Received 12 April 2006; received in revised form 21 June 2006; accepted 5 July 2006
Available online 10 August 2006
Abstract
We consider the scenario in which the mass matrices of the charged fermions in the SO(10) grand unified
theory are generated exclusively by renormalizable Yukawa couplings to one 10 120 representation of
scalars. We analyze, partly analytically and partly numerically, this scenario in the threegenerations case.
We demonstrate that it leads to unification of the b and masses at the GUT scale. Testing this scenario
against the mass values at the GUT scale, obtained from the renormalizationgroup evolution in the minimal
SUSY extension of the Standard Model, we find that it is not viable: either the downquark mass or the topquark mass must be unrealistically low. If we include the CKM mixing angles in the test, then, in order that
the mixing angles are well reproduced, either the topquark mass or the strangequark mass together with
the downquark mass must be very low. We conclude that, assuming a SUSY SO(10) scenario, chargedfermion mass generation based exclusively on one 10 120 representation of scalars is in contradiction
with experiment.
2006 Elsevier B.V. All rights reserved.
1. Introduction
All the fermionic multiplets of one Standard Model generation, plus one righthanded neutrino
singlet, fit exactly into the 16dimensional irreducible representation of the grand unification
group SO(10). This is the unique and distinguishing feature of the unified gauge theories (GUTs)
* Corresponding author.
based on this group [1]. As a bonus, the presence of three (for three generations, as we shall
assume in this paper) righthanded neutrino singlets allows one to incorporate into this GUT the
seesaw mechanism of type I [2].
However, when it comes to the scalar sector and to fermion mass generation the uniqueness
of the SO(10) GUT is lost and numerous ramifications exist. One possible strategy to limit the
freedom in the scalar sector is to confine oneself to renormalizable termsfor a review see, for
instance, [3]. In that case, the scalar representations coupling to the fermions are determined by
the relation [4,5]
16 16 = (10 126)S 120AS ,
(1)
where the subscripts S and AS denote, respectively, the symmetric and antisymmetric parts
of the tensor product. Thus, scalars with renormalizable Yukawa couplings to the fermions must
transform under SO(10) either as 10, 126, or 120 (the 10 and 120 are real representations; the
126 is complex). A minimal supersymmetric (SUSY) scenariowhich has builtin the gaugecoupling unification of the minimal SUSY extension of the Standard Model (MSSM)making
use of one 10 and one 126 for the Yukawa couplings [6] has recently received a lot of attention.
This attention was triggered by the observation [7] that maximal atmospheric neutrino mixing
may in this theory be related to b unification via the type II seesaw mechanism [8]. Detailed
and elaborate studies of this minimal theory have been performed for its Yukawa couplings [9
12] and scalar potential [13,14]. This minimal SUSY SO(10) GUT works very well, since its
Yukawa couplings are able to fit all fermion masses and mixings, allowing in particular for small
quark mixings simultaneously with large leptonic mixings. However, in this context a minimal
Higgs scalar sector is too constrained [15] and does not allow to produce large enough neutrino
masses [16].
As a way out, the 120 scalar representationwhich had been somewhat arbitrarily left out
may be used for a rescue [17,18]. In [10,11] that representation was only taken as a perturbation
of the minimal scenario, to cure minor deficiencies in the fermionic sector. However, in [19] it
was pointed out that the antisymmetric coupling matrix of the 120 could be responsible for the
different features of quark and lepton mixing, since that matrix has different weights in all four
Diractype mass matricesi.e., in the Dirac mass matrices for the uptype quarks, downtype
quarks, charged leptons, and neutrinos. Thinking along this line, the roles of the 120 and 126
could be interchanged in the chargedfermion sector: the brunt could be borne by one 10 120,
and the Yukawa couplings of the 126 would be just a perturbation. This thought is realized in the
model of [17], where the scalar 126 is still a protagonist in the neutrino sector, through the type I
seesaw mechanism.
In this paper we investigate the extreme form of this scenario of [17], namely we assume that
the 126 plays no role whatsoever in the Yukawa couplings to the charged fermions,1 and may
be important only in the neutrino sector, where it would responsible for large Majorana neutrino
masses. Thus, we base our investigation on the following assumptions:
(i) The chargedfermion mass matrices result solely from the Yukawa couplings of one 10 and
one 120 scalar multiplets.
(ii) The mechanism for the generation of the lightneutrino mass matrix is the type I seesaw
mechanism, possibly with some admixture of type II.
1 This idea was previously put forward in [20].
Due to the first assumption, the Yukawacoupling matrix of the 126 can be used freely for the
neutrino mass matrix and, therefore, one can accommodate any neutrino masses and lepton mixing that one wants, through either the type I or type II seesaw mechanisms. The tight connection
between the chargedfermion and neutrino sectors is lost, and the predictive power of the model
for the neutrino sector too. The subject of this paper is then only the discussion of the chargedfermion masses and of quark (CKM) mixing under the assumption (i), and the working out of
where this assumption is successful and where it might fail.
The chargedfermion sector and the Yukawa couplings in [18] coincide with ours. Still, our
results do not, in general, apply to that model. The reason is that its authors assume split supersymmetry, where the renormalizationgroup evolution of the fermion masses differs from the one
of the MSSM. Indeed, in order to test any specific scenario one must use the chargedfermion
masses and the quark mixings at the GUT scale. Having in mind a SUSY SO(10) GUT and
the MSSM, we use in this paper the values computed in [21] with the renormalizationgroup
evolution of the MSSM.
This paper is organized as follows. In Section 2 we discuss the mass matrices and count the
number of parameters. Basisinvariant quantities are introduced in Section 3. The derivation of
some inequalities, and b unification, are discussed in Section 4. In Section 5 we show that a
partly analytical treatment of our scenario is possible when the Yukawacoupling matrices are
assumed to be real. Section 6 explains our procedure for the numerical fit of the mass matrices
to the fermion masses and to the CKM mixing angles at the GUT scale. We present our results
in Section 7, which is followed by a brief summary in Section 8.
2. The chargedfermion mass matrices
The mass Lagrangian that we are concerned with is
LM = dL Md dR L M R uL Mu uR + H.c.
(2)
The symmetric and antisymmetric Yukawa couplings of one 10 and one 120 scalar representations, respectively [4], generate the mass matrices, which at the GUT scale may be parametrized
as
Md = S + ei A,
M = S + rei A,
Mu = pS + qei A,
(3)
S being symmetric while A is antisymmetric. The parameters p, q, and r are real and positive.
The matrix S is proportional to the Yukawacoupling matrix of the 10, while A is proportional
to the Yukawa couplings of the 120. The factors ei , rei , p, and qei depend on some ratios of
vacuum expectation values.
We may perform changes of weak basis
S U SU T ,
A U AU T ,
(4)
where U is unitary. In this way we may reach convenient weak bases. We may for instance use
U to diagonalize S:
a 0 0
0
z y
S= 0 b 0 ,
A = z 0
(5)
x ,
0 0 c
y x 0
with real and nonnegative a, b, and c. Alternatively, we may use U to force A to have only two
nonzero matrix elements, and moreover two matrix elements of S to vanish:
a f 0
0 0 0
S= f b d ,
A= 0 0 x .
(6)
0 d c
0 x 0
In the weak basis (6), again, we may choose a, b, and c to be real and nonnegative.
As for the number of degrees of freedom in the mass matrices (3), we consider two cases:
In the complexYukawacouplings case, the matrices S and A are a priori complex and contain nine independent matrix elements (six in S and three in A), hence nine moduli and nine
phases. One of three phases , , and may be absorbed in the definition of A. Through a
weakbasis transformation we may eliminate the three moduli and six phases which parametrize U . In that case the model has, therefore, nine real parameters and five phases.
In the realYukawacouplings case, in which CP violation is considered to be spontaneous,
the matrices S and A are a priori real and contain nine independent moduli. If we want to
preserve the reality of S and A, the matrix U of the weakbasis transformation (4) must be
chosen real (orthogonal),2 hence it contains three real parameters. One ends up with nine
real parameters as before, but only three phases.
With these 14 (in the complex case) or 12 (in the real case) parameters we must try and fit 13
observables: nine chargedfermion masses and four parameters of the CKM matrix. Even if there
is, in the complex case, an excessive number of parameters, the fitting may prove impossible,
due to the fact that a large number of those parameters are phases.
3. Fermion masses and invariants
We first confine ourselves to the masses. For brevity of notation we introduce
d = m2d + m2s + m2b ,
= m2e + m2 + m2 ,
= m2e m2 + m2 m2 + m2 m2e ,
= m2e m2 m2 ,
(7)
(8)
(9)
= s2 + 2r a2 ,
(10)
u = p s2 + 2q a2 ,
(11)
2 If S and U are assumed to be real, then one may obtain a weak basis of the form (5), but a, b, and c must be allowed
to be negative. It is only when we allow U to include some i factors that we may obtain nonnegative a, b, and c; but then
x, y, and z will not necessarily be real.
where
s2 = tr SS ,
1
a2 = tr AA ;
2
(12)
(13)
also,
d = s4 + a22 + 2z4 + 2 Re e2i z4 ,
= s4 + r 4 a22 + 2r 2 z4 + 2r 2 Re e2i z4 ,
u = p 4 s4 + q 4 a22 + 2p 2 q 2 z4 + 2p 2 q 2 Re e2i z4 ,
(14)
(15)
(16)
where
1 2
s2 tr SS SS ,
2
z4 = s2 a2 + tr SS AA ,
1
z4 = tr AS AS ;
2
finally,
2
d = s3 + e2i z3 ,
2
= s3 + r 2 e2i z3 ,
2
u = p 3 s3 + pq 2 e2i z3 ,
s4 =
(17)
(18)
(19)
(20)
(21)
(22)
where
s3 = det S,
1
z3 = tr SA2 tr S tr A2 .
2
(23)
(24)
4. b unification
In [18] the mass matrices for the chargedfermion sector are the same as in this paper, but the
discussion is confined to the twogenerations case. In that paper, approximate b unification is
traced back to some inequalities derived from the specific structure of the mass matrices. Here
we show that analogous inequalities hold in the threegenerations case.
It is convenient to use the weak basis of Eq. (5). We remind that, in that weak basis, a, b, and
c are real and nonnegative, while x, y, and z are in general complex. One has
s2 = a 2 + b2 + c2 ,
(25)
(26)
s4 = a b + b c + c a ,
2 2
2 2
2 2
(27)
(28)
(29)
s3 = abc,
(30)
z3 = ax 2 + by 2 + cz2 .
(31)
Note that z3 and z4 are in general complex, while the other parameters are real.
From Eqs. (16), (29), and (28) we derive
u p 4 s4 + q 4 a22 + 2p 2 q 2 z4 2p 2 q 2 bcx2 + cay2 + abz2
= p 4 s4 + q 4 a22 + 2p 2 q 2 a 2 bc x2 + b2 ca y2 + c2 ab z2 .
(32)
(33)
c a.
(34)
c2 ab a 2 bc.
(35)
Applying the inequalities (35) to the inequality (33) and remembering Eq. (26), we obtain
u p 4 s4 + q 4 a22 + 2p 2 q 2 a 2 bc a2 .
(36)
We next rewrite Eqs. (11) and (25) as
a2 =
1
u p 2 a 2 + b 2 + c 2 .
2
2q
(37)
We plug this equation into inequality (36) and find after some algebra that
u
2
1
u p 2 (b + c)2 + F,
4
(38)
where
p2 a 2
3p 2 a 2
2
2
F=
u + p (b + c)
.
2
2
(39)
(40)
(41)
(42)
u 2 u p 2 (b + c)2 u + 2 u .
u
d 2 d (b + c)2 d + 2 d ,
2 (b + c)2 + 2 .
(43)
(44)
(45)
Inequalities (45) should be compared with those of [18]. One reaches the same conclusion as
mt
mb
(46)
2u
.
u
2mc mb
a
.
mt
(47)
(48)
Numerically, using the values of the quark masses in the MSSM at the GUT scale, as given in
[21], one obtains for instance a 3.8 MeV for tan = 10.
5. Analytical treatment of the real case
In this section we analyze the case of real Yukawacoupling matrices, i.e., the case of real S
and A. In this case it is convenient to define
x1 = tr S,
1
x2 = x12 tr S 2 .
2
(49)
(50)
Then,
s2 = tr S 2 = x12 2x2 ,
(51)
s4 = x22
(52)
2x1 s3 .
With S and A real, z4 is real, and moreover it is not independent from z4 , rather
z4 z4 = x2 a2 x1 z3 .
(53)
= x22 2x1 s3 + a22 2x2 a2 + 2x1 z3 + 2 1 + cos(2) z4
= (x2 a2 )2 + 2x1 (z3 s3 ) + 2 1 + cos(2) z4 ,
2x1 s3 + r 4 a22 + 2r 2 z4 + 2r 2 cos(2 )z4
2x1 s3 + r 4 a22 2r 2 x2 a2 + 2r 2 x1 z3 + 2r 2 1 + cos(2 ) z4
2
= x2 r 2 a2 + 2x1 r 2 z3 s3 + 2r 2 1 + cos(2 ) z4 .
(54)
= x22
= x22
(55)
Now, plugging Eq. (51) into Eqs. (9) and (10), one obtains
d = x12 2(x2 a2 ),
= x12 2 x2 r 2 a2 .
(56)
(57)
2
1 2
2
0 = r 1 + cos(2 ) d x1 d + 2x1 (s3 z3 )
4
2
1 2
2
1 + cos(2) x1 + 2x1 s3 r z3 .
(60)
4
d
On the other hand, since s3 and z3 are real when the matrices S and A are real, Eqs. (20) and
(21) read in that case
s32 + z32 + 2 cos(2)s3 z3 = d ,
s32 + r 4 z32 + 2r 2 cos(2 )s3 z3 = .
(61)
Defining
f1 = 1 r 4 ,
(62)
f2 = cos(2) r cos(2 ),
(63)
f3 = r cos(2) r cos(2 ),
(64)
f4 = d ,
(65)
f5 = r d ,
(66)
f6 = r cos(2 )d cos(2) ,
(67)
4
2
(68)
We use as input the three chargedlepton masses, the three downtypequark masses, and
also r, cos(2 ), and cos(2). Eq. (68) allow us to compute s3 and z3 from that input. Inserting those values of s3 and z3 in Eq. (60), we obtain a quartic equation for x1 , which may be
analytically solved. The quantities s2 and a2 are then computed as
x12 r 2 d
,
+
2
2(1 r 2 )
d
.
a2 =
2(1 r 2 )
s2 =
(69)
(70)
Finally, z4 is computed from either Eq. (58) or Eq. (59), and z4 is obtained from Eq. (53). All the
invariants pertaining to the matrices S and A are thus analytically computed from the input.
One must, yet, take into account the fact that those invariants must satisfy several inequalities.
In the weak basis (5),
x1 = a + b + c,
(71)
x2 = ab + bc + ca,
(72)
s3 = abc.
(73)
The numbers a, b, and c are real, and they may be negative, see footnote 2. The quantity
x12 x22 + 18x1 x2 s3 4x23 4x13 s3 27s32
2
= (a b)(b c)(c a)
(74)
(75)
(76)
(77)
(78)
(79)
The conditions that , , and be nonnegative constitute a severe constraint on the inputted
values of the chargedfermion masses and of r, , .
After having computed the invariants, one may further input the three uptypequark masses
and therefrom derive the values of p 2 , q 2 , and cos(2 ). In practice, this involves solving a cubic equation, and thereafter imposing the constraints p 2 0, q 2 0, and  cos(2 ) 1. This
obviously translates into constraints on the inputted uptypequark masses.
In the way delineated in this section, one may analytically solve the case of real S and A
matrices, by inputting the chargedfermion masses and therefrom deriving S and A, without
having to have recourse to fits. In practice, however, doing things the other way roundtrying
to fit the chargedfermion masses numerically from some inputted values of S, A, and the other
parametersproves more effective. We turn to that procedure in the next section.
10
(80)
where
d2 =
mi (x) mi 2
,
mi
(81)
i=d,s,b
2 . The masses at the GUT scale are m m , whereas the m (x) are the
and analogously for ,u
i
i
i
masses calculated from Eqs. (3) as functions of the parameter set x = {S, A, p, q, r, , , } (see
Section 2 for the distinction between the real and the complex cases). The total 2 function
is the sum
2
2
2
total
= masses
+ CKM
,
(82)
with
2
CKM
=
i=12,13,23
2
.
(83)
We take the masses mi at the GUT scale, and their errors mi , from Table II of [21]; those masses
refer to the MSSM with tan = 10 and a GUT scale of 2 1016 GeV and have been obtained
through the renormalizationgroup evolution of the masses given in [22] at the Z 0 mass scale.
As for sines of the CKM angles, sin i sin i , we use Table 1 in [11]. We do not take into
account the CKM phase in our fitting procedure; this omission will be justified later.
In order to get a better understanding of our mass matrices, we perform separate minimizations
2 . We also test the real versus the complex case.
2
of masses
and of total
For the numerical multidimensional minimization of the 2 functions we employ the downhill simplex method [25]. Because the problem is highly nonlinear, we expect the existence of
many local minima.3 We start with randomly generated initial simplices. At the points where the
numerical algorithm stops, we iterate the procedure with random perturbations in order to find
a lower 2 . In this way we can be fairly certain about the distribution of the local minima and
about the position of the global minimum.
In the description of the fits, the concept of pull with respect to an observable O is useful.
The pull of O is defined as
pull(O) =
O(x) O
,
O
(84)
where the experimental value of the observable is O O, while O(x) is the theoretical prediction of O, given as a function of the parameter set x; x is the parameter set at a local minimum
3 The concept local minimum is not understood in a strict mathematical sense, rather it refers to a point where the
minimization algorithm successfully stops.
of 2 . Thus,
min 2 (x) 2 (x) =
2
pull(O) .
11
(85)
7. Results
We have performed all fits and tests of our scenario separately for real and complex coupling
matrices S and Asee Section 2. It turns out that there are no significant numerical differences
between the two cases. The extra two phases in the complex case are unable to significantly
improve our fits. Therefore, for simplicity in the following we confine ourselves to the real
case.
7.1. Fits of the masses alone
Firstly we omit the CKM angles and test whether, with the mass matrices (3), we are able to
fit the chargedfermion mass values at the SUSY GUT scale given in [21]. In Fig. 1 we show
2
2
for which min masses
the distribution, in the d2 u2 plane, of the local minima of masses
40. Though the density of points in that figure depends sensitively on the number of random
perturbations and on the number of restarts of the downhill simplex procedure, the overall picture
2
is located at u2 0, d2 23.3; the corresponding fit
is clear. The absolute minimum of masses
masses, and the pulls, are given in Table 1. For comparison, we also show in Table 1 the central
mass values of [21]. Looking at the pulls, we see that this mass fit fails only in the mass of the
2
Fig. 1. The distribution of local minima of masses
d2 + u2 in the d2 u2 plane (we force 2 to be always negligibly
2
small). The straight lines refer to constant values of masses
. This figure refers to the real case.
12
Table 1
Results of the fit for the real case without the CKM angles. The values of the masses in the second column, and the
corresponding errors m for the calculation of the pulls, have been taken from [21]. The third column gives our best fit
and the fourth column displays the corresponding pulls. The fifth and sixth columns refer to best fit in the region of small
d2 , i.e., the region in Fig. 1 with d2 1 and u2 25. All the masses are in units of MeV
m
me
m
m
md
ms
mb
mu
mc
mt
0.3585
75.67
1292.2
1.504
29.95
1063.6
0.7328
210.33
82 433
2
masses
= 23.3
2
masses
= 25.8
m(x)
pull(m)
m(x)
pull(m)
0.3585
75.67
1292.2
0.4112
29.54
1187.0
0.7249
212.47
78 466
6 103
0.3585
75.67
1292.2
1.430
29.35
1188.2
0.7321
214.66
8778
6 104
9 104
4 103
0.321
0.132
0.882
0.061
0.228
4.99
4 104
4 103
4.74
0.090
0.873
8 103
0.113
0.269
2
down quark; that particular pull is responsible for almost the complete masses
= 23.3. A glance
2
2
at Fig. 1 also reveals that there are local minima with u 25 and d 1; those minima, the best
of which is also displayed in Table 1, give rather good fits for all the downtypequark masses,
but fail severely in fitting the topquark mass: the fit value is about one order of magnitude, or
five , smaller than the experimental valuesee Table 1.
Thus, with our scenario we cannot even fit all the chargedfermion masses. However, as
stressed in the introduction, our scenario is extreme in that it allows only for the Yukawa couplings of one 10 120 scalar representation. If we allow for small perturbations of the mass
matrices, there are several ways out: there could be contributions from Yukawa couplings of one
126 [17], several 10 and/or 120 of scalars, radiative corrections, or nonrenormalizable terms.
2
can be considered acceptable, since it fails only
Consequently, the absolute minimum of masses
for md , which is small anyway. On the other hand, our philosophy of small perturbations forces
us to discard the local minimum where the fit value of mt is one order of magnitude too small.
13
2
Fig. 2. The distribution of local minima of total
(in the real case) in the d2 u2 plane. The straight lines refer to
2
2
2
constant d + u total .
Table 2
Results of the fit for the real case including the CKM angles. The best fit is described by columns three and four: this
point lies at d2 1, u2 25. Columns five and six refer to the best fit in the region of small u2
me
m
m
md
ms
mb
mu
mc
mt
sin 12
sin 23
sin 13
2 = 26.9
total
2 = 45.4
total
m(x)
pull(m)
m(x)
pull(m)
0.3585
75.67
1292.2
1.504
29.95
1063.6
0.7238
210.33
82433
0.3585
75.67
1292.2
1.563
28.24
1191.0
0.7243
215.35
8179
1 103
2 103
4 103
0.141
0.376
0.903
3 103
0.264
5.03
0.3585
75.67
1292.2
1.044
1.36
1225.4
0.7279
216.83
74145
7 103
4 103
3 103
1.993
6.29
1.145
0.030
0.342
0.56
sin
sin (x)
pull(sin )
sin (x)
pull(sin )
0.2243
0.0351
0.0032
0.2242
0.0348
0.0036
0.047
0.208
0.740
0.2243
0.0352
0.0034
2 104
0.093
0.318
14
Since we are unable to reproduce well all the quark masses, we cannot expect to obtain a
realistic CKM phase, and we have not included it in our fit.
In view of our philosophy, we have also tried a fit of the masses and of the CKM mixing angles
while allowing for artificially large errors in the lightfermion masses. Taking for instance mi =
2
5 MeV for i = e, d, u, we are able to achieve total
25.4. This is not really an improvement
when compared to the best fit in columns 3 and 4 of Table 2. However, the characteristics of this
fit are different from those of that best fit: the pulls of mb and of mt are approximately +1 and 1,
respectively, whereas the fit value of ms is 4.65 too low. Thus, the fit with drastically increased
errors in the lightfermion masses rather resembles the fit of columns 5 and 6 of Table 2.
7.3. A numerical test of b unification
In Section 4 we have traced b unification to some inequalities involving the chargedlepton
and the downtypequark masses; those inequalities are conditions on the masses necessary for
Eqs. (9), (10), (14), and (15) to have a solution. Neglecting the masses of the first generation, i.e.,
setting md me 0, those conditions are reformulated as [18]
1
m + ms
m + ms
mb
1+
.
m
m
m
(86)
mb
1.08.
m
(87)
15
We have performed a 2 analysis to check the inequalities (86). For this purpose, we consider
mi (x) mi 2 mb (x) mb 2
2
(x, mb )
+
.
d
(88)
mi
0.01mb
i=e,,,d,s
This 2 function is identical with d2 + 2 apart from the term corresponding to the bottomquark mass, wherein we leave mb free and assign to it a very small error bar of 1%. Then we
define a minimal 2 as a function of mb :
2
2
d
(mb ) min d
(x, mb ).
x
(89)
This function allows one to test the downtypequark and chargedlepton mass fits with respect
2 (m ) against m /m ; we have used the
to variations of mb [24]. In Fig. 3 we have plotted d
b
b
mean value m = 1292.2 MeV given in [21]. We see that exactly in the range of Eq. (87) the
2 (m ) is, for all practical purposes, zero. This confirms our analytic derivation of
minimum of d
b
b unification.
We can also test Eq. (46) against our numerics. We find that equation is reproduced fairly well
whenever the fit of mt is good.
8. Summary
In this paper we have investigated a SUSY SO(10) scenario in which the chargedfermion
masses are generated exclusively by the renormalizable Yukawa couplings of the fermions to
one representation 10 120 of scalars. We have studied the threegenerations case, confirming
the b unification which had previously been proved for two generations [18]. However, our
tests of this scenario against the chargedfermion masses and against the CKM mixing angles at
the GUT scale show that it is not satisfactory: the fit value of mt comes out much too low for
2 , we are able to obtain a good fit of m , at the price of m
the best fit; allowing for a larger total
t
s
turning out one order of magnitude too low and of md also being too small. We thus find that the
scenario investigated here is too restrictive: an additional mechanism for chargedfermion mass
generation is required.
Acknowledgements
We thank Thomas Schwetz for advice and discussions on the fitting procedure. The work of
L.L. has been supported by the Portuguese Fundao para a Cincia e a Tecnologia through the
project U777Plurianual.
References
[1] H. Fritzsch, P. Minkowski, Ann. Phys. 93 (1975) 193.
[2] P. Minkowski, Phys. Lett. B 67 (1977) 421;
T. Yanagida, in: O. Sawata, A. Sugamoto (Eds.), Proceedings of the Workshop on Unified Theory and Baryon
Number in the Universe, KEK report 7918, Tsukuba, Japan, 1979;
S.L. Glashow, in: J.L. Basdevant (Ed.), Quarks and Leptons, Proceedings of the Advanced Study Institute, Cargse,
Corsica, 1979, Plenum, New York, 1981;
M. GellMann, P. Ramond, R. Slansky, in: D.Z. Freedman, F. van Nieuwenhuizen (Eds.), Supergravity, NorthHolland, Amsterdam, 1979;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
16
[3] G. Senjanovic, in: J. Orloff, S. Lavignac, M. Cribier (Eds.), Proceedings of SEESAW25: International Conference
on the Seesaw Mechanism and the Neutrino Mass, Paris, France, 1011 June 2004, World Scientific, Singapore,
2005, hepph/0501244.
[4] R.N. Mohapatra, B. Sakita, Phys. Rev. D 21 (1980) 1062.
[5] R. Slansky, Phys. Rep. 79 (1981) 1.
[6] K.S. Babu, R.N. Mohapatra, Phys. Rev. Lett. 70 (1993) 2845, hepph/9209215.
[7] B. Bajc, G. Senjanovic, F. Vissani, Phys. Rev. Lett. 90 (2003) 051802, hepph/0210207;
B. Bajc, G. Senjanovic, F. Vissani, Phys. Rev. D 70 (2004) 093002, hepph/0402140.
[8] G. Lazarides, Q. Shafi, C. Wetterich, Nucl. Phys. B 181 (1981) 287;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. D 23 (1981) 165;
R.N. Mohapatra, P. Pal, Massive Neutrinos in Physics and Astrophysics, World Scientific, Singapore, 1991, p. 127.
[9] H.S. Goh, R.N. Mohapatra, S.P. Ng, Phys. Lett. B 570 (2003) 215, hepph/0303055;
H.S. Goh, R.N. Mohapatra, S.P. Ng, Phys. Rev. D 68 (2003) 115008, hepph/0308197.
[10] S. Bertolini, M. Frigerio, M. Malinsk, Phys. Rev. D 70 (2004) 095002, hepph/0406117.
[11] S. Bertolini, M. Malinsk, Phys. Rev. D 72 (2005) 055021, hepph/0504241.
[12] K.S. Babu, C. Macesanu, Phys. Rev. D 72 (2005) 115003, hepph/0505200.
[13] C.S. Aulakh, B. Bajc, A. Melfo, G. Senjanovic, F. Vissani, Phys. Lett. B 588 (2004) 196, hepph/0306242.
[14] B. Bajc, A. Melfo, G. Senjanovic, F. Vissani, Phys. Rev. D 70 (2004) 035007, hepph/0402122;
C.S. Aulakh, A. Girdar, Nucl. Phys. B 711 (2005) 275, hepph/0405074;
T. Fukuyama, A. Ilakovac, T. Kikuchi, S. Meljanac, N. Okada, J. Math. Phys. 46 (2005) 033505, hepph/0405300;
C.S. Aulakh, Phys. Rev. D 72 (2005) 051702, hepph/0501025.
[15] C.S. Aulakh, Expanded version of the plenary talks at the Workshop Series on Theoretical High Energy Physics,
IIT Roorkee, Uttaranchal, India, March 1620, 2005, and at the 8th European Meeting From the Planck Scale to the
Electroweak Scale (PLANCK05), ICTP, Trieste, Italy, 2328 May 2005, hepph/0506291;
B. Bajc, A. Melfo, G. Senjanovic, F. Vissani, hepph/0511352.
[16] C.S. Aulakh, S.K. Garg, hepph/0512224.
[17] C.S. Aulakh, hepph/0602132.
[18] B. Bajc, G. Senjanovic, Phys. Rev. Lett. 95 (2005) 261804, hepph/0507169;
B. Bajc, hepph/0602166.
[19] N. Oshimo, Phys. Rev. D 66 (2002) 0950, hepph/0206239;
N. Oshimo, Nucl. Phys. B 668 (2003) 258, hepph/0305166.
[20] K. Matsuda, Y. Koide, T. Fukuyama, Phys. Rev. D 64 (2001) 053015, hepph/0010026.
[21] C.R. Das, M.K. Parida, Eur. Phys. J. C 20 (2001) 121, hepph/0010004.
[22] H. Fusaoka, Y. Koide, Phys. Rev. D 57 (1998) 3986, hepph/9712201.
[23] R. Dermek, S. Raby, Phys. Lett. B 622 (2005) 327, hepph/0507045.
[24] T. Schwetz, private communication.
[25] J.A. Nelder, R. Mead, Comput. J. 7 (1965) 306;
W.H. Press, B.P. Flannery, S.A. Teukolsky, W.T. Vetterling, Numerical Recipes in C: The Art of Scientific Computing, Cambridge Univ. Press, Cambridge, 1992.
Abstract
The overlap hypercube fermion is a variant of a chirally symmetric lattice fermion, which is endowed
with a higher level of locality than the standard overlap fermion. We apply this formulation in quenched
QCD simulations with light quarks. In the pregime we evaluate the masses of light pseudoscalar and vector
mesons, as well as the pion decay constant and the renormalisation constant ZA . In the regime we present
results for the leading low energy constants of the chiral Lagrangian, and F . To this end, we perform
fits to predictions by chiral random matrix theory and by different versions of quenched chiral perturbation
theory, referring to distinct correlation functions. These results, along with an evaluation of the topological
susceptibility, are also compared to the outcome based on the standard overlap operator.
2006 Elsevier B.V. All rights reserved.
1. Overview
QCD at low energy cannot be handled by perturbation theory. Therefore it is often replaced
by chiral perturbation theory (PT) as an effective theory [1]. PT deals with fields for the
light mesons (occasionally also nucleons) which are the relevant degrees of freedom in that
regime. Here we present nonperturbative results for QCD itselfthe fundamental theoryat
low energy, and we also establish a connection to the most important parameters in PT. For
such systems, chiral symmetry plays a central rle, even though it is only realised approximately.
* Corresponding author.
18
In general, it is notoriously problematic to regularise quantum field theories with massless (or
light) fermions in a way, which keeps track of the (approximate) chiral symmetry. For instance,
in the framework of dimensional regularisation this issue is analysed carefully in Ref. [2]. However, in the framework of the lattice regularisation there was substantial progress in this respect
at the end of the last century (for a review, see Ref. [3]). At least for vector theories a satisfactory
solution was found, which enables the simulation of QCD close to the chiral limit. But since this
formulation is computationally demanding, production runs are limited to the quenched approximation up to now.
In this work, we present simulation results employing a specific type of chiral lattice fermions, denoted as the overlap hypercube fermion (overlapHF). In Section 2 we briefly review its
construction and discuss some properties, which are superior compared to the standard overlap
formulation.
Section 3 is devoted to the regime, where the pexpansion of PT [4] is applicable (pregime).
Here we present simulation results for bare quark masses in the range from 16.1 to 161 MeV.
We measure the masses of light pseudoscalar and vector mesons, and the quark mass according
to the PCAC relation. The latter fixes the axialcurrent renormalisation constant ZA , which has
a stable extrapolation to the chiral limit. We also present a matrix element evaluation of the pion
decay constant F , where, however, such an extrapolation is less stable.
In a fixed volume V (1.48 fm)3 (2.96 fm), we further decrease the bare quark mass to
mq 8 MeV, which takes us into the regime, i.e., the domain of the expansion [5] in PT.
The study of overlapHFs in the regime, and also the comparison to the standard overlap fermions under identical conditions, is the main issue of this work, which we present in Section 4. In
that regime, finite size effects and the topological sectors play an extraordinary rle.
Therefore, in Section 4.1 we first discuss the distribution of topological chargesdefined
by the fermion indexand the corresponding susceptibility. Then we address the low energy
constants (LECs) of the chiral Lagrangian, which parametrise the finite size effects. Hence they
can be extracted even from the regimei.e., from a relatively small volumewith their values
in infinite volume, which are relevant in physics. In Section 4.2 we compare our results for the
leading nonzero eigenvalue of the Dirac operator to the prediction by chiral random matrix
theory (RMT), which allows for a determination of the chiral condensate .
Finally we return to F , which we evaluate in the regime by means of two methods. In
Section 4.3 we fit the correlation function of the axialcurrents to a prediction by quenched
PT, making use of the previously obtained values for ZA and . At last we work directly
in the chiral limit (i.e., at zero quark mass) and consider the zeromode contributions to the
pseudoscalar correlation functions (Section 4.4). Again, a value for F emerges from a fit of
our data to a prediction by quenched PT. However, the two versions of quenched PT that we
apply to determine F differ by a subtlety in the counting rules for the quenched terms in the
expansion.
We summarise and discuss our results in Section 5. Part of them were included previously in
a PhD thesis [6] and in several proceeding contributions [7].
2. The overlap hypercube fermion in QCD
The lattice regularisation usually introduces a UV cutoff /a, where a is the lattice spacing.
However, a block variable renormalisation group transformation (RGT) generates a lattice action
on a coarser latticesay, with a spacing a = 2awhich leaves the system unchanged, so that
19
still the original cutoff matters [8]. Hence in this formulation the lattice artifacts are controlled
by 2/a , in contrast to the situation for a standard action. An infinite iteration of this blocking
procedure leads in principle (for suitable RGT parameters) to a perfect lattice action, which is
free of any cutoff artifacts.
In practice, for most systems perfect actions can only be constructed in some approximation,
such as a classical RGT step [9,10], and a truncation of the longrange couplings is needed.
Exceptions are, for instance, the quantum rotor [11], the GrossNeveu model at large N [12] and
the free fermion, where explicitly parametrised perfect actions are known [13,14]. Let us start
from free Wilsontype fermions and iterate the simple (blocking factor n) RGT
S [ , ]
e
= D D exp S[, ]
1
1
x .
x (d+1)/2
x x (d+1)/2
a
n
n
x
xx
(2.1)
xx
Here x (x ) are the sites of the original, fine (blocked, coarse) lattice, populated by the spinor
fields , ( , ), with the lattice action S (S ), and x x are the sites x in the nd block with
centre x (in d dimensions, and n = 2, 3, . . .). = 0 is a real, dimensionless RGT parameter.
After an infinite number of iterations, a perfect, free lattice action S [, ] emerges, with a
Dirac operator D consisting of a vector term plus a scalar term,
S [, ] = a d
(2.2)
x Dxy y , Dxy = (x y) + (x y)
x,y
(where we denote the spacing on the final lattice again by a). For a finite RGT parameter ,
the scalar term is nonzero. Then this action is local, since the couplings in and decay
exponentially at large distances x y, even at zero fermion mass (x and y are lattice sites)
[13]. This is satisfactory from the conceptual perspective, but in view of practical applications
the couplings need to be truncated to a short range. A useful truncation scheme works as follows
[15]: one first optimises the RGT so that the locality of the operator D is maximal, hence the
truncation is minimally harmful. Then one constructs the perfect fermion action on a small,
periodic lattice of size 34 (where we now specify d = 4). The couplings obtained in this way are
finally used also in larger volumes. This is a truncation to a hypercube fermion (HF) operator
DHF with
supp (x y) ,
supp (x y) x, y x y  a, .
(2.3)
a
D5 D
(2.4)
for a simple RGT of the type (2.1), with a transformation parameter 1. The GWR is obvious
for free perfect fermions, but it also persists under gauge interaction. Classically perfect Dirac
operators solve the GWR as well [18], and the free HF operator DHF does so to very good
approximation, as the spectrum shows [19].
20
In our HF formulation of lattice QCD [6,15,20],1 the gauge field is treated rather ad hoc,
since it is the fermionic part which is most delicate. Hence we use the standard Wilson gauge
action, and the Dirac operator is gauged as follows: if the lattice spinors x and y are coupled
by DHF,xy i.e., if they are located in the same lattice unit hypercubewe connect them by the
shortest lattice paths and multiply the compact link variables on these paths. The normalised sum
of these link products is denoted as a hyperlink between x and y (a fully explicit description
is given in Refs. [6,20]).
In addition, we
replace the simple links in the above procedure by fat links, U (x)
(1 )U (x) + 6 [staples], where is a parameter to be optimised.
This HF is nearly rotation invariant, but its mass is subject to a significant additive renormalisation [15]. Criticality can be approximated again by amplifying all the link variables by a
suitable factor u > 1. At last, in the vector term only they are also multiplied by an extra factor
v 1; this hardly alters the critical point, but it improves chirality, i.e., it moves DHF closer to a
GWR solution. The fat link, e.g., with in the range 0.30.6, helps further in this respect [6,22].
An optimisation of that kind was performed previously for quenched QCD at = 6 [23]. Here
we present the more difficult construction at = 5.85, which corresponds to a lattice spacing of
a 0.123 fm.2 As a compromise between various optimisation criteria, we chose the following
parameters [6]
u = 1.28,
v = 0.96,
= 0.52,
= 1,
(2.5)
and the HF couplings are still those identified for the massless, truncated perfect, free
fermion [15]. The physical part of the spectrum of this DHF for a typical configuration at
= 5.85 is shown in Fig. 1. We also mark some of the eigenvalues with maximal real part;
they reveal a striking difference from Wilson fermion spectra, which extend to much larger real
parts. Moreover, we include the low eigenvalues for an alternative set of parameters, which moves
the HF close to criticality. In that case we recognise a good approximation to the circle in the
complex plane with centre and radius = 1, which characterises an exact solution to the GWR
(2.4). However, for the practical purposes in this paper, we prefer the parameters in Eq. (2.5)
(the advantages are a faster convergence and a better locality if the HF operator is inserted in the
overlap formula, which we are going to discuss next).
Of course, due to the truncation and the imperfect gauging, the scaling and chirality of this
HF are somewhat distorted. At least chirality can be corrected again by inserting DHF into the
overlap formula [25]
Dov =
(2.6)
1 + A/ A A , A := D0 .
a
a
This formula yields a solution to the GWR (2.4), provided that D0 is a 5 Hermitian lattice
Dirac operator, D0 = 5 D0 5 . Another condition is that a value for must be available, which
separatesfor typical configurations at the gauge coupling under considerationthe (nearly)
real part of the spectrum of D0 in a sensible way into small (physical) eigenvalues and large
eigenvalues (at the cutoff scale). Thus the physical eigenvalues are separated from the doublers,
and we obtain the correct number of flavours. 5 Hermiticity holds for example for the Wilson
1 For recent applications in thermodynamic systems, see Ref. [21].
2 Throughout this work, we refer to the Sommer scale [24] for physical units. We do not keep track of possible errors
in that scale.
21
Fig. 1. The low lying part of the spectrum of DovHF , plus eigenvalues with maximal real part, for a typical configuration
at = 5.85 on a 84 lattice. We show one set of parameters which provides a good approximations to criticality and to
the spectral GinspargWilson circle. However, for practical purposes we prefer the parameterisation (2.5), which leads
to the other set of eigenvalues shown in this figure.
operator DW and for DHF , and for instance at = 6 good values for can easily be found in
both cases [23,26].
D0 = DW is the kernel of the standard overlap operator formulated by H. Neuberger [25];
we denote it by DN . In this case, DW changes drastically as it is inserted into the overlap formula (2.6). In contrast, DHF is already an approximate GinspargWilson operator, hence the
transition to the corresponding overlap operator, DHF DovHF , is only a modest modification,
DovHF DHF ,
(2.7)
particularly in view of the eigenvalues with large real parts. In both cases, a lattice modified but
exact chirality is guaranteed [27] through the GWR. The correct axial anomaly is reproduced
in all topological sectors for the standard overlap fermion [28], as well as the overlapHF [29].
But other important properties are strongly improved for DovHF compared to DN , due to the
construction of DHF and relation (2.7).3
Let us first consider the condition number of the Hermitian operator A A, i.e., the argument
of the square root in the overlap formula (2.6). In our simulations, we project out the lowest
30 modes of A Awhich are treated separatelyand we approximate the remaining part by a
Chebyshev polynomial to an absolute accuracy of 1012 . The polynomial degree which is needed
for a fixed precision is proportional to the square root of the condition number
c31 :=
max
,
31
(2.8)
3 The concept of improving properties of D beyond chirality by choosing a more favourable kernel than D was
ov
W
suggested in Ref. [19] and meanwhile implemented in several variants [30,22,31,10,32].
22
Table 1
The characteristic indicators for the kernel condition number and for the locality of the overlap operators DN (standard) and DovHF (described in the text),
on a 123 24 lattice
DN
DovHF
c11
c21
c31
5.85
5.85
5.85
4266(194)
1723(46)
1149(15)
179(10)
73(2)
49.2(8)
loc
loc
loc
5.85
5.7
5.6
0.63(1)
0.56(2)
0.73(1)
0.76(5)
0.53(3)
The values for c31 , given in Table 1, show that the required polynomial degree for DovHF is a
factor 5 lower. The computational effort is roughly proportional to this degree (the computation of the lowest modes of A A and their separate treatment are minor issues in terms of the
computing time). On the other hand, the use of the HF kernel is computationally more expensive
than DW by about a factor 15, so that an overhead by a factor 3 remains. We hope for this factor
to be more than compensated by the virtues of DovHF , in particular by an improved scaling due to
the perfect action background of DHF and relation (2.7). That property is very well confirmed for
the free fermion [19] and for the 2flavour Schwinger model (with quenched configurations, but
measurement data reweighted by the fermion determinant) [30]. But the corresponding scaling
behaviour in QCD is not explored yet.
An important aspect that we do explore here is the locality. For this purpose, we put a unit
source at the origin, x = x,0 , and measure the expectation value of the function [26]
fmax (r1 ) := max Dov,yx (U )x 
y
1 = r1 .
y
(2.9)
Following the usual convention we take here the distance in the taxi driver metrics
y
1 :=
y . A comparison, still at = 5.85, is given in Fig. 2 (on top) and Table 1; we see a clear
gain for DovHF , i.e., a decay of f (r1 ) exp(loc r1 ) (beyond short r1 ) where loc is larger for
DovHF than for DN . If we refer to the Euclidean distance r = y, the maximum is taken over
different sets, but the resulting decay is still exponential. In the lower plot of Fig. 2 we see that
the decay for DovHF is not only faster, but it also follows in a smoother way the exponential shape
in r, which hints at a good approximation to rotation symmetry.
Locality is a vital requirement for a safe continuum limit in lattice field theory [26]. This poses
a limitation on the overlap operator (2.6), since for decreasing , at some point an exponential
decay is not visible any longer.4
Fig. 3 shows the situation at = 5.7 (i.e., a 0.17 fm), where both operators are still local
(with optimal parameters and u for DN respectively DovHF ), but again the locality of DovHF
is superior. As we push further to = 5.6, we do not observe an exponential decay for DN
anymore, but DovHF is still manifestly local (even without adjusting the parameters in Eq. (2.5),
except for u). Therefore, the overlapHF provides chiral fermions on coarser lattices.
4 The question, in which sense locality still persists in a conceptual sense is discussed in Ref. [33]. In any case, if no
exponential decay is clearly visible, the operator should certainly not be applied in simulations.
23
Fig. 2. The locality of different overlap fermions, measured by the maximal impact of a unit source x over a taxi driver
distance r1 (on top), and over a Euclidean distance r (below), at = 5.85. We see that the overlapHF is clearly more
local. The upper plot also shows that the slope hardly changes if we proceed from a 104 to a 123 24 lattice. (The
increasing slope beyond r1 = 18 is a consequence of the anisotropy of the 123 24 lattice.) The lower plot illustrates
additionally that the overlapHF decay follows closely an exponential in the Euclidean distance, which indicates a good
approximation to rotation symmetry.
24
Fig. 3. The locality of different overlap fermions, measured by the maximal impact of a unit source x over a taxi driver
distance r1 , on a 123 24 lattice. For the overlapHF we see a superior locality at = 5.7. Its locality persist up to
= 5.6, where no exponential decay is visible for the standard overlap formulation.
The effective theory does have its predictive power, but the values of the LECs cannot be determined within PT. To this end, one has to return to QCD as the fundamental theory, and it is
our main goal in this work to investigate how far the LECs can be obtained from lattice QCD
simulations.
In PT, one usually considers a finite spatial box, which has a volume L3 , and one expects
the meson momenta p to be small, so that
p
2
4F .
L
(3.2)
.
LF
QCD QCD
(3.3)
In this section we present our simulation results in the pregime. We apply the overlapHF
described in Section 2, at = 5.85 on a lattice of size 123 24, which corresponds to a physical
volume of V (1.48 fm)3 2.96 fm. We evaluated 100 propagators for each of the bare quark
masses
amq = 0.01, 0.02, 0.04, 0.06, 0.08 and 0.1
(in physical units: 16.1161 MeV), using a Multiple Mass Solver. We will see that the smallest
mass in this set is at the edge of the pregimeeven smaller quark masses will be considered in
Section 4.
25
Fig. 4. The pion mass evaluated from overlap HFs in the pregime in three different ways, as described in Section 3.
(3.4)
which leaves the largest real overlap Dirac eigenvalue invariant. mq represents the bare mass for
the quark flavours u and d.
We first evaluate the pion mass in three different ways:
m,P P is obtained from the decay of the pseudoscalar correlation function P (x)P (0), with
P (x) = x 5 x . This is the most obvious method, but it is not the best one in this case, as
we will see.
m,AA is extracted from the decay of the axialvector correlation function A4 (x)A4 (0),
with A4 (x) = x 5 4 x .
m,P P SS is obtained from the decay of the difference P (x)P (0) S(x)S(0), where
S(x) = x x is the scalar density. This subtraction is useful especially at small mq , where
configurations with zeromodes ought to be strongly suppressed by the fermion determinant.
In our quenched study, this suppression does not occur as it should, but the above subtraction
in the observable eliminates the zeromode contributions, which are mostly unphysical.
We present the results in Fig. 4 and Table 2 (the latter collects all the results of this section).
The pion masses follow to a good approximation the expected behaviour m2 mq . Deviations
show up at the smallest masses, where we observe the hierarchy
m,P P > m,AA > m,P P SS ,
(3.5)
in agreement with Ref. [10]. This shows that the scalar density subtraction is in fact profitable,
since it suppresses the distortion of the linear behaviour down to the lightest pion mass in Fig. 4,
m,P P SS (amq = 0.01) = (289 32) MeV.
(3.6)
26
Table 2
Our results in the pregime for the pion mass, the meson mass, the PCAC quark mass, the renormalisation constant
ZA and the pion decay constant F , as described in the text and plotted in the figures of Section 3 (Figs. 49). These
results are obtained from 100 propagators at each of the bare quark masses mq . All the dimensional numbers in this table
are given in lattice units at a 0.123 fm = (1610 MeV)1
amq
0.1
0.08
0.06
0.04
0.02
0.01
am,P P
am,AA
am,P P SS
am
amPCAC
ZA
aF,P P
aF,P P SS
0.491(5)
0.483(4)
0.513(7)
0.745(7)
0.089(3)
1.13(4)
0.117(2)
0.120(2)
0.438(5)
0.432(5)
0.453(8)
0.717(8)
0.070(3)
1.14(5)
0.114(3)
0.114(3)
0.380(5)
0.378(5)
0.381(9)
0.694(10)
0.052(3)
1.15(6)
0.110(3)
0.109(3)
0.314(4)
0.317(6)
0.313(11)
0.677(15)
0.035(3)
1.16(9)
0.105(4)
0.104(5)
0.248(5)
0.240(7)
0.221(15)
0.674(32)
0.017(2)
1.16(14)
0.103(4)
0.010(10)
0.210(11)
0.189(9)
0.181(20)
0.672(55)
0.0087(13)
1.14(18)
0.098(4)
0.082(12)
That mass corresponds to a ratio L/ 2, hence around this point we are indeed leaving the
pregime. Based on the moderate quark masses in Fig. 4, we find an impressively small intercept
in the chiral extrapolation,
m,P P SS (mq 0) = (2 24) MeV.
(3.7)
On the other hand, at our larger mq values the hierarchy changes due to the interference of the
scalar correlator.
Due to quenching, one expects at small quark masses a logarithmic behaviour of the form
am2
= C1 + C2 ln amq + C3 amq
mq
(C1 , C2 , C3 : constants).
(3.8)
Corresponding results are given for instance in Refs. [34,35]. Fig. 5 shows the fits of our
data to Eq. (3.8). m,AA is best compatible with this rule (this property was also hinted at in
Ref. [35]). At least the deviation for m,P P SS (although inside the errors) could be expected,
since the scalar subtraction alleviates the quenching artifacts, which give rise to the logarithmic
corrections according to formula (3.8).
In Fig. 6 (and Table 2) we consider the vector meson mass m based on the same 100 configurations. A chiral extrapolation leads to
m = (1017 40) MeV.
(3.9)
This agrees well with a study using the standard overlap operator on the same lattice [36]. It is
quite large, however, not only in view of phenomenology, but also compared to other quenched
results in the literature, which were summarised in Ref. [37].5
On the other hand, if we insert our measured ratio mpseudoscalar /mvector at mq = 0.01 into the
phenomenologically motivated interpolation formula presented in Ref. [38], then we arrive at a
significantly lower value of m 789 MeV.
5 Note that the difference from the data by other groupsobtained with various (mostly nonchiral) lattice
formulationsis approximately constant in m over the range shown here, hence finite size effects can hardly be blamed
for it.
27
Fig. 5. The pion masses of Fig. 4, fitted against the form (3.8), which is expected for the logarithmic quenching artifacts
in the absence of an additive mass renormalisation.
Fig. 6. The mass of the meson against the squared pion mass m2,P P SS .
Fig. 7 shows the quark mass obtained from the PCAC relation,
x (4 A4 (x))P (0)
,
mPCAC =
x P (x)P (0)
(3.10)
which follows closely the bare quark mass mq (we use a symmetric nearestneighbour difference
for 4 ). As a consequence, the axialcurrent renormalisation constant
mq
ZA =
(3.11)
mPCAC
28
Fig. 7. The PCAC quark mass, evaluated for DovHF according to Eq. (3.10), from stable plateaux in the time direction.
We find mPCAC values close to mq , in contrast to the results with the standard overlap operator.
Fig. 8. The axialcurrent renormalisation constant ZA , determined from Eq. (3.11), which is found to be close to 1 for
the overlapHF. ZA is plotted against m2,P P SS .
(3.12)
which is in contrast to the unpleasantly large values found for the standard overlap operator: e.g.,
at the same = 5.85 and = 1.6 (a preferred value in that case) it amounts to ZA 1.45 [36,39,
40], and (somewhat surprisingly) at = 6, = 1.4 it even rises to ZA 1.55 [40]. According
to Ref. [32], the fat link could be especially helpful for the property ZA 1, which is favourable
for a connection to perturbation theory.
29
Fig. 9. The pion decay constant based on a matrix element evaluation in the pregimegiven by Eq. (3.13)using the
overlapHF.
As a last observable in the pregime, we measured the pion decay constant by means of the
(indirect) relation
F =
2mq
,
0P

m2
(3.13)
based on P (x)P (0), or based on P (x)P (0) S(x)S(0) (in Eq. (3.13) this affects both, the denominator and the pion state).6 The results are given in Fig. 9 and Table 2. In particular the value
at amq = 0.01 (the lightest quark mass in this plot) is significantly lower for the case of the
scalar subtraction. Hence this pushes the result towards the chirally extrapolated phenomenological value of 86 MeV [42].
However, a chiral extrapolation based on these data for F would be riskyapparently we
are too far from the chiral limit for this purpose. An extrapolated value of F,P P would come out
clearly too large, as it is also the case for DN [36]. But in particular the instability of F,P P SS at
amq = 0.04, 0.02, 0.01 calls for a clarification by yet smaller quark masses. We did consider still
much smaller values of mq in the same volume. As the results for the pion masses suggest, we
are thus leaving the pregime. For the tiny masses amq 0.005 we enter in fact the regime,
where observables like F have to be evaluated in a completely different manner. This is the
subject of Section 4.
4. Applications in the regime
The regime of QCD is characterised by a relatively small volume, i.e., the correlation length
exceeds the linear box size L. On the other hand, the box still has to be large compared to the
scale given by 4F , as in Eq. (3.2), so that the partition function is saturated by the lowest
states onlyhigher states (above about 1 GeV) do not contribute significantly. This amounts to
6 A recent alternative attempt to evaluate LECs from simulations in the pregime was presented in Ref. [41].
30
the condition
1
1
(4.1)
>L
.
m
4F
In such a box, the pexpansion of PT fails, in particular due to the dominant rle of the zeromodes. However, the latter can be treated separately by means of collective variables, and the
higher modesalong with the pion massare then captured by the expansion [5]. One now
counts the ratios
p2
1
m
2
(4.2)
(for Nf flavours),
(4.3)
is worked out in Ref. [43]. The corresponding nonlinear O(N) model (with a symmetry
breaking to O(N 1)) in a small box was studied with the FaddeevPopov method to two loops
[44], and with the Polyakov functional measure to three loops [45].
This setting cannot be considered a physical situation. Nevertheless there is a strong motivation for its numerical study: the point is that the finite size effects are parametrised by the LECs
of the effective chiral Lagrangian as they occur in infinite volume, hence the physical values of
the LECs can (in principle) be evaluated even in an unphysically small box.
We recall that the effective chiral Lagrangian includes all terms which are compatible with the
symmetries, ordered according to suitable low energy counting rules, in this case the counting
rules (4.2) of the expansion. We further presented in Section 3 the LECs as coefficients of
these terms, for instance F and in the leading order. Their determination from QCDthe
fundamental theoryat low energy is a challenge for lattice simulations. As a test, such a lattice
determination in the regime has been performed successfully for the O(4) model (which
describes chiral symmetry breaking with two flavours) some years ago [46], but in QCD this
method [47] had to await the advent of chiral lattice fermions. Unfortunately, the quenched results
for the LECs are affected by (mostly logarithmic) finite size effects [48], so that the final results
by this method still have to wait for the feasibility of QCD simulations with dynamical, chiral
quarks.
A peculiarity of the regime is that the topological sector plays an essential rle [49]: if one
measures observables in a specific sector, the expectation values often depend significantly on
the (absolute value of the) topological charge in this sector. In particular for the evaluation of the
LECs, a numerical measurement inside a specific sector and a confrontation with the analytical
predictions in this sector is in principle sufficient. This requires the collection of a large number
of configurations in a specific sector. The topology conserving gauge actions [50] are designed
to facilitate this task. However, here we stay with the Wilson gauge action, which allows us
to investigate also the statistical distribution of the topological charges, which we are going to
address next.
4.1. The distribution of topological charges
A priori, it is not obvious how to introduce topological sectors in the set of lattice gauge
configurations. However, if one deals with GinspargWilson fermions, a sound definition is given
31
Fig. 10. Index histories for DovHF (see Section 2) and for DN (at = 1.6) for the same set of configurations.
by adapting the AtiyahSinger Theorem from the continuum and defining the topological charge
by the fermionic index [18],
!
topological charge = := n+ n ,
(4.4)
where n is the number of zeromodes with positive/negative chirality. We remark that these
numbers are unambiguously determined once a GinspargWilson Dirac operator is fixed (and
that in practice only chirality positive or chirality negative zeromodes occur in one configuration7 ). However, for a given gauge configuration, the index for different GinspargWilson
operators does not need to agree. Albeit the level of agreement should be high for smooth configurations, i.e., it shouldand it does8 increase for rising values of .
In our study, still at = 5.85 on a 123 24 lattice, we compared the charges for the overlapHF operator described in Section 2 and for the standard overlap operator DN at = 1.6. As an
example, the histories of about 200 indices for the same configurations are compared in Fig. 10.
Of course, these two types of indices are considerably correlated, but only 41% really coincide.
We obtained a mean deviation of
ovHF N  = 0.80(2),
(4.5)
and we observed over more than 1000 configurations a maximal index difference of
ovHF N  = 5. Still, the similarity is of course much closer than the accidental agreement
7 This property even holds in cases where cooling deforms the configuration into a form, where a semiclassical
picture suggests the presence of topological objects with opposite chiralities [51]. Regarding the fermion index, a cancellation happens for instance for free GinspargWilson fermions, but in a realistic gauge background such an unstable
constellation is very unlikely.
8 For instance, we observed at = 6.15 on a 164 lattice that the index of D is very stable as rises from 1.3 to 1.7;
N
this changes less than 2% of the indices.
32
Fig. 11. The histogram of the overlapHF indices (on the left) and for the standard overlap indices. In both cases, 1013
indices are included.
for independent indices, since they follow essentially the expected Gaussian distributions, with a
width 3.3, see Fig. 11 and Table 3. This width fixes the topological susceptibility
top =
1 2
,
V
(4.6)
which is of importance to explain the heavy mass of the meson [52]. A discussion of that
point, as well as a measurement of top based on DN indices on L4 lattices with a continuum
extrapolation, is given in Ref. [53]. A compilation of earlier lattice results for top with various
methods is given in the first work quoted in Ref. [10].
In Fig. 12 and Table 3 we present our results with DovHF and DN on the lattice used so far,
plus a result for DN at = 6 in the same physical volume (lattice size 163 32). We also mark
the continuum extrapolation according to Ref. [53], which is fully consistent with our results.
This value for top is compatible with the WittenVeneziano scenario that much of the mass
is generated by a U (1) anomaly.
In principle, the charge histogram could give insight into the possibility of a spontaneously
broken parity symmetry in QCD, which is not fully ruled out [54]. This question was also studied
in Ref. [55] with a different definition of topological charges on the lattice. Here we observe in
all cases that the number of neutral configurations is about half of the corresponding number with
 = 1 (see Table 3). Based on this observation, it cannot be decided if the charge distribution
(at small ) favours a (discretised) Gaussian, or a double peak structure (or something else).
Hence the fate of parity symmetry remains open.
4.2. Determination of the chiral condensate
Chiral RMT conjectures predictions for the low lying eigenvalues, ordered as n , n = 1, 2, . . .
(excluding possible zero eigenvalues) of the Dirac operator in the regime (for a review, see
Ref. [56]). More precisely, the conjectured densities are functions of the dimensionless variables
V n , where is the chiral condensate in the effective Lagrangian (3.1). Here we focus on
the variable z := V 1,P , where 1,P emerges from the leading nonzero eigenvalue 1 if the
33
Fig. 12. The topological susceptibility measured by indices of DovHF and of DN , in a volume V = (1.48 fm)3 2.96 fm,
with two different lattice spacings a. Our data agree well with the continuum extrapolation reported in Ref. [53].
Table 3
Our results for the topological susceptibility and for the chiral condensate in the regime for a fixed physical volume
V = (1.48 fm)3 2.96 fm. We consider two types of overlap operators (DovHF as described in Section 2, and the standard
overlap operator DN at = 1.6), and two lattice spacings. We first give the total statistics and the resulting topological
susceptibility (see Section 4.1) in a dimensionless form, with r0 = 0.5 fm (according to the Sommer scale [24]).
Below we give separately the statistics in the sectors  = 0, 1 and 2. We extract the chiral condensate from the
density of the lowest Dirac eigenvalue in the neutral charge sector, which is most reliable since it involves the smallest
values of z, see Section 4.2 and Figs. 13, 14. As an alternative we considered the mean values of the lowest nonzero
Dirac eigenvalues 1 in the sectors  = 05. For the value of in the last line all these results for 1  match the
RMT predictions, see Fig. 15
Dirac operator
DovHF
DN
DN
a
lattice size
5.85
0.123 fm
123 24
5.85
0.123 fm
123 24
6
0.093 fm
163 32
total # of confs.
2
top r04
1013
10.81 0.47
0.071(3)
1013
11.49 0.51
0.076(3)
506
10.49 0.66
0.069(4)
# of confs. with = 0
118
124
59
# of confs. with  = 1
# of confs. with  = 2
1/3 from sector = 0
238
210
298(4) MeV
237
192
301(4) MeV
115
95
279(7) MeV
a1 =0
a1 =1
a1 =2
a1 =3
0.0069(4)
0.0130(4)
0.0184(5)
0.0247(6)
0.0067(4)
0.0136(4)
0.0193(5)
0.0242(8)
0.0059(4)
0.0093(3)
0.0130(5)
0.0155(6)
a1 =4
a1 =5
1/3 from 1 =0...5
0.0293(9)
0.0338(12)
0.0312(10)
0.0360(13)
290(6) MeV
0.0215(9)
0.0246(17)
34
Fig. 13. The cumulative density of the lowest Dirac eigenvalue 1,P of the overlapHF operator, in the topological sectors
 = 0, 1 and 2. We compare the chiral RMT predictions to our data for z = V 1,P with 1/3 = 298 MeVthe
optimal value in the neutral sector ( = 0). This value works well up to z 3 for all topological sectors.
spectral circle of the overlap operator is mapped stereographically on the imaginary axis, 1,P =
1 /(1 a1 /2).9
These RMT predictions depend on , the absolute value of the topological charge. Here we
make use of the explicit formulae [57] for the density of the first nonzero (rescaled) eigenvalues
()
in the sectors , which we denote by 1 (z). For the lowest eigenvalues, the particular density
(0)
1 was first confirmed by staggered fermion simulations (results are summarised in Ref. [56]),
but the charged sectors yielded the very same density, in contradiction to RMT. The distinction
between the topological sectors was first observed to hold for DN to a good precision [58], if the
linear box size exceeds a lower limit of about L 1.1 fm (of course, the exact limit depends on
()
the criterion).10 Once the predicted density 1 is well reproduced, we can read off the value
of , which is the only free fitting parameter for all topological sectors. It is most instructive to
plot the cumulative densities [60], which we show in Figs. 13 and 14. We compare here the predictions to the eigenvalues 1,P , which we measured in various topological sectors. The statistics
involved in each case is included in Table 3. This table also displays the values obtained in
the sector = 0, which we consider most reliable, since it deals with the lowest eigenvalues
respectively energies. As a theoretical bound, one often refers to the Thouless energy F2 /( V ),
below which these predictions should hold. In our case, it translates into zThouless 1, but the
eigenvalue distributions follow the chiral RMT behaviour up to larger z values. Clearly, in this
range it is the neutral sector ( = 0) which contributes in a dominant way, but Fig. 14 shows
that in the case of DN (for both values of ) the charged sectors  = 1 and 2 alone would
favour a different value. This ambiguity also occurs for smeared staggered fermions [59]. In
9 Alternatively, one could simply consider  , which are eigenvalues of D, but for the small eigenvalues that we
1
5
35
Fig. 14. The cumulative density of the lowest Dirac eigenvalue 1,P of the standard overlap operator DN at = 5.85
(on top) and = 6 (below), in the topological sectors  = 0, 1 and 2. We compare the chiral RMT predictions to our
data for z = V 1,P with 1/3 = 301 MeV (on top), and with 1/3 = 279 MeV (below)the optimal values in the
neutral sector ( = 0). Considering also  = 1, 2 would decrease (increase) at = 5.85 ( = 6) which is the trend
towards the result with the method of Fig. 15.
the case of the DovHF , however, a unique works well for all the three sectors  = 0, 1, 2, up
to about z 3, as we see from Fig. 13.
As an alternative approach to test the agreement of our data with the chiral RMT, and to extract
a value for , we now focus on the mean values of the leading nonzero Dirac eigenvalues 1
in all the charge sectors up to  = 5. In physical units, the results 1,P agree remarkably well
for the different overlap operators and lattice spacingssee Fig. 15although this consideration
extends beyond very low energy. Each single result for 1,P  can then be matched to the RMT
value for a specific choice of . Amazingly, all these 18 results are in agreement with the RMT
36
Fig. 15. The mean values of the first nonzero Dirac eigenvalue in the charge sectors  = 05, in physical unit. All the
measured results agree with chiral RMT if we choose 1/3 = 290(6) MeV.
if we fix
3
= 290(6) MeV ,
(4.7)
as Fig. 15 also shows. This value is between the results obtained from the eigenvalue densities at
= 0 alone, and we recognise from Figs. 13 and 14 that this is the trend if we take the charged
sectors into account.
A renormalisation procedure for obtained in this way is discussed in Ref. [61]. However,
we will only use this quenched lattice result as a fitting input in Section 4.3, so here we stay with
the bare condensate for our fixed lattice parameters.
We add that in the one flavour case values for around (270 MeV)3 have recently been
obtained based on a large Nc expansion [62] and on numerical data [63].
4.3. Evaluation of F based on the axialvector current correlator
As we mentioned before, QCD simulations with chiral quarks can only be performed in the
quenched approximation for the time being. In order to relate simulation results to the effective
low energy theory, we therefore refer to quenched PT. In that framework, mesonic correlation
functions were evaluated to the first order in Refs. [64,65]. It turned out that the vector current
correlation function vanishes; this property actually extends to all orders [66]. The scalar and
pseudoscalar correlators involve already in the first order additional, quenching specific LECs,
which obstruct the access to the physical LECs in the Lagrangian (3.1). Therefore we first focus
on the axialvector correlator, which only depends on and F in the first order (as in the
dynamical case). In particular we are going to compare our data to the quenched PT prediction
in a volume V = L3 T [65],
2
F
2
ZA A4 (t)A4 (0) = 2
+ 2mq  (zq )T h1 ( ) ,
T
1 2
1
t
h1 ( ) =
+
, = ,
2
6
T
(zq ) = zq I (zq )K (zq ) + I+1 (zq )K1 (zq ) +
,
zq
where
A4 (t) = a 3
(t > 0)
37
(4.8)
(4.9)
x
before.
12 In Ref. [67] we gave a value around 130 MeV, but the analysis did not handle the renormalisation constant Z very
A
carefullybeing more precise in this aspect reduces the value to about 120 MeV.
38
Table 4
Our results in the regime for the pion decay constant F , based on the axialcurrent correlation function. The results
are obtained at = 5.85 on a 123 24 lattice. We give our statistics of the propagators in the different topological sectors
at various bare quark masses in the regime. (Of course, different configurations where used for DovHF and for DN .)
The results for F were determined from fits to the quenched PT formula (4.8) in the range ta [11, 13], see Figs. 16
and 17
Dirac operator
DovHF
DN
amq = 0.001
# of propagators at  = 1
# of propagators at  = 2
F
50
50
(110 8) MeV
50
50
(109 11) MeV
amq = 0.003
# of propagators at  = 1
# of propagators at  = 2
F
50
50
(113 7) MeV
50
50
(110 11) MeV
amq = 0.005
# of propagators at = 0
# of propagators at  = 1
# of propagators at  = 2
F
50
50
(115 6) MeV
100
100
100
(111 4) MeV
Here we present again results at = 5.85 on a 123 24 lattice, where also the latter condition
admits  = 2, cf. Table 3. We evaluated for both, DovHF and DN , the axialvector correlators
at amq = 0.001, 0.003 and 0.005, which turns out to be safely in the regime. Our propagator
statistics at these quark masses is given in Table 4. We then fitted the data to Eq. (4.8) by using
the chirally extrapolated factors ZA (1.17 for DovHF and 1.45 for DN ), along with the values
that we obtained from the microscopic Dirac spectra (see Section 4.2 and Table 3). For each of
the overlap operators we performed at each of the quark masses a global fit over the topological
sectors that we considered, which is shown in Figs. 16 and 17. In particular our results for DovHF
reveal for the first time a quite clear distinction between the sectors  = 1 and  = 2this
property could not be observed for DN up to now. For DN at amq = 0.005 we also include the
neutral sector; as expected it has by far larger error bars than the charged sectors, but it is helpful
nevertheless to reduce the error on F in the global fit.
The emerging values for the pion decay constant are consistent,
F = (110 6) MeV (using DovHF ),
F = (109 4) MeV (using DN ).
(4.10)
39
Fig. 16. Lattice data with DovHF vs. predictions by quenched PT for the axialcurrent correlation functions in the
regime, measured separately in the topological sectors  = 1 and 2. The global fit at each mass corresponds to the
values of F given in Table 4.
40
Fig. 17. Lattice data with DN vs. predictions by quenched PT for the axialcurrent correlation functions in the regime,
measured separately in the topological sectors  = 0, 1 and 2. The global fit at each mass corresponds to the values of
F given in Table 4.
41
Fig. 18. Lattice data with DovHF vs. predictions by quenched PT for the axialcurrent correlation functions, as in
Fig. 16, but here the data are reweighted with the first nonzero Dirac eigenvalue, i.e., with the factor (4.11). This
reweightingwhich is most powerful at minimal mq improves the agreement with the theoretical prediction at relatively large t T /2. In this case, the fit leads to F = (112 7) MeV.
m2
0
0 0 + 0 02
2Nc
2Nc
(4.12)
(2)
to LqPT , which brings in 0 and m0 as another two quenching specific LECs, in addition to K.
The inclusion of 0 supplements the quenching effects; in the dynamical case it decouples form
the NambuGoldstone field.
It is somewhat ambiguous how to count these additional terms in
the quenched expansion.
Ref. [72] assumes the action terms with the coefficients 0 and K Nc to be of O(1), whereas
the one with m0 is in O(). In particular the last assumption is a bit unusual; for instance, it
disagrees with the framework referred to in Section 4.3. However, it is an acceptable possibility,
which simplifies this approach since it removes the auxiliary mass term from the dominant order.
If one further defines the dimensionless parameter
= 0
4Nc2 KF
,
(4.13)
42
For Nf valence quark flavours, we now consider the correlation function of the pseudoscalar
density P (x), which can be decomposed into a connected plus a disconnected part,
V 2 P (x)P (y) = Nf P1 (x, y) Nf2 P2 (x, y),
P1 (x, y) = Tr i5 (D + mq )1 (x, y) i5 (D + mq )1 (y, x) ,
P2 (x, y) = Tr i5 (D + mq )1 (x, x) i5 (D + mq )1 (y, y) .
(4.14)
Then one performs a spectral decomposition of the propagators and obtains the residuum in terms
of the zeromodes,
(1)
(2)
lim (mq V )2 P (x)P (0) = Nf C
(x) + Nf2 C
(x),
mq 0
(1)
(x) = vj (x)vk (x) vk (0)vj (0)  ,
connected: C
(2)
disconnected: C (x) = vj (x)vj (x) vk (0)vk (0)  .
(4.15)
The vectors vj denote the (exact) zeromodes of the GinspargWilson operator at mass zero,
(i)
DGW vj = 0. In the terms for C
the zeromodes are summed over.
Next we consider the spatial integral d 3 x P (x)P (0). Now the above procedure for the cor(i)
relation function leads to functions C (t), i = 1, 2, which are given explicitly in Ref. [72]. In
principle, these functions could be measured and fitted to the predictions in order to determine
F and . In practice, however, it is much better to consider instead just the leading term in the
expansion at t = T /2,
V d (i)
T
(i)
(4.16)
(t)
= D
s + O s 3 , s = t , i = 1, 2.
C

2
2
L dt
t=T /2
(i)
The slopes D tend to be stable over a variety of fitting ranges s [smax , smax ],
smax = a, 2a, 3a, . . . . To be explicit, the slope functions [72] in a volume V = L3 T take
the form [6]
2
1
(1)
D
=
 +
2
2Nc F2 V
(F L)
1
1 T 2
1 7
+
(4.17)
+ 2 2 2 2 +
,
2
24 3
2 F2 V
2
1
(2)
D =
1
+

2Nc F2 V
(F L)2
2
T
1 13
1
2 2
,
+ 
(4.18)
2
24 3
F2 V
where in our case
1 = 0.1314565,
1 =
1
1
= 0.083291.
2
12
sinh (T p/2)
p=0
1 is a shape coefficient, which we computed for our anisotropic volume according to the prescription in Ref. [44].
43
Table 5
Our results in the regime for the pion decay constant F along with the quenching specific LEC based on the
zeromode contributions to the pseudoscalar correlation function, see Section 4.4. The joint statistics in the sectors
 = 1 and 2, given in Table 3, contributes. We give the results at fitting range smax = a, which is most adequate in view
of Eq. (4.16)
Dirac operator
DovHF
DN
DN
lattice size
F
5.85
123 24
(80 14) MeV
17 10
5.85
123 24
(74 11) MeV
19 8
6
163 32
(75 24) MeV
21 15
Fig. 19. The results for F based on a global fit of our data to a quenched PT prediction for the zeromode contributions
to the pseudoscalar correlations function. Here and in Fig. 20 we show the results of a two parameter fit over the ranges
s [T /2 smax , T /2 + smax ].
We evaluated the LECs F and from fits to the linear term in Eq. (4.16). We used all the
zeromodes that we identified in the topological sectors  = 1 and 2the statistics is given in
(i)
Table 3. For 2 (which enters the expressions for D through WittenVeneziano relations) we
inserted the result that we measured in each case, which is also given in Table 3. For each of our
lattice sizes and each type of overlap operator we performed a global fit over both topological
sectors involved, in a fitting range smax . The emerging optimal values for F and are shown
in Figs. 19 and 20, and the values at smax /a = 1 are given in Table 5. We see that the results for
different lattice spacings and overlap Dirac operators are in good agreement, and we obtain the
most stable plateau for DovHF .
The value that we now obtain for F is below the one of Section 4.4, which used a different
observable and a different counting rule for the quenched terms. In fact, the result of this section is close to the phenomenological value (we repeat that the latter amounts to 86 MeV if
44
Fig. 20. The results for the quenching specific LEC , based on a global fit of our data to a quenched PT prediction for
the zeromode contributions to the pseudoscalar correlations function. Here and in Fig. 19 we show the results of a two
parameter fit over the ranges s [T /2 smax , T /2 + smax ].
one extrapolates to the chiral limit [42]). This result, as well as the negative value for , are also
somewhat below the values reported in Ref. [72] based on the same method. Some differences
are that Ref. [72] always used DN (with various values of ), cubic volumes, a continuum extrapolated value for 2 and partial fits were performed. We suspect that the anisotropic shape
of our volumes, T = 2L, could be the main source of the deviation from those results [73].
5. Conclusions
We have constructed an overlap hypercube Dirac operator DovHF , which is especially suitable
at a lattice spacing of a 0.123 fm. It has a strongly improved locality compared to the standard
overlap operator DN . This operator defines chiral fermions on coarser lattices than DN .
We performed quenched simulations with DovHF and with DN in a volume V (1.48 fm)3
(2.96 fm) at = 5.85 and at = 6.
In the pregime we applied DovHF and measured the meson masses m and m , the PCAC
quark mass mPCAC and the pion decay constant F at bare quark masses ranging from 16.1 MeV
to 161 MeV. The results for m and m are similar to the values found previously with DN on
the same lattice, which confirms their validity. On the other hand, mPCAC turned out to be much
closer to mq than in the standard overlap formulation. This implies an axialcurrent renormalisation constant close to 1, ZA = 1.17(2), which is favourable for the connection to perturbation
theory. Regarding F , it turned out that the data obtained in the pregime can hardly be extrapolated to the chiral limit.
45
We considered a large number of topological charges defined by the fermion indices of DovHF
or of DN (which coincide in part), and we found histograms which approximate well a Gaussian.
The resulting topological susceptibility is in good agreement with the literature.
In the regime we determined a value for the chiral condensate from the distribution of the
lowest eigenvalues. For both, DovHF and DN we obtained values around (300 MeV)3 .
We evaluated F in the regime in two ways, from the axialcurrent correlation and from the
zeromode contributions to the correlation of the pseudoscalar density. These two methods handle
the counting of the quenched terms differently, and they yield different values for F . The
axialcurrent method leads to F 110 MeV, which is consistent with various other quenched
results in the literature. The zeromode method (which might be more sensitive to the anisotropy
of the volume) leads to a lower F around 84 MeV, in the vicinity of the phenomenological value.
The final result of Ref. [69]using again a different method, based on the I = 1/2 rule, still
in the regimeis in between. We add that recently further methods were proposed to evaluate
F in the regime, involving s = 1 transitions [74] and a chemical potential [75].
From the current results, we conclude that the methods applied here work in the sense that
they do have the potential to evaluate at least the leading LECs from lattice simulations in the
regime. The quenched data match the analytical predictions qualitatively (if the volume is
not too small) andin the setting we consideredthey lead to results in the magnitude of the
LECs in Nature. However, the quenched results are ambiguous: different methods yield different
values.
For precise values and a detailed comparison to phenomenology, simulations with dynamical quarks will be needed. In particular the regime requires then dynamical GinspargWilson
fermions. For instance, quenched reweighting already leads to a distribution of the microscopic Dirac eigenvalues and the topological charges as it is expected for one dynamical quark
flavour [71]. In view of truly dynamical QCD simulations, first tests show that it is hard to arrange
for topological transitions [76], but fortunately the regime investigations can work even in a
fixed topological sector. Therefore, and also in view of the lattice size, the regime is promising,
if one is able to handle sufficiently small quark masses in a Hybrid Monte Carlo simulation, and
if ergodicity inside a fixed topological sector is achieved.
Acknowledgements
We are indebted to M. Papinutto and C. Urbach for numerical tools. We also thank A. Ali
Khan, S. Drr, H. Fukaya, P. Hasenfratz, S. Hashimoto, E. Laermann, M. Laine, K.I. Nagai,
K. Ogawa, L. Scorzato, A. Shindler, H. Stben, P. Watson and U. Wenger for useful comments.
This work was supported by the Deutsche Forschungsgemeinschaft through SFB/TR903. The
computations were performed on the IBM p690 clusters of the Norddeutscher Verbund fr
Hoch und Hchstleistungsrechnen (HLRN) and at NIC, Forschungszentrum Jlich.
References
[1] S. Weinberg, Physica A 96 (1979) 327;
J. Gasser, H. Leutwyler, Ann. Phys. (N.Y.) 158 (1984) 142.
[2] F. Jegerlehner, Eur. Phys. J. C 18 (2001) 673.
[3] S. Chandrasekharan, U.J. Wiese, Prog. Part. Nucl. Phys. 53 (2004) 373.
[4] J. Gasser, H. Leutwyler, Phys. Lett. B 184 (1987) 83.
[5] J. Gasser, H. Leutwyler, Phys. Lett. B 188 (1987) 477;
H. Neuberger, Nucl. Phys. B 300 (1988) 180.
46
[6] S. Shcheredin, Simulations of lattice fermions with chiral symmetry in quantum chromodynamics, PhD thesis
Berlin, 2004, heplat/0502001.
