Nuclear Physics B 593 (2001) 330

www.elsevier.nl/locate/npe

New QCD-estimate of the kaon

penguin matrix elements and  0/
Stephan Narison a,b
a Laboratoire de Physique Mathmatique et Thorique, Universit de Montpellier II Place Eugne Bataillon,

34095, Montpellier Cedex 05, France

b KEK Theory group, 1-1-Oho, Tsukuba-city, Ibaraki 305, Japan

Received 1 May 2000; accepted 16 October 2000

Abstract
Firstly, we use the recent ALEPH/OPAL data on the V A spectral functions for fixing the
continuum threshold with which the first and second Weinberg sum rules should be satisfied in
the chiral limit. Then, we predict the values of the low-energy constants m + m 0 , L10 , and

3/2 

test the values of the electroweak kaon penguin matrix elements Q8,7 2 obtained from DMO-like

3/2
sum rules. Secondly, we use the data on the -total hadronic width R,V /A for extracting Q8 2 ,

3/2 

in the MS-scheme, and propose some new sum rules for Q7 2 in the chiral limit, where the
latter require more accurate data for the spectral functions near the -mass. Thirdly, we analyze

1/2
component of the I = 0 scalar
the effects to the matrix element Q6 2 , of the S2 (uu
+ dd)
meson, with its parameters fixed from QCD spectral sum rules. Our results should stimulate a further
attention on the role of the (expected large) gluonium component of the I = 0 scalar meson and
of the associated operator in the K amplitude. Finally, using our previous determinations,
we deduce, in the Standard Model (SM), the conservative upper bound for the CP-violating ratio:
 0 / 6 (22 9) 104 , which is in agreement with the present measurements. 2001 Elsevier
Science B.V. All rights reserved.

1. Introduction and generalities

CP-violation is one of the most important weak interactions phenomena in particle
physics [1,2], where in their presence, stable (under strong interactions) K 0 (s d) and
particles with definite strangeness eigenvalues 1 become unstable. Therefore,
S 0 (ds)
K
E-mail address: [email protected] (S. Narison).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 1 8 - 0

S. Narison / Nuclear Physics B 593 (2001) 330

the decay of a long-lived kaon 1 into a two-pion final state is an evidence for CP-violation.
The first observation of such a transition to the + mode, was discovered 36 years ago
S0 mixing. Since then, the transition into 0 0 and the phases of the ratio
by [3] from K 0 K
of the amplitudes:
+

A(KL + )
,
A(KS + )

00

A(KL 0 0 )
,
A(KS 0 0 )

(1)

have been also observed. For a phenomenological analysis, it is convenient to work with
quantities where the final pion states are in a definite isospin. Then, one introduces the
S 0 mixing) CP-violation parameter , and the quantity governing
indirect (through K 0 K
the so-called = 1/2 rule (enhancement of the I = 0 over the I = 2 transitions):


A[KL ()I =0 ]
,
A[KS ()I =0 ]

A[KS ()I =2 ]
,
A[KS ()I =0 ]

and the direct (in the amplitude) CP-violation parameter:



1 A[KL ()I =2 ]
0
 .

2 A[KS ()I =0 ]

(2)

(3)

In terms of these quantities, one can express the measured + and 00 quantities as:
+ =  +

0

,
1 + / 2

00 = 

2 0
,
1 2

(4)

from which one can deduce the experimental value [4]:

 ' (2.280 0.013) 103 ei(43.50.1) .
0

To proceed further, one introduces the isospin amplitude:

S 0 ()I = iAI eiI ,
A K
A K 0 ()I = iAI eiI ,

(5)

(6)

and makes use of the mass-difference 1m mL mS , between KS and KL , and of

S 0 complex mass matrix: Re M12 ' 1m/2).
M12 (off-diagonal dispersive part of the K 0 K
Therefore, one can write, within a good approximation [1]:


Re A2
1 Im M12 Im A0
+
(7)
,
ei(2 0 )
 ' ei 4
1m
Re A0
Re A0
2
and



Re A2
1
Im A2 Im A0

.
 0 ' ei(2 0 + 2 )
Re A0 Re A2 Re A0
2

(8)

Experimentally [4], 2 0 ' (42 4)0 , and then, using the data on (KS
+ )/ (KS 0 0 ), one can deduce:
exp '

1 i(424)0
e
,
22

(9)

1 |K i and |K i are very close to the CP-eigenstates |K 0 i 1 (|K 0 i |K

S 0 i) and |K 0 i 1 (|K 0 i+ |K
S 0 i)
L
S
1
2
2
2
0
0
0
0
8
with CP|K1 i = +|K1 i and CP|K2 i = |K2 i. Their lifetime are L ' 5.2 10 s 15.51 m 5.8 102 S .
Their mass-difference is 1m ML MS = (3.522 0.016) 1012 MeV.

S. Narison / Nuclear Physics B 593 (2001) 330

while the recent measurement reported by the kTeV and NA48 experiments [5] on the
direct CP-violation ratio is 2


Re  0 / exp ' (21.4 4.0) 104 .
(10)
It is fair to say that a simultaneous explanation of these two previous experimental numbers
remains a challenge for the present theoretical predictions within the Standard Model (SM)
[1,2,710].
2. Theory of  0 /
In the SM, it is customary to study the 1S = 1 process from the weak Hamiltonian:
X
GF

Ci ()Qi (),
Heff = Vud Vus
2
i=1
10

(11)

where Ci () are perturbative Wilson coefficients known including complete NLO QCD
corrections [7], which read in the notations of [7]:
Ci () zi ()

Vt d Vts
y (),
i
Vud Vus

(12)

where Vij are elements of the CKM-matrix; Qi () are non-perturbative hadronic matrix
elements which need to be estimated from different non-perturbative methods of QCD
(chiral perturbation theory, lattice, QCD spectral sum rules, . . .). In the choice of basis
of [7], the dominant contributions come from the four-quark operators which are classified
as:
Currentcurrent:
Q1 (s u )V A (u d )V A ,
QCD-penguins:
Q3 (s d)V A

V A ,
()

Q2 (s u)V A (ud)
V A .

Q4 (s d )V A

u,d,s

Q5 (s d)V A

(13)

( )V A ,

u,d,s

V +A ,
()

Q6 (s d )V A

u,d,s

( )V +A .

(14)

u,d,s

Electroweak-penguins:
Q7

X
3
V +A ,
(s d)V A
e ()
2
u,d,s

X
3
e ( )V +A ,
Q8 (s d )V A
2
u,d,s

2 If one includes the preliminary NA48 result Re( 0 /) ' (12.2 2.9 (stat.) 4 (syst.)) 104 from the 98
data sample, the preliminary new experimental world average becomes Re( 0 /) ' (19.3 2.4) 104 [6].

S. Narison / Nuclear Physics B 593 (2001) 330

Q9

X
3
V A ,
e ()
(s d)V A
2
u,d,s

Q10

X
3
(s d )V A
e ( )V A ,
2

(15)

u,d,s

where , are colour indices; e denotes the electric charges 3 reflecting the electroweak
nature of Q7,...,10 , while V (+)A (1 (+)5 ) . Using an OPE of the amplitudes,
one obtains:


0
' Im t P (1/2) P (3/2) ei ,
(16)

where  0  0 [see Eqs. (7) and (8)]; t Vt d Vts can be expressed in terms of
the CKM matrix elements as ( being the CKM phase) [7,11]:
Im t |Vub ||Vcb | sin ' (1.33 0.14) 104 ,
from B-decays and . The QCD quantities P (I ) read:
  

GF || X
Ci () ()I =0 Qi K 0 0 (1 IB ),
P (1/2) =
2|| Re A0
i
X
  

GF
(3/2)
=
Ci () ()I =2 Qi K 0 2 .
P
2|| Re A2

(17)

(18)

IB ' (0.16 0.03) quantifies the SU(2)-isospin breaking effect, which includes the one
of the 0 mixing [12], and which reduces the usual value of (0.25 0.08) [7] due to
0 mixing. It is also expected that the QCD- and electroweak-penguin operators:
3/2

3/2

Q8 B8 /m2s + O(1/Nc ),

1/2

1/2

Q6 B6 /m2s + O(1/Nc ),

(19)

give the dominant contributions to the ratio  0 / [13]; B are the bag factors which are
expected to be 1 in the large Nc -limit. Therefore, a simplified approximate but very
informative expression of the theoretical predictions can be derived [7]:
2

110
0
13 Im t

m
Ss (2) [MeV]

 (4) 

mt
1/2
3/2
MS
B6 (1 IB ) 0.4 B8
,
(20)
165 GeV
340 MeV
bK = 0.80 0.15 of the 1S = 2 process has been used. This value
where the average value B
includes the conservative value 0.58 0.22 from Laplace sum rules [14]. The values of the
top quark mass and the QCD scale (4) [4,15] are under a quite good control and have
MS
small effects. A recent review of the light quark mass determinations [16] also indicates
that the strange quark mass is also under control and a low value advocated in the previous
literature to explain the present data on  0 / is unlikely due the lower bound constraints
3 Though apparently suppressed, the effect of the electroweak penguins are enhanced by 1/ as we shall see
later on in Eq. (18).

S. Narison / Nuclear Physics B 593 (2001) 330

Table 1
Penguin B-parameters for the 1S = 1 process from different approaches at = 2 GeV. We use the
value ms (2) = (119 12) MeV from [16], and predictions based on dispersion relations [24,25] have
been rescaled according to it. We also use for our results f = 92.4 MeV [4], but we give in the text
their ms and f dependences. Results without any comments on the scheme have been obtained in
the MSNDR-scheme. However, at the present accuracy, one cannot differentiate these results from
the ones of MSHV-scheme
Methods

1/2

B6

3/2

B8

3/2

B7

Comments

Lattice [9,17,18]

0.6 0.8
unreliable

0.7 1.1

0.5 0.8

Huge NLO
at matching [19]

Large Nc [20]

0.7 1.3

0.4 0.7

0.10 0.04

1.5 1.7

O(p0 /Nc , p2 )
scheme?
O(p2 /Nc ); mq = 0
scheme?

Models
Chiral QM [10]

1.2 1.7

0.9

ENJL + IVB [21]

2.5 0.4

1.4 0.2

0.8 0.1

1.2

1.6 3.0

0.7 0.9

Dispersive
Large Nc + LMD
+ LSD-match. [24]

0.9
strong -dep.

NLO in 1/Nc ,

DMO-like SR [25]

1.6 0.4
huge NLO

0.8 0.2

mq = 0
Strong s, -dep.

1.4 0.3

0.7 0.2

Debate for fixing

the Slope [27]

This work
DMO-like SR:
[25] revisited

2.2 1.5
inaccurate

0.7 0.2

mq = 0
Strong s, -dep.

-like SR

inaccurate

tc -changes

RV A

1.7 0.4

mq = 0

1.0 0.4
6 1.5 0.4

MS-scheme
ms (2) > 90 MeV

L -model [22]
NL -model [23]

FSI [26]

S2 (uu
+ dd)
from QSSR

3/2

B8

= 0.8 GeV
rel. with MS?
NLO in 1/Nc
mq = 0
Not unique
1 GeV; scheme?
M : free; SU(3)F trunc.
1 GeV; scheme?

S. Narison / Nuclear Physics B 593 (2001) 330

from the positivity of the QCD spectral function or from the positivity of the m2 corrections
to the GMOR PCAC relation. For a consistency with the approach used in this paper, we
shall use the average value of the light quark masses from QCD spectral sum rules (QSSR),
e+ e and -decays given in [16]:
m
Ss (2) ' (119 12) MeV,

m
Sd (2) ' (6.3 0.8) MeV,

m
Su (2) ' (3.5 0.4) MeV.

(21)

Using the previous experimental values, one can deduce the constraint in [16] updated:
1/2

B68 B6

3/2

0.48 B8

' 1.73 0.50 (resp. > 1.0 1.2),

(22)

if one uses the value of ms in Eq. (21) (respectively the lower bound of (90 100) MeV
reported in [16]). This result shows a possible violation of more than 2 for the leading
1/2
3/2
1/Nc vacuum saturation prediction 0.52 corresponding to B6 B8 1. Consulting
the available predictions reviewed in [7], which we will summarize and update in Table 1,
one can notice that the values of the B-parameters have large errors. One can also see that
results from QCD first principles (lattice and 1/Nc ) fail to explain the data, which however
can be accommodated by various QCD-like models. We shall come back to this discussion
when we shall compare our results with presently available predictions. It is, therefore,
clear that the present estimate of the four-quark operators, and in particular the estimates
of the dominant penguin ones given previously in Eq. (19), need to be reinvestigated.
Due to the complex structures and large size of these operators, they should be difficult
to extract unambiguously from different approaches. In this paper, we present alternative
theoretical approaches based as well on first principles of QCD ( -decay data, analyticity),
for predicting the size of the QCD- and electroweak-penguin operators given in Eq. (19).
In performing this analysis, we shall also encounter the electroweak penguin operator:
3/2

Q7

3/2

B7 /m2s + O(1/Nc ),

(23)

and some other low-energy constants (m m 0 , L10 ) though not directly relevant
to  0 /.

3. Tests of the Sacrosante Weinberg and DMO sum rules in the chiral limit
3.1. Notations
Before estimating these condensates, we shall test the procedure used in [25] by
analyzing the classics DMO- and Weinberg-like sum rules [28,29]. This analysis will also
allow us to fix the cut-off parameter tc until which the data on V A spectral functions
from ALEPH/OPAL [30,31] are known. We shall be concerned here with the two-point
correlator:
Z

LR (q) i d 4 x eiqx h0|T JL (x) JR (0) |0i = g q 2 q q LR q 2 , (24)

built from the left- and right-handed components of the local weak current:

(1 5 )d,
JL = u

JR = u
(1 + 5 )d,

(25)

S. Narison / Nuclear Physics B 593 (2001) 330

and/or using isospin rotation relating the neutral and charged weak currents:
1
1
Im LR
(v a).
V A
2
4 2
The first term is the notation in [25], while the last one is the notation in [30,31].

(26)

3.2. The sum rules

The sacrosante DMO and Weinberg sum rules read in the chiral limit: 4
Z
S0

1
Im LR = f2 ,
ds
2

Z
S1

ds s

1
Im LR = 0,
2

Z
S1

ds 1
Im LR = 4L10 ,
s 2

Z
Sem





s
4
1
Im LR = f2 m2 m2 0 ,
ds s log 2
2
3

(27)

where f |exp = (92.4 0.26) MeV is the experimental pion decay constant which should
be used here as we shall use data from -decays involving physical pions; m m 0 |exp '
4.5936(5) MeV; L10 f2 hr2 i/3 FA [hr2 i = (0.439 0.008) fm2 is the mean pion
radius and FA = 0.0058 0.0008 is the axial-vector pion form factor for e ] is
one the low-energy constants of the effective chiral Lagrangian [1]. In order to exploit
these sum rules using the ALEPH/OPAL [30,31] data from the hadronic tau-decays, we
shall work with their Finite Energy Sum Rule (FESR) versions (see, e.g., [32,35] for such
= hs si), this is equivalent to
= hddi
a derivation). In the chiral limit (mq = 0 and huui
truncate the LHS at tc until which the data are available, while the RHS of the integral
remains valid to leading order in the 1/tc expansion in the chiral limit, as the breaking of
these sum rules by higher dimension D = 6 condensates in the chiral limit which is of the
order of 1/tc3 is numerically negligible [33].
3.3. Matching between the low and high-energy regions
In order to fix the tc values which separate the low and high energy parts of the spectral
functions, we require that the 2nd Weinberg sum rule (WSR) S1 should be satisfied by the
present data. As shown in Fig. 1, this is obtained for two values of tc : 5
tc ' (1.4 1.5) GeV2

and tc ' (2.4 2.6) GeV2 .

(28)

Though the 2nd value is interesting from the point of view of the QCD perturbative
calculations (better convergence of the QCD series), its exact value is strongly affected
4 Systematic analysis of the breaking of these sum rules by light quark masses [32] and condensates [33,34]
within the context of QCD have been done earlier.
5 One can compare the two solutions with the t -stability region around 2 GeV2 in the QCD spectral sum rules
c
analysis (see, e.g., Chapter 6 of [36]).

10

S. Narison / Nuclear Physics B 593 (2001) 330

Fig. 1. FESR version of the 2nd Weinberg sum rule versus tc in GeV2 using the ALEPH/OPAL data
of the spectral functions. Only the central values are shown.

by the inaccuracy of the data near the -mass (with the low values of the ALEPH/OPAL
data points, the 2nd Weinberg sum rule is only satisfied at the former value of tc ).
After having these tc solutions, we can improve the constraints by requiring that the
1st Weinberg sum rule S0 reproduces the experimental value of f 6 within an accuracy
2-times the experimental error. This condition allows to fix tc in a very narrow margin due
to the sensitivity of the result on the changes of tc values 7
tc = (1.475 0.015) GeV2 .

(29)

4. Low-energy constants L10 , m m 0 and f in the chiral limit

Using the previous value of tc into the S1 sum rule, we deduce:
L10 ' (6.26 0.04) 103 ,

(30)

which agrees quite well with more involved analysis including chiral symmetry breakings [31,37], and with the one using a lowest meson dominance (LMD) of the spectral
integral [24].
Analogously, one obtains from the Sem sum rule:
1m m m 0 ' (4.84 0.21) MeV.

(31)

This result is 1 higher than the data 4.5936(5) MeV, but agrees within the errors with
the more detailed analysis from -decays [31,38] and with the LMD result of about 5
MeV [24]. We have checked that moving the subtraction point from 2 to 4 GeV slightly
decreases the value of 1m by 3.7% which is relatively weak, as expected. Indeed, in the
chiral limit, the dependence does not appear (to leading order in s ) in the RHS of the
Sem sum rule, and then, it looks natural to choose:
2 = tc ,

(32)

6 Though we are working here in the chiral limit, the data are obtained for physical pions, such that the
corresponding value of f should also correspond to the experimental one.
7 For the second set of t -values in Eq. (28), one obtains a slightly lower value: f = (84.1 4.4) MeV.
c

S. Narison / Nuclear Physics B 593 (2001) 330

11

as tc is the only external scale in the analysis. At this scale the result increases slightly by
2.5%. One can also notice that the prediction for 1m is more stable when one changes the
value of tc = 2 . Therefore, the final predictions from the value of tc in Eq. (29) fixed from
the 1st and 2nd Weinberg sum rules are:
L10 ' (6.42 0.04) 103 ,

1m ' (4.96 0.22) MeV,

(33)

which we consider as our best predictions.

For some more conservative results, we also give the predictions obtained from the
second tc -value given in Eq. (28). In this way, one obtains:
f = (87 4) MeV,

1m ' (3.4 0.3) MeV,

L10 ' (5.91 0.08) 103 ,

(34)

where one can notice that the results are systematically lower than the ones obtained in
Eq. (33) from the first tc -value given previously, which may disfavour a posteriori the
second choice of tc -values, though we do not have a strong argument favouring one with
respect to the other. 8 Therefore, we take as a conservative value the largest range spanned
by the two sets of results, namely:
f = (86.8 7.1) MeV,
L10 ' (5.8 0.2) 10

1m ' (4.1 0.9) MeV,

3

(35)

which we found to be quite satisfactory in the chiral limit. The previous tests are very
useful, as they will allow us to jauge the confidence level of the next predictions.

3/2

5. Soft pion and kaon reductions of h()I =2 |Q7,8 |K 0 i

An interesting approach combining pion and kaon reductions in the chiral limit with
dispersion relation techniques have been proposed recently [25] 9 in order to estimate the
matrix element:
 3/2  

3/2 
(36)
Q7,8 2 ()I =2 Q7,8 K 0 .
In the chiral limit mu,d,s m2 ' m2K = 0, one can use soft pion and kaon techniques in
order to relate the previous amplitude to the four-quark vacuum condensates:

3/2 
2

Q7

3/2 
2

Q8

4
3/2 
O ,
f3 7


4 1
3/2  1
3/2 
O
' 3
+ O8
,
f 3 7
2
'

(37)

8 Approach based on 1/N expansion and a saturation of the spectral function by the lowest state within a
c
narrow width approximation (NWA) favours the former value of tc given in Eq. (29) [39].
9 A similar approach based on large N and a lowest meson dominance (LMD) of the spectral functions is also
c
done in [24].

12

S. Narison / Nuclear Physics B 593 (2001) 330

3/2

3/2

where we use the shorthand notations: h0|O7,8 |0i hO7,8 i, and f = (92.42
0.26) MeV. 10 Here:
X

3/2
3/2
S 3
S 5 3
S 5 3 ,
S 3

O7 O1 [25] =
2
2
2
2
u,d,s
X

3/2
S a 3
S 5 a 3
S 5 a 3 ,
S a 3
(38)

O8 =
2
2
2
2
u,d,s

where 3 and a are flavour and colour matrices. Using further pion and kaon reductions
in the chiral limit, one can relate this matrix element to the B-parameters:
3/2

B7

3/2

B8

3 (mu + md ) (mu + ms ) 1
3/2 
M2 '
Q7 2 M2 ,
2
2
4
f
m
mK




1
(m
+
m
)
(m
+
m
)
u
d
u
s 1
3/2 
M2 '
Q8 2 M2 ,
2
2
4
m
f
mK

(39)

where all QCD quantities will be evaluated in the MS-scheme and at the scale M .
3/2

6. h()I =2 |Q7,8 |K 0 i from DMO-like sum rules in the chiral limit

3/2

In a previous paper [25], the vacuum condensates hO7,8 i which are related to the
3/2

weak matrix elements h()I =2 |Q7,8 |K 0 i through the soft pion and kaon reduction
techniques (see previous section) have been extracted using DasMathurOkubo(DMO)and Weinberg-like sum rules based on the difference of the vector and axial-vector spectral
functions V ,A of the I = 1 component of the neutral current:
2

3/2 
s O8 (2 )

Z
=

ds s 2
0

16 2
3/2  2
O7 ( ) =
3

2
(V A )(s),
s + 2



s + 2
(V A )(s),
ds s log
s
2

(40)

where is the subtraction point. Due to the quadratic divergence of the integrand, the
previous sum rules are expected to be sensitive to the high energy tails of the spectral
functions where the present ALEPH/OPAL data from -decay [30,31] are inaccurate. This
inaccuracy can a priori affect the estimate of the four-quark vacuum condensates. On the
other hand, the explicit -dependence of the analysis can also induce another uncertainty.
En passant, we check below the effects of these two parameters tc and . After evaluating
the spectral integrals, we obtain at = 2 GeV and for our previous values of tc in Eq. (29),
the values (in units of 103 GeV6 ) using the cut-off momentum scheme (c.o.):

3/2 

3/2 
O7 c.o. ' (0.11 0.01),
(41)
s O8 c.o. ' (0.69 0.06),
10 In the chiral limit f would be about 84 MeV. However, it is not clear to us what value of f should be used

here, so we shall leave it as a free parameter which the reader can fix at his convenience.

S. Narison / Nuclear Physics B 593 (2001) 330

13

where the errors come mainly from the small changes of tc -values. If instead, we use the
second set of values of tc in Eq. (28), we obtain by setting = 2 GeV:

3/2 

3/2 
O7 c.o. ' (0.10 0.03),
(42)
s O8 c.o. ' (0.6 0.3),
which is consistent with the one in Eq. (41), but with larger errors as expected. We have also
3/2
3/2
checked that both hO8 i and hO7 i increase in absolute value when increases where a
3/2
stronger change is obtained for hO7 i, a feature which has been already noticed in [24].
In order to give a more conservative estimate, we consider as our final value the largest
range spanned by our results from the two different sets of tc -values. This corresponds to
the one in Eq. (42) which is the less accurate prediction. We shall use the relation between
the momentum cut-off (c.o.) and MS-schemes given in [25]:



3/2 

3/2 

3/2 
3
3
+ 2ds O8 ,
O7 MS ' O7 c.o. + as
8
2


2 

3/2 

3/2 

3/2 
119
119
as
as
(43)
O8 c.o. as O7 ,
O8 MS ' 1
24
24
where ds = 5/6 (resp. 1/6) in the so-called Naive Dimensional Regularization (NDR)
(resp. t HooftVeltmann HV) schemes; 11 as s / . One can notice that the as
coefficient is large in the 2nd relation (50% correction) [40], and the situation is worse
because of the relative minus sign between the two contributions. Therefore, we have added
a rough estimate of the as2 corrections based on the naive growth of the PT series, which
here gives 50% corrections of the sum of the two first terms. For a consistency of the whole
approach, we shall use the value of s obtained from -decay, which is [30,31]:
s (M )|exp = 0.341 0.05 H s (2 GeV) ' 0.321 0.05.
104

Then, we deduce (in units of

3/2 
O7 MS ' (0.7 0.2),

GeV6 )

(44)

at 2 GeV:

3/2 
O8 MS

' (9.1 6.4),

(45)

3/2
hO8 i

comes from the estimate of the as2 corrections appearing in

where the large error in
Eq. (43). In terms of the B factor and with the previous value of the light quark masses in
Eq. (21), this result, at = 2 GeV, can be translated into:
 
4

ms (2) [MeV] 2
92.4
3/2
,
B7 ' (0.7 0.2)
119
f [MeV]
 
4

ms (2) [MeV] 2
92.4
3/2
.
(46)
B8 ' (2.5 1.3)
119
f [MeV]
Our results in Eqs. (45) compare quite well with the ones obtained by [25] in the
MS-scheme (in units of 104 GeV6 ) at 2 GeV:

3/2 

3/2 
O7 MS ' (0.70 0.10),
(47)
O8 MS ' (6.7 0.9),
using the same sum rules but presumably a slightly different method for the uses of
the data and for the choice of the cut-off in the evaluation of the spectral integral.
11 The two schemes differ by the treatment of the matrix.
5

14

S. Narison / Nuclear Physics B 593 (2001) 330

Our errors in the evaluation of the spectral integrals, leading to the values in Eqs. (41)
and (42), are mainly due to the slight change of the cut-off value tc . 12
The error due to the passage into the MS-scheme is due mainly to the truncation of the
3/2
3/2
QCD series, and is important (50%) for hO8 i and B8 , which is the main source of
errors in our estimate.
As noticed earlier, in the analysis of the pion mass-difference, it looks more natural
to do the subtraction at tc . We also found that moving the value of can affects the
3/2
value of B7,8 .
3/2

For the above reasons, we expect that the results given in [25] for hO8 i though interesting
are quite fragile, while the errors quoted there have been presumably underestimated.
Therefore, we think that a reconsideration of these results using alternative methods are
mandatory.

3/2

7. h()I =2 |Q8 |K 0 i from the hadronic tau total decay rates

In the following, we shall not introduce any new sum rule, but, instead, we shall
exploit known informations from the total -decay rate and available results from it, which
have not the previous drawbacks. The V A total -decay rate, for the I = 1 hadronic
component, can be deduced from [35] (hereafter referred as BNP), and reads: 13
X (D)
3
R,V A = |Vud |2 SEW
V A .
(48)
2
D=2,4,...

|Vud | = 0.9753 0.0006 is the CKM-mixing angle, while SEW = 1.0194 is the electroweak
corrections [41]. In the following, we shall use the BNP results for R,V /A in order to
deduce R,V A :
The chiral invariant D = 2 term due to a short distance tachyonic gluon mass [42,43]
cancels in the V A combination. Therefore, the D = 2 contributions come only
from the quark mass terms:


25
(49)
M2 V(2)A ' 8 1 + as (M ) mu (M )md (M ),
3
as can be obtained from the first calculation [32], where mu (M ) ' (3.5 0.4) MeV,
md (M ) ' (6.3 0.8) MeV [16] are respectively the running coupling and quark
masses evaluated at the scale M .
The dimension-four condensate contribution reads:




9 2 2 2
4 (4)
2
(50)
M V A ' 32 1 + as m f + O m4u,d ,
2
12 A slight deviation from such a value affects notably previous predictions as the t -stability of the results
c
(tc 2 GeV2 ) does not coincide with the one required by the 2nd Weinberg sum rules. At the stability point the
predictions are about a factor 3 higher than the one obtained previously.
13 Hereafter we shall work in the MS-scheme.

S. Narison / Nuclear Physics B 593 (2001) 330

15

and the Gell-MannOakes

where we have used the SU(2) relation huui
= hddi
Renner PCAC relation:
= 2m2 f2 .
+ ddi
(mu + md )huu

(51)

By inspecting the structure of the combination of dimension-six condensates entering

in R,V /A given by BNP [35], which are renormalization group invariants, and using
a SU(2) isospin rotation which relates the charged and neutral (axial)-vector currents,
the D = 6 contribution reads:


2

235
2
3/2 
235
(6)
as
as 2 O8
M6 V A = 2 48 4 as 1 +
48
48
M


3/2 
+ as O7
,
(52)
where the overall factor 2 in front expresses the different normalization between
the neutral isovector and charged currents used respectively in [25] and [35], whilst
all quantities are evaluated at the scale = M . The last two terms in the Wilson
3/2
coefficients of hO8 i are new: the first term is an estimate of the NNLO term by
assuming a nave geometric growth of the as series; the second one is the effect of a
tachyonic gluon mass introduced in [43], which takes into account the resummation of
the QCD asymptotic series, with: as 2 ' 0.06 GeV2 . 14 Using the values of s (M )
given previously, the corresponding QCD series behaves quite well as:

3/2 
(53)
Coefficient of O8 ' 1 + (0.53 0.08) 0.28 + 0.18,
where the first error comes from the one of s , while the second one is due to the
unknown as2 -term, which introduces an uncertainty of 16% for the whole series. The
last term is due to the tachyonic gluon mass. This leads to the numerical value:


3/2 

3/2 
(6)
. (54)
M6 V A ' (1.015 0.149) 103 (1.71 0.29) O8 + as O7
If, one estimates the D = 8 contribution using a vacuum saturation assumption, the
relevant V A combination vanishes to leading order of the chiral symmetry breaking
terms. Instead, we shall use the combined ALEPH/OPAL [30,31] fit for V(8)/A , and
deduce:
(8) 
(55)
V A exp = (1.58 0.12) 102 .
We shall also use the combined ALEPH/OPAL data for R,V /A , in order to obtain:

R,V A exp = (5.0 1.7) 102 ,

(56)

Using the previous informations into the expression of the rate given in Eq. (48), one can
deduce:
V A ' (4.49 1.18) 102 .
(6)

14 This contribution may compete with the dimension-8 operators discussed in [44].

(57)

16

S. Narison / Nuclear Physics B 593 (2001) 330

This result is in good agreement with the result obtained by using the ALEPH/OPAL fitted
(6)
mean value for V /A :

(58)
V(6)A fit ' (4.80 0.29) 102 .
We shall use as a final result the average of these two determinations, which coincides with
the most precise one in Eq. (58). We shall also use the result:


3/2
hO7 i
1
3
'
resp.
,
(59)
3/2
16
hO i 8.3
8

3/2

3/2

where, for the first number we use the value of the ratio of B7 /B8 which is about
0.7 0.8 from, e.g., lattice calculations quoted in Table 1, and the formulae in Eqs. (37)
to (39); for the second number we use the vacuum saturation for the four-quark vacuum
condensates [34]. The result in Eq. (59) is also comparable with the estimate of [25] from
the sum rules given in Eq. (40). Therefore, at the scale = M , Eqs. (52), (58) and (59)
lead, in the MS-scheme, to:

3/2 
(60)
O8 (M ) ' (0.94 0.21) 103 GeV6 ,
where the main errors come from the estimate of the unknown higher order radiative
corrections. It is instructive to compare this result with the one using the vacuum saturation
assumption for the four-quark condensate (see, e.g., BNP):

3/2 
32
2 (M ) ' 0.65 103 GeV6 ,
(61)
O8 v.s. ' huui
18
which shows a 1 violation of this assumption. This result is not surprising as analogous
violations have been obtained in other channels [36]. We have used for the estimate of
Si the value of (mu + md )(M ) ' 10 MeV [16] and the GMOR pion PCAC relation.
h
However, this violation of the vacuum saturation is not quite surprising, as a similar fact
has also been observed in other channels [30,31,36], though it also appears that the vacuum
saturation gives a quite good approximate value of the ratio of the condensates [30,31,36].
The result in Eq. (60) is comparable with the value (0.98 0.26) 103 GeV6 at =
2 GeV M obtained by [25] using a DMO-like sum rule, but, as discussed previously,
the DMO-like sum rule result is very sensitive to the value of if one fixes tc as in Eq. (29)
according to the criterion discussed above. Here, the choice = M is well-defined, and
then the result becomes more accurate (as mentioned previously our errors come mainly
from the estimated unknown s3 term of the QCD series). Using Eqs. (37) and (59), our
previous result in Eq. (60) can be translated into the prediction on the weak matrix elements
in the chiral limit:
3

 3/2 0  2

92.4
3


,
(62)
()I =2 Q8 K M ' (2.58 0.58) GeV
f [MeV]
normalized to f , which avoids the ambiguity on the real value of f to be used in a such
expression. Our result is higher by about a factor 2 than the quenched lattice result [9,17].
A resolution of this discrepancy can only be done after the inclusion of chiral corrections
in Eqs. (37) to (39), and after the uses of dynamical fermions on the lattice. However,

S. Narison / Nuclear Physics B 593 (2001) 330

17

some parts of the chiral corrections in the estimate of the vacuum condensates are already
included into the QCD expression of the -decay rate and these corrections are negligibly
small. We might expect that chiral corrections, which are smooth functions of m2 will not
affect strongly the relation in Eqs. (37) to (39), though an evaluation of their exact size
is mandatory. Using the previous mean values of the light quark running masses [16], we
deduce in the chiral limit and at the scale M :

 
4


ms (M ) [MeV] 2
92.4
3/2
2
.
(63)
B8 M ' (1.70 0.39)
119
f [MeV]
One should notice that, contrary to the B-factor, the result in Eq. (62) is independent to
leading order on value of the light quark masses.
3/2

8. New alternative sum rules for hO7 i

3/2

Here, we shall attempt to present new sum rules for extracting O7 . In so doing,
we work with the renormalized D = 6 condensate contributions to the difference of the
vector and axial-vector two-point correlators. Using the expression given in the appendix
of BNP [35], one can deduce in the MS-scheme and in the chiral limit:




3/2 

3/2 
119
2 3
2
as () O8 + as () O7
(q ) LR (q ) = 4s () 1 +
24



q 2
8 3/2
3/2
2s2 () log 2
O8 + O7 ,
(64)

3
which has the generic form:
q 2
,
(65)
2
where, we remind that as s / is the renormalized coupling and A, B, C are
constant numbers. One can notice that by working with the different q 2 -derivatives of
(q 2 )3 V A (q 2 ), one can eliminate the effects of the A and B terms, and derive the
Laplace sum rule:
q 6 LR (q 2 ) As () + Bs2 () + Cs2 () log

Ztc

3 s

ds s e




3/2  8
3/2 
1
2
tc
Im LR (s) ' 2s ( )(1 e
) O8 + O7
,

(66)

where we have transferred in the RHS the QCD continuum effect starting from the
threshold tc . Alternatively, we can also derive a -like sum rule: 15
Ztc
0







3/2  8
3/2 
1
ds 3
s n1
2
Im LR (s) ' 2s (tc )
O8 + O7
s 1
. (67)
tc
tc
n+1
3

Formally, these sum rules are much better than the one proposed in [24,25], as the leading
-dependence has disappeared after taking different derivatives or after performing the
15 The coefficient has been checked using a compilation of [45].

18

S. Narison / Nuclear Physics B 593 (2001) 330

Cauchy integral. However, unlike the ones in [24,25], one has gained one power of s,
which renders the analysis more sensitive to the high-energy tail of the spectral functions
where the data near the -mass are quite bad [30,31]. One should also notice that here
the RHS starts at order s2 , which means that chiral corrections can affect dangerously the
RHS of the sum rules, and can compete with the dimension-six condensate contributions.
Using the present ALEPH/OPAL data on the V A spectral functions, and a traditional
-stability analysis of the Laplace sum rule, we realize that the -stability is reached at
exceptional large -values of about (1.8 3) GeV2 M2 , where the OPE can already
break down. In the case of the -like sum rule, one finds that for a given value of tc , one has
a n-stability where the value of n increases with tc (n > 2.5). However, at a such value of
n, one needs more and more information (by duality) on the non-perturbative contributions
3/2
to the sum rules. From our analysis and using the value of hO8 i obtained in Eq. (60), we
obtain the conservative range:

3/2 
(68)
2 103 6 O7 6 102 GeV6 ,
where the lower (respectively higher) value corresponds to the first (respectively second)
set of tc -values given in Eq. (28). Therefore, we conclude that in order to extract more
reliable informations on the previous sum rules, one needs to have much more informations
on these sum rules both experimentally and theoretically. In particular, these sum rules can
be more useful when accurate data near the -mass is available and all chiral symmetry
breaking terms are included both in the RHS of the sum rule and in the derivation of the
relation between the kaon matrix elements and the vacuum condensates.
9. I = 0 scalar meson contribution to h( + )I =0 |Q6 |KS0 i
1/2

We study the effect of a direct production of a I = 0 scalar meson intermediate state,

to the K 0 ( + )I =0 decay process. Before
+ dd),
which we shall denote by S2 (uu
S
doing our analysis, let us present the status of the scalar meson spectrum below 1 GeV.
9.1. A short review on the scalar meson spectrum below 1 GeV
A much more complete review and analysis of the complex structure of scalar mesons
spectra is given in [46]. Here, we shall be concerned with the spectrum of light scalar
mesons below 1 GeV:
a0 (980),

f0 (975),

(400 1200),

(69)

as quoted by PDG [4]. We associate the isovector state a0 (980) to the divergence of the
vector current:

Vud
= (mu md )u(i)d,

(70)

which is natural from the point of view of chiral symmetry [36,4648] and in the
construction of an effective chiral Lagrangian including resonances [49]. In this scheme the
K0 (1.43) [4] is the s u partner of the a0 . The predicted hadronic and two-photon widths

S. Narison / Nuclear Physics B 593 (2001) 330

19

of the a0 using vertex sum rules [36,46,48] are in good agreement with present data [4,
50]. These data indeed confirm the qq
nature of the a0 and disfavour some other exotic
interpretations, which can further be tested through measurements of the radiative decay
at Daphne.
In our analysis of the I = 0 isoscalar channel, we consider the trace of the energy
momentum tensor:

X
1
mi i i ,
(71)
= (s )G2 + 1 + m (s )
4
u,d,s

where and m are the function and mass anomalous dimension. We shall consider the
bare states before mixing:

+ dd),
which is degenerate to the a0 because of
The meson S2 is associated to (uu
the good SU(2) symmetry.
The S3 s s state is above 1 GeV due to SU(3) breaking [36,46], and will not give
significant effects in our analysis.
The meson B is a gluonium state, which has been needed for solving [51] the
inconsistencies between the substracted [53] and unsubstracted [54] gluonium sum
rules. 16 The B mass is expected to be around 0.7 1 GeV, but it will not play a
significant role in this analysis, as the gluoniumquarkonium mixing in the propagator
(mass mixing) via the off-diagonal two-point correlator is small [36,52].
The separation of the quark and gluon components of the current is allowed by

is RGI, while s G2 only mix with m

renormalization group invariance (RGI) as m
to higher order in s [55]. The hadronic couplings, decay constants and masses of these
mesons have been estimated using vertex sum rules [46,48,52], low-energy theorems [36,
46,51], and/or some SU(3) symmetry relations among the meson wave functions [48]. It
comes out that:
The B couples strongly and universally to pairs of Goldstone bosons (large violation
of the OZI rule) 17 [36,46,51], which invalidates the lattice results in the quenched
approximation. 18
The S2 is relatively narrow with a width of about 120 MeV and couples almost equally
S
to and KK.
The I = 0 scalar spectrum below 1 GeV, i.e., the observed wide -meson seen below
1 GeV [4,56] and the narrow f0 (980) [4] states, is expected in our approach to come
and the gluonium B bare states.
+ dd)
from a maximal mixing between the S2 (uu
16 The resolution of such inconsistencies has been also improved recently by the inclusion of the new 1/q 2 -term
induced by the tachyonic gluon mass in the OPE [43].
17 This decay mixing which occurs via 3-point function should not be confused with the mass mixing via
an off-diagonal 2-point function. There is not a contradiction between a large decay mixing and a small mass
mixing.
18 The gluonium mass obtained in the quenched approximation of about 1.5 GeV can be identified with the one
of about 1.6 GeV obtained from the unsubtracted sum rule [46,54], which is shown [46,51] to couple weakly to
Goldstone pairs but strongly to glue rich U (1)A states like 0 0 , through mixing to 0 , and to 4 through
B B pairs.

20

S. Narison / Nuclear Physics B 593 (2001) 330

Recent data favour such a maximal gluoniumquarkonium mixing scheme [50],

together with the qq
nature of the isovector a0 (980) state, though further refined tests
are still needed.
scalar meson
+ dd)
9.2. Parameters of the S2 (uu
In the following, we shall give the values of the decay constant and couplings of the S2
which is the relevant particle in the present analysis.
Its decay constant has been fixed using Laplace sum rule for the associated two-point
correlator. At the minimum of the sum rule variable 0 , 19 and at the inflexion point
of its change versus the QCD continuum threshold of about 2.6 GeV2 , it has the
value [36,46]:
fS /(mu + md )(0 ) ' (0.32 0.08),

(72)

where, presented in this way, the number in the RHS is not sensitive to the change of
quark mass values, but has an anomalous dimension, and it runs like the inverse of the
quark mass. It is normalized as:

1
2 i = 2 fS MS2 .
+ dd|S
(73)
(mu + md )h0|uu
2
One should notice that using the value of (mu + md )(0 ) given in [16], the value of fS
is about 2 MeV, which is much smaller than f , and which invalidates the estimate
fS f often proposed in the literature. Instead, it is the quantity MS2 fS m2 f ,
which is almost constant.
The S2 hadronic coupling to has been fixed using leading order results from vertex
sum rules. It reads [36,46,48,52]:


16 3
0
0 exp MS2
' 2.5 GeV,
(74)
gS + huui
2
3 3
corresponding to 0 ' 1 GeV2 . It leads to the decay width (S2 + ) ' 120
MeV, with the normalization:


|gS + |2
4m2 1/2
.
(75)
1
(S2 + ) =
16MS
MS2
This result is in good agreement with the one obtained from the a0 coupling, by
using SU(3) symmetry for the meson wave functions [48]:
r
3
ga ' (2.50 0.15) GeV,
(76)
gS + '
2 0
where we have used the peak data: (a0 ) ' (57 7) MeV [4]. One should
notice that this coupling does not vanish in the chiral limit because it behaves like
19 The inclusion of the effect of tachyonic gluon increases the value of from 0.5 GeV2 [36,46] to
1 GeV2 [43], improving the duality between the resonance and the QCD sides of the sum rules, but does
not almost affect the result, like in the case of the pion sum rule.

S. Narison / Nuclear Physics B 593 (2001) 330

21

huui,

and, up to SU(3)-breakings, one expects an universal coupling of the S2 to

Goldstone boson pairs.
We shall use these parameters as inputs in the following analysis.
scalar meson contribution to KS ( + )I = 0 decay
+ dd)
9.3. The S2 (uu
We shall work with on-shell kaon, such that the tadpole diagram will not contribute in
20
our analysis [57]. We can write, in the chiral limit: mu = md = 0 and hs si = hddi:

1/2
Q6 2
 1/2  

( + )I =0 Q6 K 0


+ uu|0ih0|
5 d|0ih |s u|K 0 i + h + |dd

s 5 d|K 0 i . (77)
' 2 h + |u
For convenience, we shall evaluate the matrix elements at the scale = mc because
the Wilson coefficients are also given at this scale [7]. We shall use the value of mc
from combined sum rule analysis of the charmonium and D-meson [58] and the QCD
spectral sum rule average of the light quark masses quoted in [16]. They read at the
scale = mc :
mc (mc ) ' (1.20 0.05) GeV,
md (mc ) ' (8 1) MeV,

ms (mc ) ' (147 15) MeV,

mu (mc ) ' (4.4 0.5) MeV.

(78)

The first term of the weak matrix element is well-known, and can be related to the
K ll semi-leptonic form factors (see, e.g., [13]):
  



2
m2 + f m2 /(ms mu ),
(79)
h |s uK 0 = f+ MK
with, to leading order in the chiral symmetry breaking terms: f+ 1 and f 0. It
leads to:
5 d|0ih |s u|K 0 i(mc )
h + |u

2f m2 m2K m2
'
(md + mu ) (ms mu )
2

142.6
GeV3 ,
(80)
' (0.323 0.032)
(ms mu ) [MeV]
where mi are the running quark masses evaluated at the scale mc . Chiral corrections
to these terms are known to be about 10% in the literature (see, e.g., [23]), which have
been included into the error estimate.
For the second term, we assume that it is dominated by the direct production of the
S2 -scalar meson in the s-channel for an on-shell kaon p2 = m2K . Therefore, it can be
decomposed as:
+ uu|0i
+ uu|0i

= h + |S2 ihS2 |dd

h + |dd
2fS
gS +
MS2 ,

(p2 MS2 ) (mu + md )

20 We follow the notations and conventions of [13].

(81)

22

S. Narison / Nuclear Physics B 593 (2001) 330

where for on-shell kaon. With the values of the parameters given previously, we
conclude that the scalar meson contribution is:
+ uu|0ih0|

s 5 d|K 0 i(mc )
h + |dd


155
GeV3 ,
' (0.53 0.13)
(ms + md ) [MeV]

(82)

where we take into account the fact that QSSR cannot fix the signs of the S2 coupling
and decay constant, which will be fixed later on from chiral constraints on the weak
amplitude [59]. The error in this determination comes mainly from the one of decay
constant fS .
For on-shell kaon, and neglecting, to a first approximation, mu,d (respectively m2 )
versus ms (respectively m2K ), we deduce from Eqs. (77) to (81), the approximate
relation:



2 1 

1/2 
MK
2 2f m2 m2K
fK gS + fS
1+
1 2
, (83)
Q6 2 (mc )
(md + mu ) ms
f
m2
MS
which we can consider as an updated version of the expression given by VSZ in [13].
satisfying the double chiral constraint conditions (vanishing of the amplitude when
2 0 and f = f ) [57], which are recovered if
MK
K


2 1
MK
gS + fS
1.
1 2
m2
MS

(84)

This value is obtained within the errors from the values of the S2 -parameters given
previously.
For the numerics, we shall use more precise values of the different parameters by
keeping corrections to order m2 and mu,d . Therefore, taking Eqs. (80) and (82) into
Eq. (77), one obtains the final value of the weak matrix element:

1/2
Q6 2 (mc )
2

142.6
'
(ms mu ) [MeV]



(ms mu ) [MeV]
GeV3 .
(0.65 0.09) (0.53 0.13)
(85)
142.6
Using the relation [7]: 21

1/2 
Q6 2 (mc ) ' 4

r 
2

m2K
3
1/2
2 (fK f )B6 (mc ),
2 ms + md

(86)

where fK ' 1.22 f , the previous result can be translated into:

21 We shall use the usual parametrization in terms of m2 , but it can be misleading in view of the m -dependence
s
s
of our results in Eqs. (80) and (82).

S. Narison / Nuclear Physics B 593 (2001) 330

23

1/2
B6 (mc )


ms + md 2
' 3.7
ms mu



(ms mu ) [MeV]
. (87)
(0.65 0.09) (0.53 0.13)
142.6

Evaluating the running quark masses at 2 GeV, with the values given previously, one
deduces:
1/2

B6 (2) ' (1.0 0.4) for ms (2) = 119 MeV,

6 (1.5 0.4) for ms (2) > 90 MeV.

(88)

The errors added quadratically have been relatively enhanced by the partial cancellations of the two contributions.

10. Comparison of our results with some other predictions

3/2

1/2

In this section, we shall compare our values of B7,8 and B6

with the results in Table 1.

3/2

10.1. Value of B7

Our value in Eq. (46):

3/2
B7 ( = 2

ms (2) [MeV]
GeV) ' (0.7 0.2)
119

2 

92.4
f [MeV]

4
,

(89)

comes from a reanalysis of the DMO-like sum rule used in [25], and in [24] within a
large Nc expansion and a lowest meson dominance (LMD). One can notice in Table 1
a quite good agreement between the results from different approaches. However, due
to the strong -dependence of the result, one should be careful when giving its value.
Our analysis from the Laplace and -like sum rules are unfortunately inconclusive
using present -decay data and present theoretical approximation (chiral limit).
3/2

10.2. Value of B8

Our result in Eq. (63):

3/2
B8

M2

ms (M ) [MeV]
' (1.7 0.4)
119

2 

92.4
f [MeV]

4
,

(90)

comes from the analysis of the total hadronic width R,V A .

The closest comparison to be made is the one with [25] where soft pion and kaon
reductions together with DMO sum rules have been used. We have stressed that in
order to obtain our value, we have not introduced any new sum rule but took advantage
of the existing measurement of the vector and axial-vector components of the -total
width and the measured values of the corresponding D = 6 vacuum condensates.

24

S. Narison / Nuclear Physics B 593 (2001) 330

Our result agrees numerically within the errors with the one of [25], but we have also
shown that the DMO sum rules lead to an inaccurate value due mainly to the bad
convergence of the QCD series.
Our result includes NLO corrections, an estimate of the NNLO terms and the
effect of a tachyonic gluon mass which phenomenologically takes into account the
resummation of the QCD asymptotic series.
Our result is in the range given by some linear [22] and non-linear [23] models, but
is 1 to 2 higher than the largest values obtained from the present lattice [9,18], large
Nc [20], chiral quark model [10] and the one including final state interactions [26].
Though the agreements with the results from the models are interesting, it is not
clear to us how to connect the two approaches, as in these models, an I = 0 scalar
resonance has been introduced with the parameters of the observed -meson which
we expect [36,46,51] to have a large gluon component in its wave function. The
difference with the result from final state interactions [26] is more rewarding, and
needs a much better understanding in connection to the comments raised in [27].
1/2

10.3. Value of B6

scalar meson contribution, through

Assuming a dominance of the I = 0 S2 (uu
+ dd)
1/2
the operator Q6 , to the KS ( + )I = 0 decay amplitude, we have obtained:
1/2

B6 (mc )

 


(ms mu ) [MeV]
ms + md 2
, (91)
(0.65 0.09) (0.53 0.13)
' 3.7
ms mu
142.6
leading to:
1/2

B6 (2) ' (1.0 0.4) for ms (2) = 119 MeV,

6 (1.5 0.4) for ms (2) > 90 MeV,

(92)

which we give in Table 1.

Contrary to conventional approaches [1,7,9,13], which do not consider the effect of
a scalar meson, our result shows that the qq
component of the scalar meson tends to
1/2
brings the value of B6 to the one of the leading 1/Nc expectation.
In our analysis, the mass of the S2 is fixed from SU(2) symmetry arguments,
and obtained from the two-point function sum rule to be about the one of the
a0 (980), which is relatively high compared with the kaon mass. Therefore, the main
contribution observed here comes from the values of the S2 -coupling to and of
its decay constant. However, for a more definite conclusion, it is important to look
for the effect of the gluon component of the , which can eventually give a sizeable
contribution in the amplitude through a new operator other than the one discussed
here. This new feature may clarify the observed enhancement from a direct final state
interactions (FSI) analysis of the amplitude. We plan to come back to this point in a
future work.

S. Narison / Nuclear Physics B 593 (2001) 330

25

Present lattice results [9,18] are still unreliable [19], as the NLO QCD corrections at
the matching scale between the lattice and continuum results are huge. Measuring
the effects of the scalar meson on the lattice seems to be difficult due to the
propagator [60].
Due to the partial cancellations of the two contributions in the weak amplitude, taking
into account the alone effect of the qq
component of the isoscalar meson is not
sufficient for explaining the large enhancement obtained from some other approaches
quoted in Table 1 which we list below:
Some incomplete large Nc result [20] including higher order corrections O(p2 /Nc )
in the chiral limit.
A version of the ENJL-model [1] with an intermediate vector bosons [21] and the
chiral quark model [10] where both models are based on the 1/Nc expansion.
However, for the chiral quark model the predictions correspond to a lower value
of the scale . A clear connection with these results with the MS scheme as well
as the relation of the parameters used there with lattice and QSSR calculations is
needed.
Final state interactions [26] where there is still a debate for fixing the slope of the
amplitude [27].
Enhancements due to the isoscalar meson have also been found from the -model
approaches [22,23], which is mainly due to the small value of the -mass used in
the -propagator appearing in Eq. (81). However, the uncertainties come from the
fact that, in the linear sigma models [22] the Lagrangian is not unique, while in the
non-linear models [23], the mass is a free parameter which is usually identified
with the observed wide -meson having a mass in the range (0.4 1.2) GeV and
a width of about (600 1000) MeV [4,56]. On the contrary, in the present work,
for the reasons previously explained, the scalar meson entering into the analysis
is not the observed where, within our framework, the comes from a maximal
and gluonium (B ) states. Indeed, as explained
(decay) mixing between a S2 (qq)
in the previous sections, only the qq
component (the hypothetical S2 state) of the
is relevant for the present operator. 22 However, some advanced versions of the
effective Lagrangian approach, which can separate explicitly the qq
from the gluon
component of the scalar meson, are needed for the present problem. 23
A clever explanation showing the connections of the different results reviewed here
in Table 1 in order to have an unified explanation of these different determinations is
still needed.

22 The effect of the gluon component (called ) through the scalar propagator is negligible (mass mixing)
B
[36,46,52] as it comes from the off-diagonal quarkgluon two-point function. This fact does not contradict the
large gluonium decay into (decay mixing) which comes from a vertex function [36,46,51].
23 Some attempts to introduce the I = 0 scalar meson within the effective Lagrangian framework exist in the
literature [22,23,49,61].

26

S. Narison / Nuclear Physics B 593 (2001) 330

1/2

3/2

10.4. Values of B6 /B8

3/2

1/2

Using our previous determinations of B8 (M ) in Eq. (63) and B6 (2) in Eq. (88), and
the previous value of ms (2), one can deduce the ratio:
1/2

R68

B6

3/2

B8

' 0.6 0.3,

(93)

and their combination:

3/2

B68 B6

3/2

0.48 B8

' (0.3 0.4) for ms (2) = 119 MeV,

(94)

where we have added the errors quadratically.

Instead, using the lower bound ms (2) > 90 MeV reported in [16] into the expressions of
3/2
1/2
B8 (M ) in Eq. (63) and B6 (2), one can deduce the conservative upper bound:
3/2

B68 B6

3/2

0.48 B8

6 (1.0 0.4) for ms (2) > 90 MeV,

(95)

where again we have added the errors quadratically.

11. Value and upper bound of  0 /
The estimated value in Eq. (94) does not satisfy the constraint required in Eq. (22) for
explaining the present data on the CP-violation ratio  0 / given in Eq. (10). It leads,
for ms (2) = 119 MeV, to the prediction:
0
' (4 5) 104.
(96)

The failure for reproducing the data may indicate the need for other contributions than
the alone qq
scalar meson S2 (not the observed )-meson for explaining, within the
standard model (SM), these data. Among others, a much better understanding of the
effects of the gluonium (expected large component of the -meson [46,48,51]) in the
amplitude, through presumably a new operator needs to be considered.
If we use instead the conservative upper bound for B68 in Eq. (95) corresponding
to the lower bound for the strange quark mass ms (2) > 90 MeV, we can deduce the
bound: 24
 0 / 6 (22 9) 104 .

(97)

The errors come mainly from B68 (40%) and Im t (10.5%), which we have
1/2
added quadratically. In B6 , the large error is due to the partial cancellation of the
contributions from the semi-leptonic form factors and the S2 resonance. This bound
agrees within the errors with the data on the CP-violation ratio  0 / given in Eq. (10).
24 The real value of f

the result.

3/2
(92.4 or 87 MeV) used in the chiral limit expression of B8 does not affect significantly

S. Narison / Nuclear Physics B 593 (2001) 330

27

12. Comments on the 1I = 1/2 rule

However, unlike Ref. [22], we do not expect that our result will affect significantly the CPconserving 1I = 1/2 rule process. Indeed, according to the analysis in [1,7], the amplitude
Re A0 of this process is dominated by the pure I = 1/2 combination:
Q Q2 Q1 ,

(98)

where its Wilson coefficient is relatively enhanced compared with the ones of Q+
can contribute,
+ dd)
Q2 + Q1 and Q6 . Moreover, the one of Q6 , where the S2 (uu
can even be zero at the subtraction point = mc for a given renormalization scheme (socalled HV-scheme), but still remains negligible at larger values of where the perturbative
calculations of these Wilson coefficients can be trusted. Instead, octet scalar may play a rule
in this process as has been emphasized in the first study of this process on the lattice [62].
We plan to analyze carefully this process in a future work.

13. Summary
We have used the recent ALEPH/OPAL data on the V A hadronic spectral functions
from -decays for fixing the matching scale separating the low and high-energy
regions (continuum threshold), at which the first and second Weinberg sum rules
should be realized in the chiral limit [Eq. (29)]. We have used this information
for predicting and for testing the accuracy of the low-energy constants m + m 0
[Eq. (31)] and L10 [Eq. (30)], and the electroweak kaon penguin matrix elements
3/2
h()I =2 |Q8,7 |K 0 i [Eqs. (45) and (46)] obtained from DMO-like sum rules in the
chiral limit.
3/2
We have estimated the value of the weak matrix element h()I =2 |Q8 |K 0 i
using the measured V /A -decay rate and the experimentally fitted value of the
dimension six-operators, without introducing any additional sum rules. Our results, in
Eqs. (62) and (63), indicate a deviation from the vacuum saturation of the four-quark
condensates, where analogous violations have been already found from the analysis
of other channels [30,31,36].
We have introduced some alternative new sum rules in Eqs. (66) and (67) in order
3/2
to estimate h()I =2 |Q7 |K 0 i, which require improved data in the region near the
-lepton mass and more theoretical inputs in order to be useful. At present, the result
from the DMO-like sum rule obtained in Eq. (46) is more meaningful.
3/2
We remind that the results for h()I =2 |Q8,7 |K 0 i matrix elements have been
obtained in the chiral limit as we have taken advantage of the soft pion and kaon
reductions techniques in order to express them in terms of the vacuum condensates.
Main improvements of our results need the inclusion of these chiral corrections.

+ dd)
In the last part of the paper, we have analyzed the effect of the S2 (uu
1/2

component of the I = 0 scalar meson into the h()I =0 |Q6 |K 0 i matrix element.
We found that its main contribution is due to the values of its decay constant

28

S. Narison / Nuclear Physics B 593 (2001) 330

and coupling to giving the predictions in Eqs. (85) and (91), but not on the
enhancement due to its mass in the propagator.
We have expressed our predictions on the matrix elements in terms of the Bparameters, with the absolute values given in Eqs. (63), (91) and rescaled at 2 GeV
in Table 1, where we have used the value ms (2 GeV) ' (119 12) MeV [16]. These
results are compared with the ones from different approaches. The ratio of these Bparameters is also given in Eq. (93) and compared with the existing values. Their
combination given in Eq. (94) is also compared with the constraint in Eq. (22).
We have used our previous determinations of the penguin matrix elements in order
to predict the value of the CP-violating ratio  0 /. Our conservative upper bound in
Eq. (97) corresponding to ms (2) > 90 MeV, agrees with the present day experiments
given in Eq. (10). However, our estimate in Eq. (96) corresponding to ms (2) ' 119
MeV fails to explain the data, which is mainly due to the partial cancellation of
1/2
the contributions implied by the chiral constraints governing the h()I =0 |Q6 |K 0 i
matrix element. This failure may not be quite surprising as the observed -meson is
expected to have a large gluonium admixture responsible for its large width [46,
48,51], which can manifest in the K amplitude through an eventual new
operator not considered until now, but most probably along the line of dimension-8
operators discussed recently [44]. Further study of the gluonium effect is therefore
mandatory before an eventual consideration of possible effects due to new physics.

Acknowledgements
It is a pleasure to thank the KEK-theory group and Kaoru Hagiwara for their hospitality.
Stimulating discussions with Guido Martinelli in a preliminary stage of this work
are acknowledged. The original form of the manuscript has been improved thanks to
the different questions and discussions during the Japan Tour at the universities of
Ochanomizu (Tokyo), Hiroshima, Kanazawa, Tsukuba, Sendai and Kyoto, and at the
Riken Institute (Tokyo); during the visit at the KIAS Institute (Seoul), during the QCD
00 conference (Montpellier), and thanks to the referees report.

References
[1] For an introduction to CP-violation for kaons, see, e.g.:
E. de Rafael, Nucl. Phys. Proc. Suppl. B, A 7 (1989) 1;
E. de Rafael, hep-ph/9502254, Lectures given at the Tasi-school on CP-violation and the limits
of the Standard Model, Boulder, CO (1994).
[2] J. Steinberger, in: J. Tran Thanh Van (Ed.), CP-Violation in Particle Physics and Astrophysics,
Editions Frontires, 1990.
[3] J.H. Christenson et al., Phys. Rev. Lett. 13 (1964) 128.
[4] PDG96, R.M. Barnett et al., Phys. Rev. D 54 (1996) 1;
PDG97, L. Montanet et al., Phys. Rev. D 54 (1997) 1;
PDG98, C. Caso et al., Eur. Phys. J. C 3 (1998) 1.

S. Narison / Nuclear Physics B 593 (2001) 330

29

[5] kTeV Collaboration, B. Hsiung, QCD99 Euroconference (Montpellier), Nucl. Phys. Proc.
Suppl. B 86 (2000) 312;
NA48 Collaboration, B. Peyaud, QCD99 Euroconference (Montpellier), Nucl. Phys. Proc.
Suppl. B 86 (2000) 303.
[6] NA48 Collaboration, G. Tatishvili, private communication.
[7] For a review, see, e.g., S. Bosch et al., hep-ph/9904408;
G. Buchalla, A.J. Buras, M. Lautenbacher, Rev. Mod. Phys. 68 (1996) 1125.
[8] M. Jamin, hep-ph/9911390;
U. Nierste, QCD99 Euroconference (Montpellier), Nucl. Phys. Proc. Suppl. B 86 (2000) 329.
[9] For a review, see, e.g., G. Martinelli, hep-ph/9910237.
[10] S. Bertolini, J.O. Eeg, M. Fabbrichesi, hep-ph/0002234;
S. Bertolini, J.O. Eeg, M. Fabbrichesi, Rev. Mod. Phys. 72 (2000) 65.
[11] See, e.g., A. Ali, D. London, Eur. Phys. J. C 9 (1999) 687.
[12] G. Ecker et al., hep-ph/9912264.
[13] A.I. Vainshtein, V.I. Zakharov, M.A. Shifman, Sov. Phys. JETP 45 (4) (1977) 670;
V.I. Zakharov, private communication.
[14] S. Narison, Phys. Lett. B 351 (1995) 369.
[15] S. Bethke, Nucl. Phys. Proc. Suppl. B, C 39 (1995);
S. Bethke, Nucl. Phys. Proc. Suppl. B, A 54 (1997), hep-ex/0004201;
M. Schmelling, ICHEP96, Varsaw, 1996;
I. Hinchliffe, A. Manohar, hep-ph/0004186.
[16] For a review, see, e.g., S. Narison, QCD 99 Euroconference (Montpellier), Nucl. Phys. Proc.
Suppl. B 86 (2000) 242, and original references therein.
[17] Rome Collaboration, A. Donini et al., hep-lat/9910017;
L. Giusti, QCD99 Euroconference (Montpellier), Nucl. Phys. Proc. Suppl. B 86 (2000) 299.
[18] For a review, see, e.g., R. Gupta, Nucl. Phys. Proc. Suppl. B 63 (1998) 278.
[19] D. Pekurovsky, G. Kilcup, hep-lat/9812019.
[20] T. Hambye et al., Phys. Rev. D 58 (1998) 014017, hep-ph/9906434;
T. Hambye, P. Soldan, hep-ph/9908232.
[21] J. Bijnens, J. Prades, JHEP 01 (1999) 23;
J. Prades, QCD 00 Euroconference (Montpellier), and private communication.
[22] T. Morozumi, C.S. Lim, A.I. Sanda, Phys. Rev. Lett. 65 (1990) 404;
T. Morozumi, A.I. Sanda, A. Soni, Phys. Rev. D 46 (1992) 2240;
Y.Y. Keum, U. Nierste, A.I. Sanda, Phys. Lett. B 457 (1999) 157;
E.P. Shabalin, Sov. J. Nucl. Phys. 48 (1) (1988) 172.
[23] M. Harada et al., hep-ph/9910201;
Y. Kiyo, T. Morozumi, N. Yamada, private communications.
[24] M. Knecht, S. Peris, E. de Rafael, Phys. Lett. B 457 (1999) 227;
M. Knecht, S. Peris, E. de Rafael, QCD 99 Euroconference (Montpellier), Nucl. Phys. Proc.
Suppl. B 86 (2000) 279, and references therein;
S. Peris, E. de Rafael, private communications.
[25] J.F. Donoghue, E. Golowich, hep-ph/9911309.
[26] E. Pallante, A. Pich, hep-ph/9911233.
[27] A.J. Buras et al., hep-ph/0002116;
A. Pich, private communication;
G. Mennessier, private communication.
[28] T. Das, V.S. Mathur, S. Okubo, Phys. Rev. Lett. 19 (1967) 859;
T. Das et al., Phys. Rev. Lett. 18 (1967) 759.
[29] S. Weinberg, Phys. Rev. Lett. 18 (1967) 507.
[30] ALEPH Collaboration, R. Barate et al., Eur. Phys. J. C 4 (1998) 409.
[31] OPAL Collaboration, K. Ackerstaff et al., Eur. Phys. J. C 7 (1999) 571.

30

S. Narison / Nuclear Physics B 593 (2001) 330

[32] E.G. Floratos, S. Narison, E. de Rafael, Nucl. Phys. B 155 (1979) 115;
S. Narison, E. de Rafael, Nucl. Phys. B 169 (1979) 253.
[33] S. Narison, Z. Phys. C 14 (1982) 263.
[34] M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys. B 147 (1979) 385448.
[35] E. Braaten, S. Narison, A. Pich, Nucl. Phys. B 373 (1992) 581.
[36] For a review, see, e.g.: S. Narison, in: Lecture Notes in Physics, Vol. 26, World Scientific, 1989.
[37] M. Davier et al., hep-ph/9802447.
[38] R.D. Peccei, J. Sol, Nucl. Phys. B 281 (1987) 1.
[39] E. de Rafael, QCD 00 Euroconference (Montpellier).
[40] An analogous remark has been mentioned by Y. Okada.
[41] W.J. Marciano, A. Sirlin, Phys. Rev. Lett. 56 (1986) 22.
[42] For a review, see, e.g.: V.A. Zakharov, Nucl. Phys. Proc. Suppl. B 74 (1999) 392;
F.V. Gubarev, M.I. Polikarpov, V.I. Zakharov, hep-th/9812030.
[43] K. Chetyrkin, S. Narison, V.I. Zakharov, Nucl. Phys. B 550 (1999) 353.
[44] J.F. Donoghue, QCD 00 Euroconference (Montpellier);
V. Cirigliano, J.F. Donoghue, E. Golowich, hep-ph/0007196.
[45] K. Chetyrkin, private communication.
[46] S. Narison, Plenary talk given at Hadron 99, Nucl. Phys. A 675 (2000) 54c;
S. Narison, Nucl. Phys. B 509 (1998) 312;
S. Narison, Plenary talk given at QCD 97 Euroconference (Montpellier), Nucl. Phys. Proc.
Suppl. B 64 (1998) 210.
[47] S. Narison et al., Nucl. Phys. B 212 (1983) 365.
[48] A. Bramon, S. Narison, Mod. Phys. Lett. A 4 (1989) 1113.
[49] See, e.g., G. Ecker et al., Nucl. Phys. B 321 (1989) 311.
[50] L. Montanet, QCD 99 Euroconference (Montpellier), Nucl. Phys. Proc. Suppl. B 86 (2000) 381.
[51] S. Narison, G. Veneziano, Int. J. Mod. Phys. A 4 (1989) 2751.
[52] G. Mennessier, S. Narison, N. Paver, Phys. Lett. B 158 (1985) 153.
[53] V.A. Novikov et al., Nucl. Phys. B 191 (1981) 301.
[54] S. Narison, Z. Phys. C 28 (1985) 591.
[55] R. Tarrach, Nucl. Phys. B 196 (1982) 45.
[56] G. Mennessier, Z. Phys. C 16 (1983) 241.
[57] M.B. Gavela et al., Phys. Lett. B 148 (1984) 225, and references therein;
J.F. Donoghue, Phys. Rev. D 30 (1984) 1499;
M.D. Scadron, Eur. Phys. J. C 6 (1999) 141;
E.B. Shabalin, Sov. J. Nucl. Phys. 48 (5) (1988) 864.
[58] S. Narison, Nucl. Phys. Proc. Suppl. B 74 (1999) 304;
S. Narison, Phys. Lett. B 341 (1994) 73.
[59] We thank John Donoghue and the referee for mentioning this important point.
[60] G. Martinelli, private communication.
[61] A. Andrianov, D. Espriu, R. Tarrach, QCD 99 Euroconference (Montpellier), Nucl. Phys. Proc.
Suppl. B 86 (2000) 275;
M. Nagy, M.K. Volkov, V.L. Yudichev, hep-ph/0003200.
[62] M.B. Gavela et al., Phys. Lett. B 211 (1988) 139.

Nuclear Physics B 593 (2001) 3175

www.elsevier.nl/locate/npe

BPS solutions and new phases of finite-temperature

strings
I. Bakas a , A. Bilal b , J.-P. Derendinger b, , K. Sfetsos b
a Department of Physics, University of Patras GR-26500 Patras, Greece
b Institut de Physique, Universit de Neuchtel rue Breguet 1, CH-2000 Neuchtel, Switzerland

Received 17 July 2000; accepted 18 October 2000

Abstract
All high-temperature phases of the known N = 4 superstrings in five dimensions can be described
by the universal thermal potential of an effective four-dimensional supergravity. This theory, in
addition to three moduli s, t, u, contains non-trivial winding modes that become massless in certain
regions of the thermal moduli space, triggering the instabilities at the Hagedorn temperature. In
this context, we look for exact domain wall solutions of first order BPS equations. These solutions
preserve half of the supersymmetries, in contrast to the usual finite-temperature weak-coupling
approximation, and as such may constitute a new phase of finite-temperature superstrings. We present
exact solutions for the type IIA and type IIB theories and for a self-dual hybrid type II theory.
While for the heterotic case the general solution cannot be given in closed form, we still present
a complete picture and a detailed analysis of the behaviour around the weak and strong coupling
limits and around certain critical points. In all cases these BPS solutions have no instabilities at any
temperature. Finally, we address the physical meaning of the resulting geometries within the contexts
of supergravity and string theory. 2001 Elsevier Science B.V. All rights reserved.

1. Introduction
In the conventional description of string theory, where one has only very little
information about the dynamics of all possible string states, the main framework is
provided by an effective supergravity for the massless fields which is obtained by
integrating out the massive modes. As a result, many stringy effects that are attributed to
the massive modes cannot be addressed in a systematic way; we can only appreciate their
relevance in certain regions of the moduli space where massive states can become massless.
In those cases, the conventional supergravity approach has to be enlarged to include the
* Corresponding author.

E-mail addresses: [email protected], [email protected] (I. Bakas), adel.bilal (A. Bilal),

[email protected] (J.-P. Derendinger), [email protected] (K. Sfetsos).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 2 8 - 3

32

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

massless as well as all the relevant would-be massless states on an equal footing, for
otherwise the effective theory will break down due to the appearance of singularities.
Put differently, the possibility to have extra massless states in string theory signals the
limitations of the perturbative field theory description of string dynamics in all corners
of the moduli space. Conversely, including the would-be massless fields into the effective
field theory can teach us important non-perturbative lessons about string theory.
There are specific interesting problems which can be addressed systematically, based
on supersymmetry, by isolating the dynamics of a few relevant modes that can become
massless. An important example of this kind is provided by the conifold singularity
in field theory and its string-theoretical resolution [1,2]. In particular, in Calabi
Yau compactifications which admit a non-trivial 3-cycle, the moduli space develops a
singularity that invalidates the applicability of the conventional low-energy effective theory
when the cycle shrinks to zero size. Incorporating a D-brane that wraps the 3-cycle,
with mass proportional to its period, resolves the problem of conifold singularities as the
enlarged theory is appropriate for describing cycles of all sizes.
Superstrings at finite temperature provide another interesting example that has been
studied on and off for quite some time, but admittedly has not been fully explored yet.
In particular, it is known that string theory exhibits an exponential growth of the number
of states at high energy, which gives rise to a limiting temperature, known as the Hagedorn
temperature TH [311]. From the world-sheet point of view, where one uses a periodic
Gaussian model to describe the propagation of strings on tori, there is a Kosterlitz
Thouless phase transition at a critical radius where vortices can condense [12,13], thus
leading to a limiting temperature for string thermodynamics (see also [811]) associated
with a phase transition. It has been further realized that there are string states with nontrivial winding number that can become light close to this temperature, and then turn
tachyonic beyond it, thus signaling thermal instabilities of string theory at very high
temperatures [14]. Some implications of this effect have been examined in the context
of strings in the very early universe, where one tries to find ways that avoid the initial
singularity, and at the same time explain why the dimensionality of the physical space
time is four [15] (see also [16] and [17] for a more recent exposition that takes into account
D-branes). In an recent development, the contribution of the relevant winding modes was
explicitly described in terms of an effective four-dimensional supergravity using a universal
thermal potential that incorporates all phases of N = 4 superstrings, taking into account the
thermal deformation of the BPS mass formula in N = 4 supersymmetric theories [18,19].
However, no solutions have been found so far with a concrete physical interpretation and
the ability to resolve the problems of quantum cosmology or the subtle issues raised at the
end point of the black-hole evaporation; these are situations where the temperature grows
to infinity and one encounters singularities in the context of any perturbative field theory.
The present paper grew out of the attempt to construct explicit solutions of the effective
theory of supergravity which was proposed to include the dynamics of the lightest relevant
winding modes that could become massless at the Hagedorn temperature. We construct
for the first time families of non-trivial solutions with varying winding fields in the type II
and the heterotic sectors of string theory. We focus attention primarily on domain wall

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

33

solutions, one of the reasons being that they are technically the simplest ones to consider;
all the fields in the bosonic sector of the theory, namely the metric and the collection of
scalar fields, are taken to depend on a single spatial coordinate only. Ultimately, of course,
it is necessary to search for other types of solutions, for instance solutions with spherical
symmetry, in order to discuss models appropriate for quantum cosmology or black-hole
physics. We hope to report on such generalized solutions of the effective supergravity
elsewhere in the future.
There are also some other compelling reasons for being interested in the existence
and the explicit construction of domain wall solutions of the effective supergravity that
describes strings at finite temperature. The first is provided by the important rle of domain
walls, as spacetime defects, in a variety of cosmological applications, though we do
not focus here on these types of physical problems. The second is provided by the fact
that the domain walls, by their nature, break the four-dimensional Poincar invariance to
an effective three-dimensional invariance. Then, one can have a working framework for
studying the issue of supersymmetry breaking by non-perturbative effects, as it was done
in [18,19], taking into account the peculiarities of supersymmetry in three dimensions,
namely, that even if local supersymmetry is unbroken the massive multiplets will not
necessarily exhibit mass degeneracy [20].
The domain wall ansatz allows us to consider solutions of certain first order differential
equations derived from a prepotential. As we explicitly show, these solutions by
construction preserve half of the supersymmetries, and in this sense they are BPS. The
full set of equations allows consistent truncations to subsectors which correspond to the
heterotic string, type IIA or type IIB, or the type II string at the self-dual radius (which
we call the hybrid type II). For all type II cases we can find the exact general solutions in
closed form. For the heterotic sector, however, no solution in terms of known functions is
available, and we instead give a detailed analysis of the solutions around the weak and the
various strong coupling limits, as well as around certain critical points, which still allows
us to obtain a reasonably complete overall picture.
It should be stressed that in the standard finite-temperature treatment of perturbative
superstrings all supersymmetries are broken due to the different boundary conditions on
bosons and fermions one has to impose in the periodic imaginary time direction. For this
reason, it is not surprising that some modes become tachyonic beyond certain temperatures.
On the contrary, the solutions we find within the domain wall ansatz of the effective
supergravity are non-perturbative, containing regions of strong coupling, and preserve half
of the supersymmetries. As such they are expected to be stable solutions, and indeed we
show that, although the temperature can be arbitrarily high, no tachyonic instabilities ever
develop. Even though we are only considering solutions of an effective supergravity, they
may well point to a new finite-temperature phase of superstrings which is BPS and has no
thermal instabilities, i.e., no Hagedorn temperature.
To further probe the physical meaning of these domain wall solutions we first study the
propagation of a quantum test particle in the corresponding supergravity backgrounds. This
allows to discriminate between wave-regular and non-regular geometries, thus narrowing
down the physically meaningful set of solutions. The analogous question could be

34

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

addressed within string theory. The consistency of string thermodynamics seems to require
the compactness of all spatial directions since otherwise one would encounter problems
with negative specific heat (see, for instance, [15] for a delicate analysis, without making
reference to the geometrical details of the space, and references therein). It is clear from
our explicit solutions in the type II cases, and also true for the heterotic case (with
the possible exception of certain solutions that are never weakly coupled), that none of
them is periodic in the spatial dimension that defines the domain walls. However, this
does not mean that our solutions of the effective supergravity theory are insufficient to
extrapolate to string theory. Indeed, the compactness criterion is established within the
micro-canonical treatment of string thermodynamics, and this assumes that the theory
remains weakly coupled throughout all space. Furthermore, it concerns the phase where
supersymmetry is completely broken by the finite temperature. On the other hand, almost
all our solutions always contain a region of strong coupling. Most important, our solutions
are BPS, preserving half of the supersymmetries. These differences are enough to cast
serious doubt on whether the arguments about compactness of all spatial dimensions should
apply. Furthermore, as already noted, the fields corresponding to the winding modes of our
solutions never become massless even if the temperature modulus approaches the wouldbe Hagedorn temperature, which is yet another way to see that these solutions belong to a
different phase.
This paper is organized as follows: In Section 2, we review the main aspects of the
effective theory of supergravity that was proposed to describe the thermal phases of all
N = 4 superstrings and present the form of the universal thermal potential for the six
scalar fields that correspond to the s, t, u moduli and the winding states z1 , z2 , z3 that can
become massless in the heterotic, type IIA and type IIB sectors. In Section 3, we consider
the domain wall ansatz for the metric and all the relevant scalar fields. We first discuss quite
generally when a D = 4, N = 1 supergravity theory admits domain wall solutions that
are derived from first order differential equations. Indeed, this follows from some simple
reality assumptions about the scalar fields and the Khler potential. We show that these
solutions automatically preserve half (or all) of the supersymmetries and hence are BPS.
Then we specialise to the effective thermal supergravity and explicitly derive the coupled
system of six first order non-linear differential equations for them. Consistent truncations
of this system lead to various type II or heterotic sectors. In Section 4, this system is
analysed explicitly in the type IIA, type IIB as well as in the hybrid type II sectors where
the general solutions can be derived in closed form. In Section 5, we study the equations
for the heterotic sector which cannot be solved analytically, apart from a very special
solution. We are able to study the behaviour of the solutions in the vicinity of the weak and
the strong coupling regions, as well as around certain critical points, and present a fairly
complete general picture. The construction of domain wall solutions for the whole system
of six equations is beyond the scope of the present paper and can probably only be done
numerically. In Section 6, we study the rle of boundary conditions in selecting physical
solutions in the effective theory of supergravity and comment on string thermodynamics.
Finally, in Section 7, we present our conclusions and outline some directions for future
work. In Appendix A, we address the issue of having periodic solutions using criteria from

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

35

the general theory of dynamical systems. This may turn out to be useful when studying the
complete system of six equations.

2. Thermal potential of effective supergravity

In this section we present the essential ingredients for constructing an effective
N = 1 supergravity in four dimensions 1 that describes the thermal phases of all N =
4 superstrings in a universal way, following earlier work [18,19,21]. The construction
takes into account the dynamics of the would-be tachyonic winding modes of the N =
4 superstrings which are responsible for inducing a phase transition at high temperatures
in string thermodynamics. As such, the effective N = 1 supergravity theory provides a
systematic framework for quantifying arguments about the phase transition occurring at
the Hagedorn temperature due to the special form of the effective potential of the relevant
winding modes , which includes a trilinear coupling with the modulus describing
the temperature [14].
The starting point of the construction is provided by five-dimensional N = 4 theories
that are effectively four-dimensional at finite temperature. The crucial observation is that
d-dimensional superstrings at finite temperature look like (d 1)-dimensional strings with
spontaneously broken supersymmetry. Concretely, putting strings at finite temperature T
amounts to compactifying the (Euclidean) time on a circle of inverse radius R 1 = 2T .
As the Euclidean time direction is compact, one imposes boundary conditions to take
particle statistics into account; modular invariance of the thermal partition function dictates
then specific phase factors in the related GSO projection [14]. Technically, this procedure
is equivalent to a ScherkSchwarz compactification from d = 5 to d 1 = 4 with a
well-defined gauging associated to the temperature modification of the effective theory of
gauged supergravity [18,22]. This gauging of N = 4 spontaneously breaks supersymmetry.
2.1. Thermal dyonic modes
We restrict ourselves to the study of N = 4 strings, because in this case the
supersymmetry algebra and its central extensions suffice to determine the masses of all
BPS states. Recall that the states which become tachyonic above a certain Hagedorn
temperature TH are 1/2-BPS states, at least in dimensions where there are no other BPS
states with smaller fractions of supersymmetry. Then, in N = 4 strings, one can identify
completely all the perturbative and the non-perturbative BPS states that can induce thermal
instabilities and construct from first principles an exact effective supergravity for these
states, as it was done in Refs. [18,19]. It is useful to observe that the odd dyonic modes
exhibit the same finite-temperature behaviour as the odd winding string states. This follows
from the action of dualities on the N = 4 strings, relating the dyonic modes to perturbative
winding states of dual strings. These windings in turn induce thermal instabilities in the
1 We always count four-dimensional supersymmetries, i.e., N = 1 supersymmetry has four supercharges.

36

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

dual strings. Describing the complete set of thermal phases of N = 4 theories requires then
both perturbative and non-perturbative states from all string viewpoints.
We define the context of the present work, following [18,19], by first considering string
theories in six dimensions obtained by compactification on T 4 (heterotic string) or K3
(type II strings). We further compactify one dimension S 1 , with radius R6 , and so the
resulting five-dimensional theory exhibits T-duality between type IIA and type IIB (under
0 /R ) and S-dualities between heterotic and type II strings. Moreover, since we
R6 het
6
are putting strings at finite temperature T , the time dimension is taken Euclidean and
compactified on S 1 with radius R = 1/(2T ) with twisted fermionic boundary conditions
that account for the thermal effects.
In order to obtain the resulting four-dimensional thermal mass formula and examine
which states can become tachyonic above a certain temperature, thus inducing instabilities,
we first consider the usual BPS mass formula in N = 4 supersymmetric theories [2327]
written in heterotic variables:

2
1 
+ ntu
i(m
0 u + n 0 t)  .
(2.1)
M2 = 0
m + ntu + i(m0 u + n0 t) + is m
het tu
Here s, t and u are defined in terms of the compactification radii R6 , R and the heterotic
string coupling ghet as follows,
s=

1
,
2
ghet

t=

RR6
0 ,
het

u=

R
,
R6

(2.2)

0 = 4s, in units where the four-dimensional gravitational

supplemented by the relation het

coupling has been normalized to 2. These expressions will help us later to understand
the physical meaning of the various domain wall solutions in terms of the three physical
parameters in the problem R6 , R and ghet . The integers m, n, m0 , n0 are the electric
momentum and winding quantum numbers associated to the compactification on the 2torus with radii R6 and R. The tilded integers are the corresponding magnetic nonperturbative counterparts. Then, string dualities are simply described by the interchanges
s t (heteroticIIA), s u (heteroticIIB) and t u (IIAIIB), which leave invariant
the mass formula and the temperature radius R 2 = 4stu which is common to all three string
theories when measured in Planck units.
As discussed in Refs. [18,19], the thermal deformation of the mass formula (2.1) is very
simple: the momentum quantum number m is replaced by m + Q0 + n/2, Q0 being the
helicity charge. It also reverses the GSO projection, to account for the modified boundary
conditions in the temperature deformation of the theory. For our present purposes, we may
= n = 0, since the states that can
restrict our attention to only the states having m0 = n0 = m
become tachyonic first as the temperature increases are contained within this subset [19].
The thermal spectrum of these light states is then given by

 
2
1
1
2
0
m + Q + kp + k Tp,q,r R 2Tp,q,r |k|,1 Q0 ,0 ,
(2.3)
MT =
R
2

where we have defined the integers k, p, q and r by the condition (n, m

0 , n 0 ) = k(p, q, r),
p, q and r being relatively prime numbers, and Tp,q,r denotes the effective string tension

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

Tp,q,r =

p
q
r
+ 0 ,
0 + 0
het

IIB
IIA

37

(2.4)

which can be written in terms of s, t and u using the identifications

0
= 4s,
het

0
IIA
= 4t,

0
IIB
= 4u,

(2.5)

again in the Planck units with normalization = 2. The mass formula (2.3) is an
extension of the known perturbative finite-temperature mass formulas with the correct
duality and zero temperature behaviour. In particular, the last term is generated in
perturbative strings by the GSO projection related to the breaking of supersymmetry.
The various constraints imposed among the quantum numbers require that p, q, r are
all positive and that mk > 1.
Analysing the four-dimensional mass formula for M2T , we note that tachyons can appear
0 , n 0 . The
when Q0 = 0 and |k| = 1 for certain values of the four quantum numbers m, n, m
critical temperature (and hence the critical radius R) can be found by locating the zeros of
M2T . The first tachyons and the critical temperatures are:
s
0

het
,
(2.6)
(m, n, m
0 , n 0 ) = (1, 1, 0, 0) with R = 2 + 1
2
q
0 ,
(2.7)
(m, n, m
0 , n 0 ) = (0, 0, 1, 0) with R = 2IIA
q
0 .
(2.8)
(m, n, m
0 , n 0 ) = (0, 0, 0, 1) with R = 2IIB
Since the winding numbers in the heterotic, the type IIA and the type IIB strings are
respectively n, m
0 and n 0 , these three pairs of states have winding numbers 1, as expected
on general grounds, in their respective perturbative superstring theory. There are also two
other series of states occurring for Q0 = 0 and |k| = 1, with quantum numbers m = 1 and
(p, q, r) arbitrary, which can become tachyonic when p = 1 or p = 2; however, in either
series the critical temperatures that result from M2T = 0 are higher than at least one of
the perturbative Hagedorn temperatures of the heterotic, type IIA or type IIB strings given
above.
This analysis of the mass formula suggests the field content to be used in the effective
field theory description. We certainly need the three moduli s, t and u and the six scalar
fields (actually three conjugate pairs of fields) able to generate the thermal transitions.
These states will be embedded in the generic scalar manifold of N = 4 supergravity,
SO(6, m)
SL(2, R)

,
U (1)
SO(6) SO(m)
where m is the number of vector multiplets. To study the thermal phase structure however,
it is consistent and sufficient to truncate this N = 4 theory to N = 1 and to retain only chiral
N = 1 multiplets describing the relevant moduli and winding states. After truncation, the
effective theory includes three complex moduli scalars S, T and U , with scalar manifold






SL(2, R)
SL(2, R)
SL(2, R)

,
(2.9)
U (1)
U (1)
U (1)
S
T
U

38

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

and six complex windings ZA

, A = 1, 2, 3, living on




SO(2, 3)
SO(2, 3)

.
SO(2) SO(3) Z +
SO(2) SO(3) Z
A

(2.10)

The resulting scalar manifold is a Khler manifold for chiral multiplets coupled to N = 1
supergravity that arises by two successive Z2 projections applied on the N = 4 scalar
manifold.
2.2. The effective supergravity
The construction of the effective N = 1 supergravity Lagrangian proceeds as follows
[19]. A generic four-dimensional N = 1 supergravity theory is characterized by a Khler
potential K and a holomorphic superpotential W . The bosonic sector of the theory is given
by the following Lagrangian density:






1
1
SJ V ,
S ,
(2.11)
e1 L = R KI J I
4
2
SJ are the components of the Khler metric for the collection
where KI J = 2 K/ I
S
of complex scalar fields { I } present in chiral multiplets. The scalar potential V (, )
assumes the special form


1 K I J
S 3W W
S ,
e K W;I W
;J
4
using the notation of covariant derivatives
V=

(2.12)

W
K
+
W.
(2.13)
I

I
The form of the Khler potential K follows from the constraints defining the N = 4
scalar manifold, truncated to N = 1. One finds, in particular,
W;I =

with

S ) ln Y+ ln Y ,
K = ln(S + S
S ) ln(T + TS ) ln(U + U

(2.14)

S

S S

ZA ZB ZB ,
ZA + ZA
Y = 1 2ZA

(2.15)

where summation over repeated indices A or B = 1, 2, 3 is implicitly assumed. The

expression of the superpotential W follows from the gauging applied to the N = 4 theory.
This gauging encodes the ScherkSchwarz supersymmetry breaking mechanism generated
by the introduction of a finite temperature [18,19] and yields

1


+ +

1 ZA
ZA 1 ZB ZB + (T U 1)Z1+ Z1
W =2 2
2

+
+
+ SU Z2 Z2 + ST Z3 Z3 .
(2.16)
The scalar potential of the theory defined by K and W is complicated, but focusing

= Im S = Im T = Im U = Re (ZA
ZA
)=0
on the real directions defined by Im ZA

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

39

leads to important simplifications. Furthermore, a study of the mass spectrum in the lowtemperature limit shows that these directions correspond precisely to the possible phase
transitions occurring in the theory. We are then led to consider only the relevant would-be
tachyonic states using the field variables
s = Re S,

t = Re T ,

= Re ZA
.
zA = Re ZA

u = Re U,

(2.17)

Let us introduce for later convenience the quantities

3
X

x2 =

2
zA
,

HA =

A=1

zA
.
1 x2

(2.18)

Then, in this case, the components of the Khler metric in the field space become
KS S =

1
,
4s 2

KA B =

KT T =

1
,
4t 2

KU U =

1
,
4u2

2
B ,
(1 x 2 )2 A

(2.19)

; in writing the kinetic terms for the

i.e., the metric becomes diagonal in the directions ZA
B
2
2
fields zA we have to take KAB = 4A /(1 x ) , as there is a factor of 2 picked up by
+

= ZA
in this case.
ZA
In order to present the Lagrangian density for all remaining six scalar fields, we find it
helpful to trade s, t and u for 1 , 2 and 3 as follows:

s = e21 ,

t = e22 ,

u = e23 .

(2.20)

Note that 1 + corresponds to the strong coupling limit s 0, whereas the weak
coupling limit is attained for 1 . With these definitions, we obtain a simplified
form of the effective supergravity with bosonic Lagrangian density
!
3
3
X
X
1
1
4
( i )2 +
( zA )2 V ,
(2.21)
e1 L = R
4
2
(1 x 2 )2
i=1

A=1

where the scalar potential is now given by the sum

V = V1 + V2 + V3 .

(2.22)

Each term VA is a function of the moduli i and a polynomial in HA . Explicitly, we have

1 21 2
e H1 cosh(22 + 23 ) 4H12 + 1 3 ,
(2.23)
2




1
(2.24)
V2 = e22 H22 e21 23 4H22 + 1 4 ,
4




1
(2.25)
V3 = e23 H32 e21 22 4H32 + 1 4 .
4
Our normalizations have been chosen so that the kinetic terms of the fields i assume
their
canonical form, with an overall factor 1/2 as coefficient, since the normalization =
2 has been used. Then, the scalar potential V provides the universal thermal effective
potential that describes all possible high-temperature instabilities of N = 4 superstrings.
V1 =

40

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

At this point, it is very useful to observe that the scalar potential (2.22) for the six real
fields s, t, u and zA can also be cast into the special form
V=

 3
 e 2

3
e 2 1 X

2 W
1 X W
e2 .
+
3W
1 x2
4
i
4
zA
i=1

(2.26)

A=1

This property only depends on the structure of the Khler potential K. The prepotential
e given by
W


e = 1 e1 +2 +3 2e1 sinh(2 + 3 )H12 + e1 e2 3 H22 + e3 2 H32
W
2


1
1
2
2
2
+ (tu 1)H1 + suH2 + stH3 .
(2.27)
=
stu 2
The second expression indicates that the prepotential is actually given by

e = eK/2 W 
,
W
real directions

(2.28)

as can be inferred from the general form of the supergravity potential and from our specific
Khler potential 2 (2.14) and superpotential (2.16). As we will see in Section 3.2, this is
not an accident. The prepotential will be of particular importance when writing the system
of first order differential equations for the domain wall configurations of the theory. It
should be emphasized that given the potential (2.22), we may view (2.26) as a differential
e . Hence, in general, there can be solutions other than (2.27).
equation for the prepotential W
However, the latter choice is the one that leads to one-half supersymmetric solutions, as
e satisfying (2.26) would break
we will see in Section 3.3. In general, another choice for W
supersymmetry completely.
2.3. Thermal phases of N = 4 strings
The thermal phase structure of N = 4 superstrings can be deduced from the form of
the potential V . First of all note that the low-temperature phase, i.e., large R, or large
stu, or large and negative 1 + 2 + 3 , is common to all strings. It is characterized by
the condition H1 = H2 = H3 = 0, in which case V1 = V2 = V3 = 0, and so the potential
vanishes for all values of the moduli fields s, t and u. Since the four-dimensional couplings
of the three strings are 3

2
2 RR6
2 R
s= 2 ,
t= 2 = 2 0 ,
u= 2 = 2
,
(2.29)
het
R6
ghet
gIIA
gIIB
we see that this phase exists in the perturbative regime of each string theory. Notice also
that the configuration HA = 0 and i constant (but arbitrary) is an exact solution of the

2 In particular, exp(K/2) = 2 2stu(1 x 2 )2 upon restriction to the real directions of the theory.

3 Notice that there is a factor of

2 that rescales the right hand sides of (2.29) as compared with the

corresponding equation for zero temperature (2.2). This is due to the new spin structure introduced by
temperature. However, Eq. (2.5) for the three string scales remains the same.

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

41

theory, with all supersymmetries broken by the thermal deformation. The reason is that in
this phase the supersymmetry transformation is proportional to
K
1
,
eK/2 W;I =
2 stu I

(2.30)

which does not vanish in the moduli directions, except of course in the limit stu
(zero-temperature limit).
There are three high-temperature phases generated by each of the three contributions to
the scalar potential (2.22), and which are characterized by a non-zero value of a winding
state zA . Which phase is selected depends on the values of the moduli i , corresponding
to the coupling, the temperature radius R and the sixth-dimension radius R6 in each string
0 , 0
0
theory (with string units het
IIA or IIB ). None of them is a solution to the theory with
constant scalars. The detailed analysis of the thermal phase structure [19] leads to the
following classification.
2.3.1. Heterotic high-T phase
0 T 2 ), while it is
In heterotic-string units, the temperature modulus is tu = 1/(2 2 het
2
stu = 1/(2T ) in Planck units. A heterotic tachyon H1 appears first if

2

2
2 1 < tu < 2 + 1 ,
su > 4,
st > 4,
(2.31)
where the last two condition guarantee the absence of type II tachyons. Then, this implies
for the four-dimensional heterotic coupling that

1
2+1
2
2
(2.32)
2 s = ghet < gcrit = .
2 2
The upper and lower critical temperatures in (2.31) are related by T-duality in the
temperature radius. Within the interval (2.31), the theory has a (non-tachyonic) solution
for H1 = 1/2, tu = 1. And, since the scalar potential reduces to
1
(2.33)
V = e21 ,
8
a linear dilaton background is a solution. This background breaks one half of the
supersymmetries [19]. This particular solution will be rederived as a special case of a
more general class of domain wall backgrounds that break half of the supersymmetry in
the heterotic sector, in Section 5.
In the strong coupling regime of the heterotic theory, where s 0, there are only type II
instabilities; the high-temperature heterotic phase cannot be reached for any value of the
2 > g 2 . Put differently, the high-temperature heterotic phase can only exist
radius R6 if ghet
crit
in the perturbative heterotic regime, which by the heterotictype II duality corresponds to
the non-perturbative regime of the type II string theory.
2.3.2. Type-IIA/B high-T phases
0 T 2 ) for type IIA or st =
The relevant temperature modulus is either su = 1/(2 2 IIA
2
0
2
1/(2 IIB T ) for type IIB strings. The high-temperature phase for type II strings is then
defined by su < 4 (for IIA) or st < 4 (for IIB). Thus, type II instabilities arise in the strong

42

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

coupling
q region of the heterotic string. The analysis of the problem shows that the value
0 determines whether a type-IIA or a IIB tachyon comes first as the temperature
R6 / het
increases. We have in particular the following two high-temperature type II phases:

q
2+1
1 1
2
0 ,
, ghet > , R6 < het
(2.34)
IIA: 2T > 2
2ghet R6
2 2

q
2+1
1 R6
2
0 .
(2.35)
IIB: 2T > 2 0 , ghet > , R6 > het
2ghet het
2 2
Note that the type II high-temperature phases differ from the heterotic one by the absence
of an exact solution with constant windings and/or moduli. In these phases, the theory
depends on a non-zero, field-dependent winding and two non-trivial moduli, t and su in
the IIA case, u and st for IIB strings. This concludes our general review of the thermal
aspects of N = 4 strings at finite temperature.

3. BPS domain wall solutions

We are now in the position to look for solutions of the effective supergravity that describe
the thermal phases of N = 4 superstrings. We will use the domain wall ansatz for two
reasons. First, this probably is the simplest possible class of solutions one may construct in
supergravity theories whose bosonic sector is described by gravity coupled to a selection
of scalar fields with non-trivial potential terms, as the universal thermal effective potential
we have at our disposal. Second, with all fields depending only on a single variable one
can look for solutions that satisfy more stringent, but easier to solve, first order differential
equations rather than second order ones. We will show quite generally that all solutions
of these first order equations preserve (at least) half of the supersymmetries and hence are
BPS.
We will first discuss in a general setting when gravity coupled to scalars admits solutions
of first order equations. Then we will see under which circumstances this is true for the
bosonic sector of supergravity, and that in these cases the solutions are BPS. Finally, we
specialise to the case of present interest and derive the set of coupled first order equations
of the effective supergravity describing the thermal phases of N = 4 superstrings.
3.1. First order equations for gravity coupled to scalars
Consider the following D-dimensional Lagrangian density representing gravity coupled
to N scalars
1
1
1
(3.1)
L = R GI J ()g mn m I n J V (),
g
4
2
where summation over the repeated indices I, J = 1, 2, . . . , N and m, n = 1, 2, . . . , D is
implied. The equations of motion that follow from varying the metric and the scalar fields
are:

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

g :
:

1
1
1
Rmn GI J m I n J =
gmn V ,
4
2
D2
D 2 I + JIK g mn m J n K = GI J J V ,

43

(3.2)
(3.3)

where JIK is the Christoffel symbol formed using the scalar-field space metric GI J . Let
e () as:
us assume that the potential V () is given in terms of a prepotential W


2
e2 ,
e J W
e 2D 1W
GI J I W
(3.4)
V=
8
D2
where is a dimensionful constant.
Let us make the ansatz for a metric of the form
ds 2 = dr 2 + e2A(r) dx dx ,

, = 1, 2, . . . , D 1,

(3.5)

that preserves the flat spacetime symmetries in D 1 dimensions. We also assume that
all scalars depend only on the variable r, i.e., I = I (r). Then, it can be shown by a
straightforward calculation that if the following first order equations hold

d I
e,
= GI J J W
dr
2

e
dA
=
W,
dr
D2

(3.6)

the second order equations are also satisfied. 4 In proving this statement we use the fact
that the non-vanishing components of the Riemann tensor for the metric (3.5) are


R = e2A A00 + (D 1)(A0 )2 ,


(3.7)
Rrr = (D 1) A00 + (A0 )2 ,
where the prime denotes the derivative with respect to r. Note that the freedom in the choice
e . Equivalently
of sign in (3.6) is due to the fact that (3.4) does not determine the sign of W
it expresses the freedom to change r r.

Throughout the remainder

of this paper we restrict to D = 4 and choose = 2

(consistent with = 2 ).
3.2. First order equations for the bosonic sector of supergravity
Consider now the bosonic sector of a generic four-dimensional N = 1 supergravity
theory as given by Eqs. (2.11), (2.12) and (2.13). Assume that all Im( I ) vanish and that
the Khler potential is such that



K 
=
K real directions .
(3.8)
2


Re
I real directions

This is indeed the case for our thermal supergravity, but for the time being we do not
want to consider any specific model. The definition of the covariant derivative (2.13) of the
superpotential W and the assumption (3.8) imply that if we define

e = eK/2 W 
,
(3.9)
W
real directions
4 Recently, such first order equations with flat metric in the scalar-field space appeared in the context of
supersymmetric solutions in five-dimensional gauged supergravity (for a first example, see [28]). For cases where
the metric in the scalar-field space is non-trivial, five-dimensional examples have been given in [29].

44

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

we will have
e
(3.10)
W = eK/2 W;I .
I
As a consequence, the scalar potential (2.12) of the supergravity can be rewritten using
e as
only W


e e
1
I J W W
2
e
K
3W .
(3.11)
V=
4
I J

We see that this is precisely of the form (3.4) with D = 4 and = 2.

We conclude that if we find solutions to the domain wall ansatz (3.5) satisfying the first
order equations
e
1
d I
W
= KIJ
,
dr
J
2

1 e
dA
,
= W
dr
2

(3.12)

and take the fermions to vanish, then all equations of motion of the supergravity will be
satisfied. We now proceed to show that these solutions of (3.12) precisely preserve half or
all of the four-dimensional supersymmetries.
3.3. Supersymmetry of the solution
It is very simple to obtain the number of supersymmetries left unbroken by a solution
of the first order equations. The only additional assumptions we need is that the Khler
potential satisfies (3.8), Im( I ) vanishes, and that the fields only depend on a single
coordinate, say r, with g rr = 1.
Consider then the supersymmetry transformation of the (left-handed) fermionic component LI of a chiral multiplet with scalar I :


1
1

(3.13)
LI = I R L eK/2K I J W;J + fermionic terms.
2
2
Since, by assumption, the solution depends only on a single coordinate r with metric g rr =
1, we have I = r r I where r satisfies ( r )2 = 1. Then, using the first order
equation (3.12) (and (3.10)) we have
e
1
1
W

= r eK/2K I J W;J .
I = r K I J
J
2
2
It follows that for vanishing fermions

(3.14)

(3.15)
LI = PL (1 r )eK/2K I J W;J ,
2
where PL is the left-handed chirality projector (of course there is a similar equation for
RI ). Clearly, since ( r )2 = 1, the solution will preserve one half of the supersymmetries if
W;J 6= 0, and all supersymmetries if W;J = 0, as it should.
In order to compute the dependence of the Killing spinor  on the spacetime
coordinates, one in principle has to solve the gravitino equation. Equivalently, for static
configurations (as in our domain wall ansatz (3.17)) one can use an argument based on the

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

45

1/4

supersymmetry algebra and deduce that  = g00 0 , where 0 is a constant spinor subject
to model-dependent projection conditions, which reduce the number of its independent
components [30]. Applying this to our case we find
(r) = eA(r)/20 ,

(1 r )0 = 0.

(3.16)

This result should be interpreted with some care when considering strings at finite
temperature. Formulating a theory at finite temperature requires a rotation to Euclidean
compact time and, with a trivial background, supersymmetry is completely broken. As
already discussed in Section 2.3, this is the case of the low-temperature phase. Our domain
wall solutions are, however, backgrounds depending on a spacelike variable r with a
manifest three-dimensional residual spacetime symmetry, with signature (3, 0) at finite
temperature. The supercharges preserved by these backgrounds are then supersymmetries
inherited from the original five-dimensional supersymmetry algebra, with Minkowski
signature (4, 1).
We conclude that the solutions of the first order equations (3.12) are 1/2 BPS solutions,
e / I = 0, in which case the scalar fields are constant and
except of course if W
supersymmetry is unbroken.
3.4. The first order equations for the thermal supergravity
Next, we specialise the discussion to the supergravity theory that describes the thermal
phases of the N = 4 superstrings, by applying the restrictions and identifications explained
e are given in
in Section 2.2. The Khler metric, the potential V and prepotential W
Eqs. (2.21), (2.22) and (2.27). There are six real scalar fields and the Khler metric is
diagonal Kij = K(i) ij . For convenience we write the relevant equations again, for D = 4.
The metric is assumed to be conformally flat and written in the form (see for instance [31])
ds 2 = dr 2 + e2A(r) dx dx ,

, = 1, 2, 3,

(3.17)

and the BPS domain wall configurations are solutions of the following system of first order
differential equations:
e
1 1 W
di
= K(i)
;
dr
i
2

1 e
dA
= W
,
(3.18)
dr
2
taking into account the inverse of the diagonal metric in field space for each one of the
components i . Note that we have used the freedom r r to make a definite choice of
signs in these equations.
The six scalar fields are naturally separated into two groups: {i ; i = 1, 2, 3}
corresponding to s, t, u and the three winding fields {zi ; i = 1, 2, 3}. For the first set
the metric in field space is flat, i.e., K(i) () = 1, whereas for the second set all components
are equal but non-trivial, namely K(i) (z) = 4/(1 z12 z22 z32 )2 , see Eq. (2.21). The
e was given in (2.27). As a result, the domain walls of the theory correspond
prepotential W
to solutions of the non-linear system
d1


1
= e1 ++ + 2e1 sinh + H12 + e1 e H22 + e H32 ,
2
(3.19)
dr
2

46

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

d2


1
2
= e1 ++ + 2e1 cosh + H12 e1 e H22 e H32 ,
dr
2
d3


1
= e1 ++ + 2e1 cosh + H12 + e1 e H22 e H32 ,
2
dr
2
d ln H1


= e1 sinh + (4H12 + 1) 2e1 e H22 + e H32 ,
2
dr
d ln H2


1
= e1 + + 4e1 sinh + H12 2e1 e H22 + e H32 ,
2
dr
2
d ln H3


1
= e1 + 4e1 sinh + H12 2e1 e H22 + e H32 ,
2
dr
2

(3.20)
(3.21)
(3.22)
(3.23)
(3.24)

where we found it more convenient to work with the fields Hi = zi /(1 z12 z22 z32 )
instead of zi and we defined = 2 3 . As soon as a solution has been found, the
conformal factor of the metric can be obtained by a simple integration of the resulting
e , as it has been prescribed above.
expression for the prepotential W
Although these equations are first order, their explicit solution is still a difficult task
because they are highly non-linear in the general case. We note that there are examples in
the scalar sector of maximal gauged supergravity in 4, 5 and 7 spacetime dimensions
where the resulting system of equations can be linearized by introducing appropriate
combinations of fields plus an algebraic constraint. This led to a systematic classification
of the possible domain wall solutions in terms of Riemann surfaces with varying genus
(for details, see the method presented in [32,33]). However, things do not always work that
way; as it turns out the above system of equations that describe strings at finite temperature
can only be partially studied by similar methods.
It can be easily seen from the last three equations that the fields Hi can be solely
expressed in terms of the fields i , but one should avoid substituting their integral
expressions into the first three equations because it would lead to a complicated system
of integro-differential equations among the fields i alone. In either case, the system of
equations at hand is rather complicated and difficult to solve in general. Instead, we will
focus attention on subsectors obtained by consistent truncations of the field content. Each
one of these consistent truncations results into a system of three first order equations that
correspond to the various type II and heterotic theories. Luckily, it will turn out that the
general solutions for all type II theories can be found explicitly. However, in the heterotic
case a general solution cannot be given in closed form. Hence in this case, apart from
extracting the behaviour of the fields in the strong and weak coupling regions as well as
around certain critical points, one has unavoidably to rely on numerical studies. Combining
these pieces of information, we can nevertheless give a rather complete picture of how the
different solutions behave.

4. Type II string theories

The simplest truncations of the domain wall equations lead to different type II sectors.
We will first examine the type IIA and IIB theories, which can be treated simultaneously

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

47

using the appropriate field identifications, and then examine a hybrid type II sector, which
describes type II strings at the self-dual radius, and turns out to be exactly solvable, too.
4.1. Type IIA and IIB sector
According to the field identifications made in the effective supergravity, the type IIA and
type IIB sectors of the theory are obtained by setting
z1 = z2 = 0,

for type IIB,

(4.1)

z1 = z3 = 0,

for type IIA,

(4.2)

in which case H1 = H2 = 0 and H1 = H3 = 0, respectively. It follows from (3.19) and

(3.20) or (3.21) that 1 equals 2 or 3 , respectively, up to an irrelevant additive constant
which we will ignore. Then, it is convenient to set

1 = 2 = ,
2

3 = ,

z3 = z,

for IIB,

(4.3)

1 = 3 = ,
2

2 = ,

z2 = z,

for IIA,

(4.4)

or

and introduce in either case the field for the winding field
 

,
z = tanh
2

(4.5)

for which we have 2H = sinh . The appropriate string coupling is gII e and the
temperature field is T e 2 , when it is normalized in type II string units that remove the
-dependence. We treat both cases together because the (pre)potential assumes the same
form for IIA and IIB, and the same is true for the truncated system of differential equations
for the domain walls. Hence, no distinction will be made in the following between type IIA
or type IIB. Note also that the kinetic terms for the fields , and all assume their
canonical form.
Explicit calculation shows that in terms of the new variables the prepotential (2.27)
becomes



eII = 1 e e 2 + 1 e 2 sinh2 ,
(4.6)
W
2
2
whereas the corresponding potential (2.22) is
VII =

1 2
e sinh2 e2 2 cosh2 4 .
16

(4.7)

It is clear from this potential VII that becomes tachyonic if e2 2 < 4 at = 0, in

eII , the
accordance with the discussion of Section 2.3. Using the truncated prepotential W
type II domain walls obey the simpler system of first order equations


1 2 1 2
d
2
= e e
+ e
sinh ,
dr
2
2 2

48

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175



1
d
1
= e e 2 e 2 sinh2 ,
dr
2
2
1 2
d
= e
sinh 2.
dr
4 2

(4.8)

These equations can as well be obtained from the full set of first order equations (3.19)
(3.24) by imposing the above field identifications for the type IIA or IIB strings. The
equation for the conformal factor of the metric (3.17) is easily integrated and gives A =
(up to a constant that can be absorbed into a redefinition of the x s). The system of
equations (4.8) is invariant under . Therefore we will consider the case > 0
without any loss of generality.
We note for completeness that in the type II sector one may fully absorb the dependence that appears on the right hand side of these differential equations by simply
changing the coordinate variable r to defined as d = e dr. Put differently, plays the
rle of a Liouville field in that the gravitational dressing of an auxiliary massive model
made out of the fields and

1
1
1
sinh2 e2 2 cosh2 4 ,
(4.9)
L(, ) = ()2 ()2
2
2
16
yields the scalar field sector of the truncated type II theory with all three fields , and
having kinetic terms in canonical form and a dressed potential that assumes the form VII
above. 5
It is an interesting finding that the domain wall equations can be completely integrated
for the type II sector, and one has for the first time a one-parameter family of explicit
solutions with non-trivial winding . It is convenient for this purpose to present the solution
by treating the field as an independent variable, which is legitimate since the third
equation in (4.8) implies that is a monotonous function of r. Then, the differential
equation for can be easily integrated and also the equation for . The final result is a
family of BPS solutions, as shown in Section 3.3, with

e2 2 = 2 cosh2 ln coth2 + c 2,

e2 = cosh2 e

(4.10)

parametrized by an arbitrary integration constant c. As we will see, the physical

interpretation of the solutions depends
crucially on whether c is positive, negative or zero.
2
The behaviour of the functions e 2 and e2 for the three choices of c is sketched in
Fig. 1. We also note that we have omitted a multiplicative integration constant on the right
hand side of the second equation in (4.10), since it can always be absorbed into
trivial field
2 e 2 2 as 0
redefinitions.
We
see
that,
independently
of
c,
the
behaviour
of
T

is e2 2 ' 2 ln 2 . Hence, the winding mode cannot be tachyonic for all these
solutions.
5 This interpretation is motivated by two-dimensional gravity coupled to massive -models. However, unlike
other examples, where non-abelian Toda theories result in this fashion [34], the present model does not have a
special meaning in integrable systems.

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

49

Fig. 1. Qualitative behaviour of e2 2 (curves e) and e2 (curves f) as functions of for type IIA
and type IIB.

Performing the integration of the conformal factor, we find that the metric takes the form
ds =

8e

sinh2 cosh4

d +
2

cosh2

dx dx .

(4.11)

As for the relation between the variables r and in (3.17) and (4.11) this turns out to be
given by the relation of differentials
dr =

2 2e/ 2

sinh cosh2

d.

(4.12)

Clearly, this integration cannot be performed in closed form and so we lack the explicit
dependence of (r). However, the dependence can easily be spelled out in some limiting
cases to which we now turn.
4.1.1. Universal behaviour for small windings
First consider the limit of vanishing winding field . Then the asymptotic behaviour of
the fields is
2/3


3
(3r/8)4/3
2
2
,
e
'e
' r
, as r .
(4.13)
'e
8

50

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

Hence, in the limit of vanishing winding field , the fields and approach and the
value of the constant c plays no rle 6 . In addition, the metric becomes
2/3

3
2
2
dx dx , as r .
(4.14)
ds ' dr + r
8
The behaviour of the solution for larger values of the winding does depend on the
parameter c. We distinguish the following three cases.
4.1.1.1. c > 0
In this case the reality conditions for the fields and allow for 0 6 < . Then, the
function of in its full range of values.
string coupling e2 is a monotonically increasing

2
2
first increases until it reaches a maximum
However, the temperature field squared e
at = 0 and then decreases to zero as . In the limits of small and large c the
constant 0 can be found analytically:

 
c
1

ln
for c 0+ ,

4
8
(4.15)
0 =
1

for c .

c
For intermediate
values of c, 0 ranges between the above two extremes. The maximum

value of e2 2 is given by
e2

=0

1
sinh2 0 .
2

(4.16)

The asymptotic behaviour of the various fields is found by noting that due to (4.12) we
have the relation
r'

32 23/4 7/2
e
,
7c1/4

as r 0+

and ,

up to an additive constant that has been fixed in an obvious way. Then,




2/7

16c3/2 2/7
7
,
e2 ' c2 r
, as r 0+ .
e 2 '
7r
2
Similarly, the asymptotic expression for the metric is

2/7
7
dx dx , as r 0+ .
ds 2 ' dr 2 + c2 r
2

(4.17)

(4.18)

(4.19)

4.1.1.2. c = 0
As before, the reality
conditions for the fields and allow for 0 6 < . However,
both fields e2 and e2 2 are now monotonically increasing functions of in the entire
6 The exponential vanishing of the winding field is related to the fact that setting = 0 from the very
beginning is consistent with the system of Eqs. (4.8). In fact, then the solution (4.13) for and and the metric
in (4.14) below become exact (see also (5.7) below).

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

51

range of values. The asymptotic behaviour of the various fields is found by noting that due
to (4.12) we have the relation
r'

32 5/2
e
,
5

r 0+

as

and .

(4.20)

Then,

1
1
' (5r)2/5 ,
e2 ' (5r)6/5 , as r 0+ .
2
8
Similarly, the asymptotic expression for the metric is
e

1
ds 2 ' dr 2 + (5r)6/5 dx dx ,
8

(4.21)

as r 0+ .

(4.22)

4.1.1.3. c < 0
In this case the reality conditions for the fields and do not allow to exceed
a
maximum value for the winding field, i.e., 0 6 6 max . In this range, e2 and e2 2
are monotonically increasing functions of . Similarly, in the limit of small or large |c| the
constant max can be found analytically, as before,
 

1 ln c
for c 0 ,
(4.23)
max =
4
8
c1
for c ,
e 2
whereas for intermediate values of c, max ranges between the two above extremes. The
asymptotic behaviour of the various fields can be found by noting that due to (4.12) we
have the relation
8

,
(4.24)
r ' 3/4 7/4 (max )3/4 , as r 0+ and max
3sm cm
by introducing the notation
sm = sinh max ,
Then,
e


'

3cm
r
2 2

cm = cosh max .
2/3
,


'

(4.25)

r
2 2c2m

2/3
,

as

Similarly, the asymptotic expression for the metric is


2/3
3
r
dx dx , as r 0+ .
ds 2 ' dr 2 +
2 2 c2m

r 0+ .

(4.26)

(4.27)

In all cases
above, we note that the main qualitative difference is in the shape of the
function e2 2 representing the temperature field square: for c 6 0 it starts from zero and
evolves monotonically to + following the range of r from to 0, whereas for c > 0
we see a dramatic change in that the function starts and ends at zero, thus developing a
maximum along the way. The end point at r = 0 corresponds to a curvature singularity. As
we will see in Section 6, there are physical ways to differentiate among the allowed values
of c by requiring consistent propagation of a quantum test particle on the corresponding

52

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

supergravity backgrounds which have a curvature singularity. Of course, string theory (and
not supergravity alone) holds the ultimate answer for the physical relevance of our solutions
in thermodynamics.
4.2. A hybrid type-II sector
There is another truncation of the domain wall equations with H1 = 0, as in type II,
which is exactly solvable. We set for this purpose H2 = H3 and observe that the full
system of equations is consistent provided that 2 = 3 . Actually, this sector can be viewed
as a hybrid of type IIB and IIA in that we impose the diagonal constraint H2 = H3
instead of choosing the axes H2 = 0 or H3 = 0 respectively in H -field space; as such, it is
a hybrid truncation. This truncation amounts to choosing the self-dual radius so that there
is no distinction between the type-IIA and type-IIB theories.
We proceed further by setting
1
2 = 3 = ,
2
1
H2 = H3 = sinh ,
2 2

1 = ,

(4.28)

and so z2 = z3 z can be chosen as

1
z = tanh ,
2
2

(4.29)

in order for the kinetic terms of the fields

, and to assume their canonical form. As

2
. Then, the prepotential (2.27) and the potential
before, the temperature field is T e
(2.22) become


1 +2 1
2
e
+ e sinh
(4.30)
e
Whyb =
2
2
and
Vhyb =



1
sinh2 e2 (1 + cosh2 ) 8e 2 ,
32

(4.31)

22

respectively. We see that will become tachyonic, if at = 0 we have e

The truncated system of equations is



1
1
d
+ 2
2
= e
e sinh ,
dr
2
2 2

1
d
= e+ 2 ,
dr
2
1
d
= e sinh 2.
dr
4 2

< 4.

(4.32)

The equation for the conformal factor of the metric (3.17) can be easily integrated; it gives
A = ln(cosh ), up to a constant that can be absorbed into a redefinition of the x s.
Notice that the result is not the same as in the genuine type II case that we examined

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

53

Fig. 2. Qualitative behaviour of e2 2 (curves e) and e2 (curves f) as functions of for the hybrid
type II.

before. The system of Eqs. (4.32) is invariant under , and as before we only need
to consider the case of > 0. Note also that, unlike the genuine type-II case, the field
cannot be viewed as a Liouville field in this hybrid sector that provides the gravitational
dressing of a simpler massive model for the scalar fields and .
Luckily, the general solution can be found in parametric form by employing as an
independent variable, as before. We find the following family of BPS solutions
 



1
2 2
= b + 2a ln coth

,
e
2
cosh

a
e 2 ,
(4.33)
e2 =
2 cosh
where a and b are integration constants with a being positive for reality reasons. For
convenience, in the rest of this subsection, we set the positive constant a = 2, since its
precise value can be adjusted by field redefinitions and hence it has no physical relevance.
The other integration constant b cannot be fixed mathematically, but we will see later that
physical considerations in the context of supergravity may impose some restrictions on
it. Of course, as before, it is string theory that holds the answer for having an acceptable
geometrical background for domain walls. The behaviour of the functions e2 2 and e2
for the three choices of
b is sketched in Fig. 2. Again, as 0, the temperature field
behaves as T 2 e2 2 ' 2 ln 2 and e2 e 2 so that cannot be
tachyonic.
After computing the conformal factor we may write the metric as

54

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

2e 2
dx dx .
d
ds =
+
(4.34)
a cosh
sinh2 cosh3
The connection between the variables r and in (3.17) and (4.34) is given by the relation
of differentials

2 a e/ 2
d.
(4.35)
dr =
sinh cosh3/2
As before, the integration cannot be performed in closed form, apart from a few limiting
cases, as we will see next.
2

4ae

4.2.1. Universal behaviour for small windings

Let us consider first the case of vanishing winding field . It turns out that the asymptotic
behaviour of the fields is
2/3


3
(3r/8)4/3
2
2
,
e
'e
' r
, as r .
(4.36)
' 2e
8
Hence, in the limit of vanishing winding field , the fields and approach and the
value of the constant b plays no rle 7 . The metric becomes
2/3

3
dx dx , as r .
(4.37)
ds 2 ' dr 2 + r
8
Note that this asymptotic behaviour is almost identical to the one found for the pure type II
case. A similar behaviour will also exist in the heterotic case to be discussed in the next
section. There we will see that it is related to an exact solution with zero winding present
in all cases.
The behaviour of the solution for larger values of the winding , however, depends on
the parameter b. As in the genuine type II solution (4.10), we also distinguish here the
following three cases.
4.2.1.1. b > 0
In this case the reality conditions
for the fields and allow for 0 6 < . Then, the

temperature field square e2 2 is a monotonically increasing function of until it reaches

the asymptotic value 1/b. However, the string coupling e2 first increases until it reaches
a maximum at = 0 , and then decreases to zero as . In the limits of small and
large b, the constant 0 can be found analytically
 

1
3b

ln
for b 0+ ,

3
16
(4.38)
0 = r

for b ,
b
whereas for intermediate values of b, 0 ranges between the above two extremes. The
maximum value of e2 is given by

sinh 0
.
(4.39)
e2 = =
0
2 cosh 0
7 A similar comment, as in footnote 6, applies here as well.

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

55

The asymptotic behaviour of the various fields is extracted by noting that, due to (4.35),
we have the relation
32
(4.40)
r ' 1/4 e5/2 , as r 0+ and .
5b
Then, we find that
 

1
1 5r 2/5
2 2
1/4 6/5
2
' b + 5b r
,
e '
, as r 0+ .
(4.41)
e
6
2 b
Similarly, the asymptotic expression for the metric is

2/5
1
dx dx , as r 0+ .
ds 2 ' dr 2 + 5b3/2r
2

(4.42)

4.2.1.2. b = 0
As before, the reality
conditions for the fields and allow for 0 6 < . However,
both fields e2 and e2 2 are now monotonically increasing functions of in the entire
range of values. The asymptotic behaviour of the various fields is found using
 
32 3 1/4 7/4
e
, as r 0+ and ,
(4.43)
r'
7 2
which is due to (4.35). Then, we find that
 
 1/7

7 75 1/7 12/7
98
r
,
e2 '
r 2/7 ,
e2 2 '
24 54
81
Similarly, the asymptotic expression for the metric is


7 3 73 1/7 10/7
2
2
r
dx dx ,
ds ' dr +
24
4

as r 0+ .

as r 0+ .

(4.44)

(4.45)

4.2.1.3. b < 0
In this case the reality conditions for the fields and do not allow to exceed
a
maximum value for the winding field, i.e., 0 6 6 max . In this range, e2 and e2 2
are monotonically increasing functions of . In the limit of small and large |b| we find for
max the analytic expressions



1
3b
ln 32
for b 0 ,
(4.46)
max =
3
b/41
for b .
2e
For intermediate values of b, max ranges between the two above extremes. The asymptotic
behaviour of the various fields is extracted by noting that, due to (4.35), we have the relation
r'

8
3/4
3sm cm

(max )3/4 ,

as r 0+

and max
,

where we have used the same notation as in (4.25). Then,



2/3


3
3cm 2/3
2
2
'
r
,
e
'
,
e
r

2 2cm
2 2

as

r 0+ .

(4.47)

(4.48)

56

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

Similarly, the asymptotic expression for the metric is


2/3
3
r
dx dx , as r 0+ .
ds 2 ' dr 2 +
2
2 2cm

(4.49)

Note that in the hybrid type II sector, as for the genuine type II solution, the end point
at r = 0 represents a curvature singularity. We will see in Section 6 that the three different
cases, which correspond to the values of b > 0, 0 or < 0, can be distinguished using certain
physical criteria.
Finally, note that for all type II sectors (IIA, IIB and the hybrid type II), decreases
monotonically from its value at r = 0 (which may be infinite) to = 0 at r = . In
particular, this implies that (r) cannot be a periodic function of r, and hence r necessarily
is a non-compact coordinate.

5. Heterotic sector
The heterotic limit is obtained by setting H2 = H3 = 0, while keeping H1 H free
to vary. It then follows from (3.20) and (3.21) that 2 = 3 up to an irrelevant additive
constant that we will ignore. Trading the z corresponding to H for a winding field and
introducing appropriate variables

z = tanh ,
(5.1)
2 = 3 = ,
2
2
for which 2H = sinh , we obtain the expression for the truncated prepotential




ehet = 1 e e 2 sinh 2 sinh2 ,
(5.2)
W
2
and the potential




1
(5.3)
Vhet = e2 sinh2 cosh 2 2 cosh2 3 .
8

We note
for completeness that e is the string coupling and the temperature field is
2
(normalized in heterotic string units that remove the -dependence) and the
again e
kinetic terms of the fields , and assume their canonical
form with our choice of
normalizations. We see that can be tachyonic only if cosh(2 2) < 3 at = 0.
We remark that although the potential Vhet is invariant under and
ehet is invariant under the first transformation only. The second
, the prepotential W
transformation corresponds to T-duality in the temperature radius. Since the prepotential
breaks this temperature T-duality one expects the same to be true for non-trivial solutions
of our equations.
The truncated system of differential equations is




1
d
= e e 2 sinh 2 sinh2 ,
dr
2 2



1 
d
= e e 2 cosh 2 sinh2 ,
dr
2
1 = ,

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

57

d
1
(5.4)
= e sinh 2 sinh 2.
dr
2 2
The equation for the conformal factor of the metric (3.17) is easily integrated and gives
A = , as in the case of the genuine type IIA or type IIB theories. Also, similarly to this
case, we may drop the -dependence from the right hand side of the resulting first order
system by simply changing variables to defined by d = e dr. Again, may be viewed
as a Liouville field whose coupling provides the gravitational dressing" of an auxiliary
massive model of the fields and with Lagrangian density




1
1
1
(5.5)
L0 (, ) = ()2 ()2 sinh2 cosh 2 2 cosh2 3 .
2
2
8
This prescription produces naturally the canonical kinetic term for the field and dresses
the potential into Vhet .
It turns out that it is not possible to solve the domain wall equations in closed form in
the heterotic sector. Nevertheless, we can still determine the leading small asymptotics,
similarly to the analogous universal behaviour for the type II theories. This corresponds
to

large r. As it turns out, is again exponentially suppressed as r , while e and e 2

have a power-law behaviour. Hence, we can safely drop the sinh2 part in the first two
equations in (5.4) and replace sinh 2 by 2 in the third equation. The resulting system is
easily integrated and yields in this limit

9 2/3
2 4/3
e2 '
r ,
' 0 eC r , as r , (5.6)
e 2 ' 2Cr 2/3,
32C 2

where C and 0 are two positive constants of integration. Note that cosh(2 2) '
2C 2 r 4/3 ' 2 ln 2 as 0 and cannot be tachyonic. As one decreases the value
of the constant 0 , the range of approximate validity of this asymptotic solution extends to
larger intervals of r. Eventually, if 0 = 0 one obtains a particular exact solution 8

9 2/3
e 2 = 2Cr 2/3,
e2 =
r ,
= 0,
(5.7)
32C 2
which is valid for all values of r. However, since we lack the most general solution, even
in some parametric form (), we will only provide results about the dominant behaviour
of the fields in the weak coupling (i.e., ) and the strong coupling (i.e., +)
regions. In fact, we will succeed to identify these regions and perform an in depth analysis
in their vicinity using standard techniques from dynamical systems. The remaining part of
the two-dimensional space (, ) can only be studied numerically.
Before delving into details, we study a bit more the exact solution for = 0. This
solution represents a straight orbit in the parameter space (, ) with = 0 everywhere,
extending from = (identified as a weak coupling point) to = + (identified as
a strong coupling point), as r ranges from + to 0. The Liouville coordinate , which
trades r as d = e dr, is given for this special solution by

(5.8)
= 2 2 Cr 2/3 ,
8 We note that this is an exact solution for the type II sector as well. Indeed, when = 0, the three different
truncated systems of equations (4.8), (4.32) and (5.4) become identical. Note that for the corresponding type II
solution we had chosen the integration constant C = 32/3 /4.

58

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

where we absorbed an integration constant into .

Introducing small fluctuations and around this special solution, to test its stability,
we find to first order that they satisfy the linearized system of equations
1
d
= ,
d



1

d
=

,
d
2
4

(5.9)

which are written for convenience using . Therefore, we have the following solution of
the linearized problem
p
A
2
= B e /8 ,
(5.10)
= ,

where A, B are some other constants. These fluctuations are small provided that ,
otherwise the linearized approximation is not valid. We conclude this analysis by noting
that the exact solution we found is stable against perturbations in the parameter space
(, ) provided that we are in the weak coupling region where . In fact, as we
will see next, the point (, 0) is an asymptotic critical point in the two-dimensional
parameter space, and so we can talk about asymptotic stability in its vicinity.
5.1. The critical points
It is more convenient to work with the coordinate instead of r in order to drop the dependence from the right hand side of the equations. Then, as in the theory of dynamical
systems, we have to study the solutions of the system
d
d
= P (, ),
= Q(, ),
(5.11)
d
d
either as functions of , i.e., as ( ) and ( ), or in a parametric form (). For the
moment we have not prescribed a set of physical boundary conditions that select certain
orbits among the infinite many that arise in the two-dimensional parameter space (, ),
as this will be done in the next section. Here, we only focus on general properties of the
solutions around the critical points of the system.
Recall that the critical points are defined as the common zeros of the functions
P and Q.
These occur in the present case either when = 0 and sinh = 1 or when e 2 = 0 and
= 0. The second case has already been analyzed and lies in the weak coupling region
of the exact solution that was described above with approaching asymptotically .
Therefore, we focus on the other critical points

0 = ln( 2 1),
(5.12)
0 = 0,
which appear symmetrically on the -axis because of the invariance of the equations under
. Then, the dilaton equation in (5.4) is easily integrated to give at these critical
points
 
2 2
1
,
(5.13)
= = ln
r
2 2

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

59

where we have absorbed integration constants into and r in an obvious way. Hence, the
dilaton field exhibits a logarithmic dependence on r and a linear in . Recall now that the
string frame metric in four dimensions is obtained by (3.17) after multiplying with e2 .
Hence, the resulting background has a flat string metric and a linear dilaton, which is the
exact solution preserving half of the supersymmetries found earlier in [18,19]. Note that,
in this case, it makes sense to pass from the Einstein to the string frame since, when and
assume their critical values (5.12), we are left with only massless fields (graviton and
dilaton) that couple to a 2-dimensional string world-sheet action.
Next, if we linearize around these critical points (5.12) by introducing small fluctuations
and , we will find the following evolution for them

 
 
1
d
1
2

=
,
(5.14)
0

d
2 2
whereas for the dilaton d()/d = 0. Computing the two eigenvalues of the corresponding matrix we have in either case of 0 that


1
= 1 17 ,
2 2

(5.15)

and since + > 0 and < 0, we see that the pair of critical points (0, 0 ) are both saddle
points. We may solve the linearized system of equations and obtain
= Ae+ + Be ,



A +
B
e
+
e
,
= 2
+

(5.16)

with depending on the choice of critical point 0 , whereas the dilaton does not change,
to linear order, from its value in (5.13).
If we demand that the critical points are reached as +, we conclude for the
integration constants that A = 0. Hence, the variations of and have a power law decay,
namely,

 17+1
2 2
17 + 1
= B
,
=
4
r

as r .

(5.17)

On the other hand, if we demand instead, that the critical points are reached as ,
we conclude for the integration constants that B = 0. Now Eq. (5.17) is replaced by


171
r
17 1
= A
,
=
4
2 2

as r 0+ .

(5.18)

Note the different behaviour of the two solutions around the critical points (5.12) as
given by (5.17) and (5.18). In particular, from (5.13), they correspond to weak and strong
string couplings, respectively. We will see that if trajectories in the entire plane are
considered, both solutions will appear.

60

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

Fig. 3. Qualitative behaviour of different trajectories in the upper-half plane for the heterotic
case.

5.2. Strong coupling points

Having extracted the behaviour of the fields around the critical points, which can be
either at weak or at strong coupling, we will examine next their behaviour in the vicinity of
the other strong coupling points, where the dilaton field tends to +. We have identified
three such regions in the parameter space (, ), which of course have a mirror counterpart
under . They are (cf. Fig. 3):
region 1: + and 0,

(5.19)

region 2: + and ,

(5.20)

region 3: and .

(5.21)

To justify this claim and extract the details of the fields in these regions, we shall treat each
case separately.
First, it is convenient to examine the heterotic equations in the limit +, which is
common to both regions 1 and 2. Using the coordinate (instead of r) we obtain in this
limit the simplified system


d
1 2
1
d
2
= 2
= e
1 sinh ,
d
d
2
2
1 2
d
= e
sinh cosh ,
(5.22)
d
2 2

and so we see immediately that 2 = (up to an additive constant). The remaining

equations can be easily integrated to yield the expressions

3
(b a )1/3 1 (b a )2/3 ,
2a 2

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

cosh2 =

1
,
(b a )2/3

61

(5.23)

in terms of some constants a > 0 and b. However, recall at this point that we are
considering
the behaviour of the equations for +, and so we should expand

exp( 2 ) around zero. Naturally, we face two possibilities, which will be identified
with regions 1 and 2.
Region 1 follows in this context by assuming (b a )2/3 ' 1. Defining the parameter
1 = (b 1)/a we have
r

1
2a
2
( 1 )1/2 , as 1+ ,
' ( 1 ), '
(5.24)
e
3
2

and so 0 as advertized for region 1. Since 2 = (up to a constant shift), we

conclude that it is indeed a strong coupling region. One easily sees that ( 1 )
(r r1 )2/3 , for some constant r1 which we henceforth absorb into r so that the metric
behaves like
ds 2 ' dr 2 + const r 2/3 dx dx ,

as r 0+ .

(5.25)

Region 2 follows by expanding the fields around b a ' 0. If we let 2 = b/a, we find
the following behaviour

3
'
(2 )1/3 ,
2 2 a 2/3

e '

2
a 1/3

1
,
(2 )1/3

as 2 . (5.26)

In this vicinity we have , depending on the branch in the upper or lower

half-plane, and since we also have +, which indeed describes the strong
coupling region 2 as advertized. Using (2 ) (r2 r)6/7 , for some positive r2 , one
finds that
ds 2 ' dr 2 + const (r2 r)2/7 dx dx ,

as r r2 .

(5.27)

Region 3 arises by first letting . Then, the heterotic equations simplify in this
limit to the following system:
d
1
d
= 2
= e 2 sinh2 ,
d
d
4
1 2
d
= e
sinh cosh .
(5.28)
d
2 2

Clearly, 2 = (up to a constant) and so this is also a strong coupling region. Since
we have , the dominant dependence on is easy to extract. We find
1/3


3
( 3 )1/3 ,
e 2 '

2c2 2
 1/3
2 2
1

, as 3+ ,
(5.29)
e '2
3c
( 3 )1/3

62

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

where c and 3 are constants of integration. In this case we see that as it was
initially stated. Finally, 3 (r r3 )6/7 for some constant r3 which we again absorb
into r. The behaviour of the metric near r = 0 is
ds 2 ' dr 2 + const r 2/7 dx dx ,

as r 0+ .

(5.30)

The behaviour of different trajectories in the (, ) plane is depicted in Fig. 3. Because

of the symmetry we have restricted to windings > 0 without loss of generality,
in particular solutions that start in the upper-half plane cannot end somewhere in the lower
half-plane. The qualitative characteristics of the various trajectories can be deduced by
treating (or ) as an independent variable and consider the differential equation for
d/d. Before we examine the different trajectories in some detail, note by inspection of
the differential equation d/d and from the strong coupling behaviour that we already
know, that there can be trajectories along which the string coupling e develops a minimum
at a point in the plane where
e2

sinh2
sinh 2
2

> 0 and > 1+ ln

3+ 2 .

(5.31)

For < 0 there is no such minimum and the string coupling e keeps increasing with
increasing . Let us now discuss the different types of trajectories as depicted in Fig. 3.
5.2.1. Trajectories with an extreme
These are trajectories where for some we have = 0 and hence d/d = 0 showing
that these trajectories intersect perpendicularly the axis = 0. Since

sinh 2
d 2 
,
(5.32)
=

2
d =0 sinh2 1

we have for > 0+ = ln( 2 + 1) (< 0+ ) that this is a minimum (maximum) in the
trajectory. Two such typical trajectories are drawn in Fig. 3. The lower one connects the
weak coupling region (, ) = (, 0) with the strong coupling region 1, after reaching
a maximum for the winding max (0, 0+ ). Since (5.31) can never be satisfied, it can be
easily seen that d/d > 0 throughout the trajectory and hence the string coupling e
is progressively increasing with . Another such trajectory connects the strong coupling
regions 2 and 3 going through a minimum for the winding min (0+ , ). For this
trajectory there is of course a minimum for the string coupling at a point described by
(5.31). We also note that for the trajectories with 6 max that connect the weak coupling
region to region 1, the variable and hence also r decrease as increases, with r going
from + to 0. For the trajectories with > min that connect regions 3 and 2, and r
increase as increases, with r going from 0 to some r2 > 0. Note that the value of r2
depends on the chosen trajectory.
5.2.2. Trajectories with an extreme
There are trajectories for which d/d = 0 and hence there is an extreme value for .
We find that the condition for this is

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

e2
Since

sinh2
2 sinh2


H


2
d 2 
,
=
d2 d/d=0 sinh2 1

> 0 for 0+ < < 1+ ,

< 0 for 0 < < 0+ .

63

(5.33)

(5.34)

we have for > 0 that this is a minimum (indicated by min ) occurring for (0+ , 1+ ),
whereas for < 0 it is a maximum (indicated by max ) occurring for (0, 0+ ). Two
such typical trajectories are drawn in Fig. 3. The one with < 0 connects the weakly
coupled region having (, ) = (, 0) with the strongly coupled region 3. Since < 0
there is no minimum for the string coupling, which increases with in accordance with
(5.31). The trajectory with > 0 connects the two strongly coupled regions 1 and 2. For
such trajectories, besides the depicted minimum value for , there is also a minimum value
for the string coupling according to (5.31). Note that for the trajectory with > 0 ( < 0)
the variables and r increase (decrease) as increases, with r ranging from 0 to some
r2 > 0 (from to 0). Again, the value of r2 depends on the chosen trajectory.
We also note that region 2 contains trajectories that come either from region 1 or from
region 3. 9 The family of such trajectories is parametrized by the constant a in (5.26).
The values of r2 and a are clearly related, but we do not know exactly how. There exists a
critical value, which is numerically found to be acrit ' 0.9, such that for a > acrit (a < acrit )
we connect to region 1 (3). Similarly, in region 3 the different trajectories are parametrized
by the constant c in (5.29) and there is also a critical value ccrit ' 1.2; for c > ccrit (c <
ccrit ) these trajectories end up in the weakly coupled region (region 2). Of course, for the
trajectories that connects the regions 2 and 3 the constants a and c are related, but how
precisely, cannot be answered without knowledge of the explicit solution.
5.2.3. Trajectories ending at critical points
These are the trajectories depicted with the dashed lines and occur when the constants
a and c assume the critical values we mentioned. We emphasize that despite appearances,
these trajectories have the critical points as their end-points, and they do not simply pass
through them. In other words, they do not connect the strongly coupled regions 1 and
3 nor the weakly coupled region (, ) = (, 0) with the strongly coupled region 2.
The critical point indicated in Fig. 3, represents a weak coupling point for trajectories that
come either from region 1 or from region 3. In these cases the solution in the vicinity of the
critical point is given by (5.17) and the variables or r increase to + as we approach the
critical point. In contrast, for trajectories that go either to region 2 or to the weakly coupled
region with (, ) = (, 0), this critical point is at strong coupling. Then, the solution
in its vicinity is given by (5.18) and the variables or r increase as we get away from
the critical point. We note that the trajectory going to region 2 connects strong coupling to
strong coupling so that must have a minimum along its way. This is consistent with the

2 ' + ln( 2a/3)

(for region 2) and 2 ' + ln(2/c)

(for region 3). For region 1 we have instead 2 ' 2 ln ln(3/2 2a).
9 In these regions we have the linear behaviour

64

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

fact that such a minimum occurs if and only if > 0 and > 1+ . This trajectory is the
only one among the four for which r takes values in a finite interval [0, r2 ].
In Section 6, we will discuss certain criteria based on the propagation of a free
quantum particle on the corresponding backgrounds that render some of these trajectories
unphysical.
We conclude the discussion of the heterotic equations by mentioning the change of
variables

sinh
,
(5.35)
x = tanh ,
y = cosh 2
cosh2
which cast the differential equations for () into a simpler looking form for y(x) (after
taking its square)
x 2 (y 0 )2 y 2 = x 2 (x 2 1).

(5.36)

Unfortunately, we are still lacking its general solution in analytic form. Nevertheless,
the form of the Eq. (5.36) enables to perform a systematic study of the perturbative
expansion around the critical points (5.12), which are located at (x, y) = ( 1 , 12 ) in
2
the parametrization (5.35). We have the following perturbative expansion, with the upper
(lower) signs referring to the corresponding signs in (5.12),



1 n+2
1 X
an x
,
(5.37)
y = +
2
2
n=0
where the coefficient a0 of the expansion is given by
2a02 a0 2 = 0.

(5.38)

We also have the following recursive relations for n > 1:

n1

1X
(k + 2)(n k + 2)ak ank
2(n + 2)a0 1 an = 2 2 n,1 n,2
2
k=1

n1
X
2
(k + 2)(n k + 1)ak ank1
k=0

n2
X

(k + 2)(n k) 1 ak ank2 .

(5.39)

k=0

Note that the freedom to choose the constant a0 in (5.38) corresponds to the two
independent solutions of (5.36) that arise from taking the square root. Since the constant
a0 determines the sign of y 0 , we have for > 0 that a0 > 0 for > 0, whereas a0 < 0 for
< 0. The first few terms of the expansion are given by





1 3
1 2 2 2(a0 1)
1

+ .
(5.40)
x
y = + a0 x
2
6a0 1
2
2
One could develop a similar perturbative expansion around any of the strongly coupled
regions 1, 2 and 3 or in the vicinity of the weakly coupled region (, 0). The

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

65

correspondingpoints, in the xy plane, around which one should expand are (x, y) =
(0, +), (1, 2a/3), (1, c/2) and (0, 0), respectively.
Finally, let us comment on the possibility to have a compact coordinate r as suggested
by the microcanonical analysis of Refs. [1517]. This would imply that all fields must
come back to their initial value after some period in r. We have seen that there are two
types of trajectories. On the one hand, we have those for which r takes values in a semiinfinite interval. These are the trajectories that connect the weak-coupling region (, ) =
(, 0) to any other region, as well as those connecting regions 3 or 1 to the critical point.
Clearly, these solutions are not, and cannot be made periodic in r. On the other hand, we
have the trajectories ending in region 2 for which r takes values in the finite intervals
[0, r2 ] with the value of r2 depending on the chosen trajectory. This is crucially different
from what we found in the type-II cases. Note that trajectories in this second class always
go from strong to strong coupling. It may well be that these solutions can be continued into
periodic functions of r beyond these intervals, tracing the corresponding trajectories back
and forth. The situation is examplified by the parabola y = x 2 in the x y plane with x a
r
function of r given by x(r) = ln 1cos
1+cos r . This clearly is a periodic function in r and hence
r can be taken as a compact coordinate with values on the circle [0, 2]. On the other
hand, if we only look at the interval r [0, ], x(r) goes from to + and we trace
the parabola just once. This is somewhat similar to our trajectories going into region 2.
Without an explicit solution at our disposal, it is probably impossible to decide whether or
not we can continue these solutions into periodic ones or not.

6. Physical boundary conditions

Next, we examine the boundary conditions which are physically relevant for the domain
wall backgrounds. Note that until now the integration constants of the corresponding
first order equations were left undetermined, and so any solution could be useful
mathematically, as there is no preference among the different orbits in the parameter space,
say (, ). We now apply some general criteria for restricting the physical range of the
parameters within the context of supergravity using the propagation of a quantum test
particle on the corresponding backgrounds. Then, we discuss the criteria that string theory
may impose, although the situation is less clear in that case.
Curvature singularities lead to singular classical dynamics for test particles. If they
persist at the quantum level, the theory is considered as unphysical. Unfortunately, we
do not know enough about string theory in such backgrounds in order to answer this
question definitely. Instead, we can ask the simpler question whether the propagation of
a quantum test particle is well defined in the presence of a curvature singularity. A singular
classical propagation, as indicated by an incomplete geodesic motion, does not necessarily
lead to a singular wave propagation since the spacetime can produce an effective barrier
that shields the classical singularity. In general terms, a singularity will be physical if the
evolution of any state is uniquely defined for all times. The background is then called
wave-regular. This criterion is identical to finding a unique self-adjoint extension of the

66

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

wave-operator [3537]. This can be quickly investigated by looking at the two solutions
of the wave equation locally near the singularity. Note that the wave equation arises, for
instance, when one considers linearized graviton fluctuations along the three-dimensional
brane embedded into the four-dimensional domain wall ansatz (3.17), i.e., +
h . 10 In such cases, in addition to having a unique evolution of initial data, one has to
ensure that these fluctuations never become strong, which would invalidate the linearized
approximation.
A remark is in order. Since we are at finite temperature we have compact euclidean time
and it is not clear right away what is meant by dynamics and evolution. Our coordinates
r and x with = 1, 2, 3 are all spatial. To study the adiabatic dynamics of a single
test particle one should add a test-particle time coordinate with gt t = 1, gt = gt r = 0.
Indeed, a single test particle only propagates in the thermal background specified by the
metric but does not interact with the thermodynamical system. Thus in the following, we
use the five coordinates: test-particle time t and four space coordinates x i split into r and
x , = 1, 2, 3.
The relevant part of the wave equation is


d
d
g g rr
= 0.
(6.1)
dr
dr
We will soon justify the need to keep only this term in the wave equation for our class
of examples, where the metric behaves in general as in (6.4) below. In order to check the
normalizability of the two independent solutions of the wave equation, we will use the
Sobolev norm
Z
Z
1
q2
g g t t +
g g ij i j ,
(6.2)
2
2

where q is a constant and the integrals are performed on a constant-time hypersurface

. The Sobolev norm is bounded from above by the energy of the fluctuation associated
with the wave, and guarantees that the back-reaction of the fluctuation is indeed small [37].
According to this criterion there exists a unique self-adjoint extension of the wave-operator
and a unique evolution of the system, thus rendering the space wave-regular, if only one of
the two independent solutions of (6.1) is normalizable near the singularity. This solution is
then kept, whereas the other is discarded. We should note that this criterion is non-trivial
for time-like singularities. For null singularities the spacetime is globally hyperbolic and
there should be a unique self-adjoint extension of the wave-operator since the evolution is
ordinary Cauchy, and hence unique in this case.
The choice of Sobolev norm is compatible in our cases with an alternative criterion,
namely, that in geodesically incomplete spacetimes there should be no leak of conserved
quantities through the singularity. For the translational isometries that remain unbroken by
our metric ansatz (3.17), this amounts to the condition (see, for instance, [41])
10 This has been shown in the context of the AdS/CFT correspondence and in warped factor compactifications
in [38] (for an exhaustive exposition see [39,40]). The proof in the present case can also be given along similar
lines.

