment (0, i), into the half-strip > 0, /2 < < /2,
= i cosh( + i).
(A.1)
In the plane u = + i the curve (74) becomes

tanh + cot = 0.
(A.2)
Figs. 11 and 12 show, in the u and planes, the arcs defined by Eqs. (A.2) and (74). The
point O ( = 0 + i) is mapped into the origin (u = 0), where the field h vanishes. In fact
h( ), as a function of u, is simply h = 20 u. The point O ( = 0 i) is the mirror
reflection of O. The arc OA is the one that corresponds to the actual numerical solution for
506
Fig. 11. Half-strip in the u = + i plane that is conformally equivalent to the upper half -plane, with a cut in
the interval [0, 0 + i]. The arcs OA and OB are given by Eq. (A.2). This figure is borrowed from Ref. [10].
Fig. 12. Location of the roots E in the -plane. The complex roots condense in the arc OAO , given by Eq. (74),
while the real ones are distributed in the segment X A, with a real root between every two energy levels. The
point A is given by Eq. (A.3). This figure is borrowed from Ref. [10].
Im > 0, while its mirror image, O A, is the one for Im < 0. We thus see that the whole
arc OAO cuts the real axis at the point A with an energy
A = 0 sinh 0 ,
0 tanh 0 = 1.
(A.3)
These results are expected from numerical studies, which show that for weak couplings
only a fraction of energies E form complex conjugated pairs, which in the limit N
organize themselves into arcs, while the other energies remain real and located between
the lower energy levels (see Fig. 6). We can check this result computing the number of real
and complex roots E . From Eq. (A.3) we deduce that the real roots occupy the interval
[, sinh 0 ], with a density which is 20 . Hence the total number, Mreal of real roots
507
is
Mreal = 20 ( sinh 0 ).
(A.4)
Since the total number of roots is M = Mreal + Mcomplex = N/2 = 20 , we deduce that
Mcomplex = 20 sinh 0 .
(A.5)
This result agrees with the one obtained integrating the field h( ) along the curve L
surrounding (recall Eq. (44)). This integration is more easily done in the u-strip
0
+i/2

Mcomplex =
h( ) d = 4
L
(20 u)(i sinh u du),
(A.6)
where the factor 4 comes from the contribution of the up and down, and left and right pieces
of the contour L. Using contour arguments, it can be shown that the BCS Eqs. (43)(45)
also hold in cases where cuts the energy interval . As a side comment, we observe that
Mcomplex = 2d
sinh 0 , agrees with the heuristic argument stating that there are roughly
BCS /d energy levels around the Fermi level that are strongly affected by the pairing
interactions [39]. See also Ref. [34] for an approximate equation giving the number
of complex pairs (Eq. (49)).
A.2. Generic case
In order to compare with the numerical results presented in the Section 7, it is
convenient not to make the approximation . The appropriate change of variables
that generalizes (A.1) is given by
=
i cosh u
,

2
1 sinh u
(A.7)

in terms of which h = 20 u. The integral h( ) d can be more easily done in the
u-plane, yielding for the equation of the curve

iu cosh u
i
= 0,
R
(A.8)

2 arcsin sinh u
1 sinh u
which in the -plane gives Eq. (75), namely

1 + 2 /2
2 + 2
R i arcsin
= 0.
+ Argsinh
1 2 /2
2 2
(A.9)
In the limit this equation turns into Eq. (74). We can ask when this curve cuts the
interval [, ]. A convenient parametrization of the cutting point is (recall Eq. (A.3))
A =
sinh 0

2 .
1+
cosh
(A.10)
508
Plugging (A.10) into (A.9) we deduce the equation for 0 ,

0 sinh 0

2 = Argsinh cosh 0 ,

1+
cosh
(A.11)
which only has real solutions provided . When > the arc does not touch the
interval [, ] and all the roots E are complex. This happens for g > g0 = 1.13459.
This value of g0 is obtained solving the equation sinh 1/g0 = 1.
To complete this appendix we shall show that Eq. (A.7) yields a conformal mapping
that generalizes the one found by Gaudin for generic values of /. To do so, let us first
perform the change of variables

,
u=i
(A.12)
2
and introduce the parameters
k=
2 + 2
k =
1
2 + 2
1 k2,
(A.13)
in terms of which (A.7) becomes

=
ik sin
(A.14)
1 k 2 sin2
Using the elliptic functions with modulus k, and conjugate modulus k , we define a new
variable z through [40]
sin = sn(z, k),
cos = cn(z, k),

1 k 2 sin2 = dn(z, k).
(A.15)
The relation between z and is given by the equation

z(k, ) =
0
(A.16)
1 k 2 sin2
while the inverse function is denoted as = am(z, k). The relation between and z reads
= ik
sn(z, k)
.
dn(z, k)
(A.17)
Finally, using the relation sn(K + v, k)/ dn(K + v, k) = cn(v, k)/k , where K is the halfperiod of the elliptic integrals, Eq. (A.17) turns into
= i cn(v, k),
(A.18)
where z = K + v. The relation between the variables u and v is given by
iu =
v
n(K + v)
2q n
+ am(K + v, k) =
+
sin
,
2n
2
2K
n(1 + q )
2K
n=1
(A.19)
509
Fig. 13. Graphical representation of the conformal map (A.18), with the correspondences given in Table 3.
Table 3
Correspondences of the boundary regions of the conformal map (A.18)
Cut
Re = 0
0 < Im <
K < Re v < K
Im v = 0
Lower band
< Re < 0
Im = 0
Re v = K
0 < Im v < K
Upper band
0 < Re <
Im = 0
Re v = K
0 < Im v < K
Off band
| Re | >
Im = 0
K < Re v < K
Im v = K
where q = eK /K . In the limit / 1 we get k 0, K /2, and q 0, which

implies that iu v and Eq. (A.18) reduces to (A.1).
Eq. (A.18) gives a conformal map of the rectangle K < Re v < K, 0 < Im v < K
onto the upper half-plane Im > 0, with a cut along the linear segment 0 < Im (see
Ref. [41] and Fig. 13). Some examples of this mapping are given by
v = 0, K, K + iK
= i, 0, ,
(A.20)
which imply the identifications between the sides of the rectangle in the v-plane and the
relevant energy regions in the -plane, given in Table 3.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
J. Bardeen, L.N. Cooper, J.R. Schrieffer, Phys. Rev. 108 (1957) 1175.
R.W. Richardson, Phys. Lett. 3 (1963) 277.
R.W. Richardson, Phys. Lett. 5 (1963) 82.
R.W. Richardson, N. Sherman, Nucl. Phys. B 52 (1964) 221.
R.W. Richardson, J. Math. Phys. 6 (1965) 1034.
R.W. Richardson, Phys. Rev. 141 (1966) 949.
M. Gaudin, J. Physique 37 (1976) 1087.
510
[10] M. Gaudin, tats propres et valeurs propres de lHamiltonien dappariement, unpublished Saclay preprint,
1968, included in: Travaux de Michel Gaudin, Modles Exactament Rsolus, Les ditions de Physique,
France, 1995.
[11] M.C. Cambiaggio, A.M.F. Rivas, M. Saraceno, Nucl. Phys. A 624 (1997) 157.
[12] L. Amico, A. Di Lorenzo, A. Osterloh, Phys. Rev. Lett. 86 (2001) 5759.
[13] J. Dukelsky, C. Esebbag, P. Schuck, Phys. Rev. Lett. 87 (2001) 66403.
[14] J. Dukelsky, P. Schuck, Phys. Rev. Lett. 86 (2001) 4207.
[15] J. Dukelsky, S. Pittel, Phys. Rev. Lett. 86 (2001) 4791.
[16] M. Asorey, F. Falceto, G. Sierra, hep-th/0110266, Nucl. Phys. B, in press.
[17] P.P. Kulish, N. Manojlovic, nlin.SI/0103010.
[18] H.M. Babujian, J. Phys. A 26 (1993) 6981.
[19] H.M. Babujian, R. Flume, Mod. Phys. Lett. 9 (1994) 2029.
[20] G. Sierra, Nucl. Phys. B 572 [FS] (2000) 517.
[21] L. Amico, G. Falci, R. Fazio, J. Phys. A 34 (2001) 64256434.
[22] H.-Q. Zhou, J. Links, R.H. McKenzie, M.D. Gould, cond-mat/0106390.
[23] J. von Delft, R. Poghossian, cond-mat/0106405.
[24] G. Sierra, Integrability and conformal symmetry in the BCS model, in: A. Cappelli, G. Mussardo (Eds.),
Proc. of the NATO Advanced Research Workshop on Statistical Field Theories, Como, Italy, June 2001,
Kluwer Academic, in press, hep-th/0111114.
[25] G. Sierra, J. Dukelsky, G.G. Dussel, J. von Delft, F. Braun, Phys. Rev. B 61 (2000) 11890.
[26] A. Di Lorenzo, R. Fazio, F.W.J. Hekking, G. Falci, A. Mastellone, G. Giaquinta, Phys. Rev. Lett. 84 (2000)
550.
[27] G. Falci, A. Fubini, A. Mastellone, cond-mat/0110457.
[28] J. Dukelsky, C. Esebbag, S. Pittel, Phys. Rev. Lett. 88 (2002) 062501.
[29] J. Hogaasen-Feldman, Nucl. Phys. 28 (1961) 258.
[30] J. von Delft, D.S. Golubev, W. Tichy, A.D. Zaikin, Phys. Rev. Lett. 77 (1996) 3189.
[31] A. Mastellone, G. Falci, R. Fazio, Phys. Rev. Lett. 80 (1998) 4542.
[32] F. Braun, J. von Delft, Phys. Rev. Lett. 81 (1998) 4712.
[33] J. Dukelsky, G. Sierra, Phys. Rev. Lett. 83 (1999) 172;
J. Dukelsky, G. Sierra, Phys. Rev. B 61 (2000) 12302.
[34] M. Schechter, Y. Imry, Y. Levinson, J. von Delft, Phys. Rev. B 63 (2001) 214518.
[35] J. von Delft, D.C. Ralph, Phys. Rep. 345 (2001) 61.
[36] F. Braun, J. von Delft, in: Advances in Solid State Physics, Vol. 39, Vieweg, Wiesbaden, 1999, p. 341,
cond-mat/9907402.
[37] J. von Delft, F. Braun, in: I.O. Kulik, R. Ellialtioglu (Eds.), Proc. of the NATO ASI Quantum Mesoscopic
Phenomena and Mesoscopic Devices in Microelectronics, Ankara/Antalya, Turkey, June 1999, Kluwer
Academic, cond-mat/9911058.
[38] G. Falci, A. Fubini, cond-mat/0012339.
[39] M. Tinkham, Introduction to Superconductivity, 2nd edn., McGrawHill, 1996.
[40] E.T. Whittaker, G.N. Watson, A Course on Modern Analysis, Cambridge Univ. Press, Cambridge, 1980.
[41] Z. Nehari, Conformal Mapping, Dover, New York, 1975.

Operator product expansion in logarithmic

conformal field theory
Michael Flohr
Institute for Theoretical Physics, University of Hannover, Appelstrae 2, D-30167 Hannover, Germany
Received 10 October 2001; accepted 26 March 2002
Abstract
In logarithmic conformal field theory, primary fields come together with logarithmic partner fields
on which the stress-energy tensor acts non-diagonally. Exploiting this fact and global conformal
invariance of two- and three-point functions, operator product expansions of logarithmic operators
in arbitrary rank logarithmic conformal field theory are investigated. Since the precise relationship
between logarithmic operators and their primary partners is not yet sufficiently understood in
all cases, the derivation of operator product expansion formulae is only possible under certain
assumptions. The easiest cases are studied in this paper: firstly, where operator product expansions
of two primaries only contain primary fields, secondly, where the primary fields are pre-logarithmic
operators. Some comments on generalization towards more relaxed assumptions are made, in
particular, towards the case where logarithmic fields are not quasi-primary. We identify an algebraic
structure generated by the zero modes of the fields, which proves useful in determining settings in
which our approach can be successfully applied. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction
During the last few years, so-called logarithmic conformal field theory (LCFT)
established itself as a well-defined new animal in the zoo of conformal field theories in
two dimensions. To our knowledge, logarithmic singularities in correlation functions were
first noted by Knizhnik back in 1987 [41]. Although many features such as logarithmic
divergences of correlators and indecomposable representations were observed in various
places, most notably in [73,74], it took six years, until the concept of a conformal field
* Research supported by EU TMR network no. FMRX-CT96-0012 and the DFG String network (SPP no.
1096), Fl 259/2-1.
E-mail address: flohr@itp.uni-hannover.de (M. Flohr).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 3 5 - 3
512
M. Flohr / Nuclear Physics B 634 [FS] (2002) 511545
theory with logarithmic divergent behavior was introduced by Gurarie [27]. From then
one, there has been a considerable amount of work on analyzing the general structure
of LCFTs, which by now has generalized almost all of the basic notions and tools of
(rational) conformal field theories, such as null vectors, characters, partition functions,
fusion rules, modular invariance, etc., to the logarithmic case, see, for example, [13,16,21,
23,25,34,38,46,58,60,67,68,72] and references therein. Besides the best understood main
example of the logarithmic c = 2 theory and its cp,1 relatives, other specific models were
considered such as WZW models [1,22,45,64,65] and LCFTs related to supergroups and
supersymmetry [2,11,37,39,51,55,71,73].
Also, quite a number of applications have already been pursued, and LCFTs have
emerged in many different areas by now. Sometimes, longstanding puzzles in the
description of certain theoretical models could be resolved, e.g., the HaldaneRezayi state
in the fractional quantum Hall effect [7,28,70], multi-fractality [12], or two-dimensional
conformal turbulence [18,66,76]. Other applications worth mentioning are gravitational
dressing [5], polymers and abelian sandpiles [8,33,57,74], the (fractional) quantum Hall
effect [17,31,49], andperhaps most importantlydisorder [3,4,9,10,29,30,43,56,69].
Finally, there are even applications in string theory [42], especially in D-brane recoil [6,
14,15,26,44,52,53,59], AdS/CFT correspondence [24,35,40,47,48,62,63,75], as well as in
SeibergWitten solutions to supersymmetric YangMills theories, e.g., [19]. Last, but not
least, a recent focus of research on LCFTs is in its boundary conformal field theory aspects
[32,36,50,54,60].
However, the computation of correlation functions within an LCFT still remains
difficult, and only in a few cases, four-point functions (or even higher-point functions)
could be obtained explicitly. The main reason for this obstruction is that the representation
theory of the Virasoro algebra is much more complicated in the LCFT case due to the
fact that there exist indecomposable but non-irreducible representations (Jordan cells).
This fact has many wide ranging implications. First of all, it is responsible for the
appearance of logarithmic singularities in correlation functions. Furthermore, it makes
it necessary to generalize almost every notion of (rational) conformal field theory, e.g.,
characters, highest-weight modules, null vectors, etc. We note here that indecomposable
representations need not occur with respect to the Virasoro algebra, but may occur with
respect to (part of) a extended chiral symmetry algebra such as current algebras or Walgebras. For the sake of simplicity, we will confine ourselves in this paper to the case
where Jordan cells are with respect to the Virasoro algebra.
In particular, what was lacking so far is a consistent generic form of operator product
expansions (OPEs) between arbitrary rank logarithmic fields. Although such OPEs can be
derived from co-product considerations in the purely representation theoretical framework
[21,34], a direct approach trying to fix the generic form from global conformal covariance
of the fields is clearly desirable. For the simple case of a rank two LCFT, where Jordan cells
are two-dimensional, it was known since some time [9,27] that the two-point functions of
a primary (h;0)(z) and its only logarithmic partner (h;1)(z) are

(h;0)(z)(h;0)(w) = 0,

(h;0)(z)(h;1)(w) = D(h,h;1)(z w)2h ,

(h;0)(z)(h;1)(w) = D(h,h;2) 2D(h,h;1) log(z w) (z w)2h .
(1.1)
513
However, as we shall see, even in this simple case OPEs turn out to be more complicated,
and one needs all possible three-point functions as well. First results in this direction can be
found in [23,38,61,67]. Here we will start to close this gap and provide the general structure
of OPEs for fields constituting arbitrary rank Jordan cells under certain assumptions.
Let us briefly outline the basic problem: in ordinary conformal field theory, the generic
structure of the operator product expansion is fixed up to structure constants which depend
only on the conformal weights of the fields involved,

k,{n}
k
hk
|{n}| ({n})
hk (w) +
Cij (z w)
ij (z w) hk
(w) .
hi (z)hj (w) =
k
{n}
(1.2)
Here, the fields h are primaries, and the coefficients
k,{n}
ij
of the descendant contribu-
({n})
h
tions
= L{n} h = Ln1 Ln2 Lnl h are entirely fixed by conformal covariance. The point is that the structure constants Cijk can be easily determined if the two- and
three-point functions are known. In fact, these define constants Dij =
hi ()hj (0) and
Cij k =
hi ()hj (1)hk (0), respectively, where Dij is usually a diagonal matrix, i.e.,
Dij hi ,hj . The two-point functions define a metric on the space of fields, such that this
metric Dij and its inverse can be used to lower and raise indices (in field space), respectively. In particular, the OPE structure constants are simply given by
Cijk = Cij l D lk .
(1.3)
Now, in logarithmic conformal field theory, the metric induced by the two-point functions
is no longer diagonalsee (1.1) for the simplest case, where the metric, restricted to fixed

0
D1
conformal weight h, is of the form D D 2D log()()
2h . Note that, even worse, the metric
1
2
1
cannot any longer be factorized in a coordinate dependent part and a purely constant part.
It is the purpose of this paper to work out this metric together with all needed three-point
functions in order to find the correct equivalent to (1.3) in the logarithmic case.
The paper will proceed as follows. In the next section, we will set up our notation and
outline the different cases we will study in the following, progressing from the simplest
setting to some more involved indecomposable structures. Although we restrict ourselves
to indecomposable representations with respect to the Virasoro algebra alone, we will
encounter a surprisingly rich set of possibilities of which only the easier ones can be treated
with the methods presented in this paper.
In Section 3, we start the investigation in the simplest possible setting, i.e., where all
logarithmic partner fields in a Jordan cell are assumed to be quasi-primary. This setting is
as close as possible to the ordinary CFT case, and the generic form of one-, two-, and threepoint functions can be computed explicitly, as long as the primary fields within the Jordan
cells are assumed to be proper primaries. We also comment on the appropriate definition
of a Shapovalov form, and show that it is well defined under the above assumptions on the
structure of two-point functions.
In Section 4 we derive the generic structure of OPEs in this setting. Although we usually
work only with the holomorphic part of the LCFT, we briefly discuss locality in this section.
We also compute a fully elaborated example of a rank four LCFT with only proper primary
514
fields involved, demonstrating some perhaps unexpected features of operator products in

the logarithmic case.
Then, in Section 5, we consider two more general cases, where the primary fields are
not proper primaries, meaning that an OPE between two primary fields may contain a
logarithmic field. Both examples, fermionic fields and twist fields, already occur in the
best known LCFT, namely, the c = 2 system, which will serve as our prime example. We
show that OPEs can be computed for these cases as well, since these primary fields are not
part of Jordan cells.
Finally, Section 6 is devoted to the question, how the so far quite restrictive assumptions
can be relaxed. In particular, we investigate, under which circumstances logarithmic fields
can be non-quasi-primary without affecting the results of the preceding sections. We also
briefly address the other assumptions we made and discuss the difficulties one would
have to face attempting to derive the generic structure of OPEs and correlation functions
without them. In particular, we demonstrate that the case of Jordan cells containing primary
fields which are not proper primaries cannot be solved without further assumptions on the
structure of the CFT. Thus, we conclude that global conformal covariance fixes the generic
form of correlation functions and OPEs in a similar fashion as in the ordinary CFT case
only when the indecomposable representations have a rather simple form.
We conclude with a brief discussion of our results and possible directions for future
research.
2. Definitions and preliminaries

To start with, we fix some notation. In general, a rank r Jordan cell with respect to the
Virasoro algebra is spanned by r states {|h; r 1, . . . , |h; 1, |h; 0} with the property
L0 |h; k = h|h; k + (1 k,0 )|h; k 1.
(2.1)
These states are defined via limz0 (h;k) (z)|0 = |h; k, where |0 denotes the SL(2, C)
invariant vacuum with Ln |0 = 0 n 1. The fields (h;k) (z) with 0 < k < r are the
so-called logarithmic partner fields. The field (h;0) (z) is a primary field which forms the
only proper irreducible sub-representation within the module of descendants of the Jordan
cell. In the following we will denote the primary by h (z) (h;0) (z), if and only if it is a
proper primary field. We call a primary field proper primary, if its OPE with other proper
primary fields never produces a logarithmic field on the right-hand side. This definition is
motivated from examples of LCFTs, where the primary fields within Jordan cells do share
precisely this property. For instance, in the prime LCFT example, the c = 2 theory, there
exists a Jordan cell of rank two for h = 0. The primary field is the identity field which by
definition is a proper primary.
For completeness, we note that within a logarithmic CFT, Jordan cells of different
rank might occur, i.e., r = r(h) might be a function of the conformal weight of the
corresponding (proper) primary field. Of course, if r = 1, the Jordan cell reduces to an
ordinary highest weight state, and its module of descendants to an ordinary Verma module.
For more precise definitions see [72]. However, we will see later that consistency of the
515
operator algebra makes it virtually impossible that Jordan cells of different rank occur
within the same LCFT.
We have to distinguish between the proper primary fields h (z) (h;0)(z) in a
Jordan cell and so-called pre-logarithmic primary fields. Pre-logarithmic fields are Virasoro
primary fields, whose operator product expansions among themselves might lead to
logarithmic fields [46]. Typically, pre-logarithmic fields turn out to be twist fields.
Although no counter example is known, we cannot exclude that pre-logarithmic fields
might occur as primaries within a non-trivial Jordan cell. However, for the purposes of
this paper, we will not assume so for the beginning. Thus, initially we will assume that
primary fields in non-trivial Jordan cells are all proper primaries. Later, in section five, we
will discuss the case of pre-logarithmic fields.
We also say that the field (h;k) has Jordan level k in its Jordan cell, abbreviated as
J-level k. Proper primary fields have J-level zero by definition. Twist fields do not possess
a well-defined J-level. Instead, they carry a fractional charge q = /n whose denominator
denotes the branching number. Logarithmic operators can appear in OPEs of twist fields
h(q) and h(q ) , whenever q + q Z.
As discussed by F. Rohsiepe [72], the possible structures of indecomposable representations with respect to the Virasoro algebra are surprisingly rich. Besides the defining
condition (2.1), further conditions have to be employed to fix the structure. The simplest
case is defined via the additional requirement
L1 |h; k = 0,
0 k < r.
(2.2)
This condition means that all fields spanning the Jordan cell are quasi-primary. It will be
our starting point in the following.
The next complicated case is where logarithmic partners are not necessarily quasiprimary. An example is again provided by the simplest explicitly known LCFT, the c = 2
model. It possesses a rank two Jordan cell at h = 1 where the logarithmic partner is not
quasi-primary, i.e., L0 |h = 1; 1 = |h = 1; 1 + |h = 1; 0 and L1 |h = 1; 1 = | = 0, see
[21] for details. In general, this spoils attempts to make use of the global conformal Ward
identities, since these are consequences of global conformal covariance or, in other words,
of quasi-primarity of the fields. However, it seems that the condition of quasi-primary fields
can be relaxed under certain circumstances. If the state | generated by the action of L1
on |h = 1; 1 does not have a non-vanishing product with any of the states involved so far,
it will not affect the form of the conformal Ward identities. In the given example, this is the
case. We will postpone a more detailed discussion of this to the last section, but will give
some definitions motivated by an example, which will prove useful for this later discussion.
2.1. An example and zero mode content
In all known examples of LCFTs, the occurrence of indecomposable representations
can be traced back to certain pairs of conjugate zero modes. Let us explain this within
the example of the LCFT with central charge c = c2,1 = 2. It may be described, see, for
instance, [28], by two anti-commuting spin zero fields or ghost fields , = , with the
516
SL(2, C) invariant action
.
S i d2 z
(2.3)
In order to quantize this theory, one has to compute the fermionic functional integral
Z = D0 D1 exp(S).
(2.4)
This fermionic path integral, when computed formally, vanishes due to the zero modes of
the fields, which do not enter the action. To make it non-zero, the zero-modes have to be
added by inserting the fields into correlation functions,
1
D0 D1 exp(S) = 1.
Z =
(2.5)
2
As an immediate consequence, the vacuum of this theory behaves in an unusual way, since
its norm vanishes,
0|0 = 0, while the explicit insertion of the fields produces a nonzero result, 12
(z) (w) = 1. More generally, these insertions are also necessary
when arbitrary correlation functions involving only derivatives are computed, since
derivatives cancel the zero modes, i.e., the constant parts of the fields. Therefore, we
have

1
(z) (w) = 0,
2
1
1
(z) (w) (0) (0) =

,
but
(2.6)
4
(z w)2
where the latter correlator is computed in analogy to the free bosonic field. From the
viewpoint of conformal field theory, this strange behavior can be explained in terms of
logarithmic operators which naturally appear in the c = 2 theory. As shown in [27], this
CFT must necessarily possess an operator I of scaling dimension zero in addition to the
= I with L0 (half of) the Hamiltonian. This property alone
unit operator I, such that [L0 , I]
necessarily leads to the correlation functions

I(w)
II = 0,
I(z)I(w)
= 1,
(2.7)
I(z)
= 2 log(z w),
which can be proved by general arguments of conformal field theory such as global
conformal invariance and the operator product expansion. Furthermore, it follows that the
field I can indeed be identified with the normal ordered product of the fields, i.e.,
= 1 : :(z).
I(z)
2
The stress energy tensor of this CFT is given by the normal-ordered product
(2.8)
1
T (z) = : :(z),
(2.9)
2
and is easily seen to fulfill the correct operator product expansion with itself yielding the
central charge c = 2. To fix notation, the mode expansions of the fields read

,n zn ,
(z) = + ,0 log(z) +
(2.10)
n=0
517
where the s are the crucial zero modes. The above mode expansion is valid in the
untwisted sector (periodic boundary conditions), where n Z. In the twisted sector (antiperiodic boundary conditions) n Z + 1/2, and no zero modes are present. The anticommutation relations read for the case = in both sectors
1
{,n , ,m } = n+m,0 , for n = 0,
n
{,0 , ,0} = { , } = 0,
{ , ,0} = 1,
(2.11)
with all other anti-commutators vanishing. Note that the -modes become the creation
operators for logarithmic states. Indeed, the highest weight conditions of the standard
SL(2, C) invariant vacuum are
,n |0 = 0
n 0,
(2.12)
such that
1
= I(0)|0
|0
= |0.
2
(2.13)
It is instructive to conclude our

example with the little exercise to compute L0 |0 by using
the mode expansion Ln = 12 m :,nm ,m : as follows:
= 1
L0 |0
4

: m,0 m2 ,m ,m : |0
m

1
m2 (,m ,m ,m ,m ) ,0 ,0 |0
= +
4
m>0
1
1
= ,0 ,0 |0 = |0 = |0,
(2.14)
4
2
where the third equality follows from the highest weight condition (2.12), and otherwise
the anti-commutation relations (2.11) were used. This clearly demonstrates that the states
and |0 form an indecomposable Jordan cell with respect to the Virasoro algebra. The
|0
action of other Virasoro modes can be computed in the same fashion.
What we learn from this example is the crucial role of conjugate pairs of zero modes,
i.e., pairs , ,0 . It turns out that logarithmic fields are precisely those fields, whose mode
expansion contains . Therefore, it makes sense to talk of the logarithmicity of a
field, more precisely of its zero mode content. It is clear that there might be more pairs
of conjugate modes in more general LCFTs (e.g., in higher spin ghost system) and, hence,
a field might possess a higher zero mode content. The phrase zero mode is ambiguous:
we do not mean the zeroth mode in the mode expansion of a field, but pairs of conjugate
modes an , cn such that an annihilates to both sides, i.e.,
0|an = an |0 = 0, and cn is a
creator to both sides. In the example, ,0 are the annihilators and the creators. The zero
mode content of a field counts the number of creation operator zero modes. If the modes
are anti-commuting, we will also talk of an even or odd zero mode content of fields. Fields
of even zero mode content are called bosonic, fields of odd zero mode content fermionic,
respectively. It is important to note that a correlation function can only be non-vanishing,
518
if its total zero mode content is large enough to kill all annihilator zero modes. In our
example, any non-vanishing correlator must contain . Moreover, our example is of
fermionic nature, such that the zero mode content must always be even.
To fix notation, we will denote the number of creator zero modes of a field by Z0 ( ).
If the modes are fermionic, we further introduce Z ( ) and Z+ ( ) which count the
modes with respect to the spin doublet label = , respectively. Of course, Z0 ( ) =
Z ( ) + Z+ ( ) in this case. We stress that correlation functions
1 (z1 ) n (zn ) of
fields i (zi ) with Z0 (i ) = 0 for all i = 1, . . . , n must vanish. Moreover,
if the minimal
zero mode content of a given LCFT is N , all correlators vanish, whenever i Z0 (i ) < N .
Thus, such fields behave almost as null fields. They are not entirely null, since we may
insert those fields in a correlator with already sufficiently high total zero mode content
without necessarily forcing it to become zero. In our example, N = N + N+ = 1 + 1 = 2.
The rank of Jordan cells of a LCFT generated by anti-commuting fields is N/2 + 1,
otherwise it should be N + 1, as in the case of LCFTs from puncture operators in Liouville
theories [14,43,46].
2.2. Zero mode content and J-level
We will discuss in section six that the zero mode content allows to put bounds on the
J-levels of fields appearing on the right-hand side of OPEs. However, it is not clear whether
all LCFTs admit to assign a zero mode content in the above defined sense to all of its fields.
Thus, the basic problem one is faced with is that there is no a priori rule that restricts the
J-level in an OPE of the form
r(h)1
(h;k)
C(h1 ;k1 )(h2 ;k2 ) (z w)hh1 h2
(h1 ;k1 ) (z)(h2 ;k2 ) (w) =
h
k=0
fkk1 k2 (z w)(h;k) (w),
(2.15)
where the functions fkk1 k2 (x) collect possible logarithmic terms. What one would definitely
wish for would be something like a gradation such that the J-level of the right-hand side
of (2.15) is bounded as
k k1 + k2 .
(2.16)
Unfortunately, we know that this is not always true. Pre-logarithmic fields, for instance,
are true primary fields which produce a logarithmic field in their OPE which means that
1 = k > k1 + k2 = 0 + 0 in contradiction to the above bound.
Now, if the LCFT under consideration admits to assign a well-defined zero mode content
to each of its fields, we can trace back the origin of logarithms and of Jordan cells to
precisely that zero mode content, as indicated in the preceding subsection. Since the Jordan
cell structure is in this case generated by the existence of certain creator zero modes, it is
clear that the OPE of two fields can never produce fields on the right-hand side whose
zero mode content exceeds the initial one. This follows by the simple fact that the OPE
can be computed on the level of modes explicitly by contraction. Thus, it follows that the
existence of a well-defined zero mode content puts a natural bound on the maximal zero
mode content of OPEs. Therefore, instead of the above inequality (2.16), we then have for
519
the right-hand side of the OPE (2.15)

Z0 ((h;k) ) Z0 ((h1 ;k1 ) ) + Z0 ((h2 ;k2 ) ).
(2.17)
Since, on the other hand, the zero mode content determines the Jordan cell structure via the
off-diagonal action of the generators of the chiral symmetry algebra, the inequality (2.17)
implies the desired inequality (2.16).
It is clear that we are now faced with a dilemma. If we do not know a priori which fields
at which J-levels may contribute to the right-hand side of OPEs, we do not have a way to
compute the generic form of such OPEs, since we do not have enough information to fix its
structure. Thus, we are forced to make even more assumptions on the structure of LCFTs.
We will, therefore, assume throughout the whole paper that a condition of the form (2.16)
holds, as long as fields from Jordan cells are concerned. We will justify this assumption as
reasonable in section six for the case where the LCFT admits a description in terms of a
zero mode content. However, we do not know whether all LCFTs can be described in this
way.
As we will briefly discuss at the end of Section 6, there are indications that LCFTs may
not be consistent, if no such condition restricting the maximal J-level of OPEs exists. In
general, we do not have any means to compute the structure of arbitrary n-point correlation
functions in this case, but it is still possible to look at the two-point functions. Our results
indicate that these may indeed be inconsistent if, for example, primary members of Jordan
cells can produce logarithmic partners in their OPE.
3. SL(2, C) covariance
In ordinary CFT, two- and three-point functions are determined up to constants which
determine the operator algebra and must be fixed by the associativity of the operator
product expansion. Moreover, one-point functions are trivial, i.e.,
h (z) = h,0 , although
Zamolodchikov pointed out a long time ago, that in non-unitary CFTs, non-vanishing onepoint functions might be possible. For the beginning, we consider only the chiral half of
the theory, but keep in mind that LCFTs are known not to factorize entirely into chiral and
anti-chiral halfs.
In order to find the generic structure of two- and three-point functions in logarithmic
CFTs, we must consider different cases. We start with the simplest setting, as outlined in
the preceding section. Throughout this section we will, therefore, assume the following:
we consider correlation functions of fields (hi ;ki ) (zi ) from Jordan cells where (hi ;0)(zi )
are proper primary fields. This assumption guarantees that the operator product expansion
of two of the primaries will contain only primary fields (and their descendants). Thus,
the primaries behave exactly as in an ordinary CFT. Furthermore, we assume that all the
logarithmic fields in the Jordan cells are quasi-primary, i.e., that L1 (hi ;ki ) (0)|0 = 0 for all
ki = 0, . . . , r(hi ) 1. Such Jordan cells will be called proper Jordan cells in the following.
We remark that in any sensible LCFT there is at least one (possibly trivial) Jordan cell
which satisfies these assumptions, namely the h = 0 Jordan cell where the primary is the
identity field. The identity should exist in any sensible CFT, since it is the unique field
associated with the SL(2, C)-invariant vacuum.
520
Under these assumptions, as shown in the latter two references in [16], the action of the
Virasoro modes receives an additional non-diagonal term, namely,

Ln (h1 ;k1 ) (z1 ) (hn ;kn ) (zn )

=
(3.1)
zin zi i + (n + 1)(hi + hi ) ((h1 ;k1 ) (z1 ) (hn ;kn ) (zn ) ,
i
where n Z and the off-diagonal action is hi (hj ;kj ) (z) = ij (hj ;kj 1) (z) for kj > 0 and
hi (hj ;0) (z) = 0. This little extension has tremendous consequences. As we are going to
show, even the simplest quantities, namely the one-point functions, are severely modified in
their behavior. To start with, we recall that only infinitesimal conformal transformations in
the algebra sl(2, C) can be integrated to global conformal transformation on the Riemann
sphere. Thus, only the generators L1 , L0 , and L1 of the Mbius group admit globally
valid conservation laws, which usually are expressed in terms of the so-called conformal
Ward identities

L1 G(z1 , . . . , zn ) = i i G(z1 , . . . , zn ),
0 = L0 G(z1 , . . . , zn ) = i (zi i + hi + hi )G(z 1 , . . . , zn ),
(3.2)

L1 G(z1 , . . . zn ) = i zi2 i + 2zi [hi + hi ] G(z1 , . . . , zn ),

where G(z1 , . . . , zn ) denotes an arbitrary n-point function
(h1 ;k1 ) (z1 ) (hn ;kn ) (zn ) of
primary fields and/or their logarithmic partner fields. Here, we already have written down
the Ward identities in the form valid for proper Jordan cells in logarithmic conformal field
theories. We will see in Section 6, that the assumption of quasi-primary logarithmic fields
can be relaxed under certain circumstances.
3.1. One-point functions
Let us now apply the Ward identities (3.2) to an arbitrary one-point function G(z) =
(h;k) (z) of a field in a rank r Jordan cell. The identity for L1 states translational
invariance such that G(z) = E(h;k) must be a constant independent of the position z. But
the identity for L0 , stating scaling and rotational invariance, leads to the condition
hE(h;k) + (1 k,0 )E(h;k1) = 0.
(3.3)
In case of the one-point functions, special conformal transformations do not yield an

additional constraint. However, the above condition immediately results in the recursive
relation, E(h;r1l) = (h)l E(h;r1) , such that, if E(h;r1) is non-zero, automatically all
other one-point functions in this Jordan cell also do not vanish, as long as h = 0. For
h = 0, the only non-vanishing one-point function is the one of highest possible J-level,
i.e., E(h;r1) = 0, E(h;k) = 0 for 0 k < r 1. Note that E(h=0;r1) must be non-zero.
Otherwise, the whole Jordan module to fields of scaling dimension zero could be removed
from the theory, since it were orthogonal to all other states. Then, the remaining CFT would
not have a vacuum state. To be specific, we from now on normalize E(0;r1) = 1.
We can learn one more thing from the one-point functions: if a field (h;k) (z), 0 < k < r,
were not quasi-primary, special conformal transformations yield a non-zero result as long
521
as the expectation value of the non-quasi-primary contribution is non-zero. Namely,

L1
(h;k) (z) = (h)r1k L1 E(h;r1) =
(h1)
(z)
for a suitable (not necessarily quasi-primary) field (z) with conformal weight h 1
(which is not necessarily part of a Jordan cell, which is why we omit the J-level). We will
see in Section 6 that one property of the off-diagonal action of the Virasoro modes is to
reduce the zero mode content, i.e., Z0 ( ) < Z0 ( ). Hence, instead of quasi-primarity,
n(h,k)
a weaker condition will always hold, namely L1
(h;k)(0)|0 = 0 for a certain n(h, k)
depending on the conformal weight h and the J-level k . Moreover, the reduced zero mode
content may already force the vacuum expectation value to vanish, i.e., the situation may
arise that L1 (h;k) (0)|0 = 0, but L1
(h;k) (z) = 0. This is precisely the case for the
example given in Section 2.1, namely the Jordan cell at h = 1. In particular, n(h = 1, 1) = 2
in the c = 2 case, i.e., L21 |h = 1; 1 = L1 | = 0 but
= 0.
3.2. Two-point functions
The next step is to consider two-point functions G =
(h1 ;k1 ) (z1 )(h2 ;k2 ) (z2 ) of two
fields belonging to Jordan cells of ranks r1 , r2 , respectively. Translational invariance tells
us that G = G(z12 ) is a function of the distance only. Scaling invariance then leads to the
ordinary first order differential equation

(z12 z12 + h1 + h2 )G(z12 ) + (h1 ;k1 1) (z1 )(h2 ;k2 ) (z2 )

+ (h1 ;k1 ) (z1 )(h2 ;k2 1) (z2 ) .
(3.4)
The generic solution to this inhomogeneous equation is already surprisingly complicated.
Let us introduce some nomenclature to denote where in a correlator logarithmic partners
of a primary are inserted by writing

(h1 ;k1 ) (z1 )(h2 ;k2 ) (z2 ) (hn ;kn ) (zn ) = Gk1 ,k2 ,...,kn (z1 , z2 , . . . , zn ).
(3.5)
The above equation then becomes (z12 z12 + h1 + h2 )Gk1 ,k2 (z12 ) = Gk1 1,k2 (z12 )
Gk1 ,k2 1 (z12 ) with solution

Gk1 ,k2 (z12 ) = (z12 )h1 h2 D(h1 ;k1 )(h2 ;k2 )
z12
d
1h1 h2

Gk1 1,k2 ( ) + Gk1 ,k2 1 ( ) .
(3.6)
An explicit solution can be found in a hierarchical way, starting with the two-point function
of proper primary fields, G0,0 (z1 , z2 ). The conformal Ward identities then reduce to the
common CFT case with the well-known solution

h1 (z1 )h2 (z2 ) = D(h1 ;0)(h2 ;0)h1 ,h2 (z1 z2 )h1 h2 .
(3.7)
However, to be consistent with insertion of an OPE, the constant must satisfy D(h;0)(h;0) =
(0;0)
E(0;0) = 0, due to our results on the one-point functions. Hence, G0,0 (z1 , z2 ) = 0
C(h;0)(h;0)
522
and, moreover, E(h;0) = 0 for h = 0 since the form of the two-point function does not admit
contributions from other one-point functions. We conclude that the only non-vanishing
one-point function of fields in Jordan cells is
(0;r(0)1).
The reader should note that in the above reasoning we crucially made use of the
assumption that the primary fields are proper, i.e., that their OPE does not contain any
logarithmic fields. We will come back to this point later. However, if the identity of the CFT
belongs to a Jordan cell, then the assumption that the primary be proper is automatically
satisfied. Moreover, as already stressed earlier, it is assumed throughout this section that
all logarithmic partner fields are quasi-primary.
We can go on and consider G1,0 (z1 , z2 ) next. The Ward identities now yield an
additional term proportional to G0,0 , which luckily vanishes as just shown. Therefore,
we can conclude that G1,0 is non-zero, if and only if E(0;1) is non-zero, i.e., if and only if
r 1 = 1. Going on in this manner, we finally arrive at the general statement

(h1 ;k) (z1 )(h2 ;0) (z2 ) = (h1 ;0)(z1 )(h2 ;k) (z2 )
= h1 ,h2 k,r1 D(h1 ,h1 ;r1) (z12 )2h1 ,
(3.8)
which does not depend on which of the two fields is the field of maximal J-level. It is
more complicated to compute two-point functions where both fields have J-level larger
zero, except when the Jordan rank is r = 2. Then the only other possibility is G1,1 (z1 , z2 ),
where the Ward identities yield contributions proportional to G1,0 = G0,1 with solution
(h;1) (z1 )(h;1)(z2 ) = (z12 )2h [D(h,h;2) 2D(h,h;1) log(z12 )].