[7] W. Bietenholz, S. Shcheredin, Rom. J. Phys. 50 (2005) 249, heplat/0502010;
W. Bietenholz, S. Shcheredin, PoS (LAT2005) 138, heplat/0508016;
W. Bietenholz, S. Shcheredin, Nucl. Phys. B (Proc. Suppl.) 153 (2006) 17;
S. Shcheredin, W. Bietenholz, PoS (LAT2005) 134, heplat/0508034.
[8] K.G. Wilson, J.B. Kogut, Phys. Rep. 12 (1974) 75.
[9] P. Hasenfratz, F. Niedermayer, Nucl. Phys. B 414 (1994) 785;
T. DeGrand, A. Hasenfratz, P. Hasenfratz, F. Niedermayer, Nucl. Phys. B 454 (1995) 587;
M. Blatter, R. Burkhalter, P. Hasenfratz, F. Niedermayer, Phys. Rev. D 53 (1996) 923;
R. Burkhalter, Phys. Rev. D 54 (1996) 4121.
[10] P. Hasenfratz, S. Hauswirth, T. Jrg, F. Niedermayer, K. Holland, Nucl. Phys. B 643 (2002) 280;
BGR Collaboration, Nucl. Phys. B 677 (2004) 3.
[11] W. Bietenholz, R. Brower, S. Chandrasekharan, U.J. Wiese, Phys. Lett. B 407 (1997) 283.
[12] W. Bietenholz, E. Focht, U.J. Wiese, Nucl. Phys. B 436 (1995) 385.
[13] W. Bietenholz, U.J. Wiese, Nucl. Phys. B 464 (1996) 319.
[14] W. Bietenholz, R. Brower, S. Chandrasekharan, U.J. Wiese, Nucl. Phys. B 495 (1997) 285.
[15] W. Bietenholz, R. Brower, S. Chandrasekharan, U.J. Wiese, Nucl. Phys. B (Proc. Suppl.) 53 (1997) 921.
[16] W. Bietenholz, U.J. Wiese, Phys. Lett. B 426 (1998) 114;
W. Bietenholz, Nucl. Phys. A 642 (1998) 275.
[17] P.H. Ginsparg, K.G. Wilson, Phys. Rev. D 25 (1982) 2649.
[18] P. Hasenfratz, V. Laliena, F. Niedermayer, Phys. Lett. B 427 (1998) 125;
P. Hasenfratz, Nucl. Phys. B 525 (1998) 401.
[19] W. Bietenholz, Eur. Phys. J. C 6 (1999) 537.
[20] SESAM Collaboration, Comput. Phys. Commun. 119 (1998) 1.
[21] S. Wissel, E. Laermann, S. Shcheredin, S. Datta, F. Karsch, PoS (LAT2005) 164, heplat/0510031.
[22] W. Bietenholz, in: Proceedings of the International Workshop on NonPerturbative Methods and Lattice QCD,
Guangzhou, China, 2000, p. 3, heplat/0007017.
[23] W. Bietenholz, Nucl. Phys. B 644 (2002) 223.
[24] M. Guagnelli, R. Sommer, H. Wittig, Nucl. Phys. B 535 (1998) 389.
[25] H. Neuberger, Phys. Lett. B 417 (1998) 141;
H. Neuberger, Phys. Lett. B 427 (1998) 353.
[26] P. Hernndez, K. Jansen, M. Lscher, Nucl. Phys. B 552 (1999) 363.
[27] M. Lscher, Phys. Lett. B 428 (1998) 342.
[28] D.H. Adams, Ann. Phys. 296 (2002) 131.
[29] D.H. Adams, W. Bietenholz, Eur. Phys. J. C 34 (2004) 245.
[30] W. Bietenholz, I. Hip, Nucl. Phys. B 570 (2000) 423.
[31] T. DeGrand, Phys. Rev. D 63 (2001) 034503;
W. Kamleh, D.H. Adams, D.B. Leinweber, A.G. Williams, Phys. Rev. D 66 (2002) 014501.
[32] S. Drr, C. Hoelbling, U. Wenger, JHEP 0509 (2005) 030;
S. Drr, C. Hoelbling, Phys. Rev. D 72 (2005) 071501.
[33] M. Golterman, Y. Shamir, Phys. Rev. D 68 (2003) 074501.
[34] P. Hasenfratz, K.J. Juge, F. Niedermayer, JHEP 0412 (2004) 030.
[35] M. Grtler, T. Streuer, G. Schierholz, D. Galletly, R. Horsley, P. Rakow, PoS (LAT2005) 077, heplat/0512027.
[36] LF Collaboration, JHEP 0412 (2004) 044.
[37] A. Ali Khan, heplat/0507031.
[38] E. Laermann, C. DeTar, O. Kaczmarek, F. Karsch, Nucl. Phys. B (Proc. Suppl.) 73 (1999) 447.
[39] H. Fukaya, S. Hashimoto, K. Ogawa, Prog. Theor. Phys. 114 (2005) 451.
[40] R. Babich, et al., JHEP 0601 (2006) 086.
[41] C. Gattringer, P. Huber, C. Lang, Phys. Rev. D 72 (2005) 094510.
[42] G. Colangelo, S. Drr, Eur. Phys. J. C 33 (2004) 543.
[43] F.C. Hansen, Nucl. Phys. B 345 (1990) 685.
[44] P. Hasenfratz, H. Leutwyler, Nucl. Phys. B 343 (1990) 241.
[45] W. Bietenholz, Helv. Phys. Acta 66 (1993) 633.
[46] A. Hasenfratz, et al., Z. Phys. C 46 (1990) 257;
A. Hasenfratz, et al., Nucl. Phys. B 356 (1991) 332;
I. Dimitrovic, P. Hasenfratz, J. Nager, F. Niedermayer, Nucl. Phys. B 350 (1991) 893.
47
[47] L. Giusti, C. Hoelbling, M. Lscher, H. Wittig, Comput. Phys. Commun. 153 (2003) 31;
LF Collaboration, physics/0309072.
[48] P.H. Damgaard, Phys. Lett. B 608 (2001) 162.
[49] H. Leutwyler, A. Smilga, Phys. Rev. D 46 (1992) 5607.
[50] M. Lscher, Nucl. Phys. B 549 (1999) 295;
H. Fukaya, T. Onogi, Phys. Rev. D 68 (2003) 074503;
H. Fukaya, S. Hashimoto, T. Hirohashi, K. Ogawa, T. Onogi, Phys. Rev. D 73 (2006) 014503;
W. Bietenholz, K. Jansen, K.I. Nagai, S. Necco, L. Scorzato, S. Shcheredin, JHEP 0603 (2006) 017.
[51] K.I. Nagai, W. Bietenholz, T. Chiarappa, K. Jansen, S. Shcheredin, Nucl. Phys. B (Proc. Suppl.) 129 (2004) 516.
[52] E. Witten, Nucl. Phys. B 156 (1979) 269;
G. Veneziano, Nucl. Phys. B 159 (1979) 213.
[53] L. Del Debbio, C. Pica, JHEP 0402 (2004) 003;
L. Del Debbio, L. Giusti, C. Pica, Phys. Rev. Lett. 94 (2005) 032003.
[54] V. Azcoiti, A. Galante, Phys. Rev. Lett. 83 (1999) 1518.
[55] B. Alles, M. DElia, A. Di Giacomo, Phys. Rev. D 71 (2005) 034503.
[56] J.J.M. Verbaarschot, T. Wettig, Annu. Rev. Nucl. Part. Sci. 50 (2000) 343.
[57] S.M. Nishigaki, P.H. Damgaard, T. Wettig, Phys. Rev. D 58 (1998) 087704;
P.H. Damgaard, S.M. Nishigaki, Nucl. Phys. B 518 (1998) 495;
P.H. Damgaard, S.M. Nishigaki, Phys. Rev. D 63 (2001) 045012.
[58] W. Bietenholz, K. Jansen, S. Shcheredin, JHEP 0307 (2003) 033;
L. Giusti, M. Lscher, P. Weisz, H. Witting, JHEP 0311 (2003) 023;
QCDSFUKQCD Collaboration, Nucl. Phys. B (Proc. Suppl.) 129130 (2004) 456.
[59] E. Follana, A. Hart, C.T.H. Davies, Phys. Rev. Lett. 93 (2004) 241601;
S. Drr, C. Hoelbling, U. Wenger, Phys. Rev. D 70 (2004) 094502;
K.Y. Wong, R.M. Woloshyn, Phys. Rev. D 71 (2005) 094508.
[60] W.H. Press, S.A. Teukolsky, W.T. Vetterling, B.P. Flannery, Numerical Recipes, Cambridge Univ. Press, Cambridge,
1992.
[61] J. Wennekers, H. Wittig, JHEP 0509 (2005) 059.
[62] A. Armoni, M. Shifman, G. Veneziano, Phys. Lett. B 579 (2004) 384.
[63] T. DeGrand, R. Hoffmann, Z. Liu, S. Schaefer, hepth/0605147.
[64] P.H. Damgaard, M.C. Diamantini, P. Hernndez, K. Jansen, Nucl. Phys. B 629 (2002) 445.
[65] P.H. Damgaard, P. Hernndez, K. Jansen, M. Laine, L. Lellouch, Nucl. Phys. B 656 (2003) 226.
[66] P.H. Damgaard, P. Hernndez, K. Jansen, M. Laine, L. Lellouch, Nucl. Phys. B (Proc. Suppl.) 129 (2004) 754.
[67] W. Bietenholz, T. Chiarappa, K. Jansen, K.I. Nagai, S. Shcheredin, JHEP 0402 (2004) 023.
[68] P. Hernndez, K. Jansen, L. Lellouch, Phys. Lett. B 469 (1999) 198.
[69] L. Giusti, P. Hernndez, M. Laine, P. Weisz, H. Wittig, JHEP 0404 (2004) 013.
[70] H. Fukaya, S. Hashimoto, K. Ogawa, Prog. Theor. Phys. 114 (2005) 451.
[71] K. Ogawa, S. Hashimoto, Prog. Theor. Phys. 114 (2005) 609.
[72] L. Giusti, P. Hernndez, M. Laine, P. Weisz, H. Wittig, JHEP 0401 (2004) 003.
[73] M. Laine, private communication.
[74] L. Giusti, P. Hernndez, M. Laine, C. Pena, J. Wennekers, H. Wittig, PoS (LAT2005) 344, heplat/0510033.
[75] P.H. Damgaard, U.M. Heller, K. Splittorff, B. Svetitsky, Phys. Rev. D 72 (2005) 091501;
P.H. Damgaard, U.M. Heller, K. Splittorff, B. Svetitsky, D. Toublan, Phys. Rev. D 73 (2006) 074023;
J. Bloch, T. Wettig, heplat/0604020.
[76] Z. Fodor, S.D. Katz, K.K. Szabo, JHEP 0408 (2004) 003;
N. Cundy, S. Krieg, A. Frommer, Th. Lippert, K. Schilling, Nucl. Phys. B (Proc. Suppl.) 140 (2005) 841;
T. DeGrand, S. Schaefer, Phys. Rev. D 71 (2005) 034507;
N. Cundy, Nucl. Phys. B (Proc. Suppl.) 153 (2006) 54.
Abstract
We present a new scheme for extracting approximate values in the improved perturbation method,
which is a sort of resummation technique capable of evaluating a series outside the radius of convergence.
We employ the distribution profile of the series that is weighted by nthorder derivatives with respect to
the artificially introduced parameters. By those weightings the distribution becomes more sensitive to the
plateau structure in which the consistency condition of the method is satisfied. The scheme works effectively even in such cases that the system involves many parameters. We also propose that this scheme has
to be applied to each observable separately and be analyzed comprehensively.
We apply this scheme to the analysis of the IIB matrix model by the improved perturbation method
obtained up to eighth order of perturbation in the former works. We consider here the possibility of spontaneous breakdown of Lorentz symmetry, and evaluate the free energy and the anisotropy of spacetime
extent. In the present analysis, we find an SO(10)symmetric vacuum besides the SO(4) and SO(7)symmetric vacua that have been observed. It is also found that there are two distinct SO(4)symmetric
vacua that have almost the same value of free energy but the extent of spacetime is different. From the
approximate values of free energy, we conclude that the SO(4)symmetric vacua are most preferred among
those three types of vacua.
2006 Elsevier B.V. All rights reserved.
PACS: 02.30.Mv; 11.25.w; 11.25.Yb; 11.30.Cp; 11.30.Qc
* Corresponding author.
49
1. Introduction
String theory is the unique theory that contains massless spintwo particles, i.e., gravitons [1],
and thus it is considered to provide an unified microscopic description of the universe including
gravitational interactions. For this reason the string theory has been subjected to intensive studies. However, it is recognized that the perturbative string theory fails to single out our universe as
the unique vacuum of the theory [2]. Therefore we are forced to pursue nonperturbative formulations. The IIB matrix model (also called the IKKT matrix model) is proposed as a constructive
formulation of the superstring theory [3,4].
A significant feature of the IIB matrix model is that the spacetime itself is expressed by the
eigenvalue distribution of 10 bosonic matrices, and thus it is treated as a dynamical variable of
the model. The origin of our fourdimensional spacetime can be argued in the context of the
IIB matrix model as a spontaneous breakdown of Lorentz symmetry. In this regard, we have to
understand nonperturbative properties of the model.
The mechanism of the spontaneous breakdown of Lorentz symmetry in reduced matrix models
has been examined in various approaches. It has been recognized from those works that the
fermionic part of the action plays a crucial role [511].
Unveiling dynamical aspects of the model is, in general, quite a difficult problem. The Monte
Carlo method is a powerful tool for exploring such nonperturbative properties of a model. However, it is not applicable (or at least quite difficult to apply) to the IIB matrix model due to
complex phase of the action derived from the fermionic part.1 The improved perturbation method
(also called the Gaussian expansion method) is an alternative approach. It is considered as a sort
of variational method [1318]. It has been successfully applied to various models [1922], and
applications to matrix models were done in Refs. [2325].
The application to the IIB matrix model was first achieved in Ref. [26] in which various
patterns of symmetry breaking that preserve SO(d) subgroup of the original tendimensional
rotational symmetry (ansatz) were examined, and as a conclusion, fourdimensional universe is
the most preferred among them based on the comparison of free energy. The Gaussian expansion method was reformulated as an improvement of perturbative series expansion in Ref. [27].
The improved Taylor expansion (ITE), as it is referred, opened a way toward more general applications that incorporate quadratic and other types of interactions. The ITE prescription was
employed for the IIB matrix model up to fifth order of perturbation in Ref. [27]. It has been
proceeded to even higher orders and extended ansatz [2830]. The mechanism of the spontaneous symmetry breakdown is further examined in a simplified model via the Gaussian expansion
method in Ref. [31].
In the present paper, we investigate the nonperturbative solutions of the IIB matrix model by
this technique (which we call the improved perturbation method here). We focus on the possibility that the original SO(10) symmetry of the IIB matrix model may be spontaneously broken
to result in our universe which spreads in four directions and has SO(4) rotational symmetry.2
In such cases it is important to see that the method is applicable to the models that exhibit phase
transitions. In the former work [22], we applied this method to the Ising model and found that the
improved perturbation method extracts the information of the ordered phase from an expansion
about the vacuum in the disordered phase. It is also observed that an unstable vacuum is identi1 A novel technique called the factorization method is proposed to resolve the complex action problem in the Monte
Carlo simulations [11,12].
2 We perform the Wick rotation to the IIB matrix model and discuss in Euclidean spacetime.
50
fied even though it has larger value of free energy than that of the stable vacuum. We expect that
the improved perturbation method reveals different patterns of symmetry breaking that may be
developed as unstable or metastable vacua.
The improved perturbation method is considered as a sort of resummation of perturbative
series by introducing the artificial parameters into the model. The approximate value of the series
is obtained by evaluating it in the region of the artificial parameter space that realizes the principle
of minimal sensitivity [16]. We call such a region as plateau, in which the dependence on those
parameters would vanish effectively and the exact value would be reproduced. It works as a
consistency condition for the parameters.
The concept of plateau is rather obscure, and there is not yet any proper treatment of mathematical rigor. We need a practical scheme for extracting the approximate value of the series. In
addition, it should be free from the ambiguity due to the subjectivity of recognition as much as
possible. In the former works, mainly two approaches have been propounded. One is to take the
values at the extrema of the improved series with respect to the artificial parameters as approximate values [26]. The other is to take the values which corresponds to the mode of distribution
of the evaluated values around the accumulation of extrema. The latter is called the histogram
technique [32,33].
In this article we propose a new prescription which enables us to extract a good approximation from the improved series. It is achieved by incorporating the appropriate weightings in the
histograms that are given by the inverse of derivatives. This method will be applicable to various
models which have many parameters.
It has also been customary to evaluate the various observables based on the information of
plateau of the free energy. On extracting the approximate value of any observable, the information
of minimal sensitivity of only improved free energy is taken into account. To be more precise,
one takes the value of improved perturbative series of an observable at extrema of improved free
energy with respect to the artificial parameters as approximate values of that observable. We
insist that the improved perturbation method has to be applied independently for each observable
of interest.
We apply the new scheme to the analysis of the IIB matrix model by means of the improved
perturbation method. It is based on the improved series that is obtained in former works up to
eighth order of perturbation for SO(d)preserving configurations as ansatz (d = 4 and 7) [29].
This paper is organized as follows. In Section 2, we provide a description of the resummation technique called the improved perturbation method, including a review of the method. In
Section 3, we propose a new prescription which extracts the approximate value from improved
perturbative series. In Section 4, we apply the improved perturbation to the IIB matrix model and
survey the nonperturbative solutions with various patterns of symmetry breaking. Section 5 is
devoted to the conclusion. In Appendix A, we present a detail of analysis performed in Section 4.
2. Improved perturbation
It is generally believed that the perturbative expansion (also including N1 expansion and expansion) is asymptotic, and diverges beyond some finite orders. Nevertheless, what we can
evaluate for most of the theories are only those series expansions, from which we have to draw
physical information of the theory. Therefore, we need a method to estimate the exact value from
the series with the parameter (coupling constants, and so on) outside the original convergence
radius. For this purpose we introduce here a method which has been successfully applied to the
IIB matrix model and other general models.
51
2.1. Prescription
We implicitly use as coupling constant and m as parameters of the model collectively such as
masses. Let us assume that the observables of a theory would be exactly described by a function
F (, m). Perturbation theory provides an expansion of F as a power series of about = 0,
with nth coefficient denoted as fn (m):
F (, m) =
n fn (m).
(1)
n=0
In the actual cases, we only have finite portion of the series up to order N ,
F N (, m) =
N
n fn (m).
(2)
n=0
The question is whether we can presume the exact value F (, m) at the given parameters and
m from the series F N above. In many cases the convergence radius for is zero, so we cannot
expect that F N gives a reliable approximation.
Now we consider a modification of the series along the following prescriptions. First we
perform a shift of parameters:
g p ,
m m0 + g q (m m0 ),
(3)
where we have introduced g as a formal expansion parameter, and m0 as a set of artificial parameters. p and q are taken arbitrarily. We deform the series by the substitution (3), and then we
reorganize the series in terms of g, drop the O(g M+1 ) terms, and finally set g to 1. We obtain the
improved perturbative series F N as
F N (, m) F N (, m; m0 ) = F N g p , m0 + g q (m m0 ) g M ,g1 .
(4)
Here, we adopt a notation g M ,g1 to represent the operation that we disregard the O(g M+1 )
terms and then put g to 1.
It should be noted that by simply setting g = 1, the modification itself becomes trivial and
the series would be independent of the parameters m0 . However, due to dropping the O(g M+1 )
terms, the deformed series does depend on m0 .
To turn the argument around, we adopt here the principle of minimal sensitivity that the exact
value F will be reproduced when the improved series depends least on the artificial parameters
m0 . It provides a sort of consistency condition on m0 ; by tuning the parameters to the solution of
the condition, we will have a good approximation of F .
In the above we have introduced the arbitrary parameters p, q, M. If the region emerges in
the parameter space of m0 that realizes minimal sensitivity, the approximate value must be independent of the choice of parameters p, q, M. However, with the limited orders of perturbation,
the signal of minimal sensitivity is often weak in actual cases. So we have to find the optimal
values of parameters p, q, M in order to make the signal clear. We have to keep in mind that large
values of p and q turn to throw away the information of higher order terms of perturbative series,
and taking larger value of M than that of N is meaningless. Therefore, we set the parameters as
p = q = 1 and M = N throughout this paper.
52
2.2. Example
For example, we apply the improved perturbation method to a simple 1dimensional function,
1
,
1 + m
and try to estimate F (, m) at = 1 and m = 3/2.
The Taylor expansion about = 0 up to order N gives the following series:
F (, m) =
F N (, m) =
N
(1)n n mn .
(5)
(6)
n=0
This series at N has finite convergence radius m < 1. The situation is depicted in
Fig. 1(a) where F N of various N are shown as a function of m when = 1.
Now we deform the series along the prescription in the previous section, to obtain the improved series F N :
F N (, m; m0 ) =
N
n
(1)n (g)n m0 + g(m m0 ) g N ,g1 .
(7)
n=0
Here, g N ,g1 denotes disregard for the O(g N+1 ) terms followed by setting g = 1.
The improved series behaves as shown in Fig. 1(b), when the artificial parameter m0 is taken
to be 1.1. In this case, the functions F N ( = 1, m; m0 = 1.1) become close to the exact value
in the region m 1.5, and we will have a good approximation of the original function F at
m = 3/2.
To find the optimum of parameter m0 , we consider F N (, m; m0 ) as a function of m0 with m
and other original parameters fixed to the specified values, (e.g. = 1 and m = 3/2 in this case).
It can be seen that in some region of parameter m0 = 0.7 1.1, F N stays stable and gives good
approximation to the exact value, F = 0.4 (Fig. 2). We call such a stable region as plateau,
where the principle of minimal sensitivity is realized and the exact value will be reproduced.
(a)
(b)
1
and (b) improved series at m0 = 1.1.
Fig. 1. (a) Taylor series of F (, m) = 1+m
53
N
n fn (m).
(8)
n=0
m m0 + g(m m0 ),
(9)
F N is deformed as
F N (, m)
N
(g)n fn m0 + g(m m0 )
n=0
k=0
gk
min(N,k)
n=0
1
(m m0 )kn fn(kn) (m0 ).
(k n)!
(10)
(k)
Here, each coefficient fn is expanded in Taylor series about m = m0 , with fn as kth derivative
of fn (m) with respect to m. The series is then reorganized in powers of g. Then we drop the
O(g N +1 ) terms, and finally set g to 1. As a result, we will have the improved series F N :
F N =
N
n=0
N
n
k=0
1
(m m0 )k fn(k) (m0 ).
k!
(11)
54
It turns out that the improved perturbation method replaces each coefficient of the original
series, fn (m), by its Taylor expansion about a shifted point m0 ,
fn (m) fn (m; m0 ) =
Nn
k=0
1
(m m0 )k fn(k) (m0 ).
k!
(12)
It should also be noted that the expansion in Eq. (12) is taken only up to (N n)th order. This
feature works to suppress the large fluctuations coming from higher order coefficients, which
have caused divergent behavior of the original series. Thus we could expect better estimates by
this improved perturbation method.
The important point is that the reorganized series should be truncated at finite order N . If we
kept all the terms and if we could switch the order of two summations in Eq. (10), the procedure
should become trivial, only to obtain the original series.
2.3.2. Convergence
The convergence property of the improved series with respect to is, however, not affected
significantly. The ratio of (n + 1)th coefficient against nth one,
Nn1 1
k (k)
fn+1
k=0
k! (m m0 ) fn+1 (m0 )
=
,
(13)
(k)
Nn 1
fn
(m m0 )k fn (m0 )
k=0 k!
has no particular structures such as singularities, at least in obvious manner. There, the parameters
m0 are determined according to the following argument; minimal sensitivity condition will be
realized ideally when pth or lower derivatives of F N with respect to the parameters m0 become
zero:
dp N
0=
p F (, m; m0 )
dm0
p1
N
p 1
1
(Nn+pq1)
n
=
(14)
(m0 ).
(m m0 )Nnq fn
()q
(N n q)!
q
n=0
q=0
1
1
,
+
1 + m 1 + m + (m)3
(15)
and generate a Taylor series about = 0 up to order N . The radius of convergence of the original
series is (m) 0.7. Then, we apply the improved perturbation method to this series. As shown
(a)
55
(b)
Fig. 3. (a) Large fluctuation of F as a function of m0 . (b) Deviation between F N and the exact value. (m0 is chosen to
be 0.95.)
in Fig. 3(a), the improved series at = 1, m = 3/2 fluctuates violently with respect to m0 . The
deviation from the exact value grows larger as the order increases. (The artificial parameter m0
is chosen to be m0 = 0.95.) The improved perturbation method fails to reproduce the exact result
in this case at high enough orders (Fig. 3(b)).
In order to obtain a reliable estimate by the improved perturbation method, we have to combine various information of the series of different orders, and examine synthetically the whole
profile of the series.
2.3.4. Ring properties
In this section, we examine how the improved perturbation method affects the function ring
C . We are especially interested in the derivative operation, for the expectation value of an
operator is related to the derivative of the free energy with the corresponding source term.
Usually we can introduce an equivalence relation into a set of elements of function ring,
and classify them into equivalence classes. Now we insist that the plateau criteria works for such
classification; two elements of the function ring are equivalent if there exists intersection between
plateau of each function where minimal sensitivity is realized.
First we consider the add operation. Assume that we have two functions, F and G, and the
improved series, F N and GN , respectively. The procedure to obtain the improved series is basically Taylor expansion, and thus linear. If F N and GN have intersecting region where minimal
sensitivity condition for each series is realized simultaneously, then the improved series of sum
of the functions, (F
+ G)N also stays stable and reproduces the exact value there.
On the other hand, when F N and GN do not have overlapped parameter region, the sum
(F
+ G)N does not in general bear stable region in their plateaux.
Thus, the improved perturbation method preserves the add operation among the elements of
function ring that belong to the same equivalence class based on the presence of intersection of
plateau.
56
Unlike the case of the add operation, there is no reliable discussion on the product operation
of function ring. However, from several examples, it seems that the procedure also preserves the
product operation among the elements of the same equivalence class.
Next we consider the derivative operation. Similar to the ordinary function, the derivative of
the improved series, F N (, m; m0 ), with respect to m will be defined as follows:
d N
F N (, m + ; m0 ) F N (, m; m0 )
(16)
F (, m; m0 ) = lim
.
0
dm
We will designate as R(, m) the region in the parameter space m0 where the minimal sensitivity
condition is satisfied. We also call it as region of minimal sensitivity below.
If two regions of minimal sensitivity for physical parameters at m + and m have overlap,
R(, m + ) R(, m) = ,
(17)
we would conclude from the argument for the add operation, that the derivative function defined
in Eq. (16) bears the region of minimal sensitivity in the intersection of the two regions. In short,
for a function and its derivative, the region of minimal sensitivity will be expected to emerge
somehow close to each other.
Otherwise, it may happen that the region of minimal sensitivity of F N and its derivatives are
uncorrelated. Then, in what situation does the overlap vanish, i.e., R(, m + ) R(, m) = ?
This question is deeply related to the critical behavior of the underlying physical model. In usual
cases, free energy and various expectation values are regular with respect to moduli parameters,
and the region R(, m) evolves regularly as well. However, at singular points of moduli space
such as those where the symmetry of the model is enhanced, R(, m) should emerge in a quite
different manner; thus even for infinitesimally small deviation of moduli parameter, R(, m) and
R(, m + ) may not have overlaps.
Such a situation often occurs in the physical models of our interests. For example, in Ising
model, the region of minimal sensitivity drastically changes between high temperature phase and
low temperature phase. In the IIB matrix model (which will be examined extensively in later
sections), we are interested in the massless region, where the parameters become singular. In
these cases, the free energy and the expectation values of various observables show different
behavior, and therefore we have to treat them separately.
3. Plateau conditions
In this section, we will elucidate the criteria how to determine desirable regions of artificial
parameters, and to estimate the approximate value from the improved series. A guiding principle, minimal sensitivity, will be realized in regions in the parameter space where the improved
function stays rather stable. We call such region as plateau.
The subject of this section is to develop a concrete and objective scheme for identifying
plateau. Besides, the notion, stable is obscure, and so the procedure may also provide an inductive definition of plateau itself.
3.1. Minimal sensitivity and plateau
The improved perturbation method introduces artificial parameters, m0 , which we need to
determine by some means. Here we adopt as a guiding principle, minimal sensitivity, that the
improved function should depend least on those parameters; this is because m0 were originally
57
introduced as the nominal shift of parameters, though the dependence on them appeared due to
dropping the O(g N+1 ) terms. If there exists a region in the parameter space m0 where the dependence on m0 vanishes effectively, the exact value should be reproduced there. In the following we
denote simply by F N the improved series F N ( = , m = m ; m0 ) where we substitute each
desirable values and m for and m. This series is function of artificial parameters m0 .
The principle of minimal sensitivity is realized through the emergence of the region in the
parameter space where the improved series stays stable against any variation of artificial parameters; we call such a region as plateau. What we have to do for determining m0 is to identify
where the improved series becomes stable.
When the number of parameters are one (or at most two), the stable region of the function will
be pointed out by just drawing a graph of it. However, the visualization will not be possible for
the cases with more than two parameters. Moreover, even though it is possible, it may sometimes
lead to misjudgment by changing the scale arbitrarily. Therefore we need a concrete scheme for
identifying the plateau applicable to multiparameter space, which is not affected by intuitive
guesses.
3.2. Identifying plateau
The ideal realization of the principle of minimal sensitivity may have such a property that
the improved series is totally independent of the artificial parameters m0 in a region. It involves
the situation that all orders of derivatives of the improved series with respect to m0 are zero in
the region. However, such an ideal plateau is not realized in the actual cases because we have
only finite order of series. Typical profile of the improved series that forms plateau exhibits a flat
region in which the series fluctuates bit by bit; it would accompany a number of local maxima
and minima there. Thus, to turn the argument around, we consider the accumulations of extrema
as indications of plateau (or its candidates).
It should be noted that there is also a case when the series becomes stable without forming
any extrema in that region; we may miss such plateau by the above speculation. We will discuss
this type of plateau in later section.
3.2.1. Cluster identification
Assume that we have already found extrema of the improved series in the (multidimensional)
parameter space. Next issue is to identify the accumulation among them. We denote the coordinate of ith extrema by x i , and the distance between ith and j th extrema by dij . The definition of
distance in the parameter space will be presented later.
The distribution (r) of the distances dij shows characteristic behavior according to the presence of clusters of points. Here, (r) is given by the number of pairs (i, j ) whose distance dij
falls on to r < dij < r + r. As depicted in Fig. 4, when one or more clusters of points exist, (r)
has a large peak near r = 0 and some small bumps at larger r. Let n the number of points in the
cluster. Then the peak height is proportional to n2 , while those of other bumps are of order n. If
there are no such clusters and the points are scattered rather uniformly, (r) shows continuous
distribution. This concept can be extended straightforwardly to the case in which more than one
cluster are formed. From those distinctive properties, we are able to identify the formation of
clusters and the set of points belonging to them.
There is no intrinsic definition of distance dij in multidimensional parameter space. Therefore we have to choose a suitable definition. It introduces metric space and must obey the triangle
inequality.
58
(a)
(b)
Fig. 4. Distributed points (above) and distributions of distance between points (below) for each case: (a) forming the
cluster, and (b) straggling uniformly.
(18)
As another choice, if we can define a reasonable metric gab from some argument such as dimensional analysis, we will have
dij =
(19)
g ds,
geodesic
along geodesic between x i and x j . The geodesic equation may not be solved nor have any solution; we would alternatively choose a linear interpolation with weights by metric as
dij =
g(tk )x (tk ) x (tk+1 ),
x (tk ) = x i +
k
(
xj x i ),
N
(20)
59
It is not always that two extrema which close to each other in the meaning of the above
distances belong to one stable region. This is because it is possible that between these extrema
improved series varies its value drastically and vicinity of two extrema encounters by chance. In
order to avert this situation, we consider one other choice of distance. This choice comes from
extending the concept of moduli space. By the definition of distance we intend to express the
degree to which two of the extrema of a function F N resemble each other. Thus we include the
deviation of F N and its derivatives at these two points into the notion of distance:
d k F N
d k F N 2
2
2
dij = 
(21)
xi x j  +
wk
.
dx k x i
dx k x j
k
xi x j 2 +
dij2 = 
M,M =0
MM
k
k M
k M 2
d F
d F .
wk
dx k x i
dx k x j
(22)
The relative weights wk and MM above should be chosen according to the models considered.
In general, we cannot introduce a particular choice of metric naturally. Then we try some
definitions of metric to a model and investigate the improved perturbative series carefully.
3.2.2. Weighted histogram analysis
If the improved series F N bears a plateau, the distribution of the improved series in a region
enclosing the plateau should form a peak corresponding to the value of F N , which gives an
estimate of the function on the plateau.
Therefore, we choose a region (for ease of operation, we usually consider rectangular one)
which encloses accumulation of extrema, and evaluate distribution (F ) of F N [32],
(F ) = dx 1, A = x  F < F N (x) < F + F .
(23)
A
In multidimensional parameter case, the extent of the plateau is not always large nor the
shape isometric in the parameter space. Then we need a method to extract the estimated value of
F N on plateau effectively even when the region to be examined is taken roughly.
The plateau is, if ideally realized, characterized by the feature that the first and higher derivatives of F N should be zero or small. To enhance the contribution from such flat region, we
introduce weight function to the distribution as
(F ) = dx w.
(24)
A
1
d F N 2
dxi 
1
 det
(25)
d 2 F N 2
dxi dxj 
(26)
60
The histogram of weighted distribution shows sharp peak corresponding to the zeros of kth
derivatives of the improved series. Even though the region in question is taken roughly, the contribution from the fluctuating part of F N is well suppressed, and we will have an estimate of F N
on the plateau.
There is also another choice of weight function based on the convergent behavior of the series
on the plateau. If the improved series converges to some value on the plateau, it implies that the
difference between the improved series of order N and order N + 1 diminishes. To reflect such
speculation, we would better choose a weight function by
w3 =
1
F N+1 F N 2
(27)
In general, we combine the above weight functions on a casebycase basis. In the following
investigation into the IIB matrix model, we use the two weight functions w1 (25) and w2 (26).
3.2.3. Signal and noise
When the number of parameters are one (or at most two), we can easily recognize the flat region from the graph. However, there is no such simple scheme in multiparameter case in general.
We are trying to identify plateau from the accumulation of the extrema as a clue. The information
we have so far for the improved series F N is just the location of extrema and profile of (weighted)
distribution of F N in some specified regions. The accumulations of extrema provides candidates
of the plateau. Among these candidates we have to distinguish the true plateaux, though yet we
do not have definite argument on this subject. We exemplify one typical false signal, overshoot.
It often occurs when a discrete function is approximated by a series of polynomial (consider,
for example, Fourier transform of rectangular waves) that the series deviates relevantly at discontinuities, which remains with even higher orders taken into account. Such behavior is called
overshoot. In the current case, the improved series turns to be an approximation of a constant
function by a set of polynomials because the exact value is totally independent of artificial parameters. Therefore the improved series are apt to show typical overshoot behavior as well (Fig. 5).
The overshoot is rather isolated from other extrema that belong to a plateau; this should be
reflected in the characteristic profile of weighted distributions. The overshoot gives minimum
(or maximum) in its neighbor, so the histogram of the distribution weighted by w1 shows sharp
peak, and it is cut away on lower (or upper) side (Fig. 6(a)). Moreover it often happens that
the peaks in the distributions with second and higher derivatives deviate from peaks in that with
the first derivative. It is because the overshoot appears as isolated extrema away from those
forming flat region. On the contrary, in the case of plateau, the histogram of the distribution
weighted by w1 and w2 shows symmetric shape and a peak of the distribution weighted by w2
lies between (or ideally on) peaks of the distribution weighted by w1 (Fig. 6(b)). For the reason
stated above, a cluster of extrema which corresponds to overshoot can be distinguished from
plateau by investigating the shape of distributions.
Besides, it seems that the overshoots of different orders tend to form a line in the parameter
space. If there is such sequential structure, we had better suspect that they are overshoots.
3.3. Pitfalls and special cases
3.3.1. Smooth plateau
In the prescription thus far explained, we assumed that the improved series F N shows fluctuating behavior on the plateau, accompanied by the accumulation of extrema. If there is a region
61
(a)
(b)
Fig. 6. (a) Distribution weighted by w1 , w2 in overshoot region and (b) distribution weighted by w1 , w2 in plateau region.
62
where F N becomes stable but varies gently without forming extrema (Fig. 7), it should also
be considered as plateau, though the above procedure based on the existence of extrema will
not be applicable to such cases. We call this type of plateau as smooth plateau. However, the
histogram will show the peaked structure corresponding to this smooth plateau. The weighted
distribution should provide a reasonable estimate of the value.
The problem is that we do not have any guide to locate the position of plateau in this case. One
clue is the consistency with the plateau of other physical quantities. Though the improved series
of different observables may show distinctive behavior, the plateau corresponding to the physical
states should somehow be realized in each series. We have to guess the region of smooth plateau
in parameter space based on those information. If we find a plateau in analysis of an observable,
we must suspect the appearance of smooth plateau in neighborhood of that region in the case of
the other observables even for absence of any extremum.
It should be noted that we have to take care not to be confused with mere asymptotic behavior
of the series. A concrete example will be shown in the analysis of the IIB matrix model in later
sections.
3.3.2. Plateau on special hypersurface
Consider a hypersurface in the parameter space where one of the parameters is zero. If we
change the coordinate from m0 to x0 = m10 , the first derivative with respect to the new coordinate
becomes,
d F N
d F N
= m20
.
dx0
dm0
N
(28)
N
F
F
does not vanish, ddx
becomes zero at m0 = 0. Thus we have to investigate
Even though ddm
0
0
the case separately when some of parameters take zero or . This is because we do not know
what choice of the artificial parameters is preferred.
Those cases are significant in that m0 = 0 or occasionally corresponds to the point where
symmetry of the original physical model enhances; therefore the improved series may have singular structure.
63
the region which encloses each of the candidates, and distinguish the plateau among them. The
histograms also provide estimates of the exact value from the improved series.
This procedure assumes typical profile of the improved function. It will not be applicable
when the function stays stable without forming extrema. In these cases, we have to locate the flat
region referring to the information of other physical quantities, such as symmetry considerations.
We also have to take care of the special hypersurface where some of parameters take zero or
infinity; such cases may be relevant in the original physical models.
4. IIB matrix model
In this section we apply the improved perturbation method to the IIB matrix model and investigate its nonperturbative solutions. In particular we concentrate on the free energy and the
second moment of eigenvalue distribution. By comparing the values of free energy of the solutions, we can see which solution is most preferred. In the IIB matrix model, the distribution
of eigenvalues is interpreted as the spacetime itself. Therefore we can recognize the shape of
the universe by computing the eigenvalue distribution. To be more precise, the square root of
the second moment of eigenvalues distribution gives the scale of extent to each direction of universe. The deviation of the ratio of spacetime extent between directions from unity indicates the
degree of anisotropy of the universe.
4.1. Model setup
The action of the IIB matrix model is
2
gYM
gYM
2
a
[Xa , Xb ]
Xa , ,
S0DYM = Tr
4
2
(29)
N 1/8 ,
2
gYM
N .
(30)
(31)
In order to carry out the perturbative expansion, we add the following propagator terms to the
action
10
10
1
1
S0 = N Tr
(32)
mab Xa Xb +
mabc abc .
2
2
a,b=1
a,b,c=1
They are the most general quadratic terms. The fermionic bispinors are expanded by antisymmetric tensors by using gamma matrices. In the case of tendimensional MajoranaWeyl spinor,
by considering a chirality, only gamma matrices of rank three are allowed.
64
The scaling behavior of the Feynman rules for the amplitude with respect to N and is given
as follows:
1
,
N
1
propagator for fermionic fields : ,
N
3point vertex:
N ,
4point vertex:
N .
If we denote the genus of dual surface by g and the number of faces of dual surface by NF , we
find that the contribution of a single diagram is
amplitude N 2(1g) NF /2 .
(33)
If we take the limit N with fixed, the planar diagrams (g = 0) give the leading contributions. In this case the amplitude is given by
amplitude N 2
(34)
n f n .
n
Free energy F is evaluated by the sum of connected bubble diagrams in terms of propagators ma
and mabc as
1
F
=
ln
+ .
(35)
2
m
N
The expectation value of the second momentum of eigenvalue distributions is obtained by the
derivatives of free energy with respect to the parameters.
1 F
1
1
2
(36)
=
Tr
X
+
.
=
ln
a
N
ma
m
N 2 ma
The original IIB matrix model does not have the propagator term Eq. (32). Therefore we
evaluate the observables at ma = mabc = 0. They are singular in this region of parameters in the
above perturbation. We use the improved perturbation to extrapolate to the massless case.
In general, all parameters mab , mabc have to be taken into account. We may perform SO(10)
rotation to reduce mab to the diagonal form at the stage of action. We will impose further restrictions from the practical reason in the later sections.
Before going into a concrete computation, we make some assumptions. In computing the
free energy we have to evaluate all connected bubble diagrams. As we proceed to the higher
order of perturbation, the number of diagrams becomes enormous. We consider such a situation
that the U(N ) symmetry stays intact. Throughout this article we concentrate on the spontaneous
breakdown of Lorentz symmetry.3 Then we can omit the tadpole amplitude
a
X ij = 0
(37)
for the matrix corresponding to the adjoint representation of SU(N ) subalgebra. By the argument
of the gauge symmetry alone we cannot forbid the tadpole amplitude of U(1) part. If it takes non3 The breakdown of U(N ) gauge symmetry will be discussed in the subsequent paper [35].
65
(38)
(39)
As we show later the effective action bears dynamical mass term. Then the zero point of free
energy is shifted due to the redefinition of fields by Eq. (38). Thus we cannot use the zero of free
energy as a characteristics of supersymmetry.
Under the assumption above, the sum of connected bubble diagrams reduces to the sum of
oneparticle irreducible (1PI) bubble diagrams. Thus far we have to evaluate 1PI bubble diagrams
in order to obtain free energy.
In the estimation of 1PI bubble diagrams we can drastically simplify the computation by using
2PI free energy [27]. At first we compute 2PI graph amplitude G2PI by using full propagators
1
1 1
(Xa )ij (Xb )kl = cab il j k ,
(40)
( )ij ( )kl =
uabc C abc il j k ,
N
N 3!
in place of perturbative propagators ma , mabc . Here we use double line notation, and C is the
charge conjugation matrix.
From G2PI we obtain 1PI bubble amplitude F1PI by Legendre transformation [36]4
F(ma , mabc ; ca , uabc ) G2PI (ma , mabc ) +
0=
F
,
ma ca =c(m)
10
10
1
1
ma ca
mabc uabc ,
2
2
a=1
(41)
a,b,c=1
(42)
(43)
Eq. (42) corresponds to the SchwingerDyson equation which relates full propagators ca and
uabc to perturbative propagators ma and mabc , and Eq. (43) produce the 1PI amplitude using
bare propagators ma , mabc .
In the present article, we use the result of the perturbative expansion up to eighth order in
of Ref. [29].
4.2. Ansatz
In this subsection, we make assumptions for the pattern of symmetry breakdown. In the case
of the IIB matrix model, the total number of artificial parameters is quite large, namely, 10 real
numbers for mab (assumed to be diagonalized by SO(10) rotation), and 120 for mabc . It will
demand an enormous effort to search for plateau in this vast space of parameters. Therefore, we
impose restrictions on the configuration by considering symmetry to diminish the number. In
concrete manner, we make same assumptions for the full propagators cab and uabc because we
compute 2PI bubble diagrams by using these parameters.
In the former works [2629], the configurations called SO(d) ansatz have been intensively
examined which preserve SO(d) rotational symmetry. The guideline of choice is described as
4 For details see also Ref. [27].
66
follows. First, SO(d) subgroup of SO(10) is chosen to which directions the expectation values of fermionic twopoint function uabc are zero. d is taken from 1 to 9. Toward the rest of
the directions uabc may have nonzero value. Since uabc is a rank three antisymmetric tensor,
a single nonzero component of uabc brings out threedimensional subspace by permutation of
indices. Thus, (10 d) dimensional part would naturally be decomposed into multiples of threedimensional blocks. Furthermore, those blocks are subjected to the permutation symmetry under
the interchange with each other. In this way, SO(d) (d = 1, . . . , 7) symmetric ansatz has been
investigated in former works.
Among those choices shown above, we examine d = 4 and d = 7 cases in particular. It is
reported in Ref. [27] that d = 5 and 6 cases reduce to SO(7) ansatz, while d = 2 and 3 cases to
SO(4) ansatz. d = 1 had no solution.
The preserved symmetry and the explicit forms of the full propagators for d = 4 and d = 7
cases are given as follows.
SO(7) ansatz: SO(7) SO(3)
cab = diag(7c1 s, 3c2 s),
u
/ = u 8,9,10 ,
(44)
{m} and {m0 } represent collectively the set of parameters {m1 , m2 , m} and {(m0 )1 , (m0 )2 , m0 },
respectively. We put m1 = m2 = m = 0 so that it describes the original IIB matrix model. At this
stage, we set = 1. It is because the coupling constant can be absorbed by the redefinition of
the fields.
We introduce the new variables xi as
x1
1
,
(m0 )1
x2
1
,
(m0 )2
x3
1
,
m0
(47)
67
Fig. 8. Distribution of extrema of improved free energy for SO(4) ansatz plotted in x1 x2 plane.
candidates of plateau. We compute the weighted distribution of the functions in those regions.
We use here the first derivatives (25) and the second derivatives (26) in particular. Finally, we
identify the region as plateau in which two distributions overlap with each other. The graphs of
distributions in individual regions are shown in Appendix A.
4.3.1. Free energy
Fig. 8 shows the distribution of extrema of the improved series of free energy. As is seen
in Fig. 8, there are several accumulations of extrema. We pick up some accumulations as the
candidates of plateaux and compute the distributions with weight functions w1 and w2 . Besides
most of accumulations show the hopeless distributions which spread randomly, two regions A
and B which are depicted in Fig. 8 have a hopeful distribution. In the following, we refer to only
hopeful candidates.
Two regions A and B are hopeful candidates of plateau, although the behavior of the function
of the seventh order in region A is rather unstable. The similar situation also occurred in the
previous work [29] in which the positions of extrema show a different pattern for the seventh
order of perturbation. In the region A we read the value of free energy to be 12. And in the
region B, we also read the value of free energy to be 12.
From this accordance between region A and B about the value of free energy, we may expect
that two regions belong to one plateau. However from the examination of the other observables
in the later subsections, we conclude that two regions are distinguished each others and there are
two plateaux.
4.3.2. c1
Fig. 9 shows the distribution of extrema of the improved series of c1 . As mentioned in former
section, although c1 is the derivative function of free energy with respect to m1 , we have to apply
the improved perturbation separately. It is because the behavior of the function becomes singular
in the region of moduli space that corresponds to the massless case. It is seen by comparing Fig. 8
with Fig. 9. Two distributions of extrema show the different pattern.
68
Among accumulations, the regions A, B and C are promised candidates of plateau. The distribution in the region A shows a strange behavior. As the order of perturbation increases, the value
at the peak of distribution shifts to the right. When we approximate an infinitely large value by
a finite series, such a behavior may often be observed. Therefore this property indicates that the
exact value is quite large. Up to the eighth order of perturbation, we conclude that c1 takes value
of 4060. From the distribution in the region B, we read the value of c1 to be 1.52.1. Although
regions A and B point the similar value of free energy, these regions shows clearly different
distributions with respect to c1 . Therefore we conclude that these regions belong to different
plateaux. On the other hand, from the region C, we read the value of c1 to be 0.250.26. Thus
we conclude that there are three plateaux. In particular, it is interesting that the region C shows
the same value as the region C of c2 and SO(7) ansatz which will be shown in later subsections.
The significance of this accordance will be discussed later in Section 4.5.
4.3.3. c2
Fig. 10 shows the distribution of extrema of the improved series of c2 . We consider the regions
A, B, C and D as candidates of plateau. From the distribution in the region A, we read the value
of c2 to be 0.00.2. From the distribution in the region B, we read the value of c2 to be 0.130.15.
From the distribution in the region C, we read the value of c2 to be 0.250.26.
In the similar fashion, region D indicates that c2 takes the value 0.210.25. From the similarity
of the distributions in the region C and D, we conclude that these two regions belong to one
plateau. Thus we choose the region C as a representative of this plateau.
4.4. Results in SO(7) ansatz
In this subsection, we show the results for the case of SO(7) ansatz. The method of analysis
is the same as that in the SO(4) case. The graphs of distributions in individual regions are shown
in Appendix A.
69
Fig. 11. Distribution of extrema of the improved free energy for SO(7) ansatz plotted in x1 x2 plane.
70
4.4.2. c1
Fig. 12 shows the distribution of extrema of the improved series of c1 . Regions A and C
are hopeful accumulations. From the distribution in the region A, we read the value of c1 to be
0.70.8. From that in the region C, we read the value of c1 to be 0.250.26.
4.4.3. c2
Fig. 13 shows the distribution of extrema of the improved series of c2 . Regions A and C are
hopeful regions. From the distribution in the region A, we read the value of c2 to be 0.080.12.
From that in the region C, we take the value of c2 to be 0.250.26.
71
Fig. 14. Dependence of the improved free energy with respect to x1 and x2 at x3 = 0.1 (left), and x3 dependence of the
improved free energy at x1 = x2 = 0.15 (right) for SO(4) ansatz.
72
a property that the degree of anisotropy is large (probably infinite). Next plateau (we call this
region as region B) indicates that the value of free energy is 12 and that of c1 is 1.52.1 and
that of c2 is 0.130.15. This plateau also corresponds to an anisotropic vacuum that preserves
SO(4) subgroup. The ratio of extent of the eigenvalue distributions between the 4dimensional
part and the 6dimensional part is obtained by 3.14.0. This solution has a property that the
degree of anisotropy is finite.
The third candidate appears in a different manner. This is the plateau corresponding to a
SO(10)symmetric vacuum which was discussed in the previous subsection. For the observables
c1 and c2 , it indicates that both of c1 and c2 take the value 0.250.26. For the free energy this
plateau appears as smooth plateau. By computing the distribution of free energy we conclude
that the approximate value of free energy is 67. The graphs of distributions in this regions are
shown in Appendix A.
Similar argument can be applied to the case of SO(7) ansatz. There are two plateaux; one
of which indicates that the value of free energy is 57 and that of c1 is 0.70.8 and that of c2 is
0.080.12, and it corresponds to an anisotropic vacuum that preserves SO(7) subgroup. The ratio
of the extent between the 7dimensional part and the remaining 3dimensional part is 2.43.1.
The other plateau corresponds to a SO(10)symmetric vacuum. This plateau indicates that the
value of free energy is 67 and that of c1 and c2 is 0.250.26. Those values are the same as those
obtained for the third plateau of SO(4) ansatz. It is consistent with the speculation that they are
actually identical, and SO(10) symmetry is realized.
5. Conclusion
We have investigated the properties of the improved perturbation method in much detail. This
method is a sort of systematic optimization of a series expansion obtained by the perturbation
theory, achieved by reorganizing the series with the artificially introduced parameters. In particular, we focus on the realization of the principle of minimal sensitivity, which provides a condition
for the parameters that the improved series should be least dependent on them. A region in the
parameter space that satisfies the above condition is called as plateau. The exact value of the
series would be reproduced on the plateau. Thus the efficient scheme for the identification of the
plateau and evaluation of the approximate value becomes a major issue in the application of the
improved perturbation method.
We proposed a new scheme for evaluating the improved series which yields good approximate values. It relies on the distribution profile of the series that is weighted by the first and the
second derivatives with respect to the artificial parameters. By incorporating those weightings,
the distribution becomes more sensitive to such a behavior that the series stays almost constant
in a certain region of the parameters, i.e. the plateau condition be satisfied. It is also noted that
this scheme works effectively even when there are a number of parameters.
We insist that the improved perturbation method should be applied independently for each
individual observable of interest. It is because the prescription of improvement is quite nonlinear and the formation of plateau may occur in a different manner especially at low orders of
perturbation. If they are supposed to coincide at high enough orders, the information of plateau of
other observables may help to clarify the behavior of the improved series. Thus those information
should be referred synthetically.
73
We studied the IIB matrix model by the improved perturbation method in order to examine the nonperturbative solutions of the model. We concentrate on the possibility of the
spontaneous breakdown of Lorentz symmetry in the present article. We applied our scheme
to the analysis of the improved series up to eighth order of perturbation for SO(4) and SO(7)
ansatz that have been obtained in former works. As a result, we found a new solution that
corresponds to the tendimensional universe besides the four and sevendimensional solutions
already found previously. A key to this detection resides in that we performed the analysis independently for individual observables and combined the information on the plateau of
them.
The existence of the plateau of the SO(10)symmetric vacuum is quite substantial because it
provides an evidence for the validity of application of the improved perturbation method to the
IIB matrix model. It ensures us that the investigation encloses the whole moduli space of the
model including the SO(10) symmetry and that the improved series indeed describes the original
SO(10)symmetric IIB matrix model. It is a nontrivial issue when the model exhibits phase
transitions, as was discussed in the application to the Ising model [22].
Based on our new scheme for the improved perturbation method, we found that the value of
free energy becomes large in the order of four, seven and tendimensional universe. Thus we
conclude that the fourdimensional universe is more preferred than seven and tendimensional
universe.
It is interesting that there are two distinct solutions that correspond to fourdimensional universe. One is the solution in which the degree of anisotropy is quite large, and the other is the
solution in which the degree of anisotropy is finite. The estimated values of free energy are
almost the same for those two solutions. It may suggest an interesting possibility that the inflation is caused by the phase transition from one SO(4) vacuum with finite degree of anisotropy
to the other SO(4) vacuum that has infinitely large extent in four directions, i.e., our universe.
Acknowledgements
The authors are grateful to H. Kawai and T. Matsuo for valuable discussions. Y.S. is supported
by the Special Postdoctoral Researchers Program at RIKEN.
In this appendix we show the graphs of distributions in individual ansatz and these regions
which are candidates of plateaux.
As we can see in Figs. 813, for each observables there are many accumulations of extrema.
We must compute the distribution with weight function w1 , w2 for each region. Most of regions
show hopeless distributions. Then we select only the hopeful regions here and depict the graphs
of the distributions in these regions. The locations of these regions in the artificial parameter
space are depicted in Figs. 813.
74
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
Region B
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
75
76
A.1.2. c1
Region A
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
Region B
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
77
78
Region C
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
A.1.3. c2
Region A
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
79
80
Region B
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
Region C
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
81
82
Region D
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
83
84
Region B
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
A.2.2. c1
Region A
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
85
86
Region C
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
A.2.3. c2
Region A
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
87
88
Region C
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
(a) order 3
(b) order 4
(c) order 5
(d) order 6
(e) order 7
(f) order 8
89
90
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
Abstract
Recently it has been suggested that nonGaussian inflationary perturbations can be usefully analysed in
terms of a putative dual gauge theory defined on the future conformal infinity generated by an accelerating
cosmology. The problem is that unitarity of this gauge theory implies a strong constraint (the Strominger
bound) on the matter fields in the bulk. We argue that the bound is just a reflection of the equation of
state of cosmological matter. The details motivate a discussion of the possible relevance of the dS/CFT
correspondence to the resolution of the Big Bang singularity. It is argued that the correspondence may
require the Universe to come into existence along a nonsingular spacelike hypersurface, as in the theories
of creation from nothing discussed by Firouzjahi, Sarangi, and Tye, and also by Ooguri et al. and others.
The argument makes use of the unusual properties of gauge theories defined on topologically nontrivial
spaces.
2006 Elsevier B.V. All rights reserved.
92
There are, however, some doubts as to whether a precise de Sitter analogue of the AdS/CFT
correspondence can actually be constructed. It is generally agreed that a de Sitter phase can only
be metastable in string theory [10,11]; generically it would decay to a cosmological spacetime
with a Crunch; so the future boundary on which the dual theory is to be defined may not exist.
This could simply mean that the dual theory can only be defined on some spacelike hypersurface
at large but finite proper time, or that the dual theory can only be formulated in the Euclidean version of the spacetime (as in the work of Maldacena and Maoz [12]), or it may imply more serious
limitations. A concrete proposal for understanding the limitations of dS/CFT was suggested by
van der Schaar [6], who argued that dS/CFT is effective only at the level of a coarsegraining in
which each lattice site of the field theory corresponds to an entire static patch in the de Sitter bulk.
One way of stating the case would be to suggest that, on the bulk side of the correspondence,
dS/CFT is primarily relevant to specifically cosmological features of spacetime and its matter
content, not to their detailed structure.
In this spirit, we wish to make an observation regarding a curious and apparently unphysical
feature of the dS/CFT correspondence: it appears to impose a severe upper bound on the masses
of particles. In detail, the limit of a massive pform field amplitude (for a pform field of mass
m ) in de Sitter spacetime defines a CFT twopoint function at de Sitter conformal infinity. In the
p = 0 case, the conformal weight corresponding to the boundary operator defined by is given,
in four spacetime dimensions, by
1
h+ = 3 + 9 4L2 m2 .
(1)
2
One sees immediately that the weight will be complex unless m satisfies
m2 9/ 4L2 .
(2)
Presumably a violation of this bound would therefore signal a failure of unitarity in the field
theory. We stress that while this conclusion is derived for scalar matter, it applies to all kinds of
pform matter: see the discussion of the analogous extension in the AdS case in [13].
This Strominger bound [2,14,15] on particle masses is analogous to the wellknown
BreitenlohnerFreedman bound [16] on the masses of pform fields in antide Sitter spacetime;
but, being an upper bound, it is at first sight much less reasonable. For some purposes, it can be
ignored: it has been argued by Seery and Lidsey [7] that only very lowmass scalars could give
rise to nontrivial perturbations at late times, so the Strominger bound is not a problem for the
recent studies of inflationary perturbations from the dS/CFT point of view. However, one would
certainly prefer to have a more precise demonstration that the Strominger bound really does not
impose unreasonable conditions on any of the matter fields which are relevant on the scales to
which the dS/CFT correspondence applies.
We shall suggest a solution of this problem inspired by the arguments of van der Schaar
[6] mentioned earlier: we shall assume that dS/CFT should only be expected to yield useful
information at the cosmological level, and not on smaller scales. The relevant forms of matter
at this level, apart from the dark energy/inflaton (both of which we represent by a cosmological
constant) are zeropressure (nonrelativistic) matter, radiation, and, in the preinflationary era,
fluids representing strings and other extended objects. The question is then whether there is
some sense in which these specifically cosmological forms of matter do in fact satisfy the
Strominger bound. If this is not the case, then dS/CFT would imply that the apparently reasonable
demand of unitarity on the part of the field theory imposes unphysical conditions on bulk physics,
and this would call the whole approach into question.
93
We argue here that cosmological matter does (barely) satisfy the Strominger bound: interestingly, zeropressure matter actually saturates the bound. We conclude that the bound is not
unreasonable, provided that the limits to the validity of dS/CFT, of the kind pointed out by van
der Schaar, are kept in mind. The agreement of the demand of unitarity on the boundary with the
requirement that the bulk field reproduce the classical equation of state is rather remarkable.
One of the major hopes for the AdS/CFT correspondence is that it will allow us to probe the
singularities in black holes [1719]. Similarly, one might hope to use AdS/CFT or dS/CFT to
probe the Big Bang singularity [20]. Unfortunately, even in the AdS case, it is extremely difficult
to do this in an even partially realistic way. For some recent examples: in [21] an ingenious
attempt is made to study de Sitter space by embedding it inside an AdS black hole, but the
resulting cosmology has no Big Bang; conversely, the AdS/CFT correspondence has been used to
study Bang singularities in certain cosmological models [22], but these are not asymptotically de
Sitter to the future. For dS/CFT the case is still worse, since so little is known about the putative
boundary theory. In the concluding section we draw attention to the fact that the topologically
nontrivial spacetime structure assumed here (we take the spatial sections to be tori) can be
expected to give rise to some unfamiliar physics, of the kind discussed some time ago by Krauss
and Wilczek [23] and by Preskill and Krauss [24]. If the field theory at infinity is of this kind, in
which parallel transport around noncontractible loops can cause discrete transformations, then
this can give rise to very puzzling behaviour deep in the bulk, near to the initial singularity. We
speculate that this kind of topological holography could have some role to play in resolving
that singularity, when the dS/CFT correspondence is better understood.
2. Representing cosmological matter by a massive field
In this section, we show how to represent the familiar forms of cosmological matter in a way
that can be connected with the Strominger bound. At the same time, we must incorporate the
effect of such matter on the spacetime geometry.
We begin with the version of de Sitter spacetime with flat spatial sections; we compactify
these sections to copies of the (cubic) threetorus T 3 . It has been argued elsewhere that this
is the most suitable global structure for investigations of quantum cosmology [2527]. One of
the many virtues of this toral version of de Sitter spacetime is that its boundary at infinity is
compact and connected, so that we can avoid the subtle issues [2] which arise when the dS/CFT
correspondence is constructed on the spatially spherical version of de Sitter spacetime, which
has two spheres at infinity. Here, instead, we have one torus at (future) infinity. Henceforth, then,
our working hypothesis is that the spacetime topology is R T 3 (or possibly some nonsingular
quotient).
Thus the metric, before we introduce any matter, is that of spatially toral de Sitter:
g(STdS) = dt 2 K 2 e(2t/L) d12 + d22 + d32 ,
(3)
where 1,2,3 are angular coordinates on a cubic threetorus, where K defines the spatial length
scale, L is the scale defined by the cosmological constant, and we use (+) signature.
Now we wish to introduce into this spacetime the familiar cosmological matter fieldsnonrelativistic matter, radiation, cosmic strings and so on. A common property of all these forms
of matter is that they have a constant equationofstate parameter w, where w specifies the ratio
of pressure to density in the homogeneous case. In order to determine whether the Strominger
bound is satisfied, we have to find a formal representation of these matter fields by some pform
94
field with a definite mass. In view of the isotropy and homogeneity of cosmological matter, the
obvious choice is to try to do this using a scalar field (p = 0).
The solution is as follows. We take STdS spacetime, and introduce into it a homogeneous
scalar field with the usual kinetic term and with a potential
3
1
V (, ) =
(4)
1 sinh2 ( 2);
2
6
8L
here is a positive constant. Bear in mind that STdS spacetime in this signature has a negative cosmological constant 3/L2 which gives rise to a positive energy density 3/(8L2 ). The
precise numerical factor in the potential is chosen for later convenience.1
We now make the following claim: we assert that, as far as the spacetime geometry is concerned, the field exactly mimics the effects on STdS spacetime of a homogeneous matter field
with positive energy density and a constant equationofstate parameter related to by
1
w = 1.
(5)
3
Thus for example if we insert nonrelativistic matter (zero pressure, hence w = 0) into the STdS
spacetime and allow it to act on the spacetime geometry, this will have the same effect as introducing with = 3, while with the value = 4 mimics the effects of radiation; = 1
corresponds to a static network of planar domain walls, increasing to 1.5 for the scaling regime
[28]; = 2 for a static network of strings, and so on. We stress that we are not primarily interested in using this field to violate the strong energy conditionasymptotically at least, the
acceleration is due to the negative contribution made by the STdS cosmological constant to the
total pressure. Instead, just represents any form of homogeneous cosmological matter, with a
constant equationofstate parameter w 1, which is to be superimposed on the STdS spacetime.
We now proceed to justify these claims. We shall consider Friedmann cosmological models
with metrics of the form
g = dt 2 K 2 a(t)2 d12 + d22 + d32 ;
(6)
this generalizes the STdS metric in an obvious way. Adding the energy density of the field to
that of the initial STdS space, we have a Friedmann equation of the form
2
8 1 2
3
a
=
V (, ) +
.
(7)
a
3 2
8L2
The equation for itself is
a
dV (, )
+ 3
= 0.
a
d
(8)
Surprisingly, these equations have very simple solutions: one finds that (with natural initial conditions) is given by
1
tanh1 et/2L ,
=
/2
1 This choice normalizes the cosmological scale factor in a convenient way.
(9)
g(, K, L) = dt K sinh
2
(4/)
t 2
d1 + d22 + d32 .
2L
From these results one can compute the energy density and pressure of the field alone:
3
2 t
=
cosech
,
2L
8L2
3
1
2 t
p =
1 cosech
,
2L
8L2 3
95
(10)
(11)
(12)
from which Eq. (5) above is immediate. Note that the density decays more rapidly with increasing
time for larger values of .
Since we are using the standard kinetic term, automatically satisfies the null energy condition. To see the conditions under which the strong energy condition is satisfied, we have to
determine whether the quantity = + 3p is nonnegative, where and p are the total energy
density and total pressure (including the contributions due to the background cosmological constant, which total to 3/4L2 ), respectively. For the metrics we are concerned with here, is a
function of , L, and t given by
3
3
2 t
(, L, t) =
(13)
+ + 3p =
cosech
.
1+ 1
2
2L
4L2
4L2
We see that the SEC is always violated eventually, but that it is satisfied in the early Universe
provided that > 2; otherwise it is violated at all times, just as it is in de Sitter spacetime.
In physical terms, this means that, for all > 2 the Universe decelerates in its earlier (postinflationary) stages, and only later begins to accelerate: this is the case for the postinflationary
phase of our Universe. For 2, by contrast, the spacetime always accelerates, and in fact for
< 2 the acceleration actually diverges as t = 0 is approached. We shall have more to say about
this below.
If = 3, we should have the local metric for a spacetime containing nonrelativistic matter
and a de Sitter cosmological constant, and indeed g(3, K, L) givespurely locallythe classical
Heckmann metric (see [29] for a recent discussion). In the general case it agrees (again locally)
with the results reported in [30], where it is obtained by postulating a linear equation of state
(without giving a matter model). For large t we have
g , 22/ K, L dt 2 K 2 e2t/L d12 + d22 + d32 ,
(14)
which is the STdS metric given in Eq. (3); notice that effectively drops out. Thus our metric is
asymptotically STdS, for all .
All of the metrics in (10) appear to be singular at t = 0, but we remind the reader that such
appearances can be deceptive: for example, consider the metric
g(SHdS4 ) = dt 2 sinh2 (t/L) dr 2 + L2 sinh2 (r/L) d 2 + sin2 ( ) d 2 ,
(15)
with hyperbolic spatial sections (which we do not compactify here). This appears to be singular
in the same way, but in fact this is just a version of de Sitter spacetime, the (noncompactified)
2 The reader who wishes to undertake the task of verifying these solutions will find the following simple fact helpful:
if A and B are quantities related by tanh(A) = eB , then cosh(2A) = coth(B).
96
spatially hyperbolic version [27], which is, of course, entirely nonsingular. The question as to
which members of (10) are really singular is important, and we shall return to it below.
We have shown how to represent cosmological matter in a way that is relevant to the Strominger bound. Let us now show that the bound is in fact satisfied by such matter.
3. Cosmological matter satisfies the Strominger bound
At late times, the metric g(, K, L) (Eq. (10)) is locally indistinguishable from that of de
Sitter spacetime, and, furthermore, is very small (Eq. (9)); hence we see, from (4), that can
be regarded as a scalar field of squared mass
3
1
m2 =
(16)
1
6
2L2
propagating on a local de Sitter background. By computing
9
m2 = ( 3)2 /4L2 0,
(17)
4L2
we see at once that the Strominger bound is automaticallybut only barelysatisfied for all
values of . The conformal weight is real: from (1) we have
/2,
3,
h+ =
(18)
3 (/2), 3.
Notice that by varying , either from 0 to 3 or from 3 to 6, one can obtain all values of the weight
allowed by the Strominger bound (that is, all values between 3/2 and 3): in fact, apart from 3/2,
all allowed values of the weight can be obtained in two different ways. Thus for example h+ = 2
can be obtained either by choosing = 2 or by taking = 4.
From (17) we see that the Strominger bound is most nearly violated when = 3; it is saturated
in this case, that is, in the case of zeropressure, nonrelativistic matter. This has a deep mathematical significance, as follows. Because de Sitter spacetime and Euclidean hyperbolic space
have the same isometry group, the relevant representation theory has been extensively developed
[31,32]. The scalar representations fall into three families, the principal, complementary, and
discrete series. The principal representations are those which, under contraction of the de Sitter
group to the Poincar group, correspond to the familiar flat space representations. They are of
two kinds, which in fact are classified precisely by the weight h+ which appears in the dS/CFT
correspondence. The first kind is the case where h+ is complex: these of course violate the Strominger bound. The only other principal representation corresponds precisely to the case where
the bound is saturated.
The complementary series consists of representations where the Strominger bound is satisfied
but not saturated3 ; these representations have no flatspacetime analogue. (The discrete series
corresponds to the special, massless case.) We conclude that = 3 is the only value which corresponds to a welldefined flat space representation. It is interesting that = 3 is singled out in this
way, for this corresponds to the kind of matter that, in classical cosmology, is the most important
at late times. It is also interesting that the field we have used to represent asymptotically de Sitter
3 The relevant Hermitian form in the complementary case ([31], p. 518) involves a gamma function which is illdefined
if the Strominger bound is saturated, so the = 3 representation certainly does not belong to this series.
97
cosmology with parameter corresponds to precisely the same de Sitter representation as the
cosmology with parameter 6 .
We conclude that the Strominger bound is in fact satisfied by all kinds of cosmological matter. We have shown this by representing cosmological matter by fields which, except in the case
of nonrelativistic matter, correspond exactly to the complementary series of representations of
the de Sitter group. The perfect agreement herenote the fact that the most familiar kind of cosmological matter just exactly saturates the boundis rather remarkable, in that the Strominger
bound is required by a quantum condition on the boundary (unitarity of the field theory) while
the form of the potential (4) is dictated by the demand that the classical equation of state of the
bulk matter should be reproduced.4
Our discussion shows that the cosmological models with metrics given by Eq. (10) do define a
wellbehaved CFT at infinity. Since these models are both asymptotically de Sitter and (unlike de
Sitter spacetime itself) have large energy densities and pressures in the remote past, they are more
realistic than de Sitter spacetime, and they are the natural models to consider if one hopes to use
the dS/CFT correspondence to try to understand the distant past of the Universe. We conclude
with some observations regarding these spacetimes and the possible relevance of dS/CFT to the
problem of the initial singularity.
4. Comments on the initial singularity and dS/CFT
The recent work of Seery and Lidsey [7] is based on the idea that the dS/CFT correspondence
may be able to teach us something about the inflationary era. It is natural to extend this idea to ask
whether the correspondence may be brought to bear on the preinflationary era, about which very
little is known. The most important problem which arises here is that of the initial singularity, a
problem which is not solved by inflation [34]. In the case of AdS/CFT, good behaviour (such as
unitarity) on the part of the boundary field theory can be used to argue [17] for good behaviour
in the bulk (avoidance of information loss in black hole evaporation). Ultimately one hopes to
construct an analogous argument in the dS/CFT case.
Unfortunately, very little is understood about the field theory at infinity in this case, so we
cannot yet hope to replicate the progress that has been made in the AdS case. However, the
spacetimes we have been discussing do have some unusual properties which may suggest novel
approaches to the problem of the initial singularity.
First let us consider the field theory on the boundary. The key point to note here is that the
conformal boundary is geometrically very simple (it is flat) but topologically nontrivial: it is
either a torus or perhaps some nonsingular quotient of a torus [35,36]. The fundamental group
is infinite, that is, there are noncontractible curves which wind around the space arbitrarily many
times.5
Gauge theories on such spaces can have a number of very interesting properties, going far
beyond the need to impose periodic boundary conditions. For example, extending the work of
Krauss and Wilczek [23] on the theory of discrete gauge hair on black holes,6 Preskill and
Krauss [24] were led to study the extraordinary phenomena arising when gauge theories with
4 In this regard it may be of interest to examine the status of the Strominger bound when more complex equations of
state are considered; see for example [33].
5 The boundary can of course have a nontrivial fundamental group even in the locally spherical case, but this group is
always finite in that case.
6 See [37] for a comprehensive review of this subject.
98
discrete or disconnected gauge groups are constructed on spaces with nontrivial fundamental
groups. These occur very naturally when gauge groups are embedded in larger groups, as in
grand unified theories. Let us take a concrete example, though we stress that phenomena of the
kind we are about to describe can occur in very many other ways.
When the SU(3) of QCD is embedded in the Spin(10) grand unified group (usually known,
not quite accurately, as SO(10)), one finds that there are many disconnected subgroups of
Spin(10) with SU(3) as identity component. An interesting example of such a subgroup has the
following form: it may be expressed as a semidirect product of Z4 with SU(3):
SU(3) Z4 = SU(3) SU(3) (1) SU(3) ( ) SU(3),
(19)
where 1 is the element of Spin(10) which is factored out to obtain SO(10), and where is
a certain Spin(10) element with 2 = 1, such that conjugation by has the effect of complex conjugation on SU(3). Conjugation by also has the effect of complex conjugation on the
electromagnetic U(1) embedded in Spin(10); hence it is related to charge conjugation.
Such disconnected groups are of physical interest if one can prove that there is a gauge connection with a holonomy groupthe group generated by gauge parallel transport around closed
loopswhich contains an element representing . It is in fact possible to prove [38] that, with
topology T 3 , there do exist SU(3) Z4 gauge connections with holonomy groups isomorphic to
the Z4 generated by . This is analogous to parallel transport on a (flat) Klein bottle: despite the
fact that the metric is flat, parallel transport around certain noncontractible loops on the Klein
bottle can give rise to a transformation which reverses orientation. The (linear) holonomy group
of the Klein bottle is just Z2 , generated by an element of the orthogonal group O(2) which does
not lie in the connected component of the identity.
If we had a gauge connection on the torus at infinity with holonomy group Z4 generated by
as above, then moving an object around certain closed, noncontractible curves would cause
particles to transform to antiparticles, and vice versa: the system is nonorientable with respect
to charge conjugation instead of parity. The very remarkable properties of such gauge theories
are described in detail by Preskill and Krauss in [24], to which the reader is referred for further
details. Here we shall not need to consider a specific gauge group like SU(3) Z4 : all we need
is to understand that we should expect parallel transport around certain noncontractible loops
on the torus to implement discrete transformations, analogous to (but probably different from)
charge conjugation.
There are three specific points to be made here. First, with spatial topology T 3 , there is nothing
artificial about gauge connections with holonomy groups of this kind; indeed, such behaviour
should be regarded as generic when gauge theories are constructed on the torus, provided that
a disconnected gauge group arises. Second, in the cosmological context, one might require a
gauge field to have vanishing field strengths; but phenomena of the kind discussed by Preskill
and Krauss can still be present even in this case, in the manner of the AharonovBohm effect.7
Third, such gauge fields are in no sense pathological; that is, we have no reason whatever to
expect any kind of unphysical behaviour on the part of the boundary field theory, though the
details may be unfamiliarfor example, conservation of charge (or whatever conservation law
is relevant to the discrete symmetry in question) is enforced in a very unusual way, through
Cheshire charges [24]. In fact, nontrivial holonomies of this kind appear routinely in string
theory in the form of Wilson lines, which have been used recently in string cosmology [40].
7 See [39] for a recent discussion of such effects in cosmology.
99
We conclude that, if anything resembling the AdS/CFT correspondence is valid here, then it
should be possible to construct a model of bulk physics which remains nonpathological even
when the boundary theory fully incorporates effects due to the toral topology. As we shall see,
this leads us in a very interesting direction.
In order to give a more concrete basis for discussion, let us construct a very simple explicit
model of the preinflationary era. In fact, some of the spacetimes discussed above are ideally
suited to this. For all of them evolve naturally to an inflationary state, that is, they are all
asymptotically de Sitter to the future, and they all have a (single) torus as their conformal boundaries. On the other hand, some of them can be used to set up a semirealistic model of conditions
in the very early Universe, which may well have contained networks of extended objects of the
kind encountered in, for example, string gas cosmology [41]. Such networks correspond to matter fields with energy densities that decay relatively slowly; in fact, using Eq. (5), one can show
that the relevant range of is 2. We stress that it is the preinflationary era that is being
discussed here; the usual lore regarding the cosmology of extended objects is not relevant.
For example, take the case of a spacetime of topology R T 3 containing dark energy and a
static network of planar domain walls, described by (10) with = 1:
t 2
g(1, K, L) = dt 2 K 2 sinh4
(20)
d1 + d22 + d32 .
2L
This appears to be singular at t = 0, but, as we mentioned earlier, this could be deceptive. In
the metric given in (15), what is contracting to zero size is not a spatial section, but rather the
separation of neighbouring members of the congruence of timelike curves which defines the
given coordinates. The same might happen here. Physically, one can argue very strongly that the
metric in (20) should not be singular, as follows.
The metrics in (10) with > 2 are in fact singular; but this is just what we would expect. For
we saw (Eq. (13)) that in those cases the strong energy condition holds in the early Universe, and
a singularity is then demanded by the classical singularity theorems. In physical terms, gravity
is attractive when the SEC holds, and so the presence of a singularity is to be expected. (Since
the boundary theory is assumed to be wellbehaved, this presumably just means that such matter
fields are not relevant in the very early Universewhich is reasonable.) By contrast, the dark
energy corresponding to a de Sitter cosmological constant generates gravitational repulsion, and
this is how we understand the fact that pure de Sitter spacetime is not singular.
By this logic, it would not make sense for the spacetime with metric g(1, K, L) to be singular:
here, gravitation is always repulsive, just as it is in the case of the de Sitter spacetime (Eq. (3)) of
which it is a deformation. For the latter, the quantity + 3p, which (when negative) measures the
extent of violation of the strong energy condition, is given by the negative constant 3/4L2 ;
whereas here (from (13)) we have
1
t
3
2
cosech
.
1
+
(1, L, t) =
(21)
2
2
2L
4L
Evidently the function (1, L, t) is always negative, and in fact it is more negative than in the
case of de Sitter spacetime. As this quantity measures gravitational repulsion, there is in fact
even less reason for the spacetime to be singular here than in the de Sitter case. It is true that
(1, L, t) tends to negative infinity as t approaches zero, but this could be due to a singularity
of the congruence of timelike curves being used here, and not of the spacetime. We saw that this
accounted for the apparent singularity in the metric in Eq. (15). Our discussion leads us to expect
something similar here.
100
It is therefore very surprising to find that g(1, K, L) is in fact singular. The scalar curvature
of this metric is
12
9
t
2
R g(1, K, L) = 2 2 cosech
(22)
,
2L
L
L
so indeed we have a curvature singularity at t = 0; in fact, all of the metrics with 2 are singular. To appreciate fully what this means, take this metric and define it on the interval (, 0), so
that it describes a contracting cosmology. The force of gravitational repulsion increases without limit as the Universe contracts, and yet it manages to contract to zero size. This appears to be
selfcontradictory behaviour.
Ones first reaction to this extraordinary situation is to suggest that the singularity is a result
of the special boundary conditions or special symmetries of the metric, and that a more realistic
model would bounce, just as the spatially spherical version of de Sitter spacetime does, and
for the same reason. This is not correct, however: Andersson and Galloway [42,43]8 have proved
results which essentially imply that every future asymptotically de Sitter metric defined on a
manifold of topology R T 3 must be geodesically incomplete in the past if the null energy
condition and the Einstein equations hold at all times. Thus boundary conditions and symmetries
cannot be blamed for the fact that g(1, K, L) is singular.