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

g g rr i r = 0,

67

(6.3)

at the singularity. In order to have a unique evolution of the system, only one of the two
independent solutions of Eq. (6.1) should satisfy (6.3). For the other solution, which is
discarded, the right hand side of (6.3) should diverge.
Returning to our metrics, we have seen that they develop a curvature singularity as r
+
0 or r r2 . We will discuss the case of a singularity at r = 0+ , the other case being
analogous. Then the asymptotic behaviour of the metric is of the form
ds 2 ' dr 2 + const r 2 dx dx ,

as r 0+ ,

(6.4)

with being a positive constant that can be read from the appropriate expressions in
Sections 4 and 5. In the case of the strongly coupled solution (5.18) around the critical
point (, ) = (0, 0+ ), we have that = 1, and for all other cases 0 < < 1. Applying
the general formalism for a background behaving as in (6.4) we first see that the two
independent solutions of (6.1) are 1 1 and 2 r 13 (or 2 ln r if = 1/3).
Note that if we had treated the full wave equation in (6.1), we would have added a term
proportional to r on the left-hand side. Such term perturbs the two independent solutions
that we have just mentioned by a term proportional to r 22 and r 35 , respectively.
Therefore, near the singularity occurring at r = 0, this extra term can be neglected if < 1,
or does not change the power of r if = 1.
Applying either one of the above criteria, we find that backgrounds with a single
singularity behaving as in (6.4) are wave-regular for > 1/3. 11 Hence, for the genuine
type II, as well as for the hybrid type II solutions, the backgrounds are wave-regular
when the constants c and b, respectively, are negative or zero, while positive c or b are
unphysical.
For the heterotic case, we have to distinguish solutions with semi-infinite intervals of
r (those not going into region 2) and solutions with finite intervals of r (those going
into region 2). For the former there is only one singularity at r = 0+ and we find that
all solutions that contain in some limit the region 3 do not satisfy the criteria, and hence
are unphysical. The only physical solutions in this class are those corresponding to the
trajectories that connect region 1 to the weak coupling region, or the special trajectory that
go from the critical point 0+ to region 1 or to the weak coupling region. Note that for all
these trajectories remains bounded by 0+ . Now consider the solutions that have a finite
interval of r. They have two singularities, one at r = 0+ and one at r2 . If at each singularity
one of the two solutions to the wave equation is discarded, no solution will be left, as in
general the solution discarded at one singularity will not be the same as the one discarded
at the other. Such a background can be wave-regular only if one of the singularities discards
one solution and the other singularity does not impose any condition, i.e., we need > 1/3
at one singularity and < 1/3 at the other. We have = 1/3, 1/7, 1/7 and 1 for regions
1, 2, 3 and the critical point. We conclude that solutions corresponding to trajectories from
11 This condition also guarantees that the solution r 13 that diverges at the singularity occurring at r = 0,
2
is discarded. Hence, if the wave equation that we are examining corresponds to linearized graviton fluctuations,
these fluctuations will indeed remain small, thus not invalidating the approximation.

68

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

region 2 to region 1 or to the critical point are wave regular. The solution corresponding
to the trajectory from region 2 to region 3 however is not wave-regular. It is evident that a
better understanding of the string rather than the particle propagation is desirable on such
backgrounds. 12
From the string thermodynamic point of view it has been argued [1517] that the
exponential growth of the number of states forces one to work in the microcanonical
ensemble, in which case the specific heat may turn negative unless all spatial dimensions
are compact; of course some compact dimensions may well be large (see also [43,44]).
Within our domain wall ansatz, compactness or not of the space coordinates x is not an
issue since the solutions do not depend on those directions. However, compactness of the
r coordinate is possible only if the solutions (r), (r) and (r) at some r equal their
values at some other r. As we discussed in the previous section, this is not the case for
the type II solutions. For the heterotic case we have some solutions that are parametrized
by a finite interval in r. However, without the explicit solution at our disposal, it is not
clear whether they can be continued to periodic solutions or not. In any case, before
jumping into conclusions about our solutions being physical or unphysical within string
theory, one should bear in mind that the compactness criterion was established within the
microcanonical treatment of string theory, which assumes that the theory remains weakly
coupled everywhere. 13 It is clear from our solutions that they always contain regions of
arbitrarily strong coupling, thus invalidating the assumptions that could sentence some or
all of them unphysical.
Another argument in favour of our solutions is their supersymmetry. As shown
above, all solutions of the first order equations preserve half of the supersymmetries,
by their construction, while the standard high-temperature phase of strings breaks all
supersymmetries. Clearly, at this point we cannot say anything further as definite about
the relevance and physical properties of our different domain wall solutions within string
thermodynamics. Ideally, a string inspired criterion would render physical all or a proper
subset of those backgrounds, which have already been named physical using the criteria of
supergravity in this section.

7. Conclusions and discussion

In this paper we have constructed domain wall solutions of the effective supergravity
which describes all possible high-temperature phases of the known N = 4 superstrings.
We have used a universal thermal potential that contains the s, t, u moduli and their
couplings to the three (lightest) winding modes, which can become tachyonic above the
12 The above criteria have been successfully applied within the AdS/CFT correspondence in many backgrounds
related to the Coulomb branch of N = 4 and N = 2 super YangMills theories [42]. In those cases the lack of
information on string theory beyond the supergravity approximation is compensated by the information related
to gauge theory expectations.
13 It is somewhat ironical that the only solutions that perhaps could be continued periodically actually are never
weakly coupled.

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

69

Hagedorn temperature, thus triggering the thermal instabilities. Our solutions contain nontrivial winding fields, which vary with respect to the domain wall variable r, and exhibit
a common property in that they always extend to regions of strong coupling. We have
presented the exact solutions for the type IIA and type IIB sectors of the theory and a
certain (self-dual) hybrid sector, and investigated the general structure of the solutions
for the heterotic sector by extracting their behaviour in the vicinity of the weak and the
strong coupling points. From a geometrical point of view we found that none of the type-II
solutions has support on a compact spatial dimension, namely, the direction parametrized
by the variable r. For the heterotic case, the same is true for all solutions that contain a
weakly coupled region, while solutions that never are weakly coupled have support only
on finite intervals in the variable r. Whether or not these latter solutions can be continued
into periodic functions of r could not be answered without having an explicit solution at our
disposal. From a supergravity point of view, for each type of theory, there are subclasses
of domain wall backgrounds which are selected by imposing specific boundary conditions
that lead to consistent propagation of a quantum test particle.
The main property of the domain wall solutions is their supersymmetry; they all
satisfy first order differential equations, which arise as BPS conditions for gravitational
backgrounds that preserve 1/2 of the supersymmetries. The supersymmetry of our
solutions is the key to understand why, for all of them, tachyonic instabilities never
occur, even though the temperature may become arbitrarily high. Of course, we are only
considering solutions of an effective supergravity and not of the full string theory, but it
may well be that they point to a new finite-temperature phase of superstrings which is
supersymmetric and has no thermal instabilities, i.e., no Hagedorn temperature. One may
then speculate that, as superstrings are heated up from zero temperature, they prefer to
go into this more symmetric phase which is stable due to supersymmetry, and that the
ordinary high-temperature phase with the Hagedorn instability is never reached.
Let us now address a few points one might object to our solutions. First, one might
worry about the status of the domain wall ansatz in theories of gravity coupled to scalar
fields and having a potential with run-away directions that account for the usual instabilities
at the Hagedorn temperature. Note however, that this is not a priori forbidden by any
general considerations alone. Recall that domain wall solutions with infinite tension
(corresponding to infinite central charge of the supersymmetry algebra) have already found
numerous applications in supersymmetric gauge theories, most notably in supersymmetric
quantum chromodynamics with one massless flavor, where one has models with runaway vacua. In fact, these are models with rigid supersymmetry which admit stable field
configurations that restore one half of supersymmetry, as opposed to the stable but nonsupersymmetric ground state, and which are characterized by constant positive energy
density [45]. The domain wall solutions that we have constructed in supergravity for the
problem of string thermodynamics could be viewed as the gravitational analogue of such
supersymmetric configurations, having their own characteristic properties. Their stability
is guaranteed by the saturation of the BPS bound and can be further supported by analysing
the spectrum of the graviton fluctuations on the domain wall backgrounds. Based on
general principles, one can show that the graviton spectrum is obtained by computing

70

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

the energy levels of an equivalent non-relativistic problem in supersymmetric quantum

mechanics with a Schrdinger potential that is determined entirely by the conformal factor
of the metric. The details are not important as the only relevant point here is that the
spectrum is bounded from below (by zero) by the general properties of supersymmetric
quantum mechanics, thus rendering a physical spectrum for the graviton fluctuations on
the supersymmetric domain wall backgrounds that preserve three-dimensional Poincar
invariance.
Second, one should also study metric fluctuations taking us away from the class of
conformally flat metrics of the domain wall backgrounds. We have not investigated this
issue, but we expect that an analysis along the line that can, e.g., be found in [39,40], is
possible.
Third, a well-known instability of hot flat space was analysed in [46]. In this simplest
example it was found that the classical Jeans instabilities arise as a tachyon in the graviton
propagator, using small fluctuations about hot flat space, and it was further suggested that
hot flat space will nucleate black holes. Inevitably, this is the fate of gravitational systems
due to the attractive nature of gravitation. In the case of a relativistic medium, the Jeans
instability implies that a thermal ensemble of sufficiently large volume will collapse into
a black hole. Of course, it remains to be seen how these results generalize to domain
wall backgrounds of our effective supergravity by performing calculations as in ordinary
quantum gravity. One might fear that most of our solutions will be afflicted by Jeans
instabilities because they do not have support in small volumes. However, their defining
BPS property ensures their stability quantum mechanically as well. Put differently, the
BPS equation which is interpreted as a no force condition, although this interpretation
is more appropriate for asymptotically flat backgrounds with a definite Newtonian limit,
ensures that no gravitational collapse will occur.
Clearly, spherically symmetric black hole type solutions of the effective supergravity
have to be analysed in the future before addressing the long standing problems of quantum
gravity within string thermodynamics. These general remarks put in perspective future
attempts to construct other non-trivial solutions of the effective supergravity for the s, t,
u moduli and the string winding modes beyond the supersymmetric domain wall ansatz.
An important lesson that we already learned, even from the simplest solutions we have
constructed here, is that they cannot be entirely confined in regions of weak coupling only.
It would be interesting to study the most general domain wall solution of the effective
supergravity that describes all possible phases of N = 4 superstrings at finite temperature
and includes all six scalar fields. The resulting system of first order non-linear differential
equations does not seem tractable by analytical methods alone, but a combination of
numerical analysis and analytical calculations around certain points may well provide a
reasonable global picture, similar to what we did in the heterotic case. In Appendix A we
summarize for completeness some general remarks about the search for periodic orbits
in order to appreciate the difficulty to establish analytic criteria for their existence in the
general case.
Finally, it will also be interesting to study the higher dimensional interpretation of our
domain wall configurations, as in other theories of gauged supergravity. This might be

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

71

related to thermodynamics of string theory in D-brane backgrounds as has already been

discussed in the weak-coupling limit in [48,49]. One may also try to extend the formalism
to construct domain wall junctions.

Acknowledgements
One of us (I.B.) wishes to thank the members of the Institute of Physics in the University
of Neuchtel for the warm hospitality extended to him during a visit and the collaboration
that initiated the present work. We also thank G. Katsiaris for his assistance with the
numerical aspects of the heterotic sector. This research was supported by the Swiss
National Science Foundation, the European Union under contracts TMR-ERBFMRXCT96-0045 and -0090, by the Swiss Office for Education and Science and by RTN contract
HPRN-CT-2000-00122.

Appendix A. Remarks on periodic orbits

Although we have seen quite explicitly that all our type-II solutions are non-periodic in
the domain wall coordinate r, and the same probably is true also for the heterotic solutions,
it is much less clear whether this will still be true for the solutions of the full set of six
equations (3.19)(3.24). Since for the latter one probably has to rely heavily on numerical
methods, it would be nice to have at least some analytical tools at ones disposal. One
possibility is to try to establish the existence or non-existence of periodic orbits in the
solution space of our non-linear system of first order differential equations for the domain
wall configurations. Clearly if there are such periodic orbits in the solution space (e.g., in
the plane in the heterotic case) then the corresponding solution will also be periodic
in r. Note however that the converse is not necessarily true as would be examplified by the
heterotic trajectories going into region 2, should it turn out that they can be periodically
continued in r. Of course, then one might still consider them as periodic orbits in the
plane, although degenerate ones. Finally, for reasons already discussed at length in
the introduction and discussion section, it is not clear whether periodicity may serve as a
criterion within string thermodynamics to distinguish physical from unphysical solutions.
It is well known from the theory of dynamical systems that the existence of periodic
solutions, i.e., closed loops in the two-dimensional parameter space (, ), of a first order
system = P (, ) and = Q(, ) is a very important and delicate question that often
can be established only numerically (unless the general solution is known in closed form).
The boundary conditions select specific orbits () by fixing the integration constants, and
so by singling out the periodic orbits, if there are any among the solutions, corresponds to
specifying some physical boundary conditions.
To appreciate the difficulty in establishing the existence of periodic orbits analytically,
we recall briefly some basic elements of Poincars theory (see, for instance, [47] for an
elementary account). There, one has the notion of a limit cycle, i.e., a stable closed curve
in the parameter space, independent of initial conditions, towards which solutions tend in

72

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

an asymptotic sense, or from which they unwind, as it were, as . A well known

theorem states that if a limit cycle exists in a given system, then the existence of periodic
orbits is guaranteed, even though their explicit construction could be a difficult task. On
the other hand we have Bendixsons theorem stating that if one considers the function
Q
P
+
,
(A.1)
J=

in any given domain in (, ) which is bounded by a simple curve C, the system will have
no limit cycles inside that region if J has constant sign in it. Therefore, all we can learn
on general grounds is for which regions there are no limit cycles; if J changes sign in
a bounded region (and hence it is bound to vanish somewhere in it), the system will not
necessarily have limit cycles in it, but there is a good chance that it will. Furthermore, there
are systems with periodic orbits that have no limit cycles.
According to these results, we can already see a difference between the type II and
heterotic sectors concerning the possibility to have (or not to have) limit cycles in
their parameter space. Computing Bendixsons function for these two sectors (using the
coordinate that decouples the -dependence from the flows of the fields and ) one
finds the result


1 
JII = 2e 2 + e 2 1 + 3 sinh2 ,
2 2

1 
Jhyb = 2e 2 + 1 + 2 sinh2 ,
2 2





1 
(A.2)
Jhet = cosh 2 3 sinh 2 sinh2 .
2
It is obvious that JII and Jhyb are strictly negative everywhere in the parameter space
(, ), and so according to Bendixsons theorem we learn that there are no limit cycles
in these cases. Of course, this by itself does not rule out the existence of periodic orbits,
but we already know from our explicit solutions that for the pure type II and hybrid type II
sectors there are none. On the other hand, the heterotic sector could support limit cycles,
since Jhet vanishes along the curve


(A.3)
coth 2 = 3 sinh2 .
This curve has two branches, one in the upper (, ) half-plane and the other in the lower,
which are related to each other by the discrete symmetry . Drawing the curve in
the upper half-plane we see that it
starts asymptotically from = + at = 0 and drops
monotonically to the value = ln 3 < 0+ , which is approached asymptotically as
+. Therefore, if one considers bounded domains in the parameter space which intersect
with Bendixsons curve Jhet = 0, they will potentially contain limit cycles, although their
existence is not at all guaranteed. Note that this curve is crossed precisely by the trajectories
going to region 2. Of course, these trajectories are not contained in any bounded domain.
Nevertheless, this is still suggestive that maybe these trajectories can indeed be periodically
continued.
It will be interesting to look for closed trajectories in the higher dimensional parameter
space with all the relevant fields turned on.

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

73

References
[1] A. Strominger, Massless black holes and conifolds in string theory, Nucl. Phys. B 451 (1995)
96, hep-th/9504090.
[2] H. Ooguri, C. Vafa, Two-dimensional black holes and singularities of CY manifolds, Nucl.
Phys. 463 (1996) 55, hep-th/9511164;
H. Ooguri, C. Vafa, Summing up D instantons, Phys. Rev. Lett. 77 (1996) 3296, hepth/9608079.
[3] R. Hagedorn, Statistical thermodynamics of strong interactions at high-energies, Nuovo
Cimento Suppl. 3 (1965) 147.
[4] S. Fubini, G. Veneziano, Level structure of dual resonance models, Nuovo Cimento A 64 (1969)
811.
[5] K. Huang, S. Weinberg, Ultimate temperature and the early universe, Phys. Rev. Lett. 25 (1970)
895.
[6] S. Frautschi, Statistical bootstrap model of hadrons, Phys. Rev. D 3 (1971) 2821.
[7] R. Carlitz, Hadronic matter at high density, Phys. Rev. D 5 (1972) 3231.
[8] M. Bowick, L. Wijewardhana, Superstrings at finite temperature, Phys. Rev. Lett. 54 (1985)
2485.
[9] B. Sundborg, Thermodynamics of superstrings at high-energy densities, Nucl. Phys. B 254
(1985) 583.
[10] P. Salomonson, B. Skagerstam, On superdense superstring gases: a heretic string model
approach, Nucl. Phys. B 268 (1986) 349.
[11] E. Alvarez, Strings at finite temperature, Nucl. Phys. B 269 (1986) 596.
[12] B. Sathiapalan, Vortices on the string world sheet and constraints on toral compactification,
Phys. Rev. D 35 (1987) 3277.
[13] Ya. Kogan, Vortices on the world sheet of a string; critical dynamics, JETP Lett. 45 (1987) 709.
[14] J. Atick, E. Witten, The Hagedorn transition and the number of degrees of freedom of string
theory, Nucl. Phys. B 310 (1988) 291.
[15] R. Brandenberger, C. Vafa, Superstrings in the early universe, Nucl. Phys. B 316 (1989) 391.
[16] D. Mitchell, N. Turok, Statistical properties of cosmic strings, Nucl. Phys. B 294 (1987) 1138.
[17] S. Alexander, R. Brandenberger, D. Easson, Brane gases in the early universe, hep-th/0005212.
[18] I. Antoniadis, C. Kounnas, Superstring phase transitions at high temperature, Phys. Lett. B 261
(1991) 369.
[19] I. Antoniadis, J.-P. Derendinger, C. Kounnas, Non-perturbative temperature instabilities in N =
4 strings, Nucl. Phys. B 551 (1999) 41, hep-th/9902032.
[20] E. Witten, Is supersymmetry really broken?, Int. J. Mod. Phys. A 10 (1995) 1247, hepth/9409111;
E. Witten, Strong coupling and the cosmological constant, Mod. Phys. Lett. A 10 (1995) 2153,
hep-th/9506101.
[21] I. Antoniadis, J.P. Derendinger, C. Kounnas, Non-perturbative supersymmetry breaking and
finite-temperature instabilities in N = 4 superstrings, in: Proceedings of 6th Hellenic School
and Workshop on Elementary Particle Physics, Corfu, Greece, September 1998, pp. 626, hepth/9908137.
[22] C. Kounnas, B. Rostand, Coordinate dependent compactifications and discrete symmetries,
Nucl. Phys. B 341 (1990) 641.
[23] J.H. Schwarz, A. Sen, Duality symmetries of 4-D heterotic strings, Phys. Lett. B 312 (1993)
105, hep-th/9305185;
J.H. Schwarz, A. Sen, Duality symmetric actions, Nucl. Phys. B 411 (1994) 35, hep-th/9304154.
[24] M.J. Duff, J.T. Liu, J. Rahmfeld, Four-dimensional stringstringstring triality, Nucl.
Phys. B 459 (1996) 125, hep-th/9508094.

74

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

[25] M. Cvetic, D. Youm, Dyonic BPS saturated black holes of heterotic string on a six torus, Phys.
Rev. D 53 (1996) R584, hep-th/9507090;
M. Cvetic, D. Youm, Singular BPS saturated states and enhanced symmetries of fourdimensional N = 4 supersymmetric vacua, Phys. Lett. B 359 (1995) 87, hep-th/9507160.
[26] G.L. Cardoso, G. Curio, D. Lst, T. Mohaupt, S.-J. Rey, BPS spectra and nonperturbative
gravitational couplings in N = 2, N = 4 supersymmetric string theories, Nucl. Phys. B 464
(1996) 18, hep-th/9512129.
[27] E. Kiritsis, C. Kounnas, Perturbative and nonperturbative partial supersymmetry breaking: N =
4 N = 2 N = 1, Nucl. Phys. B 513 (1997) 117, hep-th/9703059.
[28] D.Z. Freedman, S.S. Gubser, K. Pilch, N.P. Warner, Renormalization group flows from
holography, supersymmetry and a c-theorem, hep-th/9904017.
[29] O. DeWolfe, D.Z. Freedman, S.S. Gubser, A. Karch, Modeling the fifth dimension with scalars
and gravity, hep-th/9909134.
[30] R. Kallosh, J. Kumar, Supersymmetry enhancement of Dp-branes and M-branes, Phys. Rev.
D 56 (1997) 4934, hep-th/9704189.
[31] M. Cvetic, S. Griffies, S. Rey, Static domain walls in N = 1 supergravity, Nucl. Phys. B 381
(1992) 301, hep-th/9201007;
M. Cvetic, H.H. Soleng, Supergravity domain walls, Phys. Rep. 282 (1997) 159, hepth/9604090.
[32] I. Bakas, K. Sfetsos, States and curves of five-dimensional gauged supergravity, Nucl. Phys.
B 573 (2000) 768, hep-th/9909041.
[33] I. Bakas, A. Brandhuber, K. Sfetsos, Domain walls of gauged supergravity, M-branes, and
algebraic curves, hep-th/9912132, Adv. Theor. Math. Phys., in press.
[34] I. Bakas, Q-Han Park, Gravitational dressing of massive soliton theories, Phys. Lett. B 387
(1996) 707, hep-th/9607243.
[35] R. Wald, Dynamics in nonglobally hyperbolic, static spacetimes, J. Math. Phys. 21 (1980)
2802.
[36] G.T. Horowitz, D. Marolf, Quantum probes of spacetime singularities, Phys. Rev. D 52 (1995)
5670, gr-qc/9504028.
[37] A. Ishibashi, A. Hosoya, Whos afraid of naked singularities? Probing timelike singularities
with finite energy waves, Phys. Rev. D 60 (1999) 104028, gr-qc/9907009.
[38] A. Brandhuber, K. Sfetsos, Non-standard compactifications with mass gaps and Newtons law,
JHEP 9910 (1999) 013, hep-th/9908116.
[39] O. DeWolfe, D.Z. Freedman, Notes on fluctuations and correlation functions in holographic
renormalization group flows, hep-th/0002226.
[40] G. Arutyunov, S. Frolov, S. Theisen, A note on gravity-scalar fluctuations in holographic RG
flow geometries, hep-th/0003116.
[41] A.G. Cohen, D.B. Kaplan, Solving the hierarchy problem with noncompact extra dimensions,
Phys. Lett. B 470 (1999) 52, hep-th/9910132.
[42] A. Brandhuber, K. Sfetsos, An N = 2 gauge theory and its supergravity dual, hep-th/0004148.
[43] N. Deo, S. Jain, C.-I. Tan, String distributions above the Hagedorn energy density, Phys. Rev.
D 40 (1989) 2626.
[44] N. Deo, S. Jain, O. Narayan, C.-I. Tan, Effect of topology on the thermodynamic limit for a
string gas, Phys. Rev. D 45 (1992) 3641.
[45] G. Dvali, M. Shifman, Surviving on the slope: supersymmetric vacuum in theories where it is
not supposed to be, Phys. Lett. B 454 (1999) 277, hep-th/9901111.
[46] D. Gross, M. Perry, L. Yaffe, Instabilities of flat space at finite temperature, Phys. Rev. D 25
(1982) 330.
[47] H. Davis, Introduction to Nonlinear Differential and Integral Equations, Dover Publications,
New York, 1962.

I. Bakas et al. / Nuclear Physics B 593 (2001) 3175

75

[48] S.A. Abel, J.L.F. Barbon, I.I. Kogan, E. Rabinovici, Some thermodynamical aspects of string
theory, hep-th/9911004;
S.A. Abel, J.L.F. Barbon, I.I. Kogan, E. Rabinovici, String thermodynamics in D-brane
backgrounds, JHEP 9904 (1999) 015, hep-th/9902058.
[49] S.A. Abel, K. Freese, I.I. Kogan, Hagedorn inflation of D-branes, hep-th/0005028.

Nuclear Physics B 593 (2001) 7698

www.elsevier.nl/locate/npe

Open branes in spacetime non-commutative little

string theory
Troels Harmark
The Niels Bohr Institute, Blegdamsvej 17, DK-2100 Copenhagen , Denmark
Received 3 August 2000; accepted 16 October 2000

Abstract
We conjecture the existence of two new non-gravitational six-dimensional string theories, defined
as the decoupling limit of NS5-branes in the background of near-critical electrical two- and threeform RR fields. These theories are spacetime non-commutative Little String Theories with open
branes. The theory with (2, 0) supersymmetry has an open membrane in the spectrum and reduces to
OM theory at low energies. The theory with (1, 1) supersymmetry has an open string in the spectrum
and reduces to (5 + 1)-dimensional NCOS theory for weak NCOS coupling and low energies. The
theories are shown to be T-dual with the open membrane being T-dual to the open string. The theories
therefore provide a connection between (5 + 1)-dimensional NCOS theory and OM theory. We study
the supergravity duals of these theories and we consider a chain of dualities that shows how the
T-duality between the two theories is connected with the S-duality between (4 + 1)-dimensional
NCOS theory and OM theory. 2001 Elsevier Science B.V. All rights reserved.

1. Introduction
Recently, it has been discovered that the world-volume theory of a Dp-brane with a nearcritical electrical NSNS B-field is a spacetime non-commutative open string (NCOS)
theory [1,2]. Subsequently, it was shown that the world-volume theory of the M5-brane
with a near-critical electrical three-form C-field is a non-commutative open membrane
(OM) theory [3,4]. 1 OM theory has been shown [3,4,18] to encompasses all the (p + 1)dimensional NCOS theories with p 6 4, along with their strong coupling duals. In this
sense, we can see OM theory as a unified framework for all these lower dimensional
theories, in much the same way as M-theory can be seen as a unified framework of lower
dimensional string theories. Another close analogy to OM theory is the way that the (5+1)dimensional (2, 0) SCFT encompasses all of the (p + 1)-dimensional YangMills (YM)
theories with p 6 4.
E-mail address: [email protected] (T. Harmark).
1 For related papers about NCOS theory, OM theory and spacetime non-commutativity, see [517].

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 2 1 - 0

T. Harmark / Nuclear Physics B 593 (2001) 7698

77

However, the (5 + 1)-dimensional NCOS theory does not appear to be directly related
to OM theory. If we use the analogy to (2, 0) SCFT and YM theories, we know that the
ultraviolet completion of the (5 + 1)-dimensional YM theory is the (5 + 1)-dimensional
(1, 1) Little String Theory (LST) [19,20] 2 living on the world-volume of type IIB NS5branes. The T-dual of the (1, 1) LST is the (2, 0) LST living on type IIA NS5-branes, and
the low energy limit of this theory is the (2, 0) SCFT. Thus, the two (5 + 1)-dimensional
LSTs provide a relation between (5 + 1)-dimensional YM and (2, 0) SCFT. This also
means that we can consider the LSTs as encompassing both the (2, 0) SCFT and the YM
theories.
In this paper, we find a relation between (5 + 1)-dimensional NCOS and OM theory by
defining two new theories which we call (1, 1) and (2, 0) Open Brane Little String Theories
(OBLSTs). The (1, 1) OBLST is defined as the world-volume theory of N type IIB NS5branes with a near-critical two-form RR-field, and the (2, 0) OBLST is defined as the
world-volume theory of N type IIA NS5-branes with a near-critical three-form RR-field.
The (1, 1) OBLST inherits the closed string from (1, 1) LST but has in addition the open
string of (5 + 1)-dimensional NCOS along with the spacetime non-commutativity, since
the decoupling limit of (1, 1) OBLST is in fact identical to that of (5 + 1)-dimensional
NCOS, as can be seen from type IIB S-duality. The (2, 0) OBLST has also the closed
string of (2, 0) LST and in addition the open membrane of OM theory, again with a
non-commutative geometry. Thus, the OBLSTs have open branes and are spacetime noncommutative. For low energies the (2, 0) OBLST reduces to OM theory, while the (1, 1)
OBLST reduces to (5 + 1)-dimensional NCOS theory for weak NCOS coupling and low
energies. We show that the (1, 1) and (2, 0) OBLST are related by T-duality, in the sense
that a T-duality in one of the open membrane directions in (2, 0) OBLST gives the open
string of (1, 1) OBLST. The (1, 1) and (2, 0) OBLST therefore provide a relation between
(5 + 1)-dimensional NCOS and OM theory, and we can consider them as encompassing
OM theory and all the NCOS theories.
In order to explore the OBLSTs we find their supergravity duals. As part of this we also
find the supergravity dual of OM theory. We subsequently examine the phase structure and
thermodynamics of the supergravity duals. From this, we see that the (1, 1) OBLST only
has an NCOS phase when the NCOS coupling is small. For strong coupling, the closed
string from LST dominates. The (2, 0) OBLST reduces to OM theory at low energies in
the supergravity description, as it should. At high energies the closed string inherited from
LST dominates in both of the OBLSTs, just as for ordinary LST.
We test the consistency of our decoupling/near-horizon limits of the OBLSTs by
connecting five different bound-states and their decoupling/near-horizon limits through
S- and T-dualities. The chain of theories we relate is: OM-theory from M2M5, D =
4 + 1 NCOS from F1D4, D = 5 + 1 NCOS/(1, 1) OBLST from F1D5, D = 5 + 1
NCOS/(1, 1) OBLST from D1NS5 and (2, 0) OBLST from D2NS5. Since the (2, 0)
OBLST from D2NS5 is related directly to OM theory, we have a closed chain of bound
states and limits. Thus, we can start at any point in the chain and then move on to other
2 See also [2123] and see [24] for a brief review of LST.

78

T. Harmark / Nuclear Physics B 593 (2001) 7698

points. The S- and T-dualities are also seen to induce corresponding dualities in the worldvolume theories.
It is important to note that instead of working in terms of decoupling limits we work
mostly with near-horizon limits in this paper. The decoupling limits can easily be read off
from the near-horizon limits. Therefore, when considering a particular near-horizon limit
we also consider this limit as defining the theory which the corresponding near-horizon
supergravity solution is dual to.
2. (1, 1) OBLST and D = 5 + 1 NCOS theory
2.1. Introduction to (1, 1) OBLST
In [2] a new theory was found from the F1D5 bound-state in the decoupling limit
gb ,

gb l 2s = fixed,

critical
B01 B01
,

(1)

where gb is the string coupling, ls the string length and B is the two-form NSNS field. This
theory was subsequently shown to be a (5 + 1)-dimensional theory of open strings, known
as (5 + 1)-dimensional NCOS theory, living in a spacetime geometry with space and time
being non-commutative.
In the following, we shall see that this theory also can be seen as a spacetime noncommutative version of the (1, 1) LST. In fact, using type IIB S-duality we can define the
same theory from the D1NS5 bound-state in the decoupling limit
gb 0,

ls = fixed,

A01 Acritical
,
01

(2)

where gb = 1/gb , ls2 = gb l 2s and A is the RR two-form field. Thus, just like for ordinary
2 =
(1, 1) LST, the low energy gauge theory on D1NS5, which has gauge coupling gYM
3
2
2
2
2
(2) ls , should have a solitonic string of tension (2) /gYM = 1/(2ls ). Since gb = 0 in
the decoupling limit the string cannot leave the brane. In order to study the behavior of this
LST-string, as we will call it in this paper, at higher energies, we turn to the supergravity
dual description of the theory and in particular the thermodynamics computed from this.
As we will explain in the following, for weak NCOS coupling and low energies the (1, 1)
OBLST reduces to what we call D = 5 + 1 NCOS, being a theory of weakly coupled open
strings. Thus, D = 5 + 1 NCOS can be regarded as a low energy limit of (1, 1) OBLST. 3
On the other hand, when the NCOS coupling is large, the LST tension is small, so we
instead have a spacetime non-commutative LST governing the dynamics of the theory.
(1, 1) OBLST reduces to YangMills theory when the effective YangMills coupling is
small.
2.2. The F1D5 and D1NS5 bound states
In this section we give the F1D5 and D1NS5 bound-states so that we can find the
supergravity dual description of (1, 1) OBLST in the next section.
3 This was also discussed in [25].

T. Harmark / Nuclear Physics B 593 (2001) 7698

79

We introduce here the notation that the S-dual string couplings and string lengths are
connected as gb = 1/gb and ls2 = g b l 2s . Our notation for the string couplings are further
clarified in Section 5.
The non-extremal F1D5 bound-state has the string frame metric [25,26]




ds 2 = H 1/2 D f dt 2 + (dx 1 )2 + (dx 2 )2 + + (dx 5 )2


(3)
+ H 1/2 f 1 dr 2 + r 2 d32 ,
the dilaton
e2 = H 1 D,

(4)

and potentials
B01 = sin coth DH 1 ,
A2345 = tan H 1 ,
1
coth (H 1 1),
A012345 =
cos

(5)
(6)
(7)

with B being the NSNS two-form field, A being the RR four-form field and A
being the RR six-form field. We also define
r02
,
r2
r 2 sinh2
,
H =1+ 0 2
r
D 1 = cosh2 sinh2 H 1 .
f =1

(8)
(9)
(10)

We use the two sets of variables , and , related by

sinh2 = cos2 sinh2 ,

cosh2 =

1
cos2

(11)

Using charge quantization of the N D5-branes we get

r02 sinh

p
l2N
gb l 2s N
= s
.
sinh2 + cosh2 =
cosh
cosh

(12)

We now use type IIB S-duality on the F1D5 solution (3)(5). This gives the D1NS5
solution



ds 2 = D 1/2 D f dt 2 + (dx 1 )2 + (dx 2 )2 + + (dx 5 )2


+ H f 1 dr 2 + r 2 d32 ,
(13)
e2 = H D 1 ,

(14)

A01 = sin coth DH 1 ,

A2345 = tan H 1 .

(15)
(16)

80

T. Harmark / Nuclear Physics B 593 (2001) 7698

2.3. Supergravity description of (1, 1) OBLST

The near-horizon limit of the F1D5 bound-state is [2,25]
ls 0,

b
r,
r =
ls

gb l 2s = fixed,

b
r0 =
r ,
ls 0

ls
x = x i ,
b
i

i = 0, 1,

b = l 2s cosh ,

b j
x =
x ,
ls
j

j = 2, . . . , 5,

(17)
(18)

where we use the notation x 0 = t. We have

L2 = r02 sinh2 = gb l 2s N = ls2 N.
We get the near-horizon solution [25]



l 2
r 2
ds 2 = s H 1/2 H 2 f d t 2 + (d x 1 )2 + (d x 2 )2 + + (d x 5 )2
b
L



1
2
2
2
+ H f d r + r d3 ,
gb2 e2 =
B01 =

ls4 r 2
,
b2 L2

l 2s r 2
,
b L2

(19)

(20)
(21)
(22)

with
r02
L2
,
H
=
1
+
.
r 2
r 2
Using S-duality, the near-horizon limit of D1NS5 is

b
ls = fixed,
r =
r,
gb 0,
gb ls

b
r0 ,
b = gb ls2 cosh ,
r0 =
gb ls

gb ls
b j
x j =
x , j = 2, . . . , 5.
x i = x i , i = 0, 1,
gb ls
b
f =1

The limit (24)(25) gives the near-horizon solution 4




r 2
L
H 2 f d t 2 + (d x 1 )2 +(d x 2 )2 + + (d x 5 )2
ds 2 = H 1/2
r
L



1
2
2
2
+ H f d r + r d3 ,

(23)

(24)
(25)

(26)

4 The string metric does not go to zero in our notation since we define the string metric via e instead of g e .
b

T. Harmark / Nuclear Physics B 593 (2001) 7698

gb2 e2 =

b2 L2
,
ls4 r 2

81

(27)

gb ls2 r 2
.
(28)
b L2
We now give the mapping from our supergravity parameters to the parameters of (1, 1)
OBLST. The (1, 1) OBLST lives on a non-commutative spacetime with the commutator
[t, x 1 ] = ib. The energy coordinate u is related to the rescaled radial coordinate r as u =
r /b. As we discuss in the following, the (1, 1) OBLST has three different phases: a weakly
coupled YangMills phase, a weakly coupled NCOS phase and a phase with LST strings.
There are special parameters for each of these phases.
The open string coupling of the (5 + 1)-dimensional NCOS is [2] 5
A01 =

l2
gb l 2s
= s.
(29)
b
b
The tension of the open string is 1/b. The YangMills coupling constant of the (5 + 1)2 = (2)3 gb.
This gives the effective YM coupling
dimensional YangMills theory is gYM
2
3
2

. The LST-string has the tension 1/(2ls2 ).

constant geff = (2) gNbu
We now consider the phase structure of (1, 1) OBLST in terms of phase diagrams with
the energy coordinate u as variable. The three possible phase diagrams for N  1 are
depicted in Figs. 13. For N 1 we have instead only two phase diagrams.
We observe that the supergravity dual of
(1, 1) OBLST reduces to the supergravity dual
of (1, 1) LST given in [27,28] when u  N ls /b.
We consider first gN
 1 which gives the two possible phase diagrams depicted in
2 1, which is equivalent to u
Figs.
1 and 2. We have three transition points. At geff
1/(ls N ), we flow from a perturbative YM description
to a near-horizon D5-brane
description. At gb e 1, which is equivalent to u N/ ls , we go either from a D5
g =

Fig. 1. Phase diagram for (1, 1) OBLST with 1/N  g  1.

Fig. 2. Phase diagram for (1, 1) OBLST with g  1.

5 For convenience we call g = G2 the NCOS open string coupling where G is the NCOS open string coupling
o
o
of [2].

82

T. Harmark / Nuclear Physics B 593 (2001) 7698

to a NS5 description,
or from a delocalized F-string to a delocalized D-string description.

At u L/b = N ls /b we flow from the ordinary (1, 1) LST to (1, 1) OBLST, and we
go either from a D5 to a delocalized F-string description, or from a NS5 to a delocalized
D-string description.
The third possible phase diagram, depicted in Fig. 3, has gN
 1. At energies u 

1/ b we have a weakly coupled (5 + 1)-dimensional NCOS

theory description, which
b we flow to a delocalized
reduces to perturbative YM theory
at
low
energies.
At
u

1/

F-string description and at u N / ls we flow to a delocalized D-string description.

In order to understand these phase diagrams, it is useful first to consider the
thermodynamics of the supergravity description. The near-horizon solutions (20)(22) and
(26)(28) give the leading order thermodynamics [25]
T=

1
,
2ls N

E=

e5
1 2
V
r ,
5
2
(2) b ls4 0

S=

e5

1 2
V
N
r ,
(2)4 b2 ls3 0

(30)

F = 0.

(31)

This thermodynamics
describes (1, 1) OBLST for u  uSG , where uSG = 1/(ls N ) for
 1, since the string corrections to the thermodynamics
gN
 1 and uSG = 1/ b for gN
are small in this region. The Hagedorn temperature of ordinary (1, 1) LST is
TLST =

1
.
2ls N

(32)

So, we have that T TLST for u  uSG . This suggest that the LST-strings dominates
the dynamics of (1, 1) OBLST for u  uSG , since the thermodynamics (30)(31) has the
same leading order Hagedorn behaviour as ordinary (1, 1) LST [2931]. That the LSTstrings live on a spacetime non-commutative geometry can be seen by the fact that the
critical behavior of the entropy at very high energies are different, as shown in [25]. For
ordinary (1, 1) LST we have that [31,32] S(T ) (TLST T )1 while for (1, 1) OBLST
we have [25] S(T ) (TLST T )2/3 .
That the LST-string dominates for u  uSG is not in contradiction with the existence of
open strings in (1, 1) OBLST, as we now explain.
Consider first the case 1/N  g  1 with phase diagram depicted in Fig. 1. This case
corresponds to strongly coupled open strings, since gN

1. Though the LST-strings are
not lighter in this case, there is the other energy scale 1/(ls N ) in LST which is connected
with LST Hagedorn behavior. As suggested in [29,31], this could be a LST-string scale
connected with fractional strings. The LST-modes corresponding to this scale are clearly

Fig. 3. Phase diagram for (1, 1) OBLST with g  1/N .

T. Harmark / Nuclear Physics B 593 (2001) 7698

83

lighter than the open string modes, thus explaining why the LST-string modes dominates

for u  1/(ls N ).
The second case has g  1 which also corresponds to strongly coupled open strings.
The phase diagram is depicted in Fig. 2. We have that 1/b  1/ ls2 thus the LST-strings are
lighter than the open strings and are therefore expected to dominate, which is confirmed
the thermodynamics.
The final case has gN
 1 corresponding to the phase diagram in Fig. 3. Since we
have that 1/b  1/(Nls2 ) the open strings of NCOS theory are much lighter than the
LST-modes connected with LST Hagedorn behaviour and we should therefore expect
them to dominate the dynamics. For the NCOS Hagedorn temperature TNCOS we know

that TNCOS 1/ b. Thus, we have that TNCOS  TLST . Clearly, the NCOS Hagedorn
temperature is not limiting, and we should have a Hagedorn phase transition at a certain

energy uNCOS . Since we have just shown that for u  1/ b we had T TLST , we get that

uNCOS . 1/ b. Thus, the reason that the LST-strings can dominate at high energies, even
though they are heavier than the open strings, is that the open strings have been subject to a
Hagedorn transition at these energies. We note that the lower dimensional NCOS theories
exhibit similar behaviour in the sense that, when they are weakly coupled, the supergravity
dual describes them only when the temperature are above the NCOS Hagedorn temperature
and a Hagedorn phase transition has occurred [25].
In summary, we have learned that at weak NCOS coupling gN
 1 the (1, 1) OBLST

has an NCOS phase for energies u  1/ b, with LST-strings dominating at higher

energies, as depicted in Fig. 3. For strong NCOS coupling gN
 1 we have perturbative
YM for low energies and LST-strings at high energies as depicted in Figs. 1 and 2. Thus,
in this sense one can say that strongly coupled (5 + 1)-dimensional NCOS theory gives a
spacetime non-commutative version of (1, 1) LST.
2.4. Branes in (1, 1) OBLST
In the ordinary (1, 1) LST we have besides the LST-string with tension 1/(2ls2 ) the d0,
d2 and d4 branes [23]. These origins from having open D1, D3 and D5 branes stretching
between NS5-branes.
In the (1, 1) OBLST we still have the LST-string, but now the D-string stretching
between two D1NS5 bound-states induces an open string. We note that the zero modes of
the open D-string is what gives the YangMills theory at low energies, which fits with the
picture that the NCOS theory at low energies reduce to YangMills theory.
Since there is not any electric field on the ends of an open D3-brane stretching between
D1NS5 bound-states we expect that we still have the same d-membrane in (1, 1) OBLST
as in (1, 1) LST, but presumably now moving in a spacetime non-commutative geometry.
Also the d4-brane seem to be part of (1, 1) OBLST.
In other words, only the open D-brane for which the potential goes to its critical value,
gives an open brane in the world-volume theory. The rest of the spectrum is unchanged.

84

T. Harmark / Nuclear Physics B 593 (2001) 7698

3. Supergravity dual of OM theory

In this section we find and study the supergravity dual of OM theory. We find the OM
supergravity dual by uplifting the (4 + 1)-dimensional NCOS supergravity dual. Apart
from being interesting in its own right, it is also important to understand the OM theory
near-horizon limit in order to understand the decoupling and near-horizon limit of (2, 0)
OBLST.
The near-horizon and decoupling limits of the M2M5 bound state have previously been
studied in [3,4,3339].
3.1. The M2M5 and F1D4 bound states
In this section we describe the supergravity solutions that we use for OM theory and
(4 + 1)-dimensional NCOS theory.
The non-extremal M2M5 brane bound-state has the metric [40]





2
bD
b 1/3 f dt 2 + (dx 1 )2 + (dx 2 )2 + D
b (dx 3 )2 + (dx 4 )2 + (dx 5 )2
= H
ds11


b f 1 dr 2 + r 2 d42 ,
(33)
+H
and the three- and six-form potentials
b1 coth ,

C012 = sin H
1
bH
b ,
C345 = tan D
b1

bH
C012345 = cos D

(34)
(35)

1 coth .

(36)

r03 sinh2
,
r3

(37)

We have
f =1

r03
,
r3

b=1+
H

b1 .
b1 = cos2 + sin2 H
D

(38)

The charge quantization for N M5-branes gives

r03 cosh sinh =

Nlp3
.
cos

(39)

In the new variables

sinh2 = cos2 sinh2 ,

cosh2 =

1
cos2

(40)

we have




2
= (H D)1/3 D f dt 2 + (dx 1 )2 + (dx 2 )2 + (dx 3 )2 + (dx 4 )2 + (dx 5)2
ds11


+ H f 1 dr 2 + r 2 d42 ,
(41)
r03 sinh2
,
r3
D 1 = cosh2 sinh2 H 1 .

H =1+

bD
b1 and D = D
b1 .
b = H D 1 , H = H
We note that H

(42)
(43)

T. Harmark / Nuclear Physics B 593 (2001) 7698

85

We now dimensionally reduce the M2M5 on the electric circle with the coordinate x 2 .
This gives the F1D4 bound-state solution. The relation between the eleven dimensional
2 and the ten-dimensional string-frame metric ds 2 and dilaton e is
metric ds11
10
2
2
= e 3 ds10
+ e 3 (dx 2 )2 .
ds11
2

Thus, we get the metric




2
= H 1/2 D f dt 2 + (dx 1 )2 + (dx 3 )2 + (dx 4 )2 + (dx 5 )2
ds10


+ H f 1 dr 2 + r 2 d42 ,

(44)

(45)

the dilaton
e2 = H 1/2 D,

(46)

and the NSNS two-form potential

B01 = sin coth DH 1 .

(47)

This solution coincides with the one given in [25].

3.2. Near-horizon limit of OM theory from D = 4 + 1 NCOS
The near horizon limit of (4 + 1)-dimensional NCOS is [2,25]

b
b
r =
r,
r0 =
r0 ,
ls 0,
ls
ls
= fixed,
b = l 2s cosh ,

ls i
b j
i
j
x =
x , j = 3, 4, 5.
x = x , i = 0, 1,

ls
b

(48)
(49)

We have the open string coupling squared

ga l 2s
.
b
Using (48)(49) on (45)(47) we get [25]



l 2
r 2
2
= s H 1/2 H 2 f d t 2 + (d x 1 )2 + (d x 3)2 + (d x 4 )2 + (d x 5 )2
ds10
b
L



1
2
2
2
+ H f d r + r d4 ,
g =

1/2

ga2 e2

= g H

B01 =

l 2s r 3
,
b L3



r 3
1+ 3 ,
L

(50)

(51)
(52)
(53)

with
3/2,
L3 = r03 sinh2 = N gb

(54)

86

T. Harmark / Nuclear Physics B 593 (2001) 7698

r03
L3
,
f
=
1

.
(55)
r 3
r 3
We now introduce the open membrane length scale lm in OM theory defined by stating
3 . As shown in [3,4] we then have
that the open membrane has tension 1/ lm
H =1+

3
= gb
3/2.
lm

(56)

This can be understood as follows. Since the radius of the electric circle with coordinate
x 2 is RE = ga ls the rescaled radius is
2

a l s
eE = g
= g b,
(57)
R
b

eE = RE ls / b follows from the fact that the x 2 coordinate should

where the rescaling R
scale the same way as the x 1 coordinate in (49). We can then write the relation 6
eE
1 R
= 3 ,
b
lm

(58)

where 1/b is the tension of the open string in NCOS. Thus, the relation (56) is the statement
that the open string in NCOS is the open membrane in OM theory wrapped on a circle of
eE .
radius R
We now want to use (56) to write the near-horizon limit (48)(49) in terms of the eleven
dimensional variables lm and lp , where lp3 = ga l 3s . Using (56) we have

3
3/2 lm
b gb
=
= 3.
(59)
lp
ls
gb
ls
Using this together with (48)(49) we can write the eleven dimensional near-horizon limit
of OM theory as
lp 0,
x i =

lp3
3
lm

r =
xi ,

3
lm
r,
lp3

i = 0, 1, 2,

r0 =

3
lm
r0 ,
lp3

x j =

3
lm
xj ,
lp3

6
lm
= lp6 cosh ,

(60)

j = 3, 4, 5.

(61)

This is a purely eleven dimensional near-horizon limit of OM theory, meaning that it can
describe OM theory with the Lorentz symmetry SO(1, 2) SO(3).
The near-horizon limit (60)(61) is the same limit of the M2M5 brane bound state
as in [35,36]. Keeping r/ lp3 fixed means that the membrane modes for open M2-branes
stretching between M5-branes are kept finite.
Using (60)(61) on (41) and (34)(36) we get the supergravity dual

lp2


r 3
L
2
H 3 f d t 2 + (d x 1 )2 + (d x 2 )2
= 2 H 2/3
ds11
r
lm
L



+ (d x 3 )2 + (d x 4 )2 + (d x 5 )2 + H f 1 d r 2 + r 2 d42 , (62)
6 In these type of relations we ignore factors of 2 .

T. Harmark / Nuclear Physics B 593 (2001) 7698

C012 =

lp3 r 3
,
3 L3
lm

H =1+

L3
,
r 3

C345 =
f =1

lp3
3
lm

H 1 ,

r03
.
r 3

87

(63)

(64)

eE , the energy coordinate u

When OM theory is on an electric circle of rescaled radius R
is
u=

eE r
R
r
= 3 .
b
lm

(65)

3.3. Phase structure of OM and D = 4 + 1 NCOS theory

In this section we examine the phase structure of OM theory and the (4 + 1)-dimensional
NCOS theory via their supergravity duals.
The OM theory near-horizon solution (62)(63) is valid when the curvature in units of
lp2


N r 3 1/3
2
C = N + 3
,
(66)
lm
is small. Thus, if N  1 this is clearly satisfied and we can describe OM theory for all
energies. If N is of order 1, we instead need that r  N 1/3 lm since we need that r  L.
From the supergravity dual (62)(63) we see that for r  N 1/3 lm the solution reduces
to AdS7 S 4 describing the six-dimensional (2, 0) SCFT [41]. Thus, for r  N 1/3 lm OM
theory reduces to (2, 0) SCFT. For r  N 1/3 lm the open membrane is large enough to have
an effect and the underlying non-commutative geometry is detectable. For r  N 1/3 lm the
solution is described by M2-branes delocalized in 3 directions. The phases are depicted in
Fig. 4.
For (4 + 1)-dimensional NCOS the curvature of the supergravity dual in units of l 2
s is
[25]

Fig. 4. Phase diagram of OM theory.

Fig. 5. Phase diagram for D = 4 + 1 NCOS with 1/N  g  1.

88

T. Harmark / Nuclear Physics B 593 (2001) 7698

b 1/2
H
.
(67)
r 2

while for gN
> 1 we need r  b in order for
Thus, for gN
< 1 we need r  b/(gN)
C  1.
Consider first gN
 1. When r L we flow from YM to NCOS with the spacetime
commutator [t, x 1 ] = ib. The (4 + 1)-dimensional NCOS theory flows into OM theory
when ga e 1. This gives two possible phase diagrams, which we have depicted in the
Figs. 5 and 6.
The case with gN
 1 is depicted in Fig. 7. This case corresponds to weakly coupled
NCOS theory. The NCOS theory flows to OM theory at u N 1/3 / lm .
The thermodynamics of the (4 + 1)-dimensional NCOS theory from the supergravity
dual is [25]
p

1 e N gb
r0
3/2 5/2
3
p
,
S=
r0 ,
(68)
V4
T=
4
2
4 N gb
12
g b4
3/2
e4 3
e4 3
1 V
V
5
r

,
F
=

r ,
(69)
E=
0
96 5 g 2 b4
96 5 g 2 b4 0

27 4
e4 T 6 .
(70)
F = 7 N 3 g b V
3
As noted in [25] this thermodynamics is equivalent to that of ordinary (4 + 1)-dimensional
YM for large N and strong t Hooft coupling.
The thermodynamics of OM theory from its supergravity dual is
p

3
1 e Nlm
r0
3
5/2
p
V
,
S=
r0 ,
(71)
T=
5
9
3
4 Nlm
24 5
lm
C=

E=

e5 3
5 V
r ,
9 0
192 6 lm

F =

F =

e5 3
1 V
r ,
9 0
192 6 lm

26 3 3 e 6
N V5 T .
37

Fig. 6. Phase diagram for D = 4 + 1 NCOS with g  1.

Fig. 7. Phase diagram for D = 4 + 1 NCOS with gN

 1.

(72)
(73)

T. Harmark / Nuclear Physics B 593 (2001) 7698

89

We see that the thermodynamics of OM theory is equivalent to that of (2, 0) SCFT for
large N .

4. (2, 0) OBLST and OM theory

4.1. Introduction to (2, 0) OBLST
The (2, 0) OBLST is defined as the D2NS5 bound-state in the decoupling limit
ga 0,

ls = fixed,

A012 Acritical
012 .

(74)

We show in the following that this limit follows both from using the near-horizon/decoupling limit of OM theory found in Section 3.2 and from doing a T-duality on the (1, 1)
OBLST. The (2, 0) OBLST has an LST-string and since (2, 0) OBLST reduces to OM
theory for low energies, it also has an open membrane. The T-duality between the two
OBLSTs is shown to relate the open membrane to the open string of (1, 1) OBLST.
At high energies the LST-string dominates and the thermodynamics has LST Hagedorn
behavior. The R-symmetry is SO(4) for these energies, but at low energies we get OM
theory and the R-symmetry is enhanced to SO(5). This we show using the supergravity
dual of (2, 0) OBLST.
The decoupling and near-horizon limits of the D2NS5 bound state have previously been
studied in [35,42].
4.2. D2NS5 bound state from M2M5 on a transverse circle
The D2NS5 bound-state in type IIA string theory can be considered as an M2
M5 bound-state localized on a transverse circle. Thus, from the M2M5 bound state in
Section 3.1 we get the metric 7



2
ds11
= (H D)1/3 D dt 2 + (dx 1)2 + (dx 2 )2 + (dx 3 )2 + (dx 4 )2 + (dx 5 )2


+ H dz2 + dr 2 + r 2 d32 ,
(75)
with
H =1+

X
n=
2

Nlp3
(r 2 + (z + 2nRT )2 )3/2

D 1 = cosh sinh2 H 1 ,

(76)
(77)

where z is the coordinate of the transverse circle with the asymptotic radius RT .
For r  RT we have
H =1+

1 Nlp3
.
RT r 2

(78)

7 We write only the extremal version of this solution so in comparing with the non-extremal M2M5 solution
(41) and (34)(36) one should use that r03 sinh2 = N lp3 for r0 0.

90

T. Harmark / Nuclear Physics B 593 (2001) 7698

We now consider the z coordinate as the eleven dimensional coordinate. Thus by the
usual M/IIA correspondence we have RT = ga ls and lp3 = RT ls2 . The D2NS5 bound-state
2 and the dilaton e given by the formula
has the string-frame metric ds10
2
2
= e 3 ds10
+ e 3 dz2 .
ds11
2

(79)

This gives the D2NS5 solution




2
= D 1/2 D dt 2 + (dx 1 )2 + (dx 2 )2 + (dx 3 )2 + (dx 4 )2 + (dx 5 )2
ds10


+ H dr 2 + r 2 d32 ,

(80)

e2 = H D 1/2 ,

(81)

A012 = sin DH 1 ,
A345 = tan H 1 .

(82)
(83)

4.3. Supergravity dual of (2, 0) OBLST

The near-horizon limit of the solution (75) is
lp 0,
x i =

lp3
3
lm

r =
xi ,

3
lm
r,
lp3

z =

i = 0, 1, 2,

3
lm
z,
lp3

x j =

6
lm
= lp6 cosh ,

3
lm
xj ,
lp3

j = 3, 4, 5.

(84)

(85)

This we obtained from the OM near-horizon limit (60)(61) since the (2, 0) OBLST limit
should be consistent with the OM theory limit.
The eleven dimensional supergravity dual of (2, 0) OBLST is therefore given by 8
2
ds11
=

lp2
2
lm

C012 =
J=




H 2/3 J 1/3 H J 1 d t 2 + (d x 1)2 + (d x 2 )2 + (d x 3 )2


+ (d x 4 )2 + (d x 5)2 + H d z 2 + d r 2 + r 2 d32 ,
lp3
3
lm

J 1 ,

C345 =

lp3
3
lm

H 1 ,

3
Nlm
,
eT )2 )3/2
(r 2 + (z + 2nR
n=

(86)
(87)

H = 1 + J.

(88)

The limit (84)(85) is precisely consistent with our requirements of a limit of (2, 0)
OBLST. We need that r/RT is finite in the near-horizon limit so since lp3 = ga ls3 and RT =
ga ls we see that the OBLST limit is gs 0 and ls kept fixed. But this is the limit of
8 The construction of the supergravity dual of (2, 0) OBLST presented here is similar to that of the D2NS5
near-horizon solution presented in [42]. The D2NS5 near-horizon solution of [42] describes (2, 0) OBLST on a
magnetic circle. We thank M. Alishahiha for a discussion about this point.

T. Harmark / Nuclear Physics B 593 (2001) 7698

91

ordinary LST so we should have closed strings of tension 1/(2ls2 ) in (2, 0) OBLST. Note
3 / l2 .
eT = lm
that R
s
We observe that the supergravity dual of (2, 0) OBLST given by (86)(87) reduces to
the supergravity dual of (2, 0) LST given in [27,28] when r  N 1/3 lm .
eT we should consider (86)(87) as an approximate description of (2, 0)
For r  R
OBLST which for small r continuously should flow to the OM theory supergravity dual
given by (62)(63).
eT and ga e  1 we can use a ten-dimensional description via weakly
When r  R
coupled type IIA string theory. The limit (84)(85) translates into the limit
ga 0,
x i =

ls = fixed,

ga ls3 i
x,
3
lm

r =

i = 0, 1, 2,

3
lm
r,
ga ls3

x j =

6
lm
= ga2 ls6 cosh ,

3
lm
xj ,
ga ls3

j = 3, 4, 5.

(89)
(90)

The type IIA near-horizon solution is then




L
r 2
2
= H 1/2
H 2 d t 2 + (d x 1 )2 + (d x 2 )2
ds10
r
L




+ (d x 3 )2 + (d x 4 )2 + (d x 5 )2 + H d r 2 + r 2 d32 ,

ga2 e2 =

6
lm
L
H 1/2 ,
6
ls
r

A012 =

ga ls3 r 2
,
3 L2
lm

L2 = ls2 N,

(91)
(92)

A345 =

H =1+

ga ls3 1
H ,
3
lm

L2
.
r 2

(93)
(94)

4.4. T-duality on an electric circle: from open membranes to open strings

3 and that
Since we believe that (2, 0) OBLST has an open membrane of tension 1/ lm
(1, 1) OBLST has an open string of tension 1/b it is natural to ask whether this is consistent
with T-duality. From point of view of the bulk, T-duality on an electric circle in (2, 0)
OBLST would give (1, 1) OBLST, since it takes D2NS5 into D1NS5. In this section
we test that this is also consistent with the decoupling limits. In Section 5 we develop this
further and connect 5 different bound states and their decoupling limits in a duality-chain.
We take x 2 as the coordinate of the electric circle with radius RE . From (84)(85) and
(89)(90) we get
3
eE = ga ls RE ,
R
3
lm

3
eT = lm RT .
R
ga ls3

(95)

Since RT = ga ls we have
3
eT = lm .
R
ls2

(96)

92

T. Harmark / Nuclear Physics B 593 (2001) 7698

A T-duality in x 2 gives
gb = ga

RT
g2 l 4 1
ls
=
= a3 s
.
eE
RE
RE
lm R

(97)

Since ls is fixed in both (1, 1) and (2, 0) OBLST this means that gb ga2 . By comparing
(24)(25) with (89)(90) we see that this is exactly what we need for the decoupling/nearhorizon limits of (2, 0) and (1, 1) OBLST to be consistent with each other. Moreover, we
see that we need

3
b
lm
=
.
(98)
3
gb ls
ga ls
Using (97) and (98) we get that
2 4
3
eE = ga ls = lm .
R
3
gb lm
b

(99)

Thus, the T-duality between the decoupling limits of (1, 1) and (2, 0) OBLST requires that
eE
1 R
= 3 .
b
lm

(100)

This relation means that the open string of (1, 1) OBLST with tension 1/b is the open
eE . Thus, the
membrane of (2, 0) OBLST wrapped around the electric circle of radius R
open membrane in (2, 0) OBLST and the open string in (1, 1) OBLST are related by
T-duality.
We elaborate further on this in Section 5.
4.5. Phase structure and thermodynamics
As already mentioned, the (2, 0) OBLST has both the open membrane with tension
3 as in OM theory, and also the LST-string with tension 1/(2l 2 ). We now consider
1/ lm
s
the phase structure of (2, 0) OBLST. We parameterize the phase diagrams with the rescaled
radial coordinate r . This is not an energy coordinate, but any energy coordinate should be
increasing with r and we can therefore use it to find the succession of transition points.
We consider two possible phase diagrams depicted in Figs. 8 and 9. For both diagrams
eT we have a transition point where for lower energies we have SO(5)
we have that at r R
R-symmetry and for higher energies SO(4) R-symmetry. In the supergravity solution this
eT the radius of the S 3 in the metric
can be understood from the observation that at r R
eT the
eT of the transverse circle. Thus, at r  R
(86) is of the same order as the radius R
supergravity dual of (2, 0) OBLST is in fact the supergravity dual of OM theory, given in
Section 3.2.
eT is
The curvature in units of ls2 of the supergravity dual for r  R
C=

1
1
q
N 1+

r 2
ls2 N

eT the curvature is given by (66). In the following we work with N  1.

For r  R

(101)

T. Harmark / Nuclear Physics B 593 (2001) 7698

93

Fig. 8. Phase diagram for (2, 0) OBLST.

Fig. 9. Phase diagram for (2, 0) OBLST.

We first consider the phase diagram of Fig. 8. For low energies we have (2, 0) SCFT
with SO(5) as the R-symmetry group. This is described by the AdS7 S 4 supergravity
dual. From Section 3.3 we know that at r N 1/3 lm we have OM theory, described by
eT the R-symmetry is broken to SO(4) and we go into
delocalized M2-branes. At r R
(2, 0) OBLST. However, we do not enter the
weakly coupled type IIA description before
6 / l 5 . Thus, we have either delocalized
the transition point ga e 1, which is at r N lm
s
M2-branes or delocalized D2-branes describing the (2, 0) OBLST phase.
In the second phase diagram, depicted in Fig. 9, we also start at low
energies with (2, 0)
3 / l 2 we enter
e
SCFT. We then proceed to the ordinary (2, 0) LST at r RT . At r N lm
s
the weakly
coupled
type
IIA
description
so
that
the
(2,
0)
LST
is
described
by
NS5-branes.

At r N ls we enter the (2, 0) OBLST phase with a non-commutative spacetime.

From a non-extremal version of the metric (86) the thermodynamics of (2, 0) OBLST
eT is found to be
for r  R
e5 1 2

V
1
S= N
r ,
(102)
T=
,
6 0
(2)4 ls lm
2ls N
E=

e5
1 2
V
r ,
5
2
6 0
(2) ls lm

F = 0.

(103)

The Hagedorn temperature of (2, 0) OBLST is

1
(104)
TLST =
.
2ls N
eT we have T TLST so that the LST-strings dominate for
Thus, we see that for r  R
e
r  RT . As discussed in [2931] the thermodynamics (102)(103) exhibits leading order
Hagedorn behavior, and one can calculate [43] that since the supergravity dual consist
of delocalized D2-branes in the UV-region, the entropy has the critical behavior S(T )
(TLST T )2/3 , just as for (1, 1) OBLST. Thus, the critical behavior of the entropy for
high energies in (2, 0) OBLST is different from that in (2, 0) LST.
From comparing the phases and thermodynamics of (1, 1) OBLST and (2, 0) we see
that there are many similarities, as one would expect from T-dual theories. The LST-strings

94

T. Harmark / Nuclear Physics B 593 (2001) 7698

dominate the thermodynamics for high energies in both cases, and the open string and open
membrane only appears as phases in the phase diagrams when they are sufficiently light.
4.6. Branes in (2, 0) OBLST
In the ordinary (2, 0) LST we have besides the LST-string with tension 1/(2ls2 ) the d1-,
d3- and d5-branes [23]. These origins from having open D2-, D4- and D6-branes stretching
between NS5-branes.
The (2, 0) OBLST still have the LST-string, but the D-membrane stretching between two
D2NS5 bound-states induces an open membrane which gives OM theory at low energy.
At low energies this open membrane reduces to a d1-brane in (2, 0) LST, or to a tensionless
string in (2, 0) SCFT. Using similar arguments as for (1, 1) OBLST, we expect that the d3and d5-branes are part of (2, 0) OBLST.
Thus, we see that the brane-spectra of (1, 1) and (2, 0) OBLST are related by T-duality.

5. A duality-chain of theories
In this section 9 we systematically explore how the T-dualities and S-dualities connects
the various bound-states and their decoupling limits that we have been discussing in
Sections 24, in order test the consistency of these limits and also to relate the parameters
of the theories. Some of the discussion has already appeared in earlier sections, but we
repeat it here for clarity. The duality-chain is depicted in Fig. 10.
All of the bound-states in the chain can be seen as the M2M5 bound-state on an electric
circle and a transverse circle. In eleven dimensions the M2M5 bound-state on a transverse
circle is given by (75)(76). We take x 2 to be the coordinate for the electric circle of radius
RE and z to be the coordinate for the transverse circle with radius RT .
For all the various decoupling limits we have

z = cosh z,
(105)
cosh ,
r = cosh r,

1
x j = cosh x j , j = 3, 4, 5,
(106)
x i , i = 0, 1, 2,
x i =
cosh
which gives

eT = cosh RT .
eE = 1
R
(107)
RE ,
R
cosh
Thus, all the decoupling limits are specified by the relation between cosh and the
parameters of M/string theory, and the parameters of the world-volume theories.
Thus, starting from the top with the M2M5 bound-state, we have
cosh =

6
lm
,
lp6

3 is the tension of the open membrane in OM theory and (2, 0) OBLST.

where 1/ lm
9 The content of this section was developed in collaboration with N.A. Obers.

(108)

T. Harmark / Nuclear Physics B 593 (2001) 7698

95

Fig. 10. The chain of theories and bound-states related by S- and T-dualities. SA (E) and SA (T )
means the type IIA S-duality in the electrical and transverse direction, respectively. T (E) and T (T )
means a T-duality in the electrical and transverse direction, respectively. SB means type IIB S-duality.

Choosing x 2 as the eleventh direction we go to the F1D4 bound-state and we have

RE = ga ls ,

lp3 = ga l 3s ,

cosh =

b
,
l 2s

g 4 =

ga l 2s
,
b

(109)

with the NCOS open string coupling g4 and tension 1/b. This gives the relations

eE = g4 b,
R

eE
1 R
= 3 .
b
lm

(110)

Thus, as already mentioned in Section 3.2, this is interpreted [3,4] as the fact that the NCOS
open string in 4 + 1 dimensions is an open membrane in 5 + 1 dimensions wrapped on the
eE , and for strong coupling the electric circle is large and we flow
electric circle of radius R
into decompactified OM theory.
Making a T-duality in the transverse direction z we go to the F1D5 bound-state. We
have
RE
b
gb l 2s
ls
,
(111)
=
,
cosh = 2 ,
g5 =
gb = g a
RT
RT
b
l s
where g5 is the (5 + 1)-dimensional NCOS open string coupling. We also define the T-dual
e0 . This gives
radius RT0 = l 2s /RT along with its rescaled version R
T

96

T. Harmark / Nuclear Physics B 593 (2001) 7698

g5 = g4

b
,
eT
R

eT0 R
eT = b.
R

(112)

We see that this can be interpreted as a NCOS T-duality (similar interpretations for other
cases have been done in [4,18]). Thus, the bulk T-duality on the F1Dp bound states
induces a world-volume T-duality in the NCOS theories relating the T-dual string couplings
and radii by the NCOS string tension 1/b.
The type IIB S-duality takes us into the D1NS5 bound-state for which we have
gb =

1
,
gb

ls2 = gb l 2s ,

cosh =

b
,
gb ls2

g 5 =

ls2
,
b

(113)

as explained in Section 2.3.

If again start from the top and instead choose z as the eleventh direction we go from the
M2M5 to the D2NS5 bound-state. We have
RT = ga ls ,

lp3 = ga ls3 ,

cosh =

6
lm
.
ga2 ls6

(114)

eE .
Here the world-volume theory is (2, 0) OBLST with the parameters ls , lm and R
Making a T-duality in the x 2 direction we obtain the D1NS5 bound-state. As already
explained in Section 4.4, we have
gb = ga

RT
ls
=
,
RE
RE

cosh =

3
lm
.
eE
gb ls2 R

(115)

We again note that cosh exactly has the right dependence on the string coupling gb that
makes us able to compare this T-dualized limit to the limit obtained in (113) by going the
0 = l 2 /R of the electric circle.
other way in the chain. We now define the T-dual radius RE
E
s
0 scales oppositely to R . Using
Since x 2 after the T-duality is a magnetic coordinate, RE
E
this and comparing with (113) we therefore get
eE
1 R
= 3 ,
b
lm

0 e
eE
RE = ls2 .
R

(116)

Thus, we see that the T-duality in the bulk now induces a little-closed-string-T-duality in
the OBLST, since contrary to (112) the T-duality is in terms of the length scale ls of the
closed strings in OBLST. One could speculate that this means that open strings of tension
1/(2ls2 ) can end on the open string/membrane in OBLST. Certainly, this would make
sense from the bulk point of view, since the open string/membrane origins from D1- or
D2-branes stretching between the D1NS5 or D2NS5 bound-states.

Note added
While writing this paper, we received the paper [44] which also considers a supergravity
dual of OM theory using a different approach than the one we have in Section 3.
In the final stages of writing this paper we were made aware that N. Seiberg and
A. Strominger have presented similar ideas as of this paper in their talks at Strings 2000,

T. Harmark / Nuclear Physics B 593 (2001) 7698

97

July 1015. We were subsequently informed that these ideas will appear in a revised
version of the paper [3] by R. Gopakumar, S. Minwalla, N. Seiberg and A. Strominger.

Acknowledgements
We thank N.A. Obers for early participation and useful discussions and M. Alishahiha,
D. Berman, E. Cheung, J. Correia, S. Minwalla, J. Nielsen and J.L. Petersen for useful
discussions.

References
[1] N. Seiberg, L. Susskind, N. Toumbas, Strings in background electric field, space/time
noncommutativity and a new noncritical string theory, JHEP 06 (2000) 021, hep-th/0005040.
[2] R. Gopakumar, J. Maldacena, S. Minwalla, A. Strominger, S-duality and noncommutative
gauge theory, JHEP 06 (2000) 036, hep-th/0005048.
[3] R. Gopakumar, S. Minwalla, N. Seiberg, A. Strominger, OM theory in diverse dimensions,
JHEP 08 (2000) 008, hep-th/0006062.
[4] E. Bergshoeff, D.S. Berman, J.P. van der Schaar, P. Sundell, Critical fields on the M5-brane and
noncommutative open strings, hep-th/0006112.
[5] N. Seiberg, L. Susskind, N. Toumbas, Space/time non-commutativity and causality, JHEP 06
(2000) 044, hep-th/0005015.
[6] O.J. Ganor, G. Rajesh, S. Sethi, Duality and non-commutative gauge theory, hep-th/0005046.
[7] J.L.F. Barbon, E. Rabinovici, Stringy fuzziness as the custodian of timespace noncommutativity, Phys. Lett. B 486 (2000) 202, hep-th/0005073.
[8] J. Gomis, T. Mehen, Space-time noncommutative field theories and unitarity, hep-th/0005129.
[9] D.J. Gross, N.A. Nekrasov, Monopoles and strings in noncommutative gauge theory, JHEP 07
(2000) 034, hep-th/0005204.
[10] G.-H. Chen, Y.-S. Wu, Comments on noncommutative open string theory: V-duality and
holography, hep-th/0006013.
[11] I.R. Klebanov, J. Maldacena, (1 + 1)-dimensional NCOS and its U (N) gauge theory dual, hepth/0006085.
[12] J.X. Lu, S. Roy, H. Singh, ((F, D1), D3) bound state, S-duality and noncommutative open
string/YangMills theory, JHEP 09 (2000) 020, hep-th/0006193.
[13] J.G. Russo, M.M. Sheikh-Jabbari, On noncommutative open string theories, JHEP 07 (2000)
052, hep-th/0006202.
[14] J. Gomis, K. Kamimura, J. Llosa, Hamiltonian formalism for spacetime non-commutative
theories, hep-th/0006235.
[15] O. Aharony, J. Gomis, T. Mehen, On theories with light-like noncommutativity, JHEP 09 (2000)
023, hep-th/0006236.
[16] S.-J. Rey, R. von Unge, S-duality, noncritical open string and noncommutative gauge theory,
hep-th/0007089.
[17] R.-G. Cai, N. Ohta, (F1, D1, D3) bound state, its scaling limits and SL(2, Z) duality, hepth/0007106.
[18] T. Kawano, S. Terashima, S-duality from OM-theory, hep-th/0006225.
[19] M. Berkooz, M. Rozali, N. Seiberg, Matrix description of M theory on T 4 and T 5 , Phys. Lett.
B 408 (1997) 105110, hep-th/9704089.
[20] N. Seiberg, New theories in six-dimensions and matrix description of M theory on T 5 and
T 5 /Z2 , Phys. Lett. B 408 (1997) 98104, hep-th/9705221.

98

T. Harmark / Nuclear Physics B 593 (2001) 7698

[21] R. Dijkgraaf, E. Verlinde, H. Verlinde, BPS quantization of the five-brane, Nucl. Phys. B 486
(1997) 89113, hep-th/9604055.
[22] R. Dijkgraaf, E. Verlinde, H. Verlinde, 5d black holes and matrix strings, Nucl. Phys. B 506
(1997) 121, hep-th/9704018.
[23] A. Losev, G. Moore, S.L. Shatashvili, M and ms, Nucl. Phys. B 522 (1998) 105124, hepth/9707250.
[24] O. Aharony, A brief review of little string theories, Class. Quant. Grav. 17 (2000) 929, hepth/9911147.
[25] T. Harmark, Supergravity and spacetime non-commutative open string theory, JHEP 07 (2000)
043, hep-th/0006023.
[26] J.X. Lu, S. Roy, Non-threshold (F, Dp) bound states, Nucl. Phys. B 560 (1999) 181, hepth/9904129.
[27] N. Itzhaki, J.M. Maldacena, J. Sonnenschein, S. Yankielowicz, Supergravity and the large N
limit of theories with sixteen supercharges, Phys. Rev. D 58 (1998) 046004, hep-th/9802042.
[28] O. Aharony, M. Berkooz, D. Kutasov, N. Seiberg, Linear dilatons, NS5-branes and holography,
JHEP 10 (1998) 004, hep-th/9808149.
[29] J.M. Maldacena, Statistical entropy of near extremal five-branes, Nucl. Phys. B 477 (1996)
168174, hep-th/9605016.
[30] J.M. Maldacena, A. Strominger, Semiclassical decay of near-extremal fivebranes, JHEP 12
(1997) 008, hep-th/9710014.
[31] T. Harmark, N.A. Obers, Hagedorn behaviour of Little String Theory from string corrections to
NS5-branes, Phys. Lett. B 485 (2000) 285, hep-th/0005021.
[32] M. Berkooz, M. Rozali, Near Hagedorn dynamics of NS fivebranes, or a new universality class
of coiled strings, JHEP 05 (2000) 040, hep-th/0005047.
[33] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 09 (1999) 032, hepth/9908142.
[34] J.M. Maldacena, J.G. Russo, Large N limit of non-commutative gauge theories, JHEP 09 (1999)
025, hep-th/9908134.
[35] M. Alishahiha, Y. Oz, M.M. Sheikh-Jabbari, Supergravity and large N noncommutative field
theories, JHEP 11 (1999) 007, hep-th/9909215.
[36] T. Harmark, N.A. Obers, Phase structure of non-commutative field theories and spinning brane
bound states, JHEP 03 (2000) 024, hep-th/9911169.
[37] S. Chakravarty, K. Dasgupta, O.J. Ganor, G. Rajesh, Pinned branes and new non Lorentz
invariant theories, hep-th/0002175.
[38] E. Bergshoeff, D.S. Berman, J.P. van der Schaar, P. Sundell, A noncommutative M-theory fivebrane, hep-th/0005026.
[39] S. Kawamoto, N. Sasakura, Open membranes in a constant C-field background and noncommutative boundary strings, JHEP 07 (2000) 014, hep-th/0005123.
[40] J.G. Russo, A.A. Tseytlin, Waves, boosted branes and BPS states in M-theory, Nucl. Phys.
B 490 (1997) 121144, hep-th/9611047.
[41] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231252, hep-th/9711200.
[42] M. Alishahiha, On type II NS5-branes in the presence of an RR field, hep-th/0002198.
[43] J. Correia, T. Harmark, N.A. Obers, in preparation.
[44] D.S. Berman, P. Sundell, Flowing to a noncommutative (OM) five brane via its supergravity
dual, JHEP 10 (2000) 014, hep-th/0007052.

Nuclear Physics B 593 (2001) 99126

www.elsevier.nl/locate/npe

On supersymmetry breaking in string theory and its

realization in brane worlds
P. Mayr
CERN Theory Division, CH-1211 Geneva 23, Switzerland
Received 26 July 2000; accepted 11 September 2000

Abstract
We use string duality to describe instanton induced spontaneous supersymmetry breaking in string
compactifications with additional background fields. Dynamical supersymmetry breaking by space
time instantons in the heterotic string theory is mapped to a tree level breaking in the type II string
which can be explicitly calculated by geometric methods. The point particle limit describes the
non-perturbative scalar potential of a SYM theory localized on a hypersurface of spacetime. The
N = 0 vacuum displays condensation of magnetic monopoles and confinement. The supersymmetry
breaking scale is determined by Mstr , which can be in the TeV range, and the geometry transverse to
the gauge theory. 2001 Elsevier Science B.V. All rights reserved.

1. Introduction and summary

The remarkable success in the understanding of non-perturbative properties of string
theories triggered by string dualities cannot conceal the fact that their impact on the
formulation of theories with N < 2 supersymmetry and dynamical supersymmetry
breaking has been much less effective. In fact the step to N = 1 supersymmetry may be
already a vital one for the description of the non-supersymmetric world as it is known
that dynamical supersymmetry breaking may appear in these theories as a consequence of
strong gauge interactions [1]. It is interesting that supersymmetry breaking in the string
theory might be so closely related to its low energy sector. In fact in the realization of
confinement as result of monopole condensation [2,3], anomaly considerations imply a
non-zero gaugino condensate [4] which breaks supersymmetry in the theory coupled to
gravity [1]. In this sense the lack of supersymmetry observed in our world could be linked
in an intriguing way to the existence of confining gauge theories.
One of the most useful N = 2 dualities has been the one between type II and heterotic
strings [5,6]. As the heterotic coupling maps to a geometric modulus in the type II theory,
E-mail address: [email protected] (P. Mayr).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 5 2 - 6

100

P. Mayr / Nuclear Physics B 593 (2001) 99126

non-perturbative effects of the heterotic theory are related to tree level of the type II string.
In particular spacetime instanton effects map to geometric instantons in the type II theory
which can be calculated by mirror symmetry. More fundamentally the instantons in the
type II theory can be viewed as honest spacetime instantons in the RR-sector, which has
the special property to not depend on the type II string coupling and thus is described by
string tree level.
This opens the fascinating possibility that after a supersymmetry breakdown to
N = 1, dynamical supersymmetry breaking to N = 0 in the RR-sector of the type II
theory is calculable as the non-perturbative effects are still governed by a geometric
coupling constant and not the string coupling. In particular condensation of fermions
will appear at string tree level. This non-supersymmetric type II theory will be dual
to a N = 1 supersymmetric heterotic theory where supersymmetry is broken only in
the non-perturbative sector of the string theory. Such a picture was advocated in [7]
using the special constructions of [8] to obtain N = 1 dual pairs from N = 2 ones by
freely acting orbifolds. 1 Based on qualitative arguments on the general properties of the
orbifolding procedure it was argued that indeed gaugino condensation in the heterotic
theory should map to a tree level supersymmetry breaking in the type II theory which,
remarkably enough, was related to monopole condensation and the associated realization
of confinement. Unfortunately the orbifold duals of [7] are not accessible to instanton
calculations by mirror symmetry and thus the study of string duals with supersymmetry
breaking has been restricted to a rather qualitative level, far from the success in the N = 2
supersymmetric case. To provide a calculable framework of string duals with dynamical
supersymmetry breaking is one of the goals of this paper.
In a seemingly unrelated development in the supergravity part of the world, the no-go
theorem [10] on partial local supersymmetry breaking N = 2 N = 1 was deprived its
validity [11]. The argument is that for a special class of supergravity theories, the tensor
calculus of Ref. [12], which was the basis for the derivation of the no-go theorem, is
inappropriate. In these special cases, the no-go theorem does not apply and spontaneous
partial supersymmetry breaking has indeed be shown to be possible [11]. In string theory
compactifications, scalar potentials that may trigger spontaneous supersymmetry breaking
can be induced by RR background fields in the type II string [13] or magnetic backgrounds
in the dual heterotic theory [14]. However, as we will argue in this paper, partial breaking
fails to exist, although in an interesting way. The reason is that even though the conditions
of partial supersymmetry breaking may be satisfied in the classical string effective theory,
there will always be instanton effects in compactifications with finite volume, that restore
the foundations of the no-go theorem. We will also argue that the effect of these instantons
is to break supersymmetry completely, rather then to restore N = 2.
An interesting picture begins to emerge by observing that the classical string effective
theories associated to N = 2 type II/heterotic duals fall precisely into the class of
supergravities that satisfy the necessary conditions for partial supersymmetry breaking.
1 The relation between spontaneous supersymmetry breaking in string theory and freely acting orbifolds with
various supersymmetries has been studied in Ref. [9].

P. Mayr / Nuclear Physics B 593 (2001) 99126

101

Would it not be for the instanton effects, we would get N = 1 supersymmetric dual pairs
in this way. Turning to the instanton corrections one observes that they represent nonperturbative spacetime effects in the heterotic string and tree level geometric instantons
in the type II theory. In other words, the instanton effects that restore the no-go theorem
embody precisely the sought for description of non-perturbative dynamical supersymmetry
breaking. Using mirror symmetry in the tree level type II theory to determine the geometric
instantons we are thus in the remarkable situation to be able to systematically calculate
the instanton effects of a phenomenologically relevant non-perturbative supersymmetry
breaking.
Interestingly, and in full agreement with the observations made in the free orbifold
models of [7], the supersymmetry breaking is intimately related to strong gauge
interactions in a SYM theory embedded into the string theory. In fact this theory is of
the world on a brane type, with the gauge fields localized on a hypersurface of tendimensional spacetime. These theories have been very successfully studied in the N =
2 context in terms of curved brane geometries in string theory, either in the geometric
engineering of type II strings on singularities of CalabiYau 3-folds [1517] or, for simple
enough gauge group G, in terms of the T-dual type IIA 5-brane [15]. The latter realization
can also be reached starting from field theories living on the world-volume of flat branes
which are bent by quantum corrections in a way determined by the non-perturbative duality
to M-theory [18]. Both, the flat and the curved brane constructions, give a string theory
realization of the intensively discussed extra large dimension scenarios [19]. In particular
the flat branes have become a most popular tool in qualitative phenomenological studies
due to their conceptional simplicity as compared to the geometric approach.
So finally, taking the point particle limit of the eventually non-supersymmetric dual pair,
leads us right to an extension of geometric engineering 2 to more realistic SYM theories on
the brane, with broken supersymmetry and patterns of confinement and chiral symmetry
breaking. 3 Using the results from mirror symmetry in the type II theory allows to derive
the explicit non-perturbative scalar potential in the SYM theory on the brane.
It is worth noting that the description of supersymmetry breaking in the SYM theory
on the brane is quite relevant for the large dimension scenarios one of whose primary
motivations has been the hierarchy problem. A serious attempt to explain the apparent
weakness of gravity should be based on a framework of quantum gravity such as string
theory, and not just field theories on flat branes. Note that the resurrection of the no-go
theorem implies that the complete breakdown to N = 0 (rather than N = 1) is inevitable
when the SYM is coupled to string theory. It would be quite interesting to develop the
understanding of the field theories based flat branes with N = 1 supersymmetry [21] to
a level where the supersymmetry breakdown can be seen, at least qualitatively. The reason
that the flat branes miss this effect is apparently the lack of a satisfying description of the
coupling of the world volume theory to gravity and string theory.
2 A different approach uses F-theory compactifications [20].
3 In the following we will use the definition of a brane in the general sense of specifying a hypersurface on

which the gauge fields localize. This notion is independent of the specific string theory construction in terms of
dual D-branes, M-branes or geometrically engineered theories at a singularity.

102

P. Mayr / Nuclear Physics B 593 (2001) 99126

Note that we argued that the restoration of the no-go theorem relies on stringy instantons
on a finite volume manifold. This does not exclude partial supersymmetry breaking
in diverse non-compact limits. In particular, taking all, transverse as well as tangential
dimensions of infinite volume gives a pure field theory decoupled from strings and gravity.
It should be clear that neither there is a no-go theorem in pure field theory nor do we claim
so. 4 In fact, while we completed this work, an interesting paper appeared [23], which also
describes supersymmetry breaking by RR fluxes in the context of geometric engineering.
In Ref. [23] the limit of infinite string mass Mstr is taken which results in a pure field
theory decoupled from string theory where partial breaking is possible. Moreover this
reference contains an elegant derivation of how the relevant N = 1 superpotentials follow
from soliton masses in the presence of RR potentials.
This paper is organized in two parts. In the first part we consider the issue of partial
supersymmetry breaking in the string effective supergravity. In Section 2 we argue that for
string compactifications on manifolds with finite volume, any classical partial breaking to
N = 1 is followed by a further breakdown to N = 0 by instanton effects. We resurrect
the general no-go theorem on partial breaking for string effective theories. In Section 3
we relate these results to the class of string effective theories which reduce to ordinary
super-YangMills theories in the point particle limit. The instantons that are relevant for
the supersymmetry breakdown turn out to be non-perturbative in the field theory coupling.
In Section 4 we describe in more detail the mechanism of supersymmetry breaking and
the calculation of the non-perturbative gravitino masses by mirror symmetry. Moreover we
observe a general link between the breakdown of supersymmetry and that of conformal
invariance.
In the second part, which starts with Section 5 and can be read almost independently, we
focus further on the pattern of supersymmetry breaking in the point particle limit, namely
in the SYM on the brane embedded in the string theory. The general non-perturbative
scalar potential contains apart from the N = 1 adjoint mass term extra soft breaking terms
that arise from the coupling to the string sector. We use special geometry to proof that these
terms are mandatory and there is no partial breaking to N = 1 in the field theory coupled to
string theory. The scale of supersymmetry breaking terms is determined by the string scale
1
,
Mstr , which is argued to be favorably in the TeV range, multiplied by a factor b1 VTV
where b1 is the one-loop beta function coefficient of the SYM theory and VTV the volume
of the dimensions transverse to the gauge theory. Details on the supergravity calculations
are relegated to the appendices.

2. Partial supersymmetry breaking in N = 2 supergravity

( or Two into one still wont go)
Before focusing on the string theory embeddings of SYM theories let us ask in general
what patterns of supersymmetry breaking may appear in the N = 2 string effective theory.
4 E.g., gaugino condensation does not break supersymmetry in field theory [22].

P. Mayr / Nuclear Physics B 593 (2001) 99126

103

As far as partial local supersymmetry breaking by a super-Higgs effect is concerned, there

used to be a very strong result, the Two into one wont go theorem of [10]. It states that
a zero eigenvalue of the gravitino mass matrix implies that the second one is zero, too.
However, the no-go theorem is based on the existence of a holomorphic prepotential
F (zi ) that defines the effective N = 2 supergravity action of the nV vector multiplets. At
that time this was indeed the only known description of the N = 2 supergravity theory
[12]. Since then, an alternative definition has been formulated which is not based on a
prepotential [24]. It thus appeared that the absence of a prepotential might possibly allow
to evade the no-go theorem and indeed it was shown in [11] that partial supersymmetry
breaking is possible in this special situation. The globally supersymmetric version of this
partial breaking has been described independently in [25].
The necessary condition for the absence of a prepotential and thus partial supersymmetry
breaking is the following one. The framework of [24] starts from a section = (X , F )T
of a Sp(2nV +2, Z) bundle over MV . It is invariant under symplectic transformations acting
on :
M,

M Sp(2n + 2, Z).

(2.1)

The components X , F of the section , also called periods, depend on the nV scalar
components zi of the vector multiplets which parametrize an nV dimensional special
Khler manifold MV . In a generic situation the upper components X can be thought
of as homogeneous coordinates on MV , while the lower half F of is related to a
prepotential F (X ) of homogeneous degree two by F = F /X . The transition to the
generic inhomogeneous variables zi on MV is described by the matrix
Ai =

X
.
zi

(2.2)

For a special form of the prepotential F and a special choice of the section , the
matrix Ai may be degenerate. Then the X cannot serve as homogeneous coordinates
on MV and no prepotential F (X ) exists. Note that this statement is not invariant under
e = M one may
Sp(2n + 2, Z) transformations and by choosing a different section

e(X
e ) exists.
e
always transform to homogeneous coordinates X in which a prepotential F
A string effective supergravity related to the geometric type II compactification on a
CalabiYau manifold M is classically described by a cubic prepotential
F=

2 1
1
Cij k Xi Xj Xk /X0 = X0
Cij k ti tj tk ,
3!
3!

(2.3)

where ti = Xi /X0 are so-called special coordinates that parametrize volumes of homology
2-cycles in M. Depending on the intersection matrix Cij k of M it may be possible to choose
a section where the X are dependent, the matrix Ai is degenerate and partial breaking
to N = 1 supersymmetry is possible. 5
However in a finite volume there are stringy instanton corrections to the classical
prepotential (2.3) that change this picture. The exact instanton corrected prepotential can be
5 We will argue that this happens for string theory embeddings of SYM theories in the next section.

104

P. Mayr / Nuclear Physics B 593 (2001) 99126

determined [26] using mirror symmetry. 6 Specifically, the components of are identified
under mirror symmetry with the period integrals of the holomorphic 3-form 3,0 on the
mirror manifold W of M:
Z
Z

(3,0)

(W ),
F =
(3,0)(W ).
(2.4)
X =
A

Here {A } denotes an integral basis of homology 3-cycles and {B } their duals. The period
integrals F have the large ti expansion
F =

1
Cij k tj tk + f (tn , qn ),
2!

(2.5)

where q n = exp(2itn ) are the instanton corrections suppressed by the exponential of the
Khler volumes of 2-spheres.
We want to argue now that even in a situation where the geometrical data Cij k allow
for a partial supersymmetry breaking to N = 1, the stringy instanton corrections always
lead to a complete breakdown of supersymmetry. To do so we have to assure that after
b of any section
b gives good
the inclusion of the instanton corrections, the upper part X
homogeneous coordinates on MV . By the relation (2.4) this is the same as asking whether
for any choice of a basis of A-cycles in (2.4), the period integrals represent locally good
homogeneous coordinates on MV . Luckily this type of question is a prominent one in
algebraic geometry and is part of the so-called infinitesimal Torelli problem. In the present
context it has been shown in Ref. [29] that the period integrals indeed have the required
property.
In conclusion, although partial supersymmetry breaking may appear to be possible in
the classical string theory, there will be always stringy instanton corrections that lead to a
complete breakdown of supersymmetry and the mechanism of Ref. [11] cannot be realized
in the string effective theory.

3. Localized SYM theories and their string theory embedding

Form the point of physics, the situation that the classical section allows for partial
supersymmetry breaking while the instanton corrected one does not is quite a remarkable
one. Specifically the supersymmetry breaking scale will depend non-perturbatively on the
Khler moduli. We will argue now that this is the case in the class of string effective
theories realizing SYM theories localized on a six-dimensional hypersurface of space
time. As we will discuss in the following, the relevant modulus is identical to the field
theory coupling constant and the supersymmetry breakdown is thus interpreted as a nonperturbative breaking by spacetime instantons.
6 For an overview and references, see [27]. For a recent proof in the supersymmetric sigma model, see [28].

P. Mayr / Nuclear Physics B 593 (2001) 99126

105

3.1. Geometry and couplings

The geometric engineering approach [1517] starts from a type II string compactified
on an almost singular CalabiYau manifold X. 7 It is useful to think about this
compactification as a two step process of a compactification to six dimensions on a singular
two complex-dimensional manifold Y followed by a further compactification on the base
sphere B to four dimensions. At the singular point p of Y there are r = rk G small 2spheres Ci with an intersection matrix equal to the negative of the Cartan matrix CijG . D2branes wrapped on the 2-spheres Ci represent N = 2 vector multiplets in six dimensions
with a gauge kinetic term (in a tree level approximation)
Lkin =

1
(j )
Cij F (i) F
.
4

(3.1)

In particular the effective action is independent of the directions transverse to the

singularity, as is expected from the fact that the gauge fields are localized on the sixdimensional hypersurface specified by p Y . This gives a natural string theory realization
of the (possibly large) extra dimension scenario of [19].
Upon compactification on the sphere B one obtains an N = 2 supersymmetric QFT
with gauge group G in four dimensions. Dimensional reduction implies that the fourdimensional tree level gauge coupling is given by the (complexified) Khler volume VB
of B:
Z
4i
F T
+ 2 = BNS + i VB s.
(3.2)
2
gF T
B

The real part of s describes the integral of the NeveuSchwarz B-field on the base 2-sphere
B.
At this point it seems useful to fix the notation for the coupling constants. In the
following we will refer to the field theory coupling constant s as the coupling constant
and use perturbative with respect to the the field theory which is designed to describe
our world. The string theory embedding with s identical to the string coupling constant is
a heterotic string on K3 T 2 . It is dual [5,6] to the geometric type II compactification
where s is the geometric volume (3.2) and not the string coupling. This is the usual
story used in geometric engineering: we can calculate non-perturbative phenomena of
the effective field theory by using the geometric type II picture. In the present paper
we extend this correspondence to also calculate non-perturbative quantities of the string
effective supergravity theory to which the SYM theory is coupled.
3.2. The string effective supergravity
The four-dimensional tree level coupling (3.2) fixes the classical piece of the prepotential
to be
7 For a basic introduction and further references, see [30].

106

P. Mayr / Nuclear Physics B 593 (2001) 99126

F = s Cij ti tj + ,

(3.3)

where the dots denote 0 and finite coupling corrections. It is easy to see that the string
effective supergravity obtained from geometric engineering has precisely the property to
allow for a partial supersymmetry breaking. In the inhomogeneous coordinates ti , s, the
standard section = (X , F /X ) is


X
F = 2F sFs
tk Fk , Fs , Fi ,
(3.4)
X = (1, s, ti ),
k

where subscripts denote differentiation. From the prepotential (3.3) one obtains after a
e with
simple symplectic transformation (s, Fs ) (Fs , s) a section

T
X
e = 2F sFs
e = (1, Fs , ti )T ,
F
tk Ftk , s, Fi .
(3.5)
X
k

In particular
F
= Ckl tk tl .
(3.6)
s
e do not depend on s and thus the necessary condition for partial
Note that the X
supersymmetry breaking is satisfied.
However, as we argued on general grounds in Section 2, instanton corrections will
change this picture. In the present situation the relevant instantons have an action e2is
which is characteristic of spacetime instantons as interpreted in the field theory. It is
one of the most powerful aspects of the geometrically engineered quantum field theories
(GEQFTs) that one may determine these non-perturbative corrections from the dual,
geometric type II instantons by mirror symmetry. We will use this information in the
next section to study the pattern of supersymmetry breaking in the exact string effective
supergravity theory.
Fs =

4. Non-perturbative supersymmetry breaking in the string effective theory

Apart from the vector multiplets, the general N = 2 supergravity can be coupled to N =
2 matter hypermultiplets. These couplings may generate a non-trivial scalar potential which
determines the vacuum structure. We review a few facts and definitions in Appendix A. In
string theory these couplings may arise as the consequence of background fields such as
RR fields in type II strings [13] and magnetic backgrounds in the heterotic string [14].
In the following we study the contribution of instanton effects in the presence of such
backgrounds.
4.1. The SU(2) GEQFT on a compact manifold
Let us first consider a global type II compactification associated to the SU(2) GEQFT as
an explicit example for the supersymmetry breaking by instantons on a compact manifold.
It will be rather obvious how to generalize to the generic situation subsequently.

P. Mayr / Nuclear Physics B 593 (2001) 99126

The prepotential of the string effective supergravity is of the form

a, b, ic R.
F (s, t) = s t 2 + 1 a t 3 + b t + c + f (s, t),

107

(4.1)

The modulus t is related to the single 2-sphere of the singularity supporting the SU(2)
theory. In the tree level approximation s , Eq. (4.1) is the same as (3.3) for G = SU(2)
up to a shift of t by a constant. This shift is an effect of the global geometry. It has no further
consequences except for the translation of the origin of the SU(2) theory to t = i. The
function f depends only on the exponentials (qs , qt ) = (e2is , e2it ). The s-independent
piece of F (s, t) describes one-loop effects in the heterotic string theory whereas a term
qsk corresponds to an k instanton effect.
The presence of background fields induces charges of a hypermultiplet w.r.t. to a gauge
symmetry gauged by the vector multiplets. Let us first ask what kind of gauge charges
will be relevant. In the super-Higgs effect, one gravitino becomes part of a massive super
multiplet with spin content (3/2, 1, 1, 1/2) [12]. It is natural to identify these two spin one
vectors with the ones in the universal sector of the graviphoton and the dilaton s, = 0, 1.
Note that this choice is universal and is valid for any gauge group G. In addition there
may be an ordinary Higgs effect in the magnetic field theory U (1) factors related to the
condensation of monopoles.
To proceed we perform a symplectic transformation on the standard section (3.4) and
e of the form
obtain a period vector
1 2

2 (t 1) + 12 fs
,
e =
X
t
1 2
1
2 (t + 1) + 2 fs
2

t s + at 3 bt + s 2c 2f + tft + sfs
.
e =
(4.2)
F
2ts + 3at 2 + b + ft
2
3
t s at + bt + s + 2c + 2f tft sfs
This basis is the equivalent of (3.5) in the global compactification. Note that the se is entirely due to the spacetime instantons correction fs =
dependent piece of X
s
2is
+ . Therefore in the perturbative, tree-level plus one-loop corrected
f/s q = e
supergravity, no prepotential exists and partial supersymmetry breaking appears to be
possible.
4.2. Non-perturbative gravitino masses
However, as asserted already, including the spacetime instantons, the X should be
always good homogeneous coordinates and the arguments of the no-go theorem [10] apply.
To be concrete, let us consider a breaking to N = 1 near the origin t = i of the field theory
Coulomb branch. The gravitino mass matrix is proportional to the supersymmetry variation
SAB :


i K/2 P01 iP11 iP01 P12
P03 + iP13
.
(4.3)
iSAB = e
P03 + iP13
P01 + iP11 iP02 P12
2

108

P. Mayr / Nuclear Physics B 593 (2001) 99126

A representative choice for a breaking to N = 1 with a zero single eigenvalue of the matrix
S is
P01 = P12 = m,

Px = 0,

for = 0, 1, else.

With this choice the mass matrix for generic moduli is



imeK/2 (t i)2 fs
0
.
iSAB =
0
(t + i)2 + fs
4

(4.4)

(4.5)

For qs 0 there is a N = 1 vacuum at t = i. For qs 6= 0 the instantons lift the zero

eigenvalue. Since the no-go theorem applies, there must be either N = 2 or N = 0
supersymmetry. An N = 2 vacuum requires t = 0, fs = 1. Even if such a point in the
moduli space would exist, it would not be connected to the large s vacuum. Thus the
2 2
instantons break further N = 1 N = 0 at a scale e8 /g .
The exciting fact about this instanton generated supersymmetry breaking is that the
non-perturbative effects are determined by mirror symmetry in the geometric type II
theory. Since the non-perturbative gravitino masses arise from the coupling of the GEQFT
to the string and gravity sector, the precise instanton series depends on the individual
compactification manifold M. As an example let us consider the mirror of the CalabiYau
manifold 8 M defined by the polynomial
p = x112 + x212 + x36 + x46 + x52 12 x1x2 x3 x4 x5 2 x16 x26 .

(4.6)

Here the xi are coordinates of a weighted projective space in which M is embedded as

the hypersurface p = 0 and (, ) are coordinates on MV (W ) related to (t, s) by the
mirror map. By calculating the period integrals and using the mirror map 9 we obtain the
m()
following result for the instanton expansion np = fs that governs the non-perturbative
gravitino masses:
= e2is 2 + 2496qt1 + 1941264qt2 + 1327392512qt3 + 861202986072qt4
m()
np


+ 540194037151104qt5 +
+ e4is

+ 448128qt2 + 2654785024qt3 + 5718020769540qt4

+ 8494210810708992qt5 +

+ e6is

2
9

+ e8is

1
8

+ 36401011968qt4 + 2160776148604416qt5 + + .

1
2

+ 2583608958216qt4


+ 12741316216063488qt5 +
347738368 3
qt
3

(4.7)
It would be interesting to study the global non-perturbative vacuum structure using this
exact information. In fact the coefficient functions fk (qt ) of the kth instanton term are
generally modular functions of the heterotic modular group, which is known for certain
8 This CalabiYau manifold served already as a prominent example in the understanding of type IIA/heterotic
duality [5,31] and GEQFTs [32].
9 We refer again to Refs. [26,27] for details and references on calculations of this type.

P. Mayr / Nuclear Physics B 593 (2001) 99126

109

compactifications. 10 This information might be sufficient to determine global properties of

the associated scalar potential. In the following we will take a different route and analyze
the vacuum structure directly in an expansion around the point particle limit of the SYM
theory.
4.3. Generalizations
It is easy to argue that the pattern of non-perturbative supersymmetry breaking described
in the above SU(2) case is generic for any gauge group G. First note that we used only the
universal sector of the dilaton and the graviphoton in the gauging and the generalization to
e of the field theory
general G is therefore trivial. Consider now the electric component X
section (3.5). On general grounds it is clear that the entries 1 and ti will not receive s
dependent instanton corrections. So the instanton corrections that lift the N = 1 vacuum
are entirely due to the modifications of the remaining period
e2 = F = Cij ti tj + .
(4.8)
X
s
This period describes tree-level gauge kinetic terms, and its s corrections non-perturbative
field theory corrections to it plus additional non-perturbative gravitational and string effects
in the full theory. Thus on pure field theory grounds, the N = 1 vacuum will be lifted in
any N = 2 QFT with non-perturbative corrections coupled to string theory. 11 The only
case left over are conformal QFTs with the exact gauge coupling being equal to the tree
level coupling. Still in this case we expect that, after adding the coupling to gravity and
breaking of conformal invariance as a consequence, s dependent instantons appear in the
string/gravity sector. 12 It is interesting to observe that the lift of the N = 1 vacuum seems
to be closely related to the breaking of conformal invariance.

5. Supersymmetry breaking in the point particle limit

In the following we want to pin down the structure of supersymmetry breaking as seen
by the low energy observable gauge theory. The string theory embedding predicts a special
form of the non-perturbative scalar potential of the low energy gauge theory with the
supersymmetry breaking parameters linked in a useful way to the string geometric moduli.
5.1. What can be expected from the field theory
Let us first ask what might be expected from what is known about supersymmetry
breaking in field theory. Obviously, since we consider spontaneous breaking in the
N = 2 supergravity theory which encodes the exact non-perturbative N = 2 SYM theory,
10 A study of the modular functions for special cases can be found in [33].
11 We emphasize once again that the no-go theorem applies to the string effective supergravity, not to a pure

field theory decoupled from string theory.

12 It follows from the general arguments in Section 2 that this is indeed the case.

110

P. Mayr / Nuclear Physics B 593 (2001) 99126

the supersymmetry breakdown should have a consistent formulation in terms of the

latter. There are three known patterns of supersymmetry breaking consistent with the
holomorphic structure of the N = 2 theory.
The first one is the addition of the N = 1 supersymmetric adjoint mass term discussed
in the original work of Seiberg and Witten [3]. It was argued there that this term drives
the theory to the point in the Coulomb moduli where the monopole gets massless. The
combined superpotential including the monopole hypermultiplet (m, m
e) is

m + madj u.
(5.1)
W = 2 aD me
m

du
e = adj da
, where the
The minimum of the scalar potential is at aD = 0, m = m
D
2
monopole becomes massless and condenses. The condensation of the magnetically charged
monopoles leads to a mass for the magnetic gauge field and confinement of the electric
fields la t Hooft [2].
Another very elegant and interesting approach to supersymmetry breaking starting
from the holomorphic properties of the N = 2 theory has been considered in Ref. [34].
Compatibility with the analysity properties of the exact field theory can be implemented
by the introduction of so-called spurion fields with vevs that trigger the breaking. It was
already argued there that the structure of the potential obtained in this way is in qualitative
agreement with what one expects from a supergravity theory.
Indeed, we will find that the supersymmetry breaking in string theory is morally
a generalization of the two mechanisms. 13 In particular the supersymmetry breaking
parameters will depend in a specific way on the geometry of the transverse dimensions.

5.2. String effective action

To study the string effective theory near the point particle limit, we consider an
2 , keeping the exact quantities including the infinite instanton
expansion in 0 1/Mstr
series at each order. This approach has been introduced in [32] to derive the nonperturbative field theory results from string theory. Specifically one considers the limit
of small  with

= ,
Mstr

Mstr e8

2 /b

1g

= fixed,

(5.2)

where b1 is the one-loop beta coefficient of the SYM theory. Note that the tree level
coupling g at the string scale equal to the volume of the base B behaves as
Im s b1 ln . It was further argued in [32] that in this limit the string effective action
is described by a supergravity period vector with building blocks
k
,  2 u, s,  2 su,
1, a k , aD

k = 1, . . . , r,

(5.3)

k ) are the vevs of the scalar fields in the electric (magnetic) vector
where the a k (aD
multiplets of the SYM theory. A crucial fact for the following is the appearance of
13 There are extra terms that resemble the soft breaking terms of the N = 1 spurion approach [35].

P. Mayr / Nuclear Physics B 593 (2001) 99126

111

u = hTr 2 i at order  2 . The dependence of on u will be generate the N = 1

supersymmetric mass term for the adjoint scalar in (5.1).
The structure of the string effective action around the field theory limit turns out to be
quite intricate. We will banish details of the supergravity calculations in the appendix and
content ourself with an outline of the qualitative structure in the following. We start with
the following ansatz for the symplectic section = (X , F ):




f +  2 c u
2 f (s + const)
=
,
F
,
(5.4)
X =

k
ca k
caD
where = , +, +, . . . . The entries X , = 1, . . . , 2 + m describe (together with the
dual periods F ) the universal graviphoton/dilaton sector and m extra scalar fields ta that
parametrize the geometry of the transverse dimensions. On the other hand the entries
Xk , Fk , k = 1, . . . , r are associated to the r = rk G field theory periods. The expressions
f , c are so far arbitrary functions of the moduli ta but will have to satisfy some
constraints imposed on the section by N = 2 special geometry. The Khler potential
obtained from has the form
2
K,
(5.5)
V
where the precise expression for K can be found in Eq. (B.1) and the leading term is
K = ln(V) +

V = VB VTV ,

VB = i(s s + const) ,
2
X
|f |2 2 .
VTV = 2

(5.6)

In fact V is the volume of the CalabiYau manifold M. Note that the field theory dependent
terms are contained in K and, apart from the inverse powers of Mstr , suppressed by the
overall inverse power of the volume V. One can distinguish two different geometric scales
contained in V which will be enter the supersymmetry breaking scales in the following.
These are the volume of the base VB ln  on which the gauge fields propagate and
the volume VTV of the dimensions transverse to the field theory.
5.3. A first look at the scalar potential
The coupling of the hypermultiplets to the vector multiplets 14 induces a non-trivial
scalar potential V for the scalars [12]

u
v

huv k
+ Px Px U ,
V = V + V + V = 3Px Px V + 4k
V

(5.7)

where the three terms are due to the gravitino, hyperino and gaugino variations,
u
) describe the interactions of the
respectively. The couplings Px , x = 1, 2, 3 (and k
u
hypermultiplets q with the four-dimensional vector multiplet A
, where is a gauge
14 See Appendix A for more details.

112

P. Mayr / Nuclear Physics B 593 (2001) 99126

index and x an index of the SU(2)R symmetry. In the present context the couplings Px
in the universal dilaton/graviphoton sector parametrize the super-Higgs effect discussed
in the previous section. Since we will be interested in a region of the field theory moduli
space where the monopoles are light, we must also add their couplings to the magnetic
ei ), i = 1, . . . , r denote the r = rk G monopole
U (1) factors of the field theory. 15 If (mi , m
i
hypermultiplets with charges qk , these couplings are described by Pkx = Qxk with



T
S
ei x mi , m
ei .
(5.8)
Si , m
 2 Qxk = qki m
The matrices U and V depend on the scalars zi in the vector multiplets. As their general
form is quite involved we will restrict to give explicit expressions along the way when
needed.
The cosmological term
The leading term in the scalar potential arises from the super-Higgs effect in the
universal sector and does not depend on the field theory moduli. It represents the
cosmological constant from the view of the brane world. Its moduli dependence is
described by the  0 -piece of the matrices U and V :

eK V0 = f f ,

eK U0 = w, g w
S,
,

(5.9)

where we used t = {ta , s} to denote the non field theory moduli and w, = K f + f, .
To avoid a cosmological constant of the order of the string scale, the sum of the leading
contributions from the gravitino, hyperino and gaugino variations in (5.7) should vanish.
There are various possibilities to achieve such a cancellation at special values of the vector
and/or hypermultiplets.
In the following we will mostly separate the question of why the cosmological constant
is so small compared to Msusy from the study of the pattern supersymmetry breaking on
the brane as a function of the couplings Px that determine the super-Higgs effect. Note that
providing a solution to the cosmological constant problem in the present context would be
rather significant as even non-perturbative effects are included. Although we will not solve
the problem of the cosmological constant we will observe some interesting interrelations
between the supersymmetry breaking on the brane and the cosmological term and keep an
eye on possible mechanisms for a cancellation of the latter.
The N = 1 preserving adjoint mass term
Let us consider for a moment a naive cancellation of the individual contributions to the
cosmological term by imposing
Px w, = 0,

Px f = 0.

(5.10)

Assuming that Eq. (5.10) holds, one finds for the leading piece of the scalar potential in
the field theory limit:
15 We will write all following equations in the appropriate local magnetic variables which are equal to a k =
D
/a k F in the UV region.

P. Mayr / Nuclear Physics B 593 (2001) 99126

 4 V =

1 X x
Sxadj u k
Ql + mxadj ul l k Qxk + m
2 x
 i 2

 2
e ,
+ 2qli qki a l a k mi  + m

113

(5.11)

where
mxadj = Mstr

Px c
.
c

(5.12)

Eq. (5.11) describes precisely the scalar potential of the N = 2 field theory with a mass
term madj for the adjoint scalars in the vector multiplets. As argued in Ref. [3], the
N = 2 theory is driven to the point a l = 0, l in the Coulomb moduli, where r mutually
p
local monopoles become massless and condense with a vev madj in a new N = 1
supersymmetric vacuum.
Relations from special geometry and breaking to N = 0
Imposing the condition (5.10) not only cancels the leading order contribution to the
scalar potential but also a number of extra terms at O( 4 ). As the expression in Eq. (5.11) is
the most general one compatible with N = 1 supersymmetry these terms must necessarily
break N = 1 N = 0 in the SYM theory.
An generic solution to (5.10) is the limit of infinitely large extra dimensions, as will be
discussed below. It might appear that partial supersymmetry breaking to N = 1 with the
adjoint mass term (5.12) is possible for any such solution. However this is in fact not true.
The reason is that the general ansatz (5.4) for the supergravity section must be subject to
further conditions in order that is a valid symplectic section of N = 2 special geometry.
In Appendix B we find that these relations have the important implication
Px f = 0 mxadj Px c = 0.

(5.13)

Thus cancellation of the N = 1 supersymmetry breaking terms at O( 4 ) implies also

the cancellation of the leading N = 1 adjoint mass term. Therefore, contrary to what
might have been naively expected, there will be no excessively large separation of the
two breaking scales to N = 1 and N = 0, respectively.
As we pointed out already, the no-go theorem in the string effective supergravity does not
imply a no-go theorem for the pure field theory completely decoupled from string theory.
This decoupling can be achieved by taking a strictly infinite string scale Mstr = which,
by Eq. (5.2), amounts to a non-compact limit in the (tangential) direction of the base. In
this case certain terms dressed by inverse powers of ln /Mstr drop from the effective
field theory potential and Eq. (5.10) can be replaced by a weaker condition that can be
satisfied and connects to the N = 1 pure field theory vacuum discussed in [23]. However
note that in the field theory coupled to string theory the phenomenologically acceptable
regime is ln /Mstr . 20 and supersymmetry is broken by the terms with extra powers of
(and in fact we will argue that the phenomenologically interesting region is in the range
of moderately small ). This leads again to a quite restricted hierarchy for the breaking
scales to N = 1 and N = 0, respectively

114

P. Mayr / Nuclear Physics B 593 (2001) 99126

In Appendix B we collect a few additional relations derived from special geometry

which contain some useful information. In particular they imply that the scale of the
supersymmetry breaking is given by
P f
,
(5.14)

which is also equal to the scale of the monopole condensate. The fact that mQ is
proportional to the one-loop beta function coefficient fits well the discussion in Section 4.3
where we observed that the supersymmetry breakdown is apparently linked to the
violation of conformal invariance. Moreover Eq. (5.14) displays the important fact that
the supersymmetry breaking scale tends to be suppressed by the volume of the transverse
dimensions.
mQ b 1

Non-compact transverse dimensions

The defining data of the string effective theory is the period vector of a K3 fibered
CalabiYau manifold. A special property of the leading periods f in the universal sector
is that they are free of any instanton corrections arising from a finite transverse volume .
The reason is that these s-independent pieces are identical to the periods of a K3 manifold
which is of the same type as the generic fiber in the fibration. Moreover on the K3 manifold
there are no instanton corrections to the period integrals as a consequence of the (4, 4)
supersymmetry of the CFT. Explicitly this means that the periods f are polynomial in the
ta of degree two with the precise form dictated by the K3 intersection matrix. Note that
there are instanton corrections with an action proportional to the Khler volumes {ta } in
the K3 fibration.
Obviously, a solution to both conditions in (5.10) cannot exist for generic moduli. It
would be interesting to classify possible solutions on sub-slices of the moduli of all K3s,
which would be tantamount for finding solutions for a vacuum with vanishing leading
cosmological constant. In the next section we will study in detail the dependence of the
scalar potential on the universal volume modulus t of the transverse dimensions. In this
case it is easy to verify that the only solution to (5.10) is the limit of infinitely large
transverse dimensions, t = . For the reasons described above this does not lead to partial
supersymmetry breaking, however. Note that non-compact transverse volume should not
be confused with the pure field theory case, which requires infinitely large tangential
dimensions VB ln /Mstr .
6. Supersymmetry breaking on the brane
After having outlined some general features we give now details of the supersymmetry
breaking in the effective SYM theory on the brane as a function of the couplings Px
parametrizing the super-Higgs effect in the universal string sector. 16 For concreteness
16 Note that the following expressions describe the supersymmetry breaking on the brane for arbitrary general
couplings and thus the distinction of patterns with partial or complete classical super-Higgs effect in the string
sector arises only after making a specific choice for them.

P. Mayr / Nuclear Physics B 593 (2001) 99126

115

we will restrict the description of the deformations of the transverse geometry to the
generic volume modulus t. Introducing extra moduli will not change the effective theory
in the sector of the SYM theory; the way that these extra moduli enter is in the leading,
cosmological term.
The universal sector of the graviphoton, the dilaton and the universal modulus t is
described by the following ansatz for the functions f :




1
1
(6.1)
f1 = 1 t 2 ,
f2 = t.
f0 = 1 + t 2 ,
2
2
Moreover it follows from the relations of special geometry that the coefficients of the u
dependent terms in the symplectic section (5.4) are
c0 = b,

c1 = b,

c2 = 0,

c0 = c0 ,

(6.2)

where b b1 with a numerical coefficient specified in (B.5). With this information one can
determine the matrices U and V and the scalar potential Eq. (5.7). Since its general form
is at first sight involved, we will perform the analysis in various steps.
6.1. Leading terms in a large expansion
Let us start with the leading terms of the field theory potential in an expansion in powers
of
b1
ln /Mstr .
(6.3)

The motivation for this expansion is that for a very large string scale, the terms with
lower powers of are suppressed. 17 In particular the pure field theory potential will
correspond to keeping only the leading terms in the limit = . However for the theory
coupled to gravity, the logarithmic dependence on Mstr implies relative small values of
bounded from above by the value 10b1 for Mstr Mpl . In fact, as we will argue that the
supersymmetry breaking scale is proportional to the string scale and moreover the latter
can well be in the TeV range in this world on a brane scenario, the value of can be
actually quite small for a moderate to low value of the string scale. In this case the terms
with lower powers in are relevant. However we find it still useful to display the structure
of the potential by organizing it in powers of , keeping in mind that this will only give a
good physical hierarchy for extremely large values of Mstr .
From the general expressions in Appendix C one obtains the following leading term of
the scalar potential in the SYM theory:



1
1 3
+
Q + Ml l k Q+
Q + Ml3 l k Q3k + Mk3
 4 V =
k + Mk +
2 l
2 l


 i 2

 
i
ib
i i l k  i 2


us + c.c. .
e
(us u s ) +

+ 2ql qk a a m + m
2 2
2
(6.4)
=

17 In fact the relative contribution of these terms might be underestimated by the naive formula (6.3), since it
does not take into account threshold effects which are known to be sizable in string theory compactifications with
large compact dimensions.

116

P. Mayr / Nuclear Physics B 593 (2001) 99126

Here the superscripts refer to complex combinations x1 ix2 of the real fields. In
i ei q i is the holomorphic bilinear in the monopole fields and Q+ =
particular Q
l
l = 2m m
l

(Q
l ) . The supersymmetry breaking terms in the potential can be divided into
(i) F-type terms: The vev of the holomorphic bilinear term in the monopole fields will
be determined by the quantity Ml which can be split into its holomorphic and harmonic
pieces in the SYM fields, respectively:
Ml = mN =1 ul + mN =0 (ul c.c.).

(6.5)

The holomorphic piece

b
P f f ,
(6.6)
c

represents the N = 1 preserving adjoint mass term depending on the transverse volume as
dictated by (6.1). On the other hand, the harmonic piece
mN =1 =

b
P f ,
(6.7)
c
induces a soft breaking of supersymmetry.
(ii) D-type terms: The alignment of the vevs of the scalar components of the individual
chiral N = 1 multiplets in the monopole hypermultiplet are determined by the remaining
component Ml3 = ul + c.c. with
mN =0 =

b 3
P f .
(6.8)
c
(iii) Remaining terms: In addition there are further soft-breaking terms that depend on
us with coefficients


1
2 2 1
2
2
2
|m
| + |mN =0 + mN =1 | + | | ,
= 2 |c|
b
2 N =0
2


1
1
+

3
c(mN =1 + mN =0 ) C + cS
m
C c C ,
(6.9)
=
b 2
2 N =0
=

where C x = Px c . A nice heuristic interpretation of these terms can be given as follows.

Let us consider the N = 1 supersymmetric term in (6.5). In the field theory this term can
be associated with a superpotential term umN =1 in Eq. (5.1). At the point where the N = 1
vacuum branches off, u = G 2 , where G is a c2 (G)th root of unity. Moreover the scale
of the N = 1 theory is related to that of the N = 2 theory by 3N =1 = mN =1 2 . Thus the
superpotential term umN =1 , evaluated at the N = 1 point of the Coulomb branch is
W = G 3N =1 .

(6.10)

So remarkably enough, evaluating the superpotential of Ref. [3] at the extremum with
respect to the SYM fields, gives precisely the expected dynamical generated superpotential
in the N = 1 theory. In the string theory context there are further contributions from the
superpotential (6.10) to the scalar potential because ei s is to be treated as a field

P. Mayr / Nuclear Physics B 593 (2001) 99126

117

rather than a constant. These terms have the form of the us dependent terms in the scalar
potential (6.4).

To determine the scale of supersymmetry breaking, note that l k /a n is very large in a

region with light monopoles, a k 0. As a consequence, minimization with respect to the
fields a k will require an adjustment of the monopole bilinears Qxl Mlx to a high precision.
Thus the generic scale of the monopole condensate and the supersymmetry breakdown is
given by
b1 x
b1
P f Mstr Mstr ,
(6.11)

where in the second expression we have assumed that the couplings in the universal string
sector are of order one in string units. As we will discuss in the next section, this is indeed
the case if these couplings Px arise from certain background fields in the string theory
compactification.
Msusy

The N = 1 pure field theory vacuum 18

If we are not interested in string theory but pure field theory, we can take the decoupling
limit of infinite string mass which implies by Eq. (5.2). Only the leading terms
(6.4) survive in this pure field theory case. This case was considered in [23] and it was
asserted that a solution with N = 1 supersymmetry exists. Note that there are many choices
for the super-Higgs effect that lead to partial supersymmetry breaking in the classical
supergravity and break the remaining supersymmetry only by instanton effects. However
only a very special subset will be related to a vacuum of the N = 1 pure field theory at
Mstr = .
A choice of or RR-fluxes (or super-Higgs effect) that cancels all supersymmetry
breaking terms in the leading part of the potential must correspond to = mN =0 = 0 and
values for and such that the last terms in (6.4) cancel. From Eqs. (6.7)(6.9) we find
that the solution is given by
P0 =

1 + t02
1 t02

P1 ,

P2 =

2t0
P ,
1 t02 1

P3 = 0,

(6.12)

The meaning of this equations is that for fixed t0 , the choice (6.12) of the super-Higgs
cancels the supersymmetry breaking terms in the leading potential at t = t0 .
Let us summarize some relevant properties of the solution (6.12) that represents a special
configuration of RR-fluxes that preserve N = 1 in the decoupled field theory at Mstr = .
(1) It corresponds to a special fine tuning of fluxes associated to 0- , 2- and 4-form
charges on the non-compact manifold. The leading contribution in the non-compact
transverse limit relevant for geometric engineering of N = 1 QFTs is from the flux
dual to the 0 and 4 brane charges.
(2) The solution exists already at finite t. The relevant decoupling is due to the infinite
volume of the dimension parallel to the brane, Im s ln /Mstr which gives
an infinite mass to the string states. This is in nice agreement with our previous
18 I thank Cumrun Vafa for conversations on the following issues.

118

P. Mayr / Nuclear Physics B 593 (2001) 99126

assertion that the breaking to N = 0 is indeed only due to the s-dependent instantons
in the string sector, which are non-perturbative in the field theory coupling (rather
than instantons from the, in string units, finite transverse volume).
(3) For finite Mstr there are extra terms discussed in the next section that come with
additional powers of 1/ and break supersymmetry. Canceling the leading terms by
(6.12) leads to an extra logarithmic dependence of the supersymmetry breaking scale
on Mstr . It is interesting to mention that the overall coefficient of these subleading
terms depends on the hypermultiplet involved in the super-Higgs effect (the value of
defined in the next paragraph). For the hypermultiplet involved in the string theory
super-Higgs effect this coefficient is non-zero [13,36]. If the coefficient would have
been zero, supersymmetry breaking would have been postponed to the next order of
/Mstr .
6.2. Subleading terms and a vanishing leading cosmological constant
Subleading terms in the expansion can be straightforwardly added using the
expressions in Appendix C. For the purpose of displaying their structure however it is
useful to temporarily make a concrete and simple choice for the super-Higgs effect. For
a physically motivated illustration let us cancel the leading cosmological constant in the
large volume limit by imposing Eq. (5.10) on the leading terms. This is achieved by
m
,
Px = 0, else,
(6.13)
P01 = P11
Mstr
were we have used a SU(2) rotation to fix the direction of the triplet of Killing
prepotentials. To treat the hyperino contribution in (5.7) on the same footing with the
u
v = P x P x . This simplification
huv k
gaugino and gravitino we will use the relation k

is justified for the couplings in string theory with = 4, as will be outlined in the next
section. The full field theory dependent scalar potential for the super-Higgs determined by
(6.13) is

b I  k l 1
b R  1 k l 2 2


1
M
M
4
1
Qk 2 uk Ql 2 ul + Qk Ql + Q3k Q3l
 V =
2
|c|
|c|
2


2
b
 i 2





iM

 2
e
+ c.c. + 2 a l qli a k qki mi  + m
+ 2u a l ul
2b




2i
2M 2
1
u + c.c.





M2
M2
M 1 l
l l
Ql a + c.c. + 2i|c|2 2 2 aD
a c.c. + 2 2 K. (6.14)
2c
b
b
b
Here uIk = cuk + cu k and
M=

b m
,



b = 1 4i M.
M

(6.15)

Let us compare this result with the soft breaking terms discussed in the context of SYM
theories in the literature. The first two lines of Eq. (6.14) have the form of soft breaking

P. Mayr / Nuclear Physics B 593 (2001) 99126

119

terms generated by the spurion approach of [34]. The term in the third line is of the form
generated by the N = 1 spurion of [35]. The coefficients of these breaking terms are related
in the string effective theory which can be thought of explicitly determining the vevs of
the spurion fields. The fourth line contains some additional soft breaking terms.
Note that the N = 1 adjoint mass term is absent only because of the choice of couplings
(6.13) and as a consequence of Eqs. (5.12) and (6.2).
6.3. Minima of the potential
Ultimately, all moduli including the numerical values of and should be fixed by
the minima of the potential. The minimization with respect to the field theory moduli is
relatively straightforward and reduces approximately to Qxl Mlx with small corrections
from the additional finite terms near the monopole point. In the special case with mN =1 = 0
above, the leading potential in a large expansion is of the form of the soft supersymmetry
breaking considered in Ref. [34], which include a detailed numeric study of the vacuum
structure for the gauge group SU(2). By comparison, the spurion vev should be identified
with the scale Msusy in (6.11).
To determine the string vacuum, one needs also to vary with respect to the dilaton s and
the transverse volume modulus t. This brings us back to the question of the cosmological
term. We will not solve the question of minimization for these fields, which we leave for
the future, but point out some qualitative features. As for the dilaton, a stabilization is
possible due to the subleading terms in with explicit s dependence. Note that there are
also further calculable terms at order  6 and higher which depend non-trivially on s. Thus
although we do not know in the moment which value of s corresponds to a minimum, there
is no obstruction in principle to answer this question.
A similar situation holds for the volume modulus t of the large extra dimensions. For a
fixed Msusy 1 TeV, the behavior of the typical inverse radius vs. Mstr is plotted in Fig. 1
below.
1
while
In particular, for small a Mstr in the TeV range, the average radius is close to Mstr
4
it is larger by a factor of 10 for a string scale equal to the Planck scale. Since the scale

Fig. 1.

120

P. Mayr / Nuclear Physics B 593 (2001) 99126

1
of supersymmetry breaking is Msusy VTV
one suspects there is no local minimum for the
volume modulus and thus t will run away to infinity to restore supersymmetry. A natural
idea to fix the minimization of the transverse volume is to link it to the minimization of
the field theory moduli space [37]. In other words, we can use, as in the compactification
discussed in Section 4, the fact that the volume modulus itself can correspond to one of the
vevs of the field theory gauge group. Since the transverse volume is of order one in string
units in such a model, one expects a relation

Msusy Mstr R 1 .

(6.16)

Therefore this class of compactifications requires the string scale to be equal to the
supersymmetry breaking scale in the TeV range and predicts it to be equal to the scale
of the extra large dimensions.

7. Geometric charges in string theory and why Mstr should be small

In the previous sections we have described the supersymmetry breakdown as a
consequence of a super-Higgs effect in the universal sector, parametrized by the couplings
Px . Let us finally comment on some important details of the relation of these couplings to
the background fields in string theory.
In Ref. [13] it was shown that backgrounds of RR 2p-forms in the type IIA theory induce
charges of the dilaton hypermultiplet w.r.t. a gauge symmetry of the vector multiplets
specified by the cohomology class of the background. From the equations in [13] it also
follows that = 4 for this case. Due to the special property of the RR sector in the type II
theory these couplings are suppressed by a factor of the string coupling II .
This is an extremely important detail from the world on a brane point of view since (i) the
string scale Mstr may be in the TeV range in this scenario [38]; (ii) the supersymmetry
breaking scale (6.11) is of the order of Mstr rather than Mpl . In fact the relation between
these two scales is given by dimensional reduction as
2

Mpl

2
Mstr .
2II

(7.1)

The special origin of the SYM theory its origin in the RR sector which makes its
coupling independent of II and its localization on a hypersurface which makes its coupling
independent of the transverse volume has the consequence that the values of and
II remain largely undetermined by phenomenological requests. This allows for a large
separation of the string scale and the Planck scale arising from a large transverse volume
and/or a small string coupling II .
After supersymmetry breaking, even a small mismatch in the cancellation of the
cosmological constant in the field theory will drive the string scale to as small a value
as possible. In other words not only is it consistent to have strings at the TeV scale in this
world on a brane scenario, but, after the supersymmetry breakdown, a low string scale may
well be energetically favored. In fact the solution to the cosmological constant problem
should tell us why Mstr a few TeV instead of being zero.

P. Mayr / Nuclear Physics B 593 (2001) 99126

121

Hidden sectors
A type IIB version of the backgrounds of Ref. [13], with a generalization to also
include NS backgrounds has been given in Refs. [23,36]. 19 In Ref. [23] the coupling of
the universal hypermultiplet to the vector gauge symmetries was derived from the BPS
tension of NS and RR 5-branes in the presence of such backgrounds. As the tension
of the NS brane lacks the suppression factor II , NS backgrounds will naturally lead
to a supersymmetry breaking scale Mpl . A low supersymmetry breaking scale could
still be obtained by breaking first supersymmetry on a hidden singularity and then
transmitting the breakdown by gravitational interactions to the observable branes, similarly
as in Ref. [1].
Outlook
Using the appropriate geometric singularities of Ref. [17], there are no obvious obstacles
to model a N = 2 supersymmetric string theory embedding of the gauge group of the
standard model on a brane and its matter spectrum. Breaking supersymmetry in the way
described in this paper leads to a generalized, calculable scalar potential depending on the
parameters of this standard model. Apart from questions of details, e.g., how natural
the standard model spectrum appears in the context of geometric singularities, higher
derivative corrections etc., this appears to be an unexpectedly accomplishable scenario for
the study of a phenomenologically relevant string vacuum with calculable, dynamically
fixed mass and coupling parameters.

Acknowledgements
I am grateful to Sheldon Katz, Wolfgang Lerche and Cumrun Vafa for essential
comments and conversations. I would also like to thank Luis lvarez-Gaum, Costas
Kounnas, Hans-Peter Nilles, Antoine van Proeyen and Stephan Stieberger for valuable
discussions and comments.

Appendix A. A brief review of the N = 2 scalar potential

Apart from the vector multiplets, the N = 2 supergravity may contain a number nH of
matter hypermultiplets q u , with the 4 nH scalars components parametrizing a quaternionic
manifold MH . The N = 2 supergravity allows for couplings of the hypers to the vectors
19 To compare the results of these two papers, which in fact seem to disagree in that the scalar potential of
the [23] is SL(2, Z) invariant whereas the one of [36] is not, one needs to rewrite the N = 2 scalar potential in
terms of a N = 1 superpotential, a task that appears to have been settled only in very special cases, see [39]. We
1 + iP 2
propose the general form of the N = 1 superpotential to be W = P X Pe F , where P = P

e for the
is the complex combination of the real Killing prepotentials and we have included a similar term P
couplings to the magnetic fields to ensure symplectic covariance. To verify the consistency of this ansatz, the use
of the supersymmetric Ward-identity in Eq. (2.97) of Ref. [40] is essential.

122

P. Mayr / Nuclear Physics B 593 (2001) 99126

by gauging isometries of MH . In other words, hypermultiplets can be charged under the

U (1)nV +1 gauge symmetry
u
 ,
q u q u + k

(A.1)

u defining the charge of q u under the th U (1) symmetry. 20

with the Killing vector k
We refer to [41] for a most general and modern account of the combined N = 2 effective
action and a detailed list of references.
The supersymmetry variations of the gauginos aA , the hyperinos and the gravitino
A depend on the scalars zi as:
a
B ,
aA = WAB

= NA A ,

A = iSAB B ,

(A.2)

with
a
= ieK/2 x 
WAB

P x (
AB b
u
X ,
NA = 2UuB BA eK/2 k
SAB =

+ b K)X g ba ,

1 x

 AB Px eK/2 X .
2

(A.3)

Here A = 1, 2, = 0, . . . , nV , a = 1, . . . , nV , K is the Khler potential, UuB the

symplectic vielbein and moreover the Px , x = 1, 2, 3 a triplet of Killing prepotentials
associated to the gauging (A.1).
The coupling (A.1) of the hypermultiplets to the vector multiplets induces a non-trivial
scalar potential V for the scalars [12] which can be written in terms of the supersymmetry
variations as


1
V = 6 tr SS + tr N N + ga b tr W a W a
2

x x
u
v
= 3P P + 4k huv k V + Px Px U ,
(A.4)
with

V = eK X X ,

U = fi g i j fj ,

fi = eK/2 (i + Ki )X .

(A.5)

Appendix B. Special geometry of the string effective GEQFTs

To determine the effective string theory we need the precise form of the section .
The large s behavior (3.3) and the consistency of the supergravity monodromies with the
monodromies of the embedded field theory imply the general form (5.4)




f +  2 c u
2 f (s + const)
=
,
F
.
X =

k
ca k
caD
The Khler potential has the  expansion
20 In particular FI terms can be included in this description by gauging a trivial hypermultiplet.

P. Mayr / Nuclear Physics B 593 (2001) 99126

123

2


S F X F
S = ln(V) +  K,
K = ln i X
V
P

with V given in (5.6), A = f c and

k
k
u,
a k aD
+ 2i(As u c.c.) + h(u, s) + h(
s ).
(B.1)
K = i|c|2 a k a D
P
2
The moduli dependent quantity = |f | describes the volume of the transverse
dimensions. From the above expressions we can determine the matrices U and V entering
the scalar potential (5.7).

Special geometry of
Special geometry implies some important conditions on the functions appearing in the
ansatz for (5.4). From the supergravity identity
1
S
S
(B.2)
U = (Im N ) + V ,
2
and the fact that U is Hermitian we see that U V must be a real symmetric matrix. Let
. After some algebraic manipulations we obtain
us consider the leading  1 term of U k

= V1k
+ A g kl ul + B g k l u l,
U1k


cb
2i
ab

V f + w
S,
A =
a g b ,
V


 



cA

Ab b
ab

g
2is
V f c w

S,
B =
a
V

A
A


b
hu h ub
,
+

(B.3)

where w,a = Ka f + f,a . Moreover b and are two constants that are defined in the
field theory:
2
bu,
= e s ,
|c|2
1X k k
b
a aD + 2 u.
FFT =
2
|c|
s FFT =

(B.4)

rk G

In particular the constant b is proportional to the one-loop beta function coefficient [42,43].
From (A ) = B one can derive the following useful relations satisfied by the functions
f , c and h appearing in the ansatz for and in K:


1
+ u(c1 s + c2 ), ci C,
h = 2iAu s

X
f2 ,
0=

c
B = f c = b,
c

 2  2
A  = B ,
and

(B.5)

124

P. Mayr / Nuclear Physics B 593 (2001) 99126

P f = 0 P c = 0.

(B.6)

Appendix C. The scalar potential for the universal volume modulus

e = U V for the section in (6.1) is given
The  expansion of the matrices V and U
by


O 0 :

eK V0

O 1 :

eK V1 k = f ca k ,

O 2 :

eK V2

= f f ,

e k =
eK U
1

eK U0l k =

V

kl A ul + c.c. ,
2
2|c|

= c f u + c f u,

e
eK U
2

eK U0 = w, g w
S,
,

V l k

eK V2k k = |c|2 a k a k ,




V
ib
f
k l S
=
X Xl +

f us + h.c.
2|c|2 k
2





2i
1
Ks + K f w
S,
+
t + h.c.

2





K
3K
f w
w,t w
S,
S,
+ i 2 bu +
t + h.c.
t,

(C.1)

where t = {t, s} and



cb
2 2
w
S,
f +
A =
t ,

2 2 i|c|2
1

w,t k la l ,
A uk + c.c. +
Xk =

c

k
k,l = Im aD,l .

References
[1] H.P. Nilles, Phys. Lett. B 115 (1982) 193;
H.P. Nilles, Nucl. Phys. B 217 (1983) 366;
S. Ferrara, L. Girardello, H.P. Nilles, Phys. Lett. B 125 (1983) 457;
For a review and further references see, H.P. Nilles, Int. J. Mod. Phys. A 5 (1990) 4199.
[2] G. t Hooft, Nucl. Phys. B 190 (1981) 455.
[3] N. Seiberg, E. Witten, Nucl. Phys. B 426 (1994) 19;
N. Seiberg, E. Witten, Nucl. Phys. B 430 (1994) 396, Erratum.
[4] K. Konishi, Phys. Lett. B 392 (1997) 101.
[5] S. Kachru, C. Vafa, Nucl. Phys. B 450 (1995) 69.
[6] S. Ferrara, J.A. Harvey, A. Strominger, C. Vafa, Phys. Lett. B 361 (1995) 59.
[7] S. Kachru, E. Silverstein, Nucl. Phys. B 463 (1996) 369.
[8] C. Vafa, E. Witten, Nucl. Phys. Proc. Suppl. 46 (1996) 225.

(C.2)

P. Mayr / Nuclear Physics B 593 (2001) 99126

125

[9] E. Kiritsis, C. Kounnas, Nucl. Phys. B 503 (1997) 117;

E. Kiritsis, C. Kounnas, P.M. Petropoulos, J. Rizos, Nucl. Phys. B 540 (1999) 87.
[10] S. Cecotti, L. Girardello, M. Porrati, Phys. Lett. B 145 (1984) 61.
[11] S. Ferrara, L. Girardello, M. Porrati, Phys. Lett. B 366 (1996) 155;
S. Ferrara, L. Girardello, M. Porrati, Phys. Lett. B 376 (1996) 275.
[12] B. de Wit, P. Lauwers, A. Van Proeyen, Nucl. Phys. B 255 (1985) 269.
[13] J. Polchinski, A. Strominger, Phys. Lett. B 388 (1996) 736.
[14] C. Bachas, A way to break supersymmetry, hep-th/9503030;
I. Antoniadis, E. Gava, K.S. Narain, T.R. Taylor, Nucl. Phys. B 511 (1998) 611.
[15] A. Klemm, W. Lerche, P. Mayr, C. Vafa, N. Warner, Nucl. Phys. B 477 (1996) 746.
[16] S. Katz, A. Klemm, C. Vafa, Nucl. Phys. B 497 (1997) 173.
[17] S. Katz, P. Mayr, C. Vafa, Adv. Theor. Math. Phys. 1 (1998) 53.
[18] E. Witten, Nucl. Phys. B 500 (1997) 3.
[19] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257.
[20] S. Katz, C. Vafa, Nucl. Phys. B 497 (1997) 196;
M. Bershadsky, A. Johansen, T. Pantev, V. Sadov, C. Vafa, Nucl. Phys. B 505 (1997) 153;
P. Mayr, P. Berglund, Adv. Theor. Math. Phys. 2 (1999) 1307.
[21] J.L.F. Barbn, Phys. Lett. B 402 (1997) 59.
[22] H.P. Nilles, Phys. Lett. B 112 (1982) 455;
G. Veneziano, S. Yankielowicz, Phys. Lett. B 113 (1982) 231.
[23] T.R. Taylor, C. Vafa, RR flux on CalabiYau and partial supersymmetry breaking, hepth/9912152.
[24] A. Ceresole, R. DAuria, S. Ferrara, A. Van Proeyen, Nucl. Phys. B 444 (1995) 92.
[25] I. Antoniadis, H. Partouche, T.R. Taylor, Phys. Lett. B 372 (1996) 83.
[26] P. Candelas, X.C. De La Ossa, P.S. Green, L. Parkes, Nucl. Phys. B 359 (1991) 21.
[27] S.T. Yau (Ed.), Essays on Mirror Manifolds, International Press, Hong Kong, 1992.
[28] K. Hori, C. Vafa, Mirror symmetry, hep-th/0002222.
[29] R.L. Bryant, P.A. Griffiths, Some observations on the infinitesimal period relations for regular
threefolds with trivial canonical bundle, in: M. Artin, J. Tate (Eds.), Arithmetic and Geometry
Volume II, Progress in Mathematics, Vol. 39, Birkhuser, Boston, MA, 1983.
[30] P. Mayr, Fortschr. Phys. 47 (1998) 39.
[31] A. Klemm, W. Lerche, P. Mayr, Phys. Lett. B 357 (1995) 313.
[32] S. Kachru, A. Klemm, W. Lerche, P. Mayr, C. Vafa, Nucl. Phys. B 459 (1996) 537.
[33] B.H. Lian, S.-T. Yau, Nucl. Phys. Proc. Suppl. 46 (1996) 248;
M. Henningson, G. Moore, Nucl. Phys. B 482 (1996) 187.
[34] L. lvarez-Gaum, J. Distler, C. Kounnas, M. Mario, Int. J. Mod. Phys. A 11 (1996) 4745;
L. lvarez-Gaum, M. Mario, Int. J. Mod. Phys. A 12 (1997) 975;
L. lvarez-Gaum, M. Mario, F. Zamora, Int. J. Mod. Phys. A 13 (1998) 403;
L. lvarez-Gaum, M. Mario, F. Zamora, Int. J. Mod. Phys. A 13 (1998) 1847.
[35] N. Evans, S.D.H. Hsu, M. Schwetz, Phys. Lett. B 335 (1995) 475;
N. Evans, S.D.H. Hsu, M. Schwetz, S.B. Selipsky, Phys. Lett. B 456 (1995) 205.
[36] J. Michelson, Nucl. Phys. B 495 (1997) 127.
[37] Work in progress.
[38] I. Antoniadis, B. Pioline, Nucl. Phys. B 550 (1999) 41.
[39] J.P. Derendinger, S. Ferrara, A. Masiero, A. Van Proeyen, Phys. Lett. B 140 (1984) 307;
E. Cremmer, C. Kounnas, A. Van Proeyen, J.P. Derendinger, S. Ferrara, B. de Wit, L. Girardello,
Nucl. Phys. B 250 (1985) 385.
[40] R. DAuria, S. Ferrara, P. Fre, Nucl. Phys. B 359 (1991) 705.
[41] L. Andrianopoli, M. Bertolini, A. Ceresole, R. DAuria, S. Ferrara, P. Fr, J. Geom. Phys. 23
(1997) 111;

126

P. Mayr / Nuclear Physics B 593 (2001) 99126

L. Andrianopoli, M. Bertolini, A. Ceresole, R. DAuria, S. Ferrara, P. Fr, Nucl. Phys. B 476

(1996) 397.
[42] M. Matone, Phys. Lett. B 357 (1995) 342.
[43] J. Sonnenschein, S. Theisen, S. Yankielowicz, Phys. Lett. B 367 (1996) 145.

Nuclear Physics B 593 (2001) 127154

www.elsevier.nl/locate/npe

Supersymmetric ZN ZM orientifolds in 4D with

D-branes at angles
Stefan Frste , Gabriele Honecker, Ralph Schreyer
Physikalisches Institut, Universitt Bonn, Nussallee 12, D-53115 Bonn, Germany
Received 13 September 2000; accepted 13 October 2000

Abstract
We construct orientifolds of type IIA string theory. The theory is compactified on a T 6 /ZN ZM
orbifold. In addition worldsheet parity in combination with a reflection of three compact directions
is modded out. Tadpole cancellation requires to add D-6-branes at angles. The resulting four
dimensional theories are N = 1 supersymmetric and non-chiral. 2001 Elsevier Science B.V. All
rights reserved.

1. Introduction
One of the major issues in string theory is to classify consistent theories in especially
3 + 1 dimensions. Insights into strong coupling regions of string theory provide reasons
to hope that apparently different models are actually equivalent and can be mapped
onto each other by duality transformations. Often, strong and weak coupling regions
are interchanged in this process. Open strings with Dirichlet boundary conditions, e.g.,
provide a perturbative description of solitonic (non-perturbative) objects (D-p-branes) in
type II string theories [1]. This observation was crucial for one of the first conjectures
about string dualities the heterotic/type I duality [2]. Since compactifications of
the heterotic string are of particular phenomenological interest one expects the same
for type I compactifications. Here, one typically starts with a type II theory on an
orbifold. In addition, worldsheet parity (possibly combined with discrete targetspace
transformations) is modded out. The consistency requirement of modular invariance is
replaced by tadpole cancellation conditions. The resulting models are called orientifolds.
This kind of construction has been considered already some time ago [37]. The first
formulation using the modern language of D-branes and orientifold fixed planes was given
* Corresponding author.

E-mail address: [email protected] (S. Frste).

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 1 6 - 7

128

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

in [8]. Subsequently, several models with different numbers of non-compact dimensions

and unbroken supersymmetries have been constructed, e.g., in [917].
In orientifolds the amount of unbroken supersymmetry depends on the orbifold group
and the arrangement of D-branes and O-planes needed for consistent compactifications.
In [18] it was pointed out that also D-branes intersecting at angles can leave some
supersymmetries unbroken. A concrete realization of this observation in orientifold
constructions was worked out in [19,20]. These authors considered type IIB (IIA) string
theory on T 4 /ZN (T 6 /ZN ) orbifolds with N = 3, 4, 6. In addition, they gauged worldsheet
parity together with the reflection of two (three) directions of the T 4 /ZN (T 6 /ZN )
orbifold such that this reflection leaves O-planes intersecting at angles fixed. To cancel
the corresponding RR-charges, D-branes intersecting at angles need to be added. In the
present paper we are going to supplement this class of models by compactifying type IIA
on a T 6 /(ZN ZM ) orbifold together with imposing invariance under worldsheet parity
inversion combined with the reflection of three orbifold directions. We will discuss only
models with N = 1 supersymmetry in four dimensions. All possible compactifications of
this kind yield non-chiral four dimensional models with different gauge groups and matter
content.
In the next section we describe general features of the construction. The third section is
devoted to a detailed study of the Z4 Z2 orientifold. Subsequently, we briefly give results
for all other consistent models, viz. the Z2 Z2 , Z6 Z3 and the Z3 Z3 orientifolds.
We conclude by summarizing our results. Three appendices provide technical details of the
considered models: Appendix A is addressed to the computation of loop channel diagrams,
Appendix B contains tables of massless spectra, and Appendix C describes a projective
representation used for the Z4 Z2 orientifold.

2. General setup
Throughout the paper we will discuss models with four non-compact dimensions labeled
by x , = 0, . . . , 3. In addition, there are six compact directions which we describe by
three complex coordinates,
z1 = x 4 + ix 5 ,

z2 = x 6 + ix 7 ,

z3 = x 8 + ix 9 .

(1)

Each of those coordinates describes a torus T 2 . In addition, points on these tori are
identified under rotations
1 : zj e2ivj zj ,

2 : zj e2iwj zj ,

(2)

where (1 , 2 ) denotes an element of the orbifold group ZN ZM . (The action on the

complex conjugated coordinates just follows from the conjugation of (2).) We will discuss
type IIA theories on these manifolds. In addition, we gauge the symmetry generated by
R where reverses worldsheet parity, and R reflects the imaginary parts of the zi ,
R : zi z i .