When generalizing to arbitrary rank Jordan cells, the following picture emerges for
the two-point functions: the structure constants depend only on the total J-level, i.e.,
D(h;k)(h;l) = D(h;k )(h;l ) D(h,h;k+l) for k + l = k + l , and they vanish, if the total Jlevel is less than the rank of the vacuum representation, i.e., D(h;k)(h;l) = 0 for k + l + 1 <
r(h = 0). Another consequence is that the only non-vanishing one-point function of type
E(h;k) is E(h=0;r(h=0)1). This, in turn, implies that a logarithmic CFT is only consistently
possible, if the vacuum representation is a Jordan cell representation of maximal rank
r(h = 0) r(h) h = 0. We then say that the LCFT has rank r. Putting things together,
the complete solution for the two-point function must have the form
(h1 ;k1 ) (z1 )(h2 ;k2 ) (z2 ) = h1 ,h2
k +k
1
2

=0
(2)
log (z12 )
D(h1 ,h2 ;k1 +k2 )
!
(z12 )h1 h2 ,
(3.9)
where we have indicated the implicit condition h1 = h2 and where for a rank r LCFT all
constants D(h,h;k) = 0 for k < r 1. This result was first obtained in [67]. In this way,
the two-point functions define for each possible conformal weight h matrices G(2)
k1 ,k2 of
size r(h) r(h). However, these matrices depend only on 2r(h) r yet undetermined
constants D(h,h;k) , r 1 k 2r(h) 2. Moreover, all entries above the anti-diagonal
are zero. This last property, i.e., that D(h,h;k) = 0 for k < r 1, is due to the one-point
523
functions since
D(h,h;k) =
1
2r(h) k 1

01 ,2 r(h)1
1 +2 =k
C(h,h,0;k+r1),
0,
(0;r1)
C(h;1 )(h;2 ) E(0;r1)
for r 1 k < 2r 1,
else.
(3.10)
Note that the three-point structure constants do, in effect, only depend on the total J-level,
as we have tried to indicate in our notation. The special form of the two-point structure
constant matrices ensures that they are always invertible.
It is often very convenient to work with states instead of the fields directly, in particular
when purely algebraic properties such as null states are considered. As usual, we have
an isomorphism between the space of fields and the space of states furnished by the
map |h; k = (h;k)(0)|0. Although one does not necessarily have a scalar product on
the space of states, one can introduce a pairing, the Shapovalov form, between states and
linear functionals. Identifying the out-states with (a subset of) the linear functionals equips
the space of states with a Hilbert space like structure. As in ordinary conformal field
theory, we have
h; k| = (|h; k) = limz0
0|(h;k) (1/z). Using now that logarithmic
fields transform under conformal mappings z f (z) as
(h;k) (z) =

k

1 l f (z) h
(h;kl) f (z)
l
l! h
z
l=0

k

f (z) f (z) h

1

logl
(h;kl) f (z) ,

l!
z
z
l=0
the out-state can be re-expressed in a form which allows us to apply (3.9) to evaluate the Shapovalov form. In ordinary conformal field theory, we simply get
h| =
limz
0|z2h h (z) such that
h|h = h,h up to normalization. Interestingly, the transformation behavior of logarithmic fields yields a very similar result, canceling all logarithmic divergences. Thus, we obtain for the Shapovalov form
h; k|h ; k = h;h D(h,h ;k+k ) ,

which is a lower triangular matrix. To demonstrate this, we consider the example of a rank
two LCFT. Then we clearly have
h; 0|h; 0 = 0,
h; 1|h; 0 =
h; 0|h; 1 = D(h,h;1) and
with
lim
0|(h;1)(1/z)(h;1)(0)|0

= lim
0|z2h (h;1)(z) + 2 log(z)(h;0)(z) (h;1)(0)|0
z0
the desired result

h; 1|h; 1 = D(h,h;2) . Hence, the Shapovalov form is well defined and
non-degenerate for the logarithmic case much in the same way as it can be defined for
ordinary CFTs. Note that the definition of the Shapovalov form does not depend on whether
the CFT is unitary or not.
524
For completeness, we mention that the Shapovalov form is not uniquely defined in
LCFTs, because the basis {|h; k: k = 0, . . . , r(h) 1} of states is not unique. The reason
is that we always have the freedom to redefine the logarithmic partner fields, or their states
respectively, as

(z) = (h;k) (z) +
(h;k)
k

i (h;ki) (z)
i=1
with arbitrary constants i . At this state, there are no further restrictions from the structure
of the LCFT which could fix a basis within the Jordan cells. Only the proper primary field,
or the proper highest-weight state respectively, is uniquely defined up to normalization.
3.3. Three-point functions
The three-point functions can be fixed along the same lines, although the procedure is
now more complicated. For each triplet h1 , h2 , h3 of conformal weights, we find a set
of r(h1 ) r(h2 ) r(h3 ) functions Gk1 ,k2 ,k3 (z1 , z2 , z3 ). From now on we will restrict
ourselves to the case where r(h) = r for all Jordan cells in the LCFT. We will see shortly
that otherwise no consistent definition of OPEs seems possible. With this restriction, we
can collect the set of three-point functions into r matrices, each of size r r, namely the
(3)
matrices (Gk1 )k2 ,k3 .
A closed formula of the type as given above for the two-point function is extremely
lengthy. However, the three-point functions can all be given in the form:

(h1 ;k1 ) (h2 ;k2 ) (h3 ;k3 )
=
k1 +k
2 +k3

C(h1 ,h2 ,h3 ;k)
j1 +j2 +j3 ,k1 +k2 +k3 k
j1 =0 j2 =0 j3 =0
k=r1
k3
k2
k1

h h h h h h h h h
1
(h1 )j1 (h2 )j2 (h3 )j3 z123 1 2 z132 1 3 z231 2 3 .
j1 !j2 !j3 !
(3.11)
The corresponding formula for the two-point function can be rewritten in the same manner
involving derivatives with respect to the conformal weight,
1 +k2
k
1
h1 ,h2 D(h1 ,h2 ;k1 +k2 k) (h2 )k (z12 )2h2 ,
(h1 ;k1 ) (h2 ;k2 ) =
k!
(3.12)
k=r1
which evaluates to exactly the form given in (3.9). Note that again the yet free structure
constants depend only on the total J-level. This agrees with what one might expect from the
total symmetry of the three-point structure constants under permutations. Differentiation
with respect to the conformal weights reproduces precisely the logarithmic contributions
to satisfy the inhomogeneous Ward identities.
These expressions can be made even more suggestive, if one treats the structure
constants as (analytic) functions of the conformal weights [67]. This is actually true in the
case of minimal models, where all structure constants can be given explicitly as functions
of the charges within a free field representations, and hence in terms of the conformal
525
weights. Putting simply

C(h1 ,h2 ,h3 ;k+r1) =
1
k!
(h1 )i1 (h2 )i2 (h2 )i2 Ch1 ,h2 ,h3 ,
(3.13)
i1 +i2 +i3 =k(r1)
allows to rewrite (3.11) entirely in terms of derivatives with respect to the conformal
weights. Here, C(h1 ,h2 ,h3 ;r1) is then the pure, not differentiated, structure constant.
4. Operator product expansions

With the complete set of two- and three-point functions at hand, we can now proceed
to determine the operator product expansions in their generic form. To do this, we
first consider the asymptotic limit limz1 z2 G(3)
k1 ,k2 ,k3 (z1 , z2 , z3 ) and define the matrices
(3)
(Gk1 )k2 ,k3 in this limit. This essentially amounts to replacing z13 by z23 . Next, we take
(2)
the two-point functions G(2)
k1 ,k2 (z2 , z3 ), collect them into a matrix (G )k1 ,k2 and invert the
latter to obtain (G(2))1 ,2 . Finally, the matrix product

(h ;k )
C(h13,h23 ;k1 +k2 ) = G(3)
k1 k
2 ,k
G(2)
k,k3
(4.1)
yields matrices (C(h31 ;k1 ),h2 )k32 encoding all the OPEs of the field (h1 ;k1 ) (z) with fields
of arbitrary J-level. An immediate consequence of (4.1) is now that associativity of the
operator algebra can only hold if the rank of all Jordan cells is equal. Indeed, assuming
the contrary, the matrices (G(3)
k )lm were not always square matrices, and the rank of the
matrices (G(2))kl would depend on the conformal weight. It is now easy to see that the
associativity conditions such as crossing symmetry

(h;k)
(h ;k )
C(h
C
= C(h
C

i ,hj ;ki +kj ) (h,hl ,hm ;k+kl +km )
i ,hl ;ki +kl ) (h ,hj ,hm ;k +kj +km )
(4.2)
cannot any longer hold, since the matrices on both sides of the equation were not always of
equal rank. In effect, associativity can only be kept if the ranks of the Jordan cells appearing
implicitly on both sides of the equation can consistently be restricted to the minimal rank
of the product matrices. This minimal rank will automatically define the maximal rank of
the LCFT under consideration. This justifies our earlier restriction.
To see, how this formula works, we will give a more explicit version of (4.1).
(2)
Let us denote the complete set of two-point functions as
, k = G,k (z2 , z3 ) =
(h;) (z2 )(h;k)(z3 ) and correspondingly the three-point functions as

, k1 , k2 =
limz1 z2 G(3)
k1 ,k2 , (z1 , z2 , z3 ) = limz1 z2
(h1 ;k1 ) (z1 )(h2 ;k2 ) (z2 )(h;) (z3 ), all essentially
given by formulae (3.9) and (3.11). The reader should not confuse this notation with the
notation for the Shapovalov form introduced earlier. Then, the OPEs take the structure
1

r1 r1

h1 ;k1 (z1 )h2 ;k2 (z2 ) =
i, r 1 i
h k=0
i=0
526

0, 0

..

.

, 0
..

r .1, 0

...
..
.
...
..
.
...
0, k 1
0, k1 , k2
0, k + 1
..
..
..
.
.
.
, k + 1
, k 1
, k1 , k2
..
..
..
.
.
.
r 1, k 1
r s, k1 , k2
r 1, k + 1
...
..
.
...
..
.
...

r 1, r 1
0, r 1
..
.
, r 1
(h;k)(z2 ),
(4.3)
which in passing also proves that the matrix of two-point functions can be inverted without
problems. Of course, the denominator is written here in a particularly symmetric way, it
equals
j, r 1 j r for any 0 j r 1. Note that the only non-zero entries above the
anti-diagonal stem from the inserted column of three-point functions. The formula (4.1)
or (4.3), respectively, are the sought after generalization of (1.3) to the case of logarithmic
CFTs.
With this result, we obtain in the simplest r = 2 case the well-known OPEs
(h1 ;0) (z)(h2 ;0) (0) =
C(h
1 ,h2 ,h;1)
D(h,h;1)
(h;0)(0)zhh1 h2 ,
(4.4)
(h1 ;0) (z)(h2 ;1) (0)

C(h ,h ,h;1)
1 2
=
(h;1)(0)
D(h,h;1)
h

D(h,h;1) C(h1 ,h2 ,h;2) D(h,h;2) C(h1 ,h2 ,h;1)
+
(h;0)(0) zhh1 h2 ,
2
D(h,h;1)

C(h ,h ,h;2) 2C(h ,h ,h;1)
1 2
1 2
(h1 ;1) (z)(h2 ;1) (0) =
log(z) (h;1) (0)

D(h,h;1)
D(h,h;1)
h

D(h,h;1)C(h1 ,h2 ,h;3) D(h,h;2) C(h1 ,h2 ,h;2)
+
2
D(h,h;1)
2D(h,h;2) C(h1 ,h2 ,h;1) D(h,h;1) C(h1 ,h2 ,h;2)
log(z)
2
D(h,h;1)

D(h,h;1) C(h1 ,h2 ,h;1)
2
log
(z)
(0)
zhh1 h2 .
(h;0)
2
D(h,h;1)
(4.5)
(4.6)
Note that, for instance, the OPE of a proper primary with its logarithmic partner necessarily
receives two contributions. One might naively have expected that proper primary fields
do not change the J-level, although already the OPE of the stress-energy tensor with a
logarithmic field will have an additional term involving the primary field. At the end of
this section we will give a complete non-trivial example, namely, the full set of generic
OPE forms for a LCFT with rank four Jordan cells.
But before doing so, we want to remark on the question of locality. The two- and
three-point functions and the OPEs can easily be brought into a form for a local LCFT
constructed out of left- and right-chiral half. The rule for this is simply to replace each
log(zij ) by log |zij |2 , and to replace each power (zij )ij by |zij |2ij . This yields a LCFT
527
where all fields have the same holomorphic and anti-holomorphic scaling dimensions and
the same J-level. Such an ansatz automatically satisfies both, the holomorphic as well as the
anti-holomorphic Ward identities, if z and z are formally treated as independent variables.
It is important to note, however, that the resulting full amplitudes do not factorize into
holomorphic and anti-holomorphic parts. This is a well-known feature of LCFTs. For
example, the OPE equation (4.6) would read in its full form
(h1 ;1) (z, z )(h2 ;1) (0, 0)

C(2) 2C(1) log |z|2
=
|z|2(hh1 h2 )
(h;1) (0, 0)
D(1)
h

D(1) C(3) D(2) C(2)
+
2
D(1)

2D(2) C(1) D(1) C(2)
D(1) C(1)
2
2
2
+
log |z|
log |z| (h;0) (0, 0)
2
2
D(1)
D(1)
(4.7)
with an obvious abbreviation for the structure constants. The reader is encouraged to
convince herself of both, that on one hand this does indeed not factorize into holomorphic
and anti-holomorphic parts, but that on the other hand this does satisfy the full set of
conformal Ward identities.
4.1. A non-trivial example
Now, we wish to present a fully worked out non-trivial example in order to demonstrate
that even the generic structure of OPEs in arbitrary rank LCFTs is indeed more complicated
than naively thought. Therefore, we present the OPEs for a rank four LCFT. Although all
explicitly known LCFTs such as c = 2 and all the other cp,1 models [16,27], or certain
non-trivial c = 0 models [29,30] are only rank two LCFTs, there are many indications
that higher rank LCFTs exist. For instance, null-vectors for higher rank LCFTs have been
noted in the latter two references in [16], and general considerations on higher rank LCFTs
have been made in [23,26]. Examples were found in the first work in [9] and in [56].
As a rule of thumb, one can reasonably conjecture that a CFT with a degenerate vacuum
structure due to the existence of non-trivial zero-modes can be extended to a logarithmic
CFT, whose maximal J-level (i.e., r 1) precisely equals the number of zero-modes (or
of anti-commuting pairs of zero modes). Again, c = 2 is here the prime example, since
the well-known bc system of conformal spins 1 and 0 can indeed be expressed in terms
of the + system briefly mentioned in Section 2.1. The c = 2 ghost system has one
crucial zero-mode such that
0|0 = 0,
0|c0 |0 = 0. With the identification c(z) = + (z),
b(z) = (z), the + system on one hand reproduces as a subset all the correlators of
the bc system when evaluated sandwiched between
| and |0, and on the other hand
constitutes an enlarged CFT which contains logarithmic fields. As discussed above, this
CFT is logarithmic of rank two, as we would expect from the number of zero-modes.
Work in this direction will appear elsewhere [20].
In order to keep the formulae readable, we will skip all the factors (z12 )hh1 h2 as
well as all arguments of the fields. Moreover, we only denote the J-levels in the structure
528
constants omitting all references to the conformal weights. Hence, we put Ck C(h1 ,h2 ,h;k)
and Dk D(h2 ,h;k) . Furthermore, k is a shorthand for logk (z12 ).
(h1 ;0)(h2 ;0) = h
C3
(h;0),
D3
(h1 ;0)(h2 ;1) = h
C3
C4 D3 C3 D4
(h;1) +
(h;0) ,
D3
D32
(h1 ;0)(h2 ;2) = h
C3
C4 D3 C3 D4
(h;2) +
(h;1)
D3
D32
C5 D32 C4 D4 D3 + C3 (D42 D5 D3 )
D33
(h1 ;0)(h2 ;3) = h

+
+
(h;0) ,
C3
C4 D3 C3 D4
(h;3) +
(h;2)
D3
D32
C5 D32 C4 D4 D3 + C3 (D42 D5 D3 )
D33
(h;1)
C6 D33 C5 D4 D32 + C4 (D42 D5 D3 )D3 C3 (D6 D32 2D5 D4 D3 + D43 )

D34
(h1 ;1)(h2 ;1) = h

C3
C4 D3 C3 D4
C3
(h;2) +

(h;1)
D3
D3
D32
(h;0),

C4 D3 C3 D4
C3 2
(h;0),

2D3
D32
D33

C3
C4 D3 C3 D4
C3
(h1 ;1)(h2 ;2) = h
(h;3) +
(h;2)
D3
D3
D32

C5 D32 C4 D4 D3 + C3 (D42 D5 D3 ) C4 D3 C3 D4
+

(h;1)
D32
D33

+
D34

C5 D32 C4 D4 D3 + C3 (D42 D5 D3 )
C3 3

+

(h;0) ,
6D3
D33

2C3
C4
(h1 ;1)(h2 ;3) = h
(h;3)
D3
D3

C5 D3 C4 D4 C4 D3 2C3 D4
C3 2
+

(h;2)
2D3
D32
D32

C5 D32 C4 D4 D3 + C3 (D42 D5 D3 )
C6 D32 C5 D4 D3 + C4 (D42 D5 D3 )
D33
529
C5 D32 C4 D4 D3 + 2C3 (D42 D5 D3 )

D33

C3 D4 2
C3 3
+

(h;1)
6D3
2D32

D34
C6 D33 C5 D4 D32 + C4 (D42 D5 D3 ) 2C3 (D6 D32 2D5 D4 D3 + D43 )

D34

C3 D4 3
C3 4
+

(h;0),
12D3
6D32
2D33

2C3
C4
(h1 ;2)(h2 ;2) = h
(h;3)
D3
D3

C5 D3 C4 D4 C4 D3 2C3 D4
+
(h;2)
D32
D32
C3 (D42 D5 D3 )
C6 D32 C5 D4 D3 + C4 (D42 D5 D3 )
D33
C5 D32 C4 D4 D3 + 2C3 (D42 D5 D3 )

D33
+

C4 2
(h;1)
2D3

D34
C6 D33 C5 D4 D32 + C4 (D42 D5 D3 ) 2C3 (D6 D32 2D5 D4 D3 + D43 )

D34

C4 3
C3 4
C5 D3 C4 D4 2

+

(h;0),
6D3
12D3
2D32

C5
2C4
2C3 2
+
(h;3)
(h1 ;2)(h2 ;3) = h
D3
D3
D3

C6 D3 C5 D4 C5 D3 2C4 D4
2C3 D4 2 5C3 3
+

+

(h;2)
6D3
D32
D32
D32
+
C7 D32 C6 D4 D3 + D5 (D42 D5 D3 )
D33
+
C5 D32 + 4C3 (D42 D5 D3 )

2D33
C6 D32 C5 D4 D3 + 2C4 (D42 D5 D3 )

D33

C4 D3 + 5C3 D4 3
C3 4

+
(h;1)
4D3
6D32
2
530

D34
C7 D33 C6 D4 D32 + C5 (D42 D5 D3 )D3 2C4 (D6 D32 2D5 D4 D3 + D43 )

D34
C6 D33 C5 D4 D32 4C3 (D6 D32 2D5 D4 D3 + D43 )
2D34
2

C4 D3 + 3C3 D4 4
C3 5

+

(h;0),
12D3
12D32
6D33

C6
2C5
2C4 2 4C3 3
+

(h;3)
(h1 ;3)(h2 ;3) = h
D3
D3
D3
3D3

C7 D3 C6 D4 C6 D3 2C5 D4
2C4 D4 2
+

D32
D32
D32

2C4 D3 4C3 D4 3 3C3 4
+

(h;2)
4D3
3D32
+
3
C8 D32 C7 D4 D3 + C6 (D42 D5 D3 )
D33
C4 D4 D3 + 5C3 (D42 D5 D3 )
C7 D32 C6 D4 D3 + 2C5 (D42 D5 D3 )

D33
+
C5 D32 + 2C4 D4 D3 + 4C3 (D42 D3 D5 )

3D33
C6 D32 + 4C4 (D42 D5 D3 )

2D33
2
3

C4 D3 + 3C3 D4 4
C3 5
(h;1)

4D3
4D32

D34
C8 D32 C7 D4 D32 + C6 (D42 D5 D3 )D3 2C5 (D6 D32 2D5 D4 D3 + D43 )

D34
C7 D33 C6 D4 D32 4C4 (D6 D32 2D5 D4 D3 + D43 )
2D34
2
C6 D33 2C5 D4 D32 4C4 (D42 D5 D3 ) 8C3 (D6 D32 2D5 D4 D3 + D43 )
6D34
C5 D32 + 3C4 D4 D3 9C3 (D42 D5 D3 )
12D33

C4 D3 + 3C3 D4 5
5C3 6
+

(h;0).
72D3
12D32
4
3
531
Although this example seems tedious and lengthy, it is worth mentioning that it yields some
surprises. For instance, the careful reader will note that the OPE (h1 ;1) (z)(h2 ;2)(w) does
not contain a term proportional to log(z12 )2 . Of course, it is clear from general arguments
that this particular OPE may contain terms proportional to log(z12 )k for k = 0, 1, 2, 3,
where 3 is the total J-level involved. The fact that the square term is missing is due to the
general structure of the OPEs as required by global conformal covariance.
It is illuminating to check the following: the three-point functions can be viewed as
polynomials in the three variables = log(z12 ), = log(z23 ) and = log(z13 ). The twopoint functions can then be seen as polynomials
in the one variable = log(z23 ). Here,

we again skip the trivial dependency on i<j zijij factors. The interesting fact, which also
(3)
provides an excellent consistency check, is that the matrix product of Gk1 ,k2 ,k (z1 , z2 , z3 )
with the inverse of the matrix G(2)

k,k3 (z2 , z3 ) yields sums of products of polynomials,
(3)
namely, (Gk1 [, , ])k2 ,k (G(2)[])k,k3 = (Ck13+k2 [, , ]), which always reduce for
= to polynomials of the one variable only. To be more specific, the full set of twoand three-point functions as derived in this paper is, indeed, consistent with the above given
OPE formula. In fact, since we have for the two-point functions
k +k

2
3

(2)m m (h2 +h3 )
(2)
e
D(h2 ,h3 ;k2 +k3 m)
,
Gk2 ,k3 [] = h2 ,h3
(4.8)
m!
m=0
and the three-point functions read

G(3)
k1 ,k2 ,k3 [, , ]
=
k1 +k
2 +k3

m=r1
C(h1 ,h2 ,h3 ;m)
k3
k2
k1

j1 +j2 +j3 ,k1 +k2 +k3 m
j1 =0 j2 =0 j3 =0

1
(h1 )j1 (h2 )j2 (h3 )j3 e(h3 h1 h2 ) e(h2 h1 h3 ) e(h1 h2 h3 ) , (4.9)
j1 !j2 !j3 !
the matrix product for the computation of the OPE functions,

1
(3)
Gk1 [, , ] G(2) []
= Ck1 []e(h3h1 h2 ) ,
=
(4.10)
yields a structure matrix with polynomial coefficients solely in the variable = log(z12 ).
In our example, we may for instance look at (h1 ;3) (h2 ;3) and there at the factor in
front of the leading term (h;3) on the right-hand side. This factor results from the sum of
appropriate products of the following expressions (notation as above):

4
(h1 ;3) (h2 ;3) (h;0) = C6 2C5 + 2C4 2 C3 3 ,
3

1
(h1 ;3) (h2 ;3) (h;1) = C7 C6 ( + ) C3 ( )4 C6 2C5 ( + )
12

2
3
1
2
C3 ( ) + 2C4 ( + ) 2 + C4 + 2C3 ( + ) 3 C3 4 ,
2
3
4

1
1
(h1 ;3) (h2 ;3) (h;2) = C8 C7 ( + ) + C6 ( + )2 C4 ( )4
2
12
532

1
1
C3 ( + )( )4 C7 C6 ( + ) + C5 ( + )2 + C3 ( )4
12
4

2

1
+ C6 + C4 + 2 + 6 + C3 ( + )( )2 2
2

1
1
C5 + 2C4 ( + ) + C3 52 + 52 + 6 3
3
2
4 1
1
+ C4 + 3C3 ( + ) C3 5 ,
4
4

1
1
(h1 ;3) (h2 ;3) (h;3) = C9 C8 ( + ) + C7 ( + )2 C6 ( + )3
2
6

1
1
1 2
4
4
+ C5 ( ) + C4 ( + )( ) C3 5 + 52 + 2 ( + )4
12
12
72

1
1
2
C8 C7 ( + ) + C6 ( + ) C5 ( + )3
2
3

1
1
4
4
+ C4 ( ) C3 ( + )( )
4
4

1
1
+
C7 C6 ( + ) C5 ( )2 + C4 2 + 2 10 ( + )
2
3

1 2
C3 3 + 32 + 2 ( )2 2
4

1
3C6 6C5 ( + ) 3C4 2 + 2 + 6
18

C3 72 + 72 + 2 ( + ) 3

1 1
3
C5 + C4 ( + ) + C3 ( + )2 4
4 3
2
5
1
5
C4 + 3C3 ( + ) C3 6
+
12
72
for the three-point functions. Note that these expressions are all symmetric under the
exchange as they should be. One needs also the two-point functions for which
we have

(h;0)(h;3) = D3 ,

(h;1)(h;3) = D4 2D3 ,

(h;2)(h;3) = D5 2D4 + 2D3 2 ,

4
(h;3)(h;3) = D6 2D5 + 2D4 2 D3 3 .
3
+
5. Non-proper primary fields

We now turn to the next complicated case, where we explicitly allow that the OPE
of two primary fields might yield a logarithmic partner field on the right-hand side. It is
well-known that such primary fields exist, in particular the so-called pre-logarithmic fields.
533
In the c = 2 theory, the field with conformal weight h = 1/8 is such a field, since
2 log(z w)I).
it is a Virasoro primary field with OPE (z)(w) = (z w)1/4 (I(w)
However, the primary field is not itself part of a Jordan cell. The c = 2 theory provides
namely, the so-called
another example of primary fields whose OPE also yields the field I,
fermionic fields . With our notation introduced in Section 2.1, the former pre-logarithmic
fields are twist fields, i.e., fields with non-trivial boundary conditions. Such fields do not
have a zero mode content. The latter fields, however, have a zero mode content with the
property that Z+ ( ) = Z ( ), since Z ( ) = . These fields are again not members
of Jordan cells.
Of course, one might imagine a situation where non-proper primary fields do form part
of a Jordan cell. The problem then is, that it is no longer possible to solve the conformal
Ward identities in a hierarchical manner, as in Section 3, without further knowledge about
the operator algebra. Basically, our approach in Section 3 was to find the first non-vanishing
two- and three-point functions, the ones with minimal zero mode content, and to derive
correlators with higher zero-mode content by solving the inhomogeneous Ward identities
step by step. If the zero mode content of all fields is known, we can estimate which twoand three-point functions might be non-zero, since the OPE must satisfy the bound
Z0 (3 ) Z0 (1 ) + Z0 (2 )
(5.1)
for all fields 3 (w) in the operator product of 1 (z) with 2 (w). This would provide us
with a starting point for the hierarchical solution scheme. Unfortunately, this does not work
for the twist fields, because no zero mode content can be defined for them. There is so far
no example known where twist fields form part of a Jordan cell. We do not see an easy way
to extend our description of Jordan cells in terms of zero mode content of fields such that it
would encompass twist fields within Jordan cells. Thus we leave investigation of this case
for future work.
5.1. Fermionic fields
Let us now concentrate on the best known case of a rank r = 2 LCFT, i.e., where
the maximal rank of Jordan cells is two, as in the prime example of the c = 2 theory.
There, logarithmic operators, which together with their proper primary partners span the
= (0;1)(z). As long as no twist fields are
Jordan cells, are created by the operator I(z)
considered, we can construct all fields in terms of the pair of anti-commuting scalar
fields defined in Section 2.1. Remember that the modes become the creation operators
for logarithmic states. It is easy to see that proper primary fields do not possess any of
the zero modes, while logarithmic fields possess precisely the zero mode contribution
1
2 . Since the modes do anti-commute, we call fields with just one zero mode
fermionic, and fields which are ratic in bosonic. This coincides with the fact that for
c = 2 all logarithmic fields and all proper primary fields have integer conformal weights.
However, nothing prohibits us from considering the fields (z) themselves which also
have zero conformal weight, but are fermionic. Many of the above arguments remain valid
when we consider correlation functions involving fields. A further restriction is that the
total number of fields must be even, since otherwise the correlation function vanishes
identically. The reason is that consistency with the anti-commutation relations enforces to
534
put
+ =
= 0. Only when the total number of fields is even, do we have a chance
that a term + will survive after contraction. Moreover, the number of + and fields
must be equal, since otherwise ,0 zero modes will survive.
More generally, a correlator of fields
1 (z1 ) n (zn ) can only be non-zero, if it
satisfies the conditions

(5.2)
Z+ (i ) =
Z (i ) and
Z0 (i ) 2
i
in the rank two case. This statement is valid for fields i which are either proper primaries,
logarithmic partners or fermionic fields (i.e., fields with fermionic zero mode content).
Correlation functions involving fermionic fields can be computed along the same lines
as set out above. The only difference is that the action of the Virasoro algebra on fermionic
fields does not have an off-diagonal part in the rank two case. More precisely, this is true
for the L0 mode of the Virasoro algebra, since this mode reduces always the total zero
mode content symmetrically, i.e., it reduces Z by the same amount as Z+ . Other modes,
such as L1 , can reduce the zero mode content unevenly, as we will see in the next section.
As long as we still assume that all fields in a Jordan cell are quasi-primary, we do not have
to worry about this possibility. Of course, in LCFTs of higher rank, fermionic fields can
easily admit an off-diagonal action of L0 , since Z can both be larger than one. To keep
the following formulae reasonable simple, we will not consider this case here.
The important fact is that the OPE of two fermionic fields produces a logarithm, i.e.,

+ (1 + log z)I(0) .
(z) (0) = I(0)
(5.3)
This follows on general grounds, since
(z) (w) = such that a three-point function
of two fermionic and one logarithmic field necessarily involves a logarithm. The argument
remains valid in the general rank two case and fields of arbitrary scaling dimension. Each
Jordan cell is extended by two fermionic sectors such that we have the four fields (h;0) ,
(h;1) , and (h;) . It is then an easy task to compute all their OPEs from the two- and
three-point functions

(h;+) (z1 )(h;) (z2 ) = + D(h,h;) (z12 )2h ,

(h;0)(z1 )(h;1)(z2 ) = D(h,h;1) (z12 )2h ,

(h;1)(z1 )(h;1)(z2 ) = D(h,h;2) 2D(h,h;1) log z12 (z12 )2h ,

(h1 ;0)(z1 )(h2 ;0) (z2 )(h3 ;1) (z3 )
= C(h1 ,h2 ,h3 ;1) (z12 )h3 h1 h2 (z13 )h2 h1 h3 (z23 )h1 h2 h3 ,

(h1 ;0)(z1 )(h2 ;+) (z2 )(h3 ;) (z3 )
= + C(h1 ;0)(h2 ,h3 ;) (zij )hk hi hj ,

(h1 ;0)(z1 )(h2 ;1) (z2 )(h3 ;1) (z3 )

= C(h1 ,h2 ,h3 ;2) 2C(h1 ,h2 ,h3 ;1) log z23 (zij )hk hi hj ,

(h1 ;1)(z1 )(h2 ;+) (z2 )(h3 ;) (z3 )

= + C(h1 ;0)(h2,h3 ;) (log z23 log z12 log z13 ) + C(h1 ;1)(h2,h3 ;)
535
(zij )hk hi hj ,

(h1 ;1)(z1 )(h2 ;1) (z2 )(h3 ;1) (z3 )

= C(h1 ,h2 ,h3 ;3) C(h1 ,h2 ,h3 ;2) (log z12 + log z13 + log z23 )

+ 2C(h1 ,h2 ,h3 ;1) log z12 log z13 + log z12 log z23 + log z13 log z23

1
1
1
log2 z12 log2 z13 log2 z23 (zij )hk hi hj ,
2
2
2
and permutations. Note that we have explicitly indicated the anti-symmetry under
exchanging the order of the fermionic fields. These results agree in the special case where
all hi = 0 with the explicit calculations for the c = 2 LCFT by Kausch [34]. The singular
terms of the corresponding additional OPEs read
(h1 ;0) (z1 )(h2 ;) (z2 ) =
C(h
1 ;0)(h2 ,h;)
D(h2 ,h;)
(z12 )hh1 h2 (h;) (z2 ),
(h1 ;) (z1 )(h2 ;) (z2 )

C(h1 ,h2 ;)(h;0)D(h2 ,h;2) C(h1 ,h2 ;)(h;1)D(h2 ,h;1)
=
2
D(h
h
2 ,h;1)

C(h1 ,h2 ;)(h;0)
log(z12 ) (h;0) (z2 )

D(h2 ,h;1)

C(h1 ,h2 ;)(h;0)
(h;1) (z2 ) (z12 )hh1 h2 ,
+
D(h2 ,h;1)
(h1 ;1) (z1 )(h2 ;) (z2 )
C(h ;1)(h ,h;) C(h ;0)(h ,h,) log(z12 )
1
2
1
2
(z12 )hh1 h2 (h;) (z2 ).
=
D(h2 ,h;)
(5.4)
The above statement shows that rank two LCFTs naturally allow for fermionic fields. It
has been suggested in [61] to formally collect these rumples of fields in superfields
h (z, + , ) of N = 2 Grassmann variables such that

h z, + , = (h;0)(z) + + (h;) (z) + (h;+) (z) + + (h;1) (z), (5.5)
which in the c = 2 case resembles the zero mode contributions. It is tempting to
conjecture that a rank k LCFT will naturally incorporate the analog of anti-commuting
scalars for Zk para-fermions, whose OPEs among them create logarithmic fields of
according J-levels. However, an investigation of this is beyond the scope of the present
paper.
5.2. Twist fields
As already explained, there is (at least) one more sort of fields which may occur in
LCFTs. In the standard c = 2 example, the two fields (z) and (z) with conformal
536
weights h = 1/8 and h = 3/8, respectively, are not yet accounted for. These fields
are twist fields. They can be treated much along the same lines as fermionic fields. The
difference is that their mode expansion is in Z + with a certain rational depending on
the boundary conditions and the ramification number of the twists. The fields and are
Z2 twists. Despite the difference in the mode expansion, twist fields behave quite similar
to the (para-)fermionic fields mentioned above. In particular, their two-point functions are
non-zero if and only if they involve a twist and its anti-twist , which resembles the
fact that for fermionic fields only the two-point function of two different fermions is nonzero. Higher twist fields are then analogous to para-fermions.
One might attempt to extend the definition of zero mode content to rational numbers
such that a twist field with mode expansion in Z + would get assigned Z0 ( ) =
Z+ ( ) + Z ( ) = + . A necessary condition for a correlator
1 . . . n to be nonzero would then read

(5.6)
Z+ (i ) Z and
Z (i ) Z.
i
However, such an assignment is obviously only determined modulo integers, and it is not
a priory clear how to implement the condition for a minimal zero mode content in general.
At least, one should now always consider separately the zero mode content Z+ and Z ,
as indicated above. In the rank two case, however, we can incorporate the condition for a
minimal zero mode content into (5.6) by simply replacing Z by N. Indeed, the Z2 twist
fields and have twist numbers (, ) = (1/2, 1/2) and (1/2, 3/2), respectively, from
which we immediately can read off, which two- and three-point functions of and fields
can be non-zero.
To emphasize the common features of fermionic and twist fields, we contrast their
possible two- and three-point functions with the ones for fermionic fields (there are no
non-vanishing two- or three-point functions involving both, fermionic and twist fields,
simultaneously). The notation means the anti-twist 1 with respect to , and one always
has h = h . The only non-trivial two-point function then reads

(z1 ) (z2 ) = D (z12 )2h ,
(5.7)
with D = D . Note that in contrast to the fermionic fields, twist fields are symmetric.
The three-point functions are easily computed and the results are
(h1 ;0) (h2 ;k ) 3 = 0,
(h1 ;0) 2 3 = 3 ,2 C(h1 ,0)2 3 (zij )hk hi hj ,

(h1 ;1) 2 3 = 3 ,2 C(h1 ,1)2 3 + C(h1 ,0)2 3 (log z23 log z12 log z13 )
(zij )hk hi hj ,
1 2 3 = (3 ,1 +2 + 3 ,11 2 )C1 2 3 (zij )hk hi hj ,
(h1 ;1) (h2 ;1) 3 = C(h1 ,h2 ;2)3 (zij )hk hi hj .