The next suggestion is that the metrics with 2 might perhaps be ruled out by some internal contradiction in dS/CFT itself, and indeed it was this possibility that inspired the present
investigation. But we have seen that the relevant forms of matter define a wellbehaved unitary
field theory at infinity; matter with, for example, = 2 is precisely as far from violating unitarity
at infinity as ordinary radiation ( = 4), and indeed both are represented formally in our analysis
by fields in the same representation of the de Sitter group. Hence there is no hint that the spacetimes with 2 are unacceptable from a holographic point of view. Furthermore, it has been
shown [26] that these spacetimes are stable against known [44] perturbative and nonperturbative
instabilities in string theory.
The AnderssonGalloway results imply that it is the topology of conformal infinity that is to be
blamed for the singularities in these metrics, as long as the Einstein equations are assumed to
hold everywhere. But we have stressed that, while the topology of the boundary can be expected
to influence the boundary gauge theory in important ways, it certainly should not give rise to
any unphysical effects. We conclude that if anything like a dS/CFT correspondence is valid,
the bulk singularity in the metrics with 2 will be resolved, and this will happen because
the boundary topology will enforce a suitable modification of the Einstein equations.9 This is
certainly a reasonable prediction since, as we saw, it is not physically reasonable for the metrics
with 2 to be singular in the first place.
The boundary topology can have physical consequences in a variety of ways, but we wish to
propose that the strange effects discussed by Preskill and Krauss [24], which directly link the
boundary topology to gauge theory physics, are relevant here. How can these effects enforce a
nonsingular spacetime structure? Only a much more detailed understanding of the correspondence can tell us exactly how this is accomplished, but a further examination of the metrics in
(10) does reveal some striking hints.
8 See [26] for a discussion.
9 The only alternative is a violation of the null energy condition, which we consider to be implausible; see [26,45,46]
101
Let us assume that the field theory at infinity has a discrete or disconnected gauge group, so
that parallel transport around noncontractible loops can impose some discrete transformation.
For the sake of clarity we shall refer to this discrete transformation as charge conjugation, but
we stress that charge conjugation is just an example of the kind of transformation that can arise
in this way.
If a dS/CFT correspondence is valid, we can expect a global particle/antiparticle ambiguity
at infinity to be inherited by the spatial sections of the bulk. At the present time, and indeed by
the end of inflation, this would be physically irrelevant, since circumnavigations of the Universe
along noncontractible curves are of course by that time not possible. They may well, however,
have been possible in the preinflationary era, and this we now investigate.
There is in fact a major difference between the metrics of the form g(, K, L) with 2
(which we originally hoped to use to describe the preinflationary era) and those with > 2; this
t
may be seen as follows. As sinh(2/) ( 2L
) is approximately proportional to t (2/) when t is small,
t
it follows that the Lorentzian [angular] conformal time (defined by d = dt/K sinh(2/) ( 2L
))
(2/) t
converges as t tends to zero, provided that > 2. On the other hand, if t is large, sinh
( 2L )
resembles a multiple of et/L , and so always converges as t tends to infinity. Since the spatial
sections are always finite, it follows that, in the case > 2, our spacetime is conformal to a piece
of Minkowki space which is finite in both space and time.
The Penrose diagram for this case is shown in Fig. 1. Future conformal infinity is spacelike,
and there is a Big Bang singularity which is also spacelike. The horizontal direction represents
only half of the range of any angular coordinate on the torus, namely the half corresponding to
angles between zero and . Thus the vertical lines represent the origin and the point which in
this direction is most distant from the origin. (The reader may find it helpful to compare this with
the diagrams in [47].)
The width of the diagram is given by . Its height, in the > 2 case, is given by
2L
H (L/K, ) =
K
0
dx
sinh
(2/)
(x)
(23)
that is, it is proportional to L/K, and it is related to in a more complicated way. For example,
for = 3 (nonrelativistic matter), the height is approximately 2.81 L/K.
The shape of this diagram determines whether the Universe can be circumnavigated. For example, with the choice of parameters leading to the particular shape shown in Fig. 1, it is clear
that no particle or signal can circumnavigate the spatial torus, even in the infinite proper time
102
available. However, if we allow to descend towards 2, the integral in Eq. (23) becomes steadily
larger. For values just above 2, H (L/K, ) will be much larger than , even if L and K are of
similar magnitudes. In that case, the Penrose diagram will resemble Fig. 1, but it will be much
taller than it is wide. In this case, circumnavigation is easy; in fact, for very tall diagrams, a
generic worldline will wrap around the torus.10
Circumnavigations of the early Universe can be of interest for a variety of reasons. For example, they can help to preserve any initial homogeneity, a fact which is the basis of Lindes
model of lowscale inflation [48,49]. They are also of interest to us here, however, because we
are assuming that circumnavigations can convert particles to antiparticles (or cause some other
similar discrete transformation). Suppose that is just above 2, so that the Penrose diagram is
much taller than it is wide; then a generic timelike worldline, representing the history of (say)
a quark, will execute a definite, nonzero but finite number of circumnavigations as we trace it
back to the singularity, and so we can specify whether this particle emerged from the singularity
as a quark or as an antiquark. But now let us ask what happens when 2, as we wish to assume
here.
If 2, then conformal time still converges as t tends to infinity, but it diverges as t tends to
zero. In this case, the Penrose diagram has an unusual structure pictured in Fig. 2. The triangle
OAC represents the lower half of the Minkowski conformal diagram, and the line EFC represents
a typical spacelike hypersurface defined by a fixed value of t . The point F represents the twot
) in that spacelike surface. This twosphere can be enclosed in a
sphere of radius K sinh(2/) ( 2L
(2/) t
cube of side length 2K sinh
( 2L ), and this cube can be allowed to expand or contract as we
follow the sphere either into the future or the past along the geodesic OFB. We now perform the
10 Recall again that we are dealing with the situation before inflation; by the end of inflation, circumnavigations will no
longer be possible.
103
usual identifications of the faces of these cubes, to obtain tori. The effect on the Penrose diagram
is (to a good approximation) to cut away all parts of the original triangle which lie to the right of
the chosen geodesic OFB; that is, we cut away the dotted region in Fig. 2. Note that conformal
infinity is also compactified: it is represented by the solid part, AB, of the upper horizontal line,
and of course it is also a torus. The singularity is represented by a single point, O.
Now note that we do not expect holography to resolve the singularity in Fig. 1, at least not
directly. Instead, the boundary theory must in some way inform us that values of greater than
2 are not appropriate for describing the preinflationary era. For, as we discussed earlier, the
matter content of those spacetimes is such that gravity is attractive in the earliest era, and so a
singularity is natural. The singularity should be directly resolved only in those cases where the
unusual matter content of the early universe (dominated by extended objects such as strings) is
such as to lead us to expect that a singularity is not physically reasonable. That is, holography
should work in a way so that the singularity at O in Fig. 2, but not the one in Fig. 1, is revealed
to be unphysical. How can the boundary physics distinguish the singularities in Figs. 1 and 2?
Consider a timelike geodesic extending to the past from the point D. In the diagram it appears
to bounce off the line OFB, because it reenters the torus on the opposite side when it passes
beyond the most distant point from the origin; it then proceeds towards the line OA (the worldline of the origin), and again appears to bounce as it passes through the origin. It reaches the
singularity at O in a finite proper time, but it must circumnavigate the torus a literally infinite
number of times in order to do this. It is not entirely clear what this statement means, and it is
particularly confusing if we recall that circumnavigations can have the effect of some discrete
transformation such as charge conjugation, so that this transformation is being applied an infinite
number of times in a finite period of proper time. It follows, for example, that it simply does not
make sense to ask whether a given object, such as a quark, was initially a quark or an antiquark.
We find ourselves getting entangled in the traditional paradoxes associated with infinity.
This is the classical way of stating the case. In reality, quantum effects would render this
situation unphysical well before the singularity is reached: quantum uncertainties, combined
with the effects of parallel transport, would make it impossible to specify whether a given object
is a particle or an antiparticle for some interval of time after the singularity. The nature of the
system and the physics of its interactions cannot be properly formulated under such conditions.
These paradoxes simply indicate that we are dealing with a procedure that cannot actually be
realized: that is, the singularities in the metrics with 2 cannot really be attained by any actual
particle or antiparticle, and they should be regarded as a mathematical fiction. The question then
is: what takes the place of the region near O in Fig. 2?
The most straightforward assumption is that the paradoxes we have been discussing are
avoided because the spatial sections of the Universe are never in fact smaller than a certain
size; this is also compatible with the ideas of string gas cosmology [41]. That is, the Universe
must be born along some spacelike hypersurface large enough so that the status of a given object
is always welldefined modulo a finite number of discrete transformations.11
This is just the creation from nothing scenario [51,52], recently revived and improved in
various ways by Tye and coworkers [5355] and by Ooguri et al. [56]; see also [57] for similar
approaches. However, if we wish to avoid violating the null energy condition, it is difficult to
arrange for the Universe to be created smoothly along a threetorus; in fact, it can only be done
by modifying the Einstein equations [27]. A simple way of performing such a modification, using
11 Alternatively, the Universe could be asymptotic in the past to some quasistatic state, as in [50].
104
the ideas of Gabadadze and Shang [58,59], was explained in [26]; the resulting creation from
nothing on a torus metric, defined for t 0, is given by
2
2
2
(2/3) 3t
gCFNT (K, L) = dt K cosh
(24)
d1 + d22 + d32 .
L
It is completely nonsingular and smooth.
In this roundabout way, we find that the topology of the boundary, expressed by the effects
discovered by Krauss, Wilczek, and Preskill, does seem to have profound consequences for bulk
physics in the case depicted in Fig. 2 (and only in that case). In fact, the effects due to nontrivial
gauge holonomy become more pronounced as we penetrate deeper into the bulk, that is, farther
into the past and closer to the apparent initial singularity. By requiring that the bulk geometry
should account for the boundary gauge holonomy group in a reasonable way, we may be able
to resolve the singularity portrayed in Fig. 2. One might call this an example of topological
holography.
Obviously this is a mere sketch of a novel way in which dS/CFT might guide us towards a
resolution of the initial singularity. What this discussion really tells us is that as we work towards
a better understanding of de Sitter holography, we should be aware of the possibility that the
gauge group may be disconnected or discrete, since this may well have profound consequences
both for the boundary field theory and for the bulk.
5. Conclusion
We have seen that, once its limitations are understood, the dS/CFT correspondence may have
much to teach us about cosmology. We found that the apparently unreasonable Strominger bound
is seen to be perfectly reasonable when it is related to the equation of state of cosmological matter.
We then asked whether dS/CFT can have any bearing on the problem of the initial singularity in
cosmology: can a gauge theory at future infinity influence the physics of the deep bulk? While
we were not able to answer this question definitively, we have uncovered some hints suggesting
that the answer may be positive.
The specific examples of gauge theories and spacetimes we have discussed here are of course
oversimplified. They should be regarded as suggestions as to the lines along which one might
work when the dS/CFT correspondence is better understood. What this discussion does clearly
show is that the topology of the future conformal boundary has a strong influence both on the
gauge theory at infinity and on bulk physics. On one side, nontrivial topology leads to unusual
physics due to holonomy effects, while on the other it may well enforce a modification of the
Einstein equations, leading to a replacement of the singularity shown in Fig. 2 by creation from
nothing. Clarifying the connections between these two ideas may help us to understand both the
dS/CFT correspondence and the problem of the initial singularity.
Acknowledgements
The author is extremely grateful to Soon Wanmei for preparing the diagrams and for many
helpful discussions.
References
[1] G.T. Horowitz, J. Polchinski, Gauge/gravity duality, grqc/0602037.
105
[2] A. Strominger, The dS/CFT correspondence, JHEP 0110 (2001) 034, hepth/0106113;
A. Strominger, Inflation and the dS/CFT correspondence, JHEP 0111 (2001) 049, hepth/0110087.
[3] E. Witten, Quantum gravity in de Sitter space, hepth/0106109.
[4] J. Maldacena, NonGaussian features of primordial fluctuations in single field inflationary models, JHEP 0305
(2003) 013, astroph/0210603.
[5] F. Larsen, R. McNees, Inflation and de Sitter holography, JHEP 0307 (2003) 051, hepth/0307026;
F. Larsen, R. McNees, JHEP 0407 (2004) 062, hepth/0402050.
[6] J.P. van der Schaar, Inflationary perturbations from deformed CFT, JHEP 0401 (2004) 070, hepth/0307271.
[7] D. Seery, J.E. Lidsey, NonGaussian inflationary perturbations from the dS/CFT correspondence, astroph/0604209.
[8] X. Chen, M.X. Huang, S. Kachru, G. Shiu, Observational signatures and nonGaussianities of general single field
inflation, hepth/0605045.
[9] J. Polchinski, The cosmological constant and the string landscape, hepth/0603249.
[10] S.B. Giddings, The fate of four dimensions, Phys. Rev. D 68 (2003) 026006, hepth/0303031.
[11] S. Kachru, R. Kallosh, A. Linde, S.P. Trivedi, de Sitter vacua in string theory, Phys. Rev. D 68 (2003) 046005,
hepth/0301240.
[12] J. Maldacena, L. Maoz, Wormholes in AdS, JHEP 0402 (2004) 053, hepth/0401024.
[13] E. Witten, Antide Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hepth/9802150.
[14] B. McInnes, Exploring the similarities of the dS/CFT and AdS/CFT correspondences, Nucl. Phys. B 627 (2002)
311, hepth/0110062.
[15] M. Cadoni, P. Carta, Tachyons in de Sitter space and analytical continuation from dS/CFT to AdS/CFT, Int. J. Mod.
Phys. A 19 (2004) 4985, hepth/0211018.
[16] P. Breitenlohner, D.Z. Freedman, Positive energy in antide Sitter backgrounds and gauged extended supergravity,
Phys. Lett. B 115 (1982) 197.
[17] J.M. Maldacena, Eternal black holes in AdS, JHEP 0304 (2003) 021, hepth/0106112.
[18] L. Fidkowski, V. Hubeny, M. Kleban, S. Shenker, The black hole singularity in AdS/CFT, JHEP 0402 (2004) 014,
hepth/0306170.
[19] K. Maeda, M. Natsuume, T. Okamura, Extracting information behind the veil of horizon, hepth/0605224.
[20] G.L. Alberghi, D.A. Lowe, M. Trodden, Charged false vacuum bubbles and the AdS/CFT correspondence,
JHEP 9907 (1999) 020, hepth/9906047.
[21] B. Freivogel, V.E. Hubeny, A. Maloney, R.C. Myers, M. Rangamani, S. Shenker, Inflation in AdS/CFT, JHEP 0603
(2006) 007, hepth/0510046.
[22] D. Bak, Dual of bigbang and bigcrunch, hepth/0603080.
[23] L.M. Krauss, F. Wilczek, Discrete gauge symmetry in continuum theories, Phys. Rev. Lett. 62 (1989) 1221.
[24] J. Preskill, L.M. Krauss, Local discrete symmetry and quantum mechanical hair, Nucl. Phys. B 341 (1990) 50.
[25] B. McInnes, The most probable size of the universe, Nucl. Phys. B 730 (2005) 50, hepth/0509035.
[26] B. McInnes, Preinflationary spacetime in string cosmology, Nucl. Phys. B 748 (2006) 309, hepth/0511227.
[27] B. McInnes, The geometry of the entropic principle and the shape of the universe, hepth/0604150.
[28] A. Friedland, H. Murayama, M. Perelstein, Domain walls as dark energy, Phys. Rev. D 67 (2003) 043519, astroph/0205520.
[29] R.J. Adler, J.M. Overduin, The nearly flat universe, Gen. Relativ. Gravit. 37 (2005) 1491, grqc/0501061.
[30] E. Babichev, V. Dokuchaev, Yu. Eroshenko, Dark energy cosmology with generalized linear equation of state, Class.
Quantum Grav. 22 (2005) 143, astroph/0407190.
[31] N.J. Vilenkin, Special Functions and the Theory of Group Representations, Translations of Mathematical Monographs, vol. 22, American Mathematical Society, Providence, 1988.
[32] J.P. Gazeau, J. Renaud, M.V. Takook, Class. Quantum Grav. 17 (2000) 1415, grqc/9904023;
A.J. Tolley, N. Turok, hepth/0108119;
V. Balasubramanian, J. de Boer, D. Minic, Class. Quantum Grav. 19 (2002) 5655;
V. Balasubramanian, J. de Boer, D. Minic, Ann. Phys. 303 (2003) 59, hepth/0207245;
A. Guijosa, D.A. Lowe, Phys. Rev. D 69 (2004) 106008, hepth/0312282;
A. Dolgov, D.N. Pelliccia, Nucl. Phys. B 734 (2006) 208, hepth/0502197.
[33] S. Nojiri, S.D. Odintsov, The new form of the equation of state for dark energy fluid and accelerating universe,
hepth/0606025.
[34] A. Borde, A. Vilenkin, Phys. Rev. D 56 (1997) 717, grqc/9702019;
A. Borde, A.H. Guth, A. Vilenkin, Phys. Rev. Lett. 90 (2003) 151301, grqc/0110012.
[35] J.H. Conway, J.P. Rossetti, Describing the platycosms, math.DG/0311476.
[36] B. McInnes, Inflation, large branes, and the shape of space, Nucl. Phys. B 709 (2005) 213, hepth/0410115.
106
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
S.R. Coleman, J. Preskill, F. Wilczek, Quantum hair on black holes, Nucl. Phys. B 378 (1992) 175, hepth/9201059.
B. McInnes, Alice universes, Class. Quantum Grav. 14 (1997) 2527.
R. Jeannerot, Grand unified models and cosmology, hepph/0604270.
A. Avgoustidis, D. Cremades, F. Quevedo, Wilson line inflation, hepth/0606031.
T. Battefeld, S. Watson, String gas cosmology, Rev. Mod. Phys. 78 (2006) 435, hepth/0510022.
L. Andersson, G.J. Galloway, dS/CFT and spacetime topology, Adv. Theor. Math. Phys. 6 (2003) 307, hepth/0202161.
G.J. Galloway, Cosmological spacetimes with > 0, grqc/0407100.
M. Kleban, M. Porrati, R. Rabadan, Stability in asymptotically AdS spaces, JHEP 0508 (2005) 016, hepth/0409242.
P. Creminelli, M.A. Luty, A. Nicolis, L. Senatore, Starting the universe: Stable violation of the null energy condition
and nonstandard cosmologies, hepth/0606090.
R.V. Buniy, S.D.H. Hsu, B.M. Murray, The null energy condition and instability, hepth/0606091.
T. Chiba, R. Takahashi, N. Sugiyama, Classifying the future of universes with dark energy, Class. Quantum Grav. 22
(2005) 3745, astroph/0501661.
A. Linde, Creation of a compact topologically nontrivial inflationary universe, JCAP 0410 (2004) 004, hepth/0408164.
A. Linde, Inflation and string cosmology, eConf C040802 (2004) L024;
A. Linde, J. Phys. Conf. Ser. 24 (2005) 151, hepth/0503195.
Y. Nakayama, S.J. Rey, Y. Sugawara, The nothing at the beginning of the universe made precise, hepth/0606127.
A. Vilenkin, Creation of universes from nothing, Phys. Lett. B 117 (1982) 25.
A. Vilenkin, The birth of inflationary universes, Phys. Rev. D 27 (1983) 2848.
H. Firouzjahi, S. Sarangi, S.H.H. Tye, Spontaneous creation of inflationary universes and the cosmic landscape,
JHEP 0409 (2004) 060, hepth/0406107.
S. Sarangi, S.H.H. Tye, The boundedness of Euclidean gravity and the wavefunction of the universe, hepth/0505104.
S. Sarangi, S.H.H. Tye, A note on the quantum creation of universes, hepth/0603237.
H. Ooguri, C. Vafa, E. Verlinde, HartleHawking wavefunction for flux compactifications: The entropic principle,
Lett. Math. Phys. 74 (2005) 311, hepth/0502211.
A. Kobakhidze, L. MersiniHoughton, hepth/0410213;
L. MersiniHoughton, Class. Quantum Grav. 22 (2005) 3481, hepth/0504026;
Q.G. Huang, hepth/0510219;
R. Brustein, S.P. de Alwis, Phys. Rev. D 73 (2006) 046009, hepth/0511093;
R. Holman, L. MersiniHoughton, hepth/0511112;
L. MersiniHoughton, hepth/0512304;
K.S. Aoyanagi, K.I. Maeda, JCAP 0603 (2006) 012, hepth/0602149;
A.O. Barvinsky, A.Yu. Kamenshchik, hepth/0605132;
A. Gorsky, hepth/0606072.
G. Gabadadze, Y. Shang, Classically constrained gauge fields and gravity, hepth/0506040.
G. Gabadadze, Y. Shang, Quantum cosmology of classically constrained gravity, Phys. Lett. B 635 (2006) 235,
hepth/0511137.
Received 31 May 2006; received in revised form 3 July 2006; accepted 12 July 2006
Available online 7 August 2006
Abstract
We consider QCD instantoninduced contributions to lepton pair production in hadronhadron collisions.
We relate these contributions to those known from deep inelastic scattering and demonstrate that they can
be calculated reliably for sufficiently large momentum transfer. We observe that the instanton contribution
to the angular distribution of the lepton pairs at finite momentum transfer strongly violates the LamTung
relationa relation between coefficient functions of the angular distribution which is valid within the framework of ordinary perturbation theory. The drastic violation of this relation, as seen in experimental data,
might be related to such instantoninduced effects.
2006 Elsevier B.V. All rights reserved.
1. Introduction
The Standard Model of electroweak (quantum flavor dynamics (QFD)) and strong (QCD)
interactions is extraordinarily successful. This success is largely based on the possibility to apply ordinary perturbation theory to the calculation of hard, shortdistance dominated scattering
processes, since the relevant gauge couplings are small. Certain processes, however, cannot be
described by ordinary perturbation theory, no matter how small the gauge coupling is. These
processes are associated with axial anomalies [1] and manifest themselves as anomalous violation of baryon plus lepton number (B + L) in QFD and chirality (Q5 ) in QCD [2]. They are
induced by topological fluctuations of the nonAbelian gauge fields, notably by instantons [3].
* Corresponding author.
05503213/$ see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2006.07.020
108
2 This is often called the DrellYan process [19]. Instanton contributions to this process have been first discussed in
Ref. [20] at a qualitative level.
109
pair production at large transverse momentum may be viewed as a distinguished process for
studying manifestations of QCD instantons.
We put special emphasis on the angular distribution of the lepton pairs at finite momentum
transfer. We observe that the instanton contribution strongly violates the LamTung relation [21]
between coefficient functions of the angular distribution, which has been verified within the
framework of ordinary perturbation theorythe QCD improved parton modelup to O(s2 )
[22,23] and even holds for the inclusion of parton transverse momentum and soft gluon effects
[24,25]. Indeed, it has been argued that the drastic violation of this relation, as seen in experimental data [2628], might be due to a nontrivial structure of the QCD vacuum [22], and in
particular could be related to instantoninduced effects [29].
The outline of this paper is as follows: in Section 2 we introduce the instantoninduced contribution to lepton pair production. Afterwards we review the known results in the related process
in DIS in an instanton background. The crucial part of this section is the continuation of these
results to hadron collisions which leads us to the photon production tensor on partonic level. In
Section 3 we will use these results to calculate the angular distribution of the produced leptons
on the partonic (Section 3.1) and the hadronic (Section 3.2) level. In Section 3.3 we give an outlook on the inclusion of multi gluon processes which lead to an enhancement of the instanton
contributions. We present our conclusions in Section 4.
2. From deep inelastic scattering to lepton pair production
We start with the derivation of the instantoninduced contribution to lepton pair production on
the parton level. We will concentrate on the case with quarks in the initial state. These contributions dominate over the ones involving initial state gluons, at least for scattering processes where
valencelike quarks and antiquarks contribute, e.g., in pp or N collisions. This is certainly
different at very high energies where very small parton momentum fractions x dominate. Since
the lower bound on x is set by M 2 /S, where M 2 is the invariant mass squared of the lepton pair
and S is the hadronhadron center of mass energy squared, the contributing values of x considered in our study are not so small for our chosen values of M 2 and S, see Section 3.2. For the
main case of phenomenological interest, i.e., nf = 3 light flavours (mq 1, for q = u, d, s),
instantoninduced quarkantiquark annihilation involves in the final state at least two quarks and
two antiquarks of different flavour, such that the chirality is violated by 2nf = 6 [2], plus an
arbitrary number of gluons (g), e.g., (cf. Fig. 2)
uL + uL + dR + dR + sR + sR + ng g
+ + .
(1)
110
Fig. 3. Instantoninduced process for nf = ng = 1, (q) + g(p) qR (k1 ) + qR (k2 ), in leading semiclassical approximation (left) and the analogous process, (q) + g(p) qR (k1 ) + qL (k2 ), from ordinary perturbation theory (right).
Both figures from Ref. [12].
(2)
have been derived, in leadingorder semiclassical approximation, in Ref. [12]. For clarity and
simplicity, let us concentrate here on the explicit result for the simplest appropriate3 case nf =
ng = 1 (cf. Fig. 3 (left)),
a
T
(q) + g(p) qR (k1 ) + qR (k2 )
13/2
2
2
b+1
b+3
a 2 3
exp
d
2b
= ieq
8
s (r )
s (r )
2
2
1
r
k
(q k)
+
(q k)2 2q k
2r
(q k)2
b+1
2
2 b+1
2
k
r
,
2
2q k q
(4)
and confront it with its chirality conserving counterpart from ordinary perturbation theory (cf.
Fig. 3 (right)),
a
T
(q) + g(p) qR (k1 ) + qL (k2 )
a
(q k1 )
(q k2 )
= eq gs L (k2 )
(5)
L (k1 ).
2
(q k1 )2
(q k2 )2
Here, eq is the quark charge in units of the electric charge e, gs is the strong coupling, a ,
a = 1, . . . , 8, are the GellMann SU(3) generators, and and are the fourvector indices of
the photon and gluon, respectively (cf. Fig. 3). The twocomponent Weylspinors L,R in Eqs. (3)
and (5) satisfy the Weylequations, kL (k) = 0, kR (k) = 0, and the relations L (k)L (k) = k,
R (k)R (k) = k. We used the abbreviations, k k , k k , for any fourvector k , with
the familiar matrices, = (1, ) and = (1,
), with being the Pauli matrices.
3 Note that the even simpler case n = 0 is not relevant for us, since the corresponding process in lepton pair production
g
would contribute only at vanishingly small transverse momentum where, anyhow, instanton perturbation theory is not
applicable (see below).
111
The other parameters in Eq. (3) arose from (the integration over) the instanton size distribution
[2,30], whose twoloop renormalization group improved form [31],
6
s (r )
d
2
2
exp
D(, r ) = 5
(6)
(r )0 + 4 (1 120 ) ,
s (r )
s (r )
has been exploited, where r is the renormalization scale and
2
gs2 (r )
4
1 ln ln r2
s (r )
=
2 1 2
4
0 ln 2r2
0 ln r2
(7)
C1 3C2 +nf C3
,
e
2
(9)
with C1 = 0.466, C2 = 1.51, and C3 = 0.292, in the MSscheme [3234]. The variable b in
Eq. (3) is a shorthand for the effective power of r in the instanton size distribution (6),
s (r )
(1 120 ).
(10)
4
It is important to note that the perturbative expression (6) for the size distribution is valid for
small 1 where is the fundamental scale in QCD. Indeed, a comparison with lattice
data from quenched (nf = 0) QCD [35] yields 0.4 for the fiducial region of instanton
perturbation theory [36]. This can be translated into a fiducial kinematical region for instanton
perturbation theory in deep inelastic scattering. On account of the fact that the main contribution
to the integration over the instanton size comes from [12]
b 0 +
b + 3/2
b + 3/2
b + 3/2
,
,
,
2
2
q
(q k1 )
(q k2 )2
(11)
corresponding to different
terms in Eqs. (3) and (4), one has to require that all virtualities, q 2 ,
(q k1 )2 , and (q k2 )2 , exceed Qmin (4 6) GeV, in order to stay in the realm of
instanton perturbation theory.
It is now straightforward to obtain the corresponding amplitudes relevant for lepton pair
production via quarkantiquark annihilation, namely the one for the chirality violating instantoninduced process qL (k1 ) + qL (k2 ) (q) + g(p) (cf. Fig. 4) and the one for the analogous
ordinary perturbative process qL (k1 ) + qR (k2 ) (q) + g(p). In fact, these processes are basically T conjugates of the deep inelastic processes from Eqs. (3) and (5), and the respective
Fig. 4. Instantoninduced process for nf = ng = 1, qL (k1 ) + qL (k2 ) (q) + g(p), in leading semiclassical approximation.
112
n=0
(n)
(k1 , k2 ; q) =
w
dPS(n) =
1
4
(n)
w
(k1 , k2 ; q),
(12)
n
d4 pj (+) 2
pj (2)4 (4) (k1 + k2 q p1 pn ).
(2)3
(13)
(14)
j =1
K k + K k
2
w3 + k k w4 ,
(15)
i = 1, 2, 3, 4,
(16)
from the partonic tensor with the help of the following projectors,
0
p
= g ,
4
1
=
k1 k2 q k1 k2 q ,
p
stu
2s
0
q
2
1
+ k k ,
=
2p
k 2 p
p
tu
2s
k K + K k
0
2q
3
1
+
=
2p
(k K ) p
,
p
tu
2
q 2 s 2 0
4
1
+ K K ,
=
K p 2p
p
tu
(17)
(18)
(19)
(20)
(21)
(1)
i
w
(22)
=
wi = p
w ,
spins,colours T (k1 , k2 ; q, p)T (k1 , k2 ; q, p),
4
d p (+) 2
1
p (2)4 (4) (k1 + k2 q p)wi
wi(1) =
4
(2)3
1
= (+) (k1 + k2 q)2 wi .
(23)
2
4 Averaging over colour and spin of the initial state is implicitly understood in Eq. (12); the index n is to label besides
the final state partons also their spin and colour degrees of freedom.
113
Along these lines, we find that the simple instantoninduced process qL + qL + g contributes as follows to the functions wi ,
2 b+1
M 2 b+1
M
2tu
=
+
+
g
t
u
(t M 2 )(u M 2 )
2 b+1 2 b+1
2
2
M
M
+1
t
u
2 b+1 2 b+1
2
2
b+1
M
M
2
,
Re(1)
(24)
+
t
u
2 b+1 2 b+1 2
2
2
M
(I)
M
(I)
w1 =
(25)
+
,
2
t
u
2 b+1
2 b+1
M
M
sM 2
(I)
sM 2
tu(t u)2
(I)
w2 =
+
2
(t M 2 )2 t
(u M 2 )2 u
(t M 2 )2 (u M 2 )2
b+1
b+1
b+1
sM 2 (4M 2 s + t 2 + u2 ) M 2 2 M 2 2
(t u) Re(1) 2
+
u
tu(t M 2 )(u M 2 ) t
(t M 2 )2 (u M 2 )2
2 b+1
2
M
t M 2 2M 4 M 2 t tu 3M 2 u + u2
u
2 b+1
2
M
,
u M 2 2M 4 M 2 u tu 3M 2 t + t 2
(26)
t
2 b+1
2 b+1
sM 2
M
M
sM 2
(s + M 2 )(t u)tu
(I)
(I)
w3 =
(t M 2 )2 t
(u M 2 )2 u
(t M 2 )2 (u M 2 )2
b+1
b+1
b+1
sM 2 (s + M 2 )(t u) M 2 2 M 2 2
Re(1) 2
+
u
(t M 2 )(u M 2 )tu t
(t M 2 )2 (u M 2 )2
b+1
6
M2 2
2
4
4
2 2
2 2
2
3
t M 2M 2M t 4M u + M t + 3M u + t u u
u
6
2
4
4
2 2
u M 2M 2M u 4M t + M u
b+1
M2 2
2 2
2
3
,
+ 3M t + tu t
(27)
t
2 b+1
2 b+1
M
M
(I)
sM 2
tu(s + M 2 )2
sM 2
(I)
w4 =
2
(t M 2 )2 t
(u M 2 )2 u
(t M 2 )2 (u M 2 )2
2 b+1 2 b+1
b+1
2
2
M
M
sM 2 (t 2 + u2 )
(s + M 2 ) Re(1) 2
+
u
tu(t M 2 )(u M 2 ) t
(t M 2 )2 (u M 2 )2
b+1
M2 2
t M 2 M 2 u + M 2 t + tu + u2
u
2 b+1
2
M
,
+ u M 2 M 2 t + M 2 u + tu + t 2
(28)
t
(I)
w
(I)
114
where
13
2 b
s
4
r
2
exp
,
2
s (r )
s (r )
M
M2
1
b+1
b+3
N 2 d2b
.
2
2
2
(I)
2 eq2 N 2
(29)
(30)
We have obtained these results starting from Eq. (3), contracting it with the gluon polarization
vector (p) and taking the modulus squared, exploiting FORM [37] for the spinor traces. The
results (24)(28) for the contribution of our simple instantoninduced process to the partonic
tensor for inclusive quarkantiquark annihilation into a timelike photon look quite similar to
the contribution of the analogous simple instantoninduced process to the deep inelastic structure tensor of a gluon found in Ref. [12]. A notable difference is the appearance of the factor
Re(1)(b+1)/2 , which reduces to unity in the spacelike kinematics of deep inelastic scattering
and was therefore not visible in the results of Ref. [12]. On the other hand, the full instanton
contribution to the deep inelastic structure tensor of a gluon can be obtained from Eqs. (24)
(28) by replacing the combinatorial factor 1/((2 Nc )2 ) by 1/(2 (Nc2 1)) and by substituting
Re(1)(b+1)/2 by 1.
As a check, let us also quote the corresponding perturbative contributions to the inclusive
partonic tensor arising from Eq. (5),
t
2M 2 s
u
(pt)
g w = (pt)
(31)
+ +
,
t
u
tu
1
(pt)
(pt)
w1 = g w ,
(32)
2
sM 2
(pt)
w2 = (pt)
(33)
,
tu
(pt)
w3
= 0,
(34)
(pt)
w4
(pt)
= w2 ,
(35)
where
(pt) 2eq2 s .
(36)
We obtained this wellknown result, see, e.g., [42], by exploiting the same FORM routines as the
ones for the instantoninduced contribution, except for replacing the input amplitude (3) by the
perturbative amplitude (5).
3. The angular distribution of the lepton pairs
In this section, we will concentrate on the angular distribution of lepton pairs in instantoninduced processesmainly concentrating on the simple one from the previous sectionand
compare it to the one predicted from ordinary perturbation theory.
In general, the angular distribution of the charged lepton + in lepton pair production,
h1 (K1 ) + h2 (K2 ) (q) + X
+ (q+ ) + (q )
(37)
115
is described by three functions, , , and , which may depend on the kinematic variables of (37),
1 d
3 1
(38)
=
1 + cos2 + sin 2 cos + sin2 cos 2 ,
d
4 + 3
2
and being the polar and azimuthal angles of + , respectively [38]. These coefficient functions
may be conveniently expressed in terms of hadronic helicity structure functions [39],
W T WL
(39)
,
W T + WL
W
,
=
(40)
W T + WL
2W
.
=
(41)
WT + WL
Here, we exploit the socalled CollinsSoper frame [38], in which the frame dependent helicity
structure functions read, in terms of the hadronic counterparts Wi of the previously introduced
invariant functions wi ,
=
r2
(q K)2 W2 (q K)(q k)W3 + (q k)2 W4 ,
2
2
2SM (1 + r )
1
(q k)2 W2 (q K)(q k)W3 + (q K)2 W4 ,
WL = W1 +
2
2
SM (1 + r )
r
(q K)2 + (q k)2
W =
W3 ,
(q K)(q k)(W2 + W4 ) +
2
SM 2 (1 + r 2 )
r2
W =
(q K)2 W2 (q K)(q k)W3 + (q k)2 W4 ,
2
2
2SM (1 + r )
where
WT = W1 +
(42)
(43)
(44)
(45)
2
q
T U + M 2 (M 2 S T U )
=
,
(46)
2
M
SM 2
determines the transverse photon momentum q with respect to the hadronic reaction plane. The
kinematic variables S = (K1 + K2 )2 , T = (q K1 )2 and U = (q K2 )2 refer to the hadron level
(cf. Eq. (37)). Similarly, K = K1 + K2 and k = K1 K2 .
r2
(pt)(1)
(pt)(1)
wL
wT
(pt)(1)
(pt)(1)
+ wL
wT
2 r2
,
2 + 3r 2
(47)
1
q 2s(s + M 2 )M 2 (t u)
,
2
2
2
2
2
M (t M ) + (u M ) 2M s + 3tu
(48)
(pt)(1)
(pt)(1)
w
(pt)(1)
wT
(pt)(1)
+ wL
116
(pt)(1)
2w
(pt)(1)
wT
(pt)(1)
+ wL
2r 2
.
2 + 3r 2
(49)
1 (pt)(1) 2 (pt)(1) = 0,
(50)
= 0, as long as no intrinsic
which, of course, holds also on the hadron level,
transverse momentum for the initial state quarks is invoked. It is nearly left intact even if one
includes O(s2 ) corrections [23].
The contributions arising from our simple instantoninduced process (cf. Eqs. (24)(28)) are
readily calculated along the same lines. They yield quite lengthy expressions, and we do not quote
them all analytically, but will illustrate them, instead, graphically. We stress, however, that the
LamTung relation is violated by instantons. This is apparent from the following, nonvanishing
expression,
2 b+1 2 b+1
2
2
(I)
b+1
4 (I) tu
M
M
(I)
2
+
1
Re(1)
2 wL 2w =
t
u
(t M 2 )(u M 2 )
b+1
b+1
sM 2 M 2 2 M 2 2
(51)
,
tu
t
u
1 (pt)
2 (pt)
Note, that the factor (M 2 / tu)b+1 in the asymmetry (51) arises from a nonplanar diagram. That
is in accordance with [29] where the importance of nonplanar interference terms for the violation
of the LamTung relation were discussed.
It is useful to view the partonic coefficient functions, for fixed M, as a function of r = q /M
and the partonic Feynman variable
1 2 =
xF
t u
.
s
(53)
In fact, , , and depend, for fixed M, only on r and xF . Their dependence on these kinematical
variables is illustrated in Figs. 5 and 6. We observe the following features:
(i) (top panels) approaches 1 for large r for pure instantoninduced processes (dashed),
i.e., these processes tend to be purely longitudinal (cf. Eq. (39)) for large transverse momenta,
in contrast to ordinary perturbative processes (dotted and Eq. (47)). The total result for (solid),
taking into account both instanton and ordinary processes in the numerator and denominator
of the partonic equivalent of Eq. (39), shows little deviation from ordinary perturbation theory.
Indeed, there are experimental hints for longitudinally polarized photons in hadron collisions
towards larger xF [28]. Note that, even if the instantoninduced process q q g is suppressed,
for larger xF , instanton effects might be relevant for this effect since gluon resummation leads to
an enhancement in this kinematic region, see Section 3.3.
(ii) The total result (solid) for (second panels from top) shows a quite significant deviation
from ordinary perturbation theory (dotted) for sizeable xF and intermediate values of r.
117
Fig. 5. Coefficient functions of the angular distribution of lepton pairs from quark antiquark annihilation, as a function
of q /M, for various values of xF , for M = 7 GeV. Dotted: result from pure ordinary perturbation theory; dashed:
result from pure instanton perturbation theory; solid: total result from ordinary and instanton perturbation theory. Also
shown, in the last row, is the ratio of the crosssections of instanton perturbation theory to ordinary perturbation theory
(dasheddotted). In the numerical results shown we have chosen r = M, = 0.346 GeV, and nf = 3.
(iii) (third panels from top) behaves quite differently in pure instantoninduced processes
(dashed) and ordinary perturbative processes (dotted and Eq. (49)). In fact, instantoninduced
processes have a value of5 (I) 2 at small, but finite r and small xF , much larger as ordinary
perturbative processes ( (pt) 1). Also in the total result for (solid) we observe a strong enhancement at small, but finite r and small xF in comparison to ordinary perturbation theory.
Correspondingly, we find a strong violation of the LamTung relation, which we display in the
forth panels from top in terms of the parameter
1
(1 2).
4
(54)
5 In fact, it follows from general arguments [22,29] that as long as instanton processes dominate over ordinary perturbative processes, one expects (I) 2.
118
Whereas this parameter is identical zero in ordinary perturbative processes (dotted), it is about
one, at small r, xF , for instantoninduced processes, leading to a drastic violation of the Lam
Tung relation in the total result (solid).
(iv) Clearly, instanton effects in the coefficient functions are most visible in kinematical regions where the instanton induced crosssection (I) dominates over the perturbative, (pt) , one.
The instantoninduced features in Figs. 5 and 6 are indeed located where the ratio (bottom panels;
dasheddotted)
(I)
(I)
2wT + wL
(I)
=
(pt)
(pt)
(pt)
2wT + wL
(55)
becomes large. Obviously, it gets large towards small momentum transfer. The dominance of
instantons is seen to set in like a brick wall. This sudden onset occurs practically
at the
or
t
Q
boundary
of
the
fiducial
kinematical
region
of
instanton
perturbation
theory,
min
u Qmin , for M > Qmin (46) GeV. Therefore, the instanton features in the coefficient
functions at very small momentum transfer, to the left of the sudden onset of instanton dominance
in Figs. 5 and 6, lie strictly speaking outside the range of validity of the semiclassical approximation. Fortunately, however, the coefficient functions are ratios of helicity structure functions (cf.
119
Fig. 7. Illustration of the dependence of our prediction of the LamTung parameter = 14 (1 2) on the choice of nf
and , for fixed M = 10 GeV and various values of xF . The thicker lines correspond to our default choice nf = 3 and
(3)
(1)
Eqs. (39)(41)), and therefore the main uncertainties coming from the extrapolation of the perturbative expression of the instantonsize distribution cancel in them. Therefore, it is expected that
our predictions of the coefficient functions remain also valid at smallish, but nonzero r. For very
small q = Mr, namely up to around 1 GeV, the simplest perturbative and instantoninduced
subprocess q q contributes and may change the angular distributions.
(v) One feature of (pt) and (pt) that translates unchanged even to the hadron level is the
scaling behavior: they depend only on the ratio r. This is not the case when instanton effects are
included since they vanish in comparison to the perturbative contributions for larger M 2 . This
is basically triggered by the ratio (I) / (pt) which leads to a M 2 dependent weighting of the
perturbative and instanton contribution. In addition, also (I) and (I) depend already slightly
on M 2 .
(vi) Remarkable is also the behavior for xF = 0. For vanishing q , one expects to recover
the wellknown leadingorder angular distribution (1 + cos2 ), that is = 1 and , = 0.
As one can see from Fig. 5 and 6, the function tends towards 2. Note that still vanishes
in the limit q 0 for very small but finite xF . For large xF , the violation of the LamTung
relation is suppressed even in a region where the ratio (I) / (pt) is not small. The very strong xF
dependence will disappear after folding with the parton distributions as we will see in the next
section.
Strictly speaking, we should take nf = 1 and a corresponding =
(I) ,
(1)
MS
120
unrealistic case of) one massless flavour, the other flavours being integrated out. In the numerical
(3)
results shown in Figs. 5 and 6 we have chosen, instead, nf = 3, and = = 0.346 GeV,
MS
for the calculation of the effective coupling parameters (pt) , (I) , and b. This value of
correspondsaccording to the standard threeloop perturbative flavour reductionto an nf = 5
(5)
value = 0.219 GeV, leading to a running QCD coupling s (mZ ) = 0.119 at the Zboson
MS
mass [14]. As illustrated in Fig. 7, our results for the coefficient functions are not largely affected
if we choose instead the nominal value nf = 1 and a corresponding value for the parameter, (1) = 0.241 GeV. This value was obtained by a linear interpolation between the central
MS
MS
(0)
MS
= 0.245 GeV [41]. Again, such details cancel to a great extend in the ratios of structure
functions. This also refers to the dependence on the renormalization scale r , for which we have
chosen M in the numerical results presented in Figs. 5 and 6.
3.2. Hadron level
Since we have to deal with collisions of hadrons, the partonic Mandelstam variables s, t, u
in Eqs. (24)(28) are not observable. Firstly, we have to calculate the tensor (12) on the hadron
level which involves a folding with the usual parton distributions, see, e.g., Ref. [42],
W (S, T , U )
dx1 dx2
qi
16
=
(56)
S
w (s, t, u) qi (x1 )qi (x2 ) + qi (x1 )qi (x2 )
3
x1 x2
i
dx1 dx2 s + t + u M 2
qi
8
=
6 Constant factors are for our purposes actually not important since we are only interested in ratios of functions W .
i
121
For fixed M, the angular distributions on the partoniclevel depend only on r and xF , but
they are independent of the centerofmass (c.m.) energy s, because the latter is fixed due to
the relation s + t + u = M 2 . On the hadron level, however, the momentum fractions x1 and x2
are variable,
and the angular distributions depend, correspondingly, for fixed M, on the hadronic
c.m. energy S. Furthermore, the variable xF has to be replaced by the hadronic one,
T U
(59)
,
S
which can be interpreted as the longitudinal photonmomentum fraction with respect to the momentum of the hadron h1 .
Fig. 8 shows the resulting angular distributions for protonproton collisions at S = 15 GeV
for M = 7 GeV, XF = 0, 0.1, 0.3, and varying values of r = q /M. For the renormalization
scale we have chosen r = M and for the parton distributions the CTEQ6 dataset [43].7
The main difference to the partonic quantities, shown in Fig. 5, concerns the xF respectively XF dependence. The strong xF dependence is smeared out on the hadron level. This
smearing leads in particular to a suppression of the instantoninduced effect at small XF . This has
nothing to do with the parton distribution functions entering the hadronic angular distribution.
Actually, since the angular distributions are ratios of two foldings with parton distribution functions, their dependence on the type of hadrons in the initial states is rather weak. The difference
is just a consequence of the x1,2 dependence of the partonic Feynman variable xF that leads to
the smeared out XF behavior after integrating over x1,2 . For similar reasons, the M 2 dependence
of the angular distribution in the instanton background is stronger than on the parton level, since
for smaller ratios M 2 /s also smaller values of xF contribute, see Figs. 8 and 9; but note that due
2
to kinematical reasons only smaller values of r = q /M are accessible for larger
M /s.
ratios
The fiducial region of instanton perturbation theory on the parton level t, u, M >
Qmin (46) GeV can be mapped on the hadronic variables. One can check that theses relations are fulfilled for all x1 and x2 for large enough values of M and q , namely M Qmin and
q Qmin . Also on the hadron level our results should hold for even smaller q , since the uncertainty towards smaller q drops out in the ratios of the angular distributions, see the discussion
in Section 3.1.
XF =
7 Actually a consistent treatment of instanton induced effects requires also parton distributions including instantoninduced parton evolution. Since this modification enters the perturbative and instanton contribution in the same way this
effect would change the angular distribution only in subleading order of instanton perturbation theory.
122
Fig. 8. The plot shows the angular structure functions similar to Fig. 5 but on hadron level. Therefore an integration
over parton distributions, e.g., for the proton, is included. Due
to the variable momentum fractions x1 and x2 one has to
specify an additional kinetic variable, e.g., the c.m. energy S.
A complete calculation of the angular distribution for this general instantoninduced process
is beyond the scope of this paper and will be attempted in the future. Let us roughly sketch the
general features of the complete process. Whereas in perturbative processes additional finalstate
gluons are certainly suppressed by an order of s , every additional gluon in an instanton background leads to an enhancement of the order 1/s . Summing over all processes with an arbitrary
number of gluons ng leads to an exponentiation of the inverse coupling constant. The resulting
factor, combined with the tunneling factor, exp[4/s ] (cf., e.g., Eq. (29)), can be written as
exp[4/s F (x )]. Here, the Bjorken scaling variable x = Q2 /(Q2 + MX2 ) appears, where Q
is the relevant momentum transfer and MX is the invariant mass of the produced partonic final
state. The socalled holygrail function F (x ) [8] is normalized to one for x = 1 and decreases
towards smaller x and therefore larger MX . Let us mention that in the electroweak theory, where
the coupling constant is much smaller, this mechanism is absolutely necessary for the process
eventually becoming observable in the high energy limit [7].
For the process discussed in the present paper, Q2 is given by the partonic quantities t, u
or M 2 , whereas, in general, MX2 = (k1 + k2 q)2 . Therefore, the integrands of the functions
123
Wi (57) involve a factor exp[4/s F (x )]. For MX2 = 0 (x = 1), and therefore also for the
process with one finalstate gluon, the factor exp[4/s ] in Eq. (29) is recovered. For positive
XF , the smallest x is given for Q2 = t ,
2M 2 /S + x1 ( 4M 2 /S(1 + r 2 ) + XF2 XF )
t
t
=
=
. (60)
x =
s + u M2
t + MX2
2x2 x1 x2 ( 4M 2 /S(1 + r 2 ) + XF2 + XF )
It is easy to check that x rises slightly with r and decreases towards the largest accessible values
of XF . Therefore, we can conclude that the instantoninduced effect in and (see Figs. 5, 6,
8 and 9) will shift to slightly smaller r. Furthermore, we expect a significant enhancement of
the instanton effect for larger XF . Correspondingly, the suppression of the simplest instantoninduced process at large XF , which we observed before, might be compensated. Note that the
applicability of instanton perturbationtheory now requires in addition a cut x xcut , where xcut
is approximately 0.35, see Ref. [36]. One can check that this requirement can be fulfilled for all
x1 , x2 and r as long as XF is not too large, or for all XF for large enough ratios M 2 /S.
Beside the discussed instantoninduced multigluon process also other perturbative processes
may contribute. Firstly, we have not taken into account an enhancement of perturbative contribu
124
tions due to soft gluon resummation [24,25] at small q since the instantoninduced contribution
that we have calculated is not reliable in this region anyway. In higher order s also new processes
contribute to the angular distribution which lead to a small violation of the LamTung relation
already in the purely perturbative framework [23]. For small transverse momenta q the usual
factorization is not reliable anymore and transverse parton momentum distributions become important. However, as already mentioned in the introduction, higher order contributions, soft gluon
effects and parton transverse momentum are not able to explain the observed strong violation of
the LamTung relation [2325].
4. Conclusions
We have calculated the angular distribution of the produced leptons in hadronhadron collision in an instanton background. It turns out that, for large enough photon virtualities M 2
2 , only small instantons contribute. Therefore, the
and transverse photon momenta squared q
instantoninduced contribution is fiducially calculable in this kinematic region using techniques
of instanton perturbation theory. The most remarkable property of the resulting angular distribution is the violation of the LamTung relation which is conserved to very high accuracy in
usual perturbation theory, but violated in experiments. This effect is a direct consequence of chirality violation in the background of QCD instantons which leads to a nontrivial spindensity of
the quarkantiquark pair in the initial state as it has been argued in [29]. Therefore, lepton pair
production in hadron collisions is potentially a very good testing ground for instantoninduced
processes: the violation of the LamTung relation is reliably calculable in instanton perturbation
theory and absent in usual perturbation theory.
I
of the more realistic general processes q q + (nf 1)q q + ng g was beyond the scope
of this paper, we cannot compare our results directly with the available data. However, the small
violation of the LamTung relation on the hadron level arising from the simplest partonic process
is already quite promising, notably in view of the expectation that additional gluons lead to a
substantial enhancement of the instantoninduced effect, as known from analyses of the related
processes in deepinelastic scattering and from the general arguments presented in this paper.
Finally, let us mention that, beside further theoretical efforts, more experimental data are required for testing instantons in the angular distribution of produced leptons at a hadron collider.
Fortunately, there are new medium energy projects under way that are also dedicated to study the
DrellYan process, e.g., at the forthcoming facilities GSIFAIR [44] and JPARC [45].8 Experiments at RHIC may also give further information on lepton pair production. In general, it seems
that fixed target experiments are especially well suited for our purposes, since on the one hand the
involved momenta are smaller and on the other hand the luminosities are larger. Therefore, a huge
amount of lepton pairs should be observable which is absolutely necessary for reconstructing a
whole angular distribution.
S 10 GeV) will be used for fixed target experiments and at the FAIR
experiment
a
29
GeV
antiproton
beam
will
be
available
for fixed target experiments or collisions with low energy protons
( S 615 GeV). Clearly, protonantiproton collisions are perfectly suited for studying DrellYan since the rate is
higher as in protonproton collisions.
125
Acknowledgements
One of us (A.R.) would like to thank Sven Moch and Fridger Schrempp for sharing their insights into QCD instantoninduced hard scattering processes with him. A.U. would like to thank
Fridger Schrempp for countless illuminating discussions about instantons and beyond. A.B. and
A.U. would like to thank also Otto Nachtmann and Danil Boer for numerous discussions about
lepton pair production in hadron collisions as a testing ground for perturbative QCD. We thank
also Markus Diehl for valuable information on future experiments on DrellYan production.
The work of A.B. was partially supported by a Heisenberg grant of the Deutsche Forschungsgemeinschaft. The work of A.U. was supported by the research program of the Stichting voor
Fundamenteel Onderzoek der Materie (FOM), which is financially supported by the Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO).
References
[1] S.L. Adler, Phys. Rev. 177 (1969) 2426;
J.S. Bell, R. Jackiw, Nuovo Cimento A 60 (1969) 47;
W.A. Bardeen, Phys. Rev. 184 (1969) 1848.
[2] G. t Hooft, Phys. Rev. Lett. 37 (1976) 8;
G. t Hooft, Phys. Rev. D 14 (1976) 3432;
G. t Hooft, Phys. Rev. D 18 (1978) 2199, Erratum.
[3] A. Belavin, A. Polyakov, A. Shvarts, Y. Tyupkin, Phys. Lett. B 59 (1975) 85.
[4] E.V. Shuryak, Nucl. Phys. B 203 (1982) 93;
D. Diakonov, V.Y. Petrov, Phys. Lett. B 147 (1984) 351;
D. Diakonov, V.Y. Petrov, Nucl. Phys. B 272 (1986) 457;
T. Schfer, E.V. Shuryak, Rev. Mod. Phys. 70 (1998) 323;
D. Diakonov, Prog. Part. Nucl. Phys. 51 (2003) 173.
[5] F.R. Klinkhamer, N.S. Manton, Phys. Rev. D 30 (1984) 2212;
V.A. Kuzmin, V.A. Rubakov, M.E. Shaposhnikov, Phys. Lett. B 155 (1985) 36;
P. Arnold, L.D. McLerran, Phys. Rev. D 36 (1987) 581;
A. Ringwald, Phys. Lett. B 201 (1988) 510.
[6] V.A. Rubakov, M.E. Shaposhnikov, Usp. Fiz. Nauk 166 (1996) 493, Phys. Usp. 39 (1996) 461.
[7] H. Aoyama, H. Goldberg, Phys. Lett. B 188 (1987) 506;
A. Ringwald, Nucl. Phys. B 330 (1990) 1;
O. Espinosa, Nucl. Phys. B 343 (1990) 310;
V.V. Khoze, A. Ringwald, Phys. Lett. B 259 (1991) 106.
[8] L.D. McLerran, A.I. Vainshtein, M.B. Voloshin, Phys. Rev. D 42 (1990) 171;
J.M. Cornwall, Phys. Lett. B 243 (1990) 271;
P.B. Arnold, M.P. Mattis, Phys. Rev. D 42 (1990) 1738;
S.Y. Khlebnikov, V.A. Rubakov, P.G. Tinyakov, Nucl. Phys. B 350 (1991) 441;
A.H. Mueller, Nucl. Phys. B 348 (1991) 310;
A.H. Mueller, Nucl. Phys. B 353 (1991) 44;
V.V. Khoze, A. Ringwald, Nucl. Phys. B 355 (1991) 351.
[9] G.R. Farrar, R.B. Meng, Phys. Rev. Lett. 65 (1990) 3377;
A. Ringwald, F. Schrempp, C. Wetterich, Nucl. Phys. B 365 (1991) 3;
M.J. Gibbs, A. Ringwald, B.R. Webber, J.T. Zadrozny, Z. Phys. C 66 (1995) 285;
D.A. Morris, A. Ringwald, Astropart. Phys. 2 (1994) 43;
Z. Fodor, S.D. Katz, A. Ringwald, H. Tu, Phys. Lett. B 561 (2003) 191;
T. Han, D. Hooper, Phys. Lett. B 582 (2004) 21;
M. Ahlers, A. Ringwald, H. Tu, Astropart. Phys. 24 (2006) 438.
[10] A. Ringwald, Phys. Lett. B 555 (2003) 227;
F. Bezrukov, D. Levkov, C. Rebbi, V.A. Rubakov, P. Tinyakov, Phys. Rev. D 68 (2003) 036005;
A. Ringwald, JHEP 0310 (2003) 008.
126
[11] I.I. Balitsky, V.M. Braun, Phys. Lett. B 314 (1993) 237;
A. Ringwald, F. Schrempp, in: Quarks 94, Vladimir, Russia, 1994, hepph/9411217.
[12] S. Moch, A. Ringwald, F. Schrempp, Nucl. Phys. B 507 (1997) 134.
[13] A. Ringwald, F. Schrempp, Phys. Lett. B 438 (1998) 217.
[14] A. Ringwald, F. Schrempp, Comput. Phys. Commun. 132 (2000) 267.
[15] H1 Collaboration, C. Adloff, et al., Eur. Phys. J. C 25 (2002) 495;
ZEUS Collaboration, S. Chekanov, et al., Eur. Phys. J. C 34 (2004) 255.
[16] D.E. Kharzeev, Y.V. Kovchegov, E. Levin, Nucl. Phys. A 690 (2001) 621;
M.A. Nowak, E.V. Shuryak, I. Zahed, Phys. Rev. D 64 (2001) 034008;
F. Schrempp, J. Phys. G 28 (2002) 915;
F. Schrempp, A. Utermann, Acta Phys. Pol. B 33 (2002) 3633;
F. Schrempp, A. Utermann, Phys. Lett. B 543 (2002) 197;
F. Schrempp, A. Utermann, hepph/0301177;
F. Schrempp, A. Utermann, hepph/0401137;
F. Schrempp, A. Utermann, hepph/0407146.
[17] F. Schrempp, in: A. De Roeck, H. Jung (Eds.), Proc. HERA and the LHCA Workshop on the Implications of
HERA for LHC Physics, CERN2005014, p. 1, hepph/0407146.
[18] M. Petermann, F. Schrempp, in preparation;
M. Petermann, PhD thesis, in preparation.
[19] S.D. Drell, T.M. Yan, Phys. Rev. Lett. 25 (1970) 316;
S.D. Drell, T.M. Yan, Phys. Rev. Lett. 25 (1970) 902, Erratum.
[20] J.R. Ellis, M.K. Gaillard, W.J. Zakrzewski, Phys. Lett. B 81 (1979) 224.
[21] C.S. Lam, W.K. Tung, Phys. Rev. D 21 (1980) 2712.
[22] A. Brandenburg, O. Nachtmann, E. Mirkes, Z. Phys. C 60 (1993) 697.
[23] E. Mirkes, J. Ohnemus, Phys. Rev. D 51 (1995) 4891.
[24] P. Chiappetta, M. Le Bellac, Z. Phys. C 32 (1986) 521.
[25] D. Boer, W. Vogelsang, hepph/0604177.
[26] NA10 Collaboration, S. Falciano, et al., Z. Phys. C 31 (1986) 513.
[27] NA10 Collaboration, M. Guanziroli, et al., Z. Phys. C 37 (1988) 545.
[28] J.S. Conway, et al., Phys. Rev. D 39 (1989) 92.
[29] D. Boer, A. Brandenburg, O. Nachtmann, A. Utermann, Eur. Phys. J. C 40 (2005) 55.
[30] C.W. Bernard, Phys. Rev. D 19 (1979) 3013.
[31] T.R. Morris, D.A. Ross, C.T. Sachrajda, Nucl. Phys. B 255 (1985) 115.
[32] A. Hasenfratz, P. Hasenfratz, Nucl. Phys. B 193 (1981) 210.
[33] M. Lscher, Nucl. Phys. B 205 (1982) 483.
[34] G. t Hooft, Phys. Rep. 142 (1986) 357.
[35] UKQCD Collaboration, D.A. Smith, M.J. Teper, Phys. Rev. D 58 (1998) 014505.
[36] A. Ringwald, F. Schrempp, Phys. Lett. B 459 (1999) 249;
A. Ringwald, F. Schrempp, Phys. Lett. B 503 (2001) 331.
[37] J.A.M. Vermaseren, mathph/0010025.
[38] J.C. Collins, D.E. Soper, Phys. Rev. D 16 (1977) 2219.
[39] C.S. Lam, W.K. Tung, Phys. Rev. D 18 (1978) 2447.
[40] ALPHA Collaboration, S. Capitani, M. Lscher, R. Sommer, H. Wittig, Nucl. Phys. B 544 (1999) 669.
[41] ALPHA Collaboration, M. Della Morte, R. Frezzotti, J. Heitger, J. Rolf, R. Sommer, U. Wolff, Nucl. Phys. B 713
(2005) 378.
[42] C.S. Lam, W.K. Tung, Phys. Lett. B 80 (1979) 228.
[43] J. Pumplin, D.R. Stump, J. Huston, H.L. Lai, P. Nadolsky, W.K. Tung, JHEP 0207 (2002) 012.
[44] PAX Collaboration, Antiprotonproton scattering experiments with polarization, Letter of Intent, Darmstadt, 2004,
available at http://www.fzjuelich.de/ikp/pax.
[45] More information about the Japan Proton Accelerator Research Complex (JPARC) are available at http://jparc.jp/NuclPart/index_e.html.
Received 18 April 2006; received in revised form 26 June 2006; accepted 12 July 2006
Available online 11 August 2006
Abstract
We calculate the quantum corrections to the mass of the zero mode of the fifth component of the gauge
field at twoloop level in a fivedimensional massless QED compactified on S 1 . We discuss in detail how
the divergences are exactly canceled and the mass becomes finite. The key ingredients to obtain the result
are the shift symmetry and the WardTakahashi identity. We also evaluate the finite part of corrections.
2006 Elsevier B.V. All rights reserved.
1. Introduction
GaugeHiggs unification [1] is considered to be one of the attractive frameworks since it provides a solution to the gauge hierarchy problem without supersymmetry [25]. In this scenario,
the Higgs filed is identified with extra components of the gauge field in higherdimensional gauge
theories. A remarkable feature in the scenario is that quantum corrections to the Higgs mass become finite and are independent of the cuttoff scale of the theory thanks to the gauge invariance
in the higher dimensions nevertheless we consider nonrenormalizable theories. The Higgs mass
is generated through the dynamics of the Wilson line for an extra component of the gauge field.
Noting that the dynamics is nonlocal, we find no counter term in the Lagrangian, which is assumed to be local, to cancel the divergence if the Higgs mass diverges. This implies that the
* Corresponding author.
128
Higgs mass should be finite under quantum corrections at all order of the perturbations. (See
Ref. [6] for attempts to prove the finiteness.) Actually, its finiteness at oneloop level was discussed by several authors [2]. (In gravitygaugeHiggs unification, the finiteness is guaranteed
by the general coordinate invariance, see [7].)
Although the concept for the finiteness of the Higgs mass is very clear, there are subtleties
if we consider higher loop corrections to the Higgs mass beyond oneloop level. For instance,
generally there appear divergences in the subdiagrams even if we consider the gaugeHiggs unification scenario. These divergences should be subtracted by adding the counter terms determined
by the lower loop calculations. After such a subtraction, the Higgs mass becomes finite at any
order of perturbations without any additional counter terms. This means that the Higgs mass can
be predicted even within nonrenormalizable theories. In fact, a Higgs mass at twoloop level are
calculated in a fivedimensional (5D) supersymmetric theory [8], where the linear divergences
appear in the oneloop subdiagrams and are subtracted by adding oneloop counter terms.
It is also very important to calculate the Higgs mass beyond oneloop level from the phenomenological viewpoint. It is known that the physical Higgs mass and the KaluzaKlein (KK) mass
tend to be too small in the scenario. To get a large KK mass, or in other words to get a small
vacuum expectation value (VEV) of the Higgs fields compared to the KK mass, we rely on a
mild tuning to cancel the Higgs mass corrections among oneloop contributions [4]. A large KK
mass helps to enhance the physical Higgs mass. However, if the KK mass is taken so large, twoloop contributions can be important. Thus, we cannot make the KK mass larger than O(10 TeV)
reliably if we do not know the twoloop corrections. In this case, the physical Higgs mass cannot
exceed the present bound [9] if the low energy effective theory is just the standard model [10].
On the other hand, if we control the twoloop corrections, the KK mass can be enlarged up to
the scale where threeloop contributions become important, say O(100 TeV). Then, the physical
Higgs mass can pass the experimental test without additional low energy degrees.
As far as we know, there seems no calculation of the Higgs mass beyond oneloop order in the
context of gaugeHiggs unification. Therefore, it is worthwhile to check explicitly the finiteness
of the Higgs mass for higher order loop corrections. In this paper, we explicitly calculate the
twoloop quantum corrections to the mass of the zero mode of the fifth component of the gauge
field in a 5D massless QED compactified on S 1 . As expected from the general argument of the
renormalization theory, the mass is shown to be finite. A key ingredient to show the finiteness
is the shift symmetry and WardTakahashi identity. Although there appear linearly divergent
vertex corrections and the wave function renormalizations in subdiagrams, these divergences are
exactly canceled as expected from WardTakahashi identity. In this simple model, there is no
need to take into account counter terms. We will discuss in detail the structure of cancellation of
the divergences and also evaluate the finite part of the corrections.
This paper is organized as follows. In Section 2, we introduce our setup and derive Feynman
rules. Section 3 is the main part of this paper. Before calculating the twoloop corrections, we
calculate the oneloop wave function renormalization and the vertex corrections to observe that
these contributions are linearly divergent and have the same magnitude but an opposite sign.
Then, the twoloop corrections to the mass of the zero mode of the fifth component of the gauge
field are shown and the structure of canceling divergences is clarified. The details of this calculation and a physical interpretation are described in Appendix A. The last section is devoted to
summarize this paper.
129
y
1 (n)
A x exp 2in
,
A x , y =
L
L n=
y
1 (n)
A5 x exp 2in
,
A5 x , y =
L
L n=
(2.4)
(2.5)
(n)
and L = 2R is the circumference of the S 1 , it is written as
where A(n)
M = AM
2
1
+ 1 1 A(n) 2 + M 2 A(n) 2
SG = d 4 x
A(n)
2
n=
2 M 2 A(n) 2 ,
+ A(n)
n
5
5
(2.6)
p p
Mn2
2
p Mn2
mn
.
Mn2
p p
1
,
Mn2 p 2 Mn2
p2
y
1 (n)
x , y =
x exp i2n
,
L
L n=
y
1 (n)
x exp i2n
,
x ,y =
L
L n=
(2.7)
(2.8)
(2.9)
(2.10)
130
(a)
(b)
(a)
(b)
(c)
Fig. 2. Feynman diagrams relevant for the fermion. (a), (b) and (c) are the fermion propagator, the gauge interaction
vertex and the vertex of fermionfermionA5 , respectively.
(l)
(n) ,
+ Mn 5 ) +
m,l+n g4A
/ (l)
ig
A
S m = d4 x
(m) inm (/
4 5 5
m,n
(2.11)
where the 4D gauge coupling constant g4 is defined as g4 = g/ L. This leads to the following
Feynman rule (see Fig. 2):
mn
p
/ + iMn 5
= mn 2
,
p
/ + iMn 5
p Mn2
(b) = g4 m,l+n ,
(a) =
(2.12)
(2.13)
(2.14)
3. Loop calculations
3.1. Oneloop
Before calculating twoloop corrections, we clarify the nature of divergences at oneloop level
since the divergences appearing in the subdiagrams of twoloop diagrams have to be subtracted
by adding the counter terms generally. The possible relevant counter terms at this order are
those of the fermion propagator, the gaugefermionfermion vertex and the gauge propagator.
The first one corresponds to that for fermion wave function renormalization. The second one
corresponds to that for the gauge interaction vertex correction. The last one should correspond to
the renormalization of the gauge coupling.
3.1.1. Fermion wave function renormalization
The wave function renormalization of the fermion is calculated as
k/ + iMm 5
d4 k 2
g
Fig. 3 =
4
2
2
i(2)4 m
k 2 Mm
(l k)2 Mnm
k/ + iMm 5
1
+ (i)5 2
(i)5
2
2
k Mm
(l k)2 Mnm
g2
d4 k
3(/k + im5 )
,
= 4
4
2
2
R
i(2) m (k m )((l k)2 (n m)2 )
(3.1)
131
Fig. 3. Wave function renormalization of the fermion. The corresponding 4D and KK momenta are denoted as (k, m) for
example.
where we normalized all the dimensionful parameters by 1/R in the last equation so that all the
parameters become dimensionless. The gauge parameter is taken to be = 1.
By using the Feynman integral
1
dx
0
1
b + (a b)x
2
=
1
,
ab
(3.2)
d4 k
3(/k + im5 )
dx
4
2
i(2) m
((k xl) (m xn)2 + (x x 2 )(l 2 n2 ))2
1
g2
= 4
R
d4 k
3(x/l + im5 )
dx 2
.
i(2)4 m
(k (m xn)2 + (x x 2 )(l 2 n2 ))2
1
(3.3)
1
1
f (m) dz
f (z) = dz 1 +
f (z),
(3.4)
1 exp(2iz)
exp(2iz) 1
m
C0
C0
where C0 is a contour that encircle the real axis clockwise. If Im z exp(2 Im z)f (z) vanishes
at  Im z and f (z) has no poles on the real axis but has poles {mi+ } in the upper half plane
j
and poles {m } in the lower half plane, the contour integral can be expressed by the summation
of the residues at each pole and integration on the real axis:
2if (z)
2if (z)
j
i
Res
Res
; z = m+ +
; z = m
exp(2iz) 1
1 exp(2iz)
i
+
(3.5)
dz f (z).
Note that if f (z) is a real function, each mi+ has a counter part of mi = mi+ , which means that
(3.5) can be reduced to
2 Re
i
Res
2if (z)
; z = mi+
exp(2iz) 1
+
dz f (z).
(3.6)
132
An important point is that the residues always contain the exponential suppression exp(2
Im mi+ ) for a large Im mi+ , leading to finite contributions. Thus, as far as we concern the divergent
contributions, it is enough to evaluate the integration on the real axis in (3.6). In other words,
we can replace the summation with respect to m by the integration on the real axis. Then, the
correction (3.3) is written as
g42
R
d4 k
i(2)4
g2
= 4
R
1
dzm
dx
(k 2
d4 k
i(2)4
dzm
3(x/l + izm 5 )
(zm xn)2 + (x x 2 )(l 2 n2 ))2
1
dx
(k 2
3(x/l + ixn5 )
.
+ (x x 2 )(l 2 n2 ))2
2
zm
This shows that the divergent parts of the wave function renormalization and the mass renormalization (times R) for the fermion mode with (l, n) are commonly given by
Wf =
g42
= g42
g42
d4 k
i(2)4
dkE
8 2
dzm
1
dx
0
3x
2 + (x x 2 )(l 2 n2 ))2
(k 2 zm
2
3kE
2 + l 2 + n2
4kE
E
2
dkE
3
(kE ),
+
O
k
E
4
8 2
(3.7)
d4 k
2
(i)
Fig. 4 = g4
i(2)4 m
k 2 m2
k 2 m2 k 2 (n m)2
k/ + im5
k/ + im5
1
+ (i)5 2
(i)5 2
(i)5 2
k m2
k m2
k (n m)2
4
2
2
d k
3(k + m )
,
= g42
(i)5 2
4
i(2) m
(k m2 )2 (k 2 (n m)2 )
Fig. 4. Vertex correction. The corresponding 4D and KK momenta are denoted as (k, m) for example.
1 In the next subsection, we use another notation.
(3.8)
133
2!(1 x)
1
= 2 ,
3
((1 x)a + xb))
a b
(3.9)
g42
g42
3
dkE kE
8 2
dzm
1
dx
0
2
2
dkE 3kE (4kE n2 )
2 + n2 )2
8 2 (4kE
2 + z 2 + (x x 2 )n2 )3
(kE
m
2
dkE 3
(kE ).
+
O
k
E
8 2 4
(3.10)
We find that it is linearly divergent and is the same as the minus of that of Wf , as expected
from WardTakahashi identity. This fact is very important to cancel divergences appearing in the
subdiagrams, as will be seen in the next subsection.
3.1.3. Gauge self energy
In this subsubsection, we calculate the wave function renormalizations of A5 and A . If we
denote them as Z5 and Z , respectively, these can be expressed at oneloop level symbolically,
Z5 = 1 + g42 ( + c) + O g44 ,
(3.11)
4
2
Z = 1 + g4 ( + c ) + O g4 ,
(3.12)
where is a cutoff scale of the theory. These factors are linearly divergent. c and c mean
the physical renormalization factors after subtracted the divergence. Taking into account these
renormalizations, the physical Higgs mass at twoloop level includes
g42 2
g4
mH @1loop + 4 m2H @2loop + O g46
Z5
Z5
Z 2 2
=
gR mH @1loop + gR4 m2H @2loop + O gR6
Z5
= 1 + gR2 (c c ) gR2 m2H @1loop + gR4 m2H @2loop + O gR6 ,
m2phys@2loop =
(3.13)
where the renormalized gauge coupling gR is defined as g42 = gR2 Z . m2H @1loop is a oneloop
finite mass of the zero mode of A5 arising from the diagram in Fig. 5(a) with zero external momentum. Here we define such that m2H @1loop does not include the gauge coupling. m2H @2loop
is a twoloop mass which we will evaluate in the next subsection. Note that the ultraviolet
(UV) divergences appearing in (3.11) and (3.12) are guaranteed to be the same by the fivedimensional Lorentz invariance. Below, we show it concretely. In addition, we will obtain, apart
from m2H @2loop , a finite mass of the zero mode of A5 at twoloop level which is proportional
to m2H @1loop and to the difference of the finite part, c c . Thus, we would like to evaluate
also the finite parts of Z and Z5 and m2H @1loop , not only the divergent part. However, note
134
(a)
(b)
Fig. 5. Oneloop renormalizations for twopoint function of A5 (a) and the photon (b).
that this contribution should be discriminated from m2H @2loop . This is because this contribution
does not modify essentially the structure of the oneloop effective potential which is written in
terms of cos(qgA5 ), reflecting the phase structure of the Wilson line, where q is a constant. In
other words, the effect merely scales the effective potential
in the horizontal direction and it is
understood by replacing gR in the potential by gRH = gR Z /Z5 .
The wave function renormalizations of A5 and A are shown in Fig. 5(a) and (b), and are
calculated as
d4 k
(/k + iMm 5 )
(/k /l + iMmn 5 )
(a) = (1)
tr (ig4 5 )
(ig4 5 )
2
2
(2)4 i m
k 2 Mm
(k l)2 Mmn
4g 2
= 24
R
2 x(x 1)l 2 + m2
d 4 kE
kE
E
dx
2 + m2 + (x x 2 )l 2 )2
(2)4 m
(kE
E
1
(3.14)
and
(b) = (1)
4g 2
= 24
R
d4 k
(/k + iMm 5 )
(/k /l + iMmn 5 )
tr (g4 )
(g4 )
2
2
(2)4 i m
k 2 Mm
(k l)2 Mmn
d 4 kE
N
dx 2
,
4
2
(2) m
(kE + m + (x x 2 )lE2 )2
1
(3.15)
respectively, where
2
N = 2kE
kE
+ 2x(1 x)lE lE + g kE
+ m2 x(1 x)lE2 .
Here, we performed Wick rotation, omitted the terms that vanish after the angular integration of
and put n = 0 since we are interested in the wave function of the zero modes. In the following,
kE
we consider only the term proportional to lE lE and set lE2 = 0 to evaluate Z .
Let us show the divergent parts of (3.14) and (3.15), which are evaluated by replacing the
summation to the integral as before, are the same. Carrying out the integration, we find
dzm
0
dzm
0
2 x(x 1)l 2 + z2
kE
m
E
2 + z2 + x(1 x)l 2 )2
(kE
m
E
2x(1 x)
x(1 x)
=
.
2
3
2
2
(kE + m )
kE
x(1 x)lE2
(3.16)
(3.17)
135
From (3.16), we can see that this part does not contribute m2H @1loop , and the contribution to the
wave function renormalization is
x(1 x)
(3.18)
3
kE
which, including the finite part, is exactly same as (3.17). Note that the integration over zm
corresponds to the calculation in the case where the fifth momentum is continuous, i.e., the limit
R . In this decompactification limit, the 5D Lorentz symmetry, which is softly broken by
the compactification, recovers. Therefore, the cancellation among these contribution is natural.
Next, we evaluate the residue parts which are free from UV divergences. As for the Z5 , we
get
2 x(x 1)l 2 + z2
kE
2i
m
E
2 + x(1 x)l 2
;
z
=
i
k
2 Re Res
E
E
2 + z2 + x(1 x)l 2 )2 m
exp(2izm ) 1 (kE
m
E
2x(1 x)lE2
=
2 +x(1x)l 2
2 kE
2
2
3/2
E 1)2
(kE + x(1 x)lE ) (e
+
2 +x(1x)l 2
2 kE
2
E
4kE e
2
2
2 + x(1 x)l 2 )(e2 kE +x(1x)lE
(kE
E
(3.19)
.
1)2
4g42
R2
3g42
d 4 kE
2 2
(3).
=
))
(2)4 (1 + cosh(2kE
4 4 R 2
(3.20)
The wave function renormalization comes from the lE2 term, therefore we obtain by differentiating (3.19) with respect to lE2 and setting lE2 = 0,
4g42
1
d 4 kE
(2)4
dx ()
0
4x(1 x)e2kE
2 (e2kE 1)2
kE
2x(1 x)
3 (e2kE 1)
kE
+
8 2 x(1 x)e4kE
4 2 x(1 x)e2kE
(e2kE 1)2
kE
(e2kE 1)3
kE
(3.21)
The overall factor 1/R 2 disappears on the dimensional grounds in the differentiation. The contribution to Z is calculated as
2i
2x(1 x)
2 Re Res
;
z
=
ik
E
2 + z2 )2 m
exp(2izm ) 1 (kE
m
= 4g42
d 4 kE
(2)4
1
0
dx ()
2x(1 x)
3 (e2kE 1)
kE
4x(1 x)e2kE
2 (e2kE 1)2
kE
.
Note that these terms have the same for as the first term and the third term in (3.21).
(3.22)
136
Fig. 6. Twoloop diagrams for the mass of the zero mode of A5 . The corresponding 4D and KK momenta are denoted as
(k, m) for example.
Z
Z5
= 1 4gR2
finite
+O
gR4
d 4 kE
(2)4
1
0
2
4 x(1 x)e2kE
8 2 x(1 x)e4kE
dx ()
+
(e2kE
(e2kE
kE
1)2
kE
1)3
(3.23)
which is in fact free from UV divergences but contains infrared (IR) divergences. This is because
we consider exactly massless charged fermion for simplicity. However, we usually consider the
case where A5 which is identified as the Higgs field get nonvanishing VEV in the gaugeHiggs
unification scenario. Then, the charged fermions acquires nonvanishing mass, and the IR divergences disappear. Thus, we recalculate Z /Z5 under the nontrivial background, A5 = a/(gR),
leading to
4
)(cos2 (2a) + cos(2a) cosh(2k ) 2)
d kE 3 sinh(2kE
Z
E
= 1 4gR2
(cosh(2k ) cos(2a))3
4
Z5 finite
(2)
3kE
E
+ O gR4
=1
gR2
ln(2a) + O gR4 ,
12
(3.24)
in the limit a 0.