(3)

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

129

The gauging of R creates orientifold fixed planes (O-planes). The location of those
planes is given by sets of points fixed under the elements of R ZN ZM . These fixed
planes carry RR-charges which must be canceled by adding D-branes to the model [1,8].
One of these O-planes is extended along the non-compact directions and the real parts of
the zi . It is invariant under the R reflection. To visualize the remaining O-planes we note
that (I = 1, 2)
1

RI : zi = (I ) 2 R(I ) 2 : zi

(4)

12

indicate that the rotations are performed with plus-minus half

where the powers of
the angle as compared to (2). Thus, a fixed plane under RI is obtained by acting with
1
(I ) 2 on the set of points fixed under R. Therefore (starting from the O-6-plane at zi =
z i , i = 1, 2, 3), one obtains a set of O-6-planes intersecting at angles, whose values are
given by half the order of the corresponding ZN ZM element. Since all these O-6-planes
have compact transverse directions and carry RR-charges, one expects that for consistency
one needs to include a certain amount of D-6-branes canceling exactly those charges. These
D-6-branes need to be parallel to the O-6-planes and hence also intersect at angles. Indeed,
for ZN orbifolds this has been shown to be the case [19,20]. In the following sections we
will generalize those models to ZN ZM orbifolds. This turns out to be a straightforward
modification of the discussion given in [19,20]. A new ingredient, however, is that in some
cases more complicated projective representations of the orientifold group in the open
string sector are needed. This has been observed before in some ZN ZM orientifolds of
type IIB models [11,15]. In fact, for the Z2 Z2 model to be discussed in Section 4 we
will obtain the T-dual version of the model of [11].
In the next paragraphs we are going to review some of the technical aspects necessary
for orientifold constructions. The main consistency requirement comes from RR charge
conservation. Technically, it translates into the tadpole cancellation condition [8]. The RR
charges describe the size of the couplings of O-planes and D-branes to RR gauge fields.
The numerical value of these couplings can be computed by extracting the RR exchange
contribution to the forces acting between O-planes and D-branes [1]. A convenient way of
computing these forces is to move to the open string loop channel. There, the RR charge
of O-planes and D-branes can be extracted from the UV-limits of the following parts of the
Klein bottle, Mbius strip and annulus diagrams [21]:
Klein bottle: Closed string NSNS states with PR()F insertion;
Mbius strip: Open string R states with PR insertion;
annulus: Open string NS states with P()F insertion.

(5)

Here, ()F is the fermion number to be defined below. (For closed strings ()F =
()FL = ()FR because of the presence of in the trace.) Further, we denote by P the
projector on states invariant under the orbifold group ZN ZM . The requirement of tadpole
cancellation determines the number of D-6-branes and part of the representation of the
orientifold group on the ChanPaton indices.
Another essential consistency condition is what is called completion of the projector
in the tree channel in [19]. Let us briefly recall their arguments. The important diagrams

130

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

Fig. 1. (a) Klein bottle, (b) Mbius strip, (c) Cylinder.

are drawn in Fig. 1. The notation is taken from [8]. Orbifold group elements are denoted
by h or g. In the tree channel picture the crosscaps correspond to O-planes invariant under
Rh. D-branes are assigned a letter i or j . The orbifold group element g denotes the twist
sector of the closed strings propagating in the tree channel. (For further details see [8].)
Consistency of the boundary conditions requires
(Rh1 )2 = (Rh2 )2 = g

(6)

in the Klein bottle diagram and

(Rh)2 = g

(7)

in the Mbius strip. For the class of orientifolds discussed here, the lhs of (6) and (7) are
the identity. Hence, in the tree channel only untwisted closed strings propagate in the Klein
bottle and in the Mbius strip. Since these must be invariant under the orbifold group, the
tree channel amplitude must contain the insertion of the complete projector on invariant
states. The actual calculation of the diagrams depicted in Fig. 1 will be done in the loop
channel, where worldsheet time is vertical. Transforming back to the tree channel one
must recover the insertion of the complete projector mentioned above. For the Klein bottle
amplitude this requirement leads to restrictions on the possible orbifold lattices. In the
Mbius strip one obtains further conditions for the representation of the orientifold group
on the ChanPaton matrices.
Finally, let us introduce some notation and definitions which we mainly borrow
from [20]. The action of the orientifold group on closed string degenerate ground states 1
is specified by
R : |s0 , s1 , s2 , s3 i |s0 , s1 , s2 , s3 i,
1 : |s0 , s1 , s2 , s3 i e2i vEEs |s0 , s1 , s2 , s3 i,
1 All states are in light cone gauge. The first entry in the state corresponds to two non-compact directions,
whereas the other three belong to the three complex compact directions.

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

E s
2 : |s0 , s1 , s2 , s3 i e2i wE
|s0 , s1 , s2 , s3 i.

131

(8)

Worldsheet parity inversion interchanges the left with the right moving sector. Under
GSO projection states with fermion numbers (1)FL = (1)FR = 1 are kept, where
(1)FL |s0 , s1 , s2 , s3 i = ei(s0 s1 s2 s3 ) |s0 , s1 , s2 , s3 i,
(1)FR |s0 , s1 , s2 , s3 i = ei(s0 +s1 +s2 +s3 ) |s0 , s1 , s2 , s3 i.

(9)

In the loop channel the expressions for the Klein bottle, Mbius strip and annulus are
(c = V4 /(8 0 )2 and V4 is the regularized volume of non-compact momentum space)
Z
K = 4c





1 + (1)F
R
dt
S 2t (L0 +L 0 )
P
TrU +T
(1) e
,
t3
2
2

(10)

Z
A=c


 

1 + (1)F
dt
1
S 2t L0
P
(1) e
Tropen
,
2
2
t3

(11)

Z
M=c





1 + (1)F
dt
R
S 2t L0
P
Tr
e
(1)
.
open
t3
2
2

(12)

Here

P=

1 + 1 + + 1(N1)
N



1 + 2 + + 2(M1)
M


(13)

is the projector on states invariant under the orbifold group ZN ZM . S denotes the space
time fermion number. In order to compute the contribution due to RR exchange in the
tree channel one needs to compute the parts of expressions (10), (11) and (12) which
are given in (5). (The spacetime fermion number insertion has been taken care of by
the minus sign in the second line in (5).) The transformation back to the tree channel is
performed by replacing t = 4l1 for the Klein bottle, t = 8l1 for the Mbius strip and t = 2l1
for the annulus [8]. Finally, RR-charge conservation is imposed by demanding the infrared
(l ) limit of the tree channel expression to be finite.

3. The Z4 Z2 R-orientifold
In this section we discuss the Z4 Z2 model in great detail because in this case
all possible subtleties show up; therefore the other models can be treated briefly in the
following sections.
E =
The lattice described by the shifts vE = (1/4, 1/4, 0) for the Z4 -factor and w
(0, 1/2, 1/2) for the Z2 -factor of the orbifold-group is essentially an SU(2)6 -lattice,
i.e., a product of three tori in the notation of complex compact coordinates. The two
crystallographically allowed orientations A and B of one of the three tori with respect
to the reflection R are shown in Fig. 2. As will be explained in the following, the only

132

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

Fig. 2. Lattices for Z4 Z2 . Black circles denote the Z4 fixed points and white circles the additional
Z2 fixed points.

perturbatively consistent models are given by the choices ABA and ABB for the compact
directions 2 .
3.1. The Klein bottle amplitude
We begin with evaluating the general expression (10) for the Klein bottle 1-loop
amplitude of the Z4 Z2 model by considering the compact momenta. The KaluzaKlein
(KK) and winding (W) states are generally given by

2
m1 eE1 + m2 eE2 ,
(14)
P =
r


r
n1 eE1 + n2 eE2 ,
(15)
W =
0
2
where mi and ni are integers, eEi are the basis vectors of the corresponding torus with radius
r and eEi are the basis vectors of the dual torus (i = 1, 2). The SU(2)2 lattices in Fig. 2 are
spanned by

 

0
2
A
A

,
(16)
,
eE2 =
eE1 =
2
0


 
1
1
,
eE2B =
,
(17)
eE1B =
1
1
with the corresponding dual basis. For the KaluzaKlein and winding states invariant under
R one gets
 
 
m 1
nr 0
A
A
,
W = 0
,
(18)
P =
1
r 0

 
 
2m 1
2 nr 0
,
WB =
,
(19)
PB =
0
1
r
0
2 The choices BAA and BAB are equivalent to these models.

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

133

Table 1
Multiplicities of the fixed points for Z4 Z2
K
(0, k1 )(0, k2 )
(2n1 + 1, k1 )(0, k2 )
(2, k1 )(0, k2 )
(0, 2k1 )(1, k2 )
(0, 2k1 + 1)(1, k2 )
(2n1 + 1, k1 )(1, k2 )
(2, 2k1 )(1, k2 )
(2, 2k1 + 1)(1, k2 )

ABA

ABB

1
4
8
8
16
16
16
8

1
4
8
4
8
8
8
4

where m, n are integers. As a consequence of relation (4) the states (18) and (19) are
invariant under the insertions R1k1 2k2 for k1 = 0, 2 and k2 = arbitrary; when A and B
are exchanged, these states are invariant under insertions with k1 = 1, 3 and k2 = arbitrary.
From
pL,R = P W

(20)

for the closed string, it follows that the lattice contribution to the 1-loop amplitude for the
Klein bottle is L[1, 1] for A-states and L[2, 2] for B-states, where the notation is taken
from [19,20] and explained in Appendix A. In general, lattice contributions only appear
for untwisted tori.
The calculation of the oscillator contributions to (10) simplifies if one takes into account
that the RR-exchange in the tree level is given by the trace over the NSNS-sector with
the insertion (1)F in the 1-loop channel. Furthermore, the elements of the orbifold group
Z4 Z2 act as the unit operator on the oscillator states which contribute to the trace because
R-invariance leads to a cancellation of the phases given in Eq. (8) between left- and
right-movers. This means that the oscillator contributions are equal for any insertion from
the orbifold group. Although the numerical results may be zero from case to case, all the
twisted sectors formally show up in the amplitude, because R does not exchange them.
(n ,k )(n ,k )
The last ingredients we need are the multiplicities K 1 1 2 2 of the 1n1 2n2 -twisted
fixed points which are invariant under the insertion R1k1 2k2 . Consider, e.g., the second
torus T2 twisted by 2 . In the A-lattice, two of the four fixed points are interchanged under
R1k1 2k2 when k1 = 0, 2 and k2 = arbitrary. In the B-lattice all four fixed points are
(0,0)(1,k2)
(0,2)(1,k2 )
= K
= 2(4) for an A(B)invariant under these insertions, such that K
type T2 . The resulting multiplicities are shown in Table 1. Considering all this, we can
evaluate the Klein bottle 1-loop amplitude (10). The notation is similar to [19,20] and
defined in Appendix A. We use the fact that for the oscillator contributions K(n1 ,k1 )(n2 ,k2 ) =
K(n1 ,0)(n2 ,0) is valid for all ni , ki (i = 1, 2) as explained above and simplify the notation
by defining K(n1 ,n2 ) K(n1 ,0)(n2 ,0)) . For the ABA-lattice we get

134

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

Z
K = c(1RR 1NSNS )

dt
t3

L[1, 1] L[2, 2]K(0,0) + 4L[1, 1]K(1,0) + 8L[1, 1]K(2,0) + 4L[1, 1]K(3,0)

+ 8L[1, 1]K(0,1) + 16K(1,1) + 16L[2, 2]K(2,1) + 16K(3,1) .
(21)
2

For the ABB-lattice, one L[1, 1] in every term in the second line of Eq. (21) has to be
exchanged for L[2, 2] and the prefactors in the third line have to be divided by two.
The modular transformation to the tree-channel t = 4l1 yields (see Appendix A)
e = 32c(1RR 1NSNS )
K

Z
dl
0

e 2]K
e (0,0) 2L[4,
e 4]K
e (1,0) 4L[4,
e 4]K
e (2,0) 2L[4,
e 4]K
e (3,0)
e 4]2L[2,
L[4,


e 4]K
e (0,1) + 4K
e (1,1) 4L[2,
e 2]K
e (2,1) 4K
e (3,1) .
4L[4,
(22)
e 4] in every term in the second line of Eq. (22) again has to
For the ABB-lattice, one L[4,
e 2] and Eq. (22) has to be multiplied by an overall factor of 1/2. We
be exchanged for L[2,
realize that the complete projector in the sense of [19] and Section 2 shows up, because all
possible insertions of the orbifold group appear, only untwisted sectors contribute and the
prefactors are given by
Y


(23)
2 sin(k1 vi + k2 wi ) ,
k1 vi +k2 wi 6=0; i=1,2,3

as expected. At this point we can clarify, why only the models ABA and ABB (and the
equivalent models BAA and BAB) are perturbatively consistent. The lattice contribution
of, e.g., AAA is changed to BBA by the insertion of 1 and there is no way to get
the complete projector. From the same argument it follows that only the orbifold groups
Z2 Z2 , Z4 Z2 , Z3 Z3 and Z6 Z3 can lead to perturbatively consistent solutions.
3.2. The annulus amplitude
To cancel the tadpoles which arise in the Klein bottle amplitude we need to introduce
D-branes. As was found in [19,20] and explained in Section 2 for the orientifold models
under consideration we have to introduce D-6-branes rotated by half the angles which are
given by the elements of the orbifold group. For Z4 Z2 this leads to a configuration of
eight different D-6-branes whose locations in the three compact tori are shown in Fig. 3.
For simplicity we restrict ourselves to the case where the D-branes are located at the fixed
points. Writing down the mode expansions, e.g., for an open string stretching from brane
1/2
with respect to the brane
(0, 0) to brane (1, 0) (where the brane (1, 0) is rotated by 1
(0, 0)) one realizes that the modings are the same as for the closed string twisted by 11 .
Therefore it is convenient to call these kinds of open strings twisted sectors [19], as we
will do in the following. Using these conventions, an open string stretching from brane
(i1 , i2 ) to brane (i1 n1 , i2 n2 ) belongs to the 1n1 2n2 -twisted sector.

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

135

Fig. 3. Arrangement of branes and action of the orbifold group for Z4 Z2 . The branes are labelled
by (i1 , i2 ) = (0, 0), . . . , (3, 1) mod (4, 2). This is convenient in the sense that the brane (i1 , i2 ) is
i /2 i /2
obtained by rotating the compact real axes with 1 1 2 2 , see also Section 2.

For the open string the compact momenta popen are given by the distance of parallel D6-branes in the corresponding directions. Therefore we consider the location of the branes
in the fundamental cells of the lattices A and B, see Fig. 4. Starting from brane (0, 0), we
A
A
= P A for the directions 4, 6, 8 and popen
= W A for the directions 5, 7, 9 as well
get popen
1 B
1 B
B
B
as popen = 2 P (4, 6, 8) and popen = 2 W (5, 7, 9), see Eqs. (18) and (19). It follows,
that the lattice contribution to the 1-loop amplitude for the annulus is given by L[2, 2] for
an A-torus and by L[1, 1] for a B-torus. The compact momenta are nonzero again only for
untwisted tori, i.e., tori where the twist acts trivially. But in addition, the D-branes have
to be invariant under insertions of the elements of the orbifold group, therefore only the
insertions 1, 12 , 2 and 12 2 yield non-vanishing compact momenta on untwisted tori.
Calculating the oscillator contributions to (11) we again focus on the RR-exchange in
the tree channel which is given by the trace over the NS-sector with (1)F insertion in the
1-loop amplitude. All twisted sectors appear, each in combination with the insertions 1,
12 , 2 and 12 2 which leave the D-branes invariant, as stated above. In contrast to the
Klein bottle amplitude the phases arising from the insertions are not cancelled. Moreover,
the representation matrices of the orbifold group have to be taken into account. These
matrices are unitary M M matrices, where M is the number of arrangements that will be
fixed by the tadpole cancellation conditions. In the amplitude the matrices appear as
(i n1 ,i2 n2 )

tr k11k2

(i ,i )1

tr k11k2 2

(24)

where (i1 , i2 ) labels the eight different branes and n1 and n2 indicate the 1n1 2n2 -twisted
(i ,i )
(i ,i )
sector, as explained above. k11k2 2 is an abbreviation for k11 2 k2 . This leads to a factor of
1 2

8M 2 for the untwisted and twisted sectors without insertions (without means insertion
of 1 in the case of the annulus), where the 8 arises from the number of branes in the
arrangement (see Fig. 3). In fact, all the calculations can be done starting with brane
(0, 0) and inserting factors of 8 appropriately, because the other branes lead to the same
amplitudes.

136

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

Fig. 4. Location of branes in the fundamental cells for Z4 Z2 .

Table 2
Intersection numbers in the annulus for Z4 Z2
A
(0, 0)
(1, 0)
(2, 0)
(3, 0)
(0, 1)
(1, 1)
(2, 1)
(3, 1)

ABA

ABB

1
1
2
1
2
1
1
1

1
1
2
1
4
2
2
2

The terms with insertions lead to twisted sector tadpoles in the tree-channel, which
cannot be cancelled by the other diagrams. This yields the twisted sector tadpole
cancellation conditions
(i ,i2 )

tr 201

(i ,i2 )

= tr 011

(i ,i2 )

= tr 211

=0

(25)

for all (i1 , i2 ), similar to [8].

The analogue to the multiplicities of fixed points in the closed string are the intersection
numbers of the D-branes in the open string case. The intersection numbers are given by the
number of times that the branes intersect within the fundamental cell and can be read off
easily from Fig. 4. Starting with brane (0, 0), a Z2 -twisted B-type torus contributes a factor
of two, whereas a Z2 -twisted A-type torus and Z4 -twisted tori of both types contribute a
factor of one 3 . Again, only the points invariant under insertions contribute, thus in the case
of the annulus only sectors without insertions appear in the amplitude and it is sufficient
(n ,n )
(n ,0)(n2 ,0)
which are given in Table 2. Now we
to consider the multiplicities A 1 2 A 1
have all the ingredients to write down the annulus 1-loop amplitude for the ABA-lattice
3 Remember that we are discussing strings starting on brane (0, 0).

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

c
A = M (1RR 1NSNS )
4

137

dt
t3

0
(0,0)

L[1, 1]L[2, 2] A

+ L[2, 2]A(1,0) + 2L[2, 2]A(2,0) + L[2, 2]A(3,0)

+ 2L[2, 2]A(0,1) + A(1,1) + L[1, 1]A(2,1) + A(3,1) ,
(26)
2

where we used the simplified notation A(n1 ,n2 ) A(n1 ,0)(n2 ,0) (see Appendix A). All the
terms with insertions vanish due to the twisted tadpole cancellation condition (25) and
therefore do not appear in Eq. (26). For the ABB-lattice one L[2, 2] in every term in the
second line of Eq. (26) has to be exchanged for L[1, 1] and the prefactors in the third line
have to be multiplied by two.
Performing the modular transformation t = 2l1 leads to
e = c M 2 (1RR 1NSNS )
A
8

Z
dl
0

e 1]A (1,0) 4L[1,

e 1]A (2,0) 2L[1,
e 1]A (3,0)
e 2]A (0,0) 2L[1,
e 1]2L[2,
L[1,


e 1]A (0,1) + 4A (1,1) 4L[2,
e 2]A (2,1) 4A (3,1) .
4L[1,
(27)
e 1] in every term in the second line of Eq. (27) has to be
For the ABB-lattice, one L[1,
e
exchanged for L[2, 2] and the whole amplitude has to to be multiplied by two. Again, the
complete projector shows up.
3.3. The Mbius strip amplitude
In the case of the Mbius strip, the compact momenta for a B-torus have to be doubled
in the 5,7,9 directions because of the R-projection, therefore one gets L[2, 2] for an Atorus and L[1, 4] for a B-torus. Again, the lattice contributions only appear for untwisted
tori and for insertions which leave the D-6-branes invariant.
Which strings contribute to the Mbius strip one-loop amplitude? Let us denote a
1n1 2n2 -twisted string by [(i1 , i2 )(i1 n1 , i2 n2 )] as explained in Section 3.2 and
k
k
consider the action of the insertion R1 1 2 2 thereupon:


(i1 , i2 )(i1 n1 , i2 n2 )
k
k

1 1 2 2 

(i1 + 2k1 , i2 + 2k2 )(i1 n1 + 2k1 , i2 n2 + 2k2 )



R
(i1 2k1 , i2 2k2)(i1 + n1 2k1 , i2 + n2 2k2 )



(i1 + n1 2k1 , i2 + n2 2k2)(i1 2k1 , i2 2k2 ) .

(28)

Since n2 , k2 = 0, 1 (mod 2), the condition i2 = i2 + n2 2k2 (mod 2) is equivalent to

2i2 = n2 (mod 2), thus only Z2 -untwisted sectors (i.e., n2 = 0) with k2 = arbitrary contribute to the amplitude. The condition i1 = i1 + n1 2k1 (mod 4) implies, e.g., for the
brane with i1 = 0 that the sectors n1 = 0 with k1 = 0, 2 and n1 = 2 with k1 = 1, 3 contribute to the amplitude. To summarize, for (i1 , i2 ) = (0, 0) the sectors (n1 , k1 )(n2 , k2 ) =

138

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

(0, 0)(0, 0), (0, 0)(0, 1), (0, 2)(0, 0), (0, 2)(0, 1), (2, 1)(0, 0), (2, 1)(0, 1), (2, 3)(0, 0) and
(2, 3)(0, 1) contribute. The other branes of the arrangement in Fig. 3 get contributions
from different sectors, but the resulting amplitude is the same, therefore we can restrict the
calculation to the case (i1 , i2 ) = (0, 0) and insert factors of 8 appropriately, again.
Now the representation matrices of the orientifold group have to be taken into account.
For the brane (i1 , i2 ) in the sector (n1 , k1 )(n2 , k2 ) they appear as
 (i n ,i n )T (i1 ,i2 ) 
Rk1 k2 .
(29)
tr 1Rk11k2 2 2
Since only untwisted and 12 -twisted sectors appear, we abbreviate
 1 ,0)T (0,0) 
ak(n1 k12) tr (nR
k1 k2 Rk1 k2 . The multiplicities M can be obtained in the same way
as in the case of the annulus, because all intersection points are invariant under R. We
get M = 2 in the 12 -twisted sectors and M = 1 in the other sectors.
This leads to the Mbius strip 1-loop amplitude for the ABA-lattice
c
M = (1RR 1NSNS )
4

dt
t3

0
(0)
(0)
a00 L[1, 4]L[2, 2]2M(0,0)(0,0) + a01
L[2, 2]M(0,0)(0,1)
(2)
(2)
(0)
+ 2a10
L[2, 2]M(2,1)(0,0) + 2a11
M(2,1)(0,1) + a20
L[2, 2]M(0,2)(0,0)


(0)
(2)
(2)
+ a21 L[1, 4]M(0,2)(0,1) + 2a30 L[2, 2]M(2,3)(0,0) + 2a31 M(2,3)(0,1) .

(30)
(n )

(n )

For the ABB-lattice, one L[2, 2] in each term of equation (30) with ak1 k12 = ak1 01 has to be
exchanged for L[1, 4].
Performing the transformation to the tree-channel t = 8l1 (see Appendix A) yields
f = 4c(1RR 1NSNS )
M

Z
dl

0
(0) e
e
f(0,0) 2a (2)L[4,
e 4]M
f(1,0)
a00 L[8, 2]L[4, 4]2M
30
(0) e
f(2,0) 2a (2)L[4,
e 4]M
f(3,0) + 4a (0)L[4,
e 4]M
f(0,1)
L[4, 4]M
+ 4a20
10
01


(2) f(1,1)
(0) e
f(2,1) + 4a (2)M
f(3,1) .
+ 4a31 M
+ 4a21 L[8,
2]M
11

(31)

For the ABB-lattice, one L[4, 4] in each term of Eq. (31) with ak(n1 k12) = ak(n1 01 ) has to be
exchanged for L[8, 2].
To obtain the complete projector and to cancel the untwisted tadpoles from the other
diagrams, the -matrices have to fulfill the conditions
(0)

(2)

(0)

(2)

(0)

(2)

(0)

(2)

a00 = a10 = a20 = a30 = a01 = a11 = a21 = a31 = M,

(32)

and the untwisted tadpole cancellation condition reads

ABA: [M 16]2 = 0,
ABB:

[M 8]2 = 0,

(33)

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

139

which fixes the number of arrangements shown in Fig. 3 to be 16(8) for the ABA (ABB)
lattice, respectively. The conditions (32) are valid for the brane (i1 , i2 ) = (0, 0). For some
other brane (i1 , i2 ) one has to replace the insertions 1k1 2k2 in (32) by 1k1 +i1 2k2 +i2 .
3.4. The closed string spectrum
The massless spectrum is found by symmetrizing the massless states which satisfy the
GSO-projection conditions (9) with respect to R, 1 and 2 . exchanges left- and
right-movers and is defined following the convention of [8]
r 1 = r ,

r 1 = r ,

r 1 = r

(34)

for integer and half-integer r. For the action of R, 1 and 2 see Section 2. In the
following we denote the NSNS vacuum by |0i and, e.g., the R state | 12 , 12 , 12 , 12 iL by
| + + + +iL . The states are given up to normalization. In the untwisted sector we find
the massless states


NSNS: + |0i, graviton + dilaton (m.i.),

i i |0i, i i |0i,


3 3 + 3 3 |0i,

2,
3)
6 scalars (m.i.),
(i = 1, 2, 3; i = 1,
1 scalar,

RR: | + + + +iL | + + +iR | iL | + iR ,

| + + iL | + + + iR | + +iL | +iR ,

axion (m.i.),
1 scalar,

where (m.i.) stands for model independent states, i.e., states which are present
independent of the orbifold group. To summarize, the untwisted massless closed string
spectrum contains the N = 1 supergravity multiplet in D = 4 and 4C, where C denotes the
chiral multiplet.
In the 1n1 2n2 -twisted sectors the masses are given by
1 2
1
0 2
mL,R = NL,R + qL,R
+ Evac ,
4
2
2
with

(
qL,R =

0, (n1 vE + n2 w)
E ,
1 1
2, 2

(NS)


(n1 vE + n2 w)
E , (R)

(35)

(36)

and
Evac =

1 X
|n1 vi + n2 wi | 1 |n1 vi + n2 wi | ,
2

(37)

i=1,2,3

where one has to take care of 0 6 |n1 vi + n2 wi | < 1. We discuss the 12 -twisted sector
explicitly, the other sectors are obtained in a similar manner.
In the 12 -twisted NSNS sector, the massless states which fulfill the GSO-projection (9)
are found to be

 1 1 

0, 1 , 1 , 0 |2iNS ,
0, , , 0 |1iNS ,
2 2
2
2
L
L




0, 1 , 1 , 0 |1i
NS , 0, 1 , 1 , 0 |2i
NS ,
2
2
2 2
R
R

140

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

NSNS and |22i

NSNS . The other two
such that we get the two massless ground states |11i
possible combinations are not invariant under 2 . Furthermore, we have to consider the
action of R, 1 and 2 on the fixed points. Since the torus T3 is untwisted, the discussion
is valid for both the ABA and the ABB lattice. For the fixed point structure see Fig. 2. In the
tori T1 and T2 , the fixed points 3 and 4 are interchanged by 1 . In addition, the fixed points
3 and 4 in the torus T2 are interchanged by R. Under the action of 2 all the fixed points
are invariant. In the following, e.g., {13} denotes the fixed point built from fixed points 1 of
T1 and 3 of T2 . The fixed points {11}, {12}, {21} and {22} are invariant under R and 1 .
The fixed points {31}, {41} as well as {32}, {42} form pairs under 1 . The fixed points {13},
{14} as well as {23}, {24} form pairs under 1 and R. The remaining fixed points {33},
{34}, {43} and {44} form a quartet under 1 and R. Symmetrization in the NSNS sector
leads to two scalars for each fixed point, each pair and the quartet, i.e., 18 scalars altogether.
The discussion of the 12 -twisted RR sector is similar, but here the action of R gives
an additional minus sign. Therefore we have to antisymmetrize between left and right
movers, such that we get no states from the fixed points and pairs mentioned above. The
quartet contributes one vector (V).
Adding the superpartners from the NSR sector, we find 9C + 1V in the 12 -twisted
sector. The remaining twisted sectors can be treated in a similar fashion and we obtain the
massless twisted closed string spectrum
ABA: 57C + 1V,
ABB:

47C + 11V.

(38)

3.5. The open string spectrum

In order to determine the open string spectrum we have to count the degrees of
freedom of the ChanPaton factors for the massless states. Therefore, we have to
find a representation of the orientifold group which satisfies the tadpole cancellation
conditions (25) and (32). In general, we have to consider the action of the orientifold group
(a,b)
on massless states of the form |, ij ij i , where represents the vacuum together with
some combination of oscillators, i, j = 1, . . . , M (M is the number of arrangements of
branes) and (a,b) is the ChanPaton matrix for a string starting on brane a and ending on
brane b, with a, b = 1, . . . , 8, i.e. 4 ,





T
R1k1 2k2 , ij (b)Rk1 k2 (a)1
R1k1 2k2 : |, ij i(a,b)
ji
Rk1 k2 j i . (39)
To check whether the twisted tadpole cancellation conditions (25) are satisfied, we need
the representation matrices of the orbifold group which can be obtained via, e.g.,




= 1 : |, ij i(a,b)
(40)
R1k1 1 2k2 R1k1 2k2 : |, ij i(a,b)
ji
ji ,
(a)
which implies 1(a) ' (b)T
Rk1 1,k2 Rk1 k2 , where ' means equal-up to an irrelevant
phase. Taking into account all these constraints, we find the -matrices listed in
4 For notational simplicity in the following we label the eight branes in the arrangement shown in Fig. 3 with
single numbers as given in Table C.5 of Appendix C.

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

141

Appendix C. They form a projective representation of the orientifold group, 5 where the
Z2 Z2 substructure is similar to the model discussed in [11], which in turn is T-dual to
the Z2 Z2 R-orientifold discussed in Section 4.
In the untwisted NS sector (i.e., strings which start and end on the same brane) the
(a,a)
m
|0, ij ij i (m = 0, . . . , 9). The string (1, 1) is invariant
massless states are given by 1/2
under R, R12 , R2 , and R12 2 , where the last symmetry is not independent
of the first three. Applying Eq. (39), we get the following constraints on the ChanPaton
matrix (1,1) in the noncompact directions = 0, . . . , 3
(1)
(1)1
T
(1)
(1)1
T
(1,1) = R00 (1,1) R00 = R20 (1,1) R20
(1)
(1)1
T
= R01 (1,1) R01 ,
(41)
where the minus signs arise from the action of R on the massless state. Using the matrices listed in Appendix C this leads to
(1,1) = (1,1)T = M1 (1,1)T M1 = M2 (1,1)T M2 .

(42)

(1,1)

Counting the remaining degrees of freedom of (1,1) we find that 1/2 |0, ij ij i
is a vector in the adjoint representation of the gauge group Sp(M/4). In a similar
(1,1)
i,i
2,
3)
yield 3C in the
|0, klilk (i = 1, 2, 3; i = 1,
manner the compact directions 1/2
antisymmetric representation of Sp(M/4).
For the (2, 2)-string, the symmetries are R1 , R13 and R1 2 . Inserting the
corresponding -matrices also leads to Eq. (42) for (2,2) , i.e., the same result as for the
(1, 1)-string, but now the gauge group Sp(M/4) constitutes a second factor of a product
gauge group, since the strings are not mapped onto each other by any symmetry of the
theory.
The invariances of the (3, 3)-string are the same as for the (1, 1)-string and lead to the
same degrees of freedom. But since the (1, 1)-string is mapped onto the (3, 3)-string by a
symmetry of the theory, namely 1 , the (3, 3)-string is charged under the same factor of
the gauge group and does not contribute any further matter. Furthermore, we have to check
that the additional identities (3,3) = 1(1) (1,1) 1(1)1 ( = 1 in the directions 0, . . . , 3,
= i on T1 , = i on T2 and = 1 on T3 ) are consistent with the degrees of freedom
found so far, which is the case, indeed.
Proceeding in a similar manner for the strings (4, 4), . . . , (8, 8) we find that the untwisted
open string spectrum contains 1V in the adjoint of the gauge group [Sp(M/4)]4 and
3C in the [(A, 1, 1, 1) (1, A, 1, 1) (1, 1, A, 1) (1, 1, 1, A)], where A denotes the
antisymmetric representation of Sp(M/4) and the four factors of the gauge group arise
from the strings (1, 1), (2, 2), (5, 5) and (6, 6), respectively. The strings (3, 3), (4, 4), (7, 7)
and (8, 8) are related to these strings by the symmetry 1 and do not contribute any further
degrees of freedom, as explained above.
We begin the discussion of the open string twisted sectors with the (1, 2)-string. In the
sense of the explanation in the beginning of Section 3.2, this string forms the 13 -twisted
5 For a summary on projective representations and further references see the appendix of [22].

142

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

(1,2)

(1,2)

1
2
sector and the massless states are given by 1/4
|013 , ij ij i and 1/4
|013 , ij ij i ,
3
where 013 denotes the 1 -twisted NS vacuum. The (1, 2)-string is invariant under 12
and 2 . This leads to the constraints on the ChanPaton factors
(2) (1,2) (1)1
(2) (1,2) (1)1

20
= 01

01 ,
(1,2) = 20

(43)

where the signs are unphysical, since we only know that 14 and 22 act trivially on the
(1, 2)-string and inserting the corresponding -matrices yields
(1,2) = M1 (1,2)M1 = M2 (1,2) M2 ,

(44)

where the signs are unphysical and thus arbitrary. Calculating the degrees of freedom
we obtain that the (1, 2)-string transforms in the bifundamental (F, F, 1, 1) of the gauge
group. The strings (2, 3), (3, 4) and (4, 1) are related to the (1, 2)-string by the symmetries
R and 1 . Since the massless state is twofold degenerated, these strings form 2 scalars in
the (F, F, 1, 1). Analogous the strings (5, 6), (6, 7), (7, 8) and (8, 5) form 2 scalars in the
(1, 1, F, F ). The 1 -twisted sector is equivalent to the 13 -twisted sector, which contains
the strings (1, 4), (2, 1), etc. Therefore, these sectors together yield 2C in the (F, F, 1, 1)
(1, 1, F, F ). Since only Z4 -twisted intersection points appear, the multiplicity is one.
The (1, 3)-string, i.e., the 12 -twisted sector, possesses the additional symmetries
R1 and R13 , which lead to the constraints
(3)
(1)1
T
(3)
(1)1
T
(45)
(1,3) = R10 (1,3) R10 = R30 (1,3) R30 ,
where the minus signs again arise from the action of R on the massless states. Inserting
the corresponding -matrices yields
(1,3) = AT (1,3)T A = B T (1,3)T B.

(46)

Counting the degrees of freedom, we find that the (1, 3)-string transforms in the
(A, 1, 1, 1), where A denotes the antisymmetric representation of Sp(M/4). The (1, 3)string is related to the (3, 1)-string by 1 , just as (2, 4) to (4, 2), (5, 7) to (7, 5) and (6, 8)
to (8, 6). The massless states are twofold degenerated and since T1 and T2 are Z2 -twisted,
the multiplicity is 2 in both the ABA- and the ABB-lattice. Thus the 12 -twisted sector contributes 2C in the [(A, 1, 1, 1) (1, A, 1, 1) (1, 1, A, 1) (1, 1, 1, A)] for both lattices.
The (1, 5)-string, i.e., the 2 -twisted sector, is invariant under 12 and 2 . Again
imposing that the square of these symmetries act trivially on the massless states, we get
the constraints
(1,5) = iN2 (1,5) M1 = M1 (1,3)T M2 .

(47)

This implies that the strings (1, 5), (5, 1), (3, 7) and (7, 3), which are related by R and
1 , transform in the (F, 1, F, 1). Analogous the strings (2, 6), (6, 2), (4, 8) and (8, 4)
transform in the (1, F, 1, F ). The massless states are twofold degenerated. Now the tori T2
and T3 are Z2 -twisted, thus the multiplicity is 2 in the ABA-lattice and 4 in the ABB-lattice.
Altogether the 2 -twisted sector contributes 2C (4C) in the (F, 1, F, 1) (1, F, 1, F ) for
the ABA (ABB) lattice.

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

143

Table 3
Open string massless spectrum of Z4 Z2
Twist-sector

ABA

ABB

untwisted

1V
3C

1 + 13
12

2C
2C

2
1 2 + 13 2
12 2

2C
1C
1C

4C
2C
2C

Gauge group/matter
[Sp(M/4)]4
(A, 1, 1, 1) (1, A, 1, 1)
(1, 1, A, 1) (1, 1, 1, A)
(F, F, 1, 1) (1, 1, F, F )
(A, 1, 1, 1) (1, A, 1, 1)
(1, 1, A, 1) (1, 1, 1, A)
(F, 1, F, 1) (1, F, 1, F )
(F, 1, 1, F ) (1, F, F, 1)
(F, 1, F, 1) (1, F, 1, F )

Proceeding in a similar manner for the remaining twisted sectors we obtain the open
string twisted massless spectrum shown in Table 3. The only pecularity is the fact that the
1 2 -twisted sector (and the sector twisted by 13 2 ) has only one massless ground state.

4. The Z2 Z2 R-orientifold
Since the orbifold group Z2 Z2 is contained as a substructure in the Z4 Z2 model
discussed above, the calculation is similar and we do not have to go into the details again.
The lattice is described by the shifts vE = (1/2, 1/2, 0) for the first Z2 -factor and
w
E = (0, 1/2, 1/2) for the second one and shown in Fig. 2. The A- and B-lattice are
not interchanged by any insertion from the orientifold group, thus we obtain perturbatively
consistent and inequivalent solutions for the lattices AAA, AAB, ABB and BBB. The
arrangement of rotated D-6-branes we have to introduce is shown in Fig. 5. For the location
of the branes in the fundamental cells, consider Fig. 4 and discard the diagonal branes.
Performing the calculation of the amplitudes, we obtain the untwisted sector tadpole
cancellation conditions
AAA: [M 32]2 = 0,

ABB:

[M 8]2 = 0,

[M 16]2 = 0,

BBB:

[M 4]2 = 0,

AAB:

(48)

which fix the number M of arrangements of branes shown in Fig. 5. The remaining tadpole
cancellation conditions yield the representation matrices of the orientifold group, which
can be read off from the Z2 Z2 substructure of Tables (C.5) and (C.6) in Appendix C.
Using these matrices, we again solve the constraints arising from the symmetries of the
various strings and obtain the massless open string spectrum shown in Table (B.2) in
Appendix B. The massless closed string spectrum is shown in Table (B.1) in Appendix B.
Considering the spectrum we can explicitly verify that for the AAA lattice the Z2
Z2 R-orientifold is T-dual to the model discussed in [11]. Applying T-duality in the

144

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

Fig. 5. Arrangement of branes and action of the orbifold group for Z2 Z2 . Inserting n = 0, 1 yields
the four branes (i1 , i2 ) = (0, 0), (1, 0), (0, 1), (1, 1) mod (2, 2).

directions of the imaginary axes of the compact dimensions transforms R to and the
four D-6-branes in Fig. 5 into one D-9-brane and three types of D-5-branes. The AAB,
ABB and BBB models are T-dual to the Z2 Z2 orientifolds with discrete B field of
rank 2, 4 and 6, respectively, listed in [23]. To our knowledge, orientifolds with discrete B
field have been discussed first in [24]. For max(N, M) > 2 the T-duals are asymmetric
orientifolds with nonzero B field [25]. Interestingly, for heterotic theories, which are
believed to be connected to open string theories, similar observations have been made
in [26,27].
5. The Z6 Z3 R-orientifold
The lattice for this model is generated by the shifts vE = (1/6, 1/6, 0) and w
E =
(0, 1/3, 1/3) and shown in Fig. 6. The A- and B-lattice are interchanged by R1 , thus
we get solutions for the lattices ABA and ABB. To cancel the tadpoles from the closed
string we have to introduce the arrangement of 18 rotated branes shown in Fig. 7 and the
number M of arrangements is fixed by the untwisted tadpole cancellation condition
[M 4]2 = 0,

(49)

which we obtain for both the lattices ABA and ABB. Since this model contains less Z2
substructure than the two models discussed above, we expect the projective representation
of the orientifold group to be less complicated. In fact it turns out that we get by with the matrices of [8], taking into account the relative signs arising from the tadpole cancellation
conditions. Calculating the massless open string spectrum shown in Table (B.4) in
Appendix B we have to take care of the fact that some of the intersection points are
not invariant under various insertions. The massless closed string spectrum is shown in
Table (B.3) in Appendix B.
6. The Z3 Z3 R-orientifold
The orbifold shifts are now given by vE = (1/3, 1/3, 0) and w
E = (0, 1/3, 1/3). The
lattice is the same as shown in Fig. 6, discarding the additional Z2 fixed points. The choices

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

145

Fig. 6. Lattices for Z6 Z3 . The black circle in the origin denotes the Z6 fixed point, the other black
(white) circles denote the additional Z3 (Z2 ) fixed points.

Fig. 7. Arrangement of branes and action of the orbifold group for Z6 Z3 . Inserting n = 0, . . . , 5
and m = 0, 1, 2 yields 18 branes.

AAA, AAB, ABB and BBB lead to perturbatively consistent solutions. All these four
models yield the untwisted tadpole cancellation condition are displayed in Eq. (49). The
calculation of the massless open string spectrum is particularly simple in this case, because
the orbifold group contains no Z2 -substructure. Therefore we can choose the -matrices
to be the identity-matrix (up to a phase), while the relative signs are again fixed by the
tadpole cancellation conditions. Altogether we obtain the massless spectrum shown in the
Tables (B.5) and (B.6) in Appendix B.
7. Conclusions
In this paper we presented a class of orientifolds of type IIA string theory with orbifold
group ZN ZM . In addition, symmetry under worldsheet parity combined with the
reflection R of three directions was imposed. Further, we considered only cases leading
to unbroken N = 1 supersymmetry in the four non compact directions. We found that
(N, M) = (4, 2), (2, 2), (6, 3) and (3, 3) are the only perturbatively consistent solutions.
For these models we gave the solutions to the tadpole cancellation conditions and the
massless spectra. In addition to the universal supergravity fields there are various gauge
and matter fields living on D-branes intersecting at angles. The smallest intersection angle
is given by / max(N, M). Explicit results are presented only for the cases where the
gauge groups are maximal, i.e. all the D-branes sit at the corresponding orientifold fixed

146

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

planes. These gauge groups can be Higgsed to smaller groups by moving certain numbers
of D-branes off the O-planes.
The type IIA orientifolds considered here can be dualized to type IIB orientifolds by
performing T-duality in the directions where R acts in a non-trivial way. As stated in the
end of Section 4 the resulting type IIB orientifolds have in most cases a non-trivial discrete
B field background. (With B we denote the NSNS antisymmetric tensor.) Constant B field
backgrounds have received some attention in the recent past because they can lead to a
microscopic description of non-commutative field theories [2831]. In this context it may
be also interesting whether the IIA orientifolds considered here can be modified to include
non trivial B field backgrounds. In order to study this question one should investigate
whether the tadpole cancellation conditions can be solved by projective representations of
the orientifold group which are not equivalent to the ones given here [32,33]. Note that Bij
is quantized only if R acts with the same sign on x i and x j . Otherwise Bij is a modulus
and instead Gij (the off-diagonal component of the metric) is quantized [34].
Acknowledgements
This work has been supported by TMR programs ERBFMRX-CT96-0045 and CT960090. We acknowledge discussions with Ralph Blumenhagen, Jan Conrad, Boris Krs and
Hans Peter Nilles.
Appendix A. Computation of one-loop diagrams
In this appendix we will give the details of the computation of the diagrams Fig. 1 in
the loop channel, i.e., evaluate the expressions (10), (11), and (12). We follow closely the
notation of [20]. First, we introduce abbreviations by identifying these expressions with
!
Z
M
N
X
X
dt
1
(n1 ,k1 )(n2 ,k2 ) (n1 ,k1 )(n2 ,k2 )
K
LK
,
(A.1)
K = (1 1)4c
t 3 4NM
0

n1 ,k1 =0 n2 ,k2 =0

M = (1 1)c

N
X
dt
1
t 3 4NM

M
A = (1 1)c

dt
t3

(N1,M1)
X

n1 ,k1 =0 n2 ,k2 =0 (i1 ,i2 )=(0,0)

M
X

LM

M
X

(N1,M1)
X

n1 ,k1 =0 n2 ,k2 =0 (i1 ,i2 )=(0,0)

LA

(A.2)




1 ,i2 n2 ) 1
tr k(i11k,i2 2 ) tr k(i11kn
2

(n1 ,k1 )(n2 ,k2 ) (n1 ,k1 )(n2 ,k2 )(i1 ,i2 )

(n1 ,k1 )(n2 ,k2 ) (n1 ,k1 )(n2 ,k2 )(i1 ,i2 )

N
X
1
4NM



(i ,i )
1 (i1 n1 ,i2 n2 )
T
tr 1Rk21 k2
Rk1 k2

!
.

(A.3)

Let us first explain the meaning of the various symbols in words and later on give the
explicit expressions. The letters K()(), M()() and A()() stand for oscillator contributions.

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

147

The upper quadruple index describes the twist sector and the insertion of a ZN ZM
element as follows: the contribution corresponds to the 1n1 2n2 twisted sector with a
1k1 2k2 insertion in the trace. (For simplicity, open strings ending on different types of
D-branes are called twisted as stated in the text.) The s are the matrix representations
of the orientifold group as in [8]. The lower index stands for the corresponding group
element, e.g., Rk1 k2 corresponds to R1k1 2k2 . The upper double index (i1 , i2 ) labels
the different types of D-6-branes as described in the text. Finally, L stands for the lattice
contribution (i.e., sums over discrete momenta and windings). The indexing is like in the
oscillator contributions given above.
A.1. Lattice contributions
The explicit expressions for the lattice contributions are
(n ,k1 )(n2 ,k2 )

1 ,k1 )(n2 ,k2 )

= K 1
L(n
K

(n1 ,n2 )
TrKK+W
R1k1 2k2 e2t (L0+L0 ) ,

(n1 ,k1 )(n2 ,k2 )(i1 ,i2 )

1 ,k1 )(n2 ,k2 )(i1 ,i2 )
= M
L(n
M

(i ,i ),(i1 n1 ,i2 n2 )

1 2
TrKK+W

(n ,k1 )(n2 ,k2 )(i1 ,i2 )

LA1

k
k
R1 1 2 2 e2t L0 ,

(A.4)

(A.5)

(n ,k1 )(n2 ,k2 )(i1 ,i2 )

= A 1

(i1 ,i2 ),(i1 n1 ,i2 n2 )
TrKK+W
1k1 2k2 e2t L0 .

(A.6)

In the Klein bottle is the number of the corresponding fixed points 6 whereas in the open
string amplitudes is the intersection number of the branes involved. (The indexing is
analogous to the one described above.) The upper index at Traces gives the twist sector.
Sums over windings and momenta lead to expressions of the form
!
!
X
X
t m2 /
t n2
e
e
,
(A.7)
L[, ]
mZ

where

nZ

= r 2 / 0 .

In the tree channel the corresponding function is defined as

!
!
X
X
lm2
ln2 /
e
e
e
.
L[, ]
mZ

(A.8)

nZ

The transformation from the loop channel to the tree channel is performed by Poisson
resummation,
X
X n2 t
2
en /t = t
e
.
(A.9)
nZ

nZ

The exact form of lattice contributions depends on the model and the results are given in
the text.
6 In more detail: It is the number of n1 n2 fixed points which are left invariant under R k1 k2 .
1
2
1
2

148

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

A.2. Oscillator contributions

The general expressions for the oscillator contributions are


(n1 ,n2 )
R1k1 2k2 (1)F e2t (L0+L0 ) ,
K(n1 ,k1 )(n2 ,k2 ) = TrNSNS


(0,0)(n1 ,n2 )
k
k
R1 1 2 2 e2t L0 ,
M(n1 ,k1 )(n2 ,k2 ) = TrR


(0,0)(n1 ,n2 )
1k1 2k2 (1)F e2t L0 .
A(n1 ,k1 )(n2 ,k2 ) = TrNS

(A.10)
(A.11)
(A.12)

Because of the R insertion in the Klein bottle partition function the expression (A.10) is
actually independent of k1 , k2 and we define
K(n1 ,n2 ) K(n1 ,k1 )(n2 ,k2 ) .

(A.13)

The upper index at the open string amplitudes indicates the boundary conditions, i.e.,
the computation is done for a string stretching between brane (0, 0) and (n1 , n2 ).
Equivalently, one could trace over open strings stretching between branes (i1 , i2 ) and
(i1 n1 , i2 n2 ).
The oscillator contributions can be expressed in terms of Jacobi theta functions and the
Dedekind eta function,
 
X (n+)2

q 2 e2i(n+) ,
(A.14)

(t) =

nZ

(t) = q 24

(1 q n ),

(A.15)

n=1

with q = e2t . One finds



K(n1 ,n2 ) =

0
1/2

M(n1 ,k1 )(n2 ,k2 ) =

1/2
0




1/2
0

(n1 vi +n2 wi ,k1 vi +k2 wi )Z2

k1 vi + k2 wi

1/2 + n1 vi + n2 wi
1/2 + k1 vi + k2 wi

0
1/2

0
1/2

!

(A.16)
!

eihn1 vi +n2 wi i

(A.17)


n1 vi + n2 wi
1/2 + k1 vi + k2 wi


+ n1 vi + n2 wi
1/2
1/2 + k1 vi + k2 wi

(2i)

(n1 vi +n2 wi ,k1 vi +k2 wi )Z

/ 2

!

n1 vi +n2 wi Z

eihn1 vi +n2 wi i

(2i) 1/2 + n1 vi + n2 wi

(n1 vi +n2 wi ,k1 vi +k2 wi )Z2

0
1/2

1/2 + n1 vi + n2 wi
1/2

n1 vi + n2 wi
1/2

(n1 vi +n2 wi ,k1 vi +k2 wi )Z

/ 2

n1 vi +n2 wi
/Z


(n1 ,k1 )(n2 ,k2 )

!
e

ihn1 vi +n2 wi i

!

(A.18)

The arguments in the theta and eta functions are 2t in the Klein bottle, t + 2i in the Mbius
strip, and t in the annulus. Further, we used the notation [20],

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

149

1
hxi x [x] ,
(A.19)
2
where the brackets on the rhs denote the integer part, and

1
(A.20)
= 1 if (n1 vi + n2 wi , k1 vi + k2 wi ) Z Z + 2 ,
0 else.
e() , M
e() and A () can be evaluated with the help of the
The tree channel expressions K
modular transformation properties,


 
2i

(t),
(A.21)

(1/t) = t e

(A.22)
(1/t) = t (t).
As usual, there is a subtlety in the Mbius strip. Before performing the modular
transformation one writes the theta functions with complex arguments as a product of theta
functions with real arguments [8]. Since the calculation is a straightforward modification
of the one presented in the appendix of [20] and the formulas are rather lengthy, we do not
give the explicit tree channel expressions here.

Appendix B. Tables of massless spectra

In this appendix we collect tables giving the massless spectra of Z2 Z2 , Z6 Z3 and
Z3 Z3 R-orientifolds. The spectrum of the Z4 Z2 model is given in the text.
B.1. The Z2 Z2 model
Closed spectrum of Z2 Z2
Twist-sector

AAA

AAB

untwisted
1
2
1 2

16C
16C
16C

16C
12C + 4V
12C + 4V

ABB
SUGRA + 6C
12C + 4V
10C + 6V
12C + 4V

BBB
10C + 6V
10C + 6V
10C + 6V

(B.1)
Open spectrum of Z2 Z2
Twist-sector

AAA

AAB

untwisted

1
2
1 2

ABB

BBB

1V
3C
1C
1C
1C

1C
2C
2C

2C
4C
2C

4C
4C
4C

Gauge group/matter
[Sp(M/4)]4
(A, 1, 1, 1) (1, A, 1, 1)
(1, 1, A, 1) (1, 1, 1, A)
(F, F, 1, 1) (1, 1, F, F )
(F, 1, F, 1) (1, F, 1, F )
(F, 1, 1, F ) (1, F, F, 1)

(B.2)

150

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

B.2. The Z6 Z3 model

Closed spectrum of Z6 Z3
Twist-sector

ABA

ABB

untwisted
1 + 15

2C

2C

8C +2V

8C + 2V

5C + 1V

5C + 1V

8C + 4V

12C

2C + 1V

3C

8C + 4V

12C

4C + 2V

6C

9C + 6V

12C + 3V

4C + 2V

6C

SUGRA + 3C

12 + 14
13
2 + 22
1 2 + 15 22
12 2 + 14 22
13 2 + 13 22
14 2 + 12 22
15 2 + 1 22

(B.3)
Open spectrum of Z6 Z3
Twist-sector

ABA

ABB

Representation of U (2) U (2)

untwisted

1V
1C
4C

(4, 10 ) (10 , 4)
(4, 10 ) (10 , 4)
(1, 10 ) (10 , 1)

1 + 15

4C

(2,
2)
(2, 2)

12 + 14

2C
8C

(4, 10 ) (10 , 4)
(1, 10 ) (10 , 1)

13

4C

(2,
2)
(2, 2)

2 + 22

2C
4C
2C

6C
12C
6C

(4, 10 ) (10 , 4)
(1, 10 )
(4, 10 )

1 2 + 15 22

2C

6C

(2,
2)
(2, 2)

12 2 + 14 22

2C
4C
2C

6C
12C
6C

(1, 10 ) (10 , 1)
(10 , 1)
(10 , 4)

13 2 + 13 22

4C

12C

(2,
2)
(2, 2)

14 2 + 12 22

2C
2C

6C
6C

(4, 10 ) (10 , 4)
(3, 10 ) (10 , 3)

15 2 + 1 22

4C

12C

(2,
2)
(2, 2)

(B.4)

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

151

B.3. The Z3 Z3 model

Closed spectrum of Z3 Z3
Twist-sector

AAA

AAB

ABB

BBB

untwisted
1 + 12

10C + 8V

SUGRA + 3C
10C + 8V

12C + 6V

18C

2 + 22

10C + 8V

12C + 6V

18C

18C

1 2 + 12 22

10C + 8V

12C + 6V

12C + 6V

18C

12 2 + 1 22

14C + 13V

15C + 12V

18C + 9V

27C

(B.5)
Open spectrum of Z3 Z3
Twist-sector

AAA

AAB

untwisted

ABB

BBB

Rep. of SO(4)

1V
3C

1 + 12

2C

2C

6C

18C

6
6
6

2 + 22

2C

6C

18C

18C

1 2 + 12 22
12 2 + 1 22

2C

6C

6C

18C

1C

3C

9C

27C

10

(B.6)
Appendix C. Projective representations of R Z4 Z2
In this appendix we give the explicit expressions for the projective representation used in
the Z4 Z2 orientifold in Section 3. The projective representation of the Z2 Z2 subsector
is chosen to be equivalent to the one presented in [11]. Mi , Ni and D are as defined in [11],


 
 
0 1
i2 0
0 i2
,
,
,
Mi =
i2 0
1 0
0
i2


0 i2
,
D =
i2
0

 
 

i2 0
0 1
1 0
,
,
,
(C.1)
Ni DMi =
1 0
0 1
0 i2


2
2
2
2
2
where 2 = 0i i
0 and the matrices fulfill Mi = N1 = D = N2 = N3 = 1. In
addition we use the notation

152

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

1
a = (1 2 ),
2
to define

A =


a
b

b
a

b = ia T


(C.2)

A1 = A,


b a
,
a b


a b
C = i
,
b a


b a
E = i
,
a b
B =

B 1 = B,


a
C =i
b

b
E 1 = i
a
1

b
a

a
b

,

.

(C.3)

The latter satisfy C T = iE. Furthermore we need the set of matrices





1
1
1 1
i i
,
H=
,
F =
2 i i
2 1 1




1
1
1 1
i
i
,
K=
,
G=
2 i i
2 1 1

(C.4)

which satisfy F T = G and H T = K. The representation matrices of the orientifold

group R Z4 Z2 are listed in Table (C.5). In the first column, we list the systematic
numbering of the branes as explained in Section 3.2, and in the second column we give the
simplified numbering used in Section 3.5. All relevant signs are listed explicitly whereas
the others are arbitrary.
The representation matrices of the orbifold group Z4 Z2 are obtained as explained in
Eq. (40) of Section 3.5. They are listed in Table (C.6) up to an irrelevant phase.
Representation-matrices of R Z4 Z2
R

R1

R12

R13

R2

R12

R12 2

R13 2

+A

M1

+B

M2

+C

M3

+E

brane
(0, 0)
(1, 0)

+B

+A

M1

+E

M2

+C

M3

(2, 0)

N1

+AT

+B T

M2

C T

M3

+E T

(3, 0)

+B T

N1

+AT

+E T

M2

C T

M3

(0, 1)

N1

+F

M3

+G

+H

M2

+K

(1, 1)

+G

N1

+F

M3

+K

+H

M2

M2

+GT

M3

+H T

+K T

F T

M2

+K T

M3

+H T

(2, 1)

N1

F T

(3, 1)

+GT

N1

(C.5)

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

153

Representation-matrices of Z4 Z2
brane
(0, 0)
(1, 0)
(2, 0)
(3, 0)
(0, 1)
(1, 1)
(2, 1)
(3, 1)

1
2
3
4
5
6
7
8

12

13

12

12 2

13 2

B T
B T
AT
AT
GT N1
GT N1
N1 F T
N1 F T

M1
M1
M1
M1
N2
N2
N3
N3

AT
AT
BT
BT
F T N1
F T N1
N1 GT
N1 GT

M2
M2
N3
N3
M1
M1
N2
N2

E T
E T
CT
CT
K T N1
K T N1
N1 H T
N1 H T

M3
M3
N2
N2
N3
N3
M1
M1

C T
C T
ET
ET
H T N1
H T N1
N1 K T
N1 K T

(C.6)

References
[1] J. Polchinski, Phys. Rev. Lett. 75 (1995) 4724, hep-th/9510017.
[2] J. Polchinski, E. Witten, Nucl. Phys. B 460 (1996) 525, hep-th/9510169.
[3] A. Sagnotti, ROM2F-87/25, Talk presented at the Cargese Summer Institute on NonPerturbative Methods in Field Theory, Cargese, Italy, July 1630, 1987.
[4] G. Pradisi, A. Sagnotti, Phys. Lett. B 216 (1989) 59.
[5] J. Govaerts, Phys. Lett. B 220 (1989) 77.
[6] M. Bianchi, A. Sagnotti, Phys. Lett. B 247 (1990) 517.
[7] P. Horava, Nucl. Phys. B 327 (1989) 461.
[8] E.G. Gimon, J. Polchinski, Phys. Rev. D 54 (1996) 1667, hep-th/9601038.
[9] A. Dabholkar, J. Park, Nucl. Phys. B 472 (1996) 207, hep-th/9602030.
[10] E.G. Gimon, C.V. Johnson, Nucl. Phys. B 477 (1996) 715, hep-th/9604129.
[11] M. Berkooz, R.G. Leigh, Nucl. Phys. B 483 (1997) 187, hep-th/9605049.
[12] J.D. Blum, A. Zaffaroni, Phys. Lett. B 387 (1996) 71, hep-th/9607019.
[13] Z. Kakushadze, G. Shiu, Phys. Rev. D 56 (1997) 3686, hep-th/9705163.
[14] G. Zwart, Nucl. Phys. B 526 (1998) 378, hep-th/9708040.
[15] S. Frste, D. Ghoshal, Nucl. Phys. B 527 (1998) 95, hep-th/9711039.
[16] D. ODriscoll, hep-th/9801114.
[17] G. Aldazabal, A. Font, L.E. Ibanez, G. Violero, Nucl. Phys. B 536 (1998) 29, hep-th/9804026.
[18] M. Berkooz, M.R. Douglas, R.G. Leigh, Nucl. Phys. B 480 (1996) 265, hep-th/9606139.
[19] R. Blumenhagen, L. Grlich, B. Krs, Nucl. Phys. B 569 (2000) 209, hep-th/9908130.
[20] R. Blumenhagen, L. Grlich, B. Krs, JHEP 0001 (2000) 040, hep-th/9912204.
[21] J. Polchinski, Y. Cai, Nucl. Phys. B 296 (1988) 91.
[22] M. Klein, R. Rabadan, JHEP 0007 (2000) 040, hep-th/0002103.
[23] Z. Kakushadze, Nucl. Phys. B 535 (1998) 311, hep-th/9806008.
[24] M. Bianchi, G. Pradisi, A. Sagnotti, Nucl. Phys. B 376 (1992) 365.
[25] R. Blumenhagen, L. Grlich, B. Krs, D. Lst, Nucl. Phys. B 582 (2000) 44, hep-th/0003024.
[26] J. Lauer, J. Mas, H.P. Nilles, Phys. Lett. B 226 (1989) 251.
[27] J. Lauer, J. Mas, H.P. Nilles, Nucl. Phys. B 351 (1991) 353.
[28] M.R. Douglas, C. Hull, JHEP 9802 (1998) 008, hep-th/9711165.
[29] A. Connes, M.R. Douglas, A. Schwarz, JHEP 9802 (1998) 003, hep-th/9711162.
[30] V. Schomerus, JHEP 9906 (1999) 030, hep-th/9903205.
[31] N. Seiberg, E. Witten, JHEP 9909 (1999) 032, hep-th/9908142.

154

S. Frste et al. / Nuclear Physics B 593 (2001) 127154

[32] M.R. Douglas, hep-th/9807235.

[33] M.R. Douglas, B. Fiol, hep-th/9903031.
[34] C. Angelantonj, R. Blumenhagen, Phys. Lett. B 473 (2000) 86, hep-th/9911190.

Nuclear Physics B 593 (2001) 155182

www.elsevier.nl/locate/npe

On D-branes from gauged linear sigma models

Suresh Govindarajan a, , T. Jayaraman b , Tapobrata Sarkar c
a Department of Physics, Indian Institute of Technology, Madras, Chennai 600 036, India
b The Institute of Mathematical Sciences, Chennai 600 113, India
c Department of Theoretical Physics, Tata Institute of Fundamental Research, Homi Bhabha Road,

Mumbai 400 005, India

Received 7 August 2000; accepted 12 October 2000

Abstract
We study both A-type and B-type D-branes in the gauged linear sigma model by considering
worldsheets with boundary. The boundary conditions on the matter and vector multiplet fields are first
considered in the large-volume phase/non-linear sigma model limit of the corresponding CalabiYau
manifold, where we find that we need to add a contact term on the boundary. These considerations
enable to us to derive the boundary conditions in the full gauged linear sigma model, including the
addition of the appropriate boundary contact terms, such that these boundary conditions have the
correct non-linear sigma model limit. Most of the analysis is for the case of CalabiYau manifolds
with one Khler modulus (including those corresponding to hypersurfaces in weighted projective
space), though we comment on possible generalisations. 2001 Elsevier Science B.V. All rights
reserved.

1. Introduction
The improved understanding of non-perturbative aspects of string theory in recent years
has shown that all five (perturbative) superstrings appear to be different corners in the
moduli space of a single theory [1]. A consequence of this is that while the existence of
a perturbative string theory description at these corners singles out strings as fundamental
objects, at generic points in this moduli space, there is a certain democracy among all
objects, fundamental as well as solitonic, as seen in the perturbative string theory. Thus,
the heterotic string appears as a soliton (D1-brane) in the type I theory. A more general
analysis indicates that an object which is a soliton at one point in the moduli space can
become a fundamental excitation at another point in the moduli space [2].
* Corresponding author.

E-mail addresses: [email protected] (S. Govindarajan), [email protected] (T. Jayaraman),

[email protected] (T. Sarkar).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 1 1 - 8

156

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

Another question of interest is the nature of spacetime at short distances. It turns out that
the answer is related to the kind of probe which is used. Given the democracy among all
objects one can probe spacetime using both fundamental strings as well as solitons such as
D-branes. Fundamental strings probe objects which are of the string scale ls while D-branes
in perturbative string theory probe much shorter scales (gs )a ls , where gs is the string
coupling constant and a is a positive constant [3]. From earlier studies of closed strings, it is
known that strings can propagate in apparently singular spaces such as orbifolds. D-brane
probes see these spacetime geometries in a manner different from that of closed string
theories. For example, it was shown that for D0-brane probes of the orbifold C3 / (where
is a discrete subgroup of SU(3)), the non-geometric phases seen by the closed string are
projected out [4].
While much is known about the nature of fundamental strings probing various spacetime
geometries, our understanding in the case of D-brane probes is in a much more primitive
state, apart from the cases of flat space and toroidal and orbifold backgrounds. In the
last couple of years, there has been considerable progress in understanding D-branes in
the context of string compactification on CalabiYau threefolds [518]. Unlike the case
of toroidal compactifications, these correspond to fewer unbroken supersymmetries and
thus fewer constraints follow. For example, the BPS conditions leave open the possibility
of having lines of marginal stability in the moduli space, where a D-brane can decay.
A D-brane which fills the non-compact spacetime while also wrapping some cycle of the
CY three-fold can possess a non-trivial superpotential in its worldvolume gauge theory. It
is of interest to derive this superpotential and its dependence on closed string moduli.
D-branes on CY manifolds fall into two distinct categories: A-type branes are those
which wrap special Lagrangian submanifolds while B-type branes wrap holomorphic
submanifolds of the CY manifold. In the worldsheet description [5], A-type branes and
B-type branes differ in the worldsheet supersymmetry that they preserve. In the openstring channel, A-type branes are compatible with the topological theory obtained with the
A-twist and B-type branes are compatible with the B-twist. In the closed-string channel, the
roles are reversed due to a change of sign in the boundary conditions on the U (1) currents
of the (2, 2) worldsheet supersymmetry algebra. In the closed-string case [19], correlation
functions of the observables in the topological A-model are independent of complex
structure moduli (of the CY) while those in the topological B-model are independent of
the Khler moduli. The modified geometric hypothesis proposed in [7,8] is in a sense the
open-string version of this statement. Based on this, one (loosely speaking) expects the
lines of marginal stability of A-branes and the superpotential of B-branes to be independent
of Khler moduli and thus calculable in the large volume limit where classical geometry
can be applied. (See [8] for a more detailed and careful discussion.)
Tests of the modified geometric hypothesis as well as the extended version of mirror
symmetry that includes D-branes and their worldvolume theories will need a worldsheet
description of CY manifolds where both Khler and complex moduli have simple
realisations. The gauged linear sigma model (GLSM) is a suitable worldsheet description
in this regard. As shown by Witten, this model has several phases of which the Calabi
Yau phase is one. Thus, the enlarged Khler cone which is required by mirror symmetry

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

157

naturally fits into the setup [24]. The price one pays for this choice is that conformal
invariance on the worldsheet is obtained only at the infrared fixed point of the GLSM.
One of the advantages of the GSLM in the closed string case is the fact that it unifies the
different techniques that are preferred in different regions of Khler moduli space. In the
CalabiYau phase or the large volume phase, the GLSM description tends to the non-linear
sigma model description of strings moving on CY manifolds. In a LandauGinzburg phase,
the description would be in terms of N = 2 supersymmetric LandauGinzburg theory. In
particular cases, the description at this point in the moduli space is even more explicit
when the LG theory is equivalent to the tensor product of a set of N = 2 minimal model
conformal field theories. One may hope to see a similar situation in the case of D-branes.
In this paper, as a first step towards the eventual goal described earlier, we study the
GLSM with (2, 2) supersymmetry on worldsheets with boundary. Due to the presence of a
boundary, one has to specify boundary conditions on the various fields in the GLSM such
that the appropriate linear combination of supersymmetry is preserved. In order that these
boundary conditions correspond to D-branes wrapped around various cycles of the Calabi
Yau manifold, we first construct the boundary conditions in the nonlinear sigma model
limit of the GLSM and look for boundary conditions in the GLSM which reduce to sensible
ones in the NLSM limit. We also find the need to introduce a boundary (contact) term in
order to obtain consistent boundary conditions. This contact term vanishes when the theta
term in the GLSM is turned off and and its presence is justified by considering the NLSM
limit. By consistent boundary conditions, we mean those for which the boundary terms
which appear in the ordinary and supersymmetric variations vanish (modulo equations of
motion) and further that the set of boundary conditions are closed under the action of the
supersymmetry that is preserved on the boundary.
The organization of the paper is as follows: in Section 2 we begin by reviewing the d = 2,
N = 2 supersymmetric GLSM for closed strings following [20]. We then add a boundary
and compute the boundary terms generated in computing the equations of motion and the
variations of the action under supersymmetry. We also review and extend the work of [9,
10] describing boundary conditions for d = 2, N = 2 supersymmetric LandauGinzburg
(LG) models of conformal field theories: this will be useful in understanding the boundary
conditions in LG phases of CalabiYau compactifications. We close with a few words about
the justification for using the GLSM; in particular we discuss the topological twisting of
the GLSM with boundary. In Section 3 we construct boundary conditions describing branes
on supersymmetric cycles in the e2 of the GLSM, as a guide to understanding the
physical meaning of boundary conditions at finite e2 ; along the way we will find certain
boundary terms that we must add for consistency. In Section 4 we finally construct and
identify boundary conditions at finite e2 . We also discuss the significance of these boundary
conditions. In Section 5 we present our conclusions.
Parts of this work have been reported earlier elsewhere [21,22]. While readying this work
for publication, other papers [14,15,17] have also appeared that, in part, study D-branes
in the GLSM approach. While [17] use a combination of the GLSM and worldvolume
techniques , [15] is closer in spirit to the techniques of this paper and the results of Section 6
of their paper have some overlap with Section 4 of this work.

158

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

2. The gauged linear sigma model

In type II string theories compactified on CalabiYau threefolds, the moduli space of
Khler classes includes regions where the size of the manifold is of order ls . Often the
CFTs appear non-geometric and are better described via LandauGinzburg orbifolds [20,
23]; or they may mediate smooth passage to CalabiYau manifolds with different topology
[24]. We are particularly interested in studying the physics of D-brane probes as one moves
through large distances in the moduli space of such compactifications, across phases or
towards singular compactifications. Unfortunately even in the geometric phases of these
models the CalabiYau metric is not known, and physical objects (such as vertex operator
correlation functions) receive corrections from worldsheet instantons. At best there are a
few points in the moduli space where the conformal field theory is well understood: in
particular at large-radius limits, and exactly solvable Gepner points, the latter being deep
in the LandauGinzburg region.
The technique introduced in [20] to study motion between these regions was to write a
2d supersymmetric field theory with a known UV Lagrangian whose infrared fixed point is
believed to be a CalabiYau compactification. (In fact, this technique was an important
part of the development of the above picture.) This model is simply a d = 2, N = 2
supersymmetric gauge theory with some number of vector multiplets and some number
of charged chiral multiplets, and is called the gauged linear sigma model (GLSM). This
is much in the spirit of using LandauGinzburg models as UV Lagrangian descriptions of
minimal models: indeed, LandauGinzburg orbifolds appear in non-geometric phases of
the GLSM.
2.1. GLSM for closed strings
For ease of reference, we will review the lagrangian and supersymmetries of the
GLSM following [20]. We work in Minkowski space with the metric (, +). We are
interested in describing compactifications of string theory with eight supercharges; the
worldsheet conformal field theory must then have N = (2, 2) superconformal symmetry.
We expect that a nonconformal theory with such an infra-red fixed point should have N = 2
supersymmetry as well.
Our candidate theory can be obtained by dimensional reduction from d = 4, N = 1
abelian gauge theory with chiral multiplets. It contains s U (1) vector multiplets, described
by the vector superfields Va (a = 1, . . . , s) and k chiral multiplets described by the chiral
superfields i (i = 1, . . . , k). Written in components, the vector multiplet consists of the
vector fields va ( = 0, 1), the complex scalar field a , complex chiral fermions a , and
the real auxiliary field D a . The chiral multiplet consists of a complex scalar i , complex
chiral fermions i , and a complex auxiliary scalar field Fi . They are charged under the
U (1)s with charge Qai . In component notation, the supersymmetry transformations of the
vector multiplet are:


v0a = i + a+ +  a + + a+ +  a ,


v1a = i + a+  a + + a+  a ,

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

159

a = i 2 + a i 2  a+ ,

a = i 2 + a i 2 a+ ,

D a = + (0 1 )a+  (0 + 1 )a + + (0 1 ) a+ +  (0 + 1 ) a ,

a
+ ,
a+ = i+ D a + 2 (0 + 1 ) a  v01

a
 ,
a = i D a + 2 (0 1 ) a + + v01

a
a
a
a
+ ,
+ = i + D + 2 (0 + 1 )  v01

a
a
a
a
(2.1)
= i  D + 2 (0 1 ) + + v01  ,
where  and  are the Grassmann parameters for SUSY transformations. The
transformation rules for the chiral multiplet are:

i = 2 (+ i  +i ),

+i = i 2 (D0 + D1 )i  + 2 + Fi 2Qai i a + ,

(2.2)
i = i 2 (D0 D1 )i + + 2  Fi + 2Qai i a  ,

Fi = i 2 + (D0 D1 )+i i 2  (D0 + D1 )i

+ 2Qai + a i +  a +i + 2iQai i  a+ + a .
(2.3)
The supersymmetric bulk action can be written as a sum of four terms,
S = Sch + Sgauge + SW + Sr, .

(2.4)

The terms on the right-hand side are, respectively: the kinetic term for the chiral
superfields; the kinetic terms for the vector superfields; the superpotential interaction; and
the FayetIliopoulos and theta terms. Sch is:

XZ

d 2 x D i D i + i i D0 + D1 i + i +i D0 D1 +i
Sch =
i

+ |Fi |2 2

X a a

2


a a Qai i i 2
Qi +i i + a i +i

+D

Qai i i

X a


i 2
Qi i i a+ +i a
a


a


a
a

i 2Qi i +i + i ,

(2.5)

where


1
ADi B (Di A)B .
(2.6)
2
This symmetrized form of the fermion kinetic term is Hermitian in the presence of a
boundary. Meanwhile, Sgauge is:
X 1 Z




2
a 2
+ 12 D a a a
d 2 x 12 v01
Sgauge =
2
e
a
a

+ i a+ 0 1 a+ + i a 0 + 1 a .
(2.7)

A Di B

160

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

The superpotential term is:



Z
2S
S
2W
W
Si W W i +j .
+
i +j + F
SW = d 2 x Fi
i
i j
i
i j
Finally, the FayetIliopoulos D-term and theta term are:
Z
Z
a
2
a
a
d 2 x v01
.
Sr, = ra d x D +
2

(2.8)

(2.9)

We wish to describe a theory with an N = (2, 2) superconformal fixed point. Thus

we wish anomaly-free vector and axial U (1) R-symmetries: these may be constructed if
P a
i Qi = 0 [20]. These R-symmetries may also be used to topologically twist the theory,
as we will discuss below.
Let us review the manifestation of the target space geometry, and of the phase structure
of the moduli space of compactifications, in the class of examples which we will use for
most of this paper, namely hypersurfaces in weighted projective space with a single Khler
modulus. The spectrum is a single abelian vector multiplet and 5 chiral multiplets. The
latter consist of 5 chiral superfields i with positive charge Qi=1,...,5 and an additional
P
superfield 6 = P with charge Q6 = Qp = 5i=1 Qi . Furthermore, we choose the
superpotential
W (, P ) = PG(),

(2.10)

where G is a quasi-homogenous transverse polynomial. There are four such examples of

4
= P 4 (the quintic); a degree
Calabi-Yau threefolds: a degree five hypersurface in P1,1,1,1,1
4
4
; a degree eight hypersurface in P1,1,1,1,4
and a degree ten
six hypersurface in P1,1,1,1,2
4
hypersurface in P1,1,1,2,5.
We wish to find the moduli space of supersymmetric ground states, which should flow
to the target space of the infrared CFT. We do so by setting the bosonic potential energy
X

X
1
Fi2 + 2 D 2 + 2| |2
Q2i |i |2 + Q2p |p|2
(2.11)
U=
2e
i

to zero. We substitute the equations of motion for the auxiliary fields D and Fi :

X
Qi |i |2 + Qp |p|2 r ,
D = e2
i

W
,
Fi =
i
to find that:
U = |G(i )|2 + |p|2

(2.12)


X

X  G 2
2
2
2
2
 + D + 2| |2
i 
Q
|
|
+
Q
|p|
,
p
i i
i 
2e2
i

(2.13)

where p and i represent the scalar components of P and i=1,...,5 , respectively.

Let us begin with the case r  0. The D term requires that the i cannot all
simultaneously vanish. Thus = 0; transversality of G requires that p = 0. One must
also set the FayetIliopoulos D-term to zero:

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

Qi |i |2 = r.

161

(2.14)

This condition together with dividing out the U (1) gauge symmetry means that the i
4
describe the weighted projective space PQ
. Finally, the condition G = 0 means that
1 ,...,Q5
4
.
the live on a degree |Qp | hypersurface in PQ
1 ,...,Q5
For r  0, vanishing of the D term requires that p 6= 0. Transversality of G then implies
that all i = 0. This theory has a unique classical vacuum; the massless excitations are
governed by a superpotential with a degenerate critical point, i.e., it is a LandauGinzburg
theory. The residual gauge invariance (for instance, Z5 in the quintic) in fact implies
that it is a LandauGinzburg orbifold. At r this is believed to be the exactly
solvable Gepner model for the quintic [25] (see [26] for a review and references). In
this way the trajectory r  0 r  0 interpolates between geometric and non-geometric
compactifications.
Beginning with the flat metric on C5 and imposing the gauge invariance and D-term
conditions, one can see that r is essentially the size of the ambient projective space. In
spacetime this Khler parameter is complexified by an NSNS two-form potential; in this
model this flows from the theta angle. We will show this explicitly in the next section, but
we can note for now that the fact that is a periodic variable reflects the periodicity of the
2-form flux. Furthermore one can show that for 6= 2n, the GLSM is nonsingular even
at r = 0 [20].
2.2. GLSM with boundary
Supersymmetric D-branes configurations will preserve four of the eight spacetime
supercharges of the compactification. The boundaries of the string worldsheet must
therefore preserve half of the N = (2, 2) superconformal symmetry of the closed strings.
We take this to mean that boundaries in the corresponding GLSM should also break half
of the supersymmetries.
We will work on the half-plane (x 0 , x 1 ) with x 1 > 0, and impose boundary conditions
at x 1 = 0. As with the conformal sigma models, the possible boundary conditions fall
into two classes [5], A-type and B-type. Roughly these correspond to branes wrapped
on special Lagrangian submanifolds and on holomorphic cycles, respectively. In our case,
A-type boundary conditions correspond to setting  =  , where = 1; B-type
conditions correspond to  =  .
The variation of the action in the presence of a boundary will generate boundary
terms in addition to the bulk terms proportional to the equations of motion. One chooses
boundary conditions on the fields such that these boundary terms vanish. In addition,
SUSY variations of the fields will also generate boundary terms; upon choosing the
preserved supersymmetries one requires that these boundary terms also vanish. We list
these boundary terms here; in the next section we will use them to derive boundary
conditions.

162

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

1. The boundary terms in the action generated by general variations of the fields are:

Z





dx 0 1 i + iQai v1a i i + 1 i iQai v1a i i
ord Sch =





i
i i +i +i +i +i i i
,
+
2

Z






1
a
v0a 1 a a + 1 a a
ord Sgauge = 2 dx 0 v01
ea





i a a
a+ a+ a+ a+ a a ] ,
+
2
Z
a
(2.15)
dx 0 v0a .
ord Sr, =
2
2. The boundary terms generated by the transformation (2.1), (2.3) are:

Z





1 
0
dx (D0 i ) + i +  +i D0 i (+ i +  +i )
susySch =
2





1 
+ (D1 i ) + i  +i D1 i (+ i  +i )
2


+ iQai i a + +i + i a + +i + i a  i + i a  i





i 
Si + + +i  i Fi ,
+ + +i  i F
2

Z








1
i 
susySgauge = 2 dx 0 0 a + a  a+ 0 a  a+ + a
ea
2









i
+ 1 a + a +  a+ + 1 a  a+ + + a
2

i a  a
v01 + + +  a + + a+ +  a
2


Da 
+ a+  a +  a + a+ ,
+
2


S
Z

W


0 W

 i + +i +
 i + +i ,
susySW = i 2 dx
i
i
Z


ia
dx 0 + a+ + + a+ +  a +  a .
(2.16)
susySr, =
2
2.3. LandauGinzburg theories with boundary
Finding appropriate boundary conditions is fairly complicated; in addition, we would
like to know their physical import. One tool is to try and understand sensible boundary
conditions for r  0 and r  0, where gauge theory/worldsheet instanton corrections
are small [20,26]. The former limit is described by a nonlinear sigma model and we will
discuss this in the next section. The latter limit is described by a LandauGinzburg theory,

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

163

and we review here supersymmetric boundary conditions for such theories [9,10,15]. 1
2.3.1. A-type boundary conditions
Consider a LandauGinzburg model with n chiral superfields i and arbitrary superpotential G(). For A-type boundary conditions, we impose n independent conditions


(2.17)
fa , = 0,
where fa are real functions. We will use the indices i, j, . . . , to denote the superfields and
the indices a, b, c, . . . , to indicate the boundary conditions. Let denote the sub-manifold
in Cn (with complex coordinates i and ) obtained by imposing these conditions. We
will in addition impose the compatibility condition:


fb (, )

= 0,
(2.18)
fa (, ),
PB
where:

{A, B} = g i j i Aj B j Ai B .

(2.19)

We will assume that on , the normals nE a (i fa , i fa ) span the normal bundle N .

The vanishing of the Poisson bracket can be rewritten as
nEa tEb = 0,

(2.20)

where tEb (i fb , i fb ) are tangent vectors to the curve fb = 0. It follows that they span
the tangent bundle T . is thus a Lagrangian submanifold of Cn by construction [29].
The induced metric (first fundamental form) on is given by
hab = tEa tEb = nEa nE b .

(2.21)

The following set of additional boundary conditions on the fields in the LG model
are consistent with the vanishing of the boundary terms which occur in the general and
fa
i . Then:
supersymmetric variations of the LG Lagrangian. Define a
i


+a + a = 0,

fa
fa
b c

1 i
1 i iKabc = 0,
i
i


S )
G() G(

= 0.
fa (, ),
PB


(2.22)
(2.23)
(2.24)

The complex conjugate conditions are also hold. Kabc is the extrinsic curvature tensor
(second fundamental form) given by

fc fb 2 fa
fc fb 2 fa

Kabc =
i j i j
i j i j

2
fc fb fa
fc fb 2 fa

+
.
(2.25)
j i i j
i j i j
This full set of boundary conditions is equivalent to requiring that be Lagrangian.
Without a superpotential, this corresponds to the microscopic (worldsheet) realisation of
1 Previous work on LandauGinzburg theories with boundary can be found in [28].

164

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

situations considered by Harvey and Lawson [29]. In the presence of a superpotential, there
is an additional condition that the real conditions Fa have a vanishing Poisson bracket with
S This suggests that one must choose one of the conditions to be F = (G G)
S ic
(G G).
where c is a real constant, as there can only n independent commuting constants of motion
in a 2n real-dimensional phase space
2.3.2. B-type boundary conditions
Under B-type boundary conditions, the unbroken N = 2 supersymmetry is given by the
condition
+ =  ,

(2.26)

where = 1. The following linear boundary conditions were constructed in the LG

model [9]


+i + Bi j j x 1 =0 = 0,


1 i + Bi j j x 1 =0 = 0,


0 i Bi j j x 1 =0 = 0,


G 
G
+ Bi j
= 0,
(2.27)

 1
i

x =0

where the boundary condition is specified by a hermitian matrix B which satisfies B 2 = 1.

Since B squares to one, its eigenvalues are 1. An eigenvector of B with eigenvalue of +1
corresponds to a Neumann boundary condition and 1 corresponds to a Dirichlet boundary
condition. Associated with every eigenvector with eigenvalue +1, there is a non-trivial
condition involving the superpotential which is given by the last of the above boundary
conditions.
More general possibilities are given by boundary conditions corresponding to a
holomorphic submanifold Cn defined by the transverse intersection of the r
conditions
fm () = 0

(m = 1, . . . , r),

(2.28)

where fm are quasi-homogeneous holomorphic functions of the i . Under supersymmetric

variation with  + =  , one obtains the conditions
X
nim ()(+i i ) = 0,
(2.29)
i

where nim (fm /i ) are the (holomorphic) normals to the surface fm = 0. Let tia ()
(a = 1, . . . , n r) be a basis of tangent vectors to such that in local holomorphic
coordinates za , tia = i /za . One needs to impose further boundary conditions in order
to cancel fermionic boundary terms arising in the ordinary variation of the action:

(2.30)
i j tia () +j + j = 0.
Supersymmetric variation of the above equation leads to the conditions

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182





i tia

i j tia ()1 j +
(+k k ) +j + j
= 0,
2 k
G
= 0.
tia ()
i
The first equation can be rewritten in the following form

tia ()i j 1 j mab b m = 0,

165

(2.31)
(2.32)

(2.33)

where mab is the extrinsic curvature of the submanifold (second fundamental form) given
by (see Ref. [30] for a discussion)
mab tia tjb

2 fm
i j

and ha b tEa tEb is the induced metric (first fundamental form). We have also defined
fermionic linear combinations a and m
(+i i ) = tia a ,


j
+i + i ij = nm m .
These are fermionic combinations which are sections of the tangent bundle and normal
bundle, respectively. The boundary terms under ordinary variations of the LG action vanish
for the above choice of boundary conditions.
The boundary conditions involving the superpotential given by Eq. (2.32) is always
satisfied if one chooses one of the boundary conditions to be G|x 1 =0 = 0. For instance,
all examples of B-type boundary conditions in LG models considered in [9] can be seen to
imply G|x 1 =0 = 0. This requirement has also been observed independently in [15]. This is
S x 1 =0 = 0 condition seen in A-type boundary conditions.
the analogue of (G G)|
2.4. Topological aspects of the GLSM
As already mentioned, the GLSM is not conformally invariant and flows to a
conformally invariant theory in the infrared (IR) limit [20]. It follows that one must be
careful in naively extrapolating results in the GLSM to the conformally invariant fixed
point. For example, in the NLSM limit of the GLSM for the quintic, the metric is given by
the pullback of the FubiniStudy metric on P 4 . This metric is clearly not the correct one.
The fact that both the GLSM and its IR fixed point have worldsheet (2, 2) supersymmetry is
quite useful in obtaining some control. To be precise, by appropriately twisting the theories,
one constructs topological theories whose observables are insensitive to such differences
between the two theories and one can make predictions.
There are two possible twists of the (Euclidean) (2, 2) model: the A-twist corresponds
R
S+ (where Q = G are
to the case when the supersymmetry charge: Q = Q + Q
the supersymmetry charges associated with the supersymmetry generators G ) becomes
S+ is a scalar. Physical states of the
S + Q
a scalar and the B-twist is where Q = Q
topological theory correspond to cohomology classes of Q and the observables are given by
correlation functions of vertex operators that are Q-closed. Observables of the topological

166

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

A-model vary holomorphically in t = 2

+ ir while those in the B-model are independent
of t. Correlation functions in the A-model can receive corrections from gauge theory
instantons which do not quite coincide with the worldsheet instantons corrections seen
in the conformally invariant NLSM. The difference arises because the instanton moduli
space for the GLSM is compact while that of the NLSM is non-compact [20]. However, it
has been shown that singularity structure of the moduli space is correctly predicted in the
GLSM [20,27].
For the case of GLSM with boundary, one may hope to apply similar techniques. As has
been pointed out earlier [5], A-type boundary conditions are compatible with the A-twist
and B-type boundary conditions with the B-twist in the open-string channel. For instance,
in the topological A-model, the field of the vector multiplet is Q-closed. This can be
easily seen by the fact that = 0 under the supersymmetry transformation generated by
Q, i.e., + =  . Further, in the NLSM limit (cf. Section 3) we will see that
P
Qi +i i
.
= i
2 K[]
In the topological A-model, +i is a (1, 0) form (to be precise, a section of (T 1,0 (X)))
and i a (0, 1) form and hence is proportional to the Khler form on the CalabiYau
manifold. The proportionality constant K[] can be seen to be non-vanishing everywhere.
It is known that A-branes wrap special Lagrangian submanifolds of the CalabiYau
manifold. Lagrangian submanifolds satisfy the condition that the restriction of the Khler
form to the submanifold vanishes. Thus, for A-type boundary conditions, it follows that

|x 1 =0 = 0.

(2.34)

We will see that our analysis in the sequel will be consistent with this condition.
Just as in the closed string case, the interpretation of results in the case of GLSM with
boundary should be done with care. For instance, in the case of A-type branes, worldsheet
instanton effects lead to a stringy notion of the topology of the cycle which can differ
significantly from the topology computed by geometric means [16]. For the case of Bbranes, as we move around in the Khler moduli space, the branes can undergo monodromy
transformations; thus, if one writes down boundary conditions far from the large-radius
limit and tries to understand it by following that boundary condition out to the largeradius limit, the result can depend on the path one takes. It is also possible that a boundary
state at the Gepner point has no stable large-radius analog [8,12]. Finally, although taking
monodromies and lines of marginal stability into account when studying D-branes far from
the large radius limit is nontrivial, some progress has been made [7,8,12]. It would be
interesting to study these effects in the GLSM.

3. The nonlinear sigma model

Our eventual goal is to use the boundary GLSM as a tool for understanding the boundary
CFT to which it should flow in the infrared. To begin with, we would like to understand
what a given set of boundary conditions for the GLSM might correspond to in the infrared.

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

167

As discussed in [20], the theory flows to strong coupling in this limit. It is therefore
tempting to simply take the limit e2 and use these results as a physical guide.
In particular, in this limit the gauge kinetic terms drop out; upon integrating out the
nonpropagating gauge fields one is left with a nonlinear sigma model.
We will now look for consistent boundary conditions for the one-modulus examples in
the limit r  0. We hope this will provide a simple guide to finding boundary conditions
for finite e2 , as well as a crude guide to the infrared physics of the GLSM. We begin
by describing the results of integrating out the vector multiplets; along the way we will
have to add a contact term to reproduce sensible N = 2 NLSM results. Following this we
will discuss A-type boundary conditions and Neumann B-type boundary conditions in this
limit.
3.1. e2 limit of the bulk linear sigma model
In the e limit of the GLSM, the kinetic energy terms for the vector multiplet
vanish, so that the component fields behave as Lagrange multipliers. This leads to the
following constraints for general U (1) charge.
1. The D-term constraint:
X


(3.1)
Qi |i |2 + Qp |p|2 r = 0,
i

when r  0, |p| is very massive due to Eq. (2.13); we will set p = 0 for the remainder
of the section.
2. The constraints imposed by integrating out the gauginos are:
X
Qi i i = 0.
(3.2)
i

3. The equations of motion for and :

P
Qi +i i
,
= i
2K[]
P
Qi i +i
,
= i
2K[]
P
where K[] j Q2j |j |2 .
4. The equations of motion for the gauge fields:
X 



Qi i i 0 i i 0 i + i i + +i +i ,
2K[] v0 =
i

2K[] v1 =





Qi i i 1 i i 1 i i i + +i +i ,

The equation for v0 is simply Gauss law.

5. Further supersymmetric variation of the above equations leads to
X 


Si +i
Qi i (D0 + D1 )i + i F
2K[]+ =
i

(3.3)
(3.4)

(3.5)
(3.6)

168

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

i X 2
Qi i +i ,
K[]
i
X 

Si i
=
Qi +i (D0 D1 )i + i F
+

2K[]

i X 2
Qi i i .
K[]

(3.7)

Of course we have integrated out , but these last equations can be used to take the
e2 limit of equations which are functions of .
In N = 2 supersymmetric type II string compactifications on CalabiYau threefolds, the
Khler parameters are complexified by the fluxes of closed NSNS two-form gauge fields
Bij . In the GLSM, this is reflected by the presence of a -term for every FayetIliopoulos
D-term [20]. This can be seen easily in the e2 limit. Substituting Eqs. (3.5), (3.6)
into the theta term S of Eq. (2.9), we find:
Z
X  D B i D B i D B i D B i

0
0
1
Qi i 1
d 2x
S =
2
K[]
i




i i
+i +i
+ (0 1 )
.
(3.8)
(0 + 1 )
2K[]
2K[]
We define v0B to the fermion-independent part of v0 in (3.5); D0B is the covariant derivative
with v0 replaced by v0B .
In the case of the quintic, K = r and the bosonic part of (3.8) corresponds to:
Bi j =

i
.
4r i j

(3.9)

This is closed and topologically nontrivial in P 4 ; it should thus correspond to the NSNS
B-field modulus for type II compactification on the quintic. It remains to understand the
fermion bilinear term. Clearly it is a boundary term, and it can be discarded in the closedstring case. We claim that in the case at hand it is sensible to subtract this off, by adding an
explicit boundary term:
Z


X
quintic
+i +i + i i .
(3.10)
Sboundary = dx 0
4r
i

We will now consider the case of more general cases involving weighted projective
spaces. In d = 2, N = 2 supersymmetric NLSMs, the fermion bilinear terms which scale
with B are [31]:
Z

k
(3.11)
d 2 x j i k Hjik
+ Hji k ,
where H = dB. In N = 2 supersymmetric compactifications of type II string theory on
CalabiYau threefolds, we must have H = dB = 0 and so these bilinear terms must vanish.
Clearly H = 0 in Eq. (3.9). In fact this is true for the B-field arising from more general Qi .
The general expression for the B-field in the NLSM limit is given by

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

169

X ia  [(Qai )2 Qaj Qai (Qaj )2 ]i j 

Bij =
,
(3.12)
2
4(K a [])2
a

X ia  Qai i j
[(Qai )2 Qaj + Qai (Qaj )2 ]i j

,
(3.13)
Bi j =
2 2K a []
4(K a [])2
a
P
where K a [] i (Qai )2 |i |2 . An explicit calculation of Hij k and Hi jk leads to the
vanishing of the fermion bilinear in Eq. (3.11) on imposing the NLSM constraints given
by (3.1) and (3.2). This implies that H vanishes on the subspace given by the D-term
constraint.
For the general case of weighted projective spaces, the B-field corresponds to a nonconstant B-field. Further, Bij and Bij are non-vanishing. 2 Is it possible to find a spacetime
gauge transformation of the form
B =

(3.14)

such that Bij and Bij vanish? Under such a gauge transformation, the worldsheet
Lagrangian transforms as
Z





(3.15)
SB-field = 2 d 2 x 0 1 1 0 .
The following choice of gauge transformation does this:

X  ia Qa i
ia i
i

,
i =
8K a [] 8ra
a

X  ia Qa i
ia i
i
+

.
i =
8K a [] 8ra
a
Of course, this also modifies Bi j to the following form
X  ia 
,
Bi j =
4ra i j
a

(3.16)
(3.17)

(3.18)

which is a constant B-field as in the quintic! There are total derivative pieces given by the
gauge transformation (see Eq. (3.15)). We discard them and this corresponds to a further
addition to the existing contact term we obtained by discarding the fermion bilinears. For
the one-modulus case, the final contact term takes the following simple form
Z
X


i
NLSM
=
dx 0
i D0 i i D0 i ,
(3.19)
Sboundary
4r
i

where we have used Eq. (3.5) to rewrite the fermion bilinear in terms of the bosonic fields.
The attentive reader may observe that the requirement that Bij and Bij vanish does not
completely fix the required gauge transformation . An additional requirement is that the
2 One reason to require these to vanish is to observe that for D-branes, gauge invariance dictates that the field
strength F and the B-field occur (loosely) in the combination (F B). For holomorphic connections on B-branes,
one has Fij = Fij = 0. Even in the closed string case, it is preferable to work in this gauge since the complex
Khler moduli involve only Bi j .

170

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

boundary conditions in the NLSM and the LSM agree. Our choice of gauge transformation
as given above precisely achieves this. The important point to note is that the choice
of boundary conditions in the NLSM is dictated by the form of the B-field, whereas in
the LSM it is the contact term (i.e., the terms we discarded in the NLSM limit as total
derivatives) which dictates the choice of boundary conditions. This will become clearer in
the following sections.
As a final note, such a term arises in an almost identical context in [32]. In that work
the issue is the construction of a BRST-invariant vertex operator for the NSNS B-field on
the disc, for D-instantons in flat space. Whether the vertex operator describes a fluctuation
B for which H = 0, or one for which H is nonzero, one must add a contact term
on the boundary which is a fermion bilinear, in order that the integrated operator is BRST
invariant. If one writes this boundary term as a total derivative in the interior of the disc,
the total fermion part of the vertex operator has the right symmetry structure in its Lorentz
indices to describe a fluctuation of H .
3.2. A-type boundary conditions in the NLSM
In Section 2, we considered a LG model with n chiral superfields and arbitrary
superpotential. The same techniques give us sensible A-type boundary conditions for the
n-chiral superfields in the e2 NLSM. In this limit, the following boundary conditions
lead to the vanishing of (2.15) and (2.16), in the limit of infinite gauge coupling:



= 0,
Fa (, )

(3.20)

+a + a = 0,


Fa
Fa
b c

D1 i
D1 i iKabc = 0,
i
i


W () W
S ()

= 0.
Fa (, ),
PB

(3.21)
(3.22)
(3.23)

We may use the equations of motion for the vector multiplet to infer sensible boundary
conditions on the component fields, via Eqs. (3.3)(3.7). For and ,
|x 1 =0 = |x 1 =0 = 0,
(+ + )|x 1 =0 = 0.

(3.24)
(3.25)

This implies the following boundary condition on the twisted chiral superfield (with + =
)
|x 1 =0 = 0.

(3.26)

Indeed, it happens that v0 = 0 in the NLSM limit. This is consistent with supersymmetry
since susyv0 = 0 using (3.25). Thus, one can choose the gauge condition
v0 = 0
in the LSM.

(3.27)

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

171

3.3. B-type boundary conditions in the NLSM

Recall that B-type boundary conditions are defined by requiring that + =  . In this
case the term complicates the story. We will begin with the case = 0.
3.3.1. = 0
As we stated above, we will work with fully Neumann boundary conditions in this
section. Let us begin with the boundary condition on (justified a posteriori):
(+ + )|x 1 =0 = 0.
The supersymmetric variations of this condition lead to:
 




Q 
= 0,
D1 i
2
x 1 =0

S
W

= 0.
F |x 1 =0 =
x 1 =0

(3.28)

(3.29)
(3.30)

The vanishing of boundary terms in the ordinary variations of the action requires further
that:
D1 |x 1 =0 = 0,

(3.31)

( )|x 1 =0 = 0.

(3.32)

It is easy to see that the conditions on are consistent with Eqs. (3.3), (3.4).
The remaining boundary conditions on the vector multiplet can be derived either from
the solutions in Eqs. (3.3)(3.7), or from SUSY variations of (3.32). The results are:
(+ )|x 1 =0 = 0,

(3.33)

1 ( + )|x 1 =0 = 0,

(3.34)

v01 |x 1 =0 = 0.

(3.35)

Following the notation in [20], these conditions can be written in superfield notation as:
S x 1 =0 = 0,
( )|
+

(3.36)

=
on the boundary. ( s are the Grassmann parameters in
where we recall that
d = 2, N = 2 superspace.) Note that holomorphic boundary conditions on the complex
scalars of the chiral superfields leads to reality conditions on the scalars in the twisted chiral
multiplet; this might have been anticipated from mirror symmetry, which exchanges A-type
and B-type boundary conditions, as well as (roughly) chiral and twisted chiral superfields.
3.3.2. 6= 0
In the presence of the theta term, D1 i |x1 =0 = 0 is clearly no longer the correct bosonic
boundary condition. 3 This is to be expected; even for constant NSNS B-fluxes through
D-branes in flat space, the boundary conditions for the string worldsheet are roughly:
1 X + B 0 X = 0.
3 Some of the material in this subsection was developed in collaboration with Albion Lawrence.

(3.37)

172

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

We find the same effect here in the e2 limit.

Ordinary variation of Eq. (3.8) then implies that:
Z
X


i
ord
dx 0
0 i i 0 i i ,
Sr, =
2r

(3.38)

where we have ignored the total derivative pieces since they have been cancelled by the
addition of the contact term. Adding (3.38) to the boundary term coming from the ordinary
variation of the bosonic kinetic energy term in the bulk action, we get the new boundary
term,


Z
X 
i
ord
0 i i + c.c. .
= dy
(3.39)
D1 i +
Skin
2r
i

However before we claim that the term in brackets multiplying i provides the boundary
conditions for the s, we note that the boundary conditions may be written in different
P
ways upto the addition of terms that vanish when we use the fact that 0 ( i Qi i i ) =
P
P
1 ( i Qi i i ) = 0, since i Qi i i = r in the bulk. For example, Eq. (3.39) can be
rewritten as


Z
X 
i B
ord
D0 i i + c.c. ,
= dy
(3.40)
D1 i +
Skin
2r
i

where D0B = (0 + iv0B ). Care should be taken when we write the boundary conditions
in the full GLSM where D 6= 0. The way to do this is to choose boundary conditions that
are closed under supersymmetry. Let us begin with the fermionic boundary conditions that
are appropriate in the presence of a constant B-field. In such situations fermions obey the
rotated boundary condition


(3.41)
+i + ei i x 1 =0 = 0,
where ei is to be determined later by requiring consistency with the boundary conditions
on the bosons. Supersymmetric variation of (3.41) implies
D1 i +

1 ei
ei
D0 i + i 2Qi
i = 0,
i
1+e
1 + ei
Fi = 0,

(3.42)
(3.43)

where we note that the covariant derivatives involve both the bosonic and fermionic
components of v0 and v1 . The vanishing of the bosonic boundary variation term Eq. (3.39)
clearly requires
i
1 ei
.
(3.44)
=
1 + ei
2r
Now by judicious use of the fact that the variation of D is zero and the explicit expressions
for the gauge fields and the field in the NLSM limit the boundary condition (3.42) indeed
reduces to



i B
B
D i 
= 0.
(3.45)
D1 i +
2r 0
1
x =0

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

173

This boundary condition is indeed the one suggested by the boundary variation terms
in Eq. (3.40) (with the above fermion boundary conditions v1F = 0). In the GLSM it is
Eq. (3.42) which will be the natural boundary condition on the bosonic boundary fields.
Eq. (3.45) may be obtained by showing that


ei 
1 ei F

iv + i 2
= 0,
(3.46)
1 + ei 0
1 + ei x 1 =0
where v0F is the fermion bilinear part of v0 . With these boundary conditions, in the e2
limit, all of the boundary terms arising in the equations of motion and the supersymmetric
variations of the action cancel out.
Finally we wish to use the equations of motion for the vector multiplet to arrive at
sensible boundary conditions for the component fields. Eqs. (3.3) and (3.4) imply the
boundary condition:


(3.47)
e2i x 1 =0 = 0.
This is consistent with Eq. (3.46) and the boundary conditions for the chiral multiplets.
The SUSY variation of (3.47) requires


(3.48)
+ e2i x 1 =0 = 0,
which in turn requires







1 e2i iD 1 + e2i v01 + 2 1 + e2i x 1 =0 = 0.

(3.49)

Our analysis has been for the case of (all Neumann boundary conditions) in one-modulus
examples in weighted projective spaces. However, the generalisation to many moduli case
is straightforward and we will not enter into the details here. We now turn to the case of
mixed boundary conditions, i.e., we impose Dirichlet boundary conditions on some of
the fields.
3.3.3. Mixed boundary conditions
We illustrate the general situation by imposing Dirichlet boundary conditions on one of
the fields, say
1 = 0.
Supersymmetric completion requires the condition
(+1 1 ) = 0.
All other fields have Neumann boundary conditions imposed on them as in Section 3.3.2.
One can check that all boundary terms in the ordinary and supersymmetric variations of
the action vanish as they did in the all Neumann case.
What about the boundary conditions implied on the fields in the vector multiplet? By
considering the expressions for and as given in Eqs. (3.3) and (3.4), respectively,
it appears that one cannot obtain simple boundary conditions as in (3.47) for the all
Neumann case. However, we claim that the problem may be resolved by requiring that on
the boundary the bulk expressions for the fields in the vector multiplet have to be modified.

174

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

The modification requires that the fields in the vector multiplet depend only on fields in the
chiral multiplet which have Neumann boundary conditions and not on those with Dirichlet
boundary conditions (1 , 1 in our case). Note that this is trivially true for the bosonic
part and refers only to the fermionic part such as the bilinear expression for in the NLSM
limit (see Eq. (3.3)). Another way to state this modification is to impose a modified Gauss
law constraint on the boundary such that
X 



Qi i i D0 i i D0 i + i i + +i +i
J0
i

= Q1 1 1 + +1 +1 .

(3.50)

For the general case, the RHS of the Gauss law constraint is given by a summation over
fermions in the Dirichlet directions. Consistency with supersymmetry forces as well as
to also depend only on fields with Neumann boundary conditions.
Thus, the boundary conditions on the fields in the vector multiplet are identical to the
all Neumann case considered in Section 3.3.2 above. This will enable us to now carry over
these boundary conditions to the GLSM in a straightforward fashion.

4. Boundary conditions in the GLSM

In this section, we propose to derive supersymmetry preserving boundary conditions
for the gauged linear sigma model, that will define appropriate D-branes wrapping
supersymmetric cycles of the CalabiYau. These boundary conditions that we derive for
the GLSM, would have to be consistent with those in the infra-red limit, i.e., should
appropriately reduce to the ones we have described in the last section.
Let us begin by recalling that whereas in the non-linear sigma model limit of the GLSM
the fields in the gauge multiplet become very massive and effectively decouple from the
theory, this is not the case with the GLSM. As a consequence the analogue of the Dirichlet
and Neumann boundary conditions in the GLSM are more general boundary conditions that
depend on fields of the gauge multiplet as well. In particular, it is clear that the boundary
conditions must be gauge covariant.
4.1. A-type boundary conditions in the GLSM
In order to extend the boundary conditions from the NLSM to the GLSM, we will have
to include boundary conditions for the p-field in addition to the ones we obtained for
the fields in the vector multiplet. One can check that the following are a consistent set of
boundary conditions in the GLSM at finite gauge coupling
Im p|x 1 =0 = 0,
(+p + p )|x 1 =0 = 0,

(4.1)

Re D1 p|x 1 =0 = 0.

(4.3)

(4.2)

All other boundary conditions are as given in the NLSM discussed earlier. It is interesting
that the A-type boundary conditions are identical in form in both the GLSM in general

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

175

(taking into account of course the need to include boundary conditions on the p-field) and
in the LG and CY phases. This is however not the case with the B-type as is clear even
from the considerations of such boundary conditions in the NLSM limit.
4.2. B-type boundary conditions in the GLSM
4.2.1. = 0
In the NLSM limit, we have seen that for both Dirichlet as well as Neumann boundary
conditions imposed on the matter fields, the boundary conditions implied on the fields in
the vector multiplet are summarised by the simple boundary condition on the twisted chiral
superfield
S x 1 =0 = 0,
( )|

(4.4)

where we impose + = and + = as well in the above condition. We will

continue to require this set of boundary conditions in the GLSM. However we will also
need to impose boundary conditions on the p-field as well as its supersymmetric partners
p such that boundary terms in the ordinary as well as supersymmetric variation of the
GLSM action vanish. It is useful to observe at this point that once we have fixed the above
boundary conditions on the fields in the vector multiplet, the choices of consistent boundary
conditions is in fact identical to that of an extended LG model involving the p-field and the
fields i with superpotential W = PG() as in Section 2 (with the condition that ordinary
derivatives in the LG model are replaced by covariant derivatives in the GLSM). This leads
to two possible classes of boundary conditions
1. Dirichlet boundary condition on p with p = 0. Since the superpotential in the
GLSM is given by W = P G(), for this choice, Fi = (W/i ) = p(G/i ) = 0.
Thus, the condition Fi = 0 which occurs whenever we impose Neumann boundary
conditions on the i is trivially satisfied. Thus, any boundary condition (Neumann or
Dirichlet) involving scalar fields other than p goes through subject to the condition
that all Dirichlet boundary conditions are specified by homogeneous polynomials.
This includes the all Neumann case which did not appear in the LG phase. In the
LG phase, (which occurs for large negative r) where p has non-vanishing vev, the
boundary condition p = 0 cannot be imposed. This also suggests that this boundary
condition is acceptable only in the CY phase where p = 0 is the ground state
condition.
2. Another possibility is that one imposes Neumann boundary condition on the p-field.
However, the Fp = 0 condition now requires G = 0. Further, one is not allowed to
choose to impose Neumann boundary conditions on individual fields for the quintic at
the Fermat point. A possible consistent choice of boundary conditions at the Fermat
point of the quintic is given by 1 + 2 = 0 and i = 0 (i = 3, 4, 5).
One can verify that for both classes of boundary conditions discussed above, the boundary
terms in the ordinary as well as supersymmetric variations vanish.

176

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

4.2.2. 6= 0
The strategy that we will pursue is to extend the boundary conditions we obtained in the
NLSM to that of the GLSM. The ordinary and supersymmetric variations of the kinetic
energy terms of the fields in the vector multiplet will now have to be considered. We will
first consider the contact term which we derived in the NLSM limit as given by Eq. (3.19).
Z
X
i
NLSM
=
(i D0 i i D0 i ).
(4.5)
dx 0
Sboundary
4r
i

However, before we choose this to be the contact term in the GLSM we must remember
that there may be a need for other terms which vanish in the NLSM limit. In fact, we do
need such a term in order to ensure that there is a smooth NLSM limit to the boundary
conditions chosen in the GLSM. The full boundary term that we need in the GLSM turns
out be


Z


D i
i X
GLSM
= dx 0

e
i D0 i i D0 i +
,
(4.6)
Sboundary
4r
2r e2
i

where tan( /2) = /2r.

In the presence of this boundary term, we now list the boundary conditions required
(using the NLSM as a guide) for cancelling ordinary as well as supersymmetric variations
of the action. In the matter sector, these are


(4.7)
+i + ei i x 1 =0 = 0,


i
i


1e
e
D0 i + i 2Qi
i 
= 0,
(4.8)
D1 i +
1 + ei
1 + ei
x 1 =0
(4.9)
p|x 1 =0 = 0,
(+p p )|x 1 =0 = 0,

(4.10)

where we have included the p-field as well as its fermionic partners p .

The boundary conditions on the fields in the vector multiplet in the GLSM are chosen to
be


(4.11)
e2i x 1 =0 = 0,


2i

(4.12)
+ e 1 = 0,
x =0

where, in keeping with the results obtained in the NLSM limit, we have chosen the phase to
be twice the phase in the matter boundary conditions. The remaining boundary conditions
are



v01
+
= 0,
(4.13)
D
2r x 1 =0




ei D 
= 0.
(4.14)
1 + e2i
2 r
x 1 =0
Notice that the last two boundary conditions are indeed a convenient split of a single
equation (given below) arising from the supervariation of Eq. (4.12) in order to make
boundary terms in the ordinary variation vanish.

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182






1 e2i iD 1 + e2i v01 + 2 1 + e2i x 1 =0 = 0.

177

(4.15)

Hence, they are really dictated by our insistence that the rotated boundary conditions on
and have a consistent NLSM limit. The full set of boundary conditions on the fields in the
vector implies the following boundary condition on the twisted chiral superfield (subject
to + = and + = )


S  1 = 0.
(4.16)
e2i
x =0
There exists another solution to the vanishing of the boundary variation terms that
however will however involve rotated boundary conditions on the fields that do not
agree with the NLSM limit. As always, we begin with the fermions and choose (+i +
ei i ) = 0 and then derive other conditions by supersymmetric variation of the
condition. The first variation leads to Eq. (4.8) which one can rewrite as two separate
conditions



1 ei
D0 i 
= 0,
(4.17)
D1 i +
1 + ei
x 1 =0

ei 
= 0,
(4.18)
1 + ei  1
x =0

Fi |x 1 =0 = 0.

(4.19)

The second equation is clearly in contradiction with the rotations implied on in the
NLSM. Further supersymmetric variation implies the following boundary condition on the
twisted chiral superfield (subject to + = and + = )
S x 1 =0 = 0.
( ei )|

(4.20)

It is easy to verify that the boundary terms in the ordinary and supersymmetric variations
vanish. This solution has also been recently proposed in [15]. As emphasized earlier, this
solution does not agree with the boundary conditions on the fields in the vector multiplet
as derived in the NLSM limit.
The case of mixed boundary conditions (i.e., some fields have Dirichlet boundary
conditions imposed on them) also goes through. The boundary contact term is again
given by Eq. (4.6). The boundary condition on the fields in the vector multiplet are as
in the all Neumann case and one can verify that all boundary terms in the ordinary and
supersymmetric variations vanish.
4.3. Discussion
We would now like to discuss some of the implications of the analysis of boundary
conditions in the GLSM for the case of the quintic at the Fermat point. In particular we
would like to address the question of whether we can identify the branes corresponding to
our various choices of boundary conditions.
The choice of all Neumann boundary conditions on the matter fields is clearly suggestive
of a D6-brane. However we have to decide what boundary conditions are appropriate for
the p field. It is useful to consider for this purpose a slightly different example, without

178

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

a superpotential, namely that of the O(3) line bundle over P2 . In this case, clearly the
construction of a D4-brane wrapping the P2 would require that there be Dirichlet boundary
conditions on the charge 3 field. Otherwise we would not obtain a compact D4-brane.
A similar situation obtains in the quintic and hence we choose p = 0 for the D6-brane in
the large volume. 4 The G = 0 condition is imposed on the ground state by continuity from
the bulk.
Similarly we could identify the cases of the mixed NeumannDirichlet boundary
conditions with lower-dimensional branes while we always maintain Dirichlet boundary
conditions on the p field. However the charges of these lower-dimensional branes will not
be the minimum charge.
However there could be other situations where we impose Neumann boundary
conditions that involve the p field. For instance we could impose conditions of the form
p15 = c where c is a complex constant. Associated to this condition would be another
Neumann condition involving the p field and the 1 . We can then choose individual
boundary conditions on the rest of the fields. The interpretation of these boundary
conditions would be different.
Let us now consider the quintic at a point in its Khler moduli space where it admits
a description as a Gepner model. B-type boundary states arising from the Recknagel
Schomerus construction [6] have been discussed in [7]. The important point to note with
regard to the RecknagelSchomerus construction is that the boundary states arise from
tensoring boundary states of individual minimal models. Thus, they can only arise in our
construction by imposing boundary conditions on individual fields. We will for the moment
focus our attention on the boundary states labelled |00000iB . (There are five such states
forming a Z5 orbit.) The analysis in [7] shows that one of these boundary states corresponds
to the pure six-brane in the large volume limit.
The Gepner point is in the LG phase (see Section 2) where we have seen that the only
allowed boundary conditions on individual fields are Dirichlet and hence all RS states
(including the one which carries pure six-brane charge) must arise in this class. (See [17]
for a related discussion.) Consider now the all Neumann case that we considered in the
GLSM with p = 0. The boundary condition on the bosonic fields as we have described
earlier are (for i = 1, . . . , 5)


ei

i
D0 i + i 2
i 
= 0.
D1 i +
2r
1 + ei
x 1 =0
It is of interest to ask what happens to these boundary conditions as r 0 while keeping
fixed at some non-zero value. In this limit ei 1 and the above boundary conditions
tend to a Dirichlet boundary condition.

( + ) 
i 
= 0.
D0 i + i
2
x 1 =0
4 An alternate possibility is to introduce a boundary fermion which enforces P G() = 0 only in the NLSM
limit. We will examine this possibility in the general context of boundary fermions and vector bundles on D-branes
in a future publication [35].

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

179

Interestingly, the part in the above equation comes out precisely of the form required
because the fermionic boundary condition becomes precisely the Dirichlet combination.
However quantum corrections (due to worldsheet instantons) become important for small
values of r and further lines of marginal stability may be crossed in taking this limit.
Though our classical analysis may thus need to be modified, the above result is suggestively
in agreement with the pure six-brane appearing in the LG phase as an all Dirichlet state.
The limitations of our argument become clear once we consider the cases of lower branes
(such as the pure D4-brane) which do not appear as a RS boundary state at the Gepner
point.
We now comment on the extension of the results of this paper to other examples. There
are a number of straightforward extensions that require very little beyond the techniques of
this paper itself. All examples which are given by complete intersections of hypersurfaces
in weighted projective spaces or a tensor product of such spaces can be dealt with. In such
cases the analogues of the p-field are such that they can be decoupled in the NLSM limit.
Thus the rest of the analysis would proceed much as in the single modulus case.

5. Conclusion
In this paper we have taken the initial steps in what appears to be an useful programme
of trying to understand, in a microscopic description, D-branes in large domains of the
moduli space of the CalabiYau backgrounds in which they live. The explicit nature of the
boundary conditions that we impose on the matter fields may be appropriately translated
into the more general characterisations of branes in the large-volume phase as zero sections
of the corresponding bundles on the CY manifold. Thus contact could be made with other,
more geometric techniques for understanding the various properties of these branes in the
large volume limit.
The analysis of the GLSM in this paper has been restricted to the open-string channel.
More information can be extracted by also investigating the closed-string channel. Some
related issues have already been considered in [14,15].
Another question which needs to be considered is the addition of (marginal) deformations corresponding to gauge fields on the brane as well as its moduli including the introduction of ChanPaton factors. This will involve describing vector bundles on CalabiYau
manifolds or their submanifolds. One construction which easily fits the boundary GLSM
is the use of monads (as suggested in [7]). It is clear that this will involve techniques which
appeared in the context of (0, 2) versions of the closed string GLSM [20,34] given that
only half of the (2, 2) supersymmetry is preserved on the boundary. For example, one has
to introduce boundary fermions which are sections of the appropriate vector bundle [35].
The fermions m which appeared in Section 2.3.2 are objects of this type (they are sections
of the normal bundle).
A related problem is the classification of boundary topological observables in the twisted
version of the GLSM. This naturally leads to the next step in this program, i.e., the use of
the GLSM description of D-branes to determine the superpotential in the worldvolume

180

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

theory of these branes. We believe that the results of this paper are a useful step in
proceeding towards this goal. This (superpotential) can be used as a check of mirror
symmetry by computing the same in the mirror manifold. Some aspects of these issues
from a slightly different viewpoint have been considered in [13,16,33].

Acknowledgements
We would like to thank Albion Lawrence and Hirosi Ooguri for collaboration during the
initial stages of this work. We are also indebted to them for numerous useful discussions
and detailed correspondence. We would also like to thank A.L. for a critical reading of an
earlier version of this paper and related correspondence. T.S. would like to thank S. Das
and A. Sen for useful discussions. T.S. also acknowledges a Junior Research Fellowship of
the Institute of Mathematical Sciences, Chennai where a part of this work was done. S.G.
is supported in part by the Department of Science and Technology, India under the grant
SP/S2/E-03/96.

References
[1] J. Schwarz, Lectures on superstring and m theory dualities, Nucl. Phys. Proc. Suppl. B 55 (1997)
1, hep-th/9607201 and references therein.
[2] C. Hull, String dynamics at strong coupling, Nucl. Phys. B 468 (1996) 113, hep-th/9512181.
[3] S.H. Shenker, Another length scale in string theory?, hep-th/9509132;
C. Bachas, D-brane dynamics, Phys. Lett. B 374 (1996) 73, hep-th/9511043;
U.H. Danielsson, G. Ferretti, B. Sundorg, D-particle dynamics and bound states, Int. J. Mod.
Phys. A 11 (1996) 5463, hep-th/9603081;
D. Kabat, P. Pouliot, A comment on zero-brane quantum mechanics, Phys. Rev. Lett. 77 (1996)
1004, hep-th/9603127;
M.R. Douglas, D. Kabat, P. Pouliot, S. Shenker, D-branes and short distances in string theory,
Nucl. Phys. B 485 (1997) 85, hep-th/9608024.
[4] M. Douglas, B. Greene, D. Morrison, Orbifold resolution by D-branes, Nucl. Phys. B 506
(1997) 84, hep-th/9704151.
[5] H. Ooguri, Y. Oz, Z. Yin, D-branes on CalabiYau spaces and their mirrors, Nucl. Phys. B 477
(1996) 407, hep-th/9606112.
[6] A. Recknagel, V. Schomerus, D-branes in Gepner models, Nucl. Phys. B 531 (1998) 185, hepth/9712186.
[7] I. Brunner, M.R. Douglas, A. Lawrence, C. Rmelsberger, D-branes on the quintic, hepth/9906200.
[8] M.R. Douglas, Topics in D geometry, Class. Quant. Grav. 17 (2000) 1057, hep-th/9910170.
[9] S. Govindarajan, T. Jayaraman, T. Sarkar, Worldsheet approaches to D-branes on supersymmetric cycles, Nucl. Phys. B 580 (2000) 519547, hep-th/9907131.
[10] S. Govindarajan, T. Jayaraman, On the LandauGinzburg description of boundary CFTs and
special Lagrangian submanifolds, J. High Energy Phys. 07 (2000) 016, hep-th/0003242.
[11] M. Gutperle, Y. Satoh, D-branes in Gepner models and supersymmetry, Nucl. Phys. B 543
(1999) 73, hep-th/9808080;
D.-E. Diaconescu, C. Rmelsberger, D-branes and bundles on elliptic fibration, hepth/9910172;

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

[12]

[13]

[14]
[15]
[16]
[17]
[18]

[19]
[20]
[21]
[22]

[23]
[24]
[25]
[26]

181

P. Kaste, W. Lerche, C.A. Lutken, J. Walcher, D-branes on K3-fibrations, hep-th/9912147;

E. Scheidegger, D-branes on some one- and two-parameter CalabiYau hypersurfaces, hepth/9912188;
M. Naka, M. Nozaki, Boundary states in Gepner models, hep-th/0001037;
I. Brunner, V. Schomerus, D-branes at singular curves of CalabiYau compactifications, hepth/0001132;
J. Fuchs, C. Schweigert, J. Walcher, Projections in string theory and boundary states for Gepner
models, hep-th/0003298.
M.R. Douglas, B. Fiol, C. Rmelsberger, Stability and BPS branes, hep-th/0002037;
M.R. Douglas, B. Fiol, C. Rmelsberger, The spectrum of BPS branes on a noncompact Calabi
Yau, hep-th/0003263.
S. Kachru, J. McGreevy, Supersymmetric three-cycles and (super)symmetry breaking, hepth/9908135;
S. Kachru, S. Katz, A. Lawrence, J. McGreevy, Open string instantons and superpotentials,
hep-th/9912151.
K. Hori, C. Vafa, Mirror symmetry, hep-th/0002222.
K. Hori, A. Iqbal, C. Vafa, D-branes and mirror symmetry, hep-th/0005247.
S. Kachru, S. Katz, A. Lawrence, J. McGreevy, Mirror symmetry for open strings, hepth/0006047.
M.R. Douglas, D.-E. Diaconescu, D-branes on stringy CalabiYau manifolds, hep-th/0006224.
J. Fuchs, C. Schweigert, Branes: from free fields to general backgrounds, Nucl. Phys. B 530
(1998) 99, hep-th/9712257;
J. Fuchs, C. Schweigert, Symmetry breaking boundaries I. General theory, hep-th/9902132;
J. Fuchs, C. Schweigert, Symmetry breaking boundaries II. More structures; examples, hepth/9908025;
D.-E. Diaconescu, J. Gomis, Fractional branes and boundary states in orbifold theories, hepth/9906242;
T. Takayanagi, String creation and monodromy from fractional D-branes on ALE spaces, hepth/9912157;
F. Roose, Boundary states and non-abelian orbifold, hep-th/000212;
F. Denef, Supergravity flows and D-brane stability, hep-th/0005049;
B. Fiol, M. Marino, BPS states and algebras from quivers, hep-th/0006189;
W. Lerche, A. Lutken, C. Schweigert, D-branes on ALE spaces and the ADE classification of
conformal field theories, hep-th/0006247.
E. Witten, Mirror manifolds and topological field theory, hep-th/9112056.
E. Witten, Phases of N = 2 theories in two dimensions, Nucl. Phys. B 403 (1993) 159, hepth/9301042.
T. Sarkar, Aspects of D-brane physics and quantum gravity, Ph.D. thesis, September 1999,
submitted to the University of Madras.
T. Sarkar, Talks at the Puri Workshop on Quantum Field theory, Quantum Gravity and Strings,
December 919, 1998 and at the Millennium Meeting on String Theory, January 37, 2000,
Bangalore, India.
P.S. Aspinwall, B.R. Greene, On the geometric interpretation of N = 2 superconformal field
theories, Nucl. Phys. B 437 (1995) 205, hep-th/9409110.
P. Aspinwall, B. Greene, D. Morrison, CalabiYau moduli space, mirror manifolds and
spacetime topology change in string theory, Nucl. Phys. B 416 (1994) 78, hep-th/9309097.
D. Gepner, Exactly solvable string compactifications on manifolds of SU(N) holonomy, Phys.
Lett. B 199 (1987) 380.
P.S. Aspinwall, The moduli space of N = 2 superconformal field theories, Lectures at the 1994
Trieste summer school hep-th/9412115.

182

S. Govindarajan et al. / Nuclear Physics B 593 (2001) 155182

[27] D.R. Morrison, M.R. Plesser, Summing the instantons: quantum cohomology and mirror
symmetry in toric varieties, Nucl. Phys. 440 (1995) 279, hep-th/9412236.
[28] N. Warner, Supersymmetry in boundary integrable models, Nucl. Phys. B 450 (1995) 663, hepth/9506064.
[29] R. Harvey, H.B. Lawson Jr., Calibrated geometries, Acta Math. 148 (1982) 47157.
[30] P. Candelas, Yukawa couplings between (2, 1)-forms, Nucl. Phys. B 298 (1988) 455492.
[31] B. Zumino, Supersymmetric sigma models in 2 dimensions, in: D.I. Olive, P.C. West (Eds.),
Duality and Supersymmetric Theories, Cambridge Univ. Press, 1999, pp. 4961.
[32] M. Gutperle, Contact terms, symmetries and D instantons, Nucl. Phys. B 508 (1997) 133, hepth/9705023.
[33] C. Vafa, Extending mirror conjecture to CalabiYau with bundles, hep-th/9804131.
[34] J. Distler, S. Kachru, (0, 2) LandauGinzburg theory, Nucl. Phys. B 413 (1994) 213, hepth/9309110;
J. Distler, Notes on (0, 2) superconformal field theories, hep-th/9502012.
[35] S. Govindarajan, T. Jayaraman, T. Sarkar, in preparation.

Nuclear Physics B 593 (2001) 183228

www.elsevier.nl/locate/npe

Renormalisation and off-shell improvement

in lattice perturbation theory
S. Capitani a,b , M. Gckeler c , R. Horsley d , H. Perlt e , P.E.L. Rakow c, ,
G. Schierholz b,f , A. Schiller e
a MIT, Center for Theoretical Physics, Laboratory for Nuclear Science, 77 Massachusetts Avenue,

Cambridge, MA 02139, USA

b Deutsches Elektronen-Synchrotron DESY, D-22603 Hamburg, Germany
c Institut fr Theoretische Physik, Universitt Regensburg, D-93040 Regensburg, Germany
d Institut fr Physik, Humboldt-Universitt zu Berlin, D-10115 Berlin, Germany
e Institut fr Theoretische Physik, Universitt Leipzig, D-04109 Leipzig, Germany
f Deutsches Elektronen-Synchrotron DESY, John von Neumann-Institut fr Computing NIC,

D-15738 Zeuthen, Germany

Received 4 August 2000; accepted 25 September 2000

Abstract
We discuss the improvement of flavour non-singlet point and one-link lattice quark operators,
which describe the quark currents and the first moment of the DIS structure functions respectively.
Suitable bases of improved operators are given, and the corresponding renormalisation factors
and improvement coefficients are calculated in one-loop lattice perturbation theory, using the
SheikholeslamiWohlert (clover) action. To this order we achieve off-shell improvement by
eliminating the effect of contact terms. We use massive fermions, and our calculations are done
keeping all terms up to first order in the lattice spacing, for arbitrary m2 /p2 , in a general covariant
gauge. We also compare clover fermions with fermions satisfying the GinspargWilson relation, and
show how to remove O(a) effects off-shell in this case too, and how this is in many aspects simpler
than for clover fermions. Finally, tadpole improvement is also considered. 2001 Elsevier Science
B.V. All rights reserved.
PACS: 11.15.Ha; 12.38.Bx
Keywords: Lattice perturbation theory; Improved actions

1. Introduction
There has been considerable progress in obtaining realistic results from numerical
simulations in lattice QCD. A new generation of massively parallel computers promises
* Corresponding author.

E-mail address: [email protected] (P.E.L. Rakow).

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 9 0 - 3

184

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

results that can be compared to a wide class of experimental data. Nevertheless, the
finiteness of the lattice spacing always leads to systematic errors in the simulations.
Therefore there is great interest in improving lattice QCD calculations. A systematic
improvement scheme, removing discretisation errors order by order in the lattice spacing a,
has been suggested by Symanzik [1], and developed by Lscher and Weisz [2] for on-shell
quantities.
An O(a) improved fermionic action which is widely used in lattice Monte Carlo
simulations is that proposed by Sheikholeslami and Wohlert [3]:
X 1
X
MC

= a4
(x)(x)

(x + a )
U (x)[1 + ](x)
Simp
a
a
x



X
a )
1 (x)
(x
U (x a )

a

Xa
clover

(1)
(x) F (x)(x) ,
2cSWg
4

clover denotes the standard clover-leaf

where is known as the hopping parameter and F
form of the lattice field strength. 1 If the parameter cSW , which gives the strength of the
higher-dimensional operator, is correctly chosen this action has no O(a) errors for onshell quantities such as hadron masses. For perturbative calculations it is simpler to use a
slightly different normalisation:
X


1 X
pert

(x + a )
U (x) 1 + (x)
(m + mc ) (x)(x)
Simp = a 4
2a
x



1 X
a )
(x
U (x a )
1 (x)
2a

Xa
clover

(x)
F
(x) (x) .
cSW g
4

(2)

The parameters of the two forms of the action are related by

1
,
(3)
2c
1
1

,
(4)
am =
2 2c
where c is the critical value of the hopping parameter, at which chiral symmetry is
approximately restored.
Simply improving the action does not remove O(a) errors from operator matrix
elements. To do this the operators must also be improved by adding higher dimensional
irrelevant operators with appropriate improvement coefficients. The operators also need
to be renormalised. In this paper we will discuss the perturbative renormalisation and
improvement of bilinear quark operators.
amc =

1 We will use the convention

= (i/2)( ).

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

185

However, the action (2) with its single tunable improvement parameter cSW only
improves on-shell quantities. Off-shell quantities still have O(a) errors, which arise from
short-distance contact terms. We will show how the contact terms can be removed at
the one-loop level of lattice perturbation theory, and off-shell quantities free of O(a)
discretisation errors can be extracted from Greens functions.
There are several reasons why it would be desirable to understand the improvement
of off-shell quantities. In particular the non-perturbative renormalisation suggested in [4]
involves comparing lattice measurements of off-shell Greens functions with continuum
perturbation theory results [5] in order to relate lattice quantities to conventional
renormalisation schemes such as MS. This matching will work best at large virtualities,
where the running coupling constant is small, and the effects of non-perturbative
phenomena such as chiral symmetry breaking have died away. It is obviously desirable
to remove the discretisation errors in the off-shell lattice Greens functions before making
the comparison with the continuum. Even within perturbation theory it is easier to calculate
Greens functions at p2  m2 than in the region where p2 and m2 are comparable.
Our strategy is to look at the tree-level results for the Greens functions, and see what
O(a) effects are present, and what has to be done to remove them. We then look at the
one-loop perturbative results, and see whether the tree-level procedure still works. We find
that at one particular value of the clover coupling the O(a) effects are of the same form
as in tree-level, and that then we can remove O(a) effects completely, and find improved
Greens functions that are free of O(a) discretisation errors, both on-shell and off-shell.
Our aim is to find perturbative expressions at one-loop for the MS-scheme renormalisation factors and for the improvement coefficients. To do this we have to compute each
Feynman diagram including all O(a) terms. These results are applicable to both quenched
and dynamical calculations of flavour non-singlet matrix elements.
In this paper we consider the complete set of point operators

(x)
i (x),

(5)

i = 1, 5, , 5 , 5 .

(6)

with

For one-link operators we discuss the physically interesting case of the leading-twist
operators occurring in the operator product expansion for the moments of the hadronic
structure functions [6]. We consider the operators which measure the first moment of the
unpolarised and polarised structure functions:

1
1
(7)
D ,
5 D ,
2
2
where symmetrisation over and and removal of trace terms is always to be understood.
The perturbative renormalisation of improved point operators has been discussed
by several groups [79]. They use the tree-level values for the operator improvement
coefficients ci (defined below), and for the coefficient cSW in the SheikholeslamiWohlert
action. The same settings have been used to calculate the renormalisation factors for the
one-link [10] and two-link [11] quark operators in the chiral limit, performing on quark

186

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

operators the transformation discussed in [9]. In this paper we present the Z factors
with coefficients ci and cSW kept arbitrary. This allows us to determine the perturbative
contributions of the various terms and their relative magnitudes. Moreover, this will enable
us to implement tadpole improved perturbation theory.
This paper is organised as follows. In Section 2 we give the operator bases for the
improvement of the point and one-link operators. In Section 3 we present a method with
which to improve the lattice quark propagator off-shell by taking care of contact terms, and
in Section 4 we extend this procedure to improve off-shell quark bilinear operators as well.
In Section 5 we compare with fermions satisfying the GinspargWilson relation, and show
how to remove O(a) effects off-shell in this case too. Finally, in Section 6 we apply tadpole
improvement to our perturbative results, and in Section 7 we present our conclusions. The
(sometimes cumbersome) complete results for renormalisation factors and improvement
coefficients are collected in Appendix A.

2. Bases for improved operators

In this section we write down a general operator basis for the improvement of quark
operators. The base operators must have the same symmetry properties as the unimproved
ones, i.e., their transformations under the hypercubic group and charge conjugation are
determined by the original operator.
First we consider the five point operators of Eq. (5). Subject to the symmetry constraints
we find the following improved operators:

imp
1
imp = (1 + amc0)
ac1 D
=
/ ,
(8)
OS
2





imp
1
5 imp = (1 + amc0 )
5 + ac2
5 ,
=
(9)
OP
2



imp

1 ac1 D
imp = (1 + amc0)
=
OV
2


1
,
(10)
+ aic2
2



imp
5 imp = (1 + amc0 )
5
=
OA

1
1
5 ,
5 D + ac2
(11)
aic1
2
2




H imp
5 imp = (1 + amc0 )
5
=
O

1
1
,
(12)
+ aic1 D D 5 + aic2 
2
2

where m is the bare fermion mass and D D D is the symmetric covariant derivative.
We have used the lattice definitions


1
U (x)(x + a )
U (x a )(x

a )
,
D (x) =
2a


1
a )U
+ a )U

(x a )
.
(13)
(x
(x) (x
(x)
D =
2a

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

187

Table 1
The improvement bases for the scalar (S), pseudoscalar (P), vector (V), axial (A) and tensor (H)
operators
Improvement basis

1 ac1 D
(1 + amc0 )
/
2
5
(1 + amc0 )

S
P

1 ac1 D
(1 + amc0 )
2

5 D
5 1 aic1
(1 + amc0 )
2


5 + 1 aic1 D D 5
(1 + amc0 )
2

V
A
H

The c2 terms in the above equations are irrelevant for forward matrix elements, which are
all that we consider. Therefore we are left with the expressions in Table 1.
We include the terms proportional to c0 so that we can get m-independent renormalisation constants and at the same time maintain O(a) improvement.
Using the scalar operator as an example, there is an equation of motion that says that for

= 0, where is a coefficient that depends on g. So

/ + m
on-shell measurements D
we can compensate for changes in c1 by making changes in c0 , which would allow us to
eliminate one of the improvement terms if we were only interested in on-shell quantities.
This equation of motion means that c0 is linear in c1 if we parameterise our operators as in


1 ac0 D
Eq. (8). If we use other parameterisations, for example (1 + abm)
2 1 / [12],
we would no longer find that b was linear in c10 .
We also consider the conserved vector current:
1
1
+ (x
+ a )(
+ 1)U (x)(x)
(J )imp = (x)(
1)U (x)(x + a )
2
2


1
.
(14)
+ aic2
2
We know that this should need no improvement for forward matrix elements, because the
only improvement term, c2 , is the coefficient of a total derivative, and so has no effect on
forward matrix elements. Thus J provides a useful check of our improvement method.
Next we consider the one-link operators of Eq. (7). Here we choose as a basis for the
improved unpolarised operator
X

imp
 

1
1
D + aic1
D , D

= (1 + amc0 )
O
2
8

X


1
1
D .

(15)
ac2 D , D + aic3
8
4

This operator basis is the same for the two possible irreducible representations of the lattice
(6)
hypercubic group to which the original operator may belong: 3 (non-diagonal, 6= )
and 1(3) (diagonal, = ). (Our notation for the irreducible representations of the lattice

188

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

hypercubic group follows [13].) In the case of the polarised structure function we find
that the improvement terms allowed by the hypercubic symmetry are different for the
representations 4(6) (non-diagonal, 6= ) and 4(3) (diagonal, = ). When 6= the
improved operator has the form
5
O

imp

1
1
5 D iac1
5 D 2
= (1 + amc0 )
2
4
X


1
1
5 D ,
5 D , D + ac3

aic2
8
4

(16)

6=,

whereas in the traceless diagonal case one has only one improvement term in the forward
case, so there is no c2 , and thus
X


1
1
5 imp
5 D aic1
5 D , D

= (1 + amc0 )
O
2
8

1
5 D .
(17)
+ ac3
4
Here repeated and indices are not summed over. We will always construct a traceless

operator when the indices are equal by using the combination 12 12 ( 5 D 5 D

), and a similar one in the unpolarised case.
The coefficients c0 , . . . , c2 can be appropriately determined using the method explained
in the following sections so that the desired improvement is achieved.

3. Improving the quark lattice propagator

3.1. Method
Even when the fermion action has been improved for on-shell quantities there are still
O(a) effects present in off-shell quantities such as the fermion propagator at a general
Euclidean momentum p. In this section we will discuss how to find an improved fermion
propagator off-shell. First we will look at the tree-level Wilson propagator and show how to
remove its off-shell discretisation errors, then we will generalise this improvement method
to the interacting case.
We are used to writing down expressions for S 1 , the inverse quark propagator. In
S 1 the main O(a) effect is the addition of the momentum-dependent Wilson mass term.
However, it is also instructive to look at the quark propagator S itself, rather than the inverse
propagator.
Let us start by looking at the propagator at tree-level. From the expression for the inverse
propagator


iX
1X
sin(ap ) + m +
1 cos(ap )
(S tree )1 (p, m) =
a
a
1
= i/
p + m + ap2 + O(a 2)
2

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

189





1
1
p + m) + O(a 2), (18)
i/
p + m am2 (1 + am) 1 a(i/
2
2

we derive that
1 am

a
+ O(a 2)
2
i/
p+m
a
(1 am)S?tree (p, m? ) + + O(a 2 ).
2

S tree (p, m) =

1
2
2 am

(19)

The tree-level lattice propagator consists then of two parts, one part is proportional to
a normal continuum propagator with a mass m? m 12 am2 , and the other part is a
momentum independent term. The nature of these two parts becomes even more clear
when we write them in position space: 2
1
S tree (x, y, m) = (1 am)S?tree (x, y, m? ) + a(x y) + O(a 2).
2

(20)

We see that (except at short distances, where an additional contact term appears) the lattice
Wilson-fermion propagator is proportional to S? , which has the form of a continuum
propagator with an improved mass m? , and which has no O(a) discretisation errors.
We will always use ? to mark bare quantities which have been O(a) improved.
This concentration of the O(a) effects at short distance is what we should expect, in fact
the fermion propagator at |x y|  a is an on-shell quantity, so it should be automatically
improved when the action is improved. It is only at short distances of order a that the
lattice propagator has a different form from the continuum propagator. The necessity of
subtracting a function from the lattice propagator to obtain an improved propagator has
been discussed in [4].
What should we expect beyond tree level? Let us write the inverse fermion propagator
as a series in the lattice spacing a:
S 1 (p, m) = 0 (p, m) + a1 (p, m) + O(a 2),

(21)

where the coefficients 0 and 1 are power series in g 2 . On-shell improvement tells us that
the lattice fermion propagator should be proportional to a continuum fermion propagator
except at short distances, so we expect equations of the same form as Eqs. (19) and (20)
to hold, though only at the value of cSW corresponding to on-shell improvement. Thus the
propagator should satisfy
1 + ab m a
+ + O(a 2 )
0 (p, m? )
2
a
(1 + ab m)S? (p, m? ) + + O(a 2 ),
2

S(p, m) =

(22)

where the improved bare mass m? is related to m through

m? = m(1 + abm m).
2 On the lattice we define (x y)
4
x,y /a , where x,y is the Kronecker delta function.

(23)

190

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

The improvement coefficients b , bm and are independent of p and m, and should only
depend on the coupling constant g 2 . By comparing with Eq. (19) we see that the tree-level
values are
b = 1,

bm = 1/2,

= 1.

(24)

The propagator S? is free of O(a) effects so we call it the improved fermion propagator.
Later, when we come to define renormalisation constants, we will always define them in
terms of the improved bare propagator S? and improved bare mass m? .
Taking the inverse of Eq. (22) gives


1
S 1 (p, m) = 0 (p, m? )(1 ab m) 1 a 0 (p, m? ) + O(a 2 )
2



0 (p, m)
= 0 (p, m) + abm m2
m


1
(25)
(1 ab m) 1 a 0 (p, m) + O(a 2 ),
2
so that dropping terms of order a 2 we get


2
1
S 1 (p, m) = 0 (p, m) + a 0 (p, m) b m0 (p, m)
2


0 (p, m) + O(a 2 ).
+ b m m2
m

(26)

Comparing with Eq. (21) we see that the improvement prescription (22) can only work if
the non-linear relation

2
1

0 (p, m),
(27)
1 (p, m) = 0 (p, m) b m0 (p, m) + bm m2
2
m
is satisfied. In Subsection 3.2 we shall see that this is indeed the case in one-loop
perturbation theory.
The pole mass of the fermion is the p value where 0 (p, m? ) vanishes, so for an on-shell
fermion the factor (1 12 a 0 (p, m? )) simply reduces to 1. Therefore we can see from
Eq. (25) that the improvement coefficient only has an effect when the fermion is offshell, so we only need to know if we are interested in extracting numbers from off-shell
lattice measurements.
From Eq. (22) we can find an explicit expression for S? :


1
a
1
,
(28)
S(p, m) =
S? (p, m? ) =
1 + amb
2
0 (p, m? )
and



a
S?1 (p, m? ) = (1 + amb ) S 1 (p, m) + S 1 (p, m) S 1 (p, m)
2
= 0 (p, m? ).

(29)

The reason we are interested in S? (p, m? ) is that it is a quantity free of O(a) effects which
can be constructed from the quark propagator S(p, m), and the latter is something we can

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

191

measure from non-perturbative simulations on the lattice. We will find that Eq. (27) is
satisfied at one particular value of the clover coefficient cSW . At this cSW value one can
use Eq. (28) or equivalently Eq. (29) to extract 0 from lattice measurements. The clover
action does not have enough tunable parameters to make the off-shell fermion propagator
free of O(a) effects, but this does not really matter, because equations such as Eq. (29)
nevertheless allow us to recover the improved off-shell propagator from quantities which
we can measure.
Up till now, we have only discussed the improvement of the fermion propagator. The
propagator and mass still have to be renormalised. The renormalised improved quark mass
and propagator are given by
mR (2 ) = Zm (2 )m? = Zm (2 )m(1 + ambm ),


1
S? (p, m? )
a
2
=
S(p, m) .
SR (p, mR ; ) =
Z2 (2 )
Z2 (2 )(1 + amb )
2

(30)
(31)

3.2. One-loop results for the quark propagator

We now want to see if the propagator improvement scheme suggested in Eq. (22)
holds in one-loop perturbation theory, and to calculate the improvement coefficients and
renormalisation factors to O(g 2 ).
Our calculations are carried out in a general covariant gauge, where the gluon propagator
is
k k

(1 )
,
(32)
G(k) =
k 2
(k 2 )2
with k 2 sin(ak /2). The Feynman gauge corresponds to = 1, the Landau gauge to
a

= 0.
We can write the inverse propagator in the form
a
p + p2
S 1 (p, m) = m + i/
2


g 2 CF

(33)
0 (p, m) + a1 (p, m) + O(a 2 , g 4 ),
2
16
where CF = (Nc2 1)/(2Nc ) for gauge group SU(Nc ). Comparing Eq. (33) with Eq. (21)
we see that
g 2 CF
0 (p, m) + O(g 4 ),
16 2
1
g 2 CF
1 (p, m) + O(g 4 ).
1 (p, m) = p2
2
16 2
We also expand the improvement coefficients to first order in g 2 :


g 2 CF
g 2 CF
4
4
d + O(g ),
b = 1 +
db + O(g ) ,
= 1 +
16 2
16 2


1
g 2 CF
db + O(g 4 ) .
bm = 1 +
2
16 2 m
p+m
0 (p, m) = i/

(34)

(35)

192

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

If we now substitute (34) and (35) into the quadratic equation (27), we find that the
functions must obey the following linear condition if the improvement procedure suggested
in Eq. (22) works:
m2
0 (p, m)
2 m



m2
p2
+ d db i/
.
(36)
pm + d 2db + dbm
= d
2
2
The explicit expressions for 0 and 1 can be read from the one-loop expression for the
fermion propagator up to O(a):
a
p + m + p2
S 1 (p, m) = i/
2
g 2 CF 
2
i/
p
+ L(ap, am)
16.64441 2.24887cSW 1.39727cSW
16 2



m2
+ 2 1 T p2 /m2
p
g 2 CF 
2
11.06803 2 9.98679 cSW 0.01689 cSW
m
16 2


+ (3 + )L(ap, am) + (3 + )T p2 /m2
p0 (p, m) +
1 (p, m) + i/

g 2 CF 
2
7.13891 0.07187 + 0.48567cSW 0.08173 cSW
16 2

1
(3 2 3cSW )L(ap, am)
2
g 2 CF 
2
6.34664 + 0.14375 1.48503 cSW + 1.28605 cSW
iam/
p
16 2


1
(3 + 2 + 3cSW )L(ap, am) ( + 3cSW )T p2 /m2
2



m2
1
(3 + 4 3cSW ) 2 1 T p2 /m2
2
p

2
2 g CF
2
14.03413 + 1.07187 + 15.48574 cSW 1.52344 cSW
am
16 2



1
1
2
2
(9 + 6cSW )L(ap, am) (12 + 5 3cSW )T p /m .
2
2
(37)
ap2

They are


2
p 16.64441 2.24887cSW 1.39727cSW
0 (p, m) = i/




m2
+ L(ap, am) + 2 1 T p2 /m2
p

2
+ m 11.06803 2 9.98679 cSW 0.01689 cSW


+ (3 + )L(ap, am) + (3 + )T p2 /m2 ,

(38)

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

and

193


2
1 (p, m) = p 7.13891 0.07187 + 0.48567cSW 0.08173 cSW

1
(3 2 3cSW )L(ap, am)
2

2
+ im/
p 6.34664 + 0.14375 1.48503 cSW + 1.28605 cSW
2

1
(3 + 2 + 3cSW )L(ap, am) ( + 3cSW )T p2 /m2
2




m2
1
(3 + 4 3cSW ) 2 1 T p2 /m2
2
p


2
+ m2 14.03413 + 1.07187 + 15.48574 cSW 1.52344 cSW
1
(9 + 6cSW )L(ap, am)
2



1
(12 + 5 3cSW )T p2 /m2 ,
2

(39)

where
T (x) ln(1 + x)/x,
L(x, y) E F0 + ln(x 2 + y 2 )

(40)

with F0 = 4.369225 . . . and E = 0.577216 . . .. Previously [16] we calculated the fermion

propagator in the limit m2  p2 , but Eqs. (38) and (39) are valid for any ratio m2 /p2 (but
a 2 m2 and a 2 p2 must both be small).
Despite the complicated form of Eqs. (38) and (39) it can be checked that at cSW = 1,
and only at this value, Eq. (36) is satisfied by 0 and 1 , and hence Eq. (27) is fulfilled.
This allows us to fix the improvement coefficients, which in a general covariant gauge have
the values
cSW = 1 + O(g 2 ),
g 2 CF
(10.91085 1.85625) + O(g 4 ),
= 1 +
16 2


g 2 CF
4
16.39210 + O(g ) ,
b = 1 +
16 2


1
g 2 CF
4
22.79406 + O(g ) .
bm = 1 +
2
16 2

(41)

Both b coefficients are gauge invariant. 3

In addition to the propagator, Eq. (37), our calculation also gives us a one-loop
expression for the critical coupling [14,15]
3 There was a mistake in b in [16], which has been corrected here.
m

194

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

c =





g 2 CF
1
2
4
1+

1.4288
c
)
.
12.8587

3.4333
c
+
O(g
SW
SW
8
16 2

(42)

In perturbation theory the quark propagator has a pole in the complex momentum plane
at p2 = m2pole . We look for a value of p2 where S 1 (p, m) in Eq. (21) has a zero
eigenvalue. Using Eqs. (38) and (39) gives for the pole mass



g 2 CF 
1 + am(cSW 1) 6 ln(ampole)
mpole 1 +
16 2




g 2 CF
2

1.38038
c
16.95241
+
7.73792
c
=m 1+
SW
SW
16 2





g 2 CF
am
2

8.23318
c
12.32005
+
18.70718
c
1+
1
. (43)
SW
SW
2
16 2
Note that mpole is gauge invariant, as it should be [17]. At cSW = 1 the pole mass is given
by


g 2 CF
6
ln(am
)
mpole 1 +
pole
16 2




am
g 2 CF
g 2 CF
23.30995 1
22.79406
1+
=m 1+
16 2
2
16 2


g 2 CF
23.30995
.
(44)
= m? 1 +
16 2
The pole mass becomes a function of m? , with the same value of bm as in Eq. (41). The
unwanted am ln(ampole) term vanishes, and so the logarithm has the same coefficient
as in the continuum. We will see that this is always the case, that when cSW 6= 1 the
coefficients of the logarithm are changed by an amount proportional to am, but at cSW = 1
all logarithmic terms have their correct values.
We define our renormalisation constants Z in two different renormalisation schemes,
MS, and a momentum subtraction scheme, which we will call MOM. In both cases we
define the Zs in terms of the improved fermion propagator S? .
The MS renormalisation constants are defined from




MS
(2 )m? ,
(45)
S? (p, m? ) = Z2MS (2 )SMS p, mMS = Z2MS (2 )SMS p, Zm
where SMS is the continuum fermion propagator calculated perturbatively in the MS scheme
at the scale :
1
(p, mMS )
SMS

= i/
p + mMS



 2




p + m2MS
m2MS
g 2 CF
2
2
/m
1

T
p

ln
+

MS
16 2
2
p2






p2 + m2MS
g 2 CF
2
2
mMS
4 2 + (3 + ) ln
+ (3 + )T p /mMS .
16 2
2
i/
p

(46)

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

195

In the MS scheme at the scale we find for the renormalisation coefficients Z2MS and
Zm as defined in Eq. (45):
MS

Z2MS (2 ) = 1 +

MS
(2 ) = 1
Zm

g 2 CF 
2 ln(a) + 16.64441 2.24887cSW
16 2

2
1.39727cSW
3.79201 ,
g2
16 2

(47)

CF 6 ln(a) 12.95241


2
.
7.73792 cSW + 1.38038 cSW

(48)

In MOM we define the Zs at the subtraction scale M through

S? (p, m? ) =

Z2MOM (M 2 )
,
MOM (M 2 )m
i/
p + Zm
?