Note that some of the introduced constants may be zero, e.g., C1 2 3 = 0 whenever
the three twists do not add up to an integer. Most remarkably is perhaps the fact that
537
(h1 ;1) (h2 ;1)3 might be non-zero. This does not happen in the c = 2 theory, since it
implies that the OPE of two logarithmic fields has a contribution
(h;1) (z)(h ;1)(0) = +
C(h,h ;2) h hh

z
(0) + ,
D
(5.8)
which is not the case in the c = 2 theory. However, already the next theory in the cp,1
series of LCFTs, namely the c3,1 = 7 model, shows precisely this feature, where the
fusion rule of the h = 0 logarithmic field with itself involves the twist field with h = 1/3
on the right-hand side. Since the main focus of this paper lies on logarithmic fields, we will
not go into further detail here. The OPEs involving fields read correspondingly
(h;0) (z) (0) =
C(h;0) h h h
z
(0),
D
(h;1) (z) (0)

C(h,h ;2) h hh
C(h;1) C(h;0) log z h h h
=
z
(h ;0) +
z
(0)
D(h;1)
D
C(h;1) h h h
z
(0),
+
D
C(h;1) h h h
C h h h

(0),
(z) (0) =
z
z
(h;0) (0) +
D(h;1)
D
(z) (0)
C(h;1) D(h;1) C(h;0) D(h;2) + C(h;0) D(h;1) log z hh h
=
z
(h;0)(0)
2
D(h;1)
+
C(h;0) hh h
z
(h;1)(0),
D(h;1)
where in the last two equations = . As remarked above, some of the structure constants
may vanish, as they do in the c = 2 LCFT. One sees that even the simple rank two case
gets quite complicated and needs a cumbersome notation. The situation is slightly better
in the particular case for the c = 2 theory where all amplitudes involving up to four twist
fields as well as amplitudes with an arbitrary number of fermionic fields were computed
in [34].
6. Non-quasi-primary fields
Our discussion of correlation functions and operator product expansions in logarithmic
CFTs heavily relies on the following assumption which we so far have made: that all
logarithmic partner fields within a Jordan cell be quasi-primary. This means in particular
that L1 |h; k = L1 (h;k) (0)|0 = 0. As a consequence, we could make elaborate use of
the Ward identities of global conformal transformations in the form (3.2). This section is
devoted to the question under which more relaxed circumstances our results still hold.
It is by no means clear that logarithmic partner fields are, indeed, all quasi-primary.
On the contrary, even the simplest known LCFT, the c = 2 model, features a Jordan
538
cell where the logarithmic partner is not quasi-primary [21]. Actually, as already outlined
in Section 2, in this model exists a Jordan cell for conformal weight h = 1, built from
a primary field1 (z) and its logarithmic partner field :+ :(z). Here, we again
used the realization of the c = 2 theory in terms of two anti-commuting scalar fields
(z) along the lines (2.10) and (2.11). It is easy to see that |1; 0 = ,1 |0 and that
|1; 1 = ,1 (+ + 1)|0. It follows that L1 |1; 0 = 0, while L1 |1; 1 = |0 = 0,
where one uses that the stress-energy tensor is given by (2.9). We remind the reader that
is one of the two zero-modes of the field (z). Hence, |1; 1 is not a state corresponding
to a quasi-primary field.
The point is that it does not matter. The global Ward identities are not affected by this
non-zero term. More generally, all correlation functions involving the field (h=1;1)(z), the
logarithmic partner of the primary h = 1 field (h=1;0)(z) = (z), behave exactly as if the
field were quasi-primary. The reason for this is simply that the state L1 |1; 1 is fermionic
with respect to the number of zero modes. More precisely, it has Z+ = 1, Z = 0 (or
vice versa). Any correlation function can only be non-zero if the numbers (Z+ , Z ) of
zero modes are, after all contractions are done, exactly (1, 1). Hence, any correlation
function involving (h=1;1)(z), which has a chance to be non-zero, must initially have
fermion numbers (Z+ , Z ) with Z+ + Z even and Z+ 1, Z 1. Applying L1 to it
results in the global conformal Ward identity up to additional terms with fermion numbers
(Z+ 1, Z ) or (Z+ , Z 1). Since Z+ + Z 1 is then necessarily an odd number, the
additional terms must vanish. It follows that the fact that (h=1;1)(z) is not quasi-primary
does not influence the correlation functions, because the spoiling term does not lead to any
non-vanishing contributions. Note that this statement is only true as long as we consider
the effect within correlation functions. The deeper reason is that the action of the Virasoro
algebra changes the fermion numbers unevenly in this case.
Of course, not all correlation functions involving a field (z) corresponding to the
state |0 automatically vanish. For example,
0|+ (z) (w)|0 = 0. What is meant in
the above discussion is that all correlation functions vanish, which result from applying L1
to a correlator involving (h=1;1)(z) and other fields such that the initial correlator might
be non-zero. Since L1 acts as a derivation, it only changes one of the inserted fields at
a time, so that starting with an admissible number of zero modes leads to terms with
non-admissible numbers of zero modes.
There are strong indications that this structure of Jordan cells with non-quasi-primary
fields is more generally true. At least, there is so far no LCFT explicitly known where
logarithmic partner fields are not quasi-primary in a way which would affect correlation
functions and, therefore, our general conclusions on their general structure. All LCFTs
which can be constructed or realized explicitly in terms of fundamental free fields, such as
the c = 2 model, receive their peculiar logarithmic fields ultimately due to the existence
of conjugate pairs of zero-modes. In the c = 2 model, these are the two pairs , ,0 ,
respectively (see Section 2.1). In this situation, we only need that the logarithmic partner
fields are quasi-primary up to terms with the wrong number of such zero modes. Here,
1 More precisely, we are dealing with a doublet of two such fields, distinguished by the -label.
539
wrong means in the above explained sense that the number of zero modes gets changed
unevenly.
Following the lines of [61], the effective zero mode content is equivalently described
in terms of nilpotent variables with which the fields spanning a Jordan cell are grouped
together in a superfield like fashion. In the rank two case this is easily accomplished by
introducing for each Jordan cell two Grassmann variables and a bosonic nilpotent
variable = + with the property 2 = 0. Then, we may group together all
fields, the primary (h;0) (z), its logarithmic partner (h;1) (z) and the two fermionic
fields (h;) (z) as h (z) = (h;0)(z) + + (h;+) (z) + (h;) (z) + + (h;1)(z).
For higher rank Jordan cells, an analogous procedure applies. For a Jordan cell of
even rank 2r one needs r 1 pairs of Grassmann variables. The member of J-level k
in the Jordan cell itself is given by the elementary completely symmetric polynomial
+
k (1 , . . . , r1 ) = k (1+ 1 , . . . , r1
r1
) of total degree 2k in the s. So the primary
field (h;0) (z) belongs to 0 1, while the top J-level field (h;r1) (z) belongs to r1
+
1+ 1+ r1
r1
. The
elementary completely symmetric polynomial k (x1 , . . . , xn )
is defined as k = i1 <i2 <<ik xi1 xi2 xik up to normalization. Other polynomials

p({i }), whose monomials m{i} = k ikk have differing partial degrees deg+ (m{i} ) =
j
deg (m{i} ) for at least one j , belong to fields which are the higher rank analogous of the
j
fermionic fields described above. Overall symmetry in the Grassmann variables demands
that all polynomials must be symmetric polynomials in the s. However, what may be
varied is how the total degree is split into + variables and variables. Hence, we
use the symmetric polynomials in two colors, kl,kl , instead. These can be written as
kl,kl ({+ , }) = max(l,kl)({})min(l,kl) ({min(l,kl) }). Note that kl,kl = 0 for l or
k l larger r 1. The Jordan cell is then spanned by the fields corresponding to the
k,k
symmetric polynomials 2k
({+ , + }) = k ({}).
For example, in the rank four case we have the following possibilities: the Jordan cell is
spanned by

h (z) = (h;0)(z) +
i+ i (h;1) (z) +
i+ i j+ j (h;2) (z)
1i<j 3
1i3
+ 1+ 1 2+ 2 3+ 3 (h;3) (z)
= 0 (h;0)(z) + 21,1 (h;1) (z) + 42,2 (h;2) (z) + 63,3 (h;3) (z).
Furthermore, we have symmetrized non-bosonic fields according to the following
diamond of two-color symmetric polynomials:
33,0
22,0
43,1
11,0
32,1
53,2
00
21,1
42,2
63,3
10,1
31,2
52,3
20,2
41,3
30,3 .
540
Of course, it is not clear whether all higher rank LCFTs fall into this pattern, but the crucial
role of conjugate zero mode pairs suggests so.
It is now easy to see, which correlation functions may be non-zero. Clearly, each
inserted field comes with an associated polynomial . Consequently, an arbitrary n-point

function can only be non-vanishing, if the product ni=1 klii ,ki li meets the conditions
n

li =
i=1
n

(ki li ) r 1.
(6.1)
i=1
Of course, this is nothing else than our initial conditions on the zero mode content, since
any field with associated polynomial kl,kl has zero mode content Z+ = l, Z = k l.
The action of symmetry generators such as the modes of the Virasoro algebra may have
off-diagonal contributions meaning that they change a field into another and thus change
a given associated polynomial into another. However, this off-diagonal term contributes
to the correlator only if the resulting product of associated polynomials again satisfies the
above condition (6.1). Now, symmetry generators always act as derivations on correlators
such that an immediate necessary condition on an off-diagonal term is that the off-diagonal
action of the generator moves vertically in the -diamond. Otherwise, the off-diagonal term
has no effect in the correlator. Coming back to our initial example in the c = 2 theory, we
have that |1; 1 has associated polynomial 21,1 , while L1 |1; 1 has associated polynomial
11,0 . Thus, L1 does not move vertically in the -diamond, and the non-quasi-primary term
hence does not contribute.
We, therefore, arrive at the following generalized picture: the action of symmetry
generators such as the Virasoro algebra in a logarithmic CFT possesses off-diagonal
additional terms with accompanying moves in the associated -diamond. However, all
such terms with a non-vertical move are irrelevant when considered within correlation
functions. All moves in the -diamond are obviously generated by the two basic moves
l,k1l
Q+ : kl,kl k1
l1,kl
and Q : kl,kl k1
.
(6.2)
The careful reader might notice that these moves always go upwards within the diamond.
However, since symmetry generators are presumably free of zero modes which do not
annihilate the vacuum (otherwise, the vacuum is no longer invariant under the considered
symmetry), we do not expect that a symmetry generator will move downwards within the
diamond. These two basic moves may be viewed as generating a BRST like symmetry,
l ,k l
since with
i kii i i = 0, we certainly have

l1 ,k1 l1

l ,k 1lj
Q+
(6.3)
klii ,ki li =
kjj 1j
klnn ,kn ln = 0,
k1
i
Actually, one may introduce operators Q

and analogously for
with 1 r 1
which act by setting formally to one by contracting with a conjugate Grassmann

variable, i.e., by acting with { , }. These latter operators have the nice property to
+
2
automatically satisfy (Q
) = 0, while Q Q = 0. These latter operators may then
indeed be considered as BRST symmetries on correlation functions. We introduce Q as
Q .
541
symbols for classes of such symmetries, denoting arbitrary moves upwards in the diamond,
+
which are not vertical. Thus, these classes contain also products such as Q+
1 Q2 , etc.
We may thus finally generalize our assumptions on the fields in LCFTs such that the
action of the Virasoro algebra only needs to satisfy our basic conditions (3.1) and (3.2) up
to terms, which can be written as Q of an admissible configuration, i.e., can be written
l ,k l
l ,k l
as Q
i kii i i with
i kii i i satisfying (6.1). If this is the case, all our results on
two- and three-point functions and on the OPE structure remain valid in precisely the form
as we have given them. All explicitly known LCFTs seem to possess only Jordan cells (and
associated non-bosonic fields) which fall into this pattern. However, we do not have a proof
that this has always to be the case. The strongest indications in favor of such a conjecture
stem from the classification of possible Jordan structure like modules in the representation
theory of the Virasoro algebra as given in [72]. A deeper study, which also should consider
Jordan cell structures with respect to other chiral symmetries has to be left for future work.
Last but not least, we remark that the zero mode content further implies (2.17) and
hence (2.16). Let us momentarily assume that these bounds would not be strict. Then, for
example, a two-point function
(h;0) (z)(h;0)(w) could be non-zero, if the OPE of the
primary field (h;0) with itself contained (0;r1) in the rank r case. It is an easy task to
check that this is, however, inconsistent, since no solution for, say,
(h;1) (z)(h;0)(w) can
be found that satisfies all three Ward identities (3.2). On the other hand, two-point functions
are insensitive to non-quasi-primary contributions, since these always would result in a
two-point function of fields of differing conformal weights. Similar inconsistencies appear
when other cases violating (2.16) are considered.
The higher rank case is more complicated, since OPEs might yield terms of too high
J-level k, which do not influence the two-point functions as long as k < r 1. One
has to consider three-point functions: assume that, e.g., the OPE (h1 ;0) (h2 ;0) produces
a term of maximal J-level k with conformal weight h3 . Consider
(h1 ;0)(h2 ,0) (h3 ,l)
with l being the minimal J-level such that this three-point function is non-zero, and thus
has the canonical form C(h1 ;0)(h2;0)(h3;l) 8(zij )hk hi hj . Then, one easily finds solutions
for
(h1 ;0)(h2 ,0) (h3 ,l+1) and
(h1 ;0) (h2 ,1) (h3 ,l) . With these solutions as inhomogeneities in the Ward identities (3.2), no consistent solution for
(h1 ;0) (h2 ,1) (h3 ,l+1)
can be found. Thus, we believe that LCFTs without a gradation in the J-level are not consistent, although we do not have a complete proof for this conjecture.
7. Conclusion
Taking into account the proper action of the Virasoro algebra on logarithmic fields,
i.e., working with Jordan cell representations as generalizations of irreducible highestweight representations [72], allows to evaluate OPEs in LCFT in a similar fashion as
in ordinary CFT. The main difference is that each n-point function represents a full
hierarchy of conformal blocks involving s = 1, . . . , n logarithmic fields with varying Jlevels ki = 0, . . . , r 1. While in ordinary conformal field theory it suffices to know
correlation functions of primary fields only (since everything else is fixed by conformal
covariance), in logarithmic CFT one needs to know the full hierarchy of r n correlations
functions of primaries and all their logarithmic partner fields. Actually, there are only
542

different such correlation functions, since the total J-level must be at least
r n n+r2
r2
r 1. If one additionally takes into account, that all correlation functions with total J-level
precisely equal to the minimal value r 1 are identical (and do not involveany logarithms),

the total number of different correlation functions reduces finally to r n n+r1
r1 + 1.
The solution of this hierarchy can be obtained step by step, where the case with one
logarithmic field only of maximal J-level is worked out in the same way as in ordinary
CFT. The same holds for correlation functions with several logarithmic fields, such that
the total J-level adds up to r 1. In each further step, the differential equations, which
result from the existence of null vectors, are inhomogeneous, with the inhomogeneity
determined by the conformal blocks of correlators with fewer logarithmic fields. Details on
the computation of four-point functions have been presented in the last reference in [16].
Of course, since the OPEs are entirely determined in terms of the two- and three-point
functions, the situation is simpler. As we have shown, these arecompletely analogous to
the case of ordinary CFTalready fixed up to constants by global conformal covariance.
The formulae (3.9) and (3.11) yield the full hierarchy of these functions in a direct way.
With the help of a full set of two- and three-point functions, OPEs are then simply
obtained through certain matrix products. Their non-trivial structure is essentially due to
the fact that the matrix of two-point functions must be inverted first. The long example
given in Section 4 demonstrates that this inversion results in many additional terms
and non-trivial linear combinations of the involved structure constants which were not
accounted for in older approaches, e.g., the one taken in [61]. In particular, it is not yet
clear, how the otherwise elegant method of [61] to write correlation functions as formal
power series expansions in nilpotent conformal weights h + , r = 0, can be transferred to
operator product expansions. Nonetheless, we believe that our discussion on the structure
matrix coefficients in terms of polynomials might be of help here.
This fills one of the few remaining gaps to put LCFT on equal footing with better known
ordinary CFTs such as minimal models. The success of conformal field theory is mainly
rooted in the fact that correlation functions can be computed effectively and exactly. The
basic tools to achieve this are operator product expansions and differential equations due to
the existence of null-vectors. Now, all these tools are also available in the logarithmic case.
It is worth mentioning the following difference between the ordinary and the logarithmic
case: in the ordinary case, any OPE hi hj one wishes to compute depends only on
two constants per primary field hk occurring on the right-hand side, namely Chi hj hk and
D hk hk . This remains true even in the case of multiplets of fields of equal conformal weight,
as long as their two-point functions can be diagonalized. If this latter situation does no
longer hold, which precisely constitutes the logarithmic CFT case [27], then things get
more complicated. Now, when computing an OPE (hi ;ki ) (hj ;kj ) , one needs for each
field from a Jordan block {(hl ;k) }k=0,...,r1 occurring on the right-hand side complete
knowledge of all structure constants C(hi ,hj ,hl ;r1+s) and D(hl ;r1+t ) for 0 s ki + kj ,
0 t min(r 1, ki + kj ). Thus, for each triplet of conformal weights (hi , hj , hl ), one
needs in fact 2r 1 three-point structure constants plus r two-point structure constants,
i.e., in total 3r 1 constants, for a rank r LCFT.
Moreover, we discussed how (para-)fermionic and twist fields can be incorporated into
our framework. A LCFT can be viewed as an extended ordinary CFT where the space
of states has additional sectors. In particular, we showed that any rank two LCFT can
543
consistently accommodate two additional fermionic sectors, and we expect that rank r
LCFTs will allow for Zr para-fermionic sectors. Furthermore, LCFTs naturally incorporate
twist fields which considerably enrich the structure of such theories. These twist fields
can be considered as the basic entities from which all other fields can be generated by
successive application of operator products. Since the OPE of a twist with its anti-twist
produces, among others, a logarithmic field, twist fields are also called pre-logarithmic
fields [46]. We expect that a more detailed and careful treatment of twist fields in higher
rank LCFTs will shed new light on the still mysterious geometric aspects underlying
logarithmic conformal field theories (see [19] for some initial remarks on this).
We also presented an BRST-like algebraic argument on the zero mode content, which
yields necessary conditions for correlators to be non-zero. This could be used to consider
Jordan cells with non-proper primary fields and non-quasi-primary logarithmic partners
in more detail. In particular, our argument provides conditions for both, the minimal zero
mode content of correlators, as well as the maximal zero mode content of the right-hand
side of OPEs, thus yielding a starting point for a generalized hierarchical scheme.
To conclude, one should remark that we have not yet discussed the most general case.
Within this paper, we assumed that the only non-vanishing one-point function is given
by the maximal J-level logarithmic partner of the identity,
(h=0;r1) . However, this
might be too restrictive. In particular, a full discussion of logarithmic CFT should include
the case of boundaries with their induced spectrum of boundary operators, which possess
non-vanishing one-point functions. It would be interesting to compute boundary-boundary
OPEs and boundary-bulk OPEs in the LCFT case along the lines of our approach.
Acknowledgement
We thank Shahin Rouhani and in particular Matthias Gaberdiel for many valuable
discussions and comments.
References
[1] D. Bernard, Z. Maassarani, P. Mathieu, Mod. Phys. Lett. A 12 (1997) 535, hep-th/9612217.
[2] M.J. Bhaseen, Nucl. Phys. B 604 (2001) 537, cond-mat/0011229.
[3] M.J. Bhaseen, J.-S. Caux, I.I. Kogan, A.M. Tsvelik, Disordered Dirac fermions: the marriage of three
different approaches, cond-mat/0012240.
[4] M.J. Bhaseen, Random Dirac fermions: the su(N ) gauge potential and ZN twists, cond-mat/0012420.
[5] A. Bilal, I.I. Kogan, Nucl. Phys. B 449 (1995) 569, hep-th/9503209.
[6] A. Campbell-Smith, N.E. Mavromatos, Phys. Lett. B 476 (2000) 149, hep-th/9908139;
A. Campbell-Smith, N.E. Mavromatos, Phys. Lett. B 488 (2000) 199, hep-th/0003262.
[7] A. Cappelli, L.S. Georgiev, I.T. Todorov, Commun. Math. Phys. 205 (1999) 657, hep-th/9810105.
[8] J. Cardy, Logarithmic correlations in quenched random magnets and polymers, cond-mat/9911024.
[9] J.-S. Caux, I.I. Kogan, A.M. Tsvelik, Nucl. Phys. B 466 (1996) 444, hep-th/9511134.
[10] J.-S. Caux, N. Taniguchi, A.M. Tsvelik, Nucl. Phys. B 525 (1998) 671, cond-mat/9801055.
[11] J.-S. Caux, I.I. Kogan, A. Lewis, A.M. Tsvelik, Nucl. Phys. B 489 (1997) 469, hep-th/9606138.
[12] J.-S. Caux, Phys. Rev. Lett. 81 (1998) 4196, cond-mat/9804133.
[13] W. Eholzer, L. Feher, A. Honecker, Nucl. Phys. B 518 (1998) 669, hep-th/9708160.
544
[14] J.R. Ellis, N.E. Mavromatos, D.V. Nanopoulos, Int. J. Mod. Phys. A 12 (1997) 2639, hep-th/9605046;
J.R. Ellis, N.E. Mavromatos, D.V. Nanopoulos, Int. J. Mod. Phys. A 13 (1998) 1059, hep-th/9609238;
J.R. Ellis, N.E. Mavromatos, D.V. Nanopoulos, Gen. Relativ. Gravit. 32 (2000) 943, gr-qc/9810086;
J.R. Ellis, N.E. Mavromatos, D.V. Nanopoulos, Gen. Relativ. Gravit. 32 (2000) 1777, gr-qc/9911055;
J.R. Ellis, N.E. Mavromatos, D.V. Nanopoulos, Phys. Rev. D 62 (2000) 084019, gr-qc/0006004.
[15] J.R. Ellis, N.E. Mavromatos, E. Winstanley, Phys. Lett. B 476 (2000) 165, hep-th/9909068.
[16] M. Flohr, Int. J. Mod. Phys. A 11 (1996) 4147, hep-th/9509166;
M. Flohr, Int. J. Mod. Phys. A 12 (1997) 1943, hep-th/9605151;
M. Flohr, Nucl. Phys. B 514 (1998) 523, hep-th/9707090;
M. Flohr, Null vectors in logarithmic conformal field theory, hep-th/0009137.
[17] M. Flohr, Mod. Phys. Lett. A 11 (1996) 55, hep-th/9605152.
[18] M. Flohr, Nucl. Phys. B 482 (1996) 567, hep-th/9606130.
[19] M. Flohr, Phys. Lett. B 444 (1998) 179, hep-th/9808169.
[20] M. Flohr, M. Krohn, in preparation.
[21] M.R. Gaberdiel, H.G. Kausch, Nucl. Phys. B 477 (1996) 293, hep-th/9604026;
M.R. Gaberdiel, H.G. Kausch, Phys. Lett. B 386 (1996) 131, hep-th/9606050;
M.R. Gaberdiel, H.G. Kausch, Nucl. Phys. B 538 (1999) 631, hep-th/9807091.
[22] M.R. Gaberdiel, Fusion rules and logarithmic representations of a WZW model at fractional level, hepth/0105046.
[23] A.M. Ghezelbash, V. Karimipour, Phys. Lett. B 402 (1997) 282, hep-th/9704082.
[24] A.M. Ghezelbash, M. Khorrami, A. Aghamohammadi, Int. J. Mod. Phys. A 14 (1999) 2581, hep-th/9807034.
[25] G. Giribet, Mod. Phys. Lett. A 16 (2001) 821, hep-th/0105248.
[26] E. Gravanis, N.E. Mavromatos, Impulse action on D-particles in RobertsonWalker spacetimes, higherorder logarithmic conformal algebras and cosmological horizons, hep-th/0106146.
[27] V. Gurarie, Nucl. Phys. B 410 (1993) 535, hep-th/9303160.
[28] V. Gurarie, M. Flohr, C. Nayak, Nucl. Phys. B 498 (1997) 513, cond-mat/9701212.
[29] V. Gurarie, Nucl. Phys. B 546 (1999) 765, cond-mat/9808063.
[30] V. Gurarie, A. Ludwig, Conformal algebras of 2D disordered systems, cond-mat/9911392.
[31] K. Ino, Phys. Rev. Lett. 81 (1998) 1078, cond-mat/980333;
K. Ino, Phys. Rev. Lett. 82 (1999) 4902, cond-mat/9812053;
K. Ino, Phys. Rev. Lett. 83 (1999) 3526;
K. Ino, Phys. Rev. Lett. 84 (2000) 201, Erratum, cond-mat/9906147;
K. Ino, Phys. Rev. Lett. 86 (2001) 882, cond-mat/0008228.
[32] Y. Ishimoto, Boundary states in boundary logarithmic CFT, hep-th/0103064.
[33] E.V. Ivashkevich, J. Phys. A 32 (1999) 1691, cond-mat/9801183.
[34] H.G. Kausch, Curiosities at c = 2, hep-th/9510149;
H.G. Kausch, Nucl. Phys. B 583 (2000) 513, hep-th/0003029.
[35] K. Kaviani, A.M. Ghezelbash, Phys. Lett. B 469 (1999) 81, hep-th/9902104.
[36] S. Kawai, J.F. Wheater, Phys. Lett. B 508 (2001) 203, hep-th/0103197.
[37] F. Kheirandish, M. Khorrami, Eur. Phys. J. C 18 (2001) 795, hep-th/0007013;
F. Kheirandish, M. Khorrami, Eur. Phys. J. C 20 (2001) 593, hep-th/0007073.
[38] M. Khorrami, A. Aghamohammadi, M.R. Rahimi Tabar, Phys. Lett. B 419 (1998) 179, hep-th/9711155.
[39] M. Khorrami, A. Aghamohammadi, A.M. Ghezelbash, Phys. Lett. B 439 (1998) 283, hep-th/9803071.
[40] J.Y. Kim, H.W. Lee, Y.S. Myung, Origin of logarithmic corrections in three-dimensional anti-de Sitter space,
hep-th/9812016.
[41] V.G. Knizhnik, Commun. Math. Phys. 112 (1987) 567.
[42] I.I. Kogan, N.E. Mavromatos, Phys. Lett. B 375 (1996) 111, hep-th/9512210.
[43] I.I. Kogan, C. Mudry, A.M. Tsvelik, Phys. Rev. Lett. 77 (1996) 707, cond-mat/9602163.
[44] I.I. Kogan, N.E. Mavromatos, J.F. Wheater, Phys. Lett. B 387 (1996) 483, hep-th/9606102.
[45] I.I. Kogan, A. Lewis, O.A. Soloviev, Int. J. Mod. Phys. A 13 (1998) 1345, hep-th/9703028.
[46] I.I. Kogan, A. Lewis, Nucl. Phys. B 509 (1998) 687, hep-th/9705240;
I.I. Kogan, A. Lewis, Phys. Lett. B 431 (1998) 77, hep-th/9802102.
[47] I.I. Kogan, Phys. Lett. B 458 (1999) 66, hep-th/9903162.
[48] I.I. Kogan, D. Polyakov, Int. J. Mod. Phys. A 16 (2001) 2559, hep-th/0012128.
545
[49] I.I. Kogan, A.M. Tsvelik, Mod. Phys. Lett. A 15 (2000) 931, hep-th/9912143.
[50] I.I. Kogan, J.F. Wheater, Phys. Lett. B 486 (2000) 353, hep-th/0003184.
[51] I.I. Kogan, A. Nichols, SU(2)0 and OSp(2|2)2 WZNW models: two current algebras, one logarithmic
CFT, hep-th/0107160.
[52] G.K. Leontaris, N.E. Mavromatos, Phys. Rev. D 61 (2000) 124004, hep-th/9912230;
G.K. Leontaris, N.E. Mavromatos, Phys. Rev. D 64 (2001) 024008, hep-th/0011102.
[53] A. Lewis, Phys. Lett. B 480 (2000) 348, hep-th/9911163.
[54] A. Lewis, Logarithmic CFT on the boundary and the world-sheet, hep-th/0009096.
[55] A.W.W. Ludwig, A free field representation of the OSP(2|2) current algebra at level k = 2, and Dirac
fermions in a random SU(2) gauge potential, cond-mat/0012189.
[56] Z. Maassarani, D. Serban, Nucl. Phys. B 489 (1997) 603, hep-th/9605062.
[57] S. Mahieu, P. Ruelle, Scaling fields in the two-dimensional abelian sandpile model, hep-th/0107150.
[58] N.E. Mavromatos, R.J. Szabo, Phys. Lett. B 430 (1998) 94, hep-th/9803092.
[59] N.E. Mavromatos, R.J. Szabo, Phys. Rev. D 59 (1999) 104018, hep-th/9808124;
N.E. Mavromatos, R.J. Szabo, D-brane dynamics and logarithmic superconformal algebras, hepth/0106259.
[60] S. Moghimi-Araghi, S. Rouhani, Lett. Math. Phys. 53 (2000) 49, hep-th/0002142.
[61] S. Moghimi-Araghi, S. Rouhani, M. Saadat, Nucl. Phys. B 599 (2001) 531, hep-th/0008165;
S. Moghimi-Araghi, S. Rouhani, M. Saadat, Current algebra associated with logarithmic conformal field
theories, hep-th/0012149.
[62] S. Moghimi-Araghi, S. Rouhani, M. Saadat, On the AdS/CFT correspondence and logarithmic operator,
hep-th/0105123.
[63] Y.S. Myung, H.W. Lee, J. High Energy Phys. 9910 (1999) 009, hep-th/9904056.
[64] A. Nichols, S. Sanjay, Nucl. Phys. B 597 (2001) 633, hep-th/0007007.
[65] A. Nichols, Logarithmic currents in the SU(2)0 WZNW model, hep-th/0102156.
[66] M.R. Rahimi-Tabar, S. Rouhani, Europhys. Lett. C 37 (1997) 447, hep-th/9606143;
M.R. Rahimi-Tabar, S. Rouhani, Logarithmic correlation functions in two-dimensional turbulence, hepth/9606154.
[67] M.R. Rahimi-Tabar, A. Aghamohammadi, M. Khorrami, Nucl. Phys. B 497 (1997) 555, hep-th/9610168.
[68] M.R. Rahimi-Tabar, S. Rouhani, Phys. Lett. B 431 (1998) 85, hep-th/9707060.
[69] M. Reza Rahimi-Tabar, Nucl. Phys. B 588 (2000) 630, cond-mat/0002309.
[70] N. Read, D. Green, Paired states of fermions in two dimensions with breaking of parity and time reversal
symmetries, and the fractional quantum Hall effect, cond-mat/9906453.
[71] N. Read, H. Saleur, Exact spectra of conformal supersymmetric nonlinear sigma models in two dimensions,
hep-th/0106124.
[72] F. Rohsiepe, On reducible but indecomposable representations of the Virasoro algebra, hep-th/9611160.
[73] L. Rozansky, H. Saleur, Nucl. Phys. B 376 (1992) 461;
L. Rozansky, H. Saleur, Nucl. Phys. B 389 (1993) 365, hep-th/9203069.
[74] H. Saleur, Nucl. Phys. B 489 (1992) 486, hep-th/9111007.
[75] S. Sanjay, Logarithmic operators in SL(2, R) WZNW model, singletons and AdS3 /(L)CFT 2 correspondence, hep-th/0011056.
[76] S. Skoulakis, S. Thomas, Phys. Lett. B 438 (1998) 301, cond-mat/9802040.