3.2. Twoloop
In this subsection, we calculate twoloop corrections to the mass of the zero mode of A5 . In 5D
massless QED, all the divergences at oneloop level are expected to cancel out. In fact, we have
seen explicitly in the previous subsection that the divergences from the wave function renormalization and the vertex correction are exactly canceled as expected from WardTakahashi identity.
Hence, we calculate twoloop diagrams without any counter terms. Straightforward calculation
of Fig. 6 is given by
4 4
g44
/l + in5 k/ + im5
k/ + im5 /l + in5
d ld k
(1)
tr
(i)
(i)
5 2
2
5 2
2
R4
(2)8 n,m
l n2
k m2
k m2
l n2
(l k)2 (n m)2
/l + in5 k/ + im5 k/ + im5 /l + in5
1
+ tr 5 2
5 2
5 2
5 2
2
2
2
2
2
l n
k m
k m
l n (l k) (n m)2
/l + in5 k/ + im5 /l + in5
/l + in5
+ 2 tr (i)5 2
2
2
(i)5 2
l n2
k m2
l n2
l n2
2
(l k) (n m)2
/l + in5 k/ + im5 /l + in5 /l + in5
1
+ 2 tr 5 2
5 2
5 2
5 2
l n2
k m2
l n2
l n2 (l k)2 (n m)2
4
4
4
g
d lE d k E
= 12 42
R
(2)8
2 + m2 ) 4nml k )
((lE2 n2 )(kE
E E
2
2
2 )2 (k + m2 )2 ((l k )2 + (n m)2 )
(l
+
n
E
E
E
E
n,m
2
2
2
((lE n )(kE lE + nm) 2n kE lE + 2nmlE2 )
+2 2
,
2 + m2 )((l k )2 + (n m)2 )
(lE + n2 )3 (kE
E
E
137
(3.25)
where we note that the contributions from the last two diagrams in Fig. 6 are the same as those
from the third and the fourth diagrams. In the last equation, we carry out the Wick rotation.
Now we perform the summations with respect to n and m. For this purpose, we replace the
summations by the integrations on the real axis and the summations of residues, as was done in
the oneloop calculation. In other words, we decompose the summations of the function f (m, n)
to the following four parts:
I:
dzn dzm f (zm , zn ),
2if (zm , zn )
i
Res
; zm = m+ ,
II:
dzn 2 Re
exp(2izm ) 1
i
2i
i
III: 2 Re
Res
dzm f (zm , zn ); zn = n+ ,
exp(2izn ) 1
i
2i
IV: 2 Re
Res
exp(2izn ) 1
i
2if (zm , zn )
i
i
2 Re
Res
; zm = m+ ; zn = n+ .
exp(2izm ) 1
i
Note that we always carry out the operation of m before doing that of n.
Here we list only the results of calculation of each part to clarify the cancellation of divergences. Detailed calculations and its physical interpretations are described in Appendix A.
I:
dzn dzm f (zm , zn ) = 0,
(3.26)
2if (zm , zn )
Res
; zm = mi+
II:
dzn 2 Re
exp(2izm ) 1
=
i
3
4 (1 + e
k)
ek2 l((l + )2 k 2 )
4 3 (1 + ek )
(k ),
ek2 k((k )2 l 2 )
(3.27)
138
III:
2 Re
i
2i
Res
exp(2izn ) 1
dzm f (zm , zn ); zn = ni+
4 3 (1 + el )
4 3 (1 + ek+ )
2
,
+ )2 l 2 ) ek+ k((k + )2 l 2 )
2i
Res
2 Re
exp(2izn ) 1
i
2if (zm , zn )
2 Re
Res
; zm = mi+ ; zn = ni+
exp(2izm ) 1
=
IV:
el2 k((k
(3.28)
16 3 (1 + el )
,
e el2 (k + l + )(k + l )(k l + )(k l )
4 3 (1 + ek+ )
4 3 (ek + ek+ + 2ek ek+ )
,
2 k((k + )2 l 2 )
2 k((k + )2 l 2 )
ek2 ek+
e ek+
4 3 (1 + ek )
4 3 (1 + ek )
+
(k ),
ek2 e k((k )2 l 2 ) ek2 k((k )2 l 2 )
(3.29)
139
gences from the fermion wave function renormalization and the vertex correction are the same
magnitude with an opposite sign. The latter feature is expected from WardTakahashi identity.
We also evaluated the finite part of corrections. We classified such corrections to two type:
those come from the wavefunction renormalization of A5 and those come from 1PI twoloop diagrams. The former keeps the structure of the oneloop effective potential essentially unchanged
and is obtained from the product of the ratio of the wavefunction renormalization factors Z /Z5
and oneloop finite Higgs mass. Although these wave function renormalization factors are linearly divergent, 5D Lorentz invariance ensures that these have same contributions. Therefore, the
UV divergences are exactly canceled in Z /Z5 while IR divergences appear. This is because we
consider exactly massless charged fermion, and we introduce a small VEV of A5 as an IR cutoff.
As for the latter, we found that they cancel out among themselves in our calculation. This result
seems to be accidental in our simple model because there is no clear physical reason to obtain
such a result. If we consider higher order loop corrections beyond twoloops even in 5D massless QED or calculate quantum corrections in more general models, the finite correction would
be remained to be nonzero.
We should note that the finite value itself may not be taken seriously because our regularization used in this paper does not have 4D gauge invariance. Namely, the photon has a
nonvanishing mass at oneloop level. However, we would like to emphasize that only the 5D
Lorentz symmetry2 (Z /Z5 ), the shift symmetry (Part I) and the relation expected from Ward
Takahashi identity (Parts II and III) are important to cancel all possible divergences. In fact, the
4D gauge invariance is not so important for the finiteness of the mass of A5 since the shift symmetry forbids the mass of A5 . Our regularization indeed preserves the shift symmetry by doing
the summation of KK modes and the relation expected from WardTakahashi identity. We can
conclude from these observations that the finiteness for the mass of A5 is correct even in our regularization scheme. Off course, it is desirable to calculate the mass in a full 5D gauge invariant
way to obtain a reliable finite mass. This subject is left for a future work.
Our discussion of obtaining the finite Higgs mass at any order of perturbations would be
generic in any gaugeHiggs unification models. Therefore, it would be very interesting to extend
our analysis to nonAbelian case not only from the theoretical but also from the phenomenological viewpoints. This subject will be reported elsewhere [11].
Acknowledgements
N.M. would like to thank N. Sakai for bringing my attention to this subject and C.S. Lim for
valuable discussions at an early stage of this work. He also would like to thank L. Silvestrini for
useful comments. T.Y. would like to thank K. Izawa for stimulating discussions. We would like
to thank M. Serone for valuable comments on the previous version of the manuscript. The work
of N.M. is supported by INFN, sezione di Roma. The work of T.Y. is supported by SISSA.
Appendix A. Detailed calculation of twoloop corrections to Higgs mass
In this appendix, the detailed calculations of Higgs mass at twoloop part expressed as IIV
in Section 3.2. and its physical interpretations are described.
2 In the case of explicit violation of 5D Lorentz invariance as in Ref. [4], Z /Z may be no longer finite. However, in
5
this case, there exist two counter terms to remove the divergences in both Z and Z5 . Thus, even in such a case, Higgs
mass will be finite since the shift symmetry protects its finiteness.
140
A.1. Part I
First, we evaluate the contribution from the first part, namely the summations are replaced by
integrations on the real axis. This contribution is expected to correspond to the diagrams where
both loops do not wind around S 1 and thus can be shrinked to a point. In general, such diagrams
give the strongest divergences. However, in our case, the fivedimensional gauge invariance will
forbid such a contribution.
Before evaluating the contribution, we define a new vector from lE and kE as E kE lE ,
and we use the same parameters without the index E to denote the absolute values of the vectors.
Among these three vectors, we can choose any two vectors as the integral variables. Then the
integrand of (3.25) is written as
I (m, n)
(l 2 n2 )(m2 k 2 ) 2nm(l 2 + k 2 2 )
(l 2 + n2 )2 (k 2 + m2 )2 (2 + (n m)2 )
(l 2 3n2 )(l 2 + k 2 2 ) + 2(3l 2 n2 )nm
.
+
(l 2 + n2 )3 (k 2 + m2 )(2 + (n m)2 )
(A.1)
We can integrate over zm of (A.1) by adding the integration on the large halfcircle in the upper
half plane and evaluating the residues at the poles on the plane.
I (n) =
d zm I (zm , n)
= Res 2iI (zm , n); zm = ik, n + i
(k(k + )2 l 2 + ((k 2)(k + )2 + (k + 2)l 2 )n2 + kn4 )
(l 2 + n2 )2 k((k + )2 + n2 )2
2
2
((k + )l (k + l 2 2 ) 3(k )((k + )2 l 2 )n2 2kn4 )
.
+
(l 2 + n2 )3 k((k + )2 + n2 )
(A.2)
In a similar way, we can further perform the integration over zn of the above expression to find
dzn I (zn ) = Res 2iI (zn ); zn = il, i(k + )
(k + l ) 2
(k + l ) 2
+
= 0.
2
kl(k + l + )
kl(k + l + )2
(A.3)
As expected from the fivedimensional gauge invariance, the contribution from this part vanishes
although each term potentially gives divergent correction.
A.2. Part II
Next, we evaluate the contribution from the second part. This contribution is expected to
correspond to the diagrams where one of the loops winds around S 1 while the other does not.
Some examples are shown in Fig. 7. Because the latter loop can be shrinked to a point, generally
this part gives a divergent contribution, even in the gaugeHiggs unification scenario. However, as
is well known, such divergences can be canceled by the oneloop counter terms. In other words,
after we remove all the divergences in the oneloop diagrams, the contribution from this part
141
(a)
(b)
(c)
Fig. 7. The diagrams where the fermion loop winds around S 1 but the photon loop does not. The cylinder denotes S 1 .
If the fermionphoton loop is shrinked to a point, these diagrams provide corrections of the 4point vertex of the
fermionfermionA5 A5 (a), the gauge interaction vertex (b) and the wave function renormalization (c), respectively.
will be finite. Then, we obtain the finite mass at twoloop level without any additional counter
terms. In [8], the Higgs mass at twoloop level are calculated in 5D supersymmetric theory,
where supersymmetry is broken by ScherkSchwarz mechanism. In fact, the linear divergences
appear and are canceled by the oneloop counter terms. In our particular case, WardTakahashi
identity should make the divergence in this contributions same as the minus of the divergence in
the contribution from the part III. Thus, all the divergences are expected to cancel out with each
other without any counter terms.
Now, let us evaluate the residues of the poles of zm in the upper half plane, i.e., zm = ik and
zm = n + i. We can interpret the former contribution as the one comes from the diagram where
the fermion line with the momentum k winds around S 1 (Fig. 7), while the latter as the one
from the diagram where the photon line winds. The residue of the first term in (A.1) on the pole
zm = ik is evaluated as
2i
; zm = ik
J1 (n) 2 Re Res 1st term of I (zm , n)
exp(2izm ) 1
2
= 4 ek k 2 l 2 k 2 2 + 2k 6 + k 4 l 2 32 4k 2 l 2 2 l 2 2 4 n2
+ 4k 4 k 2 l 2 + 22 2 l 2 2 2 n4 + 2k 2 l 2 + 2 n6
+ (1 + ek )k k 2 2 + n2 (k + )2 + n2 (k )2 + n2 l 2 + n2
2
2
2
/ ek2 k (k + )2 + n2 (k )2 + n2 l 2 + n2
(A.4)
and that of the second term is given by
J2 (n) 2 Re Res 2nd term of I (zm , n)
2i
; zm = ik
exp(2izm ) 1
2
2
= 2 k 2 2 k 2 + l 2 2 l 2
+ 3k 4 2k 2 4l 2 + 32 + l 2 2 l 2 32 n2 + k 2 3l 2 + 32 n4
3
/ ek k (k + )2 + n2 (k )2 + n2 l 2 + n2 ,
(A.5)
where ek exp(2k) 1. After the integration over zn , we find these respectively become
2 2 (2(1 + ek )(k + l)((k + l)2 2 ) + ek (3(k + l)2 + 2 ))
,
ek2 kl((k + l)2 2 )2
2 2 ((3(k + l)2 + 2 ))
ek kl((k + l)2 2 )2
(A.6)
(A.7)
142
2 2 (k 2 (l + 3) (l )(l + )2 )
ek kl(( + l)2 k 2 )2
(A.8)
(A.9)
for k < . Note that all terms vanish in the limit ek . This means that we do not have UV divergences in k integration. On the other hand, we may encounter divergences in l() integration.
In fact, we can see that the integration of (A.8) over lE is linearly divergent, while the one of
(A.9) converges. These are consistent with the interpretation that these contributions correspond
to the diagram where the fermion line with the momentum k winds on S 1 : (A.8) corresponds to
the vertex correction (Fig. 7(b)) while (A.9) corresponds to the correction of the four point vertex
fermionfermionA5 A5 (Fig. 7(a)).
In a similar way, contributions from the pole at zm = n + i is evaluated as
2i
; zm = n + i
J3 (n) 2 Re Res 1st term of I (zm , n)
exp(2izm ) 1
2
2
= 2 k 2 2 k 2 + 2 l 2 + k 2 2 k 2 + 52
+ 3k 4 6k 2 2 54 l 2 n2 + 3k 4 + k 2 3l 2 + 22 5l 2 2 + 34 n4
+ 3k 2 + l 2 2 n6 + n8
2
2
2
/ e (k + )2 + n2 (k )2 + n2 l 2 + n2
(A.10)
and that of the second term is given as
J4 (n) 2 Re Res 2nd term of I (zm , n)
2i
; zm = n + i
exp(2izm ) 1
2
2
2
2
2
2
2 2
= 2 k k + l l + 3 k 2 + 4 k 2 + 22 l 2 + l 4 n2
+ 5k 2 + 3l 2 + 2 n4 2n6
3
/ e (k + )2 + n2 (k )2 + n2 l 2 + n2 ,
(A.11)
where e exp(2) 1. After the integration over zn , we find these terms give the same
contributions with an opposite sign and the sum of these vanishes. This is also consistent with
the interpretation that these contributions correspond to the diagram where the photon line winds
around S 1 . Namely, such contributions correspond to the correction of the four point vertex
4
term which has vanishing contribution to the
AM AM A5 A5 . This is the correction to the FMN
mass correction because its Feynman rule contains momenta of the four lines and our interest is
zero external momenta case.
In summary, the contribution from this part is written as
4 3 (1 + ek )
4 3 (1 + ek )
(k ),
ek2 l((l + )2 k 2 ) ek2 k((k )2 l 2 )
(A.12)
143
the second term is finite since this contribution exists only when the momentum is smaller than
the momentum k.
A.3. Part III
Now, we evaluate the contribution from the third part. The integration over zm is given in
(A.2). It shows that there are two poles in the upper half plane: zn = il and zn = i(k + ). The
contribution from the pole at zn = il is calculated as
2i
2 Re Res I (zn )
; zn = il
exp(2izn ) 1
=
+
+
el kl((k + )2 l 2 )2
+ )2 l 2 )
el2 k((k
(A.13)
(A.14)
where el exp(2l) 1. These terms vanish in the limit el , and thus the l integration
is free from UV divergences. Note that if we choose lE and E as the integral variables and
rename them as LE and KE , kE is written as kE = KE LE E . Then the three new momenta
(KE , LE , E ) satisfy the same relation as (kE , lE , E ). In addition, the integration measure
under this rename is invariant. Thus, this means we can replace k and with each other. From
this observation, it is clear that the 4D momentum integral of the last term in (A.14) does not
contribute.
The first terms of (A.13) and (A.14) are linearly divergent with respect to k integration. It is
interesting to find that the divergence in (A.13) is the half of the one in (A.14) with the opposite sign, and is the same as the one in the part II. These results are again consistent with the
interpretation that these contributions correspond to the diagram where the fermion line with
the momentum l winds around S 1 : (A.13) corresponds to the vertex correction (Fig. 7(b)) and
(A.14) corresponds to the wave function correction of the fermion (Fig. 7(c)). The second terms
in (A.13) and (A.14) are canceled.
The contribution from the pole at zn = i(k + ) is summarized as
2i
4 3 (1 + ek+ )
2 Re Res I (zn )
, (A.15)
; zn = i(k + ) = 2
exp(2izn ) 1
ek+ k((k + )2 l 2 ))
where ek+ exp(2(k + )) 1. This term vanishes when ek+ , and thus this contribution is finite under both k and l integrations. This contribution is interpreted as coming from the
diagram where the fermion line with the momentum k and the photon line wind around S 1 .
In summary, the contribution from this part is written as
4 3 (1 + el )
4 3 (1 + ek+ )
.
2 k((k + )2 l 2 ))
el2 k((k + )2 l 2 ) ek+
(A.16)
The first term is linearly divergent, which originated from two wave function renormalizations
and a vertex correction, namely a wave function renormalization. One can see that this contribution and the first term in (A.12) are exactly canceled as expected from WardTakahashi identity.
144
A.4. Part IV
Finally we evaluate the contribution from the fourth part. The contribution of the part of the
residues in m is written in (A.4), (A.5), (A.10) and (A.11). They have poles on the upper half
plane at zn = il, i(k +), ik . The first and the last two parts give the following contributions;
2i
2 Re Res J1 (zn ) + J2 (zn )
; zn = il
exp(2izn ) 1
=
8 3 (1 + ek )(k 2 l 2 2 )
ek2 el l(k + l + )(k + l )(k l + )(k l )
[k l for the first term]
2 Re Res J3 (zn ) + J4 (zn )
=
8 4 (1 + el )(2 + el )
ek el3 kl
2i
; zn = il
exp(2izn ) 1
,
8 4 (1 + el )(2 + el )
16 3 (1 + el )
+
e el2 (k + l + )(k + l )(k l + )(k l )
e el3 l
(A.19)
(A.20)
2i
; zn = ik 
exp(2izn ) 1
4 3 (1 + ek )(2ek + ek2 e )
4 3 (1 + ek )
(k ),
ek2 (ek e )2 k((k )2 l 2 ) ek2 k((k )2 l 2 )
2i
2 Re Res J3 (zn ) + J4 (zn )
; zn = ik 
exp(2izn ) 1
=
(A.18)
2i
; zn = i(k + )
exp(2izn ) 1
(A.17)
4 3 (1 + ek )(1 + e )
e (ek e )2 k((k )2 l 2 )
(A.21)
(A.22)
16 3 (1 + el )
,
e el2 (k + l + )(k + l )(k l + )(k l )
(A.23)
4 3 (1 + ek+ )
4 3 (ek + ek+ + 2ek ek+ )
,
2
2
2
ek ek+ k((k + )2 l 2 )
e ek+ k((k + )2 l 2 )
4 3 (1 + ek )
4 3 (1 + ek )
+
(k ).
ek2 e k((k )2 l 2 ) ek2 k((k )2 l 2 )
145
(A.24)
(A.25)
Note that all terms above are finite because this part corresponds to the diagram where the
fermion and the photon wind around S 1 .
References
[1] N.S. Manton, Nucl. Phys. B 158 (1979) 141;
D.B. Fairlie, Phys. Lett. B 82 (1979) 97.
[2] Y. Hosotani, Phys. Lett. B 126 (1983) 309;
Y. Hosotani, Ann. Phys. 190 (1989) 233;
N.V. Krasnikov, Phys. Lett. B 273 (1991) 246;
H. Hatanaka, T. Inami, C.S. Lim, Mod. Phys. Lett. A 13 (1998) 2601;
G.R. Dvali, S. RandjbarDaemi, R. Tabbash, Phys. Rev. D 65 (2002) 064021;
A. Masiero, C.A. Scrucca, M. Serone, L. Silvestrini, Phys. Rev. Lett. 87 (2001) 251601;
I. Antoniadis, K. Benakli, M. Quiros, New J. Phys. 3 (2001) 20.
[3] M. Kubo, C.S. Lim, H. Yamashita, Mod. Phys. Lett. A 17 (2002) 2249;
C. Csaki, C. Grojean, H. Murayama, Phys. Rev. D 67 (2003) 085012;
N. Haba, M. Harada, Y. Hosotani, Y. Kawamura, Nucl. Phys. B 657 (2003) 169;
N. Haba, M. Harada, Y. Hosotani, Y. Kawamura, Nucl. Phys. B 669 (2003) 381, Erratum;
K. Takenaga, Phys. Lett. B 570 (2003) 244;
N. Haba, T. Yamashita, JHEP 0402 (2004) 059;
B. Grzadkowski, J. Wudka, Phys. Rev. Lett. 93 (2004) 211603;
B. Grzadkowski, J. Wudka, Phys. Rev. D 70 (2004) 015010;
Y. Hosotani, S. Noda, K. Takenaga, Phys. Rev. D 69 (2004) 125014;
Y. Hosotani, S. Noda, K. Takenaga, Phys. Lett. B 607 (2005) 276;
G. Martinelli, M. Salvatori, C.A. Scrucca, L. Silvestrini, JHEP 0510 (2005) 037.
[4] C.A. Scrucca, M. Serone, L. Silvestrini, Nucl. Phys. B 669 (2003) 128;
N. Haba, Y. Hosotani, Y. Kawamura, T. Yamashita, Phys. Rev. D 70 (2004) 015010;
G. Cacciapaglia, C. Csaki, S.C. Park, JHEP 0603 (2006) 099;
G. Panico, M. Serone, A. Wulzer, Nucl. Phys. B 739 (2006) 186;
N. Maru, K. Takenaga, hepph/0602149.
[5] R. Contino, Y. Nomura, A. Pomarol, Nucl. Phys. B 671 (2003) 148;
K. Agashe, R. Contino, A. Pomarol, Nucl. Phys. B 719 (2005) 165;
K.Y. Oda, A. Weiler, Phys. Lett. B 606 (2005) 408;
Y. Hosotani, M. Mabe, Phys. Lett. B 615 (2005) 257;
Y. Hosotani, S. Noda, Y. Sakamura, S. Shimasaki, hepph/0601241.
[6] Y. Hosotani, hepph/0504272;
Y. Hosotani, hepph/0607064.
[7] K. Hasegawa, C.S. Lim, N. Maru, Phys. Lett. B 604 (2004) 133.
[8] A. Delgado, G. von Gersdorff, M. Quiros, Nucl. Phys. B 613 (2001) 49.
[9] ALEPH, DELPHI, L3, OPAL Collaborations, Phys. Lett. B 565 (2003) 61.
[10] N. Haba, S. Matsumoto, N. Okada, T. Yamashita, JHEP 0602 (2006) 073.
[11] N. Maru, K. Takenaga, T. Yamashita, in preparation.
Abstract
We compute higher order contributions to the free energy of noncommutative quantum electrodynamics
at a nonzero temperature T . Our calculation includes up to threeloop contributions (fourth order in the
coupling constant e). In the high temperature limit we sum all the ring diagrams and obtain a result which
has a peculiar dependence on the coupling constant. For large values of eT 2 ( is the magnitude of the
noncommutative parameters) this nonperturbative contribution exhibits a nonanalytic behavior proportional
to e3 . We show that above a certain critical temperature, there occurs a thermodynamic instability which
may indicate a phase transition.
2006 Elsevier B.V. All rights reserved.
PACS: 11.10.Wx
1. Introduction
This paper is about the thermodynamics of quantum electrodynamics formulated in a noncommutative space (NCQED), which constitutes a system of selfinteracting gauge fields [1,2].
We employ the imaginary time formalism [3] in order to compute the free energy of the pure
gauge sector of the theory. Previous investigations on this subject have revealed interesting properties already at the lowest nontrivial order (twoloop order) [4]. The main purpose of the present
paper is to take into account the higher order corrections to the free energy.
Corrections to the free energy, higher than twoloops, in thermal field theories forces us to
take into consideration the sum of an infinite series of diagrams. This happens because the fields
* Corresponding author.
147
acquire, from interactions, an effective thermal mass. This socalled plasmon effect is known to
happen in gauge theories, like SU(N ), even without fermions (pure YangMills theory) because
of the nonAbelian character of fields [5]. Inasmuch NCQED is a gauge theory of selfinteracting
gauge fields, it is natural to consider the possibility of similar nonperturbative effects when we
take into account higher order corrections to the free energy. As we will see the nonperturbative effect is more subtle in the case of NCQED because the would be infrared singularities
are smoothed by the noncommutative scale (this would not be the case in the noncommutative
version of the U(N ) theory). However, in the present work we will show that in the regime of
temperatures much higher than the inverse of the noncommutative scale, we meet a breakdown
of the perturbative series and the need for summing an infinite series of diagrams is ineluctable.
In order to pinpoint the breakdown of the perturbative regime we first analyze in detail the
twoloop contributions. We perform the calculation in a general covariant gauge and obtain a
gauge independent result in terms of a relatively simple function of the temperature which can be
computed analytically for both the low and high temperature regimes. We also plot the result for
all intermediate values of the temperature. We show explicitly that the twoloop result converges
to a finite
limit when the temperature T is much higher than the inverse of the noncommutative
scale 1/ , or, equivalently when T 2 1.
Then we consider all the threeloop contributions and we verify that similarly to what happens
in the commutative fields theories, the socalled ring diagram is dominant in the high temperature
regime, being proportional to . We also show that this set of graphs, as well as the higher order
ring diagrams are independent of the gauge fixing parameter. We find that the relative strength of
the ring diagrams is of order (e )2 , where e is the coupling constant. After summing all the rings,
the resulting expression is nonanalytic in the coupling and behaves as e3 for large values of e .
We also investigate the behavior of the free energy as a function of e and show that, above some
critical temperature Tc , there occurs a thermodynamic instability. This is manifested through the
appearance of an imaginary part in the free energy, which can be directly related to the decay
rate of a metastable vacuum [6]. This behavior is induced by the presence of a noncommutative
magnetic mode in the theory, which is associated with the transverse component of the static selfenergy. We show that the magnetic mass associated with the noncommutative transverse mode,
for temperatures close to the critical temperature, is proportional to (Tc2 T 2 )1/2 .
In Section 2 we present the basic features of NCQED as well as our notation and conventions.
In Section 3 we compute the twoloop contributions to the free energy and consider the limits
of high and low temperature up to subleading terms. In Section 4 we compute the threeloop
contributions and obtain the dominant high temperature behavior in terms of graphs involving
selfenergy insertions. In Section 5 we compute the sum of all ring diagrams, investigate the
properties of the free energy and obtain the critical value of the temperature. In Section 6 we
discuss the main results and the connection between the instability and the noncommutative
magnetic mode. We leave to Appendices A and B the technical details of the rather involved
calculations of the Feynman graphs.
2. Noncommutative QED
The gauge invariance of the QED action
1
S=
4
dd x F (x) F (x)
(2.1)
148
under a U (1) gauge transformation implies that the gauge potentials, in terms of which the electromagnetic field F is defined, must have selfinteractions when the theory is formulated on a
noncommutative manifold [2]. The basic reason for this is that the usual product of two functions,
f (x)g(x) is replaced by the GrnewoldMoyal star product [1].
i
(2.3)
Since the U (1) transformation acts on the gauge fields in such a way that
i
A = U U 1 + U A U 1 ,
e
the extra terms in the variation of the action can be compensated using
F = A A ie[A , A ] ,
(2.4)
(2.5)
where [f (x), g(x)] f (x) g(x) g(x) f (x). Consequently, in the noncommutative version
of QED (NCQED) there are cubic and quartic interaction vertices similarly to the case of the
YangMills theory. Of course the details of the interaction vertices are quite different from the
usual gauge theories since there are no color charges in NCQED.
The quantization of this theory follows closely the usual approach employed in the case of
nonAbelian gauge fields. In an covariant class of gauges one adds the gauge fixing and ghost
actions given respectively by
1
dd x A A
Sgf =
(2.6)
2
and
Sghost =
dd x c c ie[A , c] .
(2.7)
The full quantum behavior can be described in terms the sum of (2.1), (2.6) and (2.7) which yields
the following Feynman rules for the propagators and the interaction vertices (for convenience we
are not including the factor i from exp (iS) in the path integral)
p p
1
:
(1 ) 2 ,
(2.8a)
p 2 + i
p
:
1
,
p 2 + i
(2.8b)
p 1 p2
(p1 p2 ) + (p2 p3 )
2
+ (p3 p1 ) ,
2ie sin
(2.8c)
:
2ie sin
p 2 p3
p3 ,
2
149
(2.8d)
p3 p4
p1 p 2
: 4e2 sin
sin
2
2
p1 p3
p4 p2
+ sin
sin
2
2
p2 p3
p1 p 4
sin
,
+ sin
(2.8e)
2
2
(all momenta are inward and Dirac delta functions for the conservation of momenta are understood). Our Minkowski metric convention is (+ ) so that p2 = p02 p
2 . The wavy
and dashed lines represent respectively the gauge and ghost fields and we have introduced the
notation
p q p q .
(2.9)
In order to study the thermodynamics of NCQED we will employ the imaginary time
formalism. This amounts to replace the zero components of all momenta by 2inT (n =
0, 1, 2, . . .). Then the ddimensional integration over the loop momenta is modified according
to the rule
dd p f (p0 , p)
(2.10)
T dd1 p
f (p0 = in , p),
n=
where n 2nT .
Another important consistency requirement is that i0 = 0. Then the momentum dependence
in the interaction vertices depends only on the spacial components (p q = pi ij qj ) so that
the sum over the Matsubara frequencies n can be performed using the standard techniques. As
we will see in the following sections the noncommutativity of space coordinates has interesting
consequences on the thermodynamical properties of the system.
3. The lowest order contributions to the free energy
Most of the results of this section can also be found in reference [4]. One reason to reproduce
these results here is because we need to specify our notation and conventions as well as our
basic approach. We will show how our approach allows to obtain in a simple and straightforward
way the twoloop result in terms of a relatively simple function of the temperature which can be
explicitly computed for both the low and high temperature regimes. In particular we obtain an
explicit result for the sub leading contributions at high .
Let us start by defining clearly what we are going to compute. Our first main goal is to obtain
the free energy per unit of volume [3]
(T , ) =
T
(T , )
= log Z(T , )
V
V
(3.1)
150
Fig. 1. Oneloop diagrams which contribute to the freeenergy. Wavy and dashed lines denote respectively vector and
ghost fields.
of NCQED theory defined in the previous section. Here Z(T , ) is the partition function, which
in the imaginary time formalism, has the form
Z(T , ) = DA Dc Dc exp (S + Sgf + Sghost ) ,
(3.2)
where the actions inside the exponential are given by Eqs. (2.1), (2.6) and (2.7) with the replacement x0 ix4 and x4 integrated from zero to 1/T . The argument represents the dependence
of the partition function on the noncommutative parameter .
The lowest order contribution to (T , ) can be represented diagrammatically by the graphs
shown in Fig. 1. This is the freefield contribution which of course is independent of (the
noncommutative character only shows up through the interaction vertices in Eqs. (2.8)). This
gives the known result of free QED, which, in d = 4 yields
d3 k k
(0) (T )
(k/T )
.
=2
+
T
log
1
e
(3.3)
V
(2)3 2
Notice that the factor 2 on the right side of the above equation counts the number of physical
degrees of freedom of the massless gauge boson A (the ghost loop contribution in Fig. 1 cancel
the unphysical degrees of freedom). The first term inside the square bracket of Eq. (3.3) represents the temperature independent (infinite) contribution of the zero point energy of the vacuum.
Subtracting this contribution, and performing an integration by parts, we obtain
(0)
(0) (T )
1
(T ) =
= 2
V
3
k 3 NB (k) dk,
(3.4)
where
1
e(k/T ) 1
is the BoseEinstein thermal distribution. Using the formula [7]
NB (k)
0
x n1
dx = (n) (n)
ex 1
(3.5)
(3.6)
2 4
T .
45
(3.7)
The first effects of interactions appears at O(e2 ) and can be computed from the diagrams in
Fig. 2. The individual contributions of these diagrams (including the symmetry factors and the
151
1
12
dp dq I2
=e
1
8
(3.8)
(3.9)
and
dp dq I3 .
=e
dd1 p
dp = T
.
(2)d1 n=
(3.10)
(3.11)
Using the Feynman rules given in Eqs. (2.8) the integrands I1 , I2 and I3 can be expressed as
1
1
1
2 pq
+
,
sin
I1 = (d 1) 2 2 + 2
(3.12)
2
p q
p (p + q)2 q 2 (p + q)2
1
pq
,
I2 = (d 1)d 2 2 sin2
(3.13)
2
p q
and
I3 =
1
1
1
2 pq
+
.
sin
2
p 2 q 2 p 2 (p + q)2 q 2 (p + q)2
(3.14)
Since I1 , I2 and I3 are integrands of dimensionally regularized integrals, one can perform shifts
in order to simplify the above expressions before the computation of the integrals and sums.
In this way we can simplify I1 and I3 performing the shifts q q p and p p q in their
152
second and third terms, respectively. Noticing that the sin2 factor remains unchanged under these
shifts, all the terms are reduced to a single type of momentum dependence which has the same
form of I2 . Combining the reduced integrands we obtain
1
12
1
8
1
2
1
2 pq
dp dq sin
= e (d 2) T
.
2
2
p q2
2
(3.15)
The factor (d 2)2 is to be expected from the gauge invariance of (T , ). Indeed, if we employ
an axial gauge condition in two dimensions so that A0 = A1 ( is some constant) the theory
becomes free because [A0 , A1 ] = 0. In general the gauge invariance of higher order contributions can be rather subtle because of the gauge dependence of the running coupling constant e.
However, we have computed Eq. (3.15) using the vector field propagator in Eq. (2.8a) showing explicitly that the result does not depend on , as we show in Appendix B. This can be
understood taking into account that the relation between the coupling constant computed in two
different gauges e2 ( = 1) = e2 ( ) + O(e4 ( )) does not change the e2 contribution in Eq. (3.15).
This however is not the case of contributions of higher order. Even so, the factor d 2 can be
present in the higher order contributions because the coupling constant would not be modified
by quantum corrections in the free twodimensional theory.
The computation of the sums and integrals in Eq. (3.15) can be performed in a straightforward
way. Using the standard relation [3]
i
dp0
f (p0 ) + f (p0 ) +
4i
i+
f (p0 = in ) =
where NB is given by Eq. (3.5). This result already takes into account that pieces containing
less than two Bose distributions vanish in dimensional regularization [8]. Of course there is no
need to keep an arbitrary d in Eq. (3.17) and in what follows we will consider d = 4. Even
without explicitly performing the integral we notice that the twoloop contribution in Eq. (3.17)
is positive while the free theory gives the negative value in Eq. (3.7).
Let us compute the twoloop integrals in Eq. (3.17). Introducing the quantity such that
ij k k = ij
(3.18)
cos p = p
p q = p q = p
q
q  cos p = p
q   sin q cos p ,
(3.19)
153
Fig. 3.
and performing the elementary p , p and q integrations Eq. (3.17) and using (3.6) yields
e2 T 4 e2 T 4 J ( )
,
72
2 4
where we have introduced the function
(2) (T , ) =
1
J ( ) =
2
du
dv
d
0
sin( uv sin())
.
(exp(u) 1)(exp(v) 1)
(3.20)
(3.21)
Performing the angular integration, which can be represented in terms of the Struve function H0
[7], we obtain
J ( ) =
2
du
dv
0
H0 ( uv)
.
(exp(u) 1)(exp(v) 1)
(3.22)
An equivalent form of this result can also be found in reference [4] expressed in terms of a four
dimensional integral.
For small values of Eq. (3.21) can be easily computed by Taylor expanding sin( uv sin()).
Then, at any order in the angular integration is trivial and the integrations over u and v can be
done using Eq. (3.6). The leading term of this expansion will cancel the first term in Eq. (3.20)
and the first nonvanishing term is
e2 T 4 2 2 2
(2) (T , )
(3.23)
.
2
45
This result is in agreement with reference [4] if we take into account that they use a definition of
which is twice ours.
From Eq. (3.21) one can also perform a numerical integration and obtain the functional dependence for any value of . Fig. 4 shows a plot of the function J ( ) which clearly indicates that
J ( )/ vanishes for large values of . This property can also be verified using the asymptotic
behavior of the Struve function for large values of its argument which yields
2
J ( )
(3.24)
log .
2
Since J ( ) grows logarithmically, the high temperature behavior of the two loop result, given
by Eq. (3.20), converges to a independent result. It is remarkable that the resulting large
154
behavior, given by the first term in Eq. (3.20), coincides with the twoloop SU(2) freeenergy
modulo a factor of three which counts the internal degrees of freedom associated with the SU(2)
vector fields [5,10]. This correspondence between large and SU(2) has been found before in
the computation of other thermal Green functions
in noncommutative QED [9]. Previously it has
been found that in the regime of large , or 1/T , the nonplanar sector of scalar fields
has a thermodynamics resembling that of a d = 2 dimensional field theory [11]. In the present
context of NCQED, the large behavior is consistent with the finite infrared behavior of the two
loop contribution in commutative YangMills theory. Indeed, since the twoloop contribution to
the SU(2) freeenergy is finite, it is natural that the corresponding noncommutative QED result
is independent of for large . At three loop order it is known that there are IR divergences
in the YangMills free energy. This suggests that the higher order contributions to the NCQED
free energy may be dependent on . In the following section we will consider the three loop
contributions.
4. The threeloops contributions
The threeloop contributions to (T , ) are shown in Figs. 5 and 6. We have defined r
k + p, s k + q and t q p and the arrows indicate the direction of momenta. Our basic
strategy to deal with these rather involved diagrams consists in first reduce the integrands as
much as possible using a generalization of the procedure employed to obtain to Eq. (3.15). In the
appendix we present the details of this rather technical manipulations. Here we only remark that
the shifts in momenta that can make the integrands simpler are restricted by the more involved
momentum dependence inside the trigonometric factors. Of course the algebra involved is more
complicated than in commutative gauge theories like QCD [10].
Let us first consider the diagrams in Fig. 5. Using the expressions (A.3a) to (A.3f) these
contributions combine into the following expression
(a)
(b)
(c)
(d)
(e)
(f)
155
Fig. 5.
1
4
= e4
1
(p q)2
1
16T1 (d 4)(d 2)2 4 2 2 + (d 2)2 4 2 2 2 2
4
k p q
k p q r s
1
(d 2)2 d
1
(d 2) 2 2 2 2
,
16
2 p2 q 2 r 2 s 2
k p q r
2
where
T3
dd1 k
(2)d1
dd1 p
(2)d1
(4.1)
dd1 q
(2)d1
k4 ,p4 ,q4
is a compact notation for the sums and integrations over the three independent momenta and
2 pk
2 q k
T1 sin
(4.2)
sin
.
2
2
In the lefthand side of Eq. (4.1) we are using a compact representation of the graphs in Fig. 5
in terms of two insertions of the selfenergy, shown in Fig. 8, with the proper symmetry factor.
There are interesting consequences of the relation between the free energy and the selfenergy
156
(g)
(h)
(i)
(j)
(k)
(l)
Fig. 6.
which will be explored later in this section and in the conclusion. Here we only point out that the
wellknown problems of infrared divergences associated with these socalled ring diagrams are
regulated in the noncommutative theory by the presence of the factor T1 .
Let us now consider the graphs of Fig. 6. Before taking into account all these contributions let
us first focus those terms which are proportional to
T1
p2 q 2 r 2 s 2
(4.3)
as in the last line of Eq. (4.1). The simplest example of such contributions is the one from the
graph (g). Using the Feynman rules given in Eqs. (2.8) we readily obtain
= e
2
4
2
2
T1
2
2
2
2
k +r p k +s q ,
p2 q 2 r 2 s 2 k 4
(4.4)
where we have used the kinematic relations (A.5). Performing the shifts p r and q s,
which does not alter the trigonometric factors and denominators, the terms involving r 2 p 2 and
157
1
2
= e4
T1
.
p2 q 2 r 2 s 2
(4.5)
This kind of simplification is also important in order unveil the true power counting of each
individual momenta. Indeed, the naive power counting of Eq. (4.4) would lead us to conclude
that the integrand had a denominator proportional to k 4 .
Adding Eq. (4.5) with the other two contributions from Eqs. (A.3h) and (A.3i) of Appendix A
yields
e4 8d 2(d 2)(d 4)
T1
.
p2 q 2 r 2 s 2
(4.6)
Combining Eq. (4.6) with the last line of Eq. (4.1) produces
e4 2(d 2)(d 4) (d 2)2
T1
.
2
p q 2r 2s2
(4.7)
The sum of the first three lines of the integrand of Eq. (4.1) plus the previous expression gives
the full integrand of the threeloops contributions which are proportional to T1 . As we can see
this part of the integrand vanishes for d = 2.
The remaining contributions from Fig. 6 comes from part of the graphs (h) and (i) as well as
the mercedes graphs in (j), (k) and (l). We shown in the appendix, that the trigonometric factors
of these contributions can all be reduced to a single factor given by
pq
r s
kp
kq
T2 = sin
(4.8)
sin
sin
sin
.
2
2
2
2
Although a direct application of the Feynman rules can produce other kinds of trigonometric
factors like
ps
q r
kp
kq
T3 = sin
(4.9)
sin
sin
sin
,
2
2
2
2
there are simple identities like
T2 T3
=0
e4
k 2 p2 r 2 s 2
(4.10)
(to verify this identity one just perform the shift p r) which makes it possible to express all
the contributions in terms of T2 . The final result from these contributions is given by
2(d 2)(d + 2) 8(d 2)2 p (q + s)
4
T2
e
(4.11)
.
p2 q 2 r 2 s 2
k 2 p2 q 2 r 2 s 2
Notice that the second term in the above expression would vanish in QCD because trigonometric
factor T2 would be absent (there would be a color factor instead).
Combining the first three lines of Eq. (4.1) with Eqs. (4.7) and (4.11) we obtain the following
contribution to the threeloop free energy
158
1
(p q)2
1
(d 4)(d 2)2 4 2 2 + (d 2)2 4 2 2 2 2
4
k p q
k p q r s
2
1
2(d 2)(d 4) (d 2)
(d 2)2 2 2 2 2 +
k p q r
16p 2 q 2 r 2 s 2
2(d 2)(d + 2) 8(d 2)2 p (q + s)
.
+ T2
(4.12)
p2 q 2 r 2 s 2
k 2 p2 q 2 r 2 s 2
From Eq. (4.12) one can now undertake the more challenging task of computing the sums
and integrals. This is also a welldefined expression which does not have infrared divergences.
However, one should remember that, as we have already pointed out, at the order e4 the renormalization of the coupling constant have to be taken into consideration. Therefore, for arbitrary
values of the temperature the result would be incomplete. On the other hand, we can investigate
some extreme limits of Eq. (4.12) which may be gauge invariant and have a meaning by itself.
One such limit is the high temperature limit.
In noncommutative field theory the temperature is certainly high when T 1/ or, equivalently, 1. In order to investigate the regime of large it is more appropriate to perform the
following rescalings
(4) (T , ) = e4
16T1
T (p4 , p),
(p4 , p)
(q4 , q ) T (q4 , q ),
T (k4 , k/
),
(k4 , k)
(4.13)
where the momenta variables on the righthand side are dimensionless. One obvious advantage
of the new dimensionless momentum variables is that the trigonometric factor T1 becomes independent of . Indeed, using the variables indicated in Fig. 7 (these are the only variables which
will be relevant in the limit d = 4 to be considered later) and performing the rescalings (4.13)
yields
p
p
sin cos p ,
k p = k
 sin cos p k
(4.14)
q 
q  sin cos q
sin cos q k
k q = k
(4.15)
and
which, in turn, make T1 independent of . In terms of the new momentum variables the dependence on is transfered to the denominators. Then the terms containing
1
1
=
4
2
2 / 2 ]2
k
[(2n) + k
(4.16)
Fig. 7.
159
yield a dominant 4 behavior when n = 0 (the zero mode). From this simple analysis we see that
the first two terms in Eq. (4.12) are dominant when is large. Taking into account that, under
the rescaling (4.13), the measure transforms as dd1 k T d1 dd1 k/ d1 , the first two terms
in Eq. (4.12) yield a leading contribution in for d = 4, which can be written as
dd1 k 1
dd1 p
dd1 q
(4)
4
2 d 5d
(T , ) 16e (d 2) T
T1
4
(2)d1 k
(2)d1
(2)d1
1 d 4 (p q)2
+ 4 4 ,
(4.17)
4 p2 q 2
p q
p ,q
4
where all the integration variables are dimensionless. Notice that the factor 5d can be viewed
as an infrared regularization which would not be present in QCD.
There are some interesting features about the result in Eq. (4.17). First, we notice that it has
been entirely generated from the ring contribution given in Eq. (4.1). Therefore we should be
able to reproduce the structure of the integrand in Eq. (4.17) from a direct calculation of the
following quantity
Tr (p, k) (q, k) = (p, k) (q, k),
(4.18)
where (p, k) is the integrand of the photon selfenergy, in the limit of large , or, equivalently
= 0 except inside the trigonometric factors. With this prescription, the calculation
setting k4 = k
of the graphs shown in Fig. 8 gives
k p 2p p
.
(p, k) = 4e2 (d 2) sin2
(4.19)
2
p4
p2
Then it is easy to verify that
1 d 4 (p q)2
1
+
.
Tr (p, k) (q, k) = 16e4 (d 2)2 T1
4
4 p2 q 2
p4 q 4
(4.20)
(k) =
(4.21)
dd1 p
(p, k).
(2)d1 p
(4.22)
(a)
(b)
Fig. 8.
(c)
160
Another important property of the high temperature limit of the threeloop free energy as
expressed in the form Eq. (4.21), is that it may be viewed as the first term of a sequence made
by successive insertions of the selfenergy. As we have already pointed out this series of ring
diagrams is free of infrared divergences. Nevertheless, we note that each successive term in this
series will have (for d = 4) an extra power of (e )2 , since each selfenergy contributes with a
factor e2 and a factor 2 associated with the extra photon denominator. Form this point of
view
we see a breakdown of the formal perturbative series in the limit when 1 (T 1/ ).
However, as we will see in the next section we can sum the series of ring diagrams in a closed
form and obtain a nonanalytic behavior in the coupling e.
The third property of Eq. (4.21), which also generalizes to all higher order ring diagrams, is its
gauge independence. To prove this property we first remark that the result for in Eq. (4.19)
has been obtained using the general covariant gauge propagator in Eq. (2.8a) (the gauge dependent terms are suppressed by powers of k). In order to complete the proof, let us finish the
calculation of and shown explicitly that the contraction of with the general gauge propagator in Eq. (2.8a) is independent of the gauge parameter . Although the selfenergy has been
computed previously, we present here yet another derivation more akin to our present approach.
From the structure of given by Eqs. (4.19) and (4.22) we can write the following general
expression
k k
k k
k k
k k
= 00 u u + nc
+ 11 2 + 22 u u 2
, (4.23)
k
k
k 2
k 2
where 00 , nc , 11 and 22 are functions of k = k sin (structures which are odd in k or k
are not compatible with the symmetry of ). In order to perform the integration in Eq. (4.22)
we are free to choose u = (1, 0, . . . , 0), k = k(0, 1, . . . , 0) and k = k(0, 0, . . . , 1) (in the
static limit k0 = 0). Then, equating Eqs. (4.22) and (4.23) and contracting with the four tensors
in Eq. (4.23) we obtain
dd1 p
p
2
1
2 p k
sin
+
2
,
00 = 4(d 2)e2
(4.24a)
2
(2)d1
p2
p4
2
(p k)
1
dd1 p
2 p k
nc = 4(d 2)e2
(4.24b)
sin
+
2
,
2
2
(2)d1
p2
4
k p
d1 p
2
d
(
p k)
1
p
k
2
sin
+
2
,
11 = 4(d 2)e2
(4.24c)
2 p4
2
(2)d1
p2
k
2
2 (p k)
2
(p k)
1
dd1 p
2 p k
2
sin
+

p
.
22 = 4(d 2)e2
2
2
2
(2)d1
p2 p4
k
k
(4.24d)
The Matsubara sums in the previous equations can be easily performed using Eq. (3.16) and
closing the contour on the right side of the complex plane. The result can be expressed in terms
of the BoseEinstein thermal distribution, given by Eq. (3.5), and its derivative as follows
1
NB (p)
=
+ (T = 0),
2
p
p
1
1
NB (p)
=
+ (T = 0).
NB p
p
p 4 2p
2
(4.25a)
(4.25b)
161
The (T = 0) pieces yield a zero result in the dimensionally regularized momentum integral [8]
so that we may replace the sums in Eqs. (4.24a)(4.24d) by the first terms in Eqs. (4.25a) and
(4.25b). In what follows we will consider d = 4 since the thermal integrals will not need a
regularization. Using the thermal part of Eqs. (4.25a) and (4.25b), Eq. (4.24a) yields
e2
00 = 3
d p sin
p k cos p
2e2
1
sin(p k)
dp p
NB (p) = 2
p2 p
.
2
e 1
k
0
(4.26)
Integrating by parts,
2e2
00 = 2
dp
0
1
p
e 1
2p +
sin(p k)
k
+ p cos(p k) .
(4.27)
This integral can be done using the formula (3.6) as well as [7]
0
0
2m
x 2m sin bx
1
m
dx
=
(1)
cth(b)
,
ex 1
2b
b2m 2
(4.28a)
2m+1
x 2m+1 cos bx
1
m
dx
=
(1)
cth(b)
.
ex 1
2b
b2m+1 2
(4.28b)
The result is
00 (k) =
2
2e2 2
2
cth(
k)
(
k)
.
+
cth
2
2 6
2k
(4.29)
2
2e2 2
1
2
cth(
k)
(
k)
+
cth
2
2 2
2k
k 2
(4.30)
(4.31)
From this transversality property, it follows that the gauge parameter dependence of the photon
propagator will cancel in a ring diagram containing any number of selfenergies. Therefore the
sum of all the rings is also gauge independent. Notice that the cancellation of the gauge dependent
part of the photon propagator in (2.8a) takes place inside the integral (4.21) which is finite both
in the infrared and ultraviolet regime.
162
5. Nonperturbative contributions
Let us now consider the following contribution to the free energy
1
1
ring
(T , ) =
2
2
1
3
+ ,
(5.1)
n
x
n=2
= x + log(1 x)
we obtain
ring
(5.2)
(e )2 00 (k)
(e )2 00 (k)
d k
+ log 1
2
2
k
k
(e )2 nc (k)
(e )2 nc (k)
sin .
+ log 1
, k = k
+
2
2
k
k
1 T4
(T , ) =
2 (2 )3
(5.3)
The functions 00 (k) and nc (k) are such that e2 00 (k) = 00 (k) and e2 nc (k) = nc (k)
where 00 (k) and nc (k) given by Eqs. (4.29) and (4.30) respectively. Performing the elementary angular integration this expression can be written as
ring
e3 T 4 1
(T , ) =
(2)2 (e )3
/2
(e )2 00 (k)
(e )2 00 (k)
d sin dk k 2
+
log
1
k2
k2
0
(e )2 nc (k)
nc (k)
+
log
1
.
k2
k2
(e )2
(5.4)
Fig. 9, show the plots of the functions 00 (k) and nc (k). The asymptotic behavior can be
easily obtained from Eqs. (4.29) and (4.30). The result for small and large values of k are given
respectively by
4 2 2 4 4 4
k +
k + ,
45
315
2 2 2 8 4 4
k
k +
nc (k)
45
945
00 (k)
(5.5a)
(5.5b)
and
2
lim 00 (k) = ,
3
lim nc (k) = 0.
k
k
(5.6a)
(5.6b)
163
(a)
(b)
Fig. 9. The longitudinal and transverse modes of the static photon selfenergy in NCQED.
Eq. (5.4) represents a contribution to the freeenergy which is nonanalytic in the coupling
constant in the sense that it is not a simple power of e2 . In order to gain some basic understanding
of this function, let us first consider its asymptotic behavior for large values of e . In this case it
is convenient first to use integration by parts and the identity
k 3 d G(k)
G(k)
1 G(k)(2G(k) kG (k))
+
log
1
=
3 dk k 2
3
k2
k 2 G(k)
(5.7)
( denotes de derivative in relation to k). Taking into account Eqs. (5.5) and (5.6) one can easily
shown that the surface term does not contribute and we are left with
ring
e3 T 4
(T , ) =
(e )
3 (2)2
/2
)
00 (200 k 00
d sin dk
k 2 (e )2 00
0
)
nc (2nc k nc
.
k 2 (e )2 nc
(5.8)
164
ring
/2
(e k))
00 (e k)(200 (e k) k 00
d sin dk
k 2 00 (e k)
0
0
(e k))
nc (e k)(2nc (e k) k nc
.
+
k 2 nc (e k)
e3 T 4
(T , ) =
3 (2)2
(5.9)
For e 1 the integrand is dominated by the asymptotic behavior of 00 and nc given by the
Eq. (5.6). Then we can easily perform the integrals in the previous expression and obtain
e3 4
6T .
e
54
Comparing the previous expression with the SU(N ) result [10]
3
2
2 T 4 16
Ne
ring
SU(N ) = N 1
9
3 4
lim ring (T , ) =
(5.10)
(5.11)
we see that, as in the case of the twoloop contribution, the following relation holds
1 ring
lim ring (T , ) = SU(2) .
(5.12)
3
Let us now analyze the properties of ring (T , ) for intermediate values of e . It is convenient
to define the quantity
e
/2
2
6
(e )2 00 (k)
27
2 (e ) 00 (k)
d sin dk k
+ log 1
I(e ) =
12 (e )3
k2
k2
0
(e )2 nc (k)
(e )2 nc (k)
+
log
1
+
k2
k2
(5.13)
165
(a)
(b)
Fig. 10. The real (a) and imaginary parts (b) of ring , in units of 546 e3 T 4 .
6. Discussion
In this paper we have obtained all the contributions to the free energy up to the leading
nonanalytic terms which arises from the summation of ring diagrams. In the lowest nontrivial
order, namely the twoloop order, previously investigated in Ref. [4], our approach allows to
cast the results in a relatively simple form which gives both the low and high temperature limits in a straightforward way. We have also analyzed the threeloop contributions and identified
the leading high temperature terms. (Threeloop contributions are also relevant to implement
the renormalization scale independence of the free energy, so that a change in the effective running coupling constant e2 (T ) is compensated by changes in higher order contributions starting
at O(e4 ) [12].) The series of leading higher order contributions, proportional to powers of (e )2 ,
has been summed exactly and the result was expressed in terms of a double integral representing
the ring contributions to the free energy as a function of e . Both the twoloop contributions and
166
the sum of the rings have been computed in a general covariant gauge and the gauge independence has been verified for these contributions.
Perhaps the most interesting feature of this analysis is expressed in Eq. (5.15), which shows
that the free energy acquires an imaginary part above a critical temperature. Although this has
been found from our analysis of the behavior of the free energy, it is interesting to notice that the
critical value
3
10
(e )c = e Tc2 =
(6.1)
2
can be related with the solution of the equation which gives a selfconsistent gauge independent
definition of thermal masses through the relation [13]
2
m2A = A (k0 = 0, k)
k =m2 ,
A
(6.2)
where the subindex A in A represents either 00 or nc, as defined in Eq. (4.23). One can
now investigate the solutions of this equation as a function of the parameter (e ).
Let us first consider the electric 00 mode. Using Eq. (4.29) one can easily find that there
is a positive solution for m200 . This solution occurs for asymptotic values of the parameter (e ).
Indeed, using the asymptotic behavior given by Eq. (5.6a) and taking into account the inverse of
the rescaling transformations in (4.13), we obtain
2
m200 = e2 T 2 .
(6.3)
3
It is interesting to note that this noncommutative Debye mass is numerically identical to the result
of the SU(2) theory [3]. This is also a consequence of the correspondence between the large e
regime and the commutative nonAbelian gauge theory, as already manifested in Eq. (5.12). (We
also note that a naive definition of the thermal mass such that m2A = limk0 A (k0 = 0, k)
would be zero in the pure gauge sector of NCQED, because both 00 and nc vanish as k 0,
for fixed values of . However, in higher orders, such a mass would be gauge dependent [3].)
Proceeding similarly with the nc mode, we find that Eq. (6.2) admits negative solutions
for m2nc . This happens in the regime e > (e )c which is when the argument of the logarithm
in Eq. (5.3) becomes imaginary. For arbitrary values of e one would have to solve Eq. (6.2)
numerically. However, when e is close to its critical value, we can obtain the analytic solution
of Eq. (6.2) using nc the expression given by Eq. (5.5b), multiplied by e2 . Performing a simple
calculation, the nontrivial solution for the mass can be written as
2 2
21
e T
7e2 2
(e )c 2
2
1
T T2 ,
mnc
(6.4)
(e )
30 c
4 2
(e )2c
where Tc2 is defined in Eq. (6.1) (again, we have taken into account the inverse of the rescaling transformations in (4.13)). This solution shows explicitly that m2nc becomes negative when
T > Tc .
The contrasting behavior of the two modes is a direct consequence of the fact that while
00 is always negative, nc is always positive (see Fig. 9). It is remarkable that despite all the
similarities between NCQED and commutative YangMills theories there is such an important
difference as far as the stability is concerned. Of course the important feature here is the presence
of the nonzero static transverse mode nc , which would vanish in theories like QCD. This is
consistent with previous findings from the analysis of the dispersion relations in noncommutative
SYM theories at finite temperature [14].
167
In conclusion, we have shown that above a certain critical temperature, the system may undergo a phase transition which is induced by the noncommutative magnetic mode. This behavior
will also occur even in the presence of fermions, because the contributions from the fermion
loops are the same as in the commutative theory. Since in commutative QED the static magnetic
mode is absent and the magnetic mass vanishes to all orders [3], it follows that the fermions
will not modify the behavior of the magnetic mode which is responsible for the instability of the
system. It is interesting to note that the negative value of the squared magnetic mass m2nc is reminiscent of the Jeans mass M 2 GT 4 which arises in thermal quantum gravity [15,16]. Such
a mass leads to the appearance of an imaginary part in the free energy, which may be related to
the decay rate of the quantum metastable vacuum.
Acknowledgements
This work was supported by FAPESP and CNPq, Brazil. J.F. and F.B. would like to thank
Ashok Das, D.G.C. McKeon and J.C. Taylor for many helpful discussions.
Appendix A. Threeloop graphs
In this appendix we will present the results for the graphs in Figs. 5 and 6 as well as some
details of the calculation. The first step of this calculation is rather straightforward involving
only a direct use of computer algebra (all the calculations in this appendix as well as in the
next, have been performed using the computer algebra package HIP [17]). There are extra technical difficults compared with similar calculations in QCD, which are mainly associated with the
trigonometric factors characteristic of NCQED. As usual, in order to obtain the simplest possible
expressions for the threeloop graphs we have made use of kinematic identities in such a way
that the dot products ki kj (ki represents any combination of momenta) are reduced, whenever
possible, to quadratic terms such as ki2 and kj2 . The next step consists in identifying which momentum shifts can be done so that one can combine the momenta ki2 and kj2 as a single one,
say ki2 . These shifts, however, are restricted to the ones which preserve the trigonometric factors
or reduce two or more trigonometric factors to a single one. This method is explained bellow in
the case of the most complex graphs (c) and (l), both containing four threephoton interaction
vertices. Computation for graphs (h) is also detailed.
We shall employ the following notation:
pq
,
r = p + k,
s = q + k,
t = p q,
s(p, q) sin
(A.1)
2
T1 s2 (k, p)s2 (k, q),
T2 s(k, p)s(k, q)s(r, s)s(p, q),
T3 s(k, p)s(k, q)s(p, s)s(q, r).
(A.2)
Before presenting the details, let us list the results. Denoting the contributions of each graph in
Figs. 5 and 6 by Aa Al , we obtain
1
1
2 4
Aa = 12(d 1) e T1 2 2 2 2 + 2 4 2 2 ,
(A.3a)
p q k r
k p q
Ab = 4e4 T1
d(d 1)2
,
k 4 p2 q 2
(A.3b)
168
Ac = e4 T1
(p q)2
(6 11d d 2 )
,
4(2d 3) 4 2 2 2 2 +
k p q r s
p2 q 2 r 2 s 2
1
1
Ad = 4(d 1)e4 T1 2 4 2 2 2 2 2 2 ,
k p q
k p q r
(p q)2
1
4
Ae = 2e T1 2 4 2 2 2 2 + 2 2 2 2 ,
k p q r s
k p q r
1
1
Af = e4 T1 8 4 2 2 6(d 2) 2 2 2 2
k p q
k p q r
1
(p q)2
(d + 2) 2 2 2 2 + 8(2d 3) 4 2 2 2 2 ,
p q r s
k p q r s
2
Ag = 2
e4 T1
,
2
p q 2r 2s2
T1 + T2
p (q + s)
Ah = (d 1)e4 18 2 2 2 2 + 4(2d 5)T2 2 2 2 2 2 ,
p q r s
k p q r s
d(d 1) 4
Ai = 2 2 2 2 2 e (T1 + T2 ),
p q r s
1
p (q + s)
4
Aj = e T2 2 2 2 2 2 2 2 2 2 2 ,
p q r s
k p q r s
1
p (q + s)
4
Ak = 2e T2 2 2 2 2 + 2 2 2 2 2 2 ,
p q r s
k p q r s
23
20d
p (q + s)
4
Al = e T2 2 2 2 2 2(2d 5) 2 2 2 2 2 .
p q r s
k p q r s
(A.3c)
(A.3d)
(A.3e)
(A.3f)
(A.3g)
(A.3h)
(A.3i)
(A.3j)
(A.3k)
(A.3l)
(A.5)
169
With this procedure the function F can be expressed in terms of the much smaller set
2 2
ki kj , (ki , kj ) {r, s, p, q, k},
plus terms containing p q and (p q)2 , which cannot be simplified. There is a total of 5!/
(5 2)!2! + 5 = 15 pairs of the form ki2 kj2 and certain ones are identical inside the integrand after
a change of variables. That is, they yield the same final denominator and trigonometric factor.
Next we analyze the equivalent pairs. For example, a denominator of the form
1
k 4 p2 q 2
is generated by the binomials p 2 q 2 , p 2 s 2 , q 2 r 2 , r 2 s 2 . That is, the first three terms can be made
equivalent to the fourth via the substitutions p r, q s, p r, q s, respectively.
For these pairs we obtain, respectively, the coefficients
21
3
+ 11d 2d 2 ,
d + 2d 2 ,
2
2
which combine to
1
4(6 5d) 4 2 2 .
k p q
21
+ 11d 2d 2 ,
2
3
d + 2d 2
2
(A.6)
Next we have the equivalent terms k 2 r 2 , k 2 s 2 , k 2 p 2 , k 2 q 2 , which are multiplied by the coefficients
39
21
21
39
+ 2d 2 17d,
+ 2d 2 17d,
2d 2 5d,
2d 2 5d.
2
2
2
2
Combining these terms, we obtain
4(15 11d)
1
.
k 2 p2 q 2 r 2
(A.7)
(A.8)
And p 2 r 2 , q 2 s 2 , with
(2d 3)(d 6).
(A.9)
(p r 2 )2 + (q 2 s 2 )2
(q s 2 )2
1
(2d 3)(d 6) 4 2 2 2 2 .
(2d 3)(d 6)
2
k 4 p2 q 2 r 2 s 2
k p q r s
(A.10)
Expanding the numerator of (A.10)
2
2
q s 2 = q 4 + s 4 2q 2 s 2 2s 2 s 2 q 2 = 2s 2 k 2 k q 2s 2 k 2 ,
we see that this term can be combined with that of (A.7), yielding
2
12 7d 2d 2
.
k 2 p2 q 2 r 2
(A.11)
170
The coefficient of the term k 4 of the function F comes multiplied by the coefficient
3
5d d 2 .
2
Finally we have the term p q
(A.12)
2p q + p 2 r 2 + q 2 s 2 + k 2
k 4 p2 q 2 r 2 s 2
(p q)
p q + q 2 s 2
4(2d 3)2 (p q)
+ 2(2d 3)2 2 2 2 2 2 .
k 4 p2 q 2 r 2 s 2
k p q r s
2(2d 3)2 (p q)
(A.13)
p q(q 2 s 2 )
1 (s 2 q 2 k 2 )(q 2 s 2 )
1 (q 2 s 2 )
1 (q 2 s 2 )2
=
+
4
4 k 2 p2 q 2 r 2 s 2 4 k 4 p2 q 2 r 2 s 2
k 4 p2 q 2 r 2 s 2
k 4 p2 q 2 r 2 s 2
4
2
2
4
2
2
1 q 2q s + s
1 s (s q 2 ) 1 (k 2 + 2k q)
1
1
=
=
.
4
2
2
2
2
4
2
2
2
2
4
2
2
2
2
2
4 k p q r s
2k p q r s
2 k p q r
2 k p q 2r 2
2
In the last term of (A.13) we proceed as in (A.25), obtaining, p q k4 , and we finally have
the amplitude Ac
1
1
Ac = e4 T1 4(6 5d) 4 2 2 + 4d 2 38d + 42 2 2 2 2
k p q
k p q r
2
(p q)
1
4(2d 3)2 4 2 2 2 2 + 6 11d d 2 2 2 2 2 .
(A.14)
k p q r s
p q r s
The result for other graphs similar graphs (e), (f) and (g) can be obtained using the same procedure. The computation of graphs (a), (b) and (d) is straightforward.
A.2. The graph (h)
This graph as well as the graph (i) are such that the output expression contains combinations
of the three trigonometric factors. More specifically, Ah = e4 (hT1 T1 + hT2 T2 + hT3 T3 ), where
hT1 = 9k 2 ,
hT2 = 2k 2 5k p 5k q 10p q + d k 2 + +2k p + 2k q + 4p q ,
hT3 = k 2 + 5k p + 5k q + 10p q d k 2 + +2k p + 2k q + 4p q .
After the substitutions k p = 12 [r 2 p 2 k 2 ], k q = 12 [r 2 p 2 k 2 ] we obtain
d 1
2 9T1 + (d + 2)T3 (d 7)T2 k 2 + 4(2d 5)(T2 T3 )p q
k 2 p2 q 2 r 2 s 2
+ (2d 5)(T2 T3 ) r 2 + s 2 p 2 q 2 .
(A.15)
Ah = e4
hk 2 2
(d 1)
9T1 (d 7)T2 + (d + 2)T3 .
2
2
2
2
p q r s
171
(A.16)
(d 1)
T3 + T2
T2 + T3
hk 2 = 2 2 2 2 2 9T1 + (d + 2)
(d 7)
2
2
p q r s
(d 1)
(d 1)
= 9 2 2 2 2 [2T1 + T2 + T3 ] 18 2 2 2 2 [T1 + T2 ].
(A.17)
p q r s
p q r s
The terms proportional to p q give
hpq 4
(d 1)(2d 5)
(d 1)(2d 5)
(T2 T3 )p q 4 2 2 2 2 2 T2 p (q + s).
2
2
2
2
2
k p q r s
k p q r s
(A.18)
q
k 2 p2 q 2 r 2 s 2
(d 1)(2d 5)
= 2(T2 T3 ) 2 2 2 2 2 p 2 + q 2
k p q r s
hrem (T2 T3 )
pr
(d 1)(2d 5)
0,
4(T2 T3 )
k 2 p2 r 2 s 2
T1 + T2
p (q + s)
4
Ah = (d 1)e 18 2 2 2 2 + 4(2d 5)T2 2 2 2 2 2 .
p q r s
k p q r s
(A.19)
(A.20)
(A.21)
2
1
4
Al = e T2 6(d 1) 2 2 2 2 + 2 2 2 2
k p s t
p q r s
k2
(2d 5)
2p 2
2r 2
t2
+
+
+
+
3
p2 q 2 r 2 s 2 t 2 k 2 q 2 r 2 s 2 t 2 k 2 p2 q 2 s 2 t 2 k 2 p2 q 2 r 2 s 2
2
(2d 5)
2
2
2
2
+ 2 2 2 2+ 2 2 2 2+ 2 2 2 2+ 2 2 2 2
2
2
2
2
3
p r s t
k p r t
k p r s
k p q s
p q s t
1
1
+ 2 2 2 2+ 2 2 2 2
k p q t
k r s t
e4 (L1 + L2 + L3 ),
(A.22)
172
pq
1
L2 = 4(2d 5)T2 2 2 2 2 2 2 2 2 2 ,
k p r s
k p q r s
(2d 5)
2
2
2
2
2
L3 = T2
+ 2 2 2 2+ 2 2 2 2+ 2 2 2 2+ 2 2 2 2
2
2
2
2
3
p r s t
k p r t
k p q s
k p r s
p q s t
L1 = 18T2
(pr),(qs)
1
2
2
k p q 2t 2
1
+ 2 2 2 2
k r s t
kp
1
(2d 5)
1
1
1
1
= 2T2
+ 2 2 2 2+ 2 2 2 2+ 2 2 2 2+ 2 2 2 2
2
2
2
2
3
p r s t
k p r t
k p q s
k p r s
p q s t
kp
1
+ 2 2 2 2
k r s t
(pt),(ks)
1
(2d 5)
2
1
2
= 2T2
+
+ 2 2 2 2 + 2 2 2 2
3
p2 r 2 s 2 t 2 k 2 p2 r 2 t 2
k p q s
p q s t
pq
1
(2d 5)
2
1
2
+
+
+
= 2T2
3
p2 r 2 s 2 t 2 k 2 p2 r 2 t 2 k 2 q 2 s 2 t 2 p2 q 2 s 2 t 2
(pr),(qs)
(2d 5)
3
2
= 2T2
+
+
3
p2 r 2 s 2 t 2 k 2 p2 r 2 t 2
= 2T2 (2d 5)
(q=s,k=r)
1
p2 q 2 r 2 t 2
(pt),(ks)
1
1
+ 2 2 2 2
p2 r 2 s 2 t 2
k p r t
q=s
(2d 5)
(2d 5)
(2d 5)
= 4 2 2 2 2 T2 = 4 2 2 2 2 T2 = 4 2 2 2 2 T3 .
p r s t
k q r p
k p r s
(A.23)
Finally
Al = e4 (L1 + L2 + L3 )
(2d 5)
pq
(d 1)
= e4 18 2 2 2 2 T2 + 4 2 2 2 2 (T2 T3 ) 4(2d 5) 2 2 2 2 2 T2
p q r s
k p r s
k p q r s
pq
(d 1)
= e4 18 2 2 2 2 T2 4(2d 5) 2 2 2 2 2 T2 ,
p q r s
k p q r s
(A.24)
173
(T2 T3 )
= 0.
k 2 p2 r 2 s 2
Notice that, in the absence of the trigonometric factor (i.e., as in QCD) the second term of
(A.24) can be simplified by the relation
pr
p q
pq
kq
=
,
k 2 p2 q 2 r 2 s 2
k 2 p2 q 2 r 2 s 2
k 2 p2 q 2 r 2 s 2
1
pq
=
2
2
2
2
2
2
k p q r s
kq= 12 s 2 12 q 2 12 k 2
kq
2
k p2 q 2 r 2 s 2
1
=
4
1
p2 q 2 r 2 s 2
(A.25)
Using the same procedure in the noncommutative case one finds a more complicated expression,
however, this step is necessary to the final result.
Shifting p r and q s, T2 T3 we write the symmetrization
pr
qs
pr,qs
pq
1 T2 p q
p T2 q
T2 p q
T2 p q
T2 2 2 2 2 =
+ 2 2 2 2+ 2 2 2 2+ 2 2 2 2
2
2
2
2
4 k p r s
k p r s
k p r s
k p r s
k p r s
1
pq
r q
ps
r s
= T2 2 2 2 2 T3 2 2 2 2 T3 2 2 2 2 + T2 2 2 2 2 .
4
k p r s
k p r s
k p r s
k p r s
pq
1 1
p (q + s)
1
T2 2 2 2 2 =
+
,
2 2 p2 q 2 r 2 s 2 k 2 p2 q 2 r 2 s 2
k p r s
p (q + s)
23 20d
Al = e4 2 2 2 2 2(2d 5) 2 2 2 2 2 T2 .
p q r s
k p q r s
(A.26)
(A.27)
(A.28)
The other Mercedes graphs, (j) and (k) can be dealt with in the same fashion.
Appendix B. Gauge parameter (in)dependence
B.1. Twoloops
From the structure of the graphs in Fig. 2 we expect a polynomial of third degree in . The
highest power comes only from the graph containing three photon propagators. A direct calculation shows that such a contribution is identically zero. This can be understood as a consequence
of the Ward identity for the cubic vertex
p q
p p p 2 q q q 2
k
(B.1)
= 2ie sin
2
and the structure of the gauge dependent part of the photon propagator in Eq. (2.8a). Then the
vanishing of the 3 terms follows from the transversality of (B.1).
174
The quadratic terms involve contributions from the first and the second graphs in Fig. 2.
Performing shifts and using relations like (A.5) the individual contributions of these two graphs
can be reduced to
(p q)2
1
( 2 )
,
G1 = e2 s 2 (p, q)
(B.2)
p4 q 4
p2 q 2
(p q)2
1
( 2 )
2 2
2 2
G2 = e s (p, q)
(B.3)
p4 q 4
p q
which clearly cancels each other. This simple cancellation only occurs because we have the
freedom to perform shifts inside the regularized integrals.
The contributions which are linear in are generated by all the three graphs in Fig. 2. Proceeding similarly as in the case of the quadratic contributions, the individual graphs can be reduced
to the following expressions
s 2 (p, q)
3 s 2 (p, q)
+
2d
,
2 p2 q 2
p2 q 2
2
s 2 (p, q)
s (p, q)
( )
2
G2 = e 2 2 2 2d 2 2 ,
p q
p q
( )
G1 = e2
( )
G3 =
e2 s 2 (p, q)
2 p2 q 2
( )
( )
(B.4)
(B.5)
(B.6)
( )
so that G1 + G2 + G3 = 0. This concludes the verification of the gauge parameter independence of the free energy at twoloop order.
B.2. Threeloops
Unlike the twoloop contributions, in the present case one does not expect a gauge parameter
independence. However, there must be some cancellation, so that the residual dependence will
combine with the perturbative corrections to the coupling constant, which are not included in the
present analysis.
Indeed, the straightforward computer algebra calculation shows that the 6 contribution (from
the graphs (c) and (l)) as well as 5 (from graphs (a), (c), (h) and (l)) vanish. This is also a direct
consequence of the Ward identities like (B.1) (as well as the analogous identity involving the
three and four photon vertices) and holds at the integrand level. We hale also verified that at order
4 every graph with four or more photon propagators gives a nonzero contribution, except (f).
We have taken advantage of the fact we are working in a spacetime of d dimensions in
order to organize the gauge dependent contributions according to the power of d. Our calculation
shows that the highest power of d is three, so that an arbitrary amplitude can be written as
A3loops = e4 A0 + dA1 + d 2 A2 + d 3 A3 .
(B.7)
Since d can be arbitrary we can study gauge parameter dependence for each power of d as well
as . Usually the highest power of the dimension will have less terms to deal with, so that in
order to understand how these terms combine we will study two simple cases, namely order
and 2 in the gauge parameter and d 2 in the dimension.
175
B.2.1. Order ( 2 , d 2 )
Only the graphs (a), (b) and (h) have nonzero contributions at ( 2 , d 2 ). They all share the same
trigonometric factor, T1 , which is invariant under q p and shifts q q k, p p k.
Keeping in mind well just use these transformations, we will omit e4 T1 , so that the resulting
amplitudes are
(k 2 + 2p k)2
,
(k 2 )3 q 2 p 2 (p + k)2
1
( 2 ,d 2 )
= 4 2 2 2 2 ,
Ab
(k ) q p
A(a
2 ,d 2 )
( 2 ,d 2 )
Ah
=4
= 16
(k 2 + 2q k)2 (k 2 + 2p k)2
.
(k 2 )4 q 2 p 2 (k + p)2 (k + q)2
(B.8)
(B.9)
(B.10)
A(a
2 ,d 2 )
( 2 ,d 2 )
Ah
=8
=8
(p + q)2
(p + q)2 (p + k)2
1
4
4 2 4.
(p 2 )4 q 2
(p 2 )4 q 2 k 2
(p )
(B.11)
(B.12)
(B.13)
Discarding the terms which are odd functions of the momenta (keeping in mind that T1 is an even
function) we have
A(a
2 ,d 2 )
1
.
(p 2 )2 q 2 k 2
(B.14)
Ad
1
Ad (p, q, k) + Ad (p, q, k) + Ad (p, q, k) + Ad (p, q, k)
4
1
= 4 2 2 2 2 .
(p ) q k
(B.15)
Finally
A(a
2 ,d 2 )
+ Ac(
2 ,d 2 )
( 2 ,d 2 )
+ Ad
= 0.
(B.16)
B.2.2. Order ( , d 2 )
The graphs (a), (b), (c) and (h) contribute to order ( , d 2 ). In this case we have to deal with a
more complicated trigonometric factor for the graph (h), which has the amplitude
( ,d 2 )
Ah
= 2e4
We shall only be concerned with transformations which preserve the delta functions, namely
r p,
s q,
{s r, q p}.
(B.18)
176
Keeping in mind that we will be only using these transformations we will always write s(r, p) =
s(k, p), etc. Shifting r p in the first term in brackets and his multiples, we have
( ,d 2 )
Ah
=4
(B.19)
This expression has a trigonometric factor which is invariant by the transformations (B.18). Omitting e4 T3 and using the delta functions we obtain
( ,d 2 )
Ah
kq
pk
8 2 2 2
p 2 q 2 k 2 (p + k)2 (q + k)2
p q k (p + k)2 (q + k)2
(k q)(k p)
1
16 2 2 2 2
4 2 2
(p ) q k (p + q)2 (p + k)2
q k (p + q)2 (p + k)2
1
1
8 4 2 2 + 8 4 2
,
k q p
k q (p + k)2
8
(B.20)
Ad
pq
(k 2 )3 q 2 (p + k)2
+8
pq
(k 2 )3 (p + k)2 (q
+ k)2
pq
pq
1
+8 2 3 2 2 8 2 3
8 4 2 2.
2
2
(k ) q p
(k ) (k + q) p
k q k
(B.21)
Aa( ,d ) 8
(B.22)
(B.23)
1
1
1
1
2
8
+
T
8
.
A( ,d ) = T1 8 2 2 4 8
2
k q p
((p + k)2 )2 q 2 p 2
k 4 q 2 p2
k 4 q 2 (p + k)2
(B.24)
Next we make shifts to write each denominator as one,
metric factors, which have been altered
1
,
k 4 p2 q 2
(B.25)
S2 s(k p, q) s(k, p) ,
(B.26)
(B.27)
(B.28)
(B.29)
177
(B.30)
This calculation shows how the gauge parameter dependence of the threeloop graphs can be
reduced to a smaller class of terms. The remaining gauge parameter dependence must cancel
when we take into account the corrections to the coupling constant. However, we have verified
that all these contributions are subleading in the high temperature regime, so that only the gauge
invariant ring contributions survive.
References
[1] M.R. Douglas, N.A. Nekrasov, Rev. Mod. Phys. 73 (2001) 977;
R.J. Szabo, Phys. Rep. 378 (2003) 207.