(49)

when p2 = M 2 . This implies that

i

 1

p2
1
,
Tr p
/ S? (p, m? ) = MOM
4Nc
Z2
(M 2 )



Z MOM (M 2 )
1
.
Tr S?1 (p, m? ) = m? m
4Nc
Z2MOM (M 2 )

(50)
(51)

The advantage of the MOM scheme is that all the quantities involved can be calculated
on the lattice, so it can be used non-perturbatively too. This is different from the MS
scheme, where we need to compare with a continuum quantity which we can only find
perturbatively.
The Zs in the MOM scheme are not simple as in the MS scheme, they still have mass
and gauge dependences which cancel in the MS case:

g 2 CF
2
Z2MOM (M 2 ) = 1 +
16.64441 2.24887cSW 1.39727cSW
16 2




m2?
2
2
+ L(aM, am? ) + 2 1 T M /m?
+ O(a), (52)
M

g 2 CF
MOM
2
(M 2 ) = 1 +
5.57638 + + 7.73792 cSW 1.38038 cSW
Zm
16 2


3L(aM, am?) (3 + )T M 2 /m2?




m2?
2
2
+ O(a).
(53)
+ 2 1 T M /m?
M
MOM becomes gauge independent when the fermion is on-shell, i.e.,
Note, however, that Zm
2
at the point M = m2pole .
The dependence on the lattice spacing a and clover coefficient cSW is the same in the
MS and MOM schemes, so that the ratio Z MS /Z MOM is independent of cSW . This is as it
should be, because the ratio Z MS /Z MOM is simply the conversion factor between the two
schemes, which can be calculated in the continuum, and so should not refer to the lattice
in any way.

196

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

4. Renormalisation and improvement of quark bilinear operators

4.1. Method
We are interested in calculating the Z factors and improvement coefficients for quark
bilinear operators. Let us first set out our notation for a general operator. We consider
forward matrix elements only, so improvement operators proportional to a total derivative
will be dropped.
All the operators in Section 2 have the form

+ amc0O
+a
Oimp = O

n
X

k ,
ck Q

(54)

k=1

k are operators with the

where O
is the original unimproved operator, and the Q
same symmetries as the original, but dimension one higher. Explicit expressions for the

/ /2.
Qk can be found in Table 1. For example, Q1 for the scalar operator is D
We define the flavour non-singlet Greens function in the usual way:
GO

imp

(p, m; c0 , c1 , . . . , cn )
1 X ip(xx 0)
e
=
V
x,x 0 , y,y 0
D
E

0)
(x)(y)
O + amc0O + ac1 Q1 + y,y 0 (y 0 )(x

,,A

1 X ip(xx 0)
=
e
V
x,x 0
D
 E
M 1 (O + amc0 O + ac1 Q1 + + acn Qn )M 1 x,x 0
A
D
E
1
1
n
1
F M (O + amc0O + ac1 Q + + acn Q )M
,
A

(55)

where M is the fermion matrix, and F denotes the Fourier transform from position to
imp
momentum space. The c0 dependence of GO is simple. We can write
GO

imp

(p, m; c0 , c1 , . . . , cn ) = GO

imp

(p, m; 0, c1, . . . , cn )

+ amc0 GO

imp

(p, m; 0, 0, . . . , 0),

(56)

so we only need to give expressions for G for the case c0 = 0.

Just as we have contact terms in the fermion propagator, we should expect to see
O(a) contact terms arising in Eq. (55) when x = y or x 0 = y 0 . These will give rise to
a contact Greens function, C O , which will have to be subtracted from the operator
Greens function, just as we subtracted a function from the fermion propagator. This
contact term is thus given by
 E
1 X ip(xx 0) D 1
e
M O + OM 1 x,x 0
C O (p, m) =
A
V
x,x 0

1

F M O + OM 1 A .
(57)

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

197

Fig. 1. The one-loop lattice Feynman diagrams needed for the contact Greens function CO (p, m),
defined in Eq. (57).

Since the coefficient of C O will be O(a), we only need to calculate it for the unimproved

operator O,
and we only need the leading order in a. The Feynman diagrams needed to
O
calculate C to O(g 2 ) are shown in Fig. 1. There is no extra calculation involved. All the
graphs needed already occur in the perturbative expansion of the operator and propagator. 4
Finding the contact Greens function is simple when we consider point operators of the
i where i is any 4 4 matrix. Because there are no covariant derivatives in the
type
operator, it is unaffected by the averaging over gauge fields, and Eq. (57) simplifies to give


(58)
C i (p, m) = i S(p, m) + S(p, m)i = i , S(p, m) .
The contact Greens function are more complicated in the general case. Their expressions
for the one-link operators are given in Appendix A (Sections (A.6) to (A.9)).
So, by subtracting a contact term proportional to C O (p, m) in a Greens function and by
choosing appropriately the improvement coefficients, an improved Greens function can be
obtained. The resulting expression for a renormalised improved off-shell Greens function
is:
2
GO
R (p, mR ; ) =

ZO (2 ; c1 , . . . , cn )
Z2 (2 ) (1 + amb )


a
imp
GO (p, m; c0 , c1 , . . . , cn ) O C O (p, m) .
2

(59)

The factor 1/[Z2 (2 )(1 + amb )] accounts for the wave-function renormalisation.
The Greens function in Eq. (55) depends linearly on the ck coefficients, while the
4 A complete listing of the graphs can be found, for example, in [10,11] or [18].

198

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

renormalised Greens function is independent of the ck s. From this we can deduce that
1/ZO (2 ; c1, . . . , cn ) must depend linearly on ck too, so we can write
ZO (2 ; c1 , . . . , cn ) =

ZO (2 ; 0, . . . , 0)
P
,
1 + nk=1 k ck

(60)

where the s are coefficients depending on g 2 . At the one-loop level, using the fact that
Z = 1 + O(g 2 ) and that the s are O(g 2 ), we can write
ZO (2 ; c1 , . . . , cn ) = ZO (2 ; 0, . . . , 0)

n
X

k ck + O(g 4 ).

(61)

k=1

Our final formula for the renormalised and improved Greens function is
2
GO
R (p, mR ; ) =

ZO (2 ; 0, . . . , 0)
1
P
Z2 (2 )(1 + amb ) 1 + k k ck


a
O imp
O
G
(p, m; c0 , c1 , . . . , cn ) O C (p, m) .
2

(62)

The improvement coefficients b, c, and are independent of the renormalisation

scheme, while the renormalisation constants Z are in general scheme dependent. Therefore
it can be useful to split the renormalisation and improvement into two stages, and to define
improved bare Greens functions by
GO
? (p, m? ) =

1
1
P
1 + amb 1 + k k ck


a
O imp
O
G
(p, m; c0 , c1 , . . . , cn ) O C (p, m) ,
2

(63)

where m? = m(1 + ambm ), using the same value for bm as found from the fermion
propagator (Eq. (41)). As in the propagator section, we will use the suffix ? to denote
bare quantities which have been improved to O(a).
The second step is then to renormalise this improved Greens function multiplicatively,
2
GO
R (p, mR ; ) =

ZO (2 ; 0, . . . , 0) O
G? (p, m? ).
Z2 (2 )

(64)

It is useful to write corresponding equations for amputated Greens functions too. We

define the amputated Greens function in the standard way:
O

imp

(p, m; c0 , c1 , . . . , cn ) S 1 (p, m)
GO

imp

(p, m; c0 , c1 , . . . , cn )S 1 (p, m),

(65)

where S(p, m) is the full fermion propagator. The amputation of Eq. (65) removes
all fermion self-energy diagrams from the perturbative expansion of . The one-loop
Feynman diagrams for are shown in Fig. 2.
The improved amputated Greens function, ? , is naturally defined by
1
O
1
O
? (p, m? ) S? (p, m? )G? (p, m? )S? (p, m? ).

(66)

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

199

Fig. 2. The one-loop lattice Feynman diagrams needed for the amputated Greens function (vertex
imp
function), O (p, m; c0 , c1 , . . . , cn ), as defined in Eq. (65).

From Eq. (64) we obtain the renormalised amputated Greens function:

2
2
2
O
O
R (p, mR ; ) = Z2 ( )ZO ( ; 0, . . . , 0)? (p, m? ).

(67)

Substituting Eq. (29) and Eq. (63) and using Eq. (65) we find

1 + amb
imp
O
P
O (p, m; c0 , c1 , . . . , cn )
? (p, m? ) =
1 + k k ck
a
O S 1 (p, m)C O (p, m)S 1 (p, m)
2


a  1
O imp
(p, m; c0 , c1 , . . . , cn ) . (68)
+ S (p, m),
2
Similarly to what was done in the case of the propagator, one can now expand O , S 1
and C O in powers of a and thus obtain a non-linear condition analogous to Eq. (27),
from which the improvement coefficients can be derived. In the case of the local operators
i , the expression Eq. (58) for the contact term means that we can write the expression

for ? in the simpler form


imp

1 + amb
a 
i (p, m; c0 , c1 ) i S 1 (p, m), i
? i (p, m? ) =
i i
2
1 + 1 c1

imp

a  1
i
(p, m; c0 , c1 ) .
(69)
+ S (p, m),
2
From Eq. (68) we can see that the two improvement coefficients and O associated
with the contact terms are only needed for off-shell improvement, because the inverse
propagator S 1 (p, m) vanishes on-shell.

200

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

4.2. Results for point quark operators

We shall now calculate the matrix elements of all point operators up to O(g 2 a) including
the finite terms. This goes beyond the work of Heatlie et al. [7], who only considered the
g 2 a ln a terms. Including all O(g 2 a) terms will enable us to compute the improvement
coefficients to O(g 2 ).
The calculations are carried out for arbitrary m2 /p2 , not just for m2  p2 as in our
previous papers [16,18,19]. We give the results for the amputated Greens functions.
In this section we show how the improvement coefficients and renormalisation constants
are calculated in one particular case, that of the scalar operator

imp
1
imp = (1 + amc0 )
ac1 D
=
/ .
(70)
OS
2
The Greens functions and results for the other operators are given in the appendix. We
consider forward matrix elements, therefore we drop the total derivative terms in the
improved bases (although in the scalar case this does not make any difference), which
now all have the form

1
imp
.
ac1 D,
(71)
/ + amc0
O =
4
The one-loop expression for the amputated scalar Greens function up to O(a) is:


S p, m; 0, c1S
= 1 ai/
pc1S
g 2 CF h
2
+
11.06803 + 2 + 9.98679 cSW + 0.01689 cSW
16 2


2
+ c1S 19.17181 + 13.80068 cSW 3.53833 cSW

i
(3 + )L(ap, am) 3(3 + )T p2 /m2

i
mh
g 2 CF
2
2
4
/m
1
+
T
p
16 2 p2

g 2 CF
2
6.34664 0.14375 + 1.48503 cSW 1.28605 cSW
+ ai/
p
16 2


2
+ c1S 11.65102 1.41422 cSW + 0.78465 cSW




3
3
S
+ + c1 + cSW L(ap, am) + 3( + 3cSW )T p2 /m2
+
2
2


 2



15
15
m2
m
S
2
2
+
10

3c
c
+

/m
1

T
p
4 2
SW
1
m + p2
2
2
p2

g 2 CF
2
31.06826 4.14375 33.97148 cSW + 3.04688 cSW
+ am
16 2


2
+ c1S 6.08204 + 2.20074 cSW + 1.44647 cSW
+ i/
p

+ (9 + 6cSW )L(ap, am) 2(3 + )

m2
m2 + p 2

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228






+ 2 12 + 5 3 c1S c1S 3cSW T p2 /m2 .

201

(72)

We only need to give the expression for S when c0 = 0, since the full expression with
non-zero c0 can be recovered by using Eq. (56).
To improve the Greens functions, we need to choose the improvement coefficients so
that all O(a) terms in the expressions Eq. (63) or Eq. (68) vanish. It is not immediately
obvious that this will be possible, because there are many more O(a) terms than there
are improvement coefficients, and therefore more equations to be satisfied than there are
unknowns. For general cSW we can not satisfy all the equations, we can only remove all
O(a) effects if cSW = 1 + O(g 2 ). In this case we can derive perturbative expressions for
the improvement coefficients. The results are


g 2 CF
22.79406 8.45146 c1S ,
16 2


g 2 CF
16.39210 10.88629 c1S ,
S = 1 c1S +
2
16
2C
g
F
8.90946.
1S =
16 2
c0S = 1 c1S +

(73)

All three improvement coefficients are gauge invariant. There is one free parameter in this
system of equations. The improvement coefficient c1S can take any value, but once it is
chosen, the values of the other improvement coefficients are fixed. This freedom comes
from an equation of motion, which allows us to compensate for a change in one of the
improvement coefficients by adjusting the other two coefficients. For example, there is a
particularly interesting value of c1S where S vanishes, which means that the scalar threepoint function contains no contact terms, and so even off-shell it is simply renormalised by
a multiplicative factor. This value of c1S is
c1S = 1 +

g 2 CF
5.50582 + O(g 4 ).
16 2

(74)

The improvement coefficients c0S and S are only defined at cSW = 1, but 1S can also be
defined for general cSW values. The result is
1S =

g 2 CF
2
19.17181 + 13.80068 cSW 3.53833 cSW
.
2
16

(75)

All these results are gauge invariant.

We calculate the continuum Greens functions (needed for the Z MS factors) in the MS
(minimal subtraction) scheme. In this paper we use a totally anticommuting 5 , even when
d 6= 4. For the scalar Greens function the result in the MS scheme is


 2



p + m2MS
g 2 CF
S
2
2
4 + 2 (3 + ) ln
3(3 + )T p /mMS
MS (p, mMS ) = 1 +
16 2
2


/ g 2 CF 
mMSp
4 1 + T p2 /m2MS .
(76)
+i 2
2
p 16

202

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

We can now calculate Z MS which is defined by

Z2MS (2 )ZSMS 2 ; c1S S p, m; 0, c1S = MS
S (p, mMS ).

(77)

The result is


S

ZSMS 2 ; c1 =
=

g 2 CF
2 + 6 ln a
+
1.38038
c
12.95241

7.73792
c
SW
2
SW
16

S
2C
2
F
1 + g16
2 19.17181 + 13.80068 cSW 3.53833 cSW c1
ZSMS (2 ; 0)
.
(78)
1 + 1S c1S
1+

In the MOM scheme we define the Z for an operator O by





 Born
Born 
MOM
(M 2 ) Tr Born
Z2MOM (M 2 )ZO
O (p) O (p) = Tr O (p) O (p) , (79)
is the operators Born term. Applying this definition to the
when p2 = M 2 , where Born
O
scalar operator gives
Z2MOM (M 2 )ZSMOM (M 2 ; 0) =

4Nc
,
Tr[S (p, m; 0, 0)]

(80)

at p2 = M 2 , so
ZSMOM (M 2 ; 0) = 1 +


g 2 CF
2
5.57638 7.73792 cSW + 1.38038 cSW
16 2
+ 3L(aM, am?) + 3(3 + )T M 2 /m2?




m2
?2 1 T M 2 /m2?
+ O(a).
M

(81)

The same procedure can be repeated for all the local operators. In Tables 26 we give
the improvement coefficients and MS renormalisation constants for all point operators.
The MS renormalisation constants are defined by equations analogous to Eq. (77). The
Table 2
The renormalisation constants Z MS for the point operators (with no improvement term added, i.e.,
c1 = 0)
Z MS (2 ; 0)
S

2
2 + 6 ln a)
1 + g CF2 (12.95241 7.73792 cSW + 1.38038 cSW

2
2 + 6 ln a)
1 + g CF2 (22.59544 + 2.24887cSW 2.03602 cSW

V
A
H

16

16
g 2 CF
2 )
1+
(20.61780 + 4.74556 cSW + 0.54317cSW
16 2
2
2 )
1 + g CF2 (15.79628 0.24783 cSW + 2.25137cSW
16
2
2 2 ln a)
1 + g CF2 (17.01808 + 3.91333 cSW + 1.97230 cSW
16

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

203

Table 3
The improvement coefficients 1 . These give the renormalisation constants when c1 is non-zero,
see Eq. (61)
1
g 2 CF
16 2
g 2 CF
16 2
g 2 CF
16 2
g 2 CF
16 2

S
V
A
H

2 )
(19.17181 + 13.80068 cSW 3.53833 cSW
2 )
(9.78635 + 3.41640 cSW + 0.88458 cSW
2 )
(19.37225 + 10.31673 cSW 0.88458 cSW
2 )
(16.24376 + 6.85531cSW + 0.58972 cSW

Table 4
The improvement coefficients c0 for general c1 at cSW = 1
c0


2
1 c1S + g CF2 22.79406 8.45146 c1S
16


2
1 c1V + g CF2 18.14912 6.96476 c1V
16


2
1 c1A + g CF2 18.02539 13.36028 c1A
16


2
1 c1H + g CF2 16.47708 12.86471c1H

S
V
A
H

16

lattice s and MS-scheme s are all given in Appendix A. In several of the tables there
is no entry for the pseudoscalar operator this is because it has no c1 improvement term,
so the associated quantities are not defined. The Zs in the MOM scheme are given in
Appendix A.
The c0 values in Table 4 agree with the values given in [20] for the case c1 0.
4.3. Results for one-link quark operators
We consider now the operators in Eq. (7). We study the improvement of these operators
along the same lines used for the point operators. The expressions for the contact Greens
functions (57) will be more complicated, and are given in Appendix A.
From Eqs. (15), (16) and (17), we can see that in forward matrix elements a basis for the
(6)
(3)
improvement is given in the unpolarised case (O , 3 and O , 1 ) by
O

imp

X
 

1
D + 1 aic1
D , D
(1 + amc0)

2
8

1
ac2 D , D ,
8

(82)

204

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

Table 5
The improvement coefficients for general c1 at cSW = 1

2
1 c1S + g CF2 (16.39210 10.88629 c1S )

2
1 c1V + g CF2 (13.71395 2.29988 c1V )

16

16
g 2 CF
A
1 c1 +
(11.46014 8.37649 c1A )
16 2
2
1 c1H + g CF2 (10.56742 5.51435 c1H )
16

A
H

Table 6
The improvement coefficients c0 and c1 when the contact term is chosen to be zero
c0
g 2 CF
16 2
g 2 CF

16 2
g 2 CF

16 2
g 2 CF
16 2
g 2 CF

16 2

S
P
V
A
H

8.83678

c1
2
1 + g CF2 5.50582

16

5.21443
0.22971
1.58146
1.44071

1+

g2C

11.41407

16 2
g 2 CF
1+
3.08365
16 2
2
1 + g CF2 5.05307
16

5 , (6) ) by
in the polarised case with 6= (O
4
5
O

imp

1
1
D 2
5 D iac1
(1 + amc0)
2
4
X

1
5 D , D ,

aic2
8

(83)

6=,

5 , (3) ) by
and in the polarised case when = (O
4

1
1
5 D , D .
5 D aic1
(1 + amc0)
(84)
2
8
For the one-link operators, we calculate the Greens functions in the limit m2  p2 ,
keeping terms up to first order in m. Using the results of our O(a) calculations which we
have collected in Appendix A, we can derive the values of the renormalisation constants
and improvement coefficients which are given in the Tables 713. For each operator there
is a particular value of the ci s where the coefficient O vanishes, and therefore there are
no contact terms and even off-shell the operator is simply renormalised by a multiplicative
factor. These values are given in Table 13.
Note that in the unpolarised case we can only determine c1 to O(g 0 ) from our one-loop
calculation. This is because c1 is the coefficient of an operator which vanishes at tree-level
5
O

imp

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

205

Table 7
The MS-scheme renormalisation constants for the one-link operators
Z MS (2 ; 0, 0)
(6)

2
2 16 ln a)
1 + g CF2 (1.27959 + 3.87297cSW + 0.67826 cSW
3

(3)

2
2 16 ln a)
1 + g CF2 (2.56184 + 3.96980 cSW + 1.03973 cSW
3

O , 3
O , 1

5 , (6)
O
4
(3)

5 ,
O
4

16

16
g 2 CF
2 16 ln a)
1+
(0.34512 + 1.35931 cSW + 1.89255 cSW
3
16 2
2
2 16 ln a)
1 + g CF2 (0.16738 + 1.24953 cSW + 1.99804 cSW
3
16

Table 8
The improvement coefficients 1 for the one-link operators
1
(6)

O , 3

(3)

O , 1

(6)

5 ,
O
4

(3)

5 ,
O
4

g 2 CF
(4.27417 + 1.08793 cSW )
16 2
g 2 CF
(4.27417 + 1.08793 cSW )
16 2
g 2 CF
2 )
(5.61603 + 4.10778 cSW 0.26315 cSW
16 2
g 2 CF
2 )
(15.31376 + 8.54773 cSW 0.26036 cSW
16 2

Table 9
The improvement coefficients 2 for the one-link operators
2
(6)

O , 3

(3)

O , 1

(6)

5 ,
O
4

g 2 CF
2 )
(9.40584 + 4.60327cSW + 0.46669 cSW
16 2
g 2 CF
2 )
(6.67330 + 4.53710 cSW + 0.44621 cSW
16 2
2
g CF
2 )
(8.29791 + 4.21724 cSW 0.49384 cSW
16 2

(3)

5 ,
O
4

(because it involves [D , D ]). However, we do still know the improved Greens function
to O(g 2 ).

5. Off-shell improvement for GinspargWilson fermions

Like clover fermions, GinspargWilson fermions are free of O(a) effects on-shell. So it
is instructive to see what happens when our off-shell improvement conditions (28) and (63)
are applied to GinspargWilson fermions [21].

206

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

Table 10
The improvement coefficients c0 for general c2 at cSW = 1 for the one-link operators
c0
(6)

2
1 c2 + g CF2 (16.34500 12.24534 c2 )

(3)

2
1 c2 + g CF2 (17.20377 8.69045 c2 )

(6)

2
1 c2 + g CF2 (16.28373 10.73103 c2 )

O , 3

16

O , 1

16

5 ,
O
4

16
g 2 CF
1 c1 +
(16.95724 11.33434 c1 )
16 2

5 , (3)
O
4

Table 11
The improvement coefficients c1 for general c2 at cSW = 1 for the one-link operators
c1
(6)

c2 + O(g 2 )

(3)

c2 + O(g 2 )

(6)

2
c2 + g CF2 (1.18321 + 0.84682 c2 )

O , 3
O , 1
5 ,
O
4

16

(3)

5 ,
O
4

Table 12
The improvement coefficients O for general c2 at cSW = 1 for the one-link operators
O
(6)

2
1 c2 + g CF2 (8.13135 3.04784 c2 )

(3)

2
1 c2 + g CF2 (9.28735 0.52613 c2 )

(6)

2
1 c2 + g CF2 (8.48845 1.84428 c2 )

O , 3

16

O , 1

16

5 ,
O
4

16
g 2 CF
1 c1 +
(7.96628 3.12321 c1 )
16 2

5 , (3)
O
4

Table 13
The values of c0 , c1 , c2 for which the one-link operators have no contact terms (O = 0)

(6)

O , 3

(3)

O , 1

(6)

5 ,
O
4

(3)

5 ,
O
4

c0

c1

c2

2
g CF2 0.98385

c2 + O(g 2 )

2
1 + g CF2 5.08351

c2 + O(g 2 )

2
1 + g CF2 8.76121

2
1 + g CF2 8.67421

2
1 + g CF2 6.64417

2
1 + g CF2 4.84307

16
2
g CF2 0.24789
16
2
g CF2 1.09147
16
2
+ g CF2 0.77983
16

16
16

16
16
16

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

207

The defining GinspargWilson relation is

DGW 5 + 5 DGW = aDGW 5 DGW ,

(85)

where DGW is a GinspargWilson fermion matrix. From this matrix we can (at least in
principle) define a related matrix [22]

1
a
DGW .
(86)
KGW 1 DGW
2
The eigenvalues of DGW lie on a circle of radius 1/a and centre 1/a, while the eigenvalues
of KGW lie on the imaginary axis. From Eq. (85) and Eq. (86) we find that
KGW 5 + 5 KGW = 0.

(87)

The propagator which we would really like to know is the fermion propagator corresponding to KGW . It should have the correct chiral properties, and be free from O(a) discretisation errors. However, we cannot work directly from KGW , because it is non-local. Therefore, we need to write down a formula for the propagator we would get from KGW , but
expressed in terms of DGW . This propagator will satisfy chirality even at zero distance, so
we expect it to be improved off-shell too.
Let us now add mass to the problem in the same way as it is added in the clover case,
to the action, giving
simply by adding a mass term m
GW + m],
[D

(88)

as the fermionic part of the action. Another way to add mass effects would be to use the
alternative action


(89)
(1 am? /2)DGW + m? ,
which has the advantage that there is no mass improvement needed. That is the method
we used in [21]. Here we have added the simple mass term, Eq. (88), because we want to
preserve the analogy with the clover action.
If we reexpress the unimproved massive propagator (DGW + m)1 in terms of KGW , we
find




1
a
1
1
(x y). (90)
=
+
DGW + m A (1 + a/2m)2 [KGW + m/(1 + a2 m)] A 2 + am
Fourier transforming, we get
1
a
,
S? (p, m? ) +
SGW (p, m) =
(1 + a2 m)2
2 + am

(91)

where
m?

m
,
1 + a2 m

(92)

and S? is the Fourier transform of h(KGW + m? )1 iA . Eqs. (91) and (92) are the analogues
of Eqs. (22) and (23) respectively, remembering that terms of O(a 2 ) are dropped in
Section 3.1. Solving Eq. (91) for S? we find
2 


a
a
,
(93)
SGW (p, m)
S? (p, m? ) = 1 + m
2
2 + am

208

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

which has the same form as Eq. (28). Note that the only matrix we need to invert to
calculate this improved propagator is the matrix (DGW + m), which is well-defined, and
local (in the sense that its elements decrease exponentially with separation). Comparing
these formulae with those in Section 3.1 we see that
= 1,

1
b = 1 and bm = .
2

(94)

These results are independent of g 2 . These all-order results coincide with the tree-level
limit of the clover fermion result Eq. (41). The values depend on the fact that in this paper
we have added mass term to the GinspargWilson action in the same way as to the clover
action.
Next we want to improve the Greens function corresponding to a flavour non-singlet

operator O O,
where O includes Dirac structure and covariant derivatives. We want
our improved Greens function GO
? (p, m? ) to be given by


1
1
(p,
m
)
=
F
O
,
(95)
GO
?
?
KGW + m? KGW + m? A
where F denotes the Fourier transform. However, we need to reexpress it in a form that
involves only DGW , not KGW . This can be shown to be equivalent to the expression
2 


a
a
O imp
O
m

(p,
m
)
=
1
+
(p,
m)

C
(p,
m)
,
(96)
G
GO
?
O
?
GW
2
2
where
imp
GO
GW (p, m) F

1
DGW + m

imp

1
DGW + m


,

(97)

a
a2
O imp O + amc0 O c1 (DGW O + ODGW ) + DGW ODGW ,
2
4


1
1
O
+
O ,
C (p, m) F O
DGW + m DGW + m A

(98)
(99)

with
c 0 = 1 c1 ,

(100)

O = 1 c1 .

(101)

Eq. (96) has the same form as Eq. (63) (up to terms of O(a 2)). A more general form of
Eq. (96) can be found in [21].
Comparing Eqs. (100) and (101) with Tables 4 and 5, we see again that the allorders GinspargWilson result is just the tree-level result for clover fermions. A particular
point to note is that the GinspargWilson improvement coefficients are the same for
all operators, while the clover action improvement coefficients are operator-dependent.
A further simplification in the GinspargWilson case is that there are no coefficients
needed, they are all zero. This means that the renormalisation constant ZO is independent
of the improvement coefficients ci (see Eq. (60)).

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

209

6. Tadpole improvement
6.1. Analytic results
It is well known that many results from (naive) lattice perturbation theory are in poor
agreement with their counterparts determined from Monte Carlo calculations. One main
reason for this is the appearance of gluon tadpoles, which are typical lattice artifacts. They
make the bare coupling g into a poor expansion parameter. Therefore, it was proposed [23,
24] that the perturbative series should be rearranged in order to get rid of the numerically
large tadpole contributions. This rearrangement will be done by using the variable u0 ,
derived from the measured value of the plaquette

1/4
1
Tr U2
.
(102)
u0 =
Nc
Its value depends on the coupling g 2 = 6/ where it has been measured.
There are two main steps involved in tadpole improvement.
Firstly, we know [24] that in the mean field approximation the Z for an operator with
nD derivatives is
D
,
ZO u1n
0

(103)
(ZO u0nD 1 )

will converge more

so it is reasonable to hope that a perturbative series for
rapidly than a series for ZO itself. Secondly, instead of writing our series in terms of the
bare parameters, we reexpress it in terms of the tadpole improved, TI, parameters
2
g 2 u4
gTI
0 ,

TI
cSW
cSW u30 ,

ciTI ci un0 ,

(104)

where n is the difference between the number of covariant derivatives in the higher
dimensional operator multiplying ci and the number of covariant derivatives in the operator
to be improved (n is always 1 for our choice of improvement terms). The new coupling gTI
is called the boosted coupling constant. Other choices of boosted coupling are possible,
for example one could also use the renormalised coupling constant at some scale 1/a 2 .
2 u4 (g )], where
To carry out this rewriting of the series, we simply replace every g 2 by [gTI
0 TI
u0 (gTI ) is the perturbative expansion for u0 :
2


gTI
4
CF 2 + O gTI
.
(105)
2
16
Formally, this cannot change the all orders result, but it should improve the rate with which
the series converges. The same procedure is followed with the improvement coefficients
TI
u3
cSW and ci , for example cSW is to be replaced by [cSW
0 (gTI )].
In this paper we will look at the tadpole improvement for operators with no anomalous
dimension. The interplay between tadpole improvement and the renormalisation group,
needed when considering operators with an anomalous dimension, will be considered in a
future paper. The result of this procedure is rather simple for the one-loop Z factors. If the
original Z is given by

u0 (gTI ) = 1

ZO = 1 +

g 2 CF
BO (cSW , ci ) + O(g 4 ),
16 2

(106)

210

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

then the tadpole improved Z is given by





2




gTI
TI
TI
TI
2
4
D
= u1n
C
,
c
)
1
+
B
+
O
g
c
+
(1

n
ZO
F
D
O SW i
TI
0
16 2




g2
TI
4
D
+ O gTI
u1n
1 + TI 2 CF BO
.
0
16

(107)

For the V and A operators (nD = 0) in the MS scheme we get the following tadpole
improved BO terms:

2
TI
TI
+ 0.54317 cSW
BVTI,MS = 10.74819 + 4.74556 cSW

2

TI
TI
+ c1TI 9.78635 3.41640 cSW
0.88458 cSW
,
(108)


2
TI
TI
+ 2.25137 cSW
BATI,MS = 5.92668 0.24783 cSW

2

TI
TI
+ c1TI 19.37225 10.31673 cSW
+ 0.88458 cSW
.
(109)
Tadpole improvement is not just applicable to renormalisation factors it can also
be used to give improved values for the improvement coefficients. The improvement
coefficients for the fermion propagator, Eq. (41), become


2


gTI
4
C
(1.04124

1.85625
)
+
O
g
1
+
,
= u1
F
TI
0
16 2


2


gTI
1
4
CF 6.52250 + O gTI ,
b = u0 1 +
16 2


2


1
gTI
4
C
12.92445
+
O
g
1
+
.
(110)
bm = u1
F
TI
2 0
16 2
An unfortunate ambiguity is that there is of course considerable freedom in the choice
of boosted g. At one-loop none of the numerical coefficients are affected by this choice, so
if one prefers another boosted g, all the formulae in this section can still be used, the only
change is that every gTI has to be replaced by the alternative boosted g.
6.2. Comparison with non-perturbative results
To test the validity of tadpole improved perturbation theory, we will now compare our
results with known non-perturbative results in the quenched theory. The local vector current
is best suited for this purpose, because the renormalisation constant ZV and improvement
coefficient c0V are known non-perturbatively for a wide range of values of c1V .
The comparison is done at g 2 = 1. At this value of the coupling one finds nonperturbatively [25] cSW = 1.769. We will use this number in both the perturbative formulae
and the numerical calculations. For u0 we obtain the value 0.8778. We then get
ZVTI = 0.8242 + 0.0486 c1V ,
V ,TI

c0

= 1.2733 1.0990 c1V .

(111)
(112)

In Fig. 3 we show ZVTI and c0V ,TI as a function of c1V . We compare the results with the
numbers of three independent non-perturbative calculations. The first calculation [27] uses

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

211

Fig. 3. The renormalisation constant ZV and the improvement coefficient c0V as a function of c1V .
The solid lines are the results of tadpole improved perturbation theory. The dashed lines refer to the
non-perturbative results of [27], and the symbols mark the non-perturbative results of [26] (solid
circle) and [28] (solid diamond).

the nucleon matrix element of the local vector current to determine ZV and c0V . The second
one is based on the Schrdinger functional [26], and the last calculation [28] uses chiral
Ward identities to improve the current and renormalisation following [4,5]. In the latter
case we calculated only at the value of c1V where V = 0, as in Table 6. It should be noted
that the results still have errors of O(a 2 ), which can be as large as 10% [15], so that we
cannot expect the results to agree completely. For ZV we find good agreement between
the tadpole improved perturbative numbers and all the non-perturbative results. For c0V our
numbers agree with [27] and [28]. The Schrdinger functional result, on the other hand,
lies 10% above the other numbers. (It is important to remember that different definitions
of Z may give results differing by O(a 2), so both results could be consistent.)
In Fig. 4 we show the renormalisation constant ZV as a function of g 2 . At smaller values
of the coupling (higher values of ) the agreement between tadpole improved perturbation
theory and non-perturbative results becomes even closer, as one might expect. In those
cases where we could check this, we found the discrepancy to reduce to 4% at = 6.4.
Thus we may say that the non-perturbative results agree with those of tadpole improved
perturbation theory within the expected O(a 2) and O(g 4 ) uncertainties.
7. Conclusion
In this paper we have presented extensive one-loop perturbative calculations of lattice
Greens functions, in which we have kept all O(a) terms. This allows us to investigate
operator improvement, firstly to see what sort of improvement terms are needed, and
secondly to calculate values of the improvement coefficients.

212

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

Fig. 4. The renormalisation constant ZV as a function of g 2 . The solid line is the result of tadpole
improved perturbation theory, while the dotted line shows the one-loop perturbative result with no
improvement. The dashed line is the non-perturbative result of [26].

We find that we can produce off-shell O(a) improved Greens functions, to all orders
in the GinspargWilson case, and at least to O(g 2 ) in the clover case. In our one-loop
calculations we find that we only need gauge-invariant improvement terms. No extra
improvement terms associated with BRST symmetry are required at this level.
Off-shell improvement doesnt mean that there are no contact terms. As long as we
know the form of the contact terms, we can remove them by using the improvement
coefficients . Contact terms, responsible for the off-shellness of the propagators and
Greens functions, can be removed using a well-determined procedure. There are always
particular values of the improvement coefficients for which the contact terms vanish, so
that one still has a multiplicative renormalisation.
In the GinspargWilson case improvement is particularly simple, because the improvement coefficients are universal, they do not depend on the operator considered, the coupling
constant, or even on which theory we are simulating (we assume that the bosonic sector
has no O(a) discretisation errors). This is not so in the clover case, the coefficients depend
on the coupling, and are different for each operator.
We have the tadpole improved one-loop values for Z factors calculated at arbitrary
cSW , and for improvement coefficients calculated at cSW = 1 + O(g 2 ), which is the
only place where O(a) improvement is possible. Numerical test cases show that tadpole
improvement works well down to 6.0 for operators with no anomalous dimension.
We are investigating tadpole improvement in the case of operators with an anomalous
dimension.

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

213

Acknowledgements
S.C. is supported in part by the US Department of Energy (DOE) under cooperative
research agreement DE-FC02-94ER40818.
Support by the Deutsche Forschungsgemeinschaft and by the BMBF is also gratefully
acknowledged.

Appendix A
In this appendix we give the perturbative expressions for the amputated three-point
functions (vertex functions) for all the operators we have considered (apart from the scalar
density, which is given in Section 4.2 of the main text), calculated to O(a), and the values
of their improvement coefficients. In order to make transparent the transformation of these
numbers into the popular MS scheme the corresponding continuum quantities are also
given.
In order to shorten the expressions for the Greens functions we will use the functions T
and L defined in Eq. (40).
A.1. Pseudoscalar vertex
The pseudoscalar operator is simpler than the other operators because there is no c1
improvement term possible (and also none needed) for the forward three-point function,
/ and 5 anti-commute. The one-loop expression for this vertex up to O(a) is:
because D
P (p, m; c0 = 0)
g 2 CF 
2
1.42500 + 2 + 3.43328 cSW
= 5 + 5
16 2


(3 + )L(ap, am) (3 + )T p2 /m2
+ am5

g 2 CF 
2
3.82788 0.14375 2.49670 cSW
16 2


+ (3 + )L(ap, am) + 2(3 + )T p2 /m2 .

(A.1)

In the MS scheme we have

PMS (p, mMS )
= 5 + 5

 2

p + m2MS
g 2 CF h
4
+
2

(3
+
)
ln
16 2
2
i


(3 + )T p2 /m2MS .

(A.2)

The MOM scheme renormalisation factor is defined by

Z2MOM (M 2 )ZPMOM (M 2 ) =
at p2 = M 2 . The result is

Tr[5

4Nc
,
P
(p, m; 0)]

(A.3)

214

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

ZPMOM (M 2 ) = 1 +

g 2 CF h
2
15.21941 + 2.24887 cSW 2.03602 cSW
16 2


+ 3L(aM, am?) + (3 + )T M 2 /m2?


i
m2
+ O(a).
?2 1 T M 2 /m2?
M

(A.4)

As well as the lack of a derivative improvement term, another special feature of the
pseudoscalar operator is that the improvement terms amP and a{S 1 , 5 }, which appear
in Eq. (69), have the same functional form to this order, so there is no natural way of
determining c0P and P separately. We choose to improve the operator by setting P = 0,
and making all the improvement through the mass-dependent term. Then the improvement
coefficients are (at cSW = 1)
c0P =

g 2 CF
5.21443,
16 2

P 0.

A.2. Local vector

The one-loop expression for the local vector vertex up to O(a) is:


V p, m; 0, c1V
= aip c1V

g 2 CF
2
+
3.97338 + 2.49670 cSW + 0.85410 cSW
16 2

2
c1V 9.78635 3.41640 cSW 0.88458 cSW




m2
L(ap, am) 2 1 T p2 /m2
p


2
2



m
p
/ p g CF
2
2
2 + 4 2 1 T p /m
+ 2
p 16 2
p


i
mp g 2 CF h
2
2
/m
2(3
+
)
1

T
p
+i 2
p 16 2

g 2 CF
2
8.66505 + 2.85625 + 9.52789 cSW 0.39053 cSW
+ aip
16 2


2
+ c1V 18.59361 2.24887 cSW 0.16098 cSW


3 c1V 3cSW L(ap, am) + 2(3 + )

m2
m2 + p 2

2

m



V
V
2
2
18 + 8 6c1 3c1 3cSW 2 1 T p /m
p

2
g CF
2
7.64168 + 0.85625 + 7.74287cSW 1.38589 cSW
+ am
16 2


2
+ c1V 13.96523 5.54358 cSW 0.36162 cSW

(A.5)

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

215

1
3 6c1V 2 3cSW L(ap, am)
2




3 3cSW 3c1V c1V T p2 /m2



2



1
V m
2
2
1 T p /m
+ 3 + 6 3cSW 2c1
2
p2

m/
pp g 2 CF
3 + 4 2c1V + 3cSW
+a
p2 16 2


m2
2( + 3cSW )T p2 /m2 + 4 2
m + p2


m2



V
2
2
6 + 12 4c1 6cSW 2 1 T p /m
.
p

(A.6)

In the MS scheme we have

VMS (p, mMS )


 2





p + m2MS
m2MS
g 2 CF
2
2
ln
2 1 T p /mMS
= +
16 2
2
p


2
2



m
p
/ p g CF
MS
2
2
/m
1

T
p
2
+
4
+ 2
MS
p 16 2
p2


i
mp g 2 CF h
2
2
/m
2(3
+
)
1

T
p
.
+i 2
MS
p 16 2

We consider two ways to define Z MOM for the vector,





p p
1
1
1X
Tr ? MOM MOM ,

2
3
4Nc
p
Z2
ZVtrans
X p 1
 ?
1
Tr p
/ MOM MOM ,
2
p 4Nc
Z2
ZVlong

(A.7)

(A.8)
(A.9)

so that
(M 2 ; 0)
ZVMOM
trans

g 2 CF 
2
20.61780 + 4.74556 cSW + 0.54317cSW
+ O(a),
(A.10)
2
16

g 2 CF
2
2
(M
;
0)
=
1
+
20.61780 + 2 + 4.74556 cSW + 0.54317cSW
ZVMOM
long
16 2




m2?
2
2
4 2 1 T M /m?
+ O(a).
(A.11)
M
=1+

A.3. Conserved vector

The one-loop expression for the conserved vector vertex up to O(a) is:
J (p, m) = aip

216

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228


g 2 CF
2
16.64441 + + 2.24887cSW + 1.39727cSW
16 2




m2
2
2
L(ap, am) 2 1 T p /m
p


2
2



p
/ p g CF
m
2
2
2 + 4 2 1 T p /m
+ 2
p 16 2
p


i
mp g 2 CF h
2(3 + ) 1 T p2 /m2
+i 2
2
p 16

g 2 CF
2
11.27782 + 1.85625 + 3.97134 cSW 0.16345 cSW
+ aip
16 2
+

m2
(3 2 3cSW )L(ap, am) + 2(3 + ) 2
m + p2

2



m
(12 + 5 3cSW ) 2 1 T p2 /m2
p

2
g CF
2
6.34664 0.14375 + 1.48503 cSW 1.28605 cSW
+ am
16 2


1
+ (3 + 2 + 3cSW )L(ap, am) + (3cSW + )T p2 /m2
2




m2
1
2
2
+ (3 + 4 3cSW ) 2 1 T p /m
2
p



m2
m/
pp g 2 CF
2
2

2(
+
3c
)T
p
/m
3
+
2
+
3c
+
4
+a
SW
SW
p2 16 2
m2 + p 2




m2
(6 + 8 6cSW ) 2 1 T p2 /m2 .
(A.12)
p
In the MS scheme we have


 2




p + m2MS
m2MS
g 2 CF
2
2
/m
1

T
p

ln

MS
16 2
2
p2





m2MS
p
/ p g 2 CF
1 T p2 /m2MS
+ 2
2
+
4
2
2
p 16
p
h
2


i
mp g CF
(A.13)
2(3 + ) 1 T p2 /m2MS .
+i 2
2
p 16

JMS (p, mMS ) = +

We can define longitudinal and transverse Z MOM according to Eqs. (A.8) and (A.9), giving
us
(M 2 ) = 1 + O(a),
ZJMOM
trans
(M 2 ) = 1 +
ZJMOM
long





2

g 2 CF
m2
2 4 ?2 1 T M 2 /m?
2
16
M

(A.14)
+ O(a).

(A.15)

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

217

The conserved vector current satisfies Eq. (63) without any improvement terms, i.e.,
c0J = 0,

J = 0.

(A.16)

This is as it should be, the conserved current is already improved for forward matrix
elements, and no further improvement terms are needed.
A.4. Axial vector
The one-loop expression for the axial vector vertex up to O(a) is:


A
A
p, m; 0, c1
i
p 5 + 5p
/ )c1A
= 5 a (/
2

g 2 CF
2
+ 5
0.84813 + + 2.49670 cSW 0.85410 cSW
16 2


2
c1A 19.37225 10.31673 cSW + 0.88458 cSW
L(ap, am)

2





m
2
2
2
2
2(1 + )T p /m (2 ) 2 1 T p /m
p







m2
/ g 2 CF
pp
2
2
2
2
/m
/m
+
4(2

)
1

T
p
2

4(1

)T
p
+ 2 5
p
16 2
p2



m g 2 CF 
i
p 5 + 5p
/) 2
2(1 ) 1 T p2 /m2
+ (/
2
2
p 16

g 2 CF
i
p 5 + 5p
/)
1.34275 + 0.85625 1.71809 cSW
+ a (/
2
16 2


2
2
+ 0.13018 cSW
+ c1A 13.96522 + 0.54301cSW + 0.05366 cSW




m2
+ 1 + c1A L(ap, am) + 2( + cSW )T p2 /m2 + 2(1 ) 2
m + p2


m2



1 4 2c1A + c1A + 2cSW 2 1 T p2 /m2
p

2
g CF
2
7.47851 1.14375 8.74287cSW + 1.38589 cSW
+ am 5
16 2


2
+ c1A 10.92421 4.44321cSW + 0.36162 cSW


1
+ 6c1A + 3 + 2 3cSW L(ap, am)
2




m2
+ 3 + 6 + (1 )c1A cSW T p2 /m2 4 2
m + p2


m2



1
A
2
2
+ 13 6 2(2 )c1 cSW 2 1 T p /m
2
p

2
2
/ g CF
pp
m
+ am 2 5
3 + 4 2c1A + cSW + 4 2
2
p
16
m + p2

218

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

+ 2(6 5 2(1 )c1A cSW )T p2 /m2

2 13 6

2(2 )c1A

cSW

m2
p2





1 T p /m
.
2

(A.17)
In the MS scheme we have
(p, mMS )
A
MS


 2



p + m2MS
g 2 CF

ln
2(1 + )T p2 /m2MS
2
2
16


2



mMS
2
2
(2 ) 2 1 T p /mMS
p







m2MS
/ g 2 CF
pp
2
2
2
2
/m
/m
+
4(2

)
1

T
p
2

4(1

)T
p
+ 2 5
MS
MS
p
16 2
p2



mMS g 2 CF 
i
p 5 + 5p
/) 2
2(1 ) 1 T p2 /m2MS .
(A.18)
+ (/
2
p 16 2

= 5 + 5

Again we can define both transverse and longitudinal Z MOM ,





p p
1
1X
1
Tr 5 ? MOM MOM ,

3
p2
4Nc
Z2
ZAtrans
X p 1


1
Tr 5p
/ ? MOM MOM .
2 4N
p
Z2
ZAlong
c

This gives the MOM scheme renormalisation factors




g 2 CF
MOM
2
2
15.79628 0.24783 cSW + 2.25137cSW
ZAtrans M ; 0 = 1 +
16 2


+ 2(1 + )T M 2 /m2?




m2
+ O(a),
+ 2(1 ) ?2 1 T M 2 /m2?
M



g 2 CF
MOM
2
2
15.79628 0.24783 cSW + 2.25137cSW
ZAlong M ; 0 = 1 +
16 2


+ 2 + 2(3 )T M 2 /m2?




m2
2(3 ) ?2 1 T M 2 /m2?
+ O(a).
M
A.5. Tensor vertex
The one-loop expression for the tensor vertex up to O(a) is:


H
H
p, m; 0, c1
i
p 5 + 5p
/)
= 5 ac1H (/
2

(A.19)
(A.20)

(A.21)

(A.22)

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

+ 5


g 2 CF
2
4.16568 1.66446 cSW 0.57503 cSW
16 2

219

2
c1H 16.24376 6.85531 cSW 0.58972 cSW
+ (1 )L(ap, am)

2





m
2
2
2
2
(1 )T p /m + 2(1 ) 2 1 T p /m
p







m2
g 2 CF
p
/
2
2
2
2
/m
/m

4(1

)
1

T
p
2(1

)T
p
+ i 2 ( p p )5
p
16 2
p2


i
m g 2 CF h
i
2
2
p 5 + 5p
/) 2
/m
4
1

T
p
+ (/
2
p 16 2

g 2 CF
i
/)
p 5 + 5p
3.66115 + 1.85625 + 0.96286 cSW
+ a (/
2
16 2


2
2
+ 0.42868 cSW
+ c1H 16.27942 + 0.26479 cSW 0.26155 cSW




1
+ 2c1H 3 + 2 + cSW L(ap, am) + ( cSW )T p2 /m2
2


m2



m2
1
+ 2(4 + )c1H 19 4 + 3cSW 2 1 T p2 /m2 + 4 2
2
p
m + p2

g 2 CF
2
7.42480 + 1.85625 + 5.16191cSW 0.09170 cSW
+ am 5
16 2


2
+ c1H 17.60663 2 7.02465 cSW 0.24108 cSW


2 4c1H cSW L(ap, am)




+ 2 (1 + )c1H + cSW T p2 /m2


m2



5 6 2(1 )c1H + cSW 2 1 T p2 /m2
p

2
g CF
p
/
3 + cSW
+ iam 2 ( p p )5
p
16 2




m2
+ 2 (1 )c1H (3 2 + cSW ) T p2 /m2 2(1 ) 2
m + p2


m2



+ 2 2(1 )c1H + 5 6 + cSW 2 1 T p2 /m2 .
(A.23)
p
In the MS scheme we have
(p, mMS ) = 5
H
MS

h  p 2 + m2 


g 2 CF
MS
(1

)
ln
T p2 /m2MS
2
2
16

2


i
m
2
2
/m
1

T
p
+ 2 MS
MS
p2
h
2


g CF
p
/
(1

)
2T p2 /m2MS
+ i 2 ( p p )5
2
p
16
+ 5

220

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

i
m2MS
2
2
/m
1

T
p
MS
p2
2


g CF mMS 
i
p 5 + 5p
/)
4 1 T p2 /m2MS .
(A.24)
+ (/
2
2
2
16 p
4

We define Z MOM by


1
1 X 1
Tr 5 ? = MOM MOM
12 4Nc
Z2
ZH

(A.25)

at the scale p2 = M 2 , which gives


g 2 CF
MOM
2
2
+
ZH (M ; 0) = 1 +
20.81009 + 3.91333 cSW + 1.97230 cSW
16 2




m2
+ O(a). (A.26)
L(aM, am?) ?2 1 T M 2 /m2?
M
A.6. First unpolarised moment, off-diagonal ( 6= ), symmetrised
For the one-link operators, we calculate the Greens functions in the limit m2  p2 ,
keeping terms up to first order in m. This is sufficient to calculate the improvement
coefficients.
(6)
Our improved operator for the first unpolarised moment in the 3 representation of the
hypercubic group is
(6)

1 1
3
= (1 + amc0) D + D
O
2 2
X 1
 
 

1
D , D + D , D
+ aic1
8
2


1
ac2 D , D .
8
The result for the amputated operator Greens function is:
(6)
1
3 (p, m; 0, c1, c2 ) = i ( p + p ) + c2 ap p
2


1
g 2 CF
8

ln a 2 p2 18.80927 + 3.79201
+ i ( p + p )
2
3
16 2

2
+ c1 (4.27417 + 1.08793 cSW)
1.62411cSW + 0.71900 cSW



2
+ c2 9.40584 + 4.60327 cSW + 0.46669 cSW



p p g 2 CF
p p g 2 CF
1

m
/
4

+i 2 p
p
16 2
3
p2 16 2


g 2 CF
4
+ ap p
3 cSW ln a 2 p2 2.80639 + 1.43576
2
16
3

(A.27)

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

221



1
2
0.49196 cSW + 0.10443 cSW
+ c1 ln a 2 p2 + 0.86075 0.30348 cSW
3


4
ln a 2p2 33.31690 + 4.29201
+ c2
3

2
+ 1.16439 cSW + 0.04840 cSW
g 2 CF
1
+ iam ( p + p )
2
16 2



1
13
+ + cSW ln a 2 p2
6
2

2
+ 1.76784 2.93576 + 2.56080 cSW 0.92094 cSW


+ c1 ln a 2 p2 2.06261 0.62935 cSW


11
2
ln a 2 p2 + 0.56785 4.93253 cSW 0.19761 cSW
+ c2
3


2
p p g 2 CF 7
+ 2 + cSW + c2
2 .
/
+ ami 2 p
p
16 2 3
3

(A.28)

The contact Greens function is given by

O

C O (p, m) = O (p, m)S(p, m) + S(p, m) (p, m),

(A.29)

where
(6)
1
3 (p, m) = i ( p + p )
2


g 2 CF
1
1
2 2

+
0.19740
c
p

14.27168

ln
a
+ i ( p + p )
SW
2
2
16 2

g 2 CF 1
m1
(
(1 + ),
p
+

p
)/
p

p2 2
16 2 2

(A.30)

(6)
1
3 (p, m) = i ( p + p )
2


g 2 CF
1
1
2 2

+
0.19740
c
p

14.27168

+ i ( p + p )
ln
a
SW
2
16 2
2

g 2 CF 1
m1
p
/ ( p + p )
(1 + ).
2
p 2
16 2 2

(A.31)

In the MS scheme we have

1
MS (p, mMS ) = i ( p + p )
2



p2 31
g 2 CF
8
1

ln

+ i ( p + p )
2
3
9
16 2
2


2
2
p p g CF
1
p p g CF
/
4.
m 2
+i 2 p
3
p
16 2
p 16 2

(A.32)

222

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

A.7. First unpolarised moment, diagonal, traceless

In the 1(3) (i.e., diagonal, traceless) representation of the hypercubic group is
(3)

1 1
1
= (1 + amc0) D D
O
2 2
X 1
 
 

1
D , D D , D
+ aic1
8
2

1
ac2 D D D D .
8
In this expression, repeated and indices are not summed over:

(A.33)

(3)

1 (p, m; 0, c1 , c2 )
1
1
= i ( p p ) + c2 a (p p p p )
2
2


g 2 CF
8
1
ln a 2 p2 17.52702 + 3.79201
+ i ( p p )
2
3
16 2
2
+ c1 (4.27417 + 1.08793 cSW)
1.72093 cSW + 0.35754 cSW



2
+ c2 6.67330 + 4.53710 cSW + 0.44621 cSW



1 (p p p p ) g 2 CF
1
1 (p p p p ) g 2 CF

m
p
/
4

+i
2
p2
16 2
3
2
p2
16 2


g 2 CF
4
1
2 2
c
3

+ a (p p p p )
SW ln a p
2
16 2
3
2
4.58870 + 1.43576 + 1.14260 cSW 0.07987cSW


1
+ c1 ln a 2 p2 + 1.88483 + 1.45305 cSW
3


4
ln a 2 p2 35.68903 + 4.29200
+ c2
3

2
+ 0.97769 cSW 0.04918 cSW

g 2 CF
1
+ iam ( p p )
2
16 2



1
13
+ + cSW ln a 2 p2
6
2

2
+ 1.00108 2.93576 + 2.62151 cSW 0.74162 cSW


+ c1 ln a 2 p2 2.06261 0.62935 cSW


11
2 2
2
ln a p + 0.38293 4.95277 cSW 0.20168 cSW
+ c2
3



2
1 (p p p p ) g 2 CF 7
+ 2 + cSW + c2
2 .
p
/
+ ami
2
3
p2
16 2 3

The contact Greens function is given by

(A.34)

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

O

C O (p, m) = O (p, m)S(p, m) + S(p, m) (p, m),

223

(A.35)

where
(3)
1
1 = i ( p p )
2


g 2 CF
1
1
2 2
ln a p 14.27168 + 0.19740 cSW
+ i ( p p )
2
2
16 2

g 2 CF 1
m
(1 + ),
(A.36)
( p p )/
p
2
p
16 2 2
(3)
1
1 (p, m) = i ( p p )
2

g 2 CF  2 2
1

+
0.19740
c
p

14.27168

ln
a
+ i( p p )
SW
16 2
2
2
g CF 1
m 1
(1 + ).
(A.37)
/ ( p p )
+ 2p
p 2
16 2 2
+

In the MS scheme we have

1
MS (p, mMS ) = i ( p p )
2
 p2 31 i
g 2 CF h 8
1

ln 2
+ i ( p p )
2
16 2 3

9


2
1
1 (p p p p ) g CF
p
/

+i
2
2
2
3
p
16
m

1 (p p p p ) g 2 CF
4.
2
p2
16 2

(A.38)

A.8. First polarised moment, off-diagonal ( 6= ), symmetrised

For the 4(6) representation we use the operator

1 1
1
1
4(6)
5 D 2 D 2
= (1 + amc0) 5 D + 5 D iac1
O
2 2
4
2
X 1

 
1
(A.39)
5 D , D + 5 D , D .
aic2
8
2
6=,

Repeated or indices are not summed over. The amputated Greens function is:
(6)

4 (p, m; 0, c1 , c2 )
1
1
= i ( 5 p + 5 p ) + c1 ai 5 (p p p p )
2
2
1 X
( 5 p p + 5 p p )
+ c2 ai
2
6=,


g 2 CF
8
1
ln a 2 p2
+ i ( 5 p + 5 p )
2
16 2
3

224

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

2
19.74374 + 3.79201 + 0.88956 cSW 0.49529 cSW


2
+ c1 5.61603 + 4.10778 cSW 0.26315 cSW



2
+ c2 8.29791 + 4.21724 cSW 0.49384 cSW



g 2 CF
p p
1
/ 5
+i 2 p

p
16 2
3
g 2 CF
1
+ ai 5 (p p p p )
2
16 2




1
4
+ cSW ln a 2 p2
3
3

2
4.96628 + 2.43576 + 0.86287cSW + 0.06741cSW

 8

2
ln a 2 p2 34.73952 + 3.62534 0.23227cSW + 0.01777cSW
+ c1
 3

2
5
2
+ c2 ln a 2 p2 + 2.39179 + + 0.14576 cSW 0.00452 cSW
3
3


X
1
g 2 CF  4
1

+
c
( 5 p p + 5 p p )
ln a 2 p2
+ ai
SW
2
16 2 3
3
6=,

2
3.13860 + 2.43576 + 0.28852 cSW 0.00271cSW


2
5
2 2
2
+ c1 ln a p + 1.53155 + 0.30307cSW + 0.01158 cSW
3
3



8
2 2
2
ln a p 32.34394 + 3.62534 0.43312 cSW 0.03719 cSW
+ c2
3
g 2 CF
m 1X
( 5 p p + 5 p p )
[2 + 2]
+i 2
p 2
16 2



7
g 2 CF
1
1
+ cSW ln a 2 p2
+ iam ( 5 p + 5 p )
2
16 2
2
6
2
+ 4.77328 3.93576 1.36644 cSW + 1.05432 cSW


1
7
2 2
2
ln a p 2.30755 2.28022 cSW + 0.07834 cSW
+ c1
3
3


5
7
2 2
2
ln a p + 0.38901 2.29812 cSW + 0.21844 cSW
+ c2
3
3


2
2
g CF
1
p p
/ 5
1 + 2 + cSW (c1 + c2 )
+ ami 2 p
p
16 2
3
3


2C


8 4
g
1m
F

.
(c

c
)
+ ai 2 5 p3 + 5 p3
1
2
2p
3 3
16 2

(A.40)

The contact Greens function is given by

O

C O (p, m) = O (p, m)S(p, m) + S(p, m) (p, m),

where
(6)

4 (p, m)

(A.41)

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

225

1
= i ( 5 p + 5 p )
2


g 2 CF
1
1
2 2
+ i ( 5 p + 5 p )
ln a p 14.27168 + 0.19740 cSW
2
16 2
2
+

g 2 CF 1
m1
(
(1 + ),

p
+

p
)/
p

p2 2
16 2 2

(A.42)

(6)

4 (p, m)
1
= i ( 5 p + 5 p )
2


g 2 CF
1
1
2 2
ln a p 14.27168 + 0.19740 cSW
+ i ( 5 p + 5 p )
2
16 2
2
+

g 2 CF 1
m1
p
/
(
(1 + ).

p
+

p
)

p2 2
16 2 2

(A.43)

In the MS scheme we have

1
MS (p, mMS ) = i ( 5 p + 5 p )
2



p2 31
g 2 CF
8
1

ln

+ i ( 5 p + 5 p )
2
3
9
16 2
2


2
p p
g CF
1
/ 5
+i 2 p

3
p
16 2
g 2 CF
1mX
(

p
p
+

p
p
)
[2 + 2]. (A.44)
+i

2 p2
16 2

A.9. First polarised moment, diagonal, traceless

(3)

For the 4 representation we use the operator

(3)

1 1
4
= (1 + amc0) 5 D 5 D
O
2 2
X 1

 
1
5 D , D 5 D , D .
aic1
8
2

Repeated or indices are not summed over. The operator vertex is:
(3)

4 (p, m; 0, c1 )
1X
1
( 5 p p 5 p p )
= i ( 5 p 5 p ) + c1 ai
2
2



g 2 CF
8
1

ln a 2 p2
+ i ( 5 p 5 p )
2
3
16 2
2
19.92148 + 3.79201 + 0.99934 cSW 0.60077cSW

(A.45)

226

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

2
+ c1 15.31376 + 8.54773 cSW 0.26036 cSW



g 2 CF
1
1 (p2 p2 )
p
/ 5

+i
2
p2
16 2
3
g 2 CF
1X
( 5 p p 5 p p )
+ ai
2
16 2



1
4
+ cSW ln a 2 p2

3
3
2
2.72696 + 2.43576 + 0.35052 cSW 0.127617cSW

+ c1 (1 ) ln a p

2 2

2
30.32684 + 4.29201 0.64293 cSW 0.00597cSW



g 2 CF
1m X
( 5 p p 5 p p )
[2 + 2]
2
2p
16 2



7
g 2 CF  1
1
+

ln a 2 p2
+ ami ( 5 p 5 p )
SW
2
2
6
16 2

+i

2
+ 3.94429 3.93576 1.41190 cSW + 0.90653 cSW


14
2 2
2
ln a p 2.33015 2 4.68331cSW 0.03571cSW
+ c1
3


4
g 2 CF
1
1 (p2 p2 )
c
c
p
/

1
+
2
+
.
+ ami
5
SW
1
2
p2
16 2
3
3

(A.46)

The contact Greens function is given by

O

C O (p, m) = O (p, m)S(p, m) + S(p, m) (p, m),

(A.47)

where
(3)

4 (p, m)
1
= i ( 5 p 5 p )
2


g 2 CF
1
1
2 2

+
0.19740
c
p

14.27168

+ i ( 5 p 5 p )
ln
a
SW
2
16 2
2
+

g 2 CF 1
m1
(
(1 + ),

p
)/
p

p2 2
16 2 2

(A.48)

(3)

4 (p, m)
1
= i ( 5 p 5 p )
2


g 2 CF
1
1
2 2
ln a p 14.27168 + 0.19740 cSW
+ i ( 5 p 5 p )
2
2
16 2
+

g 2 CF 1
m1
p
/
(
(1 + ).

p
)

p2 2
16 2 2

(A.49)

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

227

In the MS scheme we have

MS (p, mMS )
1
= i ( 5 p 5 p )
2

 p2 31 
g 2 CF  8
1

ln 2
+ i ( 5 p 5 p )
2
9
16 2 3



2
2
2
1 (p p )
g CF
1
p
/ 5
+i

2
p2
16 2
3
g 2 CF
1m X
( 5 p p 5 p p )
[2 + 2].
+i 2
2p
16 2

(A.50)

References
[1]
[2]
[3]
[4]
[5]
[6]

[7]
[8]
[9]
[10]
[11]
[12]
[13]

[14]
[15]
[16]
[17]
[18]
[19]
[20]

K. Symanzik, Nucl. Phys. B 226 (1983) 187.

M. Lscher, P. Weisz, Commun. Math. Phys. 97 (1985) 59.
B. Sheikholeslami, R. Wohlert, Nucl. Phys. B 259 (1985) 572.
G. Martinelli, C. Pittori, C.T. Sachrajda, M. Testa, A. Vladikas, Nucl. Phys. B 445 (1995) 81.
M. Gckeler, R. Horsley, H. Oelrich, H. Perlt, D. Petters, P.E.L. Rakow, A. Schfer, G.
Schierholz, A. Schiller, Nucl. Phys. B 544 (1999) 699.
M. Gckeler, R. Horsley, E.-M. Ilgenfritz, H. Perlt, P. Rakow, G. Schierholz, A. Schiller, Phys.
Rev. D 53 (1996) 2317;
M. Gckeler, R. Horsley, E.-M. Ilgenfritz, H. Perlt, P. Rakow, G. Schierholz, A. Schiller, Nucl.
Phys. B Proc. Suppl. 53 (1997) 81.
G. Heatlie, G. Martinelli, C. Pittori, G.C. Rossi, C.T. Sachrajda, Nucl. Phys. B 352 (1991) 266.
E. Gabrielli, G. Martinelli, C. Pittori, G. Heatlie, C.T. Sachrajda, Nucl. Phys. B 362 (1991) 475.
A. Borrelli, C. Pittori, R. Frezzotti, E. Gabrielli, Nucl. Phys. B 409 (1993) 382.
S. Capitani, G. Rossi, Nucl. Phys. B 433 (1995) 351.
G. Beccarini, M. Bianchi, S. Capitani, G. Rossi, Nucl. Phys. B 456 (1995) 271.
M. Lscher, S. Sint, R. Sommer, P. Weisz, Nucl. Phys. B 478 (1996) 365.
M. Gckeler, R. Horsley, E.-M. Ilgenfritz, H. Perlt, P. Rakow, G. Schierholz, A. Schiller, Phys.
Rev. D 54 (1996) 5705;
M. Baake, B. Gemnden, R. Oedingen, J. Math. Phys. 23 (1982) 944.
M. Gckeler, R. Horsley, H. Perlt, P. Rakow, G. Schierholz, A. Schiller, P. Stephenson, Phys.
Lett. B 391 (1997) 388.
M. Gckeler, R. Horsley, H. Perlt, P. Rakow, G. Schierholz, A. Schiller, P. Stephenson, Phys.
Rev. D 57 (1998) 5562.
S. Capitani, M. Gckeler, R. Horsley, H. Perlt, P. Rakow, G. Schierholz, A. Schiller, Nucl.
Phys. B Proc. Suppl. 63 (1998) 874.
A.S. Kronfeld, Phys. Rev. D 58 (1998) 051501.
M. Gckeler, R. Horsley, E.-M. Ilgenfritz, H. Perlt, P. Rakow, G. Schierholz, A. Schiller, Nucl.
Phys. B 472 (1996) 309, and references therein.
M. Gckeler, R. Horsley, E.-M. Ilgenfritz, H. Oelrich, H. Perlt, P. Rakow, G. Schierholz, A.
Schiller, P. Stephenson, Nucl. Phys. B Proc. Suppl. 53 (1997) 896.
S. Sint, P. Weisz, Nucl. Phys. B Proc. Suppl. 63 (1998) 856.

228

S. Capitani et al. / Nuclear Physics B 593 (2001) 183228

[21] S. Capitani, M. Gckeler, R. Horsley, P.E.L. Rakow, G. Schierholz, Phys. Lett. B 468 (1999)
150.
[22] T.-W. Chiu, C.-W. Wang, S.V. Zenkin, Phys. Lett. B 438 (1998) 321.
[23] G. Parisi, in: L. Durand, L.G. Pondran (Eds.), High-Energy Physics 1980, XX Int. Conf.
Madison (1980), American Institute of Physics, New York, 1981.
[24] G.P. Lepage, P.B. Mackenzie, Phys. Rev. D 48 (1993) 2250.
[25] M. Lscher, S. Sint, R. Sommer, P. Weisz, U. Wolff, Nucl. Phys. B 491 (1997) 323.
[26] M. Lscher, S. Sint, R. Sommer, H. Wittig, Nucl. Phys. B 491 (1997) 344.
[27] S. Capitani, M. Gckeler, R. Horsley, H. Oelrich, H. Perlt, D. Pleiter, P.E.L. Rakow, G.
Schierholz, A. Schiller, P. Stephenson, Nucl. Phys. B Proc. Suppl. 63 (1998) 233.
[28] S. Capitani, M. Gckeler, R. Horsley, H. Oelrich, H. Perlt, D. Pleiter, P.E.L. Rakow, G.
Schierholz, A. Schiller, P. Stephenson, Nucl. Phys. B Proc. Suppl. 63 (1998) 871.

Nuclear Physics B 593 (2001) 229242

www.elsevier.nl/locate/npe

Comments on perturbative dynamics of

non-commutative YangMills theory
Adi Armoni
Centre de Physique Thorique de lcole Polytechnique, 91128 Palaiseau cedex, France
Received 29 May 2000; revised 14 August 2000; accepted 29 August 2000

Abstract
We study the U (N) non-commutative YangMills theory at the one-loop approximation. We check
renormalizability and gauge invariance of the model and calculate the one-loop beta function. The
interaction of the SU(N) gauge bosons with the U (1) gauge boson plays an important role in the
consistency check. In particular, the SU(N) theory by itself is not consistent. We also find that the
0 limit of the U (N) theory does not converge to the ordinary SU(N) U (1) commutative
theory, even at the planar limit. Finally, we comment on the UV/IR mixing. 2001 Elsevier Science
B.V. All rights reserved.

1. Introduction and conclusions

Non-commutative gauge field theories lately attracted a lot of attention, mainly due to
the discoveries of their relation to string theory [1]. It was also found that the perturbative
structure of these theories has an interesting pattern. It was shown [2], in the case of
scalar theory, that planar diagrams of the non-commutative theory are the same as planar
diagrams of ordinary commutative theory, up to global phases. For an earlier related work
see Ref. [3]. It was then suggested [4] that non-planar graphs are UV finite, due to the
oscillatory Moyal phase which regulates the integrals. It was found later [5] that these
contributions actually lead to divergences, which were interpreted as infra-red divergences.
These contributions are singular in the 0 limit and they occur also in gauge theories
[6,7].
This paper is devoted to the study of U (N) non-commutative gauge theory. The U (1)
case was already studied by several authors [610]. The renormalization of the model, at
the one-loop approximation, was studied first in [8]. The UV/IR mixing, in the U (1) case
was studied by [6] and [7]. Related works about perturbative dynamics of non-commutative
E-mail address: [email protected] (A. Armoni).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 5 7 - 5

230

A. Armoni / Nuclear Physics B 593 (2001) 229242

field theories are [1123]. Perturbative aspects of non-commutative field theories from
string theory were discussed in [2429].
The non-commutative U (N) YangMills action is
Z
1
(1)
d 4 x tr 2 F ? F ,
2g
where F is
F = A A i(A ? A A ? A ),

(2)

and A is a N N matrix. The ?-product between two functions f and g is defined by

i

f ? g(x) = e 2

() ()

f (x + )g(x + )|,0 .

(3)

The action (1) is invariant under U (N) gauge transformation

A = i(A ? ? A ).

(4)

The gauge transformation (4) is different from the commutative gauge transformation in
the sense that it mixes the U (1) gauge boson with the SU(N) gauge bosons. In fact, the
non-commutative YangMills action (1) also mixes the U (1) and the SU(N)s. It cannot
be written as a sum of a SU(N) and a U (1) theories as the ordinary YM theory, since
there are interaction terms between the SU(N) gluons and the U (1) photon. In order
to demonstrate this point we list in Fig. 1 below the Feynman rules which describe the
3-gluons interaction (the full list of Feynman rules is written in Appendix A).
The complicated structure of the action (1) raises the question of gauge invariance and
consistency of the non-commutative YangMills theory at the quantum level. The action
(1) consists of many interaction terms with a single coupling g due to gauge invariance.
It is not clear, a priori, that the relations among the various couplings in the action is kept
at the quantum level. There are two limits of the theory which hint that the full U (N)

c
Fig. 1. Contributions to the 3 gluons vertex. (a) SU(N)SU(N)SU(N) interaction. (b)
SU(N)SU(N)U (1) interaction. (c) U (1)U (1)U (1) interaction.

A. Armoni / Nuclear Physics B 593 (2001) 229242

231

gauge invariance might be broken. The first limit is 0. In this case, the theory is
expected to reduce to the ordinary commutative theory. However, the commutative theory
has a SU(N) U (1) gauge symmetry and the SU(N) coupling is not related to the U (1)
coupling by gauge symmetry. Moreover, at the quantum level the SU(N) coupling runs
and the U (1) coupling is kept fixed. The second limit, is the planar limit. Since it looks
as if the non-commutative theory and the commutative are identical at the planar level, the
same question about the U (1) coupling should be raised.
As we shall see the U (N) gauge symmetry is not broken quantum mechanically. The
renormalization procedure does not violate the relations between the various couplings (at
least at the one loop level). The resolution of the puzzles mentioned above, is the following:
the limit 0 does not lead to the ordinary commutative theory. Though the resulting
action looks like the ordinary YM action (note that the U (1) and the SU(N) seems to
decouple in the 0 limit, see Fig. 1b), the U (1) and SU(N) couplings have exactly the
same beta function.
The fact that the limit 0 of the U (1) theory is singular was already pointed out in
[7]. It was with relation to the non-planar contributions, which are manifestly singular in .
We claim, however, that the theory is not smooth in even in the planar limit. In this case,
indeed the SU(N) sector theory looks like the commutative theory, but the interaction of
the U (1) with the U (N)s survives the limit. In particular, the U (1) gauge coupling runs.
Thus, the planar sector of the 0 theory does not correspond to the planar sector of the
commutative theory. In this way, the puzzle about U (N) gauge invariance at the quantum
level is also resolved.
The main results of the paper are the following: in Section 2 we calculate the counter
terms which are needed to regulate the divergences in the planar graphs of the SU(N) and
U (1) gluons propagators. We find that they are the same and equal to the ordinary commutative counter term of the SU(N) propagator. The non-planar contributions, however,
are different. There is a non-planar finite contribution to the U (1) propagator [7], but there
is no such contribution for the SU(N). In Section 3 we calculate the counter terms of the
various 3 gluons vertices. Our results in this section are similar to those of Section 2. The
divergent (planar) part of the various 3 gluons vertices is the same, but the finite (nonplanar) part is different. Finally, in Section 4, we calculate the beta function, discuss our
results and more general cases where also matter fields are present.
We shall use the following conventions: capital letters (A, B, C, . . .) denote U (N)
indices, small letters
(a, b, c, . . .) denote SU(N) indices. The U (1) generator is normalized
as follows t 0 = 1/ 2N , such that tr t A t B = AB /2. Finally, [t a , t b ] = if abc t c and
{t a , t b } = ab /N + d abc t c . Thus d abc represents the symmetric tensor for the fundamental
representation. In addition, we shall use the notation p = p /2.

2. Corrections to the gluon propagator

In order to check the renormalizability and gauge invariance at the quantum level, let
us start with the one-loop correction to the gluon propagator. The various contribution are

232

A. Armoni / Nuclear Physics B 593 (2001) 229242

drawn in Fig. 2
We consider first the case where the external legs carry U (1) indices (photons). The
calculation is a straightforward generalization of the calculations which were performed
for the U (1) non-commutative YangMills theory [6,7]. The three contributes are drawn
in Fig. 2. Let us focus on diagram 2a. The only difference in comparison with the U (1)
theory is that now all the U (N) gluons can circulate in the loop. We will use the Feynman
rules 1b and 1c. We denote the external momentum by p and the internal momentum by q.
The resulting expression is
Z
1
d 4 q i i
T1
A =
2 (2)4 q 2 (p + q)2


g (p q) + g (2q + p) + g (q 2p)


(5)
(q p) + g (2q p) + (q + 2p) ,
2g 2 AB AB 2
sin pq.

N
By using the identity
T1 =

(6)

(7)
= (1 cos 2pq),
sin2 pq
2
we can isolate the planar contribution which comes from the 1/2, from the non-planar
contribution (the cosine)
T1 = g 2 N + non-planar term.

(8)

The planar part of the contribution is divergent and its value is exactly the same as the
value of the divergent part of the gluon propagator in ordinary SU(N) YangMills theory.
The same pattern occurs in the other diagrams in Fig. 2. Indeed, after the summation of
the three diagrams in Fig. 2 we find that in order to cancel the divergent part of the U (1)
propagator the following counter term is needed
3(11) =

g2 N
5 2
,
(4)2 3 

(9)

where dimensional regularization was used and  = 4 d. Note that the counter term does
not depend on . As long as is non-zero, a counter term (9) is needed. Otherwise the
U (1) theory is free. Therefore, though the Feynman rules of the theory are smooth in ,
the limit 0 is singular.

Fig. 2. One-loop corrections to the gluon propagator.

A. Armoni / Nuclear Physics B 593 (2001) 229242

233

For completeness let us quote the result for the finite part of the correction to the U (1)
propagator [7]. It is calculated by replacing the cosine of (7) by an exponent and by looking
at the high momentum regime in the integrals of (5) and the two other diagrams in Fig. 2.
The result is [7]
Z

d 4 q 2q q g q 2 i2pq

2 p p
e

g
N
.
(10)
Afinite = 2g N
(2)4
q4
p 4
Note that this term is singular in . When inserted into higher loops it behaves as
ordinary infra-red divergences [5]. Thus, an effect which was originally due to high
momentum turns out to be an IR effect. This is the UV/IR mixing which was found in [5].
Let us turn now to the calculation of the correction to the SU(N) part of the gluon.
Again, let us start with diagram 2a. The coupling of the SU(N) bosons which circulate
in the loop contains the symmetric tensor d abc . The integral is the same as (5), but T1 is
replaced by T2




T2 = g 2 f xya cos pq
+ d xya sin pq
f xyb cos pq
+ d xyb sin pq
.
(11)
By using the SU(N) identities (see Appendix B)
f axy f bxy = N ab ,


4 ab
d axy d bxy = N
.
N
T2 can be written as


4
2
2
.
T2 = g N 1 2 sin pq
N

(12)
(13)

(14)

Interestingly there is another contribution to the gluon propagator which doesnt occur in
ordinary YangMills theory, due to the existence of new vertices (Fig. 1b). It is possible to
exchange a U (1) boson in half of the loop and SU(N) boson in the other half. It contributes
2
sin2 pq,

(15)
N
(the factor 2 in (15) represents two possible exchanges of the U (1)). Collecting the two
terms (14) and (15), we find that the one-loop correction to the SU(N) propagator is
exactly the same as in the commutative case. Thus the divergences can be compensated
by the commutative counter term
T20 = 2 g 2

3(NN) =

g2 N
5 2
.
2
3 
(4)

(16)

Remarkably (9) is identical to (16) (except that (16) in needed also when = 0, in contrast
to (9)). This fact is crucial to ensure gauge invariance of the model at the quantum level, as
we shall see later.
It is interesting to note that the corrections to the SU(N) propagator does not contain a
non-planar finite part. Therefore, though the propagators are identical in their divergent
part, they differ in their finite part. It is due to non-planar graphs which exist in the
corrections to the U (1) propagator but do not exist for the SU(N) one.

234

A. Armoni / Nuclear Physics B 593 (2001) 229242

3. Corrections to the 3-gluons vertex

In this section we calculate the one-loop corrections to the 3-gluons vertex. The relevant
diagrams are listed in Fig. 3 . The external momenta are p1 , p2 , p3 with p1 + p2 + p3 = 0.
We begin with the simplest case in which all external legs are U (1)s. We focus on
diagram 3a. The gluons which circulate in the loop belongs to U (N). Similar calculations
were made in Refs. [6,7].
Z
1
i
i
d 4 q i

=
V1
M
2
(2)4 q 2 (q p2 )2 (q + p1 )2


g 1 (q p1 )2 + g 2 (2p1 + q)1 + g 2 1 (p1 2q)


1 (p2 + q)3 + 13 (2q + p2 ) + g 3 (q 2p2 )1


g3 2 (2q p1 + p2 ) + 2 (p1 + q p3 )3 + 3 (p3 + p2 q)2 ,

V1 = g 3

2
N

(17)

3/2
XY Y Z ZX sin p1 q sin p 2 q sin p 3 (q + p1 ).

(18)

By using the following identity

sin p 1 q sin p 2 q sin p 3 (q + p1 )
1
= cos p3 p1 (sin 2p 1 q + sin 2p 2 q + sin 2p 3 q)
4
1
(19)
sin p 3 p1 (1 cos 2p 1 q cos 2p 2 q + cos 2p 3 q),
4
we can isolate the divergent parts from the finite parts. The divergent part comes from
the (1/4) sin p 3 p1 contribution. Since this part does not depend on q it would lead to a
contribution which is similar to the commutative case. The other diagrams in Fig. 3 are
similar. Thus the needed counter term to cancel the divergent part of the U (1)U (1)U (1)
vertex is
(111)

g2 N
2g 2 N
1 4 2
2 2

.
2
2
(4)
4 3 
(4)
3 

Fig. 3. One-loop corrections to the 3-gluons vertex.

(20)

A. Armoni / Nuclear Physics B 593 (2001) 229242

235

The finite part of the correction, which arise from the (1/4) cos p 3 p1 sin 2p i q terms in
(19), is calculated by replacing the sin by an exponent. The procedure is exactly the same
as in the U (1) case [7]. The result is

Mfinite

d 4q 1
4q q q q 2 q g + q g + q g
(2)4 q 6


ei2p 1 q + ei2p 2 q + ei2p 3 q


p p p
p p p
p p p
(21)
g 3 N cos p 3 p1 1 14 1 + 2 24 2 + 3 34 3 .
p 1
p2
p3
=g

2N cos p 3 p1

The calculation of the correction to the 3-gluons vertex when the external legs are in
SU(N), is a bit more complicated, as there are many contributions in the non-commutative
case. The first contribution is when SU(N) gluons circulate in the triangle of Fig. 3a. We
will use the Feynman rules in Fig. 1a. The calculation of the diagram is performed by
replacing V1 by




V2 = g 3 f axy cos p1 q + d axy sin p1 q f byz cos p2 q + d byz sin p 2 q


(22)
f czx cos p 3 (q + p1 ) + d czx sin p 3 (q + p1 ) .
In order to simplify (22) we use the following SU(N) identities (see Appendix B for
derivation):
f axy f byz f czx f axy d byz d czx d axy f byz d czx d axy d byz f czx


3
= 2N 1 2 f abc ,
N
d axy d byz d czx d axy f byz f czx f axy d byz f czx f axy f byz d czx


3
= 2N 1 2 d abc ,
N
and trigonometric identities similar to (7). Hence, V2 can be written as follows



N
3
3
abc
abc
1 2 + other terms,
cos p3 p1 + d sin p3 p1
V2 = g f
2
N

(23)

(24)

(25)

where other terms means additional contributions which do not lead to divergences.
Apart from the V2 contribution, there is another contribution to the SU(N)SU(N)SU(N)
vertex. It is due to SU(N) bosons flowing in two of the sides of the triangle in Fig. 3a and
a U (1) boson in the third side. The contribution is


2 axy
f
cos p1 q + d axy sin p 1 q
N
xb sin p 2 q yc sin p 3 (q + p1 ).

V20 = 3 g 3

The part that leads to divergences in (26) can be written as follows

N 3
.
V20 = g 3 f abc cos p 3 p1 + d abc sin p 3 p1
2 N2

(26)

(27)

236

A. Armoni / Nuclear Physics B 593 (2001) 229242

Thus, collecting the two contributions V2 and V20 , we find that the counter term which is
needed to cancel the divergences in the 3-gluons vertex with external legs in SU(N) is
(NNN)

g2 N
2 2
,
(4)2 3 

(28)

as in ordinary commutative YangMills theory. Note that the interaction with the U (1)s
was needed to cancel the 1/N 2 terms in (22). Another comment is that the finite
contribution (21) in the U (1)U (1)U (1) cancels in the present case.
We turn now to the renormalization of 3-gluons vertex with one external leg in U (1)
and two external legs in SU(N) (Fig. 1b). The first contribution to the diagram 3a is when
SU(N) bosons circulate in the loop. We should use the Feynman rules 1a and 1b. The
contribution is
r


2
3
f axy cos p1 q + d axy sin p1 q
V3 = g
N


(29)
f byz cos p 2 q + d byz sin p 2 q zx sin p3 (q + p1 ),
which can be simplified (by using (12), (13) and (19)) and rewritten as
r


4 1
2
3
N 2 2
sin p 3 p1 ab + other terms.
V3 = g
N
N 4

(30)

In addition there are two other diagrams which correct the SU(N)SU(N)U (1) vertex.
In one of the diagrams there are two U (1) bosons and one SU(N) bosons which flow in
the triangle (Fig. 3a) and in the other there are two SU(N) bosons and one U (1). The two
diagrams contributes the same. Their contribution is
V30


= 2g

= 2 g 3

2
N
r

3/2
sin p 1 q sin p 2 q sin p 3 (q + p1 ) ab
2 2 1
sin p 3 p1 ab + other terms.
N N4

(31)

The contribution V30 exactly compensate the 1/N 2 part in (30). Hence the needed counter
term is
(NN1)

g2 N
2 2
,
2
(4)
3 

(32)

exactly as (20) and (28).

The other terms in Eqs. (30), (31) leads to finite terms which take exactly the same
form as (21).
The calculation of the 4-gluon vertices is straightforward, though tedious. Adding matter
in the adjoint representation is also straightforward. The counter terms which are needed
in all these cases are exactly the same as the ones which are needed in ordinary SU(N)
theory.

A. Armoni / Nuclear Physics B 593 (2001) 229242

237

4. Renormalizability and gauge invariance

In the previous sections we calculated the counter terms which are needed to renormalize
the theory. Since we are dealing with a gauge theory, gauge symmetry imposes some
constraints on the various counter terms. In ordinary YangMills theory the three-gluons
vertex and the four-gluons vertex are multiplied by g and g 2 , respectively. Gauge
invariance tells us that the two couplings should be the same also at the quantum level.
In the present case the situation is even more involved. A priori, there are two types of
propagators with different wave functions renormalization. There are also three types of
vertices (even four, if we consider the f abc cos and the d abc sin parts of the SU(N) vertex
as two independent vertices).
Gauge invariance imposes the following relations, at one loop
(111)

3 (11)
3 (NN)
(NNN)
3
= 1
3
2
2
1
= 1(NN1) 3(NN) 3(11).
2

(33)

We have found that in fact all 1i are equal and 3i are equal. Clearly, (33) is satisfied.
The calculation of the beta function is also straightforward. The 2/ in the expressions
2
and the beta function is computed by
for should be replaced by log
2


3

1 + 3 .
(34)
(g) = g

2
The result is
(g) =

g 3 11
N,
(4)2 3

(35)

as expected [5]. Note that our result for the U (1) case differs by a factor of 2 from [8] due
to a different definition of the U (1) coupling.
Let us comment about various limits and some special cases. In contrast to the
commutative theory, where the U (N) theory contains two couplings: a U (1) coupling
which doesnt run and an asymptotically free SU(N) coupling, we showed that noncommutative theory can (and as we shall see in a moment must) contain a single
coupling. The theory is asymptotically free and the value of the beta function is
independent of . It is a bit unusual at first sight, since the commutative theory should
be a limit of the non-commutative theory. However, this limit is singular. As long as
is non-zero the U (1) coupling runs, independently of the value of and exactly as the
SU(N) coupling. When is zero, the U (1) is frozen. Thus though the Feynman rules of
the non-commutative theory are smooth in , the renormalization procedure makes the
limit singular.
The planar limit is also interesting. It was suggested [2,4] that the planar limit of the
non-commutative theory is the ordinary theory. However, the planar limit of the U (1)
theory is not the ordinary commutative theory but rather an interacting theory and the
counter term (9) is still needed. Similarly, the planar limit of the U (N) theory is not

238

A. Armoni / Nuclear Physics B 593 (2001) 229242

the planar limit of the SU(N) U (1) ordinary theory. In order to be more precise, let
us give an example which clarifies the difference between the commutative and the noncommutative planar theories. Correlation functions which involve only tr F would yield
trivial answers in the commutative theory, since the U (1) part is decoupled and free. On the
other hand, such correlation functions are highly non-trivial in the planar non-commutative
case.
Another remark is about the SU(N) theory. It was argued in [30] (see also [31,32] and
[24] for a derivation from string theory) that the non-commutative version of this theory
is not consistent, since the closure of the Moyal commutator is violated. Here we find
another evidence for the inconsistency of the SU(N) theory. Had we ignored the U (1)
part of the theory, we would have found that the value of the beta function is gauge
dependent. In order to see that one should calculate the counter terms in a general gauge
and to observe that a 1/N 2 gauge dependent piece is left in the beta function of the SU(N)
theory.
Finally let us comment about the N = 4 theory. Since this theory is finite in the ordinary
commutative case, it seems that for this specific case the limit 0 is smooth. Let us
focus on the planar theory first. Both the U (1) and the SU(N) gauge couplings take their
classical value and no counter terms are needed. Therefore the 0 limit is the same as
the = 0 theory. The non-planar sector of the theory is more subtle. The finite UV effects
are manifestly singular in . It was suggested in [7], that in the specific case of N = 4 these
contributions cancel and thus also this sector of the theory is smooth in .
We would like to note that there is another class of theories which are UV finite [33]
and maybe even smooth in . These are the orbifold truncations of N = 4. These theories
share the same planar diagrams as N = 4 [34]. Therefore, this sector of the theory is finite.
The non-planar sector is anyways UV finite in non-commutative theories. Moreover, since
these theories admit BoseFermi degeneracy, it is likely that the non-planar contributions
cancel as in the N = 4 case.

Acknowledgements
I would like to thank C. Angelantonj, I. Antoniadis, E. Gardi, F. Hassan and R. Minasian
for discussions and comments. This research was supported in part by EEC under TMR
contract ERBFMRX-CT96-0090.

Appendix A. Feynman rules for the non-commutative U (N ) YangMills theory

The non-commutative YangMills action including gauge fixing and ghosts takes the
following form


Z

2
1
(A.1)
S = d 4 x tr F ? F + A c ? D c + D c ? c .
2
We use the Feynmant Hooft gauge = 1.

A. Armoni / Nuclear Physics B 593 (2001) 229242

239

Fig. 4. Feynman rules. Wavy lines and dotted lines denote gluons and ghosts, respectively. Capital
letters and small letters denote U (N) indices and momenta.

Appendix B. SU(N ) identities

In this section we derive SU(N) identities which are used in the paper.
Identity A.

f axy f bxy = N ab .

We denote the adjoint representation by capital letters. We use tr T a T b = N ab . Also

a T b = if axy if byx = f axy f bxy = N ab .
= if axy . Therefore Txy
yx

a
Txy

Identity B.

d axy d bxy = (N 4/N) ab .

tr t x t x t a t b =

N 2 1 1 ab
.
2N 2

By using t a t b = 12 (if abc t c + N1 ab + d abc t c ), (B.1) reads



1
1 xa
x x a b
xay y
xay y b x
if t + + d t t t
tr t t t t = tr
2
N


1 xay
1 1 xa bx
if

=
+ d xay tr t y t b t x +
2
4N

(B.1)

240

A. Armoni / Nuclear Physics B 593 (2001) 229242

1 axy bxy
1 ab
f f
+ d axy d bxy +
.
8
4N
Thus (B.1) and (B.2) leads to


4 ab
axy bxy
axy bxy
+d d
= 2N
,
f f
N
=

(B.2)

(B.3)

and by using identity A, identity B is proven.

Identity C.
f axy f byz f czx f axy d byz d czx d axy f byz d czx d axy d byz f czx


3
= 2N 1 2 f abc ,
N
d axy d byz d czx d axy f byz f czx f axy d byz f czx f axy f byz d czx


3
= 2N 1 2 d abc .
N

(B.4)

(B.5)

We begin with


N 2 1 1 abc
+ d abc .
if
2N 4
Eq. (B.6) can be written also as follows


1
1
if xay t y + xa + d xay t y t b t c t x
tr t x t x t a t b t c = tr
2
N


1 xay
1
tr t b t c t a
=
+ d xay tr t y t b t c t x +
if
2
2N






1
if xay + d xay if ybz + d ybz if zcx + d zcx
=
16


2
if abc + d abc .
+
8N
By Eqs. (B.6) and (B.7) we arrive at






if axy + d axy if byz + d byz if czx + d czx




6
if abc + d abc .
= 2N
N
tr t x t x t a t b t c =

(B.6)

(B.7)

The real and imaginary parts of (B.8) prove identity C.

References
[1] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032.
[2] T. Filk, Divergencies in a field theory on quantum space, Phys. Lett. B 376 (1996) 53.
[3] A. Gonzalez-Arroyo, M. Okawa, A twisted model for large N lattice gauge theory, Phys. Lett.
B 120 (1983) 174;
A. Gonzalez-Arroyo, M. Okawa, The twisted EguchiKawai model: a reduced model for
large N lattice gauge thoery, Phys. Rev. D 27 (1983) 2397.

A. Armoni / Nuclear Physics B 593 (2001) 229242

241

[4] D. Bigatti, L. Susskind, Magnetic fields, branes and noncommutative geometry, hepth/9908056.
[5] S. Minwalla, M. Van Raamsdonk, N. Seiberg, Noncommutative perturbative dynamics, hepth/9912072.
[6] M. Hayakawa, Perturbative analysis on infrared aspects of noncommutative QED on R 4 , Phys.
Lett. B 478 (2000) 394;
M. Hayakawa, Perturbative analysis on infrared and ultraviolet aspects of noncommutative
QED on R 4 , hep-th/9912167.
[7] A. Matusis, L. Susskind, N. Toumbas, The IR/UV connection in the non-commutative gauge
theories, hep-th/0002075.
[8] C.P. Martin, D. Sanchez-Ruiz, The one-loop UV divergent structure of U (1) YangMills theory
on noncommutative R 4 , Phys. Rev. Lett. 83 (1999) 476.
[9] T. Krajewski, R. Wulkenhaar, Perturbative quantum gauge fields on the noncommutative torus,
hep-th/9903187.
[10] M.M. Sheikh-Jabbari, One loop renormalizability of supersymmetric YangMills theories on
noncommutative two-torus, JHEP 9906 (1999) 015.
[11] I. Chepelev, R. Roiban, Renormalization of quantum field theories on noncommutative R d ,
I. scalars, hep-th/9911098.
[12] I.Ya. Arefeva, D.M. Belov, A.S. Koshelev, Two-loop diagrams in noncommutative 44 theory,
Phys. Lett. B 476 (2000) 431.
[13] H. Grosse, T. Krajewski, R. Wulkenhaar, Renormalization of noncommutative YangMills
theories: a simple example, hep-th/0001182.
[14] I.Ya. Arefeva, D.M. Belov, A.S. Koshelev, A note on UV/IR for noncommutative complex
scalar field, hep-th/0001215.
[15] M.V. Raamsdonk, N. Seiberg, Comments on noncommutative perturbative dynamics,
JHEP 0003 (2000) 035.
[16] F. Ardalan, N. Sadooghi, Axial anomaly in non-commutative QED on R 4 , hep-th/0002143.
[17] J.M. Gracia-Bondia, C.P. Martin, Chiral gauge anomalies on noncommutative R 4 , Phys. Lett.
B 479 (2000) 321.
[18] C.-S. Chu, Induced ChernSimons and WZW action in noncommutative spacetime, hepth/0003007.
[19] I.Ya. Arefeva, D.M. Belov, A.S. Koshelev, O.A. Rytchkov, UV/IR mixing for noncommutative
complex scalar field theory, II (interaction with gauge fields), hep-th/0003176.
[20] A. Rajaraman, M. Rozali, Noncommutative gauge theory, divergences and closed strings, hepth/0003227.
[21] K. Furuta, T. Inami, Ultraviolet property of noncommutative WessZuminoWitten model, hepth/0004024.
[22] A.A. Bichl, J.M. Grimstrup, V. Putz, M. Schweda, Perturbative ChernSimons theory on
noncommutative R 3 , hep-th/0004071.
[23] J. Gomis, T. Mehen, Spacetime noncommutative field theories and unitarity, hep-th/0005129.
[24] O. Andreev, H. Dorn, Diagrams of noncommutative 3 theory from string theory, hepth/0003113.
[25] Y. Kiem, S. Lee, UV/IR mixing in noncommutative field theory via open string loops, hepth/0003145.
[26] A. Bilal, C.-S. Chu, R. Russo, String theory and noncommutative field theories at one loop,
hep-th/0003180.
[27] K. Gomis, K. Kleban, T. Mehen, M. Rangamani, S. Shenker, Noncommutative gauge dynamics
from the string worldsheet, hep-th/0003215.
[28] H. Liu, J. Michelson, Stretched strings in noncommutative field theory, hep-th/0004013.
[29] C.-S. Chu, R. Russo, S. Sciuto, Multiloop string amplitudes with B-field and noncommutative
QFT, hep-th/0004183.

242

A. Armoni / Nuclear Physics B 593 (2001) 229242

[30] K. Matsubara, Restrictions on gauge groups in noncommutative gauge theory, hep-th/0003294.

[31] J. Madore, S. Schraml, P. Schupp, J. Wess, Gauge theory on noncommutative spaces, hepth/0001203.
[32] S. Terashima, A note on superfields and noncommutative geometry, hep-th/0002119.
[33] A. Armoni, A note on non-commutative orbifold field theories, JHEP 0003 (2000) 033.
[34] M. Bershadsky, A. Johansen, Large N limit of orbifold field theories, Nucl. Phys. B 536 (1998)
141.

Nuclear Physics B 593 (2001) 243266

www.elsevier.nl/locate/npe

Perfect fluid of p-branes, 2D dilaton gravity and

the big-bang
Javier Borlaf
Instituto de Fisica Terica U.A.M.-C.S.I.C., Universidad Autnoma de Madrid, 28049 Madrid, Spain
Received 21 February 2000; accepted 28 September 2000

Abstract
This paper starts by building the energymomentum tensor of a perfect fluid of p-branes coupled
to (p + 4)-dimensional general relativity. Having three homogeneous and isotropic macroscopical
spatial dimensions, the system gravity/fluid can be reduced to an effective theory over the branes.
For the string fluid (p = 1) the effective theory is nothing but the 2D dilaton gravity where the
potential for the scalar field, which is the scale factor of the macroscopical space, is fixed by the state
equation and the three-dimensional geometry. This theory can be solved allowing us to compare some
relevant aspects in our homogeneous and isotropic string cosmologies with those of the Robertson
Walker ones. In particular, unlike the point-particle models, the existence of an initial singularity is
strongly sensitive to the state equation, and it is remarkable that this model picks out the radiation
state equation as the canonical case where the big-bang is kinematically forbidden. Moreover, we
cannot reduce the RobertsonWalker cosmologies to the limit when the string size approaches to
zero, because the existence of an upper bound on the string size is not compatible with the big-bang.
Some examples are presented. 2001 Elsevier Science B.V. All rights reserved.
1991 MSC: 83E30; 83E15; 83F05
Keywords: p-branes; Perfect fluid; String cosmology; 2D dilaton gravity

1. Introduction
The successful cosmological standard model [1] couples the pure point-particle gravity,
i.e., the four-dimensional general relativity to a very simple approximation to the pointparticle matter filling the universe, i.e., the perfect fluid:
1
R RG = T ,
2
T u u + P (G + u u ).
E-mail address: [email protected] (J. Borlaf).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 0 9 - X

(1)

244

J. Borlaf / Nuclear Physics B 593 (2001) 243266

The energymomentum tensor T describes a continuum of free-falling point-particles

of density (X), or at most, feeling an external force simulating their own interaction
and proportional to the four-dimensional pressure gradient (P (X)). The u = u is
a time-like vector field normalised to u2 = 1 and, by virtue of the energymomentum
covariant conservation T = 0, is tangent to the family of geodesics.
This model is supplemented with the assumption that there are not privileged points and
directions in the space at large scales. This approach reduces the set of allowed geometries
to the family given locally by the line element




dr 2
2
2
2
2
2
2
2
+ r d + sin d
ds = d + a ( )
1 Kr 2
2
.
d 2 + a 2 ( ) dsK

(2)

The particular kind of matter we are working with determines the state equation, that is,
the relation density/pressure, P = P (). In the universe present state we can neglect the
interaction between the continuum of point-particles taking part in the fluid, that is, we can
make P = 0, and the resulting scale factors are
s
1
a(sinh

),
Hyperbolic 3-space (K < 0) a = a(cosh

1), =
|K|
Flat 3-space (K = 0) a = C 2/3 ,

1
a(
sin ).
(3)
K
In the initial state of the universe radiation dominated matter and the time evolution
of the scale factor according to the corresponding state equation P = /3 is given by
a = C 1/2 in the flat case.
Our main objective in this paper is to broach the p-brane cosmology following the same
pattern as in the standard point particle model. Two basic ideas guide this approach; on
the one hand, we want to describe the p-brane matter content of the universe by means
of its corresponding perfect fluid, which we assume is the simplest right-hand side we
can place in the cosmological equations. On the other hand, we keep homogeneity and
isotropy in the three macroscopical space-like dimensions, assuming that p-branes taking
part in the fluid wrap around a compact (and small) space-like manifold. Therefore, in
Section 2 our first task will be to build the energymomentum tensor of a perfect fluid of
p-branes. A p-brane is a (p + 1)-dimensional hypersurface embedded in the spacetime
M and having p space-like dimensions. We can describe it explicitly through a parameter
space called worldvolume {a }, a = 1 to p + 1, and the map {X (a )} of such space
onto the ordinary spacetime M. Of course, {X }, with = 1 to D, are coordinates in the
manifold M. The coupling to the geometry in M, ds 2 = G (X) dX dX , follows the
same pattern as in the point-particle case: the time evolution of the p-brane is an stationary
point of the action
Z

(4)
SX = d p+1 .
Spherical 3-space (K > 0) a = a(1
cos ),

J. Borlaf / Nuclear Physics B 593 (2001) 243266

245

Given that det ab and ab = G (X)a X b X is the metric induced over the
brane, this action is nothing but the hypervolume swept by the extended object in its time
evolution.
The equations derived from the action (4) are the simplest ones satisfying the property
of being independent of the choice of parameters in the worldvolume 1 and coordinates in
the spacetime. 2
Therefore the minimal action principle provides the p-brane equations 3
SX

= U
1X = 0,
X
where the target-space diff. invariant object 1X is defined 4

1X = ab a b X +
a X b X

(5)

(6)

and the orthogonal projector U

G U ,

U = ab a X b X

(7)

guarantees the invariance of (5) under worldvolume reparametrizations. As an example, the

projector in the point particle case is U = X X /X 2 and therefore U 1X =

d
log X 2 = 0 are the reparametrization-invariant geodesic
X + X X X d
equations.
After building the energymomentum tensor for a continuum of such a p-branes, that
we can interpret as a perfect fluid of them, in Section 3 we couple that matter content to
the (p + 4)-dimensional general relativity, restricting ourselves to the case where we have
four macroscopical dimensions (one of them is time-like and the other three are spacelike homogeneous and isotropic) and p space-like small dimensions where the p-branes
wrap around. The effective theory describing the geometry over the branes and the scale
factor of the macroscopical space is obtained. We also study in detail the relation between
the state equation and the scale factor implied by the energymomentum conservation and
the field equations. The effective theory for the string cosmology, which is nothing but the
2D dilaton gravity, is solved in Section 4, where we get the general features emphasizing
the very special connection among the big-bang, the state equation and the string size. We
conclude that the big-bang is not natural to this model. In Section 5 we list the pressure-free
string cosmological solutions allowing a direct comparison with their RobertsonWalker
counterparts. And finally we jump into the conclusions.
1 The action and the equations are invariant under the transformation a 0a ( b ).
2 The action and the equations are invariant under the transformation X X0 (X ) together with the

corresponding tensorial transformation in the background fields.

3 S/X (L/ X ) L/X .
a
a
4 Indices are lowered (raised) with the metric components G

(and its inverse G ). Moreover, the covariant

derivation = dX is always built from the Levi-Civita connection.

246

J. Borlaf / Nuclear Physics B 593 (2001) 243266

2. The p-brane perfect fluid energymomentum tensor

As mentioned in the introduction, the simplest approximation we can pose to represent
the matter filling the universe is the perfect fluid of point-particles, whose energy
momentum tensor results to be as in Eq. (1). To simplify this introductory comment, let us
pay attention to the pressure-free case, that is, P = 0. Covariant conservation 5


T = u u + u ( u + u log ) = 0
(8)
means, on the one hand, that the velocity field u = u (X) is a geodesic field. 6 In this
way, the point-particles taking part in the fluid are free falling, or, in other words, they
do not interact between themselves as we expect from P = 0. On the other hand, the
component tangent to u in Eq. (8), gives the rate of change of the matter density in the
geodesic direction u log = u.
We know that the coupling between the energymomentum tensor and general relativity
in four dimensions accounts for very relevant features of the universe when observed at
large scales. Therefore, our goal in this section is to build the counterpart to (1) for a perfect
fluid of p-branes, having in mind their cosmological utility when coupled to D = (p + 4)dimensional general relativity. In this paper we will solve the string case (p = 1) leaving
the p > 1 ones to a future work.
2.1. Pressure-free fluids and invariance under field-redefinitions
If the universe is filled with a continuum of (p + 1)-dimensional hypersurfaces, the

Frobenius theorem guarantees that any complete set of vector fields {ua = ua (X) }
(a, b, c = 1, . . . , p + 1) tangent to them is closed under commutation. 7
c
[ua , ub ] = Cab
uc .

(9)

The condition (9) is invariant under local basis changes

ua (X) Mab (X)ub (X),

(10)
g
Mad Mbe Cde (M 1 )cg

provided that the set of scalar fields transforms

d
e
1
c
Mb ud Ma )(M )e .
Condition (9) allows us to pick a basis {ua } and a coordinate frame {Xa , Xi } in such
a way that ua = a . 8 It is obvious that in these coordinates each (p + 1)-dimensional
hypersurface is spanned by the subset of coordinates {Xa } and labelled by the transverse
ones {Xi }. We call this choice the commoving gauge.
The point is that the closure relation (9) does not ensure that we have foliated the
spacetime into a family of p-branes. We need an additional condition over the basis {ua }
c
Cab

U ab ua ub = 0,
5 The conventions are: a and a a .

6 Therefore the integral curve X

= u(X) is a geodesic one due to u u = 0.
7 [A, B] B A
A
B
8 Therefore C c = 0.
ab

+ (Mad ud Mbe

(11)

J. Borlaf / Nuclear Physics B 593 (2001) 243266

247

where

ab = G u
a ub ;

ac cb = ab ,

ab
U = u
a ub ,

U = G U .

(12)

are the components of the projector over the branes (U 2 = U and U ua = ua ), and U
is the orthogonal projector. Of course the equation (11) is invariant under basis changes in
the tangent space (10). In the commoving gauge we can check that the hypersurfaces {Xi =
Constanti } are p-branes. In other words, the equations (11) imply the p-brane equations
(5) for those hypersurfaces.
The natural candidate for the pressure-free perfect-fluid energymomentum tensor must
be proportional to the projector over the branes, i.e., T U with the density of
the matter in the fluid. Using condition (11), the covariant divergence results to be tangent
to the branes


(13)
U = ab u
a (ub log + Jb ),

where we define the set of p +1 scalars {Ja ua U }. In fact, the orthogonal component

of (U ) is nothing but U ab ua ub = 0 due to Eq. (11). Therefore, to demand

energymomentum tensor conservation equals to
Ja = ua log .

(14)

cJ .
But the above equation needs the integrability condition ua Jb ub Ja = Cab
c
Fortunately, we can derive this integrability condition by means of the basic Frobenius
equation (9). In this way, we have showed how the requirement of a spacetime foliation
through p-branes implies that the symmetric tensor proportional to the projector over the
those branes

T U

(15)

is covariantly conserved provided that the matter density scalar satisfies the integrable
equations
ua (ln ) = Ja T = 0.

(16)

In the commoving gauge, the energymomentum tensor has got only non-vanishing
components in the brane directions 9
T ab = ab ;

T ai = T ij = 0,

(17)

as we expect in the pressure-free case. Moreover, in this coordinates Ja = a

log G/ and we can integrate the density as

r

a
i
i
,
(18)
X , X = (X )
G
9 In the gauge = G . In this gauge, the components of the induced metric coincide with the matrix
ab
ab
ab
elements ab .

248

J. Borlaf / Nuclear Physics B 593 (2001) 243266

where det ab , G det G and (Xi ) > 0 is an arbitrary function depending on the
transverse coordinates.
As a simple check, fixing p = 0 we get = u2 and therefore U = u u /u2 . Choosing
the (1 1) matrix (10) in such a way that the time-like vector field is normalized u2 =
1 we recover the old expression T = u u for the pressure-free perfect fluid of pointparticles. For strings, i.e., p = 1, we agree with Leteliers description of a cloud of strings
[2] provided that we identify his elementary object, the bivector , with our composite

one  ab ua ub . 10
It was shown in Ref. [3] that the simplest version of field theories invariant under fieldredefinitions provides an explicit realization of the perfect fluid of p-branes. Specifically,
if we have a set p(=
D p 1) scalar fields { i (X )}, the simplest action which is
invariant under field redefinitions i (X) 0 i ( j (X)) and coupled to the geometry on
the manifold M is
Z

(19)
S [F ] = d D XF () G h,
M

where h det(hij ), hij (X) = G(X) i (X) j (X) and F () F ( 1 , . . . , p ) is

an arbitrary function on the scalar fields 11 transforming under redefinitions as F 0 ( 0 )
| det(/ 0 )|F (( 0 )). The field equations do not depend on the F () and they imply
that the hypersurfaces { i (X) = Constanti } are p-branes. Therefore any classical solution
i (X)} foliates the spacetime into a continuum of p-branes. Finally, the energy
{cl
momentum tensor results to be proportional to the projector over the branes, and it can
be written in terms of the fields (U = G hij i (X) j (X)):

1 S [F ] 1
= F () h U .
(20)
T [F ]
2
G G

The density = 12 F () h coincides in commoving coordinates with the one obtained

above when the identification 12 F (i ) (i ) is done.
2.2. Non-vanishing pressure in the perfect fluid of p-branes
In cosmological models the pressure introduces an external force simulating the interaction between the constituents of the fluid. Given that the pressure affects the transverse
components in the energymomentum tensor, its contribution must be accompanied with
the orthogonal projector U . Therefore, it is very natural to postulate the following form
for the energymomentum tensor of a p-brane perfect fluid:

.
(21)
T = U + P U
2
The whole covariant divergence reads




P

ab

ab ub ua () + ( + P )Ja . (22)
T = (P + )U ua ub
+P
10  ab is totally antisymmetric in two dimensions and  1 2 = 1.
11 But not in the derivatives of the fields.

J. Borlaf / Nuclear Physics B 593 (2001) 243266

The transverse component of T = 0 modifies the condition (11) to



P

ab

=0
U ua ub
+P

249

(23)

just introducing an external force which is proportional to the pressure gradient. In other
words, in the comoving gauge we can check that Eq. (23), derived from the energy
momentum conservation, implies the p-brane equations for the hypersurfaces {Xi =
Constanti } in presence of the external force F = P /( + P ). 12 The remaining equation
is just
ua + ( + P )Ja = 0

(24)

which is again integrable by virtue of Frobenius equation. Again p = 0 reduces to the usual
point-particle energymomentum tensor (1).
As an example, let us take the kind of geometry we will work with. Naming the set
of coordinates spanning the p-brane Xk {Xa }, and X {Xi } the transverse ones, the
metric we use for cosmological purposes is:
k

2
,
ds 2 = dsk2 + e(X ) ds
2 are defined as
where the metric over the branes dsk2 and the transverse ds


dsk2 Gab Xk dXa dXb ,


2
Gij X dXi dXj .
ds

(25)

(26)

To that metric (25), the hypersurfaces spanned by the basis {ua = a } are p-branes, it
2 they are. Moreover, if we impose
does not matter what (Xk ) and the metrics dsk2 and ds
over (21) the conservation condition we get the following equations



p
a + ( + P )a = 0,
Ta =
2
2

(27)
Ti = i P = 0,
2
where p = D p 1 is the dimension of the transverse space. Eqs. (27) can be integrated






p
k
Xk , X = k Xk + X e 2 (X ) ,




p
2
k
(28)
P = P Xk = k 0 Xk e 2 (X ) ,
p
where k () and (X ) are arbitrary functions, and we define k 0 () d k ()/d.
3. The p-brane perfect fluid in the (p + 4)-dimensional general relativity
Our point of view assumes that the universe consists of a perfect fluid of p-branes.
Then, we question whether we can get the most relevant aspects of the RobertsonWalker
12 Then, the p-brane equations are U (X F ) = 0.

250

J. Borlaf / Nuclear Physics B 593 (2001) 243266

cosmologies as the p-brane cosmologies when the brane size vanishes compared to the
universe size, or there are qualitative differences in the effective evolution of our four
macroscopical dimensions. Therefore, once the energymomentum tensor for a perfect
fluid of p-branes is built (21), we couple that tensor to the corresponding pure gravity in
p + 4 dimensions. But, what is pure gravity when the fundamental constituents of matter
are p-branes? In the point-particle case p = 0 we agree with the idea that general relativity
would be the privileged choice and the resulting cosmology would be the old Robertson
Walker one. In the string cosmology (p = 1) it would be natural to choose the effective
string theory required by conformal invariance or the perturbative calculation of string
amplitudes. For higher branes (p > 1) we could use the actions postulated in Ref. [4] in
basis of the S-duality between the different branes. However, the construction made in
the previous section carries an implicit choice: the (p + 4)-dimensional general relativity.
In this way, although we deliberately set aside other non-massive contributions (mainly
dilaton and (p + 1)-form), that we get it is the simplest cosmological model the one we
can postulate where the cosmic fluid consists of a continuum of extended objects instead
of point-particles.
Now we can justify the metric choice (25). Given that we want a direct comparison with
the RobertsonWalker cosmologies, we assume that our spacetime is, locally, the direct
product of a three-dimensional (p = 3) homogeneous and isotropic space, a non-compact
time-like dimension, and a small (compact) space-like manifold where the p-branes
wrap around. In this way, the isotropic observers are p-branes and the three-dimensional
2 is labelled by a single real number, i.e., the constant curvature scalar R =
metric ds
6K. The transverse space is nothing but the ordinary three-dimensional space. Now,
k
e(X )/2 a(Xk ) is the scale factor measuring the size of our macroscopical space and
now depends on the proper time as the same time as the spatial brane coordinates. 13
Finally, the metric over the branes is dsk2 .
The coupling gravity/fluid is described by means of the equations: 14


1

R RG = U P U
.
2
2
With our metric (25), the relations

=0
Gai = Rai = Uai = Uai

(29)

(30)

identically hold in the commoving gauge. As expected, the transverse equations in

(29) imply the homogeneity and isotropy of the three-dimensional macroscopical space.
Moreover they provide another equation for the scalar field over the branes (Xk ).
k
Designing k , 1k = ( k )2 , Rab and R k the covariant derivatives and laplacian, the Ricci
13 For strings, i.e., p = 1, our physical point of view differs from those appearing in Refs. [2]. There, the
string fluid is coupled to four-dimensional G.R. Here, we incorporate the fact that we do not see strings in our
four-dimensional macroscopical world, and we confine them to be wrapped around an additional small fifth
dimension, keeping the basic assumptions of homogeneity and isotropy.
14 We follow the Weinbergs [1] conventions with R = + and, as

usual, R = R is the Ricci tensor and R = R is the curvature scalar.

J. Borlaf / Nuclear Physics B 593 (2001) 243266

251

tensor and the curvature scalar built from the brane metric dsk2 , and using the usual
conventions to low and raise the indices with the metric components Gab and its inverse
2 , we get:
Gab , and repeating with Rij and R and the transverse metric ds
1
Gij R ,
3
= 6K = constant

Rij =
R

(31)

and the equation



3
3 k
2
2 k 0
k
k
()e 2 2Ke = 0.

R +2 1 +
4
3

(32)

The brane-components in Eq. (29) imply the following conditions over the brane geometry

2


1
1
k
Rab R k Gab = Gab 31k + 3 k + eff ()
2
2 

1
3
k
b + a b ,

2 a
2

(33)

where the effective potential is defined

3

eff () 6Ke + k ()e 2 .

(34)

Moreover, the transverse contribution in the density (28) must vanish

(X ) = 0.

(35)

We can observe that the field equations for the scale factor (32) and the brane
geometry dsk2 (33) derive from the following effective action.
Z
p

2


3
(36)
Seff = d p+1 Xk Gk e 2 R k 32 k + eff () .
In particular, the invariance of this action under brane diffeomorphismes Xk Xk 0 (Xk )
implies the identity


1 k
Seff ab
Seff
k Seff
G +3
=
(37)
a
Gab 2 a

Gab
and therefore, the equation for the scale factor (32) can be derived from the metrics
equations dsk2 .
3.1. The state equation
In our equations for the perfect fluid of p-branes coupled to the (p + 4)-dimensional
general relativity, the unique freedom we have is the arbitrary function k (). Firstly, we
redefine the scale factor in the usual way
a 2 e

(38)

and the arbitrary function as k (2 ln a) 0 (a). Given that the transverse contribution
vanishes (35), the energymomentum tensor conservation (28) dictates

252

J. Borlaf / Nuclear Physics B 593 (2001) 243266

= (a) = 0 (a)a 3 ,
1
P = P (a) = 00 (a)a 2
3

(39)

with 00 (a) d0 /da. Of course, the minimal conditions we must impose on 0 (a) to have
> 0 and P > 0 are:
0 (a) > 0,

00 (a) 6 0.

(40)

In short, the 0 (a) choice determines the state equation for the fluid. For example, the
simplest choice 0 (a) = constant = 0 is equivalent to the pressure-free fluid (P = 0)
fluid. Another significant example, 0 (a) = 0 /a corresponds with the radiations state
equation P = /3. We can interpolate continuously both cases through the choice 0 (a) =
0 /a q (q > 0), getting the state equation P = q/3. It is important to note that the state
equation P = 0 corresponds to = 0 a 3 whatever the p-brane it is (particles, strings,
membranes or higher-dimensional branes). The same feature occurs for any state equation:
the density/scale factor relation is independent of brane dimensions. Thus, the radiation
state equation always corresponds with the expression = 0 a 4 accounting for the red
shift due to the expansion of the three-dimensional space. What depends on p is the relation
scale-factor/proper time a( ). From now, we will assume the minimal conditions (40) on
0 (a). It will be enough for getting a lot of information in the string (p = 1) case.
Conversely, given the state equation P = P () we can integrate the relation (39)
Z2
1

a2
d
= 3 ln
+ P ()
a1

(41)

to obtain the explicit dependence of the matter density on the scale factor. Again, this
relation is macroscopical because it does not depend on internal dimensions.

4. String cosmology in the 5-dimensional general relativity

In this work our main interest is focussed in the string cosmology. Firstly, we can relate
the equations for the scale factor and the geometry induced over the strings with the 2D
dilaton gravity. To this end, we simply redefine the scale factor and the induced geometry
in the way
2
e2/3 d sk2 +
e2/3 ds
,
ds 2 =

e = e2,

(42)

eab (Xk )dXa dXb . In these new variables we immediately recognize the
where now d s2 G
action for the 2D dilaton gravity
Z
p


e
e()
e
eR
ek V
(43)
Seff = d 2 Xk G
e as
where we define the potential V

e()
e
e1/3 .
e 6K
e1/3 ()
V

(44)

J. Borlaf / Nuclear Physics B 593 (2001) 243266

253

Therefore, the perfect fluid of strings provides an explicit realization of the dilatonic gravity
for every choice of the potential, provided that we choose properly the state equation. 15
The point is that, as it is well known, the 2D dilaton gravity is an integrable field theory
[5]. This fact will allow us to get many information about the general features in the 5D
string cosmology within our model. We start deriving the equations from the (43) effective
action:
e
e(),
e
e = Gab V
eak b
(45)

2
e0 (),
ek = V
e
(46)
R
e dV
e()/d
e .
e The key point to find the explicit
e0 ()
where a, b = , and we define V
solutions is to realize the existence of a killing vector k = k a a implied by the equation
(45): 16
e
k a = t0  ab b ,

(47)

t0 is a constant whose physical meaning will be clear later. In this context, our point of view
consider the time ( ) evolution of a three-dimensional homogeneous and isotropic space of
size a( ), together with a small compact dimension of size astring as ( ) where the string
wraps around. Therefore the isotropic observers are small strings, i.e., they have a finite
size as . Our purpose is to investigate the effect of this size in the scale factor a( ) which
measures the size of the three macroscopical dimensions, and then to compare with the
RobertsonWalker point-particle models listed in the introduction and assumed to contain
most of the basic features of the large scale structure of our universe.
With this in mind, the five-dimensional metric takes the form
2
,
ds 2 = d 2 + as2 ( ) d 2 + a( )2 ds

dr 2
2
2
2
2
+
r
+
sin

d
d
.
(48)
1 Kr 2
Of course, the above scenario corresponds with the choice of an space-like killing k =
and therefore the formula (47) gives a = 0 a = a( ) a time-dependent scale factor.
Moreover, the killing provides a nice relation between the relative size of the string and the
Hubble constant H ( ) = a/a
17
as
= 2t0 |H |.
s
(49)
a
2
2
dsK
=
ds

In other words, the relative string size is proportional to the Hubble constant. As a direct
consequence, the string size vanish in the rebound points (a = 0) of our three-dimensional
space. The formula (49) is independent on the state equation and therefore must be
15 It must be noted that this is not the first time that the 2D dilaton gravity is related with the string cosmological
scenario, although in a very different context. In Refs. [6] particular versions of the two-dimensional quantum
model are used to address the graceful exit problem.
p
16  ab is totally antisymmetric in two dimensions and  1/ G.
e
17 d/d .

254

J. Borlaf / Nuclear Physics B 593 (2001) 243266

considered as a kinematical relation. Together with this kinematic relation, Eqs. (45) and
(46) can be integrated providing the time evolution of our effective universe measured
through the scale factor a( )
Za2
a1

da
= (2 1 ),

Vs (a)

1
Vstring(a) Vs (a) = 2 E 3Ka 2 +
3a

Za

!
0

0 (a )da ,

(50)

where E is the real integration constant in the model, as stated in the two-dimensional
version of Birkhoffs theorem [7].
4.1. Bounded versus unbounded universe
4.1.1. Unbounded
Given that a 2 = Vs (a), the allowed scale factors are constrained to satisfy Vs (a) > 0.
Then, the bounce points (a = 0) correspond with zeros in the string potential. In the
hyperbolic geometries it can exit at most a single bounce point. 18 Moreover, if present,
that bounce point, a , is a minimum in the scale factor. 19 In that case, the solution extends
over the range a [a , ). On the other hand, if there is no a zero in the string potential
the solution extends over the whole range a [0, ). We conclude that the evolution of
the universe size is not bounded in the hyperbolic geometries. From Eq. (50) we get the
asymptotic behavior
p
(51)
K < 0: a( ) = |K| H unbounded solution
which, besides the independence on the state equation, coincides with the asymptotic value
in the point-particle RobertsonWalker cosmologies.
The same argument applies when the three-dimensional space is flat. The only difference
is that the asymptotic behavior is strongly sensitive to the state equation. However, using
the minimal properties on 0 (a), we can show that the expansion is not bounded.
K = 0: a( ) = H unbounded solution.

(52)

4.1.2. Bounded
In the spherical geometries, Vs (a = ) = holds due to K > 0 and the physical
conditions (40) on the state equation. Therefore we always have a bounce point which is
a finite maximum, a+ , in the universe size. Moreover, we can have at most another one
bounce point. 20 Using Vs (a+ ) = 0 we can rewrite the string potential as
18 Because d(3a 2 V 2 )/da = 6Ka + (a) is always positive, a 2 V (a) (and hence V ) can vanish just once
s
s
0
s
at most.
19 a 2 V (a) increases and V () = . Therefore V (a < a ) < 0.
s
s
s

20 The second derivative d 2 (3a 2 V 2 )/da 2 = 6K + 0 (a) is negative, implying that a 2 V (a), and hence V ,
s
s
s
0
can have at most two zeros.

J. Borlaf / Nuclear Physics B 593 (2001) 243266

1
2
Vs (a) = 2 3K a+
a2
3a

!
Za+
0 (a 0 )da 0 .

255

(53)

Expanding around the maximum size a = a+ a with a > 0





2
2 6K
0
+ 0 (a+ ) +
3a Vs (a) = a 6Ka+ a+ 0 (a+ ) a
a+

(54)

we see the term (a)2 is negative. Given that Vs (a < a+ ) must be positive for some
neighborhood, we get the following inequality
6Ka+ 0 (a+ ) > 0

(55)

to be satisfied for the maximum size in order to have a solution in the hyperbolic geometry.
Having in mind that the maximum size in the perfect fluid of particles 21 must satisfy
pp
pp
6Ka+ 0 (a+ ) = 0, we can deduce 22
strings

a+ = amaximum a+

point-particle

> a+

(56)

that is, the maximum scale factor in the perfect fluid of strings is always greater than the
maximum scale factor in the perfect fluid of particles. Moreover, the maximum is reached
in a finite proper time that we can limit to
Za+
= (a+ ) (a) =
a

2 3a+
da
(a+ a)1/2
6q

pp
Vs (a)
6Ka+ 0 (a+ )

= finite,

(57)
pp

where we have restricted the inequality to the interval a+ < a 6 a+ .

In short, just like in the point-particle fluid, it is the three-dimensional geometry which
determines whether the expansion of the universe is bounded or not:
flat and hyperbolic geometries unbounded expansion,
spherical geometry bounded expansion.

(58)

4.2. The initial singularity versus the state equation

4.2.1. Non-singular cosmologies
When the state equation satisfies the minimal physical conditions, the zero size (a = 0)
is always allowed in the perfect fluid of particles. 23 Moreover, also conditions (40) ensure
that zero size is reached in a finite proper time. Therefore the singularity (1/a = = P =
) is a generic feature in the cosmological perfect fluid of particles.
21 In the same three-dimensional geometry and having the same state equation.
22 Using 0 (a) 6 0.
0
23 Conditions (40) guarantee that lim
a0 Vpp (a) > 0.

256

J. Borlaf / Nuclear Physics B 593 (2001) 243266

However, this is not the case in the perfect-fluid string cosmology. If

Za
lim0+ 0 (a 0 ) da 0 =


holds, 24
Za

we must carefully rewrite in the string potential (50),

Za
0
0
0 (a ) da 0 (a 0 ) da 0
a

with a > 0

arbitrary. 25
Za

When we approach zero size the integral goes to

0

Za

0 (a ) da = lima0

lima0
a

0 (a 0 ) da 0 =

dominating over remaining terms in the string potential and resulting in lima0 Vs (a) =
. We conclude that a = 0 cannot be kinematically reached and we have not the initial
singularity.
Za
(59)
lim0+ 0 (a 0 ) da 0 = a > aminimum a > 0 no singularity.


Therefore we have a lower bound a > a > 0 in the scale factor. Of course, it satisfies
Vs (a ) = 0 allowing us to rewrite the string potential as:
!
Za


1
2
2
0
0
(60)
Vs (a) = 2 3K a a + 0 (a ) da .
3a
a

Expanding around the minimum size, a = a + a with a > 0, we have





6K
+ 00 (a ) + .
3a 2 Vs (a) = a 6Ka + 0 (a ) + a 2
a

(61)

When we consider the spherical geometry, the term (a)2 is negative and, given that Vs (a >
a ) > 0 for some neighborhood, we can get the following inequality for the minimum size:
6Ka 0 (a ) < 0.

(62)

Having in mind again that the maximum scale factor in the point-particle fluid is just
pp
pp
given by the equation 6Ka+ 0 (a+ ) = 0, we deduce that the minimum scale
factor in the spherical (K > 0) fluid of strings is always minor than the maximum size
in the fluid of particles, when both share the same state equation and three-dimensional
geometry.
strings
point-particle
< a+
when K > 0.
(63)
a = aminimum a
24 Given that (a) is a positive definite and decreasing function, its only divergence could be placed at a = 0.
0
For example, 0 (a) = 0 /a, i.e., the radiation state equation.
25 That arbitrariness is reabsorbed in the integration constant E.

J. Borlaf / Nuclear Physics B 593 (2001) 243266

257

The previous considerations allow us to state that the cosmologies where the singularity
is forbidden by the state equation can be just labelled by the minimum size a , 26 which
pp
is bounded to be a+ > a > 0 in the spherical geometries, and to be a > 0 in the
hyperbolic and flat ones. Moreover, the minimum size is reached in a finite proper time
that we can limit to
pp
Za
2 3a+
da
6q

(a a )1/2
1 = (a) (a ) =
pp
Vs (a)
(a
) 6Ka
0

= finite
pp

in the spherical case (and taking a+

have also the inequality
Za
1 = (a) (a ) =
a

(64)
> a > a ). In the hyperbolic and flat geometries we

2 3
da

6
(a a )1/2 = finite,
Vs (a)
0 (a)

(65)

where now it is enough that a > 0. The canonical example of non-singular string
cosmology is just the one governed by the radiation state equation P = /3, or, in other
words, 0 (a) = 0 /a. Now, writing the string potential as,




1
a
2
a 2 + 0 ln
,
(66)
Vs (a) = 2 3K a
a
3a

it is easy to see that a = Vs is not defined for a = 0 and therefore the singular point a = 0
is kinematically excluded. In fact, in the hyperboloid (K < 0) the moduli a is unrestricted
and scales a < a are manifestly forbidden. In the sphere (K > 0) the minimal size is

restricted by inequality (62) to be a < 0 /6K. Then, a < a are again kinematically
forbidden.
The radiation state equation is not just the simplest non-singular example, but can be
considered the limiting state equation from which the perfect fluid string cosmologies are
non-singular. Indeed, if we use Eq. (41) we realize that
 4
Za

a1

>
H lim0 0 (a) = non-singular cosmology.
P > H
3
1
a


(67)
Of course, to conclude that the cosmology avoids the big-bang we just need the
inequality (67) holding in a 1/ = 0 neighborhood. In that way, is the crucial balance
in the universe between the gravitatorial effect of the matter and the expansive effect of the
pressure which determines whether or not the collapse of the universe in the singularity
is possible. If the pressure becomes equal or greater than /3 the universe bounces when
reaching a certain minimal size a .
26 This is the parameter of the 2D Birkhoffs theorem.

258

J. Borlaf / Nuclear Physics B 593 (2001) 243266

But the troubles for the singularity do not finish at that point. If
Za
lim

0+

0 (a 0 ) da 0 <

and we choose
Za
Za
0
0
0 (a ) da 0 (a 0 ) da 0,
0

when E < 0, Vs (0) = E < 0 and again a = 0 is forbidden.

Za
lim0+

0 (a 0 ) da 0 < and E < 0 no singularity.

(68)

Once again, bounds (62) and (63) in a and the one (64) in the proper time needed to
reach the minimum, still hold in the spherical geometry. On the other hand, in the flat
and hyperbolic cases, the bound (65) also allows to reach the minimal size in a finite
proper time. As we see, in all non-singular solutions the minimal size is reached in a
finite proper time which allow us to extend the solution backwards in time, that is, from
= 0 to = , and therefore the universe has not a beginning. We must note that,
due to the kinematical relation (49), the string size as vanishes in the bounce points, and
specifically when a = a . This makes singular the equations (45) and (46) given that the
five-dimensional metrics determinant vanishes too. But this is not a physical singularity
because the physical quantities (density and pressure) acquire finite values in the minimum.
In the non-singular and bounded solutions, the parameters a are not independent, but
they are forced by conditions Vs (a+ ) = Vs (a ) = 0 to satisfy the equation:
3K

2
a+

2
a

Za+
=

da 0 (a).

(69)

Using the properties on the string potential derived from the minimal properties on the
state equation, it can be shown that the single static solution of this model is contained
pp
in the spherical case, as the particular moduli a+ = a = a+ . Flat and hyperbolic
geometries do not possess static solutions.
4.2.2. Singular cosmologies
In the preceding section we have assumed that the models for which a = 0 is reached
are singular. To prove this, we take the trace in the basic equations (29) to get the 5D
curvature scalar: 27

(70)
R = (3P 2).
3
27 U = 2 and therefore U = U = 5 2 = 3.

J. Borlaf / Nuclear Physics B 593 (2001) 243266

259

Given that we must restrict ourselves to > 3P (67) in order to allow a = 0, we can
bound the curvature scalar to:

(71)
R = 3P + 2( 3P ) < P .
3
Obviously, on physical grounds we expect the pressure divergence when the scale factor
a goes to zero and, in view of the above inequality, the divergence of the curvature scalar
itself. Even in the pathological cases where the pressure keeps bounded in the a = 0
limit, 28 the minimal conditions (40) over the state equation guarantee the divergence of the
density 29 and therefore, the divergence of the curvature scalar again. So, we have showed
that our minimal conditions (40) imply
R(a 0)

(72)

and therefore a = 0 is a true singularity.

As an example, the curvature scalar corresponding to the state equation P = q/3
(0 6 q < 1 to allow a = 0) is: 30
(q 2)0 a0
H .
(73)
3a 3+q
Once stated that a = 0 is a singularity, from the previous section we conclude that its
appearance is constrained to the conditions
R(a) =

Za
lim0

0 (a 0 ) da 0 < and E > 0 big-bang

(74)

whatever the three-dimensional curvature it is. Of course, the a = 0 point is reached in a

finite proper time bounded by
r
Za
da
3 2
a = finite,
6
(75)
1 = (a) (0) =
E
Vs (a)
0

pp

for E > 0 and a > 0 in the flat and hyperbolic geometries, and E > 0 and a+
in the spherical ones. In the E = 0 case, the bound is
s
Za
da
3a 5
= finite,

1 = (a) (0) =
0 (a)
Vs (a)

>a>0

(76)

for any K. To end this section we remark that, in this simple model of universe filled
with a perfect fluid of strings, the big-bang is strongly sensitive to the state equation, and
specifically to the fluids behavior at high density and pressure. Moreover, it is a very
28 For example, take us (a) f (a 3 ) with f (w) a positive definite and non-increasing function having
0
a finite f 0 (0) value. It is clear that 0 (a) satisfies the minimal conditions but the pressure in a = 0 is
P (0) = f 0 (0) < .
29 If (a) is positive and non-increasing from a = 0, we have lim
3
0
a0 (a) = lima0 0 (a)/a = .
30 (a) = /a 3+q .
0
0

260

J. Borlaf / Nuclear Physics B 593 (2001) 243266

surprising fact that the radiation state equation, which dominates the matter behavior at
high densities, is the canonical case where the singularity is forbidden. On the contrary,
the pressure-free state equation, P = 0, representing the matter at the present stage in the
universe, allows the singularity. But that state (the singularity) falls out of the physical
conditions described by the pressure-free fluid.
4.3. The singularity versus the string size
The kinematical relation (49) together with the differential version of Eq. (50) provide
an explicit relation between the string size and the effective universe size:
p
(77)
as as (a) = 2t0 Vs (a).
If the solution is singular (a = 0 and E > 0), we write:
v
u
Za
2t0 u
u
2
as = tE 3Ka + 0 (a 0 ) da 0.
3a

(78)

When we approach the singularity ( 0, a 0), the string size diverges, i.e.,
as (a 0) = . In fact, if E > 0 the divergence has the simple form: 31
 1/2
E
2t0
.
(79)
as (a 0) =

3
a
On the other hand, when E = 0 the divergence has the form: 32
r
0 (a)
.
(80)
as (a 0) 2t0
3a
Therefore we can conclude that the string size diverges in the singularity in all singular
solutions. Moreover, from (78) it is easy to convince ourselves that the converse is also
true: if the string size keeps bounded, the scale factor cannot vanish, that is, the singularity
is forbidden. In short,
big-bang as (a 0) = ,
as < no big-bang.

(81)

This result is again surprising because it makes incompatible the big-bang with our
physical point of view of a fluid constituted of small strings (as  a). Moreover, given
that the string size is bounded in all non-singular string cosmologies, we conclude that
our picture of strings wrapped around an small compact dimension implies the absence
of the initial singularity. In other words, the singular feature of the RobertsonWalker
cosmologies cannot be taken as fundamental in this stringy context, but a drawback in the

R
31 When a 0, the term R0 a 2 + a (a 0 ) da 0 goes to zero as well.
0 0
2
32 Remember that conditions (40) guarantee lim
a0 0 (a) > 0.

J. Borlaf / Nuclear Physics B 593 (2001) 243266

261

approximation as = 0, i.e., when the string shrinks to a point-particle. If we keep the string
size bounded but non-vanishing, i.e., 0 < as < , the scale factor of the universe cannot
vanish (0 < a 6 a) and the big-bang smoothes giving rise to the branch < 6 0. It
is remarkable that the preceding conclusions are obtained imposing the minimal conditions
on the state equation.
Now, we can understand the physical meaning of the time t0 in the singular solutions.
Let us assume the scale factor approaches to zero in the way a C with > 0. As
we saw, the string size as 2Ct0 1 (Eq. (49)) must diverge letting constrained the
exponent to the interval 1 > > 0. The divergence of the compact dimension should mean
the dominance of the string effects from the beginning of the universe until a certain time.
We can agree that this stringy period of time finish when the scale factor has grown to
reach the string size a(0 ) = as (0 ), implying 0 = 2t0 within our approximation. Due
to the bound over the exponent , our constant t0 is nothing but the amount of time from
the beginning of the universe when the strings effects are relevant. In other words, t0 is the
time needed to decouple the stringy effects in the evolution of the universe.
singular solutions
0 <  t0 stringy regimen,
 t0 point-particle regimen.

(82)

In the non-singular solutions these arguments are invalid and t0 has another physical
interpretation. In particular, when the macroscopical space is hyperbolic, it can be shown
that there is a single maximum between s (a ) = 0 and s (a+ = ) = 0 in the relative
string size (49). That maximum should correspond with a string driven cosmic phase. Given
that is proportional to the Hubble constant, maybe this period could be a candidate to an
inflationary expansion in the non-singular cosmologies. The value depends on the moduli
t0 which therefore should determine the strength on the string effects in that phase.
To end this section, in the unbounded cosmologies, expression (77) gives the asymptotic
string size which goes to the constant:
p
(83)
K 6 0 H as ( ) 2t0 |K|.
5. The pressure-free fluid of strings
If the whole energy-matter content in the universe keeps constant through its time
evolution, the relation density/scale factor must be a 3 = constant 0 , therefore
implying (39) the state equation P = 0. In other words, the continuum of strings in the fluid
is free-falling, i.e., there is no a direct interaction between them. Comparing the potentials
for the fluid of strings and the fluid of particles, both in the pressure-free case, 33
Vpoint-particle (a) =

1
pp

6Ka 2 + 0 a ,
2
6a

33 Of course we assume the same homogeneous and isotropic three-dimensional space, i.e., K strings =
K point-particle K.

262

J. Borlaf / Nuclear Physics B 593 (2001) 243266

1
(84)
E 3Ka 2 + 0 a
2
3a
we realize that the evolution of the three-dimensional space in the RobertsonWalker
cosmology coincides with the moduli E = 0 in the our string cosmology, provided that
p-p
we identify the constants in both models in the way 0 0 /2.
Vstrings(a) =

Dust RobertsonWalker P = 0 perfect-fluid string,

p-p 2 string.

(85)

view 34

we can consider our string

Therefore, from the four-dimensional point of
cosmolgy as a continuous deformation of the RobertsonWalker one, with E the
deformation parameter.
Flat case
For example, in the flat case (K = 0), we have, on one hand, the singular branch of the
deformation 35


30 1/2
= (a + a)
3/2 3a(a
+ a)
1/2 + 2a 3/2,
(86)
4
where the moduli E > 0 is redefined across the length a > 0. The point-particle case (3)
is obtained doing a = 0. That fact makes tempting to think on a as a typical length related
with the string size. However, this is not a correct guess because the string size evolves
(78) passing through all possible values as (0) = to as () = 0. On the other hand, the
non-singular branch is


30 1/2
= (a a )3/2 + 3a (a a )1/2 ,
(87)

4
where now the moduli a > 0 (E < 0) has an obvious physical meaning as the minimal
size reached by the universe in its whole evolution. As we mention before, due to the nonexistence of a physical singularity in = 0 (a(0) = a > 0), this solution can be extended
to = . 36 Therefore, we find an eternal universe whose initial state ( = )
corresponds with a four-dimensional Minkowskian empty space ((a = ) = 0). 37 In
this way, the matter has a geometrical origin, being the result of the three-dimensional
contraction from = .
Hyperboloid
If K < 0, the singular branch in the deformation consists of two branches separated by the
solution




a
a
|K| 1/2

ln
1
+

=
,
(88)
a 2
a
a
where a 0 /6|K| for all non-flat solutions (K 6= 0). The branch connected continuously
with the point-particle solution (3) can be parametrized
34 Setting aside the compact dimension.
35 Normalised to a(0) = 0.
36 The in (87) is just due to the extension < 0.
37 The = three-dimensional space is flat because its total curvature R ( ) = 6K/a 2 vanishes just due

to a() = .

J. Borlaf / Nuclear Physics B 593 (2001) 243266

263

a = (a/
cosh 0 )(cosh cosh 0 ),
cosh )(sinh cosh 0 ),
1 = |K|1/2 (a/

> 0 .

(89)

The moduli is given by 0 6 0 < and the singularity corresponds with = 0 . We

recover the point-particle solution in the particular value 0 = 0 of the moduli. The singular
branch non connected with the point-particle solution can be parametrized as:
a = (a/
sinh 0 )(sinh sinh 0 ),
sinh 0 )(cosh sinh 0 ),
1 = |K|1/2 (a/

> 0

(90)

The moduli of solutions is given by 0 < 0 < and again the singularity is the point
= 0 .
The non-singular branch has again moduli 0 < 0 < and each solution is parametrized
as:
a = a(cosh

0 cosh 1),

0 sinh ),
1 = |K|1/2 a(cosh

< < ,

(91)

0 1) corresponds with parameter

where the manifest minimal scale factor a = a(cosh
= 0.
Sphere
As proved in the preceding section, all spherical solutions have a finite maximum size a+ .
Again we have the split into two branches. The singular one is described by the moduli
0 6 0 < /2 and parametrized as
a = (a/
cos 0 )(cos 0 cos ),
cos 0 )( cos 0 sin ).
1 = K 1/2 (a/

(92)

+
The singularity is placed at = 0 and the maximum universe size is a+ = a(1
cos 0 )/ cos 0 .
The non-singular branch is labelled by the moduli 0 < 0 < /2 and each solution is given
by:
a = a(1
cos 0 cos ),
cos 0 sin ),
= K 1/2 a(

< < .

(93)

2 (0 /2) and the minimum is a = 2asin

2 (0 /2)
The maximum size is now a+ = 2acos
> 0. This branch is connected to the point-particle singular solution through the moduli
0 = 0 and, on the other hand, with the static solution a( ) = a = a+ = a through 0 =
/2.

6. Conclusions
It is well known [8] that the usual string cosmological point of view starts with the
low energy effective action for the massless string states and pays special attention to

264

J. Borlaf / Nuclear Physics B 593 (2001) 243266

the search of cosmological solutions. But in the same way we get the simplest pointparticle cosmologies coupling the Einstein gravity to the perfect fluid of particles, it
seems natural to couple string gravity, i.e., the low energy effective action for massless
strings, to a perfect fluid of NambuGoto strings. Following this point of view, first of
all we build the energymomentum tensor for a perfect-fluid of p-branes coupled to the
(p + 4)-dimensional general relativity. The geometric properties of that tensor as well as its
covariant conservation allow us to identify it as the energymomentum of a continuum of
p-branes. As a simple check, p = 0 recovers the standard point-particle perfect fluid and
p = 1 recovers the Leteliers string fluid now including pressures contribution. We have
called that the simplest field theories invariant under the arbitrary redefinitions of the fields
provide an explicit realization of this kind of fluids [3].
Our interest has been focused on explaining whether or not the RobertsonWalker
cosmologies can be understood as the limiting case when typical p-brane size (p > 0)
goes to zero in our brane cosmologies. Therefore, our spacetime consists of four macroscopical dimensions and p small space-like dimensions where the p-brane wraps around.
Moreover, we keep the homogeneity and isotropy in the three macroscopical space-like
dimensions. Following the same pattern as in the RobertsonWalker cosmologies, the
second step consists of coupling the energymomentum tensor of our perfect fluid of pbranes to the (p + 4)-dimensional general relativity. As in the point-particle cosmology, the
state equation becomes an input. The branes/geometry coupling can be reduced to a simple
effective field theory for the geometry over the branes and the scale factor for the threedimensional space. Once the state equation is given, the pressure and the matter density of
the fluid can be obtained explicitly in terms of the scale factor (41). As we expect, those
relations are macroscopic, i.e., they are independent of the brane dimension.
In the string cosmology (p = 1) the effective theory is nothing but the two-dimensional
dilaton gravity (43) where the potential for the dilaton (the scale factor) is totally
determined by means of the state equation and the constant scalar curvature of our
macroscopical three-dimensional space (44). Given that we can solve the theory, imposing
the minimal conditions over the state equation we are able to get a lot of information about
this string cosmology. First of all, like in the point-particle cosmologies, the expansion
of the three-dimensional space is not bounded in the flat and hyperbolic geometries and
it is bounded in the spherical one. Moreover the divergence in the hyperbolic universes
size follows the same asymptotic behavior as in the RobertsonWalker cosmologies, being
independent of the state equation.
The main differences with the point-particle models that we want to remark in this paper
concern the big-bang singularity. On the one hand, due to the kinematical relation (78), our
physical picture of a fluid consisting of small strings 38 is not compatible with the initial
singularity, which is a general feature in the point-particle cosmologies. Conversely, in
the initial singularity, if present, the string size diverges. We conclude that the Robertson
Walker cosmologies are not the limit of our cosmology of perfect fluid of strings when the
string size goes to zero.
38 The typical string size has an upper bound.