Boundary one-point functions, scattering theory and

vacuum solutions in integrable systems
V.A. Fateev a,b , E. Onofri c
a Laboratoire de Physique Mathmatique, Universit Montpellier II, Pl. E. Bataillon, 34095 Montpellier, France
b On leave of absence from Landau Institute for Theoretical Physics, ul. Kosygina 2, 117940 Moscow, Russia
c Dipartimento di Fisica, Universit di Parma, and INFN, Gruppo Collegato di Parma, 43100 Parma, Italy
Abstract
Integrable boundary Toda theories are considered. We use boundary one-point functions and
boundary scattering theory to construct the explicit solutions corresponding to classical vacuum
configurations. The boundary ground state energies are conjectured. 2002 Elsevier Science B.V.
All rights reserved.
1. Introduction
There is a large class of 2D quantum field theories (QFTs) which can be considered
as perturbed conformal field theories (CFTs). These theories are completely defined if
one specifies its CFT data and the relevant operator which plays the role of perturbation.
The CFT data contain explicit information about the ultraviolet (UV) asymptotics of the
field theory, while the long distance behavior requires careful analysis. If a perturbed CFT
contains only massive particles it is equivalent to a relativistic scattering theory and it is
completely defined by specifying the S-matrix. Contrary to CFT data, the S-matrix data
exhibit information about the long distance properties of the theory in an explicit way while
the UV asymptotics have to be derived.
A link between these two kinds of data would provide a good viewpoint for
understanding the general structure of 2D QFT. In general this problem does not look
tractable. Whereas the CFT data can be specified in a relatively simple way the general Smatrix is a rather complicated object even in two dimensions. However, there exists a rather
E-mail address: enrico.onofri@unipr.it (E. Onofri).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 2 0 - 6
V.A. Fateev, E. Onofri / Nuclear Physics B 634 [FS] (2002) 546570
547
important class of 2D QFTs (integrable theories) where the scattering theory is factorized
and the S-matrix can be described in great detail. These theories admit a rather complete
description in the UV and IR regimes.
In this paper we consider the application of the boundary one-point functions and
boundary scattering theory to the explicit construction of the classical solutions describing
the boundary vacuum configurations. As an example of an integrable QFT we study the
simply laced affine Toda theory (ATT), which can be considered as a perturbed CFT (a
non-affine Toda theory). This CFT possesses an extended symmetry generated by a W algebra. We consider the boundary conditions that preserve this symmetry. This permits us
to apply the reflection amplitude approach for the calculation of the boundary vacuum
expectation values (VEVs) of exponential fields in ATTs. The reflection amplitudes in a
CFT define the linear transformations between different exponential fields corresponding
to the same primary field of the full symmetry algebra of theory. They play a crucial role
in the calculation of VEVs of the exponential fields in perturbed CFTs [1] as well as for
the analysis of UV asymptotics of the observables in these QFTs [2,3]. The boundary
VEVs, in particular, contain the information about the boundary values of the solutions of
classical boundary ATTs equations. The information about the long distance behavior of
these solutions can be extracted from the boundary S-matrix. The explicit solutions which
are constructed in this paper provide us the consistency check of CFT and S-matrix data in
integrable boundary ATTs.
The plan of the paper is as follows: in Section 2 we recall some basic facts about
Toda theories and one-point functions in ATTs defined on the whole plane. In Section 3
we consider integrable boundary ATTs. The boundary one-point functions are used to
derive the boundary values of the solutions describing vacuum configurations in ATTs.
We calculate the classical boundary ground state energies and give the conjecture for them
in the quantum case. In Section 4 we describe the boundary scattering theory, which is
consistent with this conjecture, and construct the boundary state. The vacuum solutions
describe the semiclassical asymptotics of the one-point functions of the bulk operators in
boundary QFTs. This makes it possible to derive the long distance asymptotics of vacuum
solutions. These asymptotics determine completely the explicit form of solutions which
we construct in Section 5. The solutions can be written in terms of -functions, associated
with multisoliton solutions of ATTs. In the last section we consider the opposite (dual)
limit of the quantum correlation function of the bulk field with a boundary and check that
its boundary value coincides with that given by boundary VEV. At the end of the section
we consider the solutions for non-simply laced ATTs which can be obtained by reduction
from the solutions for simply laced cases.
2. Affine and non-affine Toda theories

The ATT corresponding to a Lie algebra G of rank r is described by the action:

r

1
2
2
bei
be0
A= d x
( ) +
e
+ e
,
8
i=1
(1)
548
where ei , i = 1, . . . , r, are the simple roots of G and e0 is a maximal root:

e0 +
r
ni ei = 0.
(2)
i=1
The fields in Eq. (1) are normalized so that at = 0

a (x) b (y) = ab log |x y|2 .
(3)
We will consider later the simply laced Lie algebras A, D, E.

For real b the spectrum of these ATTs consists of r particles which can be associated
with the nodes of the Dynkin diagram (or fundamental representations) of G. The masses
of these particles mi (i = 1, . . . , r) are given by:
1 2
m =
mi ,
2h
r
mi = mi ,
(4)
i=1
where h is the Coxeter number of G and i2 are the eigenvalues of the mass matrix:
Mab =
r
ni (ei )a (ei )b + (e0 )a (e0 )b .
(5)
i=1
The exact relation between the parameter m characterizing the spectrum of physical
particles and the parameter in the action (1) can be obtained by the Bethe ansatz method
(see, for example, [4,5]). It can be easily derived from the results of Ref. [5] and has the
form:

2(1+b2)
x

mk(G) ( 1x
h ) ( h )x
2
,
1 + b =
(6)
2 ( h1 )h
where as usual (z) = (z)/ (1 z); x = b2 /(1 + b2 ) and
1/2h
r
n
i
ni
k(G) =
(7)
i=1
with ni defined by Eq. (2).

The ATTs can be considered as perturbed CFTs. Without the last term (with root e0 ) the
action (1) describes NATTs, which are conformal. To describe the generator of conformal
symmetry we introduce the complex coordinates z = x1 + ix2, z = x1 ix2 and the vector
1
Q = (b + 1/b), =
(8)
,
2
>0
where the sum in the definition of the Weyl vector runs over all positive roots of Lie
algebra G.
The holomorphic stress energy tensor
1
T (z) = (z )2 + Q z2
2
(9)
549
ensures the local conformal invariance of the NATT.

Besides conformal invariance NATT possesses an additional symmetry generated by
(G). The full chiral W (G)-algebra

two copies of the chiral W (G)-algebras: W (G) W
contains r holomorphic fields Wj (z) (W2 (z) = T (z)) with spins j which follow the
exponents of the Lie algebra G. The explicit representation of these fields in terms of
fields z a can be found in [6]. The exponential fields:
Va (x) = ea(x)
(10)
are spinless conformal primary fields with dimensions (a) =

(a
The
fields Va (x) are also primary fields with respect to full chiral algebra W (G). It was shown
in Ref. [6] that fields Va (x) and Vs(a)(x) correspond to the same representation of W (G)algebra if their parameters are related by any transformation of the Weyl group W of the
Lie algebra G acting on the vector a as follows:
(Q2
a s(a) = Q + s(a Q),
Q)2 )/2.
s W .
(11)
This means that the fields Vs(a) for different s W are the reflection image of each other
and are related by a linear transformation. The functions that define this transformation
are known as reflection amplitudes. Reflection amplitudes play an important role in the
analysis of perturbed CFTs. In particular they permit to calculate the vacuum expectation
values of exponential fields in ATTs:
G(a) = 0| exp(a )|0.
(12)
These VEVs were calculated in Ref. [7] and they can be represented in the form:

a 2

x

mk ( 1x
dt 2 2t
h ) ( h )x
a e
exp
F (a, t)
G(a) =
t
2 ( h1 )h
(13)
with

sinh(ba t) sinh((b(a 2Q) + h(1 + b2 ))t)
F (a, t) = sinh 1 + b2 t
sinh t sinh(b2 t) sinh((1 + b2 )ht)
(14)
>0
here and below subscript denotes the scalar product of the vector with a positive root ,
i.e.:
a = a ,
(a 2Q) = (a 2Q) .
(15)
This expression satisfies many possible perturbative and non-perturbative tests for onepoint function in ATT. For example, it can be easily derived from Eq. (6) and from the
equation of motion that the bulk vacuum energy E(b) in ATT is expressed in terms of
function G(a) as:

ni E(b) = h 1 + b 2 G(bei ).
(16)
The values of the function G(a) at the special points bei can be calculated explicitly and
the result coincides with the known expression for the bulk vacuum energy [8]:
E(b) =
m2 sin(/ h)
,
4 sin(x/ h) sin((1 x)/ h)
(17)
550
where x = b2 /(1 + b2).

It is easy to see from Eq. (16) that function E(b) (and the function G(a) as well) is
invariant under duality transformation b 1/b or x 1 x. The same is true for the
bulk scattering S-matrix in ATT. This remarkable property of duality is very important for
the analysis of Toda theories.
In the limit b 0 the field = b can be described by the classical equations. In
particular it can be derived from Eq. (13) that the VEV of this field 0 = 0|b|0 at that
limit can be expressed in terms of the fundamental weights i of G as:
0 = b0 =
r

log ni 2 log k(G) i
(18)
i=1
and coincides with a classical vacuum of ATT (1). At this limit we have (see Eq. (6)) that
= (mk(G))2 /4b2 + O(1) and after rescaling and shifting of the field the action (1)
can be expressed in terms of the field = b 0 as:

r

1
1
2
2
2
ei
+ O(1).
Ab =
(19)
ni e
d x ( ) + m
2
4b2
i=0
The VEVs given by Eq. (13) and scattering theory data were used in Ref. [7] to describe
the asymptotics of the cylindrical solutions of classical Toda equations following from the
action given by Eq. (19). In the next sections we use a similar strategy to explicitly find the
vacuum solutions in integrable boundary ATT.
3. Boundary Toda theories, one-point functions and classical vacuum solutions

In this section we consider simply laced boundary Toda theories defined on the halfplane H = (x, y; y > 0). The integrability conditions for classical simply laced ATT on H
were studied in the paper [12]. It was shown there that the action of integrable boundary
ATT can be written as:

r

1
2
2
bei
be0
( ) +
e
+ e
Abound = d x
8
H
+ B
i=1

dx
r
si ebei /2 ,
(20)
where either all the parameters si vanish, corresponding to the Neumann boundary
conditions:
y (x, 0) = 0,
(21)
or they are given by si = 1 and the parameter B is related to the parameter in the bulk
(in the classical case) as:
2B = /b2 .
(22)
551
For the Lie algebra A1 (sinh-Gordon model) the integrability conditions are much less
stringent and the parameters s0 and s1 can have arbitrary values [11]. Here we discuss
this problem for other Lie algebras where the choice of integrable conditions is rather
restrictive. We consider Toda theories with Neumann boundary conditions and in the case
when all parameters si = 1 (with s0 = 0 in the non-affine case). Really, these two quite
different classical theories in the quantum case are described by the same theory and are
related by a duality transformation (b 1/b) [13,14]. The cases corresponding to different
signs of the parameters si are more subtle and will be considered elsewhere.
We start from a consideration of the boundary NATTs described by the action (20)
without the last term in the bulk action and with parameter s0 = 0 in the boundary term. The
boundary ATTs will be considered as perturbed boundary CFTs. On the half-plane with W
j (z)
invariant boundary conditions we have only one W -algebra. In this case the currents W
should be the analytical continuations of the currents Wj (z) into the lower half-plane.
In particular, they should coincide at the boundary. These conditions impose very strong
restrictions on the form of the boundary terms in the action. It is rather easy to derive from
the explicit form of W -currents [6] that the Neumann boundary conditions (21) preserve
W -symmetry. The boundary condition (22) whose quantum modified version has the form
[15]
2B =
2
b
cot
,
2
2
(23)
(with all si = 1, i = 1, . . . , r, and s0 = 0) describes the dual theory and, hence, also
preserves W -symmetry.
In the W -invariant boundary NATT we have two kinds of exponential fields. The bulk
fields Va (x, y) and the boundary fields Ba (x) are defined as:
Va (x, y) = exp{a (x, y)},

1
Ba (x) = exp a (x) .
2
(24)
These fields are specified by the same quantum numbers as those for the corresponding
fields (10) defined on the whole plane. In particular, they possess the same properties with
respect to the action of the Weyl group of G (see Section 2), defined by Eq. (11). Following
the standard lines one can introduce and calculate boundary reflection amplitudes which
relate the fields Ba (x) and Bs(a) (x) (see Ref. [9] for details).
Boundary reflection amplitudes can be used for the calculation of vacuum expectation
values of the boundary exponential fields in ATTs. One-point functions:

1
GB (a) = exp a
2
B
(25)
for ADE ATTs were calculated in [17] and have the form:

GB (a) =
x
mk ( 1x
h ) ( h )x
2 ( h1 )h
a 2 /2

dt a 2 2t
e
exp
FB (a, t)
t 2
0
(26)
552
with
FB (a, t) = f (t)
sinh(ba t) sinh((b(a 2Q) + h(1 + b2))t)

>0
sinh 2t sinh(2b2t) sinh((1 + b2 )ht)
where for the boundary conditions (23) the function f (t) is:

f (t) = 2et sinh 1 + b 2 t cosh b 2 t
(27)
(28)
and for the dual theory which corresponds to Neumann boundary conditions (21) we should
make the substitution f (t) f (t):

2
f (t) = 2et b sinh 1 + b2 t cosh t.
(29)
It is easy to see from the explicit form of GB (a) that in the classical limit (b 0 with
b is fixed) the boundary VEV 0,B of the field b for the Neumann boundary conditions
coincides with the classical vacuum 0 (18) in the bulk. For the boundary conditions (23)
this happens only for Lie algebra Ar , where both these values vanish. For other cases we
can derive from Eqs. (26)(28) and Eqs. (13), (14) that difference of VEVs of the field b
at the boundary and in the bulk (see Section 2) has the form:
= bB 0|b|0 =

x sinh((h 2 )t) sinh t
.
dt
sinh xt cosh(1 x)t sinh ht
>0
(30)
Vector can be expressed in terms of elementary functions at the semiclassical limit

x 0, at the self-dual point x = 1/2 and in the dual limit x 1. We discuss the dual
limit in the last section. Here we note that at x 0 vector has a limit:
dt sinh((h 2 )t)
=
tanh t.
t
sinh(ht)
>0
(31)
For D and E series, which we consider below the numbers 2l = h 2 are always even.
For the following it is convenient to define:

1
Ei = exp ei =
(32)
(pl )ei ,
2
>0
where pl = 1/pl can be written in terms of trigonometric functions as:

l
dt sinh 2lt
cos((l + 1 2k)/ h)
2
tanh t =
.
pl = exp
t sinh ht
cos((l 2k)/ h)
0
(33)
k=1
In particular:
E0 =
pq1 ph/2q
,
ph/21
here q(G) = maxi ni ; q(D) = 2, q(E6) = 3, q(E7 ) = 4, q(E8) = 6.
(34)
553
The vector is simply related to the boundary soliton solution that describes the
classical vacuum configuration. The classical problem for this solution (y), y > 0 can
be formulated in the following way. We are looking for a solution to the classical Toda
equations, which decreases at y and satisfies at y = 0 the boundary conditions that
follow from the action (20). After rescaling and shifting (see Section 2) the field = b
0 satisfies the equation:
y2 = m2
r
ni ei exp(ei ),
n0 = 1,
(35)
i=0
and the boundary condition at y = 0:

r
1
y = m
ni ei exp ei .
2
(36)
i=0
The vector is equal to the boundary value of this solution and hence, it completely fixes
the boundary soliton. At y = 0 we have
(0) = ,
y (0) =
r
ni ei Ei .
(37)
i=0
This gives us the possibility to study Eq. (35) numerically. The numerical analysis of
this equation shows that for these and only for these boundary values a smooth solution
decreasing at infinity exist (see Appendix C).
It is natural to expect that solution (y) to the Eqs. (35), (36) can be expressed in terms
of tau-functions associated with multi-soliton solutions of the classical ATT equations (see,
for example, [16,18] and references there). For the D and E series of algebras (besides the
cases D4 and D5 [16]) the explicit form of these solutions was not known. We postpone the
discussion of these solutions to Section 5. Here we consider the classical boundary ground
state energy which can be defined as:

r
r

e

1
1
(cl)
2
2
i
Ebound =
ni Ei + dy (y ) + m
ni e
1
. (38)
2m
4b2
2
i=0
i=0
We note that numerical values of 2 , defined by Eq. (31) are rather small for all G, and the
functional (38) can be calculated to a good accuracy using the bilinear approximation:

m
ni Ei (2 log Ei ) + O ||3 .
2
4b
r
(cl)
Ebound
=
i=0
More careful analysis of Eqs. (35), (36) (see Section 2) gives us the reasons to write the
expression for boundary ground state energy in the following form. Namely, we denote as
m (G) the sum of the masses of all particles in the ATT:
m (G) =
r

i=1
mi = m
r

i=1

i = m tr M 1/2 .
(39)
554
This has the following values for simply laced Lie algebras:
2m cos(/4 /2h)
,
sin(/2h)
cos(/8)
2m sin(2/9)
,
m (E7 ) =
,
m (E6 ) = m(6 2 3 )1/2
sin(/2h)
sin(/2h)

1/2 cos(/5)
.
m (E8 ) = 4m 3 sin(/5) sin(/30)
sin(/2h)
m (An ) = 2m cot(/2h),
m (Dn ) =
(40)
The classical boundary ground state energy (38) can be written in terms of these values as:
(cl)
Ebound
(G) =
h
tan(/2h)m (G).
4b2
(41)
(q)
In the quantum case the boundary ground state energy Ebound will have contributions
coming from the boundary term in the Hamiltonian and from the bulk fluctuations around
the background solution. The contributions of the first type can be calculated using the
explicit expression for vacuum expectation values (26). For small b the first quantum
correction of the second type can be expressed through the boundary S-matrix at b = 0
(see, for example, [19,20]). We will discuss these boundary S-matrices below. Here we
note that in the strong coupling region b 1 our theory is described by the weakly coupled
dual ATT with Neumann boundary conditions (21). In the strong coupling limit the dual
theory is a set of r free bosonic theories with masses mi . The boundary ground state energy
for the free massive bosonic theory with Neumann boundary conditions and mass mi can
be easily calculated and is equal to mi /8. The first perturbative correction in the weakly
coupled dual theory can also be evaluated; since this calculation is outside the scope of the
present paper, and will be published elsewhere, we only quote the result:

m (G)
(q)
4
1+
cot(/2h)
+
O
1/b
Ebound(G) =
(42)
.
8
2hb2
Both asymptotics b 0 (41) and b (42) are in agreement with the following
conjecture for boundary ground state energy:
(q)
Ebound(G) =
sin(/2h)m (G)
,
8 sin(x/2h) cos((1 x)/2h)

x = b2 / 1 + b2 .
(43)
We note that for A1 ATT (sinh-Gordon model with special boundary conditions (23)) this
equation coincides with the expression proposed by Al. Zamolodchikov [21] (see also
[22]).
The nonperturbative check of this conjecture can be made using the boundary
Thermodynamic Bethe ansatz equations [23]. The kernels in these nonlinear integral
equations depend on the bulk S-matrices Sij ( ) and the source (inhomogeneous) terms
on the boundary S-matrices Rj ( ). The boundary ground state energy can be expressed
through the main terms of the asymptotics of these values at [24]. In particular
boundary ground state energy in ATT can be expressed in terms of the mass mj of the
particle j multiplied by the ratio of Fourier transforms of logarithms of boundary S-matrix
Rj and bulk amplitude Sjj taken at = i. To check our conjecture and to construct the
explicit solution of Eqs. (35), (36) we need the information about the boundary S-matrix.
555
4. Boundary S-matrix and boundary state

The boundary S-matrix (reflection coefficient) in the ATT for the particle j corresponding to the fundamental representation j (G) can be defined as:
|j, B,out = Rj ( )|j, B,in
(44)
where is the rapidity of particle j .

The reflection coefficients for the Ar ATT with boundary conditions (23) were
conjectured in [25,26]. They can be written in terms of function:
(z) =
sin( 2i +
sin( 2i
2h z)
2h z)
(45)
in the following way:

Rj ( ) =
(a 1)(a h)(a + x)(a + h + 1 x).
(46)
a=1
Unfortunately, we were not able to find in the literature any conjecture for other Lie
algebras consistent with duality properties discussed above. So we will give here a
conjecture which naturally generalizes the reflection coefficients (46) to other simply laced
Lie algebras. To do this we rewrite Eq. (46) in the form:

Rj ( ) = exp ij ( ) ,
(47)
where

j =

2h t
dt
sinh (1 x)t sinh (h + x)t j (Ar , t) 2
sinh
t
(48)
and
8 sinh j t sinh(h j )t
.
sinh t sinh 2ht
The natural generalization of these equations can be written as:

Rj ( ) = j ( ) exp ij ( ) ,
j (Ar , t) =
(49)
(50)
where j ( ) are the CDD factors, satisfying the conditions:

j ( )j ( ) = 1,
( )( + i) = 1
(51)
and the function j ( ) is defined by Eq. (48), where we should do the substitution
j (Ar , t) j (G, t) with:
1
4
(2 cosh t 2)mn + em en
(52)
.
jj
cosh ht
The most important part of this conjecture is that for the particles j corresponding to
fundamental representations j (G) (or the node j of Dynkin diagram) with nj in Eq. (2)
equal to 1, the CDD factors j ( ) in Eq. (50) are equal to one.
j (G, t) =
556
This statement fixes completely the boundary S-matrix for Lie algebras Dr , E6 , E7 .
We denote by Rf (G, ) the fundamental reflection coefficients. This means that all other
amplitudes Rj ( ) can be obtained from the fundamental reflection factors Rf (G, ) by
application of the boundary bootstrap fusion procedure [11,27]. It is easy to see that
Rf (Dr , ) = Rr ( ) = Rr1 ( ), where particles r and r 1 correspond to the spinor
representations of Dr ; Rf (E6 , ) = R1 ( ) = R1 ( ), where particles 1 and 1 form the
doublet of lightest particles in the E6 ATT and Rf (E7 , ) = R1 ( ), where 1 is the lightest
particle in the E7 ATT. For all these three cases the CDD factors f ( ) = 1 and the
reflection coefficients are given by Eqs. (47), (48) with j = f and
f (G, t) =
8 sinh (ht/2) sinh((h/2 + q 1)t)

,
sinh(qt) sinh(2ht)
(53)
where q(G) = maxi ni ; q(D) = 2, q(E6) = 3, q(E7) = 4.

As an example of the application of the boundary bootstrap equations we give here the
reflection coefficients Rj for the particles j = 1, 2, . . . , n 2 in the Dr ATT, which can
be obtained from the amplitude Rf ( ) = Rr ( ) = Rr1 ( ). These functions Rj have the
form (50), where:
16 sinh j t cosh((h/2 j )t) sinh(ht/2)
sinh t sinh 2ht
and the CDD factor can be written as:

dt
sinh(2h t/) sinh(1 x)t cosh(xt)j (Dr , t) ,
j = exp i
t
j (Dr , t) =
(54)
(55)
where j (Dr , t) = 8 sinh(j 1)t sinh j t/(sinh 2t cosh ht).

The Lie algebra E8 has no fundamental representations with nj = 1. The lightest
particle in this case is associated with the adjoint representation and Rf (E8 , ) = Rad ( ).
The adjoint representations for Lie algebras D, E have nad = 2 and, hence, CDD factors
should appear. The reflection coefficients Rad ( ) for Lie algebras D and E can be written
in the form (50), with:
ad (G, t) =
8 cosh((q 1)t) cosh((h/2 q)t)

,
cosh(ht/2) cosh ht
(56)
where q(G) is defined above and the CDD factor ad ( ) is defined by Eq. (55) with
function
8 sinh t
.
ad (G, t) =
(57)
cosh ht
The amplitudes Rj ( ) following from the scattering theory described above (see also
Section 5) define the minimal solution of boundary bootstrap equations and crossing
unitarity [11] conditions. They have no poles corresponding to the bound states of particles
j with boundary. The only poles with positive residues which appear in some of the
functions Rj ( ) are the poles at = i/2
Rj ( ) =
iDj2 (b)
i/2
(58)
557
corresponding to the particle boundary coupling with zero binding energy. For example,
the lightest particle whose boundary amplitude possesses such pole is the particle mad
corresponding to the adjoint representation. It is easy derive from Eqs. (56), (57) that the
corresponding residue has the form:

(1 x)
x
2
tan
tan
T0 Tq1 Th/2q ,
Dad (b) = 2 cot
(59)
2h
2h
2h
where

(k + 1)
(k + 1 x)
Tk = tan
tan
4
2h
4
2h

k
(k + x)
cot
.
cot
4
2h
4
2h
For the following it is convenient to consider the theory in the crossed channel, where
the variable y > 0 plays the role of Euclidean time. The Hamiltonian and Hilbert space
in this case coincide with those for the theory defined on the whole plane (see Ref. [11] for
details). The amplitudes Rj ( ) and their residues can be used for the construction of the
boundary state in integrable QFT. It was shown in Ref. [11] that this state can be written in
terms of numbers Dj and functions Kj ( ) = Rj (i/2 ) as:
r

1
|B = exp
(60)
Dj (b)Aj (0) +
d Kj ( )Aj ( )Aj ( ) |0,
4
j =1
where Aj ( ) are the operators creating the asymptotic states with rapidity , i.e.,
Aj ( )|0 = |j, in .
We will be mostly interested by the linear in the operators Aj part in the expansion of
boundary state |B

r

Dj (b)Aj (0) |0 +
|B = 1 +
j =1
which determines one particle contributions to the correlation function:

Fb (y) = 0|b(y)|B 0|b|0,
(61)
where the variable y > 0 plays the role of Euclidean time. These one particle
contributions can be written as:
Fb (y) =
r
Dj (b)0|b(0)|j, 0in exp(mj y) + .
(62)
j =1
The boundary soliton (y) (see Section 3) coincides with the semiclassical limit of this
correlation function, i.e., (y) = limb0 Fb (y). In the weak coupling limit one particle
matrix elements can be easily calculated and have the form:

0|(0)|j, 0in = j + O b2 ,
558
where j are the eigenvectors of mass matrix: Mj = j2 j , satisfying the conditions:

i j = ij ,
j 0.
(63)
The corresponding one particle contributions to the boundary solution (y) will be:
(y) =
r
dj j exp(mj y) + ,
(64)
j =1
where
dj = lim
b0
bDj (b).
(65)
For example the coefficient dad corresponding to the lightest particle mad in expansion (64)
and defining the main term of the asymptotics at y can be extracted from Eq. (59)
and has the form:

dad = 2 h tan
(66)
T0 Tq1 Th/2q ,
2h
where Tk2 = limb0 Tk :

(k + 1)
cot
.
Tk = tan
4
2h
4
2h
(67)
The coefficients dj as well as the vector (or numbers Ej ) fix completely the solution
(y) of Eqs. (35), (36). They determine the contribution of the zero modes of the linearized
Eq. (35) and make it possible to develop in a standard way the regular expansion at large
distances. If our scattering theory is consistent with conformal perturbation theory this
expansion should converge to the boundary values at y = 0.
At the end of this section we note that the analysis of the boundary Thermodynamic
Bethe ansatz equations with the reflection coefficients written above gives an exact
agreement with Eq. (43) for the boundary ground state energy.
5. Boundary solutions
It is natural to assume that boundary vacuum solutions can be expressed in terms of functions associated with multisoliton solutions of ATTs equations. This means that field
can be written as:
(y) =
r
ei log i (y),
i (y) 0,
y ,
(68)
i=0
where functions i (y) satisfy the equations:

r

Iij

2
2
i (y) + (i ) =
j i ni ,
j =0
here Iij is the incidence matrix of the extended Dynkin diagram of G.
(69)
559
The classical boundary state energy can be expressed in terms of numbers Ei and the
boundary values of -functions [16] as:

i (0)
h
(cl)
, i = 0, . . . , r.
ni Ei +
Ebound =
(70)
i (0)
2b2 ni
Functions i corresponding to multisoliton solutions to the ATT equations are given by
finite order polynomials in the variables
Zj = exp(mj y) = exp(mj y).
(71)
The general ansatz for these functions has the form:

i (y) =
ni
ni
k1 =0
kr =0
Yk(i)
(t1 Z1 )k1 (tr Zr )kr ,
1 kr
(72)
(j )
where Y0...0 = 1 and all other coefficients can be derived from Eq. (69) if we fix the
normalization for the linear in variables Zj terms in the expansion (72):
i (y) = 1 +
r

j =1

(i)
Y0...1
tj Zj + O Z 2 .
j ...0
(73)
These coefficients should satisfy the condition:
r
(i)
ei Y0...1j ...0 = Nj j .
(74)
i=0
After we fix the factors Nj all coefficients Y (i) in Eq. (72) are universal numbers depending
only on G. The parameters tj in Eq. (72) are simply related with parameters dj defined by
the expansion (64). It is easy to see from Eqs. (64), (68), (73), (74) that
tj = dj /Nj .
(75)
The parameters dj can be extracted from the boundary scattering theory described in the
previous section. This gives us the possibility to derive the explicit solution to Eqs. (35),
(36).
5.1. Dr boundary solution
The boundary values of this solution can be described by the numbers Ek (see Section 3)
which in this case are:
2 2
E0 = E1 = Er1 = Er =
,
h sin(/ h)
2
cos(k/ h) cos((r k)/ h)pk2 prk
Ek = Erk =
(76)
.
cos((2k 1)/2h) cos((2r 2k 1)/2h)
These numbers possess all the symmetries of extended Dynkin diagram for Dr . The same
is true for the boundary solution (y). It means that in Eq. (68) 0 = 1 and k = rk . For
560
the Lie algebra Dr the following relation holds:
r

ni cos j (2i 1)/ h ei = 2 h sin(j/ h)j
(77)
i=0
so that it is natural to choose Nj (D) = 2 h sin(j/ h). It is convenient also to introduce

the functions:
0 = 1 = r1 = r = 02 ,
k = k ,
k = 2, . . . , r 2.
(78)
The particles which are invariant with respect

to symmetries of Dynkin diagram are
particles 2l; 2l r 2 with masses m2l = m 8 sin(2l/ h). The analysis of the boundary
S-matrix (54) gives that only amplitudes R2l ( ) have a pole at = i/2. The parameters
t2l = d2l /N2l (D) can be derived from the Eqs. (54), (55) and have the form:
tan (/4 l/ h)
tan2 ((2i 1)/2h)
t2l =
(79)
.
2 cos2 (l/ h)
tan2 (i/ h)
i=1
Parameters t2l and functions Z2l = exp(ym 8 sin(2l/ h)) are defined for 2l r 2.
To write the solution in the most short form it is convenient, however, to continue these
values to l r 2. To continue the parameters t2l to l r 2 we can use Eq. (79). In this
way we obtain:
l
th2l = t2l ,
Zh2l = Z2l .
(80)
We introduce also = exp(2i/ h). Then the boundary solution can be written as:
1
ni ei log i (y),
2
r
(y) =
(81)
i=0
where
i (y) =
1

1 =0
1
r2
r2 =0 k=1
(2i1)kk
(t2k Z2k )
r2

m<n
sin((m n)/ h)
sin((m + n)/ h)
2m n
. (82)
As an example of application of the last equation we give here the result for D7 -solution.
This solution is characterized by three functions 0 , 2 , 3 (78) which are:

2

3
3 2
1
t1 Z1 + t2 Z2
tan
t1 Z1 t2 Z2 ,
0,3 = 1
2
2
4
12

3
3(t1 Z1 )2 (t2 Z2 )2
+
+
tan4
2 = 1 2t2 Z2 +
(t1 Z1 t2 Z2 )2 .
4
4
16
12
We note that in the usual basis of roots of Dr : e0 = 1 2 ; ek = k k+1 ,
k = 1, . . . , r 1; er = r1 + r the solution can be written as k = log(k1/ k ).
It can be checked that the boundary value of (y) defined by Eqs. (81), (82) coincides
with , defined by Eq. (76) and the classical boundary ground state energy (70) can be
derived using Eq. (41).
561
Fig. 1.
5.2. E6 boundary solution

The boundary values for this solution are:

3+1
3 + 3 1/2
(3 + 5/ 3 )1/2 ,
E0 = E1 = E1 =
,
E4 =
6
6
E2 = E3 = E3 = (E0 E4 )1/2 .
(83)
Here and later we use the numeration of nodes j of Dynkin diagram corresponding to the
particles mj (see Fig. 1).
The solution possesses Z3 symmetry of this diagram. It means that: 0 = 1 = 1 and
2 = 3 = 3 . The particles that are invariant with respect this symmetry are m2 = mad and
m4 . The analysis of boundary S-matrix shows that only amplitude R2 = Rad ( ) possesses
the pole at = i/2 with the residue (59). In
this case only one particle contributes to the
solution. It is convenient to take N(E6 ) = 2 3h. Then the parameter t2 = dad /N is:

t2 = tan
(84)
T0 T2 T3 / 3 = tan
tan
tan
.
24
24
24
12
The variable Z2 = exp(m2 y), where 22 = 2(3 3 ) and the solution has the form:

2
2 = 1 2t2 Z2 + (2 + 3 )t2 Z2 ,
0 = 1 + (2 + 3 )t2 Z2 ,
3

4 = 1 3 3t2 Z2 3 3(2 + 3 )(t2 Z2 )2 + (2 + 3)t2 Z2 .
(85)
It is easy to check that (0) = and the classical boundary ground state energy (70)
coincides with that given by Eq. (41).
The boundary values (32) for this solution are:

p3 p5 2 2
p4 p6
cot
=
,
E0 = E1 =
,
E2 = E4 =
p8
9
9
p3 p7
p5 p6 p7
p1 p2 p7
p3 p4 p8
,
E5 = E6 =
,
E7 =
.
E3 =
p1 p2 p3 p4 p8
p4 p6
p1 p2 p5 p7
(86)
Numbers Ek possess Z2 symmetry of the extended Dynkin diagram (see Fig. 2).
This means that in Eq. (68) 0 = 1 , 2 = 4 and 5 = 6 and only Z2 even particles
m2 = mad , m4 , m6 and m7 can contribute to this solution. The analysis of the boundary Smatrix (which can be obtained by boundary bootstrap fusion procedure from the amplitude
Rf ( ) (53)) shows that only amplitudes R2 = Rad , R4 and R7 possess the poles at
562
Fig. 2.
= i/2. These amplitudes have a form (50) where the phases j and CDD factors j are
defined by the functions j (E7 , t) and j (E7 , t) in Eqs. (48), (52), (55). For amplitude
R2 = Rad these functions are given in Section 4. For amplitudes R4 and R7 they are:
16 cosh t cosh 3t cosh 4t
,
cosh(ht/2) cosh ht
8 cosh t cosh 2t sinh 6t
,
7 =
sinh t cosh(ht/2) cosh ht
4 =
16 sinh 5t cosh 4t
,
cosh ht
8 sinh 3t sinh 5t sinh 8t
7 =
.
sinh t sinh 2t cosh ht
4 =
The coefficients djcan be extracted from these amplitudes using Eq. (65). It is convenient
to take N(E7 ) = 2 3h. Then the parameters tj = dj /N can be written in terms of numbers
Tk (67) as:

T0 T3 T5 ,
T0 T1 T3 T5 T7 ,
3t2 = tan
3t4 = tan
2h
2h

3t7 = tan
(87)
T0 T12 T2 T32 T52 T72 (T6 )1 .
2h
The corresponding variables Zj = exp(mj y), where
2 = 32 sin(/9) sin(2/9),
4 = 32 sin(/9) sin(4/9),
7 = 32 sin(2/9) sin(4/9).
(88)
We give here the explicit form for function 0 . All other -functions can be derived from
it using Eq. (69). For example, it holds:
2 = 02 0 0 + (0 )2
(89)
with similar relations for higher -functions. As an example of application of this equation
we give the explicit expression for 2 in the form (72) in Appendix A.
For all simply-laced algebras the number n0 is equal to one and function 0 in Eq. (72)
can be written as:
0 =
1
1
1

1 =0 2 =0 3 =0
Y(0)
(t Z )1 (t4 Z4 )2 (t7 Z7 )3 .
1 2 3 2 2
(90)
It is convenient to simplify the notations. The variables i take the values 0 or 1. We can
introduce the variable 1 i 3 indicating i which are equal to one. The corresponding
(0)
(0)
coefficients Y(0)
1 ,...,3 we denote as yi , yij , y123 . For example, y1 = Y100 , y13 = Y101 and so
on. With this notation we have:
y1 =
2 cos2 (2/9)
,
cos(4/9)
y2 =
2 cos2 (/9)
,
cos(2/9)
y3 =
2 cos2 (4/9)
,
cos(/9)
8 cos4 (2/9)
8 cos4 (4/9)
,
y13 =
,
cos(/9)
cos(2/9)

4
y123 = 2 cos2 (4/9)/ cos(/9) .
y12 =

7
y13 = 2 cos(4/9) ,
563
(91)
It can be checked that (0) = , where is defined by Eq. (86) and the classical boundary
ground state energy (70) coincides with that given by Eq. (41).
The boundary values for this solution follow from Eqs. (32), (34). The extended Dynkin
diagram (see Fig. 3) in this case has no symmetries and all particles can contribute to
the solution; however the analysis of the boundary S-matrix shows that only amplitudes
R1 = Rad , R2 , R5 and R8 have the poles at = i/2. The amplitude R1 = Rad is given in
Section 4. Three other amplitudes with these poles are defined by the functions j (E8 , t)
and j (E8 , t), which have the form:
16 cosh 3t cosh 5t cosh 6t
16 sinh 7t cosh 6t
, 2 =
,
cosh(ht/2) cosh ht
cosh ht

8 cosh 4t cosh 5t sinh 6t
, 5 =
1 ,
5 =
sinh t cosh(ht/2) cosh ht
cosh ht
sinh t sinh 4t
2 sinh 6t sinh 10t
.
8 =
, 8 =
sin t sinh 2t cosh ht
sinh2 t cosh(ht/2) cosh ht
2 =
These amplitudes
determine the coefficients dj (see Eq. (65)). It is convenient to take
N(E8 ) = 2 h. Then the parameters tj = dj /N can be written in terms of numbers Tk (67)

as:

t2 = tan
t1 = tan
T0 T5 T9 ,
T0 T3 T5 T6 T9 T11 (T8 )1 ,
2h
2h

3
tan
T7 T13 (T4 T6 T8 T10 T12 )1 ,
t5 = tan
2h
2h

2 2 1
T12 .
t8 = tan2
(92)
T1 T3 T5 T7 T9 T11 T13 T6 T8 T10
2h
The corresponding variables Zj = exp(mj y), where

2
1,5 = 3(5 5 )/2 3(5 + 5 )/2

2
= 3(5 + 5 )/2 3(25 + 11 5 )/2.
2,8
(93)
We give here the explicit form for function 0 . All other -functions can be derived from
it using Eq. (69). For example, function 1 is related with 0 by Eq. (89). Function 0 can
be written in the form:
0 (y) =
1

1 =0
1

4 =0
Y(0)
(t Z )1 (t2 Z2 )2 (t5 Z5 )3 (t8 Z8 )4 .
1 4 1 1
(94)
564
Fig. 3.
The coefficients Y(0)

1 4 for this function are given in Appendix B. It was checked with the
accuracy allowed by Mathematica1 that (0) = and the classical boundary state energy
(70) coincide with that given by Eq. (41).
1. In the previous sections we studied the semiclassical asymptotics of integrable
boundary ATTs. In particular, we derived the solution (y) describing the limit b 0 of
one-point function Fb (y) of the bulk operator with a boundary. Here we shortly consider
the opposite (dual) limit b of this function. The boundary value of function Fb (y)
is given by Eq. (30), and at this limit is equal

=
(95)
cot
.
2h
h
>0
For example, for the Lie algebra Dr this equation can be written as:
e0 = e1 = er1 = er = /(2h) cot(2/ h),

ek = /(2h) cot(2k/ h) + cot 2(r k)/ h .
(96)
At the limit b our theory is described by the weakly coupled dual theory, which is
boundary ATT with Neumann boundary conditions and function (y) = F (y) can be
calculated using perturbation theory. It can be written in the form:

r
r
j
j

d fii sin(ii )e2mi y cosh
mj y
.
j dj e
(y) =
(97)
4 h cosh2 cos2 (iij /2)

j =1
i=1
j
Here fii are the fusion constants and ii are the fusion angles. The constants fii are
different from zero and equal to 1 or 1 only if particle j is a bound state of two particles i.
j
In this case the fusion angle is defined by relation: mj = 2mi cos(ii /2). The complete list
j
j
of constants fii and angles ii can be found in Ref. [28]. The constants dj describing one
particle contributions are different from zero exactly for the same particles j that and the
constants dj calculated in previous section. For these particles they can be expressed in
terms of residues of amplitudes Rj ( ) (see Eq. (65)) as:
dj = lim bDj (b) = .