[2] M. Hayakawa, Phys. Lett. B 478 (2000) 394;
M. Chaichian, A. Tureanu, hepth/0604025.
[3] J.I. Kapusta, Finite Temperature Field Theory, Cambridge Univ. Press, Cambridge, 1989;
M.L. Bellac, Thermal Field Theory, Cambridge Univ. Press, Cambridge, 1996;
A. Das, Finite Temperature Field Theory, World Scientific, New York, 1997.
[4] G. Arcioni, M.A. VazquezMozo, JHEP 0001 (2000) 028.
[5] J.I. Kapusta, Nucl. Phys. B 148 (1979) 461.
[6] I. Affleck, Phys. Rev. Lett. 46 (1981) 388.
[7] I.S. Gradshteyn, M. Ryzhik, Tables of Integrals, Series and Products, Academic Press, New York, 1980.
[8] F.T. Brandt, A. Das, J. Frenkel, Phys. Rev. D 65 (2002) 085017.
[9] F.T. Brandt, J. Frenkel, D.G.C. McKeon, Phys. Rev. D 65 (2002) 125029;
F.T. Brandt, A. Das, J. Frenkel, S. Pereira, J.C. Taylor, Phys. Rev. D 67 (2003) 105010.
[10] P. Arnold, C. Zhai, Phys. Rev. D 50 (1994) 7603;
P. Arnold, C. Zhai, Phys. Rev. D 51 (1995) 1906.
[11] W. Fischler, E. Gorbatov, A. KashaniPoor, S. Paban, P. Pouliot, J. Gomis, JHEP 0005 (2000) 024.
[12] J. Frenkel, Alberto Saa, J.C. Taylor, Phys. Rev. D 46 (1992) 3670.
[13] A.K. Rebhan, Phys. Rev. D 48 (1993) 3967.
[14] K. Landsteiner, E. Lopez, M.H.G. Tytgat, JHEP 0106 (2001) 055.
[15] D.J. Gross, M.J. Perry, L.G. Yaffe, Phys. Rev. D 25 (1982) 330;
A. Rebhan, Nucl. Phys. B 351 (1991) 706.
[16] F.T. Brandt, B. CuadrosMelgar, F.M. Machado, Phys. Rev. D 67 (2003) 125006.
[17] A. Hsieh, E. Yehudai, Comput. Phys. 6 (1992) 253.
Abstract
ep
The significance of NNLO (3loop) QCD contributions to the flavor nonsinglet sector of F2 and F2ed
has been studied as compared to uncertainties (different factorization schemes, higher twist and QED contributions) of standard NLO (and LO) QCD analyses. The latter effects turn out to be comparable in size
to the NNLO contributions. Therefore the minute NNLO effects are unobservable with presently available
(precision) data on nonsinglet structure functions.
2006 Elsevier B.V. All rights reserved.
1. Introduction
In a recent publication [1] a nexttonexttoleading order (NNLO) QCD analysis of
ep
F2 (x, Q2 ) and F2ed (x, Q2 ) in the flavor nonsinglet sector was presented. Here our purpose
is to study the significance of the NNLO contribution as compared to other, possibly important,
contributions such as redundant terms in the NLO analysis (which arise when the NLO evolved
parton distributions are multiplied by the coefficient function), higher twist and QED contributions to NLO, and effects due to choosing different factorization schemes. In Section 2 we
present the relevant theoretical expressions required for our analysis, and Section 3 contains the
quantitative results. Our conclusions are summarized in Section 4.
2. Theoretical formalism
ep,d
4
1
5
1 +
ep
+
F2 = xuv + xdv xqNS,8
+ xqNS,3
,
9
9
18
6
5
5
+
,
F2ed = x(uv + dv ) xqNS,8
18
18
179
(1)
(2)
+
where d = (p + n)/2 and qNS,3
= uv dv . For x < 0.3 one analyzes the genuine NS combination
p n
F2
1
ep
2
1 +
2 F2 F2ed = x(uv dv ) + x(u d) xqNS,3
,
3
3
3
(3)
+
= uv dv + 2(u d) since sea quarks cannot be neglected for x smaller than
where now qNS,3
about 0.3. For definiteness we adopt for d u the choice [1,2]
x(d u) x, Q20 = 1.195x 1.24 (1 x)9.1 1 + 14.05x 45.52x 2
(4)
at Q20 = 4 GeV2 which gives a good description of the DrellYan dimuon production data [3],
but plays a marginal role in our analysis.
At NLO(MS) the nth Mellin moments of the above NS combinations of valence parton distributions, for brevity denoted by v + , symbolically evolve according to the wellknown expression
(see, e.g., [4,5])
(0)
v + Q2 = 1 (a a0 )R1 (a/a0 )PNS /0 v + Q20 ,
(1)+
(5)
(0)
where R1 = PNS /0 (1 /02 )PNS and a = a(Q2 ) s (Q2 )/4 with a0 = a(Q20 ). The moments of the above NS structure functions F2NS are then given by
(1)
F2NS Q2 = 1 + aC2,NS v + Q2
(6)
and this expression is commonly compared with experiment. Inserting v + (Q2 ) from (5) into
(1)
this equation one observes a redundant O(a 2 ) contribution, i.e., a(a a0 )C2,NS R1 , which in
fact belongs to a NNLO analysis and is assumed to be small at NLO. If, however, one chooses to
work to a NNLO (3loop) accuracy, such a redundancy at NLO might become significant as compared to the full NNLO contribution. This will be investigated quantitatively below. Similarly,
the choice of a factorization scheme, other than the MS scheme used thus far, might imply larger
differences than additional NNLO contributions in the MS scheme. For example, in the deep
inelastic scattering (DIS) factorization scheme [4,6] the Wilson coefficient in (6) is absorbed into
(1)
:
the parton distributions, i.e., into their evolutions in (5) with R1 R1DIS = R1 C2,NS
(0)
+ 2
vDIS
Q = 1 (a a0 )R1DIS (a/a0 )PNS /0 v + Q20
(7)
+
at NLO.
and, instead of (6), F2NS = vDIS
Taking into account QCD 3loop NNLO s3 effects, one also has to consider QED LO contributions which are of comparable size [7]. The latter can easily be implemented by changing
the nth moments of the NLO valence input distribution qv (Q20 ), qv = uv , dv , in (5) according to
qv Q20
with
1
137
(a/a0 )
1
4 2 P
0
1
1
exp
P qv Q20
40 a a0
(0)
(8)
180
(9)
(0)
(10)
(2)+
Convenient expressions for the relevant 2loop Wilson coefficient C2,NS can be found in [5]
(2)+
(Eq. (A.2)) and for the 3loop splitting function PNS in [8] (Eq. (4.22), which can be easily
Mellintransformed using [9]). The strong coupling now evolves according to da/d ln Q2 =
2=0 a +2 where 0 = 11 2f/3, 1 = 102 38f/3 and 2 = 2857/2 5033f/18 +
325f 2 /54, which refers to the MS renormalization scheme, and f denotes the number of active
flavors. Here the redundant O(a 3 , a 4 ) contributions in (10) turn out to be marginal and do not
influence our (fit) results. Furthermore, the running coupling a(Q2 ) is appropriately matched at
Q = mb = 4.5 GeV and Q = mt = 175 GeV. To obtain F2NS in the DIS factorization scheme at
NNLO, the Wilson coefficient functions in (10) have to be absorbed into the parton distributions,
(2)+
(1) 2
+ (C2,NS
) :
i.e. into their evolutions in (9) with R2 R2DIS = R2 2C2,NS
2
1
+ 2
Q = 1 (a a0 )R1DIS a 2 a02 R2DIS R1DIS
vDIS
2
DIS 2
(0)
(a/a0 )PNS /0 v + Q20 ,
a0 (a a0 ) R1
(11)
+
where R1DIS is as in the NLODIS expression (7) and, instead of (10), we now have F2NS = vDIS
at NNLO.
Since flavor NS structure functions are mainly related to the medium and large xregion, the
relevant kinematic nucleon target mass (TM) corrections are always taken into account according
to [10]
NS
n, Q2
F2,TM
1
NS
x, Q2 dx
x n2 F2,TM
2
3
2 2
j
mN
mN
(n + j )! F2NS (n + 2j, Q2 )
=
+O
,
2
j !(n 2)! (n + 2j )(n + 2j 1)
Q
Q2
(12)
j =0
where higher powers than (m2N /Q2 )2 are negligible for the relevant x < 0.8 region, as can
straightforwardly be shown by comparing (12) with the wellknown exact expression in Bjorkenx space [10].
Despite the kinematic cuts (Q2 4 GeV2 , W 2 ( x1 1)Q2 + m2N 10 GeV2 ) used for our
analysis, we also take into account higher twist (HT) corrections to F2NS via
m2N
+
2
v x, Q 1 + 2 h(x) v + x, Q2
(13)
Q
181
in order to learn whether nonperturbative effects may still contaminate our perturbative analysis.
Here we adopt the ansatz [11]
b
x
h(x) = a
(14)
c .
1x
Notice that the input valence parton distributions v + (x, Q20 ) at LO [v + (Q2 ) =
(0)
(a/a0 )PNS /0 v + (Q20 )], NLO and NNLO in the MS as well as in the DIS scheme in Eqs. (5)(11)
and (13) can and will be different in general.
Finally, Fermi motion and nuclear effects in the deuteron are strongly model dependent and
will therefore not be considered here. They were, however, taken into account in [1,11,12] using
the specific models cited there. Comparing these results with our valence distributions obtained
and to be discussed below, as well as with other results where such effects have not been taken
into account (e.g., [2]), shows that these effects do not change the quality of the QCD fits.
3. Quantitative results
In the present analysis we used the proton and deuteron data of BCDMS [13], NMC [14] and
SLAC [15], as well as the proton data of H1 [16] and ZEUS [17] in the relevant xregions discussed above which amount to 480 data points. The valence distributions have been parametrized
at the input scale Q20 = 4 GeV2 as
xuv x, Q20 = Nu x au (1 x)bu 1 + Au x cu + Bu x ,
(15)
2
ad
bd
cd
xdv x, Q0 = Nd x (1 x) 1 + Ad x + Bd x
(16)
1
1
with the normalizations Nu and Nd being fixed by 0 uv dx = 2 and 0 dv dx = 1, respectively.
The LO, NLO and NNLO fit results without HT contributions are summarized in Table 1. The
standard NLO and NNLO fits refer to the MS scheme according to Eqs. (5), (6) and (9), (10),
respectively. Our fit results for NNLO are compared in Fig. 1 with the data used. Except perhaps
in LO, we obtained equally good and acceptable fits ( 2 /dof) in each perturbative order and
scenario. As has been already noted previously [1,12,18,19], a NNLO analysis in general results
in a slightly smaller s (m2Z ) than in NLO. This is due to the fact that the higher the perturbative
order the faster s (Q2 ) increases as Q2 decreases. In order to compensate for this increase,
a NNLO fit is expected to result in a smaller value for s (m2Z ) than a NLO fit. Notice that
the values of s (m2Z ) obtained in the usual perturbative NLO and NNLO fits in Table 1 are
comparable to the ones in [1,12,18]. Repeating the NLO and NNLO fits in the DIS factorization
scheme improves only marginally the global MS fits ( 2 ), and the QED O() contributions leave
the original NLO(MS) results practically unchanged as evident from Table 1.
On the other hand, the inclusion of higher twist contributions sizeably improves the fits, i.e.,
the value of 2 /dof as can be seen from Table 2 (s (m2Z ) is reduced as expected since the HT
term takes care already of some of the Q2 dependence of the data). In order to illustrate the
relative significance of the (model dependent) HT corrections in (13), we have performed a fit
for Q2 4 GeV2 and one for Q2 10 GeV2 , denoted by HT(10) in Table 2. Clearly, HT effects
become less important when the lower cut of Q2 is increased and the value of 2 increases,
eventually approaching the larger values obtained by purely perturbative fits. Nevertheless it is
remarkable that the fits for Q2 4 GeV2 and Q2 10 GeV2 are not significantly different as
will be illustrated below (Fig. 2). For illustration we also show in Fig. 1 the NNLO results with
182
Table 1
ep,d
Parameter values of the QCD fits in various perturbative orders based on all nonsinglet data for F2 (x, Q2 ). The
parameters of the input valence distributions refer to (15) and (16). The analysis for the DIS factorization scheme is
based on (7) for NLO and on (11) for NNLO. The QED contribution to NLO is taken into account according to (8)
2 /dof
Parameter
s (m2Z )
((4) /MeV)
au
Au
ad
Ad
bu
Bu
bd
Bd
cu
Nu
cd
Nd
0.574
9.258
0.600
0.330
3.290
7.164
4.952
3.726
0.823
1.790
0.100
1.742
0.98
0.128
(196.0)
0.600
9.900
0.600
3.294
3.364
8.504
5.163
8.303
0.667
1.521
0.181
0.542
0.93
0.112
(222.6)
0.600
9.801
0.581
0.030
3.004
9.010
4.797
4.458
0.855
2.101
0.878
1.264
0.91
0.113
(228.3)
NLO QED
0.600
9.883
0.599
3.223
3.361
8.496
5.161
8.037
0.667
1.523
0.186
0.555
0.93
0.112
(217.1)
NNLO
0.600
9.654
0.600
4.060
3.571
6.810
5.209
4.870
0.599
1.309
0.323
0.657
0.89
0.111
(177.2)
0.587
9.644
0.600
2.004
2.727
9.471
4.787
6.046
0.825
1.941
0.868
1.421
0.89
0.112
(187.2)
LO
NLO
NLO DIS
NNLO DIS
HT effects included for Q2 4 GeV2 . The results for the Q2 10 GeV2 cut are not shown,
since they are very similar to the ones for the Q2 4 GeV2 cut (dashed curves).
The actual relative size of our results can best be seen by comparing the various fit results with the pure QCD NLO fit, i.e., by considering the following ratios, depicted in Fig. 2,
which are defined as follows. The effect of NNLO contributions can be visualized via rNNLO =
ep
ep
F2,NNLO /F2,NLO with the nominal NLO structure function given by (6) and the NNLO one
ep
(1)
by (10). Similarly, rLO requires the usual F2,LO as given by (5) and (6) with R1 0 and C2 0.
ep,DIS
ep
DIS = F
Furthermore rNLO
2,NLO /F2,NLO illustrates the effects of choosing the DIS factorization
scheme instead of the MS scheme, with the DIS structure function given by (7). Similarly, the
DIS
definition of rNNLO
employs the DIS structure function in NNLO given by (11). Including the
ep
ep,QED
ep
QED
= F2,NLO /F2,NLO . Fig. 2(a)
QED O() contributions to F2,NLO according to (8) results in rNLO
demonstrates that ambiguities of standard NLO MS analyses or additional QED contributions are
comparable in size to NNLO contributions in the relevant medium and large xregion. Only in
183
Fig. 1. Comparison of our NNLO fits with all presently available flavor nonsinglet data [1317] used for our analysis.
The higher twist (HT) contribution is taken into account according to (13) and (14). The NLO fits are very similar and
practically indistinguishable from the ones shown. So is the NNLO HT(10) fit resulting from the cut Q2 10 GeV2 . The
p n
inset shows our NNLO input valence distributions at Q20 = 4 GeV2 . The scales on the left ordinate refer only to F2
p n
where for each fixed value of x we have added the constant in brackets to F2 . The scales on the right ordinate refer to
p
p
F2 and to F2d . The data sets are shown with their normalization factors in parentheses (first entry refers to F2 , second
p
d
entry to F2 ) as obtained in the fit. The ZEUS data for F2 have been shifted to the right by 5% in order to make their
error bars distinguishable from the ones of the H1 data.
the smaller xregion around x = 0.2, the NNLO results are about 1% larger than the NLO ones.
DIS and r DIS are
This difference, however, disappears in the DIS factorization scheme where rNNLO
NLO
comparable and small (about 0.5%). We therefore conclude that the DIS scheme guarantees a
better perturbative convergence than the commonly used MS scheme, except in the very large
xregion (where nonperturbative contributions are uncontrollable anyway). Unfortunately such
minute effects are not testable with presently available precision data for nonsinglet structure
functions which have a typical uncertainty of about 10% in the small and large xregion.
The redundant O(a 2 ) contribution to a NLO analysis, as discussed after (6), turns out to be
ep
marginal: repeating the NLO fit with the redundant term removed from F2,NLO in (6),
(0)
ep,rem
ep
(1)
F2,NLO = F2,NLO + a(a a0 )C2,NS R1 (a/a0 )PNS /0 v + Q20 ,
ep,rem
ep
(17)
rem 1 = F
2
one obtains rNLO
2,NLO /F2,NLO 1 0.001 for all relevant values of Q .
Finally, the relevance of HT effects is illustrated in Fig. 2(b) where the ratios riHT =
ep,HT
ep
F2,i /F2,NLO for the different perturbative orders are shown, with the HT corrections incor
184
Table 2
As in Table 1 but including HT contributions as well according to (13) with the parameters (a, b, c) referring to (14). Fits
using a lower bound Q2 10 GeV2 in (13) are denoted by HT(10), whereas HT refers to Q2 4 GeV2 as stated before
(13). NLO and NNLO always refer to the MS factorization scheme
2 /dof
Parameter
s (m2Z )
((4) /MeV)
au
Au
ad
Ad
a
bu
Bu
bd
Bd
b
cu
Nu
cd
Nd
c
LO HT
0.600
6.926
0.582
7.205
2.180
3.218
4.208
4.882
2.878
0.941
0.397
1.021
0.712
0.956
0.577
0.82
0.118
(121.8)
LO HT(10)
0.600
8.286
0.597
6.163
2.105
3.167
5.214
4.985
1.061
1.000
0.422
0.942
0.750
1.052
0.749
0.88
0.126
(176.6)
NLO HT
0.600
9.897
0.600
4.905
2.198
3.368
8.215
5.383
9.563
1.468
0.508
1.104
0.222
0.467
0.303
0.83
0.104
(136.6)
NLO HT(10)
0.600
9.900
0.598
2.383
2.313
3.429
7.786
5.507
9.824
1.740
0.522
1.126
0.190
0.642
0.284
0.90
0.106
(155.8)
NNLO HT
0.600
9.900
0.566
5.816
1.695
3.571
7.086
5.425
9.475
1.104
0.519
1.116
0.302
0.441
0.432
0.83
0.103
(109.5)
NNLO HT(10)
0.600
9.898
0.599
2.260
1.989
3.622
6.491
5.594
9.119
1.801
0.518
1.099
0.235
0.716
0.252
0.91
0.105
(124.9)
porated according to (13). In general, for each perturbative order the fit results are stable with
respect to different choices for the lower bound on Q2 in (13), i.e., Q2 4 GeV2 and 10 GeV2
denoted by HT and HT(10), respectively. In particular the more relevant NLO and NNLO fit
results are very similar. Despite the fact that these results are about twice as large as the ones
without HT contributions in Fig. 2(a) throughout the whole xregion considered, they are still
much smaller than present experimental uncertainties.
185
Fig. 2. The size of perturbative (a) and nonperturbative (b) uncertainties encountered in various perturbative s orders of
ep
QCD analyses relative to our nominal NLO analysis of F2,NLO (x, Q2 ) which always appears in the denominator of the
ratios r as defined and discussed in the text. The HT contributions to the fits shown in (b) refer to the cut Q2 4 GeV2 ,
whereas HT(10) refers to a cut Q2 10 GeV2 . The typical relative experimental accuracy is illustrated at x = 0.4 by
the vertical bar (1%). At larger and smaller values of x the experimental error increases (e.g., the uncertainty is about
2.5% at x = 0.55, and 10% at x = 0.18). All results are shown for Q2 = 40 GeV2 , but the agreement with data at
Q2 = 4 GeV2 and Q2 = 100 GeV2 is similar.
4. Conclusions
ep,d
The significance of NNLO QCD contributions to the flavor nonsinglet sector of F2 (x, Q2 )
has been studied as compared to uncertainties of standard NLO (and LO) analyses. NNLO
corrections slightly improve the fits to presently available data and imply a better perturbative
convergence in the DIS factorization scheme than in the commonly used MS scheme. However,
186
ambiguities of NLO fits such as the choice of a particular factorization scheme (MS vs. DIS) and
possible higher twist effects as well as QED O() contributions turn out to be comparable in
size to NNLO (3loop) contributions. In particular, nonperturbative higher twist effects play an
important role in obtaining optimal fits (minimal 2 ) which turn out to be rather stable with respect to different choices of the lower bound on Q2 . Their contribution is about twice as large as
purely perturbative uncertainties which are typically less than about 1%. We therefore conclude
that the rather minute NNLO QCD effects in the flavor nonsinglet sector are not observable with
present precision data for flavor nonsinglet structure functions which have sizeably larger errors.
Acknowledgements
This work has been supported in part by the Bundesministerium fr Bildung und Forschung,
Berlin/Bonn.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
J. Blmlein, H. Bttcher, A. Guffanti, Nucl. Phys. B (Proc. Suppl.) 135 (2004) 152.
A.D. Martin, et al., Eur. Phys. J. C 23 (2002) 73.
E866 Collaboration, R.S. Towell, et al., Phys. Rev. D 64 (2001) 052002.
J. Blmlein, A. Vogt, Phys. Rev. D 58 (1998) 014020.
W.L. van Neerven, A. Vogt, Nucl. Phys. B 568 (2000) 263.
G. Altarelli, R.K. Ellis, G. Martinelli, Nucl. Phys. B 143 (1978) 521;
G. Altarelli, R.K. Ellis, G. Martinelli, Nucl. Phys. B 146 (1978) 544, Erratum;
G. Altarelli, R.K. Ellis, G. Martinelli, Nucl. Phys. B 157 (1979) 461.
H. Spiesberger, Phys. Rev. D 52 (1995) 4936.
S. Moch, J.A.M. Vermaseren, A. Vogt, Nucl. Phys. B 688 (2004) 101.
J. Blmlein, S. Kurth, Phys. Rev. D 60 (1999) 014018.
H. Georgi, H.D. Politzer, Phys. Rev. D 14 (1976) 1829.
U.K. Yang, A. Bodek, Phys. Rev. Lett. 82 (1999) 2467;
U.K. Yang, A. Bodek, in: Proceedings of the 6th Int. Workshop on DIS and QCD, Brussels, 1998, hepph/9806458.
S.I. Alekhin, Phys. Rev. D 68 (2003) 014002.
BCDMS Collaboration, A.C. Benvenuti, et al., Phys. Lett. B 223 (1989) 485;
BCDMS Collaboration, A.C. Benvenuti, et al., Phys. Lett. B 237 (1990) 592.
NMC Collaboration, M. Arneodo, et al., Nucl. Phys. B 483 (1997) 3.
L.W. Whitlow, et al., Phys. Lett. B 282 (1992) 475.
H1 Collaboration, C. Adloff, et al., Eur. Phys. J. C 30 (2003) 1.
ZEUS Collaboration, S. Chekanov, et al., Eur. Phys. J. C 21 (2001) 443.
G. Parente, A.V. Kotikov, V. Krivokhizhin, Phys. Lett. B 333 (1994) 190;
A.L. Kataev, et al., Phys. Lett. B 388 (1996) 179;
A.L. Kataev, et al., Phys. Lett. B 417 (1998) 374.
A.D. Martin, et al., Phys. Lett. B 531 (2002) 216.
Abstract
We consider scattering amplitudes in string models in the Regge limit of high energies and fixed momentum transfers with the use of the unitarity in direct channels. Intermediate states are taken in the multiRegge
kinematics corresponding to the production of resonances with fixed invariant masses and large relative rapidities. In QCD such kinematics leads to the BFKL equation for the pomeron wave function in the leading
logarithmic approximation. We derive a similar equation in the string theory and discuss its properties. The
purpose of this investigation is to find a generalization of the BFKL approach to the region of small momentum transfers where nonperturbative corrections to the gluon Regge trajectory and reggeon couplings are
essential. The BFKL equation in the string theory contains additional contributions coming from a linear
part of the Regge trajectory and from the soft pomeron singularity appearing already in the tree approximation. In higher dimensions in addition, a nonmultiRegge kinematics corresponding to production of
particles with large masses is important. We solve the equation for the pomeron wave function in the string
theory for D = 4 and discuss integrability properties of analogous equations for composite states of several
reggeized gluons in the multicolour limit.maren.stein@desy.de.
2006 Elsevier B.V. All rights reserved.
1. Introduction
The derivation of
amplitudes in the Regge regime of
the BFKL equation for QCD scattering
high energies E = s and fixed momentum transfers q = t [1] is based on the fact that gluon
is reggeized in perturbation theory. In the leading logarithmic approximation (LLA) the pomeron
* Corresponding author.
188
singularity in the j plane of the tchannel partial waves appears as a composite state of two
reggeized gluons. The gluon Regge trajectory is known in two first orders of perturbation theory
and the integral kernel for the BFKL equation is calculated in the nexttoleading approximation
[2], which is important for the phenomenological applications [3].
It is reasonable to believe, that the gluon reggeization has a physical meaning even beyond the
QCD perturbation theory, although up to now its Regge trajectory is calculated only at sufficiently
large momentum transfers q where the effective coupling constant is small. For low momentum
transfers we should use nonperturbative methods. For example, one can assume that the gluon
trajectory in this region is approximately linear, as it takes place for the hadron trajectories. The
linearity of the Regge trajectories was an important property in constructing the dual model by
Veneziano [4]. Later a string interpretation of the dual amplitudes was developed [5]. In the
Born approximation the dual hadron models include only particles lying on the secondary Regge
trajectories. The pomeronlike singularity appearing in the open string scattering amplitudes in
one loop approximation was identified with a leading Regge trajectory for the closed sector. Later
it was found that four dimensional string theories meet with difficulties, which were avoided in
their superstring generalizations to spacetime dimensions D = 10 [5]. Now these superstring
models are considered as candidates for an unified theory of all elementary particle interactions
including the gravity. Moreover, all of them are supposed to be various realizations of the same
Mtheory.
In the modern interpretation the closed string sector is associated with the graviton Regge
family rather than with the pomeron singularity,2 and so the pomeron does not directly present
in the string theory. In a line with the Maldacena proposal [7] for the N = 4 superYangMills
model one might expect an appearance of a colorless composite state becoming a graviton in the
t Hooft limit g 2 Nc , where g is a coupling constant and Nc is the number of colors. The
pomeron singularity seems to be a candidate for such graviton state [8]. However, in this paper
we treat the pomeron similar to the case of perturbative QCD where it is a composite state of two
reggeized gluons. Namely, this singularity should appear in the diagrams where two open strings
are exchanged in the t channel. Such Feynman graphs lead to the Mandelstam cut in the j plane
of the t channel partial wave j (t). The sum of contributions from the laddertype diagrams in
the string theory corresponds to the BFKLlike equation. We hope that its string modification
is a reasonable model for nonperturbative effects in the region of small momentum transfers.
Indeed, the string models in extra dimensions can lead to a dual description of gauge theories
including QCD [7].
Note, that in the critical dimensions D = 10 the multiRegge kinematics for intermediate particles in the schannel is not unique even for small coupling constants. Namely, one should take
into account also the production of resonances having large masses at high energies, which leads
in particular to the graviton contribution appearing in one loop. In the last case the imaginary part
of the corresponding nonplanar diagram with the graviton
Regge pole in the t channel appears
from the production of two resonances with masses m s. Below we does not discuss this
problem in details and consider mainly the D = 4 case.
We use the superstring 4d model in the RamondNeveuSchwarz version [5], but the supersymmetry is involved only to remove the tachyon from the spectrum. It is known, that noncritical
string models have difficulties related to the absence of the Smatrix unitarity in higher loops. In
particular, to restore the unitarity in one loop approximation it is needed to introduce an addi2 Note that the Regge asymptotics was investigated also in the pure (super)gravity [6].
189
tional 2D gravity field [5]. This field provides the conformal symmetry on the tree and oneloop
levels, and restores the modular invariance of the oneloop closed string amplitudes [9]. Nevertheless, the modular invariance of higher loop closed string amplitudes remains to be broken
[10]. Thus, the higher loop amplitudes for noncritical string models cannot be constructed in a
selfconsistent way.
The difficulties of noncritical string models are related mainly to the closed string sector. In
the perturbation theory with respect to the closed string coupling constant gcl the contributions
from this sector grow with energy very rapidly s 2 gc2n lnn (s). Such behavior in the case of
hadronhadron interactions is not compatible with the schannel unitarity. So one expects that
once a relevant summation over n being performed, the high asymptotics of the amplitude is
reduced to A s. It is reasonable to omit initially the closed string sector taking into account
also, that gcl is quadratic in the YangMills coupling constant g and for Nc the open
string terms in the amplitude are enhanced comparing to closed string ones. Thus, we consider
here the contributions to the production amplitude only from the open string states in crossing
channels leading to the Mandelstam cuts for the elastic t channel partial wave j (t) in the angular
momentum plane j = 1 + .
In the discussed model the gluon trajectory (t) is given by the perturbative expansion
(t) = t + 1 (t) + ,
(1)
where n (t) g 2n are radiative corrections, and the Regge slope is a reversed square of an
characteristic mass scale. Below the correction 1 (t) g 2 to the trajectory is also taken into
account. This correction is calculated from oneloop diagrams for the scattering amplitude. Oneloop nonplanar diagrams contain also a contribution destroying the unitarity, but the correction
to the Regge trajectory appears only from the planar graphs, where the problem with the closed
string sector does not exist. Providing that t 1, the loop correction in the D = 4 case
has the infrared divergency g 2 Nc ln(t/2 ) which is canceled with the contribution from the
massless particle production. For t 1 the radiative correction to the Regge trajectory has a
complicated form.
An important difference between QCD and the string model is related to the role of intermediate states with relatively large masses: ( )1 M 2 s for produced resonances. These states
are absent in QCD. In the string theory the large mass states are responsible for the appearance
of the graviton contribution to the elastic scattering amplitude in the oneloop approximation.
Further, the impact factors for the reggeonparticle scattering vanish for planar diagrams as a
result of integration over large masses. In particular, it leads to the absence of the Mandelstam
cuts in the color octet channel. In QCD the cancellation of these cuts for the t channel with gluon
quantum numbers is provided by another mechanism related to the socalled bootstrap relations
for scattering amplitudes [11].
The large mass kinematics is responsible also for the additional term in the kernel of the
BFKL equation corresponding to the soft pomeron contribution. It is important, that in the considered string model even in the tree approximation there is a colorless state in the t channel
with vacuum quantum numbers and a positive signature. In upper orders of the perturbation theory its Regge trajectory is renormalized. At small t this state mixes with the Mandelstam cut
constructed from two reggeized gluons. The radiative corrections to its trajectory are calculated
from ladder diagrams in the tchannel. The schannel imaginary part of scattering amplitudes
appears from the intermediate states in the above considered kinematics with relatively large
masses: ( )1 M 2 s for produced resonances. Physically the j plane singularity with the
190
vacuum quantum numbers in the tree approximation corresponds to the soft pomeron which can
exist together with the hard BFKL pomeron.
Similar to the perturbative QCD, we restrict ourselves to the region g 2 Nc ln(s/M 2 ) 1. However, in the string case, the region (t) ln(s/M 2 ) 1 is possible also because the Regge slope
= 1/M 2 has no g 2 smallness. Some important properties of the BFKL equation are related
to this fact. In particular, we obtain that for t g 2 Nc its solution is concentrated near the
saddle point q /2 for the reggeon transverse momenta k . For D > 4 the fluctuations of this momentum are small (k q /2)2 1/( ln s) and therefore the transverse momenta of the
g
emitted gluons are also small k 2 (ln s)1 . In the same time there are no similar restrictions on transverse momenta of the virtual gluons entering in the loop corrections to the gluon
Regge trajectories. It means, that the contribution from the multiple saddle points ki q/2 is
suppressed by the reggeization effects.
The paper is organized as follows. In Section 2 the BFKL approach to the perturbative QCD
is briefly reviewed. In Section 3 the superstring model which will be used later is introduced. In
Section 4 the calculation of the multiRegge asymptotics of production amplitudes is presented.
In Section 5 the BFKLlike equation for the superstring model is derived. Also the vanishing of
the impact factors for planar diagrams is demonstrated. In more details this problem is considered
in Appendix D. In Section 6 the calculation of the BFKL kernel is performed. In Section 7 the
equation for the case D = 4 is discussed. Among other things, it is explained why in the space
time D = 10 the nonRegge kinematics contributes to the Regge asymptotics of amplitudes. In
Section 8 the solution of the BFKL equation at small values of t is constructed. In Section 9 an
algebraic approach to this problem is developed and integrability properties of similar equations
for composite states of several open strings in the multicolour limit including a relation with the
Heisenberg spin model are discussed. Appendices AC contain some details of calculations.
2. BFKL approach in the perturbation QCD
As it was mentioned already, in the perturbative QCD the BFKL pomeron appears as a composite state of two reggeized gluons [1]. The gluon is reggeized as a result of summing radiative
corrections to the Born amplitude A
Born for the colored particle scatteringAB A B in the
Regge kinematics of large energies s and fixed momentum transfers q = t
A(s, t) = ABorn s (t) ,
(2)
(4)
The trajectory contains logarithmic divergencies canceled in the total crosssections with the
contributions from the production of soft gluons. For example, in one loop approximation we
have
2
g2
g2
q 2 + 2
q2
2
1 q =
(5)
d
N
k
N
ln
,
c
c
16 3
(k 2 + 2 )((q k)2 + 2 )
8 2
2
191
where is a gluon mass introduced for the regularization of the infraredly divergent integral. On
the other hand, the amplitude for the production of n gluons with momenta kr in the multiRegge
kinematics
s sr = (kr1 + kr )2 qr2 ,
(6)
s (t2 ) d2
s1(t1 ) d1
gTc2 c1 C(q2 , q1 ) 2
gTc3 c2 C(q3 , q2 ) gTBc B B B ,
t1
t2
(7)
where the effective vertex C(q2 , q1 ) for an emission of the gluon with a definite helicity is
q1 q2
, k1 = q 1 q2 .
k1
Here we introduced the complex coordinates
C(q2 , q1 ) =
qr = qrx + iqr ,
(8)
kr = krx + ikr
(9)
for transverse components qr , kr of gluon momenta. The contribution to the elastic scattering
amplitude from the intermediate state having a gluon with the momentum k1 is proportional to
the expression
C(q2 , q1 )C (q2 , q1 ) + C (q2 , q1 )C(q2 , q1 )
(10)
and contains the pole 1/k1 2 . The integration over k1 cancels the infrared divergency in the gluon
Regge trajectory appearing in the virtual corrections to the production amplitudes.
It is convenient to present the elastic amplitude for the colorless particle scattering in the form
of the Mellin representation
a+i
A(s, t) = is
d
s f (t),
2i
(11)
ai
where f (t) is the tchannel partial wave analytically continued to the complex values j = 1 +
of the angular momentum. The amplitude A(s, t) contains only the contribution from the t channel state with vacuum quantum numbers and the positive signature, corresponding to the
BFKL pomeron. A positive value of the parameter a in the above representation is chosen from
the condition, that all singularities of f (t) are situated to the left from the integration contour.
The tchannel partial wave f (t) can be expressed in terms of the gluongluon scattering
amplitude f (q1 , q2 ; q) integrated with the impactfactors (qi , q qi )
2
d 2 q2 (q2 , q q2 )
d 2 q1 (q1 , q q1 )
f q =
(12)
f (q1 , q2 ; q).
(2)2 q12 (q q1 )2
(2)2 q22 (q q2 )2
The impactfactors of colorless particles vanish at small gluon momenta
(0, q) = (q, 0) = 0,
which leads to an infrared stability of f
equation [1]
(13)
(q 2 ). The partial wave f
f (q1 , q2 ; q) = f0 (q1 , q2 ; q)
g 2 Nc
Hf (q1 , q2 ; q).
8 2
(14)
192
1
1
ln 12 2 1 2 + ln 12 2 1 2 4 (1),
1 2
1 2
(15)
where (x) = (ln
(x)) and we introduced the complex coordinates and momenta
r = xr + iyr ,
r =
,
r
12 = 1 2 .
(16)
The Hamiltonian has the property of the Mbius invariance, which allows us to find its eigenfunctions [22]
m
m
12
12
Em,m (1 , 2 ; 0 ) =
(17)
Q
,
10 20
10
20
where
1
n
1
n
(18)
+ i + ,
m = + i
2
2
2
2
are conformal weights.
The high energy asymptotics of the total crosssection is parametrized by the pomeron intercept
m=
t s
(19)
q k)
fG (k,
1
g2
= 2
Nc
q + 2 8 2
d 2 k q2 + 2 fG (k , q k )
2 k 2 + 2 (
q k )2 + 2
193
2 2
2 + 2
g2
(
q k)
d k k + 2
N
+
c
2 k 2 + 2 (
8 2
q k )2 + 2
G
G
q k)
f (k , q k ) f (k,
(21)
,
2
2
(k k ) +
where the gluon mass is introduced with the use of the Higgs mechanism to regularize the
infrared divergencies.
It is obvious, that in an accordance with the bootstrap requirement the solution of the above
equation corresponds to the Regge pole ansatz
+
q k)
=
fG (k,
1
1
,
q2 + 2 (
q 2)
(22)
where (
q 2 ) is the gluon Regge trajectory.
3. String model
In the string and superstring models the scattering amplitude in the tree approximation satisfies
the duality requirement: namely, the sum over the resonances in the tchannel related to its Regge
asymptotics in the schannel is equal to the (analytically continued) sum of resonances in the sand uchannels:
ci (s) ci (t)
A(s, t, u) = A(s, t) + A(u, t) + A(s, u), A(s, t) =
(23)
=
.
t ti
s si
i
The particles with squared masses equal to ti and integer spins j = ji lie on the linear Regge
trajectories
j = j0 + t,
(24)
where j0 and are their intercept and slope, respectively. The slope is universal for all
excitations of the open string. For the closed strings it is equal to /2. As for intercepts, in
the critical dimensions D = 26 for the bosonic string and D = 10 for the superstrings, they are
integer or halfinteger numbers. In particular, for the intercepts of the leading bosonic Regge
trajectories, corresponding to the massless vector (V ) particlegluon and tensor (T ) particle
graviton we have respectively
j0V = 1,
j0T = 2.
(25)
We put j0V = 1 also for the D = 4 model to leave the gluon on the trajectory. The graviton is
absent in this case, instead one has a nonphysical cut in the j plane.
The Regge asymptotics of A(s, t) in the dual models appears as a result of summing over
the poles in the schannel. Really at large s the contributions s k with integer values of k are
canceled and we can substitute approximately the sum over i by the dispersion integral
1
A(s, t)
0
ds
A(s , t),
s s
It agrees with the Regge asymptotics A(s, t) (s)j (t) providing that
A(s, t) s j (t) .
In the Born approximation there are only stable particles in the intermediate state, but with taking into account loop corrections these particles acquire the widths due to their decay into lower
194
mass states. As a result,
A(s, t) has the like singularities only for a finite number of stable
states and the amplitude is a smooth function for large values of s. The function A(s, t) s 1+ t
can be expanded in the series over the parameter t and one can interpret the corresponding
term of the expansion s(ln s)n ( t)n /n! as a contribution from the production of n particles in
a multiRegge kinematics. In QCD such a nonperturbative contribution (t) t to the Regge
trajectory could appear from the integration region k 2 2QCD in the loop corrections of the type
of (5). In this case the common factor t would lead to the linearity of the trajectory at small t .
To begin with, let us consider the Born amplitude for the tachyontachyon scattering amplitude in the bosonic string theory
A(s, t) = g 2
((s))
((t))
,
((t) (s))
(t) = 1 + t,
where for simplicity we omitted the ChanPaton factors. Asymptotically one obtains
lim A(s, t) = g 2 s
(t) ( s) t .
s
(26)
(27)
This result corresponds to the Regge asymptotics described by the reggeized gluon exchange
in the t channel. For other colliding particles there are additional factors depending on their spins.
They are related to different residues for the corresponding Regge pole. Note, that the effective
vertices for reggeized gluon interactions in QCD were obtained also from the string amplitudes
in the limit 0 [12].
For the superstring models the multiplier
((t)) in (26) is replaced by
( t), which
leads to the absence of the tachyon pole at t = 1. At small momentum transfers both models
give the same amplitude for the massless vector boson scattering. Taking, however, into account
that one should sum over other intermediate t channel states for the scattering amplitude with
arbitrary momentum transfers, it is natural to consider only the superstring model where the
tachyon disappears from the spectrum. The Regge limit of the superstring scattering amplitude
is given in the end of this section (see [5]).
As it was said in Introduction, we use the RamondNeveuSchwarz version of the open superstring model. In this model the interaction vertices are calculated in terms of the scalar superfield
X M (z, ) where z is a worldsheet coordinate and is its superpartner. Here M labels the space
time coordinates, M = 0, 1, . . . , (D 1). The vertex V (z, ; k, ) for the emission of a massless
vector boson with its momentum k = {k M } and polarization vector = { M } is given below
[5,13]
V (z, ; k, ) = DXeikX
(28)
where kX kM
and DX M
are scalar products of the corresponding Ddimensional vectors. As usually, the relation k = 0 is valid for polarizations of
external vector bosons. In the contrast to the string tradition, in this paper we use the mostly mi The covariant superderivative D(z, ) appearing in (28) is given
nus metrics ab = a0 b0 a b.
below
X M (z, )
D(z, ) = z + ,
D(z, )X M (z, )
(29)
where is the left derivative in . Note, that the gauge invariance + ck of the amplitudes is valid due to the relation
dz d DeikX = 0.
(30)
195
M
X (z, )X N (z , ) = 2 MN ln(z z ) = 2 MN ln(z z )
, (31)
z z
where MN is the spacetime metrics. The massless boson tree amplitude is obtained by integrating the vacuum expectation of the product of the vertices Vj (28) over (zj , j ). The variables
(zj , j ) are assigned to the vertex for an emission of the boson carrying the momentum kj and
polarization j . In the amplitude we do not integrate over three of coordinates zj using the integrand invariance under SL(2, R)transformation. To conserve this symmetry after fixing the
variables (z(1) , z(2) , z(3) ) one should include in the final expression the additional multiplier
r z(1) , z(2) , z(3) = z(1) z(2) z(1) z(3) z(2) z(3) ,
(32)
leading to an independence of the Born amplitude from the choice of these variables.
Thus, the open string amplitude An ({kj , j }) for the interaction of n massless bosons in a tree
approximation is given by
T(r) A(r)
An {kj , j } =
(33)
n {kj , j } ,
(r)
where each a term corresponds to an ordering of the parameters zj : {(r): zj1 > zj2 > > zjn }
and the sum is taken over the configurations, which are nonequivalent under the cyclic transmutations of indices jr . The coefficient T(r) is the ChanPaton factor [5] for the given color group.
Further, the expression A(r) ({kj , j }) is the integral over (zj , j ) from the vacuum expectation
of the product of interaction vertices multiplied by the factor r(zj1 , zj2 , zjn ):
A(r)
n {kj , j }
n1
(zjs zjs+1 ) dzjs
= g n2 (zj1 zj2 )(zj1 zjn )(zj2 zjn ) (zj2 zj3 )
s=3
dj1 V (zj1 , j1 ; kj1 , j1 ) djn V (zjn , jn ; kjn , jn ) ,
(34)
where (x) is the step function: (x) = 1 for x > 0 and (x) = 0 for x < 0.
Since the correlator (31) is singular at z = z , the integral (34) is convergent only in a certain
(r)
region of invariants constructed from external particle momenta. Each of the terms An in (33)
is calculated for such signs of the invariants where it is convergent, and the result is analytically
continued to their physical values for the production kinematics. The integrand in (34) contains
some contributions which do not contribute to the final result because they are total derivatives in
integration variables. One can make their cancellation explicit using the fact, that the integrand
in the superstring case is invariant under the SL(2, R)SUSY transformation [13,15]:
f (z)
+ (z) 1
, z = z + (z),
z = f (z),
=
(35)
z
2
where (z) and f (z) are given below
(z) = z + ,
az + b
,
f (z) =
cz + d
f (z)
1
=
.
z
cz + d
(36)
196
Here a, b, c, d are bosonic parameters and , are their Grassmann partners. Note, that the
superinterval is transformed in a simpler way
z z = Q1 (z, )Q1 (z , )(z z ),
(37)
where
Q1 (z, ) = D(z, ).
(38)
(39)
(40)
Using relation (31) for the vacuum expectation of the product of vertices (28), one finds finally
A(r)
n {kj , j }
= g n2 (zj1 zjn )(zj2 zjn ) (zj2 zj3 ) dj1 dj2 djn djn
n1
exp 2
jm jm D(zjm , jm ) ikjm jn jn D(zjn , jn ) ikjn
m>n
ln(zjm zjn jm jn ) ,
(41)
where j1 = j2 = 0. The additional Grassmann variables js are introduced for each of the
vertices
V (z, ; k, ) = d e( Dik)X .
(42)
197
So, the tree amplitude is presented by expression (33), where An ({kj , j }) is given in
Eq. (41). Note, that under an anticyclic permutation the amplitude A(r)
n ({kj , j }) receives only
the factor (1)n . Provided that three variables are fixed as zj1 = , zj2 = 1 and zjn = 0, one
can verify this property with the use of transformation (35) for the integrand in (41) choosing the
functions f (z) = zjn1 /z and (z) = jn z(jn1 jn )/zjn1 .
The ChanPaton factor in (33) is given by
T(r) = trace[j1 jn ],
(43)
where s is a color matrix for the corresponding group generator in the fundamental representation. Below we discuss the oriented string, for which s are U (n)matrices in the fundamental
representation. In this case
trace(r s ) = rs ,
(44)
(j )ab (j )cd = ad bc .
j
Hence
r s =
trace(r s j )j .
(45)
Obviously, the tensor d is symmetric in two first indices drsj = dsrj and the structure constants
frsj are completely antisymmetric. Furthermore, f1sj = 0, d11j = dj r1 = 0, and, in addition,
drsj is symmetric in all indices provided that both s = 1, r = 1 and j = 1. Besides, ds1j =
1/nsj when s = 1 and j = 1. We obtain also
n
drsj drsl =
frsj fsrl = [j l j 1 l1 ].
(47)
2
r,s
r,s
(23)
2
t
= 2g s ( t)( s)
(a )(b )
fja ja s fsjb jb ei t + 1
a
b
j a j a j b j b
+ ei t 1
dja ja s dsjb jb ,
+
n
s
(48)
where the color index ji refers to U (n)quantum numbers of the particle carrying the momentum
pi . The spin structure described by the polarization vectors i (i = a, a , b, b ) corresponds to the
198
conservation of helicities for each of colliding particles. Various terms in (48) are associated with
different Regge contributions. Their quantum numbers are the SU(n) singlet (with the signature
+) and two adjoint SU(n)representations (having the dimension n2 1 and the signatures
+ and ). Note, that in QCD the Regge asymptotics of the scattering amplitude in the
Born approximation contains only a contribution with the negative signature, corresponding to an
exchange of the reggeized gluon. The contribution from the positive signature with octet quantum
numbers appears only in upper orders of perturbation theory. Nevertheless, for large Nc the
Regge trajectories with opposite signatures coincide each with another. The degeneracy of these
t channel states is important for the duality symmetry between the colorless composite states
with different signatures [17]. As for the Regge contribution with vacuum quantum numbers,
it also takes place in QCD only in upper orders of perturbation theory and corresponds to the
BFKL pomeron. Its appearance in the superstring model already in a tree approximation can
be considered as a manifestation of the soft pomeron having a nonperturbative nature. When
n = Nc is large, one can neglect this soft pomeron contribution in expression (48).
To derive the above asymptotic behavior of the scattering amplitudes in the superstring theory
we used the relation
(49)
which follows from (46). Note, that the soft pomeron contribution appears also for the Chan
Paton factors corresponding to the colour group O(n). Moreover, in the oneloop approximation
its Regge trajectory does not contain ultraviolet divergencies for n = 32 and D = 10 [18]. In this
model the gluon Regge trajectory 1 (t) is finite is given below (see Appendix B)
1
1 (t) = 8g n
2
1
2
d2 (sin 2 )
0
L2
1 + L1
t
(1 + L1 )1 ,
(50)
n=1
L2 =
n (1 n )2
,
(1 2n cos 22 + 2n )2
(1 n )4
(1 2n cos 22
n=1
+ 2n )2
(51)
If we consider only a contribution of the planar diagram, the low limit of integration over in
the above expression is zero and the gluon Regge trajectory contains the logarithmic divergency
at small , which can be removed by a renormalization of the slope in the Born amplitude
[18]. In a similar way for the SU(n) group we have
1
1 (t) = 8g 2 Nc
0
1
d2 (sin 2 )2
0
L2
1 + L1
t
(1 + L1 )1 1 ,
(52)
199
Fig. 1. The ladder cut determining the BFKL pomeron in the a + b a + b process. The dotted line denotes a reggeon,
the solid one denotes a particle.
200
Fig. 2. The diagram for the calculation of the BFKL kernel. The dotted line denotes massless state; the solid one denotes
the tower of string states.
z1 = , z2 = 1, z8 = 0 and 1 = 2 = 0. In amplitude (33) one should sum over the contributions of the diagrams which cannot be obtained from one configuration by cyclic or anticyclic
transmutations of gluon indices. We should take into account also the ChanPaton factors T (+)
T (+) = trace[r8 r7 r6 r5 r4 r3 r2 r1 + r5 r6 r7 r8 r1 r2 r3 r4 ].
(53)
To calculate the kernel from Fig. 2 only contributions having poles in the invariant k 2 are
essential. There are 16 diagrams of such type corresponding to the configuration
(k1 = q1 , k2 = pa ),
(k5 = q2 , k6 = pb ),
(k7 = pa , k8 = q1 )
(54)
and those obtained by the interchange (kj kl ) inside each of the above brackets, which leads
to the signature factors. As it was pointed out already, pa , pb and pa , pb are momenta of the
initial and final particles, respectively. The momenta q1 , q2 , q1 and q2 correspond to intermediate
particles. Obviously, for the calculation of a discontinuity of the elastic amplitude the relations
q1 = q1 and q2 = q2 are valid, but temporally we distinguish between qi and qi performing
later an analytical continuation in the invariants (k + q1 )2 , (k + q2 )2 , (k q1 )2 and (k q2 )2 to
their physical values. In the multiRegge configuration the momentum k on Fig. 2 obeys some
2 and (k 0 )2 in the c.m. system are assumed
kinematical constraints. Namely, the quantities k 2 , k
to be much smaller than the energy invariants s, s1 and s2 . Integral (41) for each of 16 diagrams
is calculated in the kinematics where it is convergent, and subsequently the result is analytically
continued to the physical region of the reaction.
In expression (41) several polarization structures arise, but only the term
As (1 2 )(3 4 )(5 6 )(7 8 )
contributes to the multiRegge asymptotics of the tree amplitude for Fig. 2
As = g 6 AT (+) (1 2 )(3 4 )(5 6 )(7 8 ),
(55)
where the polarization vector j is associated with the momentum kj and the ChanPaton factor
T (+) is given in Eq. (53). Fixing the parameters as follows: z1 = , z2 = 1, z8 = 0 and 1 =
2 = 0, one obtains from Eq. (41)
A =
BB(1 z3 ) (z7 ) d7 d8 dz7
(zs zs+1 ) dzs ds .
(z3 z4 3 4 )(z5 z6 5 6 )(z7 7 8 )
6
s=3
(56)
201
Similar to the case of bosonic strings [20] one concludes from Eq. (58) that in the multiRegge
kinematics the essential values of parameters are
z3 0,
x/z6 0,
z3 = z4 + x,
z5 = z6 + y,
y/z6 0,
z7 /z6 0.
(59)
B x 2 k3 k4 y 2 k5 k6 z7
(z4 z6 )
exp 2 k2 k3 x + 2 k2 k5 y + 2 k2 (k3 + k4 )z4
z7 x
z7 y
z7
+ 2 k7 (k3 + k4 ) 2 k5 k7 2
2
z4
z4
z6
z7
x
y
+ 2 k7 (k5 + k6 ) 2 k3 (k7 + ke8 ) 2 k5 (k7 + k8 )
.
z6
z4
z6
+ 2 k2 (k5 + k6 )z6 2 k3 k7
(60)
k5 (k7 + k8 ) k6 k.
(61)
k2 k5 < 0,
k7 (k5 + k6 ) < 0,
k3 k7 > 0,
k3 (k7 + k8 ) > 0,
k5 k7 > 0,
k2 (k3 + k4 ) < 0,
k5 (k7 + k8 ) > 0.
(62)
202
x
,
2 (k2 k3 )
y
.
2 (k2 k5 )
(63)
(64)
where the factor G collects all large energy invariants, and A(t5678 , 2 , t3478 , t34 , t56 , t12 , t78 )
depends only on fixed transverse momenta. We define the energy invariants si and sj k as follows
si = (k1 + k)2 2(k1 k),
s23 = 2(k2 k3 ),
s25 = 2(k2 k5 ).
(65)
(66)
tij lm = (ki + kj + kl + km )2 .
(67)
(68)
(69)
where k is the longitudinal component of the momentum k. To simplify the last factor in (64)
one can use the following relations valid in the multiRegge kinematics due to Eqs. (61)
(k5 k7 )(k6 k8 ) (k5 k8 )(k6 k7 )
1
(k5 + k6 )k7 (k6 k5 )k8 + (k5 k6 )k7 (k6 + k5 )k8
=
2
1
= (k6 k8 ) t5678 t56 t78 2
2
(70)
and
1
(k4 k7 )(k3 k8 ) (k4 k8 )(k3 k7 ) = (k3 k8 ) t4378 t43 t78 + 2 .
2
(71)
After redefinition (63) of variables in expression (56) with the use of above simplifications
one can perform the Grassmann integrations. As a result, the last factor in (64) turns out to be
A t5678 , 2 , t3478 , t34 , t56 , t12 , t78 =
dx
dy
z4
dz4
Bs Vb dz6 ,
(72)
where both Bs and Vb depend on integration parameters and external variables. The expression
for Vb is the same as in the bosonic string model, and the prefactor Bs arises due to the superstring modifications. Explicitly,
203
Vb = x t34 y t56 z7
z7 x z7 y
z7 z7
y
2 x
exp x + y + 2 + 2 z4 z6 +
+
+
.
z6 z4
z4 z6
z4
z6
(73)
xyz72
z42 z62
[t5678 t56 t78 ] 2
1
y 2 z72 ( t34 + 1)
yz7 z62
z64
[t3478 t34 t78 ] + 2
1
2 2
x z7 ( t56 + 1)
4 .
xz7 z42
z4
In expression (72) one can perform easily the integration over the variables x and y
A t5678 , 2 , t3478 , t34 , t56 , t12 , t78
=
( t34 )
( t56 )ei( t34 + t56 ) I t5678 , 2 , t3478 , t34 , t56 , t12 , t78 ,
(74)
(75)
, 2, t
I t5678 , 2 , t3478 , t12 , t78 , t34 , t56
2 t12 t78 2
dy dz df ef y f t34 1 y t56 1 z t5678 1
=
(f
+
y)
.
+
y + f z 2 (1 z)
(77)
Here all integrations are performed from 0 to . The above expression is convergent at 2 < 0.
For the factor in a front of the integral we choose the condition arg 2 = . Really the phase
arising in this case, is compensated by a similar phase in (76) and A is real for < 0.
The final result is obtained by summing the contributions of 16 diagrams listed in the beginning of this section, every term being analytically continued from the kinematical region where
the corresponding integral (41) is convergent. Taking into account the spin structure (55) for each
diagram and relations (64), (66) and (76), we derive the following expression for A(f )
204
A(f ) = g 6 s1
Fr1 ,r2 ,j1 (t12 )Fr3 ,r4 ,j2 (t34 )Fr5 ,r6 ,j3 (t56 )Fr7 ,r8 ,j3 (t78 )
(+)
Tj1 ,j2 ,j3 ,j4 I t5678 , 2 , t3478 , t12 , t78 , t34 , t56 ,
(78)
where
Frs ,rl ,j (tsl ) = trace rs rl ei( tsl +1) + rs rl j
(79)
and
(+)
(80)
For the group U (n) the index ri enumerates color states of the particle carrying the momentum ki defined in (54). After an analytical continuation we put s1 = s7 and s4 = s6 . In
a similar way, I (t5678 , 2 , t3478 , t12 , t78 , t34 , t56 ) in (78) is calculated using a similar continuation of expression (77) to the region 2 > 0. This procedure is performed by the replacement
2 2 + i with +0. We have also the condition k (0) > 0, and therefore due to (69), our
prescription corresponds to the Feynman rule for going around the singularity. After the analytic
continuation the factor in front of the integral turns out to be positive.
5. BFKL equation in the string model
Omitting the impact factors of colored particles in the left and right hand sides of the contribution of the diagram Fig 1(b) one can obtain expressions for higher order ladder diagrams by
iterating its interior part. To find the BFKL kernel in the considered string model, one should
calculate from expression (78) its contribution to the t channel partial wave for the scattering of
massless particles. Also oneloop correction 1 (t) g 2 Nc to the trajectory (1) should be taken
into account (see (50)). Thus, tj l in (78) is replaced by the expression tj l + 1 (tj l ). Due to
the presence of nonplanar diagrams, the oneloop correction to the singlet trajectory differs from
that for the octet case. However, assuming that the number of colors is large, below we neglect
this difference.
The contribution to the tchannel partial wave from the diagram Fig. 1(b) is given by the
Mellin transformation in ln s applied to the imaginary part of the amplitude. To calculate it one
should find in Eq. (78) the sum of residues for the poles in the variable k 2 and integrate the
result over a relevant phase volume. Initially we put q1 = q1 , q2 = q2 , r1 = r8 , r4 = r5 ,
1 = 8 , 4 = 5 summing subsequently over indices r1 , r4 and polarization states 1 and 4 . The
poles are situated at k 2 = m, where m is an integer number changing from 0 to .
Below we denote by l, l the transverse momenta of two neighboring reggeons and by q the
total momentum transfer related to the corresponding invariants as follows
l 2 = t12 ,
(l )2 = t34 ,
(q l)2 = t78 ,
(q l )2 = t56 .
(81)
(82)
The multiRegge kinematics implies, that the inequalities s1 /k 2 1 and s4 /k 2 1 are fulfilled. The integration over this region leads to the singularities of the t channel partial wave at
205
= j 1. Here j is the total angular momentum. The contribution F (b) (; q 2 ) to the t channel
partial wave from the diagram Fig. 1(b) including the correction to the trajectory 1 (t) (1) is
given below
F (b) ; q 2
2
2
d D2 l ra ,ra ,r1 ,r4 (q; l)
l 2
(q l)2 ei( l + (ql) )
=
r1 ,r2 ,r3 ,r4
ds1 ds4 (l 2 )((ql)2 )+2
g2
(+)
j 1
T
s
ds
( s1 )
4(2)D1 r1 r2 r3 r4
s2
m=0 s(m)
s(m)
2
2
2
d D2 l Im (q; l, l )( s4 )((l ) )((ql ) )+2 s1 s4 /s k
m
(83)
Here
q 2 = q 2 1 q 2 ,
(84)
and D is the number of spacetime dimensions. The quantity Im (q; l, l ) is the residue of the
pole at k 2 = m in the integral I (t5678 , 2 , t3478 , t12 , t78 , t34 , t56 ) appearing in expression (78)
(see also definitions (81)). Further, s(m) is the low energy cutoff: s(0) = s0 for m = 0 and
2 for m 1. In this case 2 = m + k 2 . We impose the condition s 1 because
s(m) = s0 m
0
m
the production amplitude is known only in the multiRegge kinematics. The cutoff is introduced
to have a possibility to verify that the nonmultiRegge kinematics is not essential. The factor
(+)
Tr1 r2 r3 r4 is presented in Eq. (80) and
ra ,ra ,r1 ,r4 (q; l) = (a a )
2g 2
Fra ,r,r1 (t12 )Fra ,r,r4 (t78 ),
(2)D1 r
2g 2
Frb ,r,r2 (t34 )Frb ,r,r3 (t56 ),
(2)D1 r
(85)
where the function F is defined in (79). Expressions (85) are massless state contributions to
the impact factors. The total impact factor for the planar diagram Fig. 1(b) being the sum of
a tower of string states is equal to zero, which can be verified with the use of the quasielastic
asymptotics of the production amplitude (for more details see Appendix D). Its vanishing ensures
the cancellation of the AmatiFubiniStangelini cuts in the j plane for the planar diagrams.
Once the integration over s, s1 and s4 being performed, Eq. (83) is represented as follows
F (b) ; q 2
2
2
d D2 l ra ,ra ,r1 ,r4 (q; l)
l 2
(q l)2 ei( l + (ql) )
=
r1 ,r2 ,r3 ,r4
g 2 ( )
4(2)D1
d D2 l Tr(+)
1 r2 r3 r4
206
2
2
( s0 ) l (ql) ( s0 ) (l ) (ql )
(86)
Here
I(q; l, l ) = S(0) (q; l, l ) + S (q; l, l )
(87)
and, in turn,
S (q; l, l ) =
2
m
m=1
S(0) (q; l, l ) =
02
l 2 (ql)2 (l )2 (ql )2 +2
+2
Im (q : l, l ),
I0 (q; l, l ).
(88)
In this expression the factor Im (q; l, l ) is the same as in Eq. (83). We remind, that the quantity
2 coincides with 2 on the mass shell k 2 = m (m is an integer number). Expres= m + k
sion (86) is correct only in the domain where it does not depend on the cutoff s0 , which means,
that the power of s0 in this expression should be much smaller than unity. In an accordance with
Eq. (86), it is convenient to present the contribution F (; q 2 ) to the partial wave as a sum of
contributions of the diagrams Fig. 1 starting from Fig. 1(b) written in the form
2
g2
D2
D2
2
d
F ; q 2 =
l
d
l
(q;
l)
l
(q
l)
r
,r
,r
,r
a
1
2
a
4(2)D1 r ,r ,r ,r
2
m
1 2 3 4
i( l 2 + (ql)2 )
Rr1 r2 r3 r4 (; q; l, l )
r ,r ,r ,r (q; l ).
+ ((l )2 ) + ((q l )2 ) b b 3 4
(89)
The particleparticlereggeon vertices contained in Eq. (89) can be extracted from Eq. (48).
Omitting these vertices in Eq. (89), one can verify that the amplitude Rr1 r2 r3 r4 (; q; l, l1 ) for
Fig. 1 obeys the BFKLlike equation
+ l 2 + (q l)2 Rr1 ,r2 ,r3 ,r4 (; q; l, l1 )
(+)
g2
d D2 l I(q; l, l )
= I(q : l, l1 )Tr(+)
+
Tr1 r rr2 Rrr r3 r4 (; q; l , l1 )
1 ,r4 ,r3 ,r2
D1
4(2)
r,r
i( (l )2 +1)
i( (ql )2 +1)
i( (l )2 + (ql )2
e
+ (1) e
+ (1) e
,
,
(90)
where the summation over (, ) is associated with the signatures for the corresponding trajectories having their color group quantum numbers denoted, respectively, by (r, r ). So, , are 0
for a positive signature and 1 for the negative one.
The number of colors is considered to be large and therefore one can neglect the colorsinglet
reggeons in (89). In this case ri , r, r coincide with color indices of the corresponding adjoint
representations. Because and take values 0 and 1, the expression inside the large square
brackets in Eq. (90) is equal to 4. Using Eqs. (46) and (80) one finds that
(+)
/n,
Tr(+)
(91)
Tr(+)
=
T
+
2
= 0.
r
r
rr
1
2
r1 r rr2
1 rrr2
1 r rr2
r
207
(+)
(92)
(1)
where f (q; l, l1 ) and f (q; l, l1 ) are partial waves for the vacuum channel and for the state
belonging to the adjoint representation of the SU(Nc ) group, respectively.
(s)
Using expression (90) one can derive, that the partial waves f (q; l, l1 ) with s = 0, 1 obey
the BFKLlike equation
+ l 2 + (q l)2 1 l 2 1 (q l)2 f(s) (; q; l, l1 )
g 2 N c cs
= I(q; l, l1 ) +
I(q; l, l )f(s) (q; l , l1 ) d D2 l ,
(2)D1
(93)
where c0 = 1 and c1 = 1/2. The integral kernel I(q; l, l1 ) is calculated in the next section.
6. Integral kernel
With the use of (88) one can verify, that the massless state contribution to I (77) is given by
S(0) (q; l, l )/
+ l 2 + (ql)2
2
2
df dy e(y+f ) f (l ) 1 y (ql ) 1
= (l l )2
l 2
(ql)2
y (l l )2 i
f (l l )2 i
l2y
q 2 + (l )2 + (q l )2
[f (l l )2 i]
(q l)2 f
(f + y) .
[y (l l )2 i]
(94)
The integral in the above expression can be written in terms of the Whittaker function
W, ( 2 i) defined as follows
J (a, b; z)
(95)
J (a, b; z) = za+b+1 (a + 1)
(96)
(97)
208
l 2 + (ql)2 (l )2 (ql )2 +2
I t5678 , 2 , t3478 , t12 , t78 , t34 , t56
1
2
2
2
2
df dy dv v + l + (ql) 1 e (y+f +v) z k 1
=
( + l 2 + (q l)2 )
1
V1 (z, f, y; q, l, l ) + + l 2 + (q l)2 1 V2 (z, f, y; q, l, l ) ,
(99)
v
2
2
2
2
2
(ql)2
= f (l ) 1 y (ql ) 1 (1 z) q (l ) (ql ) q1 l q2
q 2 + (l )2 + (q l )2 + l 2 + (q l)2 (l l )2 (1 z)
2
2
2 y
2f
(1 z) (q l)
(1 z) + z ,
l (q l) l
q1
q2
(100)
V2 (z, f, y; q, l, l )
2
2
2 (l )2 (ql )2
= f (l ) 1 y (ql ) 1 (1 z) q
(1 z) (f + y) .
2
(ql)2
q1 l q2
(101)
q2 = y + f z (1 z) i,
0,
(102)
where +0. In integral (99) the residue in the pole at k 2 = m depends on m only through the
exponent exp[m(f + y + v)] multiplied by the derivative zm1 V(i) (z, f, y; q, l, l )/(m 1)!
calculated at z = 0. Therefore after summing the residues over m we obtain
em(f +y+v) zm1 V(i) (z, f, y; q, l, l )/(m 1)!
m=1
z=0
(f +y+v)
= Vi e
, f, y; q, l, l Vi (0, f, y; q, l, l ).
209
(q; l, l )
(103)
S (q; l, l )/
=
1
2
2
df dy dv v + l + (ql) 1
( + l 2 + (q l)2 )
2
e (ll ) (y+f +v) V1 e(f +y+v) , f, y; q, l, l V1 (0, f, y; q, l, l )
1
+ l 2 + (q l)2 1
v
(f +y+v)
V2 e
, f, y; q, l, l V2 (0, f, y; q, l, l ) .
+
(104)
1
2
2
2
df dy dv v + l + (ql) 1 e (ll ) (y+f +v)
=
2
2
( + l + (q l) )
q 2 (l )2 (ql )2
2
2
f (l ) 1 y (ql ) 1 1 ef yv
l 2
f + yef yv 1 ef yv i
(ql)2
y + f ef yv 1 ef yv i
B ef yv , f, y; q, l, l V1 (0, f, y; q, l, l ) (l l )2 V2 (0, f, y; q, l, l ) ,
(105)
where
B(z, f, y; q, l, l )
= (l l )2 (f + y) + l 2 y + (q l)2 f
z(f + y)
l 2 fy(1 z)
+ (q l )2 + (l )2 q 2 1
1z
f + yz (1 z) i
(q l)2 fy(1 z)
.
y + f z (1 z) i
(106)
Really the leading contribution to (105) arises from the region of small integration variables.
In particular, it results in a pole at = q 2 , as well as in a Mandelstam cut term. To find the main
part of (105) we cut from below the integration variables in Eq. (105) by a parameter 1. Then
from Eq. (106), one can obtain, that the leading contribution to S is given by the expression
S (q; l, l )/
2 + (ql)2
e l
[ q 2 (l )2 (q l )2 ]
( + l 2 + (q l)2 )
210
2 1
df dy dv v f (l )
2 1
y (ql )
2 (l )2 (ql )2 1
(v + f + y) q
1
2
,
(l ) (q l )2 ( + l 2 + (q l)2 )
(107)
where the pole term arises from two last terms in (106).
To calculate the integral (107) the integration region is divided into 6 domains: v > f > y,
v > y > f , f > v > y, f > y > v, y > v > f and y > f > v. In the first domain we replace
initially y fy and then f vf . As a result, the vdependence of the integrand turns out
2
to be v + q . Integrating it over v we observe the pole at = q 2 . The similar procedure is
performed in each of the rest domains. As far as, in addition, the expression + l 2 + (q l)2
is implied to be small 1/ ln s, the factor exp[ l 2 + (q l)2 ] in (107) should be replaced by
unity. For the same reason
( + l 2 + (q l)2 ) =
(1 + + l 2 + (q l)2 )/( + l 2 +
(q l)2 ) 1/( + l 2 + (q l)2 ). Using these simplifications expression (107) is given
below
S (q; l, l )/
2
1
q l 2 (q l)2 q 2 (l )2 (q l )2
2
( + q )
F (l )2 , (q l )2 ; q 2 (l )2 (q l )2
+ F 1 q 2 , (q l )2 ; q 2 (l )2 (q l )2
+ F 1 q 2 , (l )2 ; q 2 (l )2 (q l )2 ,
(108)
where
1
F (a, b, c) =
1
df
dy f a+b1 (1 + f + fy)c1 y b1 + y a1
n,m=0
(c)
.
(c m n)
(m + 1)
(n + 1)(a + m)(b + n)
(109)
One can verify that at small momenta q 2 l 2 l 1 the first term in the large square
brackets of Eq. (108) gives the main contribution
S (q; l, l ) =
sing
[ q 2 l 2 (q l)2 ][ q 2 (l )2 (q l )2 ]
.
( + q 2 )(l )2 (q l )2
(110)
Expressions (108) and (110) are correct in a neighborhood of the pole and of zeros of the numerator with the deviations being 1/ ln s Nc g 2 . As far as the numerator does not vanish at
+ l 2 + (q l)2 = + l 2 + (q l )2 = 0, it contributes to both the Mandelstam cuts
and the pole at = q 2 .
The pole at = q 2 corresponds to the soft pomeron which exists already in the
Born expression (48) for the elastic amplitude. Relatively large masses 1 M 2 s
of produced resonances contribute to this pole. Therefore in the box diagram Fig. 1(a)
211
we expect a pole of the second order from the integration over large masses of two intermediate schannel resonances. This second order pole appears as a result of the perturbative expansion of the pomeron Regge pole over the oneloop correction 1 (t) g 2 .
In the twoloop approximation, corresponding to Fig. 1(b), we should have the third order pole with the residue proportional to 12 (t). In this diagram, apart from the pole (110)
there is a product of two pole singularities 1/( + q 2 ) from the integration over the
large masses of resonances produced in the fragmentation regions of initial particles. In the
multiRegge kinematics one obtains also the poles 1/( + l 2 + (q l)2 ) and 1/( +
l 2 + (q l )2 ) leading after the integration over l and l to the Mandelstam cuts (we put here
). Because the residue of the pole (110) in the BFKL kernel is small due to
l = k and l = k
the smallness of the expressions in the square brackets, it cancels approximately the neighboring
poles depending on l and l and therefore one can attempt to extract from the contribution for
Fig. 1(b) the third order pole being the second order term in the expansion of the soft pomeron
pole in 1 (t).
Indeed, let us present the numerator of the pole in Eq. (110) in the form
2
q l 2 (q l)2 q 2 (l )2 (q l )2
= + l 2 + (q l)2 + (l )2 + (q l )2
+ q 2 q 2 + l 2 + (q l)2 + l 2 + (q l )2 .
(111)
Then the second term in the righthand side of this equality, killing the pole 1/( + q 2 ) in
(110), contributes only to the Mandelstam cuts. As for the first term in (111), it corresponds
to the second term of expansion for the soft pomeron pole. Indeed, its numerator cancels the
neighboring propagators for Mandelstam cuts. Therefore the corresponding integrals over the
relative rapidities ln s12 and ln s23 are convergent for the large invariants s12 and s23 . So, we
should calculate these integrals exactly without simplifications corresponding to the multiRegge
kinematics. It is plausible, that as a result of such calculation the pole in expression (110) together
with additional poles 1/( + q 2 ) from two impact factors would reproduce the total oneloop
correction 12 from Fig. 1(b) in the second order expansion of the pomeron pole.
The first term in (111) is important also for a cancellation of the singularities in (110) at
(l, l ) 0 and (l, l ) q leading to a convergence of the corresponding integrals over the multiRegge region. In addition, it has a nontrivial functional dependence containing both poles and
cuts in . So, for the investigation of the BFKL equation in the D = 4 model we use the whole
expression (110) without neglecting the soft pomeron pole. Simultaneously, we add a piece from
the nonmultiRegge kinematics.
For a general case of the ladder Fig. 1 one can perform a decompositions similar to (111)
for each kernel. The contributions appearing from the first terms in the righthand sides of (111)
correspond to the particles produced in a nonmultiRegge kinematics. The form of production
amplitudes in this region cannot be extracted from our above results. Probably this contribution
corresponds to a geometric progression appearing from an expansion of the soft pomeron pole in
1 (t).
Presumably one can represent the partial wave as follows f (q 2 ) as
f q 2 = f(p) q 2 + fmr q 2 ,
(112)
where the first term corresponds to the soft pomeron contribution in the form of the geometrical
progression and the term fmr (q 2 ) results from the multiRegge kinematics. In principle there
can be a more complicated situation with an interference between the Regge pole and cut.
212
S(0) (q : l, l ) =
q2
l2
(q l)2
+
+
.
(l )2 (q l )2 (l l )2 (l )2 (l l )2 (q l )2
(115)
Expression (115) coincides with the corresponding result [11] in QCD. The massive state term
in (114) is expected to vanish at t 0. So, the radiative correction to the gluon trajectory for
small momentum transfers l and q l also can be approximated by the QCD expression (5). As a
result, the BFKL equation (93) for the vacuum channel at D = 4 and q 2 g 2 Nc is drastically
simplified
+ l 2 + (q l)2 f(0) (q; l)
g 2 Nc
= (q; l) +
S(0) (q : l, l )f(0) (q; , l )
8 3
213
(q l)2
1
l2
(0)
+
(q;
,
l)
d 2l
f
(l l )2 [(l l )2 + (l )2 ] [(l l )2 + (q l )2 ]
[ q 2 l 2 (q l)2 ][q 2 (l )2 (q l )2 ] (0)
g 2 Nc
f (q; , l ) d 2 l ,
+
8 3
( + q 2 )(l )2 (q l )2
(116)
where the contribution from Fig. 1(a) is also taken into account.
In Eq. (116) we performed a relevant subtraction of the Regge trajectory contribution to obtain
(0)
the integral kernel in the BFKL form (cf. [1]), and the expression for S (q : l, l ) is given in
(115). Eq. (116) differs from the BFKL equation in QCD only by terms linear in squared gluon
momenta at its lefthand side and by an additional pole term 1/( + q 2 ) in the kernel. The
terms l 2 and (q l)2 improve the properties of its kernel at l . As a result, unlike the
QCD case in LLA, Eq. (116) is expected to have a discrete spectrum at nonzero values of q 2 .
Comparing the largel behavior of the left and righthand sides of Eq. (116) we conclude,
(0)
that the linear terms in the gluon trajectories in Eq. (116) lead to a constant behavior of f (q; l)
at l . As a result, the integral
[q 2 (l )2 (q l )2 ] (0)
h (q) =
(117)
f (q; , l ) d 2 l
(l )2 (q l )2
in the last term on its righthand side is divergent. Taking into account, that this term plays role
(0)
of an additional inhomogeneous contribution to Eq. (116) we present f (q; l) in the form
f(0) (q; l) =
g 2 Nc [ q 2 l 2 (q l)2 ]
h (q)
3
8 ( + q 2 )[ + l 2 + (q l)2 ]
+
g 2 Nc h (q) (0)
f (q; l) + f(0) (q; l),
8 3 ( + q 2 )
(118)
(0)
(0)
where h (q) is given by (117), while f (q; l) and f (q; l) are determined from the equation
(0)
(0)
(0)
(below F (q; l) is denoted either by f (q; l) or f (q; l))
+ l 2 + (q l)2 F(0) (q; l)
g 2 Nc
= (q; l) +
S(0) (q : l, l )F(0) (q; , l )
8 3
(q l)2
l2
1
(0)
+
(119)
F (q; , l) d 2 l .
(l l )2 [(l l )2 + (l )2 ] [(l l )2 + (q l )2 ]
(l l )2 [(l l )2 + (l )2 ] [(l l )2 + (q l )2 ]
(0)
(0)
[ q 2 l 2 (q l)2 ]
,
+ l 2 + (q l)2
(120)
(121)
214
Using (118) and (117) one obtains h (q) as the solution of a linear equation
[q 2 l 2 (q l)2 ] (0)
+ q 2
h (q) =
f (q; , l) d 2 l,
l 2 (q l)2
+ q 2 (, q 2 )
where
, q
g 2 Nc
=
0 +
8 3
+
l 2 <2
2
2
[q l (q
[q 2 l 2 (q l)2 ]2
d 2 l ln 2
[ + l 2 + (q l)2 ]l 2 (q l)2
l)2 ] (0)
2
f (q; , l) d l .
l 2 (q l)2
(122)
2
(123)
We subtracted the logarithmic divergency from the second term in the brackets assuming that
subtraction term is added to the quantity 0 determined by the integration region k 2 1. So,
0 depends also on the nonmultiRegge configurations, leading to the renormalization 1 of the
soft pomeron Regge trajectory. This conclusion follows from expression (86) for the production
crosssection, where the kernel dependence from the cutoff s0 is essential, and from our discussion of Eq. (111). It is natural to expect that 0 1. So, the solution of Eq. (116) depends on the
additional parameter 0 . The equation + q 2 = (, q 2 ) allows to find the Regge trajectories. In addition, one can conclude from (123) that (, q 2 ) contains the Mandelstam cuts in the
plane.
In the region g 2 Nc q 2 1 the asymptotic behavior of the scattering amplitude is related
(0)
to singularities of the integral f (q; , l , l1 ) d 2 l near q 2 /2. They appear from the kinematics, in which the solution of Eq. (116) is concentrated at l = q/2. Let us introduce the new
momenta v and v according to the definition
l = q/2 + v,
l = q/2 + v ,
v2 q 2,
(v )2 q 2 .
(124)
2 /2
[ (l l )2 ]+ q
(l l )2
(125)
where l l = v v . The numerator in (125) is different from unity only in the region r 2 =
(l l )2 s0 /s. Due to Eq. (69) for a massless intermediate state the value of r 2 in the multiRegge kinematics s1 , s2 > s0 1/ is restricted by the condition r 2 s02 /s.
However, according to the generalized Gribov theorem the gluon production amplitude for the
momenta r 2 1/ is also large in the quasielastic regions s1 1/ and s2 1/ and equals
to the elastic amplitude multiplied by a bremsstrahlung factor (see for example [21]). Therefore
the integral over r 2 is not bounded from below by s02 /s being infraredly divergent. As usually,
this divergency is canceled with the contribution from the virtual corrections proportional to the
gluon Regge trajectories. Thus, we substitute by unity the numerator in (125) and represent the
massless contribution to the gluon trajectory correction as follows
g 2 Nc ln l 2 /2
(q l)2
4v 2
d 2l
l2
+
=
(l l )2 [(l l )2 + (l )2 ] [(l l )2 + (q l )2 ] (v 2 + (v )2 )
4v 2 d 2 l
+
(126)
.