J. Borlaf / Nuclear Physics B 593 (2001) 243266

265

On the other hand, the big-bang is strongly sensitive to the state equation. Given that the
standard approximation at very high densities assumes radiation domination, it is specially
remarkable that the radiation state equation results to be the canonical case where the
singularity is kinematically forbidden. Using the stellar equilibrium language, if we have
a neighborhood of 1/ = 0 (high density) where P > /3 holds, the expansive effect due
to the pressure is able to compensate the gravitatorial tendency and the universe does not
collapse in the singularity. On the contrary, if P 6 /3 the collapse is possible, although it
still depends on an initial condition (the integration constant E). Therefore one of the most
important conclusions we get is that, unlike the point-particle cosmologies, the big-bang is
not a general feature of our string cosmologies. Moreover, the very special dependence of
the initial singularity on the state equation seems to exclude this possibility.
The absence of the initial singularity allow us to extend the solutions from = 0
(minimal non-vanishing size) to = resulting in a eternal and non-singular universe.
In the flat and hyperbolic geometries, 39 the universe evolves from an initial ( = )
flat state (infinite macroscopical size) 40 and empty, going to a non-vanishing minimal size
state having a finite matter density ( = 0), bouncing and ending in final state ( = )
also empty and flat. We note that we can understand the presence of matter in the present
times as the result of a contraction of the initial flat and empty space. Therefore, in this
model matter has a purely geometric origin. But, what about the strings? In this evolution,
the size of the compact dimension where the strings wrap around keeps bounded. The
maximum string size and the minimum non-vanishing universe size are, together with the
state equation, the inputs of the model.
In the spherical case, the universe repeatedly fluctuates between a non-vanishing
minimal size a > 0 and a finite maximum size a+ .
Of course, the full cosmological meaning of an string fluid energymomentum tensor
must be found in its coupling with the string effective action [9] including torsion and
dilaton fields. From that point of view, this paper must be understood as a natural
previous step where basic features of fluids description are encountered allowing a direct
comparison with the standard RobertsonWalker cosmologies. Moreover, we must have in
mind that T-duality will enforce to carefully redefine the concept of small [10] as well as
of homogeneity and isotropy demanding a slight modification of our energymomentum
tensor. Then, armed with the full (gravity/matter) string model, maybe we will be able to
verify the goodness of assumptions like dilatons dominance and matter dominance
in different stages of the assumed stringy cosmological model. What is more important,
does dilatons contribution modify our non-singular scenario? This work is in progress. The
same strategy will be followed with higher branes. There, the pure gravity is postulated via
electromagnetic duality [4,11].

39 Assuming the radiation dominates the maximum density states of the universe, and the ordinary free-falling
(P = 0) matter dominates the low density states.
40 The three-dimensional curvature and the whole curvature tensor vanishes when a = .

266

J. Borlaf / Nuclear Physics B 593 (2001) 243266

Acknowledgements
I am grateful to the Universidad Autnoma de Madrid and specially to Enrique Alvarez,
the Instituto de Fisica Terica (U.A.M./C.S.I.C.) and Silvia Vargas for their support.

References
[1] S. Weinberg, Gravitation and Cosmology, Wiley, 1972;
R.M. Wald, General Relativity, Univ. of Chigago Press;
S.W. Hawking, G.F.R. Ellis, The Large Scale Structure of SpaceTime, Cambridge Univ. Press,
1973.
[2] P.S. Letelier, Clouds of strings in general relativity, Phys. Rev. D 20 (1979) 12941302;
P.S. Letelier, String cosmologies, Phys. Rev. D 28 (1983) 24142419;
L.L. Smalley, J.P. Krisch, String fluid dynamics in general relativity, Class. Quantum Grav. 14
(1997) 24052415.
[3] J. Borlaf, Perfect fluids of p-branes from theories invariant under field redefinitions, Phys. Rev.
D 60 (1999) 46001.
[4] M.J. Duff, R.R. Khuri, J.X. Lu, String solitons, Phys. Rept. 259 (1995) 213326, hepth/9412184;
M.J. Duff, Supermembranes, hep-th/9611203.
[5] T. Banks, M. OLoughlin, Two-dimensional quantum gravity in Minkowski space, Nucl.
Phys. B 362 (1991) 649664.
[6] M. Gasperini, G. Veneziano, Singularity and exit problems in two-dimensional string cosmology, Phys. Lett. B 387 715720;
S.-J. Rey, Back reaction graceful exit in string inflationary cosmology, Phys. Rev. Lett. 77
(1996) 19291932.
[7] D. Louis-Martinez, G. Kunstatter, Birkhoffs theorem in two-dimensional dilaton gravity, Phys.
Rev. D 49 (1994) 52275230.
[8] G. Veneziano, String cosmology: concepts and consequences, hep-th/9512091;
J. Barrow, K.E. Kunze, String cosmology, gr-qc/9807040.
[9] D. Lst, S. Theisen, Lectures on String theory, Lectures Notes in Physics, Springer-Verlag;
M. Green, J.H. Schwarz, E. Witten, Superstring Theory, Cambridge Univ. Press, 1987.
[10] M. Gasperini, G. Veneziano, O(d,d)-covariant string cosmology, Phys. Lett. B 277 (1992) 256
264.
[11] C. Park, S.-J. Sin, P-brane cosmology and phases of BransDicke theory with matter, Phys.
Rev. D 57 (1998) 46204628.

Nuclear Physics B 593 (2001) 269288

www.elsevier.nl/locate/npe

Analytic resummation for the quark form factor

in QCD
Lorenzo Magnea
Dipartimento di Fisica Teorica, Universit di Torino, and I.N.F.N., Sezione di Torino,
Via P.Giuria 1, I-10125 Torino, Italy
Received 14 July 2000; accepted 16 October 2000

Abstract
The quark form factor is known to exponentiate within the framework of dimensionally regularized
perturbative QCD. The logarithm of the form factor is expressed in terms of integrals over the scale of
the running coupling. I show that these integrals can be evaluated explicitly and expressed in terms of
renormalization group invariant analytic functions of the coupling and of the spacetime dimension,
to any order in renormalized perturbation theory. Explicit expressions are given up two loops. To this
order, all the infrared and collinear singularities in the logarithm of the form factor resum to a single
pole in , whose residue is determined at one loop, plus powers of logarithms of . This behavior is
conjectured to extend to all loops. 2001 Elsevier Science B.V. All rights reserved.

1. Introduction
The asymptotic behavior of the electromagnetic form factor of charged particles at high
energy has been the object of theoretical studies for almost half a century. It is the simplest
gauge theory amplitude to exhibit a doubly logarithmic behavior, as a consequence of
its infrared and collinear singularities. Thus the understanding of its asymptotic energy
dependence requires a resummation of perturbation theory, going beyond renormalization
group logarithms. This resummation was performed for the first time, at the leading
logarithm level, by Sudakov [1], who considered the off-shell form factor for an abelian
gauge theory. He found that the leading (double) logarithms exponentiate, which results in
the strong suppression of the elastic scattering of charged particles at high energy which
bears his name. A similar exponentiation for the on-shell form factor was obtained in [2],
and in later years the result was extended to nonabelian gauge groups [3].
Further extending the exponentiation to all subleading logarithms is a nontrivial task,
which requires the complex machinery of perturbative factorization theorems. This goal
E-mail address: [email protected] (L. Magnea).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 2 3 - 4

270

L. Magnea / Nuclear Physics B 593 (2001) 269288

was achieved by Mueller [4] and Collins [5] for abelian theories, and shortly later by
Sen [6] for QCD. In essence, the complete exponentiation is achieved by showing that the
energy dependence of the form factor must follow an evolution equation, which embodies
the constraints of renormalization group and gauge invariance. This evolution equation
was typically solved in terms of an undetermined initial condition, containing low-energy
information and depending on some chosen infrared regulator, say, a gluon mass. It was
later realized [7] that dimensional regularization can be used directly at the level of the
evolution equation. One can then give a simple and explicit solution for the form factor,
using as initial condition the electric charge at vanishing momentum transfer. When the
solution is written in this form, all infrared and collinear poles are explicitly exponentiated,
and the results can be directly compared with diagrammatic calculations.
The quark form factor, although divergent, is a quantity of considerable interest
for phenomenology: in fact, it appears in the computation of several cross sections,
and techniques similar to the ones used for its exponentiation have been applied to
several processes of interest in high energy physics, in particular to processes in which
the perturbative series needs to be resummed because of logarithmic enhancements
in special regions of phase space. Outstanding examples are the inclusive DrellYan
cross section [8] and the transverse momentum distribution of vector bosons produced
in hadronic collisions [9]. More recently, similar techniques have been generalized to
hadronic cross section with colored particles in the final state [10]. Strikingly, the quark
form factor enters directly in the hard partonic cross section for the DrellYan process in
the DIS scheme, which is proportional to the modulus squared of the ratio of the timelike
to the spacelike form factor [8,11]. In [7] it was shown that this ratio is given by an infinite
phase (a generalization of the Coulomb phase, expressed by a series of pure counterterms
in the MS scheme), times a manifestly finite exponential factor.
In the present paper, I will build upon the results of Ref. [7], and show that the resummed
expression for the form factor can be explicitly evaluated to the desired accuracy, thus
expressing the logarithm of the form factor as an analytic function of s (Q2 ) and ,
manifestly renormalization group invariant. The key observation is the following: typically,
in resummed expressions, integrals over the renormalization scale cannot be explicitly
evaluated, because the one-loop running coupling has a Landau pole on the integration
contour, which extends from the hard scale, 2 = Q2 , all the way down to 2 = 0, in the
region dominated by nonperturbative effects. The sensitivity to the Landau pole, which
can be estimated using various methods and models [12], is then taken as a measure of the
influence of nonperturbative effects on the given observable. In dimensional regularization,
however, the running coupling depends not only on the renormalization scale, but also
on the dimensional regularization parameter . At one loop, the coupling still exhibits
a Landau pole, but for general (complex)  the pole is located at complex values of
the renormalization scale 2 , away from the integration contour. As a consequence, all
integrals appearing in the logarithm of the form factor are well-defined, and can be
evaluated. Alternatively, one can express the one-loop running coupling as a power series
in terms of s (Q2 ), and integrate the series term by term: one finds that there is no
factorial growth in the coefficients of the series obtained after integration; in fact, the

L. Magnea / Nuclear Physics B 593 (2001) 269288

271

series can be summed to reconstruct the same analytic function of s and  that was
found by direct integration. It is easy to see how these results generalize when the running
coupling is expressed including higher orders in the QCD function. The complexity
of the answer grows, but the most relevant features remain unchanged; in particular, the
answer is manifestly invariant under changes in the renormalization scale, to the given
accuracy in the function; also, infrared and collinear poles resum, so that the logarithm
of the form factor has only a single pole in , as well as cut on the negative real axis in the
 plane; conventional perturbative results can be recovered by reexpanding in powers of s
for finite .
The outline of the paper is the following. In Section 2 I briefly review the known
results on the exponentiation of the quark form factor. In Section 3 I illustrate the power
of dimensional regularization by solving explicitly to all orders the recursion relation
arising from the renormalization group equation for the generalized Coulomb phase: all
the towers of poles (leading, next-to-leading, etc.) arising in the series of counterterms can
be expressed in terms of the perturbative coefficients of the relevant anomalous dimension,
and of the QCD function; all the resulting series can be summed, and they give functions
of  with at most a logarithmic singularity; the leading behavior as  0 can be explicitly
computed to all orders. In Section 4 I turn to the main issue, and compute the logarithm
of the form factor using one-loop results only, as well as the one-loop running coupling;
the result is remarkably simple, a single dilogarithm and a single logarithm of functions
of s and , explicitly independent of the choice of renormalization scale. In Section 5
I generalize the calculation to two loops, obtaining again a relatively simple answer,
containing only logarithms and dilogarithms. It is fairly clear that the calculation could
be formally pushed to even higher orders, adding to the complexity of the answer, but
without changing its basic features; furthermore, the perturbative coefficients of the various
functions appearing in the logarithm of the form factor are known only to two loops.
Section 6 summarizes the results and briefly discusses possible applications and future
developments.

2. Exponentiation of the form factor

The techniques needed to compute the quark form factor in perturbative QCD to all
orders in the logarithms of the energy are described in detail in Ref. [13]. Here I will just
give the relevant definitions and sketch the main results, in the spirit of Ref. [7].
The (timelike) quark form factor in dimensionally regulated massless QCD is defined
by
(p1 , p2 ; 2 , ) = h0|J (0)|p1 , p2 i


Q2
2
, s ( ),  ,
= ieeq v(p2 ) u(p1 )
2

(2.1)

where J (x) is the electromagnetic quark current, while |p1 , p2 i is a state containing a
quark and an antiquark with momenta p1 and p2 , and total energy Q2 = (p1 + p2 )2 , taken

272

L. Magnea / Nuclear Physics B 593 (2001) 269288

to be much larger than the confinement scale. In Eq. (2.1) I made use of the fact that in the
massless limit the perturbative form factor has only one spin structure, and I defined the
scalar form factor to be normalized to 1 in the absence of strong interactions. Eq. (2.1) is
written in the renormalized theory, so that all ultraviolet divergences arising in perturbation
theory have been dealt with, and one can take  = 2 d/2 < 0 to regulate infrared and
collinear poles. Notice that, as a consequence of electromagnetic current conservation, the
anomalous dimension of the form factor vanishes.
The conventional analysis of the form factor [13] proceeds through the steps that
are characteristic of the derivation of perturbative factorization theorems: one starts by
identifying the integration regions in momentum space that contribute to the form factor
at leading power 1 of Q2 ; next, one constructs a factorized expression for the form
factor, in terms of (nonlocal) operators, typically defined in terms of eikonal lines, whose
matrix elements reproduce the leading behavior in the various regions. Each operator
factor will typically depend both on the gauge choice and on the renormalization scale;
imposing gauge and renormalization group (RG) invariance leads to an evolution equation
describing the energy dependence of the whole form factor. In dimensional regularization,
this equation takes the form
  2

 2




Q
1
Q
2
2
2
2
log
, s ( ), 
=
, s ( ),  .
K , s ( ) + G
Q
Q2
2
2
2
(2.2)
The functions K and G are characterized as follows: K is a pure counterterm, thus a
series of poles in the MS scheme, which I will use throughout; G contains all the energy
dependence, and is finite as  0; the sum K + G must be renormalization group
invariant, because the form factor is; thus K and G renormalize additively, according to

  2


+ (, s )
, s ,  = K (s )
G

s
2



+ (, s )
K(, s ). (2.3)
=

s
The anomalous dimension K is well known in perturbative QCD. It is essentially the
anomalous dimension associated with a cusp in a Wilson line, corresponding here to the
hard vertex where the quark and the antiquark annihilate [14]. It is independent of , and
can be expanded perturbatively as

 
X
(n) s n
K
,
(2.4)
K (s ) =

n=1

with coefficients known to two loops [15]



67
5
(2) CA CF nf CF .
K(2) =
K(1) = 2CF ,
18
9

(2.5)

1 Technically, this is done using a massive regulator; in dimensional regularization all the energy dependence
is logarithmic, leading regions are responsible for infrared and collinear poles, and one is in fact working with
the whole form factor.

L. Magnea / Nuclear Physics B 593 (2001) 269288

273

Since we are dealing with divergent quantities, it is important to keep  < 0 throughout.
Thus the function is defined by
s
s ),
= 2s + (

s2 X  s n

bn
.
(s ) =
2

(, s ) =

(2.6)

n=0

At the leading nontrivial order, the solution of Eq. (2.6) is


 2  
 2

 


2  1

1

b0
2
2
2
1
s 0

, s (0 ),  = s 0

,

4
20
20
20

(2.7)

with b0 = (11CA 2nf )/3. Using this form of the running coupling, and its higher order
generalizations, one can solve the renormalization group equation for the function G, as



 
 2
Q2
Q
2
2
,

(
),

=
G
1,

(
),

,
G
s
s
2
2
Z

1
+
2


  2

d2
2
K
, s ( ),  ,
2
2

(2.8)

Q2

where the initial condition was chosen to emphasize that G is real for negative Q2 .
The second important consequence of dimensional regularization is the fact that it
provides a natural initial condition for Eq. (2.2), in terms of which one can construct an
explicit solution. In fact, one readily recognizes that each term in the perturbative expansion
of (Q2 ) must be proportional to a positive integer power of (2 /(Q2 )) . Thus for fixed
 < 0 all perturbative corrections to the form factor vanish in the limit Q2 0, and one
can use


(2.9)
0, s (2 ),  = 1.
One finds then

 2
Q
2
, s ( ), 

2
2
"
( Q

 2
 
Z

d 2
1
2
2
, s ( ),  , 
K , s ( ) + G 1,
= exp
2
2
2
0
2

1
+
2

#)
  2

d2
2
K
, s ( ), 
.
2
2

(2.10)

Eq. (2.10) can be used directly as a generating function of diagrammatic results in

dimensional regularization; for example, using one-loop results for the various functions
appearing in the exponent, as well as the one-loop running coupling reexpanded in terms
of s (Q2 ), one reproduces correctly the quartic and cubic poles in  at O(s2 ). Introducing

274

L. Magnea / Nuclear Physics B 593 (2001) 269288

the two-loop contribution to the anomalous dimension K , one matches the double pole
as well; the only singularity requiring a full two-loop computation (the knowledge of
G(2) , as well as O( 2 ) terms in G(1) ), is the single pole, in agreement with the general
considerations of Ref. [16]. In Ref. [7] it was shown that Eq. (2.10) can be used to evaluate
explicitly the ratio of the timelike to the spacelike form factor, in terms of a contour integral
in the complex plane of the renormalization scale 2 . The result (at 2 = Q2 ) is of the form
(1, s (Q2 ), )
(1, s (Q2 ), )
(
Z "

i

2
d G 1, ei , s (Q2 ),  , 
= exp i K , s (Q ) +
2
2
0

Z
d K ei , s (Q2 ), 

#)
.

(2.11)

Eq. (2.11) proves that the ratio of form factors is given by an infinite phase times a finite
exponential factor. The derivation of Eq. (2.11) also shows that the ratio is dominated by
ultraviolet contributions: in fact, it is expressed by a contour integral that never approaches
the origin. Thus, one may conclude that it is free of power corrections related to the Landau
pole. In particular, the integrals in Eq. (2.11) can be evaluated explicitly to the desired
accuracy in the running coupling; the result is of direct phenomenological relevance, since
the modulus squared of the ratio of form factors appears in the resummed partonic Drell
Yan cross section [8,17].
I will now concentrate on Eq. (2.10), and examine the structure of the exponent in more
detail. In particular, I will be concerned with the fate of the Landau pole, that is expected
to make the integration over the infrared region ill-defined. It will become apparent that
dimensional regularization provides a solution to this problem. I will however start with an
exercise in the application of the renormalization group, which displays in a simple context
the power of dimensional regularization.

3. The counterterm function

Let us begin by considering the Coulomb phase K(, s ). In any minimal renormalization scheme, it is a series of poles which can be written as 2
K(, s ) =

X
Kn (s )
n=1

n

Kn (s ) =

X
m=n

Kn(m)

 m
s

(3.1)

In such a scheme, K(, s ) has no explicit dependence on the renormalization scale . The
renormalization group equation (2.3) then reduces to
2 Notice that K(, ) has only one pole per loop. The leading pole in the exponent of (Q2 ), ( n / n+1 ), is
s
s
generated at each order by the integration of the anomalous dimension K .

L. Magnea / Nuclear Physics B 593 (2001) 269288

(, s )

K(, s ) = K (s ),
s

275

(3.2)

which, given the finiteness of K and the form of (, s ), turns into a recursion relation
for the perturbative coefficients of K. Explicitly
1
d
K1 (s ) = K (s ),
ds
2
1
d
d
s)
Kn+1 (s ) = (
Kn (s ).
s
ds
2
ds

(3.3)

Eq. (3.3) expresses the fact that the physics of the function K(, s ) is contained in the
residue of the simple pole, K1 (s ), which is completely determined by the anomalous
dimension K , and in turn determines the coefficients of all higher poles. One can
approximate the exact recursion relation by truncating the four-dimensional function
s ) to some chosen order.
(
It is useful to collect the terms in K corresponding to leading poles ((s /)n ), next-toleading poles (sn+1 / n ), and so on. This is done by writing
K(, s ) =

Km (, s ),

m=0

Km (, s ) =

X
n=1

Kn(n+m)

 n+m 1
s
.

n

(3.4)

According to general arguments, one expects the leading poles in K to be completely

determined to all orders by the one-loop anomalous dimension and the one-loop
function; similarly, the mth tower of poles, Km is expected to require the anomalous
dimension and the function to m + 1 loops.
In the remainder of this section, the following facts about these towers of poles will be
established:
(m)
The recursion relation for the perturbative coefficients Kn can be explicitly solved
including all orders in the function.
All the resulting series of poles, Km (, s ), can be summed.
Upon summation, Km (, s ) is an analytic function of s and , regular as  0 for
m > 0. The only singularity at  0 is logarithmic and completely determined by a
one loop calculation.
The finite limits Km (0, s ) (m > 0) can be computed for all m in terms of the
perturbative coefficients of and K . They reconstitute a power series in s .
As with most recursion relations, it is useful to start with the simplest cases.
3.1. One loop
The first of Eqs. (3.3) is readily solved and yields
(m)

K1

1 (m)
.
2m K

(3.5)

In the second of Eqs. (3.3) one can start by truncating the function at O(s2 ), keeping
only the one-loop coefficient b0 . Using Eq. (3.5) one readily finds an expression for the
generic coefficient Kn(m) ,

276

L. Magnea / Nuclear Physics B 593 (2001) 269288

Kn(m)



b0 n1 (mn+1)
1

=
K
,
2m
4

(3.6)

which, as we will verify, is exact for n = m. Then



 n 2 (1) 
X
b0 s
s
Kn(n)
= K ln 1 +
.
K0 (, s ) =

b0
4

(3.7)

n=1

Notice that this implies that the resummation of all the leading poles in the ratio (2.11)
degrades the singularity to a logarithm:

 (1)

b0 s iK /b0
(1, s (Q2 ), ) 
= 1+
.
4
(1, s (Q2 ), ) L.P.

(3.8)

Since all other poles are perturbatively weaker, one might expect that they should resum to
even weaker singularities, or to regular functions. This in fact turns out to be the case.
3.2. Two loops
To resum next-to-leading poles, one must include the next coefficient of the function,
(2)
b1 , as well as K . Eq. (3.3) may still be solved, and gives
" n1 
#




1 X n1
b0 n1k
b1 k (mnk+1)
(m)
K
,
(3.9)

Kn =
2m
4
4
k
k=0

(p)

where the sum is further truncated by the requirement that K = 0 for p < 1. One recovers
the previous answer, Eq. (3.7), for the leading poles. Next-to-leading poles are given by

 

b0 n3 b0 (2)
1
b1 (1)
(n)

K (n 2) K ,
(3.10)
Kn1 =
2n
4
4
4
Again, the corresponding series is summable, and yields
 n+1 1
s

n
n=1

 (1)
 
8 2K b1
b0 s
(2)
= 2
K ln 1 +
b0
4
b0
  (1)


2 (1)b1
4
s 2 K b1
.
K(2) + K2

b0
b0
b0 b0 s + 4

K1 (, s ) =

Kn(n+1)

(3.11)

The logarithmic singularity receives a new contribution, which, however, is supressed by a

power of . Thus K1 has a finite limit as  0, given by

(1) 
2s (2) K b1

.
(3.12)
K1 (0, s ) =
b0 K
b0

L. Magnea / Nuclear Physics B 593 (2001) 269288

277

3.3. All loops

Encouraged by the simple form of the answers obtained for K0 and K1 , one sets out to
s ). It turns out that the recursion relation (3.3) is solvable to
include higher orders in (
all orders, and the solution is a straightforward generalization of Eqs. (3.6) and (3.9). One
finds
" n1
#

P
X
Y
b pj
1
(mn+1 n1
(m)
i=1 pi )
K
.
(3.13)

Kn =
2m
4
p1 ,...,pn1 =0

j =1

To identify explicitly the towers of subleading poles, it is useful to reorganize the sum
isolating the perturbative coefficients of the anomalous dimension K . This is done by
writing Eq. (3.13) as
( q
#)
qn2 " n1

1
X Y
bqj qj+1
1 X (mn+1q1 ) X
(m)
,
(3.14)
K

Kn =
2m
4
q1 =0

q2 =0

qn1 =0

j =1

where the sum over q1 is effectively cut off by the fact that for q1 > m n + 1 the summand
vanishes. It is not trivial to reconstruct the general form of the mth tower of subleading
poles from Eq. (3.14); however, upon generating a sufficient number of examples, one
recognizes that


m+1
1
b0 nm1 X
(m+2p)
(n)
Cp (n, m) K
,
(3.15)

Knm =
2n
4
p=1

where
Cp (n, m) =

p1
X
k=0

nm1
k



b0 k (p)
Ck (b1 , . . . , bp1 ).

(3.16)

(p)

The coefficients Ck (b1 , . . . , bp1 ) are polynomial expressions in the bi s, given by

" k 
#
p1
X
Y
bmi
(p)
.
(3.17)

Ck (b1 , . . . , bp1 ) =
4
Pk

mi =1

i=1

i=1 mi =p1

The key observation concerning Eqs. (3.15)(3.17) is that the dependence on n is simple, so
that summation over n can explicitly be performed for all m. In fact, for fixed m, Eq. (3.15)
can be rewritten as


m
1
b0 nm1 X
(n)
Dr (m)nr ,
(3.18)

Knm =
2n
4
r=0

with easily determined coefficients Dr (m). All the series to be summed are thus simple
power series, in fact derivatives of the logarithmic series which resums the leading poles,
up to subtractions of a finite number of terms. It is easy to verify that the results obtained
for Km (, s ) are finite in the limit  0. One can in fact determine the form of this

278

L. Magnea / Nuclear Physics B 593 (2001) 269288

limit explicitly for all m. It is given by polynomial expressions in the coefficients of the
function, similar to Eqs. (3.15)(3.17):
m
1  s m 2 X
(m+1p)
Bp (m) K
,
m
b0

Km (0, s ) =

(3.19)

p=0

where
Bp (m) =

p
X
j =0

b0
Bj (m, p)
2

and
Bj (m, p) =

"

m
X
q =1
Pj i
i=1 qi =p

j
,

#
j 
Y
bqi
.
2

(3.20)

(3.21)

i=1

The most amusing fact here is perhaps that in the limit  0 one recovers a finite
perturbative expansion in powers of s / .
The conclusion of this exercise is that the counterterm function K(, s ) is completely
determined, as an analytic function of , s , and the perturbative coefficients of and K .
In the limit  0 this function has a logarithmic singularity, arising from the resummation
of the leading poles, and a contribution which is independent of  and can be computed
as a perturbative expansion in powers of s / , with coefficients determined again by the
functions and K . Explicitly
K(, s ) = KDIV (, s ) + KFIN (s ) + O(,  ln ),

(3.22)

with KDIV (, s ) given by Eq. (3.7), and

KFIN (s ) =

Km (0, s ) =

m=1

Am =

m
X

2 X 1  s m
Am ,
b0
m
m=1

(m+1p)

Bp (m) K

(3.23)

p=0

None of the series involved in this computation shows signs of factorial growth in the
coefficients. In the final expression, Eq. (3.23), the growth of the coefficients is implicit in
their dependence on bi and K(i) , the series in Eq. (3.23) being otherwise convergent with a
logarithmic behavior.

4. One-loop resummation
The success in the computation of the counterterm function K(, s ) is encouraging in
view of a more explicit evaluation of the full form factor. It is particularly tempting to
study the behavior of the integrals in Eq. (2.10) in the small- 2 region, where one expects

L. Magnea / Nuclear Physics B 593 (2001) 269288

279

to find obstructions due to the Landau pole, which should however cancel in the ratio of
the timelike to the spacelike factor, Eq. (2.11), which is ultraviolet dominated.
A natural place to start is a more precise characterization of the Landau pole in the
context of dimensional regularization. The one-loop running coupling for finite  can still
be characterized by a mass scale, defined as the value of the renormalization scale where
the coupling diverges, just as in conventional dimensional transmutation. In fact, Eq. (2.7)
still has a simple pole located at
1/

4
2
2
2
,
(4.1)
= Q 1+
b0 s (Q2 )
and one can use the scale 2 to define the coupling, as
s (Q2 ) =

b0

4

Q2

2

.

(4.2)

By inspection, Eqs. (2.7), (4.1) and (4.2) reduce to their four-dimensional counterparts as
 0. There are however two important observations to be made.
The running coupling defined by Eq. (4.2) vanishes as Q2 0 for fixed , provided
Re  < 0, in agreement with Eq. (2.7) and with Eq. (2.9). This behavior for small Q2
is simply due to dimensional counting: in fact, since we are keeping the regulator in
place, we are studying the renormalized coupling gR = g0  for fixed bare coupling,
in d > 4, where it vanishes as a power of the scale.
For general (complex) , the Landau pole in Eq. (4.1) is located at complex values
of the scale 2 . In particular, for real negative , the location of the pole has a
nonvanishing imaginary part provided  < b0 s (Q2 )/(4),  6= 1/n. Thus, in
general, the Landau pole in the dimensionally regulated exponent of Eq. (2.10) is not
on the integration contour. One expects, and I will verify below, that the integrals
appearing in Eq. (2.10) may be evaluated explicitly in terms of analytic functions of
s and . When  becomes close to 0, the pole migrates to the integration contour, and
correspondingly the integral will develop a cut. Nonperturbative information (if any)
will be encoded in the structure of the singularities at the cut, and in the corresponding
higher order generalizations.
Armed with these observations, one can proceed to verify the integrability of log (Q2 ),
starting with the simplest situation, in which only one-loop information is retained. The
integration can be performed using three different methods, which highlight different
features of the exponent.
4.1. Integration by series
The idea of reexpanding the one-loop running coupling, Eq. (2.7), in powers of s (Q2 )
is perhaps the first that comes to mind, in view of its frequent usage in the context of the
study of power corrections. One writes

 2

  2 
  2 



X

n
0
0

b0
1
2
2
2
s 0 ,

, s 0 ,  =
s 0
1
(4.3)

4
2
2
20
n=0

280

L. Magnea / Nuclear Physics B 593 (2001) 269288

and one integrates the series term by term, looking for possible factorial growth in the
coefficients of the series obtained after integration. This would then be interpreted by
standard methods as a sign of nonperturbative contributions. In the present case, changing
variables in each integral in Eq. (2.10) as
 2 

2
1,
(4.4)
z=
2
one finds that all integrals as easily performed. The integration of the anomalous dimension
K generates terms that are independent of , the integration variable of the outer integral
in Eq. (2.10). These terms are divergent at the lower limit of integration, 2 = 0, but they
are cancelled by the -independent terms in the expansion of the counterterm function
K(, s ), a fact that can be explicitly verified to all orders by using the results of Section 3.
Considering for simplicity the spacelike form factor (obtained from Eq. (2.10) by changing
Q2 Q2 ), the result is then


2
Q2
2
, s ( ),  =
a(2)
log
b0
2



 2  k+1

n

X
(1) X 1

a(2 ) n (1)n+1
K2
1


n+1
k+1
2
Q2
n=0
k=0





2  n+1
G(1) ()
1
,
(4.5)
1


Q2
where I defined
b0
s (2 ),
4
and I wrote the function G perturbatively as
a(2 ) =

G(1, , ) =

X
n=1

G(n) ()

 n

(4.6)

(4.7)

with G(1) () = CF (3 ( (2) 8))/2 + O( 2 ).

It is apparent that the series in Eq. (4.5) is far from pathological. It does not exhibit any
factorial growth, it is alternating in sign, and definitely has a finite radius of convergence. In
fact, one recognizes a logarithmic series in the terms with a single sum, whereas the terms
in the nested sum are finite harmonic sums of the kind encountered performing Mellin
transforms of AltarelliParisi kernels, which are related to polylogarithms.
Having found that the series arising from the integration of Eq. (2.10) is well-behaved,
clearly the simplest method to sum it is to perform the integrals directly, using Eq. (2.7)
instead of the Taylor expansion in Eq. (4.3).
4.2. Analytic expression
Direct integration proceeds through steps that are closely related to the ones outlined in
the previous subsection. Through the change of variable in Eq. (4.4) one trivially integrates

L. Magnea / Nuclear Physics B 593 (2001) 269288

281

the one-loop anomalous dimension, obtaining a logarithm. The logarithm is not integrable
over the scale 2 because of the singularity at 2 = 0, which is however subtracted by the
contribution of the counterterm function (here one must use for K(, s ) the resummation
of the leading poles, Eq. (3.7)). Using once more the same change of variables, the second
integration can also be performed. After minor reshuffling with dilogarithm identities, and
(1)
inserting the values of K and G(1) , the result reads



 2 

2CF 1
a(2 )

Q2
2
Li2
, s ( ),  =
log
b0 
2
Q2 a(2 ) + 


 2 
a(2 )

, (4.8)
C() ln 1
Q2 a(2 ) + 
where C() = G(1) ()/CF . Eq. (4.8) sums the series in Eq. (4.5), and has several rather
remarkable properties. First, since the form factor is RG invariant, and I used the exact
solution of the one-loop RG equation for to perform the integration, Eq. (4.8) should be
independent of 2 to one-loop accuracy in (s ). This is easily verified, and one obtains




Q2
2
,

(
),

= log 1, s (Q2 ), 
log
s
2






a(Q2 )
a(Q2 )
2CF 1
Li2
+ C() log 1 +
.
(4.9)
=
b0 
a(Q2 ) + 

Eq. (4.9), and its generalization including two-loop effects given in Section 5, are the
central results of the present paper. The quark form factor is compactly expressed in terms
of a simple analytic function of the coupling and of , which is manifestly independent
of the renormalization scale, and can be reexpanded in powers of s (Q2 ) to generate
perturbative results that can be directly compared with Feynman diagram calculations. In
particular, Eq. (4.9) resums all leading (sn / 2n ) and next-to-leading (sn / 2n1 ) infrared
and collinear poles in the form factor.
The second important feature of Eqs. (4.8) and (4.9) is that it is possible to study the
structure of the singularities of the form factor in the complex  plane. A detailed analysis
of this question is left for future work. Here I will simply note a few aspects that are
apparent in Eq. (4.9). First, the Landau singularity, as expected, appears in Eq. (4.9) as one
of the two branching points of a cut which is shared by both terms. The cut can be taken to
run along the negative real axis in the  plane (the physical region), from  = a(Q2 )
(the Landau singularity, see Eq. (4.1)) to  = 0. At the Landau point, both terms in Eq. (4.9)
diverge logarithmically. On the other hand, the limit  0, which is the physically relevant
one, can be studied explicitly; one finds

 




1
1
3
2CF
a(Q2 )
(2)
+
+

log

log 1, s (Q2 ),  =
b0

a(Q2 )
a(Q2 ) 2


+ O(,  log ) .
(4.10)
Eq. (4.10) has interesting features: one observes that resumming all the leading poles in the
logarithm of the form factor softens the singularity in a manner similar to what happens

282

L. Magnea / Nuclear Physics B 593 (2001) 269288

for the counterterm function (see Eq. (3.7)). In this case the leading singularity is a simple
pole, with a residue which does not depend on the coupling, an thus not on the scale either.
Including higher loops in corresponds to resumming weaker singularities, thus one might
expect that it would not affect the leading singularity in Eq. (4.10). In Section 5, I will
verify that this is indeed the case at two loops. Another interesting feature of Eq. (4.10)
is the appearance of a term in the form factor of the form exp(c/a(Q2)), which, in the
 0 limit, translates into a power behavior of the form (Q2 /2 )c . The term in question
in the present case is not of direct physical significance: it leads to a positive fractional
power of Q2 , and it is tightly connected with the infrared divergence; in fact, as we will
see, it cancels in the ratio (2.11), which is the simplest quantity of physical interest that can
be built out of the form factor. It is however interesting that such a term can arise in this
way: one sees that dimensional regularization and resummation can combine to give welldefined (and gauge-invariant) power corrections, which might be of physical interest if
the present method can be generalized to construct infrared safe quantities to the same level
of accuracy.
As a final test of the correctness of Eq. (4.9), one can compute the ratio (Q2 )/ (Q2 )
in the limit  0. To the present accuracy, one finds


 

2CF
a(Q2 )
(1, s (Q2 ), )
=
i 1 + log
log
b0

(1, s (Q2 ), )





3
1
2
+ i log 1 + ia(Q ) + O() ,

a(Q2 ) 2
(4.11)
which agrees with Eq. (3.7) for the logarithmically divergent term, and agrees with the
results of [7] for the finite contribution, which was computed there to this same accuracy
using Eq. (2.11). Notice that the the term responsible for the power behavior in Q2 has
cancelled together with the simple pole.
4.3. Integration over the coupling
Generalizing Eqs. (4.8) and (4.9) to include two- and higher loop effects would appear
to be nontrivial with the techniques described so far. In fact, even at two loops the RG
equation for cannot be solved in terms of elementary functions without approximations.
There is, fortunately, no need to do that: one can change variables according to
ds
d
=
,

(s )

(4.12)

i.e., use the coupling directly as integration variable, without the need to introduce an
explicit expression for s (2 ) in the intermediate stages. At one loop
ds
1
d2
=
.
2
s  + b0 s
4

(4.13)

One recognizes yet another way to characterize the Landau singularity in dimensional
regularization: it arises because the one-loop function with  6= 0 does not have a

L. Magnea / Nuclear Physics B 593 (2001) 269288

283

(double) zero at s = 0, as usual, but has two distinct zeroes, one at the origin and one
located at s = 4/b0 . This second zero will be on the integration contour for real
negative , provided || < b0 s (Q2 )/(4), as expected from the earlier calculations. For
 < 0, this second zero, where the function is decreasing, is the one responsible for
asymptotic freedom, while at the origin in the s plane the function vanishes with
positive derivative, as it would in a QED-like theory, such as 4 in d = 4. In fact, this
kind of behavior of dimensionally regularized theories (with the opposite sign of the
function) is familiar in statistical field theory [18], where dimensional continuation is used,
for example, to study the properties of the (nontrivial) 4 theory in d = 3 starting from the
(trivial) case d = 4. In the present case, one can see from the explicit solution, Eq. (2.7),
that below the critical point  = b0 s (Q2 )/(4) the coupling decreases smoothly to 0,
starting from the boundary value s (Q2 ). Above the critical point, on the other hand (and
thus getting closer to the physical value  = 0), the coupling develops a Landau pole, so
that it becomes impossible to evolve smoothly along real values of the scale.
Implementing the change of variables Eq. (4.12), one sees that the only remaining
dependence on the scale 2 is in the upper limit of integration of the anomalous dimension
integral, which is now s (2 ), and in the counterterm function K(, s (2 )). These two
dependences must (and do) cancel by RG invariance, in the same way in which the
singularities at 2 0 cancelled in the previous versions of the calculation. At the lower
limit of integration, 2 = 0, one must use s (0) = 0, in accordance with the arguments
given above and at the beginning of Section 4; one then recovers directly Eq. (4.9). This
third procedure of integration is clearly, and by far, the most straightforward; furthermore,
it is readily generalizable to to more than one loop, as I will now show.

5. Two-loop resummation
The generalization of the procedure outlined in Section 4.3 to include higher orders in
the function is straightforward. At l loops, (s ) is a polynomial of degree l + 1, thus
Eq. (2.10) will be expressed in terms of a double integral of a combination of rational
function of s , which in general will be computable by partial fractioning in terms of
combinations of polylogarithms. The result will be expressed in terms of the perturbative
coefficients of the functions (s ), K (s ) and G(1, , ), which are known to two loops;
therefore I will briefly illustrate the procedure and the results in that case, where all the
ingredients of the final expression are explicitly known, and the size remains manageable.
At two loops one must use, for each integral over the scale of the coupling in Eq. (2.10),
the change of variables
ds
d2
=
2
s  +

1
b0
4 s

b1 2

4 2 s

(5.1)

Partial fractioning of the denominators is achieved by introducing the two nontrivial zeroes
of the two-loop function, or, for convenience, the two zeroes of the equation b1 2 +
b0 + 4 = 0 (with = s / ),

284

L. Magnea / Nuclear Physics B 593 (2001) 269288

b0
a =
1
2b1

!
16b1
1
.
b02

(5.2)

Note that it will be possible to verify that the result reduces to the correct limit at one loop
by using the fact that, as b1 0, a becomes the one-loop zero, a 4/b0 , while a+
diverges as a+ b0 /b1 .
The cancellation of the dependence on s (2 ) between the counterterm function K and
the anomalous dimension integral takes place as usual. The final result can be written in
the form







2
s (Q2 )
2
(1)
(2)
G () + a+ G () log 1
log 1, s (Q ),  =
b1 (a+ a )
a+



2) 
2 


1
(Q

s
s (Q )
log2 1
+
+ 2 K(1) + a+ K(2) Li2
4
a+
2b1a+ (a+ a )
a+
 



2
a+ s (Q )
a+
1
Li2
Li2

b1 a (a+ a )
(a+ a )
a+ a

 

s (Q2 ) a
s (Q2 )
(5.3)
+ (a+ a ).
log
+ log 1
a+
(a+ a )
It is fairly straightforward to check that Eq. (5.3) reduces to Eq. (4.9) when one
takes {b1 , K(2) , G(2)()} 0. On the other hand, the analytic structure of Eq. (5.3) is
considerably more complicated than the one briefly discussed at one loop. In particular,
the presence of a second nontrivial zero in (s ) introduces a new branch point, so that the
structure of cuts becomes more complicated. One does not expect, however, the behavior
physical quantities related to the form factor to be qualitatively affected by the inclusion of
two-loop effects, since these are understood to modify the running of the coupling by an
amount which is a logarithmically suppressed and smooth function of the energy.
It is interesting to study the behavior of the two-loop resummed form factor, Eq. (5.3),
in the neighborhood of the physical limit,  0. One finds that, as expected, the simple
pole in the logarithm of the form factor given in Eq. (4.10) is not affected by the inclusion
of two-loop effects. The logarithmic singularities, however, are enhanced to log2 strength,
and also the constant terms in the limit  0 receive new nontrivial contributions. One
finds



(1)
(1)


K (2) 2b1K
s (Q2 )b1
4
2
2
+
log
1+
log 1, s (Q ),  =
b0 
b0 s (Q2 )
b0
b03





(2)
(1)
(1)
2K
2K
2b1K
s (Q2 )b1
4
2
(1)
1+

G0 +
log
+
b0
b0
b0
s (Q2 )b0
b0 s (Q2 )
b02

 


(2) 
(2)
(2)
2
2G0
4K
4
s (Q )b1
s (Q2 )b1

+
log
1
+
+ K2 Li2
b0 + s (Q2 )b1
s (Q2 )b0 b1
b1
b0
b0
+

4K(1)
s (Q2 )b02

+ 1 2 (2)

4b1 K(1)
b03

1 (2)

4K(2)
b02

+ O(,  log ),

(5.4)

L. Magnea / Nuclear Physics B 593 (2001) 269288

285

(2)
(i)
where G(i)
0 = lim0 G (). Also in this limit, it is easy to check that taking {b1 , K ,
(2)
G ()} 0 one gets back Eq. (4.10).
Although it is not proved here, one can conjecture rather confidently that the simple
pole in log (Q2 ) is completely determined at one loop, and receives no corrections from
higher orders in . The presence of the pole is in fact tightly connected with the one-loop
zero of the function, which is located at a distance O() from the origin. All higher-order
zeroes have locations in the s plane that are independent of  in the small  limit, thus
they are not expected to introduce further singular behavior in this region. Logarithmic
singularities, on the other hand, arise both from the anomalous dimension integral and
from the integration of the function G, thus one must expect that they get higher order
corrections.

6. Summary and perspectives

I have discussed the quark form factor in the context of dimensionally regulated
perturbative QCD. Building upon the results of Ref. [7], as well as upon the understanding
developed previously by many authors [16], I have been able to derive resummed analytic
expression for the form factor that are independent of the choice of renormalization
scale, and can be systematically improved upon by including higher orders in the QCD
function.
The main ingredient in the derivation is the consistent usage of dimensional regularization as analytic continuation to general complex values of the spacetime dimension,
d = 4 2. It turns out that this continuation does not only regulate ultraviolet (as well as
infrared) singularities in Feynman integrals: it also regulates singularities such as the Landau pole, which arise upon resummation of classes of Feynman diagrams, and are usually
interpreted as signals of nonperturbative physics. How this happens is well understood,
and this understanding has been applied as a technical tool in statistical field theory for a
long time: for d > 4, the QCD function develops an asymptotically free fixed point at a
distance O() from the origin in the s plane, while the origin itself becomes infrared free;
the coupling vanishes like a power of the renormalization scale, and it does so smoothly for
sufficiently large d; when d gets close to the physical value d = 4 the coupling develops a
Landau pole for real values of the scale; this pole is then found on the integration contour
of resummed expressions for QCD amplitudes, which, as a consequence, develop cuts with
computable branching points and discontinuities.
The versatility of dimensional regularization was exploited here also to derive compact
resummed analytic expressions for the counterterm function K(, s ): the fact that the
renormalization group equation for a pure counterterm (with a finite anomalous dimension)
turns into a recursion relation is well known; the fact that for the quark form factor the
recursion relation is completely solvable to all orders in , and the resulting towers of poles
can be summed, provided a strong motivation to pursue the calculation for the complete
form factor.

286

L. Magnea / Nuclear Physics B 593 (2001) 269288

The calculations presented in this paper are not directly applicable to physical processes:
the paper, after all, deals with a divergent quantity 3 . There are however several lines of
enquiry that are opened by these results, and I believe are worth pursuing.
First of all, it would be of great interest to extend results such as Eq. (4.9) and
Eq. (5.3) to more general QCD amplitudes, in particular to amplitudes with more than
two colored particles. This is not straightforward, because more complicated amplitudes
have a nontrivial tensor structure in color space, and evolution equations like Eq. (2.2)
correspondingly turn into matrix equations. It should be emphasized however that much of
the necessary technology has already been developed in the context of the resummation of
large-x logarithms for the production of heavy quark pairs and other colored final states in
hadronic collisions [10]. In that case one resums large corrections due to the presence of +
distributions at each order in perturbation theory, arising from the imperfect cancellation
of real and virtual contributions. In the present case one would be interested in purely
virtual corrections contributing to the cross section with terms proportional to functions.
I believe that a generalization of the present results in this direction should be possible, and
work to achieve it is under way.
A more ambitious goal would be to attempt the construction of finite, physical quantities
to the same accuracy achieved here for the form factor. This would require controlling the
infrared/collinear singularities of dimensionally regulated real emission amplitudes to all
orders. While this is not likely to be possible in a general multi-scale process, progress
might conceivably be made in the simplest cases, such as the inclusive DrellYan cross
section, by introducing suitable approximations to the d-dimensional cross section.
On a more speculative note, one may wonder whether the analytic structure uncovered
in the present paper may shed some light on nonperturbative features of QCD amplitudes.
It was already noticed in the comments to Eq. (4.10) that in the neighborhood of  = 0
one finds terms that translate into powers of the ratio Q2 /2 . More generally, the analytic
features of the form factor near the Landau cut might provide suggestions to build models
for the nonperturbative behavior of simple QCD amplitudes, as well as for the behavior of
the coupling itself, much as renormalon calculations did [12,19].
In the context of finite order perturbative QCD calculations, the existence of genuinely
RG-invariant expressions such as Eqs. (4.9) and (5.3) might prove useful as a model for the
renormalization scale dependence of more complicated QCD amplitudes. Not surprisingly,
the results for the form factor are naturally expressed in terms of s (Q2 ), since Q2 is the
only perturbative scale in the amplitude. One can however go back to equations such as
(4.8), choose a different renormalization scale, say = Q/2, reexpand the form factor to
a finite perturbative order, and study the variations of the answer with .
Finally, it should be mentioned that large corrections to perturbative QCD amplitudes
due to analytic continuation (the 2 terms) do not arise only in quark-initiated process.
A case in point is Higgs production via gluon fusion, in an effective theory in which the
top quark has been integrated out [20]. Also in that case the present techniques may well
3 The modulus squared of Eq. (4.11) is an exception, as it enters directly the resummed DrellYan cross section;
it was, however, computed already in Ref. [7].

L. Magnea / Nuclear Physics B 593 (2001) 269288

287

prove useful and provide a very accurate and rigorous evaluation of the impact of these
terms to all orders in the coupling.
In summary, I believe that the results presented here should be of interest for the study
of several aspects, both formal and practical, of perturbative QCD. Much work remains to
be done.

Acknowledgements
It is a pleasure to thank George Sterman for getting me started into this field a long time
ago, as well as for early discussions of this work. I would also like to thank Michele Caselle
for his insights concerning the use of dimensional regularization in statistical field theory.

References
[1] V.V. Sudakov, Sov. Phys. JETP 3 (1956) 65.
[2] R. Jackiw, Ann. Phys. 48 (1968) 292;
P.M. Fishbane, J.D. Sullivan, Phys. Rev. D 4 (1971) 458.
[3] Yu.L. Dokshitzer, D.I. Dyakonov, S.I. Troyan, Phys. Rep. 58 (1980) 271;
J.M. Cornwall, G. Tiktopoulos, Phys. Rev. D 13 (1976) 3370;
J. Frenkel, J.C. Taylor, Nucl. Phys. B 116 (1976) 185;
E.C. Poggio, G. Pollack, Phys. Lett. B 71 (1977) 135.
[4] A.H. Mueller, Phys. Rev. D 20 (1979) 2037.
[5] J.C. Collins, Phys. Rev. D 22 (1980) 1478.
[6] A. Sen, Phys. Rev. D 24 (1981) 3281;
G.P. Korchemsky, A.V. Radyushkin, Nucl. Phys. B 283 (1987) 342;
G.P. Korchemsky, Phys. Lett. B 220 (1989) 629.
[7] L. Magnea, G. Sterman, Phys. Rev. D 42 (1990) 4222.
[8] G. Sterman, Nucl. Phys. B 281 (1987) 310;
S. Catani, L. Trentadue, Nucl. Phys. B 327 (1989) 323;
L. Magnea, Nucl. Phys. B 349 (1991) 703;
H. Contopanagos, E. Laenen, G. Sterman, Nucl. Phys. B 484 (1997) 303, hep-ph/9604313.
[9] G. Parisi, R. Petronzio, Nucl. Phys. B 154 (1979) 427;
J.C. Collins, D. Soper, Nucl. Phys. B 193 (1981) 381;
J.C. Collins, D. Soper, Nucl. Phys. B 213 (1983) 545, Erratum;
J.C. Collins, D. Soper, G. Sterman, Nucl. Phys. B 250 (1985) 199.
[10] S. Catani, M.L. Mangano, P. Nason, L. Trentadue, Nucl. Phys. B 478 (1996) 273, hepph/9604351;
N. Kidonakis, G. Sterman, Nucl. Phys. B 505 (1997) 321, hep-ph/9705234;
R. Bonciani, S. Catani, M.L. Mangano, P. Nason, Nucl. Phys. B 529 (1998) 424, hepph/9801375;
N. Kidonakis, Int. J. Mod. Phys. A 15 (2000) 1245, hep-ph/9902484.
[11] G. Parisi, Phys. Lett. B 90 (1980) 295.
[12] M. Beneke, Phys. Rep. 317 (1999) 1, hep-ph/9807443.
[13] J.C. Collins, Sudakov form-factors, in: A.H. Mueller (Ed.), Perturbative Quantum Chromodynamics, World Scientific, Singapore, 1989, p. 573.
[14] G.P. Korchemsky, G. Marchesini, Nucl. Phys. B 406 (1993) 225, hep-ph/9210281.
[15] J. Kodaira, L. Trentadue, Phys. Lett. B 112 (1982) 66.

288

L. Magnea / Nuclear Physics B 593 (2001) 269288

[16] S. Catani, Phys. Lett. B 427 (1998) 161, hep-ph/9802439.

[17] L. Magnea, G. Sterman, in: Fermilab 1990, Proceedings, Workshop on Hadron Structure
Functions and Parton Distributions, World Scientific, 1990, p. 423.
[18] K.G. Wilson, M.E. Fisher, Phys. Rev. Lett. 28 (1972) 234;
for context, see, e.g., G. Parisi, Statistical Field Theory, AddisonWesley, 1988.
[19] See, e.g., Yu.L. Dokshitzer, G. Marchesini, B.R. Webber, Nucl. Phys. B 469 (1996) 93, hepph/9512336;
M. Beneke, V.M. Braun, L. Magnea, Nucl. Phys. B 497 (1997) 297, hep-ph/9701309.
[20] S. Dawson, Nucl. Phys. B 359 (1991) 283;
M. Kramer, E. Laenen, M. Spira, Nucl. Phys. B 511 (1998) 523, hep-ph/9611272;
E. Laenen, in: Proceedings of the 32nd Rencontres de Moriond, Les Arcs, France, March 1997,
hep-ph/9705385.

Nuclear Physics B 593 (2001) 289310

www.elsevier.nl/locate/npe

Leading twist asymmetries in deeply virtual

Compton scattering
A.V. Belitsky a, , D. Mller a,b , L. Niedermeier b , A. Schfer b
a C.N. Yang Institute for Theoretical Physics, State University of New York at Stony Brook,

NY 11794-3840, Stony Brook, USA

b Institut fr Theoretische Physik, Universitt Regensburg, D-93040 Regensburg, Germany

Received 15 June 2000; accepted 7 September 2000

Abstract
We calculate spin, charge, and azimuthal asymmetries in deeply virtual Compton scattering at
leading twist-two level. The measurement of these asymmetries gives access to the imaginary and real
part of all deeply virtual Compton scattering amplitudes. We note that a consistent description of this
process requires taking into account twist-three contributions and we give then a model dependent
estimate of these asymmetries. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.10.Hi; 12.38.Bx; 13.60.Fz
Keywords: Deeply virtual Compton scattering; Asymmetries; Skewed parton distribution

1. Introduction
The scattering of electroweak probes off hadrons serves as a clean tool, free of
complications from hadronhadron reactions on both the theoretical and experimental
sides, for the extraction of reliable information on the substructure of strongly interacting
particles. Using the photon as a probe (the absorptive part of) the forward virtual Compton
(VC) process (q)N(P ) (q)N(P ) allows to study the strong interaction dynamics
and at the same time it has a simple QCD description in the hard regime when q 2 
m2hadr . In more general settings it is instructive to address the non-forward scattering
() (q1 )N(P1 ) () (q2 )N(P2 ). The systematic approach to its calculation is established
only for the deep inelastic domain, where the QCD factorization theorems separate short
and long distance phenomena into a perturbative parton subprocess and soft functions
which encode information about the strongly coupled regime. The latter, known as the
* Corresponding author.

E-mail address: [email protected] (D. Mller).

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 8 8 - 5

290

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

skewed parton distributions (SPDs), being studied for some time [1], have attracted
increased attention in light of the conceivable opportunity to learn more about the spin
structure of the nucleon [2]. They are also of interest in their own right being hybrids
of parton densities/distribution amplitudes and form factors. They share properties of the
former in different regions of phase space [3] which have been studied perturbatively to a
great extent at one and two-loop orders [4]. Deeply virtual Compton scattering (DVCS),
with q22 = 0, proves to be an experimentally accessible reaction [5]. In electroproduction
processes of a real photon there is a strong contamination of DVCS by the BetheHeitler
(BH) process. In view of the extreme interest to extract, or at least to constrain, SPDs it
is timely to address the question of the best observables that allow to get rid of unwanted
background. Fortunately, the interference of the two processes provides a rich source of
information. It was suggested that diverse asymmetries [2,610] can be used to disentangle
the real and imaginary parts of DVCS and thus give access to SPDs. In the present
contribution we consider a number of spin, azimuthal and charge asymmetries which
share these properties and give predictions for the kinematics of the HERA and HERMES
experiment. Spin asymmetries make it possible to extract the imaginary part of the DVCS
amplitude and thus, due to the reality of SPDs, which holds owing to the spatial and time
reversal invariance of strong interactions, give directly a measurement of the shape (at
leading order in s in complete analogy to DIS) of SPDs on the diagonal t = .
The paper is organized as follows. In Section 2 we calculate the squared amplitude
for DVCS, BH, and the interference term to leading twist-two accuracy. In Section 3 we
present simplified formulae for different cross sections and give a qualitative discussion of
the feasibility to measure the DVCS leading twist-two amplitudes in different kinematical
regions. After introducing models for SPDs, we give an estimate for different asymmetries
for HERA and HERMES kinematics in Section 4. Finally, in Section 5 we summarize.

2. Cross section
To start let us discuss different contribution to the differential cross section of the
electroproduction process e(k, )N(P1 , S1 ) e(k 0 , )N(P2 , S2 ) (q2 , ), given by the
standard formula
d =

1
|T |2 (, S1 )(2)4 4 (k + P1 k 0 P2 q2 )
4k.P1
d 3P 2
d 3q2
d 3k0
.

0
3
3
2 (2) 2E2 (2) 2(2)3

(1)

The scattering amplitude squared |T |2 in the cross section, contains beside the VCS
(Fig. 1a) and BH (Fig. 1b and crossed contribution) parts also the interference term:
X 

+ TVCS
TBH .
|TVCS |2 + |TBH |2 + TVCS TBH
(2)
|T |2 (, S1 ) =
0 ,S2 ,

The BH-amplitude is purely real and is given as a contraction of the leptonic tensor, at

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

291

Fig. 1. The virtual Compton scattering amplitude and the BetheHeitler process.

leading order in the fine structure constant ,



0 , 0 ) (/k
/ )1 + (/k0 +
/ )1 u(k, ),
L = u(k
with the hadronic current



U (P1 , S1 ),
J = U (P2 , S2 ) F1 (2 ) + iF2 (2 )
2M
where

(3)

(4)

= P2 P1 = q1 q2 ,
parametrized in terms of Dirac, F1 , and Pauli, F2 , form factors normalized according to
p
p
F1 (0) = 1, F2 (0) p = 1.79, and F1n (0) = 0, F2n (0) n = 1.91, for proton and
neutron, respectively. Thus, the BH amplitude is of the form
e3
 L J .
(5)
2
The form factors are known fairly well from experimental measurements and can be parametrized by dipole formulae in the small 2 region


2 2
p
p
,
GE (2 ) = (1 + p )1 GM (2 ) = n1 GnM (2 ) = 1 2
mV
TBH =

GnE (2 ) = 0,

(6)
2

where we have introduced the electric, GiE (2 ) = F1i (2 ) 4M 2 F2i (2 ), and magnetic,
GiM (2 ) = F1i (2 ) + F2i (2 ), form factor characterized by cutoff mass mV = 0.84 GeV,
see, e.g., [13]. The hadronic tensor TVCS is:

e3
for e ,
0
) u(k), where
(7)
TVCS = 2  T u(k
+ for e+ .
q1
It is defined by the time ordered product of two electromagnetic currents
Z
T (q, P1 , P2 ) = i dx eix.q hP2 , S2 |Tj (x/2)j (x/2)|P1, S1 i,

(8)

where q = (q1 + q2 )/2 (and the index refers to the outgoing real photon). It contains
for a spin-1/2 target twelve 1 independent kinematical structures [6]. In this paper we
1 12 = 1 3 (virtual photon) 2 (final photon) 2 (initial nucleon) 2 (final nucleon). The reduction factor
2
1/2 is a result of parity invariance.

292

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

restrict ourselves to the twist-2 part of T that does not contain transversal photon spin
flip contributions. Such contributions arise in the next-to-leading order of perturbation
theory due to the gluon transversity [7,11] and are separately considered in [12]. From
the structure of the OPE we immediately learn that these contributions are contained in the
following form factor decomposition: 2
q V1
A
i q 1 + ,
(9)
P .q
P .q
where P = P1 + P2 and the gauge invariant tensors t = P t P are constructed by
means of the projection tensor P g q1 q2 /q1 .q2 . The ellipsis indicate twist-3
and higher contributions. The vectors Vi and axial-vectors Ai can be expressed by a
form factor decomposition


i

(10)
U (P1 , S1 ) + ,
V1 = U (P2 , S2 ) H1 + E1
2M


e1 5 + Ee1 5 U (P1 , S1 ) + ,
(11)
A1 = U (P2 , S2 ) H
2M
where higher twist contributions are neglected. These form factors depend on the following
variables
T (q, P , ) = g

Q2
,
P .q

1
Q2 = q 2 = (q1 + q2 )2 ,
4

2 .

Note that in general (for off-shell final photons) a second scaling variable =

.q
P .q

appears,

2
)
4Q2

. The amplitudes are given as

which is related however to , i.e., = (1
R
convolution in t, dt, of perturbatively calculable hard scattering parts with SPDs:
 
 
H
H1
(12)
(, Q2 , 2 ) = T1 (, Q2 , 2 , t)
(t, , 2 , 2 ),
E1
E
 e
e 
H1
2
2
e1 (, Q2 , 2 , t) H (t, , 2 , 2 ),
,

)
=
T
(13)
(,
Q
e
Ee1
E
with summation over the different parton species implied and 2 being the factorization
scale. The hard scattering amplitudes are available in next-to-leading order (NLO)
approximation and they read in LO for a quark of charge Qi
Q2i
(t t),
(14)
1 t/ i
with (+) for parity even (odd) cases.
In the consequent presentation we give our results in the laboratory frame (see Fig. 2) in
which we use the kinematical variables
T i(0) (, t) =

k = (E, 0, 0, E),

k 0 = (E 0 , E 0 cos e sin e , E 0 sin e sin e , E 0 cos e ),

P1 = (M, 0, 0, 0), P2 = (E2 , |P 2 | cos N sin N , |P 2 | sin N sin N , |P 2 | cos N ).

(15)
2 We adopt throughout the conventions of [14], e.g.,  0123 = 1.

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

293

Fig. 2. The kinematics of the reaction e(k)N(M) e(k 0 )N(P 2 ) (q 2 ) in the rest frame of the target.

Furthermore, we introduce the azimuthal angle r = N e between the lepton and

hadron scattering planes as well as s = N + e . The spin vector of the spin-1/2 target
for longitudinal and transverse polarization is given by
S = (0, 0, 0, ) with = 1,

S = (0, cos , sin , 0),

(16)

respectively.
From the experimental point of view one works with the variables Q2 q12 and x
q12 /(2P1 .q1 ), which are related to our variables by
2


x 1 2
2 
1

1
x
2q
1
.
(17)
=

Q2 = q12 1 2 Q2 ,
2

2
2
2x
2q1
2x 1+ 2
q1

After performing the phase space integration we obtain for the differential cross section (1)
 


y
3 xy 2
d
4M 2 x 1/2  T 2
d
= 2
=
1+
 e3  , (18)
dx dQ2 d|2 | dr
Q dx dy d|2 | dr
8Q4
Q2
where we introduced as well the conventional variable y = P1 .q1 /P1 .k [y = 1 E 0 /E in
the frame (15)].
Here we present simple analytical expressions for the amplitudes entering the cross
section for positron beam. Changing to electrons will generate a minus sign for the
interference terms. To deduce them we present at first the amplitudes squared in terms of
the form factors. The result for the DVCS amplitude reads in leading twist-2 approximation
(we set |e| = 1)


2 2y + y 2 2
q.V1 q.V1 + q.A1 q.A1
2
6
y
Q


(2 y) 2
q.A1 q.V1 + q.V1 q.A1 ,
+8
y
Q6

|TDVCS |2 = 8

(19)

where = 1 means that the spin of the lepton is parallel to the beam direction. It is obvious
that the leading contribution scales as 1/Q2 .
The exact squared amplitude for BH reads in terms of the electromagnetic currents:

294

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

|TBH |2 =

8 q.J k.J + k.J q.J q.J q.J 2k.J k.J

2
(2k. 2 )(Q2 + 2k.)


Q4 + 4
4
J.J
4 1+

2(2k. 2 )(Q2 + 2k.)



4i (Q2 + 4k. 2 )qJ J + 2 k(2q+)J J
,
+
4 (2k. 2 )(Q2 + 2k.)

(20)

with the latter being defined in Eq. (4). Instead of the exact expression we may use an
approximated one. To be consistent we have to expand the squared BH-amplitude, which
starts with 1/2 , and the interference term up to the same order as the squared DVCSamplitude, namely to 1/Q2 accuracy. However, one has to take into account that the lepton
propagator in the u-channel of the BH amplitude gives a contribution which behaves as
2
(k q2 )2 = (1y)
y Q (1 + O(1/Q)). Therefore, the Taylor expansion is not legitimate for
large y and sets, otherwise, an upper limit for y, namely 1 y  M 2 /Q2 . To avoid this
problem, we have to expand the propagator,
s
s
(

2min p
Q2
2
2
1

y
1 x cos(r )
1y +2
(k q2 ) =
y
Q2
2



2min
2 1 y
+ (1 x) 1 2 2
+
Q2
2

)



2min
3
(1 y) x + 2(1 x) 2 cos(2r ) + O(1/2 ) , (21)

and the remaining parts separately in Taylor series which results in a Pade-like approximation for the squared BH term. Here 2min is the minimal value of 2 which is defined by
the kinematical restriction 2min = M 2 x 2 /(1 x + xM 2/Q2 ). Taking only the first terms
in the expansion, thus, neglecting all contributions formally suppressed by 1/Q, we get


2(2 2y + y 2 ) J.J
q.J q.J
(2 y)y 2 ikJ J
2
+
4
,
(22)

4
|TBH | =
(1 y)
(1 y) 4 Q2
4
2 Q4
which, however, is valid for y  1.
For interference terms we adhere to the same approximation and give here only the
leading contributions in 1/Q, i.e.,



2 2y + y 2
1
q.J

q
= 4

+
2
k
J
q.V1
TBH TDVCS

1y
y
2 Q4
Q2

q.A
2ikqJ 21
Q



1
q.J
(2 y)y

+
2
k
J
q.A1
4

1 y 2 Q4
y
Q2

q.V1
2ikqJ 2 .
(23)
Q

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

295

Note that for a consistent expansion up to order 1/Q2 also twist-three contributions having
both kinematical and dynamical origins must be considered (see Ref. [15] for the case of a
scalar target). In general these contributions are poorly understood and have to be studied
in more detail.
As a next step for the calculation of the cross section we have to sum over the spin
of the outgoing proton and write the final answer in terms of SPDs and electromagnetic
form factors. Here we give the results for polarized spin-1/2 target, where = 1 means
polarization along the lepton beam, averaged over the azimuthal angle s . The DVCS
amplitude squared |TDVCS |2 = |TDVCS |2unp + |TDVCS |2pol , with pol = {LP, TP}, consists of
the elements
|TDVCS |2unp =

2(2 2y + y 2 )
y 2 (2 x)2 Q2





e x 2 H1 E + E1 H + H
e
e1 H
e1 Ee + Ee1 H
4(1 x) H1 H1 + H
1
1
1
1
1


2 
2
2
2

2 e e
E1 E1 ,
E1 E1 x
x + (2 x)
4M 2
4M 2

(24)

2(2 y)
y(2 x)2 Q2





e + H
e
e1 H x 2 H1 Ee + Ee1 H + H
e1 E + E1 H
4(1 x) H1 H
1
1
1
1
1
1

|TDVCS |2LP =


x
|TDVCS |2TP

2
1 2
x + (2 x)
2
4M 2

E1 Ee1 + Ee1 E1



s
s

2min
8 cos( r /2)(2 2y + y 2 ) 1 x 2
=
1

y 2 (2 x)2 Q2
M2
2



(2 x) (Re H1 )(Im E1 ) (Im H1 )(Re E1 )









e1 Re Ee1
e1 Im Ee1 Im H
x Re H
s
s

2min
2
2 sin( r /2)(2 y) 1 x

+
y(2 x)2 Q2
M2
2


e E1
e1 E + H
2x H1 Ee1 + H1 Ee1 2(2 x) H
1
1


2

+ x E1 Ee1 + E1 Ee1 .

For the BH term we use an analogous decomposition with

"


2min
2(2 2y + y 2 ) 1 x
2
4
1

F12 + 2(F1 + F2 )2
|TBH |unp =
(1 y)2
x2
2
 2
 #

+
1 F22 ,
2min
"
#


2min
4(2 y)y
1x
2
(F1 + F2 ) 2
1
F1 + F1 + F2 ,
|TBH |LP =
(1 y)2
x
2

(25)

(26)

(27)
(28)

296

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

|TBH |2TP

8 sin( r /2)(2 y)y 1 x

=
(1 y)xM(2)3/2


2xM 2 F1 + 2 F2 .

s
1

2min
(F1 + F2 )
2
(29)

For a polarized lepton beam the interference term is decomposed into a contribution

+
for an unpolarized and an additional one for a polarized target, i.e., I 2 TBH TDVCS

2
2
TDVCS TBH = Iunp () + Ipol (), where
s

2) 1 x
2
8(2

2y
+
y
2
p
() =
1 min
cos(r )
Iunp
2
1 yyx 2 Q2


x
2
e
(F1 + F2 )H1
F2 E1
Re F1 H1 +
2x
4M 2
s

2
8(2 y) 1 x
p
1 min
sin(r )

2
1 yx 2 Q2


2
x
e1 F2 E1 ,
(F1 + F2 )H
Im F1 H1 +
2x
4M 2
s

2
8(2 2y + y 2 ) 1 x
2
p
1 min
sin(r )
ILP () =
2
1 yyx 2 Q2
(


x
x
(F1 + F2 ) H1 + E1
Im
2x
2

 )
2
x
x

e1
F1 +
F2 Ee1
+ F1 H
2x 2
4M 2
s

2
8(2 y) 1 x
p
1 min
cos(r )
+
2
1 yx 2 Q2
(


x
x
e1
(F1 + F2 ) H1 + E1 + F1 H
Re
2x
2

 )
x
x
2
F1 +

F2 Ee1 ,
2x 2
4M 2

2
()
ITP

"


2
8(2 2y + y 2 )
cos ( 3r /2)
p
=
1 min
(1 x)
2
2
1 yyx(2 x) Q2 M 2





e1 (2 x)F1 xF2 E1 xF1 Ee1
Im 2F2 H1 + H

cos ( + r /2) 2min
(F1 + F2 )
+
2
2





e1 x 2 E1 Ee1
Im 4(1 x) H1 H

(30)

(31)

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

297




2
e1 )
+ 1 min
Im
2(1 x)F2 (H1 H
2

#


(2 x)F1 + xF2 E1 + x(F1 + xF2 )Ee1


"


2min
sin ( 3r /2)
8(2 y)
p
1
(1 x)
+
2
2
1 y(2 x)x Q2 M 2





e1 (2 x)F1 xF2 E1 xF1 Ee1
Re 2F2 H1 + H

sin ( + r /2) 2min
(F1 + F2 )

2
2





e1 x 2 E1 Ee1
Re 4(1 x) H1 H





2min
e1 [(2 x)F1 + xF2 ]E1

H
H
Re
2(1

x)F
+ 1
2
1
2
#

+ x(F1 + xF2 )Ee1
.
(32)

3. Extraction of leading twist-two amplitudes

A strong motivation for the measurement of DVCS arises from the fact that the second
moment of SPDs in the parity even sector is related to form factors appearing in the
decomposition of the symmetric QCD energy-momentum tensor . A gauge-invariant
decomposition of the matrix element of the angular-momentum tensor, M, = x
x , provides, therefore, a separate estimation of the total angular-momentum fraction
carried by quarks and gluons [2]. To achieve this goal it is necessary to interpolate the
corresponding moments of the SPDs to forward kinematics = 0. The new information
that is required (and not available from DIS) is contained in the SPDs appearing in the
spin-flip amplitude E1 . Having the spin puzzle in mind, we give special attention to the
problem of extracting E1 from measurements in different kinematical domains.
The simplified explicit expressions (24)(32) for the amplitudes squared allow us to
discuss the extraction of the leading twist-two amplitudes from future experimental data.
Moreover, they allow us to give a qualitative discussion of the ratio for the different cross
sections in more detail. Note, however, that the formulae, presented below, are only valid
in a small kinematical window and in general not sufficient for numerical estimates. In
the following we divide the kinematical region into 2 2min , |2min| < |2 | < M 2 , and
|2min | < M 2 6 |2 |. Furthermore, we separately consider the small x region.
3.1. Small x region
Let us start with the small x region, which is for instance relevant for the HERA
experiments [5]. We assume that the small x behavior of E1 and Ee1 is not one power
e1 , and, furthermore, that the SPDs in the small x region are
less than that of H1 and H

298

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

essentially determined by the usual parton densities at least in the DGLAP region. It is
easy to establish that for small values of x the ratio of the real to imaginary part of
the unpolarized amplitude H1 is about 0.3 (by means of dispersion relation the same
magnitude has been obtained in Ref. [8]). In the polarized case we find for instance for the
e1 is of order 11.7. However,
GehrmannStirling parametrization [21] that this ratio for H
the contribution of the latter is quite small as compared to H1 .
Taking into account these numbers, we find that for longitudinally polarized target
Eqs. (24), (25) can be approximated by:



2(2 2y + y 2 )
2 
2
2
2
)

)
+
(Re
E
)
(Im
H
(Im
E
(33)
|TDVCS |2
1
1
1
y 2 Q2
4M 2


4(2 y)
e1 + Re H1 Re H
e1 .
Im H1 Im H

2
yQ
2
Contributions containing Ee1 are down by two powers of x or they are proportional to x 4M
2.
The transversally polarized part offers an opportunity to measure E1 :
s
s
2 ) 2
2min
4
cos(

/2)(2

2y
+
y
r
1

|TDVCS |2TP
y 2 Q2
M2
2


(Re H1 )(Im E1 ) (Im H1 )(Re E1 )
s
s
2min
2
4 sin( r /2)(2 y)

yQ2
M2
2









e1 Im E1 + Re H
e1 Re E1 .
Im H

(34)
x 1 ,

however, since the sea quark and gluonic

The interference term starts with
contributions to the DVCS amplitudes are expected to grow with x 1 , the DVCS cross
section and the interference term have the same x dependence. For the longitudinally
polarized case it reads for small x:
r
2



8 sin(r ) 1 min
2
2
2
p
Im (2 y)y F1 H1
F2 E1
I =
4M 2
1 yyx 2 Q2


x
e1
(2 2y + y 2 ) (F1 + F2 )H1 + F1 H
2
r
2



8 cos(r ) 1 min
2
2
2
p
Re (2 2y + y ) F1 H1
F2 E1

4M 2
1 yyx 2 Q2


x
e1 .
(F1 + F2 )H1 + F1 H
(2 y)y
2
(35)
The interference term might give in future an opportunity to access the imaginary and
2
real part of the linear combination F1 H1 4M
2 F2 E1 due to the charge and single lepton

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

299

spin asymmetries for 2min < 2 . For a polarized proton beam one may extract in an
e1 ) and in combination with a polarized lepton
analogous way Im(x(F1 + F2 )H1 /2 + F1 H
beam one also gets the real part of this expression. Note that one has now the opportunity,
at least in principle, to extract the DVCS cross section (33) for unpolarized or double spin
flip experiments and thus separate H1 from the E1 contributions for the imaginary and real
part.
The squared of the BH term will generally start with x 2 (however, the spin dependent
part goes only like x 1 ) and has therefore a similar x dependence as the other ones, i.e.,
x Im H1 /Fi is of order one or so, and grows only slightly with increasing x. However, in
comparison with the squared DVCS term it is multiplied by y 2 Q2 /2 . Thus, one expects
that the BH background is not large in the small y and 2 /Q2 region. We conclude from
our discussion that the measurement of the unpolarized and longitudinal polarized cross
sections as well as single spin asymmetries it is feasible to disentangle the imaginary and
e1 , and E1 at small x and 2 of the order of one GeV2 or larger. For
real part of H1 , H
smaller values E1 is in general kinematically reduced. Since Ee1 is suppressed at least by one
power in x, we conclude that this amplitude will not be accessible in the small x region.
As mentioned, if 2 starts to be smaller, the contributions proportional to E1 die out.
In the case of its value at the kinematical boundary, i.e., 2 2min , we see that Eqs. (27)
(29) are quite simple, and, for instance, for a longitudinally polarized target we obtain
|TBH |2 =

4(2 2y + y 2 )
4(2 y)y
(F1 + F2 )2 +
(F1 + F2 )2 .
2
(1 y)min
(1 y)2min

(36)

Moreover, in this case the interference term for unpolarized or longitudinally polarized
target drops out, too. It is worth to have a closer look at the ratio of the DVCS and BH
cross section. The unpolarized cross section is essentially governed by the imaginary part
of unpolarized parton distributions:
dDVCS ( = 0, = 0) (1 y)x 2 M 2 Im H12
=
.
(37)
dBH ( = 0, = 0)
2y 2 Q2
(F1 + F2 )2
It is remarkable that the ratio of double spin flip (DF) cross sections is proportional to the
product of unpolarized and polarized parton distributions:
dDVCS ( = 1, = 1) dDVCS ( = 1, = 1)
dBH ( = 1, = 1) dBH ( = 1, = 1)
e1 + Re H1 Re H
e1
(1 y)x 2M 2 Im H1 Im H
.
(38)
=
2
2
2
y Q
(F1 + F2 )
For the sake of completeness we want also to mention that the interference term for a
transversally polarized target does not vanish for 2 2min , while both the DVCS and BH
cross section are reduced to the unpolarized ones. This interference term is proportional to
e1 :
H1 H


16(2 2y + y 2 ) cos ( + r /2)
e1
p
(F1 + F2 ) Im H1 H
I 2 () =
2
1 yyx Q2 M 2


sin ( + r /2)
16(2 y)
e1 .
p
(F1 + F2 ) Re H1 H

2
1 yx Q2 M 2

(39)

300

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

Note that at the kinematical boundary the spin flip contributions are generally suppressed
by at least a factor x 2 . Thus, in this kinematical range it is hopeless to extract E1 or Ee1 from
any experiment.
3.2. Asymmetries for |2 |  |2min|
Away from the kinematical boundary |2 |  |2min | we can consider different cross
sections to get separately information on the real and imaginary parts of the leading twisttwo amplitudes, H1 , . . . , Ee1 . Since the BH cross section depends only on the product of
the lepton and target polarization, i.e., on or f () with f ( + ) = f (), we
have
p a direct access to the interference term in polarized experiments to leading order in
1/ Q2 . Moreover, the relative sign of the interference term is determined by the charge of
the lepton beam. Thus, the following cross sections allow one to get access to the leading
twist-two structure functions in the DVCS amplitude. In the approximation used above and
neglecting terms O(2min /2 ), they read:
1. Polarized positron beam and unpolarized target:
SL d d d

16(2 y) 1 x
p
=
sin(r )
1 yx 2 Q2


2
x
e1 F2 E1 dM.
(F1 + F2 )H
Im F1 H1 +
2x
4M 2

(40)

2. Unpolarized positron beam and longitudinally polarized target:

SLN d d d

16(2 2y + y 2 ) 1 x
p
=
sin(r )
1 yyx 2 Q2



x
x
e1
(F1 + F2 ) H1 + E1 + F1 H
Im
2x
2

 
x
x
2
F1 +
+
F2 Ee1 dM.
2x 2
4M 2

(41)

3. Unpolarized positron beam and transversally polarized target: ( = {0, }):

STN d d d

16(2 2y + y 2 )
cos (3r /2)
p
(1 x)
=
2
2
2
1 yyx(2 x) Q M





e1 (2 x)F1 xF2 E1 xF1 Ee1
Im 2F2 H1 + H



cos (r /2) 
e1 (2 x)F1 + xF2 E1
Im 2(1 x)F2 H1 H
+
2


e
(42)
+ x(F1 + xF2 )E1 dM.

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

301

4. Charge asymmetry in unpolarized experiment:

C d d +
unp

unp

unp

16(2 2y + y 2 ) 1 x
p
=
cos(r )
1 yyx 2 Q2


2
x
e1 F2 E1 dM.
(F1 + F2 )H
Re F1 H1 +
2x
4M 2

(43)

5. Charge asymmetry in double spin-flip experiments with longitudinally polarized

target:

unp
DFL
C d d d C d




16(2 y) 1 x
x
x
p
(F1 + F2 ) H1 + E1
=
cos(r ) Re
2x
2
1 yx 2 Q2

 
2
x

e1 x
F1 +
F2 Ee1 dM.
(44)
+ F1 H
2x 2
4M 2

6. Charge asymmetry in double spin-flip experiments with transversally polarized

target:

unp
DFT
C d d d C d
8(2 y)
p
=
1 y(2 x)x Q2 M 2

sin (3r /2)
2(1 x)

2





e1 (2 x)F1 xF2 E1 xF1 Ee1
Re 2F2 H1 + H



sin (r /2) 
e1 (2 x)F1 + xF2 E1
Re 2(1 x)F2 H1 H
+



e
+ x(F1 + xF2 )E1 dM,
(45)

where dM =

3 xy
(1
8 Q2

4M 2 x 1/2
)
dx dQ2 d|2 | d

Q2

r.

Note that for a consequent

numerical treatment the 2min /2 dependence cannot be dropped and can be easily restored
from Eqs. (30)(32).
As we see from this list, the two single spin cross sections of longitudinally polarized
beam or target give us information on the imaginary part of two linear combinations of
the four leading twist-two amplitudes. The real part of these two quantities is accessible
via charge asymmetry in unpolarized or longitudinally double spin flip experiments.
Obviously, at low |2 |, i.e., compared to 4M 2 4 GeV2 , there is a suppression of
e SPDs, which are theoretically not well constraint:
contributions proportional to E and E
e is suppressed by x 2 .
the contribution proportional to E SPDs completely drops, while E
For the kinematics of present experiments 2 1 GeV2 , and thus it is not a good
approximation to drop afore mentioned terms, 2 /4M 2 O(1). However, in certain
asymmetries these contributions are accompanied by a factor x and thus can be safely

302

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

discarded at smaller values of x. Further constraints separately for the imaginary and
real part arise from spin and charge asymmetries for a transversally polarized target.
Fortunately, due to different angular dependences, we have indeed two further independent
linear combinations of the DVCS amplitudes. But the kinematical pre-factors and an
additional angular dependence of the denominator for large y, which we have dropped
for simplicity, makes their practical use questionable.
Finally, the whole cross section for unpolarized and polarized production tests the
leading twist-two approximation used above. However, to be consistent one has then to
expand the BH and interference terms up to order 1/Q2 . As outlined in Section 2 this is
straightforward for the BH cross section, while in the interference term new twist-three
contributions will enter. These additional terms will be worked out in a forthcoming paper.
Here we should only mention that the different azimuthal angle dependence can be used to
pick up different combinations of helicity amplitudes, entering at different twist levels. For
instance, for small y it is justified to use the derived formulae which imply that all single
spin and charge asymmetry cross sections integrated over the azimuthal angle r vanish at
twist-two level, for instance,
Z2

SL d
dr
= O 1/Q2
dr

Z2

unp

dr

or

C d
= O 1/Q2 .
dr

(46)

4. Numerical estimates
To give phenomenological predictions for the asymmetries discussed above we have
to specify models for the SPDs. Definitely, this is the main source of uncertainty for the
numerical estimates we present in this section. (Of course, the primary goal of experiments
is rather to constrain the skewed functions via the theoretical formulae.)
Let us discuss in turn spin non-flip and flip functions. In the former case SPDs have
a definite limit for forward kinematics when they reduce to the familiar parton densities.
For spin non-flip SPDs we choose an oversimplified factorized form of the 2 and (t, )
dependence of the SPDs:
H i (t, , 2 , Q2 ) = F1i (2 )q i (t, , Q2 ),
ei (t, , 2 , Q2 ) = Gi (2 )q i (t, , Q2 ),
H
1

(47)

where i denotes the quark flavour. Here F1i (2 ) and Gi (2 ) are elastic parton form
factors normalized to unity at the origin, F1i (0) = Gi (0) = 1, and q(t, , Q2) as well as
q(t, , Q2) are the non-forward functions specified below. The support of these functions
is [1, 1], where for t > 0 we have the quark distribution and for t < 0 the antiquark
, Q2 ) = q(t, , Q2 ). The
distribution, i.e., q(t,
, Q2 ) = q(t, , Q2) and q(t,
ei (t, , 2 , Q2 ) at 2 = 0, = 0 is determined by
normalization of H i (t, , 2 , Q2 ) and H
i
i
e
parton densities H (t, 0, 0) = q (t), H i (t, 0, 0) = q i (t). Since the dependence drops
out in the first moment we can constrain the 2 dependence by sum rules, e.g.,

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

Z1

303

Z1
i

dt H (t, ,

) = F1i (2 ),

dt E i (t, , 2 ) = F2i (2 ),

(48)

with the Dirac and Pauli form factor, respectively. For non-polarized SPDs the valence
(p )
u
and d quark
form factors in the proton can be easily deduced from (6) via FI n =
2
1
u
d
1 Qu FI + 2 Qd FI which results in
p

2FIu (2 ) = 2FI (2 ) FIn (2 ),

p

FId (2 ) = FI (2 ) 2FIn (2 ),

for I = 1, 2.

(49)

For s (or in general, sea) quark contribution in the parity even sector the first moment
vanishes and the sum rule (48) does not give any constraint. Nevertheless, the counting
rule for elastic form factors tells us that for large 2 we have a (2 )3 behaviour. This
suggests the following dipole fit with the mass cutoff mV chosen as for valence quarks:


1
2 3
2
sea
2
(
)
=
G
(
)
=
1

,
(50)
Gsea
E
1 + sea M
m2V
where sea will be specified below.
So far the model has been governed by the known forward densities. Unfortunately,
a similar reduction is absent for the helicity flip amplitudes. For E i we adopt nevertheless
E i (t, , 2 ) = r i (t, )F2i (2 )

with r i (t, ) = q i (t, ).

(51)

The identification of r i (t, ) with q i (t, ) ensures the sum rule for valence quark
contributions. Note that the parameter sea normalizes the sea quark contribution, for
instance, sea = 0 provides E sea = 0. In the axial vector channel the quark form factors can
(3)
be read off from the iso-triplet axial form factor G1 (2 ) of the -decay and related by
(3)
isotopic symmetry to the form factor in question. The decay constant gA is expressed by
(3)
GoldbergerTreiman relation gA 1 f gNN in terms of the pion decay constant f ,
2M

(3)

nucleon mass M and NN-coupling, gNN , and has the numerical value gA = 1.267 [13].
2
If we assume the same 2 -dependence for the iso-singlet G(0)
1 ( ) with the same cutoff
(0)
and a constant gA we get for quarks




2 2
2 3
2
, for i = {uv , dv }, and Gsea
(
)
=
1

Gi1 (2 ) = 1 2
1
mA
m2A
(52)
with the scale mA = 0.9 GeV [13].
Finally for the polarized spin-flip amplitude it was observed [16,18,19] that, similar to
e at small 2
the -decay effective pseudoscalar form factors [13], one can approximate E
by the pion pole so that one ends up with the model


ed = 1 |t| (t/ )g (2 ),
eu = E
E
2
4
and (x) = (1 x 2 ).
3

with g (2 ) =

(3) 2
M
4gA
m2 2

(53)

304

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

It is reliable to assume that the SPDs in the DGLAP region can be modelled by the
forward parton distributions measured in inclusive reactions. In the simplest case we
assume that this is the only contribution. We refer to this model as the forward parton
distribution (FPD) model which has no skewedeness dependence at the input scale Q0 :




(54)
()q i t, , Q20 = ()q i t, Q20 for t > 0.
The contributions for t < 0 are easily restored by means of symmetry. Although the
input distribution does not depend on a small evolution step does generate such a
-dependence [17]. A further model is based on an proposal for the so-called double
distribution (DD) [1], namely,
Z1
q(t, , Q ) =
2

1|x|
Z

dx

dy (x + y t)f (y, x, Q2 ).

(55)

1+|x|

The functional dependence in the x-subspace is given by the shape of the forward parton
density, f (x), while the y/[1 |x|]-dependence of the integrand has to be similar to that
of the distribution amplitude. Thus, f (y, x) is given [16] by the product of a forward
distribution f (z) (more precisely q(z) for quarks and zg(z) for gluons) with a profile
function
f (y, x) = (y, x)f (x),

(56)

where for quarks and gluons is given by


2
3 [1 |x|]2 y 2
15 [1 |x|]2 y 2
G
,
(y, x) =
.
(57)
(y, x) =
4 [1 |x|]3
16
[1 |x|]5
Now we are in a position to give numerical estimates for the cross sections defined in the
preceding sections.
Q

4.1. HERA kinematics

In the small x kinematics, 105 6 x 6 102 , first estimates for the unpolarized
azimuthal and single spin asymmetry has been given in the factorization [8] and BFKL [22]
approach. In the following we evaluate numerically different asymmetries in order to
e1 and E1 are measurable in HERA experiments. In the following
demonstrate that H1 , H
we deal with the FPD model, where we equate the SPDs with the parton densities taking
the MRS A0 [20] and the GS A [21] parametrization at the input scale Q = 4 GeV2 and
using u = d = s /2 as well as u = d = s . For simplicity we do not discuss the Q2
dependence of our predictions due to the perturbative evolution.
The unpolarized azimuthal asymmetry is defined by
R 3/2
R /2
d unp
d unp
/2 dr dr /2 dr dr
,
(58)
A=
R 2
d unp
0 dr dr
and its first signature has been seen by the ZEUS collaboration [5]. In the approximation (27) the unpolarized squared BH term is r independent. However, it turned out

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

305

Fig. 3. (a) Ratios of different approximations of the BH cross section to the exact one in dependence of the azimuthal angle r for Q2 = 4 GeV2 and x = 104 for the Pade-type approximation for 2 = 0.1 GeV2 (solid) and 2 = 0.5 GeV2 (dashed) and for the leading approximation (27) (dash-dotted and dotted, respectively). (b) Unpolarized azimuthal angle asymmetry for
2 = 0.05(0.25) GeV2 versus x. Solid (dash-dotted) and dashed (dotted) lines show the result
for the Pade and leading approximation, respectively, where sea is set to zero.

that this approximation may provide misleading results due to the neglected azimuthal
dependence of the BH process. In Fig. 3a it can be seen that the dropped terms cause a
strong r -dependence and that only for the Pade-type approximation introduced in Section 2 (see Eq. (21) for the expansion of the propagator) there is a good agreement with
the exact expression. Since the BH cross section is in general suppressed in the upper
hemisphere, one may expect quite different predictions depending on the approximations
involved. Indeed, for small values of x, where y = Q2 /xs with the center of mass energy
2
s = 4 27.5 820 GeV2 , we find a strong deviation, even
p for small . Note that the
2
2
interference term is only taken into account up to order 1/ Q and at present it is not
clear whether the 1/Q2 term is crucial or not.
2
To get information on {F1 H1 4M
2 F2 E1 } in a cleaner way one may use the azimuthal
unp
unp
unp
d +
defined in Eq. (43) (in
asymmetry of the charge asymmetry C d = d

306

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

Fig. 4. Unpolarized azimuthal charge asymmetry AC for Q2 = 4 GeV2 and x = 5 104 with
sea = 2 at LO for the complete expression (solid curve) and neglecting the E1 contribution (dashed
curve) plotted in (a) versus 2 . The same asymmetry for 2 = 0.05 GeV2 (solid curve) and
2 = 0.25 GeV2 (dashed curve) at LO is shown in (b) for the region 1 104 6 x 6 2 103 .
In (c) and (d) the same is shown for the electron single spin asymmetry.

the following we restore the 2min /2 dependence in the interference amplitudes):

R /2
AC =

R 3/2
C d
C d
/2 dr dr /2 dr dr
R 2
d unp +d + unp
0 dr
dr
unp

unp

(59)

and the single (lepton) spin asymmetry with unpolarized target SL = d d defined
in Eq. (40):
R
R 2
SL d
SL d
0 dr dr dr dr
.
(60)
ASL =
R 2
d +d
0 dr
dr
The former (later) one is proportional to the real (imaginary) part of the linear combination
that was mentioned. Both of them give sizeable effects. Although ASL is proportional to
the imaginary part which is growing as x 1 this asymmetry is suppressed by a kinematical
factor y as compared to the charge asymmetry. Therefore, we find in the considered
kinematics a two times larger value of the charge asymmetry as for the single spin one.
This should be reversed for larger values of y (see also the discussion in Ref. [10]). In
Fig. 4 we demonstrate the influence of the amplitude E1 which is quite sizeable for larger
values of 2 , where 2 /Q2 remains small.
e1 one have to consider the proton single spin asymmetry SLN = d
To measure H
d given in Eq. (41):

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

R
ASLN =

R 2
d
d
dr SLN
dr SLN
dr
dr
R 2
d +d
0 dr
dr

307

(61)

and the charge asymmetry in double spin-flip experiments, i.e., DFL

C d = d

d + C d unp , defined in Eq. (44)

R
=
ADFL
C

R 2
DFL d
DFL d
dr Cdr dr Cdr
.
R 2
d unp +d + unp
0 dr
dr

(62)

Unfortunately, for the considered Q2 value we find for our model that this single spin
asymmetry is compatible with zero. Although, we have an enhancement for small y due do
the factor 1/y in comparison to the lepton single spin asymmetry, this enhancement cannot
e1 for small x. For larger value of Q, i.e., also larger x, and
compensate the weaker rise of H
2
and small y we found a few percent effect. Note also that the ratio of proton to lepton
e1 )/(Im H1 ) times 1/y, which is sizeable at
single spin asymmetry gives us the ratio (Im H
not too small x. The azimuthal asymmetry in double spin flip experiments normalized to
the unpolarized cross section is tiny. However, if we would compare it with the extracted
double spin flip part, we would get quite sizeable effects which are of the order of 10%
or even more. Furthermore, to explore H1 separately, one can make use of the discussion
given in Section 3.1. Especially, for small y the BH cross section is suppressed by y 2 in
comparison to the DVCS one and, thus, the former one drops out.
Let us note that the perturbative NLO corrections to the imaginary part for small x are
of order of 20% for each quark species as well as for the gluon contribution.
4.2. HERMES kinematics
Now let us turn to the HERMES experiment with a E = 27.5 GeV positron beam
scattered by a hydrogen target and give predictions for the asymmetries which can be
accessed there. To give numerical predictions we take the model for SPDs deduced from the
double distribution model and forward (un-)polarized parton densities from Ref. ([23]) [24]
with sea = 2. As a starting point we choose Q2 = 4 GeV2 , x-range 0.10.4, and the tchannel momentum transfer 2 = 0.10.5 GeV2 . Since the parton densities are defined
at rather low momentum scale Q20 = 0.23 GeV2 we evolve them up to the experimental
scale using the formalism of [4]. We concentrate mostly on the cross sections proportional
e1 is concentrated in the ERBL
to the imaginary part of the DVCS amplitude. Since E
region with zero at |t| = , Im Ee1 vanishes at all order.
Similarly to the previous section we calculate the charge asymmetry for unpolarized
experiments given in Eq. (59). As we see form the Fig. 5a it reaches the level of 515%
for 2 = 0.5 GeV2 . The single lepton spin asymmetry (60) is much larger and can be as
big as 2030% (see Fig. 5b) which gives promises to measure it experimentally. Note that
in both cases the cross sections are very sensitive to the helicity flip distribution E which
is responsible for the orbital momentum of partons in the proton.

308

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

Fig. 5. Perturbative leading order results for the charge asymmetry for an unpolarized beam (a),
single spin asymmetries for a polarized positron beam (b) and an unpolarized target; as well as for an
unpolarized lepton beam and a longitudinally (c) (transversally (d)) polarized proton target versus x,
for Q2 = 4 GeV2 . The predictions for the model specified in the text are shown as solid (dotted)
curves for 2 = 0.1(0.5) GeV2 , respectively. The same model however with neglected spin-flip
contributions are presented as dashed (dash-dotted) line for the same values of 2 .

In view of limited space let us consider finally only the spin asymmetry for a
longitudinally and transversally polarized proton (42). In the former case azimuthal
averaging of Eq. (41) is the same as in Eq. (60) and leads to a sizeable asymmetry of
order 20%. In the later case in order to extract the combination of SPDs multiplied by
cos (r /2) we define the following azimuthal asymmetry
R 2/3
R 5/3
STN d
d
2/3 dr STN
/3 dr dr
dr
.
(63)
ASTN =
R 2
d +d
0 dr
dr
The numerical estimate presented in Fig. 5d demonstrates that it has a sizable effect which
in contrast to the other symmetries does not vanish at the kinematical boundary 2 = 2min .

5. Conclusions
In this paper we have given theoretical predictions for diverse asymmetries which can
be measured in exclusive leptoproduction experiments of a real photon. Our estimates
are rather encouraging, since they demonstrate a possibility to separate the contributions
coming from different leading twist-two DVCS amplitudes by means of polarized lepton
beam, and longitudinally and transversally polarized targets. This is the first step on the

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

309

way to constrain the form of the SPDs from experimental data. The models used for
numerical estimates lead to large charge and single (lepton) spin asymmetries of order
20%. However, since an asymmetry gives information on a linear combination of SPD
one has to resort to other combinations of cross sections in order to disentangle a given
distribution.
We have not discussed in the present paper the NLO correction to the DVCS amplitudes.
However, as has been shown in Ref. [25] the latter could be very sizable and therefore
change the LO predictions significantly in the valence quark region. Yet another important
issue is the study of kinematical and dynamical higher twist corrections. Both of these
problems deserve a detailed investigation and will be considered elsewhere.

Acknowledgements
We would like to thank M. Amarian, M. Diehl, and A.V. Radyushkin for constructive
correspondence. This work was supported by DFG and BMBF. D.M. thanks G. Sterman
for the hospitality extended to him at the C.N. Yang Institute for Theoretical Physics while
this work was finished.

References
[1] D. Mller, D. Robaschik, B. Geyer, F.-M. Dittes, J. Horeji, Fortschr. Phys. 42 (1994) 101.
[2] X. Ji, Phys. Rev. Lett. 78 (1997) 610;
X. Ji, Phys. Rev. D 55 (1997) 7114;
X. Ji, J. Phys. G 24 (1998) 1181.
[3] A.V. Radyushkin, Phys. Lett. B 380 (1996) 417;
A.V. Radyushkin, Phys. Rev. D 56 (1997) 5524.
[4] A.V. Belitsky, D. Mller, Nucl. Phys. B 537 (1999) 397;
A.V. Belitsky, D. Mller, L. Niedermeier, A. Schfer, Nucl. Phys. B 546 (1999) 279;
A.V. Belitsky, D. Mller, A. Freund, Nucl. Phys. B 574 (2000) 347.
[5] P.R.B. Saull (ZEUS Collaboration), Talk at EPS99, Prompt Photon Production and Deeply
Virtual Compton Scattering, Tampere, Finland; July 1999.
[6] P. Kroll, M. Schrmann, P.A.M. Guichon, Nucl. Phys. A 598 (1996) 435.
[7] M. Diehl, T. Gousset, B. Pire, J.P. Ralston, Phys. Lett. B 411 (1997) 193.
[8] L. Frankfurt, A. Freund, M. Strikman, Phys. Rev. D 58 (1998) 114001;
L. Frankfurt, A. Freund, M. Strikman, Phys. Rev. D 59 (1999) 119901, Erratum.
[9] L. Frankfurt, A. Freund, M. Strikman, Phys. Lett. B 460 (1999) 417.
[10] A. Freund, M. Strikman, Phys. Rev. D 60 (1999) 071501.
[11] P. Hoodbhoy, X. Ji, Phys. Rev. D 58 (1998) 054006.
[12] A.V. Belitsky, D. Mller, Phys. Lett. B 486 (2000) 369.
[13] L.B. Okun, Leptons and quarks, North-Holland, Amsterdam, 1982.
[14] C. Itzykson, J. Zuber, Quantum Field Theory, McGraw-Hill, New York, 1980.
[15] I. Anikin, B. Pire, O. Teryaev, Phys. Rev. D 62 (2000) 071501.
[16] A.V. Radyushkin, Phys. Rev. D 59 (1999) 014030.
[17] A.V. Belitsky, D. Mller, Nucl. Phys. B Proc. Suppl. 79 (1999) 573, hep-ph/9905263.
[18] L. Mankiewicz, G. Piller, A.V. Radyushkin, Eur. Phys. J. C 10 (1999) 307.
[19] L.L. Frankfurt, P.V. Pobylitsa, M.V. Polyakov, M. Strikman, Phys. Rev. D 60 (1999) 0140010.

310

[20]
[21]
[22]
[23]
[24]
[25]

A.V. Belitsky et al. / Nuclear Physics B 593 (2001) 289310

A.D. Martin, R.G. Roberts, W.J. Stirling, Phys. Lett. B 354 (1995) 155.
T. Gehrmann, W.J. Stirling, Phys. Rev. D 53 (1996) 6100.
I.I. Balitsky, E. Kuchina, Phys. Rev. D 62 (2000) 074004.
M. Glck, E. Reya, A. Vogt, Z. Phys. C 67 (1995) 433.
M. Glck, E. Reya, M. Stratmann, W. Vogelsang, Phys. Rev. D 53 (1996) 4775.
A.V. Belitsky, D. Mller, L. Niedermeier, A. Schfer, Phys. Lett. B 474 (2000) 163.

Nuclear Physics B 593 (2001) 311335

www.elsevier.nl/locate/npe

Light-cone representation of the spin and orbital

angular momentum of relativistic
composite systems
Stanley J. Brodsky a, , Dae Sung Hwang b , Bo-Qiang Ma c,d,e ,
Ivan Schmidt f
a Stanford Linear Accelerator Center, Stanford University, Stanford, CA 94309, USA
b Department of Physics, Sejong University, Seoul 143-747, South Korea
c CCAST (World Laboratory), P.O. Box 8730, Beijing 100080, China
d Department of Physics, Peking University, Beijing 100871, China
e Institute of High Energy Physics, Academia Sinica, P.O. Box 918(4), Beijing 100039, China
f Departamento de Fsica, Universidad Tcnica Federico Santa Mara, Casilla 110-V, Valparaso, Chile

Received 13 March 2000; accepted 17 October 2000

Abstract
The matrix elements of local operators such as the electromagnetic current, the energymomentum
tensor, angular momentum, and the moments of structure functions have exact representations in
terms of light-cone Fock state wavefunctions of bound states such as hadrons. We illustrate all
of these properties by giving explicit light-cone wavefunctions for the two-particle Fock state of
the electron in QED, thus connecting the Schwinger anomalous magnetic moment to the spin and
orbital angular momentum carried by its Fock state constituents. We also compute the QED oneloop radiative corrections for the form factors for the graviton coupling to the electron and photon.
Although the underlying model is derived from elementary QED perturbative couplings, it in fact
can be used to simulate much more general bound state systems by applying spectral integration
over the constituent masses while preserving all of the Lorentz properties, giving explicit realization
of the spin sum rules and other local matrix elements. The role of orbital angular momentum in
understanding the spin crisis problem for relativistic systems is clarified. We also prove that the
anomalous gravitomagnetic moment B(0) vanishes for any composite system. This property is shown
to follow directly from the Lorentz boost properties of the light-cone Fock representation and holds
separately for each Fock state component. We show how the QED perturbative structure can be used
to model bound state systems while preserving all Lorentz properties. We thus obtain a theoretical

Work partially supported by the Department of Energy, contract DE-AC03-76SF00515, by the National
Natural Science Foundation of China under grants No. 19605006 and No. 19975052, by Fondecyt (Chile) under
grants 1990806 and 8000017, and by a Ctedra Presidencial (Chile).
* Corresponding author.
E-mail addresses: [email protected] (S.J. Brodsky), [email protected] (D.S. Hwang),
[email protected] (B.-Q. Ma), [email protected] (I. Schmidt).

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 2 6 - X

312

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

laboratory to test the consistency of formulae which have been proposed to probe the spin structure
of hadrons. 2001 Elsevier Science B.V. All rights reserved.
PACS: 12.20.-m; 12.39.Ki; 13.40.Em; 13.40.Gp

1. Introduction
The light-cone Fock representation of composite systems such as hadrons in QCD
has a number of remarkable properties. Because the generators of certain Lorentz boosts
are kinematical, knowing the wavefunction in one frame allows one to obtain it in any
other frame. Furthermore, matrix elements of space-like local operators for the coupling
of photons, gravitons, and the moments of deep inelastic structure functions all can
be expressed as overlaps of light-cone wavefunctions with the same number of Fock
constituents. This is possible since in each case one can choose the special frame q + = 0 [1,
2] for the space-like momentum transfer and take matrix elements of plus components
of currents such as J + and T ++ . Since the physical vacuum in light-cone quantization
coincides with the perturbative vacuum, no contributions to matrix elements from vacuum
fluctuations occur [3]. Light-cone Fock state wavefunctions thus encode all of the bound
state quark and gluon properties of hadrons including their spin and flavor correlations in
the form of universal process- and frame-independent amplitudes.
Formally, the light-cone expansion is constructed by quantizing QCD at fixed light-cone
QCD
= P +P
time [4] = t + z/c and forming the invariant light-cone Hamiltonian: HLC
PE2 where P = P 0 P z [3]. The momentum generators P + and PE are kinematical;
d
generates lighti.e., they are independent of the interactions. The generator P = i d
QCD
cone time translations, and the eigen-spectrum of the Lorentz scalar HLC gives the mass
spectrum of the color-singlet hadron states in QCD together with their respective lightQCD
cone wavefunctions. For example, the proton state satisfies: HLC |p i = Mp2 |p i. The
expansion of the proton eigensolution |p i on the color-singlet B = 1, Q = 1 eigenstates
QCD
(g = 0) gives the light-cone Fock expansion:
{|ni} of the free Hamiltonian HLC
!
!
n
n
n
XY
X
X


dxi d2 kEi
3
(2)
p (P + , PE ) =
E
16 1
xi
ki

xi 16 3
n i=1
i=1
i=1



(1)
n xi , kEi , i n; xi P + , xi PE + kEi , i .
The light-cone momentum fractions xi = ki+ /P + and kEi represent the relative momentum
coordinates of the QCD constituents. The physical transverse momenta are pEi = xi PE +
kEi . The i label the light-cone spin projections S z of the quarks and gluons along
the quantization direction z. The physical gluon polarization vectors  (k, = 1) are
specified in light-cone gauge by the conditions k  = 0,  =  + = 0. The n-particle
states are normalized as



n; pi0 + , pE 0i , 0i n; pi+ , pEi , i

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

n
Y

16 3 pi+ pi0 + pi+ (2) pE 0i pEi 0i i .

313

(2)

i=1

The solutions of HLC |p i = Mp2 |p i are independent of P + and PE ; thus given the
eigensolution Fock projections hn; xi , kEi , i |p i = n (xi , kEi , i ), the wavefunction of
the proton is determined in any frame [5]. In contrast, in equal-time quantization, a Lorentz
boost always mixes dynamically with the interactions, so that computing a wavefunction in
a new frame requires solving a nonperturbative problem as complicated as the Hamiltonian
eigenvalue problem itself.
The LC wavefunctions n/H (xi , kEi , i ) are universal, process independent, and thus
control all hadronic reactions. Given the light-cone wavefunctions, one can compute
the moments of the helicity and transversity distributions measurable in polarized deep
inelastic experiments [5]. For example, the polarized quark distributions at resolution
correspond to
n
XZ Y
X  ()

2


E
dxj d2 kEj
qq /p (x, ) =
n/H xi , ki , i
QCD

n,qa

j =1

n
X
i

!
xi

(2)

n
X
i

kEi (x xq )a q 2 M2n ,
(3)

where the sum is over all quarks qa which match the quantum numbers, light-cone
momentum fraction x, and helicity of the struck quark. Similarly, moments of transversity
distributions and off-diagonal helicity convolutions are defined as a density matrix of the
light-cone wavefunctions. Applications of non-forward quark and gluon distributions have
been discussed in Refs. [6,7]. The light-cone wavefunctions also specify the multi-quark
and gluon correlations of the hadron. For example, the distributions of spectator particles
in the final state which could be measured in the proton fragmentation region in deep
inelastic scattering at an electronproton collider are in principle encoded in the light-cone
wavefunctions.
()
Given the n/H , one can construct any spacelike electromagnetic, electroweak, or gravitational form factor or local operator product matrix element of a composite or elementary
system from the diagonal overlap of the LC wavefunctions [8]. Studying the gravitational
form factors is not academic: Ji has shown that there is a remarkable connection of the
x-moments of the chiral-conserving and chiral-flip form factors H (x, t, ) and E(x, t, )
which appear in deeply virtual scattering with the corresponding spin-conserving and spinflip electromagnetic form factors F1 (t) and F2 (t) and gravitational form factors Aq (t) and
Bq (t) for each quark and anti-quark constituent [9]. Thus, in effect, one can use virtual
Compton scattering to measure graviton couplings to the charged constituents of a hadron.
Exclusive semi-leptonic B-decay amplitudes involving timelike currents such as B
A` can also be evaluated exactly in the light-cone formalism [10]. In this case, the
timelike decay matrix elements require the computation of both the diagonal matrix
element n n where parton number is conserved and the off-diagonal n + 1 n 1

314

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

convolution such that the current operator annihilates a q q 0 pair in the initial B
wavefunction. This term is a consequence of the fact that the time-like decay q 2 = (p` +
p )2 > 0 requires a positive light-cone momentum fraction q + > 0. Conversely for spacelike currents, one can choose q + = 0, as in the DrellYanWest representation of the spacelike electromagnetic form factors [8]. However, as can be seen from the explicit analysis of
timelike form factors in a perturbative model, the off-diagonal convolution can yield a nonzero q + /q + limiting form as q + 0 [10]. This extra term appears specifically in the case
of bad currents such as J in which the coupling to q q 0 fluctuations in the light-cone
wavefunctions are favored [10]. In effect, the q + 0 limit generates (x) contributions as
residues of the n + 1 n 1 contributions. The necessity for such zero mode (x) terms
has been noted by Chang, Root and Yan [11], Burkardt [12], and Choi and Ji [13]. We can
avoid these contributions by restricting our attention to the plus currents J + and T ++ .
It should be emphasized that the light-cone Fock representation provides an exact
formulation of current matrix elements of local operators. In contrast, in equal-time
Hamiltonian theory, one must evaluate connected time-ordered diagrams where the gauge
particle or graviton couples to particles associated with vacuum fluctuations. Thus even
if one knows the equal-time wavefunction for the initial and final hadron, one cannot
determine the current matrix elements. In the case of the covariant BetheSalpeter
formalism, the evaluation of the matrix element of the current requires the calculation of
an infinite number of irreducible diagram contributions.
One of the important issues in the formulation of light-cone quantized quantum field
theories is the existence of a consistent scheme for non-perturbative renormalization. A
general nonperturbative renormalization procedure for QCD has recently been outlined by
Paston et al. [14]. An alternative is to the use of broken supersymmetry as an ultraviolet
regulator [15]. Some simplified model light-cone field theories have been successfully
renormalized using generalized PauliVillars regularization [15].
As an illustration of the structure of the light-cone Fock state representation, we will
present a simple self-consistent model of an effective composite spin- 21 system based
on the quantum fluctuations of the electron in QED. The model is patterned after the
structure which occurs in the one-loop Schwinger /2 correction to the electron magnetic
moment [8]. In effect, we can represent a spin- 21 system as a composite of a spin- 21 fermion
and spin-one vector boson with arbitrary masses. We also give results for the case of
a spin- 21 composite consisting of scalar plus spin- 21 constituents, as would occur in a
composite of a photino and slepton in supersymmetric QED and in the radiative corrections
due to Higgs exchange. The light-cone wavefunctions describe off-shell particles but
are computable explicitly from perturbation theory. We will explicitly compute the form
factors F1 (q 2 ) and F2 (q 2 ) of the electromagnetic current, and the various contributions
to the form factors A(q 2 ) and B(q 2 ) of the energymomentum tensor. The model thus
provides a check on the general formulae, particularly the structure of angular momentum
on the light-cone; it provides an important illustration of J z conservation, Fock state by
Fock state; it demonstrates helicity retention between fermions and vector bosons at x
1; and it provides a template for an effective quark spin-one diquark structure of the valence
light-cone wavefunction of the proton.

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

315

The one-loop models can be further generalized by applying spectral PauliVillars

integration over the constituent masses. This representation of an effectively composite
system is particularly useful because it is based simply on two constituents but yet is
totally relativistic. The resulting form of light-cone wavefunctions provides a template for
parameterizing the structure of relativistic composite systems and their matrix elements
in hadronic physics. We thus obtain a theoretical laboratory to test the consistency of
formulae which have been proposed to probe the spin structure of hadrons. This clarifies the
connection of parton distributions to the constituents spin and orbital angular momentum
and to static quantities of the composite systems such as the magnetic moment. For
example, the model also provides a self-consistent form for the wavefunctions of an
effective quarkdiquark model of the valence Fock state of the proton wavefunction. A
similar approach has recently been used to illustrate the evolution of light-cone helicity and
orbital angular momentum operators [16]. A nonperturbative calculation of the electron
magnetic moment using the discretized light-cone quantization method [3] is given in
Ref. [17].
Many of the features of the analysis apply to arbitrary composite systems. For example,
we will explicitly prove the vanishing of the anomalous gravito-moment coupling B(0)
to gravity for any composite system. This remarkable result was first derived classically
from the equivalence principle by Kobsarev and Okun [18], and from the conservation
of the energymomentum tensor by Kobsarev and Zakharov [19]. See also the more
recent discussions in Refs. [9,20]. We will demonstrate that B(0) = 0 follows directly
for composite systems in quantum field theory from the Lorentz boost properties of the
light-cone Fock representation, and that it is valid in every Fock sector.

2. Electromagnetic and gravitational form factors

The light-cone Fock representation allows one to compute all matrix elements of local
currents as overlap integrals of the light-cone Fock wavefunctions. In particular, we
can evaluate the forward and non-forward matrix elements of the electroweak currents,
moments of the deep inelastic structure functions, as well as the electromagnetic form
factors and the magnetic moment. Given the local operators for the energymomentum
tensor T (x) and the angular momentum tensor M (x), one can directly compute
momentum fractions, spin properties, the gravitomagnetic moment, and the form factors
A(q 2) and B(q 2 ) appearing in the coupling of gravitons to composite systems.
In the case of a spin- 21 composite system, the Dirac and Pauli form factors F1 (q 2 ) and
F2 (q 2 ) are defined by


i
q u(P ),
0 ) F1 (q 2 ) + F2 (q 2 )
(4)
hP 0 |J (0)|P i = u(P
2M
where q = (P 0 P ) and u(P ) is the bound state spinor. In the light-cone formalism it
is convenient to identify the Dirac and Pauli form factors from the helicity-conserving and
helicity-flip vector current matrix elements of the J + current [8]:

316

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

 J + (0) 

P , = F1 (q 2 ),
(5)
P + q, 
+
2P
2
 J + (0) 


P , = (q 1 iq 2 ) F2 (q ) .
(6)
P + q, 
+
2P
2M
The magnetic moment of a composite system is one of its most basic properties. The
magnetic moment is defined at the q 2 0 limit,

e 
F1 (0) + F2 (0) ,
(7)
=
2M
where e is the charge and M is the mass of the composite system. We use the standard
light-cone frame (q = q 0 q 3 ):




q 2
q = q + , q , qE = 0, + , qE ,
P




M2
(8)
P = P + , P , PE = P + , + , 0E ,
P

2 is 4-momentum square transferred by the photon.

q
where q 2 = 2P q = E
e
F2 (0) can then be
The Pauli form factor and the anomalous magnetic moment = 2M
calculated from the expression
Z




F2 (q 2 ) X d2 kE dx X
0
=
ej a xi , kEi
, i a xi , kEi , i , (9)
(q 1 iq 2 )
3
2M
16
a
j

where the summation is over all contributing Fock states a and struck constituent charges
ej . The arguments of the final-state light-cone wavefunction are [1,2]
0
= kEi + (1 xi )E
q
kEi

(10)

for the struck constituent and

0
= kEi xi qE
kEi

(11)

for each spectator. Notice that the magnetic moment must be calculated from the spinflip non-forward matrix element of the current. It is not given by a diagonal forward matrix
element [21]. In the ultra-relativistic limit where the radius of the system is small compared
to its Compton scale 1/M, the anomalous magnetic moment must vanish [22]. The lightcone formalism is consistent with this theorem.
The form factors of the energymomentum tensor for a spin- 21 composite are defined by

i S( )
P q
0 ) A(q 2) ( PS) + B(q 2 )
hP 0 |T (0)|P i = u(P
2M

1
+ C(q 2 ) (q q g q 2 ) u(P ),
(12)
M
where q = (P 0 P ) , PS = 12 (P 0 + P ) , a ( b) = 12 (a b + a b ), and u(P ) is the
spinor of the system.
As in the light-cone decomposition Eqs. (5) and (6) of the Dirac and Pauli form factors
for the vector current [8], we can obtain the light-cone representation of the A(q 2 ) and

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

317

B(q 2 ) form factors of the energy-tensor equation (12). Since we work in the interaction
picture, only the non-interacting parts of the energymomentum tensor T ++ (0) need to be
computed in the light-cone formalism. By calculating the ++ component of Eq. (12), we
find
 T ++ (0) 


P , = A(q 2 ),
(13)
P + q, 
+
2
2(P )
2
 T ++ (0) 


P , = (q 1 iq 2 ) B(q ) .
(14)
P + q, 
+
2
2M
2(P )
The A(q 2 ) and B(q 2 ) form factors Eqs. (13) and (14) are similar to the F1 (q 2 ) and F2 (q 2 )
form factors Eqs. (5) and (6) with an additional factor of the light-cone momentum fraction
x = k + /P + of the struck constituent in the integrand. The B(q 2 ) form factor is obtained
from the non-forward spin-flip amplitude. The value of B(0) is obtained in the q 2 0
limit. The angular momentum projection of a state is given by
Z


1 ij k
i
d3 x T 0k x j T 0j x k
hJ i = 
2


1
) i u(P ).
(15)
= A(0)hLi i + A(0) + B(0) u(P
2
This result is derived using a wave packet description of the state. The hLi i term is the
orbital angular momentum of the center of mass motion with respect to an arbitrary origin
and can be dropped. The coefficient of the hLi i term must be 1; A(0) = 1 also follows when
we evaluate the four-momentum expectation value hP i. Thus the total intrinsic angular
momentum J z of a nucleon can be identified with the values of the form factors A(q 2) and
B(q 2 ) at q 2 = 0:




1
(16)
hJ z i = z A(0) + B(0) .
2
One can define individual quark and gluon contributions to the total angular momentum
from the matrix elements of the energymomentum tensor [9]. However, this definition is
only formal; Aq,g (0) can be interpreted as the light-cone momentum fraction carried by the
quarks or gluons hxq,g i. The contributions from Bq,g (0) to Jz cancel in the sum. In fact,
we shall show that the contributions to B(0) vanish when summed over the constituents of
each individual Fock state.
We will give an explicit realization of these relations in the light-cone Fock representation for general composite systems. In the next section we will illustrate the formulae by
computing the electrons electromagnetic and energymomentum tensor form factors to
one-loop order in QED. In fact, the structure of this calculation has much more generality
and can be used as a template for more general composite systems.

3. The light-cone Fock state decomposition and spin structure of leptons in QED
The Schwinger one-loop radiative correction to the electron current in quantum
electrodynamics has played a historic role in the development of quantum field theory.

318

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

In the language of light-cone quantization, the electron anomalous magnetic moment ae =

/2 is due to the one-fermion one-gauge boson Fock state component of the physical
electron. An explicit calculation of the anomalous moment in this framework using Eq. (7)
was give in Ref. [8]. We shall show here that the light-cone wavefunctions of the electron
provides an ideal system to check explicitly the intricacies of spin and orbital angular
momentum in quantum field theory. In particular, we shall evaluate the matrix elements of
the QED energymomentum tensor and show how the spin crisis is resolved in QED for
an actual physical system. The analysis is exact in perturbation theory. The same method
can be applied to the moments of structure functions and the evaluation of other local
matrix elements. In fact, the QED analysis of this section is more general than perturbation
theory. We will also show how the perturbative light-cone wavefunctions of leptons and
photons provide a template for the wavefunctions of non-perturbative composite systems
resembling hadrons in QCD.
The light-cone Fock state wavefunctions of an electron can be systematically evaluated
in QED. The QED Lagrangian density is



i
1
ih

S
S
S F F , (17)

+ ieA
ieA m
L=
2
4
and the corresponding energymomentum tensor is





i  S

+ ieA
ieA + [ ]
T =
4
1
(18)
+ F F + g F F .
4
Since T is the Noether current of the general coordinate transformation, it is conserved.
In later calculations we will identify the two terms in Eq. (18) as the fermion and boson

contributions Tf and Tb , respectively.

The physical electron is the eigenstate of the QED Hamiltonian. As discussed in the
introduction, the expansion of the QED eigenfunction on the complete set |ni of H0
eigenstates produces the Fock state expansion. It is particularly advantageous to carry out
this procedure using light-cone quantization since the vacuum is trivial, the Fock state
representation is boost invariant, and the light-cone fractions xi = ki+ /P + are positive:
P
0 < xi 6 1, i xi = 1. We also employ light-cone gauge A+ = 0 so that the gauge boson
polarizations are physical. Thus each Fock-state wavefunction hn|physical electroni of
the physical electron with total spin projection J z = 12 is represented by the function
z
nJ (xi , kEi , i ), where


2
E2
+ E

+ ki + mi E
, ki
(19)
ki = ki , ki , ki = xi P ,
xi P +
specifies the momentum of each constituent and i specifies its light-cone helicity in the z
direction. We adopt a non-zero boson mass for the sake of generality.
The two-particle Fock state for an electron with J z = + 12 has four possible spin
combinations:



+ E

E
two particle P , P = 0

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

Z
=

319

h



d2 kE dx

1
x, kE + 12 + 1; xP + , kE
x(1 x) 16 3 + 2 +1







x, kE + 12 1; xP + , kE + 1
x, kE  12 + 1; xP + , kE
+ 1
+ 2 1
2 +1

 1
i

+ E

E
x, k 2 1; xP , k ,
(20)
+ 1
2 1

where the two-particle states |sfz sbz ; xP + , kE i are normalized as in (2). Here sfz and sbz
denote the z-component of the spins of the constituent fermion and boson, respectively.
The wavefunctions can be evaluated explicitly in QED perturbation theory using the rules
given in Refs. [5,8]:

(k 1 + ik 2 )

,
2
x,
k
=

+ 12 +1

x(1 x)

(+k 1 + ik 2 )

,
x, kE = 2
1
+ 2 1
1x
(21)



m

,
1
x, kE = 2 M

2 +1
x

x, kE = 0,
1
2 1

where


= x, kE =

e/ 1 x
.
2 + m2 )/x (k
E 2 + 2 )/(1 x)
M 2 (kE

(22)

Similarly,



+ E

E
two particle P , P = 0
Z
h



d2 kE dx

x, kE + 12 + 1; xP + , kE
1
=
3
+ 2 +1
x(1 x) 16







x, kE  + 12 1; xP + , kE + 1
x, kE  12 + 1; xP + , kE
+ 1
+ 2 1

12 1


x, kE  12 1; xP + , kE ,

2 +1

where



E

+ 12 +1 x, k = 0,




m

2
M

x,
k
=

x
+ 2 1
(k 1 + ik 2 )

,
x, kE = 2
1

2 +1

1x

(+k 1 + ik 2 )

1
.
x, kE = 2
2 1
x(1 x)
The coefficients of in Eqs. (21) and (24) are the matrix elements of
u(k + , k , kE )
u(P + , P , PE )


k+
P+

(23)

(24)

320

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

which are the numerators of the wavefunctions corresponding to each constituent spin s z
configuration. The two boson polarization vectors in light-cone gauge are  = ( + =

E
k
E ), where E = E, = (1/ 2 )(x iy).
The polarizations also satisfy
0,  = E2k
+ ,
the Lorentz condition k  = 0. In (21) and (24) we have generalized the framework of
QED by assigning a mass M to the external electrons, but a different mass m to the
internal electron lines and a mass to the internal photon line [8]. The idea behind this
is to model the structure of a composite fermion state with mass M by a fermion and a
vector constituent with respective masses m and .
The electron in QED also has a bare one-particle component:
 ,





= Z (1 x) kE = 0E s z = 1 ,
(25)
one particle

where Z is the wavefunction normalization of the one-particle state. If we regulate the

theory in the ultraviolet and infrared, Z is finite.
We first will evaluate the Dirac and Pauli form factors F1 (q 2 ) and F2 (q 2 ). Using Eqs. (5)
and (20) we have to order e2




F1 (q 2 ) = P + , PE = qE  P + , PE = 0E
Z 2E




d k dx h
0
1
x, kE
1
x, kE
= Z+
3
+1
+1
+
+
16
2
2







i

0
0
x, kE
1
x, kE + 1
x, kE
1
x, kE ,
+ 1
+ 2 1

+ 2 1

2 +1

2 +1

(26)
where
0
= kE + (1 x)E
q .
kE

(27)

Ultraviolet regularization is assumed. For example, we can assume a cutoff in the

P kE 2 +m2
invariant mass of the constituents: M2 = i ixi i < 2 , or we can use PauliVillars
regularization by introducing a fictitious photon with a large mass .
At zero momentum transfer
Z 2E




d k dx h
1
x, kE 1
x, kE
F1 (0) = Z +
3
+1
+1
+
+
16
2
2







i

x, kE 1
x, kE + 1
x, kE 1
x, kE ,
+ 1
+ 2 1

+ 2 1

2 +1

2 +1

(28)
where the renormalization constant Z is given, using PauliVillars regularization, by

Z = 1
2

Z1
0

(1 + x 2 ) M 2 +
ln
dx
(1 x)
M 2 +


(xM + m)2
1
+

2
2
x
M + mx +

m2
x
m2
x

2
1x

+
+

2
1x
2
1x

1
M 2 +

m2
x

2
1x


.

(29)

This ensures the Ward identity and F1 (0) = 1. Further discussion of the Ward identity for
QED in light-cone perturbation theory is given in Ref. [23].

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

321

The one-loop model can be further generalized by applying spectral PauliVillars

integration over the constituent masses. The resulting form of light-cone wavefunctions
provides a template for parameterizing the structure of relativistic composite systems and
their matrix elements in hadronic physics.
The Pauli form factor is obtained from the spin-flip matrix element of the J + current.
From Eqs. (6), (20), and (23) we have




2M
+ E
P , P = qE  P + , PE = 0E
F2 (q 2 ) = 1
2
(q iq )
Z 2E




d k dx h
2M
0
x, kE
1
x, kE
1
= 1
2
3
1
1
+
+
(q iq )
16
2
2



i

0
x, kE 1
x, kE
+ 1
2 +1

2 +1

d2 kE dx (m Mx)
0
x, kE
x, kE
16 3
x
Z 2E
d k dx (m xM)
2
= 4Me
16 3 x(1 x)
1

M 2 ((kE + (1 x)E
q )2 + m2 )/x ((kE + (1 x)E
q )2 + 2 )/(1 x)
1
.
(30)

2
2
2
E
M (k + m )/x (kE 2 + 2 )/(1 x)

= 4M

Using the Feynman parameterization, we can also express Eq. (30) in a form in which the
q 2 = E
q2 dependence is more explicit as
Me2
F2 (q ) =
4 2

Z1

Z1
d

dx
0

m xM
(1 ) 1x
E2
x q

M2 +

m2
x

2
1x

(31)

The anomalous moment is obtained in the limit of zero momentum transfer:

Z 2E
1
d k dx (m xM)
F2 (0) = 4Me2
16 3 x(1 x) [M 2 (kE2 + m2 )/x (kE2 + 2 )/(1 x)]2
Me2
=
4 2

Z1
dx
0

m xM
M 2

m2
x

2
1x

(32)

which is the result of Ref. [8]. For zero photon mass and M = m, it gives the correct order
Schwinger value ae = F2 (0) = /2 for the electron anomalous magnetic moment for
QED.
As seen from Eqs. (13) and (14), the matrix elements of the double plus components of
the energymomentum tensor are sufficient to derive the fermion and boson constituents
form factors Af,g (q 2 ) and Bf,g (q 2 ) of graviton coupling to matter. In particular, we shall
verify A(0) = Af (0) + Ab (0) = 1 and B(0) = 0.
The individual contributions of the fermion and boson fields to the energymomentum
form factors in QED are given by

322

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

 T ++ (0) 


Af (q 2 ) = P + , PE = qE  f + 2  P + , PE = 0E
2(P )
Z 2E
h




d k dx

0
x 1
x, kE
1
x, kE
=
3
+1
+1
+
+
16
2
2

0
0
x, kE
1
x, kE + 1
x, kE
1
+ 1
+ 2 1

+ 2 1

2 +1

2 +1

i
x, kE , (33)

0 is given in Eq. (27), and

where kE

 T ++ (0) 


Ab (q 2 ) = P + , PE = qE  b + 2  P + , PE = 0E
2(P )
Z 2E
h




d k dx

00
(1 x) 1
x, kE
1
x, kE
=
3
+1
+1
+
+
16
2
2

00
00
E
E
x, k 1
x, k + 1
x, kE
1
+ 1
+ 2 1

+ 2 1

2 +1

2 +1

i
x, kE , (34)

where
00
= kE x qE .
kE

(35)

Note that
Af (0) + Ab (0) = F1 (0) = 1,

(36)

which corresponds to the momentum sum rule.

The fermion and boson contributions to the spin-flip matter form factor are

 Tf++ (0)  +


2M
+ E
 P , PE = 0E


,
P
=
q
E
P

1
2
+
2
(q iq )
2(P )
Z 2E




d k dx h
2M
0
x 1
x, kE
1
x, kE
= 1
2
3
1
1
+
+
(q iq )
16
2
2



i

0
x, kE 1
x, kE
+ 1

Bf (q 2 ) =

2 +1

d2 kE dx
(m Mx)
16 3

= 4M

Z
= 4Me2

2 +1

0
x, kE
x, kE

d2 kE dx (m xM)
(1 x)
16 3

1
2
2
E
((k + (1 x)E
q ) + m )/x ((kE + (1 x)E
q )2 + 2 )/(1 x)
1

2
2
2
E
M (k + m )/x (kE 2 + 2 )/(1 x)

M2

Me2

Z1
d

4 2
0

and

Z1
dx
0

x(m xM)
(1 )

1x
x

qE2 M 2 +

m2
x

2
1x

(37)

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

323

 Tb++ (0)  +


2M
+ E

 P , PE = 0E

,
P
=
q
E
P

(q 1 iq 2 )
2(P + )2
Z 2E
h




d k dx
2M

0
(1

x)
1
x, kE
1
x, kE
= 1
2
3
1
1
+
+
(q iq )
16
2
2
i

0
x, kE 1 (x, kE )
+ 1

Bb (q 2 ) =

2 +1

d2 kE dx
16 3

= 4M

Z
= 4Me

2 +1

0
(m Mx) x, kE
x, kE

d2 kE dx (m xM)
(1 x)
16 3

1
2
2
E
((k x qE ) + m )/x ((kE x qE )2 + 2 )/(1 x)
1

2
2
2
E
M (k + m )/x (kE2 + 2 )/(1 x)
Z1 Z1
x(m xM)
Me2
d dx
.
= 2
2
x
2
4
(1 ) 1x qE2 M 2 + mx + 1x

M2

(38)

The total contribution for general momentum transfer is

B(q 2 ) = Bf (q 2 ) + Bb (q 2 )
Z 2E
d k dx (m xM)
2
= 4Me
16 3
(1 x)
(

M 2 ((kE + (1 x)E
q )2 + m2 )/x ((kE + (1 x)E
q )2 + 2 )/(1 x)
)
1

M 2 ((kE x qE )2 + m2 )/x ((kE x qE )2 + 2 )/(1 x)

1

M 2 (kE2 + m2 )/x
Z1 Z1
Me2


dx x(m xM)

4 2
0

2 + 2 )/(1 x)
(kE

1
(1 )

1x
x

qE2

M2 +

1
x
(1 ) 1x
qE2 M 2 +

m2
x

2
1x


.

m2
x

2
1x

(39)

This is the analog of the Pauli form factor for a physical electron scattering in a
gravitational field and in general is not zero. However at zero momentum transfer
B(0) = Bf (0) + Bb (0) = 0,

(40)

in agreement with classical arguments based on the equivalence principle and conservation
of the energymomentum tensor [9,1820].

324

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

Fig. 1. Helicity-flip electromagnetic and gravitational form factors for spacelike q 2 = Q2 < 0 from
the quantum fluctuations of a fermion at one-loop order in units of / for QED and g 2 /4 2 for
the Yukawa theory. The fermion constituent mass is taken as mf = M. The boson constituent is
massless.

The helicity-flip electromagnetic and gravitational form factors for the fluctuations of
the electron at one-loop are illustrated in Fig. 1. The cancellation of the sum of graviton
couplings B(q 2 ) to the constituents at q 2 = 0 is evident.
(a) Helicity-flip Pauli form factor F2 (q 2 ) in QED. Notice that F2 (0) = 1/2.
(b) Helicity-flip form factor Bb (q 2 ) of the graviton coupling to the boson (photon)
constituent of the electron at one-loop order in QED. Notice that Bb (0) = 1/3.
(c) Helicity-flip fermion form factor Bf (q 2 ) of the graviton coupling to the fermion
constituent at one-loop order in QED. Notice that Bf (0) = 1/3, and thus Bf (0) + Bb (0) =
0.
(d) Helicity-flip Pauli form factor F2 (q 2 ) in the Yukawa theory. Notice that in this case
F2 (0) = 3/4.
(e) Helicity-flip form factor Bb (q 2 ) of the graviton coupling to the boson at one-loop
order in the Yukawa theory. Notice that Bb (0) = 5/12.

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

325

(f) Helicity-flip fermion form factor Bf (q 2 ) of the graviton coupling to the fermion
constituent at one-loop order in the Yukawa theory. Notice that Bf (0) = 5/12, and thus
Bf (0) + Bb (0) = 0.
In Section 5 we shall prove that the anomalous gravitomagnetic moment B(0) = Bf (0) +
Bb (0) is identically zero for arbitrary composite systems in quantum field theory. The
proof follows from the Lorentz covariance of the light-cone wavefunction and applies to
the contribution of each individual Fock component.

4. The light-cone Fock state decomposition and spin structure of composite fermions
in Yukawa theory
As a second example, we shall consider an effectively composite system composed of a
fermion and a neutral scalar based on the one-loop quantum fluctuations of this theory. The
light-cone wavefunctions describe off-shell particles but are computable explicitly from
perturbation theory. We consider the Yukawa Lagrangian

iS
S
S ) m
( ) (
2
1
1
S,
+ ( )( ) 2 + g
2
2
and the corresponding energymomentum tensor is given by
L =

i


S
S

+ ( )( ) g L,
2
which is conserved.
The J z = + 12 two-particle Fock state is given by



+ E

E
two particle P , P = 0
Z
h



d2 kE dx

1 x, kE + 12 ; xP + , kE
=
3
+
x(1 x)16
2


i

+ 1 x, kE  12 ; xP + , kE ,
T =

where



m

,
+ 1 x, k = M +
x
2
1
2

1 x, kE = (+k + ik ) .
2
x

(41)

(42)

(43)

(44)

The scalar part of the wavefunction is given in Eq. (22) with e replaced by g. The
normalization of the Fock states is as in (2).
Similarly, the J z = 12 two-particle Fock state is given by



+ E

E
two particle P , P = 0

326

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

Z
=

h



d2 kE dx

1 x, kE + 12 ; xP + , kE
x(1 x)16 3 + 2


i

+ 1 x, kE  12 ; xP + , kE ,

(45)

where

(+k 1 ik 2 )

1 x, kE =
,
+2
x 

(46)


m

E
.

x,
k
=
M
+
1

x
2
In (44) and (46) we have generalized the framework of the Yukawa theory by assigning a
mass M to the external electrons, but a different mass m to the internal electron lines and
a mass to the internal boson line [8]. The idea behind this is to model the structure of a
composite fermion state with mass M by a fermion and a boson constituent with respective
masses m and .
Using Eqs. (5) and (43) we have
Z 2E







i
d k dx h

0
0
E
E
E
E

x,
k

x,
k
+

x,
k

x,
k
, (47)
F1 (q 2 ) = Z +

+ 12
+ 12
12
12
16 3
0 is given in Eq. (27). At zero momentum transfer
where kE
Z 2E
i






d k dx h

E
E
E
E
(x,
k
)
(48)

x,
k

x,
k
+

x,
k

F1 (0) = Z +

,
+ 12
+ 12
12
12
16 3
where the renormalization constant Z in light-cone gauge is given, using PauliVillars
regularization, by
2
  Z1 "
2
M 2 + mx + 1x
1 g2
dx (1 x) ln
Z = 1
2
2
4 4
M 2 + mx + 1x
0
!#
1
(xM + m)2
1

+
+
.
(49)
2
2
2
2
x
M 2 + m +
M 2 + m +
x

1x

1x

The Pauli form factor is obtained from the spin-flip matrix element of the J + current.
From Eqs. (6), (43) and (45) we have
Z 2E







i
2M
d k dx h

0
0
E
E
E
E

x,
k

x,
k
+

x,
k

x,
k
F2 (q 2 ) = 1

+ 12
+ 12
12
12
(q iq 2 )
16 3
Z 2E




d k dx (1 x)(m + Mx)
0
x, kE
x, kE
= 2M
3
2
16
x
Z 2E
d k dx (m + xM)
= 2Mg 2
16 3
x2
1

M 2 ((kE + (1 x)E
q )2 + m2 )/x ((kE + (1 x)E
q )2 + 2 )/(1 x)
1

M 2 (kE 2 + m2 )/x (kE 2 + 2 )/(1 x)

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

Mg 2
=
8 2

Z1

Z1
d

dx
0

(1 )

1x
x (m + xM)
2
1x 2
E M 2 + mx
x q

2
1x

327

(50)

The anomalous moment is obtained in the limit of zero momentum transfer,

Z 2E
1
d k dx (m + xM)
F2 (0) = 2Mg 2
2
2
2
E
16 3
x2
[M (k + m )/x (kE2 + 2 )/(1 x)]2
Mg 2
=
8 2

Z1
dx
0

1x
x (m + xM)
2
2
M 2 + mx + 1x

(51)

The individual contributions of the fermion and boson fields to the energymomentum
form factors in the Yukawa model are given by

 


Af (q 2 ) = P + , PE = qE x  P + , PE = 0E
Z 2E







i
d k dx h

0
0
E
E
E
E
x

x,
k

x,
k
+

x,
k

x,
k
, (52)
=

+ 12
+ 12
12
12
16 3
0 =k
E + (1 x)E
q , and
where kE





Ab (q 2 ) = P + , PE = qE (1 x) P + , PE = 0E
Z 2E
h




d k dx

00
(1

x)
1 x, kE
1 x, kE
=
3
+
+
16
2
2



i

00
E
+ 1 x, k 1 x, kE ,
2

(53)

00 = k
E x qE . Note that again
where in this case kE

Af (0) + Ab (0) = F1 (0) = 1,

(54)

which corresponds to the momentum sum rule.

The fermion and boson contributions to the spin-flip matter form factor are
Z 2E
2M
d k dx
x
Bf (q 2 ) = 1
(q iq 2 )
16 3
h







i

0
0
1 x, kE + 1 x, kE
1 x, kE
1 x, kE
+2

+2

d2 kE dx (1 x)(m + Mx)
0
x, kE
x, kE
3
16
x
Z 2E
d k dx (m + xM)
= 2Mg 2
x
16 3
1

M 2 ((kE + (1 x)E
q )2 + m2 )/x ((kE + (1 x)E
q )2 + 2 )/(1 x)
1

M 2 (kE 2 + m2 )/x (kE 2 + 2 )/(1 x)

= 2M

328

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

Mg 2
=
8 2

Z1

Z1
d

dx
0

(1 x)(m + xM)
(1 ) 1x
E2 M 2 +
x q

m2
x

(55)

2
1x

and
Bb (q 2 ) =

Z 2E
2M
d k dx
(1 x)
1
2
(q iq )
16 3
h

0
x, kE
1
x, kE + 1
1
+ 2 1

+ 2 1

2 +1

0
x, kE
1

2 +1

x, kE

i

d2 kE dx (1 x)(m + Mx)
00
x, kE
x, kE
3
16
x
Z 2E
dx
(m
+
xM)
k
d

= 2Mg 2
x
16 3
1

2
2
2
E
M ((k x qE ) + m )/x ((kE x qE )2 + 2 )/(1 x)
1

2
2
2
E
M (k + m )/x (kE 2 + 2 )/(1 x)

= 2M

Mg 2

Z1

Z1
d

8 2
0

dx
0

(1 x)(m + xM)
x
(1 ) 1x
qE2 M 2 +

m2
x

2
1x

(56)

The total contribution from the fermion and boson constituents is

B(q 2 ) = Bf (q 2 ) + Bb (q 2 )
Z 2E
d k dx (m + xM)
= 2Mg 2
x
16 3

1

M 2 ((kE + (1 x)E
q )2 + m2 )/x ((kE + (1 x)E
q )2 + 2 )/(1 x)

1

M 2 ((kE x qE )2 + m2 )/x ((kE x qE )2 + 2 )/(1 x)

1

2
M 2 (kE + m2 )/x (kE 2 + 2 )/(1 x)
Z1

Z1

Mg 2
d dx (1 x)(m + xM)
8 2
0
0

1

2
1x 2
(1 ) x qE M 2 + mx +

2
1x

1
x
(1 ) 1x
qE2 M 2 +

m2
x

2
1x


.

(57)

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

329

At zero momentum transfer

B(0) = Bf (0) + Bb (0) = 0,

(58)

which is another example of the vanishing of the anomalous gravitomagnetic moment. [See
also Fig. 1(e) + Fig. 1(f).] The general proof that B(0) = 0 for any system is given in the
next section. Note that B(Q2 ) does not vanish for nonzero momentum transfer.

5. The anomalous gravitomagnetic moment for composite systems

In this section we shall show that the anomalous gravitomagnetic moment B(0) always
vanishes for each contributing Fock state of a general composite system. In order to
calculate B(0) using Eq. (14), we need to consider a non-forward amplitude. The internal
momentum variables for the final state wavefunction are given by Eqs. (10) and (11). The
1 , k 1 , and k 2 label
subscripts of xi and kEi label constituent particles, the superscripts of q

the Lorentz indices, and the subscript a in a indicates the contributing Fock state. The
essential ingredient is the Lorentz property of the light-cone wavefunctions.
It is important to identify the n 1 independent relative momenta of the n-particle Fock
state.

 T ++ (0) 


B(0)
P ,

= lim
P
+
q,

1
+
2
1
2M
2(P
)
q 0 q

 ++ 



+
E = qE  T (0)  P + = 1, PE = 0E
=
1,
P

P
1 0 q 1
2(P + )2
q

Z n1
X Y d2 kEk dxk
= lim
1 0 q 1
16 3
q
a
= lim

k=1

x1 , x2 , . . . , xn1 , (1 x1 x2 xn1 ),

" n1
X

0
0
0
0
0
0
, kE2
, . . . , kEn1
, kE1
kE2
kEn1
kE1
#

xi + (1 x1 x2 xn1 )
i=1
a x1 , x2 , . . . , xn1 , (1 x1 x2 xn1 ),

kE1 , kE2 , . . . , kEn1 , kE1 kE2 kEn1 .

(59)

Using integration by parts,

Ba (0)
2M
Z n1
Y d2 kEk dxk
a x1 , x2 , . . . , xn1 , (1 x1 x2 xn1 ),
=
16 3
k=1



kE1 , kE2 , . . . , kEn1 , kE1 kE2 kEn1

330

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

" n1
X
i=1

xj 1
(1 + xi ) 1 +
ki j 6=i kj
n1

xi

+ (1 x1 x2 xn1 )

n1
X
j =1

xj

1
kj

a x1 , x2 , . . . , xn1 , (1 x1 x2 xn1 ),
kE1 , kE2 , . . . , kEn1 , kE1 kE2 kEn1

Z n1
Y d2 kEk dxk
16 3

k=1

a x1 , x2 , . . . , xn1 , (1 x1 x2 xn1 ),
kE1 , kE2 , . . . , kEn1 , kE1 kE2 kEn1

" n1
X
i=1

n1
X

xj + (1 x1 x2 xn1 ) xi 1
1 +
ki
j =1

x1 , x2 , . . . , xn1 , (1 x1 x2 xn1 ),
kE1 , kE2 , . . . , kEn1 , kE1 kE2 kEn1

= 0.

(60)

Thus the contribution Ba (0) from each contributing Fock state a to the total anomalous
gravitomagnetic moment B(0) vanishes separately.

6. The perturbative models as a template for a composite system

We can use the structure of the one-loop QED and Yukawa calculations with general
values for M, m, and , to represent a spin- 21 system composed of a fermion and a spin-1
or spin-0 boson. Such a model describes an effectively composite system with no bare
one-particle Fock state. We can also generalize the functional form of the momentum
space wavefunction (x, kE ) by introducing a spectrum of vector bosons satisfying the
generalized PauliVillars spectral conditions
Z
(61)
d2 2N (2 ) = 0, N = 0, 1, . . . .
For example, we can simulate a proton as a bound state of a quark and diquark [24],
using spin-0, spin-1 diquarks, or a linear superposition of the two states. The model can be
made to match the power-law fall-off of the hadron form factors predicted in perturbative
QCD by the choice of sum rule conditions on the PauliVillars spectra [25,26]. The lightcone framework of the model resembles that of the covariant parton model of Landshoff,
Polkinghorne and Short [27,28], in which the power behavior of the spectral integral at
high masses corresponds to the Regge behavior of the deep inelastic structure functions.
Although the model is based on just two Fock constituents, it is relativistic and satisfies
self-consistency conditions such as in the point-like limit where R 2 M 2 0, the anomalous

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

331

moment vanishes [22]. The light-cone formalism also properly incorporates Wigner boosts.
Thus this model of composite systems can serve as a useful theoretical laboratory to
interrelate hadronic properties and check the consistency of formulae proposed for the
study of hadron substructure.

7. Spin and orbital angular momentum composition of light-cone wavefunctions

In general the light-cone wavefunctions satisfy conservation of the z projection of
angular momentum:
Jz =

n
X
i=1

siz +

n1
X

ljz .

(62)

j =1

The sum over siz represents the contribution of the intrinsic spins of the
n Fock state
constituents. The sum over orbital angular momenta ljz = i kj1 2 kj2 1 derives from
kj

kj

the n1 relative momenta. This excludes the contribution to the orbital angular momentum
due to the motion of the center of mass, which is not an intrinsic property of the hadron.
We can see how the angular momentum sum rule Eq. (62) is satisfied for the
wavefunctions Eqs. (20) and (23) of the QED model system of two-particle Fock states.
In Table 1 we list the fermion constituents light-cone spin projection sfz = 12 f , the boson
constituent spin projection sbz = b , and the relative orbital angular momentum l z for each
contributing configuration of the QED model system wavefunction.
Table 1 is derived by calculating the matrix elements of the light-cone helicity
operator
+ 5 [29] and the relative orbital angular momentum operator i k 1 k 2 k 2 k 1 [16,30,
31] in the light-cone representation. Each configuration satisfies the spin sum rule: J z =
sfz + sbz + l z .
For a better understanding of Table 1, we look at the non-relativistic and ultra-relativistic
limits. At the non-relativistic limit, the transversal motions of the constituent can be
neglected and we have only the | + 12 i | 12 + 1i configuration which is the nonrelativistic quantum state for the spin-half system composed of a fermion and a spin-1
boson constituents. The fermion constituent has spin projection in the opposite direction
to the spin J z of the whole system. However, for ultra-relativistic binding in which the
transversal motions of the constituents are large compared to the fermion masses, the
Table 1
Spin decomposition of the J z = + 12 electron
Configuration

 1  1

+ + + 1
2
2
 1  1

+  + 1
2
2
 1  1

+ + 1
2

Fermion spin sfz

Boson spin sbz

Orbital ang. mom. l z

+ 12

+1

12

+1

+ 12

+1

332

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

Table 2
Spin decomposition of the J z = + 12 fermion in Yukawa theory
Configuration

 1  1
+ +
 21   21 
+ 
2

Fermion spin sfz

Boson spin sbz

+ 12

12

+1

Orbital ang. mom. l z

| + 12 i | + 12 + 1i and | + 12 i | + 12 1i configurations dominate over the | + 12 i

| 12 + 1i configuration. In this case the fermion constituent has spin projection parallel
to J z .
The corresponding spin content in the Yukawa theory is given in Table 2. In this case,
the non-relativistic fermions spin projection is aligned with the total J z , and it is antialigned in the ultra-relativistic limit. The distinct features of spin structure in the nonrelativistic and ultra-relativistic limits reveals the importance of relativistic effects and
supports the viewpoint [29,32,33] that the proton spin puzzle can be understood as due
to the relativistic motion of quarks inside the nucleon. In particular, the spin projection of
the relativistic constituent quark tends to be anti-aligned with the proton spin in a quark
diquark bound state if the diquark has spin 0. The state with orbital angular momentum
l z = 1 in fact dominates over the states with l z = 0. Thus the empirical fact that 1q
is small in the proton has a natural description in the light-cone Fock representation of
hadrons.
The explicit formulas for the quark spin distributions 1q(x, 2) in the quarkdiquark
models can be immediately obtained for the spin-1 diquark model from Eqs. (20) and (21):
1q(x, 2)spin-1 diquark
 


Z 2E


kE2
kE2
d k dx
m 2
2
2
=

M
+

||2 , (63)
2
x
16 3
x 2 (1 x)2 (1 x)2
and for the spin-0 diquark model from Eqs. (43) and (44):



Z 2E


d k dx
m 2 kE2
2
2
2
M
2 ||2 ,
M+
1q(x, )spin-0 diquark =
x
16 3
x

(64)

where we have regulated the integral by assuming a cutoff in the invariant mass: M2 =
2 +m2
P kEi
i
< 2 . Again, one sees the transition of 1q from the nonrelativistic to
i
xi
relativistic limit. In the spin-0 diquark model 1q = 1 in the nonrelativistic limit, and
decreases toward 1q = 1 as the intrinsic transverse momentum increases. The behavior
is just opposite in the case of the spin-1 diquark.