(98)
b
h
1 Copyright Wolfram Research Inc.
565
As an example of the application of Eq. (97) we give the explicit expression for function
(y) in Dr ATTs:

[r/2]1

d
(gl gr1l + fl ) ,
2l em2l y
(y) =
4
h l=1
where
gl ( ) = sin(2l/ h)
exp(4 2 sin(l/ h)my cosh )
cosh2 cos2 (l/ h)
exp(2 2my cosh )

.
fl ( ) = 2 sin(4l/ h)
cosh2 sin2 (2l/ h)
(99)
,
(100)
A plot of the functions 0 = e0 and 0 = e0 for D7 ATT is given in Fig. 4.

It is straightforward to derive from Eqs. (99), (100) that

[r/2]1

4l
.
2l 1
(0) =
h
2 h l=1
(101)
If we take into account Eq. (77) we can easily find that boundary value (0) coincides
with vector defined by Eq. (96). The same result can be obtained for all simply laced
ATTs.
Fig. 4. The functions 0 = e0 and 0 = e0 for the D7 ATT.
566
2. In this paper we derived the weak and strong coupling limits of function Fb (y). In
the intermediate region the long distance behavior of this function can be expressed in
terms of boundary amplitudes Rj and form factors [29]. Some of form factors in ATTs
are calculated in Ref. [10]. At short distances we can calculate Fb (0) which is equal to
vector . Using boundary conditions and equations of motion we can express first two
derivatives of this function through the boundary one-point function GB (a):
F b (0) =
r
b 2 B
ej GB (bej ),
2
F b (0) = b2
j =0
r
ej GB (2bej ).
(102)
j =0
To calculate the next derivatives of this function we need the boundary VEVs of the
descendant fields. At present the problem of their calculation is not solved yet. We hope to
return to this problem in the next publications.
3. In the main part of this paper we studied boundary simply laced ATTs with dual to
Neumann boundary conditions (23). At the end of this section we consider the classical
limit of non-simply laced ATTs with the same boundary conditions. These theories are
dual to the ATTs with dual affine algebras and boundary conditions (21). In the non-simply
laced case the corresponding boundary vacuum solutions (y) satisfy the same Eqs. (35)
with modified boundary conditions at y = 0 which can be written as:
y = m
r
2ni /ei2 ei exp(ei ).
(103)
i=0
The solutions nsl of these equations can be obtained from the solutions sl for
simply laced ATTs described in the previous section by the reduction with respect to the
symmetries of extended DE Dynkin diagrams (which are the symmetries of solutions
sl ). This reduction can be described in the following way. Affine Dynkin diagrams for
non-simply laced algebras can be obtained by folding the nodes of diagrams for simply
laced algebras G. Let us the set of nodes {j } of affine Dynkin diagram for G corresponds
after the folding to the node j of the diagram for non-simply laced algebra G. We denote
as eJ the root of G with arbitrary J {j } and as ej the corresponding root of G. Then it
holds:
ej nsl (y) = eJ sl (y).
(104)
This relation defines completely the solutions of Eqs. (35), (103) for all for non-simply
laced algebras G. The ratio of corresponding boundary ground state energies is equal to the
ratio of Coxeter numbers of G and G.
Extended Dynkin diagram for the Lie algebra Br can be obtained by folding nodes r
and r + 1 of the diagram for Dr+1 . In this case E (cl) (Br ) = E (cl) (Dr+1 ). In the dual limit
(2)
Br ATT with boundary conditions (103) is described by weakly coupled Br (or A2n1 )
ATT with Neumann boundary conditions. The perturbative calculations and analysis of
boundary S-matrix for this theory gives us the reasons to conjecture that in quantum case
the boundary ground state energies for Br and Dr+1 ATTs have a similar form:
(q)
Ebound(Br ) =
m cos(/4 /2h)
,
4 sin(x/2h) cos((1 x)/2h)
(105)
567
where h now depends on x: h = 2r x.

Folding the nodes r + 1 with r + 2 and 0 with 1 of Dr+2 diagram we obtain the diagram
(2)
for Cr or (Dr+1
) ATT. The similar consideration of this theory leads us to conjecture that
Ebound(Cr ) is also described by Eq. (105) where now h = 2r + 2(1 x).

Folding the nodes k and 2r k of affine D2r diagram we obtain the classical solution
for Br boundary ATT. In this case E (cl) (Br ) = E (cl) (D2r )/2. The solution for BCr (or
A(2)
2n ) ATT can be obtained from D2r+2 solution by folding D2r+2 diagram with respect
to all symmetries. In this case all r particles contribute to the solution and E (cl) (BCr ) =
E (cl) (D2r+2 )/2.
The solution for G2 ATT can be obtained from D4 solution by folding the nodes
1, 3, 4 of affine D4 Dynkin diagram. In this case only the heavy particle m2 contributes
to the solution and E (cl) (G2 ) = E (cl) (D4 ). Contrary to that only the light particle m1
(3)
contributes to the solution for G
2 or (D4 ) ATT. This solution can be obtained by the
reduction with respect threefold symmetry of E6 extended Dynkin diagram. In this case
(cl) (E )/3. Folding the nodes 1 with 1
and 3 with 3 of the same diagram we
E (cl) (G
6
2)=E
obtain one particle solution for F4 ATT. The three particle solution for F4 or (E6(2) )
ATT can be derived by the reduction with respect to Z2 symmetry of E7 extended Dynkin
diagram. As a result E (cl) (F4 ) = E (cl) (E7 )/2. The solution for the last ATT corresponding
to the Lie algebra Cr vanishes: (y) = 0.
In this paper we found explicitly all solutions of Eqs. (35), (103). However some of
non-simply laced ATTs require more general integrable boundary conditions [12], which
form the parametric families. We intend to describe the corresponding boundary one-point
functions, scattering theory and boundary solutions in a separate publication.
(q)
Acknowledgements
We are grateful to Al. Zamolodchikov, V. Riva and B. Ponsot for useful discussions.
E.O. warmly thanks A. Neveu, director of L.P.M., University of Montpellier II, and all his
colleagues for the kind hospitality at the Laboratory. This work supported by part by the
EU under contract ERBFRMX CT 960012 and grant INTAS-OPEN-00-00055.
Appendix A
In this appendix we give the explicit form for the function 2 in E7 ATT. This function
can be obtained from the function 0 (see Eqs. (90), (91)) using Eq. (89) and is determined
(2)
(2)
by the following coefficients Yabc , a, b, c = 0, 1, 2; Y000 = 1:
2 =
2
2
2
(2)
Yabc (t2 Z2 )a (t4 Z4 )b (t7 Z7 )c ,
a=0 b=0 c=0
where
(2)
Y100
= y1(2) = 4 cos(2/9),
(2)
Y010
= y2(2) = 4 cos(/9),
(A.1)
568

(2)
(2)
(2)
Y001 = y3 = 4 cos(4/9),
(2)
Y110 = 1/ cos2 (/9),

(2)
Y101 = 1/ cos2 (2/9),
Y011 = 4 cos2 (4/9)/ cos2 (/9),
(2)
= 128 cos4 (4/9)/ cos2 (/9),
Y111
(2)
Y002
= y32 ,
(2)
Y200
= y12 ,
(2)
Y210
= y12 y2(2)y1 /y2 ,
(2)
Y102 = y13 y1 y3 /y1 ,
(2)
(2)
2
Y220 = y12
,
Y012 = y23y2 y3 /y2 ,

(2)
2
= y23
,
Y022
(2)
(2)
(2)
Y120
= y12 y1(2)y2 /y1 ,
(2)
Y201 = y13y3 y1 /y3 ,
(2)
(2)
(2)
(2)
(2)
Y122 = y123y23 y1 /y1 ,
(2)
(2)
(2)
Y021 = y23 y3 y2 /y3 ,
(2)
2
Y202 = y13
,
(2)
(2)
Y211
= y12 y13Y011
/(y2 y3 ),
Y112 = y13y23 Y110 /(y1 y2 ),
(2)
Y020
= y22 ,
(2)
(2)
Y121
= y12 y23 Y101
/(y1 y3 ),
(2)
Y221 = y123y12 y3 /y3 ,
(2)
(2)
Y212 = y123y13 y2 /y2 ,
(2)
2
Y222 = y123
.
Appendix B
In this appendix we list the coefficients Y(0)
1 2 3 4 in the Eq. (94) for 0 function in
E8 ATT. It is convenient to simplify the notations. The variables i take the values 0
or 1. We can introduce the variable 1 i 4 indicating i which are equal to one.
(0)
The corresponding coefficients Y1 4 We denote as yi , yij , yij k and y1234 . For example,
(0)
(0)
(0)
y2 = Y0100 ; y13 = Y1010 and so on. Then Y0000
= 1 and

5
(1 + 5)2
8
y1 =
+ 4+ ,
3+
8
3
5

5
(1 + 5) 7
8
+ 4+ ,
y2 =

4
3
3
5

2

5
(1 + 5)
8
4+ ,
y3 =
3+
8
3
5

7
(1 + 5)
8
5
y4 =
+ 4+ ,
4
3
3
5

19
83
458
4289 9362
298
43
44
+
,
,
y13 = + 4
y12 =
3
3
3
3
5
3 5
5
3 5

19
38
7801 14342
17
71 101
25
y23 =
+
,
+5
,
y14 =
3
3
3
3
5
3 5
5
3 5

43
278
19
125
4289 9362
52
43
,
,
y24 = + 4
y34 = + +
3
3
3
3
5
3 5
5
3 5
y123 = y12 y13y23 /(y1 y2 y3 ),
y124 = y12 y14 y24 /(y1 y2 y4 ),
y134 = y13 y14y34 /(y1 y3 y4 ),
y234 = y23 y24 y34 /(y2 y3 y4 ),
569
y1234 = y12 y13 y23 y14y24 y34 /(y1 y2 y3 y4 )2 .
Appendix C
The classical vacuum solutions of Section 5 have been cross-checked numerically.
We used Mathematicas routine NDSolve which allow good accuracy control through
the parameters Prec, AccuracyGoal, WorkingPrecision. The code for the
solution of boundary problem (35), (36) has been developed for all cases considered in
this paper; it is available at http://www.fis.unipr.it/~onofri.
The inherent instability of the boundary value problem (i.e., the existence of solutions
with exponential growth at infinity) makes the direct numerical problem rather illconditioned; however this fact is completely under control. The numerical solutions agree
with the exact ones until the unstable modes eventually take over by amplifying the
unavoidable truncation and algorithmic errors. In Fig. 5 we may see that the discrepancy
between exact and numerical solution eventually grows with exactly the slope dictated
by the square root of the largest eigenvalue of the mass matrix (the pattern is replicated
systematically for all the other Toda systems).
In case an exact solution were not available, a possible strategy, is given by backwardintegration:
starting from large values of y and assuming a general linear combination

= dj j exp{mj y} one can integrate back to y = 0 and fit the coefficients dj to
Fig. 5. The exponential growth of the difference between exact and numerical solution: the slope is given by the
square root of the highest eigenvalue of the mass matrix.
570
match the boundary conditions. This technique is not affected by the unstable modes and
can give rather accurate values for the asymptotic data.
References
[1] V. Fateev, S. Lukyanov, A. Zamolodchikov, Al. Zamolodchikov, Phys. Lett. B 406 (1997) 83;
V. Fateev, S. Lukyanov, A. Zamolodchikov, Al. Zamolodchikov, Nucl. Phys. B 516 (1998) 652.
[2] A. Zamolodchikov, Al. Zamolodchikov, Nucl. Phys. B 466 (1996) 577.
[3] C. Ahn, V. Fateev, C. Kim, C. Rim, B. Yang, Nucl. Phys. B 565 (2000) 611.
[4] Al. Zamolodchikov, Int. J. Mod. Phys. A 10 (1995) 112.
[5] V. Fateev, Phys. Lett. B 324 (1994) 45.
[6] V. Fateev, S. Lukyanov, Sov. Sci. Rev. A 212 (1990) 212.
[7] V. Fateev, Mod. Phys. Lett. A 15 (2000) 259.
[8] C. Destri, H. de Vega, Nucl. Phys. B 358 (1991) 251.
[9] V. Fateev, Normalization factors, reflection amplitudes and integrable systems, hep-th/0103014.
[10] S. Lukyanov, Phys. Lett. B 408 (1997) 192;
V. Brazhnikov, Nucl. Phys. B 542 (1999) 694.
[11] S. Ghosdal, A. Zamolodchikov, Int. Jour. Math. Phys. A 9 (1994) 3841.
[12] P. Bowcock, E. Corrigan, P.E. Dorey, R.H. Rietdijk, Nucl. Phys. B 445 (1995) 469.
[13] E. Corrigan, Int. J. Mod. Phys. A 13 (1998) 2709.
[14] G.M. Gandenberger, Nucl. Phys. B 542 (1999) 659.
[15] V. Fateev, A. Zamolodchikov, Al. Zamolodchikov, Boundary Liouville field theory 1. Boundary states and
boundary two-point functions, hep-th/0001012.
[16] P. Bowcock, JHEP 05 (1998) 8.
[17] V. Fateev, Mod. Phys. Lett. A 16 (2001) 1201.
[18] P. Bowcock, M. Perkins, Aspects of classical backgrounds and scattering for affine Toda theory on a half
line, hep-th/9909174.
[19] E. Corrigan, G.W. Delius, J. Phys. A 32 (1999) 159.
[20] E. Corrigan, A. Taormina, J. Phys. A 33 (2000) 8739.
[21] Al. Zamolodchikov, unpublished.
[22] Z. Bajnok, L. Palla, G. Takacs, Nucl. Phys. B 622 (2002) 565.
[23] A. Leclair, G. Mussardo, H. Saleur, S. Skorik, Nucl. Phys. B 453 (1995) 581.
[24] P.E. Dorey, A. Pocklington, R. Tateo, G. Watts, Nucl. Phys. B 525 (1998) 641.
[25] E. Corrigan, P.E. Dorey, R.H. Rietdijk, R. Sasaki, Phys. Lett. B 333 (1994) 83.
[26] G.W. Delius, G.M. Gandenberger, Nucl. Phys. B 554 (1999) 325.
[27] A. Fring, R. Kberle, Nucl. Phys. B 421 (1994) 159.
[28] H.W. Braden, E. Corrigan, P.E. Dorey, Nucl. Phys. B 338 (1999).
[29] P.E. Dorey, M. Pillin, R. Tateo, G.M.T. Watts, Nucl. Phys. B 594 (2001) 625.

Thermodynamics of the one-dimensional

multicomponent Bariev model
for correlated hopping
Rui-Hong Yue a,b , P. Schlottmann b
a Institute of Modern Physics, Northwest University, Xian, 710069, PR China
b Department of Physics, Florida State University, Tallahassee, FL 32306, USA
Abstract
The N-component Bariev model for correlated hopping in one dimension is investigated within
the framework of Bethes ansatz. The thermodynamic Bethe ansatz equations are derived based on
the string hypothesis for both, a repulsive and an attractive interaction. These equations are discussed
in several limiting cases, such as the ground state, high temperature limit, and weak and strong
coupling. 2002 Elsevier Science B.V. All rights reserved.
PACS: 71.10.Fd; 71.10.Pm; 71.27.+a
Keywords: Correlated hopping; Bethe ansatz; One-dimensional fermion lattice; Superconductivity
1. Introduction
The discovery of high-temperature superconductivity has further stimulated the study
of highly correlated systems. The quantum fluctuations originating from the strong
correlations between particles are particularly enhanced in low-dimensional systems. The
two-dimensional Hubbard and t J models are two excellent candidates to explain
the anomalous properties of high-Tc related materials [1,2]. In the Hubbard and t J
models, the quantum fluctuations are caused by an on-site Coulomb repulsion and
antiferromagnetic spin exchange between nearest neighbor electrons (with excluded
multiple site occupation), respectively. Of special interest are the one-dimensional versions
E-mail addresses: yue@phy.nwu.edu.cn, yuerh@phys.fsu.edu (R.-H. Yue), schlottm@phys.fsu.edu
(P. Schlottmann).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 3 3 2 - 2
572
R.-H. Yue, P. Schlottmann / Nuclear Physics B 634 [FS] (2002) 571590
of the Hubbard and (supersymmetric) t J models which are integrable and can be
diagonalized by means of nested Bethe ansatze [3,4].
A different approach to high-Tc was proposed by Hirsch [5]. The hopping of electrons or
holes is a function of the band filling and depends on the occupation of the final state. This
correlated hopping may be repulsive (i.e., it reduces the hopping amplitude) or attractive
(the amplitude is enhanced by the occupation). The latter gives rise to superconducting
fluctuations. The quantum fluctuations here are induced by the two-site interactions.
Although the one-dimensional Hirsch model is not integrable, there exists a related model
proposed by Bariev [6] which is integrable. Barievs model can be considered as half of
Hirschs model, since the correlations are in one direction only. Bariev [7] has extended
this model to include an arbitrary number of colors (components), thereby preserving the
integrability.
In this paper we investigate the equilibrium thermodynamics of the N -component
Bariev model. The Hamiltonian of the system is

L
N

cj, cj +1, + cj +1, cj, exp

nj +h( ), ,
H =
(1)
=
j =1 =1
where cj,
(cj, ) is the creation (annihilation) operator of an electron of component (or
color) located at site j and nj, is the corresponding number operator of electrons. Here
h(x) is the step function, i.e., h(x) = 1 for x > 0 and h(x) = 0 if x < 0. The parameter
describes the interaction strength. If = 0 the system reduces to an uncorrelated
tight binding band. Positive (negative) values of correspond to a repulsive (attractive)
interaction. Since for fermions the number operator is just one or zero, the interaction
terms can be rewritten as
Hint =
L
N

N1
l
cj, cj +1, + cj+1, cj,
e 1 Pl,j
,
j =1 =1
(2)
l=1
stands for the lth symmetric polynomial of (n

where Pl,j
j +1,1 , . . . , nj +1, 1 , nj, +1 , . . . ,
nj,N ). From Eq. (2) it is clear that the model includes multi-particle interactions.
Model (1) has been exactly diagonalized by means of nested Bethe ansatze in [7]. The
energy eigenvalue of the Hamiltonian is
E=
Ne
2 cos(kj ),
(3)
j =1
where the kj denote the wave number of the electrons (also called charge rapidities) and Ne
is the total number of electrons. In order to distinguish the different colors of the electrons
(l)
we have to introduce sets of spin (or color) rapidities { }, l = 1, . . . , N 1. If the number

(0) = N ,
of electrons of a given color l is Nl = M (l) M (l+1) 0, M (l) = N1
e
m=l Nm , M
(N)
= 0, then the charge and spin rapidities satisfy the following Bethe ansatz equations
M
(BAE)
e
ikj L
(1)
M

=1

e kj (1)
, ,

(2)
M

573
(1)
Ne
(1)
(1)

M

(2)

e ,
e (1)
e (1)
, 2 ,
kj , =
=1
M (l+1)
=
j =1
M (l1)
(l)
(l)
(l)
M

(l+1)
(l1)

=
(4)
e (l)
e (l)
, 2
=1
=1
=
with
e(x, a) =
sin(x/2 + ia)
,
sin(x/2 ia)
(5)
= /2 and L being the length of the chain.

The properties of the system can now be derived from the above Bethe ansatz equations.
The Bethe equations are valid for arbitrary , but for the Hamiltonian to be Hermitian it
requires that be real. The regime of real can be separated into two parts, namely 0
and 0, corresponding to a repulsive and attractive interaction, respectively. For the
repulsive case, the ground state and low-lying excitations at zero temperature were derived
in Refs. [6,8] for N = 2 and in Refs. [7,9] for general N . For the attractive case, the ground
state and low-lying excitations at zero temperature were studied in Ref. [10] just for N = 2.
Note that the attractive and repulsive cases are related by a particlehole transformation.
The mapping cj,

(1)j cj, transforms a system with coupling into one with ,
renormalizing the Hamiltonian by an overall factor of exp[(N 1)]. The classification
of Bethe states is, however, different for > 0 and < 0, such that it is useful to analyze
both cases separately.
The purpose of this paper is to investigate the equilibrium thermodynamics of the system
for both repulsive and attractive interaction. In Section 2 we classify all the solutions of
the BAE for positive according to the string hypothesis and derive the thermodynamic
Bethe ansatz (TBA) equations. In Section 3 we analyze these equations in some special
cases, such as the low and high temperature limits, the non-interacting limit and the very
strong coupling limit. In Section 4 we consider an attractive interaction. We obtain the
classification of all physical solutions of the BAE and the integral equations satisfied by the
corresponding densities. The TBA equations are derived by minimizing the free energy. In
Section 5 we discuss limiting cases of the TBA equations for an attractive . Conclusions
are presented in Section 6.
2. Thermodynamic Bethe equations for a repulsive interaction

In this section, we first classify the solutions of BAE for a repulsive interaction and then
we derive the TBA equations.
For a repulsive interaction, > 0, there are no Cooper-pair-like bound states among
electrons. Thus the charge rapidities kj must be all real. Mathematically this follows from
the first set of Eq. (4). Consider a complex charge rapidity with positive imaginary part. In
the thermodynamic limit, the left-hand side of the equation vanishes, while the right-hand
side has no zero. Hence, the equation is not satisfied if Im(kj ) > 0. With similar arguments
574
it is shown that if Im(kj ) < 0 the equation cannot be satisfied either. Hence, all kj have to
be real. This is similar to the multicomponent Fermi gas [1113].
(l)
The spin rapidities , on the other hand, can be complex, as can be seen from the
second and third sets of Eq. (4). By shifting a spin rapidity solution by 2i one obtains
another solution to the BAE. Hence, in the thermodynamic limit the spin rapidities form
strings of arbitrary length n (n = 1, 2, . . . , ), i.e.,
(l)
(l)
,n,s = ,n + i(n + 1 2s),
s = 1, 2, . . . , n,
n = 1, . . . , ,
(6)
(l)
where ,n are real parameters characterizing the motion of the center of mass of the
string of length n, with l labeling the level in the Bethe ansatz hierarchy, and being the
running index in the set. As shown below, the (l)
are all real in the ground state and the
complex solutions correspond to excited states. The string hypothesis for related models
was postulated by Takahashi [14,15].
Substituting the string solutions into the BAE, Eq. (4), we arrive at
(1)
M
n

2Ij = Lkj +
kj (1)
,n , n ,
j = 1, . . . , Ne ,
n=1 =1
(2)
(1)
2J,n
Ne
M
m

(1)

(2)
=
,n kj , n +
cnm (1)
,n ,m
j =1
m=1 =1
(1)
Mm

(1)

anm ,n (1)
,m ,
= 1, . . . , Mn(1),
m=1 =1
(l+1)
(l)
2J,n
M
m
(l1)
M
m
(l)

(l+1)
(l1)
+
cnm ,n ,m
cnm (l)
,n ,m
m=1 =1
m=1 =1
(l)
Mm

(l)

anm ,n (l)
,m ,
= 1, . . . , Mn(l) ,
l = 2, . . . , N 2,
m=1 =1
(N2)
(N1)
2J,n
M
m

cnm (N1)
(N2)
,m
,n
m=1 =1
(N1)
M
m
(N1)

,
anm ,n (N1)
,m
= 1, . . . , Mn(N1) ,
(7)
m=1 =1
(l)
where the quantum numbers Ij and J,n are sets of either integers or half-integers (they
arise from the multivaluedness of the logarithm). The remaining functions are defined by

(k, a) = 2 tan1 coth(a) tan(k/2) ,
cnm (x) =
min(n,m)
575

x, (n + m + 1 2s) ,
s=1
anm (x) =
min(n+1,m)
x, (n + m + 2 2s) +
s=1
min(n1,m)

x, (n + m 2s) ,
s=1
(8)
and the number of electrons with color l is given by

Nl = M (l) M (l+1) =
nMn(l)
n=1
nMn(l+1) .
(9)
n=1
In thermodynamic limit the size of the system L and the number of electrons of each
color Nl tend to infinity with their ratios remaining constant. The quantities Ij /L and
(l)
J,n /L can then be regarded as continuous variables and the summations over the running
indices in Eq. (7) can be replaced by integrals. We introduce the usual density functions
(l)
and n for the charge rapidities and the spin rapidity strings, respectively, and denote
(l)
with h and nh the densities for the corresponding holes. The integral equations satisfied
by the densities are formally similar to those of the multicomponent Fermi gas [1113],
except for the integration kernels and integration limits, i.e.,

2 () + h ()

d , n n(1) ,
=1+
2
n=1

(1)
(1)
n () + nh
()

d , n +

m(1) ,
d anm
m=1

m(2)
d cnm
m=1

(l)
2 n(l) () + nh ()

=
d cnm
m(l1) + m(l+1)
m=1

m(l) ,
d anm
m=1

(N1)
(N1)
n
() + nh
()

m=1

m(N2)
d cnm
576

m(N1) ,
d anm
(10)
m=1
and c denote the derivative of the functions with respect to the rapidity.
where , anm
nm
The integration limit is in all cases because all the rapidities appear in trigonometric
functions. After applying a Fourier series transformation,
1
f ()eix ,
f (x) =
2 =
f() =

f (x)eix dx,
(11)
we arrive at the following linearly coupled equations

,0 = ()
+ h ()
en|| n(1) (),
n=1
(1)
nh
() = en|| ()
nm () m(2) ()
C
m=1
(l)
() =
nh
(N1)
() =
A nm ()m(1) (),
m=1

nm () m(l1) () + m(l+1) ()
C
m=1
nh
A nm ()m(l) (),
m=1
nm ()m(N2) ()
C
m=1
A nm () m(N1) (),
(12)
m=1
where
nm () = sinh[min(n, m)] e max(n,m)|| ,
C
sinh()
nm ().
A nm () = 2 cosh()C
(13)
Here we used the identity

dx
0
a
cos(x)
=
1 + a 2 2a cos(x) 1 a 2
with a = e and being an integer.

The number of electrons of each color, the momentum and the energy are

Ne
= d (),
L

M (l)
=
n d n(l) (), l = 1, . . . , N 1,
L
n=1

E
P
= d (),
= 2 d () cos()
L
L
(14)
(15)
577
with M (0) = Ne . In order to conserve the number of electrons of each color we introduce
Lagrange multipliers, which correspond to the magnetic field, the chemical potential and
the crystalline field splittings. The Lagrange multiplier associated with M (l) is denoted Al .
The above equations are valid generally in the thermodynamic limit for an arbitrary
state. To impose thermal equilibrium we write the entropy as

S
=
( + h ) ln( + h ) ln h ln h d
L
N1

(q)
(q) (q)
(q)
(q)
(q)
(q)
(q)
n + nh ln n + nh n ln n nh ln nh d
+
n=1 q=1
(16)
and minimize the free energy F = E T S with respect to the densities subject to the
constraints enforced by the Lagrange multipliers and Eq. (12). We introduce a dressed
(q)
(q)
(q)
energy for each class of states, i.e., 0 = T ln(h /), n = T ln(nh /n ), and denote
(q)
(q)
n = exp(n /T ). The TBA equations are now
0() = 2 cos A0 T

1
,
d C1n ln 1 + n(1)
n=1

(1) 1

nA1
ln 1 + n(1) () =
+
d Anm ln 1 + m
T
m=1

d C1n ln 1 + exp 0 /T

(2) 1
,
d Cnm ln 1 + m
m=1

(l) 1

nAl
(l)
+
d Anm ln 1 + m
ln 1 + n () =
T
m=1

d Cnm
m=1
(l1) 1
(l+1) 1
ln 1 + m
1 + m
(17)
(N)
where n = is assumed for all n, and the dressed energies for the strings satisfy the
asymptotic boundary condition
lim
1 (l)
= Al .
n n
The free energy is given by

F
d
= T
ln 1 + exp 0()/T
L
2
(18)
(19)
578
and the density functions can be recovered by differentiating with respect to the chemical
potential A0 ,
=
1 0
1
,
1 + exp(0/T )
2
A0
n(l) =
1 n(l)
1
.
1 + n(l)
2
A0
(20)
(l)
An alternative expression for the equations for n is,

1

ln 1 + n(l) ()
N1

(q)
d Gclq ln 1 + n
=
q=1
+ n,1
N1

d FNl ln 1 + exp 0 /T

(q)
(q)
d Glq ln 1 + n+1 1 + n1 ,
(21)
q=1
(l)
where 0 0 and the integration kernels are the inverse Fourier series transforms of
lq () = sinh[min(l, q)] sinh[(N max(l, q))] ,
G
sinh() sinh(N)
c () = 2 cosh()G
l () = sinh(l) ,
lq ().
G
F
lq
sinh(N)
(22)
3. Special limits: repulsive interaction

3.1. Ground state
In the very low temperature limit only states with negative dressed energies are occupied
as a consequence of the Fermi statistics of the rapidities. States with positive dressed
energies are empty and the zeroes of the dressed energies define the Fermi energies.
We decompose 0 and n(l) into its positive and negative parts, 0 = 0 + + 0 and n(l) =
n(l)+ + n(l) , such that 0 + 0, n(l)+ 0 and 0 < 0, n(l) < 0. From Eq. (17) we now
obtain as T 0

0() = 2 cos() A0 +
d C1n n(1) ,
n(l)+ () = nAl
+

m=1
n=1

(l)
d Anm m
m=1

(l+1)

(l1)
+ m
,
d Cnm m
(23)
579
where we used the notation n

n,1 0 and n
0. Noting that Al 0, the
asymptotic conditions for the dressed energies, limn n(l) /n = Al , yield that the
(l)
n > 0 for n > 1 and all l. Hence, these states are not populated in the ground state.
Consequently, in the ground state there are no strings and all rapidities are real.
Without ambiguity we may suppress the subindex n = 1 in 1(l) , i.e., denote it by (l) ,
and write Eq. (23) as
(0)
B1
0() = 2 cos() A0 +
(N)

d C11 (1) ,
B1
Bl
(l)+
() = Al

d A11 (l)
Bl
B
l+1

d C11 (l+1)
Bl+1
B
l1

d C11 (l1) .
(24)
Bl1
From inspection of Eq. (24) we see that if all colors have the same population (no magnetic
or crystalline fields, i.e., Al = 0 for l 1) then Bl = for l 1, but B0 Q remains
arbitrary. Consequently there are no holes among the spin rapidities and (l)+ () 0 for
all l, such that the spin rapidities can be eliminated via Fourier transformation. The final
result is
0() = 2 cos() A0
Q
+
Q
d in |n| sinh[(N 1)n]

0
,
e
2
sinh(Nn)
n=
1
= () + h ()
2
Q
d in |n| sinh[(N 1)n]

.
e
2
sinh(Nn)
n=
(25)
Note that Q and A0 are related through 0(Q) = 0. The above results agree with those of
Ref. [9].
For an arbitrary density of electrons (general Q), the above equation must be solved
numerically. The solution yields the elementary charge excitation energy and momentum
580
for both particles |0 | > Q and holes |0 | < Q,

0

pch (0 ) = 2 d () + h () ,

7Ech (0 ) = 0(0 ),
(26)
and the spin wave excitation energy and momentum for the lth-color hole

(l)
(0 ) = (l) (0 ),
7Esp
0
(l)
psp
(0 ) = 2
d (l)(),
(27)
where 0 is the rapidity parametrizing the excitation. The Fermi velocity for the charges
can be obtained through

d7Ech (0 ) dpch (0 )
.
vF =
(28)
d0
d0
Q
Q
As found in Ref. [9] the spin excitations are gapped in zero-field. Hence, they do not have
a Fermi surface and do contribute to the low temperature properties with an exponential
activation. The gaps can be closed with external fields. The ground state susceptibility
is zero, since the system does not respond to fields smaller than a critical one. The gaps
depend on and the density of electrons. As 0 or as the electron density tends to zero,
also the gaps tend to zero. Performing a Sommerfeld expansion of the free energy and the
dressed energies we obtain that the specific heat is proportional to T . The coefficient
of the specific heat is determined by the low-energy excitations, i.e., the group velocity
vF [13],
1
=
,
L
3 vF
in agreement with the result from conformal field theory.
(29)
3.2. Special cases 0 and

For general electron density and finite interaction there is no analytic solution and the
equations have to be solved numerically. However, an analytic solution can be obtained in
the free particle limit 0 and the strong repulsive limit . The final results for
these cases are

N
2 cos() + A 0 , || Q,
0=
2 cos() + A0 ,
|| > Q,
N
1
=
(30)
,
|| Q,
h =
,
|| > Q,
2
2
if = 0, and
0 = 2 cos() A0
+ h =
(N 1)(2 sin Q + QA0 )

,
N (N 1)Q
N
,
2[N (N 1)Q]
all ,
(31)
if = . The ground state energy, number of electrons and chemical potential are

E
[2N sin Q]/,
= 0,
=
[2N sin Q]/ N (N 1)Q , = ,
L

Ne
NQ/,
= 0,
=
NQ/ N (N 1)Q , = ,
L

2 cos Q,
= 0,
= A0 = 2 cos(Q) 2(N1)Q sin(Q) cos(Q), = .
N
581
(32)
Note that in the free particle limit the results reduce to the multicomponent Fermi gas case
[11,13] with the approximations sin Q Q and cos Q Q2 /2 (neglecting the constant
term in both the Hamiltonian and the energy). For large the density of states is enhanced,
depending on the band-filling, because the correlations reduce the effective band-width. In
this limit the electrons effectively act like hard core bosons.
3.3. High-temperature limit
In the high temperature limit we assume that the band width and the chemical
potential are much smaller than T . The magnetic field and crystalline fields, however,
are comparable to T . In this limit the dressed energies can be assumed constant, such that
all integrals in Eq. (17) can be carried out, and we obtain the following set of algebraic
equations

j(l)
2
(l)
(l)
(1 + j +1 )(1 + j 1 )
(l+1) 1
(l1) 1
) ][1 + (j
) ]
(1)
(1)2
(1 + 2 )
=
,
1
(2)
[1 + (1 )1 ](1 + 1 )
(l)