(v 2 + (v )2 )(v v )2
215
Here l and v are related according to Eqs. (124). Performing the expansion in v in the righthand
side of Eq. (93) one can write it as follows
(0)
g 2 Nc 2
2
2
2
ln q /64v f (q; v, v1 )
+ q /2 + 2 v +
4 2
= (q; q/2)
g 2 Nc
2v 2
2
(0)
(0)
2
f
+
(127)
(q;
,
v
,
v
)
(q;
,
v,
v
)
f
1
1 d l .
8 3
(v v )2
(v 2 + (v )2 )
The impact factor in (127) is taken at l = q/2 because it is expected to be a smooth function of
l near l = q/2. For q 2 1 the radiative correction (5) to the gluon trajectory at a small momentum transfer should be replaced by 1 (t) taken at t = q 2 /4. Thus, the final equation valid
for both restrictions g 2 Nc q 2 1 and q 2 1 is obtained from (127) by the substitution
(q 2 ) where
g 2 Nc 2
2
q 2 = + 2 1(m) q 2 /4 +
(128)
.
ln
q
/4
8 2
The corresponding quantities are defined in Eqs. (1) and (114). Note, that the infrared divergency
at 0 in the last term is canceled with a similar divergency in the righthand side of Eq. (126)
at v v.
8. Solution of the equation at small momentum transfers
At q = 0 the integral kernel of the BFKL equation for the string theory at D = 4 is nonsingular at small momenta. In this case one can expect that for the t channel partial wave the cut
at = 0 disappears, and instead of a fixed singularity of f (q 2 ) in the plane there are only
Regge poles. Here we demonstrate this phenomenon in the case of small values of q 2 , where
there exists an analytic solution of the equation in the D = 4 string theory. The pole contribution
to the kernel corresponding to the soft pomeron will be neglected.
In the domain of relatively small q2
q2 g 2 Nc
(129)
one can divide the region of possible values of k2 into two subregions 2 q2 and 2
(g 2 Nc )1 , where = 12 . In the first subregion one can use the conformal (Mbius) invariance
and the eigenfunction in the mixed representation coincides with the Fourrie transformation in
the c.m. coordinate 0 from the function Em,m (1 , 2 ; 0 ) (17). Its asymptotics at small 2 has
the form [22]
m
1m
q , )
m + eim,m (q ) 1m
,
Em,m (
(130)
2
where
e
im,m (
q)
= (1)
q
4
4i
q
q
n
(m + 12 )
(m + 12 )
(m + 32 )
(m + 32 )
(131)
(132)
216
(133)
(134)
where
ei() = 24i
2 (1 + i)
( 12 i)
( 32 i)
2 (1 i)
( 12 + i)
( 32 + i)
(135)
On the other hand, in the region 2 k2 (g 2 Nc )1 one can put q = 0 and after the redefinition of the wave function and its argument
= k3 (z), z = ln k2
(
q , k)
(136)
the BFKL homogeneous equation in the string model can be written as the Schrdinger equation
E = H ,
g 2 Nc
E,
4 2
H = HBFKL (i/z) + ez ,
(137)
where
HBFKL () = (i + 1/2) + (i + 1/2) 2(1),
4 2
.
g 2 Nc
(138)
The analogy with the Schrdinger equation is especially fruitful in the diffusion approximation, where
HBFKL (i/z) = 4 ln 2 14 (3)(/z)2
(139)
ez
has the form of the nonrelativistic kinetic energy. The potential energy
grows rapidly at
large positive z and therefore the wave function should tend to zero in this region
z/2
.
e
lim (z) exp 2
(140)
z
14 (3)
For z the potential energy vanishes, which agrees with a possibility to neglect the string
effects at small k2 . In the momentum representation
(p) =
(141)
where p = i/z, the BFKL equation is reduced to the equation in finite differences
E HBFKL (p) (p) = (p i).
(142)
The function (p) can have the singularities (poles) only in the upper semiplane. It is analytic in
the lower semiplane to provide a rapidly decreasing behavior of (z) at z +. The positions
of the poles is given below
pr = p0 + ir
(r = 0, 1, 2, . . .),
(143)
217
(144)
(145)
(146)
up to a periodic function satisfying the relation 0 (p) = 0 (p + i). We should substitute this
function by a constant
0 (p) = const,
(147)
because in an opposite case for p i the wave function does not decrease sufficiently
rapidly due to the additional factors exp(2ip). Indeed, for 0 (p) = 1 the normalization
integral
ln 2
2 / p 2 E+4
2
14
(3)
(148)
dp (p) =
dp
E+4 ln 2
E+4 ln 2
sinh
p
+
sinh
p
14 (3)
14 (3)
is convergent at p . Moreover, for 0 (p) = 1 the wave functions (p) with different E
are orthogonal. Note, that the integrand (148) contains the poles. After their appropriate regularization it leads to the function (E E ) in the orthonormality conditions.
Let us go to the z representation
i0
(z) =
i0
dp ipz
(p).
e
2
(149)
For large positive z the contour of the integration over p should be shifted in the lower semiplane
up to the saddle point situated at
E + 4 ln 2
E + 4 ln 2
g 2 Nc
+ ip + i
+ ln 7 (3)
z = ip i
14 (3)
14 (3)
2 2
g 2 Nc
.
ln (ip)2 7 (3)
(150)
2 2
We can estimate (z) by the value of the integrand in (149) at this point
2 2
ipz
z/2
(z) e
(p) exp 2
e
7 (3)g 2 Nc
(151)
(152)
218
7 (3)g 2 Nc
2 2 k2
i
E+4 ln 2
14 (3)
7 (3)g 2 Nc
2 2 k2
i
E+4 ln 2
14 (3)
(153)
By comparing this result with expressions (134) and (135) for small in the intermediate
region /(g 2 Nc ) 2 1/
q 2 we obtain the quantization of the Regge trajectories
Er + 4 ln 2
7 (3)g 2 Nc
2
(154)
ln
= 2(r + 1/2), r = 0, 1, 2, . . .
14 (3)
2 2 q2
for n = 0 and small
E + 4 ln 2 = 4
0
1.
g 2 Nc
(155)
For comparatively large energies E in the diffusion approximation one can use the semiclassical approximation near the turning point z = z0 , where
ez0 = E + 4 ln 2,
4 2
,
g 2 Nc
(156)
E + 4 ln 2
(157)
The Fourrie transformation to the zrepresentation can be performed with the use of the saddlepoint method
3/2 2 E + 4 ln 2
(z) exp i(z)
i
3
14 (3)
4
2 E + 4 ln 2
+i
,
+ exp i(z)3/2
(158)
3
14 (3)
4
where z = z z0 . Therefore in the diffusion approximation of small the wave function at
z equals
(z) ei/4
7 (3)g 2 N
2 2 k2
i
E+4 ln 2
14 (3)
+ ei/4
7 (3)g 2 N
2 2 k2
i
E+4 ln 2
14 (3)
(159)
(160)
219
(161)
and the corresponding quantized energies can be obtained from the relation E = HBFKL () (see
(138)).
To derive an exact solution of the BFKL equation for small q2 let us introduce the new
variables
x = 2 l 2 ,
x = 2 (l )2 ,
x1 = 2 l12 .
(162)
c=
g2N
4 2
1
1
f (x )
x
f
(x)
dx ,
2
2
x x 
x x 
x
x + 4x
(163)
(0)
Here f (x) (x)/ x is F (0; l)/x averaged over the angle between l and l , and (x) is
(0; l)/x averaged over . We expect that (0; l) 0 at x 0 and (x) is finite at x = 0.
The above BFKL equation differs from that in QCD [11] by the presence of an additional term
proportional to x in its lefthand side. As in the QCD case, we search the solution in the form of
the Mellin transformation
i
f (x) =
(x) 1/2 C( )
d
,
2i
= i.
(164)
(x) 1/2 1 ( )
d
.
2i
(165)
To obtain an equation for C( ) one collects the terms proportional to x . For the contribution
xf (x) the integration contour should be moved to the line = 1 and therefore the function
C( ) cannot have any singularities inside the strip 1 < < 0. If this condition is fulfilled,
we have
C( 1) = 1 ( ) cb( ) + C( ),
(166)
where
b( ) = ( + 1/2) + ( + 1/2) 2(1)
(167)
ln
.
2i
(168)
220
(169)
where C( ) C( ( )). The calculation of C( ) is reduced to the known mathematical problem of finding a function satisfying the requirement, that its discontinuity is proportional to the
same function. Let us define an auxiliary function (, 1 ) being a solution of the homogeneous
equation
( 1, 1 ) = cb( ) + (, 1 )
(170)
with 1 being an arbitrary subtraction point, where = 1. Note, that the sign in the righthand
side of Eq. (170) is opposite in comparison with the sign in front of the corresponding term in
Eq. (166).
The explicit expression for such function is given below
i
sin (1 ) ln[cb( ) + ] d
(, 1 ) = exp
sin ( 1 ) sin ( ) 2i
(171)
where it is implied that < 0 and 1 < 0. At > 0 the result is obtained by an analytic continuation of (171) from the region < 0. Furthermore, it is implied, that the solution
for < 0 = (g 2 Nc ln 2)/ 2 can be derived also by an analytic continuation from the region
> 0 , where the argument of the logarithm has two zeros situated on the imaginary axes and
pinching the integration contour at 0 .
The integral over is convergent at large since from (167) one obtains
ln b( ) ln ln 

(172)
at
.
Let us show, that indeed expression (171) is a solution of Eq. (170). The pole at =
is situated to the left of the integration contour and can pinch only the right singularity of the
logarithm situated at the zero of its argument. The pole at = 1 being to the right from the
contour pinches with the left singularity of the logarithm. It means, that the function (, 1 )
has no singularities in the strip 1 0. To verify that solution (172) satisfies Eq. (170) it
is enough to note that after the shift 1 the pole at = + 1 of the integrand moves to
the point = which was earlier to the left from the integration contour. The initial and final
expressions differ each from another by an additional term in the exponent. This term is obtained
by taking the residue in the pole at = . As a result, relation (170) is fulfilled.
It is useful to investigate the positions of zeroes and poles of (, 1 ). Both of them are
obtained due to pinching the poles 1/ sin ( ) with the singularities of the logarithm situated
at zeros and poles of its argument [cb( ) + ]. The poles are situated at = (n + 1), where
(+)
n = 0, 1, 2, . . . . The zeros are situated between these poles. We denote their position by m for
()
()
()
m > 0 and m for m < 0. It is obvious, that m  < n  for m < n. The function
(, 1 ) has zeroes at = m() r, where r is an integer or zero for m = 1, 2, . . . and r = 0
()
for m = 0. Indeed, due to the above discussion = 0 is not a singularity of the exponent.
Furthermore, (, 1 ) has zeros in the right halfplane at = n + 1/2, where n is an integer
(+)
or zero. The poles are situated in the right halfplane at = m + n and in the left halfplane
()
at = (n + 3/2) for n = 0, 1, 2, . . . . Similar to the case 0 the point = 1/2 does not
221
corresponds to a singularity of the exponent. In the above discussion we used the relation
sin (1 )
= cot ( 1 ) cot ( ).
sin (1 ) sin ( )
(173)
(174)
( , 1 )1 ( ) d
.
2i( , 1 )[cb( ) + ] sin ( )
(175)
In an agreement with general arguments the continuation of the partial wave in the complex
plane from the integer points is performed from the region > 0 . Similar to (171) in (175)
the conditions < 0 and 1 < 0 are assumed to be fulfilled and the expression in the region
> 0 are obtained by an analytic continuation. It can be written in the equivalent form
i
C( ) =
i
1 ( )( , ) d
2i[cb( ) + ] sin ( )
(176)
C( ) 1/( 0 ),
(177)
where
the omitted factor has no singularity at small . Thus, at
x 0 the solution is
in
both
points
(178)
(179)
(180)
222
At < 0, as it was discussed above, describes the wave function of the particle with an energy
equal to c, which is rejected from the potential barrier ez . In this case z = ln x, and i is
the momentum of the colliding particle.
According to (176), the function C( ) in (164) determining the solution of Eq. (163) is given
below
a
(0 )
C( ) =
(0 + + 1)
+i
i
a 1 ( )
( + 0 ) d
.
2i
(0 + 1) sin ( )
(181)
In (181) it is implied that < 0, and so that the pole at = is twisted with the right side.
At > 0 the result is obtained by an analytic continuation in . Furthermore, in (164) the
pole at = 0 is on the righthand side from the integration contour. It is solely the solution at
0 > 0 because in this case (180) determines a function of x increasing at x . At 0 < 0
the solution is not unique because (180) might be added to (181). In an agreement with general
arguments one should chose the solution which is an analytical continuation of the solution (181)
to the region 0 < 0. The result is presented by Eq. (181) where the integration contour is defined
in an accordance with these arguments.
9. Heisenberg spin model and integrability
To investigate the region of q2 g 2 Nc it is convenient to use the conformal invariance
of the BFKL kernel in QCD (see [22]). In the coordinate representation for the wave function
describing the composite state of two reggeized gluons with the impact parameters 1 and 2 we
have the expression [22] (see (17))
m m
12
12
, 12 = 1 2 ,
Em,m (1 , 2 ; 0 ) =
(182)
10 20
10 20
where 0 is the coordinate of the pomeron, r = xr + iyr and r = xr iyr are respectively the
holomorphic and antiholomorphic variables, rs = r s and
n
1
n
1
+ i + ,
m = + i
(183)
2
2
2
2
are conformal weights related to the eigenvalues of the Casimir operators of the Mbius group
m=
2
2 = 12
.
M
1 2
Note, that in (183) the conformal spin n is integer n = 0, 1, 2, . . . and the parameter is a
real number for the principal series of the unitary representations.
2 is related to the generators of the Mbius group M
The operator M
2,
=M
1 + M
M
Mrz = r
,
r
Mr =
,
r
Mr+ = r2
.
r
(185)
(186)
223
For the solution of the BFKL equation in the string theory it is convenient to introduce also the
generators
1 M
2.
N = M
(187)
Together with the operators M
z
M , N = N ,
z
M , N = N ,
z
N , N = M ,
(188)
+
N , N = 2M z .
(189)
We can find the representation of this algebra in the space of the functions Em
m
12
Em (1 , 2 ; 0 ) =
10 20
(190)
as follows
M z Em = (0 0 m)Em ,
M + Em = 02 0 + 2m0 Em ,
02
m(m 1)
Em+1 +
Em1 ,
N Em =
2m 1
(m 1)2
z
m
N 0 N E m =
0 Em1 ,
m1
+
N + 20 N z 02 N Em = 2mEm1
M Em = 0 Em ,
(191)
.
and analogously for the representation of M and N on functions Em
The BFKL integral operator KBFKL is diagonal in the (m, m)representation and its eigenvalue
has the property of the holomorphic separability
g2
(192)
Nc m,m , m,m = m + m ,
8 2
where the holomorphic energies are the following functions of the conformal weights m and m
BFKL =
m = (m) + (1 m) 2(1),
m = (m) + (1 m) 2(1).
(193)
In the case of the string theory in the eigenvalue equation for the pomeron wave function f
in the dimension D = 4 we have the additional contribution KBFKL (neglecting the pole term
from the soft pomeron)
f = Kf,
K = KBFKL + K,
K = p12 p22 .
(194)
N = 1 2 ,
= 1 2 .
(195)
224
In this representation the pomeron wave function in QCD can be obtained by the Fourrie transformation
m
m
2
i qR
q , )
= d Re
,
Em,m (
(R + 2 )(R 2 )
(R + 2 )(R 2 )
1 + 2
R=
(196)
.
2
Let us present the solution of the BFKL homogeneous equation in the string theory as a
superposition of the above functions
f (
q , )
=
Cm,m (
q )
(m)
(m)Em,m (
q , ).
(197)
n=
q ) to simplify the relations beHere we extracted the factor
(m)
(m) from coefficients Cm,m (
tween them. The operators N and N act on the functions Em,m (
q , )
as follows
m(m 1)
q 2
N Em,m =
(198)
Em1,m ,
Em+1,m
2m 1
4(m 1)2
m(m 1)
q2
E
Em,m+1
N Em,m =
(199)
m,m1 .
2m 1
4(m 1)2
Therefore the function f (
q , )
is a solution of the homogeneous BFKL equation in the string
q ) in (197) satisfy the following recurrent relation
theory if the coefficients Cm,m (
g2
2
q
+
Nc (m + m ) +
q)
Cm,m (
2
8 2
m + 1 m 2 q 2
m 2 m 2
q)
q)
Cm1,m1 (
Cm+1,m1 (
= 2
2m 3 2m 3
2m + 1 2m 3 4
m + 1 m + 1 q 2 q 2
m 2 m + 1 q 2
q) +
q ) . (200)
Cm1,m+1 (
Cm+1,m+1 (
2m 3 2m + 1 4
2m + 1 2m + 1 4 4
By introducing the new function
m
q ) = (2m 1)1 (2m 1)1 (q/2)m q /2 Cm,m (
q)
m,m (
one can write this recurrent relation in a simpler form
2
g2
q
Nc (m + m ) +
q)
(2m 1)(2m 1)m,m (
+
2
8 2
q 2
q ) (m + 1)(m 2)m+1,m1 (
q)
(m 2)(m 2)m1,m1 (
=
2
(m 2)(m + 1)m1,m+1 (
q ) + (m + 1)(m + 1)m+1,m+1 (
q) .
(201)
(202)
One should add to this recurrent relation the information about the asymptotic behavior of the
q ) at large m and m corresponding to k q investigated above. Note, that
coefficients Cm,m (
contrary to the case of small q2 , considered in the previous section, now the eigenfunctions
contain a mixture of states with different conformal spins. Expanding m,m in the basis of the
functions x m x m one can reduce the recurrent relation (202) in the diffusion approximation to a
225
differential equation, which can be solved, for example, by the semiclassical methods similar to
those used in the previous section.
In the case of the colourless state constructed from several reggeized gluons [23] the homogeneous equation for its wave function in the string theory is given in the multicolour limit
Nc below (cf. [14])
E(1 , 2 , . . . , n ) = H (1 , 2 , . . . , n ),
g 2 Nc
E,
8 2
(203)
.
r
(204)
where
(n)
H = HBFKL
+ l2
n
(pr )2 ,
l2 =
r=1
8 2
,
g 2 Nc
pr = i
(n)
(n)
(n)
(n)
hBFKL =
(r,r+1)
hBFKL ,
(205)
r=1
(r,r+1)
(206)
and hnBFKL is the local Hamiltonian for the integrable XXX model [24] with the spins coinciding
with the generators of the Mbius group (185). Really we have two independent spin chains for
holomorphic and antiholomorphic subspaces. The term l 2 in Eq. (204) describes an additional
interaction between these two spin chains because according to (185)
(pr )2 = 4Mr Mr .
(207)
This term violates the Mbius symmetry for H and leaves only its invariance under translations
and rotations. Therefore the eigenvalues of H can depend on q 2 , which leads to the Regge trajectories for composite states of reggeized gluons. We do not know, if the corresponding Heisenberg
spin model is integrable or not. But in the region q 2 g 2 Nc it is possible to apply the integrability of the QCD Hamiltonian for calculating the Regge trajectories. Indeed, as in the previous
section, one can divide the essential momenta in two regions kr2 qr2 and kr2 1/ . In the first
region we can use the integrability of the BFKL Hamiltonian to obtain the wave function of the
composite state. For the leading singularity the integrals of motion are quantized and depend
only on the conformal weights m, m [14,24,25]. Therefore the corresponding energy EBFKL for
this leading singularity is a function of these variables
EBFKL = E(m, m).
(208)
It means, that for the solution of the equation in the second region k2 1/ we can use the same
methods which were used in the previous section for the calculation of the pomeron trajectory.
We hope to return to the problem of finding the Regge trajectories for the odderon and other
gluon composite states in our future publications.
226
Acknowledgements
The work is partially supported by the Russian State Grant Scientific School RSGSS1124.2003.2, the RFBR Grant 040217094 and the Marie Curie Grant. We thank J. Bartels,
V. Fadin, V. Kudryavtsev, A. Sabio Vera and other participants of the PNPI Winter School for
helpful discussions.
Appendix A. Conformal factor for SL(2)SUSY transformations
If the fixed variables are (z1(0) 1(0) ), (z2(0) 2(0) ) and z3(0) while the superpartner of z3(0) is not
(0) (0) (0)
(0)
(0)
fixed, then the discussed factor H (z1 , z2 , z3 , 1 , 2 , ) turns out to be [16]
(0) (0) (0) (0) (0)
H z1 , z2 , z3 , 1 , 2 ,
(0)
(0) (0)
(0)
1
= z1 z3 z2 z3
(0)
(0)
(0)
(0)
1
(0)
2(z1 z3 )
2
(0)
(0)
2(z2 z3 )
(A.1)
(0)
When 1 = 2 = 0 this factor is reduced to the expression given in Section 3 of the paper.
(0) (0) (0)
(0)
(0)
Refixing the above variables to the new values (z1 , z2 , z3 , 1 , 2 ) can be achieved by the
following transformations.
(0)
(0)
Firstly, both 1 and 2 are pushed to vanishing values. The supersymmetric SL(2) transformation (35), which preserves the variables z1 , z2 and z3 but adjusts to 1 and 2 the zero
values, is given by
f (z) = z
(z) =
(z z1 )(z z2 )
3 0 (z3 ),
(z3 z1 )(z3 z2 )
2 (z z1 )
1 (z z2 )
,
(z1 z2 ) f (z1 ) (z1 z2 ) f (z2 )
(A.2)
227
1
3
0
(I +1 I ) dI R,
(B.1)
I =1
B(1 2 , )B(3 1, )
B(1 3 , )B(2 1, )
s
B(1 1, )B(2 3 , )
B(1 3 , )B(2 1, )
t
(B.2)
1 2n cos 2 + 2n
.
(1 n )2
(B.3)
n=1
The factor K includes the colour matrices T and the products of polarization vectors. In the
Regge limit s t it equals (cf. (48))
K = 3 g 4 N T ( s)2 (a a )(b b )
(B.4)
where N = 32 is the dimension of the SO(32) group. In the same limit the region 32 = 3 2
1/( s) 1 is essential and we have the following simplifications
B(12 , )B(3 1, )
1
B(13 , )B(2 1, )
B(1 1, )B(23 1, )
B(13 , )B(2 1, )
sin 1 sin 32
432 l1 ,
sin 2 sin 31
sin 1 sin 32
l2 ,
sin 2 sin 31
(B.5)
(B.6)
where
l1 =
l2 =
n=1
n sin 22
n sin 221
,
1 2n cos 221 + 2n 1 2n cos 22 + 2n
(1 2n cos 21 + 2n )(1 n )2
.
+ 2n )(1 2n cos 22 + 2n )
(1 2n cos 231
n=1
(B.7)
(B.8)
sin 1 sin 32
,
sin 2 sin 31
x =1y =
sin 3 sin 21
sin 2 sin 31
(B.9)
(1 x) sin 2 sin 3
.
cos 2 sin 3 x cos 3 sin 2
(B.10)
228
1
dx
1
1
d3 sin 2 sin 3 sin 23
d2
1
2
32
2
(B.11)
y
1
32
(B.12)
d (1 t)
L2
1 + L1
t
1
d2
(sin 2 )2
ln(1/ s)( s)1+ t
(1 + L1 )1 .
(B.13)
Here both
L1 = 4(sin 2 )2
l1
(1 ) 1 =32 =0
and L2 are given explicitly by (51). As the result we obtain for the Regge trajectory Eq. (50) in
the text.
Appendix C. MultiRegge production amplitudes
In integral (72) we redefine z7 z7 z4 z6 /(z4 z6 ) and introduce f = z4 z6 instead of z4 .
In addition, we replace z6 f z6 . Then the integral I (t5678 , 2 , t3478 , t34 , t56 , t12 , t78 ) in (75) is
given by expression
I t5678 , 2 , t3478 , t34 , t56 , t12 , t78
2 t34 t56 2
dz7 dz6 df ef z7 f t12 2
=
t
t
t
t
z7 78 z6 5678 (1 + z6 ) t3478 q1 34 q2 56
( t78 + 1)q1 q2 ( t5678 t56 t78 2 )q1 t56 q1
+
+ 2
z62 (1 + z6 )2 z72
(1 + z6 )z62 z7
z6 q2
t34 q2
t35 + t36 + t45 + t46
( t3478 t34 t78 + 2 )q2
+
+
,
+
z6 (1 + z6 )
(1 + z6 )2 z6 z7
(1 + z6 )2 q1
(C.1)
where
q1 = f (1 + z6 ) + z7 z6 2 i,
q2 = f z6 + z7 (1 + z6 ) 2 i,
0. (C.2)
+
.
z6 (1 + z6 ) q1
q2
z6 (1 + z6 )
229
(C.3)
To derive Eq. (C.3), one integrates the first term in the brackets in Eq. (C.1) over z7 by parts. As
a result, we obtain the expression similar to Eq. (C.3) but the terms inside the brackets turn out
to be
(C.4)
2
+
z6 (1 + z6 )
z7 z62 (1 + z6 )
z6 (1 + z6 )2
z6 (1 + z6 )2 z7
1
t56 q1
t34 q2
t78 2
( 2 )2
1
+
+
+
.
+
z7
(1 + z6 )2 q1
z62 (1 + z6 ) z6 (1 + z6 )2
z62 (1 + z6 )2 z7
z62 q2
(C.5)
Further, the terms proportional to t78 are integrated by parts over z7 to remove this factor t78 .
Analogously the term t12 is integrated by parts over f to remove the factor t12 . The third term
is integrated by parts over z6 to remove both nominators ( t5678 + 1) and ( t3478 + 1) in the
corresponding contributions. After these transformations we obtain (C.3). If we shall integrate
by parts the first term in Eq. (C.5) over z6 is possible to reduce Eq. (C.3) to the expression
I t5678 , 2 , t3478 , t34 , t56 , t12 , t78
2 t34 t56 2
dz7 dz6 df ef z7 f t12 1
=
t 1
t
1 t
t
z7 78 (1 + z6 ) t3478 z6 5678 q1 34 q2 56
t3478
t34 + t56
z7 t34
+
+
(1 + z6 )
(1 + z6 )q1
(1 + z6 )2
f t56
z6
f + z7
+
+
.
(1 + z6 )q2 (1 + z6 )2 (1 + z6 )
(C.6)
(C.7)
230
(C.8)
z7 /z6 0
(D.1)
while x and y being defined by the relations z3 = z4 + x and z5 = z6 + y are now comparable
in their values with z4 and z6 . However, the poles t34 = n1 and t56 = n2 appear from the
regions x/z4 0 and y/z6 0. In this kinematics one can expand the integrand in powers of x
and y to obtain the poles at t34 = n1 and t56 = n2 . It can be verified that the corresponding
integral vanishes, and, so, the impact factor for the planar diagram is equal to zero.
For the sake of simplicity we give the corresponding proof for the boson string theory, assuming, that the external interaction states are tachyons. In this case only the leading term in x and y
is needed and expression (58) can be simplified as it was done in Eq. (60). Furthermore, similar
to the multiRegge limit we obtain
k2 (k3 + k4 ) k2 (k5 + k6 ) k1 k,
k7 (k3 + k4 ) k7 (k5 + k6 ) k8 k,
(D.2)
but relations (61) for k3 (k7 + k8 ) and for k5 (k7 + k8 ) are not valid. It is helpful to redefine again
the variables according to Eq. (63). After calculating integrals over x and y the asymptotics of
A(0) turns out to be
(D.3)
231
In the calculation of I (0) we performed the change of the integration variables as it was done
in Appendix B: z7 z7 z4 z6 /(z4 z6 ), z4 f = z4 z6 and z6 f z6 = f z/(1 z) (see
Eq. (C.7)). The impact factor is proportional to the sum over the residues of I (0) in the poles
t5678 = n for fixed values t34 = t56 = 1. The parameter n = m 1 takes integer values
from n = 1 up n = . The result contains the factor
m=0
dm
(1 z) [t3456 t12 t78 ] .
m!dzm
(D.5)
The sum is calculated in the region t3456 t12 t78 > 0 where the series is convergent. Then it
is continued analytically to physical values for t3456 t12 t78 . For z 1 this sum is equal to
(1 z) [t3456 t12 t78 ] = 0. Thus, the impact factor for the planar diagram is zero. The vanishing
of the impact factor for the higher mass states t34 = n1 and t56 = n2 is verified in a similar
way. For the superstring theory one can prove also the vanishing of the impact factors for the
planar diagrams.
References
[1] V.S. Fadin, E.A. Kuraev, L.N. Lipatov, Phys. Lett. B 60 (1975) 50;
Ya.Ya. Balitsky, L.N. Lipatov, Sov. J. Nucl. Phys. 28 (1978) 822.
[2] V.S. Fadin, L.N. Lipatov, Phys. Lett. B 429 (1998) 349;
G. Camici, M. Ciafaloni, Phys. Lett. B 430 (1998) 349;
A.V. Kotikov, L.N. Lipatov, Nucl. Phys. B 582 (2000) 19.
[3] S.J. Brodsky, V.S. Fadin, V.T. Kim, L.N. Lipatov, G.V. Pivovarov, JETP Lett. B 70 (1999) 155;
S.J. Brodsky, V.S. Fadin, V.T. Kim, L.N. Lipatov, G.V. Pivovarov, JETP Lett. B 76 (2002) 306.
[4] G. Veneziano, Nuovo Cimento A 57 (1968) 190.
[5] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, vols. I and II, Cambridge Univ. Press, Cambridge, 1987,
and references therein.
[6] L.N. Lipatov, Phys. Lett. B 116 (1982) 411.
[7] J.M. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231;
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Nucl. Phys. B 428 (1998) 105;
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253.
[8] J. Polchinski, M.J. Strassler, JHEP 0305 (2003) 012.
[9] J. Distler, H. Kawai, Nucl. Phys. B 235 (1989) 509;
A. Gupta, S.P. Trivedi, M.B. Wise, Nucl. Phys. B 340 (1990) 475.
[10] G.S. Danilov, Phys. Lett. B 342 (1995) 73.
[11] E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP 45 (1977) 641, Zh. Eksp. Teor. Fiz. 71 (1977) 840.
[12] L.N. Lipatov, Nucl. Phys. B 307 (1988) 705.
[13] D. Friedan, E. Martinec, S. Shenker, Nucl. Phys. B 271 (1986) 93.
[14] L.N. Lipatov, Phys. Lett. B 309 (1993) 394.
[15] D.V. Volkov, A.A. Zheltukhin, A.I. Pashnev, Sov. J. Nucl. Phys. 27 (1978) 131, Yad. Fiz. 27 (1978) 243.
[16] G.S. Danilov, Sov. J. Nucl. Phys. 52 (1990) 727, Yad. Fiz. 52 (1990) 1143;
G.S. Danilov, Phys. Lett. B 257 (1991) 285.
[17] L.N. Lipatov, Nucl. Phys. B 548 (1999) 328.
[18] M.B. Green, J.H. Schwarz, Phys. Lett. B 151 (1985) 21.
[19] J.H. Weis, Phys. Rev. D 5 (1971) 1043.
[20] R.C. Brower, C.E. DeTar, J.H. Weis, Phys. Rep. 14 (1974) 257, and references therein.
[21] L.N. Lipatov, Nucl. Phys. B 307 (1988) 705.
[22] L.N. Lipatov, Sov. Phys. JETP 90 (1986) 1536.
[23] J. Bartels, Nucl. Phys. B 175 (1980) 365;
J. Kwiecinski, M. Prascalowicz, Phys. Lett. B 94 (1980) 365.
[24] L.N. Lipatov, hepth/9311037;
L.N. Lipatov, DFPD/93/TH/70, unpublished.
232
Dynamics of supertubes
Stefano Giusto , Samir D. Mathur, Yogesh K. Srivastava
Department of Physics, The Ohio State University, Columbus, OH 43210, USA
Received 10 February 2006; received in revised form 17 July 2006; accepted 21 July 2006
Available online 24 August 2006
Abstract
We find the evolution of arbitrary excitations on 2charge supertubes, by mapping the supertube to a string
carrying traveling waves. We argue that when the coupling is increased from zero the energy of excitation
leaks off to infinity, and when the coupling is increased still further a new set of long lived excitations
emerge. We relate the excitations at small and large couplings to excitations in two different phases in
the dual CFT. We conjecture a way to distinguish bound states from unbound states among 3charge BPS
geometries; this would identify black hole microstates among the complete set of BPS geometries.
2006 Elsevier B.V. All rights reserved.
1. Introduction
A supersymmetric brane in type II string theory is a 1/2 BPS object. The bound state of N
identical branes (wrapped on a torus) can be mapped by duality to a massless quantum with momentum P = N/R on a circle of radius R. Thus the bound state has degeneracy 256, regardless
of N .
The situation is very different for 1/4 BPS states. Such states can be made in many duality
related ways: NS1P, NS1D0, D0D4,D1D5, etc. If the two charges are n1 , n2 then the de
generacy of the bound state is Exp[2 2 n1 n2 ] [1,2]. In the classical limit n1 , n2 this
degeneracy manifests itself as a continuous family of solutions. Examples are the 2charge D1
D5 solutions found in [3] and the supertubes constructed in [46]. These 2charge states are
important because they give the simplest example of a black hole type entropy [2].
* Corresponding author.
234
In this paper we address the question: what is the low energy dynamics of such 1/4 BPS
states? We will perform some calculations to arrive at a conjecture for the answer. The behavior of the system can depend on whether the coupling is small or large, and whether we have
bound states or unbound states. For this reason we first give an overview of possible dynamical
behaviors, and then summarize our computations and conclusions.
1.1. Possibilities for low energy dynamics
In the following it will be assumed that all compactifications are toroidal, and all branes are
wrapped on these compact directions in a way that preserves 1/4 supersymmetry.
(a) Drift on moduli space: A D0 brane can be placed at rest near a D4 brane; there is no
force between the branes. Thus we have a moduli space of possibilities for the relative separation.
If we give the branes a small relative velocity v then we get a v 2 force, and the resulting motion
can be described by motion on moduli space [7]. More generally, we can make 3charge black
holes that are 1/8 BPS and their slow motion will be described by motion on a moduli space [8,
9]. For later use we make the required limits explicit: the velocity v is O(), the time over which
we follow the motion is O(1/), and the distance in moduli space over which the configuration
drifts is O(1)
1
t ,
(1.1)
x 1 ( 0).
Note that the different branes or black holes involved here are not bound to each other.
(b) Oscillations: Consider branes carrying just one charge, and let these be NS1 for concreteness. Then the force between branes is v 4 . So for the relative motion between such branes
we again get drift on moduli space except that the moduli space is flat [10]. But we can also
focus on just one brane and study its low energy excitations. These will be vibration modes
along the brane, with the amplitude for each harmonic behaving like a harmonic oscillator. Calling the amplitude for a given harmonic An x we note that x will have the time evolution
x = x cos(t + ). Setting x = for a small deformation, the analogue of (1.1) is
v ,
v ,
t 1,
x
( 0),
(1.2)
where we have assumed that we are not looking at a zero mode = 0. For the zero mode we will
have the behavior
x = x0 + vt
(1.3)
and we get drift over configuration space with characteristics given by (1.1).
(c) Quasioscillations: Consider a charged particle free to move in the xy plane in a uniform magnetic field Fxy = B. The particle can be placed at rest at any position on the plane, and
it has the same energy at all these points. Thus far its behavior looks like that of a system with
a zero mode. But if we give the particle a small velocity then it describes a small circle near its
original position, instead of drifting along the plane. The motion is described by
v ,
t 1,
x
( 0).
(1.4)
Thus even though we may have a continuous family of energetically degenerate configurations,
this does not mean that the dynamics will be a drift along this space.
235
(d) Gravitational radiation: We are going to give our system a small energy above extremality. But the system is coupled to type II supergravity, and there are massless quanta in this theory.
Thus any energy we place on our branes can leave the branes and become radiation flowing off
to infinity. There will of course always be some radiation from any motion in the system, but
the relevant issue here is the time scale over which energy is lost to radiation. If the time scale
relevant to the dynamics is t then as 0 we have to ask what fraction of the energy is lost to
radiation in time t . If the fraction is O(1), then the system is strongly coupled to the radiation
field and cannot be studied by itself while ignoring the radiation. If on the other hand the fraction
of energy lost to radiation goes to zero as 0 then the radiation field decouples and radiation
can be ignored in the dynamics.
(e) Excitations trapped near the brane: In the D1D5 system we can take a limit of parameters such that the geometry has a deep throat region. In [3,11] it was found that excitations of
the supergravity field can be trapped for long times in this throat; equivalently, we can make
standing waves that leak energy only slowly to the radiation modes outside the throat [12].
These are oscillation modes of the supergravity fields and thus could have been listed under
(b) above. We list them separately to emphasize that the fields excited need not be the ones making the original brane state; thus the excitation is not in general a collective mode of the initial
fields.
1.2. Results and conjectures
Consider first the D1D5 bound state geometries found in [3]. These geometries are flat space
at infinity, they have a locally AdS3 S 3 T 4 throat, and this throat ends smoothly in a cap.
The geometry of the cap changes from configuration to configuration, and is parametrized by a
function F (v). All the geometries have the same mass and charges, are 1/4 BPS, and yield (upon
quantization) different bound states of the D1D5 system.
What happens if we take one of these geometries and add a small energy? The bound state
of D1D5 branes has a nontrivial transverse size, so one may say that the brane charges have
separated away from each other in forming the bound state. If the charges indeed behave like
separate charges then we would expect drift on moduli space dynamics, (type (a) in our list). Or
does the bound state fragment into a few unbound states, which then drift away from each other?
This is in principle possible, since the D1D5 system is threshold bound. Do we stay within the
class of bound geometries of [3] but have drift on moduli space (1.1) between different bound
state configurations (i.e., drift on the space F (v))? Or do we have one of the other possibilities
(b)(e)?
Now consider the opposite limit of coupling: take a supertube in flat space. The supertube carries NS1D0 charges, and develops a D2 dipole charge. This D2 brane can take on a family of
possible profiles in space, giving a continuous family of 1/4 BPS configurations. What happens
if we take a supertube in any given configuration and add a small amount of energy? Is there a
drift among the family of allowed configurations, or some other kind of behavior?
In [13] the round supertube was considered, and the low energy behavior yielded excitations with time dependence eit . Can we conclude that there is no drift among supertube
configurations? Any drift can occur only between states that have the same values of conserved
quantities. The round supertube has the maximal possible angular momentum J for its charges,
and is the only configuration with this J . So drifting is not an allowed behavior if we give a
small excitation to this particular supertube, and we must look at the generic supertube to know
if periodic behavior is the norm.
236
(1.5)
Then we can look at a small segment of the supertube which looks like a straight line. But this
segment is described by a geometry, and we look for small perturbations of the geometry. We
solve the linearized supergravity equations around this straight line supertube and note that the
resulting periods of the solutions agree with (i) above.
We note however that far from the supertube the gravity solution will be a perturbation on free
space with some frequency satisfying 2 > 0. The only such solutions are traveling waves. For
small Q/a we find that the amplitude of the solution when it reaches the approximately flat part
of spacetime is small. Thus we expect that the radiation into modes of type (d) will be suppressed
by a power of Q/a.
(iii) Now imagine increasing g to the point where
Q a.
(1.6)
In this situation we see no reason why the part of the wavefunction leaking into the radiation
zone should be suppressed. Thus we expect that the excitation will not be confined to the vicinity
of the branes, but will be a gravitational wave that will flow off to infinity over a time of order
the crossing time across the diameter of the supertube.
(iv) We increase the coupling further so that
Q
a.
(1.7)
237
Now the geometry has a deep throat and as mentioned above we find excitations which stay
trapped in this throat for long times, with only a slow leakage to radiation at infinity. What is
the relation between these excitations and those found in (ii)? We argue that these two kinds
of excitations are different, and represent the excitations in two different phases of the 2charge
system. These two phases were identified by looking at microscopic degrees of freedom as a
function of g in [14,15], and what we see here appears to be a gravity manifestation of the
transition.
(v) All the above computations were for bound states of the 2charge system. But we have
seen above that if have unbound statestwo different 2charge black holes for examplethen
we get drift modes of type (a). It looks reasonable to assume that in any coupling domain if we
have two or more different bound states then the relative motion of these components will be a
drift. For example at g = 0 we can have two supertubes that will move at constant velocity past
each other.
It is intriguing to conjecture that this represents a basic difference between bound and unbound
states: bound states have no drift modes and unbound states do have one or more such modes.
The importance of this conjecture is that 3charge systems are very similar to 2charge ones,
so we would extend the conjecture to the 3charge case as well. While all bound states can be
explicitly constructed for the 2charge case, we only know a few 3charge bound states [16].
There is a way to construct all 3charge supersymmetric solutions [17] but the construction does
not tell us which of these solutions are bound states. Since these bound states are the microstates
of the 3charge extremal black hole, it is very important to be able to select the bound states
out of all the possible supersymmetric solutions. The above conjecture says that those states
are bound which do not have any drift type modes of excitations, and the others are unbound.
If this conjecture is true, then we have in principle a way to identify all 3charge black hole
states.
Note: We will use the term supertube or just tube for 2charge bound states in all duality
frames, and at all values of the coupling. The supertube made from D0, NS1 charges carries a
D2 dipole charge, and we will call this the D0NS1 supertube or the D2 supertube. When we use
charges NS1, P we will call the object the NS1P supertube.
2. The NS1P bound state in flat space
We will find that the most useful representation of the 2charge system will be NS1P. We
compactify a circle S 1 with radius Ry ; let X 1 y be the coordinate along this S 1 . The elementary
string (NS1) is wrapped on this S 1 with winding number n1 , and np units of momentum run along
the S 1 . We are interested in the bound state of these charges. This corresponds to the NS1 being a
single multiwound string with wrapping number n1 , and the momentum is carried on this NS1
by its transverse oscillations.1
Consider first the BPS states of this system. Then all the excitations carry momentum in one direction; we set this to be the positive y direction and call these excitations right moving. In Fig. 1(a) we open up the multiwound string to its covering space
where we can see the transverse oscillation profile. As explained in [19] these oscillations
1 The momentum can also be carried by the fermionic superpartners of these oscillation degrees of freedom, but we
will not focus on the fermions in what follows. For a discussion of fermion modes in the 2charge system see for example
[18].
238
Fig. 1. (a) The NS1 carrying a transverse oscillation profile in the covering space of S 1 . (b) The strands of the NS1 as
they appear in the actual space.
cause the n1 strands to separate from each other and the bound state acquires a transverse
size. For the generic state of this 2charge system the radius of the state is and
the surface area of this region gives the entropy of the state by a Bekenstein type relation [20]
A
Smicro = 2 2 n1 n5 .
4G
(2.1)
To understand the generic state better it is useful to look at configurations that have a much larger
transverse size, and later take the limit where we approach the generic state. The relevant limit
is explained in [21]. In this limit the wavelength of the vibration on the multiwound string is
much larger than the radius of the S 1 , so locally the strands of the NS1 look like Fig. 1(b). In the
classical limit n1 np these strands will form a continuous strip, which will be described
by (i) the profile of the strip in the space transverse to the S 1 and (ii) the slope of the strands at
any point along the profile.2
An Sduality gives NS1P D1P, and a further Tduality along y gives D1P D0NS1.
But note that locally the string is slanted, and the Tduality also generates a local D2 charge.
Thus we get a supertube3 where the D0NS1 have formed a D2 [4]. There is of course no net
D2 charge; rather the D2 is a dipole charge. Note also that the slope of the NS1 in the starting
NS1P configuration implies that the momentum is partly along the direction of the strip. Since
we do no dualities in the strip direction we will end up with momentum being carried along the
D2 supertube.
In this BPS configuration the D2 supertube is stationary. If we add some extra energy to
the tube (while keeping its true (i.e., nondipole) charges fixed) then we will get the dynamics of the supertube. But we can study the dynamics in the NS1P picture and dualize to the
D2 supertube at the end if we wish. In the NS1P picture we just have to study a free, classical string. Here the left and right movers decouple and the problem can be solved exactly.
Let us review this solution and extract the dynamics of the supertube in the limit of large
charges.
2 Note that the separation between successive strands is determined by the slope, since the radius of the S 1 is fixed at
Ry .
3 In [22] the same dualities were performed in the reverse order.
239
(2.2)
where
T=
1
.
2
(2.3)
We can get an equivalent dynamics by introducing an auxiliary metric on the world sheet (this
gives the Polyakov action)
X X ab
T
d 2 g a
SP =
(2.4)
g .
2
b
The variation of gab gives
X X 1
X X cd
g =0
g
ab
a b
2
c d
(2.5)
so gab must be proportional to the induced metric. Substituting this gab in (2.4) we get back
(2.2), thus showing that the two actions are classically equivalent.
The X equations give
X ab
= 0.
a g b g
(2.6)
Note that the solution for the X does not depend on the conformal factor of gab .
We choose coordinates 0 , 1 on the world sheet so that gab = e2 ab for some .
Writing
+ = 0 + 1,
= 0 1
(2.7)
we have
g++ = 0,
g = 0.
(2.8)
X X
= 0.
(2.9)
(2.10)
(2.11)
240
We can use the residual diffeomorphism symmetry to set the harmonic function X 0 to4
1
X 0 = a + b = a + b + + .
(2.12)
2
Let the coordinate along the S 1 be called y. We can solve the constraints to express the terms
involving y in terms of the other variables. We find
b2
b2
i Xi ,
i Xi ,
+ y+ =
(2.13)
y =
+ X+
X
+ +
4
4
where X i , i = 1, . . . , 8, are the spatial directions transverse to the S 1 . The parameter b should be
chosen in such a way that the coordinate y winds nw times around a circle of length Ry when
+ 2 . There is no winding around any other direction. We also use a reference frame in
which the string has no momentum in any direction transverse to the S 1 . We let 0 < 2 .
Then the target space coordinates can be expanded as
np
+
y=
cn ein + dn ein ,
+ nw Ry +
Ry
n=0
i
i in
i in +
X =
(2.14)
cn e
.
+ dn e
n=0
Define
b2
i Xi .
(2.15)
X
4
4
From the energy and winding required of the configuration we find that the choice of signs in
(2.13) should be
S+ =
b2
i Xi ,
+ X+
+ +
+ y+ = S+ ,
S =
y = S .
(2.16)
After an interval
=
(2.17)
all the X i return to their original values. This can be seen by noting that is only a parameter
that labels world sheet points, so the actual configuration of the system does not depend on the
origin we choose for . Thus consider the change
( = 1 , ) ( = 1 + , + ).
(2.18)
From (2.14) we see that the X i are periodic with period = . The coordinate y does not return
to its original value, but in the classical limit that we have taken to get the supertube we have
smeared over this direction and so the value of y is not involved in describing the configuration
of the supertube. But the slope of the NS1 at a point in the supertube is relevant, and is given by
i
X
y
s =
(2.19)
.
4 Note that it is more conventional to set a light cone coordinate X + to be linear in . Using a light cone coordinate
allows the constraints (2.9) to be solved without square roots, but for us this is not important since we will not need to
quantize the string.
241
But
y
(in)cn ein( ) + (in)dn ein( + ) .
= n w Ry +
(2.20)
n=0
(2.21)
(2.22)
where E is the energy of the configuration. We therefore find that the motion of the supertube is
periodic in the target space time coordinate with period
t = X 0 = E.
(2.23)
For the NS1P system the dipole charge is NS1this arises from the fact that the NS1 slants as
shown in Fig. 1(b) and so there is a local NS1 charge along the direction of the supertube. The
tension of the NS1 is T = 1/(2 ). This is thus the mass of the dipole charge per unit length
1
.
2
We then see that (2.23) can be recast as
md =
(2.24)
1 E
.
(2.25)
2 md
This form for the period will be of use to us later, because we will find that it holds in other
duality frames as well.
t =
y = b .
(2.26)
we expect the
(2.27)
(2.28)
242
so it does not satisfy (2.8). (Here prime denotes differentiation with respect to ). However we
can change to new coordinates on the worldsheet
+ +
, = f ,
(2.29)
with
(X i X i )( )
.
f =
b2
This brings the metric to the conformally flat form
(2.30)
ds 2 = b2 d d + .
(2.31)
Xi = Xi
(2.32)
(2.33)
so that the X are harmonic in the coordinates ( + , ). Thus the coordinates ( + , ) are conformal coordinates for the problem and we have verified that (2.27) is a solution of the equations
of motion.
We now proceed to adding a small right moving perturbation, which was our goal. Consider
the perturbed configuration
+ + + f ( )
t = b = b
,
2
X i = X i + x i + ,
y = b = b
+ + f ( )
,
2
(2.34)
(2.35)
so that it is conformally flat to first order in the perturbation. The target space coordinates X in
(2.34) are clearly of the form (2.33) so they are harmonic, and we have found a solution of the
string equations of motion.
2.3. Summary
We can get a general solution of the NS1P system by taking arbitrary harmonic functions X i
in (2.14) and determining X 0 , y by (2.12), (2.13). Taking the classical limit where the strands of
the string forms a continuum gives the arbitrary motion of the supertube, and the period of this
motion is given by (2.25). The conformal gauge coordinate that is used on the string is not
very intuitive, since it is determined by the state of the string. We next looked at the linearized
perturbation to a BPS state, and this time we started with an intuitively simple coordinate on the
stringthe static gauge coordinate proportional to the spacetime coordinate y. We found the
explicit map (2.29) to the conformal gauge coordinates. The solution to the linearized problem
was then given by an arbitrary choice of the x i in (2.34).
243
We will now see that these solutions reproduce the behavior of the D2 brane supertube at the
exact and linearized levels respectively. The NS1P system is the easiest way to solve the problem, since it exhibits the separation of the dynamics into a left and a right mover; this separation
is not obvious in the other duality frames.
3. Perturbations of the D0NS1 supertube
In this section we will consider the more conventional definition of the supertube: the true
charges are D0NS1 and the dipole charge is a D2. The dynamics is given by the DBI action
of the D2 with worldvolume fields corresponding to the true charges. In [13] perturbations were
considered around the round supertube which has as its profile a circle in the (X1 , X2 ) plane.
This supertube has the maximum possible angular momentum J for its charges. So if we add
a small perturbation to it we know that we will not get a drift through a set of supertube
configurationsJ is conserved and there are no other configurations with this value of J . So
even though periodic excitations were found for this supertube we cannot conclude from this
that small perturbations to the generic supertube will also be periodic. Thus we wish to extend
the computation of [13] to the generic supertube. We will write the equations of motion for the
generic case, but instead of solving them directly we will note that we have already solved the
problem in NS1P language and we will just dualize the solution there and check that it solves
the equations for the D0NS1 supertube.5
We work in flat space with a compact S 1 of length Ly = 2 Ry , parametrized by the coordinate
y. We have already obtained the general motion of the supertube in the NS1P description, and
below will verify that this solves the general D2 supertube equations as well. But first we check
the behavior of small perturbations, and for this purpose we model our presentation as close to
that of [13] as possible. Thus we let the supertube lie along a closed curve in the (X1 , X2 )
plane, but need not be a circle as in [13]. The worldvolume of the D2 will be S 1 .
Let R and be the radial and angular coordinates in the (X1 , X2 ) plane. We will denote
by Za all the coordinates other than t, X1 , X2 , y. We will also sometimes use the notation
XI = {X1 , X2 , Za }. We fix a gauge in which the world volume coordinates on the D2 brane
are t, , y. Thus the angular coordinate in the supertube plane serves as the parameter along the
supertube curve . On the D2 world volume we have a gauge field, for which we adopt the
gauge
At = 0.
(3.1)
(3.2)
(3.3)
where T2 is the D2 brane tension, g is the metric induced on the D2 world volume and F is the
field strength of A. There are no background fields, so there is no ChernSimons term in the
action.
5 The fact that for given charges there is a range of possible configurations around a generic supertube was also noted
in [23].
244
Za = 0
(3.4)
(3.5)
It is known [4] that this configuration6 satisfies the equations of motion and is supersymmetric
for arbitrary R( ), B( ) if E 2 = 1 and sign B( ) = 1. Without any loss of generality, we will
take E = 1 and sign B( ) = 1 in what follows. The electric field E induces a NS1 integer charge
given by
L
1
1
d y =
d
,
n1 =
(3.6)
T
T
(t Ay )
where T is the NS1 tension. The magnetic field B induces a D0 integer charge equal to
T2
n0 =
(3.7)
dy d B( ),
T0
where T0 is the D0 brane mass.
We want to study fluctuations around the configuration described above. So we expand the
Lagrangian up to quadratic order. We will assume that the fluctuations do not depend on y. We
parametrize the D2brane world volume as
R = R( ) + r(, t),
Za = za (, t)
(3.8)
(3.9)
where lower case quantities denote the fluctuations. The metric induced on the D2 brane world
volume is
2
ds 2 = dt 2 + ( R d + rd + t r dt)2 + R( ) + r d 2 + dy 2
+ ( za d + t za dt)2 .
(3.10)
245
We wish to find the equations of motion up to linear order in the perturbation. To do this we
expand L up to second order in r, ay , a :
L
= L(0) + L(1) + L(2) .
T2
(3.12)
We find
L(0) = B,
R 2 + ( R)2
(1)
t ay + Rt r .
L = ay +
B
(3.13)
(3.14)
We see that at first order in the perturbation the Lagrangian reduces to a total derivative in
and t; this verifies the fact that our starting configuration satisfies the equations of motion. The
term quadratic in the perturbation is
1
R 2 + ( R)2 + B 2
(2)
(t r)2 2t r r
L =
2
B
R 2 + ( R)2 + B 2
R
R
2
Rt rt ay 4
rt ay 4 rt ay
B 2
B
B
(R 2 + ( R)2 + B 2 )(R 2 + ( R)2 )
R 2 + ( R)2
(t ay )2 2
t ay ay
3
B
B 2
R 2 + ( R)2 + B 2
t a 2
(3.15)
(t za )2 2t za za
.
B
B
From this Lagrangian we find the following equations of motion for the linearized perturbation:
2
R + ( R)2 + B 2 2
R 2 + ( R)2 + B 2 2
R
t r + 2t r +
t ay + 2t ay
B
B
B
R
R
t ay = 0,
2 t ay + 2
B
B
2
2
R + ( R)2 + B 2 2
R + ( R)2
t ay + 2t ay
B
B 2
2
2
2
2
R + ( R) + B 2
R
R + ( R)2
R
+
t ay = 0,
t r + 2t r
+ 2 t r +
B
B
B
B 2
R 2 + ( R)2 + B 2 2
t za + 2t za = 0,
B
t2 a = 0.
(3.16)
We have an additional equation coming from the variation of At ; this is the Gauss law which
says
E t a = 0.
(3.17)
The last equation in (3.16) and (3.17) together say that we can add an electric field along the
direction but this field will be constant in both and t . We will henceforth set this additional E
to zero, and thus a = 0 for the rest of the calculation.
246
Note that only time derivatives of fields occur in the equations; there are no terms where the
fields appear without such time derivatives. Thus any time independent perturbation is a solution
to the equations. This tells us that we can make arbitrary time independent deformations of
the supertube, reproducing the known fact that the supertube has a family of time independent
solutions.
The D0 and NS1 integer charges of the perturbed configuration are
T2
T2
dy d B( ) ay =
dy d B( ),
n0 =
T0
T0
2
R + ( R)2
T2
R
(R 2 + ( R)2 )(R 2 + ( R)2 + B 2 )
d
n1 =
t ay
+2 r +
T
B
B
B 3
R 2 + ( R)2
R 2 + ( R)2 + B 2
R
(3.18)
+
ay +
Rt r + 2
r .
B 2
B 2
B
We see that the D0 charge is unchanged by the perturbation. This charge in fact is a topological
invariant of the gauge field configuration. For the NS1 charge we can check conservation by
explicitly computing the time derivative and verifying that it vanishes.
The angular momentum in the (X1 , X2 ) plane is
J = d dy (2 X1 1 X2 ),
(3.19)
where
i =
L
,
(t Xi )
i = 1, 2.
(3.20)
(3.21)
(3.22)
f =
(X1 )2 + (X2 )2
.
b2
(3.23)
247
For the NS1P solution the target space coordinates are given by
t = b ,
y = b = b ,
Xi , = Xi + xi + ,
Za , = za + ,
i = 1, 2,
(3.24)
(3.25)
(3.26)
In this solution derived by duality from NS1P the natural coordinates on the D2 are
( , , y).7 But when we wrote the DBI action for the D2 the natural coordinates were (t, , y),
where was the angle in the (X1 , X2 ) plane
tan =
X2 ( , )
.
X1 ( , )
(3.27)
The coordinates t and are related by a constant, so there is no difficulty in replacing the t by
in converting the NS1P solution to a D0NS1 supertube solution. But the change is
more complicated, and will necessitate the algebra steps below. Inverting (3.27) gives
= (, )
(3.28)
so we see that the change depends on time as well, if the supertube is oscillating. The
variables describing the supertube configuration will be
R(, ) = X12 (, ), + X22 (, ), ,
Ay (, ) = t b (, ),
Za (, ) = Za (, ),
(3.29)
248
where the prime denotes a derivative with respect to the argument . The function ( ) will
be the solution of the equation
tan =
X2 ( )
X1 ( )
(3.31)
(3.32)
(3.33)
R 2 + ( R)2 = B 2 f ,
(3.34)
where f = f ( ( )). Now consider the small perturbation on the supertube. We will keep all
quantities to linear order in the xi , za . Inverting the relation (3.27) gives us
= + ,
(, ) =
X1 x2 X2 x1
,
X1 X2 X2 X1
(3.35)
where
xi = xi 2 f ,
i = 1, 2.
(3.36)
Using (3.35), the first identity in (3.33), and performing an expansion to first order in xi , we find
the perturbation around the static configuration
B
(x1 X2 x2 X1 )
bR
R
(x1 sin x2 cos ),
= x1 cos + x2 sin +
R
B
B
ay (, ) = b = (x1 X2 x2 X1 ) (x1 sin x2 cos ),
R 2
R
za (, ) = za 2 f
za .
r(, ) = R(, ) R( ) =
(3.37)
We would like to check that the functions r, ay and za defined above satisfy the equations of
motion (3.16). For this purpose, some useful identities are
B
B
xi = (1 + f )xi ,
za = (1 + f )za .
b
b
We can simplify some expressions appearing in the equations of motion (3.16)
(3.38)
249
R 2 + ( R)2 + B 2 2
t r + 2t r
B
4
R
=
x1 sin + x2 cos +
(x1 cos + x2 sin )
b
R
4
R
+
(x1 sin x2 cos ),
b
R
R 2 + ( R)2 + B 2 2
t ay + 2t ay
B
B
4
B
= 4 (x1 cos + x2 sin )
(x1 sin x2 cos ),
b
bR
R
R 2 + ( R)2 + B 2 2
t za + 2t za = 0.
B
(3.39)
The last identity proves that the equations for za are satisfied. For the equations involving r
and ay some more work is needed. The l.h.s. of the first equation in (3.16) is equal to
4
R
B R B R
R
+
(x sin x2 cos ) 1 +
b 1
R
R B
R B
B
R B R
4
+ (x1 cos + x2 sin )
(3.40)
b
R
R R
which, after some algebra, is seen to vanish. The l.h.s. of the second equation in (3.16) is
4
R 2 + ( R)2
B
R
R
(x sin x2 cos )
1
b 1
B 2
R
B
R
2
2
R R 1 B
R + ( R)
+
2 R
B R
B 2
B R 2 + ( R)2 ( R)2 R
4
+
+ (x1 cos + x2 sin )
(3.41)
+
b
R
B 2
B R
B
which also vanishes.
We thus find that the expressions (3.37), with xi , za arbitrary functions of their arguments,
satisfy Eqs. (3.16).
3.2. Period of oscillation
We would like to determine the period of the oscillations of the solution (3.37). The world
sheet coordinate has a period 2 . The time dependence of the solution (3.37) is contained
in functions xi (2 f ( )) and za (2 f ( )). The quantity (X )2 ( ) which
appears in the definition of f ( ) will be the sum of a constant term, R 2 , plus terms periodic in
:
an ein + c.c.
(X )2 = R 2 +
(3.42)
n=0
250
f = 2 +
bn ein + c.c. .
b
(3.43)
n=0
The functions xi , za are functions of the coordinate along the supertube. This supertube is a
closed loop, so all functions on it are periodic under the shift ( , )( , + 2). This implies
R2
xi 2 f = xi 2 f 2 1 + 2
(3.44)
,
b
where we have used (3.43) to get the change in f ( ).
We have a similar relation for za (2 f ( )). Thus the period of the oscillations is
given by
b2 + R 2
(3.45)
.
b
This form of the period is similar to that found in [13]; it reduces to the period found there when
the radius R is a constant.
To arrive at our more general form (2.25) we write
t = b =
+ f
=
d 1 + f () .
(3.46)
2
+ f ( ) when increases by 2 can be written as 0 d (1 + f ()).
So the change in
We then find that the argument of xi , za are unchanged when ( , )( + , + 2) with
2
2
1 + f d = 0.
(3.47)
2
R 2 + ( R)2
1+
d .
B 2
(3.48)
Now using the fact that B = b , t = b and changing variables from to we get
R 2 + ( R)2
1
d B +
t =
(3.49)
.
2
B
Now we express (3.49) in terms of the NS1 and D0 charges (we can use the unperturbed
values of these quantities), using
R 2 + ( R)2
T2
T2
d
dy d B.
,
n0 =
n1 =
(3.50)
T
T0
B
We get
1 n0 T0 + n1 T Ly
t =
(3.51)
,
2
T2 Ly
251
1 E
.
2 md
(3.52)
(3.53)
In this subsection X i denotes all coordinates other than t and y. We wish to prove that (3.53)
satisfies the dynamical equations of the D2 brane. The DBI Lagrangian density is given by
L
= det(g + F )
T2
= ( X)2 ( X)2 + ( X X)2 + b2 E 2 ( X)2
1/2
+ B 2 b2 ( X)2 2EB( X X)
.
(3.54)
The equations of motion are
X i [( X)2 + B 2 ] X i [( X X) EB]
L
i
2
X [( X) + E 2 b2 ] X i [( X X) EB]
= 0,
+
L
E( X)2 + B( X X)
B[( X)2 b2 ] + E( X X)
= 0.
L
L
(3.55)
(3.56)
To verify that these equations are satisfied by the configuration (3.53) we need following identities:
4
b
L
+ 4(+ X+ )2 ( X )2 + 4(+ X+ X )2
=
T2
2
+ b2 2S+ S 2(+ X+ X ) (+ X+ )2 ( X )2
1/2
b2
8S+ S (+ X+ X )
(3.57)
=
+ 2S+ S 2(+ X+ X ),
2
b2
+ 2S+ S 2(+ X+ X ),
( X)2 + B 2 = ( X)2 + E 2 b2 =
2
252
( X X) EB = 0,
E( X)2 + B( X X)
= 2S+ ( X )2 (+ X+ X ) 2S (+ X+ )2 (+ X+ X )
2
b
+ 2S+ S 2(+ X+ X ) ,
= (S+ S )
2
2
2
B ( X) b + E( X X)
2
2
b
b
2
2
( X ) (+ X+ X ) 2S + (+ X+ ) + (+ X+ X )
= 2S+
2
2
2
b
+ 2S+ S 2(+ X+ X ) .
= (S+ + S )
(3.58)
2
Then Eqs. (3.55), (3.56) become
2 X i 2 X i = 4+ X i = 0,
(S+ S ) (S+ + S ) = 2 S+ 2+ S = 0
(3.59)
which are seen to be satisfied due the harmonic nature of the fields X i , y.
3.4. Quasioscillations
In the introduction we termed the periodic behavior of the supertube a quasioscillation. In
a regular oscillation there is an equilibrium point; if we displace the system from this point
then there is a force tending to restore the system to the equilibrium point. But in the supertube
we can displace a stationary configuration to a nearby stationary configuration, and the system
does not try to return to the first configuration. The only time we have such a behavior for a
usual oscillatory system is when we have a zero mode (1.3). Such zero modes allow a drift in
which we give the system a small initial velocity and then we have an evolution like (1.1). But
the supertube does not have this behavior either; there is no drift.8
Now consider a different system, a particle with charge e and mass m placed in a uniform
magnetic field Fxy = B. With the gauge potential Ay = x we have the Lagrangian
m
2
m
2
L=
(3.60)
(x) + (y)2 + eA v =
(x) + (y)2 + ex y.
2
2
The equations of motion are
e
e
y = x.
x = y,
(3.61)
m
m
Since each term in the equation has at least one time derivative, any constant position x = x0 , y =
y0 is a solution. But if we perturb the particle slightly then the particle does not drift over this
space of configurations in the manner (1.1); instead it describes a circle with characteristics
(1.4). So while this motion is periodic the physics is not that of usual oscillations, and we call it
a quasioscillation.
8 By contrast, giant gravitons have usual vibration modes [24]. The giant graviton in AdS S 3 has a zero mode
3
corresponding to changing the radius of the giant graviton, and we find a drift over the values of this radius. In [12]
giant gravitons were studied for AdS3 and it was argued that they give unbound states where one brane is separated from
the rest [12,25].
253
Now we wish to show that the motion of the supertube is also a quasioscillation. We will
take a simple configuration of the D2 brane to illustrate the point. Let the D2 brane extend along
the zy plane and oscillate in one transverse direction x. We will restrict to motions which are
invariant in y and thus described by a field x = x(t, z). We will also turn on a (yindependent)
world volume gauge field, for which we choose the At = 0 gauge:
A = Az (t, z) dz + Ay (t, z) dy,
F = Az dt dz + Ay dt dy + Ay dz dy
Az dt dz + E dt dy B dz dy.
(3.62)
(3.63)
In order to have a qualitative understanding of the dynamics induced by this Lagrangian, let us
expand it around a classical stationary solution with x = 0, E = 1, B = B and Az = 0. We denote
by ay (t, z) the fluctuation of the gauge field Ay , so that
E = 1 + ay ,
B = B ay .
(3.64)
As the gauge field Az decouples from all other fields we will set it to zero. Keeping terms up to
second order in x and ay , we find the quadratic Lagrangian density to be
ay
1 + B 2 2
1 1 + B 2 2
x + xx + 2
ay + ay ay .
L(2) = B +
(3.65)
+ ay +
2B
2B
B
B
The terms of first order in ay are total derivatives (with respect to t and z) and do not contribute
to the action. The fields x and ay are decoupled, at this order, and both have a Lagrangian of the
form
1 + B 2 2
(3.66)
+
2B
(with = x or ay ). As we can see the Lagrangian (3.66) has no potential terms (terms independent of ) and we find that any time independent configuration solves the equations of motion.
There is however a magnetictype interaction ( ), which is responsible for the fact that all
timedependent solutions are oscillatory. Indeed, the equations of motion for are9
(2)
L =
1 + B 2
+ = 0
2B
(3.67)
9 In [6] it was pointed out that the dynamics of open strings ending on supertubes could be understood in terms of the
effective open string metric G which results from the field strength F on the supertube:
1
1
=
.
G +
2
+ 2 F
It turns out that Eq. (3.67) is just the Laplace equation in this open string metric. (We thank the referee for pointing out
this connection.)
254
whose solution is
= eikzit ,
=2
B
k.
1 + B 2
(3.68)
One can make the analogy between the interaction and the toy problem of a particle in a
magnetic field more precise by discretizing the z direction on a lattice of spacing a. Then we
have
m
m
n+1 n
n n+1
(2)
2
2
dz L a
(3.69)
+ n
a
+
,
2 n
a
2 n
a
n
n
where in the second line we have discarded a total timederivative and m = (1 + B 2 )/B. The
term n n+1 is just like the term x y in (3.60) induced by a constant magnetic field where the
variables n , n+1 play the role of x, y.
3.5. Summary
We have obtained the full dynamics of the D2 supertube, by mapping the problem to a free
string which can be exactly solved. In the D2 language it is not obvious that the problem separates into a right mover and a left mover, but (3.53) exhibits such a break up. This breakup
needs a world sheet coordinate that is a conformal coordinate on the string world sheet, and is
thus not an obvious coordinate in the D2 language. The D2 has a natural parametrization in terms
of the angular coordinate on the spacetime plane (X1 , X2 ), and the difficulties we encountered
in mapping the NS1P solution to the D2 supertube all arose from the change of parametrization.
4. The thin tube limit of the gravity solution
So far we have ignored gravity in our discussion of the supertube, so we were at vanishing
coupling g = 0. If we slightly increase g then the gravitational field of the supertube will extend
to some distance off the tube, but for small enough g this distance will be much less than the
radius of the supertube. We will call this the thin tube limit, and we picture it in Fig. 2(b).
We expect that in this thin tube the dynamics should not be too different from that found
at g0, and we will find that such is the case; we will find periodic excitations with frequency
agreeing with that found from the free string computation and the D2 brane DBI action. But by
doing the problem in a gravity description we move from the worldsheet theory to a spacetime
one, which will help us to understand what happens when we increase the coupling still further.
Let us recall the 2charge BPS geometries made in the NS1P duality frame [3]. Start with
type IIB string theory and take the compactification M9,1 M4,1 S 1 T 4 . As before the coordinate along S 1 is y and the coordinates za , a = 1, . . . , 4, are the coordinates on T 4 . The S 1 has
length Ly = 2Ry and the T 4 has volume (2)4 V . The four noncompact spatial directions are
called xi , i = 1, . . . , 4. We also write u = t + y, v = t y.
The NS1 is wrapped n1 times around the S 1 , and carries np units of momentum along the S 1 .
This momentum is carried by transverse traveling waves; we assume that the polarization of the
wave is in the four noncompact directions and is described by a function F (v). Then the string
255
Fig. 2. (a) The supertube at g0, described by a worldsheet action. (b) The thin tube at weak coupling. (c) The thick
tube reached at larger coupling. (d) At still larger coupling we get a deep throat geometry; the strands of the NS1
generating the geometry run along the dotted curve.
B=
e2
(4.1)
with
Q1
H =1+
LT
LT
dv
,
2
i (xi Fi (v))
0
Q1
K=
LT
LT
0
Q1
Ai =
LT
(Fi (v))2
dv i
,
2
i (xi Fi (v))
LT
dv
0
Fi (v)
.
2
(
x
i i Fi (v))
(4.2)
LT
dv
0
2
Fi (v) .
(4.3)
g2 3
n1 ,
V
Qp =
g2 4
np .
V Ry2
(4.4)
256
Thus when we keep other parameters fixed and take g very small then the gravitational field of
the supertube gets confined to a small neighborhood of the supertube and we get a thin tube like
that pictured in Fig. 2(b). If we increase g large then we pass to a thick tube like Fig. 2(c) and
then to the deep throat geometry of Fig. 2(d). We can thus say that Fig. 2(a) is weak coupling
and Fig. 2(d) is strong coupling but note that for strong coupling g itself does not need to
be large since the charges n1 , np are large in (4.4). Thus to be more correct we should say that
Fig. 2(d) is obtained for large effective coupling.
In this section we will consider the weak coupling case so that we have a thin tube. Then
to study the nontrivial part of the metric we have to go close to a point on the tube, so the
tube looks essentially like an infinite straight line. Let z be a coordinate along this line (not to
be confused with za , which are coordinates on T 4 ) and r the radial coordinate for the threespace perpendicular to the ring. The NS1P profile was described by a function F (v); let v = v0
correspond to the point z = 0 along the ring and choose the orientation of the z line such that z
increases when y increases. Then we have
2
z F (v0 ) (v v0 ),
(4.5)
xi Fi (v) z2 + r 2 .
i
Since we are looking at distances r from the ring which are much smaller than the size of the
ring we have
r F (v0 ) .
(4.6)
We can thus make the following approximations
1
dz
Q1
Q1
=1+
,
H 1+
2
2
LT F (v0 ) z + r
LT F (v0 ) r
K
Q1 F (v0 ) 1
,
LT
r
Az
Q1 1
.
LT r
(4.7)
Q1
,
LT F (v0 )
Qp
Q1 F (v0 )
.
LT
(4.8)
(4.9)
Q1 Qp
Qp
Q1
,
K = 1 + K = 1 +
,
A=
.
(4.10)
r
r
r
Here we use xi , i = 1, 2, 3, to denote the three spatial noncompact directions transverse to the
tube.
We are looking for a perturbation of (4.9) corresponding to a deformation of the string profile.
The profile could be deformed either in the noncompact xi directions or in the T 4 directions. We
consider deformations in one of the directions of the T 4 ; this maintains symmetry around the
H =1+
257
tube in the noncompact directions and is thus easier to work with. We thus consider deforming
the string profile in one of the T 4 directions, denoted a. We will also smear the perturbed metric
on T 4 , so that our fields will be independent on za .
The BPS geometry (4.9) carries a wave of a definite chirality: let us call it right moving. If
the deformation we add also corresponds to a right moving wave, the resulting geometry can be
generated by GarfinkleVachaspati transform [26,27]. This will alter the metric and Bfield as
follows:
2
2
dsstring
dsstring
+ 2A(1) dza ,
B B + A(2) dza ,
(4.11)
where
A(1) = A(2) = H 1 av dv
(4.12)
and av is a harmonic function on R3 Sz1 , whose form will be given in Section 4.2. If we also
add a left moving deformation, thus breaking the BPS nature of the system, we do not have a
way to generate the solution. Note, however, that the unperturbed system has a symmetry under
za za
(4.13)
and the perturbation will be odd under such transformation. We thus expect that only the components of the metric and Bfield which are odd under (4.13) will be modified at first order in the
perturbation. We can thus still write the perturbation in the form (4.11), with A(1) and A(2) some
gauge fields on R(3,1) Sz1 Sy1 , not necessarily given by (4.12).
To find the equations of motion for A(1) and A(2) we look at the theory dimensionally reduced
on T 4 , using the results of [28]. At first order in the perturbation the dimensionally reduced metric
g6 is simply given by the sixdimensional part of the unperturbed metric (4.9). The part of the
action involving the gauge fields is
2 1
2
1
1
SA =
(4.14)
g6 e2 F (1) F (2) H 2 ,
4
4
12
where all the index contractions are done with g6 . F (1) and F (2) are the usual field strengths of
A(1) and A(2) while the field strength H of the dimensionally reduced Bfield B includes the
following ChernSimons couplings:
1 (1) (2)
(1)
A F + A(2)
F + cyc. perm.,
2
1
(2)
(2) (1)
= B + A(1)
A A A .
2
H = B
B
(4.15)
Using B, A(1) and A(2) as independent fields, we find that the linearized equations of motion for
the gauge fields are
1
(1)
(2)
e2 F
+ e2 H F
= 0,
2
1
(2)
(1)
e2 F + e2 H F
= 0.
2
These can be rewritten as decoupled equations as
A = A(1) A(2) ,
(4.16)
(4.17)
258
1
e2 H F
e2 F
= 0.
2
(4.18)
Our task is to find the solutions of these equations representing nonBPS oscillations of the two
charge system (4.9).
4.1. Solution in the infinite wavelength limit
The geometry (4.1) has a singularity at the curve x = F (v), which is the location of the strands
of the oscillating NS1. Since we wish to add perturbations to this geometry, we must understand
what boundary conditions to impose at this curve. The wavelength of the oscillations will be of
order the length of the tube. Since the tube is thin and we look close to the tube, locally the tube
will look like a straight line even after the perturbing wave is added. The wave can tilt the tube,
and give it a velocity. So in this subsection we write the metric for a straight tube which has been
rotated and boosted by infinitesimal parameters , . In the next subsection we will require that
close to the axis of the tube (where the singularity lies) all fields match onto such a rotated and
boosted straight tube solution.
We will consider oscillations of the supertube in one of the T 4 directions. Since we smear on
the T 4 directions, the solution will remain independent of the torus coordinates za but we will
get components in the metric and B field which reflect the tilt of the supertube. We are using
the NS1P description. The unperturbed configuration looks, locally, like a NS1 that is a slanted
line in the yz plane, where z is the coordinate along the tube. The perturbation tilts the tube
towards a T 4 direction za . We will find it convenient to start with the NS1 along y, first add the
tilt and boost corresponding to the perturbation, and then add the noninfinitesimal tilt in the yz
plane (and the corresponding boost).
We start from the one charge system
2
= H 1 (d t )2 + (d y )2 + (d z )2 + dxi dxi + d za d za +
dza dza ,
dsstring
a=a
B = H 1 1 d t d y ,
e2 = H 1
(4.19)
and perform the following operations: an infinitesimal boost in the direction za , with parameter
t = t za ,
za = za t ,
y = y ,
z = z
(4.20)
za = za y,
t = t,
z = z.
(4.21)
B = H 1 1 d t d y + H 1 (H 1) d y d za + H 1 (H 1) d t d za ,
e2 = H 1 .
(4.22)
259
Fig. 3. A short segment of the NS1 moving at the speed of light in the y direction. This yields a velocity v for the segment
in the direction perpendicular to itself.
A = ( + )H 1 (H 1) d u
(4.23)
z = z cosh t sinh ,
za = za
(4.24)
z = z cos y sin ,
t = t,
za = za .
(4.25)
The parameters , are related. This is because the segment of string under consideration is
supposed to be a short piece of the string in a state like that in Fig. 1(a), where the traveling wave
is moving in the positive y direction with the speed of light. We depict this segment in Fig. 3.
We can ask how fast the string segment must be moving in a direction perpendicular to itself to
yield dy/dt = 1, and we find
v tanh = sin .
(4.26)
This implies
sinh = tan ,
cosh =
1
.
cos
(4.27)
260
a=a
1
1 dt dv H (H 1) sinh dv dz
sin2 cos2
dv dza + ( + ) cos dt dza
+ H 1 (H 1)
cos
+ ( + ) sin dz dza ,
B= H
e2 = H 1 .
(4.28)
(4.29)
,
cos
+ = ( + ) cos
(4.30)
(4.31)
and we have used (4.29) and (4.27). We see that, as before, A+ comes from right moving perturbations, proportional to , and A comes from left moving perturbations, proportional to
+ .
4.2. Solution for A+
Let us look for a solution of (4.18), in the A+ sector, which matches the configuration (4.30)
when k 0. We learned from (4.30) that A+ receives contributions only from the BPS (right
moving) part of the wave and that, at least in the long wavelength limit, only the component A+
v
is nonvanishing. One can thus look for a solution of the form
1 +
A+
av ,
v =H
A+
t = 0,
A+
z = 0,
A+
i = 0.
(4.32)
261
Eq. (4.18) implies the following conditions for av+ (here = i i is the ordinary Laplacian in
the 3dimensional space of the xi )
= t:
t2 av+ = 0,
= v:
av+
= z:
= i:
(4.33)
+ z2 av+
t z av+ = 0,
t i av+ = 0.
2At z av+
= 0,
(4.34)
(4.35)
(4.36)
(4.37)
solves the linearized equations of motion. The general solution of (4.37), with momentum
n
k=
(4.38)
Rz
along z, is
c+ ekr + c ekr
+ c.c.