8. Conclusions
The LC wavefunctions n/H (xi , kEi , i ) provide a general representation of a relativistic composite system. They are universal, process independent, and control all hadronic

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

333

reactions. In this paper we have constructed explicit models which are simple but yet are
completely relativistic, preserve all of the Lorentz properties of a composite system of
quantum field theory. Because of this explicit realization we can see how different hadronic
phenomena can be interrelated. For example, the matrix elements of local operators such
as the electromagnetic current, the energymomentum tensor, angular momentum, and the
moments of structure functions have exact representations in terms of light-cone Fock state
wavefunctions of bound states such as hadrons. We have shown that each Fock state of a
composite system satisfies J z conservation, component by component. We have emphasized that the correct expression for the orbital angular momentum l z involves a sum over
n 1 relative momentum contributions for a Fock state with n constituents.
We have illustrated these properties by examining the explicit form of the light-cone
wavefunctions for the two-particle Fock state of the electron in QED, thus connecting the
Schwinger anomalous magnetic moment to the spin and orbital momentum carried by its
Fock state constituents. We have also computed the QED one-loop radiative corrections
for the form factors for the graviton coupling to the electron and photon. The one-loop
model provides a transparent basis for understanding the structure of relativistic composite
systems and their matrix elements in hadronic physics. Although the underlying model is
derived from elementary QED perturbative couplings, it in fact can be used to model much
more general bound state systems by applying spectral integration over the constituent
masses while preserving all of the Lorentz properties.
We thus have obtained a theoretical laboratory to test the consistency of formulae which
have been proposed to probe the spin structure of hadrons. For example, we have computed
the quark spin distributions 1q(x, 2 ) in quarkdiquark models. In particular, the spin
projection of the relativistic constituent quark tends to be anti-aligned with the proton spin
in a quarkdiquark bound state if the diquark has spin 0. The empirical fact that 1q is
small in the proton thus has a natural description in the light-cone Fock representation of a
relativistic bound state.
We have also given general exact expressions for the matrix elements of the electromagnetic, electroweak, and graviton couplings for operators for arbitrary composite systems,
giving explicit realization of the spin sum rules and other local matrix elements in terms of
the light-cone Fock state wavefunctions.
Finally, we have given a general proof demonstrating that the anomalous gravitomagnetic moment B(0) for gravitons coupling to matter vanishes identically for any composite
system. At one loop order in QED, we can see the explicit cancellation of the graviton coupling to the lepton and photon. In fact we have shown that this remarkable property holds
generally for any composite or elementary system at all orders directly from the Lorentz
boost properties of the light-cone Fock representation.

References
[1] S.D. Drell, T.M. Yan, Phys. Rev. Lett. 24 (1970) 181.
[2] G.B. West, Phys. Rev. Lett. 24 (1970) 1206.

334

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

[3] For a review and further references see: S.J. Brodsky, H. Pauli, S.S. Pinsky, Phys. Rep. 301
(1998) 299.
[4] P.A.M. Dirac, Rev. Mod. Phys. 21 (1949) 392.
[5] The notation and calculational rules can be found in:
G.P. Lepage, S.J. Brodsky, Phys. Rev. D 22 (1980) 2157;
G.P. Lepage, S.J. Brodsky, Phys. Lett. B 87 (1979) 359;
G.P. Lepage, S.J. Brodsky, Phys. Rev. Lett. 43 (1979) 545;
G.P. Lepage, S.J. Brodsky, Phys. Rev. Lett. 43 (1979) 1625, Erratum.
[6] B. Geyer, D. Robaschik, M. Bordag, J. Horej, Z. Phys. C 26 (1985) 591;
T. Braunschweig, B. Geyer, J. Horej, D. Robaschik, Z. Phys. C 33 (1987) 275;
F.-M. Dittes, D. Mller, D. Robaschik, B. Geyer, Phys. Lett. B 209 (1988) 325;
I.I. Balitskii, V.M. Braun, Nucl. Phys. B 311 (1988/89) 541;
P. Jain, J.P. Ralston, in: Future Directions in Particle and Nuclear Physics at Multi-GeV Hadron
Beam Facilities, BNL, March 1993.
[7] S.J. Brodsky, L. Frankfurt, J.F. Gunion, A.H. Mueller, M. Strikman, Phys. Rev. D 50 (1994)
3134.
[8] S.J. Brodsky, S.D. Drell, Phys. Rev. D 22 (1980) 2236.
[9] X. Ji, Phys. Rev. D 58 (1998) 056003, hep-ph/9710290;
X. Ji, Talk presented at 12th Int. Symp. on High-Energy Spin Physics (SPIN96), Amsterdam,
September 1996, hep-ph/9610369;
X. Ji, Phys. Rev. Lett. 78 (1997) 610;
X. Ji, Phys. Rev. D 55 (1997) 7114.
[10] S.J. Brodsky, D.S. Hwang, Nucl. Phys. B 543 (1999) 239.
[11] S.J. Chang, R.G. Root, T.M. Yan, Phys. Rev. D 7 (1973) 1133.
[12] M. Burkardt, Nucl. Phys. A 504 (1989) 762;
M. Burkardt, Nucl. Phys. B 373 (1992) 613;
M. Burkardt, Phys. Rev. D 52 (1995) 3841.
[13] H.-M. Choi, C.-R. Ji, Phys. Rev. D 58 (1998) 071901.
[14] S.A. Paston, V.A. Franke, E.V. Prokhvatilov, Theor. Math. Phys. 120 (1999) 1164, hepth/0002062.
[15] S.J. Brodsky, J.R. Hiller, G. McCartor, Phys. Rev. D 60 (1999) 054506, hep-ph/9903388.
[16] A. Harindranath, R. Kundu, Phys. Rev. D 59 (1999) 116013.
[17] J.R. Hiller, S.J. Brodsky, Phys. Rev. D 59 (1999) 016006.
[18] I.Yu. Kobsarev, L. Okun, ZhETF 43 (1962) 1904, English translation: JETP 16 (1963) 1343;
L. Okun, C. Rubbia, in: H. Filthuth (Ed.), Proceedings of the 4th International Conference on
Elementary Particles, Heidelberg, Germany (1967), North-Holland, 1968.
[19] I.Yu. Kobsarev, V.I. Zakharov, Ann. Phys. 60 (1970) 448.
[20] O.V. Teryaev, hep-ph/9904376 (1999).
[21] X.-S. Chen, D. Qing, F. Wang, Chin. Phys. Lett. 16 (1999) 403, hep-ph/9802347;
D. Qing, X.-S. Chen, F. Wang, Phys. Rev. D 58 (1998) 114032.
[22] S.J. Brodsky, F. Schlumpf, Phys. Lett. B 329 (1994) 111.
[23] S.J. Brodsky, R. Roskies, R. Suaya, Phys. Rev. D 8 (1973) 4574.
[24] F.E. Close, Phys. Lett. B 43 (1973) 422.
[25] S.J. Brodsky, J.R. Hiller, Phys. Rev. D 46 (1992) 2141.
[26] S.J. Brodsky, J.R. Hiller, G. McCartor, Phys. Rev. D 58 (1998) 025005.
[27] P.V. Landshoff, J.C. Polkinghorne, R.D. Short, Nucl. Phys. B 28 (1971) 225.
[28] S.J. Brodsky, F.E. Close, J.F. Gunion, Phys. Rev. D 8 (1973) 3678.
[29] B.-Q. Ma, J. Phys. G 17 (1991) L53;
B.-Q. Ma, Q.-R. Zhang, Z. Phys. C 58 (1993) 479.
[30] B.-Q. Ma, I. Schmidt, Phys. Rev. D 58 (1998) 096008.
[31] P. Hgler, A. Schfer, Phys. Lett. B 430 (1998) 179.

S.J. Brodsky et al. / Nuclear Physics B 593 (2001) 311335

[32] B.-Q. Ma, Phys. Lett. B 375 (1996) 320.

[33] B.-Q. Ma, I. Schmidt, J. Soffer, Phys. Lett. B 441 (1998) 461.

335

Nuclear Physics B 593 (2001) 336358

www.elsevier.nl/locate/npe

A unitary model for structure functions

and diffractive production at small x
A. Capella a , E.G. Ferreiro a, , C.A. Salgado a , A.B. Kaidalov b
a Laboratoire de Physique Thorique, Universit de Paris XI, Btiment 210, F-91405 Orsay Cedex, France 1
b ITEP, B. Cheremushkinskaya ulitsa 25, 117259 Moscow, Russia

Received 11 May 2000; revised 28 August 2000; accepted 12 September 2000

Abstract
We propose a unified approach which describes both structure functions in the small-x region and
diffractive production in p-interactions. It is shown that the model, based on reggeon calculus and
a quarkparton picture of the interaction, gives a good description of available experimental data in a
broad region of Q2 (including Q2 = 0) with a single Pomeron of intercept P (0) = 1.2. Predictions
for very small x are given and the problem of saturation of parton densities is discussed. 2001
Elsevier Science B.V. All rights reserved.
PACS: 12.40.Nn; 13.60.Hb

1. Introduction
Investigation of high-energy interactions of virtual photons with nucleons and nuclei
gives information on the dynamics of high-density partonic systems as well as on the
transition between perturbative and nonperturbative regimes in QCD.
Experiments at HERA have found two extremely important properties of small-x
physics: a fast increase of parton densities as x decreases and the existence of substantial
diffractive production in deep inelastic scattering (DIS).
From a theoretical point of view there are good reasons to believe that the fast increase
of the p with energy in the HERA energy range will change to a milder (logarithmic)
increase at much higher energies. This is due to unitarity effects, which are related to
shadowing in highly dense systems of partons with eventual saturation of densities.
This problem has a long history (for reviews see [14]) and has been extensively discussed
Corresponding author.

E-mail address: [email protected] (E.G. Ferreiro).

1 Unit Mixte de Recherche CNRS UMR n 8627.

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 5 5 - 1

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

337

in recent years [518]. It is closely connected to the problem of the dynamics of very
high-energy heavy ion collisions [1927].
In this paper we will address the problem of unitarization in p-interactions using the
reggeon calculus [28] with a supercritical Pomeron (P P (0)1 > 0). In this approach
the unitarization effects mentioned above are described by multi-Pomeron exchanges.
The problem of the Pomeron in QCD is not solved yet. The calculation based on QCD
perturbation theory leads to the so-called BFKL-Pomeron [2931] and the problem of
unitarization for such Pomeron has been considered by many authors [127] in the
leading logarithmic approximation. However, recent calculations [32,33] indicate that
NLO corrections are large and modify substantially the picture of high-energy interactions.
In particular the intercept of the leading Regge pole depends on nonperturbative effects
[3235].
Here, we will use a more phenomenological approach and will assume that the Pomeron
is a simple pole with an intercept P (0) = 1.2 determined from the analysis of high-energy
hadronic interaction when all multi-Pomeron contributions are taken into account [36].
In Ref. [37] it was suggested that the increase of the effective intercept of the Pomeron,
eff = 1 + eff , as Q2 increases from zero to several GeV2 is mostly due to a decrease
of shadowing effects with increasing Q2 . A parametrization of the Q2 dependence of eff
such that eff 0.1 for Q2 0 (as in soft hadronic interactions) and eff 0.2 (our input
or bare Pomeron intercept) for Q2 of the order of a few GeV2 , gives a good description
2
of all existing data on p total cross-sections in the region of Q2 <
5 10 GeV
2
[37,38]. At larger Q , effects due to QCD evolution become important. Using the above
parametrization as initial condition in the QCD evolution equation, allows to describe the
data in the whole region of Q2 studied at HERA [37,39].
In the reggeon calculus [28] the amount of rescatterings is closely related to diffractive
production. AGK-cutting rules [40] allow to calculate the cross-section of inelastic
diffraction if contributions of multi-Pomeron exchanges to the elastic scattering amplitude
are known. Thus, it is very important for self-consistency of theoretical models to describe
not only total cross-sections, but, simultaneously, inelastic diffraction. Indeed, in the
reggeon calculus the variation of eff with Q2 is directly related to the corresponding
variation of the ratio of diffractive to total cross-sections.
In this paper we present an explicit model which leads to the pattern of energy
2
2
2
behaviour of (tot)
p (W , Q ) for different Q described above. Moreover, it allows to
describe simultaneously diffraction production by real and virtual photons. The plan of
the paper is as follows. In Section 2 the model is described. Section 3 contains the
expressions of the total cross-section tot p . Formulae for diffractive production are given
(tot)

in Section 4. Section 5 is devoted to the description of experimental data on p (W 2 , Q2 )

and diffraction dissociation of the photon. In Section 6 we compare our results with results
of other authors on this subject, discuss the problem of saturation and give predictions
for future experiments. The parameters of the model and their numerical values are given
in Appendix A.

338

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

2. Description of the model

The amplitude of elastic p scattering at very high energy can be described in terms
of the single Pomeron exchange (Fig. 1a) and multi-Pomeron exchanges (Figs. 1b,c). The
2
(tot)
p ),
Pomeron exchange contribution to tot p (or to the structure function F2 = 4 Q
2
e.m.

increases as (1/x) at small x = Q2 /(W 2 + Q2 ). The contribution of n-Pomeron

exchange behaves as (1/x)n . These contributions are important at very small-x to
restore unitarity. A resummation of multi-Pomeron contribution is provided by Gribov
reggeon calculus [28]. In this approach a two-Pomeron exchange diagram (Fig. 1b), for
example, can be represented as a sum over all intermediate states (Fig. 2) which can be
diffractively produced by a single Pomeron exchange. Thus the magnitude of the two
Pomeron exchange in the elastic amplitude is related to the cross-section of diffractive
production. The multi-Pomeron exchanges (Fig. 1c) are related by AGK-cutting rules [40]
to shadowing corrections to diffractive production.
Let us first discuss the main mechanisms of diffraction dissociation of a proton
and of a virtual photon. For a proton the main contribution to the diagram of Fig. 2
is due to the proton intermediate state. The dissociation of the proton will be taken
into account in the quasi-eikonal approximation [41,42] which works well in
hadronic interactions. As for a virtual photon, it can diffractively produce vector
mesons (Fig. 3a) (the so-called quasi-elastic processes) and large mass states. For
M 2  Q2 , the latter correspond to the triple-Regge diagrams of Fig. 3b with P and f -exchanges in the t-channel. Models of this type allow to describe diffractive

Fig. 1. Single Pomeron exchange (a) and multi-Pomeron exchanges (b, c).

Fig. 2. Intermediate states that can be diffractively produced by a single Pomeron exchange.

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

339

Fig. 3. Diagram for diffractive vector mesons production (a) and triple-Regge diagram (b) with
P - and f -exchanges in the t -channel (P P P and P Pf ).

dissociation both at Q2 = 0 [71] and at large Q2 [43,44]. A simple quasi-eikonal

ansatz for multi-Pomeron exchanges based on this type of model was used in Ref. [45]
for the description of multiparticle production in DIS. However, the quasi-eikonal
approximation for diffractive dissociation of a highly virtual photon is difficult to interpret
in terms of reggeon diagrams while an explicit calculation of these diagrams for
multi-Pomeron exchanges leads to too many parameters for the matrix of transition
between different channels. For these reasons, we adopt here a different approach,
based on the quarkparton picture of high-energy interaction of a real or virtual
photon.
Taking into account that a photon dissociates into a q q-pair

and this pair can have

multiple interactions with the target, we separate such pairs into two groups: aligned
or asymmetric configurations, which will be denoted by A and the rest or symmetric
configurations denoted by S. The A-configuration was introduced by Bjorken and
Kogut [46] in the framework of the parton model in DIS and is characterized by a strongly
asymmetric distribution of the relative longitudinal momenta of q and q (zq  1, zq 1
or vice versa). This separation into A and S components is very important at large Q2 ,
where the A-configuration has a large transverse size and a large interaction cross-section
with the proton, while the S-component has small transverse size (1/Q) and a small crosssection. On the other hand the probability of the A-configuration decreases as 1/Q2 and
2
2
both configurations lead to the same behaviour, (tot)
p 1/Q , at large Q [46]. What
is important, however, is that these two components give rise to a different behavior for
diffraction: the A-component gives the same 1/Q2 behaviour for total and diffractive crosssections, while the S-component leads to a diffractive cross-section which decreases as
(1/Q2 )2 at large Q2 . These results are well known and have been extensively discussed in
the literature [4749] in connection with HERA experiments.
The q q-state

is described, in general, by non-perturbative dynamics and only for large

Q2 and small size component it is possible to use perturbative QCD.
Finally, we stress that at present energies, due to phase space limitations, only diagrams
with none or one triple reggeon interactions are important. In particular large mass
diffraction corresponds to a triple Pomeron interaction diagram. However, to be more

340

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

Fig. 4. Fan-type diagrams with Pomeron branchings.

general we resum all fan-type diagrams of the type shown in Fig. 4. This will be
discussed in more detail in the next section.

3. The total p cross-section

We formulate the expressions of cross-sections in the impact parameter space and take
into account that A- and S-components are diagonal:
Z




(tot)
(tot)
(1)
p s, Q2 = 4 d 2 b p b, s, q 2 ,
X






2
gi Q2 i(tot) b, s, Q2 , i = A, S.
=
(2)
(tot)
p b, s, Q
i

(Q2 )

is the probability of the ith component, which according to the discussion

Here gi
above, has been chosen in the form


gA Q2 =

gA (0)
;
1 + Q2 /m2A

gS Q2 = gS (0),

(3)

where gA (0), and m2A are considered as phenomenological free parameters.

(tot)
The cross-sections i (b, s, q 2 ) are written in the form

1 exp(Ci (b, s, Q2 ))
,
b, s, Q2 =
2C
P (s, b, Q2 )




i0
f
+ i0 s, b, Q2 .
i s, b, Q2 =
2
1 + a3 (s, b, Q )

(tot)

The expressions for the eikonals i0 are:





b2
fi (Q2 )
k
exp k i
s, b, Q2 = Cik i
i0
0k (Q2 , )
40k
ith k = P , f and i = A, S where

(4)
(5)

(6)

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

s + Q2
k = k (0) 1;
= ln
,
s0 + Q2


2
Q2 + k0 .
i0k = R0ki

341

(7)

The quantity is chosen in such a way as to behave as ln x1 for large Q2 and as ln ss0 for
Q2 = 0. The functions fi (Q2 ) have been chosen in the form 2

, i = S,
2
(8)
fi Q = 1 + Q2 /m2S

1,
i = A.
2
is given in Appendix A. The slopes of the
The parametrization of the radius R0ki
0
2
0
trajectories P = 0.25 GeV and f = 0.9 GeV2 were fixed at their values known from
soft hadronic interactions.
P (0)r
We turn next to the denominator in the first term of Eq. (5). Here a = gpp
P P P /16
P
where gpp (0) and rP P P (0) are, respectively, the Pomeron-proton coupling and the triple
Pomeron vertex, both at t = 0. The expression of 3 (s, b, Q2 ) is given in Section 4. Note
that Eqs. (4)(6) correspond to a quasi-eikonal model [41,42]. For a = 0, its Born term,
Eq. (5), is a sum of Pomeron and secondary (f ) reggeon exchanges. The latter is included

in order to be able to use the model at energies as low as s 10 GeV. The coefficient C
takes into account the dissociation of the proton and is taken to be equal to 1.5 [41,42]. To
first order in a, the Born term, Eq. (5), also contains the contribution of P P P and Pf P
triple Regge terms (see Fig. 5). This can be easily seen by developing the first term of the
r.h.s. of Eq. (5) to first order in a, and using the expression of 3 in Eqs. (26) and (27).
Taken to all orders in a, Eq. (5) corresponds to the resummation of all fan-type diagrams
of the type shown in Fig. 4, using the Schwimmer formula [50]. Here, however, each
new branching contains not only a Pomeron as in Fig. 4 but also a secondary f -reggeon
exchange. These branchings are controlled by the parameters aP and af , respectively
(see Eq. (26)). 3 As discussed in the last paragraph of Section 2, diagrams with more than
one branching (i.e., more than one triple reggeon interaction) are not important at present
energies due to phase space limitations.
Note that the triple Pomeron interaction introduces a large (gluonic) size component in
our S-component. Actually, the separation between the two components is washed out at
large Q2 due to QCD evolution. Note also that the second term of Eq. (5) (f -exchange)
does not contain a denominator as does the first one. Such a denominator would add
to the f -exchange f Pf and fff terms. However, such double reggeon f P and ff
P f and ( f )2 , respectively).
exchanges are already present in Eq. (4) (terms i0
i0
i0
In the above formulae we have neglected the real parts of the reggeon exchanges given by
the signature factors. Since P (0) is not very different from unity, their effect is expected to
be rather small except when the contribution of the multi-Pomeron exchanges becomes
2 In Eqs. (3) and (8) we used the simplest parametrization of g (Q2 ) and f (Q2 ) consistent with our
i
i
requirements. In general gS (Q2 ) and fA (Q2 ) can depend on Q2 .
3 Eq. (5) is valid when these couplings are the same irrespectively of whether the branching takes place off a
Pomeron or an f .

342

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

Fig. 5. Triple Pomeron diagrams: (a) triple Pomeron P P P already represented in Fig. 3b; (b) the
interference term Pf P .

very important. As for secondary (f ) exchange, the signature factor enters only in terms
with multiple f -exchange which are not important at the comparatively high energies
of the present experiments.
Many parameters of the model such as R0ki are strongly constrained by the data on
hadronic interactions and were fixed (see Appendix A). Let us also note that the f
(tot)
exchange in p is related to the valence quarks contribution, which for small x belongs
f

to the A-component. Thus CS = 0.

The list of all parameters and their numerical values is given in Appendix A.

4. Diffractive production
Let us now introduce the expressions for diffraction dissociation of a virtual photon.
They can be obtained, using AGK-cutting rules, from the imaginary part of the p-elastic
scattering, given by Eqs. (1)(8). Neglecting s-channel iterations of the triple-Pomeron
graphs, the total diffraction dissociation cross-section can be written as a sum of three
components
X (0)
(dif)
i + P P P ,
(9)
p =
i=A,S

where
(0)
i

= 4gi2

and
P P P = 2

Z 

2
(tot)
i
Q
b, s, Q2 d 2 b
2

gi2 Q2

2
Pi P P b, s, Q2 e2Ci (s,b,Q ) d 2 b,

(10)

(11)

where Pi P P (b, s, Q2 ) = aiP (s, b, Q2 )3 (s, b, Q2 ), and iP (s, b, Q2 ) is given by the

first term in the r.h.s. of Eq. (5). For the total diffractive production cross-section, which
includes the diffraction dissociation of the proton, it is necessary to multiply expressions
(10), (11) by the factor C = 1.5, introduced in Section 3.

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

343

Note that to first order in a, P P P corresponds to the sum of the two triple reggeon
contributions in Fig. 5. Taken to all orders in a, P P P corresponds to the resummation of
all fan-type diagrams obtained by adding to the diagrams in Fig. 5, multiple branchings
with Pomeron and f -reggeon at each branching (see the discussion in Section 3). In the
following, the sum of these two types of contributions will be referred to as the triple
Pomeron (P P P ) although the diagram in Fig. 5b corresponds to an interference term.
With gaussian forms of the Born term in impact parameter, Eq. (6), the expression (10)
can be also written as follows (see [42])
  

gi2 (Q2 )Bi
Zi
(0)
f
f (Zi ) ,
(12)
i =
C
2
where
f (Z) =

X
(Z)n1
n=1

n n!

i s, b, Q2 d 2 b,
Z


8C
i2 s, b, Q2 d 2 b.
Zi =
Bi

Bi = 2

(13)
(14)
(15)

In the next sections we will compare our model with differential diffractive cross-sections
as functions of M 2 (square of the mass of the diffractively produced system) or of =
Q2 /(M 2 + Q2 ) (which is convenient at large Q2 and plays the same role for the Pomeron
as the variable x for the proton 4 ) or of the variable xP (xp = x/). Present experiments
at HERA measure the differential cross-sections integrated over the variable t (square of
(3)
is
the momentum transfer between initial and final protons) and usually the function F2D
introduced:
Z
Q2
d
(3)
dt.
(16)
xP
xP F2D =
4 2 e.m.
dxP dt
In our model it can be written as a sum of three terms
!
X (3)




(3)
(3)
2
2
F2Di x, Q , + F2DP P P x, Q , ,
(17)
F2D =
i

where
(3)

(3)B

F2Di = F2Di i .

(18)

(3)B
(x, Q2 , ) are the lowest order (Born) approximations for these functions (their
F2Di
explicit forms are given below) while i are suppression factors due to higher order
multi-Pomeron exchanges
  

Zi
8
(19)
f (Zi ) .
i = f
2
Zi
4 It should be noted that the structure function of the Pomeron F (, Q2 ) can be used only for the cases
P
when multi-Pomeron exchanges are small, because in general it is impossible to separate out this amplitude.

344

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

(3)
The functions F2Di
satisfy the condition
xZPmax

(3)
F2Di
dxP = i(0) x, Q2

xPmin

Q2
,
4 2 e.m.

(20)

where
xPmin =

x
max

xPmax = 0.1;

max =

Q2
2 + Q2
Mmin

2 = 4m2 .
with Mmin

The -dependence of the A-component was chosen according to Ref. [43,44], where
the small--dependence was determined from the P Pf -triple-Regge asymptotics with a
simple ansatz for the behaviour at 1


(3)B(P )
x, Q2 ,
xP F2DA
2
2 (Q2 ) Z


2
Q2 gA
2P f (1 )nP (Q )
2
P
2
(21)
d b A x, Q , b R d
=
max
4 2 e.m.
2p f (1 )nP (Q2 )
min

where
=

x
Q2 + S0
= ;
x
Q2 + M 2

min =

x
= 10x.
xPmax

In the following all powers of 1 are chosen according to the arguments presented in
Refs. [37] and [43,44]:




Q2
1 3
(22)
nP Q2 = +
2 2 c + Q2
with c = 3.5 GeV2 .
The S-component should be maximal at close to 1 and was chosen in the form
Z

Q2 gS2 (Q2 )

(3)B
x, Q2 , =
d 2 b S2 x, Q2 , b R d .
xP F2DS
2
max
4 e.m.

min

(23)

This component contains the contribution of comparatively small masses and should
decrease as 1 faster than P Pf contribution, Eq. (21). Our choice satisfies this
condition. Note that from perturbative QCD a behaviour 3 has been found [49]. However,
the contribution of this component at large Q2 , where perturbative QCD is applicable, is
small ( m2S /Q2 ). Our results are rather insensitive to the power of in the range 1 to 3
[51].
(3)
2
The triple-Pomeron (P P P plus Pf P ) contribution F2DP
P P (x, Q , ), is given by



P P P
(3)
(3)B
2
2
,
(24)
xP F2DP
P P x, Q , = xP F2DP P P x, Q ,
PBP P
where P P P is given by Eq. (11), its Born term, PBP P , by the same equation with C = 0,
and

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

(3)B
2
xP F2DP
P P x, Q ,
Z
X






Q2
2
2a
d
b
gi2 Q2 iP b, s, Q2 3 s, b, Q2 , ,
=
2
4 e.m.

345

(25)

with

3 s, b, Q , =
2


k exp

b2
x)
4k (/

 k
2

(1 )nP (Q )+4
.
x)
x
k (/

(26)

Here P = 1, f determines the relative strength of the Pf P -contribution and k =

2 + 0 ln(/
x).
The function 3 (s, b, Q2 ), which enters in Eqs. (5) and (11) is given by
R1k
K


3 s, b, Q2 =

Zmax

d
3 s, b, Q2 , .

(27)

min

Since the triple Pomeron formula is not valid for low masses, we use here Mmin = 1 GeV.
(3)B
Note that xP F2DP P P is obtained from PBP P replacing 3 (s, b, Q2 ) by 3 (s, b, Q2 , ).
In this way
Z
Q2
(3)
P P P .
dxP F2DP
P P dxP =
4 2 e.m.
5. Comparison with experiment
The model outlined above was used for a joint fit of the data on the structure function
(3)
(x, Q2 , ) in the region of small x (x <
F2 (x, Q2 ) and diffractive structure function F2D
102 ) and Q2 6 10 GeV2 . 5 The full list of parameters and their values either fixed or
obtained in the fit are given in the Appendix A.
In Fig. 6, the dependence of p as a function of Q2 for different energies is shown
and Fig. 7 gives the x-dependence of F2 (x, Q2 ) for different Q2 . Our results are compared
with experimental data. The description of the data is excellent. Note that the x-dependence
and its variation with Q2 is fixed in the model and is strongly correlated with the ratio of
(dif)
(tot)
p / p and its dependence on Q2 . The model confirms the Q2 -dependence of the
2

2 (x,Q )
, assumed in Ref. [37]. The function eff depends not
effective intercept eff = d lnd Fln(1/x)
2
only on Q , but also on x and it is shown in Fig. 8 for two different intervals of x. The
function eff decreases as Q2 or x decreases due to the increase of shadowing effects. It is
interesting to note that, for Q2 = 0, eff in the region of HERA energies is close to 0.13,
i.e., substantially higher than in hadronic interactions. For s the cross-section (tot)
p
has a Froissard-type behavior in ln2 ss0 .
In Figs. 9 and 10 the results of the model are compared with experimental data on
diffractive structure function F2D(3) (x, , Q2 ). The dependence for fixed values of Q2
and xP is shown in Fig. 9 and the xP -dependence for different values of Q2 and

5 Actually, for F only data with Q2 6 3.5 GeV2 were included in the fit.
2

346

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

Fig. 6. The p cross-section as a function of Q2 for different energies compared to the following
experimental data: H1 1995 [63] (black points), ZEUS 1995 [64] (black squares), E665 [65] (black
triangles) and ZEUSBPT97 [66] (white circles). The data at Q2 = 6.5 GeV2 (H1 1994 black
stars, and ZEUS 1994 white stars) are from [67,68].

is presented in Fig. 10. The model reproduces the experimental data rather well. For
comparison we limited ourselves to the data at low values of xP 6 102 , where the effect
of the non-diffractive RRP contribution (which is not included in the model) is small.
Diffractive production at Q2 = 0 is reasonably well reproduced as well (Fig. 11). The
(3)
2D
eff ) is shown in Fig. 12
dependence of diffractive cross-sections on dif
eff (F2D (1/xP )
as a function of Q2 (for fixed M). The experimental points are also shown. The model
2
2
2
predicts a weak dependence of dif
eff on Q for Q > 5 10 GeV and QCD-evolution
does not change this result [43,44].

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

347

Fig. 6. (continued).

6. Discussion and conclusions

The model we propose here for the description of total and diffractive cross-sections of
interaction of a virtual photon with a proton is a natural generalization of models used for
the description of high-energy hadronic interactions. The main parameter of the model
intercept of the Pomeron, P P (0) 1, was fixed from a phenomenological study of
these interactions (P = 0.2) and was found to give a good description of p-interactions
in a broad range of Q2 (0 6 Q2 < 10 GeV2 ). Note that a single Pomeron is present in the
model.
It should be noticed that at higher values of Q2 , QCD evolution is important and our
results should be used as initial condition in DGLAP evolution (see Refs. [39,43,44]). In
particular, due to QCD evolution, the effective Pomeron intercept in F2 will reach values
significantly larger than 1.2 at large Q2 . For diffractive process the situation is different.
In this case, the effective Pomeron intercept at large Q2 is not significantly modified as
a result of QCD evolution. Moreover, the latter has rather small effect for intermediate
values of [43,44]. For these reasons our model, without QCD evolution, can be used
here at comparatively larger values of Q2 .

348

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

Fig. 7. F2 as a function of x for different values of Q2 . The experimental data are the same of Fig. 6.

Another important parameter of the theory is the triple Pomeron vertex, which appears
in the parameter a. It is quite remarkable that the same value of this parameter allows to
describe diffractive data both at Q2 = 0 and at a few GeV2 . Thus, in our approach, the
value of this parameter can be determined from soft diffraction data [52].
The variation of the photon virtuality gives a unique possibility to study unitarization
effects for different scales. Let us discuss the qualitative features of our model and its
relation to the models introduced by other authors [518]. It has been mentioned above
that the different role of small and large size components of a virtual photon has been
emphasized by many authors [518] and perturbative calculations of unitarization effects
for p have been carried out in Refs. [518]. In our opinion the perturbative calculations
can be valid only for the small size component (S) at large Q2 . An interesting analysis
of unitarization effects for both total p-cross-section and diffractive production was
performed by Golec-Biernat and Wsthoff [518]. These authors, however, postulated
an unconventional form for the elastic p-amplitude and it is not clear which diagrams

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

349

Fig. 7. (continued).

lead to these unitarization effects. 6 We agree with their conclusion that for the structure
function F2 (x, Q2 ) the unitarization effects at HERA energies are important mostly in the
2
2
2
region of rather small Q2 <
2 4 GeV . Indeed, for Q > 2 4 GeV , the experimental
value of eff is close to the one of the input (bare) Pomeron P = 0.2. Within our approach
this is only possible if the S-component (where unitarity corrections are higher twists)
dominates. However, some unitarity corrections are present at large Q2 due to the Acomponent and the triple Pomeron interaction. Moreover, they will become more important
at higher energies (see the discussion in the next paragraph). However, we differ in many
details from their predictions on the properties of diffractive production at large Q2 and on
6 Correlations in hadronic interactions have been studied for many years. It is known that they can be described
[53] in an eikonal model where the Born term, at fixed impact parameter, is used as eikonal. In the model
of Golec-Biernat and Wsthoff [518], on the contrary, the Born term is integrated over impact parameter and
then eikonalized. The unitarity effects obtained in this way are very large. It is unlikely that they can describe the
observed features of correlations.

350

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

Fig. 8. The function eff as a function of Q2 for 2 intervals of x. The solid line corresponds to the
interval x = (Q2 + s0 )/(Q2 + s) = 1.85 104 1.16 105 and the dashed line to the interval
x = 7.41 1014 4.64 1015 .

the pattern of saturation of parton densities at large Q2 and extremely small x.

Let us now discuss the saturation patterns in our model, i.e., the properties of
(tot)
p (s, Q2 ) in the limit s . It follows from formulae (1)(8) that for the A-component,
(tot)

which has a large cross-section, p (s, b, Q2 ) will tend to a saturation limit 1/2C rather
fast as energy increases. On the other hand, for the small size (at large Q2 ) component (S)
the situation is different. If we neglect the triple-Pomeron contribution (i.e., with a = 0 in
Eq. (5)),



Cm2S
b2
exp

S s, b, Q2
P
Q2
4S0P
will increase with energy until the increase of exp(P ) does not overcompensate the
smallness of m2S /Q2 . For such extremely large energies, cross-sections in the small impact
parameter region will saturate to a Q2 -independent value and F2 (x, Q2 ) Q2 . This is
a usual picture of saturation in perturbative QCD [127]. However, the inclusion of the
P P P -fan diagrams (large distance, nonperturbative effects) according to Eqs. (5), (26),
(27) changes the picture drastically. In this case the increase of 3 (s, b, Q2 ) exp( )

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

D(3)

Fig. 9. The diffractive structure function xP F2

The experimental data (H1 1994) are from [69].

351

as a function of for fixed values of Q2 and xP .

will compensate the increase of S (s, b, Q2 ) at very large and lead, at large Q2 , to a
(tot)
behaviour p Q12 f (ln Q2 ) even for x 0. (Note that for large enough Q2 this will
happen in a region where S is still small.)
The above effects are relatively small at present energies due to the smallness of the
triple Pomeron vertex rP P P which determines the constant a. However, in the saturation
2
limit the properties of (tot)
p (s, Q ) in our model are quite different from those in other
(tot)

models [527]. Our predictions for p (s, Q2 ) for energies higher than those accessible
at HERA are shown in Fig. 13.
The fan type diagrams with triple Pomeron interaction have been considered in the
perturbative QCD in several papers [527,5461] using leading order BFKL Pomeron.
However in this approximation the Pomeron is not a Regge-pole and the triple Pomeron
vertex depends on external virtuality, contrary to the factorized case considered by us. This
difference leads to a different dependence of structure functions on Q2 in the saturation
region. We think that in the NLO approximation of perturbative QCD, when the leading
singularity in the j -plane becomes a factorizable Regge pole the properties of structure
functions in the saturation region will be closer to the behaviour considered in our model.

352

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

D(3)

Fig. 10. The diffractive structure function xP F2

as a function of xP for fixed values of Q2 and .
The experimental data (H1 1994 and H1 1995) are from [69] and [70].

In a forthcoming publication [51] we introduce a different but closely related model

with also two components for the interaction of the q q pair: a large-size component,
parametrized in the same way as the A-component in the present work, and a smallsize component computed using the q q perturbative QCD wave function with its
longitudinal and transverse components. The results are very similar to the ones obtained
here. Again, a single Pomeron with intercept 1.2 allows to describe the data on both total
p cross-section and diffractive production.
Acknowledgements
It is a pleasure to thank K. Boreskov, O. Kancheli, G. Korchemski, U. Maor and
C. Merino for discussions. This work is partially supported by NATO grant OUTR.LG
971390. E.G.F. and C.A.S. thank Ministerio de Educacin y Cultura of Spain for financial
support.

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

353

Fig. 11. Diffractive production at Q2 = 0 GeV for two different energies. The experimental data are
from [71].

Appendix A
2
In this appendix we give the list of all the parameters in the formulae for (tot)
p (s, Q ) (or
F2 (x, Q2 )) and diffractive production, given in Sections 3 and 4, respectively together
with their numerical values. The total cross-section of p interactions is related to the
structure function F2 as follows

4 2 e.m.


2
F2 x, Q2 ,
=
(tot)
p s, Q
Q2

(A.1)

where s = W 2 and x = Q2 /(1 + Q2 ). This cross-section is related by Eq. (1) to the

2
corresponding quantity, (tot)
p (b, s, Q ), in the impact parameter space. The latter is written
as the sum of two components (Eq. (2)) with probabilities given Eq. (3) where gA (0), gS (0)
and m2A are free parameters. Their values obtained from the fit are

354

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

2
Fig. 12. The effective Pomeron slope D
eff as a function of Q for MX = 5 GeV. The experimental
values are from ZEUS 1994 [72].

gA (0) = 0.15e.m. ,

gS (0) = 0.367e.m. ,

m2A = 0.227 GeV2 .

(A.2)

For a = 0 (i.e., when triple reggeon vertices vanish; see Eq. (5)) the cross-sections
are written in a quasi-eikonal form (Eq. (4)) with Born terms given by the
exchange of the Pomeron P and secondary reggeon (f ) trajectories). Their expressions
contain the intercepts k (0) = 1 + k (k = P , f ) for which we take

i(tot)(b, s, Q2 )

P = 0.2,

f = 0.3.

(A.3)
f

They also contain five other parameters treated as free ones: CAP , CSP , CA (CS = 0), s0 and
m2S . Their values obtained in the fit are:
m2S = 1.343 GeV2 ,

s0 = 0.463 GeV2 ,

CAP = 0.830 GeV2 ,

CSP = 0.807 GeV2 ,

CA = 14.28 GeV2 .
f

(A.4)

The value of the Pomeron intercept P (0) = 1.2 has been found in soft hadronic
interactions [36]. It is also consistent with the intercept of the BFKL Pomeron at NLO.
When treated as a free parameter in the fit its value turns out to be very close to the one
in Eq. (A.3). The value of the intercept of the f has been allowed to change within some

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

355

Fig. 13. Our predictions for the p cross section at energies higher than those accessible at HERA.
The experimental data are the same as in Figs. 6 and 7.

restricted limits. The value f (0) = 0.7 leads to better results than the more conventional
value f (0) = 0.5.
Our expressions contain also the triple Pomeron term P P P whose strength is controlled
by the parameter a in Eq. (5), and the interference term Pf P whose strength (relative to
the P P P term) is given by the parameter f . We use
a = 0.052 GeV2 ,

f = 8.

(A.5)

Note that the value of a obtained in the fit agrees with the value obtained from soft
diffraction (see the discussion in Section 6).
Finally we turn to the quantities


2
(A.6)
Q2 + k0
i0k = R0ki
and
2
+ k0 ln(/x)
k = R1k

(A.7)

356

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

in Eqs. (5) and (26). They are directly related to the t-dependence of the elastic p
amplitudes and of the triple reggeon vertices, respectively, and have the standard Regge
behavior as functions of or /x. We use the conventional parameters of the reggeon
slopes
P0 = 0.25 GeV2 ,

f0 = 0.9 GeV2 .

(A.8)

Our results for the diffractive cross-sections integrated over t are not sensitive to the
values of the parameters and we have fixed them to the values obtained from soft
interaction data and from vector dominance. We take [62]


2
2
2
+ Rpk
,
Q2 = RvkA
R0kA
(A.9)
2


RvkS
2
2
2
+ Rpk
R0kS Q =
Q2
1+ m
2

with
2
2
= RPf
= 2 GeV2 ,
RpP

2
Rvki
= 1 GeV2

(A.10)

and
2
2
= R1f
= 2.2 GeV2 .
R1P

(A.11)

Note that the Q2 -dependence of the P and f vertices for the S component have
been chosen in such a way that the corresponding radii tend to zero for Q2 [62].
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]

L.V. Gribov, E.M. Levin, M.G. Ryskin, Phys. Rep. 100 (1983) 1.
E. Laenen, E. Levin, Ann. Rev. Nucl. Part. 44 (1994) 199.
A.B. Kaidalov, Surveys High Energy Phys. 9 (1996) 143.
A.H. Mueller, hep-ph/9911289.
A.H. Mueller, Nucl. Phys. B 437 (1995) 107.
A.L. Ayala, M.B. Gay Ducati, E.M. Levin, Phys. Lett. B 388 (1996) 188.
A.L. Ayala, M.B. Gay Ducati, E.M. Levin, Nucl. Phys. B 493 (1997) 355.
A.L. Ayala, M.B. Gay Ducati, E.M. Levin, Nucl. Phys. B 510 (1998) 355.
E. Gotsman, E. Levin, U. Maor, Nucl. Phys. B 493 (1997) 354.
E. Gotsman, E. Levin, U. Maor, Phys. Lett. B 425 (1998) 369.
E. Gotsman, E. Levin, U. Maor, Phys. Lett. B 452 (1999) 387.
E. Gotsman, E. Levin, U. Maor, E. Naftali, Nucl. Phys. B 539 (1999) 535.
L.L. Frankfurt, W. Kopf, M. Strikman, Phys. Rev. D 57 (1998) 512.
L.L. Frankfurt, W. Kopf, M. Strikman, Phys. Rev. D 54 (1996) 3194.
M. Mac Dermott, L.L. Frankfurt, V. Guzey, M. Strikman, hep-ph 9912547.
K. Golec-Biernat, M. Wsthoff, Phys. Rev. D 59 (1999) 014017.
K. Golec-Biernat, M. Wsthoff, Phys. Rev. D 60 (1999) 114023.
A.H. Mueller, Eur. Phys. J. A 1 (1998) 19.
L. Mc Lerran, R. Venugopalan, Phys. Rev. D 49 (1994) 2233, 3352.
L. Mc Lerran, R. Venugopalan, Phys. Rev. D 50 (1994) 2225.
L. Mc Lerran, R. Venugopalan, Phys. Rev. D 53 (1996) 458.

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]

[67]
[68]
[69]

357

J. Jalilian-Marian et al., Phys. Rev. D 59 (1999) 014014, 034007.

A. Kovner, L. Mc Lerran, H. Weigert, Phys. Rev. D 52 (1995) 3809, 6231.
Yu.V. Kovchegov, A.H. Mueller, Nucl. Phys. B 529 (1998) 451.
Yu.V. Kovchegov, A.H. Mueller, S. Wallon, Nucl. Phys. B 507 (1997) 367.
A.H. Mueller, Nucl. Phys. B 558 (1999) 285.
A. Capella, A. Kaidalov, J. Tran Thanh Van, Heavy Ion Phys. 9 (1999) 169.
V.N. Gribov, ZhETF 57 (1967) 654.
L.N. Lipatov, Sov. J. Nucl. Phys. 23 (1976) 338.
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP 45 (1977) 199.
Y.Y. Balitsky, L.N. Lipatov, Sov. J. Nucl. Phys. 28 (1978) 822.
V.S. Fadin, L.N. Lipatov, Phys. Lett. B 429 (1998) 127.
G. Camici, M. Ciafaloni, Phys. Lett. B 395 (1997) 311.
L.P.A. Haackman, O.V. Kancheli, J.H. Koch, Nucl. Phys. B 518 (1998) 275.
M. Ciafaloni, D. Colferai, G.P. Salam, JHEP 9910 (1999) 017.
A.B. Kaidalov, L.A. Ponomarev, K.A. Ter-Martirosyan, Sov. J. Nucl. Phys. 44 (1986) 468.
A. Capella, A. Kaidalov, C. Merino, J. Tran Thanh Van, Phys. Lett. B 337 (1994) 358.
A. Kaidalov, C. Merino, Eur. Phys. J. C 10 (1999) 153.
A. Kaidalov, C. Merino, D. Perterman, hep-ph/9911331.
V.A. Abramovsky, V.N. Gribov, O.V. Kancheli, Sov. J. Nucl. Phys. 18 (1974) 308.
K.A. Ter-Martirosyan, Nucl. Phys. B 36 (1972) 566.
K.A. Ter-Martirosyan, Phys. Lett. B 36 (1973) 566.
A. Capella, A. Kaidalov, C. Merino, J. Tran Than Van, Phys. Lett. B 343 (1995) 403.
A. Capella, A. Kaidalov, C. Merino, D. Perterman, J. Tran Than Van, Phys. Rev. D 53 (1996)
2309.
A. Capella, A. Kaidalov, V. Nechitailo, J. Tran Than Van, Phys. Rev. D 58 (1998) 014002.
J.D. Bjorken, J.B. Kogut, Phys. Rev. D 8 (1973) 1341.
N.N. Nikolaev, B.G. Zakharov, Zeit. Phys. C 49 (1990) 607.
L.L. Frankfurt, M. Strikman, Phys. Rep. 160 (1988) 235.
J. Bartels, J. Ellis, H. Kowalski, M. Wsthoff, hep-ph/9803497.
A. Schwimmer, Nucl. Phys. B 94 (1975) 445.
A. Capella, E.G. Ferreiro, A.B. Kaidalov, C.A. Salgado, in preparation.
A. Kaidalov, Phys. Rep. 50 (1979) 157.
A. Capella, A. Krzywicki, Phys. Rev. D 18 (1978) 4120.
I. Balitskii, Nucl. Phys. B 463 (1996) 99.
Y.V. Kovchegov, Phys. Rev. D 60 (1999) 034008.
Y.V. Kovchegov, Phys. Rev. D 61 (2000) 074018.
H. Weigert, hep-ph/0004044.
J. Jalilian-Marian, A. Kovner, H. Weigert, Phys. Rev. D 59 (1999) 014015.
J. Jalilian-Marian, A. Kovner, A. Leonidov, H. Weigert, Phys. Rev. D 59 (1999) 034007.
A. Kovner, J.G. Milhano, Phys. Rev. D 61 (2000) 014012.
A. Kovner, J.G. Milhano, H. Weigert, hep-ph/0004014.
L.P.A. Haakman, A. Kaidalov, J.H. Kach.
C. Adloff et al., H1 Collaboration, Nucl. Phys. B 497 (1997) 3.
J. Breitweg et al., ZEUS Collaboration, Phys. Lett. B 407 (1997) 432.
M.R. Adams et al., E665 Collaboration, Phys. Rev. D 54 (1996) 3006.
A. Zhokin, H1 Collaboration, ZEUS Collaboration, in: I. Sarcevic, C.-I. Tang (Eds.),
Contribution to the Proc. of the XXIX International Symposium on Multiparticle Dynamics
(ISMD99), Brown University, Providence, USA, August 913, 1999, World Scientific, in press.
M. Derrick et al., ZEUS Collaboration, Z. Phys. C 72 (1996) 399.
S. Aid et al., H1 Collaboration, Nucl. Phys. B 470 (1996) 3.
C. Adloff et al., H1 Collaboration, Z. Phys. C 76 (1997) 613.

358

A. Capella et al. / Nuclear Physics B 593 (2001) 336358

[70] H1 Collaboration, in: 29th International Conference on High-Energy Physics ICHEP98,

Vancouver, Canada, July 1998.
[71] C. Adloff et al., H1 Collaboration, Z. Phys. C 74 (1997) 221.
[72] J. Breitweg et al., ZEUS Collaboration, Eur. Phys. J. C 6 (1999) 43.

Nuclear Physics B 593 (2001) 361397

www.elsevier.nl/locate/npe

Non-linear -model for gauge field disorder

T. Guhr a , T. Wilke b,
a Max-Planck-Institut fr Kernphysik, Postfach 103980, D-69029 Heidelberg, Germany
b The Niels Bohr Institute, Blegdamsvej 17, DK-2100 Copenhagen , Denmark

Received 18 August 2000; accepted 13 October 2000

Abstract
Various studies indicate similarities between the statistical features of quantum chromodynamics
(QCD) and disordered systems in condensed matter physics. To further investigate this, we propose
and analytically discuss a statistical model which merges Efetovs approach to disordered systems
with the principles of chiral symmetry and QCD. The object of our interest are the spectral
fluctuations of a Dirac particle which propagates in a finite four-dimensional box under the influence
of gauge fields. In our model, the gauge fields are replaced with a stochastic white noise potential,
which we refer to as gauge field disorder. We arrive at effective supersymmetric non-linear -models
and spontaneous breaking of supersymmetry is found. A fundamental scale of the stochastic feature
in QCD is set by the equivalent of the Thouless energy in QCD which appears naturally in our
model. We discuss various connections to other low-energy effective theories, in particular to the
NambuJona-Lasinio model and to chiral perturbation theory. 2001 Elsevier Science B.V. All
rights reserved.
PACS: 12.40.Ee; 11.30.Rd; 12.38.Lg
Keywords: Chiral symmetry breaking; Statistical models; Thouless energy; Supersymmetry

1. Introduction
Random Matrix Theory (RMT) is a widely used statistical concept to model spectral
fluctuations in a variety of different systems, ranging from chaotic billiards and light
atoms to many body systems such as molecules and nuclei, for a review see Ref. [1].
The universality of these fluctuations allows one to efficiently separate scales. Some years
ago, Shuryak and Verbaarschot [2] extended this concept to quantum chromodynamics
(QCD) by taking chirality into account. Chiral Random Matrix Theory (chRMT) has found
numerous applications since. In particular, it was shown that chRMT provides a remarkably
good description of data from lattice QCD [3].
* Corresponding author

E-mail address: [email protected] (T. Wilke).

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 1 3 - 1

362

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

More recently, a close similarity between certain features of disordered systems in

condensed matter physics and QCD was suggested and discussed. The propagation of an
electron in a disordered solid, such as a piece of wire, is diffusive due to multiple scattering
at the impurities of the crystal. The diffusive motion determines, together with the size of
the probe, the dimensionless conductance. The latter directly relates to a fundamental scale,
the Thouless energy, for a review see Ref. [1]. Employing the Gell-MannOakesRenner
relation, it was argued in Refs. [46] that the propagation of a quark under the influence
of the gauge fields should yield similar statistical features. Indeed, the equivalent of the
Thouless energy was found in numerical data of the instanton liquid model of QCD [4,5]
and of lattice QCD [7,8]. The Thouless energy was identified by analyzing the spectral
fluctuation which are of RMT type for energy separations smaller than the Thouless energy
and which deviate for larger energy separations. The identification of this fundamental
scale in lattice QCD was performed in the microscopic region near zero virtuality [8]
where chirality is very important and in the macroscopic or bulk region far away from zero
virtuality where chirality does not influence the fluctuation properties any more [7]. In the
microscopic region, the existence of a Thouless energy can be understood by employing
chiral perturbation theory [9,10]. Remarkably, in the bulk region, there is no such direct
link to an effective theory.
Our aim is to describe the propagation of a Dirac particle in the QCD vacuum by a
stochastic field theory. For this purpose, we generalize the stochastic model known from
mesoscopic physics to a covariant formulation in four space time dimensions including
chiral symmetry. We do so by merging Efetovs supersymmetric approach to disordered
systems [11,12] with the principles of QCD and chiral symmetry. This extends and
complements the semi-classical reasoning of Refs. [46]. We will not use formal analogies
to the GOR relation as in Refs. [46], rather we will derive the Thouless energy as a natural
result from our stochastic model. This holds in the bulk as well in the microscopic regime.
Within our model, the regimes are distinguished by the symmetries, but the existence of
the Thouless energy itself is generic for both.
Related alternative approaches are obtained from lattice descriptions of the Dirac
operator [13,14] or extended chiral matrix model [15].
An important remark is in order: in our context of statistical physics, supersymmetry
is used as a powerful tool to perform averages. Thus, the bosons and fermions in our
models do not relate in any direct way to physical particles. This is different from standard
applications of supersymmetry in high energy physics.
The paper is organized as follows. In Section 2, we present some heuristic arguments
which shall help the reader to develop an intuition for the physical effects that we wish to
address later on analytically. To this end we review some stochastic features of disordered
systems and establish, anticipating our analytical results, the link to QCD. In Section 3, we
discuss a relativistic fermion moving in a four-dimensional chirally symmetric stochastic
disorder potential. Thereby, we setup the model and express it in terms of a supersymmetric
field theory. We also analyze its symmetries. In Section 4, the supersymmetric field theory
is treated within a saddle point approximation. This is equivalent to a mean field approach.
Relations to the NambuJona-Lasinio model are found. We arrive at two effective theories

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

363

in Section 5, in a certain macroscopic regime, we obtain Efetovs non-linear sigma model

in four dimensions. In the microscopic regime, the functional reduces to chiral perturbation
theory. This results in another connection to low energy models of QCD. We summarize
and conclude in Section 6. Some calculations and basic definitions are shifted into the
appendices.

2. Similar stochastic features of QCD and disordered systems

We consider an electron of mass m moving in a probe, a small piece of wire of about
1 m, for example. The electron is scattered at the impurities of the crystal. We assume
that the phase coherence length is the largest scale such that quantum effects are still
significant. This roughly defines the mesoscopic regime. The system has a finite size Ld ,
where L is the linear size and d 6 3 the dimensionality. The motion of the electron is
schematically illustrated in Fig. 1. Because of the multiple scattering, the path of the
electron becomes increasingly complicated and its motion acquires stochastic features.
Thus, a proper Hamiltonian is given by
2

b = h 1 + V (r),
(1)
H
2m
where the potential V (r) is a stochastic disorder potential. It is taken as a white noise
potential with mean value zero,


(2)
V (r) V = 0
and a -correlated second moment,


1
r r0 .
V (r)V (r0 ) V =
2(F )

(3)

Fig. 1. Propagation of an electron through a finite sample. The electron is scattered at impurities
which are represented by dots. The interactions are symbolized by dashed lines. The linear size of
the system is L, as indicated in the figure.

364

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

The density of states at the Fermi energy per volume is denoted by (F ) and is the
average elastic mean free time between two scattering processes. The average h. . .iV is
taken over the ensemble of realizations of the disorder potential. Alternatively, one could
also average over the different possible paths taken by the electron. However, the physics
of the system favors the average over the potential: in mesoscopic physics, measured
quantities such as the conductance of the sample depend considerably on the particular
realization of the disorder potential. Thus, the experimentalists aim at measuring average
results. This can be achieved by averaging over different samples or, mostly, by averaging
over different realizations of the disorder which are obtained by varying some external
parameters such as a magnetic field or so. The average over the potential V (r) closely
matches this experimental situation.
The multiple scattering renders the motion of the electron diffusive. The classical
diffusion time td sets a crucial scale for the statistical properties. Thouless showed [1618]
that the associated energy scale
Ec =

h
td

(4)

distinguishes two different regimes for the spectral fluctuations. The probability density
P (x, y, t) of finding a particle which traveled from the point x to the point y within the
time t is, in the case of diffusive motion, described by the equation
D

d
X

2
P (x, y, t) = P (x, y, t),
2
t
xi
i=1

(5)

where D is the diffusion constant. The Thouless energy is related to D by

Ec =

h D
,
L2

(6)

where L is the linear size of the system. For energy separations smaller than the Thouless
energy Ec , the fluctuations are of the WignerDyson type, i.e., identical to those of chaotic
systems and therefore well described by Random Matrix Theory (RMT), for a review see
Ref. [1]. For energy separations larger than Ec , the energy levels lose their correlations and
their fluctuations are Poissonian. In other words, the electron has completely explored the
system for times larger than td . System specific properties are not resolved anymore. Above
the Thouless energy, or, equivalently, for times smaller than td , non-universal spectral
statistics arise, which are not solely determined by RMT [19]. Moreover, the Thouless
energy is, in units of the single particle mean level spacing, the dimensionless conductance
of the system and therefore a directly measurable quantity.
Spectral correlations can most conveniently be studied by Efetovs supersymmetric
technique [11,12,20] which allows one to perform the average over the ensemble of the
disorder exactly [11,20]. Efetov identified the Thouless energy in systems described by
the Hamiltonian (1) in the resulting non-linear model. Roughly speaking, it consists
of a kinetic and a potential term. The latter arises due to a certain symmetry breaking.
In the formal limit of zero dimensions, the kinetic term vanishes and RMT is obtained.

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

365

Fig. 2. Propagation of a quark through the QCD vacuum. The smooth and twisted lines represent
quarks and gluons, respectively. The propagating quark is scattered by the vacuum fluctuations.
Quark loops are not involved in the quenched approximation.

The diffusion constant and thus the Thouless energy can be viewed as a measure for the
competition between the potential term and the kinetic term.
In the sequel, we will merge Efetovs approach with the principles of chiral symmetry
and QCD. Our motivation for doing so is, first, the analogy between disorder and QCD
which was suggested and discussed in a semi-classical framework in Refs. [46], and,
second the identification of the Thouless energy in lattice QCD data [7,8]. Anticipating
the results of our calculations, we present a schematic illustration of the propagation of
a quark through the vacuum in Fig. 2. Since chiral symmetry is broken, the vacuum is
a complicated superposition of states. The quark is scattered at the vacuum fluctuations
which are generated by closed gluon and quark loops. We consider a finite volume V4 , the
linear size of the system is (V4 )1/4 . Just as in mesoscopic disordered systems, we are not
interested in details of the quark paths but rather in the average statistical properties of the
particle. Hence we are not aiming at solving the dynamics of the YangMills theory, we
rather study the stochastic features caused by the complicated gauge field interactions. We
interpret the vacuum fluctuations as impurities that generate a gauge field disorder. Thus,
we replace the gauge fields with a white noise potential. The motion of the Dirac particle is,
in this sense, also diffusive. A diffusion coefficient and a Thouless energy result naturally.
Here, in QCD, the pion decay constant will play a role similar to the conductance.

3. A Dirac particle in gauge field disorder

The concepts introduced in the previous section are now made quantitative. The model
is constructed in Section 3.1, supersymmetry is introduced in Section 3.2. The symmetries
of the model are discussed in Section 3.3.

366

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

3.1. Setup of the model

We replace the gauge fields coupled to the free Dirac operator by a stochastic potential,
we have


0
c[u]
,
(7)
/ =
iD[u]
0
c [u]
where

c[u] = i i + gu (x) ,

(8)

and i , i = 1, 2, 3, are the Pauli matrices, 4 = 12 and g is the gauge coupling. The
stochastic fields u (x) are complex. It is important to note that they are not considered
to be gauge fields, see the discussion below. With this definition, the Dirac operator is
Hermitian, and thus has real eigenvalues. As argued above, we take a white noise potential,
with


(9)
u (x) u = 0
and


u (x)u (y) u = (4)(x y),

(10)

where the constant measures the correlation strength. For the time being, we do not fix it,
we rather determine it later. The average over different realizations of the disorder potential
replaces the average of all gauge field configurations. The first and second moment,
(9) and (10), respectively, are consistent with a Gaussian probability distribution of the
potential. We write the average of a functional R[u] of the disorder potential in the form


Z
Z


1
4

(11)
d x u (x)u (x) ,
R[u] u = d[u]R[u] exp

with the measure

d[u] =

4
4
Y
Y
du du
d(Re u ) d(Im u )
=
.

=1

(12)

=1

This choice of the measure ensures the correct normalization, i.e., h1iu = 1.
Our model is presented pictorially in Fig. 3. The physical vacuum |vaci contains
quantum fluctuations. A perturbation series of the fluctuations in terms of quark and gluon
loops is shown in Fig. 3(a). The vacuum state cannot be calculated analytically, because this
would require a resummation of the perturbation series to infinite order. The propagating
quark interacts with the vacuum via gluon exchange. Our model replaces this process by the
coupling of the quark to an effective, stochastic, potential. This is symbolized in Fig. 3(b).
Some comments on the structure of the Dirac operator (7) in color space are in order.
As already emphasized, the fields u
/ (x) are not gauge fields. In our approach, they are
static fields, as obvious from Eq. (11). We are free to make this assumption, because we
wish to study the statistical properties of the system. We do not try to solve its dynamics.
Nevertheless, the original SU(Nc ) gauge group should be reflected in the operator (7) in

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

367

Fig. 3. Symbolic representation of the stochastic disorder model. In (a), a perturbative expansion of
the physical vacuum |vaci in terms of quark and gluon loops is depicted. In (b), a quark is coupled by
gluon exchange to the vacuum state. This describes the propagation of a quark through the vacuum
as it has been shown in Fig. 2. Calculating these graphs would require the full dynamics of the theory
and the summation of the perturbation series to infinite order. In the statistical model, the dynamical
theory is replaced by a disorder potential described by Eqs. (7)(12).

some way. Since we replace the gauge fields with the potential (11), the gluonic degrees
of freedom can be viewed as are integrated out, resulting in an effective potential. Our
model has some resemblance with other low energy effective theories of QCD. Firstly, the
NambuJona-Lasinio (NJL) model [21,22] describes an effective four-fermion interaction
of quarks having the same global symmetries as QCD. Gluons are not present in this model.
In Section 4.2, we will elaborate further on the relation between our approach and the
NJL model. We notice that neither the NJL model nor many other effective models of
QCD are gauge invariant. Since we aim at establishing a close connection to, in particular,
the NJL model, we use a similar route. Alternatively, a non-linear model was derived
which preserves gauge invariance [13]. However, this model starts from a lattice action
and therefore not fully preserves Lorentz invariance. Secondly, in the instanton model, the
quarks are coupled to instantons, which are the classical solutions of the YangMills field
equations. The reader is referred to Ref. [23] for a recent review. The corresponding Dirac
operator shares certain features with the one that we use. Moreover, the instanton action
induces an effective four-fermion interaction [24]. In all these models, the Dirac operator
is proportional to the unit matrix in color space. Thus, here we use the same assumption
for our Dirac operator (7).
For each realization of the disorder potential u
/ (x), a set of eigenvalues {i [u]} is
obtained,

/
iD[u]
i = i [u]i .

(13)

We are interested in calculating the spectral correlations by averaging over the disorder
potential, i.e., the spectral density
X





1
1
i [u]
=
(14)
Im tr
R1 () =

iD[u]
/
u
u
i

368

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

or, more generally, the spectral k-point functions

+
* k
+
* k
YX


1 Y
1
j i [u]
= k
Im tr
.
Rk (1 , . . . , k ) =

j iD[u]
/
j =1 i

j =1

(15)

The eigenvalues have an imaginary increment, = i, to ensure convergence.

The positive sign refers to retarded and the negative to the advanced Green function,
respectively. For simplicity, the superscript will be suppressed in the following. Instead
of the functions (15), it is technically easier to calculate the correlation functions
* k

Y
1
1
b
tr
.
(16)
Rk (1 , . . . , k ) = k

j iD[u]
/
u
j =1

For later use, we introduce the Green function

/ S(x, y; u, ) = (4) (x y),
iD[u]

(17)

to our Dirac operator.

3.2. Disorder average with supersymmetry
To perform the disorder average, we generalize Efetovs supersymmetric technique for
Schrdingers quantum mechanics to Diracs quantum mechanics. For the convenience
of the reader not familiar with supersymmetry in disordered systems, we do this rather
explicitly.
The correlation functions (16) can be written as derivatives with respect to external
sources J1 , . . . , Jk of a generating functional Zk ( + J),
+ 
* Qk
k
det(j iD[u]
/ + Jj ) 
1

j
=1
bk () =
R

Q
Qk
(2)k kj =1 Jj
/ Jj ) u J=0
j =1 det(j iD[u]


1
k
 ,
=
Z
(
+
J)
(18)
Q
k

(2)k kj =1 Jj
J=0
where we have introduced = diag(1 , . . . , k ) 12 and J = diag(J1 , . . . , Jk ) k with
k = diag(1, +1). The ratio of the determinants in (18) is expressed as a functional integral
containing an equal number of bosonic and fermionic fields. We introduce the 4 2kcomponent, spacetime dependent, superfields

T
(19)
(x) = z1 (x), . . . , zk (x), 1 (x), . . . , k (x) ,
where zj (x) and j (x) are bosonic and fermionic fields, respectively. They are further
decomposed in left and right handed parts,
!
!
zjR (x)
jR (x)
,
j (x) =
.
(20)
zj (x) =
zjL (x)
jL (x)
The fields zjL,R (x) and jL,R (x) are commuting and anti-commuting two component Pauli
spinors, respectively. Thus, zj (x) and j (x) are four component Dirac spinors. We suppress

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

369

the color indices, since the Dirac operator is diagonal in color space. We will take this into
account by introducing a factor Nc when taking traces. With the definition

/
D[u; + J] = ( + J) 14 12k iD[u],

(21)

where
( + J) = diag(1 J1 , . . . , k Jk , 1 + J1 , . . . , k + Jk ),
one can write

Z

Zk ( + J) =

(22)

 Z

4

d[] exp i d x (x)D[u; + J](x)

(23)
u

for the generating functional.

The disorder average can now easily be performed. It is a Gaussian integration and
yields an effective four-point interaction of the fields (x). It is decoupled via a Hubbard
Stratonovitch transformation by introducing supermatrix fields (x). One arrives at [25]
Z
Z
 
2
d[] d[ ] d
Zk ( + J) = 22k


 Z




1

trg

(x)
(x)

i
(x)D
,

+
J
(x)
. (24)
exp d 4 x
4 g 2
The s are 2k 2k supermatrix fields. The above expression is bilinear in the superfields
(x) allowing us to perform their integration. The Dirac operator D[, ; +J] is defined
as a direct product of super and Dirac matrices. It reads


).
(25)
D , ; + J = (x) PR + (x) PL + ( + J) 14 12k (i/
The left and right handed projectors PL,R are given by




12 0
0 0
.
,
PL =
PR =
0 12
0 0

(26)

An important observation from (25) is that (x) is coupled only to the right handed
modes of the Dirac spinors and (x) only to the left handed modes. This implies chiral
symmetry, since (x) and (x) are independent variables. To do the integration over
the supervector (x), one has to interchange the integration d[ ] and d[]. This can
cause problems concerning the convergence properties of the integrals [25]. In the present
context, however, we do not expect this to affect our calculations.
After integrating out the supervector (x), we finally obtain
Z



 
2k 2
(27)
d[ ] d exp F , ; + J
Zk ( + J) = 2
with


F , ; + J = trg





1

d x
(x) (x) + tr log D , ; , J .
4 g 2
4

(28)

The trace tr = trDirac trcolor in front of the logarithm is taken with respect to both Dirac and
color indices, while trg is the trace in superspace.

370

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

3.3. Symmetries of the supersymmetric functional

We now study the global symmetries of our model. Since our model is, by construction,
not gauge invariant, we do not consider gauge symmetries. The fermionic part of the QCD
Lagrangian Lfermion possesses chiral symmetry. We write
Ns

Lfermion =

f
X

b (x),

f (x) iD[A]
/ iMf f (x) = (x)
iD[A]
/ iM

(29)

f =1

where the fermion fields have Nfs components in flavor space

1 (x)
u(x)

..
= d(x)
(x) =
.
.

..
s
Nf (x)
.

(30)

b=
We refer to them as sea quarks. Their current masses are described by the matrix M
diag(M1 , . . . , MNfs ) = diag(Mu , Md , . . .). We identify them with the quark flavors up,
b = 0, the Lagrangian Lfermion is invariant under SUV (N s )
down, strange and so on. For M
f

SUA (Nfs ) symmetry transformations. The Lagrangian Lfermion is not invariant under the
b 6= 0, since 5 does not
axial transformation if the current sea quark masses are finite, M
b By introducing left and right handed spinors
anti-commute with the diagonal matrix M.
L,R (x) = PL,R (x) according to Eq. (26), one can rewrite the SUV (Nfs ) SUA (Nfs )
symmetry in the form




(31)
SUL Nfs SUR Nfs .
This is the chiral symmetry of the QCD Lagrangian.
/
differs by construction
In our study of the spectral correlations, the operator D[u]
/ [A] in the fermionic part of the QCD Lagrangian (29). The operator
from the operator D
appearing in our spectral correlation functions (16) enters our model Lagrangian in the
form

/ (x).
(32)
L = (x)
iD[u]
By comparing this with Eq. (29), we see that Eq. (32) describes Nfv = k quarks with
imaginary masses. We refer to them as valence quarks. The fermion fields have k
components in the space of the valence quarks. Finite values of the i s are equivalent to
non-zero, imaginary, current masses. Using the same line of arguing as above, we conclude
that the Lagrangian (32) has the symmetry




(33)
SUL (k) SUR (k) = SUL Nfv SUR Nfv ,
if = 0. The difference between valence and sea quarks lies in the different ways in which
they enter the spectral correlators (16). The valence quarks are given as an additional
degree of freedom of the Dirac operator in the resolvent. Sea quarks alter the weight of the
averaging process h. . .iu since they enter as fermionic determinants, i.e. they determine the
symmetries of the Dirac operator in the partition function. In the quenched approximation

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

371

the fermionic determinants are omitted, i.e., Nfs = 0. We are mainly interested in the
/ 1 .
symmetries of the valence quarks, i.e., the symmetries of the Green function ( iD)
Since all manipulations made in Section 3.2 are exact, the symmetries are respected
throughout the calculation and must also be present in the final functional. The superfields
(x) are readily decomposed in terms of left and right handed modes as we can see
from Eq. (20). Thus, the Lagrangian in Eq. (23) possesses symmetries similar to the ones
discussed above for the fermionic field (x). The difference is that the symmetries are
defined in superspace. Taking into account the equal number of bosonic and fermionic
variables, we see that for + J = 0 the Lagrangian
L = (x)D[u; + J](x),

(34)

where D[u; + J] is defined in Eq. (21), is invariant under the transformation

L (x) uL L (x),
R (x) uR R (x),

uL uL = uL uL = 12k ,
uR uR = uR uR = 12k .

(35)

The unitary supermatrices uL,R are elements of the group SU(k/k). Thus, we find
SUL (k/k) SUR (k/k)

(36)

as the corresponding supersymmetric version of the chiral symmetry (33). If we take into
account the fermionic determinants of the weight, we have to increase the number of
fermionic variables of the supervector by Nfs additional entries. Thus, in the unquenched
case we have the symmetry




(37)
SUL k/k + Nfs SUR k/k + Nfs .
However, here we restrict ourselves to the quenched limit, Nfs = 0.
The final generating functional preserves the chiral invariant combination (36) if the
explicit symmetry breaking is turned off, i.e., + J = 0 in Eq. (25). Indeed, one has
invariance under the transformation
(x) uL (x)uR ,
(x) uR (x)uL .

(38)

This can be seen by writing the functional in (28) in the form



Z





1

, (39)

(x)
(x)
+
2N
log
D
;
,
J
,

F , ; + J = trg d 4 x
c
KG
4 g 2
which is obtained from the original form after performing the trace trDirac trcolor . The Klein
Gordon operator is given by
 





(40)
DKG , ; + J = (x) + ( + J) (x) + ( + J) + 12k .
No direct product appears. If + J = 0, the functional contains only the combination
(x) (x), which transforms like (x) (x) uR (x) (x)uR . This proves the
invariance, since trg log DKG = log detgDKG and detg(uR uR ) = 1.

372

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

We see that non-zero values of the valence quark masses, ( + J) 6= 0, explicitly

break the chiral symmetry (36). But there is a remnant symmetry. Consider the case of k
degenerate imaginary valence quark masses, 1 = = k = , i.e.,
+ J = 12k ,

(41)

where the source terms are set to zero. Then the functional (39) is invariant under the
transformation
(x) u (x)u ,
(x) u (x)u ,

(42)

where u is a unitary supermatrix, uu = u u = 12k . Thus, we have the symmetry group

SU(k/k).

(43)

The invariance under the transformation (42) follows if we insert Eqs. (42) and (41) into
the KleinGordon operator (40). It transforms like DKG uDKG u . This symmetry which
remains as we move away from zero virtuality with the choice (41) we refer to as the bulk
symmetry. It is broken if the degeneracy between the valence quark masses is removed,
i.e., i 6= j for any two i 6= j .
At zero virtuality, where all of the eigenvalues of the k-point function are zero, we have
exact chiral symmetry, see Eq. (36). The symmetry is broken by a non-zero value of any of
the eigenvalues. This corresponds to explicit breaking of chiral symmetry by current quark
masses. The larger the eigenvalue, the stronger the symmetry breaking. If one increases
the absolute values of the i s, the Lagrangian still exhibits symmetry even for large
eigenvalues as long as the k eigenvalues of the k-point function are degenerate, i = j for
all i, j . One may identify this bulk symmetry with the isospin symmetry between up and
down quarks for the case of the two-point function or with the symmetries of the hadronic
particle multiplets in the EightfoldWay in the case of the three-point function. The bulk
symmetry is broken if we have non-degenerate eigenvalues. This is reminiscent of isospin
breaking by the difference in the up and down quark masses or the mass splitting of the
particles in the EightfoldWay. We conclude that the one-point function in the bulk does
not have any symmetries. For the two-point function in the bulk, the symmetry breaking
is controlled by one parameter, the distance between the two levels, such that a non-zero
value leads to symmetry breaking, corresponding to the mass difference of the current up
and down masses, Mu Md . In contrast to the bulk region, the one-point function in the
microscopic regime exhibits symmetry, namely the chiral symmetry (36) for k = 1. The
symmetry breaking is induced by the eigenvalue itself for the case that it is non-zero, just
as explicit symmetry breaking by non-vanishing current quark masses.
Consequently, there is a close relation between the one-point function in the microscopic
regime and the two-point function in the bulk. In both cases the symmetry breaking is
controlled by a single parameter, although the symmetry groups are different. We will
return to this issue in Section 5.

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

373

4. Saddle point approximation

To make analytical progress, we have to resort to a saddle point approximation. After
expanding the action in Section 4.1, we calculate the saddle point equation and establish
the link to the NJL model in Section 4.2. The solutions and symmetries of the saddle point
equation are discussed in Section 4.3. In Section 4.4, we discuss the validity of the saddle
point approximation. Physically, it is a weak disorder expansion which will be shown to
coincide with the t Hooft limit, i.e. the limit Nc with Nc g 2 = const.
4.1. Expansion of the action
The partition function (27) defines a quantum field theory of the superfields (x) with
an action F [, ; + J] given in Eq. (28). The classical equations of motion for the
fields are found from the principle of least action, which requires that the first variation
of F [, ; + J] vanishes. We express the fields (x) and (x) by the classical fields
0 (x) and 0 (x) and fluctuations around them, (x) and (x), as
(x) = 0 (x) + (x) and (x) = 0 (x) + (x).

(44)

The functional in (28) is then expanded in powers of (x) and (x). For convenience,
we define
(x) = (x) PR + (x) PL .

(45)

With the definition 4 = 12k 4 , we have (x) = 4 (x)4 , with the Dirac matrix 4 .
Expanding the functional yields


F , ; + J = F [; + J]
= F [0 ; + J] + F [0 , ; + J]
1
+ 2 F [0 , ; + J] + .
2

(46)

The classical fields are found from the condition of stationarity

F [0 , ; + J] = 0.

(47)

The variation is performed by varying the matrix elements of the supermatrices. We

introduce the Green function S(x, y; ; + J) associated with the Dirac operator
D[; + J] = D[, ; + J], of Eq. (25). It satisfies
D[; + J]S(x, y; ; + J) = (4) (x y).

(48)

We define the trace tf

rg by f
trg(A B 1Nc ) = Nc (trgA)(trDirac B) where A is a super, B a
Dirac matrix and 1Nc is the color unit matrix, respectively. To zeroth order, we have


Z


1

rg d 4 x

(x)
(x)
+
log
D[
;

+
J]
.
(49)
F [0 ; + J] = tf
0
0
0
8 Nc g 2

374

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

The first variation is given by

F [0 , ; + J]

Z
= tf
rg d 4 x (x)


1

(x)
+
S(x,
x;

+
J)
,
0
8 Nc g 2 0

(50)

while the second variation is found to be

2 F [0 , ; + J]

Z
Z
= tf
rg d 4 x d 4 y

1
(x) (y) (4)(x y)
8 Nc g 2


S(x, y; 0 ; + J)(y)S(y, x; 0 ; + J)(x) . (51)
The expansions (49)(51) are the starting points for all of the following considerations.
4.2. Saddle point equation and relation to NambuJona-Lasinio model
According to the condition (47) the saddle point is found from Eq. (50) by


1
0 (x) + Nc trDirac S(x, x; 0; + J) 12k PR = 0,
2
4 g


1
0 (x) + Nc trDirac S(x, x; 0 ; + J) 12k PL = 0,
2
4 g

(52)

where we have used that (x) is a free variational differential. Equivalently, we have



1

(x)
+
S(x,
x;

+
J)

P
0 = trDirac
0
2k
R,L
8 Nc g 2 0


1
(x) + S(x, x; 0 ; + J) .
(53)
trL,R
8 Nc g 2 0
The traces trR and trL are taken over Dirac matrices and project out the right and left
handed parts, respectively. We note that we still have two equations in (53). In momentum
representation, the saddle point equation reads
cut


Z
d 4p
S(p;

+
J)
= 0.
trR,L 0 + 8 Nc g 2
0
(2)4

(54)

The integral over the momentum occurring in this expression has to be regularized, since it
is not convergent. The four-momentum covariant regularization scheme appears to be most
convenient since we work with a Euclidean metric, i.e., a cutoff cut in momentum space
has to be introduced. This corresponds to the ultraviolet cutoff in lattice gauge theory due
to the finite lattice spacing. We note that 0 does not depend on the momenta, because we
started with a potential which is -correlated in space time. The momentum representation
of the Green function in Eqs. (50) and (51), defined from
Z
(55)
S(p; 0 ; + J) = d 4 x eipx S(x, y; 0; + J)

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

375

is given by
S(p; 0 ; + J)




1
/ (0 + 4) 0 + 4 12k 14 p2 ,
= 0 + 4 + 12k p

(56)

with 4 = ( + J) 14 . To perform the trace trR,L S(p; 0 ; , J), we note that (0 + 4)

is block diagonal in Dirac space. This holds as well for the inverse of the operator in the
p [. . .]1 ) = 0, because there are only off-diagonal
square brackets in Eq. (56). Thus, trR,L (/
blocks. The remaining expression has only diagonal blocks and the trace trR,L can be
dropped since it projects out the diagonal blocks only. It follows that


0 = 8 Nc g 2 0 + ( + J)
cut
Z

1
d 4p 
(0 + 4) 0 + 4 12k 14 p2 ,
4
(2)

(57)

which are still two coupled equations. For convenience, we rewrite Eq. (57) in a slightly
different way. From the definition = i(0 + 4) and = 4 4 = i(0 + 4),
Eq. (57) acquires the form

= i4 + 8Nc g
2

cut
Z

1
d 4p
.
(2)4 + 12k 14 p2

(58)

For scalar fields, this equation is well known from the NJL model. It is the so called gap
equation for the self-energy.
The NJL model is an effective low energy theory of QCD. See Ref. [26] for a review.
It contains a four-point interaction between the fermions, and the interaction term is
constructed to be chirally invariant. In its simplest version, Nf = 2, the NJL Lagrangian is
LNJL = L0 + Lmass + Lint

2




2 
b (x) + G (x)(x)

5 (x) ,
= (x)
i/
iM
+ (x)

(59)

b = diag(Mu , Md ) contains the current

where = (1 , 2 , 3 ) are the Pauli matrices and M
b
up and down quark masses. For M = 0, Eq. (59) defines a SUL (2) SUR (2) invariant
theory where the four-point interaction has the effective coupling strength G. The effective
interaction is assumed to arise from the complex non-Abelian interaction of the gluons.
Thus, there are no explicit gluons in this model. From the NJL Lagrangian (59), we can
see the close relationship between the supersymmetric description of spectral correlations
and the NJL model. In the former case, one obtains, after the disorder average, a functional
with four-point interaction of superfields (x) which is chirally invariant. In the latter
case, fermion fields (x) interacting via an effective four-point interaction. In both cases,
the free part of the Lagrangian is described by the free Dirac operator i/
. Explicit chiral
b and ( + J).
symmetry breaking is generated by the mass matrices i M
In Section 3.2, the four-point interaction of the superfields (x) was decoupled via a
HubbardStratonovitch transformation, which lead to the introduction of the superfields
(x). In the NJL model, a similar concept is known, which is called bosonization.
The Lagrangian (59), which contains quark degrees of freedom only, is transformed via

376

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

a HubbardStratonovitch transformation to a Lagrangian which contains only mesonic

degrees of freedom. For Nf = 2 it reads
Lmeson =

1
e
2 (x) + a (x) a (x) i tr i/
e
(x) + i a (x)t a .
4G

(60)

the isosinglet, not to

The pion fields a build the isotriplet of SU(2) and the sigma field e
be confused with the superfields. Again, we see the close similarity between both models,
when comparing the bosonized NJL model with our model.
One of the main results of the NJL model is the dynamical breaking of chiral symmetry.
One finds that the chirally invariant, perturbative, vacuum is not the minimum in energy.
The physical vacuum formed by Cooper-like quarkantiquark pairs. It can be proven by
the BogoliubovValatin variational approach that the latter indeed has lowest energy and
is thus the real ground state of the model, see, for example, Ref. [26]. An alternative way
of understanding chiral symmetry breaking is to calculate the self-energy of the quarks. In
the case of the NJL model, the scalar part of the self-energy and the dynamically created
mass are identical. Thus, if the particle obtains a physical mass, chiral symmetry is broken
because chirality is not preserved for massive particles. In the Hartree approximation,
one can derive a self-consistent equation for the dynamically generated masses of the
quarks, which we will refer to as constituent or effective quark mass M . This equation
is called gap equation in analogy to the BCS theory where a similar model describes the
transition between normal and superconducting state. If we set Mu = Md = = M0 the
gap equation is [26]
cut
Z

M = M0 2iNc G trDirac
= M0 + 8Nc GM

Zcut

d 4p
1
4
(2) p
/ iM

1
d 4p
,
(2)4 p2 + M 2

(61)

where the factor Nc arises from the trace over the color indices. A non-zero solution for
the effective mass M indicates chiral symmetry breaking, since this breaks the axial
flavor symmetry. The remaining symmetry is (SUL (Nf ) SUR (Nf ))/SUA (Nf ). If the
current masses are complex or purely imaginary, the solutions of Eq. (61) are analytically
continued into the complex plane. The comparison between both mean field equations (58)
and (61), shows once more the relationship between the NJL and our supersymmetric
model. In the supersymmetric model, the supermatrices are the analogue of the
effective mass M . Explicit symmetry breaking is introduced by a finite current mass M0
for the NJL and by a non-zero matrix ( + J) in the case of supersymmetry. This coincides
with the discussion of the symmetries of the supersymmetric functional, see Section 3.3.
For scalar fields, the saddle point equation (57) is
M

cut
Z

= 2Nc g trDirac
2

d 4p
S p; + M
4
(2)

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

cut
Z

= 2Nc g trDirac
2

= 8Nc g

+

377

1
d 4p

4
(2) M + p
/
Zcut

1
d 4p
,

4
(2) (M + )2 p2

(62)

where is a real number. For non-vanishing the solution M is complex in general. It is

related to the effective quark mass,
iM = lim M ,

(63)

where M is the solution of the gap equation (61). This can be seen by comparing Eqs. (61)
and (62). Employing the relation to the NJL model, we identify the effective coupling
strength through
G = g2 .

(64)

The real part of M can be absorbed by redefining the virtual mass, Re M .

This leads only to a relative shift in the bulk, but does not cause any problems near zero
virtuality, since the solutions are purely imaginary there, lim0 Re M = 0, and we are
left with real quark masses. Taking the imaginary part of Eq. (62), we obtain
Im M

cut
Z

= 2G Im tr

d 4p
S p; + i Im M ,
4
(2)

(65)

where tr = trDirac trcolor . The spectral density per volume is found as

() =

1
1
R1 () = Im tr S x, x; + i Im M ,
V4

(66)

where S(x, y; + Im M ) is the coordinate representation of the Green function S(p; +

Im M ) in Eq. (62). From this we find
() =

Im M
,
2G

(67)

in the bulk, and, using the BanksCasher relation [27], we arrive at



 M

=
2G

(68)

at zero virtuality. We note that this relation between the condensate and the dynamically
generated mass coincides with a similar relation derived from the NJL model [26]. Eq. (68)
determines the constant in Eq. (10):
=

Im M 0
M
.
2
2

g ()
g |hi|

(69)

This is similar to the considerations in mesoscopic physics, see Eq. (3), with the
identification = 1/2 Im M .

378

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

4.3. Solutions and symmetries of the saddle point equation

Before we turn to the matrix equation, we investigate the solutions of the scalar gap
equation (61). If the current quark masses are put to zero, this equation possesses the
obvious solution M = 0. In this case, chiral symmetry is not broken. To find the other,
non-trivial and energetically favored, solution, the integral on the right-hand-side is solved
explicitly. Ruling out the trivial case M = 0, the equation reads



2cut
Nc G
2
2

M
log
1
+

.
(70)
1=
cut
2 2
M 2
A solution M 2 6= 0 only exists if the coupling G exceeds a critical value Gc , one has the
condition
G > Gc =

2 2
.
Nc 2cut

(71)

For smaller values of G, the gap equation has only the symmetry preserving solution
M 2 = 0. If the coupling strength exceeds its critical value, G > Gc , one unique solution
exists. For the case of non-zero current quark mass, one always has a non-trivial solution
with M > M0 .
In Fig. 4, we show the spectral density () in the infinite volume limit obtained from
the numerical solution of the saddle point equation for arbitrary G. On both axis the scale
is set by cut . If the coupling strength is below the critical value, G < Gc , the spectral
density at zero virtuality vanishes, (0) = 0, and chiral symmetry is therefore not broken.
For G  Gc the density increases like the spectral density of the free Dirac operator,
() 3 , with a drop at = cut . On the other hand, if the coupling exceeds the critical

Fig. 4. Spectral density () in the infinite volume limit for different coupling strengths.

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

379

value, G > Gc , the density at zero virtuality is non-vanishing, indicating chiral symmetry
breaking.
For very large couplings, G  Gc , the spectral density at zero scales like (0)

1/ G, which can be seen from Eq. (77).

In the following, we distinguish the bulk and the microscopic region. In case of
the one-point function in the bulk, the generating functional does not exhibit further
symmetries. The saddle point is a single point. In the microscopic region, however, the
one-point function results from a manifold of saddle points. The points are transformed
into each other by rotation in superspace. The symmetry is not the full, chiral, symmetry
of the original theory. We will interpret this as leading to spontaneous breaking of chiral
symmetry. For the case of the two-point function in the bulk, we also find an invariance
under rotations of the supermatrices due to the degeneracy of two eigenvalues, although
the symmetry is different from the symmetry near zero virtuality. This is related to the
symmetry of the original theory which is not spontaneously broken at the saddle point.
4.3.1. One-point function in the bulk
For the one-point function, k = 1, we write 1 = and J1 = J and thus

( + J) = 12 + J k,
where
k =

1 0
0 1

(72)


,

(73)

breaks the symmetry between bosonboson and fermionfermion block. The 0 fields of
the saddle point equation (57) are now 2 2 supermatrices.
It is instructive to examine the zero dimensional case first. If we neglect the source term,
the saddle point equation reads


2Nc G0
= 0.
(74)
0 + 0 + 12 14
(0 + 12 14 )(0 + 12 14 )
The coupling constant G0 for the zero dimensional limit differs from the physical coupling
constant G. It is measured in energy units squared whereas G has the dimension of inverse
energy squared. This compensates the dimensions of the momentum integration which
disappears in zero dimensions. Writing 0 = 4 0 4 , one arrives at
4 0 4 0 + 4 0 4 + 2Nc G0 12 14 = 0.

(75)

A solution is

s
!

2
12 14 ,
0 = i 2Nc G0
2
4

(76)

where we notice that (4 )2 = 12 14 . The spectral density R1 () is according to Eq. (18)

given by


1
4 p
Im tf
rg 0 (k 14 ) =
8Nc G0 2 .
(77)
R1 () =
Nc G0
G0
This is the well-known Wigner semi circle law, the standard result of RMT.

380

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

For the four-dimensional case a solution of the saddle point equation (57) is
0 = i Im M 12 14 ,

(78)

where M is the complex solution of Eq. (62). The solution (78) is just a single point, and
there are no further symmetries.
4.3.2. Two-point function in the bulk
The two-point function


b2 (1 , 2 ) = S R [u]S A [u] = tr
R
u

1
1
tr
+

i
D[u]
/

iD[u]
/
1
2


(79)
u

has one pole above and one below the real axis. The supermatrices 0 and 0 are now 4 4
matrices. Introducing relative coordinates
1 = + ,

2 = ,

(80)

one has for J = 0

+ J = 14 + ( + i)L.
The 4 4 diagonal matrix

1
L=

(81)

(82)

1
breaks the symmetry between the retarded and advanced blocks of the supermatrices. The
functional F [, ; + J] in Eq. (28) is, as discussed in Section 3.3, invariant under
the transformation u u, with u being a unitary 4 4 supermatrix, u SU(2/2),
provided the eigenvalues are degenerate, i.e., = 0. In this sense, the rotational symmetry
of the saddle point manifold is broken by the matrix L.
At the saddle point the source terms and the symmetry breaking parameter are neglected,
i.e., J = 0 and = 0. A solution of the saddle point equation (57) is
0 = i Im M L 14 ,

(83)

where M is the solution of Eq. (62). One also has the solutions 0 L 14 , 14 14 ,
14 14 , but they are discarded. They change the analytical properties of the two-point
function (79), shifting the imaginary parts of the Green functions across the real axis.
This suppresses the contribution from this saddle points by one order in the asymptotic
expansion.
The saddle point solution is degenerate, since the saddle point equation is invariant
under the transformation 0 (u 14 )0 (u 14 ), with u SU(2/2). This symmetry
is intimately related to the symmetry of the supersymmetric functional (28). Every element
of SU(2/2) can be written as u = ru0 where r commutes with L, [r, L] = 0, whereas u0

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

381

does not commute with L, [u0 , L] 6= 0 [11,20]. Thus, we find the solution of the saddle
point equation for the two-point function in the bulk,
0 = i Im M (Q 14 ),

(84)

where Q = u0 Lu0 is in the coset manifold U (1, 1/2)/(U (1/1) U (1/1)) [11,20].
4.3.3. Microscopic one-point function
The saddle point solution for the two-point function in the bulk has rotational symmetry
which is broken by the matrix L if the distance between two eigenvalues does not vanish,
6= 0. In the case of the microscopic one-point function, the absolute value of and the
unit matrix 12 break the chiral symmetry, in analogy to and L for the two-point function
in the bulk.
Setting the symmetry breaking matrix (72) to zero, + J = 0, the saddle point equation
reads
Z
1
d 4p
,
(85)
0 = 2Nc G0
4

(2) 0 p2 12 14
0
which has the trivial solution
0 = 0 = 0.

(86)

It will be ruled out in the following. As argued above, there exist solutions 0 6= 0,
provided G > Gc . To understand the microscopic symmetries, we first turn to the zero
dimensional limit. In this case, one has
0 0 = 2Nc G0 12 14

(87)

or, equivalently,
0 0 = 0 0 = 2Nc G0 12 .

(88)

The saddle point is degenerate, in contrast to the one-point function in the bulk. Thus, we
have to integrate over the whole saddle point manifold. For the zero-dimensional case, this
was done in Ref. [28]. The result for the microscopic density for Nf = 0 is then the same
as obtained from the matrix model. Coming back to four dimensions, the above symmetry
argument still holds, since the momentum operator in the saddle point equation (85) is
proportional to the unit matrix, both in super and in Dirac space. The matrices 0 are
measured in units of energy. We make them dimensionless by setting
0 = iM U

and 0 = iM U ,

(89)

where M is the non-trivial solution of the gap equation (61). The supermatrices U are
unitary, i.e.,
U U = U U = 12 ,

(90)

and are in a proper symmetric space to ensure convergence.

According to Eqs. (36) and (38), the original functional is invariant under global chiral
transformations. This chiral symmetry is not preserved at the saddle point. Instead, one has

382

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

Fig. 5. Perturbative expansion of the disorder Green function S(x, y; u, ), defined in Eq. (17). The
solid line represents the free Green function S0 (x, y; ) and the dashed line interaction with the
disorder potential u
/(x).

the degeneracy of the saddle point according to the unitarity condition (90). Thus, chiral
symmetry is broken at the saddle point, in analogy to the breaking of chiral symmetry
for non vanishing solutions of the NJL gap equation (61). This is not only true for the
microscopic one-point function but also for the microscopic k-point function in general
provided that all eigenvalues vanish, 1 = = k = 0. We conclude that chiral symmetry
is spontaneously broken by means of
SUR (k/k) SUL (k/k) SU(k/k).

(91)

This is the supersymmetric version of chiral symmetry breaking in our model.

4.4. Weak disorder limit and the large Nc limit in QCD
Formally, the asymptotic expansion of the action is a weak disorder limit as taken in
mesoscopic systems. Here, we will clarify the nature of this limit by extending arguments
from diagrammatic perturbation theory, see, for example, Refs. [2931].
Fig. 5 shows the disorder Green function S(x, y; u, ), defined in Eq. (17), expanded in
a perturbation series in the disorder potential u
/ (x). The free Green function S0 (x, y; )
is associated with the free Dirac operator, ( i/
)S0 (x, y; ) = (4) (x y), and is
represented by a thin solid line. A dashed line symbolizes a scattering process with the
disorder potential. The particle propagates from x to y and it is scattered k times at
the intermediate points x1 , . . . , xk . The average over the disorder potential according to
Eqs. (9)(11) corresponds graphically to connecting the space time points xi and xj with
interaction lines. This is shown in Fig. 6. All combinations of xi and xj have to be taken
into account. This means that each point of the corresponding diagram has to be connected
with all the others. Graphs with odd numbers of interaction lines vanish upon performing
the Gaussian average. Each interaction line gives a factor , cf. Eq. (10). The proper selfenergy is obtained by constructing the sum of all irreducible diagrams and amputating
external legs. The resulting diagrams with the self-energy insertions are resummed and a
Dyson equation


S(x, y; u, ) u =

is obtained.

1
[S0 (x, y; )]1

(92)

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

383

Fig. 6. Perturbative expansion of the average Green function hS(x, y; u, )iu . As in Fig. 5, the solid
line symbolizes the free Green function. The interaction is represented by dashed lines. Due to the
disorder average, space time points are linked by interaction lines. The perturbation series is shown
up to fourth order in the potential.

In a mesoscopic probe, the degree of disorder is characterized by the mean free time ,
cf. Eq. (3), or, equivalently, by the mean free path l = vF , where vF is the Fermi velocity.
The condition
kF l  1

(93)

determines the limit of weak disorder, where kF is the Fermi wave number. In that
case, not all of the diagrams in Fig. 6 give the same contribution: The diagrams with
crossed interaction lines, Fig. 6(b), are of order O(1/kF l) compared to the ones without,
Figs. 6(a) and 6(c), respectively [11,29,31]. The self-energy can be calculated within this
approximation scheme. Translating the diagrams into the corresponding integral equation,
one obtains Eq. (62), and we identify = M . In the NJL model, the approximation
scheme leading to the gap equation is the large Nc limit. The large Nc limit to QCD
was developed and applied to mesons by t Hooft [32] and later extended to baryons by
Witten [33]. t Hooft realized that QCD is considerably simplified in the limit Nc .
Precisely, the approximation is
Nc ,

Nc g 2 = const,

(94)

where g is the dimensionless coupling constant. In this t Hooft limit only a certain class
of diagrams in the perturbative expansion give a contribution. First, the lowest order,
one-loop, diagrams, are, by construction, finite. Second, quark loops are suppressed by
a factor of 1/Nc with respect to gluon loops. Third, non-planar diagrams are suppressed
by a factor of 1/Nc2 with respect to planar diagrams. Planar diagrams are diagrams that
have no crossing of interaction lines. In our model, we have a further, very important,

384

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

simplification. The gluons in the distribution (11) do not have kinetic terms, i.e., they
serve as a static potential without dynamics. Hence, there are no internal gluon loops
involved. The remaining diagrams in this limit are exactly the same that contribute
in mesoscopic systems in the in the limit of weak disorder. Hence, we conclude the
equivalence of the large Nc limit in our model and the weak disorder limit.

5. Effective theories
We now use the results of the previous sections to derive effective theories. In Section 5.1
we work out the fluctuations around the saddle points. In Sections 5.2 and 5.3, we arrive
at two different effective non-linear models in the bulk and in the microscopic region,
respectively.
5.1. Quadratic fluctuations
The zeroth order term of the expansion (46) is zero at the saddle point F [0 ; + J] = 0,
if the symmetry breaking parameters are switched off, i.e. if we set = 0 in the bulk and
= 0 in the microscopic regime, respectively. This is not true if either of the symmetry
breaking parameters is finite. In that case, the degeneracy of the saddle point is removed
and the saddle point manifold shrinks to a point, yielding on non-zero value of the
functional. We will thus expand it up to linear order in the symmetry breaking parameter,
in the bulk as well as in the microscopic region, which gives the symmetry breaking terms.
They describe the RMT properties in the corresponding region of the spectrum. For details
of the calculation see Appendix A.
In the bulk, we have to linear order in
Z


(95)
F [0 ; , ] = i() trg d 4 x LQ(x) ,
where Q(x) has the symmetries as discussed near Eq. (84).
In the microscopic region we have
Z


i
F [0 , ] = (0) trg d 4 x U (x) + U (x)
2

(96)

where U (x) obeys Eq. (90).

If we neglect the dependence on x and restrict the spacetime integration to a finite
domain, both functionals would reproduce the random-matrix results. This is so because
the quadratic fluctuations around the saddle point cannot give a contribution in this limit.
The equal number of bosonic and fermionic variables yields unity when integrated over.
This is not the case for the spacetime dependent fields in Eq. (51) and we have a
contribution from the quadratic fluctuations. We will make a further approximation by
restricting ourselves to fields which are slowly varying in spacetime. This allows us to
make a gradient expansion, tantamount to an expansion in powers of the momenta.

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

385

At this point, we have to distinguish between Goldstone and massive modes. Goldstone
modes, GM (x), lie in the saddle point manifolds (84) and (89), while the massive modes
mass (x) do not. In the quadratic approximation these modes decouple [11] and one has
2 F [0 , ; + J] = 2 F [0 , GM ; + J] + 2 F [0 , mass ; + J],

(97)

2F

has the form given in Eq. (51). In Appendix B, we discuss both terms on the
where
right-hand-side of (97) employing a gradient expansion. From this, we see that the massive
modes do not contribute to the functional. In contrast, the Goldstone modes will yield
a term quadratic in both the fields and derivatives which generate the dynamics of the
superfields.
Setting the symmetry breaking parameters to zero, we have in the bulk region
Z

1 2
F [0 , ; ] = ()D() trg d 4 x Q(x) Q(x),
(98)
2
2
whereas in the microscopic region, we obtain
Z

1 2
F [0 , ] = (0)D(0) trg d 4 x U (x) U (x).
(99)
2
2
In both cases the matrix fields have the same symmetries as in Eqs. (95) and (96),
respectively.
5.2. Bulk model: Efetovs model in four dimensions
Adding the symmetry breaking term and the kinetic term, we construct the generating
functional for the two-point function in the bulk. Reinserting the source terms, we obtain,
after dropping an irrelevant normalization constant,
Z


(100)
Z2 (, , J) = d(Q) exp F [Q; , , J]
with
F [Q; , , J] =

() trg
2

Z
d 4 x D() Q(x) Q(x)




2i ( + i)L + J Q(x) .

(101)

The fields Q(x) are the Goldstone modes of the saddle point manifold. They are in
the coset manifold U (1, 1/2)/(U (1/1) U (1/1)). The source terms are included in the
matrix J = diag(J1 , J2 , J1 , J2 ). The two-point correlation function is obtained from
the generating functional according to Eq. (18).
The functional (101) exactly reproduces Efetovs result [11,12], now formulated in four
spacetime dimensions. The fields Q(x) exhibit the same symmetry properties as those
in Refs. [11,12]. Chiral symmetry is not present. This is in agreement with its explicit
breaking by the virtual current mass i. From the interplay between kinetic and symmetry
breaking terms we identify the equivalent of the Thouless energy in the bulk of the QCD
Dirac spectra.
In a finite volume V4 , the longest wavelength which fits inside the Euclidean box is
given by the inverse of the linear extension of the box, i.e., 1/(V4 )1/4 . Thus, the kinetic

386

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

energy cannot be smaller than D()/ V4 . For smaller energies, the wavelength exceeds
the extension of the box and does not resolve details inside it. In that case, the fields
become constant and the functional (101) is dominated by the constant modes. Hence,
only the symmetry breaking term gives a contribution. Then the level statistics follow the
predictions of RMT [11,20]. If the energy exceeds this scale, the kinetic term becomes
more and more important, giving rise to corrections to the RMT predictions. We conclude
that the range of validity in the bulk is given by
D()
.
bulk
c
V4

(102)

This is the equivalent of the Thouless energy in terms of the original energy scale. The
mean level spacing 1 = 1/R1 () = 1/(V4 ()) scales with the inverse of the Euclidean
four-volume. Thus, on the scale of the mean level spacing, we have
p
bulk
c
()D() V4 ,
1

(103)

where
(Im M )2
tr
()D() =
2

d 4 y (x y)2 S x, y; + i Im M


S y, x; i Im M

(104)

is the second moment of the probability density of the particle [34].

5.3. Microscopic model: chiral perturbation theory
In the microscopic region, the combination of the symmetry breaking and kinetic terms,
yields an effective theory in the bulk as well. The expressions are generally valid for
any chiral disorder model such as the one introduced in Section 3.1, provided that the
coupling is strong enough to cause dynamical breaking of chiral symmetry. The relevant
parameters of the non-linear sigma model are the spectral density (0) and the diffusion
coefficient D(0). When we turn to QCD, both can be related to low energy quantities of
the theory: The BanksCasher relation [27] states that the spectral density at zero virtuality
is equal to the absolute value of the chiral condensate |h i|.
Moreover, there is a further result for D(0) which relates to the pion decay constant f .
In momentum representation, the diffusion coefficient reads
Z
d 4p
1
8Nc (M )2
,
(105)
D(0) =
4
2
)2 )2

(2)
(p
+
(M
|hi|
which follows directly from the definition (104) by using the momentum representation for
the Green functions. The integral on the right-hand-side turns out to be proportional to the
square of the pion decay constant. In fact, one can derive the relation
Z

2
d 4p
1
(106)
f2 = 4Nc M
4
2
(2) (p + (M )2 )2

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

387

from the model form of the quark propagator and a low energy theorem of QCD. For a
detailed discussion see Ref. [26]. Comparing (105) and (106), we find
D(0) =

2f2
.

|hi|

(107)

In Refs. [46], this formula was obtained from the GOR relation by using formal
analogies and semi-classical considerations. Here, we gave a derivation of the diffusion
constant (105) from our stochastic model.
Finally, after dropping an irrelevant normalization constant and reinserting the source
term, we obtain for the generating functional of the one-point function in the microscopic
regime
Z


(108)
Z1 (, J ) = d(U ) exp F [U ; + J ] ,
with



 trg
F [U ; + J ] = 

Z
d 4x

f2
U (x) U (x)
|h i|





i
+ J k U (x) + U (x) .
2

(109)

The supermatrices U (x) are unitary U (x)U (x) = U (x)U (x) = 12 and in a proper
symmetric space that ensures convergence of the integrals. The functional is dominated

V4 ). The resulting spectral

by the constant modes for energies smaller than f2 /(|hi|
density in this limit, which yields chRMT, was calculated in Ref. [28]. We conclude that
the equivalent of a Thouless energy in the microscopic region is given by

micro
c

f2
.

|hi|
V4

(110)

In terms of the mean level spacing 1 = /(|hi|V

4 ) we have
p
1
micro
c
f2 V4 .
(111)
1

Below Sc , chRMT describes the spectral correlation of the QCD Dirac operator, whereas
beyond this scale corrections to the universal statistics arise.
The functional (109) bears a close relation to chiral perturbation theory (CPT) [35,36].
CPT is an effective low energy theory for mesonic degrees of freedom. Quark and gluons
do not appear as explicit degrees of freedom. We denote the pion fields with a (x) and
define


(112)
V (x) = exp i a (x) a /f SU(2),
Sc =

where a are the generators of SU(2), i.e., the Pauli matrices. In lowest order in the
momentum the Lagrangian reads
LCPT =

|hi|



f2
b V (x) + V (x) .
tr V (x) V (x) +
tr M
4
2

(113)

388

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

b = diag(Mu , Md ) have been added to describe the explicit breaking of

The quark masses M
chiral symmetry due to non-vanishing current quark masses. It was argued in Refs. [9,10]
that the Thouless energy can already be identified from the Lagrangian (113). Clearly,
there is a competition between the kinetic and the symmetry breaking term, leading, up to
an order of magnitude, to the same energy scale as given in Eq. (110). In order to find a
direct relation to the spectrum of the Dirac operator, the authors of Refs. [9,10] investigated
the Lagrangian of partially quenched chiral perturbation theory (pqCPT) [37]. We review
their arguments briefly.
One can define the partially quenched partition function of QCD in the sector of zero
topological charge by


b J
Z pq Mv , M,
N
Z


/ iMv J ) Yf
det(iD[A]
det iD[A]
/ iMf eSYM [A] ,
(114)
= d[A]
det(iD[A]
/ iMv )
f =1

with Nf sea quarks with masses Mf , and a valence quark with mass Mv together with
its supersymmetric bosonic partner. An external source term J has also been included. In
the case J = 0, the determinants containing the valence quarks cancel. This is why one
b J ) with
uses the notion partially quenched in this context. The derivative of Z pq (Mv , M,
respect to J generates the function





b = log Z pq Mv , M,
b J 
,
(115)
Mv , M

J
J =0

which is the valence quark mass dependence of the chiral condensate [9], not to be
confused with the self-energy defined before. It was argued in Ref. [9] that the spectral
b after analytically continuing it into the complex
density can be obtained from (Mv ; M),
plane, from the discontinuity across the imaginary axis:





1
b i , M
b .
lim i + , M
(116)
R1 () =
2 0
In the low energy limit, the QCD action is modeled by the effective action of CPT given
by Eq. (113). Generalizing the model to Nf flavors and taking into account the fermionic
and bosonic valence quarks, one arrives at the action [9,10]
 2

Z





c U (x) + U (x) , (117)
b = trg d 4 x f U (x) U (x) + |h i| M
F mv , M
4
2
c = diag(M1 , . . . , MNf , Mv , Mv ). In the quenched
where U (x) SU(Nf + 1/1) and M
limit, Nf = 0, and after analytically continuing Mv in the complex plane, Mv i, we
see the close similarity between Eq. (117) and our model (109) up to numerical constants.

6. Summary and discussion

We studied the propagation of a Dirac particle in a four dimensional disorder potential.
Our model is motivated by similarities of finite volume QCD to disordered systems which

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

389

were found by analogy arguments and furnished with empirical evidence from lattice
QCD data. Our ansatz was stochastic. We did not attempt to solve the dynamics of
the gluon background but rather averaged over the statistical properties induced by the
gluon dynamics. We used a supersymmetric technique. Although supersymmetry is, in
our context, solely a mathematical tool which does not directly relate to physical particles,
we showed that the symmetries of the functionals exhibit symmetries analogous to theories
built upon particles. We found that chiral symmetry is spontaneously broken in a mean field
approach, valid in the t Hooft limit of large Nc . We showed how that limit corresponds to
the weak disorder in mesoscopic physics.
Within the mean field approximation, we derived effective theories for the spectral
correlations of the particle. We identified the equivalent of the Thouless energy in the bulk
and in the microscopic region of the spectrum. This demonstrates that the Thouless energy
is generic and not only present in the microscopic regime. Moreover, we discovered several
connections of our disorder model to other low energy effective theories of QCD as well
as to Efetovs non-linear sigma model.
In Fig. 7, we show a schematic representation of the various relations among the models
that we investigated. QCD as the fundamental theory of the strong interaction is the starting
point. It describes both the quark and the gluon fields. As discussed earlier, one has to
resort to effective models, although they cannot be derived rigorously from QCD. This is
indicated in the upper part of the figure. In the leftmost column, we have chiral perturbation
theory (CPT), and partially quenched chiral perturbation theory (pqCPT) which is a
subclass of CPT. CPT implies broken chiral symmetry from the beginning. It is a theory
of pion fields, and quarks and gluons are not present as explicit degrees of freedom. In the
rightmost column, we have the NJL model. It starts with a chirally symmetric action, but
the ground state of the theory is not invariant under chiral transformations, i.e. it undergoes
chiral symmetry breaking spontaneously. The NJL model appears in two versions: one is
a theory of quark degrees of freedom with a four-point fermion interaction and the other
is the bosonized theory consisting of mesonic degrees of freedom. They are related to
each other by a HubbardStratonovitch transformation. Gluonic degrees of freedom are
not explicitly present in both cases. However, they are included in a simplified fashion in
the point-like interaction of the model. In the second column from the right, we have placed
our disorder model. It is a description of fermions coupled to a stochastic disorder potential.
After performing the disorder average, we obtain a four-point interaction of fermions.
This reveals the close similarity between the NJL model and the disorder model already
at this stage. The disorder model can be reformulated in terms of superfields which
are coupled by a four-point interaction due to the disorder average. After decoupling the
quartic terms via a HubbardStratonovitch transformation and integrating out the fields,
one arrives at a model containing the superfields . Again, we have a connection to
the NJL model. The superfields are interpreted as the supersymmetric version of the
quark fields of the NJL model and the superfields as the supersymmetric version of
the meson fields of the bosonized NJL model. Moreover, we found that the mean field
equations of both NJL and the supersymmetric model are equivalent. From the mean field
approximation to the supersymmetric functional, we obtain two effective non-linear sigma

390

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

Fig. 7. Relations among several models. The low energy effective theories of QCD which we
discussed are depicted in the second row from above. The solid lines represent rigorous calculation,
whereas dashed lines indicate close resemblance between the corresponding models. The dotted lines
symbolize the zero-dimensional limit of some of the models, leading to RMT or chRMT.

models, depending on whether the bulk or microscopic region is considered. In the bulk,
Efetovs non-linear sigma model is reproduced, but now in four spacetime dimensions. In
the microscopic region, we have two features. First, the chiral symmetry, which was present
in the supersymmetric formulation before, undergoes spontaneous breaking thus lowering
the symmetry of the ground state. Second, the microscopic non-linear sigma model is
related to pqCPT, since it has been seen that the actions in both cases are similar. In the
second column from the left, we have the chiral random-matrix models. They are obtained
as the zero-dimensional limit of several models, as indicated in the figure. Moreover,
conventional RMT is obtained from the zero-dimensional limit of the bulk model, as shown
in the lower part of the figure.
Hence, our statistical model properly describes several aspects of low energy QCD.
Importantly, our study shows that a detailed understanding of the gluon dynamics is not
needed to model the spectral fluctuations.

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

391

Fig. 8. Attempt to illustrate the relations between QCD, our non-linear models and chRMT.

In Fig. 8, we make a second attempt to illustrate the various connections and relations
between models. Here we compare the actions of the models involved. In the first line we
have the QCD action SQCD with the physical quarks q and with the YangMills action
SYM containing the gauge fields. To be general, we included here the sea quark masses M.
0
which
If we integrate out the quarks, we arrive, on the left hand side, at the action SQCD
is still exact. If one now replaces everything that contains gauge fields with a random
matrix W , one arrives at the action SRMT of chRMT. Starting from the first line, one can
argue that the integration over the gauge fields, which cannot be done exactly, will lead to
the class of models of the NJL type, represented by the action SNJL on the right-hand side.
2 . If we now formally
Here, (qq)
2 stands for any effective potential depending on (qq)
0
,
replace the physical quarks q with superfields we find supersymmetric models SSUSY
precisely of the type discussed in the present article. They can be rewritten exactly in the
form SSUSY . However, if we now take the zero-dimensional limit of this model, we are back
to the chRMT model SRMT . This shows how all theses models are connected in a loop-like
structure. The limit of weak gauge field disorder yields the two non-linear models Smicro
and Sbulk . The former is directly related to chiral perturbation theory represented by the
action SCPT . The latter, however, cannot be simply related to an effective QCD model.
Remarkably, at present, the only way to derive it is the one we gave here.

392

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

As has now been demonstrated by a wealth of studies, a detailed understanding of the

gluon dynamics is not needed to model the spectral fluctuations. Even more, a quantitative
description is obtained. We hope that it will be possible one to use these insights as an input
for lattice QCD, either by constructing effective models or by modifying the algorithms
accordingly. This is, undoubtedly, rather ambitious, but, although undoubtedly, it seems a
worthwhile goal to be pursued.

Acknowledgements
We acknowledge fruitful discussions with R. Baltin, S.P. Klevansky, E. Lutz, D. Toublan,
J.J.M. Verbaarschot and H.A. Weidenmller. We thank M.R. Zirnbauer for important
comments.

Appendix A. Expansion in symmetry breaking parameter

We expand the zeroth order term in the saddle point approximation up to linear order in
the symmetry breaking parameters.
A.1. Bulk: Expansion in
As discussed in Section 4.3, the superfields at the saddle point are given by 0 =
i Im M (u0 (x)Lu0 (x)) 14 = 0 (x) 14 , with u0 u0 = u0 u0 = 14 , cf. Eq. (84). We
note that the supertrace over any power of 0 (x) vanishes, since
(

n
i Im M trg 14 = 0, n even,

n
(A.1)
trg 0 (x) =

n
i Im M trg L = 0, n odd.
Thus, the quadratic term in Eq. (49) vanishes. We expand the logarithmic term in Eq. (49)
in powers of by first writing




)
tf
rg log 0 (x) + 14 + L 14 14 (i/







= tf
rg log 0 (x) + L 14 14 S0 (x, x; ) + 14 14


) ,
(A.2)
+ tf
rg log 14 ( i/
). The second
where S0 (x, x; ) is the Green function associated with the operator ( i/
term on the right-hand-side vanishes, because it is proportional to unity in superspace.
From the definition of the logarithm, the first term can be expanded in a series. This yields


tf
rg log 0 (x) + L) S0 (x, x; ) + 14 14
= tf
rg

X

N
(1)N
0 (x) + L S0 (x, x; )N
N

N=1

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

X
(1)N
N
N=1


2
+O ,

= tf
rg

0 (x)

N

+ NL 0 (x)

N1

393

S0 (x, x; )N
(A.3)

where in the last line terms of order higher than linear in are neglected. The
terms 0N vanish according to Eq. (A.1). Correspondingly, for odd N , the terms
L(0 (x))N1 vanish, whereas for even N they give a non-vanishing contribution, since
trg(0 (x)L) = i Im M trg(Lu0 (x)Lu0 (x)) 6= 0. Thus, after relabeling the summation
index, the logarithm in Eq. (A.3) is written to first order in ,

X



2N


S0 (x, x; )2N+2 + O 2 . (A.4)
i Im M
tf
rg log(. . .) = trg L0 (x) tr
N=0

The summation over N can be performed, yielding ()/ Im M . Collecting everything,

we get the result displayed in Eq. (95).
A.2. Microscopic region: Expansion in
The saddle point solution 0 (x) of the microscopic one-point function exhibits unitary
symmetry, according to the discussion in Section 4.3. The symmetry is broken by a nonvanishing value of . The microscopic fields fulfill the condition 0 (x)0 (x) = M 2 12 .
Thus, any power (0 (x)0 (x))N 14 vanishes. With the same line of argumentation which
leads to Eq. (A.3), we have


)
tf
rg log 0 (x) PR + 0 (x) PL + 12 14 + 12 (i/

X

N
(1)N h
0 (x) PR S0 (x, x; 0) + 0 (x) PL S0 (x, x; 0)
= tf
rg
N
N=1

N1
+ N 0 (x) PR S0 (x, x; 0) + 0 (x) PL S0 (x, x; 0)
i


12 S0 (x, x; 0) + O 2 .
(A.5)
All terms corresponding to order O(0 ) vanish. This can be seen from the chiral structure
of the free Dirac operator, which carries over to its inverse S0 (x, x; 0), see Eq. (7), and
the explicit form of the projectors PL,R , Eq. (26): odd powers N only give rise to offdiagonal elements in Dirac space, and the trace vanishes. Moreover, it is easy to see that
PL S0 (x, x; 0)PLS0 (x, x; 0) = PR S0 (x, x; 0)PR S0 (x, x; 0) = 0. Hence, only combinations
with the alternating multiplication of PL S0 (x, x; 0) and PR S0 (x, x; 0) do not vanish.
However, the supermatrices then appear only as powers of (0 (x)0 (x))N 12 and give
no contribution either.
The situation changes for the terms of order O() because there is an additional
multiplication with 12 S0 (x, x; 0), being proportional to the symmetry breaking
matrix 12 in superspace. This occurs in a similar manner as in the bulk, where
multiplication with L gives a non-vanishing contribution. Even powers of N 1 still
vanish, because they are now off-diagonal in Dirac space due to the multiplication with

394

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

S0 (x, x; 0). In contrast, odd powers do not vanish since they have diagonal elements in
Dirac space. Again, only combinations with alternating PL S0 (x, x; 0) and PR S0 (x, x; 0)
survive, yielding
tf
rg log(. . .)

"

= tf
rg

0 (x) PR + 0 (x) PL 12

+ O 2 .

iM

2N

#
S0 (x, x; 0)

2N+2

N=0

(A.6)

The summation can be performed as above, yielding the spectral density. However, one
has to take into account the projectors PR,L which project out the spectral density of the
left and right handed modes, L (0) and R (0), respectively. Due to chiral symmetry they
give the same contribution, adding up to the total density, L (0) = R (0) = (0)/2. Thus,
finally the symmetry breaking term in the microscopic region is given by Eq. (96).

Appendix B. Gradient expansion

We perform a gradient expansion of the quadratic fluctuations around the saddle point.
B.1. Bulk
From the bulk solution (84) of the saddle point equation, one has 02 (x) = const. Thus,
the variation of the Goldstone modes yields
0 (x)GM (x) + GM (x)0 (x) = 0.

(B.1)

In contrast, the massive modes commute with the saddle point solutions [11]
0 (x)mass (x) mass (x)0 (x) = 0.

(B.2)

We switch off the external sources and set the symmetry breaking parameter to zero, = 0.
From this we find
1 2
F [0 , GM ; ]
2
Z
Z
1
4
rg d x d 4 y S(x, y; 0 ; )S(y, x; 0; )GM (y)GM (x)
= tf
2
Z
1
tf
rg d 4 x GM (x)GM (x)
(B.3)
+
16Nc G
and
1 2
F [0 , mass ; ]
2
Z
Z
1
rg d 4 x d 4 y S(x, y; 0 ; )S(y, x; 0; )mass (y)mass (x)
= tf
2
Z
1
tf
rg d 4 x mass (x)mass (x).
+
16Nc G

(B.4)

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

395

The minus sign in the argument of the Green function in Eq. (B.3) compared to Eq. (B.4)
is the crucial difference between Goldstone and massive modes as we will see.
Starting with the massive modes, we find that the first term in Eq. (B.4) does not give
a contribution to the superintegral. This is because the square of the Green function S 2
enters. It can be shown [11,29] that S 2 is negligible in the same approximation leading
to the saddle point equation. This can be understood from its the analytical properties.
Since the poles are solely above or solely below the real axis, the integration contour can
be closed without enclosing poles. In our case, the momentum integration is ultraviolet
divergent, which gives a contribution from the semicircular contour. Nevertheless, since
this contribution is the same for both the square S 2 and the product S A S R , we conclude
that
Z


2
tr d 4 y S x, y; + i Im M
Z
 tr

d 4 y S x, y; + i Im M S y, x; i Im M .

(B.5)

This is also true in the case where Im M is replaced by a supermatrix. The argument
still holds for higher moments. Thus, the only contribution from the massive modes comes
from the second term of Eq. (B.4). Because of the equal number of commuting and anticommuting variables, the superintegral yields


Z
Z
1
(B.6)
d 4 x mass (x)mass (x) = 1.
d[mass ] exp
16Nc G
Hence, the massive modes are integrated out and give no contribution in the quadratic
approximation.
We now turn to the Goldstone modes. For the saddle point solution (83), the product of
Green functions in Eq. (B.3) is diagonal in the space of supermatrices,
S(x, y; 0 ; )S(y, x; 0; )




= 14 S x, y; + i Im M S y, x; i Im M .

(B.7)

For the first term in the gradient expansion, we obtain

Z




()
1
tr d 4 y S x, y; + i Im M S y, x; i Im M =
2
2 Im M
1
(B.8)
=
4G
from the integration over y, and the last equality follows from Eq. (67). We note that the
trace f
trg in front of the quadratic term in Eq. (B.4) yields a factor 4Nc from the trace over
the Dirac and color indices. From this, we see that the first term of the gradient expansion
exactly cancels with the second term in Eq. (B.3).
The linear order in the gradient expansion vanishes upon integration in spacetime.
For the same reason, terms with second derivatives vanish if they consist of combinations
(y x )(y x ), 6= , leaving the expression 1/2(x y)2 (x). With Eq. (B.7)
we see that the integration over y yields the diffusion coefficient D(),

396

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

1
tr
2

d 4 y (x y)2 S x, y; + i Im M S y, x; i Im M

()D()
.
(Im M )2

(B.9)

Collecting all terms, we finally obtain Eq. (98).

B.2. Microscopic region
Most of the arguments given in the above considerations for the bulk region can easily
be carried over to the microscopic regime. We conclude from the unitarity of the Goldstone
modes, see Eqs. (89) and (90),

(x) + 0 (x)GM (x) = 0.

0 (x)GM

(B.10)

Eq. (B.2) still holds for the massive modes. With the same line of argumentation, we
conclude that the massive modes do not contribute to the integral. We turn now to the
gradient expansion of the Goldstone modes. The contribution from the lowest order in
the gradient expansion again cancels with the second term in Eq. (B.3). The linear order
vanishes as well. Thus, one is left with the quadratic derivatives and the integration over
y yields the diffusion coefficient D(), now taken at zero virtuality, = 0. Finally, we get
Eq. (99).

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]

T. Guhr, A. Mller-Groeling, H.A. Weidenmller, Phys. Rep. 299 (1998) 189.

E.V. Shuryak, J.J.M. Verbaarschot, Nucl. Phys. A 560 (1993) 306.
J.J.M. Verbaarschot, T. Wettig, hep-ph/0003017.
J.C. Osborn, J.J.M. Verbaarschot, Phys. Rev. Lett. 81 (1998) 268.
J.C. Osborn, J.J.M. Verbaarschot, Nucl. Phys. B 525 (1998) 738.
R.A. Janik, M.A. Nowak, G. Papp, I. Zahed, Phys. Rev. Lett. 81 (1998) 264.
T. Guhr, J.-Z. Ma, S. Meyer, T. Wilke, Phys. Rev. D 59 (1999) 054501.
M.E. Berbenni-Bitsch, M. Gckeler, T. Guhr, A.D. Jackson, J.-Z. Ma, S. Meyer, A. Schfer,
H.A. Weidenmller, T. Wettig, T. Wilke, Phys. Lett. B 438 (1998) 14.
J.C. Osborn, D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 540 (1999) 317344.
P.H. Damgaard, J.C. Osborn, D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 547 (1999) 305.
K.B. Efetov, Adv. in Phys. 32 (1983) 53.
K.B. Efetov, Supersymmetry in Disorder and Chaos, Cambridge University Press, 1997.
A. Altland, B.D. Simons, Nucl. Phys. B 562 (1999) 445.
I. Ichinose, Nucl. Phys. B 575 (2000) 613.
K. Takahashi, S. Iida, Nucl. Phys. B 573 (2000) 685.
D.J. Thouless, Phys. Rep. 13 (1974) 93.
D.J. Thouless, J. Phys. C 8 (1975) 1803.
D.J. Thouless, Phys. Rev. Lett. 39 (1977) 1167.
B.L. Altshuler, B.I. Shklovskii, Zh. Eksp. Teor. Fiz. 91 (1986) 220, Sov. Phys. JETP 64 (1986)
127.
J.J.M. Verbaarschot, H.A. Weidenmller, M. Zirnbauer, Phys. Rep. 129 (1985) 367.
Y. Nambu, G. Jona-Lasinio, Phys. Rev. 122 (1961) 345.

T. Guhr, T. Wilke / Nuclear Physics B 593 (2001) 361397

[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]

397

Y. Nambu, G. Jona-Lasinio, Phys. Rev. 124 (1961) 246.

T. Schfer, E. Shuryak, Rev. Mod. Phys. 70 (1998) 323.
T. Schfer, E. Shuryak, J.J.M. Verbaarschot, Phys. Rev. D 56 (1995) 1267.
T. Guhr, T. Wettig, Nucl. Phys. B 506 (1997) 589.
S.P. Klevansky, Rev. Mod. Phys. 43 (1992) 649.
T. Banks, A. Casher, Nucl. Phys. B 169 (1980) 103.
A.V. Andreev, B.D. Simons, N. Taniguchi, Nucl. Phys. B 432 (1994) 487.
A.A. Abrikosov, L.P. Gorkov, I.E. Dzyaloshinskii, Methods of Quantum Field Theory in
Statistical Physics, Prentice-Hall, New York, 1963.
A.L. Fetter, J.D. Walecka, Quantum Theory of Many-Particle Systems, McGraw-Hill, New
York, 1971.
B.L. Altshuler, B.D. Simons, Universalities: From Anderson localization to quantum chaos, in:
E. Akkermans, G. Montambaux, J.-L. Zinn-Justin (Eds.), Les Houches, Session LXI, 1994.
G. t Hooft, Nucl. Phys. B 75 (1974) 461.
E. Witten, Nucl. Phys. B 160 (1979) 57.
A.J. McKane, M. Stone, Ann. Phys. 131 (1981) 36.
J. Gasser, H. Leutwyler, Phys. Lett. B 188 (1987) 477.
J. Gasser, H. Leutwyler, Nucl. Phys. B 307 (1988) 763.
C.W. Bernard, M.F.L. Golterman, Phys. Rev. D 49 (1994).

Nuclear Physics B 593 (2001) 398412

www.elsevier.nl/locate/npe

Regularisation: many recipes, but a unique

principle: Ward identities and normalisation
conditions. The case of CPT violation in QED
Guy Bonneau 1
Laboratoire de Physique Thorique et des Hautes Energies, Unit associe au CNRS UMR 7589, Universit
Paris 7, 2 Place Jussieu, 75251 Paris Cedex 05, France
Received 25 August 2000; accepted 10 October 2000

Abstract
We analyse the recent controversy on a possible ChernSimons like term generated through
radiative corrections in QED with a CPT violating term: we prove that, if the theory is correctly
defined through Ward identities and normalisation conditions, no ChernSimons term appears,
without any ambiguity. This is related to the fact that such a term is a kind of minor modification of
the gauge fixing term, and then no renormalised. The past year literature on that subject is discussed,
and we insist on the fact that any absence of an a priori divergence should be explained by some
symmetry or some non-renormalisation theorem. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.10.Gh; 11.30.Er; 12.20.-m
Keywords: Ward identities; Radiative corrections; CPT violation; ChernSimons

1. Introduction
In the past years, the interesting issue of a possible spontaneous breaking of Lorentz
invariance at low energy has been considered: this issue also led to CPT breaking [13]. In
particular, the general Lorentz-violating extension of the minimal SU(3) SU(2) U (1)
standard model has been discussed: as many breaking terms are allowed, people look for
possible constraints coming from experimental results as well as from renormalisability
requirements and anomaly cancellation.
In that respect, there arose a controversy on a possible ChernSimons like term
generated through radiative corrections [27]. This phenomenon was studied in QED, an
abelian gauge theory, as a part of the standard model. We think that the controversy comes
E-mail address: [email protected] (G. Bonneau).
1 Tel: 33-144277657; Fax: 33-144277990.

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 2 5 - 8

G. Bonneau / Nuclear Physics B 593 (2001) 398412

399

from some misunderstandings on regularisation and (re)normalisation of a given theory

(see a review in [8]), and this paper intends to clarify the situation.
In the next section, the basis facts are recalled, with emphasis on Ward identities,
and the distinct three concepts of counterterms (finite or not), quantum corrections or
spurious anomalies and physical radiative corrections are defined. In Section 3, breaking
terms are added to the usual QED Lagrangian density, and one loop contributions are
discussed. We check that, as soon as the theory is precisely defined through its symmetries
(Ward identities) and its physical parameters (Normalisation conditions), there remains no
ambiguity in the results. Then we prove a kind of non-renormalisation theorem that allows
for a minimally CPT broken theory (Subsection 3.2). Finally, the literature is discussed
and some comments are also offered on the invoked difference between an invariance of the
action that leaves the Lagrangian density non-invariant, and one that leaves the Lagrangian
density invariant.

2. Quantum electrodynamics
The Lagrangian density is:
1 2
1
1
m eA
(A)2 + 2 A2 ,
/ ) F

(1)
L0 = (i/
4
2
2
where is the gauge parameter and an infra-red regulator photon mass. The classical
field equations write:

A
/ (x)(x),
(x)(i
/ m) = e(x)
/ (x),
(i / m)(x) = eA






1

A (x) = e(x)
2 + 2 A (x) 1
(x),

(2)
(3)

and the Feynman rules are:

0
(p, p)class.
hA (p)A (p)iclass. = D


p p
i
i p p
= 2
g
+ 2
2
2
p
p
p 2 p2
i
h(p) (p)iclass. = S0 (p, p)class. =
p
/ m
prop.
h (p)(q)A ((p + q))iclass. = (p, q, (p + q))class. = ie .

(4)

The Ward identity resulting from gauge invariance ensures that the non-transverse part of
the 2-photon proper Green function (p, p)class. = i(g p2 p p ) + i2 g
i
2
p p is not renormalised, i.e., the parameters and are (unphysical) tree level
parameters.
By power counting, the primitively divergent graphs are here (p, p)transverse ,
(p, p) and (p, q, (p + q)), respectively the transverse photon and electron
2-points proper Green functions and the photonelectron proper vertex function. All
corresponding parameters (positions and residues of the poles in propagators, couplings

400

G. Bonneau / Nuclear Physics B 593 (2001) 398412

at zero momenta, . . . ) but for the unphysical, non renormalised ones and 2
require normalisation conditions, a point which has often been missed since the successes
of minimal dimensional regularisation scheme [8]. In perturbative quantum expansion,
this requires addition of definite counterterms into the Lagrangian density: the question
of their being finite or not being merely a question of personal taste. For example, in
the original BPHZ [9] substraction scheme, a definite Taylor expansion with respect to
external momenta is subtracted from the Feynman integrand so that the integration over
loops momenta becomes possible: some finite counterterms are then needed to implement
the wanted normalisation conditions (of course, they will depend on some renormalisation
scale used to fix the normalisation conditions). In other schemes, infinite counterterms
(plus finite ones of course!) are defined after some regularisation of the divergent integrals
in order to achieve the same aim. In the BRS approach, the number of primitive divergences
and corresponding counterterms is given by the dimension of the FadeevPopov 0-charge
sector of the cohomology space of the Slavnov operator corresponding to the isometries
that define the theory [10].
The second problem is related to the symmetries of the classical action: the physical
meaningfulness of the quantum extension requires that the symmetries still hold at the
quantum level, which is possible as soon as the corresponding Ward identity has no
anomaly. As is well known, this may involve the addition of finite counterterms to
the Lagrangian density (the so-called quantum corrections or spurious anomalies). In
particular, each time one regularises a theory without respecting its symmetries, such nonsymmetric quantum corrections are needed, and moreover the classical currents have to be
redefined (renormalised) in order that their conservation should lead to the correct Ward
Identity (the correct contact terms . . . ). A pedagogical example may be found in the
second reference in [11]. In the BRS approach, the absence of anomaly corresponds to an
empty FadeevPopov charge-1 sector of the cohomology space of the upper mentioned
Slavnov operator.
Finally, the success of a perturbative theory lies in its ability to compute with precision
some quantities such as S matrix elements whose classical values acquire radiative
corrections, for example the Bremsstrahlng in annihilation processes: e+ + e
photon + final state X, or the magnetic moment of the electron which is found to be 2
in the classical theory with Lagrangian (1) and gets some definite corrections in higher
orders in the electric charge, or the Lamb-shift that results from higher loop contributions
to the photon self-energy , etc.
One should not confuse these three concepts: Lagrangian counterterms (finite or not)
required to get a definite perturbative expansion with definite values for the physical
parameters of the theory; Lagrangian finite spurious anomalies or quantum corrections
required to compensate for the use of a non symmetry preserving regularisation scheme;
calculable radiative corrections to physical processes.
For instance, in QED the electromagnetic current is conserved, thanks to the gauge
invariance of the theory; so one obtains the Ward identity:
L+1;2N
(p, pi ; qj ; qj0 ) =
p ;
i

i 2
0
p 2 p11 L1 N

G. Bonneau / Nuclear Physics B 593 (2001) 398412

+e

N
X


401

L;2N
(pi ; qj , qk + p, . . . ; qj0 )

k=1
L;2N
(pi ; qj ; qj0

i

, qk0 + p, . . .)

(5)

where L is the number of external photons (of momenta pi : i = 1, 2, . . . , L) and N

the number of incoming and outgoing electrons (of momenta qj0 : j = 1, 2, . . . , N), all
momenta being ingoing ones. In particular, for L = 1, N = 0 this gives the announced
non-renormalisation of the longitudinal part of the photon propagator i (p2 2 ). This
Ward identity is best rewritten on , the generating functional for proper Green functions:





1

ie (x)
= 2 + 2 A .
(x)
(6)
x
A (x)
(x)

(x)
Then, if one uses the BPHZ scheme that breaks gauge invariance, the addition of finite
counterterms into the Lagrangian and a redefinition of the electromagnetic current is
required in order for the Ward Identity (6) to be satisfied to all orders of perturbation
theory. On the contrary, PauliVillars or dimensional regularisation only need symmetric
counterterms (the usual Z factors).
In the same spirit, a softly broken axial symmetry exists at the classical level; but it does
not survive quantisation because of the axial anomaly. This one is readily seen, for example
within dimensional regularisation [12,11], to result from the impossibility of defining in an
algebraically consistent way (in complex D dimensions) a matrix 5 that anticommutes
with all Dirac matrices : = D 13. So, when computing
 5 
(x)

with the help of the electron field equations, an evanescent contribution [11,12] appears:
 
D (x)
where 2 = { , 5 } and D denotes the covariant derivative + ieA . Moreover the
axial current has also to be redefined, a point often missed (see for example controversies
on Adler Bardeen theorem in super-YangMills between 1982 and 1985 [8] and a recent
paper by Jacquot [14]).

3. QED with odd-CPT Lorentz violating terms

As explained in the introduction section, let us now consider QED (Eq. (1)) with possible
presence of CPT-odd, Lorentz violating terms:
5

where b is a fixed vector,

L1 (x) = b (x)
(x),
1
(7)
L2 (x) = c  F (x)A (x), where c is a fixed vector.
2
Other breakings could be considered (see a discussion in the first paper of [2]), but we
simplify and require charge conjugation invariance, which selects L1 (x) and L2 (x). Note
for further reference that experiments on the absence of birefringence of light in vacuum

402

G. Bonneau / Nuclear Physics B 593 (2001) 398412

put very restrictive limits on the value of c , typically for a timelike c , c0 /m 6 1038
[2].
Let us consider the classical Lagrangian density L0 + L1 + L2 . In order to avoid the
difficulties resulting from the new poles in the propagators, we take into account the
smallness of the breakings and include them into the interaction Lagrangian density as
super-renormalisable couplings. Moreover, we define the photon and electron masses by
the same normalisation conditions as in ordinary QED, e.g.,

prop. 

(p) (p)
p
/ =m, b=c=0 = 0,
According to standard results in renormalisation theory, these breakings add new terms in
the primitively divergent functions and require 2 new normalisation conditions to fix their
parameters b and c :

prop. 
i 

,
b = Tr 5 (p) (p)
p=0
4
prop. 
1



.
A (p)A (p)
c =
p=0

12
p

(8)

Note that, contrary to L1 (x), the L2 (x) term also breaks the local gauge invariance of
the Lagrangian density, even if the action remains gauge invariant. If fields and the gauge
parameter function (x) vanish sufficiently rapidly at infinity, there will be no difference;
however, the literature on this subject [4] emphasizes the difference, and we want to clarify
this point. We shall prove that the variation of L2 (x) in a local gauge transformation being
linear in the quantum field, no essential difference occurs.
So it is tempting to separate the discussion into two cases:
QED with the two CPT odd, Lorentz breaking, C conserving terms: L1 (x) and L2 (x).
QED with the sole breaking term L1 (x): is it a consistent quantum theory?
3.1. Lagrangian of QED with CPT-odd, Lorentz and gauge breaking terms
Starting from the Lagrangian density L0 + L1 + L2 , let us analyse the Ward identity
corresponding to the gauge invariance of the action. The classical field equations are now:

/ (x)(x),
i / m b/ 5 (x) = eA

A

/ (x),
(9)
(x)
i / m b/ 5 = e(x)






1

A (x)  c F (x) = e(x)

2 + 2 A (x) 1
(x); (10)

](x) is conserved, and Eq. (10) the fact

then Eqs. (9) ensure that the vector current [

that the scalar field A (x) is a free field of squared mass 2 .

After addition of adequate counterterms, these equations of motion may always be
extended to the quantum level [9,13], and, as a consequence, the same is true for the vector
current conservation and the free-character of the longitudinal photon A (x).

G. Bonneau / Nuclear Physics B 593 (2001) 398412

403

Let be the classical action

Z
= d 4 x [L0 + L1 + L2 ],
the Ward identity writes:
Z




+ i(x) (x)
(x)

d x (x)

e
A (x)
(x)
(x)


Z
1
2
1
= d 4 x A (x)2(x) + A (x) (x) +  c F (x) (x)
e
e
2e




1

ie (x)
(x) = 2 + 2 A (x).

Wx

A (x)
(x)

(x)
(11)

The last equation is exactly the same as the one for ordinary QED (6): so, to select the
desired action, one needs an extra symmetry such as Lorentz invariance if one wants
ordinary QED (1), or some non-renormalisation theorem if one wants to consistently
suppress the L2 term through a normalisation condition (Subsection 3.2).
As soon as we use a regularisation that respects the symmetries (gauge, Lorentz
covariance and charge conjugation invariance), the perturbative proof of renormalisability
reduces to the check that the O(h ) quantum corrections to the classical action : 1 =
|class. + h , constrained by the Ward identity (11):




ie (x)
(x) = 0,
(12)

Wx

A (x)
(x)
(x)
may be reabsorbed into the classical action through suitable renormalisations of the fields
and parameters of the theory. Thanks to the quantum action principle [17,10], is a charge
conjugation invariant integrated local polynomial in the fields, their derivatives and the
parameters of the theory, of dimension 4 (recall that the photon field and the parameters
m, b , c have dimension 1, the electron field dimension 3/2): then the general solution of
(12) is readily shown to be of the same form as the classical action (without the gauge
fixing and photon mass terms).
Q.E.D.
The breakings (7) introduce two operators which may be defined through a modification
of the classical action: to we add two source terms for the C = +1, dimension 3
insertions




1
5
5
5

J = (x) and K =  F A (x) :

2
Z


= + d 4 x (x)J5 (x) + (x)K5 (x) .
Then, as soon as the renormalisation is properly done, the operators being bilinear in the
quantum fields, the Ward identity (11) holds true, to all orders of perturbation theory (all
orders in h , e, b , c , m), for the Green functions with one insertion of either of these
operators:

404

G. Bonneau / Nuclear Physics B 593 (2001) 398412

action on (11) of

for the gauge invariant axial current J5 (y),

(y) 

side vanishing: Wx
(y) ==0 = 0,
action on (11) of (y) for the non gauge invariant operator K5 (y),

the right-hand

the right-hand
side reducing
itself
to
a
tree
contribution
as
the
variation
is
linear
in
the
photon field:

1
(y) (y x).

=


F
Wx
(y) ==0
y
2
All Green functions are of course computed with the complete Lagrangian density
L0 + L1 + L2 . Of course, as for ordinary QED, the axial current is not coupled to
the fields of the model, and its quantum non-conservation due to the axial anomaly
is not dangerous. However, should one consider the CPT breaking extensions of the
standard SU(3) SU(2) U (1) model, the generalisation of the AdlerBardeen non
renormalisation theorem would be necessary in order for the one loop cancellation of the
axial anomaly to stay to all orders.
As particular consequences of the previous discussion, the complete proper 2-photon
Green function (p, p) satisfies
p (p, p) = p (p, p) = 0,

(13)

up to the classical longitudinal contribution, the one with one axial-current insertion
verifies:
 5 


((p + q)) prop.

p A (p)A (q)
 5 


((p + q)) prop. = 0,

= q A (p)A (q)
(14)
and the one with a K5 insertion


prop.
0

p A0 (p)A (q) 12  F A ((p + q))



prop.
0

= q A (p)A0 (q) 12  F A ((p + q))

= i p q .

(15)

Note that the validity of these equations is not a question of personal taste or choice: if
the renormalisation is correctly done, they do hold true as a consequence of the gauge
invariance of the action (moreover, as we shall show, in the absence of (14), (15), the
finiteness of the CPT-odd part of the photon self energy will remain accidental).
We now illustrate these results by a one-loop calculation. We first need a regularisation
to give a meaning to the loops integrals involved in those Green functions. Any consistent
one is as good as any other, but it is simpler to consider regularisations that respect
the invariances of the classical theory, here gauge invariance. A question arises about
Lorentz non-invariance: a priori there is no longer symmetric integration and averaging
formulas for l l (where l is a loop momentum variable): any computation will become
fairly hard; in particular, linear divergences would stay 2 and require extra subtractions
2 For instance, as:

x dx
40
p
= 0 + 2p p log
+ O(1/, 1/0 ).
2
2
m2
(x + p) + m

G. Bonneau / Nuclear Physics B 593 (2001) 398412

405

and counterterms. However, as discussed in [2], the spirit of the spontaneously broken
Lorentz invariance is that, except for the vacuum expectation values b and c of some
fields, Lorentz invariance holds: Colladay and Kosteleck speak of true observer Lorentz
invariance. So we shall use the Lorentz preserving consistent dimensional regularisation
of t Hooft and Veltmann 3 [8,1113] and, in particular, check at the one-loop order that
the correction to the self-energy of the photon of first order in b unambiguously vanishes
at p2 = 0, in agreement with the theorem in the appendix of [3].
Note that gauge invariance (13) ensures, as in standard QED, that the 4-photon Green
function is not primitively divergent. Then h L0 , the standard counterterms of QED
should be added to L0 . Let us now compute the b and c dependent counterterms, to
one-loop order.
3.1.1. The one-loop electron self-energy
To first order in the small breaking parameters b and c , a one-loop calculation gives:

prop. 

= p
/ m b/ 5 (1 + I ) 3mI 3/c 5 I
i (p) (p)

where I =

he2

16 2

+ regular (2 -independent) terms,



m2
2
C + log
;
4D
42

(16)

C is the Euler constant and 2 the scale needed in the dimensional scheme,
D
d 4k
(4D) d k

.
(2)D
(2)4

In such a calculation, 4 as there is no need for traces of Dirac matrices, there is no

unconsistency in using a fully anticommuting 5 matrix. So we have the standard QED
renormalisations of the fermionic field (wave function and mass) and a c dependent
renormalisation of the b parameter in L1 .
Higher orders in b and c are given by convergent integrals, then contribute as regular
(2 independent) functions at D = 4, and do not change the result (16). Of course, should
one implement the normalisation condition (8), extra finite counterterms renormalizing L1
should be required. Indeed, Lorentz covariance and (16) ensures that

prop. O(h )
i 

Tr 5 (p) (p)
p=0
4
 2 2

 2



b c b.c
b c2 b.c
' b + 3c I + b a1 2 , 2 , 2 + c a2 2 , 2 , 2 .
m m m
m m m

(17)

3 Let us emphasize that if one uses correct (anti-)commutation relation of Dirac matrices and 5
(Breitenlhoner and Maison [13]), there will be no ambiguity in the results and a minimal subtraction can be
safely done (it corresponds to some implicit but definite normalisation conditions).
4 We used = D and k k = k 2 g /D: this allows shifting of internal momenta and ensures gauge

invariance. Other prescription would lead to different finite counterterms (spurious anomalies). This point is
sometimes missed, in particular, in Ref. [6] (after Eq. (7)), leading to differences proportional to 4 D which
give extra finite contributions, when involved into divergent integrals.

406

G. Bonneau / Nuclear Physics B 593 (2001) 398412

Note also that, as we choose to fix the wave function and the mass of the electron by the
usual normalisation conditions at vanishing b and c , no finite new counterterms, such
]/m2 , will be required. At this point, additive renormalisation requires a
as b b [
h L1 (c, b) counterLagrangian.
3.1.2. The one-loop photon self-energy
In the same way, the self energy of the photon may be expanded in powers of b (up to
one-loop order, L2 does not modify the photon self energy, except at the tree level). The
b independent contribution comes from ordinary QED, the b linear one being given by
the two triangle-like graphs with a zero-momentum axial vertex:
Z
q + m) (/
q +p
/ + m) 5 (/
q +p
/ + m)]
d 4 q Tr[ (/
2
,
I (p) = h e b
4
2
2
2
2
2
(2)
(q m )[(q + p) m ]
Z
q + m) 5 (/
q + m) (/
q +p
/ + m)]
d 4 q Tr[ (/
h e2 b
4
2
2
2
2
2
(2)
[q m ] ((q + p) m )
Z
4 q Tr[ (/
d
/ + m) (/
q + m) 5 (/
q + m)]
q +p
= h e2 b
4
2
2
2
2
2
(2)
[q m ] ((q + p) m )
(qqp)

I (p),

(18)

as soon as shift of loop momenta is allowed and thanks to the cyclicity of the trace of a
product of matrices (this last property is independent of the dimension of the matrices and
should hold in any consistent dimensional regularisation). The only algebraic properties we
need for the calculation in D-dimension are the algebraically consistent ones [15,13,8]:
as previously emphasized, = D and k k = k 2 g /D (k being the shifted loop
momentum after the standard Feynman trick for combining propagators into a single
denominator),
the trace of 5 times an odd number of Dirac matrices vanishes,
the trace of 5 times an even number of Dirac matrices can be reduced with the
Clifford algebra to the quantity Tr[ 5 ]; consistency enforces Tr[ 5 ] =
Tr[ 5 ] = 0.
Using Feynman parameters to combine the denominators in (18), the first triangle
contribution gives

he
2 
b p Tr 5 (A + B)
2
8
where the divergent part comes from:

Z1
A =
0



dx x(4 6x) (4 D)x(2 x)
Z

q 0 2 /D
(42 )2D/2 d D q 0
,
D/2
[q 02 + x(1 x)p2 m2 ]3

and B, a convergent one, may be evaluated at D = 4:

G. Bonneau / Nuclear Physics B 593 (2001) 398412

Z1
B =
R1



dx x 2 (1 x)2 p2 + x(2 x)m2

407

d 4q 0
1
.
2
02
[q + x(1 x)p2 m2 ]3

As 0 dx x(46x) = 0, the divergent part vanishes and, after Wick rotation and integration
on q 0 , one is left with the finite quantity:
i
A+B =
2

Z1


2 + m2
x(1 x)pE
dx x(2 3x) log
x(2 x)
42

2 x(2 x)m2
x 2 (1 x)2 pE
2 + m2
x(1 x)pE


.