1
lH
n
=
,
1 + n(l) ,
lim
=
n n
T
n=1
[1 + (j
(33)
where = exp(0/T ) and n(0) = n(N) = . The corresponding equations for the densities
are

1
+
(1 + ) =
m(1) ,
2
m=1

(1) (1)
(1) (1)
(1) (1)
2 1 n n = n(2) n+1 n+1 n1 n1 n,1 ,

(l)
(l)
(l)
(l)
2 1 n(l) n(l) = n(l+1) + n(l1) n+1
(34)
n+1
n1
n1
,
(0)
(N)
(l) (l)
with n = n = 0 0 = 0. The solution of these equations is very involved for

N > 2. For N = 2 the solution is similar to the one given in [14,15]. For N = 2 we may
suppress the superscripts (1) of and without ambiguity and for A0 = 0 we obtain
n =
sinh2 [(2n + 1)a]

2
sinh (2a)
1,
sinh(a)
,
sinh(3a)
582
1
,
2
n =
cosh[(2n + 1)a] sinh2 (a) sinh(2a)

,
sinh[(2n 1)a] sinh[(2n + 1)a] sinh[(2n + 3)a]
(35)
where a = H /(4T ). The free energy of system is F /L = 2T ln[sinh(2a)/ sinh(a)],

the density of electrons is 1 (the chemical potential is zero) and the magnetization is
S z /L = 12 tanh(a), i.e., there are four states corresponding to occupation 0, 1 and 2 per
site. In the zero field limit the free energy is just T ln(4) and the magnetization vanishes.
4. Thermodynamic Bethe equations for an attractive interaction

An attractive interaction ( < 0) tends to form bound states of particles with different
colors, in analogy to Cooper pairs in the case of N = 2. These bound states are described
by solutions of the Bethe ansatz equations with complex charge rapidities. Mathematically,
this is seen from the first set of Eq. (4). As L the left-hand side of the equation
vanishes, while the right-hand side has a zero if kj = (1) + i||. Similarly, kj = (1)
i|| is also a solution. Hence, the charge rapidities are related to spin rapidities. The spin
rapidities themselves can be complex and form strings. This gives rise to charge bound
states. The size of the bound state is limited by the number of different colors. Real charge
rapidities are also solutions of the Bethe ansatz equations. As for the repulsive case, the
spin rapidities can also form strings of complex momenta, which are independent of the
charge rapidities. The general structure of solutions of Eq. (4) in the thermodynamic limit
can be summarized as follows [1113]
kj = (n1) + i(n 1 2s), s = 0, . . . , n 1, n = 1, . . . , N,
(n1) + i(n 1 l 2s), s = 0, . . . , n 1 l,
(l)
=
1 l n 1, n = 1, . . . , N,
(l)
n + i(n 1 2s), s = 0, . . . , n 1, n = 1, . . . , .
(36)
Inserting these solutions into Eq. (4), we obtain the following equations
l + 1 = 2
(l)

() + h(l) () +

d , m| | m(l+1)
m=1
N1

d (q)
q=0
min(l,q)lq

, |q l| + 2lq + 2 | | ,
=0
l = 0, . . . , N 1,

(l)
n,h
() = d C1n (l1)
d Anm m(l)
+
m=1

d Cnm m(l+1) + m(l1) ,
m=1
l = 1, . . . , N 1,
(37)

(0)
583
(N)
where n = n 0, all the integration kernels are defined the same way as for the
repulsive case, (l) and h(l) are the densities for the rapidities (l) and the corresponding
(l)
the distribution functions for the (l)
holes, and n(l) and n,h
n and their holes, respectively.
The energy of system is given by
sinh[(l + 1)]
E
= 2
L
sinh()
N1

d cos() (l) ().
(38)
l=0
With the particle and hole distribution entropy,

N1

S
=
d (l) ln 1 + h(l) (l) + h(l) ln 1 + (l) h(l)
L
q=0
N1

(l)

(l)
(l)
d m(l) ln 1 + mh m(l) + mh ln 1 + m(l) mh ,
(39)
q=1 m=1
the free energy can be expressed as F = E T S
N1
q=0
Aq Nq with

N1

(q)

Nq
(q+1)
=
m d m () m
() .
d (l) () +
L
l=q
(40)
m=1
Here the parameters Aq are Lagrange multipliers to enforce the conservation of particles
of each color. Minimizing the free energy under the constraint (37) yields
sinh[(l + 1)]
Aq
sinh()
N1
0 (l) () = 2 cos()
q=l
+T
N1

d ln 1 + exp 0 (q) /T
q=0
1 in( )|n|[max(l,q)+lq ] sinh[(1 + min(l, q) lq )n]

e
2 n=
sinh(n)

(l+1) 1
d
, m| | ln 1 + m
,
T
2
m=1
0 l N 1,
584

T ln 1 + n(l) () = n(Al Al1 )

+ T d C1n ln 1 + exp 0 (l1) /T
+T
T

m=1

m=1

(l) 1
d Anm ln 1 + m

(l+1) 1
d Cnm ln 1 + m
(l1) 1
,
1 + m
1 l N 1,
(41)
(l)
where the integral kernels are same as for the repulsive case and exp(0 (l) /T ) = h / (l) =
(l)
(l)
(l)
(l)
(l)
(l) and n = nh /n = exp(n /T ). The equation for n can be rewritten as

1
n(l) () = n,1 T d G0 ln 1 + (l1)

+T

d G0 ln
[1 + n+1 ( )][1 + n1 ( )]
(l)
(l)

,
[1 + (n(l+1)( ))1 ][1 + (n(l1) ( ))1 ]

(42)
0 (n) = 1/[2 cosh(n)], and (l) 0

where G0 is the inverse Fourier series transform of G
0
(0)
(N)
and m = m . The free energy of system is

N1

F
d
= T
ln 1 + exp 0 (q) ()/T ,
(q + 1)
L
2
(43)
q=0
and the
densities can be obtained by differentiating with respect to the chemical potential
= N1
q=0 Aq /N ,
(l) =

1 0 (l)
1
1 + exp 0 (l) /T
,
2
1 n
1
1 + n(l)
.
2
(l)
n(l) =
(44)
5. Special limits: attractive interaction

5.1. Ground state
In the low-temperature limit we decompose the 0 (l) and n(l) into their positive and
negative parts, i.e.,
0 (l) = 0 (l)+ + 0 (l) ,
n(l) = n(l)+ + n(l) ,
0 (l)+ > 0,
n(l)+ > 0,
0 (l) 0,
n(l) 0.
(45)
585
Substituting into Eq. (41) we obtain in the limit T 0

n(l) = n,1 d G0 0 (l1)

(l)+

(l)+
+ d G0 n+1 + n1 + n(l+1) + n(l1) ,
(l)
n
= Al1 Al 0.
n n
lim
(46)
(l)
The physical solution of the above equations is n 0 for all n and l. The integral
equations satisfied by the dressed energies 0 (l) are then
0 (l)+ () =
N1
Aq + (l + 1)
q=l
N1
1
Aq
N
q=0
N2
Bq
q=0 B
exp

0 (q) Gl+1,q+1 d

0 (N1)+ Fl+1 d ,
+
| |>BN1
0 (N1) () = 2e(N1)|| cos()
N1
1
Aq
N
q=0
N2
Bq
q=0B

0 (q) Fq+1 d

0 (N1)+ HN1 d ,
+
| |>B
(47)
N1
and similar equations are satisfied by the distribution functions. The integration limits are
determined by the Fermi surfaces of the excitations, i.e., 0 (q) (Bq ) = 0. Here Gexp and
exp () = G
l,q () exp(||)
HN1 are defined as the inverse Fourier series transform of G
l,q

and HN1 () = FN1 () exp(||), respectively.
In the absence of external fields the number of electrons with each color is the same,
such that Bq = 0 for q < N 1, and the equations for the ground state reduce to
0 (N1) () = 2e(N1)|| cos()

+
| |>BN1
N1
1
Aq
N
q=0

0 (N1)+ HN1 d ,
586

(N1)
(N1) () + h
() =
1
+
2
(N1)

HN1 d .
| |>BN1
(48)
The solution dependson BN1 which is related to the number of electrons Ne and the
chemical potential N1 N1
q=0 Aq . These equations have in general to be solved numerically.
For two colors, i.e., N = 2, Eq. (48) reduces to the result of Ref. [10]. The exponential
factor exp[(N 1)] reflects the change in energy scale due to the particlehole
transformation mentioned in Introduction.
In the absence of external fields all 0 (l) -excitations for l = 0, . . . , N 2, are gapped,
i.e., 0 (l) > 0. Physically, this means that a finite external field larger than a critical field,
is needed to break up the bound states. Each branch has its critical field closing the
corresponding gap. The only branch with Fermi points is the 0 (N1) -band. The zerotemperature magnetic susceptibility vanishes for fields smaller than the smallest critical
field. At low T the susceptibility is exponentially activated as a consequence of the spin
gaps. This is reminiscent of the Meissner effect in superconductivity. Here the bound states
(generalized Cooper pairs) exist at every temperature, i.e., in contrast to the BCS-theory
they are preformed bound states.
The charge sector, on the other hand, has a gapless spectrum. The system then
behaves like a one-component Luttinger liquid. The gaps vanish for || 0 and grow
monotonically with ||. The specific heat at low T is proportional to T . This can be
obtained from a Sommerfeld expansion of the free energy and the dressed energy of the
charges. The -coefficient of the specific heat is /L = /(3vc ), where vc is the group
velocity of the charges defined similarly to the repulsive case.
5.2. Special cases 0 and
In both, the free particle limit and the strongly attractive case, Eq. (48) can be solved
exactly. We limit ourselves to present the final results for the free particle limit = 0

2 cos() ,
|| BN1 ,
(N1)
0
() =
2N cos() N, || > BN1 ,
(N1) () =
1
,
2
|| BN1 ,
2
E
= N sin(BN1 ),
L
(N1)
N
,
|| > BN1 ,
2
NBN1
Ne
=
,
(49)
L
() =
= 2 cos(BN1 ),
and for the strongly attractive limit

(N1)
(N1) () + h
() =
N
1
,
2 + (N 1)BN1
2N sin(BN1 ) (N1)||
E
=
e
,
L
+ (N 1)BN1

2
(N 1)
= e(N1)|| cos(BN1 ) +
BN1 cos(BN1 ) sin(BN1 ) . (50)
N
587
Note that in this case the correlations reduce the density of states by a factor that depends
on the band-filling. The exponential factor exp[(N 1)] reflects the change in energy
scale due to the particlehole transformation mentioned in Introduction.
We now compare the results for the free particle limits for the repulsive and attractive
cases. The energy, the number of electrons and chemical potential are indeed the same. In
the repulsive case, the density = N/(2) stands for the total density of electrons, and
hence the density of electrons with one color is then 1/(2). On the other hand, in the
attractive case the density (N1) = 1/(2) corresponds to the density of electrons of one
color. Thus the two limits yield the same results.
5.3. High temperature limit
We consider temperatures much higher than the band width, an interaction parameter
limited so that e(N1) T , and Lagrange multipliers given by A0 = (N 1)H /2
and Al = lH + A0 . Under these assumption, Eq. (41) can be reduced to

n(l)
2
(l)
(l)
][1 + n1
]
[1 + n+1
(l+1) 1
(l1)
) ][1 + (n )1 ]
[1 + (n
1 l N 1,
n = 1, . . . , ,
N1

1 min(l,q)+1
1 + (q)
1 + (l) = e(l+1)/T e(l+1)(Nl1)H/T
q=0

1 1
1 + n(l+1)
,
0 l N 1,
(51)
n=1
with 0 = ( (l1))1 and n = n

(l)
(0)
(N)
= , and
N1

l +1
= (l) (l) +
(q) 1 + min(l, q) +
n(l+1) , 0 l N 1,
2
q=0
n=1

(l) (l)
(l) (l)
n+1 + n1
n1 + n(l+1) + n(l1) ,
2n(l) 1 + n(l) = n+1
1 < n, 1 l N 1,

21(l) 1 + 1(l) = 2(l) 2(l) + (l1) + 1(l+1) + 1(l1) ,
(0)
(N)
1 l N 1,
(52)
with n = n = 0. Although these equations are algebraic, it is very hard to find

a solution except in special limits. We limit ourselves to N = 2, where we may drop
the superscript in (l) and (l) without ambiguity. In this limit we obtain the following
recursion relations
(n )2 = (1 + n+1 )(1 + n1 ), n > 1,
1

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

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

1
(1)2
= e2/T 1 + (1) 1 + (0)
,
588
1
1 + (n )1 ,
(0) = e/T eH/(2T ) 1 + (1)
(53)
n=1
which have the following solution

n = f 2 (n) 1,
1
(0) = f 2 (0) 1 ,
where
sinh(b + an)
,
f (n) =
sinh(a)

exp(b 2a) =
(1) = f 2 (1) 1,
1 + ea e/T
1 + ea e/T
(54)
(55)
with a = H /(2T ). Inserting this result into Eq. (53) we obtain the densities

cosh(b)f (1)
cosh[b a]f 2 (1)

f (2) ,
(1) =
,
f (0)
f (0)

1
1
1
f (1) f (2)
n =
,
2
f (n 1)f (n) f (n)f (n + 1)
(0) =
(56)
and the energy, the magnetization and particle number are given by
sinh(b a) sinh(b 2a)
Sz
=
,
L
sinh(a)
E
Ne
Sz
=
+H .
L
L
L
The free energy is given by

sinh(a)
F
.
= 2T ln
L
sinh(2a b)
Ne
2 sinh(b a) cosh(b 2a)
=
,
L
sinh(a)
(57)
(58)
For zero magnetic field and zero chemical potential, the free energy is just T ln(4), i.e.,
there are four degrees of freedom per site corresponding to the empty state, the singly
occupied states (spin-up and spin-down) and the doubly occupied state.
6. Conclusions
In strongly correlated electron systems the hopping of the electrons may be reduced or
enhanced depending on the occupation of the final state. Bariev has proposed a model for
correlated hopping involving many-site interactions. The special interest in Barievs model
lies in its integrability in one dimension, i.e., the model is soluble in terms of a sequence of
nested Bethe ansatze. We studied the thermodynamics of the model for N spin components,
arbitrary band filling and as a function of a coupling constant , which can be attractive or
repulsive. We classified all the states according to the string hypothesis and obtained the
thermodynamic Bethe ansatz equations. We further analyzed the TBA equations in several
limiting cases, i.e., the T 0, 0, || and high temperature limits.
As mentioned in the Introduction, there is a particlehole transformation relating the
attractive and repulsive cases. However, the classification of the Bethe states is very
589
different for the two representations. For this reason we have studied the thermodynamics
within both frameworks. Although the equations are formally very different, they actually
represent the same model with only a rescaling factor of all energies of exp[(N 1)].
For a repulsive interaction there are no bound states in the ground state and the system
consists of freely propagating charges (characterized by real charge rapidities k) and spin
waves (described by the real spin rapidities). In the absence of external fields the spin
excitations are gapped. Hence, the zero-field magnetic susceptibility of the ground state
vanishes and the specific heat is linear in the temperature. The band of charge rapidities
has two well-defined Fermi points and the elementary charge excitations from the ground
state are of the particlehole type. In the limit of a very strong repulsion the charges
behave like hard-core bosons and an effective Pauli principle is obtained for all particles,
independently of their colors. In the repulsive case the hopping between sites is inhibited
by the occupation of the neighboring sites.
For an attractive interaction, on the other hand, the hopping is favored by the occupation
of a neighboring site. This leads to bound states between electrons with different spin
components, which then tend to hop together in a coherent way. In the absence of symmetry
breaking fields (magnetic or crystalline fields) the ground state consists of spin-neutral
charge bound states of N electrons, each particle with a different spin component. Hence,
for N = 2 these bound states can be interpreted as Cooper pairs. Although these charge
bound states are the consequence of a coherent collective state, there is no long-range order
in the system at T = 0. These charge bound states have a finite binding energy, i.e., there
are spin gaps, which can be closed by a magnetic field. There is a critical field required
to depair the bound state (reminiscent of the Meissner effect in the case of Cooper pairs).
These bound states exist at every temperature and not just in the ground state (preformed
Cooper pairs). The T = 0 magnetic susceptibility is zero (there is no response to a field
less than the critical one), but the low-T specific heat is proportional to T .
Acknowledgements
The support by the National Science Foundation under grant No. DMR01-05431 and
the Department of Energy under grant No. DE-FG02-98ER45797 is acknowledged. R.Y.
is also supported in part by the Chinese Scholarship Council and the NSFC under grant
19975036. R.Y. would like to thank the Department of Physics of Florida State University
for the hospitality.
References
[1]
[2]
[3]
[4]
[5]
P.W. Anderson, Science 235 (1987) 1196.

F.C. Zhang, T.M. Rice, Phys. Rev. B 37 (1988) 3759.
E.H. Lieb, F.Y. Wu, Phys. Rev. Lett. 20 (1968) 1445.
P. Schlottmann, Phys. Rev. B 36 (1987) 5177.
J.E. Hirsch, Phys. Lett. A 134 (1989) 451;
J.E. Hirsch, Physica C 158 (1989) 326.
[6] R.Z. Bariev, J. Phys. A: Math. Gen. 24 (1991) L549.
590
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
R.Z. Bariev, J. Phys. A: Math. Gen. 24 (1991) L919.

R.Z. Bariev, J. Phys. A: Math. Gen. 25 (1992) L261.
R.Z. Bariev, A. Klmper, A. Schadschneider, J. Zittartz, J. Phys. A: Math. Gen. 26 (1993) 4863.
R.Z. Bariev, A. Klmper, A. Schadschneider, J. Zittartz, J. Phys. A: Math. Gen. 26 (1993) 1249.
P. Schlottmann, J. Phys. Condens. Matter 5 (1993) 5869.
P. Schlottmann, J. Phys. Condens. Matter 6 (1994) 1359.
P. Schlottmann, Int. J. Mod. Phys. B 11 (1997) 355.
M. Takahashi, Prog. Theor. Phys. 46 (1971) 401.
M. Takahashi, Prog. Theor. Phys. 46 (1971) 1388.

The XXX spin s quantum chain and the alternating

s 1 , s 2 chain with boundaries
Anastasia Doikou
Department of Mathematics, University of York, Heslington, York YO10 5DD, United Kingdom
Received 4 January 2002; accepted 22 March 2002
Abstract
The integrable XXX spin s quantum chain and the alternating s 1 , s 2 (s 1 s 2 = 1/2) chain with
boundaries are considered. The scattering of their excitations with the boundaries via the Bethe
ansatz method is studied, and the exact boundary S matrices are computed in the limit s, s 1,2 .
Moreover, the connection of these models with the SU(2) Principal Chiral, WZW and the RSOS
models is discussed. 2002 Elsevier Science B.V. All rights reserved.
1. Introduction and review

Quantum spin chains are 1D quantum mechanical models of N microscopic degrees
of freedom, namely spins, with the Heisenberg (XXX) model solved by Bethe [13]
being the prototype. Two types of quantum spin chains exist, known as closed (periodic
boundary conditions) and open. Closed spin chains have been very well studied (see,
e.g., [24]), whereas open chains are less widely studied, although of great interest. A very
interesting aspect of these models is that in the continuum limit, they correspond to (1 + 1)dimensional integrable quantum field theories [5]. For example, the S matrix that describes
the anisotropic Heisenberg (XXZ) model in a certain regime also describes a massive
integrable quantum field theory, namely the Thirring model or the sine-Gordon model.
Also, the S matrix that describes the bulk scattering for the spin s XXX chain coincides
with the one of the WessZuminoWitten (WZW) model [6]. Finally, open spin chains
and the corresponding 2D field theories with boundaries display a rich variety of boundary
phenomena whichfor integrable modelscan be investigated exactly.
E-mail address: ad22@york.ac.uk (A. Doikou).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 5 4 - 7
592
A. Doikou / Nuclear Physics B 634 [FS] (2002) 591610
During the last years, there has been an increased research interest on integrable models
with boundaries, especially after the prototype work of Cherednik and Sklyanin [7,8].
In particular, Cherednik introduced the reflection equation in order to obtain boundary
scattering matrices for theories in half line. In the presence of boundaries in addition
to the bulk scattering, the scattering of the particles with the boundariesdescribed by
the boundary S matrixshould also be considered. The boundary S matrix satisfies a
collection of algebraic constraints namely the boundary YangBaxter (reflection) equation
[7]. Sklyanin considered this equation in the spin chain framework and he generalized
the Quantum Inverse Scattering Method for integrable models with boundaries. Moreover,
Fring and Kberle [9] obtained solutions of the reflection equation for the affine Toda
field theories with boundaries, whereas Ghoshal and Zamolodchikov [10] found solutions
of the reflection equation for the sine-Gordon model on a half line. Finally, de Vega and
(1)
Gonzalez-Ruiz made similar calculations for the XXX (SU(2)), XXZ (A1 ), XY Z and
(1)
any SU(n), An1 open spin chain [11].
In this study the integrable XXX spin s quantum chain and the alternating s 1 , s 2
chain with boundariesobtained by fusionare explored. As a warm up, both models
with periodic boundaries conditions are reviewed and their relation with 2D quantum field
theories is discussed.
To describe the models it is necessary to introduce the basic constructing element,
namely the R matrix, which is a solution of the YangBaxter equation [12,13]
R12 (1 2 ) R13 (1 ) R23 (2 ) = R23 (2 ) R13 (1 ) R12 (1 2 ).
(1.1)
We focus on the special case where the R matrix is related to the spin s representation of
SU(2), obtained by fusion [14],

R0k =
w Sk
w0 w3 Sk3
w0 + w3 Sk3
w+ Sk+

,
(1.2)
where w0 = + i/2, w3 = w = i and S 3 , S are the SU(2) generators in the spin

s representation and act, in general, on a (2s + 1)-dimensional space V = C 2s+1 . The
generators satisfy the following commutation relations

S + , S = 2S 3 ,

S 3 , S = S
(1.3)
and for, e.g., s = 1 they become,

Sk3
0
1
Sk+
= 2

1 ,
Sk
= 2 1

. (1.4)
1
For s = 1/2 we obtain the well-known XXX R matrix,

R12 ()jj,jj = ( + i),
R12 ()j k,j k = ,
R12 ()j k,kj = i,
1 j, k 2.
j = k,
j = k,
(1.5)
593
1.1. The XXX spin s quantum chain

The XXX (and XXZ) spin s integrable chain has been intensively studied by several
authors in the bulk (see, e.g., [1421]). To construct the model we derive the transfer matrix
considering the R matrix related to the spin s representation of SU(2) (1.2). We introduce a
mass scale in our system therefore, we derive the transfer matrix with inhomogeneities ,
t () = tr0 T0 (),
(1.6)
where
2s
2s
2s
2s
T0 () = R02N
( )R02N1
( + ) R02
( )R01
( + ),
(1.7)
is given by (1.2) and acts on V0 Vi . The auxiliary space V0 is a two-dimensional

space whereas the quantum space Vi , i = 1, . . . , 2N , is a (2s + 1)-dimensional space.
After we diagonalize the transfer matrix we find the following eigenvalues [14]

M

i N ( j 2i )
i N
2s
1 () = + is +
+ + is +
2
2
( j + 2i )
j =1
2s
R0i

M
i N
i N ( j + 3i2 )
.
+ is +
+ is +
2
2
( j + 2i )
j =1
(1.8)
The corresponding Bethe ansatz equations are,

eq ( )N eq ( + )N =
M

e2 ( ),
(1.9)
=1
where q = 2s and
en () =
in
2
in
2
The BAE (1.9) are similar to the ones found in [16] for the lattice analogue of the SU(2)
PCM.
In general, if we fuse the auxiliary space as well, we find that the eigenvalues of the
transfer matrix are [14],
2s
l () =
l

ak ( )N ak ( + )N
k=0
M

j =1
( j 2i )
( j + il + 2i )
( j + i(k 1) + 2i ) ( j + ik + 2i )
(1.10)
where
ak () =
k1

l1

i
i
is + + im
+ is + in +
.
2
2
m=k
n=0
(1.11)
594
Let us consider the special case where l = 2s and = + 0 = 2i (2s 1). Then
R 2s (0 ) becomes the permutation operator, therefore, we can derive a local Hamiltonian
for the system (up to an additive constant)

2s

2s

i d
i d

log t ()
log t ()
+
.
H
(1.12)
2 d
2 d
=+0
=+0
The corresponding eigenavalues follow from (1.10) and (1.12)
E=
2 M
1
n
2
n=1 j =1 (j + () +
q
iq
2 )(j
+ ()n
iq
2)
(1.13)
also the momentum and spin are given by

P=
2 M
(j + ()n +
1
log
2i
(j + ()n
n=1 j =1
iq
2)
,
iq
2)
S z = qN M.
(1.14)
(1.15)
The ground state and the low lying excitations of the model can be studied, in the
thermodynamic limit N . In this limit, the solutions of the Bethe ansatz equations
are given by the so-called string hypothesis, i.e., the solutions of (1.9) are grouped into
strings of length n with the same real part and equidistant imaginary parts
i
)
= n + (n + 1 2j ), j = 1, 2, . . . , n,
(n,j
(1.16)
2
where n is real. It is known [1418], that the ground state, i.e., the state with the least
energy, is the filled Dirac sea with strings of length q = 2s. The low lying excitations are
holes in the q sea and also strings of length n = q. One can study the scattering among
the low lying excitations of the model and show [1421] that the S matrix, as s ,
coincides with the one of the SU(2) PCM, i.e., it is the SSU(2) SSU(2) S matrix [22,23].
The SU(2) scattering amplitudes (for the singlet triplet and singlet, respectively) are

/2 d
+i
i e
,
S+ () =
S0 ().
e
S0 () = exp
(1.17)
2 cosh( 2 )
i
In the scaling limit, , the excitations become massive relativistic particles [18], with
energy and momentum,
)() 2e cosh ,
p() 2e sinh ,
(1.18)
where the momentum p is defined mod for even number of excitations.

Faddeev and Reshetikhin showed in [18] that for finite s, inconsistencies in the counting
of the states exist. Therefore, the interpretation of the S matrix as SSG (s)SSU(2) somehow
fails; SSG (s) is the sine-Gordon S matrix [24], with the triplet amplitude being

i sinh((2s 1) 2 ) d
S0 () = exp
(1.19)
e
,
2 cosh( 2 ) sinh(2s 2 )
595
and obviously SSG (s ) SSU(2) . However, Reshetikhin conjectured in [6] that the
correct S matrix for the spin s chain is the SRSOS (s) SSU(2) matrix which coincides with
the S matrix of the WZW model at level k = 2s (WZWk ) [6,22,23]. The SRSOS (s) is the
scattering matrix of the RSOS model, and the spin s is related to the restriction parameter
r of the RSOS model, i.e., r = 2s + 2 (for a more detailed analysis see [6,25]). More
specifically, the spin s = 1 chain has a hidden supersymmetry, which is described by the
RSOS part of the S matrix [6].
A key observation is that, as k = 2s , the SRSOS (s) SSU(2) matrix degenerates
to a tensor product of two rational matrices, and it coincides with the S matrix of the
SU(2) PCM without topological term, found in [16,18,21]. The later comment reflects the
fact that the perturbed WZWk as k , reduces to the PCM without topological term
(SRSOS (s ) reduces to SSU(2) ), as described in [26].
1.2. The alternating s 1 , s 2 spin chain
Alternating spin chains have been originally introduced by de Vega and Woyanorovich
in [27] and they have been also studied in the bulk by several authors (see, e.g., [2832]).
We define the transfer matrix of the chain with inhomogeneities,
t = tr0 T0 (),
(1.20)
where
1
2
1
2
( )R02N1
( + ) R02
( )R01
( + ),
T0 () = R02N
(1.21)
and R i is related to the spin s i (i = 1, 2) representation (1.2). The eigenavalues of the

transfer matrix, after we fuse the auxiliary space as well, are given by (see also [27])
l(1,2)() =
l

ak(1)( )N ak(2)( + )N
k=0
M

j =1
( j 2i )
( j + il + 2i )
( j + i(k 1) + 2i ) ( j + ik + 2i )
(1.22)
where
(j )
ak () =
k1

l1

i
i
j
j
is + + im
+ is + in +
.
2
2
m=k
(1.23)
n=0
The corresponding Bethe ansatz equations have the form,

eq 1 ( )N eq 2 ( + )N =
M

e2 ( ),
(1.24)
=1
where q j = 2s j and q 1 q 2 = 1. The BEA (1.24) coincide with the ones found by
Polyakov and Wiegmann for the lattice analogue of the SU(2) PCM with WZW term [33].
596

j
Again for l = 2s j (j = 1, 2) and = + 0 = 2i (2s j 1) the R j (0 ) matrix

becomes the permutation operator and hence we obtain a local Hamiltonian,

i d
i d
H
(1.25)
log t 1 ()
log t 2 ()
+
2 d
2 d
=+1
=+2
0
and the corresponding eigenvalues are

E=
2 M
1
n
2
n=1 j =1 (j + () +
qn
iq n
2
)(j + ()n
iq n
2 )
(1.26)
Moreover, the momentum and spin of a Bethe state are given by

2
q
M
(j + ()n +
1
P=
log
2i
(j + ()n
n=q 1 j =1

S z = q 1 + q 2 N M.
iq n
2 )
,
iq n
)
2
(1.27)
(1.28)
It has been proved [30,32], that this model has two types of low lying excitations, i.e.,
holes. The scattering among them was studied, the corresponding S matrix was computed,
for s i , and it was shown to be the SSU(2) SSU(2) plus a non-trivial leftright
scattering SLR () = tanh 2 ( 2i ). The SSU(2) SSU(2) , SLR () is also the massless S
matrix of the SU(2) principal chiral model with WZW term at level 1 (PCM1 ), conjectured
by Zamolodchikovs [22]. It has been also proved [32], that in the scaling limit ,
both excitations obey a massless relativistic dispersion relation, namely
) 1 () = p1 () e e ,
) 2 () = p2 () e e .
(1.29)
These are the energy and momentum of the right and left movers, respectively (see,
e.g., [22]) and the factor e provides a mass scale for the system.
2. Spin chains with boundaries

After the brief review on the bulk XXX spin s and alternating spin chains, we are ready
to study these models in the presence of boundaries. For both models we will derive the
Bethe ansatz equations and we will compute the exact reflection matrices. To construct the
spin chain with boundaries in addition to the R matrix another constructing element, the
K matrix, is needed. The K matrix is a solution of the reflection (boundary YangBaxter)
equation [7],
R12 (1 2 )K1 (1 )R21 (1 + 2 )K2 (2 )
= K2 (2 )R12 (1 + 2 )K1 (1 )R21 (1 2 ).
(2.1)
In what follows we are going to use Sklyanins formalism [8] in order to construct both
models with boundaries.
597
2.1. The XXX spin s open chain

2.1.1. The Bethe ansatz equations
Let as first consider the XXX spin s = q/2 quantum chain with boundaries. We define
the transfer matrix for the open chain [8]
t () = tr0 K0+ ()T0 ()K0 ()T0 (),
(2.2)
T0 () is the monodromy matrix (1.7) and

2s
2s
2s
2s
T0 () = R10
( )R20
( + ) R2N10
( )R2N0
( + ).
(2.3)
The K () = K(, ) matrix satisfies the reflection Eq. (2.1) and the K + () =
K ( )t ( + ) satisfies,
R12 (1 + 2 )K1+ (1 )t1 R21 (1 2 2)K2+ (2 )t2
= K2+ (2 )t2 R12 (1 2 2)K1+ (1 )t1 R21 (1 + 2 ),
(2.4)
where = i. We choose the diagonal solution of the reflection equation [10,11], namely
K() = diag( + i, + i ).
(2.5)
Note that in the presence of boundaries we have to fuse the K matrix as well, and we use
the following fusion hierarchy for the transfer matrix [34,35]

1
1
t j () = 2j 1 2 + (2j 1)i t j 2 ()t 2 + (2j 1)i

( + (2j 2)i)2j 2(2 + (2j 2)i) j 1
t
() ,
(2 + 2i(2j 1))
(2.6)
where

() = K + () [T ()] K () [T()],
(2.7)
and the quantum determinants are [8,14,34],

[T ()] = [T()] = ( + + i)N ( + i)N ,

K + () = g(2 3)l , +
K () = g(2 + )l , ,
moreover

i
i
() = + is +
+ is +
,
2
2
l(, ) = ( + i + )( + i )
(2.8)
() = ( + i)( + i),
g() = + i,
(2.9)
and
j () =
j

k=1
( + ik),
0 () = 1.
(2.10)
598
Then the eigenvalues of the transfer matrix, after we fuse the auxiliary space, are given by
2s
l () =
l

hk ()fk ()ak ( )2N ak ( + )2N
k=0
M

j =1
( j 2i )
( j + i(k 1) + 2i ) ( j + ik + 2i )
M

j =1
( j + il + 2i )
( + j 2i )
( + j + il + 2i )
( + j + i(k 1) + 2i ) ( + j + ik + 2i )
(2.11)
where ak have been already defined in (1.11) and

fk () = fk+ ()fk (),
(2.12)
where
fk () =
l1

+ i + im + i
k1
+ i in .
m=k
(2.13)
n=0
fk () correspond to the left (+) and right () boundary. hk () are derived by the fusion
hierarchy (2.6) and for, e.g., l = 1, l = 2 they are, respectively,1
hk () =
l1
k1

1
(2 + 2im)
(2 + 2i + 2in)
2 + i
m=k
(2.15)
n=0
and
hk () =
l1

(2 + 2im)
m=k
k1

(2 + 2i + 2in).
(2.16)
n=0
From the analyticity of the eigenvalues we obtain the Bethe ansatz equations
1
2N
ex1
eq ( + )2N e1 ( )
+ ( )ex ( )eq ( )
M

e2 ( )e2 ( + ),
(2.17)
=1
where x = 2 + 1. Recall that for l = 2s, R 2s (0 ) is the permutation operator, therefore

we obtain a local Hamiltonian

2s

i d
log t ()
H
(2.18)
2 d
=+0
1 For general l
hk ()
l1

m=k
(2 + 2im)
k1

n=0
(2 + 2i + 2in).
(2.14)
599
and the corresponding energy eigenvalues

E=
2 M
1
n
2
n=1 j =1 (j + () +
q
iq
2 )(j
+ ()n
iq
2)
(2.19)
2.1.2. The boundary S matrix

Now we can study the scattering of the excitations with the boundaries and determine
the boundary S matrix. Before we do that it is necessary to determine what is the ground
state and the low lying excitations for the model. The ground state, as in the bulk, is the
filled Dirac sea with strings of length q = 2s. The Bethe ansatz equations for the ground
state become
1
1
2N
Xqx
Xqq ( + )2N eq ( )
+ ( )Xqx ( )Xqq ( )
Mj

Eqq ( )Eqq ( + ).
(2.20)
=1
Let us consider x q 1, and define

Xnm () = e|nm+1| ()e|nm+3| () e(n+m3) ()e(n+m1) (),
2
2
() e(n+m2)
()e(n+m) ().
Enm () = e|nm| ()e|nm+2|
(2.21)
For the next we need the following notations

an () =
1 d
i log en ()
2 d
(2.22)
and the Fourier transform of an , is given by

a n () = e
n
2
(2.23)
We also need the following expressions

1 d
i log Xnm (), Enm () ,
Znm (), Anm () =
2 d
(2.24)
where their Fourier transforms are

e
max(n,m)
2
sinh((min(n, m)) 2 )
,
sinh( 2 )

max(n,m)
2
A nm () = 2 coth
sinh min(n, m)
e
nm .
2
2
nm () =
Z
(2.25)
(2.26)
Finally, the energy (2.19) takes the form

E=
2
M

n=1 =1
Zqq + ()n .
(2.27)
600
The density for the ground state is given by

0 () = Zqq ( + ) + Zqq ( ) (Aqq 0 )()
1
Zqx () Zqx + () + aq () + Aqq () ,
+
(2.28)
L
where L = 2N is the length of the chain and denotes the convolution. The solution of the
last equation is
0 () = )() +
1
r , + + r , + Q() ,
L
(2.29)
where

Q()
=
qx ()

Z
r , =
,
1 + A qq ()
a q ()
1 (),
+K
1 + A qq ()
1 () =
K
A qq ()
,
1 + A qq ()
(2.30)
and
)() =
2

s ()i ,
s () =
i=1
qq ()
Z
.
1 + A qq ()
(2.31)
We can write the above Fourier transforms in terms of trigonometric functions using the
definitions of a q and A qq ,
(q1)
2
e 2 cosh( 2 )
,
2 cosh( 2 ) sinh( q
2 )
1