(4.39)
r
Without loss of generality let us set n to be positive. To have a converging field at large r one
should take c+ = 0. Matching with (4.30) fixes c :
av+ = eikz
c = ( )Q1
(4.40)
so that
1
A+
v = ( )H
Q1 ikzkr
+ c.c.
e
r
(4.41)
The above result is consistent with the form of A+ derived by GarfinkleVachaspati transform: consider a string carrying a right moving wave described by the profile Fi (v) in the
noncompact directions xi and fa (v) in the T 4 direction za . After smearing over za , Garfinkle
Vachaspati transform predicts a gauge field
(1)
(2)
A+
v = A v + Av ,
(4.42)
(2)
1
A(1)
v = Av = H
Q1
LT
LT
fa (v)
.
2
i (xi Fi (v))
dv
0
(4.43)
Eq. (4.43) is analogous to the relation (4.2) for Ai , applied to the case in which the profile extends
in the T 4 directions. Let us take the near ring limit of (4.43) for a profile fa of the form
fa (v) = a ei kv + c.c.
Around some point v0 on the ring we write
z0 = F (v0 ) v0
z = F (v0 ) (v v0 ),
(4.44)
(4.45)
(4.46)
262
and
fa (v) = ia F (v0 ) keik(z+z0 ) + c.c.
(4.47)
2H
1 i Q1 a k ikz0
LT
+
ikz0 kr
eikz
1 i Q1 a k e
dz 2
+
c.c.
=
2H
+ c.c.
LT
r
r + z2
(4.48)
Using (4.8) to relate Q1 and Q1 we see that (4.48) coincides with (4.41), with
( ) = 2ia F (v0 ) k.
(4.49)
The timeindependent solution (4.41) represents the response of the system to a BPS right
moving wave. Since the A+ part of the gauge field should only be sensitive to BPS deformations,
we expect that Eq. (4.18) for A+ should not admit timedependent solutions consistent with the
boundary condition (4.30). In an appendix we prove this fact for the more general A+ ansatz.
4.3. Solution for A
We now look at the A sector, where we expect to find the timedependent configurations
corresponding to left moving nonBPS perturbations.
Consider an ansatz of the form
1
A
av ,
v =H
1
A
at ,
t =H
1
A
az ,
z =H
A
i = 0.
(4.50)
By spherical symmetry A
i only has a radial component Ar and we chose our gauge to set
Ar = 0. (Such a gauge can have difficulties at r = 0 but we can consider it as an ansatz and
see later that we obtain a good solution.) The equations for av , at and az , obtained by using
the ansatz (4.50) in (4.18) and using the background (4.9), are (we list the equations in the order
= t, v, z, i)
at + H t2 av + z z at t az + At z at t az = 0,
(4.51)
av + z2 av H K A2 t2 av 2At z av + H 2 i H i H 2av + Kat
H 1 i H i 2av + Kat + i at i K = 0,
(4.52)
az + H t z av + H K A2 t z at t az Az z at t az
+ 2H 2 i H i H az + Aat 2H 1 i H i az + Aat + 2i at i A = 0,
(4.53)
H t i av z i az + H K A2 t i at Az i at At i az
+ H 1 i H z az + A t az + z at H K A2 t at
i A z at t az 2i H t av = 0.
(4.54)
Inspired by the limiting solution (4.30), we make the following ansatz for
av = (Q1 Qp )eikzit f (r),
av , at
and az :
az = 2 Q1 Qp eikzit f (r).
r
This equation can be simplified by taking
f
r
after which we get
f=
263
(4.55)
(4.56)
(4.57)
2Q1 f 2 (Q1 Qp )f + 2k(kQ1 + Q1 Qp )f
Q1 f
2 Q1 Qp k + (Q1 + Qp ) = 0.
r
(4.58)
(4.60)
with
k = k = k
2
Q1 Qp
Q1 + Qp
2
(4.61)
and thus
f = c+ e+kr + c ekr .
(4.62)
In order to have a converging solution for large r one needs c+ = 0 and to match with (4.30) one
needs c = + . To summarize we find
1
A
(Q1 Qp )eikzit
v = ( + )H
1
A
Q1 eikzit
t = 2( + )H
ekr
,
r
ekr
,
r
ekr
1
A
(4.63)
Q1 Qp eikzit
.
z = 2( + )H
r
It is a lengthy but straightforward exercise to verify that (4.63) solves the remaining equations
(4.52)(4.54).
264
.
=2
v=
k
Q1 + Qp
(4.64)
The direction z used above is the coordinate along the supertube. So even though z looked like
an infinite direction in the near tube limit, this direction is actually a closed curve with a length
Lz . The time for the wave to travel around this closed curve is
Lz
Lz
Lz
Q1 + Qp
Qp
dz
Q1
1
=
dz
+
t =
(4.65)
= dz
.
v
2
Qp
Q1
2 Q1 Qp
0
We have
Q1 =
Q1 1
,
LT
1 =
1
dy
.
=
F  dz
Qp =
Q1
LT
(4.66)
with
(4.67)
This gives
Lz
1
dy dy dz 2
dz + =
dz
+
2
dz dz dy
0
0
1
1
1
2
(T n1 Ly ) +
T F  dy =
(MNS1 + MP ),
=
2T
2T
2T
1
t =
2
Lz
(4.68)
where MNS1 is the mass contributed by the NS1 charge and MP is the mass of the momentum
charge. We see that this period t agrees with the period (2.25) found from the NS1P system
at g = 0.
We offer an intuitive explanation for the time period (2.25). We have
2
dy
Q1
= 2 =
.
(4.69)
Qp
dz
Thus we can write (4.64) as
v=2
dy
dz
.
2
1 + ( dy
dz )
(4.70)
Consider a segment of the NS1 before the perturbation is added. In Section 4.1 we had seen,
(with the help of Fig. 3) that because this segment represents a wave traveling in the y direction
with dy/dt = 1, the velocity of this segment perpendicular to itself was
1
.
v = sin =
2
1 + ( dy
)
dz
(4.71)
265
So we have a segment of a NS1, moving at a certain velocity transverse to itself. Go to the rest
frame of this segment. Then any small perturbation on the segment will move to the right or to
the left with speed unity. Consider the perturbation going left.
Now return to the original reference frame, and look at this perturbation on the segment.
The distances along the segment are not affected by the change of frame
(since the boost is
2 . This means
perpendicular to the segment) but there is a time dilation by a factor = 1/ 1 v
that the perturbation will be seen to be moving along the strand at a speed
vL =
=
dy
dz
2
1 + ( dy
dz )
(4.72)
We are interested in the motion of the perturbation in the z direction, so we look at the z component of this velocity
vL,z = vL sin =
dy
dz
.
2
1 + ( dy
dz )
(4.73)
What we actually observe as the wave on the supertube is a deformation moving along the tube,
so we wish to measure the progress of the waveform as a function of the coordinate z. A given
point on our NS1 segment moves in the direction of the velocity v, so it moves towards smaller
z values at a speed
vz = v cos =
dy
dz
.
2
1 + ( dy
dz )
(4.74)
Thus if we measure the speed of the left moving perturbation with respect to a frame where z is
fixed then we find the velocity
pert
vL = vL,z + vz = 2
dy
dz
2
1 + ( dy
dz )
(4.75)
vR = vL,z + vz = 0.
(4.76)
This agrees with the fact that if we add a further right moving wave to the NS1 then we just get
another BPS tube configuration, which is stationary and so does not change with time.
5. Coupling to radiation modes
The perturbations of the thin tube in the infinite line limit is seen to fall off exponentially
with the distance from the tube axis. Note however that if we take the longest wavelengths on the
supertube, then the term ekr is not really significant. For such modes k 1/a where a is the
radius of the tube. So ekr 1 for r a, and for r a we cannot use the infinite line limit of
the thin tube anyway. If however we look at higher wavenumbers on the tube then k n/a and
then the factor ekr is indeed significant in describing the fall off of the perturbation away from
the tube axis.
266
We now wish to look at the behavior of the perturbation far from the entire supertube, i.e.,
for distances r
a. Here we use the symbol r for the radial coordinate in the 4dimensional
noncompact space, to distinguish it from the radial distance r from the tube axis that we used in
the last section when looking at the infinite line limit. For r
a we get flat space. Suppose we
were studying a scalar field = 0 in the supertube geometry. We can write
= eit R(r)Y (l) (, , ).
(5.1)
If 2 < 0 then we get solutions et ; these are not allowed because they will not conserve
energy. For 2 > 0, we get the behavior (see Appendix B)
r+ eir + r eir
1 + O r 1 .
(5.2)
3/2
r
This solution describes traveling waves that carry flux to and from spatial infinity. Thus if we
start with an excitation localized near the supertube then the part of its wavefunction that extends
to large r will lead to the energy of excitation flowing off to infinity as radiation.
Let us see how significant this effect is for the thin tube. Let us set Q1 , Qp to be of the same
order. From (4.63) we see that the magnitude of the perturbation behaves as
R=
A H 1
Q1
Q1
.
r
r + Q1
(5.3)
Thus if the perturbation is order unity at the ring axis then at distances r a we will have
Q1
.
(5.4)
a
But the thin tube limit is precisely the one where the ratio Q1 /a is small, so the part of the
wavefunction reaching large r is small. Thus the rate of leakage of energy to the radiation field is
small, and the excitations on the thin tube will be long lived. This is of course consistent with
the fact that in the limit g0 we can describe the system by just the free string action or the D2
brane DBI action, and here there is no leakage of energy off the supertube to infinity.
As we keep increasing g we go from the thin tube of Fig. 2(b) to the thick tube of Fig. 2(c).
Now Q1 /a 1 and the strength of the perturbation reaching the radiation zone is not small. We
thus expect that the energy of excitation will flow off to infinity in a time of order the oscillation time of the mode. Thus we expect that the oscillations of the supertube become broad
resonances and cease to be well defined oscillations as we go from Fig. 2(a) to (c).
In the above discussion we referred to the excitation as a scalar field, but this is just a toy
model; what we have is a 1form field in 5 + 1 spacetime. In Appendix B we solve the field
equations for this 1form field at infinity, and find again a fall off at infinity that gives a nonzero
flux of energy. We also find the next correction in 1/r, and show how a series expansion in 1/r
may be obtained in general. These corrections do not change the fact that the leading order term
carries flux out to infinity. It is important that the first correction to flat space is a potential 1/r 2
and not 1/r; this avoids the appearance of a logarithmic correction at infinity.
It is to be noted that such series solutions in 1/r are asymptotic expansions rather than series
with a nonzero radius of convergence [29], so these arguments are not a rigorous proof for the
absence of infinitely long lived oscillations. The wave equations for a given are similar in
structure to the Schrodinger equation (in 4 + 1 dimensions) with a potential V falling to zero at
infinity
A
R + V (r, )R = 2 R.
(5.5)
267
Note that because 2 > 0 our wavefunction would be like a positive energy eigenstate of the
Schrodinger equation; i.e., we need an energy eigenvalue embedded in the continuum spectrum.
For the Schrodinger equation there are several results that exclude such eigenvalues on general
grounds [30]. The required results come from two kinds of theorems. First we need to know that
there is no potential well with infinitely high walls near the origin; if there was such a well then
we can have a positive energy eigenstate which has no tail outside the well. Next, given that
there is a tail outside the well we need to know that the potential falls off to zero fast enough
and does not oscillate too much; such oscillations of the potential can cause the wavefunction
to be backscattered towards the origin repeatedly and die off too fast to carry a nonzero flux
at infinity. We cannot directly apply these results to our problem because our equations are not
exactly the Schrodinger equation, but the potential like terms in our equations do not appear to
be of the kind that will prevent flux leakage to infinity.
To summarize, we conjecture that as we increase g to go from Fig. 2(a) to (c) the periodic
oscillations present at g 0 merge into the continuum spectrum of bulk supergravity. Thus for
g > 0 the energy of excitation placed on the supertube eventually leaks off to infinity, with the
rate of leakage increasing as we go from the thin tube to the thick tube.
6. Long lived excitations at large coupling
Let us increase the coupling still further. Then the supertube geometry becomes like that
pictured in Fig. 2(d) [3,31,32]. The metric is flat space at infinity, then we have a neck region,
this leads to a deep throat which ends in a cap near r = 0. Supergravity quanta can be trapped
in the throat bouncing between the cap and the neck for long times before escaping to infinity.
We first consider the gravity description, then a microscopic computation, and finally suggest a
relation between the two.
6.1. The geometry at large effective coupling
Consider an NS1 wrapped n1 times on the S 1 with radius Ry , and give it the transverse
vibration profile
X1 = a cos
(t y)
,
n 1 Ry
X2 = a sin
(t y)
.
n 1 Ry
(6.1)
Thus the string describes a uniform helix with one turn in the covering space of the S 1 . At
weak coupling g 0 we get a ring with radius a in flat space, while at strong coupling we get
a geometry like Fig. 2(d) with the circle (6.1) sinking deep into the throat (the dotted line in the
figure).
In [3] the computations were done in the D1D5 duality frame, so let us start with that frame
and dualize back to NS1P later. We will denote quantities in the D1D5 frame by primes. The
time for a supergravity quantum to make one trip down the throat and back up is
tosc = Ry ,
(6.2)
where Ry is the radius of S 1 in the D1D5 frame. When the quantum reaches the neck there
is a probability P that it would escape to infinity, and a probability 1 P that it would reflect
back down the throat for another cycle. For low energy quanta in the lth spherical harmonic this
268
(6.3)
where is the energy of the quantum. We see that the escape probability is highest for the
swave, so we set l = 0. Then the expected time after which the trapped quantum will escape is
tescape = P01 tosc .
(6.4)
The low energy quanta in the throat have 2/tosc [3,12] so for our estimate we set
=
2
.
Ry
(6.5)
We then find
1/4 4
Ry 4
(Q1 Q5 )
tescape
1
1
=
=
,
2
2
tosc
a
(2) Q1 Q5 (2)
(6.6)
where a = (Q1 Q5 )1/2 /Ry is the radius obtained from a after the dualities to the D1D5 frame
[3]. In this frame the cap + throat region has the geometry of global AdS3 S 3 T 4 . The
curvature radius of the AdS3 and S 3 is (Q1 Q5 )1/4 . The ratio
(Q1 Q5 )1/4
(6.7)
a
gives the number of AdS radii that we can go outwards from r = 0 before reaching the neck
region.11 Thus is a measure of the depth of the throat compared to its diameter.
While all lengths in the noncompact directions are scaled under the dualities, the ratio of such
lengths is unchanged. Thus in the NS1P duality frame
(Q1 Qp )1/4
.
a
We note that
=
(6.8)
4.
(6.9)
Thus when the throat becomes deep the quanta trapped in the throat become long lived excitations
of the system.
For completeness let us also start from the other limit, where the coupling is weak and we have
a thin long tube as in Fig. 2(b). The radius of the ring described by (6.1) is a. The gravitational
effect of NS1, P charges extends to distances Q1 , Qp from the ring. In the definitions (4.8) we
put in the profile (6.1), and find
Qp
Q1
,
Qp =
.
(6.10)
2a
2a
If we take for the thickness of the ring the length scale Q1 Qp then from (6.10) we find
Q1 =
(6.11)
269
so we see again that the ring thickness becomes comparable to the ring radius when 1. For
1 we have a thin
ring and for
1 we have a deep throat.
Instead of using Q1 Qp as a measure of the ring thickness we can say that the ring is thin
when
a Q1 ,
a Qp .
(6.12)
Q1 Qp
.
Q1 + Qp
(6.13)
From the first equality in (4.65) we get an expression for t , which we equate to the expression
found in (2.25), to get
1 Q1 + Qp
= t = MT ,
Lz
2
Q1 Qp
(6.14)
where Lz = 2a is the length of the ring and MT is its total mass. Using this and (6.10) we can
rewrite (6.13) as
MT (Q1 Qp )1/2 =
(Q1 Qp )1/2
.
2a
(6.15)
Expressing the macroscopic parameters in terms of the microscopic charges and moduli12
Q1 =
g 2 3 n1
,
V
Qp =
g 2 4 np
,
V Ry2
a=
n1 np
(6.16)
g2 3
.
V Ry
(6.17)
This version of the criterion for ring thickness will be of use below.
6.2. The phase transition in the microscopic picture
We now turn to the microscopic description of the system. Consider first the BPS bound state
in the D1D5 duality frame. Suppose we add a little bit of energy to take the system slightly
above extremality. From the work on nearextremal states [3,15,33,34] we know that the energy
will go to exciting vibrations that run up and down the components of the effective string
D1D5 + E D1D5 + PP,
(6.18)
where we call the excitations P P since they carry momentum charge in the positive and negative
S 1 directions. For the geometry made by starting with the NS1P profile (6.1) we have no fractionation; i.e., the effective string formed in the D1D5 bound state has n1 n5 singly wound
circles [3]. Thus the minimum energy needed to excite the system is the energy of one left and
12 The expression for a is obtained by using the profile (6.1) in (4.3) and use the expressions (4.4).
270
1
1
2
+ = .
Ry
Ry
Ry
(6.19)
The charges D1D5P can be permuted into each other, so we can map D1D5 to PD1, and
then the dual of (6.18) is
PD1 + E PD1 + D5D5.
(6.20)
A further S duality brings the system to the NS1P system that we are studying, and then we get
PNS1 + E PNS1 + NS5NS5.
(6.21)
This may look strange, since it says that if we excite an oscillating string the energy of excitations
goes to creating pairs of NS5 branes; we are more used to the fact that energy added to a string
just creates more oscillations of the string. Dualizing (6.19) gives for the excitation (6.21) the
minimum energy threshold
NS1P
= 2mNS5 = 2
ENS5NS5
V Ry
.
g2 3
(6.22)
Thus at small g these excitations are indeed heavy and should not occur. For comparison, we find
the minimum energy required to excite oscillations on the NS1P system. For small g we use the
spectrum of the free string which gives
Ry n 1
np 2
Ry n 1
np 2
4
4
+
+
N
=
+ NR .
M2 =
(6.23)
L
Ry
Ry
2 1
,
MT
(6.24)
271
couple only weakly to the radiation modes of the gravity field. When the tube becomes very thick
then the oscillation modes are again long livedwe get
1 and by (6.9) this implies a very
slow leakage of energy to infinity.
Thus we see that the modes at small and large coupling should not be seen as the same
modes; rather the 2charge modes at weak coupling disappear at larger g because of coupling
to the radiation field, and at still larger g the 3charge modes appear. For these latter modes
one might say that the gravitational field of the system has trapped the excitations of the metric
from the region r (Q1 )1/2 , (Qp )1/2 , so that these modes have in some sense been extracted
from the radiation field.
7. A conjecture on identifying bound states for the 3charge extremal system
Consider a D0 brane placed near a D4 brane. The force between the branes vanishes. But now
give the D0 a small velocity in the space transverse to the D4. The force between the branes goes
as v 2 , and the motion of the D0 can be described as a geodesic on the moduli space of its static
configurations [7]. This moduli space would be flat if we took a D0D0 system (which is 1/2
BPS) but for the D0D4 case (which is 1/4 BPS) the metric is a nonflat hyperkhler metric.
We can look at more complicated systems, for example 3charge black holes in 4 + 1 spacetime. Now the system is 1/8 BPS. The positions of the black holes give coordinates on moduli
space, and the metric on moduli space was computed in [8]. If we set to zero one of the three
charges then we get a 1/4 BPS system.
It is easy to distinguish motion on moduli space from the kinds of oscillatory behavior that
we have encountered in the dynamics of supertubes. As mentioned in the introduction, when we
have motion on moduli space we take the limit of the velocity going to zero, and over a long
time t the system configuration changes by order unity. Using x as a general symbol for the
change in the configuration13 we have for drift on moduli space
1
t ,
(7.1)
x 1 ( 0).
On the other hand for the periodic behavior that we have found for both the weak coupling and
strong coupling dynamics of bound states, we have
v ,
v ,
t 1,
x
( 0).
(7.2)
Note that for the motion (7.1) the energy lost to radiation during the motion vanishes as 0,
so the dynamics (7.1) is unlike any of the cases that we have discussed for the bound state.
While the moduli space metric in [8] was found for spherically symmetric black holes (naive
geometries in the language of [3]) we expect that a similar drifting motion would occur even if
we took two actual geometries of the 2charge system and gave them a small relative velocity
with respect to each other.14 Thus such unbound systems would have a dynamical mode not
present for the bound states.
For the bound state 3charge geometries that have been constructed [16] the structure is very
similar to the structure of 2charge geometries. It is therefore reasonable to conjecture that 3charge geometries will have a similar behavior: Unbound systems will have drift modes like
13 For example x could be the separation of two black hole centers.
14 The motion of the centers of the two states could be accompanied by a slow change in the internal configurations of
the states.
272
(7.1) while bound systems will have no such modes. If true, this conjecture could be very useful
for the following reason. It is known how to write down the class of all 3charge supersymmetric
geometries [17,35,36]. But we do not know which of these are bound states. On the other hand
the microstates of the 3charge black hole [37] are bound states of three charges. If we can select
the bound state geometries from the unbound ones by some criterion then we would have a path
to understanding all the microstates of the 3charge black hole. This is important because the
3charge hole has a classical horizon and our results on this hole should extend to all holes.
To summarize, it seems a reasonable conjecture that out of the class of all supersymmetric
3charge geometries the bound states are those that have no drift modes (7.1). It would be
interesting to look for drift modes for the 3charge geometries constructed recently in [38,39];
here the CFT dual is not known so we do not know a priori if the configuration is a bound state.
The same applies for geometries made by adding KKmonopole charge to BPS systems carrying
a smaller number of charges [40]. It would also be useful to extend these considerations to the
suggested construction of 3charge supertubes and their geometries [41,42].
In the introduction we have also asked the question: can the bound state break up into two or
more unbound states under a small perturbation? Since the bound state is only threshold bound,
such a breakup is allowed on energetic grounds. But for the 2charge system we see that bound
states are not close to unbound states. The bound states are described by a simple closed curve
traced out by the locations x = F (v) for 0 v < LT . A superposition of two such bound states
has two such simple closed curves. The curve can break up into two curves if it selfintersects,
but in a generic state the curve is not selfintersecting. If we add a little energy to a 2charge
bound state then we have seen that the configuration does not drift through the space of bound
states, so the curve will not drift to a curve with a selfintersection and then split. Thus we expect
that generic bound states are stable to small perturbations; energy added to them causes small
oscillations for a while and the energy is eventually lost to infinity as radiation.
8. Discussion
Let us summarize our arguments and conjectures. If we have two BPS objects with 1/4 susy
then they feel no force at rest, but their low energy dynamics is a slow relative motion described
by geodesics on a moduli space. If we look at just one 1/4 BPS bound state then it has a large degeneracy, which in the classical limit manifests itself as a continuous family of timeindependent
solutions. If we add a small energy to the BPS bound state, then what is the evolution of the
system?
Based on the behavior of unbound objects one might think that there will again be a drift over
the family of configurations, described by some metric on the moduli space of configurations.
But we have argued that this is not what we should expect. We first looked at the 1/4 BPS
configurations at zero coupling, where we get supertubes described by a DBI action. We saw
that the best way to get the dynamics of such 1/4 BPS objects is to use the NS1P picture,
which is a multiwound string carrying a traveling wave. For this zero coupling limit we found
that instead of a drift over configurations we get oscillatory behavior. These oscillations are
not described by a collection of simple harmonic oscillators. Rather they are like the motion of
a charged particle in a magnetic field where each term in the equation of motion has at least
one time derivative, and there is a continuous family of equilibrium configurations. We found a
simple expression (2.25) for the period of oscillations with arbitrary amplitude, which reduced
to the period found in [13] for the case of small oscillations of the round supertube.
273
If we increase the coupling a little then we get a gravity description of the supertube, but with
gravitational field of the tube extending only to distances small compared to the circumference of
the tube. Thus we get a thin long tube. Zooming in to a point of the tube we see an essentially
straight segment, and we studied the perturbations to this geometry. We found excitations that
agree in frequency with those found from the zero coupling analysis.
We noted that the part of the excitation that leaks out to spatial infinity will have the form
of a traveling wave. As we increase the coupling the amplitude of the wave reaching this region
becomes larger. Thus there will be an energy flux leaking out to infinity, and the excitation will
not remain concentrated near the supertube. But as we increase the coupling still further we
find that the geometry develops a deep throat and we get a new kind of long lived excitation:
Supergravity modes can be trapped in this throat for long times, only slowly leaking their energy
to infinity.
We argued that the different kinds of excitations found at weak and strong coupling reflect
the phase transition that had been noted earlier from the study of black holes [14,15]. At weak
coupling the excitations on such a system creates pairs of the charges already present in the BPS
state. But at larger coupling the excitation energy goes to creating pairs of a third kind of charge.
The value of g where this transition occurs was found to have the same dependence on V , R, ni
as the value of g where the supertube stops being thin; i.e. where the gravitational effect of the
tube starts extending to distances comparable to the radius of the tube.
We have noted that bound states do not exhibit a drift over a moduli space of configurations,
while unbound states do. If 3charge systems behave qualitatively in the same way as 2charge
ones then this fact can be used to distinguish bound states from unbound ones for the class of
1/8 BPS states; such bound states would give microstates of the 3charge extremal hole.
Acknowledgements
This work is supported in part by DOE grant DEFG0291ER40690. S.G. was supported by
an INFN fellowship. We would like to thank Camillo Imbimbo and Ashish Saxena for helpful
discussions.
Appendix A. Analysis of the A+ equations
In this appendix we look at the field equations (4.18) for the field A+ . We have found the
expected timeindependent solutions in Section 4.2. We will now consider a general ansatz for
the solution and argue that there are no time dependent solutions for this field, if we demand
consistency with the long wavelength limit (4.30).
Let us write
1 +
A+
av ,
v =H
+
A+
t = at ,
+
A+
z = az ,
+
A+
i = ai .
(A.1)
Since we have spherical symmetry in the space spanned by the coordinates i all fields will be
functions only of the radial coordinate r in this space; further, the A+
i can have only the component A+
r . Putting this ansatz into (4.18) we obtain the coupled system of equations (we list the
equations in the order = t, v, z, i)
at+ + t2 av+ t i ai+ + z z at+ t az+ + At z at+ t az+ = 0,
(A.2)
av+ + z2 av+ H K A2 t2 av+ 2At z av+
+ i H K A2 i at+ 2i Ai az+ i H K A2 t ai+ 2i Az ai+ = 0, (A.3)
274
az+ + t z av+ z i ai+ + H K A2 t z at+ t az+ Az z at+ t az+ = 0, (A.4)
t i av+ z i az+ + z2 ai+ j i aj+ j ai+ + i A z at+ t az+
+ H K A2 t i at+ Az i at+ At i az+
H K A2 t2 ai+ 2At z ai+ = 0.
(A.5)
We look for a gauge field A+ having the same z and t dependence as A . We write
av+ = eikzit fv (r),
ar+ = ieikzit r ,
(A.6)
where the notation we have used for ar+ will be helpful in what follows.
Consider first Eqs. (A.2) and (A.4):
ft 2 fv (kft + fz )(k A) = 0,
fz + kfv + k + (kft + fz ) kA + H K A2 = 0.
Taking k times the first equation plus times the second we get
(kft + fz ) k 2 2kA 2 H K A2 (kft + fz ) = 0.
(A.7)
(A.8)
We note that the coefficient k 2 2kA 2 (H K A2 ) has the form c + d/r. If d = 0 then by
an argument similar to that leading to (4.59) we find that there is no solution with the required
behavior at r = 0. Setting d = 0 tells us that we need to have the same relation between and k
that we have seen before (4.59)
2 Q1 Qp
= k
(A.9)
.
Q1 + Qp
Using (A.9) one finds
k 2kA H K A
2
=k
Q1 Qp
Q1 + Qp
2
k 2
(A.10)
and thus
(kft + fz ) k 2 (kft + fz ) = 0.
(A.11)
,
i H K A2 i at+ 2i Ai az+ =
r
2
k
r
r
i H K A2 t ai+ 2i Az ai+ = 0.
kft + fz = c
(A.12)
(A.13)
275
(A.14)
(A.15)
fv will be too singular to agree with (4.30) at small r. The vanishing of c also makes (A.12)
agree with (4.30) at small r.
From (A.14) and using the short distance limit implied by (4.30) we get
ekr
(A.16)
.
r
Consider now the last Eq. (A.5) (for i = r, the only nontrivial component). The fact that c = 0
implies
fv = ( )Q1
z at+ t az+ = 0.
Moreover one has
H K A2 t r at+ Az r at+ At r az+ = ir ft ,
H K A2 t2 ar+ 2At z ar+ = i2 r .
Eq. (A.5) then gives
k2
r fv + 1 2 r (ft ) = 0.
(A.17)
(A.18)
(A.19)
(A.20)
We see that (A.19) implies (A.20). But at this stage we would like to compare the solution we
have found with the limits required by (4.30). First consider the case Q1 = Qp . Then = k and
(A.19) is not compatible with (A.16). Now consider Q1 = Qp . Eq. (A.19) implies
ft =
2
fv + const.
k 2 2
(A.21)
This is again incompatible with the limit (4.30), according to which ft should vanish for
small r. We conclude that there are no timedependent solutions for A+ consistent with the limit
(4.30).
Appendix B. Asymptotic behavior of the perturbation
In this appendix we will study the behavior of perturbations on the 2charge geometries near
spatial infinity. We wish to see how fields fall off with r, and in particular to check that they
carry energy flux off to infinity. We will first look at a scalar field to get an idea of the behaviors
involved, and then address the 1form gauge field that is actually excited in our problem.
276
h=
1+
Q1
r 2
Qp
1+ 2 .
r
(B.2)
(B.3)
where Y (l) the lth scalar spherical harmonic. The wave equation for
= 0
(B.4)
implies
Q1 + Qp Q1 Qp
1 3
l(l + 2)
2
r
R
+
+
R = 0.
1
+
R
r
r
r 3
r 2
r 4
r 2
(B.5)
(B.6)
r 3/2
(B.7)
At leading order in 1/r the terms proportional to 1/r 2 and 1/r 4 can be neglected and we have
the solution
r+ eir + r eir
(B.8)
r 3/2
which corresponds to traveling waves carrying a nonzero flux. When the terms of higher order in
1/r are included, (B.7) can be recursively solved as a formal power series in 1/r:
(n)
(n)
r+
r
eir
eir
R r+ 3/2 1 +
(B.9)
+ r 3/2 1 +
.
r n
r n
r
r
R = r+ eir + r eir
n=1
(n)
R=
n=1
277
(n)
The r are finite for any value of and any n. However, since at large n one has
(n)
r
(n1)
r
i(n 1)
2
(B.11)
the series (B.9) has zero radius of convergence. Equations like (B.5) lead instead to asymptotic
series in 1/r [29], and we expect that the above expansion is to be interpreted as an asymptotic
series, which accurately describes the behavior of R at sufficiently large r. From (B.9) we can
still conclude that the perturbation R radiates a finite amount of flux at infinity. Note that, in
order to avoid logarithms in the expansion (B.9), it is crucial that the next to leading corrections
to Eq. (B.7) are of order 1/r 2 .
B.2. The vector perturbation
We find a similar situation for the case in which the perturbation is represented by a vector
on the sixdimensional space R(5,1) S 1 . As we showed in Section 4 the perturbation on the
2charge system is a vector field with wave equation
1
e2 F e2 H F
(B.12)
= 0.
2
e2 = H 1
(B.13)
with
Qp
Q1
(B.14)
,
K = 2 .
r 2
r
The spherical symmetry of the background (B.13) allows us to expand the vector field com(l)
ponents into spherical harmonics: denoting by Y (l) and Y the scalar and vector spherical
3
harmonics on S , we can write
H = 1 +
it
RI (r)Y (l) (, , ),
A
I =e
it
A
Rs (r) Y (l) (, , ) + Rv (r)Y(l) (, , )
=e
(B.15)
Y(l) = 0,
where primed quantities refer to the metric on an S 3
(B.16)
278
1
I J it
=0
RJ J eit Rs
(B.17)
I rg e
r
and the components with = I give
1
3 KL I J it
K r g g L e
RJ J eit RL
3
r
g I J l(l + 2) it
RJ J eit Rs
e
2
r
it
it
IJK Q1
e
=0
R
R
(B.18)
J
K
K
J
r 3
with rty = 1. As expected from group theory considerations, the component Rv decouples from
all others, while Rs and RI satisfy a coupled system of differential equations. We want to show
that, in spite of these mixings, Rv , Rs and RI admit an 1/r expansion analogous to (B.9).
Putting in the explicit value of gI J in (B.17) and (B.18) and using the gauge
A
t =0
(B.19)
r
r
l(l + 2)
Q1 + Qp Q1 Qp
1 3
2
r
R
R
+
+
1
+
Ry
r
r
y
y
r 3
r 2
r 2
r 4
2
Q1 1
3 1+ 2
(Q1 Qp )(iRr + r Ry ) = 0.
(B.24)
r
r
(Eq. (B.22) is the I = r component of (B.18); Eqs. (B.23) and (B.24) are linear combinations of
the I = t, y components of (B.18).)
Eq. (B.20) for Rv is analogous to Eq. (B.7): it thus admits an analogous asymptotic expansion,
of the form
(n)
(n)
rv,+
rv,
eir
eir
+ rv, 1/2 1 +
.
Rv rv,+ 1/2 1 +
(B.25)
r n
r n
r
r
n=1
n=1
279
Rs =
Rs
,
r 1/2
Ry =
Ry
.
r 3/2
(B.27)
We find
Qp
l(l + 2) r Rs
1 l(l + 2)
Rr =
Rs + i
+ 7/2
r Ry + O r 9/2
2
5/2
2
r
r
2
(B.28)
and
2 (Q1 + Qp ) + 1/4
l(l + 2) 2
Rs +
r Rs + O r 3 = 0,
2
2
2
r
r
2 (Q + Q ) l(l + 2) 3/4
1
p
r2 Ry + 2 Ry +
Ry + O r 3 = 0.
2
r
r2 Rs + 2 Rs +
(B.29)
In (B.28) and (B.29) we have organized the powers of 1/r by assuming that Rs , Ry and their
rderivatives are of order r 0 : by looking at (B.29), we see that this assumption is actually implied
by the equations themselves. Note also that the next to leading corrections in (B.29) are of oder
1/r 2 . We can thus conclude that Rs and Ry have the form
Rs
Ry
(n)
(n)
eir rs,+ eir rs,
+
,
r n
r n
r 1/2
r 1/2
n=0
(n)
eir ry,+
r n
r 3/2
n=0
n=0
(n)
eir ry,
.
r n
r 3/2
n=0
(B.30)
(n)
(n)
Analogously to the case of the scalar perturbation, the coefficients rs, and ry, for n > 1 are
(0)
(0)
recursively determined from rs, and ry, , and are finite for any value of and any finite n.
Substituting in (B.28) we have the solution for Rr
Rr
(n)
(n)
eir rr,+ eir rr,
+
,
r n
r n
r 5/2
r 5/2
n=0
(n)
(B.31)
n=0
(n)
(n)
(0)
where rr, are determined in terms of rs, and ry, (the leading coefficients rr, vanish for
l = 0).
280
The components Rs and Ry give rise to nonvanishing energy flux at infinity while Rr does
not contribute to the flux at leading order. Note that because Rr is zero if both Rs and Ry vanish,
it is not possible to have a solution in which only Rr is excited, and thus all solutions carry some
flux at infinity.
References
[1] C. Vafa, Nucl. Phys. B 463 (1996) 435, hepth/9512078.
[2] A. Sen, Nucl. Phys. B 440 (1995) 421, hepth/9411187;
A. Sen, Mod. Phys. Lett. A 10 (1995) 2081, hepth/9504147.
[3] O. Lunin, S.D. Mathur, Nucl. Phys. B 623 (2002) 342, hepth/0109154.
[4] D. Mateos, P.K. Townsend, Phys. Rev. Lett. 87 (2001) 011602, hepth/0103030;
R. Emparan, D. Mateos, P.K. Townsend, JHEP 0107 (2001) 011, hepth/0106012;
D. Mateos, S. Ng, P.K. Townsend, JHEP 0203 (2002) 016, hepth/0112054.
[5] D.S. Bak, A. Karch, Nucl. Phys. B 626 (2002) 165, hepth/0110039.
[6] M. Kruczenski, R.C. Myers, A.W. Peet, D.J. Winters, JHEP 0205 (2002) 017, hepth/0204103.
[7] M.R. Douglas, D. Kabat, P. Pouliot, S.H. Shenker, Nucl. Phys. B 485 (1997) 85, hepth/9608024.
[8] D.M. Kaplan, J. Michelson, Phys. Lett. B 410 (1997) 125, hepth/9707021.
[9] J. Michelson, A. Strominger, JHEP 9909 (1999) 005, hepth/9908044.
[10] R.R. Khuri, Phys. Lett. B 294 (1992) 331, hepth/9205052.
[11] O. Lunin, S.D. Mathur, Nucl. Phys. B 615 (2001) 285, hepth/0107113.
[12] O. Lunin, S.D. Mathur, A. Saxena, Nucl. Phys. B 655 (2003) 185, hepth/0211292.
[13] B.C. Palmer, D. Marolf, JHEP 0406 (2004) 028, hepth/0403025.
[14] S.D. Mathur, Nucl. Phys. B 529 (1998) 295, hepth/9706151.
[15] S.D. Mathur, hepth/0510180.
[16] S.D. Mathur, A. Saxena, Y.K. Srivastava, Nucl. Phys. B 680 (2004) 415, hepth/0311092;
S. Giusto, S.D. Mathur, A. Saxena, Nucl. Phys. B 701 (2004) 357, hepth/0405017;
S. Giusto, S.D. Mathur, A. Saxena, hepth/0406103;
O. Lunin, JHEP 0404 (2004) 054, hepth/0404006.
[17] J.P. Gauntlett, J.B. Gutowski, C.M. Hull, S. Pakis, H.S. Reall, Class. Quantum Grav. 20 (2003) 4587, hepth/
0209114;
J.B. Gutowski, D. Martelli, H.S. Reall, Class. Quantum Grav. 20 (2003) 5049, hepth/0306235.
[18] M. Taylor, hepth/0507223.
[19] O. Lunin, S.D. Mathur, Nucl. Phys. B 610 (2001) 49, hepth/0105136.
[20] O. Lunin, S.D. Mathur, Phys. Rev. Lett. 88 (2002) 211303, hepth/0202072.
[21] S.D. Mathur, Fortschr. Phys. 53 (2005) 793, hepth/0502050.
[22] D. Mateos, S. Ng, P.K. Townsend, Phys. Lett. B 538 (2002) 366, hepth/0204062.
[23] D. Bak, Y. Hyakutake, N. Ohta, Nucl. Phys. B 696 (2004) 251, hepth/0404104;
D. Bak, Y. Hyakutake, S. Kim, N. Ohta, Nucl. Phys. B 712 (2005) 115, hepth/0407253.
[24] S.R. Das, A. Jevicki, S.D. Mathur, Phys. Rev. D 63 (2001) 024013, hepth/0009019.
[25] O. Lunin, S.D. Mathur, I.Y. Park, A. Saxena, Nucl. Phys. B 679 (2004) 299, hepth/0304007.
[26] Vachaspati, T. Vachaspati, Phys. Lett. B 238 (1990) 41;
D. Garfinkle, T. Vachaspati, Phys. Rev. D 42 (1990) 1960;
A. Dabholkar, J.P. Gauntlett, J.A. Harvey, D. Waldram, Nucl. Phys. B 474 (1996) 85, hepth/9511053;
C.G. Callan, J.M. Maldacena, A.W. Peet, Nucl. Phys. B 475 (1996) 645, hepth/9510134.
[27] O. Lunin, J. Maldacena, L. Maoz, hepth/0212210.
[28] J. Maharana, J.H. Schwarz, Nucl. Phys. B 390 (1993) 3, hepth/9207016.
[29] A. Erdlyi, Asymtotic Expansions, Dover, New York, 1987.
[30] M. Reed, B. Simon, Analysis of Operators, Methods of Modern Mathematical Physics, vol. 4, Academic Press, New
York, 1978, see Theorems XIII. 56, XIII. 57, XIII. 58.
[31] V. Balasubramanian, J. de Boer, E. KeskiVakkuri, S.F. Ross, Phys. Rev. D 64 (2001) 064011, hepth/0011217.
[32] J.M. Maldacena, L. Maoz, JHEP 0212 (2002) 055, hepth/0012025.
[33] J.M. Maldacena, A. Strominger, Phys. Rev. D 55 (1997) 861, hepth/9609026.
[34] C.G. Callan, J.M. Maldacena, Nucl. Phys. B 472 (1996) 591, hepth/9602043.
[35] I. Bena, N.P. Warner, hepth/0408106.
[36]
[37]
[38]
[39]
[40]
281
Abstract
Perturbation of logarithmic conformal field theories using Zamolodchikovs method is investigated. We
derive conditions for the perturbing operator, such that the perturbed model be integrable. We also consider
an example where an integrable model arises out of perturbation of the c = 2 logarithmic conformal field
theory.
2006 Elsevier B.V. All rights reserved.
Keywords: Logarithmic conformal field theory; Integrable models; c theorem
1. Introduction
Conformal field theories (CFT) [1] describe the behavior of a system at its critical point.
Also in two dimensions CFTs are integrable since the conformal algebra is infinitedimensional.
Therefore it is natural to expect that perturbation of a CFT by an operator may lead to an integrable model in two dimensions [2]. However not all perturbations may lead to integrable models.
The perturbing operator has to be chosen carefully so that an infinite number of currents remain
conserved, therefore the structure of the CFT and the perturbing operator becomes important.
6
, p = 3, 4, 5, . . . , and perturbThe case for unitary models with central charge c = 1 p(p+1)
ing field being 1,2 , 2,1 , 1,3 was analyzed in [2]. The usefulness of this approach lies in the
fact that one may use the structure of CFT to investigating the integrable model. In fact using
this device Zamolodchikov solves the Ising model in two dimension in presence of a magnetic
field [2,3]. The other example is the sinhGordon theory which can be viewed as perturbation of
a Liouville theory [4].
* Corresponding author.
284
In this paper we address the same question for a logarithmic conformal field theory (LCFT)
where perturbation leads to an integrable model with at least two continuity equations. The difference between LCFT and CFT, lies in the appearance of nondiagonizable groups of operators
which all have the same conformal weight, this leads to appearance of logarithms in the correlation functions [68]. We follow Zamolodchikovs method and observe that for some LCFTs
there are two methods for arriving at integrable models associated with the two primary fields
which are partners in LCFT structure. These two theories can be investigated in the compact
form by using the nilpotent variables [9,10] defining an algebra similar to the Zamolodchikovs
algebra [2]. In fact for investigating the integrability of these theories we need to check the consistency of S matrix of these theories but in the first step we just try to find the extra conserved
quantities. You can see the S matrix of some models in Refs. [2,3,5,11,12].
This paper is organized as follows: In section one after a brief review of integrable models,
we go on to review Zamolodchikovs approach to perturbation of CFT [2,5]. In section two we
review Zamolodchikov ctheorem in LCFTs using the nilpotent variable approach. In the last
section, we extend the Zamolodchikov approach to LCFTs and then apply the formalism to the
c = 2 theory.
1.1. Perturbed CFT and integrable models
If an infinite number of local integrals of motion survives after perturbing a CFT, then the
newly constructed model is integrable. For an infinite number of conserved charges Ps to survive
we should make some assumptions, which we investigate in this section. Suppose we perturb the
critical action S by a relevant scalar operator:
S = S + (z, z) d 2 z,
(1)
where the weight of is (h, h), and the dimension of is (1 h, 1 h). For to be a relevant
perturbation, we need y = 2 h h > 0. The integrals of motion, surviving the perturbation are:
Ps = [Ts+1 dz + s1 d z],
(2)
where Ts and s are local fields of spin s, satisfying the continuity equation;
z Ts+1 = z s1 .
(3)
Let us consider scattering of n particles Aa , a = 1, 2, . . . , n, whose masses are ma . Their momenta satisfy the massshell condition p p = p p = m2 , where the components of p are
p = p 0 + p 1 and p = p 0 p 1 . In order to ensure exact integrability we assume that the field
theory possess an infinite number of nontrivial, commutative integrals of motion. These nontrivial, i.e. other than energymomentum, conserved charges transform as sth order tensors under
the Lorentz group, we call them Ps , s = s1 , s2 , . . . , s indicates the spin of Ps . In two dimensions
spin refers to Lorentzspin, and Ps transforms under Lorentz transformations L : = +
with the following form:
Ps Ps = es Ps ,
(4)
where is the change in rapidity. Therefore we have P1 = p and P1 = p. Since parity relates
the integrals of motion Ps to Ps , we can consider only s > 0 for parity conserving theories.
285
(5)
sa (p)
to be of the
(6)
Since Ps are integrals of local densities the action of Ps on well separated multiparticle in or
out states is the sum of the oneparticle contributions so in a scattering process p1 , . . . , pn
the following equality holds:
p1 , . . . , pm
n
m
(pi )s =
(pi )s .
i=1
(7)
i=1
If at least one nontrivial conservation law exists such that s > 1, we have n = m, which means
there is no particle production and only timedelays and exchange of quantum numbers are allowed [13]. Therefore one can conclude that the final set of momenta and energies coincides with
the initial one.
The other general conclusion for integrable quantum field theory is that nparticle Smatrix is
a product of n(n 1)/2 twoparticle Smatrices. This factorization can be effected in different
ways and all of them must give the same result the consistency conditions are called YangBaxter
factorization equations [1416].
Now let us investigate the relation between integrable models and relevant perturbations of
conformal field theories (CFT). By igniting the perturbation the system will flow away from its
UV fixed point and may end up at another critical conformally invariant fixed point.
Let us go back to Eq. (1) where S describes a CFT theory, which contains as one of
its operators and we assume, that all its correlation functions are known. First of all we check
whether there exist currents J (z, z), whose conservation survives the perturbation.
The correlation functions of a particular operator J (z, z) for perturbed action are given by the
following equation
J (z, z) = J (z, z) S + d 2 z1 J (z, z)(z1 , z1 ) S + O 2 .
(8)
For the cases which this integral are finite, any z dependence come from possible singular
points z z1 so in the neighborhood of z1 we can use the operator product expansion (OPE):
ai
i (z1 , z1 ),
J (z, z)(z1 , z1 ) =
(9)
z z1 J +i
i
where = 2h and J and i are the scaling dimensions of J and i and the equation is true
for every zk . This singularities will be integrable if J + i < 2, since in a unitary theory
all dimensions are positive, only a finite number of operators i will contribute in the correlation
expansion in the first order of .
In the particular example of the energymomentum tensor the OPE is
T (z)(z1 , z1 ) =
h
1
(z1 , z1 ) +
1 (z1 , z1 ).
2
z z1
(z z1 )
(10)
Using Eq. (8) and regularizing the second term by cutting out a small section z z1 2 a 2 ,
where a is some microscopic length scale we immediately get the conservation law for the
286
energymomentum tensor as expected. Since the energymomentum tensor must remain conserved z T + z = 0, where
= (h 1)(z, z).
(11)
As shown by Zamolodchikov [2], it follows from the ward identities that there is a set of
operators Dn such that:
d
Dn (z, z) =
(12)
(, z)( z)n (z),
2
z
z = D0 ,
1
Dn1 I = zn (z, z),
n!
[Ln , Dm ] = (1 h)(n + 1) + m Dn+m .
(13)
(14)
(15)
A simple application is
z T (z, z) = D0 L2 I = (h 1)D2 I = (h 1)L1 (z, z).
(z) = :T 2 (z):
(16)
which is defined as
(17)
h3 3
= (h 1)(D2 L2 + L2 D2 )I = (h 1) 2L2 L1 +
L1 .
6
(18)
For a general the righthand side cannot be written as a derivative. However certain s might
resolve this problem. As an example take as perturbation the field 1,3 of the unitary models with
c < 1. It has the following nullvector equation at level 3:
2
1
L1 L2 +
L31 1,3 (z) = 0.
L3
(19)
(h + 2)
(h + 1)(h + 2)
Now one can use (19) to rewrite (18) in the form z T4 (z, z) = z 2 (z, z), with
h1
(h 2)(h 1)(h + 3) 2
2hL2 +
L1 1,3 ,
2 =
h+2
6(h + 1)
(20)
where T4 is in the conformal tower of the identity and 2 is in the conformal tower of the perturbing operator . In general the existence of a conservation law is equivalent to saying linear
operator z , acting between these two conformal towers has a nonvanishing kernel, up to derivative fields. So if s and s be the dimensions of the spaces of quasiprimary fields constructed
in the conformal towers of either the identity or the perturbing field at the level s then, if the
condition s+1 s + 1 is satisfied, it must have a nonvanishing kernel which is equivalent to
the existence of a conservation law. The above method is useful for finding conservation laws for
small s and named Zamolodchikov counting argument. For the above theories by using counting
argument see that for s = 1, 3, 5, 7 there is conservation law. By comparing with KdV equation
287
when c goes to infinity Zamolodchikov conjectured that there must be continuity equation for all
odd s [2]. As an other example, for the Ising model perturbed by a magnetic field, 1,2 , the application of above counting criterion illustrates that Ts is indeed conserved if s = 1, 7, 11, 13, 17, 19.
Using purely elastic scattering theory Zamolodchikov conjectured [2] the existence of integrals
of motion with spins s, s = 1, 7, 11, 13, 17, 19, 23, 29 mod 30. This method is applicable to other
models such as LeeYang model and Zn models and so on [2,5]. One can use Zamolodchikovs
counting argument for theories which are perturbed by field 1,2 which have null vector in level
3
L21 )1,2 = 0. For these theories in levels s = 1, 5, 7, 11 there are some
two, (L2 2(2h+1)
continuity equations [2]. For example the T6 has the following form:
1
18
+ h 2 L23 1.
T6 = L32 1
(21)
4 (2h + 1)
It is generally believed [2] that these first few conserved currents are just the first few representatives of the infinite sets of conserved currents with spins s = 1, 6n 1, n = 1, 2, 3, . . . .
2. Perturbation by logarithmic operators
The ctheorem [17] concerns the behavior of renormalization group flows in the subspace
of all interactions in the continuum limit. This theorem holds just for unitary, renormalizable
quantum field theories in two dimensions, it asserts that there exists a function c of coupling
constants which is monotonically decreasing along the renormalization flow and it is stationary
at the fixed point and takes as its values at these fixed points the corresponding central charge.
The proof is based on rotational invariance, positivity, the conservation of the stress tensor and
renormalizability [17]. This theorem implies the following formula for the change in the central
charge c = cUV cIR between two fixed points
c = 12
R 2 (R)(0) d R 2 ,
(22)
where is the nonzero trace of energymomentum tensor, it is given by Eq. (11) for CFT.
Eq. (22) is useful for finding the central charge of one fixed point given the central charge of the
theory at the other [18,19].
The extension of the c theorem to LCFTs has some difficulties [21,22]. First the logarithmic
theories are not unitary so there is not reflection positivity, second the logarithm in the response
function changes the renormalization equations. However Eq. (22) still holds with new s.
Before establishing the renormalization flows in LCFT let us briefly summarize LCFTs using
a nilpotent weight method introduced in [9,10]. The difference between an LCFT and CFT, lies
in the appearance of logarithmic as well as powers in the singular behaviors of the correlation
functions. In LCFT, nondiagonizable groups of operators may exist which all have the same
conformal weight [68]. They form a Jordan cell under the action of L0 . In the simplest case a
pair of operators exist which transform according to
(z) = h (z) (z) log .
(z) = h (z),
(23)
Using nilpotent variables i2 = 0, i j = j i and the construct (z, ) = (z) + (z) we
arrive at the following equation instead of (23):
(z, ) = (h+) (z, ).
(24)
288
In this formalism the two point functions have the following form:
b(1 + 2 ) + d1 2
.
(z1 , 1 )(z2 , 2 ) =
(z1 z2 )2h+1 +2
(25)
In addition one can write the OPE of T and (z, ) as the following form
T (z)(z1 , ) =
h+
1
(z1 , ) +
1 (z1 , ).
z z1
(z z1 )2
(26)
Now, we may perturb the fixed point action by or by a pair of logarithmic operators and
S = S + d d 2 z ( )(z, ),
(27)
where () = + and the integral over is the Grassmanian integral.
Similar to the previous section one can use the OPE of T by and and find the following
continuity equation
z T + z = 0, = (h 1) + (h 1) + .
(28)
Similar to the ordinary CFT case the energymomentum conservation is obtained. If = 0 then
the case is similar to ordinary CFT [20] but if = 0 then the renormalization flow will change.
To calculate the renormalization flow of and we need the OPE coefficients which in
LCFT have the following form
(z1 , 1 )(z2 , 2 ) = zh z1 z2 d z A(1 , 2 ) + B(1 , 2 ) (z, ),
A(1 , 2 ) = A + (1 + 2 )D + 1 2 G,
B(1 , 2 ) = B + (1 + 2 )E + 1 2 K.
(29)
Under a length rescaling of partition function by 1 + t, it can be shown that the lowest order
renormalization group equation for the action (27) is
() = t
d
( + ) = ( )
dt
= (2 2h 2 )( )
d1 d2 (1 )(2 ) A(1 , 2 ) + B(1 , 2 ) .
(30)
There is no potential in this case however in covariantized form there is a gradient flow for
the coupling constants [22]. For covariantization one should contract the functions with the
Zamolodchikov metric on the moduli space of perturbed CFT which is defined as the following
form in a neighborhood of fixed point action
G(1 )(2 ) (z1 z2 )2h (z1 , 1 )(z2 , 2 ) (z z )=a ,
(31)
1
where a is the short distance cutoff. Using (25) give the exact form of the metric
b(1 + 2 ) + d1 2
G S , 1 , 2 =
.
a 1 +2
Using the above metric gradient flow equation is written as
C
C
=
d1 d2 G(1 , 2 )(1 )1 ,
=
d1 d2 G(1 , 2 )(1 ).
(32)
(33)
289
By using the curlfree condition for the function C provided that the OPE coefficient satisfy the
conditions E = A, B = 0, b(K D) = dE the potential for the renormalization group flow is
therefore written as
Ln , Dm ( ) = (1 h )(n + 1) + m Dn+m ( ).
(36)
(37)
(38)
(39)
In which the first piece is similar to (16) and the second is similar to (28). This method is
useful for finding higher conservation laws for example z T4 has the following form
z T4 = ( )D0 ( )L2 L2 I = ( )(h + 1) D2 ( )L2 + L2 D2 ( ) I
h+ 3 3
L1 ().
= ( )(h + 1) 2L2 L1 +
(40)
6
Before deciding whether T4 is conserved or not we briefly recall some facts about singular vectors
in the context of LCFT. In an LCFT, the representation of the Virasoro algebra is constructed from
290
a compound highest weight vector which form a Jordan cell [9,10]. All the representations are
produced by applying Ln s to this states. There may some representation, in which some of the
descendants are perpendicular to all other vectors including themselves. For example the central
charge and highest weight are (c, h) = (1, 1), (25, 3), (1, 1/4), (25, 5/4), (0, 2), (28, 2)
the following operator becomes singular
2
1
3
L3
(41)
L1 L2 +
L
1,3 (z, ) = 0.
(h + + 2)
(h + + 1)(h + + 2) 1
In these theories T4 is related to a conservation law, where are replaces h with h + in (18)
one can repeat this calculation for T6 and find a new continuity equation, T6 is defined as the
following form
c+2 2
(42)
L3 1.
6
The above method shows that we have an integrable model if we have a null vector in level three.
The wellknown example is the c = 2 model with the following action:
1
d 2 z,
S=
(43)
4
T6 = L32 1
where and are Grassmanian variables. This theory is an example of LCFT with the c = 2
and two logarithmic operators 1 and with conformal weight h = 0 and null vectors in level
three. In this case one can write the null vector of as (L3 L1 L2 + 12 L31 ) = 0. This
theory is connected to the wellknown statistical models such as dense polymer model [23] and
sandpile model [24,25]. If we perturb this theory by then we reach a massive fermionic model
which is integrable. Some of the first conserved currents are
T = L2 I,
T4 = L22 I,
T6 = L32 I,
= m2 ,
m2 2
2 =
L1 ( ),
2
4 = 9m2 L4 ( ).
(44)
(45)
(46)
Using the equations of motion it can be shown that this field theory has an infinite series of
currents T2n satisfying the continuity equation z T2n+2 = z T2n n = 0, 1, 2, . . . , where T2n has
the following form:
T2n = m2n+2 zn zn ,
n = 0, 1, 2, . . . .
(47)
The above calculation shows the existence of conserved quantities for all odd spins in fact this
field theory is a free field theory and therefore the S matrix is free from any pole structure. For
the free ghost theories the correlation functions of some operators was derived before both for
bosonic and fermionic cases [26].
For theories which have a null vector in level two, the counting criterion of Zamolodchikov
is difficult to apply because counting the dimension of levels in LCFT is complex and this is not
known. However for proving integrability we can use the fact that an LCFT has a null vector in
level two
3
L2
(48)
L21 1,2 ( ) = 0
2(2(h + ) + 1)
291
if the central charge and the highest weight be (c, h) = (1, 1/4) = (25, 5/4) = (0, 0). Similar
to the CFT case there are some conservation laws in levels 1, 5, 7, 11. For example in level five
there is a continuity equation which one can calculate using Eqs. (36)(38) and the null vector
equation (48). The explicit expression for T6 is
1
18
+h+ 2 .
T6 = L32 1 aL23 1, a =
(49)
4 (2(h + ) + 1)
The expression for 4 is cumbersome and has the following form
27(h + 1)
3(h + 1)(h + 3)
+
4 =
4(2h + 2 + 1)
4(2h + 2 + 1)2
(h + 1)(h + 3)(h + 5)
+
5!
2a(h + 1)(2h + 2 5) 4
+
L1 1,2 ( )
5!
18(h + 1) 3(h + 1)(h + 3)
(2h + 2 + 1)
2
+ 2a(h + 1) L1 L3 1,2 ( )
9(h + 1)
h+ 1
+ 36
+ 3(h + 1)(h + 3)
2h + 2 + 1
2h + 2 + 1
8a(h + 1) L4 1,2 ( ).
(50)
The (c, h) = (0, 0) theory can be a candidate for percolation, as in this model one has a
zero weight operator and the central charge is also zero. Despite these correspondences, no one
has seen logarithmic structure in percolation explicitly, although some effort in this direction
exists [27].
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
292
Abstract
We study the flux tube junctions in the limit of large magnetic flux. In this limit the flux tube becomes a
wall vortex which is a wall of negligible thickness (compared to the radius of the tube) compactified on a
cylinder and stabilized by the flux inside. This wall surface can also assume different shapes that correspond
to soliton junctions. We can have a flux tube that ends on a wall, a flux tube that ends on a monopole and
more generic configurations containing all three of them. In this paper we find the differential equations that
describe the shape of the wall vortex surface for these junctions. We will restrict to the cases of cylindrical
symmetry. We also solve numerically these differential equations for various kinds of junctions. We finally
find an interesting relation between soliton junctions and dynamical systems.
2006 Elsevier B.V. All rights reserved.
1. Introduction
In a recent series of works [13] we studied the behavior of the AbrikosovNielsenOlesen
(ANO) vortex [4] in the large n limit, where n is the number of quanta of magnetic flux carried by
the vortex. We have seen that in this limit the ANO vortex becomes essentially a bag, analogous
to the bag models of hadrons [5,6]. We now briefly review the essential results.
The theory under consideration is the AbelianHiggs model
2
1
L = F F ( ieA )q V q ,
(1.1)
4
where the potential is chosen such that it has a minimum in the Higgs phase q = q0 = 0. The
ANO vortex is a stringlike soliton that extends in time and one spatial dimension. The simplest
* Corresponding author.
294
Fig. 1. This plot is one of the outcomes of the numerical analysis made in [3]. It shows the profile functions q(r) (red/solid
line) and A(r) (blue/dashed line) for the BPS potential ( = 1) and winding number n = 25 000. (For interpretation of
the references to color in this figure legend, the reader is referred to the web version of this article.)
way to describe it is to introduce cylindrical coordinates (z, r, ) and orientate it in the z direction.
The fields can then be put into the following form
n
(1.2)
A(r).
er
The problem is now to evaluate the profile functions q(r) and A(r) subjected to the boundary
conditions q(0) = 0, q() = q0 and A(0) = 0, A() = 1. The claim is that, for every Higgslike
potential V , in the large n limit the profiles become
2
2
lim A(r) r /RV , 0 r RV ,
lim q(r) H (r RV ),
(1.3)
n
n
1,
r > RV .
q = q0 ein q(r),
A =
This conjecture has been proved in [3] by numerical computations. We show in Fig. 1 the result
obtained for the BPS potential and n = 25 000. The step function of the profile q(r) reveals the
presence of a substructure: a domain wall interpolating between the Coulomb phase q = 0 and
the Higgs phase q = q0 . For this reason we have named this object wall vortex.
The wall vortex is essentially a bag, such as the ones used in the context of the bag models of
hadrons. A domain wall of thickness W and tension TW is wrapped onto a cylinder of radius R.
The bag model is a good approximation when the thickness of the wall is small compared to the
radius of the vortex. There are three energy terms that must be considered. The first one comes
from the tension of the wall and is proportional to the radius R. The second comes from the
energy density of the interior of the bag and is proportional to R 2 . The third energy term comes
from the magnetic flux and is proportional to 1/R 2 . When they are summed we obtain the tension
as function of the radius R:
2n2
+ TW 2R + 0 R 2 .
(1.4)
e2 R 2
The radius of the vortex RV is the one that minimizes this expression. Physically we can understand it in this way. When we derive (1.4) with respect to R we obtain the various forces that act
on the surface of the bag. The first two terms bring a collapse force that tends to squeeze the tube.
T (R) =
295
The third term acts instead as a pressure that tends to expand the tube. When the two opposite
forces are equilibrated we have a stable configuration.
The aim of the present paper is to study a more generic embedding of the wall vortex in
the threedimensional space. The configuration (1.4) has the maximal number of possible symmetries, it is invariant under cylindrical rotation and z translation. Now we want to relax the
last condition, namely we consider a wall that has only cylindrical symmetry. The radius R is
no longer a constant but a function R = f (z), and the force balance will not be an algebraic
equation but a system of differential equations.
Solitons in the Higgs phase have received great attention in the last years. A lot of different
models in different limits have been investigated and the jungle of solitons and soliton junctions
is enormously vast. Here we just mention the works that have mostly influenced the present paper: solitons in nonlinear sigma models [15,16], vortices and walls in supersymmetric models
with a FayetIlioupoulos term [79], the moduli matrix approach [1719], nonAbelian vortices
and their junction with nonAbelian monopoles [1012], the vortexmonopolevortex junction
[13,14] and also the MQCD realization of soliton junctions [2325]. We find particularly amazing the various relations between the two fundamental solitons in the Higgs phase: the vortex
and the wall. More recent works are [2022]. The large magnetic flux limit is a very general
approach, it applies to the AbelianHiggs model (1.1) that is the basic building block of all the
more sophisticated theories that contain solitons in the Higgs phase. Our result can thus be applied also to these theories and maybe explain some of the various relations between walls and
vortices. Some of these recent developments on solitons are summarized in the reviews [26,27].
The paper is divided into two main parts. In Section 2 we derive the system of differential
equations that governs the vortex junctions. In Section 3 we explicitly study the physical solutions
of the differential equations. We conclude in Section 4 with comments and possible applications.
2. The differential equations
To obtain the differential equations for the profile r = f (z) we proceed in three steps. First
in Section 2.1 we redo the wall vortex analysis interpreting the minimization of T (R) (1.4) as
a balance of forces acting on the wall. Then in Section 2.2 we find the mechanical forces (the
ones coming from the tension TW and the energy density 0 ) for a generic profile with cylindrical
symmetry. Finally in Section 2.3 we obtain the master equation governing the profile f (z).
2.1. Wall vortex redone
The wall vortex radius is obtained by the minimization of the function T (R). The derivative
of (1.4) divided by the perimeter 2R gives
TW
2n2
+
+ 0 = 0.
(2.1)
R
e2 R 4
This equation can be interpreted as a balance of forces per unit of area acting on the surface of
the wall. Preceding from the right to left of (2.1) we have:
296
Fig. 2. The magnetic field B and the magnetic current J at the boundary of the wall vortex.
2
B2
2
There is a simple way to understand the origin of the magnetic field force B2 . We can write
2n
the magnetic field as a function of the radius B(r) where B = eR
2 inside the wall vortex, B = 0
B = J
outside and the transition is encoded in the function B(r). Using the Maxwell equation
= J B thus the
.
The
force
per
unit
volume
acting
on
density
current
is
F
we find J (r) = dB(r)
dr
force per unit of area acting on the surface of the wall is
dB(r)
B 2
B(r)2
B(r)
(2.2)
,
dr = d
=
dr
2
2 inside
and is directed outwards. Note that this force is independent on the precise behavior of the function B(r).
2.2. Mechanical forces
Now we want to find the mechanical forces in the case of a generic profile r = f (z). We
proceed in this way. First we write the mechanical energy as an integral with respect to z of the
surface element times the tension TW plus the volume element times the energy density 0 :
Emech = dz 2TW f 1 + f 2 + 0 f 2 .
(2.3)
In absence of the magnetic field, the profile f (z) is the solution to the EulerLagrange equations
obtained by varying Emech :
2TW
1 + f 2 f f
+ 20 f = 0.
(1 + f 2 )3/2
(2.4)
Dividing by the perimeter 2f one obtains the forces per unit area
TW
1
f
T
+ 0 = 0.
W
f (1 + f 2 )1/2
(1 + f 2 )3/2
(2.5)
297
0 is a force per unit area due to the internal energy density and it does not depend on the profile. The other two terms can be identified with the tension divided by the two radia of curvature
of the surface, see the geometry in Fig. 3. To verify it we now compute the two radia of curvature geometrically. Take a point P = (r, , z) = (f (z), 0, z), or in the Cartesian coordinates,
(x, y, z) = (f (z), 0, z) on the surface. The tangent vector in that point is
1
f
, 0,
.
i=
(2.6)
1 + f 2
1 + f 2
The plane orthogonal to i and passing through P is
f (X f ) + (Z z) = 0.
(2.7)
The unit orthonormal vectors lying in the plane (2.7) and orthogonal to i are (j, k), where
f
1
, 0,
,
k = i j = (0, 1, 0).
j=
(2.8)
1 + f 2
1 + f 2
One radius of curvature is in the plane (i, j) and is the easiest one to compute
f
1
=
.
R1
(1 + f 2 )2/3
(2.9)
It is directed outwards in case of positive f or inwards in case of negative f . The other radius
of curvature is in the plane (j, k) defined by Eq. (2.7). The curvature inwards that lies in the plane
(j, k) is
1
1
=
.
R2 f 1 + f 2
(2.10)
In summary, the force inwards and outwards at the point P = (f (z), z) is respectively
F1 =
TW ;
f 1 + f 2
F2 =
f
TW ,
(1 + f 2 )2/3
(2.11)
298
i,
B =
3. The junctions
In this section we will discuss the physical solutions to the master equation (the system (2.12)
and (2.13)). We first make a rescaling in order to simplify the master equation. We rescale the
lengths of a quantity l RV l such that the radius of the vortex now is 1. Note that this rescaling
1
f . We rescale also the magnetic
applies to all the lengths, so f RV f , f f and f RV
field such that B = 1 in the case of the wall vortex. Thus the rescaled master equations are:
i,
= 0,
B = ,
2 f f
1
+
f
+ (1 ) = 0.
B 2 wall +
f (1 + f 2 )3/2
(3.1)
(3.2)
.
=
1 RV 0
(3.3)
The parameter can take on values from 0 to 1. If = 0 the collapse force is due only to the
internal energy density (MIT bag), while if = 1 it is due only to the tension of the wall (SLAC
bag).
3.1. Near vortex approximation
To determine the solution we start with an approximation that is valid when the junction is
near the vortex (soon it will be clear what we mean). The difficult part of the master equations is
the partial differential equation that determines the magnetic field. To overcome this problem we
use an approximation that is valid only when the profile is near to be a cone. To determine the B
field we proceed with the following steps. We take a point P = (z, r = f (z)) on the surface and
299
then we consider the cone that passes through it and is tangent to the surface (see Fig. 4). Then
we find the magnetic field that is generated by a charge c at the tip of the cone
2
1
f
c
,
B = 2 = c
(3.4)
f
d
1 + ( rff )2
where d is the distance from a generic point (z, r) on the red surface to the tip of the cone. In
order to have a constant magnetic flux as we vary z, the charge c must be a function of z. The
magnetic flux is
f
B =
Bz = 2
dr rc
f
f
2
1
(1 + ( rff )2 )3/2
.
1 + f 2
(3.5)
(3.6)
This expression gives us the function c(z). The magnetic field in the point P is what we need for
the differential equation (3.2). To obtain it we just take (3.4) evaluated at r = f (z). The function
c(z) is given by (3.6) where we have also to remember that in our rescaled units where the radius
of the vortex is 1 and the magnetic field inside is 1, the magnetic flux is . Finally the differential
equation (3.2) becomes:
1 + f 2 f f
f 4
+
+ (1 ) = 0.
f (1 + f 2 )3/2
4f 4 (1 + f 2 )( 1 + f 2 1)2
(3.7)
300
point must be a source of magnetic flux, we can call it a monopole. In Fig. 6 we have more
general junctions: (C) is a wallvortexwall, (D) is a monopolevortexwall, (F) monopole
vortexmonopole.
The overall picture becomes more clear if we consider the differential equation (3.7) as a
dynamical system. This means simply that we rewrite the second order differential equation in
the following way
f = g,
(3.8)
g = F(f, g),
301
Fig. 7. The phase diagram (f, f ) around the stationary point (1, 0).
where F(f, g) is just what remains if we isolate f in the differential equation (3.7). The differential equations (3.8) are a simple example of a dynamical system. We have a phase space (f, g)
and a flow defined on it. The time of the flow is in this case the z coordinate. The vortex is the
fixed point of the flow f = 1 and g = 0. To understand the physical behavior around the fixed
point we make a Taylor expansion
0 1
f 1
f
=
+ .
(3.9)
g
3 0
g
The diagonalization leads to
g 3(f 1) = 3 g 3(f 1) + .
(3.10)
What we read from this is the following. The vortex is a stationary point of the dynamical system.
The linear expansion shows that this is a saddle point. The solutions of Fig. 5 (with the corresponding z z reflected ones) are the orbits that at z
(z ) go into the vortex. The
orbits around the vortex are the two lines (3.10) g = 3(f 1). If we draw the solutions in a
phase plot we obtain Fig. 7.
3.2. Flux tube, domain wall and monopole ( = 1)
The first junction we want to study is the flux tube that ends on a domain wall. In this case we
have to take = 1 so the Higgs phase and the Coulomb phase are both true vacua of the theory.
First we want determine the coefficient of the logarithmic deformation of the wall when a
vortex is attached to it. What we are going to describe is summarized in Fig. 8. The piece of
junction in the figure has two forces that act on it. The first is the vortex that pulls it down with a
tension TV . The second is the wall that pulls it up. These two forces must be equal:
TV = 2rTW
1
1 + tan 2
(3.11)
302
where tan = dr/dz. At the end we obtain a differential equation for the profile r(z):
r (z) (2TW /TV )2 r(z)2 + 1 = 0.
2
(3.12)
This equation can be trusted only at large r. In fact, in making the balance of the forces we have
not considered the magnetic field inside the wall. For large r its contribution is negligible since
B fall offs as 1/r 2 which implies that the energy density B 2 r 2 falls of as 1/r 2 . Anyway (3.12)
gives the correct coefficient of the logarithmic bending at large r:
z
TV
log r + cos t.
2TW
(3.13)
TV
Since 2T
= 32 RV , and we are choosing units in which RV = 1, this means that the vortexwall
W
junction asymptotically goes like
r e(2/3)z .
(3.14)
Now it is time to finally compute the vortexwall junction. The strategy we use is the following. Eqs. (3.10) and (3.14) give us the asymptotic behavior of the profile function at z
(near the vortex) and z + (near the wall). We construct a trial function with the desired
asymptotic behaviors and with a certain number of free parameters. These parameters will be
adjusted by our numerical code in order to approximate in the best possible way the real
vortexwall solution. The more the free parameters are the bigger is the space of profile functions
that our trial function can span. For any given choice of the parameters we have a certain profile
function and we can thus solve the Laplace equation (2.12) and the mechanical forces. To solve
the Laplace equation we use a finite element method routine. Once we have the mechanical force
and the magnetic force for a given profile we can compute the total force that is just the sum of
the two. For the real vortexwall junction the two forces must be exactly equal but opposite
in direction in every point on the profile (2.13). For our trial function we thus compute the norm
of the total force (the integral of the total force squared) and then minimize it. The minimization
gives, within the space spanned by our trial functions, the best approximation to the real junction.
For the vortexwall junction the result of the computation is given in Fig. 9. We show both the
junction and the force diagram. We have used the same strategy to compute the vortexmonopole
junction. The result is given in Fig. 10. We can finally plot the global phase diagram in Fig. 11.
Fig. 11. Phase diagram for = 1. The dashed lines correspond the near vortex approximation given in Fig. 7.
303
304
And so the orbits around the vortex are the two lines g = 4
(f 1).
The first junction we want to study is the vortexwall. In the case < 1 the energy density of
the Coulomb vacuum is not zero and this means that a stable domain wall does not exist. On the
other hand we have a beautiful solution for the vortexwall junction in the case < 1 in Fig. 12.
What happens to this solution when is decreased? The fact is that a vortexwall junction can
exist if the twodimensional space orthogonal to the vortex is compactified. If we compactify the
twodimensional space on a circle then its radius RMax is given by
TV =
B2
2
2RV
2
+ TW 2RV + 0 RV
=
B2
2
2RMax
2
+ 0 RMax
.
(3.17)
The previous equation is simply the balance of forces. The tension of the vortex, that is equal
to the sum of the magnetic field, wall and bulk energies, must be equal to the tension in the
Coulomb vacuum that is given only by the magnetic field and the energy density. With our units
(B = , TW = 2 , 0 = 1
2 ) the radius RMax is determined by the following equation
2
1+
(3.18)
+
.
=
RMax
2
2
2
2RMax
For the vortexwall junction the result of the computation is given in Fig. 12.
The last junction is the vortexmonopole junction for the case < 1. This junction is the most
difficult to compute and is also the one where we have reached least precision. The difficult part
305
comes from the shape of the wall vortex surface around the point where it intersects the axial
line r = 0. Due to symmetry reasons the magnetic field in this point must be zero so the total
mechanical force must be zero. In the case = 1 the Coulomb energy density 0 is zero and
so also the curvature of the wall vortex surface must be zero (see Fig. 10). When the Coulomb
energy density 0 is not zero and it exerts a force that tends to retreat the surface. The two radia
of curvature are equal, directed outwards and with modulus given by
TW
= 0 ,
(3.19)
RCurv
that in our units means
2
.
RCurv =
(3.20)
1
For the vortexmonopole junction the result is given in Fig. 13. We also plot the global phase for
= 0.2 in Fig. 14.
2
306
307
1 T (k, N)
.
N T (1, N)
(4.2)
R(x, N ) is a quantity with a smooth saturation limit. Now comes an assumption. Inside the
su(N ) Lie algebra there is a particular generator, that up to gauge invariance is unique, that
exponentiated passes trough all the elements of the center of the gauge group. We assume that
the kstrings are a sort of dual vortices of this U (1). This is of course a nonproved assumption
but the nice thing, as we are going to see, is that it has a very clear signal that can be tested with
lattice computations [29]. The fact is that the string tension must now satisfy two constraints. The
first is that in the free string limit (k fixed while N goes to infinity) the tension must be linear
in k plus subleading corrections. On the other hand, based on the assumption we just made, the
tension for the dual U (1) vortex must also be linear when k is large. The only reasonable way
to combine these two limits is that R(x, N ) in the saturation limit is the triangular function plus
subleading corrections.
R(x, N ) = min(x, 1 x) + O(1/N).
(4.3)
If this turns out to be correct, the vortexmonopole junction studied in the present paper will be a
good description of the kstring quark junction and will maybe be visible in lattice computations
with a large number of colors.
Acknowledgements
We thank K. Konishi for useful comments and for the precious collaboration to the initial
stages of the work. We thank also D. Tong for a useful discussion. The work of S.B. supported
by the Marie Curie Excellence Grant under contract MEXTCT2004013510.
References
[1]
[2]
[3]
[4]
S. Bolognesi, Domain walls and flux tubes, Nucl. Phys. B 730 (2005) 127, hepth/0507273.
S. Bolognesi, Large N , ZN strings and bag models, Nucl. Phys. B 730 (2005) 150, hepth/0507286.
S. Bolognesi, S.B. Gudnason, Nucl. Phys. B 741 (2006) 1, hepth/0512132.
A.A. Abrikosov, On the magnetic properties of superconductors of the second group, Sov. Phys. JETP 5 (1957)
1174, Zh. Eksp. Teor. Fiz. 32 (1957) 1442;
H.B. Nielsen, P. Olesen, Vortexline models for dual strings, Nucl. Phys. B 61 (1973) 45.
[5] W.A. Bardeen, M.S. Chanowitz, S.D. Drell, M. Weinstein, T.M. Yan, Heavy quarks and strong binding: A field
theory of hadron structure, Phys. Rev. D 11 (1975) 1094.
[6] A. Chodos, R.L. Jaffe, K. Johnson, C.B. Thorn, V.F. Weisskopf, A new extended model of hadrons, Phys. Rev. D 9
(1974) 3471;
K. Johnson, C.B. Thorn, Stringlike solutions of the bag model, Phys. Rev. D 13 (1976) 1934.
308
[7] M. Shifman, A. Yung, Domain walls and flux tubes in N = 2 SQCD: Dbrane prototypes, Phys. Rev. D 67 (2003)
125007, hepth/0212293.
[8] N. Sakai, D. Tong, Monopoles, vortices, domain walls and Dbranes: The rules of interaction, JHEP 0503 (2005)
019, hepth/0501207.
[9] R. Auzzi, M. Shifman, A. Yung, Studying boojums in N = 2 theory with walls and vortices, Phys. Rev. D 72 (2005)
025002, hepth/0504148.
[10] A. Hanany, D. Tong, Vortices, instantons and branes, JHEP 0307 (2003) 037, hepth/0306150.
[11] R. Auzzi, S. Bolognesi, J. Evslin, K. Konishi, A. Yung, NonAbelian superconductors: Vortices and confinement in
N = 2 SQCD, Nucl. Phys. B 673 (2003) 187, hepth/0307287;
R. Auzzi, S. Bolognesi, J. Evslin, K. Konishi, NonAbelian monopoles and the vortices that confine them, Nucl.
Phys. B 686 (2004) 119, hepth/0312233.
[12] R. Auzzi, S. Bolognesi, J. Evslin, Monopoles can be confined by 0, 1 or 2 vortices, JHEP 0502 (2005) 046, hepth/0411074.
[13] D. Tong, Monopoles in the Higgs phase, Phys. Rev. D 69 (2004) 065003, hepth/0307302.
[14] M. Shifman, A. Yung, NonAbelian string junctions as confined monopoles, Phys. Rev. D 70 (2004) 045004, hepth/0403149.
[15] J.P. Gauntlett, R. Portugues, D. Tong, P.K. Townsend, Dbrane solitons in supersymmetric sigmamodels, Phys.
Rev. D 63 (2001) 085002, hepth/0008221.
[16] J.P. Gauntlett, D. Tong, P.K. Townsend, Multidomain walls in massive supersymmetric sigmamodels, Phys. Rev.
D 64 (2001) 025010, hepth/0012178.