An integration by parts on x for the log term finally gives:

i
2
A + B = p2 K p2 = ipE
2

Z1
dx
0

x 2 (1 x)
2 + m2
x(1 x)pE

p 2 0

' i

p2
,
12m2

a finite result, which moreover vanishes for p2 = 0. The complete one-loop 2-photon
proper Green function is:
prop.
hA (p)A (p)itransverse




4
= i g p2 p p 1 I + regular ( 2 -independent) terms
3


2
h e
 p 2c + b 2 p2 K(p2 ) ,
2
p2 K(p2 ) is analytic for p2 < 4m2 ,
p
p2
log [ + 1 + 2 ]
2
2
p
' E2
p K(p ) = 1
6m
1 + 2

s
2
when pE
0 with =

!
2
pE
.
4m2
(19)

are given
Note that, here also, contributions with higher order dependence on
by convergent integrals (then contributing as 2 independent terms), thanks to gauge
invariance (the unique logarithmically divergent polynomial term b b , cannot be
transverse on p and p ).
A few remarks are in order to understand the results:
The finiteness does not come from a cancellation between the two triangle-like
graphs: on the contrary, each of the triangle being superficially linearly divergent,
its divergent part is a dimension-1 polynomial in masses and external momentum p.
The only possible structure (in any regularisation that respects Lorentz covariance) is
 p , then the divergence is at most a logarithmic one; moreover, thanks to Bose
symmetry ( , p p), it doubles itself when the two triangles are added.
Then, to understand the convergence, it is necessary to use gauge invariance and
the Ward identity (14) for the operator J5 . First, note that the O(b ) term is given by

408

G. Bonneau / Nuclear Physics B 593 (2001) 398412

the axial vertex hA (q)A (p)J5 ((p + q))iprop. for q + p = 0. Second, the divergent
part of this vertex is a dimension-1 polynomial in masses and external momenta p and
q: A (p q) , thanks to Bose symmetry, constrained by gauge invariance (14),
then it vanishes. Third, the value of this vertex, thus given by convergent integrals, is
uniquely fixed, independently of the regularisation used, as soon as the vector current
conservation is ensured (see the nice old discussion in Adlers 1970 lectures [16]),
and of course the same occurs for the photon self energy correction b [I (p) +
I (p)].
As argued by Coleman and Glashow [3], due to analyticity, from the once well known
theory of Feynman-diagram singularities, this ChernSimons like contribution will
Q.E.D.
vanish at p + q = 0, p2 = 0.

The finiteness a fortiori holds for higher orders in b : the only superficial divergence
could come from a b b ; (p, p), but the corresponding dimension zero polynomial
divergence has to be transverse with respect to the photon momenta p , so it must
vanish. Moreover, should one implement the normalisation condition (8), no extra finite
counterterm renormalizing L2 should be required. Indeed, Lorentz covariance and a
generalisation of Coleman and Glashows argument ensure that
prop. O(h )
1



=0
A (p)A (p)
p=0

12
p
(consider the proper Green function of two photons A (p) and A (q) with one insertion
of b J5 ((p + q)), computed with the complete Lagrangian density L0 + L1 + L2 , i.e.,
to all orders in b , and the corresponding Ward identity (14)). Note also that, as we choose
to fix the wave function of the photon by the usual normalisation conditions at vanishing
b and c , no finite new counterterm, such as b2 [F F ]/m2 , etc., will be required.
At this point, additive renormalisation requires no h L2 (c, b) counter-Lagrangian.
3.1.3. The one-loop photonelectron vertex
Finally, simple power counting shows that the vertex function h(p) (q) A((p +
q))iprop. has no b or c dependent divergence: so, as we choose to fix the electron
photon vertex by the usual normalisation condition at vanishing b and c , the QED
counter-Lagrangian h L0 ensures finiteness and correct normalisation conditions.
3.1.4. Higher-loop orders
To summarise, to one-loop order, given the classical Lagrangian L0 + L1 + L2 ,

renormalisation only requires the counterterms h(L

0 + L1 ): if b is renormalised to

b1 (b, c), no h L2 correction has appeared. The ChernSimons like term is not renormalised
even if the c parameter is rescaled by a Z31 factor, to compensate for the usual photon

field renormalisation Z3 . Does this stay in higher orders?

Let us suppose that, up to N-loop order, renormalisation has been done with counterterms L0 + L1 and that the only dependence on b and c of the Lagrangian plus

5 ](x), b (b, c) being

counterterms is through the classical L2 term and bN (b, c)[
N
an order-N polynomial in h.

G. Bonneau / Nuclear Physics B 593 (2001) 398412

409

One of the primitively divergent Green functions is the electronphoton vertex. As

previously argued, its divergence and normalisation condition are the same as for
ordinary QED, and so taken into account by the counterterm L0 .
To (N + 1)-order, the b , c dependent part of the divergence of the self-energy of
the electron (subdivergences being properly subtracted) is proportional to b 5
and c 5 . Moreover, as
 2
 2
O(hN+1 )

i 

= b aN+1
b
+
c
Tr 5 h(p) (p)iprop. p=0
N+1
2
4
m
m2
 2



b c2 b.c
b2 c2 b.c

,
,

,
,
+
c
,
(20)
+ b N+1
N+1
m2 m2 m2
m2 m2 m2
the normalisation conditions (8) require only a L1 like counterterm.
On the one hand, the derivative with respect to b of the regularised photon selfenergy at (N + 1)-loop order is given [17,10] by:

(N+1) (p, p)
b
Z
prop. (N+1)
 


bN
4
5

d x (x)
A (p)A (p)
= i
b
Z
prop. (N+1l)
l=N
X  b (l) 


N
4
5

= i
d x (x)
.
A (p)A (p)
b
l=0

First, this quantity is finite: the proof is the same as the one given in Subsection 3.1.2, based on the vanishing of a dimension-one, Lorentz covariant 3-tensor
a , polynomial in masses, parameter b , external momenta p and q, transverse
with respect to p and q , and associated to the divergence of any n-loop order
5 ]((p + q))iprop. ,
(n 6 N + 1) axial (unintegrated) vertex hA (p)A (q)[
with subdivergences properly subtracted.
Second, we look for possible finite counterterms required by the normalisation
conditions (8); following Coleman and Glashows argument, 5 analyticity and
gauge invariance of the n-loop order axial (unintegrated) vertex, i.e., vector current
(n)
conservation, enforce its proportionality to some pa q b Gab; (p, q, b, m), n,
giving after integration pa pb G(n)
ab; (p, p, b, m).
On the other hand, the derivative with respect to c of the regularised photon selfenergy at (N + 1)-loop order is given [17,10] by:

(N+1) (p, p)
c
Z

 prop. N+1

1
4

 F A (x)
d x
= i A (p)A (p)
2
5 The existence of an anomaly for the axial current conservation law does not enter the game as we used only
vector current conservation.

410

G. Bonneau / Nuclear Physics B 593 (2001) 398412

= i

l=N
X
l=0

bN
c

Z

(l)
 prop. (N+1l)
1
 F A (x)
d 4x
.
A (p)A (p)
2

In that equation, the vertex insertions are at least at one loop order, so the same
argument as before holds, thanks to the Ward identity (15).
As a consequence, to (N + 1)-loop order, the quantity
prop. 
1



A (p)A (p)
p=0

12
p
is given by its value for vanishing b and c . Then the normalisation conditions (8)
may be ensured without any L2 counterterm.
The b and c independent part of the electron and photon self-energy being
correctly renormalised by the counterterm L0 , L0 + L1 ensures the correct
renormalisation up to order (N + 1).
Q.E.D.
So, to all orders of perturbation theory, the theory described by L0 + L1 + L2 is a quantum
consistent theory, with an (infinite) renormalisation of the photon field, electron mass,
electric charge and b parameter, and no L2 renormalisation.
3.2. Gauge invariant QED with a CPT-odd, Lorentz breaking term
Consider now a possible situation without L2 in the classical Lagrangian, i.e., with a
value zero for the parameter c defined by the normalisation condition (8). The analysis of
the previous Subsection shows that L2 is not renormalised: then, its absence at the classical
level is stable against perturbative expansion, to all-loop order, and the theory described by
L0 + L1 is a quantum consistent theory, with an (infinite) renormalisation of the photon
field, electron mass, electric charge and b parameter.

4. Discussion and summary

We have exemplified the fact that, as soon as a theory is correctly defined not only by
a Lagrangian density such as L0 + L1 + L2 , but by some symmetry requirement such as
gauge invariance of the action and appropriate normalisation conditions (8), the quantum
corrections are unambiguous.
The opposite conclusion often given in the literature [4,7], results from an insufficient
definition of the model and some unprecised arguments:
Jackiw and Kosteleck [4] never introduce any regulator. Then some of their relations
are delicate ones: see for example for a divergent integral (after Eq. (12)), the
commutation of a derivation with respect to external momentum and the integration.
If the integral in their Eq. (11), which is twice our tensor I (p), is computed with
dimensional regularisation, a result [ sin 1] is found, and not simply sin (with
p2 = 4m2 sin2 /2).
Moreover, in the absence of normalisation conditions or Ward identities fixing
some ambiguities, the difference of two equivalent linearly divergent integrals

G. Bonneau / Nuclear Physics B 593 (2001) 398412

411

gives an ambiguous logarithmic divergent one. Even when one uses a symmetric
integration that suppresses the linearly (and eventually the logarithmically) divergent
part, the finite part remains ambiguous. The surface term that comes from a
shift in the integration momentum in a linearly divergent integral is a regulator
dependent quantity: if one redoes the calculation in Appendix A5-2 of Jauch and
Rohrlich standard book [18] with the dimensional scheme, one easily checks that no
surface term occurs after a shift of the integration momenta [15]. Recall that this
possibility of shifting internal momenta is needed to preserve gauge invariance in loop
calculations (see, for example, [19, Subsection 17.9]).
If the gauge invariance constraint (14) is not imposed, the finiteness of the
triangles appears as purely accidental 6 [4,7] and would not stay in higher
orders. So, some authors rightly conclude that the corresponding one-loop finite
contribution is ambiguous [7]. Note that in the main text of [20], Jackiw correctly
summarizes the discussion of his Section 4 by the sentence An arbitrary value
persists only when no symmetry is enforced, but he gave his paper a misleading title:
When radiative corrections are finite but undetermined. Such unexpected finiteness
also occurs in other situations, for example in the so-called complex sine-Gordon
model: as was shown in [21] the origin lies, not in some isometry of the theory, but in
the physical property of non-production for the S-matrix.
Although local gauge invariance is lost at the level of the Lagrangian density, the
breaking is linear in the quantum fields, and then the invariance of the action ensures
the validity of the usual Ward identity that corresponds to vector current conservation.
So, the difference advocated by Jackiw and others between a term in the Lagrangian
density locally gauge invariant and one giving a gauge invariant contribution to the
action is not relevant for the present case, as, but for the tree level, such Green
functions with insertion satisfy the same usual Ward identity (11).
Finally, it is difficult to see the difference advocated in [4] between a first order (in b )
perturbative calculation and what is claimed to be a non-perturbative unambiguous
value, but is, as a matter of fact, obtained with exactly the same triangle integrals as
everyone. More precisely, after the expression given in [4, Eq. (5)] for a complete
(non-perturbative) contribution of the breaking to the 2-photon one-loop Green
function, it is argued that, as the linear divergences cancel among the two terms of
the integral, there will be no momentum routing ambiguity, and then a unique value
will be obtained, for example by an expansion in the parameter b . As we previously
explain, this argument is incorrect. Moreover, in [7] the computation is also done to
all orders in the breaking parameter b and it is explicitly verified that higher orders
do not contribute to a possible correction to c .
Note also that the Lagrangian density L0 + L2 would not lead to a coherent theory as an
(infinite) counterterm L1 appears at the one-loop order (16).
6 The vector current conservation (13) only imposes the transversality of the (divergent part of the) photon self
energy, which does not excludes a  b p infinite contribution.

412

G. Bonneau / Nuclear Physics B 593 (2001) 398412

To summarize, we have proven that a theory with a vanishing tree level ChernSimons
like breaking term is consistent as soon as it is correctly defined: thanks to the gauge
invariance of the action, the normalisation condition c = 0 may be enforced to all orders
of perturbation theory. A L2 term, bilinear in the gauge field, appears in facts as a minor
modification of the gauge fixing term as A remains a free field: then, as part of the
gauge term, it is, as usual, not renormalised.
Of course, the 2-photon Green function receives definite radiative corrections
h e2 p 2

' 12
2 m2  p b + . Recall the case of the electric charge: physically measurable
quantities occur only through the p2 dependence of the photon self-energy (as the Lambshift is a measurable consequence of a non-measurable charge renormalisation). Unfortunately, as Coleman and Glashow explained, the absence of birefringence of light in vacuum, i.e., the vanishing of the parameter c , gives no constraint on the value of the other
one b .

References
[1] S. Caroll, G. Field, R. Jackiw, Phys. Rev. D 41 (1990) 1231.
[2] D. Colladay, V.A. Kosteleck, Phys. Rev. D 55 (1997) 6760;
D. Colladay, V.A. Kosteleck, Phys. Rev. D 58 (1998) 116002, and references therein.
[3] S. Coleman, S.L. Glashow, Phys. Rev. D 59 (1999) 116008.
[4] R. Jackiw, V.A. Kosteleck, Phys. Rev. Lett. 82 (1999) 3572.
[5] W.F. Chen, Phys. Rev. D 60 (1999) 085007.
[6] J.-M. Chung, P. Oh, Phys. Rev. D 60 (1999) 067702.
[7] J.M. Chung, Phys. Lett. B 461 (1999) 138;
M. Prez-Victoria, Phys. Rev. Lett. 83 (1999) 2518.
[8] G. Bonneau, Int. J. Mod. Phys. A 5 (1990) 3831.
[9] W. Zimmermann, Local operator products and renormalization in quantum field theory, in: S.
Deser et al. (Eds.), 1970 Brandeis Lectures, Vol. 1, M.I.T. Press, Cambridge, 1970, p. 395.
[10] C. Becchi, The renormalization of gauge theories, in: B.S. de Witt, R. Stora (Eds.), Les Houches
1983, Elsevier, 1984.
[11] G. Bonneau, Nucl. Phys. B 171 (1980) 470;
G. Bonneau, Nucl. Phys. B 177 (1981) 523.
[12] G. t Hooft, M. Veltman, Nucl. Phys. B 44 (1972) 189.
[13] P. Breitenlohner, D. Maison, Commun. Math. Phys. 52 (1977) 11.
[14] J.L. Jacquot, Is the axial anomaly really determined in a continuous non-perturbative
regularization?, hep-th/9909014.
[15] K.J. Wilson, Phys. Rev. D 7 (1973) 2911, Appendix.
[16] S.L. Adler, Perturbation theory anomalies, in: S. Deser et al. (Eds.), 1970 Brandeis Lectures,
Vol. 1, M.I.T. Press, Cambridge, 1970, p. 1.
[17] J.H. Lowenstein, Commun. Math. Phys. 24 (1971) 1.
[18] J.M. Jauch, F. Rohrlich, The Theory of Photons and Electrons, Addison-Wesley, Reading, MA,
1959.
[19] J.D. Bjorken, S.D. Drell, Relativistic Quantum Fields, McGraw-Hill.
[20] R. Jackiw, When radiative corrections are finite but undetermined, hep-th/9903044.
[21] G. Bonneau, Phys. Lett. B 133 (1983) 341;
G. Bonneau, F. Delduc, Nucl. Phys. B 250 (1985) 561.

Nuclear Physics B 593 (2001) 415437

www.elsevier.nl/locate/npe

Top-quark couplings to TeV resonances at future

lepton colliders
T. Han a , Y.J. Kim a , A. Likhoded b , G. Valencia c,d,
a Department of Physics, University of Wisconsin, Madison, WI 53706, USA
b Institute for High Energy Physics, Protvino, Russia
c Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA
d Fermi National Accelerator Laboratory, Batavia, IL 60510,USA

Received 5 June 2000; accepted 18 October 2000

Abstract
We study the processes WL WL t t and WL ZL t b (tb) at future lepton colliders as probes
of the couplings of the top quark to resonances at the TeV scale. We consider the cases in which
the dominant low energy feature of a strongly interacting electroweak symmetry breaking sector is
either a scalar or a vector resonance with mass near 1 TeV. We find that future lepton colliders with
high energy and high luminosity have great potential to sensitively probe these physics scenarios.
In particular, at a 1.5 TeV linear collider with an integrated luminosity of 200 fb1 , we expect
about 120 events for either a scalar or a vector to decay to t t, tb. Their leading partial decay widths,
which characterize the coupling strengths, can be statistically determined to about 10% level. 2001
Elsevier Science B.V. All rights reserved.

1. Introduction
The mass generation mechanisms for electroweak gauge bosons and for fermions are
among the most prominent mysteries in contemporary high energy physics. In the standard
model (SM) and its supersymmetric extensions, elementary scalar doublets of the SU(2)L
interactions are responsible for the mass generation. Yet there is no explanation for the
scalar-fermion Yukawa couplings. On the other hand, if there is no light Higgs boson found
in the next generation of collider experiments, then the interactions among the longitudinal
vector bosons would become strong at a scale of O(1 TeV) and new dynamics must set
in
[1]. The fact that the top-quark mass is very close to the electroweak scale (mt v/ 2)
is rather suggestive: there may be a common origin for electroweak symmetry breaking
* Corresponding author.

E-mail address: [email protected] (G. Valencia).

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 4 3 - X

416

T. Han et al. / Nuclear Physics B 593 (2001) 415437

and top-quark mass generation. Much theoretical work has been carried out recently in
connection to the top quark and the electroweak sector [24].
Due to the Goldstone boson equivalence theorem (ET) [5], the longitudinal gauge bosons
(WL , ZL ) resemble the Goldstone modes (w , z) at energies much larger than their mass
MW and thus faithfully reflect the nature of the electroweak symmetry breaking (EWSB).
To study the EWSB sector in connection with the top quark, a sensitive probe is to produce
a top quark via longitudinal gauge boson scattering [6]
WL+ WL , ZL ZL t t,
tb.
W ZL t b,
L

(1)
(2)

These processes will receive significant enhancement if there are underlying resonances
in this sector that couple to both Goldstone bosons and the top quark. In particular, it
is interesting to note that both a scalar (Higgs-like) resonance and a vector (techni-rholike 0 ) resonance would contribute to WL+ WL t t in process (1); while only a vector
(techni-rho-like ) resonance would significantly enhance process (2).
These processes can be effectively realized at high energy lepton colliders via nearly
collinear gauge boson radiation
e+ e WL WL t t,
+

e e

WL ZL

e tb,

(3)
(4)

where e tb generically denotes e t b and e+ tb. Within the effective W -boson

approximation (EWA) [7], the t t production of Eq. (1) was first calculated in Ref. [8]
and then in Ref. [9]. In the approach of an electroweak effective Lagrangian, they were
studied in Ref. [10]. A full evaluation of the SM diagrams was performed for t t production
at a 1.5 TeV linear collider [11,12] and at a multi-TeV muon collider [13]. Effects on
WL+ WL t t from other strongly interacting dynamics were recently discussed in [14,
15].
In this paper, we carry out a comprehensive evaluation for processes (3) and (4) within
and beyond the SM. In Section 2, we first formulate the effective interactions for a stronglyinteracting electroweak sector (SEWS) including the heavy top quark (Top-SEWS). We
parameterize the sector in a (relatively) model-independent way by introducing a heavy
scalar or a heavy vector to unitarize (up to a few TeV) the universal low-energy amplitudes
[16]. We also comment on the current low-energy constraints on this sector. In Section 3,
we perform detailed numerical analyses for the Top-SEWS signal and the SM backgrounds.

We present our results at an e+ e linear collider with a center-of-mass energy s = 1.5

and also illustrate some results at a lepton collider with CM energy of 4 TeV. We find that
the future high energy lepton colliders have substantial potential to explore the Top-SEWS
sector to great precision. Section 4 contains our conclusions. Some useful formulae are
presented in Appendix A.

T. Han et al. / Nuclear Physics B 593 (2001) 415437

417

2. Effective interactions in the Top-SEWS sector

2.1. Low energy amplitudes
The low energy behavior of the scattering amplitudes for the processes VL VL q q is
determined by symmetry, and is the same in all models in which the electroweak symmetry
is broken by a strong interaction. If we parameterize the scale of this strong interaction

with the mass of its lightest resonance MR , then in the region where mW  s  MR , the
amplitudes are dominated by the low energy theorems (LET). These low energy theorems
can be obtained in a simple manner with the use of the Goldstone-boson equivalence
theorem as in Refs. [5,6].
The framework for a model-independent analysis is that of effective Lagrangians. Within
this framework, the low energy amplitudes arise from the lowest dimension operator one
can construct that respects the symmetries of the standard model. The non-renormalizable
effective Lagrangian responsible for the low energy interactions of the would-be Goldstone
bosons w , z is given by
L=

v2
Tr U U ,
4

(5)

where U = exp(iwi i /v) [16,17], with Tr i j = 2ij and v 246 GeV. The minimal
interactions between third generation fermions and the would-be Goldstone bosons are
those of the standard model in the limit MH and can be found, for example, in
Refs. [6,18]. We work in the limit mb = 0 and find (for terms with up to two w , z),


imt
mt
+ 5 )t
zt5 t +
w+ t(1 5 )b w b(1
v
2v


1
mt
+ 2 tt w+ w + zz .
2
v

L = i

(6)

As a benchmark we will consider the low energy amplitudes (LET) defined as the

dominant terms in the limits MW  ( s, mt )  (MR , 4v). Taking the scattering angle
to be that between the momentum of the w (or z) and the top quark in the center of
mass frame, we find for the neutral channels,

mt s
++

MLET (ww t t ) = MLET (ww t t ) = 2 t ,

v
2
mt
(1 t ) sin

,
M
LET (ww t t ) =
v 2 1 + t cos 2m2t /s

mt s
++

MLET(zz t t ) = MLET (zz t t ) = 2 t ,

v
m2t
sin

,
(7)
MLET(zz t t ) = 2
v 1 + t cos
where t = (1 4m2t /s)1/2 .
For the charged wz t b channel we define the angle between the momentum of the
z and the top quark in the center of mass frame, to obtain

418

T. Han et al. / Nuclear Physics B 593 (2001) 415437

3
2m
m (1 + cos )
,
= 2 t
2 (1 cos ) + 2m2 /s]
v s [m
t
2
2mt
m sin
+

.
M+
LET (w z t b) =
2 (1 cos ) + 2m2 /s]
v 2 [m
t
+

M++
LET (w z t b)

(8)

In writing these equations we took MW = mb = 0, and m = (1 m2t /s)1/2 . We present

some details in Appendix A.
The low energy amplitude for the process ww t t in Eq. (7) grows linearly with energy

and violates partial wave unitarity at an energy sww 3 TeV [19,20]. Similarly, the low
energy amplitude for the process ww ww violates unitarity around 1.2 TeV [17,21].
The energy scale at which these violations of unitarity take place can be interpreted as the
scale at which new physics must come into play. For our present purpose, the new physics
that restores unitarity will be either a scalar or a vector resonance and we discuss these two
cases in the next two sections. To maintain a model-independent discussion, we introduce
the resonances through effective (non-renormalizable) low energy Lagrangians. Therefore,
after inclusion of the resonance, the partial wave amplitudes will still violate unitarity at
higher energies. We will adjust the couplings of the resonances so that violation of unitarity
does not occur until a scale between 2 and 5 TeV. With this prescription, we describe the
phenomenology of a resonance which is the dominant dynamical feature of a strongly
interacting electroweak symmetry breaking sector (SEWS) below 2 TeV.
2.2. Scalar resonance
The interactions between the standard model gauge bosons and a generic scalar
resonance S, have been considered in Ref. [22]. The leading order effective Lagrangian
for these interactions contains two free parameters: the resonance mass MS and a coupling
constant gS that can be traded for the width of the new resonance into W and Z pairs. The
effective Lagrangian for the scalar resonance and its interactions with the w , z is given
by


1
1
1
(9)
L = S S MS2 S 2 + gs vS Tr U U
2
2
2
from which one obtains that,
3 gS2 MS3
.
(10)
32 v 2
With gS = 1, Eq. (10) reproduces the width of the standard model Higgs boson.
It is straightforward to compute the amplitudes for ww ww scattering in this model.
They are obtained from the Lagrangians, Eqs. (5) and (9). For example we find,
Sww =

A(w+ w zz) =

s s(1 gS2 ) MS2

.
v2
s MS2

(11)

All the other channels can be obtained from this one by using custodial SU(2) and crossing
symmetries [17]. From Eq. (11), we can see that the choice gS = 1 reproduces the standard
model amplitude. In this case, the amplitude takes a constant value at high energies as

T. Han et al. / Nuclear Physics B 593 (2001) 415437

419

corresponds to the renormalizable standard model. If gS 6= 1, however, the amplitude grows

with energy violating unitarity at some point. To study this issue, we construct the I = 0,
J = 0 partial wave amplitude,




Sww
s
s
0
2
1 + 2 gS 1
a0 (s) = 2
gS MS s MS2 + iSww (s/MS )
MS




2


MS
s
s
Sww
2
2
log 1 + 2
gS 1 2gS 1
.
(12)
+ 2
s
3gS MS MS2
MS
To obtain this result we have introduced a scalar-resonance width into the s-channel Higgs
propagator according to the prescription of Ref. [23],
1 + i( /MS )
1
,

2
s MS
s MS2 + i (s/MS )

(13)

which produces a well behaved amplitude both near the resonance and at high energy. 1
We first look at this result in the region above the resonance, and near the limit of validity

of the effective theory. For s  MS , Eq. (12) reduces to

a00 (s) = 1 gS2

5 Sww
s

.
16v 2 3 MS

(14)

Once again, we see that gS = 1 reproduces the standard model result, in which the unitarity
condition becomes a constraint on the Higgs boson mass [5]. The first term shows how, for
gS 6= 1, the model corresponds to a non-renormalizable effective theory. If we demand that

this term satisfy a unitarity condition |Re(a00 (s))| < 0.5 up to s = 2 (3) TeV we obtain
0.8 (0.9) 6 |gS | 6 1.2 (1.1). If we choose to consider all the effects of the resonance,
including its width according to the prescription of Ref. [23], we then find for MS = 1 TeV
that unitarity condition in the form |a00 | < 1, is satisfied up to 2 (5) TeV with
0.4 (0.9) 6 |gS | 6 1.1 (1.02).

(15)

The effective Lagrangian describing the coupling of the top quark to the scalar resonance
that we use is,
mt
(16)
L = S tt,
v
where is a new coupling that can be traded for the width of the scalar into top-quark
pairs,


3 2 m2t MS
4m2t 3/2
.
(17)
1

St t =
8 v 2
MS2
The case = 1 corresponds to the usual standard model Higgs-boson coupling to the top
quark. With Eqs. (5), (9) and (16), we have an effective theory involving a new generic
scalar resonance and its couplings to would-be Goldstone bosons and top quarks. Using
1 Both a constant width and an energy dependent width for a heavy scalar (with mass near 1 TeV) produce
undesirable features [24]. We thank S. Willenbrock for pointing out Ref. [23].

420

T. Han et al. / Nuclear Physics B 593 (2001) 415437

Eqs. (9) and (16), and the Equivalence Theorem (ET), we find the matrix element for
WL WL t t via the scalar S to be

s
mt s

(ww

t
t
)
=
M
(ww

t
t
)
=
g

t ,
(18)
M++
S
S
S
s MS2 v 2
here and henceforth, the double superscripts denote the t t helicities. For = gS = 1, it
reduces to the standard model result. We obtain identical amplitudes for zz t t. To
describe the resonance region we modify this amplitude according to the prescription in
Ref. [23]. Having chosen an SU(2) singlet, electrically neutral, scalar resonance, there is
tb in this model.
no contribution to WL ZL t b,
There are no significant, phenomenological constraints on the couplings gS and .
Following Refs. [6,19] and [25], we can demand that the process ww t t satisfy inelastic
partial-wave unitarity. By choosing the I = 0, J = 0, color singlet channel as in Ref. [20],
we find,



 GF mt s  MS2 + s(gS 1) 

 < 0.5.
3
(19)




s MS2
8 2
In analogy with our treatment of Eq. (12), we first concentrate in the region above the
resonance. The first factor in Eq. (19) corresponds to the low energy theorem, and it violates

unitarity at s 3 TeV. We can have the full amplitude satisfy unitarity up to 3 TeV by
requiring the second factor to satisfy




gS s
1 < 1
(20)

2
sM
S

or, for energies far above the resonance, 0 < gS < 2.

If, on the other hand, we introduce the width according to the prescription of Ref. [23]

and consider the full amplitude for MS = 1 TeV, unitarity is satisfied up to s = 2 (5) TeV
if
0 (0.4) < gs < 1.9 (1.6).

(21)

Finally, we consider t t t t scattering at energies above the resonance (in the same
helicity, color singlet channel). The J = 0 partial wave is,
3 GF m2t 2
(22)
.
4
2
If we follow Ref. [20] and require that |Re a0 (t t t t)| < 0.5, we obtain the unitarity
constraint
a0 (t t t t ) =

|| < 2.9.

(23)

As we will see in our numerical studies in Fig. 8, this constraint is not very restrictive.
2.3. Vector resonance
The interactions of standard model gauge bosons with new vector resonances have been
described in the literature [26,27]. In the notation of Ref. [26], two new parameters are

T. Han et al. / Nuclear Physics B 593 (2001) 415437

421

introduced a and g (these correspond to and g 00 /2 from Ref. [27], respectively). The
effective Lagrangian consists of a kinetic term for the triplet of vector resonances
i i /2 and an interaction term of the form





1
+ + i2g
,
(24)
L = av 2 Tr + + i2g
4
where = exp(iwi i /2v).
The two new constants a, g are related to the mass and width of the vector particles
via the relations,

a 2 g 2
M .
0 w+ w =
(25)
192
In addition, when the electroweak gauge bosons are added in a gauge invariant manner, they
mix with the new vector bosons [27]. In the charged sector, the physical states become,
M2 = av 2 g 2 ,

= cos + W sin ,
Vph

Wph
= W cos sin

(26)

with a mixing angle given by

tan 2 =

g
4ag g
,
(1 + a)g 2 4a g 2
g

(27)

the last expression being the limit as M . Similar expressions for the mixing in
the neutral sector can be found in the literature [27]. For our numerical studies we use
couplings a, g that have been studied in detail by the authors of the BESS model, and
that respect partial wave unitarity below a few TeV. For example, for M = 1 TeV and
= 30 GeV, the partial wave a00 for ww ww scattering satisfies |a00 | < 1 up to about
2.6 TeV, while |a11| < 1 is satisfied up to about 3 TeV with the simple prescription of using
a constant width in the s-channel propagator.
We are currently interested in effective couplings between the new vector mesons and
the top and bottom quarks. These couplings may arise either from the ordinary quark
couplings to gauge bosons through the mixing of Eq. (26), or from novel direct couplings.
The couplings induced by mixing are universal, common to the three generations, and can
be bound by low energy experiments. More interesting is the possibility of a direct coupling
to the third generation. We write these effective couplings in a generic form
(gV + gA 5 ) i i ,
Leff =

(28)

where is the third generation quark doublet (t b).

Unlike the case of the scalar resonance, there exist low energy constraints on the direct
couplings of new vector resonances to the third generation quarks because they modify
their couplings to the W and Z. These deviations can be parameterized by the low energy
effective Lagrangian

g 
Leff = (1 + L )tL bL + R tR bR W+ + h.c.
2


g
(29)
(Lt + Lt )tL tL + (Rt + Rt )tR tR Z .

2 cos W

422

T. Han et al. / Nuclear Physics B 593 (2001) 415437

Fig. 1. W V mixing induced anomalous tbW coupling.

For simplicity, 2 let us consider the case of anomalous W couplings as shown in the
diagram in Fig. 1. We find in the limit of large vector resonance mass that
2
L = (gV gA ),
g

2
R = (gV + gA ).
g

(30)

There are several constraints on these couplings. The measurement of the rate for b
s restricts |R | . 0.004 [28,29] but does not place significant constraints on L . An
analysis of precision measurements at LEP [30] (with updated data from Ref. [31]) results
in |L | . 0.1 excluding the parameter T . 3 From future experiments, it is expected that
a B factory can place significant further constraints of order |R | . 0.001, |L | . 0.03
[32]; and that single top production can yield |L | . 0.05 [33]. From all these numbers
we conclude that a right-handed coupling is severely constrained and for the rest of this
paper will concentrate on left-handed couplings exclusively. That is, we choose to satisfy
the relation
gA = gV ,

(31)

and, in keeping with the current bounds, we will concentrate on the case

gV . 0.03 g.

(32)

After these bounds are imposed, partial wave unitarity does not place additional
constraints.
When new direct couplings such as those in Eq. (28) are introduced in a gauge invariant
manner, as is done in the BESS model [27] (through their parameter b), the effective
couplings gV ,A of Eq. (28) receive contributions from both the direct couplings and
mixing. In this general case Eq. (28) is not literally correct, as the charged and neutral
couplings differ. We shall employ Eq. (28) for simplicity, assuming that the direct couplings
dominate. For example, in the heavy vector limit of the BESS model, they take the form
2 The anomalous Z-boson couplings are obtained in a similar manner, but there is the additional complication
of both 0 and Z mixing with the photon.
3 If the parameter T is included, the bound could be stronger in certain cases. In our case, however, the induced
anomalous couplings are of the form L = Lt , and are not restricted by T where they appear in the combination
L Lt [30].

T. Han et al. / Nuclear Physics B 593 (2001) 415437

gV = gA =

g b
.
4 1+b

423

(33)

Considering both the charged and neutral vectors, this results in

L = Lt =

1 b
21+b

and R = Rt = 0.

(34)

The bound |L | . 0.1 then corresponds to |b| . 0.25. In this respect, it is expected that
a future linear collider studying the reaction e+ e W + W can probe values of b some
ten times smaller than this [34]. Our present study is complementary in the sense that we
wish to observe the direct signal for a vector state production at high energies.
The couplings gV ,A can be traded in favor of the physical partial decay widths via the
relations
t t
t b


1/2 




M
m2t
m2t
m2t
2
2
=
14 2
gV 1 + 2 2 + gA 1 + 4 2 ,
4
M
M
M




2


M
m2
m2
gV 2 + gA 2 1 t2
=
2 + t2 .
4
M
M

(35)
(36)

bb is the same as Eq. (35) replacing mt by mb . In our calculations, we consistently take

MW = mb = 0.
In combining the couplings to WL in Eq. (24) with the couplings in Eq. (28), we obtain
the scattering amplitudes
+

+
M++
(w w t t ) = M (w w t t )

mt s
cos ,
= a gg
V
s M2 + iM
+
M
(w w t t ) =

a g
s sin
(gV gA t )
2 s M2 + iM

(37)

and

mt s
a g
=
m
2 s M2 + iM


(gV + gA )(1 + cos ) + (gV gA ) ,
s
a g
+

m (gV gA ) sin .
M
(w z t b ) =
2 s M2 + iM

M
(w z t b )

(38)

Unlike the case of the scalar resonance, these amplitudes do not grow with energy. For this
reason, partial wave unitarity does not place any significant constraints. Also, unlike the
case of a scalar resonance, the heavy vector resonance is narrow and a simple treatment of
the constant width as above is adequate.

424

T. Han et al. / Nuclear Physics B 593 (2001) 415437

3. Sensitivity to the Top-SEWS signal at future lepton colliders

3.1. Effective W approximation
The effective W -boson approximation (EWA) provides a viable simplification for high
energy processes involving W -boson scattering [7]. It is particularly suitable for the
calculations under consideration since the W -boson level evaluation for the SEWS signal at
high energies is adequate. The condition for the EWA to be valid is that the invariant energy
scales are all much larger than the W mass, namely s, |t|, |u|  MW . Besides, it is more
accurate for a longitudinal W -boson due to the fact that the distribution function is largely
independent of the W beam energies. To justify the EWA in our calculation, we compare
our signal evaluation for a heavy scalar resonance with a full tree-level SM calculation [35].
We define the SM signal for a heavy Higgs boson S following the subtraction procedure
[22]
S (MH ) = (MH ) (MH = 100 GeV).

(39)

The interpretation of this procedure is that the signal from a heavy Higgs boson should
be defined as the enhancement over the perturbative SM expectation with a light (MH
100 GeV) Higgs boson, which should be viewed as the irreducible SM background.
The process of e+ e Zt t followed by Z leads to the same final state as our
signal. It arises from different physics and needs to be removed from our consideration. We
follow the earlier studies [36] to introduce a recoil mass against the observable t t system
defined as

2
= s + m2t t 2 s (Et + Et),
(40)
Mrec
where Et (Et) is the top (anti-top) quark energy in the e+ e CM frame. The recoil-mass
spectrum peaks at MZ for the background. Here and henceforth, we will impose Mrec >
200 GeV to effectively remove the Z background. We will adopt the same cut for
the e tb process as well.
In Fig. 2, we present the production cross section for e+ e W + W tt versus
) on the top quark pair for MH at (a) e+ e CM energy
an invariant mass cut (mcut
t t

s = 1.5 TeV and (b) s = 4 TeV. The solid curves are from the full SM calculation
based on the subtraction scheme Eq. (39). The dotted curves are the Low Energy Theorem
(LET) result for w+ w t t by EWA. The two calculations agree very well, especially
for higher mt t values.
In Fig. 3, we show the cross section for the same process, but versus the scattering angle
cut (cos cut
t ) on the final state top quarks. In Fig. 3(a), we compare the scalar resonance
model in Section 2.2 for MS = 1 TeV (dashed curve) with the SM for MH = 1 TeV (solid
curve) based on the subtraction scheme Eq. (39). In both cases and henceforth, to preserve
unitarity near the heavy scalar resonance, we adopt the prescription given in Eq. (13).
Fig. 3(b) compares the case for LET and the full SM calculation for MH . We see
that the two calculations agree better for central scattering at low cos t . These two figures
motivate us to introduce the basic acceptance cuts,
mt t > 500 GeV,

| cos t | < 0.8.

(41)

T. Han et al. / Nuclear Physics B 593 (2001) 415437

425

+
Fig. 2. Production cross sections (in fb) for e+
tt versus the invariant mass
e W W
cut on the top quark pair for MH at (a) s = 1.5 TeV and (b) s = 4 TeV.

+ W
Fig. 3. Production cross sections (in fb) for e+ e W

tt at s = 1.5 TeV versus

the polar angle cut on the final state t and t (a) for MS = 1 TeV and (b) for MH .

These cuts not only guarantee the good agreement between the full SM calculation and the
EWA that we will adopt for signal calculations beyond the SM, but are also necessary to
enhance the signal over the SM backgrounds as we will discuss later. We impose a similar
requirement for the tb process.

426

T. Han et al. / Nuclear Physics B 593 (2001) 415437

+ W tt versus the CM energy for

Fig. 4. Production cross sections (in fb) for e+ e W

(a) MS = 1 TeV and (b) MH .

Finally we show the comparison of the two calculations versus the center of mass (CM)
energy in Fig. 4 for (a) a 1 TeV scalar resonance and (b) MH , where the basic cuts of
Eq. (41) have been imposed on both t and t. The agreement is generally at order of 50% or
better for MS = 1 TeV and almost perfect for MH . We consider this to have justified
our future use of the EWA in the signal calculations beyond the SM. However, we note that
the EWA calculations for signal processes (1) and (2) do not properly incorporate the full
kinematics of processes (3) and (4). We thus simulate the full kinematics and determine
the cut efficiencies using the complete SM calculations in the next section.
3.2. Top-SEWS signal and SM backgrounds
For the SM-like heavy Higgs model adopted in the last section, the Higgs boson mass
MH is the only parameter. We would like to explore more general classes of strongly
interacting electroweak models (SEWS) as outlined in Section 2. For a generic scalar
singlet model, there are three input parameters: MS , gS , and as described in Section 2.2.
The two couplings can be traded by the physical partial widths and they are subject to the
unitarity constraints Eqs. (15), (21), (23). In terms of the partial widths for MS = 1 TeV,
these constraints lead to
80 GeV . Sww . 600 GeV,
Sww St t . (300 GeV)2 ,
St t . 420 GeV,

where we required that partial wave unitarity be satisfied up to sww = 2 TeV.

(42)

T. Han et al. / Nuclear Physics B 593 (2001) 415437

427

The three input parameters for the vector model are gV (= gA ), g,

and a. These can
be traded by the physical parameters M , ww (or wz ), and t t (or t b ). From the
constraint of Eq. (32), we get for a neutral 0 with mass M = 1 TeV,
t tww . (8 GeV)2

(43)

and for a charged of the same mass,

t b wz . (11 GeV)2 .

(44)

For the WL WL t t signal, the SM backgrounds are

e+ e e+ e t t,
+

e e

(45)

W W t t.

(46)

Process (45) mainly comes from t t and has a large cross section, typically about

7.5 fb at s = 1.5 TeV. The charged-current process Eq. (46) occurs at a lower rate, about

1.7 fb at s = 1.5 TeV, but it is kinematically more difficult to separate from the signal.
Following earlier studies [36,37], we demand a high transverse momentum for the final
state t and t
pT (t) > 150 GeV,

(47)

which is motivated by the Jacobian peak in the pT spectrum from a heavy particle decay.
We also require a moderate momentum for the t t pair
30 GeV < pT (t t ) < 300 GeV.

(48)

This cut forces the final state leptons to be away from the forwardbackward collinear
region. Since there are no energetic electrons in the central region for the signal process,
we then veto events with final state electrons at large angle:
Ee > 50 GeV,

| cos e | < cos(0.15 rad).

(49)

We find that these cuts, along with the basic cuts (41), are very effective to suppress the SM
backgrounds. The background (45) is largely eliminated by the combination of pT (t t ) cut
and electron veto. The effects of the cut on the charged current t t process are summarized
in Table 1, again using the SM-like heavy Higgs model as the representative for the signal.
From this table, we find the cut efficiencies of Eqs. (47), (48), (49) for the signal based on
the subtraction scheme (39). They are

 = 83% for s = 1.5 TeV,

(50)
 = 84% for s = 4 TeV.
For the remainder of this paper we use these cut efficiency figures for all the signal
calculations within the EWA in the WL WL t t channel.
For the process WL ZL tb, the irreducible background is from
e+ e e W Z e tb.

(51)

where
Again, the largest contribution comes from the photon induced process e t b,
the other electron goes along the beam direction after emitting a collinear photon. For this

428

T. Han et al. / Nuclear Physics B 593 (2001) 415437

Table 1
Cross sections of process (46) for a SM-like heavy Higgs (1 TeV) and a light Higgs (0.1 TeV) with
different cuts. This gives the cut efficiencies for Top-SEWS signals as in Eq. (50)

s = 1.5 TeV

MH = 1 TeV

Background MH = 0.1 TeV

1.41 fb
1.13 fb

0.21 fb
0.14 fb

MH = 1 TeV

Background MH = 0.1 TeV

8.86 fb
7.20 fb

0.98 fb
0.55 fb

cuts (41)
cuts (41), (47), (48), (49)

s = 4 TeV
cuts (41)
cuts (41), (47), (48), (49)

case, we impose the cuts (41) and (47). Since there is a final state electron in the signal
process with a typical transverse momentum of MZ /2, we require an electron tagging at a
finite angle
| cos e | < cos(0.15 rad).

(52)

This electron tagging is very effective to remove the SM background, and more so at higher
energies. The combination of the above cuts significantly reduces the e process and brings
the background to a level below the signal rate.
To determine the signal efficiency of the electron tagging, we calculate the process
e+ e e + .

(53)

We find

s = 1.5 TeV,

 = 61% for s = 4 TeV,

 = 93% for

(54)

which will be used for other signal calculations in the WL WL tb channel based on the
EWA. The lower tagging efficiency at 4 TeV is due to the fact that the final state electron
becomes more forward at higher energies.
We now present the production cross sections for the signal and background versus the

+ W tt process with the

CM energy s after the cuts. Fig. 5(a) is for e+ e W

full cuts (41), (47), (48), (49) for the scalar model (dots), vector model (dashes), LET (dotdash), and the irreducible background MH = 0.1 TeV (solid). For illustration, we have
taken the model parameters to be
MS = 1 TeV,

Sww = 493 GeV,

St t = 50 GeV,

M = 1 TeV,

ww = 50 GeV,

t t = 1.3 GeV.

(55)

Fig. 5(b) shows the production cross sections for e+ e e W Z e tb, with the cuts
(41), (47), (52) for the vector model (dots), LET (dashes), and the SM background (solid).
The model parameters are taken as

T. Han et al. / Nuclear Physics B 593 (2001) 415437

429

+ W tt versus s for
Fig. 5. Production cross sections (in fb) for (a) e+ e W

MH = 0.1 TeV (solid), LET (dot-dash), scalar singlet model (dots), and vector model (dashes),

and for (b) e+ e e W Z e tb versus s for the vector model (dots), LET (dashes) and SM
background (solid) with the full cuts imposed. We have used the model parameters as in Eqs. (55)
and (56).

M = 1 TeV,

ww = 50 GeV,

t b = 2.5 GeV.

(56)

We see from Fig. 5 that our cuts effectively suppress the SM background below the signal
for both channels. The signal rate at a 1.5 TeV linear collider is typically between 0.5 and
1.5 fb.
3.3. Results and discussion
For the remainder of our analysis, we consider the top quark to decay hadronically t
bW bjj 0 with a branching fraction approximately 70%. In doing so, we assume that we
can fully reconstruct the t and t events, and that there are no isolated central electrons from
the decay to confuse our electron vetoing and tagging requirements. We also assume a data

sample of 200 fb1 at s = 1.5 TeV. We can see from Fig. 5 that we may reach a clear
signal observation for both a scalar and a vector resonance in the t t channel. Although
the LET channel has relatively lower rate, it may nevertheless lead to about 35 events
after including the branching fraction, which is well above the background expectation.
For the tb channel, both signal and background rates are slightly smaller than the t t case,
but the vector signal rate is still sufficiently above background to be observable. The LET
amplitude in this case is below background as was already seen in the model discussions.
It is important to know how well one can reconstruct the signal and contrast it with
the background beyond simple event counting. In Fig. 6 we show the distributions for the

signal and background versus t t invariant mass (mt t ) for (a) s = 1.5 TeV and (b) s =
4 TeV with the SM background represented by MH = 0.1 TeV (dot-dash), LET signal

430

T. Han et al. / Nuclear Physics B 593 (2001) 415437

Fig.
differential cross sections versus t t invariant mass for (a)
6. Signal and background

s = 1.5 TeV and (b) s = 4 TeV: SM background by MH = 0.1 TeV (dot-dash), LET (dots),
scalar model (dashes), and vector model (solid). We have used the model parameters as in Eq. (55).
All cuts are imposed and the branching fraction for t bjj 0 is included.

(dots), scalar model (dashes), and vector model (solid). The model parameters are taken
as in Eq. (55). All cuts are imposed and the branching fraction for t t decay is included.
We see a broad resonance around MS = 1 TeV in scalar model due to its large width. For
the vector case, however, the width is narrower in general and the signal peaks above the
scalar model for M = 1 TeV in a distinctive manner. The LET model has relatively low
rate and has no particular structure. In Fig. 7 we show the same signal and background
distributions, but for the tb channel versus the invariant mass (mt b ). The model parameters
are taken as in Eq. (56). Clearly, the observation of a signal in Fig. 7 would provide decisive
information to discriminate a scalar resonance from a vector. The authors of Ref. [15]
use, instead, the top-quark polarization information to distinguish a scalar from a vector
resonance.
We next consider the extent to which one can probe the couplings for different models.
Fig. 8(a) shows the accuracy contours to determine the partial decay widths (thus the

couplings) in the scalar resonance model with MS = 1 TeV at s = 1.5 TeV and a
luminosity of 200 fb1 . The statistical accuracy for the cross section measurement is
defined as

S +B
,
R =
S
where S and B are the number of signal and background events respectively and we have
ignored the experimental systematic effects. We find that partial widths can be probed up
to about 10% accuracy for St t 50 GeV within the unitarity bounds shown as dotted

T. Han et al. / Nuclear Physics B 593 (2001) 415437

431

Fig.
cross sections (in fb) versus tb invariant mass (mt b ) for (a)
7. Signal and background

s = 1.5 TeV and (b) s = 4 TeV: SM (dots) and vector model (solid). We have used the model
parameters as in Eq. (56). All cuts are imposed and the branching fraction for t bjj 0 is included.

Fig. 8. At a linear collider with s = 1.5 TeV and a luminosity of 200 fb1 , (a) accuracy contours
of determination of the partial widths for the scalar resonance model with MS = 1 TeV; (b) statistical
significance contours to distinguish a 1 TeV scalar from LET. Dotted lines are unitarity constraints
given in Eq. (42).

lines (Eq. (42)). In the region above the curve labeled R = 8%, the partial widths can be
determined to 8% or better accuracy.
Also shown in Fig. 8(b) are the statistical significance contours to distinguish a 1 TeV
scalar signal from the LET amplitudes. Here we define an appropriate standard deviation

432

T. Han et al. / Nuclear Physics B 593 (2001) 415437

as
=

{ (St t, Sww ) (0, 0)}L

,
(0, 0)L

where (St t, Sww ) is the cross section in the scalar model for the given widths, (0, 0)
is the non-resonance cross section with gS = = 0, and L is the luminosity. It is easy
to see that the scalar model is distinguishable at the 5 level from the non-resonant case
within the unitarity limits given in Eq. (42).
We show the same analysis for the vector model in Fig. 9, where we present the

sensitivity to the vector resonance model with M = 1 TeV at s = 1.5 TeV and a
luminosity of 200 fb1 , for (a) 13% accuracy determination of the partial widths for
W + W 0 t t and (b) 20% accuracy determination of the partial widths for W Z
t b (tb). We found in Fig. 9(a) that a 13% accuracy determination of the partial
widths is possible for the neutral vector process in the region allowed by the current bounds
given in Eq. (43). For the charged process we have a smaller signal rate, nevertheless, we
found in Fig. 9(b) that we can still determine the partial widths to 20% accuracy in the
region allowed by current constraints, Eq. (44).
Before concluding, several remarks comparing our work to that of Ref. [15] are in
order. We have taken a purely phenomenological approach to the couplings of the new
resonances. After a general parameterization, we restrict our numerical analysis to those
values that are allowed by current phenomenological and unitarity constraints. Ref. [15],
on the other hand, explores couplings that arise from consideration of technicolor and topcolor ideas for a vector (TC ) and a scalar (Ht ), respectively. Our phenomenological study

Fig. 9. At a linear collider with s = 1.5 TeV and a luminosity of 200 fb1 , determination of
the partial widths for the vector resonance model with M = 1 TeV (a) 13% accuracy contour for
tb). Dotted lines
W + W 0 t t channel; (b) 20% accuracy contour for W Z t b(
are from the current low energy constraints in Eqs. (43) and (44).

T. Han et al. / Nuclear Physics B 593 (2001) 415437

433

goes beyond Ref. [15] in that we explore the effect of using the effective W approximation
by comparing suitable processes to the exact calculations. We also discuss in some detail
the background to the signals and how to reduce it. Our conclusions may be somewhat
optimistic in that, unlike Ref. [15], we do not consider the effects of initial state radiation
and beamstrahlung. These effects, and other detector specific issues are clearly important
and should be addressed when doing detailed simulations for specific experiments. Finally,
we have demonstrated how to distinguish a vector resonance from a scalar by comparing
the two processes WL ZL t b (tb) and WL WL t t, an idea complementary to that of
Ref. [15] which hopes to make use of the top-quark polarization.

4. Conclusion
We have studied the couplings of the top quark to TeV resonances that occur in models in
which the electroweak symmetry is broken by a strong interaction. Our study is motivated
by the fact that in many models the top quark plays a special role in the breaking of
electroweak symmetry.
We have explored the possibility that the new physics is realized as a scalar resonance
or a vector resonance near 1 TeV, as well as the case where the resonance is beyond reach
(LET). For the scalar model, we used partial wave unitarity to constrain the parameters as
shown in Section 2.2. For the vector model, we used current experimental data to constrain
the parameter space as presented in Section 2.3. We found within these constraints,
that signal rates in these models at future lepton colliders can be well above the SM
backgrounds after judicial cuts are applied. The models can be distinguished from each
other by systematically studying two different final states. Typically one would see a broad
scalar resonance S, or a relatively narrow vector resonance 0 in the t t channel and a
narrow vector resonance in the tb channel. In particular, we illustrated our results at a
1.5 TeV linear collider with an integrated luminosity of 200 fb1 . One expects to observe
about 120 events for either S t t or 0, t t, tb including final state identification.
Even if the lower-lying resonances are inaccessible at the collider, one can still observe
a statistically significant deviation as predicted by the LET for the t t final state from the
perturbative SM expectation. Finally, the leading partial decay widths, which characterize
the coupling strengths, can be statistically determined to about 10% level.

Acknowledgements
The work of T.H. and Y.J.K. was supported in part by the US DOE under contract No.
DE-FG02-95ER40896 and in part by the Wisconsin Alumni Research Foundation. The
work of A.L. was supported in part by the RFBR grants 99-02-16558 and 00-15-96645. The
work of G.V. was supported in part by DOE under contact number DE-FG02-92ER40730.
G.V. thanks the Special Research Centre for the Subatomic Structure of Matter at the
University of Adelaide for their hospitality and partial support while part of this work
was completed.

434

T. Han et al. / Nuclear Physics B 593 (2001) 415437

Appendix A. SM amplitudes using the equivalence theorem

In this appendix, we present the standard model amplitudes obtained using the
From these
Equivalence Theorem for the processes w w+ t t, zz t t, and zw+ t b.
amplitudes one can derive the (LET) amplitudes by taking the large Higgs mass limit,
ignoring the electroweak couplings, and setting mb = 0.
For the process w w+ t t:

MH 2 mt s

t ,
MH (ww t t ) =
(s MH 2 ) v 2

M
H (ww t t ) = 0,

M
b (ww t t ) =

(A.1)

2m3
2 t
v s

t + cos
,
1 + t cos 2m2t /s

m2t
(1 t ) sin
,
(A.2)
2
v 1 + t cos 2m2t /s
mt
2
M
(ww t t ) = 2g xW Q cos ,
s

2
(A.3)
M (ww t t ) = g xW Q sin ,


mt
2 1

M
Z (ww t t ) = 2gV gZ 2 xW cos ,
s


1

2
xW sin ,
(A.4)
MZ (ww t t ) = (gV gA t )gZ
2
where

1/2
g
, xW = sin2 W , gZ =
,
t = 1 4m2t /s
cos W
1
1
gV = T3 QxW , and gA = T3 .
2
2
2  s and ignoring the terms of order g 2 , we find to leading order,
Taking the limit MH

mt s

(ww

t
t
)
=

t ,
LET: M
LET
v2
m2t
(1 t ) sin

.
(ww

t
t
)
=
M
LET
2
v 1 + t cos 2m2t /s

M
b (ww t t ) =

For the process zz t t:

M
H (zz t t )

mt s
=
t ,
(s MH 2 ) v 2
MH 2

M
H (zz t t ) = 0,
M
top (zz t t ) =

(A.5)

2m3
2 t
v s

M
top (zz t t ) =

t + cos
,
1 + t cos

m2t
sin
,
2
v 1 + t cos

(A.6)

T. Han et al. / Nuclear Physics B 593 (2001) 415437

mt s
t ,
v2
m2t
sin

.
(zz

t
t
)
=

M
LET
2
v 1 + t cos

435

LET: M
LET (zz t t ) =

(A.7)

Again we keep only the leading order terms and ignore the top-quark exchange diagram

for M
LET (zz t t ).

For the process zw+ t b:

2 2 mt
++
+

g m cos ,
MW + (zw t b ) =
4
s
+
+
+

= 0,
+
(zw

t
b
)
=
M
(zw
t b)
M
W
W+

= 2 g 2 m sin ,
(zw+ t b)
(A.8)
M+
W+
4
where m = (1 m2t /s)1/2 .

3
m (1 + cos )
2mt
,

2
2
v s [m (1 cos ) + 2m2t /s]
+
+
+

M
top (zw t b) = Mt op (zw t b) = 0,
2
2mt
m sin
+

,
M+
top (zw t b ) =
2
2
v
[m (1 cos ) + 2m2t /s]
+
M++
top (zw

=
t b)

is proportional to mb and is neglected.

where Mb (zw+ t b)
3
m (1 + cos )
++
+
= 2m
,
LET: MLET (zw t b)
t 2
2
v s [m (1 cos ) + 2m2t /s]
2
2m
m sin
+
+

.
MLET (zw t b) = 2 t
2
v
[m (1 cos ) + 2m2t /s]

References
[1] For a review, see, e.g., M. Chanowitz, Ann. Rev. Nucl. Part. Sci. 38 (1988) 323.
[2] C. Hill, Phys. Lett. B 266 (1991) 419;
C. Hill, Phys. Lett. B 345 (1995) 483;
E. Eichten, K. Lane, Phys. Lett. B 352 (1995) 382.
[3] B. Dobrescu, C. Hill, Phys. Rev. Lett. 81 (1998) 2634;
R.S. Chivukula, B. Dobrescu, H. Georgi, C. Hill, Phys. Rev. D 59 (1999) 075003;
G. Burdman, N. Evans, Phys. Rev. D 59 (1999) 115005;
H.-J. He, T. Tait, C.-P. Yuan, Phys. Rev. D 62 (2000) 011702R.
[4] E.H. Simmons, hep-ph/9908488.
[5] B.W. Lee, C. Quigg, H.B. Thacker, Phys. Rev. D 16 (1977) 1519;
M.S. Chanowitz, M.K. Gaillard, Nucl. Phys. B 261 (1985) 379;
Y.P. Yao, C.P. Yuan, Phys. Rev. D 38 (1988) 2237;
J. Bagger, C. Schmidt, Phys. Rev. D 41 (1990) 264;

(A.9)

(A.10)

436

[6]
[7]

[8]
[9]
[10]

[11]

[12]
[13]
[14]
[15]
[16]
[17]

[18]
[19]
[20]
[21]
[22]

[23]
[24]
[25]
[26]
[27]

T. Han et al. / Nuclear Physics B 593 (2001) 415437

H. Veltman, Phys. Rev. D 41 (1990) 2294;

H.-J. He, Y.-P. Kuang, X. Li, Phys. Rev. Lett. 69 (1992) 2619;
H.-J. He, W.B. Kilgore, Phys. Rev. D 55 (1997) 1515.
M. Chanowitz, M. Furman, I. Hinchliffe, Nucl. Phys. B 153 (1979) 402.
M. Chanowitz, M.K. Gaillard, Phys. Lett. B 142 (1984) 85;
G. Kane, W. Repko, W. Rolick, Phys. Lett. B 148 (1984) 367;
S. Dawson, Nucl. Phys. B 249 (1985) 42.
R.P. Kauffman, Phys. Rev. D 41 (1990) 3343.
M. Gintner, S. Godfrey, in: Proc. of New Directions for High-Energy Physics, June 25July 12,
1996, Snowmass, 1996, p. 824.
T. Lee, Phys. Lett. B 315 (1993) 392;
F. Larios, E. Malkawi, C.P. Yuan, Acta Phys. Polon. B 27 (1996) 3741;
F. Larios, C.-P. Yuan, Phys. Rev. D 55 (1997) 7418;
J. Wudka, Talk given at the 5th Int. Conf. on Physics Beyond the Standard Model, Balholm,
Norway, 29 April4 May 1997, hep-ph/9706434.
T. Barklow, in: Proc. of New Directions for High-Energy Physics, June 25July 12, 1996,
Snowmass, 1996, p. 819;
T. Han, hep-ph/9910495.
F. Larios, T. Tait, C.-P. Yuan, Phys. Rev. D 57 (1998) 3106.
V. Barger, M. Berger, J. Gunion, T. Han, Talk given at ITP Conf. on Future High-energy
Colliders, Santa Barbara, CA, 2125 October 1996, hep-ph/9704290.
T. Barklow et al., in: Proc. of New Directions for High-Energy Physics, June 25July 12, 1996,
Snowmass, 1996, p. 735.
E.R. Morales, M.E. Peskin, hep-ph/9909383.
M.S. Chanowitz, M. Golden, H. Georgi, Phys. Rev. Lett. 57 (1986) 2344;
M.S. Chanowitz, M. Golden, H. Georgi, Phys. Rev. D 35 (1987) 1149.
A. Dobado, M. Herrero, Phys. Lett. B 228 (1989) 495;
A. Dobado, M. Herrero, Phys. Lett. B 233 (1989) 505;
J. Donoghue, C. Ramirez, Phys. Lett. B 234 (1990) 361;
S. Dawson, G. Valencia, Nucl. Phys. B 348 (1991) 23;
S. Dawson, G. Valencia, Nucl. Phys. B 352 (1991) 27.
R.D. Peccei, S. Peris, X. Zhang, Nucl. Phys. B 349 (1991) 305.
T. Appelquist, M. Chanowitz, Phys. Rev. Lett. 59 (1987) 2405.
W. Marciano, G. Valencia, S. Willenbrock, Phys. Rev. D 40 (1989) 1725.
J. Bagger, S. Dawson, G. Valencia, Nucl. Phys. B 399 (1993) 364.
J. Bagger, V. Barger, K. Cheung, J. Gunion, T. Han, G. Ladinsky, R. Rosenfeld, C.-P. Yuan,
Phys. Rev. D 49 (1994) 1246;
J. Bagger, V. Barger, K. Cheung, J. Gunion, T. Han, G. Ladinsky, R. Rosenfeld, C.-P. Yuan,
Phys. Rev. D 52 (1995) 3878.
M.H. Seymour, Phys. Lett. B 354 (1995) 409.
G. Valencia, S. Willenbrock, Phys. Rev. D 42 (1990) 853;
G. Valencia, S. Willenbrock, Phys. Rev. D 46 (1992) 2247.
S. Jager, S. Willenbrock, Phys. Lett. B 435 (1998) 139;
R.S. Chivukula, Phys. Lett. B 439 (1998) 389.
J. Bagger, T. Han, R. Rosenfeld, FERMILAB-CONF-90-253-T, published in Snowmass
Summer Study, 1990, p. 208.
R. Casalbuoni et al., Phys. Lett. B 155 (1985) 95;
R. Casalbuoni et al., Nucl. Phys. B 282 (1987) 235;
R. Casalbuoni et al., Nucl. Phys. B 310 (1988) 181;
R. Casalbuoni et al., Phys. Lett. B 249 (1990) 130;
R. Casalbuoni et al., Phys. Lett. B 253 (1991) 275.

T. Han et al. / Nuclear Physics B 593 (2001) 415437

[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]

437

K. Fujikawa, A. Yamada, Phys. Rev. D 49 (1994) 5890.

F. Larios, M.A. Perez, C.P. Yuan, Phys. Lett. B 457 (1999) 334.
S. Dawson, G. Valencia, Phys. Rev. D 53 (1721) 1996.
C. Caso et al., Eur. Phys. J. C 3 (1998) 1.
A.A. El-Hady, G. Valencia, Phys. Lett. B 414 (1997) 173.
T. Tait, C.P. Yuan, hep-ph/9710372.
D. Dominici, Riv. Nuovo Cimento 20 (1997) 1, hep-ph/9711385.
For the full SM calculations, we have made use of the Madgraph package by T. Stelzer, W.F.
Long, Comput. Phys. Commun. 81 (1994) 357.
[36] V. Barger, K. Cheung, T. Han, R.J.N. Phillips, Phys. Rev. D 52 (1995) 3815;
V. Barger, M. Berger, J. Gunion, T. Han, Phys. Rev. D 55 (1997) 142.
[37] V. Barger, K. Cheung, T. Han, R.J.N. Phillips, Phys. Rev. D 42 (1990) 3052.

Nuclear Physics B 593 (2001) 438450

www.elsevier.nl/locate/npe

QCD dipole model and kT factorization

A. Bialas a,b , H. Navelet a , R. Peschanski a,
a Service de Physique Thorique, CEA-Saclay, Gif-sur-Yvette cedex, France
b M. Smoluchowski Institute of Physics, Jagellonian University, Institute of Nuclear Physics, Cracow, Poland

Received 22 September 2000; accepted 25 October 2000

Abstract
It is shown that the colour dipole approach to hard scattering at high energy is fully compatible
with kT factorization at the leading logarithm approximation (in log xBj ). The relations between
the dipole amplitudes and unintegrated diagonal and non-diagonal gluon distributions are given. It
is also shown that including the exact gluon kinematics in the kT factorization formula destroys the
conservation of transverse position vectors and thus is incompatible with the dipole model for both
elastic and diffractive amplitudes. 2001 Elsevier Science B.V. All rights reserved.

1. Introduction
An attractive feature of the colour dipole approach to high-energy interactions [13] is
that it provides a very simple physical interpretation of the physics at small values of the
Bjorken variable xBj . As it can also be justified in the framework of perturbative QCD in
the leading logarithm approximation [4] (following the BFKL approach [5,6]), it represents an interesting possibility for the description of hard scattering at small xBj . Indeed,
even in its simplest version, the colour dipole model turned out to be rather successful in
describing the total [7] and diffractive [8] virtual photonnucleon cross-sections. Its generalizations are now commonly used for the parametrization of data from HERA and the
Tevatron [912].
The basic ingredients of the model are the dipoledipole cross-section and the
distribution of dipoles in the projectile and in the target. 1 It is formulated in transverse
(configuration) space, taking advantage of the fact that at very high energy the transverse
positions of the colliding objects are to a good approximation conserved during
* Corresponding author.

E-mail addresses: [email protected] (A. Bialas), [email protected] (H. Navelet),

[email protected] (R. Peschanski).
1 Thus one deals only with colourless objects which gives a certain theoretical advantage in the extrapolation
to the non-perturbative region.
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 4 0 - 4

A. Bialas et al. / Nuclear Physics B 593 (2001) 438450

439

the collision. This is, in fact, the origin of the simplicity of the dipole model approach to
high-energy scattering.
It appears important to confront these attractive features of the dipole model to the
present knowledge on perturbative QCD resummations at leading and next-to-leading level
in logarithms of the energy (i.e., log xBj ). Among the basic known properties of the sum
of the corresponding Feynman diagrams, the coupling of external sources (in particular
a virtual photon in deep-inelastic reactions) is based on the theorem of kT factorization,
proven in the leading logarithmic approximation of QCD [13]. The theorem states that the
unintegrated gluon distribution, i.e., the distribution of energy and transverse momentum
of gluons in the target, factorizes from the rest of the process. The remaining factor is
the so-called impact factor. This impact factor is an universal quantity, the same in all
processes initiated by the same external source, e.g., the photon. The unintegrated gluon
distribution characterizes the target, but again the target impact factor can also be factorized
out, leaving place to an universal interaction term, given by the BFKL Pomeron in the same
leading logarithmic approach [13]. At next-to-leading level, the modified interaction term
is now known [14] but the impact factors are not yet determined.
It should be emphasized that, although both colour dipole model and kT factorization
were (till now) justified only at the leading logarithm level, in practical applications it is
necessary to go beyond this approximation in order to fix the energy scale of the problem.
Before knowing the exact next-leading approximation, the current extension beyond the
leading order is different in the two approaches. In the dipole model it is necessary to
postulate the relation between xBj and the energy available for the dipole cascade to
develop. In the kT factorization approach used in phenomenology (see, e.g., [12]) the
relation between xBj and gluon longitudinal momentum is fixed by the kinematics of the
corresponding Feynman diagram. As neither of these methods can be justified without
extensive next-to-leading order calculations, it remains an open question to know either
which one describes better the physical reality or how both are to be modified.
In the present paper we discuss the relation between these two approaches. We start
with the kT factorized expression for the total cross-section (with longitudinal polarization
of the photon, but the results extend to the transverse one in the same way) assuming
either full kinematics or its leading-log contribution. In the last case we prove the exact
equivalence with the dipole model expression as expected, since they are both based on
BFKL dynamics. In the former one, we show an explicit violation of the conservation
of transverse positions of the colliding objects during the collision. Taking as an explicit
example the diffractive production of two jets, we extend our discussion to the off-diagonal
gluon distributions, where similar differences appear when using the full kinematics.
Our main conclusion is that, although equivalent at the leading logarithm level, the
extensions of the dipole model and those based on kT factorization to the next-to-leading
order lead to results which are not compatible with each other. This conclusion emphasizes
the urgent need for the full next order calculation 2 which would settle the question
of validity of the two most favoured approaches to the hard collisions at high energy.
2 Such a calculation is under way [15].

440

A. Bialas et al. / Nuclear Physics B 593 (2001) 438450

In particular, the fate of transverse coordinate conservation in high-energy diffractive

collisions is to be examined. This seems important, as some nowadays quite popular
models [912] are based on the assumption that this property remains true not only to
all orders of the perturbation theory but even extend to the non-perturbative regime.
In the next section we remind briefly the results of the QCD dipole model for the
total (virtual) photon cross-section and show in Section 3 that they are equivalent to
the results obtained using kT factorization in the leading logarithm approximation. Next,
kT factorization with exact gluon kinematics is discussed. In Section 4 it is shown to be
incompatible with the QCD dipole model. In Section 5 we extend the arguments to inelastic
collisions involving the non-diagonal gluon distributions. Our conclusions are summarized
in the last section.

2. Total cross-section in the colour dipole approach

Let us first remind briefly for future reference the results obtained in the dipole model
for the total cross-section of the virtual photon on an arbitrary target. The cross-section
formula reads (cf., e.g., [16])
Z
2

(1)
(Y ; Q) = 2Nc d 2 r dz (r; Y ) (r, z; Q) ,
where the factor 2Nc represents summation over spins and colours of the q q states
describing the virtual photon, and (through the optical theorem)
 


(2)
(r; Y ) = 2 rE, zT (t = 0; Y )rE, z 2T (r; Y ),
is the dipole-target cross-section and (r, z; Q) is the light-cone photon wave function (our
notation is explicit on Fig. 1). It is to be noted that T is a function of only one transverse
vector rE. This is the consequence of the diagonal character of the interaction (following
from the conservation of the impact parameter in high-energy collisions).

Fig. 1. Kinematics of the dipole model.

A. Bialas et al. / Nuclear Physics B 593 (2001) 438450

441

We shall exclusively discuss longitudinal cross-sections 3 and thus give below only the
formula for the wave function of the longitudinal photon [17]:

eq
b ,
2z(1 z)QK0 Qr
(3)
L (r, z; Q) =
2
b 2 = z(1 z)Q2 .
with Q
In case the target is itself a dipole of transverse size r0 (one may more generally expand
the target wave-function over a basis of such states), the cross-section (1) is given by [18]
 
Z
 


d ( )Y r 2
e
h( ),
(4)
rE, zT (t = 0; Y ; r0)rE, z = 2s2 r02
2i
r0
where

s Nc 
2(1) (1 ) ( ) ,

is the well-known BFKL [5] kernel eigenvalue and

( ) =

h( ) =

1
4 2 (1 )2

(5)

(6)

Finally, Y is the rapidity range available for the dipole cascade to develop. Y is not uniquely
defined in the context of the dipole model, a consequence of its origin in the leading
logarithm approximation. 4 One expects that Y is of the form


1
(7)
Y0 .
Y = log
xBj
Y0 can at least in principle depend on all variables (except xBj ) which are
relevant in the process considered. In the case of the total cross-section a successful
phenomenology [7] gave Y0 = const but additional dependence on other variables, in
particular on z can also be envisaged [19,20]. As emphasized in the introduction, selecting
a definite form of Y0 cannot be justified in the leading logarithm approximation (on which
the dipole model is based) and thus represents additional assumption which can only be
supported by phenomenological arguments.

3. Total cross-section and the kT factorization

We shall now recall the results obtained for the total cross-section in the approach using
the kT factorization as the starting point, and show that, when considered in the leading-log
approximation, they are equivalent to the dipole model predictions. As mentioned in the
introduction, we introduce the unintegrated gluon distribution which we shall denote by
g(xg , k 2 ) where xg is the target (nucleon) momentum fraction carried by the gluon and kE is
the gluon transverse momentum.
3 The transverse cross-sections can be treated similarly and the obtained conclusions are identical.
4 This is one of the major ambiguities in the application of the QCD dipole model to the data [8].

442

A. Bialas et al. / Nuclear Physics B 593 (2001) 438450

Fig. 2. Kinematics of two gluon exchange in the kT factorization approach.

The cross-section is evaluated from the diagram shown in Fig. 2. The relevant formulae
can be found, e.g., in [21] where one reads 5
L =
=

4 2
FL
Q2
1
8s ef2 Q2 Z

d 2k
k4


2
dz z(1 z)


d 2p

1
1
2
2
2
b +p
b + (p k)2
Q
Q

2

g xg , k 2 .

Using twice the identity

Z


1
1
2 i pE rE
b ,
=
r
e
K
Qr
d
0
b 2 2
p2 + Q

(8)

(9)

results in the following expression:

L =

8s ef2 Q2

1
(2)2

Z1


2
dz z(1 z)

b
d 2 r K0 Qr

b 0
d r K0 Qr
2 0

E r Er 0 )
D rE, rE0 ,
d 2 p ei p(E

(10)

where
0

D rE, rE =

d 2k
E

E 0

g xg , k 2 1 ei kEr 1 ei kEr .
4
k

(11)

The formulae (10) and (11) provide the key point of our discussion. The conservation of
transverse position throughout the interaction would mean that
Z




E r Er 0 )
D rE, rE0 2 rE rE0 .
(12)
d 2 p ei p(E
E it becomes clear from (11)
As this may happen only if D(Er , rE0 ) does not depend on p,
that the question of the gluon kinematics, i.e., the determination of xg as a function of the
5 For the derivation, see [22]. To simplify the formulae, we neglect the quark masses. Our conclusions do not
depend on this simplification (see, e.g., [23]).

A. Bialas et al. / Nuclear Physics B 593 (2001) 438450

443

other variables, is crucial. Indeed, if the gluon kinematics is taken into account, xg can be
expressed by other variables (see, e.g., [21]):
b 2 + k 2 + (p0 )2
Q
E
,
pE0 = pE (1 z)k.
(13)
xg = xBj
b2
Q
E Postponing the detailed
Consequently, as seen from (11), D(Er , rE0 ) does depend on p.
discussion of this case to the next section, we shall now consider the formula (11) in the
leading logarithm approximation (for which the kT factorization can be exactly proven).
In this approximation the relation between xBj and xg is not well defined because in the
limit xBj 0, the difference between log xBj and log xg is finite and thus any contribution
depending on this difference (or equivalently on the ratio xBj /xg ) must be of higher
order. Taking this ambiguity into account we can treat xg in (8) as a constant, independent
of other variables entering the process. 6 Under this condition (8) can be transformed into
the form given by the dipole model formula discussed in the previous section [3,24].
E the integral over d 2 p gives (2)2 2 (Er rE0 )
Indeed, since D(Er , rE0 ) does not depend on p,
and thus we obtain
1
Z
16s ef2 Q2 Z

2



b 2
dz z(1 z)
d 2 r K0 Qr
L =

0
Z 2


d k
E

g xg , k 2 1 ei kEr .
(14)
4
k
Using (3) this can be written in the form of (1) with the function T (r, Y ) now given by [24]
Z 2


d k
2s
E

g xg , k 2 1 ei kEr .
(15)
T (r, Y ) =
4
Nc
k
This formula expresses the dipole-target forward elastic amplitude in terms of the
unintegrated gluon distribution in the target. It can be inverted. To this end we calculate
the moments of T . Using (15) and the identity
Z
 0( /2)0(1 /2)
du 
,
(16)
1 J0 (u) = 1+
u1+
2
[0(1 + /2)]2
0

one obtains

Z
1
e
d 2 r r 22 T (r, Y )
T ( , Y )
2


0( )0(1 )
4 2 s
(17)
g(x
g , ),
=
Nc
21+2 [0(1 + )]2
where
Z
Z 2




1
dk 2+2
d k 2
2
k
k
g
x
,
k
g x, k 2 ,
=
(18)
g(x
g, )
g
4
2
k
k
and the factor between brackets {. . .} can be interpreted as the gluon-dipole vertex
function [23].
6 This property was actually used in the proof of the k factorization [13].
T

444

A. Bialas et al. / Nuclear Physics B 593 (2001) 438450

Using the inverse Mellin transform and (17), we thus obtain

Z


d 22
2
k
g(x
g, )
g xg , k 2
2i
Z
d 22 21+2 [0(1 + )]2 e
Nc
k
T ( , Y ),
=
2i
0( )0(1 )
2 2 s
where Te( , Y ) is given by (17).

(19)

This is a general formula valid for any target. If the target is itself a QCD dipole, we can
use the formula (4) for T and obtain


Z

4Nc s
[0(1 + )]2 ( )Y
d kr0 22
e
h( )
.
(20)
g xg , k 2 =

2i 2
0( )0(1 )
4. Exact gluon kinematics
Although the validity of the kT factorization was established only at the leading order
in log xBj , it is natural to treat the diagrams of Fig. 2 as giving the correct structure
of the process (to all orders). This is usually the way kT factorization is applied for the
description of high-energy collisions [912,21]. It should be realized, however, that such
an approach takes into account only part of the higher order corrections and thus cannot
be justified from the first principles without estimating contributions from other diagrams
which appear at higher orders.
The important consequence of this procedure is, in particular, that the kinematics
of the diagram is to be fully taken into account and thus the value of xg in Eq. (8)
becomes now well-defined and given by (13). As we have already indicated in the previous
section, it turns out that these relations break the connection between the formula (8) of
kT factorization and (1) of the colour dipole model, established in Section 3 at the leading
logarithm level. Indeed, in this case we have

Z 2 
b 2 + k 2 + (p0 )2 2


d k
Q
Er

Er 0

i kE
i kE
g
x
,
k
1

e
.
(21)
1

e
D rE, rE0 =
Bj
4
b2
k
Q
Thus D does depend on pE and, consequently, the integral over d 2 p in (12) does not give
2 (Er rE0 ) (except in the particular case when g(xg , k 2 ) does not depend on xg at all,
corresponding to the energy-independent cross-section). To obtain a better insight, we
express the gluon distribution in the form of a Mellin transform:
Z




dN
2
(xg )N gN k 2 ,
(22)
g xg , k =
2i
and find (see [25])
Z


0
0
d 2 p ei pE (Er Er ) D rE, rE
q
 b 2 2 1N
Z
q
2
N
b





 Q + k 
(Q )
dN

b 2 + k 2 
gN k 2
K1N rE rE0  Q
.

 rE rE0 
2i
(2xBj )N 0(N)
(23)

A. Bialas et al. / Nuclear Physics B 593 (2001) 438450

445

One thus explicitly sees that the process is no longer diagonal in transverse size rE of
the dipole. Since the diagonal character of the dipole interaction in the transverse space
is the fundamental ingredient of the colour dipole approach, we must conclude that the
extension of the theorem of kT factorization [12,21] to non-leading order by including the
exact gluon kinematics is not compatible with the colour dipole model. This conclusion
seems important, as it emphasizes the need for a better understanding of the standard
phenomenology of hard high-energy interactions.

5. Diffractive processes: off-diagonal gluon distributions

To complete the argument, we shall discuss also inelastic diffraction. The simplest
example for which results were presented in both models is the two jet production off
any target.
In the dipole model the cross-section for this process can be readily obtained taking as a
starting point the formula for quasi-elastic diffraction given in [26]. The result is

2
Nc 
d
=
G p,
z, t; Y  ,
2
dz d p dt
2
where


1
G p, z, t; Y =
2

 





d 2 r exp i pE rE rE, zT (t; Y )rE, z rE, z; Q .

(24)

(25)

Here pE is the transverse momentum of the jet and (Er , z; Q) is the photon wave function
given by (3). 7
In the case when the target is itself a dipole of size r0 , the amplitude hEr , z|T (t; Y ; r0 )|Er , zi
was first discussed in [6] and explicitly calculated in [27]. Here we shall restrict ourselves
to the forward scattering t = 0 [28]. In this case the formula for T is much simpler and
given by (4), up to the determination of Y .
As already explained, Y is the rapidity range available for the dipole cascade to
develop. It should be emphasized again that it remains undetermined and, in particular,
need not be the same as in (4). The general formula (7) remains valid but in the case
of the diffractive production a successful phenomenology [8] required (see formula (7))
2
Y0 = log log Q2Q+M 2 , i.e.,
 
xBj
1
.
(26)
Y = log
,
xP =
xP

We shall now show that, at the leading logarithm approximation, the formula for the
diffractive two jet production obtained with the help the kT factorization can also be
expressed in the form (24), (25) derived from the dipole model. We also derive an explicit
relation between the dipole elastic amplitude and the off-diagonal gluon distribution.
7 As before, we shall discuss explicitly only longitudinal photons. The case of transverse photons can be treated
on the same lines.

446

A. Bialas et al. / Nuclear Physics B 593 (2001) 438450

We start from the formula of [29], giving the forward (t = 0) differential cross-section
for diffractive production of two jets (see also [30,31])
dL
dt dz d 2 p
Z




s2 2 d 2 k d 2 k 0
eq
f xBj , x 0 , x 00 , k 2 f xBj , x 0 , x 00 , k 02 L (z, k, k 0 , p),
=
4
04
2Nc
k k
(27)
where the function f (xBj , x 0 , x 00 , k 2 ) is the linear combination of the off-diagonal [32,33]
gluon distributions inside a proton

1




f xBj , x 0 , x 00 , k 2 = g x 0 + xP , x 0 , k 2 + g x 00 + xP , x 00 , k 2 ,
2
where x 0 and x 00 are the gluon momenta as explained in [29] and xP given by (26).
Since the expression for L (given in [29]) can be written in the factorized form


1
1
2 2
2

L = 4Q z (1 z)
b 2 (p k)2 + Q
b2
p2 + Q


1
1

,
2
2
0
b
b2
p +Q
(p k )2 + Q

(28)

(29)

b 2 = z(1 z)Q2 , one sees immediately that (27) can be cast in the form (24) with
where Q
q
s em eq2
2Qz(1 z)
G=
Nc


Z 2


1
1
d k
0 00 2
f
x
,
x
,
x
,
k

.
(30)

Bj
b 2 (p k)2 + Q
b2
k4
p2 + Q
It remains to be shown that G given by (30) can be written in the form (25).
To see this, we follow closely the argument of Section 3. Using the identity (9) we write
the first term of (30) in the form of an integral over d 2 r:
Z 2


1
d k
f xBj , x 0 , x 00 , k 2 2
4
b2
k
p +Q
Z
Z 2




1
d k
Er
b .
=
f xBj , x 0 , x 00 , k 2 K0 Qr
(31)
d 2 r ei pE
2
k4
The same operation can be applied to the second term and thus we obtain
Z
1
d 2 r eipr T (r, z, x, xP )L (r, z; Q),
G=
2
where L is given by (3) and
Z 2


2
d k
E

s
1 ei kEr f xBj , x 0 , x 00 , k 2 .
T (r, z, x, xP ) =
4
Nc
k

(32)

(33)

Thus we recover a similar formula as in the case of total cross-section (see Section 2,
Eq. (15)), except that now the unintegrated gluon distribution g(x, k 2 ) is replaced by the

A. Bialas et al. / Nuclear Physics B 593 (2001) 438450

447

unintegrated off-diagonal gluon distribution f (xBj , x 0 , x 00 , k 2 ), as appropriate for inelastic

processes [32,33].
One sees that the Eq. (33) is valid even if exact gluon kinematics is included (at no place
in the derivation we needed the assumption that x 0 and x 00 are constants, independent of pE ).
Its physical interpretation, however, depends crucially on this problem. Indeed, if x 0 and
E also T given by (33) is p-independent
E
and only in this case it can
x 00 do not depend on p,
be interpreted as the amplitude for scattering of the dipole of a given transverse size rE. If,
on the other hand, the exact gluon kinematics is taken into account (which includes partly
contribution of higher orders) one has [29]
x 0 = xP

k 2 + 2pE kE
,
z(M 2 + Q2 )

x 00 = xP

k 2 + 2pE kE
,
(1 z)(M 2 + Q2 )

(34)

and thus T is a function of both rE and p.

E This of course cannot be the case if it is to be
interpreted as the scattering amplitude.
Thus we conclude that, similarly as in the case of total cross-section, the description of
diffractive scattering in the dipole model is only compatible with the kT factorization if
one restricts to the leading logarithm level.
If one stays at the leading logarithm approximation, one can of course repeat the
argument of Section 3 and thus conclude that the relations between the dipole amplitude
and the gluon distribution of Section 3 should be valid with the replacement g f and
Y Y . In particular
Z




Nc
d 22 21+2 [0(1 + )]2 e
0 00 2
k
(35)
T ,Y .
f xBj , x , x , k =
2
2i
0( )0(1 )
2 s
Using (4), one can also write down the explicit prediction of the dipole model for the gluon
distribution in the dipole target:


Z

4Nc
[0(1 + )]2
d kr0 22 ( )Y
0 00 2
s
. (36)
e
h( )
f xBj , x , x , k =

2i 2
0( )0(1 )
6. Discussion and conclusions
In conclusion, we have shown that in the leading logarithm approximation the
kT factorization and the colour dipole model give equivalent description of hard processes
at high energy.
On the other hand, when one steps beyond the leading logarithmic approximation
by supplementing the kT factorization algorithm with the exact kinematics of the
corresponding Feynman diagrams (as is the case in practical applications [12]), the result
is incompatible with the colour dipole model: we have shown that such a procedure leads
to a violation of the conservation of transverse positions and sizes of the colliding objects
(which is fundamental for the dipole model interpretation of high-energy collisions).
This breaking of diagonality of the interaction in impact parameter should not be
surprising: indeed, since the transverse momentum and impact parameter are conjugate

448

A. Bialas et al. / Nuclear Physics B 593 (2001) 438450

variables, conserving one of them in the interaction does not allow to conserve the other
one. In fact, this effect is analogous to that discussed already in [26].
Several comments can be made.
(i) As both the colour dipole model and kT factorization (with exact kinematics) are
currently used for the description of data [12], our result indicates that such analyses
must be taken with some care.
(ii) It also emphasizes the need for a complete next-to-leading order calculation which
would elucidate the problem of compatibility of the two approaches and also
provide the necessary information on the correct energy scale.
(iii) In absence of such a complete calculation, one may only speculate about the
origin of the difficulty. One possibility is that, by including other missing higher
order contributions, one recovers the compatibility between the two pictures. This
would mean that the theorem of kT factorization should be replaced by a sort of
impact parameter factorization. Another possible scenario is that kT factorization
at higher orders will not correspond to impact parameter conservation and thus the
dipole model must be abandoned or reformulated.
(iv) Speculating about the possibilities of modification of the colour dipole model,
one may think necessary to add (at the next-to-leading order) the contributions
1 dipole 2 dipoles with the two new dipoles having energies of the same order
of magnitude. Such a contribution would certainly require a reinterpretation of the
conservation of transverse positions in high-energy scattering and thus make more
plausible the compatibility with kT factorization.

Acknowledgements
A.B. is indebted to A. Stasto and K. Golec-Biernat for useful discussions. He thanks
the Service de Physique Thorique of Saclay for support and kind hospitality. This
investigation was supported in part by the KBN Grant No 2 P03B 086 14 and by the
Subsidium of Fundation for Polish Science 1/99.

References
[1] L.L. Frankfurt, M.I. Strikman, Phys. Rep. 160 (1988) 235;
A.H. Mueller, Nucl. Phys. B 335 (1990) 115.
[2] N.N. Nikolaev, B.G. Zakharov, Z. Phys. C 49 (1991) 607.
[3] N.N. Nikolaev, B.G. Zakharov, Phys. Lett. B 332 (1994) 184;
N.N. Nikolaev, B.G. Zakharov, Z. Phys. C 64 (1994) 631;
N.N. Nikolaev, B.G. Zakharov, Z. Phys. C 53 (1992) 331.
[4] A.H. Mueller, Nucl. Phys. B 415 (1994) 373.
[5] L.N. Lipatov, Sov. J. Nucl. Phys. 23 (1976) 642;
V.S. Fadin, E.A. Kuraev, L.N. Lipatov, Phys. Lett. B 60 (1975) 50;
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP 44 (1976) 45;
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP 45 (1977) 199;
I.I. Balitsky, L.N. Lipatov, Sov. J. Nucl. Phys. 28 (1978) 822.

A. Bialas et al. / Nuclear Physics B 593 (2001) 438450

449

[6] L.N. Lipatov, Zh. Eksp. Teor. Fiz. 90 (1986) 1536;

L.N. Lipatov, Sov. Phys. JETP 63 (1986) 904.
[7] H. Navelet, R. Peschanski, C. Royon, Phys. Lett. B 366 (1996) 329;
H. Navelet, R. Peschanski, C. Royon, S. Wallon, Phys. Lett. B 385 (1996) 357.
[8] A. Bialas, R. Peschanski, C. Royon, Phys. Rev. D 57 (1998) 6899;
S. Munier, R. Peschanski, C. Royon, Nucl. Phys. B 354 (1998) 297.
[9] K. Golec-Biernat, M. Wuesthoff, Phys. Rev. D 59 (1999) 014017;
K. Golec-Biernat, M. Wuesthoff, Phys. Rev. D 60 (1999) 114023.
[10] K. Golec-Biernat, hep-ph/0006080;
A.M. Stasto, K. Golec-Biernat, J. Kwiecinski, hep-ph/0007192, and references therein.
[11] E. Gotsman, E.M. Levin, U. Maor, Eur. Phys. J. C 5 (1998) 303, hep-ph/9708275;
E. Gotsman, E. Levin, U. Maor, E. Naftali, Nucl. Phys. B 539 (1999) 535, hep-ph/9808257;
E. Gotsman, E.M. Levin, U. Maor, E. Naftali, Eur. Phys. J. C 10 (1999) 68, hep-ph/9904277;
J. Forshaw, G. Kerley, G. Shaw, Phys. Rev. D 60 (1999) 074012;
J. Forshaw, G. Kerley, G. Shaw, Nucl. Phys. A 675 (2000) 80c;
M. McDermott, L. Frankfurt, V. Guzey, M. Strikman, hep-ph/9912547;
K. Suzuki, A. Hayashigaki, K. Itakura, J. Alam, T. Hatsuda, Phys. Rev. D 62 (2000) 031501,
hep-ph/0005250;
M.G. Ryskin, Yu.M. Shabelski, A.G. Shuvaev, hep-ph/0007238;
J. Nemchik, hep-ph/0008161.
[12] For recent reviews, see, M. Wuesthoff, A. Martin, J. Phys. G 25 (1999) R309;
A. Hebecker, Phys. Rep. 331 (2000), hep-ph/9905226;
A. Hebecker, Acta Phys. Pol. B 30 (1999) 3777, hep-ph/9909504.
[13] S. Catani, M. Ciafaloni, F. Hautmann, Nucl. Phys. B 366 (1991) 135;
S. Catani, M. Ciafaloni, F. Hautmann, Nucl. Phys. B 29 (1992) 182;
J.C. Collins, R.K. Ellis, Nucl. Phys. B 360 (1991) 3;
E.M. Levin, M.G. Ryskin, Yu.M. Shabelski, A.G. Shuvaev, Sov. J. Nucl. Phys. 53 (1991) 657.
[14] V.S. Fadin, L.N. Lipatov, Phys. Lett. B 429 (1998) 127;
M. Ciafaloni, Phys. Lett. B 429 (1998) 363;
M. Ciafaloni, G. Camici, Phys. Lett. B 430 (1998) 349.
[15] J. Bartels, S. Gieseke, C.-F. Ciao, hep-ph/0009102;
V. Fadin, D. Ivanov, M. Kotsky, hep-ph/0007119;
V. Fadin, J. Bartels, private communications.
[16] A. Bialas, Acta Phys. Pol. B 27 (1996) 1163.
[17] J.D. Bjorken, J.B. Kogut, D.E. Soper, Phys. Rev. D 3 (1971) 1382.
[18] A.H. Mueller, B. Patel, Nucl. Phys. B 425 (1994) 471.
[19] A. Bialas, Acta Phys. Pol. B 28 (1997) 1239.
[20] A. Bialas, W. Czyz, W. Florkowski, Eur. Phys. J. C 2 (1998) 683.
[21] J. Kwiecinski, A. Martin, A. Stasto, Phys. Rev. D 56 (1997) 3991.
[22] J.R. Forshaw, D.A. Ross, Quantum Chromodynamics and the Pomeron, Cambridge Univ. Press,
Cambridge, 1997.
[23] S. Munier, R. Peschanski, Nucl. Phys. B 524 (1998) 377.
[24] V. Barone, M. Genovese, N.N. Nikolaev, E. Predazzi, B.G. Zakharov, Phys. Lett. B 326 (1994)
161.
[25] I.S. Gradstein, I.M. Ryzhik, Table of Integrals, Series and Products, Academic Press, 1980,
formula 8.432.5.
[26] A. Bialas, R. Peschanski, Phys. Lett. B 387 (1996) 405.
[27] A. Bialas, H. Navelet, R. Peschanski, Eur. Phys. J. C 8 (1999) 643.
[28] A. Bialas, W. Czyz, Acta Phys. Pol. B 29 (1998) 2095.
[29] K. Golec-Biernat, J. Kwiecinski, A.D. Martin, Phys. Rev. D 58 (1998) 094001.
[30] N.N. Nikolaev, B.G. Zakharov, Phys. Lett. B 332 (1994) 177.

450

A. Bialas et al. / Nuclear Physics B 593 (2001) 438450

[31] J. Bartels, H. Lotter, M. Wuesthoff, Phys. Lett. B 379 (1996) 239.

[32] A.V. Radyushkin, Phys. Rev. D 56 (1997) 5524.
[33] X. Ji, J. Phys. G 24 (1998) 1181;
X. Ji, Phys. Rev. D 60.

Nuclear Physics B 593 (2001) 451470

www.elsevier.nl/locate/npe

Three flavour neutrino oscillations in models with

large extra dimensions
R.N. Mohapatra a , A. Prez-Lorenzana

a,b,

a Department of Physics, University of Maryland, College Park, MD 20742, USA

b Departamento de Fsica, Centro de Investigacin y de Estudios Avanzados del I.P.N.

Apdo. Post. 14-740, 07000 Mxico, D.F., Mexico

Received 29 June 2000; accepted 25 September 2000

Abstract
The key challenges for models with large extra dimensions, posed by neutrino physics are: first
to understand why neutrino masses are small and second, whether one can have a simultaneous
explanation of all observed oscillation phenomena. There exist models that answer the first challenge
by using singlet bulk neutrinos coupled to the standard model in the brane. Our goal in this paper
is to see to what extent the simplest versions of these models can answer the second challenge. Our
conclusion is that the minimal framework that has no new physics beyond the above simple picture
cannot simultaneously explain solar, atmospheric and LSND data, whereas there are several ways
that it can accommodate the first two. This would suggest that confirmation of LSND data would
indicate the existence of new physics either in the brane or in extra dimensions or both, if indeed it
turns out that there are large extra dimensions. 2001 Elsevier Science B.V. All rights reserved.
PACS: 14.60.Pq; 14.60.St; 11.10.Kk

1. Introduction
Particle physics models where there are large hidden space dimensions beyond the three
familiar ones have been the focus of intense activity during the past two years [1,2]. Beyond
the simple reason that such extra dimensions are predicted by string theories, a major point
of interest in these models is that often these large extra dimensions come with a TeV scale
for the strings which leads to a plethora of new observable phenomena in collider as well
in other arenas of particle physics and cosmology.
* Corresponding author. New address: AS ICTPHE Group, Rm. 266, Strada Costiera 11, 34100 Trieste, Italy.

E-mail addresses: [email protected] (R.N. Mohapatra), [email protected]

(A. Prez-Lorenzana).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 3 4 - 9

452

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

In order for these models to provide a satisfactory description of low energy particle
physics, they must handle some obvious problems that come with the existence of a
fundamental scale in the multi-TeV range. One such problem has to do with understanding
the small neutrino masses in a natural manner. The conventional seesaw [3] explanation
which is believed to provide the most satisfactory way to understand this, requires that
the new physics scale (or the scale of SU(2)R U (1)BL ) be around 1012 GeV or higher.
Clearly, low string scale theories do not have any fundamental scale of that type. Moreover,
the low value of the string scale leads to enhanced (and unacceptable) contributions to
neutrino masses from higher dimensional operators. While there are suggestions involving
thick branes with point splitting [4] to remedy similar problems that arise from 1B 6= 0
operators, they dont work for neutrino mass operators. Therefore, a necessary ingredient
to understand small neutrino masses in the low string scale models is to assume that theory
have a BL symmetry. This will forbid higher dimensional operators LH LH /M (M is
the string scale), which are the source of the problem for neutrino masses.
Depending on whether BL is a global or local symmetry, one can have two ways to
solve the neutrino mass problem in models with large extra dimensions. 1 In the former
case, discussed in [6], one has to introduce singlet bulk neutrinos which then lead to small
Dirac masses for them. On the other hand, if BL symmetry is chosen to be a local
symmetry, anomaly cancellation requires that, right handed neutrinos be present in the
brane as in the models discussed in Refs. [7,8]. Since local BL must be broken to avoid
massless gauge bosons, one again has to deal with the induced operators of the same type
as above with M = MBL . It was shown in Refs. [7,8] that to get neutrino masses in the
desired eV range, one must have string scale M ' MBL 109 GeV range or higher.
As a result, a class of experimentally accessible phenomenological predictions are lost,
although long range gravity tests are still possible. These models are similar to the ones
discussed in [9]. We must emphasize that there may be other ways to realize local BL
symmetry, where one may be able to maintain a lower string scale. For instance, anomaly
cancellation can be achieved by putting the right handed neutrino in a distant brane, as in
a recent scheme [10]. The smallness of neutrino masses in such schemes could arise from
the exponential suppression of BL breaking effects in our brane. We do not pursue local
BL models here.
In the context of models that have global U (1)BL symmetry, one can maintain the TeV
scale for the strings and still get small neutrino masses by introducing isosinglet neutrinos
in the bulk as has already been discussed in Ref. [6]. These models are interesting because a
very minimal set of particles beyond the standard model are sufficient to get small neutrino
masses. Key reason for this result is the relation between the fundamental scale, M , the
radius of the extra dimensions, R, and effective Planck scale, MP ` ,
M 2+n R n = MP2 ` .

(1)

1 It is generally believed that string theories do not have any global symmetries [5], which would seem to imply
that BL must be a local symmetry. In our discussion, however, we will consider that BL is a global symmetry
of the theory, as in Ref. [6] and see where it leads us phenomenologically.

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

453

Since the singlet neutrino is a bulk field, the effective couplings of its Fourier modes to the
standard model fields are naturally suppressed by the ratio M /MP ` , which for a TeV M
produces the right order of magnitude for neutrino masses.
An analysis of the implications of the mixing profile in these models for solar neutrino
deficit was discussed in [11]. Also, implications for atmospheric neutrinos were discussed
in [12], and some phenomenological bounds were given in [1114]. Our goal in this
paper is to attempt a unified explanation of all known neutrino oscillation data, i.e., solar,
atmospheric as well as LSND under different assumptions for the initial input parameters
for the minimal bulk neutrino scenario.
The basic reason for embarking on such an ambitious program is the encouraging
feature that the masses of the lower KK modes of the bulk neutrinos are given by integral
multiples of R 1 which is of order 103 eV when R millimeter. This is of the right order
necessary to solve the solar neutrino problem via small angle MSW mechanism using e
to s oscillation. Thus we see that there exists a natural way to understand the lightness of
the sterile neutrino [8,11,15], a situation if realized would pose a major challenge to four
dimensional theories. Once the solar neutrino problem is understood, it would appear that
all the necessary ingredients are at hand to understand the atmospheric and LSND data
using oscillations among the familiar neutrinos, i.e., e for LSND and for
atmospheric.
This program has already been undertaken in special parameter domains and it has
already been suspected [8,12] that it does not really work. Our goal is to extend these
discussions to a somewhat larger parameter domain to see if there is chance for this
program to succeed and unfortunately our answer is also in the negative. Our work
complements the above works and extends them. Specifically, we try to give analytical
reasonings to see how the different oscillation data can (or cannot) be understood. Since
we do not take recourse to a detailed numerical analysis, we cannot rule out the possibility
that some small parameter domain exists where all data can be accommodated; but we
consider that unlikely.
The negative conclusion of our work, combined with the works of [8,12] implies with
virtual certainty that in the large extra dimension framework, understanding the neutrino
data does require new physics beyond the standard model in the brane or new physics in
extra dimensions or both.
Our basic strategy is as follows: we start by requiring that the parameters of the theory
provide an explanation for solar and atmospheric oscillations. Then we ask whether they
can account for the small LSND probability of e appearance from the beam.
This paper is organized as follows: in Section 2, we discuss neutrino oscillations with
a single flavour, which depends on two parameters, the bulk radius and a dimensionless
parameter related to the Dirac mass term in the theory. The latter defines the pattern
of neutrino oscillations. This analysis sets the stage for the three flavour case, which we
discuss in Section 3 for various possible domains of the parameter space. We end the paper
with a concluding section that summarizes the results and discusses their implications.

454

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

2. Oscillations with a single flavour

Let us begin our discussion by focusing on the simplest case with one generation of
fermions in the brane and one bulk neutrino, to understand the general profile of the
neutrino oscillations in models with large extra dimensions. We will discuss the necessary
ingredients to understand the three flavour case that is explored in the next section.
Obviously, the fields that could propagate in the extra dimensions are chosen to be gauge
singlets. Let us denote bulk neutrino by B (x , y). It has a five dimensional kinetic energy
term and a coupling to the brane field L(x ) given by
Z
BR (x, y = 0) + dy BL (x, y)5 BR (x, y) + h.c.,
(2)
L = LH
where from the five dimensional kinetic energy, we have only kept the 5th component
that contributes to the mass terms of the KK modes in the brane; H denotes the Higgs

doublet, and = h MMP ` the suppressed Yukawa coupling. It is worth pointing out that this
suppression is independent of the number and radius hierarchy of the extra dimensions,
provided that our bulk neutrino propagates in the whole bulk. For simplicity, we will
assume that there is only one extra dimension with radius of compactification as large as a
millimeter, and the rest with much smaller compactification radii. The smaller dimensions
will only contribute to the relationship (1) but its KK excitations will be very heavy
and decouple from neutrino spectrum. Thus, all the analysis could be done as in five
dimensions.
The first term in Eq. (2) will be responsible for the neutrino mass once the Higgs field
develops its vacuum. The induced Dirac mass parameter will be given by m = v, which
for M = 1 TeV is about h 105 eV. Obviously this value depends only linearly on the
fundamental scale. Larger values for M will increase m proportionally. After introducing
the expansion of the bulk field in terms of the KK modes, the Dirac mass terms in (2) could
be written as
!




0B
m
2
m
0
,
(3)
eL BL
0
BR
0 5
where our notation is as follows: B0 represents theKK excitations, the off diagonal term

2 m is actually an infinite row vector of the form 2 m(1, 1, . . .). The operator 5 stands
for the diagonal and infinite KK mass matrix whose n-th entrance is given by n/R. This
notation was introduced in [7] to represent the infinite mass matrix in a compact manner.
Using this short hand notation makes it easier to calculate the exact eigenvalues and the
eigenstates of this mass matrix [8]. Simple algebra yields the characteristic equation
(4)
2n = 2 cot(n ),

with n = mn R, = 2 mR, and where mn is the mass eigenvalue [6,11]. The eigenstates,
on the other hand are given symbolically by [8]
"
#

1
2 m5 0
BL ,
L +
(5)
nL =
Nn
m2n 52

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

455

where the sum over the KK modes in the last term is implicit. Nn is the normalization
factor given by


1
(6)
Nn2 = 2 2n + f ( ) ,

where f ( ) = 2 /2 + 2 4 /4. Figs. 1 and 2 depict the exact numerical results of n and
Nn for several choices for the value of the parameter. Using the expression (5), we can
write down the weak eigenstate L in terms of the massive modes as

X
1
nL .
L =
Nn

(7)

n=0

Thus, the weak eigenstate is actually a coherent superposition of an infinite number of

massive modes. Therefore, even for this single flavour case, the time evolution of the mass
eigenstates involves in principle all mass eigenstates and is very different from the simple
oscillatory behaviour familiar from the conventional two or three neutrino case. The time
dependent survival probability is given by
 i L 2 2
X e 2ER2 n 


Psurv (L) = |hL (L)|L (0)i|2 = 


Nn2 
n=0

L
2
2
X
sin2 4ER
2 (n k )
= 12
.
(8)
Nn2 Nk2
k,n=0
It is clear that the survival probability depends strongly on the parameter , reflecting
the universal coupling of all the KK components of B with L in (2). Figs. 3 and 4
show the profile of the survival probability obtained from the numerical solutions for three
different values of and is clearly very different from simple familiar oscillatory behaviour.
However, to better understand these results, we will follow an analytical approach in what
follows.
It is simpler to consider the two limiting cases. First let us assume that  1. As already
known [8,11], the eigenvalues in this case are given by 0 = mR, and n = n otherwise.
Therefore, to a good approximation, we may take the mixing parameters N0 = , and
Nn = (n/ ) for non zero n. Where the extra factor = (1 + 2 2 /6)1/2 is introduced to
keep the proper normalization in the expansion (7). This approximation is confirmed by
our Figs. 1 and 2. The survival probability is now given as





2 n2 L
2 (n2 k 2 )L
4 2 X sin 4ER 2
2 4 X sin
4ER 2
4
.
(9)
Psurv (L) = 1 4

n2

n2 k 2
n=1

k,n=1

It is simple to see from last expression that the probability has an oscillation length Losc =
4ER 2 . A typical profile in this case is depicted in Fig. 4. Also, the main contribution
to the oscillation pattern comes from the lowest elements of the tower, which turns out to
be the main component of L . Another way to see this result is to note that could be
made small by making R small; but this makes the KK excitation masses large so that they
decouple from the light sector of the theory leaving only the right handed component of

456

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

Fig. 1. Eigenvalues n as calculated from Eq. (4) for various values of . For comparison with the
low limit discussed in the main text we show the deviation from the integer numbers. This picture
depict how for large the eigenvalues get shifted close to semi-integer numbers. Notice also that for
small ; 0 is clearly the only non integer eigenvalue.

Fig. 2. Mixing factors for the expansion (7) for different values of the parameter . To make the effect
for large visible we have introduced an arbitrary normalization where N0 = 1. This amplifies the
value of 1/Nn by a constant scaling. Of course, the real values satisfy the normalization condition.
This figure shows how, for large , the effective number of KK mass eigenmodes contributing into
the weak eigenstate L gets larger and those modes become almost equally suppressed, while for
small the lightest mode (n = 0) become the main component, with the decoupling of all other
modes.

the zero mode of the bulk neutrino, BR to form a Dirac mass with L and stay light. The
left-handed zero mode decouples and remains massless.
In the  1 limit, the small mixing with the tower elements leads to an oscillation
pattern dominated by the lightest KK mode with a mass just about 1/R and more or less
simulate the familiar oscillatory one. This can happen, for instance, if R 0.2 mm, which

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

457

Fig. 3. Exact survival probability for L versus L/2E in units of R 2 (as it is explicit in the argument)
for three different values of . Dotted lines represent the continuous approximation discussed in the
main text. Note that only the low limit has a periodic behaviour.

Fig. 4. Here we show an amplification of the survival probability for the = 0.1 case showed in
Fig. 3. Note the large number of wiggles produced by the oscillation of consecutive levels in Eq. (9).
We also depict the continuous limit (dotted line) for comparison.

gives 1/R 2 106 eV2 , just about what it is needed to provide an explanation to the solar
neutrino problem assuming a small MSW mixing angle [11]. To use this case to compare
with solar neutrino data, one needs to include the matter effect, which has already been
discussed in Ref. [11] and we do not enter into this here.
A more direct application of the formula in Eq. (9) can be made to discuss the
atmospheric neutrino oscillations, which is a vacuum oscillation. However, to get the
equivalent of large mixing angle we need to adjust 1 and 1/R 2 1m2atm , which means

458

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

that Eq. (9) is not applicable and we must therefore use Eq. (8) and study its large limit.
To discuss this, we start with the limit when  1. First thing to note is that the pattern
of eigenvalues is very different in this case from the small case. For small values of n
the eigenvalues are well approximated by n = 2n+1
2 , while Nn becomes n independent
until certain cut off value that roughly speaking is given by n = 2 2 /4. Beyond that
point n n and we may no longer neglect the contribution of n to (6). This results in
a suppression of Nn that goes like 1/n. It should therefore be reasonable to consider as a
first approximation that the expansion (7) is cut off at n , and that all the mass eigenstates

contributing to L are equally suppressed by 1/ n . In this approximation, we obtain the

survival probability to be


n


2 X
L
2
2
2
sin
k .
(10)
Psurv (L) = 1
n 2
4ER 2 n
k,n=0

Certainly, this relation presents a oscillatory profile, however, the superposition of the
equally suppressed oscillations will result most of the time on a destructive interference,
with the exception of the very sharp resonances that appear each time L reaches a multiple
of the oscillation length. In other words, L is a superposition of a large number of mass
eigenstates, with masses covering a large range, from 1/2 to n in units of 1/R, all of them
contributing by the same amount. As a result, once L is released, one may surmise that the
time evolution of the different components will most likely wash out the original coherent
superposition and the initial L will almost disappear. As Fig. 3 shows, this conclusion is
borne out by the the numerical analysis. To get an analytical result that also supports this
conclusion, note that in the sum in Eq. (10), the dominant contributions come when n 6= k
in which case each term in the sum averages to 1/2 and on performing the double sum it
is easy to see that one arrives at PSsurv 1/n .
If n is very large, it suppresses the survival probability too much and cannot help in
the understanding of the atmospheric data. So, clearly, if we wanted an understanding of
the atmospheric data, we must assume smaller , perhaps values closer to one. In this
case, truncating the sum in the expression for the survival probability in Eq. (9), cannot be
justified and we must seek an alternative way to deal with Eq. (8). An approach suggested
in [12] is to use a continuous approximation to the sum in Eq. (8), which leads to
Z
2
 2 4 
2


p

2
eizn



 4
(11)
Psurv (z) =  dn 2
1 erf izf ( )  ,
 =

n + f ( ) 
4f ( )
0

where z = L/2ER 2 . Clearly, for large ; f ( ) 2 4 /4, and last expression in Eq. (11)
simplifies. In order to study the dependence of Psurv (z) on , we plot Psurv in Fig. 3 in
the continuous and discrete approximations. One thing that emerges is the non-oscillatory
nature of the function as we exceed = 1. We also see that the continuous limit gives
a good approximation for the slope (see Fig. 3), even for cases where 6 1 [12]. This
expression, however, does not represent the survival probability in the limit  1, since
in this case (i.e.,  1), the Psurv has an oscillatory behaviour unlike the last term in Eq.
(11) (see Fig. 4). In the overlap region close to ' 1, one may evaluate the second term in

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

459

Eq. (11) in a slightly different way which leads to an expression for the survival probability
as
Psurv ( ) = 2 ( ) + 2 ( ),

(12)

where we have introduced the new variable = zf ( ) = ( 2 /2R) L/2E and the
functions ( ) = 1 C( ) S( ); and ( ) = C( ) S( ) with S and C the sine and
cosine Fresnel integrals:
r
C( ) =

Z
2

dt cos(t )
0

and

S( ) =

Z
dt sin(t 2 ).
0

We draw attention to the fact that involves not only the model parameters and R
but also the experimental variable z. Thus for a given model, the variable has different
values for different oscillation experiments; for instance, for the atmospheric neutrino case,
2
2
1
a typical value for
2R 1.2 10 eV .
The expression (12) is valid just before Psurv ( ) reaches the average, PSsurv = 4/ 2 2 .
In Fig. 5 we present the behaviour of , and Psurv . Note the steep fall off of Psurv as
increases. For = 1, Psurv has already dropped under 20%, which is smaller than the
observed deficit in solar and atmospheric data. Therefore, if we want to fit the overall
suppression of the atmospheric neutrinos, we must remain in a very narrow range of
parameters. A rough estimate of these parameters may be obtained by expanding (12) to
first order on , which yields
r

 
2
L 1/2
2
=12
.
(13)
Psurv ( ) 1 2

R
4E

l
Fig. 5. In this figure we plot the functions ( ) and ( ) that describe the slope of the survival
probability Psurv (which we also show)in the continuous limit where > 1. The argument is the
dimensionless parameter = ( 2 /2R) L/2E defined in the main text.

460

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

Now, we may invert the last equation to get



2 (1 PSexp ) L 1/2
'
.
R
4E exp
4

(14)

Using average values for the L, E and Pexp for atmospheric neutrinos, we find 2 /R
102 eV. Note that this naive approximation gives just about the value obtained by the
numerical fitting of the data in [12]. For this value, as > 1, we get R in the range
103 eV < 1/R < 102 eV implying 1 < 2 < 10. Thus a solution to the atmospheric
neutrino puzzle using KK modes requires that R be in the sub-millimeter range. This in
turn determines how large M should be to avoid any unwanted large fine tuning. We see
that for m > 103 eV and Yukawa coupling h of order one, we need M > 100 TeV.
We also point out that in the case of > 1, extra dimensions must be much larger
than a millimeter if we want to fit the solar neutrino data either via MSW or via vacuum
oscillation. To see this note that if we take 1m2sol 106 eV2 , this would imply L/4E
1/1m2sol 106 eV2 . Putting this in Eq. (13), we get 2 /R ' 104 eV. For  1, this
implies that R  0.2 mm. Similar estimate for the case of vacuum oscillation yields R 
2 cm. Therefore, in all our discussion of solar neutrino oscillations involving the bulk
neutrinos, we will work in the approximation  1.
Let us now summarize our findings for a single neutrino case: for  1, Psurv has an
oscillatory behaviour as given by (9). As approaches 1, n starts to deviate from the
integer value n, which in turn disturbs the periodic nature of Psurv and the maximum and
minimum values of probability are not reached away from the source [6]. This picture gets
worse as one approaches  1 when a very sharp slope drives the probability near zero as
we move away from the source of the neutrinos and it remains around its average, 4/ 2 2 ,
most of the time. For this reason, whenever we try to fit the atmospheric neutrino data with
bulk neutrinos, the value of must be tuned to a very narrow range.
Let us apply the discussions of this section to study how the oscillation of known neutrinos to bulk ones would effect the current experiments such as CHOOZ-PALOVERDE,
LSND and the atmospheric neutrinos. For this purpose, we first note that in contrast with
the usual two neutrino oscillation case, where the pattern is determined by three parameters, the mixing angle , the 1m2 and L/4E for the experiment, in the case of bulk neutrino
oscillation, we have L/4E characterizing an experiment like before, bulk radius R, which
replaces 1m2 (which we will assume to be in the millimeter range) and model parameter
(which is the analog of the mixing angle). For a given R, is the only parameter characterizing the oscillation pattern. In Fig. 6, we plot the variation of the survival probability
Psurv against for the various cases mentioned above for a typical characteristic value of
L/4E. We see from this figure that to explain the observed overall deficit of atmospheric
neutrinos by nearly 50%, one needs to go to 1.5 or so. This conclusion is in accord
with our conclusion based on Eq. (14) above.

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

461

Fig. 6. Dependence of the survival probability as a function of for the characteristic experimental
values of L/4E, with R = 0.2 mm.

3. Three flavour oscillations

Let us now apply the discussion of the previous section to the case of three standard
model generations in the brane so that we have three brane neutrinos e,, . To give masses
to all of them in a minimal scenario, we will use three bulk neutrinos and allow arbitrary
Yukawa couplings between the bulk and the brane neutrinos. This leads to an arbitrary
Dirac mass matrix that involves the familiar left-handed neutrinos and the right handed
KaluzaKlein modes of the bulk neutrinos. As already discussed in [8], the most general
Dirac mass terms with three flavours may be written, after a rotation of the bulk fields, as
Z
(15)
L = L U MD BR (y = 0) + dy BL 5 BR + h.c.,
where U is a unitary matrix and MD = Diag(m1 , m2 , m3 ), in the basis where L =
(e , , )L ; and B = (B1 , B2 , B3 ). The mass parameters m are just the eigenvalues of
the Yukawa coupling matrix multiplied by v, the vacuum expectation value of the standard
model doublet field. They are of the order of eV or less since the couplings of the bulk
modes to the brane fields are naturally suppressed, as already stated on the previous section.
Now, to simplify the discussion, we rotate the weak eigenstate neutrinos a to the ones
related to the weak eigenstates by the rotation U , i.e., a = Ua , where a = e, , and
= 1, 2, 3. We then have

Z
3 
X

(16)
m L BR (y = 0) + dy BL 5 BR + h.c. .
L=
=1

This reduces the problem to a consideration of three KK towers mixing with three neutrinos
which are related to the weak interaction eigenstates by the unitary matrix U defined
above. Each
tower is characterized by its parameter defined in analogy with Section 2
as 2 m R. After diagonalization of the mass matrices in each tower, each standard

462

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

neutrino can be written as a coherent superposition of the three different towers of mass
eigenstates:
a =

3
X

Ua =

=1

3 X
X
=1 k=0

Ua

1
k .
Nk

(17)

This expression generalizes Eq. (7). It is now clear that the three flavour oscillations will
correspond to the oscillations among the three towers. In this regards, the explanation to
neutrino puzzles is not any more described in terms of three single neutrinos, as in the usual
case; instead all the KK modes can contribute (unless all the KK excitations decouple from
the spectrum).
To proceed further, let us define the partial transition probabilities
p (L) h (L)| (0)i h (L)| (0)i.

(18)

Notice that the diagonal component p may be interpreted as the survival probability of
, and it takes the form of Eqs. (9) and (12) in that case. It of course involves the s and
not the flavor eigenstates as is obvious. The transition probability among standard flavours
can be written in terms of the p s as
X

Ua
Ub Ub
Ua p .
(19)
Pab =

Neglecting all CP phases we may expand Eq. (18) to the form

X
sin2 2z (2n 2k )
.
p = 1 2
2 ( )N 2 ( )
Nn

k
k,n=0

(20)

Now, we are ready to address the oscillation problem. Our approach will be as follows:
we will first select the parameter range that provides overall reduction required to explain
the solar and atmospheric data, and then ask whether for the same range of parameters we
can explain the observed oscillation between to e reported by LSND. Without loss of
generality, we can assume the hierarchy 1 < 2 < 3 . We consider three possible scenarios:
(i) 1,2,3  1;
(ii) 1,2  1 6 3 , and
(iii) 1  1 6 2,3 .
The case where all a > 1 is already ruled out since, as we discussed earlier, it cannot
explain the solar neutrino data without implying that the extra dimensions be too large. We
therefore do not discuss this case. In the discussion of our results, the following expressions
for the partial transition probabilities will be very useful: If ,  1, then

 





1
L
2
2
2
2 2 2
2
1m + + c(z) + c (z) + s (z) ,
(21)
cos
p =
2E
2 2
with 1m2 = m2 m2 ; 2 = 1 + 2 2 /6 and z = L/2ER 2 as before, and where we have
introduced the functions



 X

1 cos(zn2 )
c(z)
.
(22)
=
s(z)
n2 sin(zn2 )
n=1

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

463

It is simple to check that the diagonal term of Eq. (21) reduces to Eq. (9). For  1 <
we get
p =

1
2
c (z) + 2 c(z) c (z) + s(z) s (z) ;
2

(23)

where we now used c (z) = cos(2 )( ) sin(2 )( ) and s (z) = sin(2 )( ) +

cos(2 )( ) with = z 2 /2, and and as defined in the previous section, we

should stress that all those functions have a similar deep behaviour. And, finally, for , >
1 we have
p = c (z)c (z) + s (z)s (z).

(24)

Again, it is straightforward to check that the diagonal component of this equation reduces
to Eq. (12).
3.1. Case I:  1
Substituting Eq. (21) into (19), we get, to leading order in


X
2 L
2
1m + O(2 ).
Pab = ab 2
Ua Ub Ub Ua sin
4E

(25)

Therefore, as expected, to this order, we obtain the standard expression for the transition
probability well known for the three neutrino case. It is clear that in this scenario, solar
and atmospheric data can be explained as in the usual three flavour neutrino models by
adjusting the spacing of the different s. For the solar neutrino puzzle, one may either
use MSW or VO solution depending on how much fine tuning one is willing to tolerate.
For intermediate values where L/4E  1/1m2 , the leading corrections in become
important. Expanding Pab up to order 4 , by introducing Eq. (21) and neglecting the
standard oscillatory term we found
 X


2
X
2 2
c(z) +
|Ua |2 2
Ua Ub 2 I (z);
(26)
Pab ab 1 2
6

where we have denoted


 4
2

+ c2 (z) + s 2 (z)
c(z) .
I (z) =
36
3

(27)

Notice that the first term between parenthesis on Eq. (26) contains the lower order
correction to the standard survival probability in Eq. (25), while the second term, of order
4 will be only relevant for flavour transitions. Taking the average on the last equations,
and using that c = 0 and c2 + s 2 = 4 /90, we get



2
2
2 X
7 4 X

|Ua |2 2 +
Ua Ub 2 .
(28)
PSab ab 1
3
180

This last expression generalizes that presented in Ref. [12], where only the contribution of
3 was assumed.

464

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

It is clear that understanding LSND results in this case would require that first
undergo a transition to the lower KK modes of the bulk neutrinos and then back to the
e . One might hope that if we adjusted the extra dimension radius to be small enough
R 1 0.22 eV, then one would get the right mass difference to fit LSND data. The key
question then is to see whether the transition rate comes out right. For this purpose, we
consider up to the lowest non-vanishing order in for Pe given in (26), which after
using the unitarity of U and the fact that 32 12 = 21m2atm R 2 turns out to have the form

2
Pe sin2 2e 1m2atm R 2 I (z),

(29)

with sin2 2e 4(U3 Ue3 )2 . It is straightforward to check that Pe (L = 0) = 0 by using

the identities c(0) = 2 /6 and s(0) = 0. From the limits obtained by the CHOOZ and
Palo Verde Collaboration, we know that |Ue3 |2 < 0.03 [16], and assuming |U3 |2 = 0.5
for maximal mixing, the largest optimistic value for the mixing angle we may take is about
sin2 2e = 0.06. By fixing L/E as for LSND, I (z) becomes only a function of R (since
z = L/2ER 2 ). In Fig. 7 we have plotted I (R) versus R. From this figure we see that for
a reasonable large R ( 10 eV1 ), the function I (R) has very small values already. The
combined effect with the factor 1m2atm R 2 < 1 will reduce Pe even more. Clearly, Pe
will be maximal for the larger possible value of R. However, the largest
allowed radius
q
that permit us to still be confident in our approach is about 1/R

1m2atm 0.06 eV.

A numerical calculation with these inputs gives for LSND Pe = 3 104 , which is
one order of magnitude smaller than the observed anomaly in the beam of LSND.
Since higher order corrections in to the probability are unlikely to introduce enough
enhancement (a factor of 10 is needed), we conclude that this scenario yields a too small
probability for e transition to explain the LSND observations.

Fig. 7. The function I (R) versus R for the experimental ratio L/2E as in LSND. The window on the
upper right shows an amplification of the region for large R, where I (R) 6 103 .

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

465

3.2. Case II: 1,2  1 6 3

To proceed with this case, it is convenient to write the survival probability Paa in the
following form:
X
X
2
2
4
(Ua Ua )2 p + 2Ua3
Ua
p3 + Ua3
p33 .
(30)
Paa =
,=1,2

=1,2

The first term in the above equation can be written to leading order in the small s as
follows:


X




L
2 2
1m221 + O 2 ,
(Ua Ua )2 p 1 Ua3
sin2 2aa sin2
(31)
4E
,=1,2

where sin2 2aa = 4(Ua1Ua2 )2 . This term by itself can explain the solar neutrino deficit.
Clearly, the standard two neutrino oscillation expression is recovered to this order if we
set Ue3 0 to satisfy the bounds imposed by the reactor data [16]. This will make the
contributions of the last two terms in the survival probability (30) negligible and the first
oscillatory term can then be used to solve the solar neutrino puzzle. One can of course use
the MSW mechanism to solve the solar neutrino problem or the vacuum oscillation. The
constraint on the parameter space is that the 1m212 be appropriately adjusted. This does
not impose any condition on the bulk radius and can be satisfied by the initial choice of
parameters (the Yukawa couplings) in the theory. For instance, M 10 TeV can lead to
the MSW-type mass differences. Let us note that, if 1m212 105 eV2 , then, we will have
|12 22 | 105 R 2 (eV cm)2 .
Now let us consider P . From Eq. (30), we see that there are several contributions to the
atmospheric neutrino deficit. First, there is the contribution of the towers labeled by 1,2 ,
which is oscillatory, although it can not be identified as in the usual oscillations.
Then, there is also the contribution induced by the term p33 which is of the form of (12).
Finally, there is also a mixed term, p3 .
To proceed with the full discussion, let us consider two cases:
Case (i): U3 = 0
If U3 = 0, the last two contributions are removed and we get the survival probability



2
2
c(z) .
(32)
1m221R 2 + 12
P 1 4 2U2
6
One might then hope that the oscillations into the lower KK
qmodes with mass differences

of about 1/R will do the job provided we choose 1/R 1m2atm to match the data. In
this case, the atmospheric muon neutrinos oscillate into the sterile neutrinos, a possibility
which has its characteristic tests. 2 For this solution to work, one needs to assume 1
1 so that one gets maximal mixing. This in turn means that, to explain both solar and
atmospheric neutrino data, an almost degeneracy 1 2 must be maintained. This alters
2 We realize that from an experimental point of view, this looks less likely to be realized in nature [17]; we take
a somewhat liberal view of the situation and still contemplate this as a viable possibility.

466

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

our explanation of the solar neutrino deficit since, now, these new contribution to it (see
the O( 2 ) terms in Eq. (28)), become more and more important and in fact of order one,
making it hard to understand the solar neutrino deficit, since c(z) 1. We will, therefore,
consider this case as an unfavorable one for understanding the neutrino puzzles.
Case (ii): U3 6= 0
Turning to the case where U3 6= 0, if we keep 1,2  1 (to maintain our understanding
2 to the transition probability
of the solar neutrino data) then, the corrections of order 1,2
become almost negligible and the dominant contribution to Pab then must come from 3
corrections. This yields


L
1m221
Pab (ab Ua3 Ub3 )2 4(Ua1 Ub2 )2 sin2
4E
+ 2Ua3Ub3 (ab Ua3 Ub3 ) c3 (z) + (Ua3 Ub3 )2 p33 .

(33)

Notice that p33 in the last equation has the same form as Eq. (12), and the function c3 is
the one defined above with the same sharp behaviour as p33 . Then specializing Eq. (33) to
the in the atmospheric case, we get




2 2
2
2
4
+ 2U3
p33 .
(34)
1 U3
c3 (z) + U3
P 1 U3
From our naive analysis in the previous section we may expect that this equation can
account for the atmospheric data without too much trouble as long as 32 /R 102 eV
or so to make the width of the slope larger than the experimental parameters and to avoid
the over washing of the flux. Indeed, it has been checked numerically in Ref. [12] that
2 0.4 and 2 /R 0.02 eV.
the atmospheric neutrino data can be fitted in this case if U3
3
It is worth mentioning that a mixed explanation could be also possible, where the three
towers contribute equally to provide atmospheric oscillations, but we will not discuss this
case here.
An important point to note however is that due to the features of the function c3 (z), the
atmospheric neutrino data will not exhibit the oscillatory behaviour that one would expect
in the conventional two neutrino models.
Lets turn now to LSND results. By taking that Ue3 0 as suggested by solar neutrino
and reactor data [16], the transition probability reduces to
X
U Ue U Ue p .
Pe =
,=1,2

This reflects the fact that the same argument that suppresses the contribution of the third
KK tower to Pee also does the same for Pe . As Ue3 = 0 remove the contributions of the
third tower, we may expand Pe to the lowest order by the same expression (26) used in
the previous case, which is now given as

2
(35)
Pe sin2 2e 1m2solR 2 I (z),
where now sin2 2e = 4(U2 Ue2 )2 , and I (z) as given in Eq. (27). This resembles our
former expression in (29). In order to estimate the magnitude of this contribution, note
that the atmospheric neutrino fitting requires 32 /R 102 eV. For 32 110, this implies

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

467

that R 2 104 106 eV2 . Thus if 1m2sol corresponds to the MSW solution (large or small
angle), then 1m2solR 2 ' 1 and one obtains Pe = sin2 2e I (z). From Fig. 7, we see again
that for relevant values of R and L/E, the function I (z) takes very small values ( 104 ),
making the Pe very small.
Notice that this argument is independent of the way we get the atmospheric deficit. We
could also imagine, for instance, that a small value of Ue3 is allowed, and then calculate
the leading correction to the above expression. However, it turns out to be of the form
P
(U3 Ue3 ) =1,2 U Ue 2 (c(z) 2 /6), where last term between parenthesis is already
smaller than 104 by itself.
3.3. Case III: 1  1 6 2,3
In this case, the transition probability can be written as
X
X
Ua Ub p1 +
Ua Ub Ub Ua p .
Pab = (Ua1 Ub1 )2 p11 + 2Ua1Ub1
=2,3

,=2,3

(36)
The simplest possibility is to let the first term in above equation is be responsible for solar
neutrino oscillations into bulk neutrinos as explained by Eq. (9) [11]. This requires that the
radius be fixed to be about 1/R 103 eV. In this case, to keep the e from mixing too
much with the other neutrinos and generate further reduction of the survival probability
for the solar neutrino, we choose Ue1 1. This essentially decouples the first tower from
the others. As a simple approximation, if we assume that Ue,2,3 = 0, then clearly this
suppresses the oscillations from into e , making it difficult to understand the LSND
results.
Moreover, in this scenario, even the explanation for atmospheric data seems to run into
some trouble. As Ue1 1, it is not unreasonable to conclude based on orthogonality that
U1 0. This implies that
X
(U U )2 p ,
(37)
P =
,=2,3

where all p are given as in Eq. (24). As a rough approximation, if we assume that
the partial transition probabilities are all almost of the same order (as they seem to be
numerically), say p33 , and use orthogonality once more to get P p33 . Therefore, we
get maximal contribution from the large deficit generated by p33 . A hierarchical p will
not help with this, since either we get equal mixing among them or one of them has a
dominant contribution. In any case, the leading order will combine both things, a large and
fast developing slope, and a large mixing angle. Nevertheless, based on the results of [12],
where they found some scenarios where those two ingredients come together (although
for small ), it is still hard to rule out the scenario without a careful numerical analysis
of the data. In any case, the simplest requirement to get a deficit not too large compared
2 /R 102 , which
with the experimental data, imposes a tuning of the main parameters 2,3
2 10.
combined with the condition for R to understand the solar neutrino data fixes 2,3

468

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

In order to understand the LSND results, we must allow for small Ue2 and/or U1 . Since,
in the present scenario, the solar neutrino deficit is being explained purely by oscillation
into the bulk, so, we get the following picture: The first tower contributions are too small
for LSND, they are even smaller than the size of those considered on the previous cases,
while the oscillations involving the other towers are mainly destructive due to large values
of involved. The conversion will not occur but for a small contribution related with the
small part of those towers contained in e . However, as one may suspect, the contributions
are all proportional to the mixing angles (U Ue )2 , which can not be larger than 102 .
So, the question is whether the behaviour of p can help. To analyze this point we notice
that
X
U Ue Ue U (1 p ).
(38)
Pe
,=2,3

Note that one can estimate the value of (1 p ) from the the information on atmospheric
neutrino deficit as follows. First point is that all p s are of same order and therefore,
one can write the Pe (U1 Ue1 )2 (1 Psurv ), where Psurv ( ) is the same function which
for = atmos gives the survival probability of muon neutrinos in the atmospheric data.
Thus, Psurv (atm ) 0.5. The value of corresponding to the LSND case is however much
smaller due to smaller oscillation distance; therefore to estimate Pe we must use the value
of which is much smaller and find the corresponding Psurv . A numerical evaluation gives
(1 Psurv) 102 . The next question is how large the mixing parameters are. In order not
to make Pee in atmospheric neutrino data different from one, U2 must be much smaller
than 1. On the other hand to fit LSND observations, we need to have U1 0.5. It therefore
appears difficult to accommodate the LSND observations in this case.

4. Concluding remarks
The analysis of the present paper shows that in minimal models for neutrino masses in
theories with large extra dimensions, it is not possible to get a simultaneous explanation of
gross overall oscillations needed to understand the solar, atmospheric and the LSND data.
Fitting the overall deficit in the atmospheric and solar neutrino data for various ranges of
the input parameters, the largest conversion probability, Pe , that we find, is around 104 .
This is too low to explain the LSND observations. One must therefore invoke new physics
beyond the standard model in the brane [8,18] or new physics outside the brane [15] for a
simultaneous understanding of all observed neutrino data.
The oscillation pattern is governed by the dimensionless parameters . Three of them
arise in the models under consideration and we get the following picture:
(i) 1,2,3  1: in this case, solar and atmospheric neutrino data are understood as in the
case of four dimensional models and the hope that the presence of the bulk neutrinos
with an appropriate KK excitation scale could provide an understanding of the LSND
data is not realized.
(ii) 1,2  1 6 3 : solar neutrino data is provided as in the four dimensional models but
atmospheric data is explained by to bulk oscillation once the appropriate values

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

469

q
of the bulk radius is chosen i.e., either 1 2 and 1/R 1m2atm or 32 /R 102 .
In this case, the atmospheric neutrino flux does not oscillate as a function of L/E, a
feature distinct from the conventional two neutrino oscillation picture.
(iii) 1  1 6 2,3 . Both, solar and atmospheric data are explained by bulk
oscillations. Therefore 1/R 2 1m2sol 103 eV with matter effects for solar and
2 10. Again, the L/E behaviour of the atmospheric flux is not oscillatory.
2,3
(iv) 1,2,3  1. There is no explanation for solar neutrino data in this case. This
parameter range is therefore ruled out.

Acknowledgements
The work of RNM is supported by a grant from the National Science Foundation
under grant number PHY-9802551. The work of APL is supported in part by CONACyT
(Mxico).

References
[1] I. Antoniadis, Phys. Lett. B 246 (1990) 377;
I. Antoniadis, K. Benakli, M. Quirs, Phys. Lett. B 331 (1994) 313;
P. Horava, E. Witten, Nucl. Phys. B 460 (1996) 506;
P. Horava, E. Witten, Nucl. Phys. B 475 (1996) 94;
J. Lykken, Phys. Rev. D 54 (1996) 3693;
K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 436 (1998) 55;
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263;
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Rev. D 59 (1999) 086004;
I. Antoniadis, S. Dimopoulos, G. Dvali, Nucl. Phys. B 516 (1998) 70;
N. Arkani-Hamed, S. Dimopoulos, J. March-Russell, hep-th/9809124.
[2] For early proposals see: V.A. Rubakov, M.E. Shaposhnikov, Phys. Lett. B 152 (1983) 136;
K. Akama, in: K. Kikkawa, N. Nakanishi, H. Nariai (Eds.), Gauge Theory and Gravitation,
Proceedings of the International Symposium on Gauge Theory and Gravitation, Nara, Japan,
August 2024, 1982, Lecture Notes in Physics, Vol. 176, Springer-Verlag, 1983, p. 267;
M. Visser, Phys. Lett. B 159 (1985) 22;
E.J. Squires, Phys. Lett. B 167 (1986) 286;
G.W. Gibbons, D.L. Wiltshire, Nucl. Phys. B 287 (1987) 717.
[3] M. Gell-Mann, P. Ramond, R. Slansky, in: P. van Niewenhuizen, D.Z. Freedman (Eds.),
Supergravity, North Holland, 1979;
T. Yanagida, in: O. Sawada, A. Sugamoto (Eds.), Proceedings of Workshop on Unified Theory
and Baryon number in the Universe, KEK, 1979;
R. N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
[4] N. Arkani-Hamed, M. Schmaltz, Phys. Rev. D 61 (2000) 033005.
[5] E. Witten, Invited talk at the Neutrino2000 Conference, Sudbury, Canada, June, 2000.
[6] K.R. Dienes, E. Dudas, T. Gherghetta, Nucl. Phys. B 557 (1999) 25;
N. Arkani-Hamed, S. Dimopoulos, G. Dvali, J. March-Russell, hep-ph/9811448.
[7] R.N. Mohapatra, S. Nandi, A. Prez-Lorenzana, Phys. Lett. B 466 (1999) 115.
[8] R.N. Mohapatra, A. Prez-Lorenzana, Nucl. Phys. B 576 (2000) 466.
[9] C. Burgess, L. Ibanez, F. Quevedo, Phys. Lett. B 447 (1999) 257.

470

[10]
[11]
[12]
[13]

[14]

[15]
[16]
[17]
[18]

R.N. Mohapatra, A. Prez-Lorenzana / Nuclear Physics B 593 (2001) 451470

N. Arkani-Hamed, M. Schamaltz, in: Beyond 4-D Workshop, ICTP, Trieste, 2000.

G. Dvali, A.Yu. Smirnov, Nucl. Phys. B 563 (1999) 63.
R. Barbieri, P. Creminelli, A. Strumia, hep-ph/0002199.
A. Faraggi, M. Pospelov, Phys. Lett. B 458 (1999) 237;
G.C. McLaughlin, J.N. Ng, Phys. Lett. B 470 (1999) 157;
G.C. McLaughlin, J.N. Ng, nucl-th/0003023;
A. Ioannisian, A. Pilaftsis, hep-ph/9907522.
A. Das, O.C.W. Kong, Phys. Lett. B 470 (1999) 149;
Y. Grossman, M. Neubert, Phys. Lett. B 474 (2000) 361;
A. Lukas, A. Romanino, hep-ph/0004130;
E. Ma, M. Raidal, U. Sarkar, hep-ph/0006046.
A. Ioannissian, J.W.F Valle, hep-ph/9911349.
CHOOZ collaboration, M. Apollonio et al., Phys. Lett. B 466 (1999) 415;
Palo Verde Collaboration, F. Boehm et al., hep-ex/0003022.
See the invited talk by T. Kajita, Neutrino2000 Conference, Sudbury, Canada, 2000.
D. Caldwell, R.N. Mohapatra, S. Yellin, in preparation.

Nuclear Physics B 593 (2001) 471504

www.elsevier.nl/locate/npe

Phenomenological study of a minimal superstring

standard model
G.B. Cleaver a,b, , A.E. Faraggi c , D.V. Nanopoulos a,b,d , J.W. Walker a
a Center for Theoretical Physics, Department of Physics, Texas A&M University,

College Station, TX 77843, USA

b Astro Particle Physics Group, Houston Advanced Research Center (HARC),

The Mitchell Campus, Woodlands, TX 77381, USA

c Department of Physics, University of Minnesota, Minneapolis, MN 55455, USA
d Academy of Athens, Chair of Theoretical Physics, Division of Natural Sciences,

28 Panepistimiou Avenue, Athens 10679, Greece

Received 7 October 1999; revised 25 May 2000; accepted 6 September 2000

Abstract
Recently, we demonstrated the existence of heterotic-string solutions in which the observable
sector effective field theory just below the string scale reduces to that of the MSSM, with the standard
observable gauge group being just SU(3)C SU(2)L U (1)Y and the SU(3)C SU(2)L U (1)Y charged spectrum of the observable sector consisting solely of the MSSM spectrum. Associated
with this model is a set of distinct flat directions of vacuum expectation values (VEVs) of nonAbelian singlet fields that all produce solely the MSSM spectrum. In this paper, we study the effective
superpotential induced by these choices of flat directions. We investigate whether sufficient degrees
of freedom exist in these singlet flat directions to satisfy various phenomenological constraints
imposed by the observed standard model data. For each flat direction, the effective superpotential
is given to sixth order. The variations in the singlet and hidden sector low energy spectrums are
analyzed. We then determine the mass matrices (to all finite orders) for the three generations of
MSSM quarks and leptons. Possible Higgs -terms are investigated. We conclude by considering
generalizations of our flat directions involving VEVs of non-Abelian fields. 2001 Elsevier Science
B.V. All rights reserved.

1. Minimal superstring standard models

Recently [13] we demonstrated that it is indeed possible for a string model [4,5] to have
exactly the minimal supersymmetric standard model (MSSM) fields as the SU(3)C
Corresponding author.

E-mail addresses: [email protected] (G.B. Cleaver), [email protected]

(A.E. Faraggi), [email protected] (D.V. Nanopoulos), [email protected]
(J.W. Walker).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 4 3 - 5

472

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

SU(2)L U (1)Y -charged matter content of its low energy effective field theory. We
propose that string models with this property be classified as minimal superstring standard
models (MSstr SM ) [3]. In our MSstr SM, decoupling of all MSSM-charged exotics from
the low energy effective field theory was accomplished by sets of vacuum expectation
values (VEVs) that eliminate the anomalous U (1)A endemic to several classes of string
models (in particular, those of bosonic lattice, orbifold, or free fermionic construction).
Besides restoring spacetime supersymmetry through cancellation of the FayetIliopoulos
(FI) D-term, these sets of VEVs also give FI-scale ( 4 to 7 1016 GEV) mass to the
MSSM exotics.
If the underlying initial state of the universe truly was an anomalous U (1)A string
model, then determination of the specific flat VEV direction chosen to cancel the FI D-term
was a result of non-perturbative dynamics. That is, the physically preferred flat direction
cannot be identified perturbatively. However, through perturbative means we can locate
and classify the possible flat directions most consistent with observed data. Classification
of non-Abelian (NA) singlet flat directions that produce the MSSM gauge group and matter
fields, while simultaneously decoupling all MSSM exotics, was performed in [3]. Based
on our stringent F -flatness constraints, we found three directions flat to all order, one
direction flat to 12th order, and around 100 remaining directions only flat to seventh order
or less.
The existence of free fermionic models with solely the MSSM spectrum below the
string scale reinforces the motivation to improve our understanding of this particular
class of string models. Both from the point of view of understanding the non-perturbative
dynamics, as well as improving the techniques that are needed in order to confront
the perturbative string models with the low energy experimental data. In this paper we
perform studies of the phenomenological features of these first four flat directions of the
FNY model of [4,5]. We explore the phenomenology of our MSstr SM flat directions
and investigate which (if any) of our four singlet directions appear most consistent with
observed phenomenological criteria.
We remark that phenomenological studies, similar to the one performed in this paper,
were done in the past for other three generation free fermionic models. The new features in
this paper are as follows. First, the FNY model is the first known example of a semi-realistic
string model, which produces solely the MSSM-charged spectrum just below the string
scale. Thus, for the first time such a phenomenological analysis is carried out in a minimal
superstring standard model. Second, and more importantly, in the phenomenological
analysis performed in this paper, we implement the systematic techniques for the analysis
of F and D flat directions that were developed over the last few years [3,611]. Relative to
the more primitive studies performed in the past, our study here has the advantage that it
incorporates in much of the analysis the non-renormalizable terms to all finite orders. In the
analysis we systematically decouple from the effective low energy field theory the fields
that become superheavy and their superpotential couplings. Furthermore, the solutions that
we study in this paper are flat to all orders. We emphasize that such an exhaustive analysis
is performed for the first time in a semi-realistic string model. Thus, our paper further
advances the methodology needed to confront potentially viable superstring models with

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

473

experimental data.
Our paper is organized as follows. In Section 2 we present a brief review of Z2 Z2
free fermionic models, the class from which the FNY model originates. In Section 3 we
review D- and F -flatness constraints. Our study and discussion of the phenomenology of
our four flat directions appears in Section 4. For each of the flat directions, we analyze
the three generation mass matrices for up, down, electron, and neutrino states, and study
the effective Higgs -terms. We also investigate coupling constant strength for high order
superpotential terms. We conclude Section 4 with study of additional F -flatness constraints
imposed when both fields in a vector-like pair (with opposite U (1) charges) acquire VEVs.
Lastly, in Section 5 we include some general discussion and overview of our singlet flat
directions. We briefly consider generalizations of them in which NA fields are also allowed
to take on VEVs.

2. Z2 Z2 free fermionic models

Constructing minimal superstring models, i.e., models with solely the MSSM spectrum
below the string scale, is clearly the coveted goal of superstring phenomenology. However,
it should be emphasized that the success of the FNY model in achieving this goal should
not be viewed as indicating that the model of Ref. [4] is the correct string vacuum. In
this respect it is important to understand that the FNY model belongs to a large class of
three generation free fermionic models, which possess an underlying Z2 Z2 orbifold
structure. Many of the issues pertaining to the phenomenology of the standard model and
supersymmetric unification have been addressed in the past in the framework of the quasirealistic free fermionic models. An important property of these models is the fact that
they produce three generation models with the standard SO(10) embedding of the standard
model spectrum. No other orbifold heterotic-string compactification has yielded a similar
structure. The FNY model should be viewed as a prototype example of a semi-realistic
free fermionic model. The success of the FNY model in producing solely the MSSM
spectrum below the string scale should then be regarded as providing further evidence
for the assertion that the true string vacuum is connected to the Z2 Z2 orbifold in the
vicinity of the free fermionic point in the Narain moduli space.
For completeness we recall the basic structure of the free fermionic superstring models.
The purpose is to highlight the fact that the free fermionic models correspond to a large
set of viable three generation models, which differ in their detailed phenomenological
characteristics. In this respect, the FNY model should be regarded as a representative
example.
A model in the free fermionic formulation [12,13] is defined by a set of boundary
condition basis vectors, and one-loop GSO phases, which are constrained by the string
consistency requirements, and which completely determine the vacuum structure of the
models. The physical spectrum is obtained by applying the generalized GSO projections.
The first five basis vectors of the Z2 Z2 free fermionic models consist of the NAHE
set [1419]. The gauge group after the NAHE set is SO(10) E8 SO(6)3 with N = 1

474

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

Table 1
The NAHE set

12

34

56

1,...,5

1,...,8

1
S

1
1

1
1

1
1

1
1

1, . . . , 1
0, . . . , 0

1
0

1
0

1
0

1, . . . , 1
0, . . . , 0

b1
b2
b3

1
1
1

1
0
0

0
1
0

0
0
1

1, . . . , 1
1, . . . , 1
1, . . . , 1

1
0
0

0
1
0

0
0
1

0, . . . , 0
0, . . . , 0
0, . . . , 0

Table 2
y 3,...,6

y 3,...,6

y 1,2 , 5,6

y 1,2 , 5,6

1,...,4

1,...,4

1
S

1, . . . , 1
0, . . . , 0

1, . . . , 1
0, . . . , 0

1, . . . , 1
0, . . . , 0

1, . . . , 1
0, . . . , 0

1, . . . , 1
0, . . . , 0

1, . . . , 1
0, . . . , 0

b1
b2
b3

1, . . . , 1
0, . . . , 0
0, . . . , 0

1, . . . , 1
0, . . . , 0
0, . . . , 0

0, . . . , 0
1, . . . , 1
0, . . . , 0

0, . . . , 0
1, . . . , 1
0, . . . , 0

0, . . . , 0
0, . . . , 0
1, . . . , 1

0, . . . , 0
0, . . . , 0
1, . . . , 1

spacetime supersymmetry, and 48 spinorial 16s of SO(10), sixteen from each sector
b1 , b2 and b3 . The three sectors b1 , b2 and b3 are the three twisted sectors of the
corresponding Z2 Z2 orbifold compactification. The Z2 Z2 orbifold is special precisely
because of the existence of three twisted sectors, with a permutation symmetry with
respect to the horizontal SO(6)3 symmetries. The NAHE set is depicted in Tables 1, 2
which highlights its cyclic permutation symmetry. The NAHE set is common to a large
class of three generation free fermionic models. The construction proceeds by adding
to the NAHE set three additional boundary condition basis vectors which break SO(10)
to one of its subgroups, SU(5) U (1), SO(6) SO(4) or SU(3) SU(2) U (1)2 ,
and at the same time reduces the number of generations to three, one from each of
the sectors b1 , b2 and b3 . The various three generation models differ in their detailed
phenomenological properties. These detailed properties depend on the specific assignment
of boundary condition basis vector for the internal worldsheet fermions {y, |y,
1,...,6 }.
However, many of the characteristics of the three generation models can be traced back
to the underlying NAHE set structure. One such important property to note is the fact
that, as the three generations are obtained from the three twisted sectors b1 , b2 and
b3 , they automatically possess the standard SO(10) embedding. Consequently, the weak
hypercharge, which arises as the usual combination U (1)Y = U (1)T3R + 12 U (1)BL , has
the standard SO(10) embedding. To date, of the three generation heterotic-orbifold models
that have been constructed, only the free fermionic models have yielded such a structure.
It should be emphasized that the success of free fermionic models in providing a viable
framework for reproducing the low energy phenomenology makes evident the need for
better understanding of this class of models, both from the phenomenological point of

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

475

view, as well as trying to understand the nonperturbative mechanism which fixes the string
vacuum. It should be further emphasized that because the free fermionic construction is
formulated at an enhanced symmetry point in the string moduli space, it is very natural to
expect that the true string vacuum should indeed be found in the vicinity of this point. The
structure of the Z2 Z2 orbifold, which underlies the free fermionic models, then seems
particularly suited for constructing three generation models. It is a very intriguing fact that
precisely where one would have expected to find the true string vacuum, indeed the most
realistic string models have been found. The further success of the free fermionic models
in producing models with solely the MSSM charged spectrum in the observable sector then
provides further evidence for the assertion that the true string vacuum is indeed located in
this vicinity. Elaborate exploration of the realistic free fermionic models is therefore vital.
The detailed massless spectrum of the FNY model and quantum charges are given
in Refs. [3,4]. Here we briefly recap the notation used in this paper. The massless
spectrum includes three generations from the sectors b1 , b2 and b3 . The Neveu
Schwarz (NS) sector produces the gravity and gauge multiplets, three pairs of electroweak
doublets {h1 , h2 , h3 , h 1 , h 2 , h 3 }, seven pairs of SO(10) singlets with observable U (1)
S12 , 23 ,
S23 , 13 ,
S13 , 56 ,
S56 , 0 ,
S0 , 4 ,
S4 , 0 ,
S0 }, and three
charges, {12 ,
56
56
4
4
scalars that are singlets of the entire four-dimensional gauge group, {1 , 2 , 3 }. The
states from the NS sector carry vector-like charges with respect to all unbroken U (1)
symmetries. The states from the sectors which are combinations of {1, b1,2,3,4, } +
2 are generically denoted by Vn , n = 1, 2, . . . . The states from the sectors with
some combination of {1, b1,2,3,4, } + are generically denoted by Hn , n = 1, 2, . . . .
A superscript s denotes when a respective H or V field carries only U (1) charges and is
a singlet for each of the non-Abelian gauge groups. The Vn and Hn states are vector-like
with respect to some U (1) currents but can be chiral with respect to others.