1 () +
Q()
=K
2 coth( 2 ) sinh( q
2 )
1 () =
K
(2.32)
and for x q 1
r (, ) =
e(x q) 2
2 cosh( 2 )
(2.33)
moreover,
s() =
1
,
2 cosh( 2 )
)() =
2

i=1
1
.
2 cosh(( ()i ))
(2.34)
Let us consider the state with holes in the q sea, where an even number. Then, in
the thermodynamic limit we obtain the density of the state form the Bethe ansatz equation,
namely
1
K1 ( i ) + K1 ( + i ) .
L
() = 0 () +
i=1
(2.35)
601
The energy of the state with holes in the q sea (2.27) is given by
E = E0 +
)( ),
(2.36)
=1
where E0 is the energy of the ground state and )() is the energy of the hole in the q sea.
Finally, we compute the spin of the holes, and we can see from (1.15) that the spin of a hole
in the q sea is (1.15) s = f/2, where f is an overall factor (see also [14]). We consider
the spin of the hole to be 1/2 for what follows. We conclude that the hole in the q sea is a
d
particle like excitation with energy ), momentum p ()() = 1 d
p())
)() =
2

n=1
1
,
2 cosh( ()n )
1 1
+
tan sinh ()n
4
2
2
p() =
(2.37)
n=1
and spin s = 1/2. We can easily check that in the scaling limit where the energy
and momentum become from (2.37),
)() 2e cosh(),
p() 2e sinh().
(2.38)
Note that the excitations in the scaling limit, as in the bulk case satisfy a massive relativistic
dispersion relation.
Having studied the excitations of the model we are ready to compute the complete
boundary S-matrix. To do so we follow the formulation developed by Korepin, and later by
Andrei and Destri [36,37]. First we have to consider the so-called quantization condition.
2iLp
e
(2.39)
S 1 | 1 , 2 = 0,
where p is the momentum of the particle, the hole in our case. For the case of (even)
holes in the q sea we compare the integrated density (2.35) with the quantization condition
(2.39). Having also in mind that,
)() =
1 d
p(),
d
(2.40)
we end up with the following expression for the boundary scattering amplitudes (the
boundary S matrix will have the form S = diag( , )),

+
= exp 2piL
d () )()

(2.41)
and

, = f ()k0 ()k0 ()k1 , ,
(2.42)
602
where

k0 ( 1 ) = exp 2i
1

1
4
K1 ( i ) + K1 ( + i ) + Q() d
(2.43)
i=1
the x dependent part is

k1 1 ,
= exp 2i
r , d .
(2.44)
We are interested in the limit that s , in this limit we can easily see from (2.30) (see
also [17]) that
e 2
,
cosh( 2 )

1 (),
Q()
K
1 ()
K
e(x q) 2
r , =
2 cosh( 2 )
(2.45)
with x q to be a fixed number as q and f () is an overall CDD factor given by

ei
f () = exp
d
1
2 cosh( 2 )

= tanh

i
+
.
2
2
(2.46)
Using the above Fourier transforms we end up with the following expressions,

2
sinh( 3
d
2 )e
sinh(2i)
,
k0 () = exp 2
sinh(2)
0
e2
d
sinh(2i)
,
k1 , = exp
cosh()

(2.47)
where
2s + 1 is the renormalized boundary parameter and k0 (), k1 (, )
are the independent and the dependent part, respectively, of the usual XXX (SU(2))
reflection matrix (see, e.g., [38]). We notice from (2.42) that there are two copies of the
independent part, whereas just one copy of the dependent part exists. Recall, that we
considered the diagonal K matrix (2.5) in order to construct our chain, therefore we need
to determine the other element of the boundary matrix. To do so we exploit the duality
symmetry [39] of the transfer matrix for . Hence, the other element is given by

, = e2 1 () , ,
(2.48)
= 2
the ratio (, ) is the same as in the XXX model [38] but with a renormalized boundary
(, )
parameter.
The above equations (2.42)(2.48), are simply combined to give the boundary S matrix
as the tensor product of two rational (XXX) boundary S matrices [38] (up to an overall
603
CDD factor), i.e.,

S() = f ()SSU(2) , 1 SSU(2) , 2 ,
(2.49)
where 1 = , 2 . This matrix coincides with the one found for the SU(2) PCM
[40]in our case is a free parameter.
However, in analogy with the bulk case, for finite s the boundary S matrix is expected
to be of the form SSU(2) (, ) SRSOS (s), where SRSOS (s) is the boundary S matrix of
the RSOS model. The SRSOS matrix is a solution of the boundary YangBaxter equation in
the RSOS representation (see, e.g., [4143]). The SSU(2) (, ) SRSOS (s) matrix should
also describe the WZWk=2s boundary scattering and it should presumably reduce to the S
matrix we found (2.49), as s .
2.2. The alternating s 1 , s 2 open spin chain
2.2.1. The Bethe ansatz equations
The corresponding transfer matrix t () for the open alternating chain of 2N sites and
s 1 = q 1 /2, s 2 = q 2 /2 spins (q 1 q 2 = 1) is (see also, e.g., [8,27]),
t () = tr0 K0+ ()T0 ()K0 ()T0 (),
(2.50)
where T0 () is the monodromy matrix defined previously in (1.21) and

2
1
2
1
T0 () = R10
( )R20
( + ) R2N10
( )R2N0
( + ).
(2.51)
Again here we use the diagonal K matrix (2.5) and the fusion hierarchy (2.6) with
[T ()] = 1 ( + i)N 2 ( + + i)N ,
[T()] = 1 ( + + i)N 2 ( + i)N
(2.52)

i
i
+ is n +
.
n () = + is n +
2
2
(2.53)
and
The eigenvalues of the transfer matrix, after we fuse the auxiliary space, are given by
(1,2)
() =
l

(1)
(1)
(2)
(2)
hk ()fk ()ak ( )N ak ( + )N ak ( )N ak ( + )N
k=0
M

j =1
M

j =1
( j 2i )
( j + il + 2i )
( j + i(k 1) + 2i ) ( j + ik + 2i )
( + j 2i )
( + j + il + 2i )
( + j + i(k 1) + 2i ) ( + j + ik + 2i )
(2.54)
604
provided that {j } satisfy the Bethe ansatz equations

2
q

1
en ( )N en ( + )N e1 ( )ex1
+ ( )ex ( )
n=q 1
M

e2 ( )e2 ( + ).
(2.55)
=1
j
For l = 2s j , R j (0 ) becomes the permutation operator, therefore, we can obtain a local

Hamiltonian for the open chain

1

2

i d
i d

log t ()
log t ()
+
H
(2.56)
4 d
4 d
=+1
=+2
0
and
E =
2 M
1
4
n=1 j =1 (j +
qn
iq n
2 )(j
2 M
1
4
n=1 j =1 (j + +
iq n
2 )
qn
iq n
2
)(j +
iq n
2 )
(2.57)
2.2.2. The boundary S matrix

The ground state consists of two filled Dirac seas with strings of length q n = 2s n ,
n = 1, 2. Then the Bethe ansatz equations for the ground state become
1 n
1 n
Xnx
+ Xnx
q

Xnj n Xnj n + en n
j =q 1
2
Mj
q

Enj n Enj n + ,
(2.58)
j =q 1 =1
where n can be q 1 , q 2 and let us consider for simplicity x q 2 . Finally, the energy (2.57)
takes the form
2
Mj
Mj
q
q

1
1 i
Zij +
Zij i .
E=
2
2
1
1
i,j =q =1
(2.59)
i,j =q =1
The density that describes the ground state is given by the following integral equations,
2
0n () =
q
q

1
j
Znj ( ) + Znj ( + )
Anj 0 ()
2
1
1
j =q
j =q

q2

1
+
Anj () ,
Znx + () Znx () + an () +
L
1

j =q
(2.60)
where n can be q 1 or q 2 . The solution of the above integral equation is given by
1 n
r , + + r n , + Qn () ,
0n () = ) n () +
L
where
2
n () =
Q
q

605
(2.61)
ni () +
a i ()R
i=q 1
q

in (),
K
1
i=q 1
q2
ni (),
ix ()R
r , =
Z
n
nm () =
K
1
i=q 1
q

mi ()
A ni ()R
(2.62)
i=q 1
and
1
s ()i ,
2
2
) n () =
(2.63)
i=1
and its Fourier transform is

2
s () =
n
q

ij R
nj )().
(Z
(2.64)
i,j =q 1
Here R is the inverse of the kernel K of the system of the linear equations (2.60),

nm () = 1 + A nm () nm + A nm ()(1 nm ),
K
nm () =
R
q2

det K
j =q 1
1 + A jj () nm (1 nj ) A nm ()(1 nm ) ,
(2.65)
(2.66)
where the determinant of K is, in terms of trigonometric functions,

q1
= 4 coth2 e 2 sinh q 2 sinh .
det K
(2.67)
2
2
2
nm has the following explicit form in terms of trigonometric functions
In particular, K
1
111 () =
K
e 2
,
2 cosh( 2 )
112 () = K
121 () =
K
122 () =
K
sinh((q 1 2) 2 )
2 cosh( 2 ) sinh((q 1 1) 2 )
1
,
2 cosh( 2 )
(2.68)
2
1 () = 1 + e
,
Q
2 cosh( 2 )
2 () = K
22 () +
Q
sinh( 2 )
1
.
+
2 cosh( 2 ) 2 cosh( ) sinh( q 2 )
2
2
(2.69)
The dependent part for x q 2 is
e(x q ) 2
r , =
,
2 cosh( 2 )
1
r 2 , = 0,
(2.70)
606
finally,
sn () =
1
,
2 cosh( 2 )
) n () =
2

i=1
1
.
4 cosh(( + ()i ))
(2.71)
Let us consider the state with n holes in the q n sea, where n is an even number.
Then in the thermodynamic limit we obtain the density of the state from the Bethe ansatz
equations, namely
n
nn
1
K1 ( i ) + K1nn ( + i )
L
n () = 0n () +
(2.72)
i=1
the energy of the state with n holes in the q n seas is given (2.59) by
E = E0 +
n

) n n ,
(2.73)
=1
where E0 is the energy of the ground state and ) n () is the energy of the hole in the q n
sea. Finally, we compute the spin of the holes from (1.28), and we can verify that the spin
of a hole in the q 1 sea is s 1 = 1/2 whereas the spin of a hole in the q 2 sea is s 2 = 0. We
conclude that the hole in the q n sea is a particle like excitation with energy ) n , momentum
d n
pn () n () = 1 d
p ())
) n () =
2

l=1
1
,
4 cosh(( ()l ))
1 1
+
tan sinh ()l ,
4
4
2
pn () =
(2.74)
l=1
and spin s 1 = 1/2, s 2 = 0. We can easily check that in the scaling limit, the energy
and momentum become from (2.74),
) n () e cosh(),
pn () e sinh(),
(2.75)
the factor e provides a mass scale for the system. Note that in the presence of
boundaries the excitations, in the scaling limit, satisfy a massive relativistic dispersion
relation (2.75) whereas in the bulk case the excitations are massless relativistic particles
(1.29). This is a very interesting phenomenon which is presumable related to the type of
the boundaries. The boundaries we impose, force a left (right) mover to reflect as a left
(right) mover. It is possible that the boundary can reflect a left mover to a right one and
vice versa. This type of boundaries would probably lead to massless excitations in the
scaling limit.
Again, we consider the quantization condition [36,37], in order to compute the exact
reflection matrices, namely
2iLpn n
e
(2.76)
S 1 | 1 , 2 = 0,
607
where pn is the momentum of the hole in the q n sea. For the case of n holes in the q n sea
we compare the integrated density (2.72) with the quantization condition (2.76). Having
also in mind that,
) n () =
1 d n
p ()
d
(2.77)
n
we end up with the following expression for the boundary scattering amplitudes (S
=
n
n
diag( , ))

n n
+
= exp 2piL
d () )()

(2.78)
and

, = k0n ()k1n , ,
(2.79)
where

k0n ( 1 ) = exp i
1
nn
K ( i ) + K nn ( + i ) + Qn () d
(2.80)
0 i=1
the x dependent part is

k1n

1

1 , = exp 2i r , d .
(2.81)
We are interested in the limit that q n , in this limit we can easily verify that
nn () +
n () K
Q
1
,
2 cosh( 2 )
e 2
2 cosh 2
nn ()
K
(2.82)
and
e(x q ) 2
,
r 1 , =
2 cosh( 2 )
1
r 2 , = 0,
(2.83)
with x q 1 to be a fixed number as q n . Using the above Fourier transforms we

end up with the following expressions,

2
sinh( 3
d
n
2 )e
k0 () = exp 2
sinh(2i)
,
sinh(2)
0
k1n
e2n
d
sinh(2i)
,
, n = exp
cosh()
0
(2.84)
608
where 21 = 2 2s 1 + 1, 22 , are the renormalized boundary parameters, k0n ()

and k1n () are the independent and the dependent part, respectively, of the usual XXX
(SU(2)) reflection matrix (see, e.g., [38]). Exactly as in the case of the spin s open chain,
we consider the diagonal K matrix, therefore we need to determine the other element of
each boundary matrix. We exploit the duality symmetry [39] of the transfer matrix for
, and we find that the other diagonal element is given by

n
n
, n = e2n 1 ()
, n .
(2.85)
The ratio
n (, )
n
n
(,n )
is the same as in the XXX model but with a renormalized boundary
parameter. We observe for the alternating chain as well as in the spin s chain that only one
free boundary parameter 1 (2 ) survives.
Two copies of the rational (XXX) reflection matrix were computed, one for each
excitation. Therefore, we conclude that the boundary S matrix should be of the structure,

S() = SSU(2) , 1 SSU(2) , 2 .

(2.86)
We assume that this matrix should also coincide with the one of the PCM1 . A calculation
of the boundary S matrix from the field theory point of view would probably confirm
our results. We have to mention that there have been some studies for the SU(2) PCM
with WZW term with boundaries, [44] but mainly in the context of quantum impurity
(Kondo) problem. In particular, in [44] the authors considered dynamical boundaries, i.e.,
they considered quantum impurities at the boundaries, and they derived the corresponding
reflection matrices.
3. Discussion
The XXX spin s and the alternating s 1 , s 2 (s 1 s 2 = 1/2) chains were explored.
For both models the Bethe ansatz equations were derived using fusion, and the exact
boundary S matrices were computed. We were particularly interested in the case that
s, s i . More specifically, for the spin s chain the boundary scattering amplitudes
were simply combined to give the boundary S matrix of the form f ()SSU(2)(, 1 )
SSU(2) (, 2 ).
For the alternating spin chain two different types of excitations exist: 1/2 and 0
spin, respectively. The boundary scattering for each excitation was studied and the
corresponding reflection matrices were derived. Two copies of the XXX boundary S
matrix were computed (s i ), and the boundary S matrix was given as a tensor product
of two rational boundary S matrices. This matrix is also expected to coincide with the one
of the SU(2) PCM1 . Note that we could end up with to same result if we started from the
anisotropic spin chains and then take the isotropic and s, s i limits.
It would be also interesting to consider dynamical K matrices [44] in order to construct
the open spin chain and then to study the reflection of the particle-like excitations with the
dynamical boundary. Another interesting aspect would be the study of the thermodynamics
of the alternating spin chain via the TBA. The main purpose would be the derivation of the
609
central charge (see, e.g., [28,29]), of the model for both bulk and boundary cases. We hope
to report on these issues in a future publication.
Acknowledgements
I would like to thank R.I. Nepomechie and E. Sklyanin for valuable discussions
and comments. Also, I am grateful to N. Reshetikhin for correspondence. This work is
supported by the EPSRC.
References
[1] H. Bethe, Z. Phys. 71 (1931) 205.
[2] L.D. Faddeev, L.A. Takhtajan, Russ. Math. Surv. 34 (1979) 11;
L.D. Faddeev, L.A. Takhtajan, J. Sov. Math. 24 (1984) 241.
[3] L.D. Faddeev, L.A. Takhtajan, Phys. Lett. A 85 (1981) 375.
[4] N.Yu. Reshetikhin, Nucl. Phys. B 251 (1985) 565.
[5] I. Affleck, In Fields Strings and Critical Phenomena, Les Houches Lectures, Elsevier, 1990.
[6] N.Yu. Reshetikhin, J. Phys. A 24 (1991) 3299.
[7] I.V. Cherednik, Theor. Math. Phys. 61 (1984) 977.
[8] E.K. Sklyanin, J. Phys. A 21 (1988) 2375;
P.P. Kulish, E.K. Sklyanin, J. Phys. A 24 (1991) L435.
[9] A. Fring, R. Kberle, Nucl. Phys. B 421 (1994) 159;
A. Fring, R. Kberle, Nucl. Phys. B 419 (1994) 647.
[10] S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 3841;
S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 4353.
[11] H.J. de Vega, A. Gonzlez-Ruiz, Mod. Phys. Lett. A 9 (1994) 2207;
H.J. de Vega, A. Gonzlez-Ruiz, J. Phys. A 26 (1993) L519.
[12] R.J. Baxter, Ann. Phys. 70 (1972) 193;
R.J. Baxter, J. Stat. Phys. 8 (1973) 25;
R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, 1982.
[13] V.E. Korepin, Theor. Math. Phys. 76 (1980) 165;
V.E. Korepin, G. Izergin, N.M. Bogoliubov, Quantum Inverse Scattering Method, Correlation Functions and
Algebraic Bethe Ansatz, Cambridge Univ. Press, 1993.
[14] A. Kirillov, N. Reshetikhin, J. Sov. Math 35 (1986) 2621;
A. Kirillov, N. Reshetikhin, J. Phys. A 20 (1987) 1565.
[15] L.A. Takhtajan, Phys. Lett. A 87 (1982) 479.
[16] A. Polyakov, P. Wiegmann, Phys. Lett. B 131 (1983) 121.
[17] P. Wiegmann, Phys. Lett. B 142 (1984) 173.
[18] L.D. Faddev, N.Yu. Reshetikhin, Ann. Phys. 167 (1986) 167.
[19] H. Babujian, Nucl. Phys. B 215 (1983) 317.
[20] G. Japaridze, A. Nersessian, P. Wiegmann, Nucl. Phys. B 230 (1984) 511.
[21] H. Babujian, A. Tsvelik, Nucl. Phys. B 265 (1986) 24.
[22] A. Zamolodchikov, Al. Zamolodchikov, Nucl. Phys. B 379 (1992) 602.
[23] D. Bernard, Phys. Lett. B 279 (1992) 78.
[24] A. Zamolodchikov, Al. Zamolodchikov, Ann. Phys. 120 (1979) 253.
[25] V.V. Bazhanon, N.Yu. Reshetikhin, Int. J. Mod. Phys. A 4 (1989) 115.
[26] C. Ahn, D. Bernard, A. Leclair, Nucl. Phys B 346 (1990) 409;
D. Bernard, A. Leclair, Phys. Lett. B 247 (1990) 309.
[27] H.J. de Vega, F. Woyanorovich, J. Phys. A 25 (1992) 4499.
610
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
S.R. Aladim, M.J. Martins, J. Phys. A 26 (1993) 7287.

H.J. de Vega, L. Mezincescu, R.I. Nepomechie, Phys. Rev. B 49 (1994) 13223.
H. J de Vega, L. Mezincescu, R.I. Nepomechie, Int. J. Mod. Phys. B 8 (1994) 3473.
B.D. Doerfel, S. Meisner, J. Phys. A 30 (1996) 6471.
A. Doikou, A. Babichenko, Phys. Lett B 515 (2001) 220.
A. Polyakov, P. Wiegmann, Phys. Lett. B 141 (1984) 223.
L. Mezincescu, R.I. Nepomechie, J. Phys. A 25 (1992) 2533.
R.I. Nepomechie, hep-th/0110116.
V. Korepin, Theor. Math. Phys. 41 (1979) 53.
N. Andrei, C. Destri, Nucl. Phys. B 131 (1984) 445.
M. Grisaru, L. Mezincescu, R.I. Nepomechie, J. Phys. A 28 (1995) 1027;
A. Doikou, L. Mezincescu, R.I. Nepomechie, J. Phys. A 30 (1997) L507.
A. Doikou, R.I. Nepomechie, Nucl. Phys. B 521 (1998) 547;
A. Doikou, R.I. Nepomechie, Nucl. Phys. B 530 (1998) 641;
A. Doikou, R.I. Nepomechie, Phys. Lett. B 462 (1999) 321.
N.J. Makcay, B.J. Short, hep-th/0104212.
C. Ahn, W.M. Koo, hep-th/9708080;
C. Ahn, W.M. Koo, J. Phys. A 29 (1996) 5845.
R.E. Behrend, P.A. Pearce, D.L. OBrien, J. Stat. Phys. 84 (1996) 1;
R.E. Behrend, P.A. Pearce, J. Phys. A 29 (1996) 7827.
M.T. Batchelor, V. Fridkin, A. Kuniba, Y.K. Zhou, Phys. Lett. B 735 (1996) 266.
P. Fendley, Phys. Rev Lett. 71 (1993) 2485;
J.N. Prata, Phys. Lett. B 438 (1998) 115.

Algebraic Bethe ansatz for a quantum integrable

derivative nonlinear Schrdinger model
B. Basu-Mallick, Tanaya Bhattacharyya
Theory Group, Saha Institute of Nuclear Physics, 1/AF Bidhan Nagar, Kolkata 700 064, India
Abstract
We find that the quantum monodromy matrix associated with a derivative nonlinear Schrdinger
(DNLS) model exhibits U (2) or U (1, 1) symmetry depending on the sign of the related coupling
constant. By using a variant of quantum inverse scattering method which is directly applicable to
field theoretical models, we derive all possible commutation relations among the operator valued
elements of such monodromy matrix. Thus, we obtain the commutation relation between creation
and annihilation operators of quasi-particles associated with DNLS model and find out the S-matrix
for two-body scattering. We also observe that, for some special values of the coupling constant, there
exists an upper bound on the number of quasi-particles which can form a soliton state for the quantum
DNLS model. 2002 Elsevier Science B.V. All rights reserved.
PACS: 11.10.Lm; 11.30.-j; 02.30.Ik; 03.65.Fd
Keywords: Derivative nonlinear Schrdinger model; YangBaxter equation; Algebraic Bethe ansatz; Soliton
1. Introduction
Quantum integrable field models and spin chains in low dimensions have recently
attracted much attention due to their close connection with diverse areas of physics as well
as mathematics. By using algebraic Bethe ansatz, which occurs naturally in the framework
of quantum inverse scattering method (QISM), one can find out the spectrum and various
correlation functions of quantum integrable models with short-range interactions [14].
The nonlinear Schrdinger (NLS) model is a well known example of such quantum
integrable field model in (1 + 1) dimension [18]. For the case of derivative nonlinear
E-mail addresses: biru@theory.saha.ernet.in (B. Basu-Mallick), tanaya@theory.saha.ernet.in
(T. Bhattacharyya).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 2 8 8 - 2
612
B. Basu-Mallick, T. Bhattacharyya / Nuclear Physics B 634 [FS] (2002) 611627
Schrdinger (DNLS) model, however, the situation is a little bit complicated due to the
following reason. There exists one type of classical DNLS model with equation of motion
like [9]

iqt + qxx 4i q q 2 x = 0,
(1.1)
where the subscripts denote the derivatives with respect to corresponding variables. By
using an equal time nonultralocal Poisson bracket (PB) structure given by {q(x), q(y)} =
{q (x), q (y)} = 0, {q(x), q (y)} = x (x y), one can show that the infinite number
of conserved quantities (including the Hamiltonian) associated with this DNLS equation
of motion yield vanishing PB relations among themselves [10]. This fact establishes
the classical integrability of DNLS model (1.1) in the Liouville sense. However, due to
the appearance of nonultralocal commutation relations among the basic field operators,
quantum version of such DNLS model cannot be handled by QISM and therefore quantum
integrability cannot be established for this case.
On the other hand, there exists another type of DNLS model with equation of motion
like [11]
it + xx 4i( )x = 0.
(1.2)
The Hamiltonian
H=
+

xx + 2i( ) x dx,
(1.3)
and an equal time ultralocal PB structure

{(x), (y)} = (x), (y) = 0,

(x), (y) = i(x y),
(1.4)
yield Eq. (1.2) through a canonical evolution: t = {, H }. By using the ultralocal PB

structure (1.4) and Lax operator given by

(x)(x) 2 /4
(x)

U (x, ) = i
(1.5)
,
(x)
(x)(x) + 2 /4
where is a spectral parameter, one can prove the classical integrability for the DNLS
model (1.2) in the Liouville sense. Furthermore, by taking advantage of ultralocal
commutation relations among the basic field operators, integrability of the corresponding
quantum DNLS model can also be established through QISM [12].
Similar to the case of other integrable systems, the monodromy matrix plays a key
role in deriving the conserved quantities and studying exact solvability of the DNLS
model (1.2) as well as its quantum analogue. Though the PB relations (commutation
relations) among various elements of classical (quantum) monodromy matrix associated
with such DNLS model have been computed earlier [12], this problem should be
reinvestigated due to the following reasons. First of all, it was assumed earlier that the
monodromy matrix of classical (quantum) DNLS model exhibits SU(2) (U (2)) symmetry
for any value of the corresponding coupling constant . However, in comparison to the
case of NLS model where the symmetry of monodromy matrix depends on the sign of
the corresponding coupling constant, this seems to be a rather questionable assumption.
613
Therefore, it is necessary to properly study the symmetry properties of the classical

(quantum) monodromy matrix of DNLS model and use those symmetries as an input for
deriving the PB relations (commutation relations) among the elements of the monodromy
matrix. Secondly, a rather cumbersome method was adopted earlier for finding out the
commutation relations among the elements of quantum monodromy matrix at the infinite
interval limit. As it will be clear from the discussions in Section 4 of the present article, the
above mentioned method yields an incorrect commutation relation between creation and
annihilation operators associated with the quasi-particles of quantum DNLS model. Since
such commutation relation plays a crucial role in finding out the norm of Bethe eigenstates,
the S-matrix for two-body scattering and various correlation functions, it is necessary
to calculate very carefully the commutation relations among the elements of quantum
monodromy matrix at infinite interval limit. Finally, it may be noted that two different
approaches were taken earlier for treating the classical and quantum DNLS model [12]. The
integrability of quantum DNLS model was established by first discretising the system on a
lattice, evaluating the commutation relations among the elements of the monodromy matrix
defined on that lattice and finally taking the continuum limit. For the case of classical
DNLS model, however, no such lattice regularisation was taken and PB relations among
the elements of monodromy matrix were evaluated directly for the continuum model.
The aim of the present article is to shed some light on the above mentioned issues
and especially study the quantum DNLS model by using a variant of QISM [2] which
can be applied to continuum field models without performing any lattice regularisation.
In Section 2 of this article, we find out the symmetry properties of monodromy matrix
associated with the classical DNLS model (1.2) and subsequently use those symmetries for
evaluating PB relations among various elements of this monodromy matrix. In Section 3,
we construct the quantum monodromy matrix of DNLS model on a finite interval and
derive all possible commutation relations among the elements of such monodromy matrix
through QISM. In Section 4, we take the infinite interval limit of these commutation
relations and construct exact eigenstates for the diagonal elements of quantum monodromy
matrix through algebraic Bethe ansatz. Furthermore, we obtain the commutation relation
between creation and annihilation operators of quasi-particles associated with DNLS
model and find out the S-matrix of two-body scattering among such quasi-particles.
Section 5 is the concluding section.
2. Integrability of the classical DNLS model

To investigate the symmetry properties of monodromy matrix associated with the
classical DNLS model (1.2), we start with a Lax operator of the form

(x)(x) 2 /4
(x)
U (x, ) = i
(2.1)
.
(x)
(x)(x) + 2 /4
Note that the off-diagonal elements of this Lax operator differ from the corresponding
elements of previously given Lax operator (1.5) through some scale factors. However,
as will be shown later, the traces of monodromy matrices associated with the Lax
614
operators (1.5) and (2.1) coincide with each other and consequently both of these Lax
operators correspond to the same DNLS model (1.2).
By using the Lax operator (2.1) along with its asymptotic form at |x| , we define
the monodromy matrices of DNLS model on finite and infinite intervals as
x2
Txx12 () = P exp
U (x, ) dx
(2.2)
x1
and

T () =
lim
x2 +
x1
x2
e(x2, ) P exp
U (x, ) dx e(x1 , ),
(2.3)
x1
i
respectively, where P denotes the path ordering and e(x, ) = e 4 3 x . To find out the
symmetries of such monodromy matrices, we note that the Lax operator (2.1) satisfies the
relations
U (x, ) = KU (x, )K 1 ,
(2.4a)
U (x, ) = LU (x, )L ,
(2.4b)
1 0
0
and L = 0 1 . By using these relations we find that the
where K = 1/ 0
symmetries of monodromy matrix (2.3) are given by
T () = KT ( )K 1 ,
1
T () = LT ()L
(2.5a)
(2.5b)
By exploiting the symmetry relation (2.5a) and restricting to the case when is a real
parameter, one may now express T () in the form

a() b ()
.
T () =
(2.6)
b()
a ()
Since the Lax operator (2.1) is a traceless matrix, we also get det T () = 1 or, equivalently,
|a()|2 + |b()|2 = 1. Moreover, by using the symmetry relation (2.5b), it is easy to see
that a() = a() and b() = b(). So, for the case of real , T () within the range
0 contains all informations about the scattering data. Therefore, in the following we
shall consider the PB relations among the elements of monodromy matrix (2.6) only within
the range 0.
In analogy with the monodromy matrix T () (2.3) which is defined through Lax
() corresponding to Lax
operator (2.1), one can also define the monodromy matrix T
operator (1.5) as

x2
(x, ) dx e(x1 , ).
T() = lim e(x2, ) P exp U
(2.7)
x2 +
x1
x1
() can be written in a SU(2) symmetric form for

It was assumed earlier that this T
any value of the corresponding coupling constant [12]. For the purpose of properly
615
(), we observe that the Lax operators (1.5)

investigating the symmetry properties of T
and (2.1) are related through a symmetry transformation given by
(x, ),
MU (x, )M 1 = U
(2.8)
1/4 0
where M =
. Consequently, the corresponding monodromy matrices would also
0 1/4
() = MT ()M 1 . By using this relation along with Eq. (2.6), we can
be related as T
express T() for real as

a()
b ()

,
T () =
(2.9)
b()
a ()
where = sign and
a()
= a(),
a () = a (),

b()
= b(),
b () = b ().
(2.10)
() takes the form of a

Now from Eq. (2.9) it is evident that the monodromy matrix T
SU(2) group valued object when > 0 and SU(1, 1) group valued object when < 0.
Thus we find that, similar to the case of NLS model, symmetry of the monodromy matrix
associated with DNLS model is also determined through the sign of the corresponding
coupling constant. As a result PB relations among the scattering data of DNLS model,
() to be a SU(2) group valued object, actually
which were derived earlier by assuming T
correspond to the case > 0.
It may be noted that, the diagonal elements of the monodromy matrices T () (2.6)
() (2.9) coincide with each other. So, by using the results of Ref. [12] where ln a()
and T
was expanded in powers of 1/, we can write

ln a() = ln a()

iCn
n=0
2n
(2.11)
and find out the first few Cn s as

+
dx,
C0 =
(2.12a)
+
C1 = 4i
x dx,
(2.12b)
C2 = 8
+

xx 2i( ) x dx.
(2.12c)
Note that the Hamiltonian (1.3) of DNLS model is related to the third expansion
1
C2 . As a result, the monodromy matrices T() (2.9)
coefficient (2.12c) as H = 8
and T () (2.6) correspond to the same DNLS model (1.2).
616
Next, we want to derive the PB relations among the elements of T () (2.6) for both
positive and negative values of the coupling constant . To this end, we apply (1.4) to
evaluate the PB relations among the elements of Lax operator (2.1) and find that

U (x, ), U (y, ) = r(, ), U (x, ) 1 + 1 U (y, ) (x y),
(2.13)
where

r(, ) = t c 3 3 + s c (+ + + )
(2.14)
+
2
c
with t c = 2(
2 2 ) , s = 2 2 . Next, by employing a standard technique [1] for deriving
PB relations among the elements of monodromy matrix (2.3) with the help of Eq. (2.13),
we obtain

T (), T () = r+ (, )T () T () T () T ()r (, ),
(2.15)
2
where r (, ) matrices are given by

r (, ) = E 1 (L, ) E 1 (L, )r(, )E 1 (L, ) E 1 (L, )

c
c
= t c 3 3 + s
+ + s
+ ,
c = 2i2 (2 2 ). By substituting the symmetric form of T () (2.6) to
with s
Eq. (2.15) and expressing it in elementwise form, we finally obtain
{a(), a()} = 0,

a(), a () = 0,
2

+ 2
{a(), b()} = 2
a()b() 2i 2 2 2 b()a(),
2

2

+ 2
a(), b() = 2
a()b ()
2

+ 2i 2 2 2 b ()a(),

b(), b() = 4i2 2 2 |a()|2 .
(2.16a)
(2.16b)
(2.16c)
(2.16d)
(2.16e)
The above PB relations among the scattering data of the DNLS model are evidently valid
for all values of the coupling constant . Since from Eqs. (2.16a) and (2.11) it follows that
{Cm , Cn } = 0 for all m, n, the DNLS model (1.2) represents a classical integrable system in
the Liouville sense. With the help of transformation (2.10), one can also recast Eqs. (2.16)
as the PB relations among the elements of monodromy matrix T() (2.9) and compare
such PB relations with their counterparts in Ref. [12]. It is easy to see that the forms of
Eqs. (2.16a)(2.16d) remain unaltered through the transformation (2.10) and coincide with
their counterparts as given earlier (after taking care of slight changes in notations and in
the definition of fundamental PB relation). However, by using Eqs. (2.16e) and (2.10) we
find that

2
b(),
b () = 4i| |2 2 2 |a()|
(2.17)
It is interesting to note that, for the case < 0, the above equation differs from its
counterpart [12] through a sign factor. Thus, we are able to derive here the correct PB
617
relation between b()

and b () for the case < 0. It is well known that the commutation
relation between the quantum analogues of b()

and b () plays a crucial role in finding
out the norm of the Bethe eigenstates, the S-matrix for two-body scattering and various
correlation functions. As a first step for evaluating this commutation relation and other
commutation relations which would represent the quantum counterparts of the classical
PB relations (2.16), in the following we shall quantise the monodromy matrix (2.2) of
DNLS model on a finite interval.
3. Commutation relations for the quantum monodromy matrix on a finite interval

In the quantised version of the DNLS model (1.2), the fundamental PB relations (1.4)
are replaced by equal time commutation relations among the basic field operators:

(x), (y) = (x y),
[(x), (y)] = (x), (y) = 0,
(3.1)
and vacuum state is defined as (x)|0 = 0. By using the ultralocal commutation
relations (3.1) and a version of QISM which can be applied to field models without
performing any lattice regularisation [2], at present we shall construct the quantum
monodromy matrix of DNLS model on a finite interval and derive all commutation
relations among the elements of such quantum monodromy matrix. To this end, we assume
that the quantised form of the classical Lax operator (2.1) is given by

(x)
f(x) 2 /4
Uq (x, ) = i
(3.2)
,
(x)
g(x) + 2 /4
where (x) = (x)(x), and f, g are two yet undetermined parameters. It will be
shown later that f , g at the classical limit and, as a result, Uq (x, ) (3.2)
correctly reproduces U (x, ) (2.1) in this limit. By using Uq (x, ), we define the quantum
monodromy matrix of DNLS model on a finite interval as
x2
Txx12 () = :P exp
Uq (x, ) dx: ,
(3.3)
x1
where the symbol :: denotes the normal ordering of operators. It is evident that this quantum
monodromy matrix (3.3) satisfies a differential equation given by
x2
T () = : Uq (x2 , )Txx12 ():
x2 x1
i
= 2 3 Txx12 () + i (x2 )+ Txx12 () + i Txx12 ()(x2 )
4
+ if (x2 )e11 Txx12 ()(x2 ) ig (x2 )e22 Txx12 ()(x2 ),
(3.4)
where e11 = 12 (1 + 3 ) and e22 = 12 (1 3 ). For the purpose of applying QISM, however,
it is needed to find out the differential equation satisfied by the product Txx12 () Txx12 ().
To this end, we borrow a notation
of Ref.
..
.. [2] where the sign for normal arrangement of
operator factors is taken as .... . The sign .... , applied to the product of several operator factors
618
(including and ), ensures the arrangement of all on the left, and all on the right,
without altering the order of the remaining factors. For example,
.
..
.X Y .. = XY ,
where X and Y may be taken as some elements of the quantum monodromy matrix (3.3).
Now by using the basic commutation relations (3.1) and the method of extension [2],
we find that the product of two monodromy matrices satisfies the following differential
equation (details of this calculation are given in Appendix A):
.
.
x2
Tx1 () Txx12 () = ..L(x2 ; , )Txx12 () Txx12 ().. ,
x2
(3.5)
L(x; , ) = Uq (x, ) 1 + 1 Uq (x, ) + L (x; , ),
(3.6)
where
with
f 2 (x)
0
L (x; , ) =
f (x)
0
f (x)
0
gf(x)
0
gf(x)
g(x)
0
0
.
g (x)
g 2 (x)
Next, let us consider a (4 4) R(, ) matrix of the form
1
0
0
0
0 s(, ) t (, ) 0
R(, ) =
0 t (, ) s(, ) 0 ,
0
0
0
1
(3.7)
(qq )

i , being an yet undeterwith t (, ) = 2 q
2 q 1 , s(, ) = 2 q2 q 1 and q = e
mined real parameter. It is easy to check that the R(, ) matrix (3.7) and L(x; , ) (3.6)
satisfy an equation given by
2
R(, )L(x; , ) = L(x; , )R(, ),
(3.8)
provided the parameters f, g, and the coupling constant are related as