[17] Y. Isozumi, M. Nitta, K. Ohashi, N. Sakai, Construction of nonAbelian walls and their complete moduli space,
Phys. Rev. Lett. 93 (2004) 161601, hepth/0404198.
[18] Y. Isozumi, M. Nitta, K. Ohashi, N. Sakai, All exact solutions of a 1/4 BogomolnyiPrasadSommerfield equation,
Phys. Rev. D 71 (2005) 065018, hepth/0405129.
[19] M. Eto, Y. Isozumi, M. Nitta, K. Ohashi, K. Ohta, N. Sakai, Y. Tachikawa, Global structure of moduli space for
BPS walls, Phys. Rev. D 71 (2005) 105009, hepth/0503033.
[20] D. Tong, Dbranes in field theory, JHEP 0602 (2006) 030, hepth/0512192.
[21] M. Shifman, A. Yung, Bulkbrane duality in field theory, hepth/0603236.
[22] M. Eto, T. Fujimori, Y. Isozumi, M. Nitta, K. Ohashi, K. Ohta, N. Sakai, NonAbelian vortices on cylinder: Duality
between vortices and walls, Phys. Rev. D 73 (2006) 085008, hepth/0601181.
[23] E. Witten, Branes and the dynamics of QCD, Nucl. Phys. B 507 (1997) 658, hepth/9706109.
[24] A. Hanany, M.J. Strassler, A. Zaffaroni, Confinement and strings in MQCD, Nucl. Phys. B 513 (1998) 87, hepth/9707244.
[25] S. Bolognesi, J. Evslin, Stable vs unstable vortices in SQCD, JHEP 0603 (2006) 023, hepth/0506174.
[26] D. Tong, TASI lectures on solitons, hepth/0509216.
[27] M. Eto, Y. Isozumi, M. Nitta, K. Ohashi, N. Sakai, Solitons in the Higgs phase: The moduli matrix approach,
hepth/0602170.
[28] J. Greensite, The confinement problem in lattice gauge theory, Prog. Part. Nucl. Phys. 51 (2003) 1, heplat/0301023.
[29] B. Lucini, M. Teper, U. Wenger, Glueballs and kstrings in SU(N ) gauge theories: Calculations with improved
operators, JHEP 0406 (2004) 012, heplat/0404008.
[30] A. Armoni, M. Shifman, Remarks on stable and quasistable kstrings at large N , Nucl. Phys. B 671 (2003) 67,
hepth/0307020.
[31] D.K. Arrowsmith, C.M. Place, An Introduction to Dynamical Systems, Cambridge Univ. Press, Cambridge, 1990.
Abstract
The spectral properties of operators formed from generators of the qOnsager nonAbelian infinitedimensional algebra are investigated. Using a suitable functional representation, all eigenfunctions are
shown to obey a secondorder qdifference equation (or its degenerate discrete version). In the algebraic
sector associated with polynomial eigenfunctions (or their discrete analogues), Bethe equations naturally
appear. Beyond this sector, where the Bethe ansatz approach is not applicable in related massive quantum
integrable models, the eigenfunctions are also described. The spinhalf XXZ open spin chain with general
integrable boundary conditions is reconsidered in light of this approach: all the eigenstates are constructed.
In the algebraic sector which corresponds to special relations among the parameters, known results are
recovered.
2006 Elsevier B.V. All rights reserved.
PACS: 02.20.Uw; 03.65.Fd; 04.20.Jb; 11.30.j
Keywords: Tridiagonal algebra; qOnsager algebra; qSturmLiouville; Bethe ansatz; Open spin chain
1. Introduction
In the context of quantum integrable systems (continuum or lattice), finding the nonAbelian
infinitedimensional algebra responsible of the existence of an (in)finite number of mutually
commuting conserved quantities is a challenge. Indeed, the success of conformal field theory essentially relies on the Virasoro algebra and its representation theory which allows to derive exact
results such as the energy spectrum and exact correlation functions [1]. For lattice models such as
Email address: baseilha@phys.univtours.fr (P. Baseilhac).
05503213/$ see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2006.08.008
310
the Ising [2], XY [3], superintegrable Zn chiral Potts model [4] and some generalizations [5], it is
the Onsager nonAbelian Lie algebra [2] which plays an analogous role and allows to derive the
exact spectrum for any conserved quantity of the model using solely its representation theory [6].
In these latter models, the transfer matrix can be decomposed on a basis of mutually commuting
conserved quantities I2k+1 , k = 0, 1, . . . which form an Abelian subalgebra of the Onsager algebra. In particular, each quantity I2k+1 is expressed in terms of certain nonlinear combinations of
the two fundamental operators A0 , A1 which satisfy the DolanGrady relations [3]
A0 , A0 , [A0 , A1 ] = 0 [A0 , A1 ] and A1 , A1 , [A1 , A0 ] = 0 [A1 , A0 ],
(1)
and generate the Onsager algebra with generators Ak , Gk , k integer [6,7]. Note that the parameter
0 as well as the realizations of A0 , A1 are modeldependent: for the Ising model 0 = 16 and for
the superintegrable Zn chiral Potts model 0 = n2 . In both models, the generators are realized in
terms of the sl2 loop algebra and the solution of the spectral problem for the transfer matrix or
any conserved quantity has been solved using solely the representation theory of (1) [6].
In the last few years, there has been some renewed interest in the Onsager algebra, Dolan
Grady integrable structure [813] and its deformation [1416] in the context of solvable lattice
models as well as in the representation theory of orthogonal polynomials [9,1720]. Related with
the subject of this paper, a qdeformed analogue of the Onsager algebra has been exhibited in
various quantum integrable systems (AzbelHofstadter model, XXZ open spin chain with the
most general boundary conditions [16], . . . ) which are usually studied using the quantum inverse
scattering method. As shown in [14,15] and similarly to the undeformed case [2,3], in these
models the integrability property is actually related with the existence of two operators W0 , W1
which satisfy the tridiagonal relations (also called the qdeformed DolanGrady relations) with
scalars ,
W0 , W0 , [W0 , W1 ]q q 1 = [W0 , W1 ],
W1 , W1 , [W1 , W0 ]q q 1 = [W1 , W0 ].
(2)
More generally, they generate a qdeformed analogue of the Onsager algebra with generators
{Wk , Wk+1 , Gk+1 , Gk+1 } introduced and studied in details in [16]. In addition, it has been discovered that the transfer matrix in these modelscalculated in the inverse scattering framework
[21]can be simply written in the form
t (u) =
N1
(3)
k=0
where F2k+1 (u) are certain rational functions of the spectral parameter u and {I2k+1 }, k =
0, 1, . . . , N 1, is a family of mutually commuting operators acting on the quantum space of
the system. Remarkably, these operators form an ordered subset,1 of the qdeformed analogue
of the DolanGrady infinite hierarchy [14,15]. Written in terms of the generators of the infinitedimensional qdeformed analogue of the Onsager algebra, this hierarchy takes the form:
+
I2k+1 = Wk + Wk+1 +
(4)
Gk+1 +
Gk+1 , k = 0, 1, . . .
k+
k
for arbitrary parameters , , , k . Although technically rather lengthy, by analogy with the
undeformed case [3] it is alternatively possible to write the operators {Wk , Wk+1 , Gk+1 , Gk+1 }
1 In the case of finitedimensional representations, the infinite hierarchy truncates due to the existence of linear relations
among the generators. For the XXZ open spin chain, see [16].
311
for k 0 in terms of nonlinear combinations of the fundamental ones {W0 , W1 }. For instance,
one has [15] (see also [16])
I1 = W0 + W1 +
[W1 , W0 ]q +
[W0 , W1 ]q ,
k+
k
(5)
where the qcommutator [X, Y ]q = q 1/2 XY q 1/2 Y X has been introduced. Note that the parameters N , , , , , , k , q and the functions F2k+1 (u) obviously depend on the physical
characteristic of the model under consideration.2
Having in mind the use of the Onsager algebra in the Ising [2], superintegrable Zn chiral
Potts model [4] or the Virasoro algebra in conformal field theory [1], it seems to us important to
analyse the models which enjoy the hidden dynamical symmetry (2) from the point of view of the
representation theory associated with the tridiagonal algebra (2). Indeed, for several modelsfor
instance the XXZ open spin chain with arbitrary integrable boundary conditionsthe standard
Bethe ansatz approach does not apply except at some special points in the boundary parameters
space. As we are going to see, the analysis based on the representation theory of (2) gives a
clear understanding of the structure of the space of states in models with transfer matrices of the
form (3).
Among the interesting problems in integrable lattice models with hidden symmetry (2), one
fundamental is to identify the eigenstates of the transfer matrix, and corresponding eigenvalues.
According to the decomposition (3) and the fact that the spectral parameter u is arbitrary, the
structure of these eigenstates is encoded in the spectral problem for the qdeformed DolanGrady
hierarchy {I2k+1 }, k = 0, 1, 2, . . . . By virtue of [16]
[I2k+1 , I2l+1 ] = 0 and I2k+1 = 2k+1
(6)
for all k, l = 0, 1, . . . , it follows that the structure of the eigenstates is completely determined
from the spectral problem for any operator I2k+1 , provided its eigenvalues are nondegenerate.
For instance, one can focus on the simplest operator, namely I1 defined by (5). If its spectrum
is nondegenerate (see details in further sections), finding the eigenstates of (4) is equivalent to
solve I1 = 1 .
In this paper, we establish a direct relationship between the spectral problem for the hierarchy (4) and a secondorder qdifference equation in one variableor its discrete version
satisfied by the states in a suitable functional representation. The necessary and sufficient
information to solve this spectral problem is shown to be encoded in the fundamental operators W0 , W1 , which properties uniquely determine the eigenfunctions (z). Some examples are
considered in details.
The paper is organized as follows. In Section 2, an infinitedimensional module of the tridiagonal algebra (2) is constructed. It is shown that the spectrum of the operator W0 can be algebraized
i.e. corresponds to (possibly degenerate) polynomial eigenfunctions which roots satisfy a set of
Bethe equations. In this eigenbasis of W0 , the action of W1 is described. Then, the spectral
2 The simplest example with a transfer matrix of the form (3) corresponds to N = 1: it is the AzbelHofstadter model
which describes the problem of Bloch electrons in a magnetic field on a twodimensional lattice. There, the elements
W0 , W1 are realized in terms of the Weyl algebra [14]. One has F1 (u) = u2 , , , , k are related with length scales
along the directions of the lattice and q = 1, related with the magnetic flux per plaquette, is a root of unity. An other
model is the XXZ open spin chain with nondiagonal boundary conditions. In this case W0 , W1 are expressed in terms of
Uq 1/2 (sl2 ) operators acting on the Hilbert space of the whole spin chain with N sites [14,15]. The parameters , , ,
k are related with integrable boundary conditions (see Section 3).
312
problem for (5) is considered: in the algebraic sector associated with special relations among
the parameters , , , k = 0, the operator I1 admits polynomial eigenfunctions. Beyond the
algebraic sector i.e. for arbitrary parameters , , , k = 0, the (nonpolynomial) eigenfunctions (z) of (5) are expanded on the (infinitedimensional) basis of polynomials eigenfunctions
of W0 . Corresponding weight functions are determined by a coupled system of recurrence relations, which coefficients follow from the fundamental operator W1 . In Section 3, the same
approach is applied to the case of finitedimensional modules of the tridiagonal algebra. In particular, we focus on the XXZ open spin chain with integrable boundary conditions. The eigenstates
(zs ) of the transfer matrix, represented by functions defined on a discrete support, are shown
to satisfy a set of discrete qdifference equations. For general values of the boundary parameters, the eigenstates are constructed. For special linear relations among the boundary parameters,
a subset of eigenstates (called Bethe eigenstates) are given by polynomial eigenfunctions defined
on a discrete support, which roots are solutions of Bethe equations in agreement with known results. An alternative construction of these eigenstates is also proposed in Appendix A. Comments
follow in the last section.
2. A secondorder qdifference equation from the infinite hierarchy
In this section, we construct an infinitedimensional module of the tridiagonal algebra (2) on
which the infinite qdeformed DolanGrady hierarchy (4) acts. Although the infinitedimensional
modules are more complicated to study than the finitedimensional ones, in our example the main
properties of the finitedimensional case (eigenvalue sequences, block diagonal structure in the
dual eigenbasis and duality of the fundamental operators) hold. These properties will be used to
derive a qdifference equation for the eigenstates of (4).
2.1. An infinitedimensional module of the tridiagonal algebra
Let V denotes an irreducible infinitedimensional module on which W0 , W1 are both diagonalizable. Suppose q is not a root of unity. Let z denote a complex variable. Define to be the
shift operator such that 1 (z) = q 1 z. Our aim is to endow the vector space of all functions
in the variable x z + z1 with the module structure of a certain tridiagonal algebra (2). By
analogy with the finitedimensional case (see [17,18,20]), for the linear transformations W0 , W1
it is natural to consider the following realization3 of the fundamental operators:
W0 : (z) + (z)1 + d z, z1 (z) (z) I,
W1 :
z + z1 .
(7)
Here, (z), (z), d(z, z1 ) are functions of z which are not necessarily unique, as the spectrum
of W1 may contain degeneracies. As we will see in Section 3, this kind of situation happens in
the case of finitedimensional modules too. The existence of such degeneracies induces dim(Vs )
possible realizations of the operator W0 .
Our purpose is now to find the necessary and sufficient conditions on (z), (z), d(z, z1 )
such that the tridiagonal relations (2) are satisfied. First, for any (z), (z), d(z, z1 ) it is easy
to check that (7) automatically solves the second equation in (2) provided one identifies
2
= q q 1 .
(8)
3 If (z) = (z1 ), we assume there exists at least one other realization of the operator W which spectrum is identical.
0
313
Then, replacing (7) in the l.h.s. of the first equation of (2) a straightforward calculation gives
W0 , W0 , [W0 , W1 ]q q 1
= (, , d) (z) (z) 2 1 (, , d) (z)1 (z) 2
(, , d)(z) (x) x + 1 (, , d) (z) 1 (x) x 1
(9)
with
2
q q z + q 2 q 1 z1
(, , d) = (z) + (z) d z, z1
q q 1 z1 z
+ (z) + (z) d z, z1
1
q q 2 z
+ 1 (z) + 1 (z) 1 d z, z1
+ q q 2 z1 ,
(10)
2
2
+ (z) + (z) d z, z1
(, , d) = (z) + (z) d z, z1
(z) + (z) d z, z1
q + q 1 (z) + (z) d z, z1
+ 2 + q + q 1 (z) (z) + (z)1 (z) + (z) 2 (z)
(1 + q + q 1 )
(z)1 (z) 1 (x) x
(x) x
+ (z) 2 (z) (x) 2 (x) .
+
(11)
(12)
Identifying (9) and (12), we immediately deduce that the operators W0 , W1 with (7) satisfy the
tridiagonal algebraic relations (2) iff the functions (z), (z), d(z, z1 ) solve the constraints:
(, , d) = 0 and (, , d) = .
(13)
Eq. (10) being a secondorder qdifference equation with simple Laurent polynomials as coefficients, we restrict our attention to the solutions (z), (z) which are rational functions of z
and d(z, z1 ) is supposed to be a Laurent polynomial in z. In this case, define the family
2N +2
1
k=1 (1 k z)
(z) =
(14)
and (z) = z1 ,
2N 2
2
2
(1 z )(1 qz ) k=1 (1 k z)
where we have introduced 4N scalars {k }, {k } in an arbitrary field K. Replacing (14) in (10),
(11), Eqs. (13) impose some constraints on the parameters:
For N = 1, it is straightforward to check that the constraints (13) are satisfied with the identification (8) and
2
d z, z1 = 1 + 1 2 3 4 q 1 ,
= q q 1 1 2 3 4 q 1 ,
(15)
for any choice of the parameters {1 , 2 , 3 , 4 }. Notice that the realization (7) for the operator
W0 in this case coincides with the secondorder AskeyWilson qdifference operator [22].
(16)
314
whereas the second constraint yields to the following relations among the parameters:
2+k
3
j
k
3k
2+kj
(1) i1 i2 i3+k (1 2 )
1 2
= 0,
k=2
2
k=3
j =0
i1 <<i3+k
(1) i1 i2 i3+k (1 2 )
3+k
2k
j
1 2 2kj
= 0,
(17)
j =0
i1 <<i3+k
with {ik } {1, . . . , 6}. So, it remains six independent parameters, say 1 , 2 , 3 , 4 , 5 , 6 .
Simplifying (13), one finds
2
= q q 1 1 2 3 4 5 6 11 21 q 1 .
(18)
Although we do not proceed further,4 for higher values of N it is possible to classify all
possible parameter sequences such that (7) with (14) and d(z, z1 ) const satisfy (2). More
generally, it is possible to find examples for , which are neither such that (z) = (z1 ), nor
rational functions of z.
2.2. A polynomial eigenbasis
For the rational functions (z), (z) satisfying (13), a basis of possibly degenerate eigenfunctions n[m] (z), n = 0, 1, 2, . . . and m = 1, . . . , dim(Vn ) of W0 with eigenvalues {n } can be
constructed. The realization (7) for W0 yields to the qSturmLiouville problem
(z)n[m] (qz) + (z)n[m] q 1 z + z, z1 n[m] (z) = n n[m] (z)
(19)
with
z, z1 = d z, z1 (z) (z).
In general, only a part of the spectrum of qdifference operators can be algebraized, i.e. corresponds to polynomial eigenfunctions in z. However, according to the structure of the spectrum
of W0 , W1 for the finitedimensional case [17,18], we expect the operator W0 in (7) can be entirely algebraized provided (13) are satisfied. For this reason, we are looking for polynomial
eigenfunctions symmetric in the variable z i.e. invariant under the replacement z z1 :
n[m] (z) =
(z zj ) z1 zj
(m)
for zj zj ,
(20)
j =1
with i = 1, 2, . . . , n.
4 The case N = 3 has been also considered, although not reported here.
(21)
315
As an example, let us consider the family of rational functions (14) with suitable parameter
sequences such that (13) are satisfied. Introducing the new parametrization
z e2 ,
l e2l ,
l e2cl ,
q e ,
(22)
l=1
2N +2
sinh(i + cl )
sinh(i l )
sinh(i cl )
sinh(i + l )
l=1
j =1,j =i
(23)
with i = 1, 2, . . . , n. For N = 1 and no special relations among the parameters {k }, one has
dim(Vn ) = 1. The polynomial eigenfunctions (20) coincide exactly with the AskeyWilson polynomials which zeros are known to satisfy (23) at N = 1. For N = 2 the parameters c1 , c2 are
determined by the relations (17), and the solutions (20) can be seen as some generalization of
the AskeyWilson polynomials which, to the best of our knowledge, are new. In both exactly
solvable cases, the spectrum can be easily derived. Replacing (20) in (19), expanding both sides
of the equation and identifying the leading terms one finds
n = 1 2 3 4 q 1 q n + q n
for N = 1,
n = 1 2 3 4 5 6 11 21 q 1 q n + q n
for N = 2.
(24)
with n = Cq n + C q n ,
(25)
dim(Vn+1 )
W1 n[m] (z) =
dim(V
n )
l=1
l=1
dim(Vn1 )
(26)
l=1
For the family of rational functions (14) and d(z, z1 ) const, the coefficients an[lm] , bn[lm] ,
cn[lm] can be determined explicitly in terms of the parameters k , k . As the simplest example,
consider the fundamental operators W0 , W1 with (7) and (14) for N = 1. To avoid degenerate situations, we assume the parameters {k } are generic. As mentioned above, in this case dim(Vn ) = 1
5 This property can be also derived starting from (2). See for instance [18].
316
i.e. m, l = 1. The coefficients in (26) are well known [22], they take the form
(1 1 2 q n )(1 1 3 q n )(1 1 4 q n )(1 1 2 3 4 q n1 )
,
1 (1 1 2 3 4 q 2n1 )(1 1 2 3 4 q 2n )
(1 q n )(1 2 3 q n1 )(1 2 4 q n1 )(1 3 4 q n1 )
,
cn[11] = 1
(1 1 2 3 4 q 2n2 )(1 1 2 3 4 q 2n1 )
bn[11] =
(27)
To resume, polynomial eigenfunctions of the form (20) with Bethe equations (21) provide an
example of irreducible6 infinitedimensional module of the tridiagonal algebra. In this basis, W0 ,
W1 act as (25) and (26), respectively, where explicit expressions of the coefficients depend on
the solutions (z), (z), d(z, z1 ) of (13). For the family of rational functions (14), the Bethe
equations take the form (23) with the identifications (22).
2.3. Structure of the eigenfunctions
For quantum integrable models with dynamical symmetry (2), the spectral problem for the
family of mutually commuting operators {I2k+1 } k = 0, 1, . . . , reduces to the one for (5) provided the spectrum of this operator is nondegenerate. Let (z) denotes an eigenfunction of the
hierarchy (4). Using the realization (7) of the fundamental operators, the spectral problem for I1
in the functional representation reads7
A(z) (qz) + A(z) q 1 z + B z, z1 (z) = 1 (z)
(28)
with
+ 1/2 1 1/2
1
q
z +
q z (z),
A(z) = + q q
k+
k
+ 1/2
1/2 1
1
q
z+
q z
(z),
A(z) = + q q
k+
k
B z, z1 = z, z1 + z + z1
+
+
z, z1 .
+ q 1/2 q 1/2 z + z1
k+ k
Here, the parameters , , , k are arbitrary. Contrary to the case of W0 , the operator I1
admits only partial algebraization of it spectrum. Depending on the parameters, we now describe
the structure of its eigenfunctions (z) in the nonalgebraic sector and the algebraic sector,
respectively.
Nonalgebraic sector: For generic values of the parameters , , , k , there are no
polynomial solutions. Indeed, the leading contribution in the numerator of (28)independent of
1 cannot be canceled out. A natural approach is then to consider (z) as an infinite power
series expansion in the variable x = z + z1 . However, for the family of rational functions (14)
6 For special relations among the parameters entering in the solutions (z), (z), d(z, z1 ), the module may become
indecomposable. This possibility is not considered here.
7 The reader familiar with the algebraic Bethe ansatz will immediately recognize the similarity of this equation with
the Baxter identity for the eigenvalues of the transfer matrix, also called TQ relations [23].
317
the problem becomes quickly rather complicated when the value of N increases. Instead, a convenient procedure consists in expanding the solutions of (28) on the basis of eigenfunctions
n[m] (z) of W0 given by (20) with (23). For generic values of the parameter , , , k we
write
(z) =
dim(V
n )
n=0
(29)
m=1
According to (7) and the action (25) of W0 and (26) of W1 on the polynomial basis, the coefficients fn[m] are determined by the following coupled system of recurrence relations:
dim(Vn1 )
(n 1 )fn[l] +
Bn1[lm] fn1[m] +
m=1
dim(V
n )
An[lm] fn[m]
m=1
dim(Vn+1 )
Cn+1[lm] fn+1[m] = 0,
(30)
m=1
for l = 1, . . . , dim(Vn ), where the coefficients An[lm] , Bn[lm] , Cn[lm] follow from the action of (5)
on (20). Using (25) and (26), they read
+ 1/2 1/2
1/2 + 1/2
q
q
q
q
n +
n+1 bn[lm] ,
Bn[lm] = +
k+
k
k
k+
+ 1/2 1/2
1/2 + 1/2
q
q
q
q
n +
n1 cn[lm] ,
Cn[lm] = +
k+
k
k
k+
+
+
n an[lm] .
An[lm] = + q 1/2 q 1/2
(31)
k+ k
Algebraic sector: This sector corresponds to polynomial eigenfunctions of I1 which exist
for special relations among the parameters, identified as follows. Let us assume that n (z) is a
polynomial symmetric in the variable z of the form
n (z) =
(z zj ) z1 zj ,
(32)
j =1
with n = 0, 1, . . . . For the family of rational functions (14) with suitable nonvanishing parameters, replacing (32) in (28) one finds that the leading contribution in the l.h.s. of the equation
(obtained by taking the limit z ) is independent of 1 . Then, the leading terms will cancel
out (and similarly if one considers the limit z 0) only if the parameters satisfy the relations
2N2
2N+2
1 + 1/2 n
1 1/2
1
q
q + q q
q
k
k q n = 0,
+ q q
(33)
q
k+
k
k=1
k=1
i.e. Bn[lm] = 0. This implies that any eigenfunction (32) can be written as a finite sum of elementary eigenfunctions (20). Choosing for simplicity the representation (32), the roots zi are now
determined by the generalized system of Bethe equations
( + (q q 1 )( k++ q 1/2 zi + k q 1/2 zi1 )) (zi )
n (qzi )
=
n (q 1 zi )
( + (q q 1 )( k+ q 1/2 zi1 + k q 1/2 zi )) (zi )
+
for i = 1, 2, . . . , n.
(34)
318
Given n fixed by (33), the spectrum can be calculated by plugging (32) into (28) and identifying
the leading terms of both sides of the qdifference equation. For the family of rational functions
(14), the result is
(N )
n
1 + 1/2 (N )
1 =
q q
q
G q n
q + F q q
k
k+
n
1/2
(N ) n + (N ) n
1
1/2
q q
zi + zi1 ,
+ q q
F+ q
F q
(35)
k
k+
(N )
F+
(N )
q 1/2 G+
i=1
(N )
(N )
(1)
4
(1)
F = 1,
G+ =
i1 i2 i3 ,
(1)
G =
i1 <i2 <i3
F+ = 1 2 3 4 5 6 11 21 q 1 ,
(2)
(2)
G+
4
(2)
F = 1,
4
i ,
i=1
(2)
G =
6
i 1 2 ,
i=1
i1 i2 i3 i4 i5 11 21
1 2 3 4 5 6 11 21 q 1 11 + 21 .
For higher values of N , a similar form of the spectrum is expected according to (14). It is however
important to remind that the parameters {k } are restricted by some relations generalizing (17).
For nonvanishing parameters , the spectrum (35) is nondegenerate. So, (32) is also a polynomial eigenfunction of the infinite hierarchy (4).
Consequently, the operator I1 admits partial algebraization of its spectrum. In the algebraic
sector associated with (33), there is a single eigenstate associated with the polynomial eigenfunction (32) with (34). For the family of rational functions (14), the spectrum can be written in terms
of the roots of the Bethe equations, and takes the form (35). The rest of the spectrum is associated
with nonpolynomial eigenfunctions which can be written in the (infinitedimensional) eigenbasis (20) with (21) of W0 . The structure of the eigenstates (29) is encoded in the coefficients fn[m]
which satisfy (30).
3. A discrete qdifference equation for the XXZ open spin chain
For quantum integrable lattice models with dynamical qOnsager symmetry, a similar analysis can be done in order to derive a qdifference equation for the eigenstates of the transfer
matrix (3). The main difference is however the Hilbert space of these models, which is now
finitedimensional.8 To give an illustration, let us consider the XXZ open spin 21 chain with gen8 Linear relations among the higher operators of the qdeformed analogue of the Onsager algebragenerated from
W0 , W1 exist that are responsible of the truncation of the hierarchy (4) (see [16] for details).
319
N1
k=1
(0)
(0)
2
(q 1/2 q 1/2 ) (+ ) 1
1
1
+
+
k
k
+ +
3
(0)
(0)
2
(q 1/2 q 1/2 )
(+ + )
1/2
(q 1/2 q 1/2 ) ( ) N
1/2
N
N
+
+ + + ,
3 + 2 q + q
( + )
2
+
(36)
where 1,2,3 and = (1 i2 )/2 are usual Pauli matrices. Here, = (q 1/2 + q 1/2 )/2 denotes the anisotropy parameter with q = exp . Integrable boundary conditions correspond to the
choice [24]
(0)
(0)
+ = = (c00 + ic01 )/2,
(left),
k+ = (k ) = q 1/2 q 1/2 ei 2
=
(right).
= (c00 + i c01 )/2,
+ = ( ) = ei 2 q 1/2 + q 1/2
(37)
The r.h.s. parametrization has been introduced in which case one immediately identifies the
Hamiltonian as defined in [25]. In total, one has six boundary parameters c00 , c01 , c00 , c01 ,
, . Except if explicitly specified, from now on we assume these parameters are generic.
As discovered in [15,16], the known integrability of the XXZ open spin chain (36) with (37)
[21,24] is related with the existence of a hidden dynamical symmetry, a qdeformed analogue of
the Onsager algebra. This nonAbelian algebra is generated by a family of nonlocal operators
(N )
(N )
(N )
(N )
{Wk , Wk+1 , Gk+1 , Gk+1 }, which fundamental ones satisfy the tridiagonal relations (2). They
read [16]
W0(N ) = (k+ + + k ) I(N1) + q 3 /2 W0(N1) ,
(N )
W1
= (k+ + + k ) I(N1) + q 3 /2 W1
(N1)
(38)
(0)
W0 = k+ + + k + + q 3 /2 ,
W1 = k+ + + k + q 3 /2 .
(1)
(0)
Note that the left boundary parameters in (36) appear in the realization (38) of the algebra (2),
whereas the right boundary parameters enter in the definition of the linear combinations (4). For
the normalization above, the structure constant of the tridiagonal algebra (2) is fixed to
= =
(q q 1 )2
.
4
(39)
(N )
320
W1 Vn Vn+1 + Vn + Vn1
(0 n N ),
and
such that
V1 = 0, VN+1 = 0;
(N )
+ Vs + Vs1
W0 Vs Vs+1
(N )
(0 s N ),
and
such that
V1
= 0, VN +1 = 0.
Also, for 0 n N , the dimensions of the eigenspaces Vn and Vs are equal and the shape vector
[17] of the pair is symmetric and unimodal. For the family of operators (38), one has9 [20]
N
, 0 n N.
dim(Vn ) =
(40)
n
(N )
(N )
+ = cosh
and = cosh ,
(0)
(41)
N
(
n+1)
l=1
(N )
n[lm] n+1[l] (zs ) +
(Nn )
l=1
(N )
n[lm] n[l] (zs ) +
N
(
n1)
(N )
cn[lm] n1[l]
(zs ),
l=1
(44)
where the explicit expressions of the matrix entries can be found in [20]. Analogous results also
hold if, instead, we had considered the dual overlap coefficients which also form a complete
N!
9 We denote N =
n
n!(N n)! .
321
(N )
eigenbasis of W1 , a phenomena due to the duality property of (2). Obviously, in this dual basis
(N )
the operator W0 has a block tridiagonal structure.
The analysis of previous section can be applied to the finitedimensional case similarly: the
(N )
operator W0 can be associated with a discrete qdifference operator acting on the (discrete)
functional space spanned by (42). Defining 1 (f (zs )) f (zs1 ) = f (q 1 zs ), the discrete analogue of the realization (7) is now given by [20]
N
(N )
(N )
(N )
(N )
1
W0 : k (s)s + k (s)s + k (s), k 1, . . . ,
,
s
(N )
zs + zs1 /2,
W1 :
(45)
where
(N )
k (s) =
N
(
s+1)
l=1
(N )
k (s) =
(Ns )
l=1
(N )
(N ) U
(s + 1)
l
,
s[lk]
(N )
Uk (s)
(N )
k (s) =
N
(
s1)
l=1
(N )
)
c(N
s[lk]
Ul
(s 1)
(N )
Uk (s)
(N )
(s)
(N ) Ul
.
s[lk] (N )
Uk (s)
The explicit expressions of the coefficients can be found in [20]. It is important to notice that
(N )
(N )
the realization of W0 is not unique (except for s = 0 and s = N ), as the spectrum of W1
N
contains s degeneracies.
3.2. Eigenstates of the XXZ open spin chain
Due to the nonAbelian qOnsager dynamical symmetry of XXZ open spin chain (36), the
transfer matrix can be written in the form (3) with (4) [15,16]. By analogy with the case of
infinitedimensional modules, all necessary and sufficient information to determine the eigenstates of (3) is encoded in the structure of the tridiagonal pair W0(N ) , W1(N ) given by (38). We
choose to take (42) as the basis on which any eigenstate, denoted (zs ), is expanded. Considering the spectral problem for I1 and using the realization (45), we find that the eigenstates of the
XXZ open spin chain with general boundary conditions (36) are the family of solutions to the
following system of discrete qdifference equations
1
)
(N )
(N )
A(N
zs , zs1 (zs ) = 1 (zs ),
k (zs ) (qzs ) + Ak (zs ) q zs + Bk
N
k = 1, . . . ,
(46)
s
with
+ 1/2 1
1/2
)
(N )
1
A(N
(z
)
=
+
q
q
q
z
+
q
z
k (s),
s
s
s
k
2k+
2k
+ 1/2
1/2 1
(N )
(N )
q
zs +
q zs
k (s),
Ak (zs ) = + q q 1
2k+
2k
+
(zs + zs1 )
(N )
(N )
(N )
Bk zs , zs1 = k (s) + + q 1/2 q 1/2
+
k (s)
,
k+ k
2
322
(45) and s = 0, 1, . . . , N .
Difference equations are known to be closely related with discrete equations. For the example
considered here, the qdifference equation10 (28) can be seen as an extension of (46) on the
whole complex plane, where the parameters {k }, {k } in (14) are chosen such that
(N )
z = zs1 k(N ) (s),
z = zs1 k (s),
1
(N )
z, z z=z1 k (s),
(47)
z = zs1 (zs ).
s
Indeed, finitedimensional module of the tridiagonal algebra (2) can be obtained from the infinitedimensional modules by imposing z to be on a discrete support z = zs , s = 0, 1, . . . , N . It follows
that the results of previous section can be used to analyse the structure of the eigenstates of the
model (36). In particular, in view of (33), one expects that these eigenstates admit a description
in terms of solutions of Bethe equations for certain relations between the boundary parameters.
Below, we call the corresponding special cases the algebraic sector of the operator (5).
Nonalgebraic sector: For generic values of the boundary parameters , , , k , there
is no nontrivial subspace of V which is left invariant by the action of (5). So, any eigenstate is a
linear combination of all the eigenfunctions (42), n = 0, 1, . . . , N . In general, it takes the form
(zs ) =
(Nn )
N
(N )
(48)
n=0 m=1
The eigenfunctions (42) being clearly identified [20], it remains to determine the nonvanishing
(N )
coefficients fn[m] . Using the block tridiagonal structure (44) of W1 in the functional basis (42),
we find that the coefficients fn[m] (1 ) satisfy the set of coupled recurrence relations (30) with
the definitions (43), (40) and the explicit expression for the coefficients taken from [20]. Note
that each eigenfunction (20) at z = zs is individually associated with a system of Bethe equations,
whereas there is no single system of Bethe equations for (zs ).
For small values of N , it is not difficult to check the recurrence relations (30) numerically.11
One finds that the 2N values of 1 agree exactly with the direct diagonalization of I1 for generic
values of the boundary parameters (37). As expected, the corresponding eigenvectors diagonalize
the higher operators (4).
Algebraic sectors: The advantage of the formulation of the transfer matrix in terms of (4)
is that the representation theory of the tridiagonal algebra (2) guarantees that the discrete qdifference equations (46) admit a subset of solutions associated with Bethe equations. This subset
is identified with the special points in (28) with (43), (40)or equivalently (46) using (47)
(N )
(N )
where either some of the coefficients Ak (zs ), Ak (zs ) or Bn[m] , Cn[m] in (31) are vanishing, or
the special points for which the eigenfunctions (42) are no longer linearly independent.
For instance, suppose Bn[lm] 0. Plugging the parametrization (37), (41), (43) into (31), one
obtains a linear relation between the left and right boundary parameters which reads
+ = i( ) (N 2n + 1)/2 mod (2i).
(49)
10 The class of qdifference equations (28) may be extended by a gauge transformation (z) f (z) (z), (z)
(z)f (z)/f (qz), (z) (z)f (z)/f (q 1 z) and (z) (z), where f (z) is a rational function of z.
11 K. Koizumi, private communication.
323
(N )
N
(p+1
)
(N )
l=1
(Np )
(N )
l=1
N
p1
( )
(N )
(50)
l=1
one has
I1 Vn Vn + Vn1 ,
...,
I1 V0 V1 + V0 .
Then,
I1 Wn Wn
where Wn =
n
Vp
p=0
is an invariant eigenspace of (4). The eigenstates n (zs ) associated with (32) at z = zs and (34)
form an eigenbasis of Wn which dimension follows from the degeneracies of each eigenspace of
(N )
W0 , given by (40). One has
n
N
.
dim(Wn ) =
(51)
p
p=0
(52)
In this case, using the substitution n N n in (32) at z = zs , the eigenstates with (34) form a
(N )
(N )
basis of Wn . It is important to notice that the conditions Ak (zs ) 0 or Ak (zs ) 0 may have
been considered alternatively, in which case the linear relations between the boundary parameters
are obtained from (49), (52) by complex conjugation. This is not surprising, in view of the duality
between the eigenbasis of (38). To complete the analysis of the eigenstates in both algebraic
sectors (49), (52), it is worth mentioning that the solutions n (zs ) can be regarded as linear
combinations of the eigenfunctions (42), truncated in comparison with (48). The coefficients in
the expansion satisfy a set of recursion relations, reported in Appendix A.
Other invariant subspaces exist if the eigenfunctions (42) entering in (50) are not linearly
independent. These are the special points where the spectrum (43)or the one in the dual basis
admits additional degeneracies, namely
+ (N 2n)/2 = iZ,
(53)
in perfect agreement with [26]. The same result can be also derived from the explicit form of the
eigenfunctions n[m] (zs ) [20].
324
To conclude this section, the approach based on the tridiagonal algebra (2) provides a new
derivation of the linear relations (49), (52), (53). In our framework, the (Bethe) eigenstates n (zs )
correspond to the algebraic sectors of the qdeformed DolanGrady hierarchy (4), where the
Bethe anstaz equations have been formulated and the spectrum of the XXZ open spin chain has
been derived [25,27,28].
4. Comments
A new approach based on the algebraic properties and irreducible modules of the infinitedimensional qdeformed analogue of the Onsager algebra has been proposed in the context of
quantum integrable models. The main result of this paper is an exact secondorder qdifference
equation (respectively its degenerate discrete version) for the eigenfunctions associated with the
infinite (respectively truncated) qdeformed DolanGrady hierarchy (4). For generic values of the
parameters , , , k and variable z (respectively discrete zs ), the structure of the eigenfunctions has been described in details. They admit an infinite (respectively finite) serie expansion in
terms of the polynomial eigenfunctions of W0 which roots satisfy a system of Bethe equations
of the form (23), whereas the weight functions satisfy a coupled system of recurrence relations
which explicit form is determined from W1 . In the algebraic sectors which correspond to special
relations between the parameters, the expansion truncates. In this case, the eigenstates admit a
formulation in terms of solutions of the system of Bethe equations (34). The approach presented
here has been applied to the XXZ open spin chain, although higher spin representations can be
obviously studied similarly.
Equations of the type (28), (46) already have important physical applications. For instance,
the method of algebraization has been applied to the AzbelHofstader problem [29] (see also
[30]). In this model, the spectral problem for the Hamiltonian (Schrdinger equation) H I1
with anisotropy parameters , is the famous Harpers equation. For certain values of the
physical parameters, it can be algebraized using its connection with Uq (sl2 ) at root of unity.
Nonpolynomial solutions have been considered in [31].
Such a connection between qSturmLiouville problems and quantum integrable systems
is not new. However, its understanding and formulation using the qdeformed Onsager algebra/tridiagonal algebra (2) studied here is rather promising. A classical counterpart of our
description clearly needs further investigations. In this direction, let us mention that a classical
analogue of (2) has been studied in [32]. The simplest example of realization of classical Leonard
pairs W0 , W1 has been proposed:
W0 = y 2 + U (x)
and W1 = (x)
with {x, y} = 1,
(54)
where the admitted potentials U (x) are the (i) hyperbolic, (ii) modified hyperbolic, (iii) trigonometric PschlTeller potential, or the (iv) Morse, (v) singular and (vi) shifted oscillator potential.
Indeed, in the continuum limit q 1 of (7) with (14) at N = 1 the qdifference equation (28)
turns into the Schrdinger equation for the generalized PschlTeller potential. Furthermore, using a suitable change of variable and gauge transformation one arrives at the Heun equation. In
view of the recent works [33], we expect our approach will provide an algebraic understanding of
the ordinarydifferential equations/integrable models (ODE/IM) (for a review, see [34] and references therein). In the quantum case, it should be also mentioned that our results exhibit some
links with a recent work relating a qSturmLiouville problem with the Bethe ansatz equations
of the XXZ open spin chain with Dirichlet boundary conditions [12]. In particular, for vanishing
parameters {k } the rational functions (14) considered here can be easily related with the ones in
325
[35]. However, for this choice the qOnsager dynamical symmetry is brokenas expectedand
(7) do no longer satisfy (2).
To conclude, the program initiated in [14] opens the possibility of studying massive quantum
integrable models from an algebraic point of view (modelindependent) which idea takes its
roots in the original work of Onsager [2]. The exact spectrum of the complete hierarchy (4) is the
subject of a separate work, among other problems mentioned above.
Note added
The explicit expressions for the entries of the matrices (38) in their dual eigenbasis are not
reported here. We refer the reader to [20] for these data.
Acknowledgements
I thank P. Forgacs, H. Giacomini, N. Kitanine, K. Koizumi and R. Weston for discussions, and
P. Terwilliger for suggestions and interest in this work. I wish to thank the organizers of the 3rd
Annual EUCLID Meeting 2005 where preliminary results were presented. Part of this work is
supported by the ANR research project Boundary integrable models: algebraic structures and
correlation functions, contract number JC0552749 and TMR Network EUCLID Integrable
models and applications: from strings to condensed matter, contract number HPRNCT200200325.
Appendix A. An alternative construction of the Bethe eigenstates
For special relations between the left and right boundary parameters, the corresponding Bethe
eigenstates can be written directly in terms of the eigenfunctions of W0(N ) . Consider the linear
combination
n (zs ) =
(Np )
n
(N )
fp[m]
(1 )p[m]
(zs ) for n N 1,
(A.1)
p=0 m=1
where fp[m]
are nonvanishing coefficients. As an immediate consequence of (50), given n this
combination is an eigenfunction of I1 iff the condition Bn[lm] 0 i.e. (49) and the coefficients
satisfy
(p 1 )fp[l]
+
N
(p1
)
Bp1[lm] fp1[m]
+
m=1
(Np )
Ap[lm] fp[m]
m=1
N
p+1
( )
Cp+1[lm] fp+1[m]
= 0,
m=1
and
(n 1 )fn[l]
(Nn )
m=1
An[lm] fn[m]
N
(
n1)
m=1
Cn+1[lm] fn+1[m]
= 0 for l = 1, . . .
N
.
n
326
Another choice is Cn[lm] 0, which corresponds to the linear relation between the left and
right boundary parameters (52). An eigenfunction of I1 takes the form
n (zs ) =
(Np )
N
fp[m]
(1 )p[m] (zs )
(N )
for n 1,
(A.2)
p=n m=1
N
)
(p1
Bp1[lm] fp1[m]
+
m=1
(Np )
N
p
Ap[lm] fp[m]
m=1
N
p+1
( )
Cp+1[lm] fp+1[m]
= 0,
m=1
and
(n 1 )fn[l]
N
(
n1)
Bn1[lm] fn1[m]
m=1
(Nn )
An[lm] fn[m]
= 0 for l = 1, . . . ,
m=1
N
.
n
1
W0 , [W0 , W1 ]q q 1 + W1
and W2 =
1
W1 , [W1 , W0 ]q q 1 + W0 .
(B.1)
Using previous results, suppose {W0 , W1 } take the form (7) and assume , and d solve
(13). Replacing (7) in (B.1), after simplifications one finds the following action on the infinitedimensional module
W1 : 1 (z) + 1 (z)1 + 1 z, z1 I,
(B.2)
W2 :
2 z, z1
with
2 z, z1 = 1
1
(z + z1 )2
z, z ,
(q 1/2 + q 1/2 )2
1
q q 2 z + q 1 q 2 z1 qz, q 1 z1
1 (z) =
+ q 1 1 z + (1 q)z1 z, z1 (z),
327
1 1
q q 2 z + q q 2 z1 q 1 z, qz1
+ (1 q)z + 1 q 1 z1 z, z1 (z),
1 2
q 1 z + q 2 1 z1 (z)(qz)
1 z, z1 =
+ q 2 1 z + q 2 1 z1 (z) q 1 z
2
+ z + z1 .
+ z + z1 q + q 1 2 z, z1
1 (z) =
The tridiagonal structure (B.2) is not surprising: similarly to the finitedimensional case,12 the
action of higher order operators can be written solely in terms of the qdifference operators 1
and the identity I . The qSturmLiouville problem associated with W1 takes the form
1 (z)(qz) + 1 (z) q 1 z + 1 z, z1 (z) = (z)
(B.3)
with the definitions above. Polynomial solutions can be constructed similarly to the case of W0 .
It is however important to notice that, for any z,
1 (z) (z)
=
,
1 (z) (z)
(B.4)
an immediate consequence of (10) and (13). Then, (B.4) implies that the system of Bethe equations associated with W0 is identical to the one for W1 . Consequently, up to some overall factors
polynomial eigenfunctions of both operators coincide.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B 241 (1984) 333.
L. Onsager, Phys. Rev. 65 (1944) 117.
L. Dolan, M. Grady, Phys. Rev. D 25 (1982) 1587.
G. von Gehlen, V. Rittenberg, Nucl. Phys. B 257 (1985) 351.
C. Ahn, K. Shigemoto, Mod. Phys. Lett. A 6 (1991) 3509.
B. Davies, J. Phys. A 23 (1990) 2245;
B. Davies, J. Math. Phys. 32 (1991) 2945;
S.S. Roan, Onsagers algebra, loop algebra and chiral Potts model, Max Planck Institute fr Mathematik, preprint,
1991.
J.H.H. Perk, Startriangle equations, quantum Lax operators, and higher genus curves, in: Proceedings 1987 Summer Research Institute on Theta Functions, Proceedings of Symposia in Pure Mathematics, vol. 49, American
Mathematical Society, Providence, 1989, pp. 341354;
H. AuYang, B.M. McCoy, J.H.H. Perk, S. Tang, Solvable models in statistical mechanics and linobreak Riemann
surfaces of genus greater than one, in: M. Kashiwara, T. Kawai (Eds.), Algebraic Analysis, vol. 1, Academic Press,
San Diego, 1988, pp. 2940.
E. Date, S.S. Roan, J. Phys. A 33 (2000) 3275.
G. von Gehlen, S.S. Roan, The superintegrable chiral Potts quantum chain and generalized Chebyshev polynomials,
hepth/0104144;
G. von Gehlen, Int. J. Mod. Phys. B 16 (2002) 2129.
S.M. Klishevich, M.S. Plyushchay, Nucl. Phys. B 628 (2002) 217;
S.M. Klishevich, M.S. Plyushchay, J. Phys. A 36 (2003) 11299.
G. von Gehlen, Int. J. Mod. Phys. B 16 (2002) 2129.
328
Abstract
We prove that the chiral propagator of the deformed N = 4 SYM theory can be made finite to all orders in
perturbation theory for any complex value of the deformation parameter. For any such value the set of finite
deformed theories can be parametrized by a whole complex function of the coupling constant g. We reveal
a new protection mechanism for chiral operators of dimension three. These are obtained by differentiating
the Lagrangian with respect to the independent coupling constants. A particular combination of them is a
CPO involving only chiral matter. Its allorder form is derived directly from the finiteness condition. The
procedure is confirmed perturbatively through order g 6 .
2006 Elsevier B.V. All rights reserved.
1. Introduction
Recently, following [13], there has been a renewed interest in the deformed N = 4 supersymmetric YangMills (SYM) theory. It has the same field content as N = 4 SYM, namely (in
an N = 1 formulation) a gauge superfield V and a set of three chiral matter superfields I ,
I = 1, 2, 3, all in the adjoint representation of the gauge group SU(N ). What distinguishes the
* Corresponding author.
330
(1)
where g is the N = 1 SYM coupling constant and the deformed commutator is defined as
[A, B] = AB BA.
(2)
The parameter can be considered real (its phase can be absorbed into a redefinition of I ),
while is in general complex. The undeformed N = 4 SYM is recovered when = = 1.
Although in principle both and can be taken as Taylor series expansions in powers of g 2
around g = 0, in most of the recent literature [48] the case of constant has been commonly
considered.
The main feature of the deformed theory is that, despite the breaking of N = 4 supersymmetry
down to N = 1, it can be made finite (and thus conformal) by imposing a condition on the
parameters g, , . The search for finite N = 1 theories has a long history [913]. In the most
general case one considers a superpotential of the Yukawa type
Yij k i j k ,
(3)
where i, j, k are combined color and flavor indices, and Yij k is a set of complex couplings. These
theories are finite if all beta and gamma functions vanish. In the matter sector Y and Y are
related through the nonrenormalization of the chiral vertex [14,15], so it is sufficient to demand
the vanishing of the matrix of gamma functions of the chiral superfields, (g, Y )ij = 0. This is a
set of conditions on the couplings which are to be adjusted order by order in perturbation theory.
The existence of a solution in the general case has been studied in [11,12].
The superpotential (1) is a particular case of (3) with the interesting feature that all matter
gamma functions are equal due to the Z3 Z3 symmetry of the potential.2 So, it is enough to
impose a single finiteness condition
(g, , ) = 0
(4)
to ensure that the matter propagators and couplings do not undergo infinite renormalization. This
feature of the socalled deformed N = 4 theory with superpotential (1) was essential for
finding its gravity dual in [3] and for the subsequent development in the context of the AdS/CFT
correspondence [17,18].
The question about the vanishing of the propagator corrections and of the beta function in the
N = 1 gauge sector of theories with the superpotential (3) is more subtle. A threeloop result is
available [10], but its generalization to all orders [1,13] relies on the existence of the socalled
exact function [19].
The first steps in the study of the perturbative aspects of the deformed theory with superpotential (1) were made in [4], with a particular accent on the chiral primary operators (CPO) in
the theory. Subsequently, the condition for finiteness were established at two loops in [5] and at
three loops in [6]. An allorder condition in the large N limit was found in [7]. The set of CPOs
(chiral ring) of the deformed theory was further studied in [58].
In the present paper we concentrate on two particular perturbative issues in the deformed
N = 4 SYM theory.
2 The most general superpotential with this property was found in [1,16]; see Section 3.2.4 for more details.
331
In Section 2 we investigate the finiteness properties of a theory with superpotential W, deformed by a gdependent deformation parameter. For future convenience we shall write it in the
form3
W, = g (g) tr 1 2 , 3 + (g) tr 1 2 , 3 ,
(5)
where is a complex constant,4 and is defined as
N 2 2 + 2
.
(6)
(N 2 2) + 2
The main result of Section 2 is the proof that for any complex constant and any complex
function (g) satisfying (0) = 0, there exists a unique function (g), such that the chiral
propagator is finite to all orders in perturbation theory, with the consequence that the chiral
field has a vanishing anomalous dimension. To be precise, since we shall always compute the
difference between the quantities in the deformed theory and in N = 4 SYM (which corresponds
to = = 1, = 0), everywhere in this paper by finite we actually mean as finite as in
N = 4 SYM. We derive the explicit form of the finiteness condition at order g 6 for any number
of colors N , and at order g 8 in the planar N limit. We also briefly discuss the corrections
to the threepoint vertices.
In Section 3 we study a particular type of CPO of dimension three, namely
OF = tr 1 2 3 + tr 1 3 2 ,
(7)
=
which is a mixture of the two terms in the superpotential. Its existence was revealed in [4] where
the oneloop value of the mixing coefficient was determined through a direct twopoint function
computation. This oneloop result was confirmed in [5,6] and in [8] it was shown that is not corrected at two loops. We compute the threeloop correction to the value of . However, the main
purpose of Section 3 is to show that can in fact be determined without graph calculations, but
directly from the finiteness condition (4). The key observation is that the quantum corrections to
the correlators of composite operators, i.e. their derivatives with respect to each independent coupling, are generated by the insertion of very special CPOs of the type I = a tr(W 2 ) + b tr()
(here tr(W 2 ) is the N = 1 SYM chiral Lagrangian). The latter are obtained by differentiation of
the chiral part of the Lagrangian, taking into account the relation among the couplings. To do
this we rewrite the superpotential in the form W = f tr( 1 [ 2 , 3 ]) + d tr( 1 { 2 , 3 }) and
treat the holomorphic couplings f, d as independent, while g is determined from the finiteness
condition (g, f, d, f, d) = 0. The derivatives with respect to f, d give rise to two CPOs If,d .
We can say that If,d generate quantum corrections along the tangent directions to the surface in
the moduli space defined by = 0. Then the operator OF is simply the linear combination of
If and Id such that tr(W 2 ) drops out. This means that the form of OF to all orders is directly
determined by the corresponding finiteness condition = 0. When restricted to three loops, the
general formula exactly reproduces the result of our graph calculation. Also, we can immediately
explain the observation of [8] that is not corrected at two loopsit simply follows from the
fact that has no twoloop contribution. In Section 3.2.4 we generalize the construction of OF
to the most general deformed theory which is made finite by a single condition [1,16].
3 The trace is over the color indices of the fundamental representation of the SU(N ) gauge group. The generators, T a ,
of the fundamental representation are normalized according to tr(T a T b ) = 12 ab .
4 The gdependence of the deformation parameter is hidden in the terms proportional to [ , ] . Indeed one can
rewrite (5) in the form of (1) with a gdependent deformation parameter (g).
332
(8)
c,d
and similarly for and . This implies also the vanishing of the color contractions of the and
(as well as the and ) terms in the superpotential W, of Eq. (5). Note also that if is a
pure phase  = 1, then also  = 1. The real function (g) and the complex function (g)
have a power series expansion in g 2 that we find useful to cast in the form
(g) =
n
(n) g 2 N ,
(9)
n=0
(g) =
n
(n)
g 2 N .
(10)
n=1
(0)
Note that by assumption (0) = 0, while (0) = = 0. At each order in g 2 the general
(n)
(n)
superpotential W, depends on 3 real parameters. For n > 0 they are (real) and (com(0)
plex), while for n = 0 we choose them to be (real) and (complex). As we shall see, this
choice allows us to express in a compact form the solution of the condition for the perturbative
finiteness of the chiral field propagator to all orders.
2.1. The chiral propagator to all orders
We start by reconsidering the order g 2 , g 4 and g 6 conditions for finiteness of the chiral propagator in the theory with the general superpotential W, of Eq. (5) (the order g 6 condition in the
case of the superpotential of Eq. (1) was found in [6]).
Let us write the action of the deformed theory in the form
S, = S0 + Sv + SW, ,
(11)
where S0 contains the free kinetic terms and Sv is the part of the standard N = 4 SYM action
involving the couplings of the gauge superfield V (including the gaugefixing term). Finally,
SW, is the part of the action involving the deformed superpotential W, given in Eq. (5). In
this notation the action of the undeformed N = 4 SYM theory reads ( = = 1, = 0)
Sg = S0 + Sv + SWg .
(12)
333
In the deformed theory the lowest components of the order g 2n correction to the propagator of
Sv +S
W,
1 (x ) (x ) ,
G2n
(13)
, (x1 , x2 ) = e
g 2n a 1 1b 2
where by eSv +SW, g 2n we denote all terms of order g 2n in the expansion of the exponent. The
similar computation, which in N = 4 SYM reads
Sv +SW
G2n
(14)
1 (x ) (x ) ,
g (x1 , x2 ) = e
g 2n a 1 1b 2
is know to give a finite result [20]. Hence, if the computation of the difference
2n
G2n (x1 , x2 ) = G2n
, (x1 , x2 ) Gg (x1 , x2 ),
(15)
tr T a T c , T d tr T b T d , T c = 0,
(18)
c,d
tr T a T c , T d tr T b T d , T c = 0,
(19)
c,d
5 Because of the Z symmetry of the action, all three chiral superfields are on the same footing, so our choice of the
3
flavor index is conventional.
334
tr T a T c , T d tr T b T d , T c = ab C .
(20)
c,d
where
(N 2 2)( + 1) + 2( + )
,
8N
N 2 (N 2 4)
.
C = C
((N 2 2) + 2)((N 2 2) + 2)
C =
(21)
(0)
2
2N 2
,
(N 2 2)( + 1) + 2( + )
(22)
(23)
exactly as for the simpler case of the potential in Eq. (1). Let us note that since the cancellation
is due to the vanishing of the numerical factor in front of one single diagram, both the logarithmically divergent and the finite part contributions vanish.
At the next perturbative order the difference between the superpotentials of Eqs. (5) and (1)
shows up. Indeed, if we impose the constraint (23) (we always choose the positive square root
for the solution) the order g 4 finiteness condition reads
(0) (1) C = 0,
(24)
implying
(1) = 0.
(25)
(1)
Note that due to the vanishing of the contractions (18) and (19) the complex coefficient
remains undetermined. Hence, contrary to what is said in previous papers the superpotential
W, is allowed to contain a different from zero order g 3 term. Since the cancellation is again
due to the vanishing of numerical factors in front of each diagram, both the divergent and the
finite parts are set to zero by Eq. (25).
The situation becomes more complicated at order g 6 , since at this order both the genuine
threeloop diagram and the oneloop diagram (multiplied by order g 4 coefficients) contribute.
Thus, if we impose Eq. (25), the order g 6 finiteness condition reads
(1)
2
C
2(0) (2) C +
3 (3)(k(0) )6 (N 2 4)
( 1)( 1)P3 (, N ) = 0,
256 (4 2 )2
N5
(26)
where
P3 (, N ) =
2 + + 1 2 + + 1 9 N 2 + 5( 1)2 ( 1)2 ,
(2)
(27)
and C and C are given in Eq. (21). Note that Eq. (26) is linear in , so there is always a
(2)
(1)
(1)
unique solution for as a function of , N and   for any . Let us stress that this is the
first case in which there are (nonvanishing) contributions from two different (super)diagrams. In
335
fact the first two terms in Eq. (26) come from the diagram with oneloop topology, while the third
term comes from the genuine threeloop diagram. Hence, the vanishing of the divergent part does
not automatically imply also the cancellation of the (potentially scheme dependent) finite parts.
Let us stress that these finite contributions modify only the normalization of the chiral superfield
propagator. This finite correction of the normalization, which is present only in the deformed
theory (remember that all terms in Eq. (26) originate from the deformed theory), will give rise
to a logarithmic divergency at the next order (g 8 ). Hence starting at order g 8 the explicit form of
the condition for finiteness of the chiral propagator will in general be scheme dependent.
It is clear that one can proceed iteratively. If we satisfy all the finiteness conditions up to
order g 2n , then the finiteness condition at order g 2(n+1) will schematically read
(p)
2(0) (n+1) C + fn+1 N, , g, , p = 1, . . . , n = 0,
(28)
where fn+1 is a computable function and we have used the lower order equations to express all
(q)
(p)
, q = 1, . . . , n in terms of the coefficients , p = 1, . . . , n 1. Eq. (28) is linear in (n+1)
and thus it has always a unique solution (since both (0) = 0 and C = 0).
To summarize we have shown that for any complex constant and any complex function
(g) satisfying (0) = 0, there exist a unique (possibly scheme dependent) real function
(g), such that the anomalous dimension of the chiral superfields is zero to all orders in perturbation theory.6
The question of the finiteness of the vector superfield propagator is more subtle and is beyond
the scope of this paper. It is closely related to the coupling constant renormalization which we
shall discuss in the next subsection. We note that the finiteness of the vector superfield propagator
up to order g 6 follows only from the order g 2 and g 4 conditions in Eqs. (23) and (25).
2.2. The threepoint vertices
Another important question which we would like to address is whether in the deformed theory
there is a coupling constant renormalization. To answer this question, one has to compute the
perturbative corrections to the threepoint vertex functions. There are four potentially different
such vertex functions, namely the triple chiral (or antichiral) vertex, the chiralantichiralvector
vertex, the triple vector vertex and the ghostghostvector vertex (see Fig. 2). All four vertices
in the action at tree level are proportional to g, but only the triple chiral vertex at higher orders
receives in addition to the standard perturbative corrections also finite corrections from (n) and
6 Since the potential (5) is a special case of the general Yukawa potential (3), our result could in principle be obtained
as a particular case of [11]. However, translating the parametrization from (3) to (5) is not an easy task. Therefore we
find it useful to give the explicit form of the finiteness condition and its solution adapted to our special case.
336
Fig. 3. The twoloop and the only planar threeloop contributions to the triple chiral vertex.
(n)
. Another peculiar feature of the triple chiral vertex is that its color structure depends on
g in a nontrivial way. Indeed, while the three vertices involving the vector field are always
proportional to the SU(N ) structure constants f abc , the triple chiral vertex is proportional to a
linear combination of the two deformed (by and ) commutators (see Eq. (5)).
We recall that the triple chiral vertex function obeys a nonrenormalization theorem [14,15]
which relates the propagator and the coupling constant renormalization factors. The triple vector vertex is related to the simpler ghostghostvector one by a SlavnovTaylor identity [22].
A similar identity relates the chiralantichiralvector vertex to the matter propagator.
For our purposes it again suffices to compute only the difference between the values of each
threepoint function in the deformed theory and in N = 4 SYM. Our results can be summarized
as follows.
At order g 3 and order g 5 all three vertices with external vector lines are exactly equal to
the corresponding ones in N = 4 SYM. Only the triple chiral vertex receives at order g 5 a
finite nonplanar correction from the first (super)diagram in Fig. 3. It affects both the [ , ] and
[ , ] structures in the effective superpotential Weff [8]. In particular, the correction to the [ , ]
structure is
(0)
Weff g 5 , =
( 1)( 1)(N 2 4)
3 (3)(k )5
32(4 2 )2 N 2 ((N 2 2)( + 1) + 2( + ))
P3 (, N ) tr 1 2 , 3 ,
(29)
( 1)((N 2 2) + 2)
3 (3)(k )5
32(4 2 )2 N 2 ((N 2 2)( + 1) + 2( + ))
2
N ( + 1) 2 6 + 2 + 4 2 + 1
+ ( 1)2 ( 1) 7( 1) 3 + 3 tr 1 2 , 3 .
Weff g 5 , =
(30)
At order g 7 vertices behave quite differently. The ghostghostvector vertex remains equal to its
N = 4 value. The other three vertices receive corrections from nonplanar diagrams. Whether
they sum up to zero or not is an open question.7 Only the triple chiral vertex receives also finite
planar corrections coming from the second (super)diagram in Fig. 3, which as explained in the
next subsection modify the chiral propagator at order g 8 .
7 To answer it one has to carefully compute the numerous threeloop super diagrams which contribute to the various
threepoint functions. We shall address this issue in a future publication.
337
(0)
2
2
,
(1 + )
g4:
(1) = 0,
g :
2(0) (2)
(1)
2
+
= 0,
(31)
(32)
(33)
because = 1/, = 1/, C /C = . After imposing these conditions, the cancellation of the order g 8 divergencies coming from the diagram with oneloop topology and the planar
4loop diagram depicted in Fig. 4, leads to the condition
(1) (2)
2(0) (3) +
(2) (1)
(0)
+
8 ( )8 (1 + 6()2 + ()4 )
= 5
( + 1)
( 1)4
,
= 85
( + 1)5
(34)
where 5 is a numerical constant proportional to (5) (see also [35]), and to obtain the second
equality we used the order g 2 condition of Eq. (31).
It follows that even in the planar limit, in order to make the theory finite we are obliged to
finetune order by order in g 2 the coefficients in the superpotential. Indeed, for generic the
above equation necessarily requires a nonvanishing correction to the superpotential. The only
exception is when  = 1, in which case the order g 2 condition alone is sufficient for all order
finiteness [7].
Let us stress that Eq. (34) is scheme independent, since it follows from requiring the cancellation of the leading logarithmic divergencies. However, as it involves two different (super)diagrams, the vanishing of the divergent part does not automatically imply the cancellation of
338
the (possibly scheme dependent) finite parts. These finite corrections change only the normalization of the chiral superfield propagator and will give rise to logarithmic divergencies at the next
order. Hence, even in the planar limit, starting from order g 10 , the explicit form of the condition
for finiteness of the chiral propagator will be scheme dependent.
Let us note also that the 5loop discrepancy in the planar maximally helicity violating (MHV)
amplitudes pointed out in [18], is actually not present thanks to the order g 8 finiteness condition
in Eq. (34). Indeed the vacuum diagram considered in [18] is equivalent to the 4loop planar
correction to the chiral propagator shown in Fig. 4, which, as we have shown, can be cancelled
by finetuning the parameters in the superpotential.
To conclude, let us briefly comment on the freedom which the conditions for finiteness of the
(n)
chiral propagator leave. As we already noted, all the coefficients
remain arbitrary. Thus one
(n)
simple choice is = 0 for all n. In this case the superpotential in Eq. (5) reduces to the simpler
expression (1), widely studied in the literature. However, as noted in [8] (see also Eq. (30)), even
if we start with the simple superpotential proportional only to [ , ] , the quantum corrections
to the effective superpotential Weff give rise also to contributions proportional to [ , ] . This
(n)
suggests that we may choose to precisely cancel the quantum corrections to the effective
superpotential proportional to [ , ] , obtaining
Weff tr 1 2 , 3 ,
(35)
(1)
=0
to all orders in perturbation theory. For the first two coefficients one finds in this case
(2)
= Weff g 5 , /N 2 , where Weff g 5 , is given in Eq. (30).
and
339
The quantum corrections to the correlator (36) can be obtained by differentiating it with respect to the couplings gi . Each such derivative leads to the insertion of a derivative of the action
into the correlator8
L(x0 , 0 , 0 ; gj )
G
d 4 x0 d 4 0
= e L
O(1) O(n)
gi
gi
L(0)
= d 4 x0 d 4 0
(37)
O(1) O(n) .
gi
In what follows we assume that the theory is (super)conformal, i.e. all the beta functions gi
vanish. As we already know, in the deformed theory this is achieved by imposing a constraint on
the couplings which should be taken into account when differentiating. We shall come back to
this essential point in Section 3.2.
Before discussing the superconformal insertion procedure, let us explain some details about
its conformal analog. To be more specific, let us consider the simplest case of scalar operators O
with n = 2. After integration over d 4 0 in (37) and setting = = 0, we obtain
L(x0 ; gj )
O(1)O (2) = d x0
O(1)O (2) ,
(38)
gi
gi
where L(x0 ) is the Lagrangian operator (the top component in the expansion of L). The bare
operators O in (38) are in general ill defined. Consequently, they must be renormalized, unless
they are protected. Let us start with the case of a single multiplicatively renormalized operator. We will use the notation [O](x) = lim0 O(x, ) with O(x, ) = Z(, , gi )O(x, ). Here
Z(, , gi ) is a renormalization factor depending on the couplings gi , on the regulator (e.g.,
a fourvector
in pointsplitting regularization) and on the renormalization scale (or subtraction
point) . O(x, ) is the regularized version of the bare operator (e.g., with the constituent fields
put at distances
). Now, suppose that the renormalized operator [O] is a conformal primary
of dimension = 0 + (gi ), where 0 is the naive and the anomalous dimension. In the
pointsplitting scheme we have [24]
(gi )/2
.
Z(, , gi ) = 2 2
(39)
Repeating the differentiation (38), but this time with the renormalization factors included, we
find
1 2 2 4
O(x1 , 1 )O (x2 , 2 ) =
ln 1 2 O(x1 , 1 )O (x2 , 2 )
gi
2 gi
d 4 x0 Lgi (x0 )O(x1 , 1 )O (x2 , 2 ) ,
(40)
where for short we have set Lgi = L/gi . Since we are taking the derivative of a finite quantity,
the apparent logarithmic singularity in the first term of the righthand side of Eq. (40) has to be
compensated by a similar singularity in the second term.9
L
e
. Connected correlators are automatically generated in
this way by differentiation. To simplify notations we will not explicitly indicate this.
9 Our discussion about this point is similar to that in [25], except for the regularization scheme used.
8 The functional integral in (36) should be divided by
340
To show how this comes about we recall that twopoint correlator of renormalized scalar
operators takes the form predicted by conformal invariance:
A(g)
[O](x1 )[O] (x2 ) = lim O(x1 , 1 )O (x2 , 2 ) = 2
,
1,2 0
(x12 )
(41)
Lgi (x0 )[O](x1 )[O] (x2 ) regular =
Bi (g)
,
2 x 2 )2 (x 2 )2
(x01
02
12
(42)
where regular means x0 = x1 = x2 . Inserting (42) into (40) leads to a divergent integral which
we regularize by splitting points 1 and 2
Bi (g)
Bi (g)
d 4 x0
= 2 ln 12 22 4
+ finite part,
(43)
2
2
2
2
2
2 )
2
(x12 )
x01 x01+1 x02 x02+2
(x12
where we have introduced the subtraction point, . We now see that this term can provide the
singularity which cancels the logarithm in the first term in the righthand side of (40), if
Bi (g)
= 2 2
.
gi
A(g)
(44)
The conclusion is that the anomalous dimension (g) is controlled by the regular part of
the (2 + 1)point correlator (42). This observation, generalized to the supersymmetric case,
explains why CPOs have no anomalous dimension. The reason is that in both the deformed
and undeformed theories superconformal invariance forbids the existence of a nonvanishing
Lgi (0)[O](1)[O] (2)regular with [O] being a CPO. Consequently, such operators keep their
naive dimension 0 , i.e., they are protected. However, this does not necessarily mean the total
absence of quantum corrections to the correlator [O][O] . Indeed, conformal invariance allows
contact term contributions to the (2 + 1)point correlator of the form
Lgi (x0 )[O](x1 )[O] (x2 ) contact = Ci (g) 4 (x01 ) + 4 (x02 )
1
2 )0
(x12
(45)
The appearance of such terms is related to the general fact that the factors 1/x 4 in (42) are singular distributions with a function type singularity.10 It is clear that such terms, integrated over
10 Note that in the case of CPOs with an integer 2, an ultralocal contact term, like 4 (x ) 4 (x ) for = 2,
0
01
02
0
could be added to (45). We need not consider such terms here since we are assuming x12 = 0.
341
the insertion point x0 , will give quantum corrections to the normalization A(g) of the correlator (41). This is precisely what happens to CPOs in the deformed theory, starting at two loops
[5,6]. In the undeformed theory the more powerful extended superconformal symmetry forbids
even the contact terms, so there the twopoint functions of CPOs are completely protected (for
more details see [26]).
We note that the above insertion procedure can be generalized in an obvious way in the presence of mixing, i.e. when the renormalized operators have the form Oi = Zij Oj .
3.2. CPOs as derivatives of the deformed N = 4 Lagrangian
3.2.1. Holomorphic form of the action. Finiteness condition
For our purposes in this section it is more convenient to rewrite the action of the deformed
N = 4 theory as follows:
d 4 x d 4 egV I egV I
S = tr
+ d 4 xL d 2 W W
+ d 4 xL d 2 f 1 2 , 3 + d 1 2 , 3
4
2
+ d xR d f 1 3 , 2 + d1 3 , 2
(46)
,
where W is the chiral fieldstrength of the gauge potential V ,
1
1
W = D 2 e2gV D e2gV = D 2 D V + O(g).
(47)
4g
2
Note that in (46) we have written the SYM kinetic term in its chiral form.
If needed, it can
be rewritten as an antichiral term according to the identity d 2 W 2 = d 2 W 2 valid in a
topologically trivial background.11 Unlike Section 2, here we prefer to parametrize the potential
by two complex coupling constants, f in front of the commutator (i.e., color tensor fabc ) and d in
front of the anticommutator (i.e., color tensor dabc ). Both of them as well as the gauge coupling
g are treated as small perturbative parameters f d g. To switch back to the notation of
Section 2 we need to perform the change of parametrization
g
f = ( + 1) (g) + ( + 1) (g) ,
2
g
d = ( 1) (g) + ( 1) (g) .
(48)
2
The deformed theory is finite (and hence conformal) up to three loops12 if the couplings satisfy
the relation
(g, f, d, f, d) = (1) + (3) = 0.
(49)
11 It is more convenient, although not essential, to work with a real gauge coupling g, therefore we do not introduce an
instanton angle.
12 We remark that now the counting of loops is somewhat different than in the preceding section. There all the
couplings were expressed in terms of g, so n loops meant perturbative level g 2n . Here nloop corrections are homogeneous polynomials of degree 2n in the couplings g, f, d.
342
This is the condition for vanishing anomalous dimension of the matter superfields involving a oneloop
16 2 (1) = 2
N2 4 2
d + 2Nf 2 2Ng 2
N
(50)
and a threeloop
3
24 (3)(N 2 4) 2
d 2(N 2 10)d4 3N 2 f 2 d 2 + d 2 f2
16 2 (3) =
(51)
3
N
contributions13 (cf. Eq. (26)). The absence of a twoloop contribution is explained by the fact
that in carrying out the two and threeloop calculation of the oneloop condition (1) = 0
has been used, after which no new condition arises at two loops. Alternatively, the calculation
may be done without imposing the oneloop condition. In this case one finds [13] a twoloop
contribution to of the form
(52)
where P (1)
is some homogeneous polynomial of degree two in the couplings whose explicit form
is not important for us. Similarly, (3) gets modified by a term of the type (1) P (2) with P (2) a
polynomial of degree four. Thus, the complete version of the finiteness condition (49) up to three
loops has the form
(1) 1 + P (1) + P (2) + (3) = 0.
(53)
Clearly, since P (1) , P (2) 1, we can multiply this equation by (1 + P (1) + P (2) )1 . In this way
we recover (49), up to terms of fourloop order which are beyond the scope of this section.
Following [13], we note that the threeloop condition (1) + (3) = 0 can be formally reduced
to a oneloop condition of the type (1) = 0 by a change of variables (or, equivalently, by a finite
coupling renormalization). For instance, one such change is
3 (3)(N 2 10) 3 2
9 (3)(N 2 4) 3
d d f ,
d d .
d d +
(54)
4
2
64 N
64 4 N 2
Finally, returning to the notation of Section 2, it is easy to see that the condition (49) with (1)
from (50) and (3) from (51), rewritten in terms of , (g) as indicated in (48) and expanded
up to g 6 , is equivalent to (23) at order g 2 , (25) at order g 4 and (26) at order g 6 .
f f
(55)
13 This condition is a particular case of the finiteness condition for the most general trilinear superpotential (3) discussed
in [13]. It was obtained in its explicit form for the case of the deformed N = 4 SYM potential in [4] at one loop, in [5]
at two loops and in [6] at three loops.
343
(56)
In this equation we have treated the holomorphic couplings f and d as independent, while the
gauge coupling is taken as a (real) function of them, g = g(f, d, f, d). This point of view is
preferable here because we want to obtain chiral operators through differentiation with respect
to the holomorphic couplings f, d. In Section 2 we adopted the alternative (i.e. perturbative)
point of view where the gauge coupling is the universal small parameter used in the perturbative
calculations.
In order
to compute the derivatives of the Lagrangian in (46) (completed with the gaugefixing
term d 4 x d 4 D 2 V D 2 V and with the ghost term) with respect to the independent holomorphic couplings, we first absorb the gauge coupling g into the gauge potential and the gaugefixing
parameter14
V g 1 V ,
g 2 .
(57)
The effect of this rescaling is that g now appears only in front of the classical SYM term W 2 in
the Lagrangian (recall (47)), while it drops out from the gaugefixing and ghost terms. So
2
Lg = tr W 2 ,
g
while the variation with respect to the holomorphic couplings gives
Ld = tr 1 2 , 3 .
Lf = tr 1 2 , 3 ,
(58)
(59)
It is now clear that the total derivative of G with respect to each holomorphic coupling gives
rise to the insertion of a chiral operator:
G
= d 4 xL d 2 0 If (0)O(1) O(n) ,
f
G
(60)
= d 4 xL d 2 0 Id (0)O(1) O(n) ,
d
where
2f 2
tr W + tr 1 2 , 3 ,
gg
2d 2
tr W + tr 1 2 , 3 .
Id =
gg
If =
(61)
In Section 3.3 we show that the insertions I are chiral primary operators of protected conformal dimension 0 = 3. Similarly, differentiating with respect to the antiholomorphic couplings
14 In [27] it is claimed that this redefinition of V may lead to the socalled rescaling anomaly, which is used to justify
the exact NSVZ beta function [19]. However, as mentioned in [27], the rescaling anomaly is not seen in dimensional
regularization. Here we adopt the point of view that there exists a scheme free from such anomalies. Note also that the
rescaling of the gaugefixing parameter in (57) has no effect on the correlators (36) of gauge invariant composite
operators O.
344
amounts to inserting antichiral operators (to this end the SYM kinetic term should be rewritten
in its antichiral form).
This result admits the following interpretation.15 The finiteness condition (49) can be viewed
as the equation of a real hypersurface in the complex moduli space of the couplings. In Section 2
we chose to describe it in a parametric fashion, by expressing the matter couplings as functions
(perturbative power series) of the gauge coupling g. In this section we use the holomorphic
couplings f and d as the independent coordinates on the surface whose tangent space is spanned
by the derivatives /f and /d. The quantum equivalents of the tangent space vectors are the
CPOs (61). Each of them generates quantum corrections to the Greens functions of operators in
the theory when moving along the corresponding tangent direction to the surface of couplings.
Finally, the operator OF (7) is nothing but the linear combination of If and Id from which
the SYM term W 2 drops out:
OF =
f + d 1 3 2
d If f Id
= tr 1 2 3 +
tr .
d f
f d
(62)
The explicit form of the relative coefficient in (62) is determined by a straightforward calculation.
From (49), (50) and (51) we find
f + d
N 2 f + (N 2 4)d 9 (3)(N 2 4)
= 2
d
f d
32 4
N f (N 2 4)d
2(N 2 4)d 3 df 3N 2 f3 d 2 N 2 d 2 ff 2 + 2(N 2 10)d 2 d 2 f
.
[N 2 f (N 2 4)d]2
(63)
The first term coincides with the oneloop result first obtained in [4]. The second term is the
threeloop correction; it has been obtained by expanding (f + d )/(f d ) in f d g up
to g 4 . The absence of a twoloop correction in (63) (independently noticed in [8]) is due to the
specific form (52) of the twoloop contribution which allows us to rewrite the finiteness condition
up to three loops in the form (49).
3.2.3. Perturbative calculation at order g 6
We have checked by an explicit computation that the operator defined in Eqs. (62), (63) has
vanishing anomalous dimension at order g 6 . Note that with the notation introduced in Eqs. (2),
(6), we can rewrite OF (up to an irrelevant rescaling) in the form
OF tr 1 2 , 3 ) + a2 g 2 + a4 g 4 + tr 1 2 , 3 .
(64)
We have computed through order g 6 the corrections to the twopoint functions of (the lowest
component of) the operator OF with all the operators it can mix with. There are only three such
operators, namely
tr 1 3 , 2 and tr W 2 .
tr 1 3 , 2 ,
(65)
All the corrections to the twopoint function of OF with tr(W 2 ), as well as the logarithmically
divergent correction to the twopoint function with tr(1 [3 , 2 ] ) vanish by the color contractions. At order g 6 the logarithmically divergent correction to the twopoint function with
tr(1 [3 , 2 ] ) comes only from the three superdiagrams depicted in Fig. 5. The first two are
15 E.S. is grateful to Ken Intriligator for this remark.
345
genuine order g 6 diagrams, while the last (order g 2 ) diagram appears multiplied by the order g 4
coefficient a4 in (64). All three diagrams lead to the same (logarithmically divergent) coordinate
space integral. The cancellation of this divergence, i.e. the vanishing of the order g 6 anomalous
dimension of