= sin ,
ei/2
,
f=
cos /2
g=
ei/2
.
cos /2
(3.9a)
(3.9b)
(3.9c)
By using Eqs. (3.5) and (3.8), we find that the monodromy matrix (3.3) of DNLS model
satisfies the quantum YangBaxter equation (QYBE):
R(, )Txx12 () Txx12 () = Txx12 () Txx12 ()R(, ).
(3.10)
619
Expressing this QYBE in elementwise form, one may explicitly obtain all possible
commutation relations among the elements of quantum monodromy matrix (3.3).
Note that the relations (3.9) are obtained as a necessary condition for satisfying
QYBE (3.10). By solving Eqs. (3.9), the parameters f, g and can be determined as
some functions of the known coupling constant . Thus QYBE fixes all undetermined
parameters in the quantum Lax operator (3.2) and corresponding R(, ) matrix (3.7).
Due to Eq. (3.9a) it is evident that, the coupling constant of quantum DNLS model
must be restricted within a range given by | | 1 and the parameter has a one-to-one
correspondence with such coupling constant for 2 2 . It may be noticed that, by
putting x2 = x1 + in Eq. (3.3) (here is a small positive parameter) and retaining terms
only up to the order , one obtains

n
fn 2 /4
x1 +
,
Tx1
(3.11)
() 1 + i
n
gn + 2 /4
x +
x +
x +
where n = 1 x11 (x) dx, n = 1 x11 (x) dx, n = 1 x11 (x) dx. It is
interesting to observe that this Txx11 + () (3.11) satisfies QYBE (3.10) up to order and
reproduces (up to a gauge transformation) the lattice Lax operator for quantum DNLS
model [12,13] when the parameter is identified as a lattice constant. In this article,
however, we shall not use any lattice discretisation and directly work with the quantum
monodromy matrix (3.3) which satisfies QYBE exactly.
Next, for the purpose of investigating the classical limit of the quantum Lax
operator (3.2), we replace the last commutation relation in Eq. (3.1) by [(x), (y)] =
h(x y), where h is the Plancks constant, and then repeat all calculations of this section.
It turns out that Eqs. (3.9b), (3.9c) remain unchanged, but Eq. (3.9a) is modified as:
h = sin . Consequently, for any fixed value of , 0 limit is essentially equivalent
to h 0 limit. Since from Eqs. (3.9b), (3.9c) it follows that f and g when
0, the quantum Lax operator (3.2) indeed reproduces the classical Lax operator (2.1)
at h 0 limit.
4. Algebraic Bethe ansatz for the quantum monodromy matrix on an infinite

interval
For the purpose of taking infinite interval limit of QYBE (3.10), we define the quantum
analogue of classical monodromy matrix (2.3) as
T () =
lim
x2 +
x1
e(x2 , )Txx12 ()e(x1 , ),
(4.1)
where Txx12 () is given by Eq. (3.3). It may be observed that, exactly like the case of
classical Lax operator (2.1), the quantum Lax operator (3.2) also satisfies the following
relations:
Uq (x, ) = KUq (x, )K 1 ,
1
Uq (x, ) = LUq (x, )L
(4.2a)
(4.2b)
620
where K and L are same matrices which have appeared in (2.4). By using Eq. (4.2a)
and assuming to be a real parameter, is easy to show that the quantum monodromy
matrix (4.1) can be expressed in a symmetric form given by

A() B ()
.
T () =
(4.3)
B()
A ()
Since from Eq. (4.2b) it follows that A() = A() and B() = B(), it is necessary
to consider the quantum monodromy matrix (4.3) only within the range 0. In analogy
with T () (4.1), which is defined by quantising the classical monodromy matrix T ()
()
(2.3), one may also define the quantum analogue of the classical monodromy matrix T
(2.7). By using arguments similar to the classical case, it is easy to show that such quantum
monodromy matrix can be written in a symmetric form given by

()
A() B
,
T() =
(4.4)

B()
A ()
= B(). Thus we find that, the quantum

where = sign and A()
= A(), B()
monodromy matrix T() can be expressed like a U (2) group valued object when > 0 and
U (1, 1) group valued object when < 0. As a result, the commutation relations which were
derived earlier by assuming T() to be a U (2) group valued object [12] should correspond
to the case > 0.
For finding out the commutation relations among the operator elements of monodromy
matrix (4.1), it is required to get rid of the oscillatory terms which arise from the
product Txx12 () Txx12 () at the asymptotic limits x1 , x2 . To this end, we split
the L(x; , ) matrix (3.6) into two parts:
L(x; , ) = L0 (, ) + L1 (x; , ),
where L0 (, ) is given by
L0 (, ) = lim L(x; , )
|x|
i (2 + 2 )
4i (2 2 )
i 2
4 (
2 )
+ 2 )
0
0
0
and L1 (x; , ) is the field dependent part of L(x; , ), which vanishes at x . Due
to Eq. (3.8) it follows that
i 2
4 (
R(, )(x; , ) = (x; , )R(, ),
(4.5)
where (x; , ) = eL0 (,)x . By using the above mentioned splitting of L(x; , ), one
can derive the integral form of differential Eq. (3.5) as
Txx12 () Txx12 () = (x2 x1 ; , )
x2
.
.
+ dx (x2 x; , )..L1 (x; , )Txx1 () Txx1 ().. .
x1
621
Due to the presence of field dependent matrix L1 (x; , ), the second term in the r.h.s. of
above integral equation vanishes at the limit x1 , x2 . Consequently, at this limit, one
gets Txx12 () Txx12 () (x2 x1 ; , ), which is an oscillatory term. To get rid of this
unwanted term, we define an operator like
W (, ) =
lim
x1
x2 +
(x2 ; , )Txx12 () Txx12 ()(x1 ; , ),
(4.6)
which is clearly well behaved at the infinite interval limit. By using (3.10) and (4.5), it is
easy to verify that this W (, ) (4.6) satisfies an equation given by
R(, )W (, ) = W (, )R(, ),
(4.7)
which represents QYBE for the infinite interval limit.

Next, we want to express QYBE (4.7) through the direct product of two monodromy
matrices of the form (4.1). To this end, we note that W (, ) (4.6) may be rewritten as
W (, ) = C+ (, )T () T ()C (, ),
(4.8)
where
C+ (, ) = lim (x; , )E(x; , ),
(4.9a)
C (, ) = lim E(x; , )(x; , ),
(4.9b)
with E(x; , ) = e(x, ) e(x, ). Substituting the explicit forms of E(x; , ) and
ikx
(x; , ) to (4.9), and extracting the limits in the principal value sense: limx P ( e k )
= i(k), we obtain
1
0
0 0
1
0 0
0
C+ (, ) =
,
0 + (, ) 1 0
0
0
0 1
1
0
0 0
1
0 0
0
C (, ) =
(4.10)
,
0 (, ) 1 0
0
0
0 1
where
(, ) =

2i
2i
+ 2 2 2 = 2
.
2
2

2 i6
By using the expression (4.8), we rewrite QYBE (4.7) for the infinite interval limit as
R(, )C+ (, )T () T ()C (, )
= C+ (, )T () T ()C (, )R(, ).
(4.11)
By inserting the explicit forms of R(, ) (3.7), C (, ) (4.10), and T () (4.3) to

QYBE (4.11) and comparing its matrix elements from both sides, we finally obtain
[A(), A()] = 0,
(4.12a)
622

A(), A () = 0,
A()B () =
2 q
(4.12b)
2 q 1
B ()A()
2 2 i6

2 q 2 q 1
=
B ()A() 2 2 2 B ()A(), (4.12c)
2 2
2 q 2 q 1
A()B()
2 2 i6

2 q 2 q 1
=
A()B() 2 2 2 A()B(),
2
2

B()B () = (, )B ()B() + 4 2 2 A ()A(),
B()A() =
(4.12d)
(4.12e)
where q = ei and

4 2 2 2
8 2 2 2
2
.
(, ) = 1 +
2
( 2 i6)(2 2 + i6)
(2 2 )
The above commutation relations among the elements of quantum monodromy matrix (4.1)
are evidently valid for both positive and negative values
of the coupling constant . With the
= B(), one can also recast Eqs. (4.12)

help of transformations like A()
= A(), B()
as the commutation relations among the elements of the monodromy matrix T() (4.4)
and compare such commutation relations with their counterparts in Ref. [12]. It is easy
to see that the forms of commutation relations (4.12a)(4.12d) remain unaltered through
the above mentioned transformation and match with their counterparts
as given earlier.
= B(), Eq. (4.12e) may be

However, by using the transformation A()
= A(), B()
expressed as

B
() = (, )B
()B()

B()
+ 4| | 2 2 A ()A(),
which does not match at all with its counterpart [12] for either positive or negative
value of . It can be shown that, due to a problem which arises while taking the infinite
interval limit of QYBE, an incorrect commutation relation was obtained earlier between
and B
() for both positive and negative values of . It is interesting to
the operators B()

()] = 0. On the other hand,
observe that, for the case = , Eq. (4.12e) gives [B(),
B
from Eq. (2.16e) it follows that {b(), b ()} = 0 for = . Thus we find that, similar to
the case of NLS model [6], the correspondence between Poisson brackets and commutators
among some elements of monodromy matrix may turn out to be a quite nontrivial one.
Due to Eq. (4.12a) it follows that all operator valued coefficients occurring in the
expansion of ln A() in powers of would commute among themselves and, as a
consequence, the monodromy matrix (4.1) of DNLS model leads to a quantum integrable
system. With the help of Eq. (4.1), it is easy to find that A()|0 = |0. By using this
relation and Eq. (4.12c), it can be shown that
A()|1 , 2 , . . . , N =

N 2

q 2 q 1
r
r=1
2r 2 i6
|1 , 2 , . . . , N ,
(4.13)
623
where j s are all distinct real numbers and |1 , 2 , . . . , N B (1 )B (2 )

B (N )|0. Thus the states |1 , 2 , . . . , N diagonalise the generator of conserved
quantities for the quantum DNLS model. However, by using Eq. (4.13), one finds that the
eigenvalues corresponding to different expansion coefficients of ln A() would be complex
quantities in general. To extract real eigenvalues, we define another operator A()

through
as
the relation: A()
A(ei/2 ) and expand ln A()
ln A()
=

iCn
n=0
2n
(4.14)
With the help of Eqs. (4.13) and (4.14) it is easy to see that Cn s satisfy eigenvalue equations
like Cn |1 , 2 , . . . , N = n |1 , 2 , . . . , N , where the first few n s are explicitly given
by
0 = N,
1 = 2 sin
2j ,
2 = sin 2
j =1
4j .
(4.15)
j =1
In analogy with the classical case, one may now define the Hamiltonian for quantum
1
C2 . Eigenvalue equations corresponding to this Hamiltonian are
DNLS model as H = 8
evidently given by
N

1
4
H|1 , 2 , . . . , N = +
(4.16)
1 2
j |1 , 2 , . . . , N .
4
j =1
Till now we have assumed that j s are some real parameters, for which |1 , 2 , . . . ,
N represents a scattering state. One can also construct the quantum N -soliton state for
DNLS model [12] by choosing complex values of j given by

N +1
j ,
j = exp i
(4.17)
2
where is a real parameter and j [1, 2, . . . , N]. Thus j s are uniformly distributed on
a circle of radius . For this choice of j , Eq. (4.13) reduces to a simple form like
2

2 q N+1
N
|1 , 2 , . . . , N ,
A()|1 , 2 , . . . , N = q
(4.18)
2 2 q N+1
where the eigenvalue of A() has only one zero and one pole on the complex 2 -plane.
By using (4.18), we obtain the energy eigenvalue corresponding to the quantum N -soliton
state as
4 sin(2N)
|1 , 2 , . . . , N .
(4.19)
8 sin
In general, we may choose any positive integer value of N (greater than one) for
constructing a quantum soliton state. However, we now consider the DNLS model with
some particular values of coupling constant given by = sin = sin( 2m
n ), where m
and n are nonzero integers which do not have any common factor. By using Eq. (4.17), one
obtains j = j +n for this case. Since all j must take distinct values, we get N n as
H|1 , 2 , . . . , N =
624
a restriction on the number of quasi-particles which form a bound state for the quantum
DNLS model corresponding to coupling constant = sin( 2m
n ).
Thus, by applying the method of algebraic Bethe ansatz, we are able to construct
the exact eigenstates for quantum DNLS model. The commutation relation (4.12e) also
plays an important role in the framework of algebraic Bethe ansatz, since by using
this commutation relation one should be able to calculate the norm of eigenstates
|1 , 2 , . . . , N and various correlation functions of the DNLS system. However, it may
be noted that the commutation relation (4.12e) contains product of generalised functions
(2 2 i6)1 (2 2 + i6)1 , which does not make sense at the limit . As a
result, the action of operators B (), B() are not well defined on the Hilbert space [8] and
eigenstates like |1 , 2 , . . . , N are not normalised on the -function. However, it is well
known that, one can avoid this type of problem in the case of NLS equation by considering
the quantum analogue of classical reflection operators [1,6,7]. So, in analogy with the case
of NLS equation, at present we consider a reflection operator given by
R () = A1 ()B(),
(4.20)
and its adjoint R(). By using Eqs. (4.12), we find that such reflection operators satisfy
well defined commutation relations like
R()R() = S 1 (, )R()R(),
R ()R () = S 1 (, )R ()R (),

R()R () = S(, )R ()R() + 42 2 2 ,
(4.21)
where
2 q 2 q 1
.
(4.22)
2 q 1 2 q
It is evident that these commutation relations among reflection operators of DNLS model
are nicely encoded in a form of ZamolodchikovFaddeev algebra [1,14] and S(, ) (4.22)
represents the nontrivial S-matrix element of two-body scattering between the related
quasi-particles. It is easy to check that this S(, ) satisfies the conditions given by
S(, ) =
S 1 (, ) = S(, ) = S (, ),
(4.23)
and remains nonsingular at the limit . As a result, the action of operators

R (), R() would be well defined on the Hilbert space and eigenstates like R (1 )
R (2 ) R (N )|0 can be normalised on the -function.
We find that the classical monodromy matrix for DNLS model can be written as a
SU(2) (SU(1, 1)) group valued object for positive (negative) value of the corresponding
coupling constant. By using such symmetric form of classical monodromy matrix, we
derive Poisson bracket relations among the scattering data of the DNLS model for all
values of the coupling constant. We also quantise the monodromy matrix of DNLS model
on a finite interval. A variant of quantum inverse scattering method, which can be applied
625
to field models without performing any lattice regularisation, fixes all parameters in the
quantum monodromy matrix of DNLS model in a nontrivial way. Similar to the classical
case, this quantum monodromy matrix exhibits U (2) (U (1, 1)) symmetry for positive
(negative) value of the coupling constant. By applying quantum inverse scattering method,
we derive all possible commutation relations among the elements of such monodromy
matrix. Infinite interval limits of these commutation relations enable us to construct the
exact eigenstates for quantum DNLS model through algebraic Bethe ansatz. In this context,
we consider the DNLS model with some special values of coupling constant given by
= sin = sin( 2m
n ), where m and n are nonzero integers which do not have any
common factor. It turns out that the number of quasi-particles, which form a bound state
for such quantum DNLS model, cannot exceed the value n.
We also obtain the commutation relation between creation and annihilation operators
associated with quasi-particles of DNLS model and find out the S-matrix for two-body
scattering. Such a commutation relation between creation and annihilation operators should
play an important role in a future study, since by using it one might be able to calculate
the norm of Bethe eigenstates and various correlation functions of the DNLS system.
It may be noted that, there exist quantum integrable multicomponent generalisations of
NLS model which can be diagonalised through algebraic Bethe ansatz [1518]. A large
class of multicomponent classical DNLS models, having infinite number of conserved
quantities, are also studied in the literature [19,20]. However, the Hamiltonian structure
of such multicomponent DNLS models have not yet received much attention. So it might
be interesting to investigate whether there exist some multicomponent generalisations
of classical DNLS model which are integrable in the Liouville sense and retain their
integrability property even after quantisation.
Appendix A
Here we give the details of deriving Eq. (3.5). A direct attempt to calculate x 2 (Txx12 ()
x2
Tx1 ()) by using Eq. (3.4) evidently leads to indeterminate expressions of the form
[Txx12 (), (x2 )]. To bypass this problem, we follow the method of extension [2] which
shifts the upper limit of one monodromy matrix (say Txx12 ()) by introducing a small
parameter 6 and takes 6 0 limit only after differentiating the product Txx12 +6 ()Txx12 ()
with respect to x2 . Thus, by using Eq. (3.4), we obtain

x2 +6
Tx1 () Txx12 ()
x2
.

.
= .. Uq (x2 + 6, ) 1 + 1 Uq (x2 , ) Txx12 +6 () Txx12 ()..
+ K+ + K ,
where

K+ = i Txx12 +6 (), (x2 ) + Txx12 ()

+ if Txx12 +6 (), (x2 ) e11 Txx12 ()(x2 )

ig Txx12 +6 (), (x2 ) e22 Txx12 ()(x2 ),
(A.1)
626

K = i Txx12 +6 () (x2 + 6), Txx12 ()

+ if (x2 + 6)e11 Txx12 +6 () (x2 + 6), Txx12 ()

ig (x2 + 6)e22 Txx12 +6 () (x2 + 6), Txx12 () .
Now we consider the case 6 > 0. Since (x2 + 6) commutes with (x), (x) for all x
lying within x1 and x2 , we can write [(x2 + 6), Txx12 ()] = 0 and K = 0 for this case. So,
we have to calculate only the nontrivial commutator [Txx12 +6 (), (x2 )] which appears in
the expression of K+ . To this end, we consider a transformation which replaces the
classical variables (x) and (x) by quantum operators (x) and (x), respectively
( 1 denotes the reverse transformation). By applying a correspondence principle [2] to
the present case, we obtain

x +6
Tx12 (), (x2 ) = i : Txx12 +6 (q; ), (x2 ) : ,
(A.2)
where Txx12 +6 (q; ) represents a classical monodromy matrix given by
Txx12 +6 (q; ) =
x2 +6
P exp
Uq (x, ) dx,
x1
and Uq (x, ) =
find that
1 U
q (x, ).
Txx12 +6 (q; ), (x2 )
By using the fundamental PB relations (1.4), it is easy to

x2 +6

dx Txx2 +6 (q; ) Uq (x, ), (x2 ) Txx1 (q; )
x1

= Txx22 +6 (q; ) f (x2 )e11

g (x2 )e22 + Txx12 (q; ).
Taking 6 0 limit of the above expression and inserting it to (A.2), we obtain

lim Txx12 +6 (), (x2 ) = i f (x2 )e11 g (x2 )e22 + Txx12 ().
60
(A.3)
Taking 6 0 limit also in Eq. (A.1) and using (A.3), we finally obtain the differential
Eq. (3.5). Note that, instead of 6 > 0, we could have chosen 6 < 0 in Eq. (A.1). Only the
commutator [(x2 + 6), Txx12 ()] gives a nontrivial contribution for this case. However,
repeating similar steps as outlined above and finally taking the 6 0 limit, we get again
the same differential Eq. (3.5).
References
[1] L.D. Faddeev, Sov. Sci. Rev. C 1 (1980) 107;
L.D. Faddeev, in: J.B. Zuber, R. Stora (Eds.), Recent Advances in Field Theory and Statistical Mechanics,
North-Holland, Amsterdam, 1984, p. 561.
[2] E.K. Skylanin, in: M. Jimbo (Ed.), YangBaxter Equation in Integrable Systems, Adv. Ser. in Math. Phys.,
Vol. 10, World Scientific, Singapore, 1990, p. 121.
[3] V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and Correlation
Functions, Cambridge Univ. Press, Cambridge, 1993 and references therein.
627
[4] Z.N.C. Ha, Quantum Many-Body Systems in One Dimension, World Scientific, Singapore, 1996 and
references therein.
[5] J. Honerkamp, P. Weber, A. Wiesler, Nucl. Phys. B 152 (1979) 266.
[6] H.B. Thacker, D. Wilkinson, Phys. Rev. D 19 (1979) 3660;
D.B. Creamer, H.B. Thacker, D. Wilkinson, Phys. Rev. D 21 (1980) 1523.
[7] K.M. Case, J. Math. Phys. 25 (1984) 2306.
[8] E. Gutkin, Phys. Rep. 167 (1988) 1.
[9] D.J. Kaup, A.C. Newell, J. Math. Phys. 19 (1978) 798.
[10] A. Kundu, J. Phys. A 21 (1988) 945.
[11] H.H. Chen, Y.C. Lee, C.S. Liu, Phys. Scr. 20 (1979) 490.
[12] A. Kundu, B. Basu-Mallick, J. Math. Phys. 34 (1993) 1052.
[13] B. Basu-Mallick, A. Kundu, Phys. Lett. B 287 (1992) 149.
[14] A.B. Zamolodchikov, A.B. Zamolodchikov, Ann. Phys. 120 (1979) 253.
[15] F.C. Pu, B.H. Zhao, Phys. Rev. D 30 (1984) 2253.
[16] F.C. Pu, B.H. Zhao, Nucl. Phys. B 275 [FS 17] (1986) 77.
[17] S. Murakami, M. Wadati, J. Phys. A 29 (1996) 7903.
[18] M. Mintchev, E. Ragoucy, P. Sorba, P. Zaugg, J. Phys. A 32 (1999) 5885.
[19] T. Tsuchida, M. Wadati, Phys. Lett. A 257 (1999) 53;
T. Tsuchida, M. Wadati, Inv. Prob. 15 (1999) 1363.
[20] P.J. Olver, V.V. Sokolov, Inv. Prob. 14 (1998) L5.

CUMULATIVE AUTHOR INDEX B631B634
Ahn, Ch.
Akemann, G.
Aliani, P.
Alvarez, O.
Antonelli, V.
Antoniadis, I.
Antoniadis, I.
Aoki, H.
Apreda, R.
Argeri, M.
Armoni, A.
B634 (2002) 141

B631 (2002) 471
B634 (2002) 393
B633 (2002) 309
B634 (2002) 393
B631 (2002) 3
B631 (2002) 66
B634 (2002) 71
B631 (2002) 342
B631 (2002) 388
B632 (2002) 240
Bantay, P.
Barnich, G.
Barto, E.
Basu-Mallick, B.
Becchi, C.
Becirevic, D.
Benakli, K.
Bertolini, M.
Bhattacharyya, T.
Bianchi, M.
Blum, J.D.
Boyanovsky, D.
Braden, H.W.
Brandt, F.
Brecher, D.
Brignole, A.
Bueno, A.
Buras, A.J.
Burkardt, M.
B633 (2002) 365

B633 (2002) 3
B632 (2002) 330
B634 (2002) 611
B633 (2002) 250
B634 (2002) 105
B631 (2002) 3
B632 (2002) 257
B634 (2002) 611
B631 (2002) 159
B634 (2002) 3
B632 (2002) 121
B633 (2002) 414
B633 (2002) 3
B634 (2002) 23
B631 (2002) 195
B631 (2002) 239
B631 (2002) 219
B632 (2002) 311
Cabrera-Carnero, I.
Cacciapaglia, G.
Cadoni, M.
Caffo, M.
Campanelli, M.
Cao, F.J.
Caravaglios, F.
B634 (2002) 433

B634 (2002) 230
B632 (2002) 383
B634 (2002) 309
B631 (2002) 239
B632 (2002) 121
B633 (2002) 193
0550-3213/2002 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 4 2 7 - 3
Carta, P.
Chetyrkin, K.G.
Chu, C.-S.
Ciocarlie, C.
Cirelli, M.
Ciuchini, M.
Clark, T.E.
Cristadoro, G.
Czarnecki, A.
Czyz, H.
B632 (2002) 383

B634 (2002) 413
B632 (2002) 219
B632 (2002) 303
B634 (2002) 230
B634 (2002) 105
B632 (2002) 3
B634 (2002) 230
B631 (2002) 219
B634 (2002) 309
Damgaard, P.H.
Degrassi, G.
Delduc, F.
Deser, S.
De Vega, H.J.
Di Vecchia, P.
Doikou, A.
Dolgov, A.D.
Dotsenko, V.S.
Dubnicka, S.
Dubnickov, A.-Z.
Dukelsky, J.
B633 (2002) 97
B631 (2002) 195
B631 (2002) 403
B631 (2002) 369
B632 (2002) 121
B631 (2002) 95
B634 (2002) 591
B632 (2002) 363
B631 (2002) 426
B632 (2002) 330
B632 (2002) 330
B634 (2002) 483
Enger, H.
B631 (2002) 95
Fateev, V.A.
Fjelstad, J.
Flohr, M.
Font, A.
Franco, E.
Franco, E.
Freedman, D.Z.
Freitas, A.
Froggatt, C.D.
Frolov, S.
Fuchs, J.
B634 (2002) 546

B633 (2002) 379
B634 (2002) 511
B634 (2002) 51
B633 (2002) 212
B634 (2002) 105
B631 (2002) 159
B632 (2002) 189
B631 (2002) 285
B632 (2002) 69
B633 (2002) 379
Gavai, R.V.
Gersdorff, G.V.
B633 (2002) 127

B634 (2002) 90
630
Giedt, J.
Gimnez, V.
Giusto, S.
Gomes, J.F.
Gorsky, A.
Gracey, J.A.
B632 (2002) 397

B634 (2002) 105
B633 (2002) 250
B634 (2002) 433
B633 (2002) 414
B634 (2002) 192
Hambye, T.
Hansen, S.H.
Harlander, R.V.
Harmark, T.
Hartnoll, S.A.
Hatsuda, M.
Hebecker, A.
Heller, U.M.
Hernndez, A.
Hollik, W.
Hwang, S.
B633 (2002) 171

B632 (2002) 363
B634 (2002) 413
B632 (2002) 257
B631 (2002) 325
B632 (2002) 114
B632 (2002) 101
B633 (2002) 97
B634 (2002) 51
B632 (2002) 189
B633 (2002) 379
Imbimbo, C.
Imeroni, E.
Iso, S.
B633 (2002) 250

B631 (2002) 95
B634 (2002) 71
Jacobsen, J.L.
Janssen, B.
B631 (2002) 426

B634 (2002) 23
Kamimura, K.
Khlebnikov, S.
Khuri, R.R.
Kleinert, H.
Koike, Y.
Krauss, F.
Khn, J.H.
Kuraev, E.A.
B632 (2002) 114

B631 (2002) 307
B633 (2002) 295
B632 (2002) 51
B632 (2002) 311
B633 (2002) 237
B634 (2002) 413
B632 (2002) 330
Laugier, A.
Lavignac, S.
Lee, P.
Lee, P.
Li, T.
Litim, D.F.
Lopez, E.
Love, S.T.
Lozano, Y.
Lozano-Tellechea, E.
L, H.
Lubicz, V.
Lukierski, J.
B631 (2002) 3
B633 (2002) 139
B632 (2002) 283
B632 (2002) 303
B633 (2002) 83
B631 (2002) 128
B632 (2002) 240
B632 (2002) 3
B634 (2002) 23
B631 (2002) 95
B633 (2002) 114
B633 (2002) 212
B632 (2002) 219
Maggiore, M.
Malbouisson, A.P.C.
Malbouisson, J.M.C.
Mangano, M.L.
Martinelli, G.
B631 (2002) 342

B631 (2002) 83
B631 (2002) 83
B632 (2002) 343
B634 (2002) 105
Masiero, A.
Masina, I.
Mastrolia, P.
Mescia, F.
Mignemi, S.
Minasian, R.
Misiak, M.
Montani, G.
Morariu, B.
Moretti, M.
Moss, I.G.
Moss, I.G.
B634 (2002) 105

B633 (2002) 139
B631 (2002) 388
B633 (2002) 212
B632 (2002) 383
B631 (2002) 43
B631 (2002) 219
B634 (2002) 370
B634 (2002) 326
B632 (2002) 343
B631 (2002) 500
B632 (2002) 173
Navas-Concha, S.
Navelet, H.
Naylor, W.
Nepomechie, R.I.
Ng, S.
Nguyen, X.S.
Niclasen, R.
Nicolis, A.
Nielsen, H.B.
Nowling, S.R.
B631 (2002) 239

B634 (2002) 291
B632 (2002) 173
B631 (2002) 519
B634 (2002) 209
B631 (2002) 426
B633 (2002) 97
B631 (2002) 342
B631 (2002) 285
B632 (2002) 3
Obers, N.A.
Odesskii, A.
Onofri, E.
Ooguri, H.
B632 (2002) 257

B633 (2002) 414
B634 (2002) 546
B632 (2002) 283
Papinutto, M.
Park, J.
Park, J.
Park, M.-I.
Pastor, S.
Pearce, P.A.
Perry, M.
Peschanski, R.
Petcov, S.T.
Picariello, M.
Pittau, R.
Pokotilov, A.
Polychronakos, A.P.
B634 (2002) 105

B632 (2002) 283
B632 (2002) 303
B634 (2002) 339
B632 (2002) 363
B631 (2002) 447
B634 (2002) 209
B634 (2002) 291
B632 (2002) 363
B634 (2002) 393
B632 (2002) 343
B633 (2002) 295
B634 (2002) 326
Quirs, M.
B634 (2002) 90
Raffelt, G.G.
Ravindran, V.
Remiddi, E.
Remiddi, E.
Reyes, J.
Richard, C.
Riotto, A.
Riotto, A.
Romn, J.M.
Roudeau, P.
B632 (2002) 363

B634 (2002) 247
B631 (2002) 388
B634 (2002) 309
B634 (2002) 105
B631 (2002) 447
B631 (2002) 342
B634 (2002) 90
B634 (2002) 483
B633 (2002) 193
Rubbia, A.
Rubtsov, V.
Ruelle, P.
B631 (2002) 239

B633 (2002) 414
B633 (2002) 345
Sakaguchi, M.
Santachiara, R.
Santana, A.E.
Savoy, C.A.
Schlittgen, B.
Schlottmann, P.
Semikhatov, A.M.
Semikoz, D.V.
Sezgin, E.
Shore, G.M.
Sierra, G.
Silvestrini, L.
Singer, I.M.
Skenderis, K.
Slavich, P.
Smith, J.
Soff, G.
Sorin, A.S.
Sotkov, G.M.
Stocchi, A.
Sturani, R.
Sundell, P.
Suyama, T.
Svetitsky, B.
B632 (2002) 114

B631 (2002) 426
B631 (2002) 83
B633 (2002) 139
B632 (2002) 155
B634 (2002) 571
B633 (2002) 379
B632 (2002) 363
B634 (2002) 120
B633 (2002) 271
B634 (2002) 483
B634 (2002) 105
B633 (2002) 309
B631 (2002) 159
B631 (2002) 195
B634 (2002) 247
B633 (2002) 237
B631 (2002) 403
B634 (2002) 433
B633 (2002) 193
B631 (2002) 66
B634 (2002) 120
B634 (2002) 71
B633 (2002) 97
631
Takanishi, Y.
Tarantino, C.
Tipunin, I.Yu.
Tomasiello, A.
Torrente-Lujan, E.
Tseytlin, A.A.
B631 (2002) 285

B633 (2002) 212
B633 (2002) 379
B631 (2002) 43
B634 (2002) 393
B632 (2002) 69
Urban, J.
B631 (2002) 219
Van den Bossche, B.

Van Neerven, W.L.
Vzquez-Poritz, J.F.
Verhoeven, O.
Vernizzi, G.
B632 (2002) 51
B634 (2002) 247
B633 (2002) 114
B633 (2002) 345
B631 (2002) 471
Waldron, A.
Walter, W.
Weiglein, G.
Westerberg, A.
Wettig, T.
Woo, K.
B631 (2002) 369

B632 (2002) 189
B632 (2002) 189
B632 (2002) 257
B632 (2002) 155
B634 (2002) 141
Yue, R.-H.
B634 (2002) 571
Zakrzewski, W.J.
Zemlyanaya, E.
Zimerman, A.H.
Zinn-Justin, P.
Zwirner, F.
B632 (2002) 219

B632 (2002) 330
B634 (2002) 433
B634 (2002) 417
B631 (2002) 195

Calculation of Nonperturbative Terms in Open String Models: Julie D. Blum

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Calculation of Nonperturbative Terms in Open String Models: Julie D. Blum

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 634 (2002) 322

Calculation of nonperturbative terms in open string

J.D. Blum / Nuclear Physics B 634 (2002) 322

Fig. 1. Toric diagram of Z2 Z2 blowup.

J.D. Blum / Nuclear Physics B 634 (2002) 322

Fig. 2. Toric diagram of Z2 Z2 flopped phase: r3 r3 .

to a two-torus along planes xi = 0, to a circle along lines xi = xj = 0, and to a point

and cyclic permutations where i = zi zi and zi = eti with ti = ri + ii , the initial

J.D. Blum / Nuclear Physics B 634 (2002) 322

and y4 = 0 to fix a constant solution of the D-term equations yields

A noncompact, supersymmetric Lagrangian three-cycle in the original manifold is

Let us write the above in the form

counts the number of domain walls wrapping the two-cycle parametrized by i mi ti

J.D. Blum / Nuclear Physics B 634 (2002) 322

Fig. 3. Riemann surface P (u, v) = 0 in mirror of Z2 Z2 blowup.

J.D. Blum / Nuclear Physics B 634 (2002) 322

Fig. 4. Phases for noncompact three-cycle in Z2 Z2 blowup.

Extracting the piece of v that is independent of eu gives v, and one has

q1a+d+e q2m+nabc+d q3m+nc+e

are integers. One finds

J.D. Blum / Nuclear Physics B 634 (2002) 322

n!(n 1)!(m n)!

Correspondingly, u = u v + t2 = t3 2 t3 + t2 2 t2 and v = v. Phase II is almost

J.D. Blum / Nuclear Physics B 634 (2002) 322

J.D. Blum / Nuclear Physics B 634 (2002) 322

Fig. 5. Toric diagram of Z2 Z4 blowup.

Fig. 6. Toric diagram of Z2 Z4 flopped phase: r7 r7 .

J.D. Blum / Nuclear Physics B 634 (2002) 322

Fig. 7. Toric diagram of Z2 Z4 flopped phase: r7 r7 , r4 r2 r2 r4 , r3 r3 .

solutions are as follows:

where is a solution corresponding to a four-cycle,

(1)n+p+r+m (n + m + p r 2q 1)!z1r z2 z3m z7n z8s z9

(1)s (2r s 1)!zr zr z2n zs zn

(1)n+r (2s n r 1)!zr zr z2n zs zn

(1)s (2n s 1)!zr zr z2n zs zn

J.D. Blum / Nuclear Physics B 634 (2002) 322

and the zi s must be found as a function of the qi s perturbatively.

In the limit that z7 = z8 = z9 = 0,

3.2. The mirror and open string calculations

J.D. Blum / Nuclear Physics B 634 (2002) 322

Fig. 8. Riemann surface P (u, v) = 0 in mirror of Z2 Z4 blowup.

Fig. 9. Phases for noncompact three-cycle in Z2 Z4 blowup.

J.D. Blum / Nuclear Physics B 634 (2002) 322

(1)b+d (2n + m 1)!

J.D. Blum / Nuclear Physics B 634 (2002) 322

the two-dimensional worldsheet with boundary and p is a boundary field which

J.D. Blum / Nuclear Physics B 634 (2002) 322

linear sigma model corresponding to the numerical invariants. This correspondence

J.D. Blum / Nuclear Physics B 634 (2002) 322

Fig. 10. Flopped phase of Z2 Z4 blowup: r1 r1 , r6 r6 .

J.D. Blum / Nuclear Physics B 634 (2002) 322

(1)n+m (3n m 1)!zn zm zp

(1)m+p (2n m p 1)!zp zn zm

Fig. 11. Toric diagram of Z7 blowup.

J.D. Blum / Nuclear Physics B 634 (2002) 322

(1)m (2n m 1)!zp zm zn

Fig. 12. Riemann surface P (u, v) = 0 in mirror of Z7 blowup.

J.D. Blum / Nuclear Physics B 634 (2002) 322

Fig. 13. Phases for noncompact three-cycle in Z7 blowup.

In phase II u is unchanged while v = v u = P . The results are shown in the

J.D. Blum / Nuclear Physics B 634 (2002) 322

Nuclear Physics B 634 (2002) 2350

Dielectric fundamental strings in matrix string

where i, j = 1, . . . , 9. Usually R is taken to be R = gs , and so one has 2 =

where T0 = 1/R = 1/(gs ) and

9 . We then define the dimensionless worldsheet coordinates

multiplied by a factor of R/ , each [X, X] term appears multiplied by a factor of 1/gs ,

= b09 + b0i DX + bi9 X + bij X DX dt dx.