3. Flat MSSM directions of the FNY model

3.1. Spacetime supersymmetry and D- and F -constraints
In supersymmetric models, each chiral spin- 21 m fermion is paired with a scalar field
m to form a superfield m . The potential V () for the scalar fields receives contributions
from D-terms
X

m
Ta m ,
(3.1)
Da
m

where Ta is a matrix generator of the gauge group g for the representation m , and from
F -terms,
W
.
(3.2)
Fm
m
The scalar potential has the form
X
X
g Da Da +
|Fm |2 .
(3.3)
V () = 12
,a

476

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

For an Abelian group, U (1)i , (3.1) reduces to

X
2
Q(i)
Di
m |m | ,

(3.4)

m
(i)

where Qm is the U (1)i charge of m . When one Abelian group U (1)A is anomalous, 1
i.e., when the trace over the massless fields of its charge is non-zero,
Tr Q(A) 6= 0,

(3.5)

then (3.4) is modified by the appearance of an additional term  on the right-hand side,
where


gs2 MP2
Tr Q(A) .
192 2

(3.6)

Here gs is the string coupling and MP is the reduced Planck mass, MP MPlanck / 8
2.4 1018 . The FI D-term  results from the standard string theory anomaly cancellation
mechanism [20,21]. The universal GreenSchwarz relations, which result from modular
invariance constraints, remove all Abelian triangle anomalies except those involving either
one or three UA gauge bosons. The string anomaly cancellation mechanism breaks UA and,
in the process, generates . 2
Spacetime supersymmetry is broken when the scalar potential acquires a positivedefinite VEV. Thus, the FI DA -term  breaks supersymmetry near the string scale, with
V gs2  2 ,

(3.7)

unless a set of scalar VEVs, {hm0 i}, carrying anomalous charges Q(A)
m0 can cancel  by
(A)
making a contribution to D of equal magnitude but opposite sign:

A  X (A)
Qm0 |hm0 i|2 +  = 0.
(3.8)
D =
m0

Further, maintaining supersymmetry also requires any set of scalar VEVs satisfying
Eq. (3.8) to also be D flat for all of the non-anomalous Abelian and non-Abelian gauge
groups as well,

(i,) 
= 0.
(3.9)
D
The appearance of a given superfield m in the superpotential W imposes additional
constraints on flat directions via the associated F -term, (3.2). F -flatness (and thereby
supersymmetry) can be broken through an nth order W -term containing m when all of
the additional fields in the term acquire VEVs,
1 If initially the anomaly is contained in two or more U (1)
A,i , then the anomaly can always be rotated into a
single U (1)A by a unique rotation.
2 The form of the FI D-term was determined from string theory assumptions. Therefore, a more encompassing
M-theory [22,23] might suggest modifications to this FI D-term. However, recently it was argued that M-theory
does not appear to alter the form of the FI D-term [24]. Instead an M-theory FI-term should remain identical to
the FI-term obtained for a weakly-coupled E8 E8 heterotic string, independent of the size of M-theorys 11th
dimension.

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504


hFm i

W
m

n hi

477

(3.10)


hi
MPl

n3
,

(3.11)

where hi denotes a generic scalar VEV. If m also takes on a VEV, then supersymmetry
can be broken simply by hW i 6= 0. Therefore we also demand that hW i = 0 for each
individual term.
The higher the order, n, of an F -breaking term, the stronger the suppression of the
supersymmetry breaking scale below the string scale. This scale suppression normally
results from a product of two effects: (i) a factor (hi/MPl )n < 1 (where typically
1
MPl in weakly coupled models) and (ii) less than factorizable growth, with
hi 10
increasing n, of the worldsheet correlation function integral In3 , which is contained
within the non-normalizable (n > 3) coupling constants, n . By this we mean that In3 
(I1 )n3 for n > 4. In the weakly coupled case, F -breaking terms with orders as high as n
17 can generate supersymmetry breaking at an energy scale too far above the electroweak
scale. Note that for our flat directions, each hi/MPl contributes a suppression factor of
approximately 1/30 to (3.11).
As the string coupling increases (and the physics becomes less perturbative in nature)
the string scale can be significantly lowered. For strong coupling, the mass scale in
the denominator of (3.11) should be replaced by the string scale Mstr . The diminished
suppression from each hi/Mstr factor then requires that F -flatness be maintained to an
even higher order.
F -flatness can be broken by two classes of superpotential terms, those composed of:
(i) only the VEVed fields and (ii) the VEVed fields, and a single field without a VEV. 3
Obviously, F -flatness is guaranteed to a specific order n in W when neither class of terms
appears at order n or below. In [3], the three all-order flat directions, and the one 12thorder flat direction satisfying our exotic decoupling requirements were found by this
technique. While lack of the appearance of either class of term for a given D flat direction
is sufficient to guarantee F -flatness, this requirement is not necessary. The non-presence of
such terms can be relaxed. Several terms can appear without breaking F -flatness, provided
that the sum over all the terms in each hFm i, and in hW i, vanishes.
3.2. MSSM flat directions
Tables 3 and 4 in Appendix A form a review of the the four basic classes of VEV
directions (denoted FD1, FD2, FD3, and FD4) presented in [3]. These directions sustain
the MSSM gauge group and produce, in the low energy effective field theory, exactly the
three standard generations of MSSM matter fields and a single set of Higgs h and h fields
as the only MSSM-charged fields. All other MSSM exotics are decoupled, acquiring FIscale masses. FD1 is the root direction contained within the three others. That is, all of
S4 ,
S0 ,
S56 , H s , H s , H s , and H s , that take on VEVs
the fields, 12 , 23 , 4 , 40 ,
4
15
30
31
38
3 The first class of terms was referred to as type A and the second as type B in [9,10].

478

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

in FD1, do so likewise in FD2, FD3, and FD4. FD2, FD3, and FD4 each contain one
S0 , H s , and H s , respectively. FD1, FD2, and FD3 are flat
additional field with a VEV:
56
19
20
to all order, while F -flatness in FD4 is broken at twelfth order. The FayetIliopoulos scales
(that is, the overall scales of the VEVs) for these four flat directions are approximately
6.7 1016 GeV, 3.9 1016 GeV, 4.8 1016 GeV, and 6.7 1016 GeV, respectively.
The above charged fields are all vector-like for all Abelian groups, while the
corresponding H and V fields are not. Thus, a possible variation of this class of flat
S to each take on a VEV.
directions is to allow both a field and its vector partner
S
For example, in FD1 both 12 and 12 can acquire VEVs, so long as




 12 2 
S12 2 = 3hi2 ,
(3.12)
where is an overall scale factor for a given flat direction and is specified in Table 4 of
S12 picks up a VEV
Appendix A. Indeed, we examine flat directions varieties wherein
in each of FD1 through FD4; we refer to these respective modified directions as FD1V
0 receives a VEV alongside
through FD4V. 4 For FD2 we also consider the case where 56
0
0
0
S
56 . This particular variation is denoted FD2 . Thus, FD2 V signifies the presence of VEVs
S12 and 0 . Generally, the appearance of VEVs for both fields in a vector-pair
for both
56
results in new flatness constraints with often important phenomenological implications. We
discuss vector-pair constraints in Section 4.3.

4. Phenomenology of flat directions

Phenomenology of a model is generally vastly altered by the turning on of a flat
direction [13]. Further, the specific phenomenological modifications from that of the unVEVed model can vary substantially for different directions. We survey the respective
phenomenological features of our three all-order and our one twelfth order singlet flat
directions (the latter considered primarily for comparison) presented in [3] for the FNY
model. We investigate, in particular, the MSSM three generation mass hierarchies and the
effective -term(s) for each direction.
4.1. Field content and decoupling of FI-scale massive fields
The functions of our singlet flat directions were twofold. In addition to cancelling
the FI-term, we required each direction to decouple all exotic fields carrying fractional
electric charge (that is, one SU(3)C triplet/anti-triplet pair, four SU(2)L doublets, 16 NA
singlets, two SU(2)H doublets, and two SU(2)H 0 doublets) from the effective low energy
field theory. The fractional fields, denoted in (4.1) as Ei0 , are decoupled from the low
energy effective field theory of each flat direction through a generalized Higgs effect. (See
Appendix B.) That is, all of the fractional fields appear in various (relatively low) nth-order
superpotential terms containing n 2 flat direction field VEVs hXj i,
4 FD1V is therefore embedded in FD2V, FD3V, and FD4V.

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

Ei01 Ei02 hXj1 i

hXj2 Xj3 Xjn2 i

.
(MPl )n3

479

(4.1)

The flat direction VEVs hXi i are generally constrained to be around the FI scale, hi. 5
Thus, masses for the fields that are decoupled through renormalizable (i.e., n = 3) terms
are also expected to be around the FI scale, while masses of fields decoupled through
non-renormalizable (n > 3) order terms are suppressed below the FI scale by factors of
(hi/MPl )n3 . All of the fractionally charged exotics decouple through renormalizable
terms, except for the SU(3)C vector-like triplet pair. The triplets appear in a fifth order
mass term [1,2]. Thus, we expect the vector triplet to receive a mass smaller, perhaps, by a
factor of around 1/10 to 1/100.
Each of our four flat directions decouples not only the fractionally-charged fields,
but also the remaining MSSM-charged exotics with integer electric charge. Like their
fractionally charged counterparts, the six integer charged SU(2)L doublets receive mass
through third order terms. Various sets of additional NA singlets and NA hidden sector
fields are also decoupled by the flat directions.
The additional fields decoupled by each direction, along with the superpotential terms
responsible for the decoupling, are also specified in Appendix B. The massless and the
massive (decoupled) superfield content and related superpotential terms are first presented
for FD1 in this appendix. Following this, the alterations to mass eigenstates resulting from
replacement of FD1 by each of the other VEV directions are listed. In our root direction
FD1, 15 of the 44 electrically uncharged U (1)-charged NA singlets and one of the totally
uncharged fields, 1 , receive masses from third order terms; four more singlets receive
mass at fifth order; eight of the 30 NA hidden sector states with no electric charge become
massive from third order terms, four such fields from fourth order terms, and four from fifth
order terms. Several fields remain massless through at least seventh order; 21 of the U (1)c
), two uncharged
charged NA singlets, (including the right-handed neutrino singlets, N1,2,3
singlets, 2,3 , one SU(3)H triplet/anti-triplet pair, six SU(2)H doublets, and six SU(2)H 0
doublets.
S0 H s H s prevents more than one of its three associated
The renormalizable term
56 19 20
fields from taking on a VEV simultaneously. Hence the distinctions in VEVs between FD2,
s and H s become FI-scale massive,
FD3, and FD4. In FD2 (and all of its variations), H19
20
S0 and H s , and in FD4(V) they
while in FD3(V) the corresponding massive states are
56
20
S0 and H s . FD3 and FD3V also give FI-scale mass both to another SU(2)H doublet
are
56
20
and to another SU(2)H 0 doublet. A novel aspect of the FD4 and FD4V classes, perhaps
identifying them from the rest, is their rendering of a near FI-scale Majorana mass to a
s
through a seventh
right-handed neutrino singlet, N1c , and to the associated singlet field V31
order term,

0 s s s s c s
(4.2)
4 H15 H20 H30H31 N1 V31 .
5 An exception to this is the overall scale of the four -related VEVs, which is unconstrained, provided the
4
difference of the norms of these fields is around the FI scale. The overall scale for VEVs of vector-like pairs of
fields provides a similar exception.

480

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

A few other eigenstate differences between FD1 and others are also especially
significant. For example, in all non-V directions (that is FD1, FD2(0) , FD3, and FD4)
S12 is a massive eigenstate and 23 is a massless eigenstate. However,
S12 i = 0,
wherein h
S12 receives a VEV. This triggers
in the V directions, FD1V, FD2(0) V, FD3V, and FD4V,
S
the rotation of the the 12 and 23 eigenstates into the massless state
S12 (1) q

1
S12 i|2 + |h23 i|2
|h





S12
S12 23 23 ,

(4.3)

and into the orthogonal FI-scale massive eigenstate,

S12 M q

1
S12 i|2 + |h23 i|2
|h





S12 23 .
S12 +
23

(4.4)

S12 also transforms the massless Higgs eigenstate from simply

Acquisition of a VEV by
h h 1

(4.5)

into
h q

1
S12 i|2 + |hH s i|2
|h
31

s


S12 h 4 .
H31 h 1

(4.6)

S12 i, undetermined perturbatively, is extremely

In Section 4.3 we will see that the scale of h
relevant to both the inter- and intra-generational mass hierarchy.
Another important phenomenological effect appears in the FD20 and FD20 V directions.
0 is massive while H s is massless. However, the VEV
In all but these two directions, 56
15
0 in FD20 and FD20 V, rotates these fields into the massless eigenstate,
of 56
0 (1)
q
56

0 i|2 + |hH s i|2

|h56
15

s s

 0
0
56
56 H15
H15

(4.7)

and into the orthogonal massive eigenstate

0 M
q
56

1
0 i|2 + |hH s i|2
|h56
15

0  s

 0
s
H15
56 + 56
H15 .

(4.8)

0 also carries important hierarchical implications.

The VEV of 56

4.2. Superpotentials
We have computed the effective low energy superpotentials, denoted herein as W FD1 ,
0
0
W FD3 , W FD4 , W FD1V , W FD2V , W FD3V , W FD4V , W FD2 , and W FD2 V , for each of
the four singlet directions FD1, FD2, FD3, FD4, and six of their vector pair variations,
respectively. We present each superpotential up through sixth order in Appendix C.
The superpotential for FD1 is listed in total. For the remaining flat directions, only the
additional terms not present in W FD1 are given. The terms in each superpotential are
categorized according to whether they contain, in addition to singlet fields: (i) nothing else,

W FD2 ,

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

481

(ii) only standard MSSM fields, and/or right-handed neutrinos, (iii) both MSSM fields and
hidden sector non-Abelian fields, or (iv) only hidden sector non-Abelian fields.
The terms in each of our 10 low energy effective superpotentials are those originating
from seventh or lower order terms in the original un-VEVed superpotential, W o .
A specific term of order no in W o generates for a given flat direction, a term of effective
order ne = no nv , when nv of the fields in the W o -term acquire VEVs. Since we expect
the effective coupling constant coming from an no -order term in W o to be smaller than
that of a comparable ne th-order W o -term when nv > 0 (except perhaps when nv = 1 and
ne = 3), we refer to such a term as a suppressed effective ne th-order term in the flat
directions effective superpotential.
To remove some trivial redundancies, the terms we list in the various effective
superpotentials are only those for which the maximal number of possible VEVs are
realized. For each such term, the existence in the same superpotential of related higher
order terms containing fewer VEVs is implied. Consider a specific effective third order
s
s
s
s
s
H30
H31
iN1c H20
V31
. This term implies the
MSSM superpotential term for FD1: h40 H15
presence of a quartet of related effective fourth order terms, including for example,
s H s H s iN c 0 H s V s , several effective fifth and sixth order terms, and the seventh
hH15
30 31
1 4 20 31
s
s
s
s
s 6
H20
H30
H31
V31
.
order one, N1c 40 H15
For each flat direction, the decoupling of the associated FI-scale massive fields
drastically simplifies the corresponding superpotential in comparison to W o . Only a
handful of third through sixth order W o -terms survive field decoupling for any direction.
Large numbers of W o -terms do not appear until the seventh order. Relatedly, seventh order
W o -terms are the primary source of the variations among our ten low energy effective
superpotentials. It it for these reasons that we include up to seventh order W o contributions
to those ten superpotentials.
The small number of surviving third through six order W o -terms results in a division
between singlet, MSSM, and NA hidden sector terms at low order. For example, the only
mixed MSSM-hidden sector terms (henceforth, simply referred to as mixed terms) in
W FD1 , besides the many with seventh order W o origin, are three effective fifth order terms
appearing at sixth order in W o . Similarly, besides gaining five additional mixed terms from
seventh order in W o , FD1V only acquires two new effective fourth order terms. The two
latter terms have their origin at fifth and sixth order in W o . FD3 merely results in three
additional mixed effective fifth order term with seventh order origin. FD2, FD20 , FD2V,
FD20 V, FD4, and FD4V generate no additional mixed terms.
Effective low order, especially renormalizable, terms with unsuppressed couplings are
few in number. For example, in the FD1 superpotential, W FD1 , the singlet sector content of
W FD1 is just two unsuppressed renormalizable terms. FD1V offers a sole additional singlet
term, while the remaining flat directions provide for no others at all.
A general property of the flat direction superpotentials is, indeed, that their more
distinguishing terms owe their origin to seventh (or higher) order in W o . This implies a
6 A term containing a VEV of an FI-scale massive field does not imply higher order terms wherein the VEV is
replaced by the field.

482

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

type of strong stability for an MSSM field theory realized from the FNY model. That
is, the MSSM low energy effective field theory is quite robust against variation of the
MSstr SM generating VEVs that are flat to at least 12th order.
4.3. MSSM three generation mass matrices
Viable three generation mass hierarchies present very strong constraints that a realistic
model should satisfy. Whether or not the FNY model in particular can face this challenge
remains to be seen. Even models in the close neighborhood of a candidate solution may
likely not be able to satisfy all mass criteria. Here we investigate the structure of the three
generation up, down, electron, and neutrino mass matrices for each of our singlet flat
solutions. First we will consider the up, down, and electron matrices and then separately
discuss the Dirac and Majorana matrices for neutrinos.
4.3.1. Quark and lepton masses
The root of all of our singlet flat directions, FD1, presents a respectable start to viable
quark and electron mass matrices. When respective components h 1 and h3 of the Higgs
pair h and h acquire VEVs, FD1 gives an unsuppressed renormalizable mass to one
generation, via the terms




(4.9)
g h 1 Q1 uc1 + L1 N1c + gh3 Q3 d3c + L3 e3c ,

where g gs 2 is the physical four-dimensional gauge coupling constant. These four

terms are, in fact, the only surviving renormalizable W o MSSM-class terms. Identifying
the bottom quark and tau lepton as the most massive down-like and electron-like states, we
see from Eq. (4.9) that the phenomenologically successful relation [25], mb = m at the
unification scale, is maintained. In this model all the heavy generation Yukawa couplings
are obtained at the cubic level of the superpotential, which differs from the case in some
other free fermionic models in which the bottom quark and tau lepton Yukawa couplings
necessarily arise from nonrenormalizable terms [17,18]. Note, however, the problematic
property of this model that the top and bottom quark mass eigenstates do not appear in the
same SU(2) doublet. Instead, the top appears in a doublet with some linear combination
of the down and strange quarks. Our study indicates that MSSM models involving only
singlet flat directions generically have this unwanted feature. To date, we have found no
singlet direction exceptions to this. In contrast, as we will show in [36], there are nonAbelian MSSM-generating flat directions for which top and bottom quark mass eigenstates
correctly appear in the same SU(2) doublet.
The additional, non-renormalizable mass terms for the up-, down-, and electron-like
states resulting from each of the other singlet flat directions appear in the mass matrices
Muc ,Q , Md c ,Q , Mec ,L . The components mij of these matrices are defined by the convention
X



T
mij fic Fj = f1c , f2c , f3c Mf c ,F F1 , F2 , F3 ,
(4.10)
i,j

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

483

where (f c , F ) ranges over (uc , Q), (d c , Q), (ec , L), and (N c , L). 7 In these matrices h1
and h3 are two components of the physical Higgs mass eigenstate h. h 1 and h 4 H34 are

the parallel for h.

The zeros in these matrices are valid to all finite orders. The terms that are generated by
(r)
a particular flat direction are marked by the superscripts (r) of the coupling constants n .
The given mass term is also generated by all flat directions that contain the one specified.
For instance, an r = 1V superscript indicates the associated mass term appears for FD1V
and for all flat direction in which FD1V is embedded. Likewise, a term with superscript
r = 20 appears for both FD20 and FD20 V. The coupling constant subscript n indicates that
the given mass term appears at nth-order in W o .
Muci ,Qj =

(
(1V)
h0 ch0 8

h 1 g

s Hs i
h23 40 H15
30

5
MPl

(
(1V)
h0 ch0 7

s Hs i
h23 H15
30

4
MPl

(
(1V)
h 4 4

s i
hH30

MPl

(20 V)

+ h 1 12

S 2 23 0 4 H s 2 H s 2 i
h
15
30
12
56

(4.11)

9
MPl

Notice that several terms in our mass matrices contain doublets denoted as either h0 and
h 0 . These are linear combinations of SU(2)L doublets generically defined as,


1
h12 ih1 + lh h23 ih3
ch0

s

1
S 
0
12 h1 + lh H31
h 4 ,
h
ch0

h0

and

(4.12)
(4.13)

q
p
S12 i|2 + |lh hH s i|2 . The
where (ch0 )1 1/ |h12 i|2 + |lh h23 i|2 and (ch0 )1 1/ |h
31
coefficients lh and lh are defined below.
For generic values of lh (lh ), h0 (h 0 ) is not a mass eigenstate. Rather, lh 6= 1 (lh 6= 1)
results in both massless and FI-scale massive components for h0 (h 0 ). Only in the lh 1
7 The component indices, i and j , of our mass matrices (4.11), (4.16)(4.18) carry the same boundary sector
interpretation as the fields themselves. Thus, the top quark mass term appears in position (1, 1) in the up-class
matrix, rather than in position (3, 3).

484

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

(lh 1) limit does h0 (h0 ) become an eigenstate, with an FI-scale mass. In the respective
limit for each,

h0 hM p
h 0 h M q

1
|h12

i|2

+ |h23
1

i|2

S12 i|2 + |hH s i|2

|h
31

h12 ih1 + h23 ih3 ,

(4.14)

s


S12 h 1 + H31

h 4 .

(4.15)

hM and h M are orthogonal to the massless eigenstates h and h given in (4.25), (4.26).
For a given mass term, lh (lh ) is determined by the ratio of the coupling constants for the
contributions from h1 and h3 (h 1 and h 4 ). Because the terms in which h0 and h0 appear are
all seventh order or higher, computation of the related lh and lh is extremely non-trivial.
(See Section 4.3.5.) However, the symmetries of the worldsheet charges of the respective
fields in both the h0 down and electron mass terms and in the h0 up and neutrino mass terms
strongly suggest lh = 1 and lh = 1 in all cases. If this is true (as we suspect), then all h0
and h 0 terms become decoupled from the mass matrices.
S12 , FD1V provides additional up mass terms, but no further
By giving a VEV to
terms for downs or electrons. To the up matrix (4.11) FD1V contributes: (i) a fourth order
and
diagonal term in m3,3 for the second generation, involving the h 4 component of h;
(ii) seventh and eighth order off-diagonal terms in m1,2 and m2,1 . The seventh and eighth
order off-diagonal terms involve h 0 . Only if the associated lh 6= 1 will the respective terms
contain the massless h field and contribute to the matrix. Should this be the case, then
(as we will argue below) physical constraints would most likely require the h 1 component
of h to strongly dominate over the h 4 component. If these off-diagonal terms are nonzero, then they offer a first generation up mass of order m1,2 m2,1 /m1,1 . This implies the
dominant mixing would be between the first and third generations which is interesting and
may have testable phenomenological implications. FD20 contributes the only additional up
mass term: a twelfth order second generation term that, based on (i) above can be ignored.
Thus, the generational up mass ratios are identical for FD1V, FD2V, FD20 V, FD3V, FD4V.
For the remaining directions, only the top quark receives mass

Mdic ,Qj

2 s

s
s
s
s
0
0
0

0
(20 ) h56 H15 H31 i
(20 ) h23 56 4 H15 H30 H31 i
= h ch0 7
h

0
,
1
10
7
4

M
M
Pl
Pl

0
0
h3 g
(4.16)

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

s
0
0 s
0 0 (20 ) h56 4 H15 H31 i
0
h
c

0
h 8

5
MPl


0
0 s 2H s H s i
0 ) h23 H

(2
56
4
15
30
31

h1 10
7
.
=
MPl
0
 0
2 s

s
s
2
0
0
S

(20 V) h12 23 56 4 H15 H30 H31 i

+ h3 12

MPl

0
0
h3 g

485

Meic ,Lj

(4.17)

With the exception of the trivial twelfth order FD20 V electron mass (which the tenth
order FD20 term dominates over) all first and second generation down and electron mass
terms are derived from FD20 . Thus, the d and e mass ratios are identical for FD20 and
FD20 V. In the other directions only the bottom and tau fields receive mass. If h0 contains
a massless h component, then FD20 produces a seventh order mass term for a second
down generation via m2,1 in Mdic ,Qj and an eighth order mass term for a second electron
generation via m1,2 in Meic ,Lj . In this case then FD20 also begins to provide for both
types of first generation masses through tenth order terms m2,2 in Mdic ,Qj and Meic ,Lj .
However, while these mass terms might suggest a first generation mass, they are insufficient
because (if the seventh order terms exist) they simply rotate the second generation d c and
e eigenstates, respectively. On the other hand, if h0 is a massive eigenstate, then it is the
m2,2 terms that are responsible for the charm and muon masses.
Our up, down, and electron mass matrices indicate that the better phenomenology is
clearly found along the flat direction FD20 V. Since FD20 V contains both FD1V and FD20 ,
all of the terms in the (4.11), (4.16), and (4.17) appear in this direction. However, as we
shall see in Section 4.4, there is a cost for this superior phenomenology: allowing both
S12 ) and (
S0 , 0 ) to acquire VEVs offers new
components of each vector pair (12 ,
56
56
dangers to low order F -breaking. Still, phenomenologically consistent resolutions to this
problem do appear possible. As we discuss in Section 4.3.4, part of the answer is suggested
independently by our mass matrices and the Higgs fields components. Before analysis of
the Higgs fields though, we investigate the neutrino mass matrices.
4.3.2. Neutrino masses
Let us now consider the three types of neutrino mass terms: Majorana doublet terms
mLi Lj , Dirac terms mLi Njc , and Majorana singlet terms mNic Njc . Simply by conservation
of gauged Abelian charges, we can show that no Majorana doublet terms appear for any of
our flat directions. This is favorable for a good seesaw mechanism.
As we have already shown in (4.9), FD1 provides for a renormalizable third generation
Dirac mass term for m1,1 from g h 1 L1 N1c . To this, FD1V would add two diagonal terms:
a trivial fifth order minor perturbation to m1,1 (containing h 4 ), and a twelfth order
contribution to m3,3 via h0 . (See the Dirac mass matrix (4.18) below.) Thus, FD1V would
imply significant mass difference between each generation of neutrinos while keeping
the first generation massless. FD20 V would raise the second generation scale through its

486

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

addition of a sixth order h 4 contribution to m3,3 , and also supplies a first generation mass
via a 14th order m2,2 term. 8 FD4V keeps the first and second generation mass scale
distinction by only providing trivial 16th and 22nd order perturbations to FD1Vs m2,2
and m3,3 terms, respectively. However, should h0 have a h component, then FD4 yields
mixing between the third generation and that associated with m3,3 via (i) a twelfth order
contribution to m1,3 and (ii) a thirteenth order contribution m3,1
MNic ,Lj =

h 1 g

h0 c 0 (4V) hX3,1 i
h 12 M 9

(4V)
h0 ch0 13

1 i
hX1,3
10
MPl

1a
n
hX1b i o
(20 V) hX2,2 i
0 c 0 (4V) 2,2
h 1 14
+
h
16
h
11
13
MPl
MPl

1a
1b
n
(1V) hX3,3 i
(4V) hX3,3 i
+ h 1 22
h0 12
9
19
MPl
MPl
4
o
(20 V) hX3,3 i
+ h 4 6
3
MPl

Pl

(4.18)
where
2
s 2 s
1
s
s
s
23 4 2 + 40 H15
H20 H30
H31
H38
,
X1,3
(0)
1a
2
0
s 2 s 2
S12

23 2 4 56 56
H15
H30 ,
X2,2


1b
02
s 2 s 2 s
s
s 2
2
2
H38
,
X2,2 23 4 + 4 H15 H20 H30 H31
1
s 2 s
s
s
s
23 40 H15
H20H30
H31
H38
,
X3,1
1a
s 2 s 2
S12 23 4 56 H15

H30 ,
X3,3


3 s 4 s 2 s 2 s 2 s 2
1b
2
S12

23 2 4 + 40 H15
H20 H30 H31 H38 , and
X3,3
4
0
s
S56
56
H30
.
X3,3

While some Dirac terms appear for our flat directions, Majorana singlet terms do not.
Local U (1) charge conservation forbids neutrino singlet Majorana mass terms of the form
Y


(4.19)
Nic Njc h i Hs ,
,

for any of our flat directions. Alternative Majorana singlet masses arising via terms of the
form,
Y


(4.20)
Nic S h i Hs ,
,
8 Each of these FD20 V terms requires a VEV for the vector-partner of
S56 .

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

487

where S is a generic singlet, are likewise forbidden, with one important exception. FD4
s H s H s iN c H s V s , into the FI-scale mass
converts the effective trilinear term, h40 H15
30 31
1 20 31
term,

0 s s s s c s
(4.21)
4 H15 H20 H30H31 N1 V31 .
Thus, a complete and viable seesaw mechanism appears for the third generation under
FD4. If they are not decoupled, the off-diagonal FD4V Dirac terms could then propagate
this seesaw mechanism to another generation.
The shortage of two generations of Majorana singlet masses for all but FD4 and FD4V,
combined with the previously discussed missing first generation down and electron masses,
present evidence that VEVs of NA fields may be necessary if viable mass matrices are to
be obtained in this model. However, FD1V offers a possible way around this for neutrinos:
s , might play the neutrino singlet role if h0 should contain a massless h

another field, V32

9
component. FD1V produces the seventh order Dirac mass term,


s
s 0
s
H38
.
(4.22)
h L3 V32
4 H15
s
s
s
s
s
H30
H20
H31
iN1c V31
, also appears in the FD4V
This term, along with the FD4 term, h40 H15
s
s
and V32
:
superpotential. Furthermore, FD1V contains an interaction term for V31


s
s
s s
S12 H30
V31
V32
(4.23)

H29

(which also appears in FD4V). The combination of these terms offers some interesting
neutrino dynamics for FD4V.
4.3.3. Proton decay
In the MSSM, the dangerous proton decay operators arise from baryon and lepton
number violating superpotential terms of the form
 


W = 1 uc d c d c + 2 Qd c L + 3 LLec + 1 QQQL + 2 uc uc d c ec MPl , (4.24)
where generational indices are suppressed (for general discussions of proton decay in string
models see, e.g., [2630]). i and j represent terms of generic order and can contain builtin suppression factors of (N c /MPl ) and/or (hi/MPl )n , where hi represents either an NA
singlet state VEV or a singlet product of NA fields, 10 such as a condensate of two hidden
sector vector-like fields. Proton decay limits imply 1 2 . 1024 and j /M(GeV) .
1025 for 1(B L) = 0 decays.
In the FNY model, the VEVs of FD1 and FD1V produce several sets of the dangerous
operators in (4.24). While these operators originate from terms of at least sixth or seventh
order in W o , the associated suppression factors in these terms do not appear strong enough
to slow proton decay sufficiently. That is, the proton lifetime would be significantly shorter
than known limits. We remark that local discrete symmetries which do forbid proton decay
mediating operators to all orders of non-renormalizable terms do appear in some three
9 FD3V (which also contains (4.22)) generates a similar h term hH s H s H s ih L V s at sixth order, while
4
15 19 31 4 1 32
not producing a h 1 counterpart.
10 For products of NA fields, the appropriate extra number of 1/M factors is implied.
Pl

488

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

generation free fermionic models [2630]. However, such a symmetry does not seem to
operate in the case of the flat directions of the FNY model. Therefore, there would still
appear the need to find a model that incorporates the attractive features of the FNY model,
while at the same time, incorporates such local discrete symmetries.
4.3.4. Effective -term for the Higgs
In all of our flat directions the massless MSSM Higgs doublets have the general
composition,
h p
h q

1
|h12

i|2

+ |h23
1

i|2

S12 i|2 + |hH s i|2

|h
31

h23 ih1 h12 ih3




s
S12 h 4 .
H31
h1

and

(4.25)
(4.26)

S12 , receives a VEV (as in FD1, FD2, FD20 , FD3, and FD4), then
Thus, when 12 , but not
h is simply h 1 . However, when h
S12 i 6= 0 (as in FD1V, FD2V, FD20 V, FD3V, and FD4V) h
S12 i 6= 0 appears a problematic issue.
becomes a linear combination of h 1 and h 4 H34 . h
As we have commented, the three possible non-renormalizable up mass terms containing
S12 i 6= 0. Thus, these terms actually
h 1 in (4.11) (including the two h0 terms) all require h

imply that h has a non-zero h4 component. Furthermore, the remaining non-renormalizable

S12 i 6= 0. Hence, either all or none of
up mass term involves h 4 , thereby also requiring h
the non-renormalizable up terms appear.
Also note that, while all of the down mass terms in (4.16) and electron mass terms
in (4.17) are seventh order or higher, the hh 4 i up term in m3,3 is fourth order. Therefore, if
we assume that seventh (or tenth) order terms can produce viable down and electron second
and first generations mass scales, then we would expect a fourth order hh 4 i-related mass
term to upset the generational mass hierarchy. However, this problem can be eliminated
if the fourth order up mass can be sufficiently suppressed to be of similar magnitude to a
seventh (tenth) order down mass. This is indeed possible: since h 4 is only a component of
the fourth order mass eigenvalue contains a suppression factor,
the eigenstate h,
q

S12 i
h
S12 i|2 + |hH s i|2
|h
31

(4.27)

s
S12 i is several orders of magnitude below the
i is an FI-scale VEV, when h
Since hH31
FI-scale, it would indeed seem possible for (4.27) to suppress the fourth order up mass,
making it comparable to a seventh order down mass. 11
Investigation of the set of possible effective Higgs -terms leads to similar conclusions
S12 i appears in a tenth order -term for h1 h 4 ,
S12 i. Evidence against a large h
regarding h


0
s 2 2
40 H30 H15
H31 h1 h 4 /(MPl )7 .
(4.28)
23 56
11 Note that in this case, the leading component of the seventh order down mass term should not contain a h
S12 i
factor.

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

489

S12 , but h 1 is replaced by

Here the set of VEVs generating an effective does not contain
h 4 . Thus, we should again expect to be significantly above the EW scale unless the h 4
S12 i is also
contribution to h is either zero or extremely small, which again implies that h
far below the FI-scale. The same conclusions are suggested by 16th order terms from FD4:


 s 2 s 2 s 3 s 2 
S56 4 2 + 40 2 H15
H20 H31 H38 (MPl )13 . (4.29)
h12 ih1 + h23 ih3 h 4 23
4.3.5. Non-renormalizable coupling constants
An important stringy aspect to the generational mass hierarchy is the numeric scale of the
the coupling constants n in nth-order non-renormalizable mass terms. In free fermionic
models, these coupling constant can be expressed in terms of an n-point string amplitude
An . This amplitude is proportional to a worldsheet integral In3 of the correlators of the n
vertex operators Vi for the fields in the superpotential terms [3135],

n3
g p
8/
Cn3 In3 /(MPl )n3 .
(4.30)
An =
2
The integral has the form,
Z


f
f
b
(zn1 )Vnb (0)
In3 = d 2 z3 d 2 zn1 V1 ()V2 (1)V3b (z3 ) Vn1
Z
= d 2 z3 d 2 zn1 fn3 (z1 = , z2 = 1, z3 , . . . , zn1 , zn = 0),

(4.31)
(4.32)
f

where zi is the worldsheet coordinate of the fermion (boson) vertex operator Vi (Vib )
of the ith string state. Cn3 is an O(1) coefficient that includes renormalization factors
in the operator product expansion of the string vertex operators and target space gauge
group ClebschGordon coefficients. SL(2, C) invariance is used to fix the location of three
Q
of the vertex operators at z = z , 1, 0. When nv of the fields li=1 Xi take on VEVs,
Ql
h i=1 Xi i, then the coupling constant for the effective ne = (n nv ) order term becomes
Q
A0ne An h li=1 Xi i.
Q
An n-point string function trivially vanishes when the correlator h i Vi i itself vanishes,
resulting from non-conservation of at least one or more gauged or global (including
Ising) worldsheet charges. When all charges are conserved, one must compute In3 to
determine the numeric value of An . It might actually be possible for an n-point function to
Q
Q
vanish upon integration of h i Vi i, even when h i Vi i is non-zero (i.e., when all gauge,
picture-changed global worldsheet, and Ising charges are conserved). Typical non-zero
values of I1 and I2 integral for 4- and 5-point string amplitudes are around 100 and 340
[3134] for free fermionic models.
As (4.16) and (4.17) indicate, when the relevant h1 - and h3 -term coupling constants
are non-equal, our second generation down and electron mass terms appear at seventh and
eight order, respectively, in the superpotential. 12 Thus, in the case of coupling constant
inequality, an order of magnitude estimate of the mass ratio between the second and
third generations requires knowing the coupling strength of the seventh order term. This
12 Alternatively, when the couplings are equal, then actual mass terms are tenth order.

490

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

necessitates numeric computation of the associated correlation function integral. To our

knowledge, this has not been done to date for sixth order or beyond. Thus, for an order
of magnitude comparison of 7th order couplings to 3rd (i.e., between third and second
generations), we have analyzed the integral I4 associated with the superpotential term,
0 H s H s i. In the zero external momentum limit we find,
h3 Q1 d2c h23 56
15 30
Z
I4 =

d 2 z3 d 2 z4 d 2 z5 d 2 z6
|z3 |5/4 |z5 |
|z 1||z z3 ||z z5 |2 |z z6 ||z3 z4 ||z3 z5 ||z4 |
{|z5 |2 |z6 |2 /z + 2|z4 |2 z }
.

(|1 z3 | |z6 |)3/2(|1 z6 | |z3 z6 |)3/4 |z4 z5 |4 |z4 z6 |2

(4.33)

The integrand in I4 , computed directly from the product of correlations functions, initially
contained many more terms than shown in (4.33). However (relating to our comments
about the possibility of an In3 = 0 integral above), the vast majority of terms were in
fact not holomorphic with regard to coordinates, but instead contained extra (zi zj ) or
(zi 0 z j 0 ) factors that could not be paired to form |zi zj | factors. Since integration of
such non-holomorphic terms results in a zero contribution to I4 , only the holomorphic
(equivalently, non-zero) terms are included in (4.33).
The second term in the integrand of I4 clearly dominates over the first term, except in
the z6 limit. Thus, for a first order of magnitude approximation, I4 reduces to,
Z
I4 = 2

d 2 z3 d 2 z4 d 2 z5 d 2 z6

|z3 |5/4|z4 ||z5 |

|z z3 ||z z5 |2 |z z6 ||z3 z4 ||z3 z5 |
1
.

(|1 z3 | |z6 |)3/2(|1 z6 | |z3 z6 |)3/4 |z4 z5 |4 |z4 z6 |2

(4.34)

Several poles and fixed factors of infinity (i.e., of z ) are found in the integrand, making
evaluation difficult at best. This could indicate that the spacetime momentum should not
be set to zero before performing the integral.
General consideration of (4.34) might appear to imply something contrary to that
expected of a seventh order coupling constant. For a weakly coupled model, when n 3
fields acquire VEVs in an nth-order term, we expect the magnitude of the effective third
order coupling constant to be far below unity, i.e., to produce strongly suppressed terms.
However, integration over a space with infinite limits of a a positive-semidefinite integrand
involving several poles makes a finite result difficult to imagine. In this case, the resolution
to obtaining a finite value seems to rest on the infinities in the denominator. The approach
to computing the integral is, perhaps, not to actually integrate, but to examine the divergent
areas and expand about them. Clearly the most highly divergent region is near z4 = z5 , but
the zeros from the 1/z factors might be enough to damp out the four resulting infinities.

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

491

What value of I4 is required of the seventh order down term to satisfy the physical
ms : mt 103 ? The coupling constants of the third order mass terms are all A3 = g.
Thus, a correct order of magnitude mass ratio requires,
0 Hs Hs i
0 Hs Hs i

4
h23 56
A7 h23 56
1 p
15 30
15 30
=
8/ C4 I4
A3
(MPl )4
2
103 .

(4.35)

Insertion of the FD20 VEVs in (4.35) results in

C4 I4 43.

(4.36)

In NAHE class Z2 Z2 models, I1 and I2 take on respective values typically around 70

and 400. Furthermore, In generally increases with n. Thus, (4.35) is in disagreement with
the probable magnitude of I4  400, given that C4 is 0(1). 13 However, if the seventh
order down mass term contains only the massive eigenstate hm (i.e., equal couplings) then
(4.16) suggests the second generation results from a tenth order term. The constraint on the
associated I7 becomes,
0 H s 2H s H s i
0 H s 2H s H s i

7
h23 40 56
A10 h23 40 56
1 p
15
30 31
15
30 31
=
8/ C7 I7
7
A3
(M
)
2
Pl
103 .
(4.37)

This implies I7 200000. While extrapolating a pattern for values of In based only on the
known ranges of I4 and I5 in FNY is untenable, a value of I7 in this range appears possible.
Thus, it is phenomenologically preferable that for the seventh order down term h0 reduces
to the the massive hM eigenstate.
4.4. F -constraints from vector pairs
The FD20 and FD20 V variations of FD2, in which vector partners of FD2 fields
0
S0 and 12 , respectively, also take on VEVs, result in the more
S
56 and of both
56
0 does not acquire a
phenomenologically viable mass matrices. Note especially that if 56
VEV, then no new up, down, or electron mass terms are produced by the FD2 class models.
The appearance of these vector partner VEVs does generate some new and dangerous (low
order) non-zero F -terms,


S12 23 ,
(4.38)
hW/13 i = g

0 s
s
(4.39)
hW/H16 i = g 56 H15 .
S0 and 12 acquire VEVs,
Examination of the FNY superpotential indicates that, when
56
survival of spacetime supersymmetry down to the EW scale requires VEVs for some NA
fields as well. No other singlet terms can lead to cancellation of F -term components. Note
S12 should not
that phenomenological issues, such as the Higgs mass scale, imply that
13 In strongly coupled models, for which substitution of M
string  MPl in (4.35) might be justified, an even
smaller C4 I4 value is found.

492

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

acquire an FI-scale VEV. Consider now (4.39). Inclusion of singlet and hidden sector terms
in (4.39) through sixth order results in

0 s

s
= g 56
H15 + 4 N3c V37 H28 MPl
W/H16

+ 5 2 N3c V37 H28 (MPl )2

s
+ 6 hH30 iN1c H29
V31 H20 (MPl )3 .
(4.40)
s
i from 2 N3c V37 H28 implies a larger contribution
A non-zero contribution to hW/H16
c
c
0 i 6= 0.
from N3 V37 H28 . Thus, N3 , V37 , and H28 must all receive FI-scale VEVs when h56
This is a viable solution because fourth order string couplings can actually be as large as
third order couplings [35].

5. Conclusions
In this paper we investigated phenomenological issues of the four flat directions
(and some of their variations) derived in [3] for the FNY model [4,5]. Using our
stringent F -flat requirements, these were the only directions found to be flat beyond
seventh order that decoupled all fractionally charged MSSM exotics from the low energy
superpotential. (It was as a serendipitous bonus that these four directions decoupled
all MSSM-charged exotics.) Approximately 100 other VEV directions were found that
removed the fractionally charged fields, but F -flatness for these was broken at only sixth
or seventh order.
The first three of our flat directions are actually flat to all finite order, while F -flatness
of the fourth is broken at twelfth order. Variations on these flat directions, in which both
components of vector pairs of fields were allowed to acquire VEVs, were also investigated.
The variations of FD2 were found to be the most phenomenologically viable with regard
to mass matrices.
We examined, for the different flat directions, the inter-generational mass hierarchy
for the MSSM quarks and leptons and possible effective -terms for the MSSM Higgs.
Reasonable, but not perfect, up, down, electron, and neutrino hierarchy-producing mass
matrices were found for particular variations of our flat directions. Possible additional
F -flatness constraints on the vector-pair flat directions were also examined.
An interesting aspect of the Higgs fields was mutually suggested by the mass matrices,
the possible -terms, and the additional F -flatness constraints. In the FNY field basis
given in [4], the MSSM Higgs eigenstates h and h each have two components. For
several phenomenological reasons, the SU(2)L doublet h 1 contribution to to h must greatly
and not the h 1 contribution, that
outweigh h 4 s. Relatedly, it is only the h 4 component of h,
couples to the second generation up field, ultimately giving mass to it. This results in an
additional cosine suppression factor for the charm mass term. This is a favorable feature
for this model, since the charm mass term originates at fourth order, while those for the
strange and muon masses appear at seventh and eight order, respectively. Thus, without the
extra cosine factor, this model would predict a several orders of magnitude mass difference
between the charm and strange fields.

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

493

Our investigation of singlet flat directions led to several pieces of evidence suggesting
that some non-Abelian fields must also acquire VEVs for the FNY model to produce viable
phenomenology. The additional F -terms arising from FD1V and FD20 especially implied
this. In particular, for FD20 it appears necessary for two SU(2)H 0 doublets (along with a
neutrino singlet) to also acquire FI-scale VEVs.
In forthcoming [36], we will generalize our current flat directions by permitting nonAbelian fields to acquire VEVs. The simplest non-Abelian extension to each of our current
direction is the addition of the only completely chargeless product of hypercharge-singlet
non-Abelian fields: V5 V40V10 V37 . We shall also examine the additional flat directions
available when our basis set of 24 non-trivial D-flat singlet directions increases to a basis
set of 53 non-trivial directions involving the 30 hypercharge-singlet hidden sector nonAbelian fields in the FNY model.
We conclude by reemphasizing that the success of the FNY model in producing minimal
superstring standard models should not be regarded as suggesting that the FNY model
is the true string vacuum. Rather, it is our firm opinion that the success of the FNY
model in this regard, together with the other properties of the free fermionic NAHE-based
models, such as the natural emergence of three generations, can be regarded as providing
evidence to the assertion that the true string vacuum will possess some of the characteristics
shared by this class of models. Furthermore, the existence of highly realistic string models
near a maximally symmetric point in the moduli space, provides further indication for
the relevance of string theory in nature. The continued development of the techniques
needed to seriously confront string models with experimental data is therefore imperative.
The main focus of the present paper being the phenomenological analysis of exact flat
directions.

Acknowledgements
This work is supported in part by DOE Grants No. DE-FG-0294ER40823 (A.F.) and
DE-FG-0395ER40917 (G.C., D.V.N., J.W.).

Appendix A. D- and F -flat directions

The four classes (Table 3) are defined by their respective set of non-Abelian singlet
field VEV components. The class identification number appears in column one. Column
two specifies the number of singlet VEVs, with its first entry specifying the number
of VEVs resulting from FI cancelation and its second entry indicating the number of
4 -related fields to which additionally VEVs are assigned. Column three indicate the
superpotential order at which flatness is broken. The s in the remaining columns
identify the field VEVs. An implies a given field takes on a VEV, while a implies
the vector partner does instead. A below {VEV1 } indicates that all fields in the set
S4 ,
S0 ), H s , H s , H s , H s } acquire VEVs.
{12 , 23 , (4 , 40 ,
4
30
38
15
31

494

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

Table 3
Classes of D- and F -flat NA singlet directions that yield an MSSM low energy effective field theory
FD#

#
VEVs

O(W )
breaking

1
2
3
4

8+3
9+3
9+3
9+3

12

{VEV1 }

S56

0
56

s
H19

s
H20

Table 4
Examples of flat directions from the four classes presented in Table 3
FD#

Q(A)
112

1
2
3
4

2
6
4
2

0 ,H s H s ,H s ,H s }
{12 ,23 ,(4)
30 38 15 31

hi
6.7 1016
3.9 1016
4.8 1016
6.7 1016

GeV
GeV
GeV
GeV

3,
10,
4,
3,

1,
2,
3,
2,

1,
2,
2,
3,

3,
8,
8,
3,

2,
6,
2,
4,

2,
4,
6,
4,

S56

0
56

s
H19

s
H20

1
3
1
2

0
1
0
0

0
0
2
0

0
0
0
2

1
2
2
3

Column one (Table 4) indicates the class to which an example direction belongs. Column
two contains the anomalous charges Q(A) /112 of the example flat directions. Column
three specifies the overall common scale of the associated VEVs. The remaining column
s
s
s
s
H38
, H15
, H31
}
entries specify the ratios of the norms of the VEVs. {12 , 23 , (4 ), H30
{VEV1 } form the set of fields possessing VEVs in all four directions. The third component
VEV involves all of the 4 -related states and is the net value of |h4 i|2 + |h40 i|2
S0 i|2 . E.g., a 1 in the 4 column for FD1 specifies that
S4 i|2 |h
|h
4

 
 
0 
S4  
S4  = 1 hi2 .
|h4 i|2 +  40  

Appendix B. Field content for flat directions

Massless fields for flat direction 1:
3 generations of MSSM fields: (Q, uc , d c , L, ec , N c )i=1,2,3 ;
2 MSSM Higgs:
h p
h q

h23 ih1 h12 ih3 ,

|h12 i|2 + |h23 i|2

1
S12 i|2 + |hH s i|2
|h
31





s
S12 H34 ;
H31
h1

21 non-Abelian singlets: 2 , 3 , 12 , 23 ,

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

495

s


1
s
s
S23 (1) q
S23 h12 iH29
+ hXiH37
H30
,

s i|2 + |hXi|2
|h12 i|2 + |hH30
S56 H s H s ,
where X 23
31 38

0 0


1
S4 4 ,
S4 4

4 (1) q
S4 i|2 + |h
S0 i|2
|h
4
4 (2) q

1
|h4 i|2 + |h40 i|2

 0

S4 40
S4 ,
h4 i

S4 i 0 h 0 i
S4 h4 i
S0 )
S0 i4 + h
(h
4
4
4
,
4 (3) q 4
0
0
2
2
2
S
S
|h4 i| + |h4 i| + |h4 i| + |h4 i|2
S56 ,
S0 , H s , H s , H s , H s , H s , H s , H s , V s , V s , V s , V s ;
56 ,
56
15
19
20
30
31
36
38
1
12
31
32
2 hidden sector SU(3)H (anti)-triplets: V4 , V13 ;
6 hidden sector SU(2)H doublets: H23 , H26 , V5 , V15 , V39 , V40 ;
6 hidden sector SU(2)H 0 doublets: H25 , H28 , V10 , V19 , V35 , V37 .
Fields with (near) FI-scale masses for flat direction 1:
(i) exotic MSSM-charged states
SU(3)C triplets: H33 , H40 ;
SU(2)L doublets: h2 , h 2 , h 3 ;
hM p

1
|h12 i|2 + |h23 i|2

h12 ih1 + h23 ih3 ,

h 4 H34 , h4 H41 V45 , V46 , V51 , V52 ;

hidden sector SU(2)H doublets (with fractional electric charge): H1 , H2 ;
hidden sector SU(2)H 0 doublets (with fractional electric charge): H11 , H13 ;
(ii) non-Abelian singlets
S12 , 13 ,
1 ,
S13 M q

s M
q
H36

S23 M

1
s i|2
|hH30

+ |hXi|2

1
s 2
|hH30
i| + |hXi|2




s S
s
H30
13 + hXiH36
,

s s

S13 H30
hXi
H36

S23 + ((hH s i + hXi)2 /hH s i)H s + (hXih12 i/hH s i)H s ]

[h12 i
30
30
29
30
37
q
,
s i + hXi)2 /hH s i|2 + |hXih i/hH s i|2
|h12 i|2 + |(hH30
12
30
30

s M
q
H37

1
s 2
|hH30
i| + |hXi|2

s s

S23 H30
hXi
H37 ,

S0 i4 + h
S4 i 0 + h 0 i
S4 + h4 i
S0 )
(h
4
4
4
,
4 M q 4
0
0
2
2
2
S
S
|h4 i| + |h4 i| + |h4 i| + |h4 i|2

496

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

0 , H s, H s, H s, H s, H s, H s, H s, H s , H s , H s , H s , H s , H s , H s , H s ,
56
7
3
4
5
6
8
9
10
16
17
18
21
22
32
37
s
s
s
s
s
s
s
s
s
s
s
s
, V2s , V11
, V21
, V22
, V41
, V42
, V43
, V44
, V47
, V48
, V49
, V50
;
H39
(iii) hidden sector non-Abelian fields:
SU(3)H (anti)-triplets: H42 , V14 , V24 , V34 , H35 , V3 , V23 , V33 ;
SU(2)H doublets: V7 , V17 , V25 , V27 ;
SU(2)H 0 doublets: V9 , V20 , V29 , V30 .
Effective FI-scale tadpole in FD1 superpotential:



S40 + 40
S4 .
(B.1)
1 4

Effective FI-scale mass terms in FD1 superpotential:

W3 terms:

 0
0

0 

S4 4 + 4
S40
S4 4 +
S4 + h4 i
1

s s

S23 + H30
S13 h12 i
+
H29


s
s
s s
V22 + V29 V30 + V25 V27 + V23 V24
+ h12 i h1 h 2 + H36 H37 + V21


S12 13 + h3 h 2 + V33 V34
+ h23 i




s
+ h4 i V45 V46 + H1 H2 + 40 V51 V52 + H7s H8s + H9s H10


s s
s s
S4 i H3s H4s + H5s H6s + V41
V42 + V43
V44
+ h


0  s s
s s
S4 V47 V48 + V49
V50 + H11H13
+


S56 i H s H s + H s H s
+ h
17 18
21 22

s 0 s
S13 H s + H s h2 h 4 + H s H41 h 3 ;
56 H16 + H30

+ H15
29
31
38

(B.2)

W4 terms:

s s
H15 H30 [V3 V14 + V7 V17 ];

(B.3)

W5 terms:



 s s


s
s
s
s
H38
H30
V2 V11 + V9 V20
23 H31
h 4 H41 + H33 H40 + 40 H15

 s s

s
s
S56 H31
H38
+
H32H39 + H35 H42 ;

(B.4)

W6 terms:


S56 H s H s H s H s .
23
31 38 29 36

(B.5)

Additional fields receiving (near) FI-scale masses for other flat directions:
Fd #

non-FD1 VEVs

New mass and tadpole terms

1V

S12

S0

S12 i[13 23 + h2 h 1 ], h
S12 23 i13
h

0  s s
S
56 H19 H20

0  s s
0 s s
56 H15 H16, 56 H15 H16

2
2V
20
20 V

56

S0 ,
S12

56
S0 , 0

56

56

S0 , 0 ,
S12

56
56

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

3
3V

s
H19
s ,
S12
H19

s
H20

4V

s ,
S12
H20

497

s 0 s
S H
H19
56 20

s 0 s
S H
H
56 19

20

s
s Hs Hs V s V s
H15 H20
31
38

0 s s s s 41 c50 s
4 H15 H20 H30 H31 N1 V31

The first column denotes the flat direction. A V in this column indicates a vector-like
partner to one of the standard fields associated with a given flat direction is also allowed
a VEV. The listings in column two indicate a flat directions VEVs that are not present
in flat direction 1. Column three lists the VEV-induced mass terms new to the respective
flat directions and not present in flat direction 1. If a mass term for a given field is already
present in a standard flat direction, respective higher order mass terms (should they exist)
are not necessarily listed.
Rotated massless (with (1) superscript) and FI-scale massive (with M superscript)
eigenstates for non-FD1 directions:
FD1V (and all embeddings):
S1 (1) q

S1 M q

1
S12 i|2 + |h23 i|2
|h
1

S12 i|2 + |h23 i|2

|h

S12 h23 i23 ,
S12 i
h


S12 + h
S12 i23 ;
h23 i

FD20 , FD20 V:

s s

0  0
1
0 (1)
q
56 56 H15
H15 ,
56
0 i|2 + |hH s i|2
|h56
15
0 M
q
56

1
0 i|2 + |hH s i|2
|h56
15

0  s

 0
s
H15
56 + 56
H15 .

Appendix C. Superpotentials for low-energy effective field theories for flat directions
Flat direction 1 low energy effective superpotential, W FD1 :
Singlet terms
W3 :
 0 s s

s s
S56 H19 H20 +
S23 V31
V32 ;
g

(C.1)

MSSM terms
W3 :




g h 1 Q1 uc1 + L1 N1c + gh3 Q3 d3c + L3 e3c
 c s s

s
s
s
H30
H31
N1 H20 V31 ;
+ 40 H15

(C.2)

498

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

W4 :


 s
s
h1 Q3 d3c + h1 L3 e3c H29
H30


S40 12 Q1 Q2 uc d c + Q1 uc L2 ec + Q1 uc L2 ec
+ 12 + 40
1 2
1
2
1
 2
+ Q2 d1c L1 N2c + Q2 d2c L1 N1c + L1 L2 e2c N1c


+ 12 40 Q2 uc1 L2 e1c + Q2 d1c L2 N1c + uc1 uc1 d2c e2c + uc1 uc2 d2c e1c

+ uc2 d1c d1c N2c + uc2 d1c d2c N1c



S40 Q1 Q1 Q2 L2 + Q1 Q1 uc2 d2c + L1 L1 e2c N2c

+ 12

s s 
H30 Q1 Q2 Q3 L3 + Q1 Q2 uc3 d3c + Q1 Q3 uc2 d3c
+ H15

+ Q1 uc2 L3 e3c + Q1 uc3 L3 e2c

 s

s
s
H38
Q2 d2c L3 + L2 L3 e2c V32
+ 4 H15


 s
s
s
+ 40 H15
H38
Q2 d3c L2 + uc2 d2c d3c V32


s
s
+ 4 H15
H30
Q3 uc1 L3 e2c + Q3 uc2 L3 e1c + uc1 uc2 d3c e3c

+ uc1 uc3 d3c e2c + uc2 uc3 d3c e1c



s
s
H30
Q2 Q3 uc1 d3c + Q2 uc1 L3 e3c + Q2 uc3 L3 e1c
+ 40 H15



s
s
H30
Q1 Q2 Q2 L2 + Q1 Q2 uc2 d2c + Q1 uc2 L2 e2c ;
+ 23 H15

(C.3)

W5 :



 s s

S40 h3 Q2 d2c + h3 L2 e2c V31

V32
1 + 40

s 
 s s
c s s
c
+ h12 ih3 Q2 d1 V1 V12 + H15 Q2 d2 L1 + L1 L2 e2c H19
V32

s 
c c
c
c
c
c
c
+ H30 Q1 Q3 u1 d3 + Q1 u1 L3 e3 + Q1 u3 L3 e1 + Q3 d1 L1 N3c
 s
+ Q3 d3c L1 N1c + L1 L3 e3c N1c H29

s 

s s


s
S40 h3 L1 e2c V1s V12
+ H38
V32 + 12
Q1 d2c L3 + Q1 d3c L2 H29



s
s
+ 12 H38
Q3 d2c L1 N2c + L1 L2 e3c N2c H20


s
S23 + h1 L3 e3c
S23 + Q1 Q2 uc1 d2c + Q1 uc1 L2 e2c
+ 23 H30
h1 Q3 d3c
 s
+ Q1 uc2 L2 e1c + Q2 d1c L1 N2c + Q2 d2c L1 N1c + L1 L2 e2c N1c H29


s
+ 4 H30
Q3 uc1 L3 e1c + Q3 d1c L3 N1c + uc1 uc1 d3c e3c + uc1 uc3 d3c e1c
 s
+ uc3 d1c d1c N3c + uc3 d1c d3c N1c H29


s
s
s
+ 4 H38
V32
L2 L3 e1c H29

0 s 
 s s
+ 4 H15 Q2 d1c L2 + uc2 d1c d2c H19
V32

0 s c c c s s
+ 4 H38 u1 d2 d3 H29V32


 s
s
S4 H30
+
Q1 Q1 Q3 L3 + Q1 Q1 d3c uc3 + L1 L1 e3c N3c H29

 

S56 H s h1 Q3 d c + L3 ec 56 H s
+
30
3
3
29

s s 
+ H15
H30 Q1 Q2 Q3 L3 + Q1 Q2 d3c uc3 + Q1 Q3 uc2 d3c

+ Q1 uc2 L3 e3c + Q1 uc3 L3 e2c 3



+ H15 H30 Q1 d2c L3 N3c + Q1 d3c L2 N3c + Q1 d3c L3 N2c 56 ;

(C.4)

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

499

W6 :


 s s s
H29 V32
Q1 d1c L2 + L1 L2 e1c H19

s

 s
c
+ H30 h1 h1 h 1 Q3 d3 + L3 e3c H29

s

s s s
+ H30
V32 H29
h1 Q2 d2c + h1 L2 e2c V31

c c
c
+ h12 i Q1 Q2 u1 d2 + Q1 u1 L2 e2c + Q1 uc2 L2 e1c

+ Q2 d1c L1 N2c + Q2 d2c L1 N1c + L1 L2 e2c N1c 3 3



0

S4 Q1 d c L1 H s H s V s
+ 40 uc1 d1c d2c +
2
19 29 32

s


s
s
+ H15
V32
Q2 d2c L1 + L1 L2 e2c 2 H19

s 

s
s
+ H38 Q1 d2c L3 + Q1 d3c L2 3 H29
V32
;

(C.5)

Mixed MSSM-hidden terms

W5 :


h12 i h3 Q2 d1c V10 V19 + h3 L1 e2c (V4 V13 + V5 V15 )

S40 h3 L1 e2c V10 V19

+ 12 40 h3 Q2 d1c [V4 V13 + V5 V15 ] + 12







s
+ 12 H38
Q1 d2c L3 + Q1 d3c L2 H25 V35 + uc1 d2c d3c + L2 L3 e1c H23V39
 c

s
N2 [H26 H26H25 V19 + H26 V15 H25 H28 + H23 H26 H28 V19
+ 23 H30
+ H23 V15 H28 H28]

s s c
s
H30 N2 56 V31
V19 V35
+ H30

s s 

c
+ H30 H38 h 1 Q1 u3 V19 V35 + h 1 Q3 uc1 V15 V39


s s 
H38 h 1 L1 N3c V19 V37 + h 1 L3 N1c V15 V40
+ H31


+ H s H s h 1 L2 V s V15 V39 ;
38

38

32

(C.6)

W6 :


 s
H25 V37
Q3 d2c L3 + Q3 d3c L2 + uc3 d2c d3c + L2 L3 e3c H19


s
c
c
c c c s
+ h12 i Q1 d1 L2 + L1 L2 e1 H19 H25 V35 + u1 d1 d2 H19H23 V39

+ h3 Q2 d1c 3 V10 V19 + h3 L1 e2c 3 V5 V15

 s
H25 V37
+ h23 i Q2 d2c L2 + uc2 d2c d2c + L2 L2 e2c H19




0


c
c
c
c
c
S4 Q3 d L3 + u d d +
S4 Q3 d L2 + L2 L3 ec H s H23V40
+
2
3 2 3
3
3
19

s
s
+ H15
H28V10
h3 Q1 d3c V32

s


s
+ H38
H25 V37 .
h 1 L3 V4 V13H23 V40 + H23 V5 V15 V40 + H25 V10 V19 V37 + V1s V12
(C.7)

Hidden terms
W3 :

s 
0 0 s  s
S4 H V H25 V35
H30 + 4
30
31

s 
0 0 s  s
S4 H31 H19 V19 V37
+ H31 + 4

500

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504


0 s 
0 0 0 s  s
S4 H31 + 4
S4
S4 H31 H19V15 V40
+

0 s 
0 0 0 s  s
S4 H + 4
S4
S4 H V H23V39 ;
+
30
30
31

(C.8)

W4 :

s
s
s
s
s
s
V32
H26 V40 + h4 iH37
V32
H28 V37 + H30
H25V35
H37
3 V31


S56 [H23H23 H28 H28 + H23 H26H25 H28 + H26 H26H25 H25 ]
+ 23

s s s s

0 s
s
S4 H30 3 V31
H23V39 + H15
H30 H20 V31 H25 V10
+

0 0 s

0 s s s s
s
S
+ 4 4 H30 3 V31 H25 V35 + 4 H15 H30 H20V31 H23 V5 ;

(C.9)

W5 :

s
s
s
s
s
V32
H26 V40 + H30
H25 V35 + H31
V19 V37
3 3 V31
3 3 H19
2 H37

0 s


s
0
s
s
S4 H 3 3 V H23 V39 +
S4 H 3 3 H V15 V40
+
30
31
31
19

s s
s
s
+ H15 H30 2 H20 V31 H25 V10 ;

(C.10)

W6 :

 s s

S56 V31

V32 H26 H26 H25H25 + H23 H26 H25H28 + H23 H23 H28H28

s
s
H25 V35 .
3 3 3 V31
+ H30

(C.11)

Flat direction 1V modifications to superpotential, W FD1V :

Singlet terms
W3 :





S12 40
S12 H s + 12
S12
S12 H s +
S40 H s H s V s V s ;

30
30
30
29 31 32
W5 :


s
s
s s
S12 H30
V31
V32 ;
3 3 H29

MSSM terms
W3 :


s 

S4 H s h 4 Q3 uc
H30 + 4
30
3



s
s
S
+ 12 H31 + 12 12 12 H31



s
S40 H31
h 4 Q1 uc1 + h 4 L1 N1c
+ 12 40

s
s
s
s
s
H30
H38
;
ch0 h 0 Q1 uc2 + 4 H15
ch0 h 0 L3 V32
+ 23 H15
W4 :



0 0 s 
s
s
s s
S12 H31
S4 H31 h3 h 4 V31
V32
+ 12
+ 4
H31



s
S23 +
S56 H s h 4 Q3 uc 56
+ 23 H30
h4 Q3 uc3
30
3

s s
s
s

+ H15 H31 h4 L1 H19 V32

(C.12)

(C.13)

(C.14)

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504


S12 Q1 Q2 uc1 d2c + Q1 uc1 L2 e2c + Q1 uc2 L2 e1c + Q2 d1c L1 N2c
+ 12 12

+ Q2 d2c L1 N1c + L1 L2 e2c N1c




+ 23 H s c 0 h0 Q1 uc + L1 N c H s ;
30 h

29

501

(C.15)

W5 :



s
0
S56
h1 h 1 h 4 Q3 uc3 + h 4 L3 N3c 56
H30


 s s

s s
s
s s
S12 h3 Q2 d2c + h3 L2 e2c V31
H31 h1 h 4 H29
V31
V32 + 12
V32
+ H30





s
c
c

+ 12 H
h4 Q1 u + h4 L1 N 3 3 ;
31

(C.16)

W6 :


 s s
s
0
S56
+ h 4 L2 N2c 56 V31
V32
h 4 Q2 uc2
H30

s
s s
s

+ H31 h3 h4 3 3 V31 V32 + H38 h1 h4 Q1 uc3 L2 V32

.

Mixed MSSM-hidden terms

W4 :




S12 + 12 40
S40 h 4 L1 H23 V5 + 12
S40 h 4 L1 H25V10 ;
h12 i + 12 12
W5 :

(C.17)

(C.18)


 s
s
s
s
H25 V37 + 23 H30
h4 L2 H19
H29 h 4 L1 H23 V5 ;
23 H31

(C.19)

h12 i3 3 h 4 L1 H23 V5 .

(C.20)

W6 :

Hidden terms
W3 :

 s
 s

s
s
S12 H30
S12 H31
V31 H25V35 + 12
H19 V19 V37
12

 s

0 s
S
S
S
S40 H s H s V15 V40 ;
+ 12 12 4 H30 V31 H23 V39 + 12 12
31 19
W4 :


s
s
S12 H30
H25 V35 .
3 V31
12

(C.21)

(C.22)

Flat direction 20 modifications to superpotential, WFD20 :

MSSM terms
W4 :


0 s s 
56 H15H30 Q1 Q3 Q3 L2 + Q1 Q3 uc3 d2c + Q1 uc3 L2 e3c
 s

0 0 s 
S56 H30 h1 Q3 d3c + h1 L3 e3c H29

;
+ 56

(C.23)

502

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

Flat direction 2V modifications to superpotential, W FD2V :

MSSM terms
W4 :

s 0 
S56 h 4 L3 N3c 56 ;
H30
W5 :


s S0
s s
56 h4 Q2 uc2 V31
V32 .
H30

(C.24)

(C.25)

Flat direction 20 V modifications to superpotential:

MSSM terms
W3 :

s 0 0 
S56 h 4 Q3 uc3 ;
H30 56
W4 :


s
0 S0
56 h4 Q3 uc3 .
56
H30

(C.26)

(C.27)

Flat direction 3 modifications to superpotential, W FD3 :

MSSM terms
W4 :
 s

s s 
H15 H19 Q2 d2c L1 + L1 L2 e2c V32


 s
s
s
+ 40 H15
H19
;
Q2 d1c L2 + uc2 d1c d2c V32

(C.28)

W5 :


 s s
s
V32
Q1 d1c L2 + L1 L2 e1c H29
H19

0 s c c c s s
0 s
s
s
S4 H19 Q1 d2c L1 H29
+ 4 H19 u1 d1 d2 H29V32 +
V32

s s 

s
+ H15 H19 Q2 d2c L1 2 + L1 L2 e2c 2 V32
;

W6 :




s s
s
V32 .
Q1 d1c L2 + L1 L2 e1c 2 + 3 H29
H19

Mixed MSSM-hidden terms

W5 :



s
Q1 d1c L2 + uc1 d1c d2c + L1 L2 e1c H25 V35 .
12 H19

(C.29)

(C.30)

(C.31)

Flat direction 3V modifications to superpotential, W FD3V :

MSSM terms
W3 :

s s s
s
;
H15 H19H31 h 4 L1 V32

(C.32)

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

W4 :


s
s
s
s
H19
H31
2 .
h4 L1 V32
H15

503

(C.33)

Flat direction 4 modifications to superpotential, W FD4 :

MSSM terms
W4 :



s
s
H20
Q3 d2c L1 N2c + L1 L2 e3c N2c .
12 H38

(C.34)

References
[1] G.B. Cleaver, A.E. Faraggi, D.V. Nanopoulos, String derived MSSM and M-theory unification,
Phys. Lett. B 455 (1999) 135, hep-ph/9811427.
[2] G.B. Cleaver, M-fluences on string model building, CTP-TAMU-46/98, hep-ph/9901203.
[3] G.B. Cleaver, A.E. Faraggi, D.V. Nanopoulos, A minimal superstring standard model I:
flat directions, ACT-2/99, CTP-TAMU-12/99, TPI-MINN-99/22, UMN-TH-1760-99, hepph/9904301.
[4] A.E. Faraggi, D.V. Nanopoulos, K. Yuan, Nucl. Phys. B 335 (1990) 347.
[5] A.E. Faraggi, Phys. Rev. D 46 (1992) 3204.
[6] M. Luty, W. Taylor, Phys. Rev. D 53 (1996) 3399.
[7] T. Gherghetta, C. Kolda, S. Martin, Nucl. Phys. B 468 (1996) 37.
[8] E. Dudas, C. Grojean, S. Pokorski, C.A. Savoy, Nucl. Phys. B 481 (1996) 85.
[9] G. Cleaver, M. Cvetic, J.R. Espinosa, L. Everett, P. Langacker, Nucl. Phys. B 525 (1998) 3.
[10] G. Cleaver, M. Cvetic, J.R. Espinosa, L. Everett, P. Langacker, Nucl. Phys. B 545 (1999) 47.
[11] G. Cleaver, Mass hierarchy and flat directions in string models, in: Proc. of Orbis Scientiae 97
II, Miami Beach, Florida, December, 1997.
[12] I. Antoniadis, C. Bachas, C. Kounnas, Nucl. Phys. B 289 (1987) 87.
[13] H. Kawai, D.C. Lewellen, S.H.-H. Tye, Nucl. Phys. B 288 (1987) 1.
[14] A.E. Faraggi, D.V. Nanopoulos, Phys. Rev. D 48 (1993) 3288.
[15] A.E. Faraggi, hep-th/9511093.
[16] A.E. Faraggi, hep-th/9708112.
[17] A.E. Faraggi, Phys. Lett. B 278 (1992) 131.
[18] A.E. Faraggi, Nucl. Phys. B 387 (1992) 239.
[19] A.E. Faraggi, Phys. Lett. B 326 (1994) 62.
[20] M. Dine, N. Seiberg, E. Witten, Nucl. Phys. B 289 (1986) 585.
[21] J. Atick, L. Dixon, A. Sen, Nucl. Phys. B 292 (1987) 109.
[22] E. Witten, Nucl. Phys. B 471 (1996) 135.
[23] D.V. Nanopoulos, M-phenomenology, CTP-TAMU-42/97, ACT-15/97, hep-th/9711080.
[24] J. March-Russell, Phys. Lett. B 437 (1998) 318.
[25] A.J. Buras, J. Ellis, M.K. Gaillard, D.V. Nanopoulos, Nucl. Phys. B 135 (1978) 66.
[26] A.E. Faraggi, Nucl. Phys. B 428 (1994) 111.
[27] A.E. Faraggi, Phys. Lett. B 398 (1997) 95.
[28] J. Pati, Phys. Lett. B 388 (1996) 532, hep-ph/9607446.
[29] G. Cleaver, Nucl. Phys. B (Proc Suppl.) 62 (1998) 161, hep-th/9708023.
[30] J. Ellis, A.E. Faraggi, D.V. Nanopoulos, Phys. Lett. B 419 (1998) 123.
[31] S. Kalara, J. Lpez, D.V. Nanopoulos, Phys. Lett. B 245 (1990) 421.
[32] S. Kalara, J. Lpez, D.V. Nanopoulos, Nucl. Phys. B 353 (1991) 650.
[33] A.E. Faraggi, Nucl. Phys. B 487 (1997) 55.

504

G.B. Cleaver et al. / Nuclear Physics B 593 (2001) 471504

[34] G. Cleaver, M. Cvetic, J.R. Espinosa, L. Everett, P. Langacker, Phys. Rev. D 57 (1998) 2701,
hep-th/9705391.
[35] M. Cvetic, L. Everett, J. Wang, Nucl. Phys. B 538 (1999) 52, hep-th/9807321.
[36] G.B. Cleaver, A.E. Faraggi, D.V. Nanopoulos, Non-Abelian flat directions in a minimal
superstring standard model, in preparation.

Nuclear Physics B 593 (2001) 505544

www.elsevier.nl/locate/npe

Worldsheet and spacetime properties of pp0

system with B field and noncommutative
geometry
B. Chen, H. Itoyama, T. Matsuo, K. Murakami
Department of Physics, Graduate School of Science, Osaka University, Toyonaka, Osaka 560-0043, Japan
Received 13 June 2000; accepted 13 October 2000

Abstract
We study worldsheet and spacetime properties of the pp0 (p < p0 ) open string system with
constant Bij field viewed from the Dp0 -brane. The description of this system in terms of the CFT
with spin and twist fields leads us to consider the renormal ordering procedure from the SL(2, R)
invariant vacuum to the oscillator vacuum. We compute the attendant two distinct superspace twopoint functions as well as their difference (the subtracted two-point function). These bring us an
integral (KobaNielsen) representation for the multiparticle tree scattering amplitudes consisting of
N 2 vectors and two tachyons. We evaluate them explicitly for the N = 3, 4 cases. Several novel
features are observed which include a momentum dependent multiplicative factor to each external
vector leg and the emergence of a symplectic tensor multiplying the polarization vectors. In the
zero slope limit, the principal parts of the amplitudes translate into a noncommutative field theory
in p0 + 1 dimensions in which a scalar field decaying exponentially in (p0 p) dimensions and a
noncommutative U (1) gauge field interact via the minimal coupling and a new interaction. A large
number of nearly massless states noted before are shown to propagate in the t-channel. 2001
Elsevier Science B.V. All rights reserved.
Keywords: Noncommutative field theory; pp 0 open string system; Renormal ordering; Subtracted two-point
function; Multiparticle scattering amplitudes; Gaussian damping factor
PACS: 11.25

This work is supported in part by the Grant-in-Aid for Scientific Research (10640268, 12640272, 12014210)
from the Ministry of Education, Science and Culture, Japan, and in part by the Japan Society for the Promotion
of Science for Young Scientists.
* Corresponding author.
E-mail addresses: [email protected] (B. Chen), [email protected]
(H. Itoyama), [email protected] (T. Matsuo), [email protected]
(K. Murakami).

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 1 5 - 5

506

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

1. Introduction
After decades of investigations, string perturbation theory has now become a wellestablished old subject. It is yet a nontrivial task to uncover spacetime properties, given a
first quantized worldsheet theory. Some of our previous endeavors on strings are regarded
as a search for a formulation in which these spacetime properties come out in a more
transparent way. String theory with constant Bij background offers an intriguing situation
in which the emerging spacetime picture is given in terms of noncommutative geometry
and is a focus of the recent intensive studies [15].
Several important steps have been taken in [6]. In particular, the proper spacetime metric
on the Dp-brane (the open string metric) has been extracted from the worldsheet theory
of an open string with its both ends on a Dp-brane (the pp open string system): the
distances measured with respect to this metric are kept finite at all scales. The attendant
noncommutative field theory in the zero slope limit lives on this metric together with the
parameter representing noncommutativity of spacetime.
In the previous paper [7], we have examined the more complex pp0 (p < p0 ) open
string system where the both ends of the open string are on a Dp-brane and on a Dp0 brane, respectively. We have obtained the open string metric and the noncommutativity
parameter on the Dp0 -brane from the worldsheet two-point function. They in fact agree
with those of the pp system. We have computed the spectrum in each case of (p, p0 ),
uncovered the emergence of a large number of nearly massless states in some cases and
clarified the connections among the GSO projection, branes at angles and supersymmetry.
Yet a number of other properties, in particular, spacetime properties on the Dp0 -brane with
Dp-brane inside have remained elusive. Elucidating upon these is a major goal of the
present paper.
It should be mentioned that, in the vanishing Bij background, several properties of the
pp0 open string system have been studied. For example, amplitudes of some scattering
processes taking place on the Dp-brane worldvolume (i.e., that of the lower-dimensional
D-brane) has been evaluated in [8] and the conformal field theory correlation functions
have been studied in [9]. In the presence of Bij background the four point tachyon
amplitudes have been given in [10] for the p0 = p + 2 case.
In [6] several properties of the 04 system have been derived. The description of
the system from the D0-brane has offered a new perspective to the moduli space of
noncommutative instantons [11] where the noncommutativity of the system is measured
by the presence of the FayetIliopoulos D term. In contrast to this moduli space point
of view, the thrust of the present paper is to uncover the spacetime properties of the
system viewed from the higher-dimensional D-brane, namely the Dp0 -brane. This can be
accomplished by placing vertex operators on the worldvolume of the Dp0 -brane and by
considering the scattering processes. This is an extension of the computation of scattering
amplitudes in the pp open string system with and without Bij background [5,1214]. This
line of reasonings has led us to carry out a systematic study which begins with evaluating
superspace two-point functions in the relevant CFT with the spin and twist fields, proceeds

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

507

to the computation of scattering amplitudes on the Dp0 -brane and ends with identifying a
proper low energy noncommutative field theory in the zero slope limit.
In the next section, we begin with quantizing an open string ending on Dp- and Dp0 (p < p0 ) branes. We exploit superspace formulation, which we find extremely efficient
in the calculation pursued in the subsequent sections. We evaluate two distinct two-point
functions on superspace and observe the importance of the renormal ordering procedure
from the SL(2, R) invariant vacuum to the oscillator vacuum. This procedure leads us to
consider the difference of these two two-point functions as well. This third quantity plays
an important role in Section 4 and will be referred to as subtracted two-point function.
The conformal weights of the twist and spin fields are readily computed from the renormal
ordering procedure.
In Section 3, we examine the tachyon vertex operator of the pp0 open string and the
vector vertex operator of the p0 p0 open string. We derive the on-shell conditions in terms
of the open string metric (p +1)-dimensional momenta of the tachyon (p0 +1)-dimensional
momenta and polarizations of the massless vector and mass of the tachyon.
In Section 4, we consider the multiparticle tree scattering amplitudes consisting of
external states of N 2 vectors and two tachyons. We are able to derive an integral
(KobaNielsen) representation of these quantities as integrals over N 3 locations of the
vertex operators as well as the N 2 Grassmann counterparts and the N 2 Grassmann
sources conjugate to the polarization vectors. (See [15] for the case of the pp string with
vanishing B field.) Several striking properties emerge from this representation. Among
other things, we find a momentum dependent exponential factor to each external vector
leg, which the subtracted two-point function is responsible for, as well as a new symplectic
tensor multiplying the polarizations of the massless vector. We observe that some parts of
the amplitudes are expressible in terms of the inner products of polarizations, momenta
and the symplectic tensor with respect to the open string metric, while there are a host of
other parts which do not permit such generic description by the inner product. We evaluate
the amplitudes for the N = 3, 4 cases explicitly.
In Section 5, we examine the zero slope limit of the system. We find that the parts of the
amplitudes expressible in terms of the inner product in this limit (the principal parts) can
be summarized as a noncommutative field theory of a scalar field and a noncommutative
U (1) gauge field in p0 + 1 dimensions in which the scalar field decays exponentially in
0
the x p+1 , . . . , x p -directions. They interact via the minimal coupling and a new interaction
which consists of the field strength and the scalar bilinear contracted with the symplectic
tensor. The contributions from the residual parts are consistent with the propagations in the
t-channel of a large number of nearly massless states found in [7].
2. Basic properties of pp0 system with Bij field
In this section, we will provide the two-point functions, and the twist and the spin fields
for a pp0 open string in constant B field background. This will also help us to establish our
notations. We introduce two types of normal ordering: the one is taken with respect to the

508

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

SL(2, R) invariant vacuum and the other is with respect to the oscillator vacuum. We will
establish the relationship between these two, which will be important for our calculation
in the subsequent sections.
2.1. Action and boundary condition
The action of the NSR superstring in the constant B background takes the form of
Z
Z


1
d 2 d d g + 2 0 B DX (z, z)DX (z, z),
(2.1)
S=
2
where z = (z, ) and z = (z, ) are the superspace coordinates on the worldsheet, D =
/ + /z and D = / + /z are the superspace covariant derivatives and g
denotes the spacetime metric which is taken to be flat. z = 1 + i 2 and z = 1 i 2
are complex coordinates on the plane which are related to the strip coordinates (, ) by
z = e +i and z = e i , respectively. The superfield X (z, z) is the string coordinate
which is expressed in terms of component fields as
r
2

e (z, z) + i F (z, z) .
X (z, z) + i (z, z) + i
(2.2)
X (z, z) =
0
In terms of the component fields the action (2.1) is given by


Z

2
1
2
0

e
e
X X + .
d g + 2 B
S=
2
0

(2.3)

Here we have eliminated the auxiliary field F (z, z) by using the equation of motion F =
0, and the operators and denote /z and /z, respectively. Since the B dependent
terms do not couple to the worldsheet metric, the energy-momentum tensor of this system
has the same form as that of the string without B field background:

with

1
T(z) TF (z) + TB (z) g DX D 2 X ,
2

(2.4)

i
2

g X ,
TF (z) =
2 0

1
1

TB (z) = 0 g X X g .

(2.5)

Let us consider a pp0 open string in the type IIA theory with p < p0 and p and
p0 being even integers. We concentrate on the situation in which a Dp-brane extends in
0
the (x 0 , x 1 , . . . , x p )-directions and a Dp0 -brane extends in the (x 0 , x 1 , . . . , x p )-directions
with the Dp-brane inside. The worldsheet of the open string corresponds to the upper halfplane: Im z > 0 ( 0 6 6 ). The Dp-brane worldvolume contains the boundary = 0
while the Dp0 -brane worldvolume contains the boundary = . As the spacetime is flat
with the metric

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

509

gij
1
..

gij = ij

(i, j = 1, . . . , p0 ),

(2.6)

1
we can bring B into a canonical form

0
b1

b1 0

0
b2
Bij =

2 0

b2 0

..

(i, j = 1, . . . , p0 ),

B = 0 otherwise.

(2.7)

In what follows we will investigate the system on this background.

The boundary conditions for the string coordinates in the NS sector are [6,7]


0
0
DXp +1,...,9 + DXp +1,...,9  =0, = = 0,
DX0 DX0  =0, = = 0,




gkl DXl DXl + 2 0 Bkl DXl + DXl  =0, = = 0 (k, l = 1, . . . , p),





DXi + DXi  =0 = = gij DXj DXj + 2 0 Bij DXj + DXj  = = = 0
(i, j = p + 1, . . . , p0 ).

(2.8)

For the bosonic components these conditions read





0
+ Xp +1,...,9  =0, = 0,
X0  =0, = 0,




gkl Xl + 2 0 Bkl + Xl  =0, = 0 (k, l = 1, . . . , p),






+ Xi  =0 = gij Xj + 2 0 Bij + Xj  = = 0
(i, j = p + 1, . . . , p0 ),

(2.9)

and for the fermionic components



p 0 +1,...,9
e0 
ep0 +1,...,9 
=
0,

= 0,
0
=0,
=0,




l
l
0
l
l
e + 2 Bkl +
e 
= 0 (k, l = 1, . . . , p),
gkl
=0,





ei 
ej + 2 0 Bij j +
ej 
= gij j
=0
i +
=0
=
(i, j = p + 1, . . . , p0 ).

The boundary conditions for the R-fermions are



e0 
e0 
= 0 +
= 0,
0
=0
=


0
0
0
ep +1,...,9 
ep0 +1,...,9 
= p +1,...,9
= 0,
p +1,...,9 +
=0
=

(2.10)

510

B. Chen et al. / Nuclear Physics B 593 (2001) 505544




el + 2 0 Bkl l +
el 
el
gkl l
= gkl l +
=0

el )
= 0 (k, l = 1, . . . , p),
+ 2 0 Bkl ( l
=





ei 
ej + 2 0 Bij j
ej 
= gij j +
=0
i +
=0
=
(i, j = p + 1, . . . , p0 ).

(2.11)

It should be noted that these boundary conditions are written on the complex plane, while
the boundary conditions in [7] are written on the strip.
2.2. Quantization of a pp0 string
In this subsection we will give the mode expansions of the string coordinates of a pp0
open string and the commutation relations among their modes.
Let us first consider the x 0 -direction. In this direction the string coordinate obeys the
Neumann boundary condition on both ends. The coordinate X0 (z, z) is expanded as
r
0


0 X m
0
0
0 0
zm + zm .
(2.12)
X (z, z) = x i p ln(zz) + i
2
m
m6=0

The modes satisfy the following commutation relations,

 0 0  0 0
 0 0
 0 0
x , x = p , p = 0,
m , n = g 00 mm+n .
x , p = ig 00 ,

(2.13)

e0 become
The mode expansions of the NS fermions 0 and
X
X
e0 (z) =
br0 zr1/2,

br0 zr1/2 .
0 (z) =

(2.14)

rZ+1/2

rZ+1/2

The oscillators satisfy

 0 0
br , bs = g 00 r+s .
In the R sector, we have
X
0 m1/2
dm
z
,
0 (z) =
mZ

(2.15)

e0 (z) =

0 m1/2
dm
z

(2.16)

mZ

and


0
, dn0 = g 00 m+n .
dm

(2.17)

Next we consider the x i -directions, i = 1, . . . , p. In these directions, the boundary

conditions on the string coordinates are the same as those on the string coordinates
along the Dp-branes in the DpDp system with B field. This implies that the mode
expansions and the commutation relations among the oscillators take the same form as
those in the DpDp system in the B field background. The mode expansions of the bosonic
coordinates Xi (z, z) are
i
i


Xi (z, z) = x i i 0 g 1 (g 2 0 B) j pj ln z i 0 g 1 (g + 2 0 B) j pj ln z

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

r
+i

511

i
0 X h 1
g (g 2 0 B) j zm
2
m6=0

i j

i
m
.
(2.18)
+ g 1 (g + 2 0 B) j zm
m
Under this expansion, the commutation relations among the oscillators are given [3,4] by
 i j
 i j
 i j
p , p = 0,
x , p = iGij ,
x , x = i ij ,
 i j
(2.19)
m , n = Gij mm+n ,
where ij is the noncommutativity parameter [6] and Gij is the inverse of the open string
metric Gij [6] defined respectively as
ij

1
1
B
,
ij = (2 0 )2
g + 2 0 B g 2 0 B
ij

1
1
ij
g
.
(2.20)
G =
g + 2 0 B g 2 0 B
We will further generalize Gij to include the time direction. This is denoted by G and
will simplify our formulas in the subsequent sections. Taking Eqs. (2.6) and (2.7) into
account, we obtain
00

G =
ij

1
1
(1 + b12 )
0

0
1
(1 + b12)

1
(1 + b22 )
0

0
1
(1 + b22 )

..

(2.21)

We obtain for the NS-fermions

X 
i j

i (z) =
g 1 (g 2 0 B) j br zr1/2 ,

rZ+1/2

ei (z) =

X 

rZ+1/2

and for the R-fermions

i j
g 1 (g + 2 0 B) j br zr1/2 ,

(2.22)

j
with bri , bs = Gij r+s ,

512

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

X
i
i j
(z) =
g 1 (g 2 0 B) j dm zm1/2 ,

mZ
X
i j
i

e (z) =

g 1 (g + 2 0 B) j dm zm1/2 ,

(2.23)
 i j
with dm
, dn = Gij m+n .

mZ

Finally, we investigate the x i -directions (i = p + 1, . . . , p0 ). We complexify the string

coordinates Xi (z, z) in these directions as
r
2 I
I
2I 1
2I
eI (z),
(z, z) + iX (z, z) =
Z (z, z ) + i I (z) + i
Z (z, z) = X
0
r
2 I
I
I
e I (z), (2.24)
2I 1
2I
Z (z, z) = X
(z, z) iX (z, z) =
Z (z, z) + i (z) + i
0

where I, I = (p + 2)/2, . . . , p0 /2 and we have eliminated the auxiliary field F i . From the
boundary conditions (2.9) and equations of motion, we find that the mode expansions of
I

Z I and Z are given by

r
I


0 X n
I
I
z(nI ) z(nI ) ,
Z (z, z) = i
2
n I
nZ
r
0


X Im+I (m+I )
I
Z (z, z) = i
z(m+I ) ,
z
2
m + I

(2.25)

mZ

where I are defined by

e2iI =

1 + ibI
,
1 ibI

0 < I < 1.

(2.26)

Now we can introduce the open string metric GI J , GI J , GI J and GI J concerning the
0
x p+1 , . . . , x p directions,
2
I J .
GI J = GJ I =
(2.27)
GI J = GI J = 0,
(1 + bI2 )
Similarly, the boundary conditions (2.10), (2.11) and equations of motion lead us to the
mode expansions of the NS-fermions,
X
X
I
I
eI (z) =
br
z(rI )1/2,

br
z(rI )1/2,
I (z) =
I
I
rZ+1/2
I

(z) =

rZ+1/2

bs+I z(s+I )1/2,

I

e I (z) =

sZ+1/2

bs+I z(s+I )1/2 ,

I

sZ+1/2

(2.28)
and that of R-fermions,
X
I
dn
z(nI )1/2,
I (z) =
I

eI (z) =

nZ
I

(z) =

mZ

I
dn
z(nI )1/2 ,
I

nZ
I
d m+I z(m+I )1/2 ,

eI

(z) =

mZ

d m+I z(m+I )1/2 . (2.29)

I

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

The commutation relations are


 I
2
nI , Jm+J = I J (n I )n+m ,

 I

2
 I
2 IJ
J
J
dnI , d m+J = I J n+m .
brI , bs+J = r+s ,

513

(2.30)

As for the x p +1 , . . . , x 9 -directions, the string coordinates obey the Dirichlet boundary
condition. Our analysis in the remaining part of this paper does not involve these directions.
2.3. Two-point functions on superspace
In this subsection we construct two-point functions of the pp0 open string coordinates
on superspace. For this purpose, we begin by defining the oscillator vacuum of the system.
As shown in the last subsection, the mode expansions in the x 0 - and the x i -directions
(i = 1, . . . , p) are similar to those of the usual open strings obeying Neumann boundary
conditions in the sense that the bosons and the R fermions have integral moding oscillators
and the NS fermions have half-integral moding ones. Therefore we can define the vacuum
in these directions in the same way as the usual open string. In the NS sector the vacuum
|0i is defined by
( 0
i
|0i = 0, for m > 0,
m |0i = 0, m
(2.31)
0
i
br |0i = 0, br |0i = 0, for r > 12 ,

where 0 = 2 0 p ( = 0, 1, . . . , p). In the R sector the vacuum |S i belongs to the

spinor representation of the SO(p, 1) group. Now that the commutation relations and
the vacuum |0i are determined, we can evaluate the two-point functions of the string
coordinates in these directions [6,12,16],
G00 (z1 , z1 |z2 , z2 ) h0|RX0 (z1 , z1 )X0 (z2 , z2 )|0i



= g 00 ln(z1 z2 1 2 ) z1 z2 1 2




+ ln z1 z2 1 2 z1 z2 1 2 ,
Gij (z1 , z1 |z2 , z2 ) h0|RXi (z1 , z1 )Xj (z2 , z2 )|0i

= g ij ln(z1 z2 1 2 ) z1 z2 1 2






+ g ij 2Gij ln z1 z2 1 2 z1 z2 1 2
2

z1 z2 1 2
ij
ln
2D ij ,
0
2
z1 z2 1 2

(2.32)

where R stands for the radial ordering. D ij are the contributions from the zero modes x i
and these will be fixed conveniently as is done in [6]. When we restrict these two-point
functions onto the Dp0 -brane worldvolume, i.e., the worldsheet boundary characterized by
z = e +i = e and = , they become



G00 e1 , 1 | e2 , 2 G00 (z1 , z1 |z2 , z2 ) =,=

2
= 2g 00 ln e1 e2 + 1 2 ,

514

B. Chen et al. / Nuclear Physics B 593 (2001) 505544




Gij e1 , 1 | e2 , 2 Gij (z1 , z1 |z2 , z2 ) =,=
= 2Gij ln e1 e2 + 1 2

2

i ij
(1 2 ), (2.33)
0

where (x) is the sign function.

In the x i -directions (i = p + 1, . . . , p0 ), the situation is more complex as the string
coordinates are expanded in non-integer power of z and z. We define the oscillator vacuum
| i for the bosonic sector so that this should be annihilated by the negative energy modes:
( I
O
nI |I i = 0, n > I ,
(2.34)
|I i with
| i =
Im+I |I i = 0, m > I .
I
For the fermions in the NS sector we define the oscillator vacuum |si by 1
( I
O
brI |sI i = 0, r > 12 ,
|si =
|sI i with
I
bs+I |sI i = 0, s > 12
I
and for the R sector we define the oscillator vacuum |Si by
( I
O
dnI |SI i = 0, n > I ,
|SI i with
|Si =
I
d m+I |SI i = 0, m > I .
I

(2.35)

(2.36)

By using the commutation relations and the defining relations of the vacua, we can
calculate the two-point functions 2 :
1 The vacuum |si is defined in order that the negative energy and the positive energy modes for 0 < < 1/2
I

should be the annihilation and the creation modes, respectively. The energy carried by the lowest creation mode
I

b1/2+I becomes negative when I becomes greater than 1/2. We could define another oscillator vacuum |e
si
by

I
br
|e
s i = 0, r > 32 ,
O
I I
|e
sI i with
|e
si=
bI
sI i = 0, s > 21 ,
s+I |e
I
which makes the negative energy and the positive energy modes for 1/2 < I < 1 to be the annihilation and the
creation modes, respectively.
2 If we adopted |e
si as the vacuum of the NS fermions instead of |s i, the two-point function of the
I

supercoordinates ZI and Z would be

J
IJ
e
G (z1 , z1 |z2 , z2 ) h,e
s|RZI (z1 , z1 )Z (z2 , z2 )|,e
si
!
!
"
2 I J
z2
z2
= (|z1 | |z2 |)
+ F 1 I ;
F 1 I ;

z1 1 2
z1 1 2
!
!#
z2
z2
F 1 I ;
F 1 I ;
z1 1 2
z1 1 2
"
!
!
z1 1 2
2 I J
z1 1 2
+ (|z2 | |z1 |)
F I ;
+ F I ;

z2
z2
!
!#
z 1 2
z 1 2
F I ; 1
F I ; 1
.
z2
z2

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

515

G I J (z1 , z1 |z2 , z2 )
J

h, s|RZI (z1 , z1 )Z (z2 , z2 )|, si

 




2 I J
z2 + 1 2
z2 + 1 2
F 1 I ;
+ F 1 I ;
= |z1 | |z2 |

z1
z1




z2 + 1 2
z 2 + 1 2
F 1 I ;
F 1 I ;
z1
z1
 


I
J

2
z1
z1
F I ;
+ F (I ;
+ |z2 | |z1 |

z2 + 1 2
z + 1 2

 2


z1
z1
F I ;
, (2.37)
F I ;
z 2 + 1 2
z2 + 1 2
where (x) is the step function, F (; z) is defined as

F (; z) =

X 1
z
F (1, ; 1 + ; z) =
zn+ ,

n+

(2.38)

n=0

and F (a, b; c; z) is the hypergeometric function. When we restrict this two-point function
onto the worldsheet boundary on the Dp0 -brane worldvolume, this becomes



G I J e1 , 1 | e2 , 2 G I J (z1 , z1 |z2 , z2 ) =,=



e2 1 2
= 4GI J (1 2 )F 1 I ;
e1



e1
.
(2.39)
+ (2 1 )F I ;
e 2 1 2
Now we would like to study the two-point function (2.39) more closely. Let us recast
the right-hand side of Eq. (2.39) into
  




e2
e1
(e1 /e2 )I
1
F 1 I ; + F I ;
1 2
4GI J
2
e1
e2
e 1 e2





4 I J
e2
e1
F 1 I ; F I ;
.
(2.40)
+ (1 2 )
1 + bI2
e1
e2
As noncommutativity of the Dp0 -brane worldvolume originates from the term proportional
to the sign function (1 2 ) in the above equation [6,7], we will also refer to this term as
noncommutativity term in what follows. Here it should be noted that by using the relations

d  c1
z F (a, b; c; z) = (c 1)zc2 F (a, b; c 1; z),
dz
F (a, b; b; z) = (1 z)a
we can obtain

 

1
d
F 1 I ;
F (I ; z) = 0.
dz
z

(2.41)

(2.42)

516

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

This implies that the noncommutativity term in Eq. (2.40) is constant. The value of this
constant can be fixed by evaluating the noncommutativity term at a certain point on the
real axis, such as e1 /e2 = 1. By using the hypergeometric series, we find that
F (1 I ; 1) F (I ; 1) =

1
= cot(I ) = bI .
n
+
I
n=

Thus we find that the noncommutativity term becomes

 



e2
e1
4 I J
4 I J
;
;
bI .
F
1

=
I
I
2

1 + bI
e1
e2
1 + bI2

(2.43)

(2.44)

When we rewrite the complex string coordinates ZI and Z into the real one Xi (i =
p + 1, . . . , p0 ), this noncommutativity term takes the same form as that in Eq. (2.33). This
means that, as is pointed out in [7], the noncommutativity on the D-brane worldvolume
in the pp0 system is the same as that in the pp system [3,4,6]. From Eq. (2.44), we
conclude that


G I J e1 , 1 | e2 , 2


e1
4 I J
bI ,
(2.45)
+ (1 2 )
= 4GI J H I ;
e 2 1 2
1 + bI2
where H(; z) is defined by using the hypergeometric series as



zn1+I

b
;
=
bI

F
1

I
I

z
2
n + 1 I
2
n=0
H(; z) =

zn+I

b
;
z)
+
=
+ bI
F
(
I
I

2
n + I
2

for |z| > 1,

(2.46)
for |z| < 1.

n=0

The two infinite series in the above defining relation should be analytically continued to
each other.
In Appendix A we give another derivation of the two-point function (2.45) from
Eq. (2.39).
2.4. Twist field and spin field
Here we would like to make further consideration on the oscillator vacuum consisting of
the bosonic sector | i and the fermionic sector |si. As is explained in the last subsection,
I

the primary fields DZI , DZI , DZ and D Z defined on the upper half plane are expanded
in non-integer powers of z and z. It follows that when we extend the defining region of
these fields to the whole complex plane through the doubling trick these fields become
multi-valued functions on the whole plane. For example, when the primary fields Z I (z),
I

Z I (z), Z (z) and Z (z) on the whole plane are transported once around the origin,

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

they gain phase factors:

Z I e2i z = e2iI Z I (z),


I
I
Z e2i z = e2iI Z (z),

Z I e2i z = e2iI Z I (z),


I
I
Z e2i z = e2iI Z (z).

517

(2.47)

This implies that a twist field I+ ( 1 ) and an anti-twist field I ( 1 ), both of which are
I

mutually non-local with respect to Z I and Z , are located at the origin and at infinity on
the plane, respectively. They create a branch cut between themselves. The twist field +
serves as a boundary changing operator from the p0 -brane to the p-brane and the anti-twist
field acts in the opposite way [8,10]. The incoming vacuum |I i defined in Eq. (2.34)
should be interpreted as being excited from the SL(2, R)-invariant vacuum |0i by the twist
field I+ :


(2.48)
|I i = lim I+ 1 |0i.
1 0

In the same way, the outgoing vacuum hI | should be regarded as

 2h


I
1
1

h0|I 1 ,
hI | = lim
e
e
1

1 0 e

(2.49)

where hI denotes the weight of the (anti-) twist field. We will later explain that hI =
I

1
I
2 I (1 I ) [17]. We can read off the OPEs of Z and Z with I from Eq. (2.34):
( I
Z I (z)J+ (0) JI z(1I )e
I+ (0),
Z (z)J+ (0) JI z(1I ) I+ (0),
(2.50)
I
I
Z (z)J+ (0) JI zI e
I0+ (0),
Z (z)J+ (0) JI zI I0+ (0),

Z I (z)J (0) JI zI I (0),

Z I (z)J (0) JI zI e
I (0),
(2.51)
I
Z I (z) (0) I z(1I ) 0 (0),
Z (z)J (0) JI z(1I )e
I0 (0),
J
J
I
where s are excited twist fields.
Similar argument holds for the fermionic coordinates. In the NS sector, the spin fields sI+
and sI are mutually non-local with respect to the fermions. They are located at the origin
and at the infinity on the worldsheet, respectively. They exchange the boundary conditions
corresponding to the p-brane and those to the p0 -brane by generating a branch cut between
themselves. The incoming vacuum |sI i and the outgoing vacuum hsI | should be regarded
as being excited from the SL(2, R)-invariant vacuum by spin fields:
 2hs


I


1
1

h0|
s
hsI | = lim

,
(2.52)
|sI i = lim sI+ 1 |0i,
I
e
e
1
1
1 0
1 0 e
where hsI is the weight of the spin fields which will be found to be hsI = 12 I 2 . The
defining relation (2.35) yields the OPEs

eI (z)s + (0) I z+I t 0+ (0),

I (z)sJ+ (0) JI z+I tI0+ (0),

J
J
I
(2.53)
I
I (z)s + (0) I zI t + (0),
e (z)s + (0) I zI t+ (0),

J
J
J
I
J

518

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

I (z)sJ (0) JI zI tI0 (0),

eI (z)s (0) I zI t 0 (0),

J
J
I

I (z)s (0) I z+I t (0),

e I (z)s (0) I z+I t (0).

J
J
I

(2.54)

In the fermionic sector, the bosonization simplifies the treatment of the spin fields. While
we will not invoke this treatment here, we include it here for the sake of completeness. We
write
r
r
2 iH I (z)
2 iH I (z)
I
I

e
e
,
(z) =
,
(2.55)
(z) =

where H I (z) are free bosons normalized as H I (z)H J (w) I J ln(z w). Then the
OPE (2.54) tells us that the spin fields sI (z) should be bosonized as 3
I
sI+ (z)
= e+iI H (z) ,

I
sI (z)
= eiI H (z) .

(2.56)

We can repeat the same analysis in the R sector. We find that the incoming vacuum
Eq. (2.36) and the outgoing vacuum are excited from the SL(2, R)-invariant vacuum by
the spin fields SI+ (z) and S (z), respectively. They are bosonized as
SI+ (z)
= ei(1/2+I )H

I (z)

SI (z)
= ei(1/2+I )H

I (z)

(2.57)

2.5. Subtracted two-point functions and weights of twist and spin fields
In the x i -directions (i = p + 1, . . . , p0 ), we have two types of vacuum: the one is the
SL(2, R)-invariant vacuum and the other is the oscillator vacuum. We can define the normal
ordering corresponding to each one. We will use the symbols : : and to denote the
normal orderings with respect to the SL(2, R)-invariant vacuum and the oscillator vacuum,
respectively.
In the other directions we have a single type of vacuum, namely SL(2, R)-invariant
vacuum. We have : :-normal ordered product only. For free bosons and free fermions in
these directions, it is defined by a subtraction:
: X (z1 , z1 )X (z2 , z2 ) : = RX (z1 , z1 )X (z2 , z2 ) G (z1 , z1 |z2 , z2 ),

(2.58)

for , = 0, 1, . . . , p. Here G (z1 , z1 |z2 , z2 ) is the two-point function defined in

Eq. (2.32).
In the same way we can define -normal ordered product for the free fields in the
x i -directions (i = p + 1, . . . , p0 ) as

ZI (z1 , z1 )Z (z2 , z2 ) = RZI (z1 , z1 )Z (z2 , z2 ) G I J (z1 , z1 |z2 , z2 ),

(p + 2)/2, . . . , p0 /2.

(2.59)

Here G (z1 , z1 |z2 , z2 ) is the two-point function defined

for I, J =
in Eq. (2.37). In addition to the -normal ordered product, we would like to define
IJ

3 We are also able to perform the same analysis on the other incoming and the outgoing vacua, |e
sI i and he
s |,
in the NS sector defined in the footnote 1. These states are excited from the SL(2, R)-invariant vacuum by e
sI+ (z)
and e
s I (z), respectively, which are bosonized as
I
e
s I+ (z)
= e+i(1+I )H (z) ,

I
e
s I (z)
= ei(1+I )H (z) .

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

519

the : :-normal ordered product for these free fields ZI and Z . In order to apply the
definition (2.58) directly to these fields, we have to evaluate their two-point functions on
the SL(2, R)-invariant vacuum. As is pointed out, the twist and the spin fields play the
role of boundary changing operators. This implies that if we delete the twist and the spin
fields and thus remove the cut generated by them, the boundary conditions imposed on the
positive real axis of the z-plane is expected to be identical to those imposed on the negative
I

real axis. This leads us to conclude that the two point function of ZI and Z evaluated
on the SL(2, R)-invariant vacuum takes the same form as that of the p0 p0 system with B
field:
J

GI J (z1 , z1 |z2 , z2 ) h0|RZI (z1 , z1 )Z (z2 , z2 )|0i




2 I J
ln(z1 z2 1 2 ) z 1 z 2 1 2
=

+ ln z1 z 2 1 2 z 1 z2 1 2

2
1 + bI2

ln z1 z 2 1 2 (z1 z2 1 2 )

z1 z 2 1 2
bI
ln
2i
2
1 + bI z 1 z2 1 2
+ D-term.

(2.60)

Computing this two point function at the boundary = and = , we obtain




GI J e1 , 1 | e2 , 2 GI J (z1 , z1 |z2 , z2 )
=,=

4 I J

ln e1 + e2 1 2

(1 + bI2 )
4 I J bI
(1
+
(1 + bI2 )

2 ).

2

(2.61)

From the fact that the normal ordered product is defined by a subtraction, we can readily
I

find that for an arbitrary functional O of the free fields ZI and Z the normal ordering is
formally expressed as (see, e.g., [18])

 Z

2
2
IJ
O,
: O := exp d z1 d z2 G (z1 , z1 |z2 , z2 ) I
Z (z1 , z1 ) J
Z (z2 , z2 )
 Z


2
2
IJ

d z1 d z2 G (z1 , z1 |z2 , z2 ) I
O, (2.62)
O = exp
Z (z1 , z1 ) J
Z (z2 , z2 )
where d 2 z is defined as d 2 z = d 2 d d . From these general definitions of the normal
orderings, we can read off the formula of the reordering between them:

Z

O . (2.63)
: O := exp
d 2 z1 d 2 z2 G sub I J (z1 , z1 |z2 , z2 ) I
Z (z1 , z1 ) J
Z (z2 , z2 )

520

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

Here G subI J (z1 , z1 |z2 , z2 ) is a subtracted two-point function defined as

J

G sub I J (z1 , z1 |z2 , z2 ) h, s| : ZI (z1 , z1 )Z (z2 , z2 ) : |, si

= G I J (z1 , z1 |z2 , z2 ) GI J (z1 , z1 |z2 , z2 ).

(2.64)

Let us compute the subtracted two-point function G sub at the worldsheet boundary
= and = . From Eqs. (2.39) and (2.61), we find that


G sub I J e1 , 1 | e2 , 2

J
h, s| : ZI e1 , 1 Z (e2 , 2 ) : |, si
"
(I ) + (1 I ) 1
8 I J
+ ln e1 +2

=
2
2
(1 + bI2 )
(



1 e1
e2 1I X (1 I )n
ln
+ (1 2 )
2 e2
e1
n!
n=1

 )
n1
X
I
e2 n
1

(m + 1)(m + 1 I )
e1
m=0
(
 1 I X

1 e2
e
(I )n
ln
+ (2 1 )
2 e1
e2
n!
n=1

 )#
n1
X
1 I
e1 n
1

(m + 1)(m + I )
e2
m=0

 1 I
1
e
8 I J
1
,
(2.65)
+ 1 2
e2
e1 + e2
(1 + bI2 )
IJ

where is Eulers constant, (w) denotes the digamma function defined as (w) =
(d/dw) ln (w) and (a)n (a + n)/ (a). Here we have used the formulas (A.1) and
(A.5).
Let us compute the conformal weights of the twist and spin fields. This helps us
appreciate these subtracted two-point functions better. The relevant part of the energymomentum tensor in this computation consists of the terms depending on the string
(z) = TB
(z) + TB
(z)
coordinates in the x i -directions, i = p + 1, . . . , p0 : TB
with

J
(Z,Z)
(z) = I0 J : Z I Z (z) :,
TB



J
J
(, )
(z) = I J : I (z) : : I (z) : .
(2.66)
TB
4
Using the subtracted two point function, we obtain

I
hI |TB (z)|I i = 0 lim hI | : Z I (w)Z (z) : |I i
wz


 I 1

1
1
1 I
1
w
+

= lim
wz
z
wz
z
wz
(w z)2
(Z, )

(Z,Z)

(, )

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

1 I (1 I )
.
2
z2

521

(2.67)

This implies that the twist fields I have the weights hI = 12 I (1 I ). In the same way,
we obtain

I
I
hsI |TB (z)|sI i = lim hsI | : I (w) (z) I (w) (z) : |sI i
4 wz
 


 I

I 1
1
1
1
1
w

+
+
= lim
wz
z
wz 2 w z
wz
(w z)2
=

1 I 2
.
z2 2

(2.68)

From this equation we can read 4 that the spin fields sI have the weights hsI = 12 I 2 .
3. Vertex operators
Let us pay some attention to vertex operators of our system before we start calculating
scattering amplitudes. We focus on two types of vertex operators. The one is the tachyon
vertex operators, and the other is the massless vector vertex operators. These are the
relevant ones in order for us to find out the spacetime processes occurring on the Dp0 brane worldvolume.
Here we make a comment on the GSO projection. In this paper we take the GSO
projection in the NS sector so that the oscillator vacuum corresponding to the tachyon
vertex operator survives. It follows that the GSO projection adopted in this paper is not
always the same as that in our previous work [7]: in the cases of p0 = p + 2, p + 6 they are
opposite to each other, while in the cases of p0 = p + 4, p + 8 they are the same. This is
attributed to the sign convention of the Dp-brane charge: in the cases of p0 = p + 2, p + 6
the Dp-branes in this paper should be referred to as anti-Dp-branes in the convention
of [7].
3.1. Tachyon vertex operator of pp0 string
First let us investigate vertex operators of the pp0 open string which contain the twist
and the spin fields. We will focus on the vertex operator that corresponds to the ground
state in the NS sector of this open string. This vertex operator is seen, for instance, in [10,
17]:


(1) 1

(0) 1

1
+ VT

V
T , = VT
!
r
p
0




X
1
1
k X , :,
(3.1)
= T , : exp i
2
=0

4 This result can also be obtained from the fact that T (, ) (z) is bosonized as T (, ) (z)
=
B
B

12 I J : H I H J (z) :.

522

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

where X ( 1 , ) is the boundary value of the superfield X (z, z): X ( 1 , )

X (z, z)z=z= 1 ,= , and T ( 1 , ) is the superfield whose lowest component T0 ( 1 )
consists of the twist and the spin fields,

Y 1
1

I sI .
(3.2)
T0 1 =
I

The upper component is obtained by applying the supercurrent TF (z) to T0 ( 1 ) [17].

(0)
(1) 1
( ) is the matter field
In Eq. (3.1), VT ( 1 ) denotes the 0-picture vertex and VT
contribution to the (1)-picture vertex
!
p
Y
X

(1) 1

(1) 1


1

I sI : exp i
k X
=e
VT
=e
:,
(3.3)
VT
I

=0

comes from the -ghost sector.

where
It is worth noting that spacetime momentum k of the tachyon vertex operator (3.1) is not
(p0 + 1)-dimensional but (p + 1)-dimensional. This is because the pp0 string coordinates
0
in x p+1 , . . . , x p -directions do not possess zero-modes and thus the momenta in these
directions are not defined. This implies that an initial/final tachyon field corresponding
to this vertex operator is frozen in these spacetime directions.
Let us study the physical state conditions for this vertex operator. If the (1)-picture
vertex satisfies the physical state condition, the 0-picture vertex is automatically physical
because of the worldsheet supersymmetry. We will therefore concentrate on the (1)picture vertex. Through the operatorstate mapping this vertex operator corresponds to the
state 5 ,
 (1) 
(1) 1

V
(3.4)
lim V
|0i = |0; ki |, si,
T

1 0

Pp
k x )|0i. Here x are the zero
=0(1) 
is a primary state. We just
modes of the bosonic coordinates. It is evident that VT
require that this state should have weight 12 . From the calculation in the last subsection, we
find that the state |, si has a weight
 X


 X I (1 I ) I 2
I
+
.
(3.5)
=
h |, si =
2
2
2
where the state |0; ki is defined as |0; k i = exp(i

Substituting the mode expansions (2.18) and (2.22) into the defining relation (2.5), we
see that the terms in TB (z) which depend on the string coordinates in x -directions ( =
0, 1, . . . , p) are
 X
p 
X
1
1
(X,)

(z)
g X X (z) + g (z)
Lm zm2 ,
TB
0
2
=0

mZ

with
Lm =

p
1X X
1

G : mn
n : +
2
4
nZ ,=0

5 Here we ignore the ghost sector.

(2r m)

rZ+1/2

p
X
,=0

G : bmr
br :, (3.6)

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

523

where 0 = 2 0 p . Here G is the open string metric including time direction, its
inverse is given in Eq. (2.21). This yields
L0 |0; k i = 0

p
X

G k k |0; ki.

(3.7)

,=0
(1)

Gathering all results obtained above, we conclude that the weight of the state |VT i is
P
Pp
L0 = I 21 I + 0 ,=0 G k k . Thus the on-shell condition L0 = 12 requires that the
mass squared of this state should be

p
X 
X
1
0 2
0

G k k = 1
I .
(3.8)
mT
2
I

,=0

3.2. Massless vector vertex operator of p0 p0 string

The vector emission vertex operator takes the form of






(1) 1
(0) 1
+ Vvec

Vvec 1 , Vvec
!
r
p0
p0
0 X

i X
1 , exp i
: (k)X
k X 1 , :,

2
2
=0

(3.9)

=0

( 1 , ) (D +D)X (z, z)|

where (k) denotes the polarization vector and X
z=z= 1 ,= .
(0) 1
(1) 1
( ) is the 0-picture vertex and Vvec
( ) is the matter field contribution
The operator Vvec
(1) 1
(1)

1
to the (1)-picture vertex Vvec ( ) = e Vvec ( ). Their explicit forms are
!
p0
p0
X
X




1
(1) 1
e exp i
: (k) +
k X :,
=
Vvec
2
=0

(0)
Vvec

=
2 0

p0
X
=0

0
: (k) i X +
2

exp i

=0

"

p
X

k X

p0
X

k +

!
e

#

=0

:,

(3.10)

=0

where X ( 1 ) is defined as X ( 1 ) = ( + )X (z, z)|z=z= 1 .

The string coordinates of a p0 p0 open string in the x i -directions (i = 1, . . . , p0 ) obey
the same boundary conditions as the x i -directions with i = 1, . . . , p of the pp0 string.
They have the same mode expansions and the commutators among the oscillating modes.
Therefore the vector emission vertex operator in each picture corresponds to the respective
state
0

p
X
 (1) 

(1) 1
V
V
(k) b1/2|0; k i0 ,

lim

|0i
=

vec
vec
1 0

=0

524

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

 (0) 


(0) 1
V
vec lim Vvec |0i
1 0
0

p
X

(k)

!
!
p0
X

+ 2 0
k b1/2
b1/2 |0; k i0 ,

=0

(3.11)

=0

Pp0
where |0; k i0 is defined as |0; k i0 = exp( =0 k x )|0i. Let us consider the physical
state conditions on these states. The relevant part of the energy-momentum tensor for this
analysis is Eq. (3.6) with p being replaced by p0 . This yields
#
"
p0
X
 (0) 


0

G k k + 1 V (0) ,
=
L0 V
vec

vec

,=0
0

p
X

 (0) 
G k (k)|0; k i0 .
= 2 0
L1 Vvec

(3.12)

,=0

From these relations we find that the physical state conditions, L0 = 1 and L1 = 0, require
that
0

m2vec

p
X

G k k = 0,

,=0

p
X

G k (k) = 0.

(3.13)

,=0

We will write the vector emission vertex operator Vvec ( 1 , ) in an exponential form
[15],


Vvec 1 ,
)
# 
" p 0 (r
Z

X


1
0

k + (k) D + D X (z, z) : 
,
= d : exp i

2
2
1
=0

z=z= ,=

(3.14)
by introducing a Grassmann parameter .
We will write the part of vertex operator (3.14) which depends on the string coordinates
in the x i -directions (i = p + 1, . . . , p0 ) as
# 
" p0 /2
0 /2
p

X
X
J

I
EI (, k, )Z (z, z) +
EJ (, k, )Z (z, z) : 
,
: exp

1
I =(p+2)/2

J =(p+2)/2

z=z= ,=

(3.15)
using the complex variables. Here EI (, k, ) and EI (, k, ) are differential operators on
the superspace defined as
r


1
0
I + ieI (k) D + D ,
EI (, k, ) = i
2
2
r


1
0
EI (, k, ) = i
I + ieI (k) D + D ,
(3.16)
2
2
with

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

1
I = (k2I 1 ik2I ),
2


1
eI (k) = 2I 1 (k) i2I (k) ,
2

1
I = (k2I 1 + ik2I ),
2


1
eI (k) = 2I 1 (k) + i2I (k) .
2

525

(3.17)

4. Scattering amplitudes
We would like to find out a proper description of the physical processes taking place on
the worldvolume of the Dp0 -brane with the Dp-brane inside (p < p0 ) in the case where
the B field is nonvanishing. In this section, we will consider multiparticle tree scattering
amplitudes consisting of the external states of N 2 vectors obtained from the mode of the
p0 p0 open string and two tachyons from the mode of the pp0 open string. The spacetime
picture of this string scattering process is depicted in Fig. 1. That this process is possible is
easy to see once we draw a spacetime diagram and map the end points of the open strings
onto a circle. See Fig. 2.
Open string tree amplitudes in general are obtained by placing vertex operators on the
boundary of the upper half plane, namely, the real axis, integrating over the positions of

Fig. 1. The spacetime picture of the process.

Fig. 2. N -point string diagram.

526

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

the vertex operators and dividing by the volume of the (super)conformal killing vectors. To
obtain the amplitudes of our concern, we first locate each of the two kinds of the tachyon

vertex operators V+
T (, ), VT (, ) discussed in the last section at = 1 and at = 2

respectively. A cut is generated on the interval between these two locations as V+

T and VT
contain the twist field and the anti-twist field respectively. The worldvolume of the Dp0 brane contains this interval on which we place the N 2 vector emission vertex operators
Vvec (, ) of the p0 p0 open string. In what follows we will obtain the integral (Koba
Nielsen) representation of the amplitudes. The explicit expressions for the N = 3, 4 cases
that we obtain will be exploited to determine the form of the low energy effective field
theory in the subsequent section.
In manifestly supersymmetric formulation on superspace, the N point tree amplitude in
question reads
Z Y
N

c
VSCKV

N
Y

da da h0| V+
T (1 , 1 ; k1 )VT (2 , 2 ; k2 )

a=1

Vvec (c , c ; kc , c )|0i,

(4.1)

c=3

where VSCKV denotes the volume of the isometry group generated by the superconformal
Killing vectors, namely, the graded extension of the SL(2, R) group. We have denoted by c
the overall constant which does not concern us in this paper. In Eq. (4.1), the domain of a
integrations is not restricted except that 3 , 4 , . . . , N are located on the cut created on the
interval between 1 and 2 . This domain falls into a sum of the (N 2)! regions. In each
region, an ordering among 3 , 4 , . . . , N is specified and integrals over each region give a
contribution corresponding to a respective open string (dual) diagram 6 . We will evaluate
the contribution from the region 2 < 3 < 4 < < N < 1 and this is denoted by AN .
In most cases below, we will not write the region of integrations explicitly.
Eq. (4.1) is invariant under the graded SL(2, R) transformations after the physical state
conditions are invoked at each vertex operator. For actual evaluation of the amplitude, we
first set 1 = 2 = 0 to fix the odd elements of the transformations. We then fix the even
elements by giving fixed values to three of the locations of the vertex operators. These
locations are chosen as 1 , 2 , and 3 . This amounts to factoring out the following volume
element from the integration,
d 3 F (1 , 2 , 3 ) d1 d2 (1 2 ),

(4.2)

where
d 3 F (1 , 2 , 3 )

d1 d2 d3
.
(1 2 )(2 3 )(3 1 )

(4.3)

6 Recall that we have (N 1)! open string (dual) diagrams in the case of the N point amplitude of a pp open
string.

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

527

Having done this, we obtain

QN
QN
Z
+(1)
(1)
a=1 da
c=3 dc
h0| VT
(1 ; k1)VT
(2 ; k2 )
AN = c
d 3 F (1 , 2 , 3 ) 1 2

N
Y

Vvec (c0 , c0 ; kc0 , c0 )|0i.

(4.4)

c0 =3

The component corresponding to the (1)-picture has been selected at each of the tachyon
vertex operators. Let us choose 1 = 0, 2 = and 3 = 1, so that the negative real
axis becomes the worldsheet boundary ending on the Dp0 -brane. Introducing positive real
variables xa a (= ea ) > 0 and adopting Eq. (3.14) for the vector emission vertex
operators, we find
QN
 Pp0 /2
I =(p+2)/2 I
0
0
dx
d
d
1
0
a
a
a
a=1
a =3
= c
d 3 F (x1 , x2 , x3 )
x1 x2
x2
" p (r
N
0
Y
X

kf X (xf , f )
h0|
: exp i
2
f =1
=0
)# 

1


:
|0i
+ f f X (xf , f )
 1 =2 =0
2
Z

AN

QN

h, s|

N
Y
c=3

1 =2 =0

"

0 /2
p
X

: exp

EI (c , kc , c )ZI (xc , c )

I =(p+2)/2

0 /2
p
X

#
J

EJ (c , kc , c )Z (xc , c ) : |, si,

(4.5)

J=(p+2)/2

where x1 = 0, x2 = and x3 = 1. Here the right-hand side consists of two of the

expectation values of the exponential operators: the one is obtained from the x -directions
( = 0, . . . , p) and the other is from the x i -directions (i = p + 1, . . . , p0 ), and we have
used
hI , sI | = lim x I h0| I sI (x).

(4.6)

Let us examine the contribution from the x i -directions (i = p + 1, . . . , p0 ). As is

explained in Section 2, the operators inside h, s| |, si in Eq. (4.5) are normal ordered
with respect to the SL(2, R) invariant vacuum and are not with respect to the oscillator
vacuum. Applying the reordering formula (2.63), we obtain
X

N
X
Y
J
I
: exp
EaI Z (xa , a ) +
EaJ Z (xa , a ) : |, si
h, s|
a=3

N
Y
a=3

X

exp
EaI EaJ G sub I J ((xa , a ))
I,J

528

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

h, s|

N
Y

N
Y

X

X
J
exp
EcI ZI (xc , c ) +
EcJ Z (xc , c ) |, si,

c=3

0P

GI J

xa

I,J

aI aJ

a=3

IJ

h, s|

N
Y

X

X
J

EcI ZI (xc , c ) +
EcJ Z (xc , c ) |, si,
exp

c=3

where
Ca (I ) =



X

I J a
0
exp Ca (I ) + 2 a
eaI aJ G
xa

X
I,J

(4.7)





1
2aI aJ G
+ (I ) + (1 I ) ,
2
IJ

(4.8)

and EcI and EcJ stand for EI (c , kc , c ) and EJ (c , kc , c ), respectively. Note that the part
in Eq. (4.7) that corresponds to self-contractions has been given by the subtracted Green
function at the coincident point:
X
EaI EaJ G sub I J ((xa , a ))
IJ

lim

c a
xc xa

X
IJ

= Ca (I ) +

EcI EaJ G sub I J (xc , c | xa , a )

X

2 aI aJ G

IJ

I,J

I J a
0
ln xa + 2 a eaI aJ G
.
xa

(4.9)

The last factor h, s| |, si in Eq. (4.7) is now calculated using the two-point function
G I J (xc , c | xc0 , c0 ):
X

N
X
Y
J

EcI ZI (xc , c ) +
EcJ Z (xc , c ) |, si
exp
h, s|
c=3

Y
36c<c0 6N

X

exp
EcI Ec0 J G I J (xc , c | xc0 , c0 )
I,J

+ EcJ E G (x , | xc , c )




X
Y
xc
IJ
0
exp
G
2 cI c0 J H I ;
=
xc 0 c c 0
36c<c0 6N
I,J


xc 0
2 0 cJ c0 I H I ;
xc + c c 0

 I


(xc /xc0 )I c0 (xc /xc0 )I 1 c

c 0 c
xc 0
0
0
+ ecJ c I
+ c 2 ecI c0 J
xc xc 0
xc
xc xc 0

 I


1
I

c 0 c
(xc0 /xc )
c0 (xc0 /xc )I c
xc
0
+ cJ ec0 I
+ c0 2 cI ec0 J
xc 0
xc xc 0
xc xc 0
c0 I

IJ

c0

c0

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

529




(xc /xc0 )I
(1 I ) + I xc0 /xc
+ c c0 ecI ec0 J
1 c c 0
xc xc 0
xc xc 0


(xc0 /xc )I
(1 I ) + I xc /xc0
1 c c 0
+ ecJ ec0 I
xc xc 0
xc xc 0

I
J
X


2 bI 0
cI c0J cJ c0 I .
(xc xc0 )

(1 + bI2 )

(4.10)

I,J

The factor coming from the x -directions ( = 0, . . . , p) is handled by the two-point

0
function G :
" p
#
N
Y
X

: exp
E (f , kf , f )X (xf , f ) : |0i
h0|
f =1

=0

"

p
X

16f <f 0 6N

,0 =0

= exp

"
h0| : exp

E (f , kf , f )E0 (f 0 , kf 0 , f 0 )

#


(zf , zf |zf 0 , zf 0 ) 
 zf =zf =xf ,
f = f

p
N X
X

E (c , kc , c )X (xc , c ) : |0i
c=1 =0
( p 
"
X
X
0 G kf kf 0 ln(xf xf 0 + f f 0 )2
= exp
16f <f 0 6N ,=0

f f 0
2 0 f G f kf 0 + f 0 G kf f 0
xf xf 0

1
+ f f 0 G f f 0
(xf xf 0 + f f 0 )
)#
p
X i
ij
kf i kf 0 j (xf xf 0 )
+
2

i,j =1

(2)p+1

p
Y

(k1 + + kN ),

(4.11)

=0

where E (f , kf , f ) is a differential operator defined as

r


1
0
kf + i f f (k) Df + D f ,
E (f , kf , f ) = i
2
2

E (f , kf , f )X (xf , f ) E (f , kf , f )X (zf , zf ) zf =zf =xf .

(4.12)

f = f

We have now evaluated the two of the expectation values in Eq. (4.5) and are ready to
present the integral representation of AN . Let us first introduce several shorthand notations.

530

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

the inner product of two (p + 1)-dimensional vectors Ai and Bj lying

We denote by (p)
on the Dp-brane worldvolume with respect to the open string metric. Namely,
p
X

B=
A (p)

G A B .

(4.13)

,=0

Similarly, we denote by (p0 ) the inner product of two (p0 + 1)-dimensional vectors Ai and

Bj lying on the Dp0 -brane worldvolume with respect to the open string metric. We will
0 ) B to denote the inner product of the last (p0 p) components of the two
also write A (p,p
vectors. For example, we have
0 /2
p
X

0) =
k (p,p

0 /2
p
X

G I e J + G J eI =
IJ

JI

I,J =(p+2)/2

I,J =(p+2)/2

p 0 /2

0) k =
k (p,p

GI J I e J + J e I ,

0) =
(p,p

IJ

2G I J ,

I,J =(p+2)/2

0 /2
p
X

2GI J eI eJ .

(4.14)

I,J =(p+2)/2

We will use the notations  0 and 0 which denote

(p,p )

k 0
(p,p )

I

(p,p )

GI J I e J + J e I ,

k 0
(p,p )

I

X 2 I J (I e eI )
J
J
J

(1 + bI2 )

(4.15)

etc. From these defining relations one can find that

X
X




0 ) ,
0 ) J ,
k  0 I = k (p,p
k 0 I = ik (p,p
I

(p,p )

(p,p )

where J is a (p0 + 1) (p0 + 1) antisymmetric matrix defined as

0
0
.
..
..

0 1
p+1

.
J = J

p+2

1 0
.

..
.

0 1 p0 1
1 0

(4.16)

(4.17)

p0

We will group the terms in the exponent by the number of a s and by the number of a s,
using the notation [0, 2], [2, 0], [1, 1], [2, 2]. The first number in the bracket indicates the
number of a s and the second number the number of a s.

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

531

Having prepared these, we write Eq. (4.5) as

!Z N
p
N
N
Y
X
Y
Y
p+1

ke
dxa
da 0 da 0 exp Ca 0 (I )
AN = c(2)
=0

x2

I I

a=4

(x2 x3 )(x3 x1 )

e=1

16c<c0 6N

(xc0 xc )

2 0 kc0

(p)

N
Y
c00 =3
kc

a 0 =3
0 kc00

xc00
Y

36c<c0 6N

(p,p 0 )

kc00


X
exp 2 0
GI J






xc
xc 0
cI c0 J H I ;
+ cJ c0 I H I ;
xc 0
xc


exp(NC) exp [0, 2] + [2, 0] + [1, 1] + [2, 2] .

I,J

Here,
X

c c 0
xc xc 0
36c<c0 6N


 I  I 


1X
xc 0
xc
kc 0 +
kc (p)
+
kc  0 kc 0 I
(p,p )
2
xc 0
xc
I
 I  I 


xc 0
xc

,
+ kc 0 kc 0 I
(p,p )
xc 0
xc

N
X

1
1
1
0
0 ) (1 + iJ )kc
c
c c c (p,p
+ k1 (p)
[1, 1] = 2
2
xc
x1 xc
c=3

1
c
+ k2 (p)
x2 xc

X

1
0
kc0 + c0 kc (p)
c0 )(c c0 )
(c c (p)
2
xc xc 0
0
36c<c 6N

 I 1  I 
X


xc 0
xc
1

0
kc I
+
c c
c
+
(p,p 0 )
2
xc 0
xc
I
 I  I  
xc
xc 0

+
c 0
xc 0
xc
 I  I 


xc 0
xc

c 0 I
+
c
+ c0 kc
(p,p 0 )
xc 0
xc
 I  I 1  
xc
xc 0

+
c 0
xc 0
xc
 I 1  I 


xc 0
xc

kc 0 I

c
+ c c
(p,p 0 )
xc 0
xc
 I  I  
xc
xc 0

c 0
0
xc
xc
[0, 2] = 2 0

(4.18)

532

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

+ c0 kc 0 c0


(p,p )



xc
xc 0

I

xc 0
xc

I 
c


 
xc0 I 1

c 0
,
xc


 I  I 


1X
xc 0
c c0
xc

c 0 +
0
+
c (p)
c
[2, 0] =
0) c I
0
0
(p,p
xc xc
2
xc
xc
I
36c<c0 6N
 I  I 


xc 0
xc

,
+ c 0 c 0 I
(p,p )
xc 0
xc
X
c c c0 c0
[2, 2] =
(xc xc0 )2
36c<c0 6N

 I  I 




1X
xc
xc 0

0
0

c
c (p) c +
c I (1 I )
+
(p,p 0 )
2
xc 0
xc
I
 I 1  I 1 
xc 0
xc
+
+ I
xc 0
xc
 I  I 



xc
xc 0

+ c 0 c0 I (1 I )
(p,p )
xc 0
xc
 I 1  I 1 
xc 0
xc

,
(4.19)
+ I
xc 0
xc
xc

xc 0
X

I

and (NC) denotes the terms containing the sign function:

X

(NC) =

16a<a 0 6N

p
X
i
(xa xa 0 )
ij kai ka 0 j
2
i,j =1

(xc xc0 )

36c<c0 6N

16a<a 0 6N

I,J

2 I J bI
(1 + bI2 )

(cI c0 J cJ c0 I )

p
X
i
(xa xa 0 )
ka ka 0 ,
2

(4.20)

,=0

with k1j = k2j = 0 for (j = p + 1, . . . , p0 ). Here we have written the noncommutativity

term in terms of the real variables and generalized the notation to include the time
components 0i = 0. We also remind the readers that H(I ; xc /xc0 ) is given in Eq. (2.46).
So far, we have not exploited that x1 = 0, x2 = and x3 = 1 except that the oscillator
vacuum |, si, h, s| is obtained from the tachyon vertex operators. Firstly, by sending
x2 = , all factors in Eq. (4.18) containing x2 are removed. In fact, this is ensured by an
equality
1

0 /2
p
X

I =(p+2)/2

I + 2 0

X
c6=2

kc = 0,
k2 (p)

(4.21)

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

533

which is obtained from the momentum conservation


p X
N
Y

kc
=0

c=1

and the on-shell condition (Eq. (3.8)) for the tachyon. Secondly, setting x1 = 0, we find
Y

(xc0 xc )

2 0 kc0

(p)

16c<c0 6N
c,c0 6=2

N
Y

kc

N
Y
c00 =3

0 sc + 0 m2T

xc

0 kc00

xc00

(xc xc0 )

(p,p 0 )

2 0 kc0

kc00

(p)

kc

(4.22)

36c<c0 6N

c=3

(kc + k1 ) and we have used the on-shell condition for the tachyon
Here sc (kc + k1 ) (p)
(Eq. (3.8)) and that for the vector (Eq. (3.13)). Finally, we would like to convert the
integrations at Eq. (4.18) into those over a set of N 3 SL(2, R) invariant cross ratios.
We choose these cross ratios as
(x1 xa+3)(x2 x3 ) xa+3
=
, a = 1, . . . , N 3.
(4.23)
x (a+3)
(x1 x3 )(x2 xa+3 )
x3
We can, therefore, accomplish this conversion by rescaling xa+3 by x3 and setting x3 = 1
in Eq. (4.18) without changing the form of the integrand.
Putting all these considerations together, we obtain from Eq. (4.18)
AN = c(2)

p+1

Z Y
p X
N
N
N
Y
Y

ke
dxa
da 0 da 0 exp Ca 0 (I )
=0

N h
Y

e=1

0 sc + 0 m2T

xc

c=4

(1 xc )

a 0 =3

a=4
2 0 k3

(p)

kc i

(xc xc0 )

2 0 kc0

kc

46c<c0 6N





X
xc
0
IJ

exp 2
G
cI c0 J H I ;
xc 0
36c<c0 6N
I,J


xc 0
+ cJ c0 I H I ;
xc


exp(NC) exp [0, 2] + [2, 0] + [1, 1] + [2, 2] 
Y

(p)

x1 =0,x2 =,x3 =1

(4.24)

This expression is regarded as an SL(2, R) invariant integral (KobaNielsen) representation

for the amplitude of our concern. Let us list several features which are distinct from the
corresponding formula in the case of a pp open string (see [15]).
1. The term denoted by exp(NC) which originated from the noncommutativity of
the worldvolume extends into both the x 1 , . . . , x p directions and the remaining
0
x p+1 , . . . , x p directions.
2. To each external vector leg, we have a momentum dependent multiplicative factor
exp C(I ).

534

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

3. A new tensor J has appeared.

4. There are parts in the expression which are expressible in terms of the momenta of the
tachyons, the momenta and the polarization tensors of the vectors and J alone, using
the inner product with respect to the open string metric. These parts come, however,
with a host of other parts which do not permit such generic description in terms of
the inner product.
Let us, finally, compute N = 3, 4 cases explicitly. For N = 3 case, we need to pick up
3 and 3 from [1, 1]. Using the transversality of the polarization vector (Eq. (3.13)) and
the (p + 1)-dimensional momentum conservation, we find that
!
p
3
Y
X
p+1

ka
A3 = c(2)
r

=0

0 
2

a=1

ij
0 ) J 3 eC3 (I ) e(i/2) k1i k2j .
3 ik3 (p,p
(k2 k1 ) (p)

(4.25)

For N = 4 case, we have





1 X
3 4
k4 +
k3 (p)
k3  0 k4 I x I + x I
[0, 2] = 2 0
(p,p )
1x
2
I




+ k3 0 k4 I x I x I
,
(p,p )


3 3 
3 ik3 (p,p
0 ) J 3 + k4 (p)
3
(k2 k1 ) (p)
[1, 1] = 2 0
2



4 4 
4 ik4 (p,p
0 ) J 4 + k3 (p)
4
(k2 k1 ) (p)
+
2x


3 4
3 + 4 k3 (p)
4
3 k4 (p)
+
1x






1 1 Xn
3 3  0 k4 I x I +1 + x I 3 x I + x I 4

(p,p )
21x
I






+ 4 k3  0 4 I x I + x I 3 x I + x I 1 4
(p,p )






+ 3 3 0 k4 I x I +1 x I 3 x I x I 4
(p,p )






o
,
+ 4 k3 0 4 I x I x I 3 x I x I 1 4
(p,p )





1 X
3 4
4 +
3 (p)
3  0 4 I x I + x I
[2, 0] =
(p,p )
1x
2
I





,
+ 3 0 4 I x I x I
(p,p )





1 Xn
3 3 4 4

+
3  0 4 I (1 I ) x I + x I

[2, 2] =
3 (p) 4
2
(p,p )
(1 x)
2
I


+ I x I +1 + x I 1

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

+ 3 0 4
(p,p )


I

535





o
(1 I ) x I x I + I x I +1 x I 1
, (4.26)

where we have set x3 = 1 and written x4 = x. By picking up terms from [2, 2] +

[0, 2][2, 0] + 12 [1, 1]2 , we obtain
A4 = c(2)

p+1

p
Y
=0

4
X
a=1

! Z1
ka

dx x t + mT (1 x)
2

2 0 k3

(p)

k4

exp C3 (I ) + C4 (I ) + (NC)



X


1
0
k3  0 k4 + k3 0 k4 I H I ;
exp
(p,p )
(p,p )
x
I



+ k3  0 k4 k3 0 k4 I H(I ; x)
(p,p )
(p,p )



1
4 1 2 0 k3 (p)
k4
3 (p)

2
(1 x)


0 1 
3 ik3 (p,p
0 ) J 3 k4 (p)
3
(k2 k1 ) (p)
+
2 x



4 ik4 (p,p
0 ) J 4 + k3 (p)
4
(k2 k1 ) (p)

1 
3 ik3 (p,p
0 ) J 3 k3 (p)
4
(k2 k1 ) (p)
1x


4 ik4 (p,p
3 (k2 k1 ) (p)
0 ) J 4
k4 (p)

X x I


4
0 k3  0 k4 + k3 0 k4 I 3 (p)
+
(p,p )
(p,p )
(1 x)2
I





1 I
0


k3 (p) k4 3
+
4 + 3
4 I
(p,p 0 )
(p,p 0 )
2

X x I


4
0 k3  0 k4 k3 0 k4 I 3 (p)
+
(p,p )
(p,p )
(1 x)2
I





1 I
k 4 3  0 4 3 0 4 I
0 k3 (p)
+
(p,p )
(p,p )
2
X I x I +1


3  0 4 + 3 0 4 I
+
2
(p,p )
(p,p )
2 (1 x)
+ 0

X I x I 1


+
3  0 4 3 0 4 I
2
(p,p )
(p,p )
2 (1 x)
I

0 X x I L

k 3  0 k 4 + k 3 0 k 4 I 3  0 4 + 3 0 4 L
2
(p,p )
(p,p )
(p,p )
(p,p )
2
(1 x)
I,L

X x I +L

k 3  0 k 4 k 3 0 k 4 I 3  0 4 3 0 4 L
(p,p )
(p,p )
(p,p )
(p,p )
2
(1 x)2
0

I,L

536

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

0 X x I +L
k 3  0 k 4 + k 3 0 k 4 I 3  0 4 3 0 4 L
2
(p,p )
(p,p )
(p,p )
(p,p )
2
(1 x)
I,L

X x I L

k 3  0 k 4 k 3 0 k 4 I 3  0 4 + 3 0 4 L
(p,p )
(p,p )
(p,p )
(p,p )
2
(1 x)2
0

I,L

0 X x I L
k 4  0 3 k 4 0 3 I k 3  0 4 + k 3 0 4 L
+
(p,p )
(p,p )
(p,p )
(p,p )
2
1x
I,L

0
2

X x I +L 1
I,L

1x

k 4  0 3 + k 4 0 3
(p,p )

(p,p )

I

k 3  0 4 k 3 0 4
(p,p )

(p,p )

L




0 X x I 1 n 
3 ik3 (p,p
0 ) J 3 k4 (p)
3
(k2 k1 ) (p)
2
1x
I


k 3  0 4 k 3 0 4 I
+

(p,p )

(p,p )





o
0 ) J 4 + k3 (p)
4 k 4  0 3 + k 4 0 3 I
4 ik4 (p,p
(k2 k1 ) (p)


(p,p )

X x I n 



3 ik3 (p,p
0 ) J 3 + k4 (p)
3
(k2 k1 ) (p)
+
2
1x
I


k 3  0 4 + k 3 0 4 I

(p,p )

(p,p )

(p,p )






o


(k2 k1 ) (p) 4 ik4 (p,p0 ) J 4 k3 (p) 4 k4

k4

,
0 3
0 3 I


(p,p )

(p,p )

(4.27)
(k4 + k1 ).
where t is defined as t s4 (k4 + k1 ) (p)

5. The zero slope limit and the low energy effective action
In the last section, we have evaluated the three and four point amplitudes with the initial
and final tachyons and N 2 vectors (N = 3, 4) present. Let us try to extract physical
significance from these.
ij
The three point amplitude Eq. (4.25) contains the two multiplicative factors e(i/2) k1i k2j
and eC (I ) both of which are listed in the last section as prominent features. The first factor
represents the noncommutativity of the Dp-brane worldvolume. The second factor will
be discussed shortly. Aside from these factors and (p + 1)-dimensional delta functions
representing momentum conservation, Eq. (4.25) is interpreted as coming from field theory
vertex of

1
A + 88 J MN FMN ,
(5.1)
8i 8 (p)
2
where 8 is a complex tachyon field and AM and FMN are the gauge field and its field
strength, respectively. The first term is the gaugescalar derivative interaction while the
second factor is a new interaction coming from our pp0 open string system.

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

537

Fig. 3. The string diagram corresponding to the s-channel and the t -channel.

The four point amplitude Eq. (4.27) is quite complex but one can still systematically
investigate the singular behavior of the integrand around its end points x = 0, 1. This
behavior is sufficient to tell us the zero slope limit of the amplitude and the content of
the low energy field theory. We will focus upon this in the remainder of this section. To
be more accurate we consider the sum of Eq. (4.27) and the one obtained from this by
k3 k4 , 3 4 in accordance with the two open string (dual) diagrams.
The nontrivial zero slope limit is given by sending 0 0 while keeping the parameter
ij of noncommutativity and the open string metric fixed. From Eqs. (2.6) and (2.7) this
means that [6]
0 1/2 0,
g 0,
|bI |

1/2

(5.2)

In this limiting procedure, 0 bI becomes finite: 0 bI I . Let us first look at the

multiplicative factor C(I ) defined in Eq. (4.8). Using (1) = and Eq. (A.5), we obtain
X


X
|I |I J GI J =
|I | k  0 k I .
(5.3)
C(I )
(p,p )
2
I,J

So this exponential multiplicative factor acts as a Gaussian damping factor when vectors
0
propagate into the x p+1 , . . . , x p directions.
Let us come back to the four point amplitude (4.27). If the integrand is regular, one could
take the 0 0 limit inside the integral and this will not give us any nontrivial contribution.
If the integrand is singular at some point, it will still not give us much as long as one can
avoid such singularity by a contour deformation. The nontrivial contribution in the 0 0
limit, therefore, is obtained only when we have end point singularities.
Let us focus on the behavior of the integrand near x = 0 from which we can read off
the mass of particles exchanged in the t-channel. Fig. 3 indicates that the t-channel poles
originate in the propagation of the pp0 open string. Thus the complicated behavior of
the integrand near x = 0 should reflect the spectrum of the pp0 open string [7]. In order
to identify the t-channel poles, we expand the integrand of the amplitude (4.27) around
x = 0:

538

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

Z1
A4 =

dx

fA x t +KA ,

(5.4)

where the coefficients fA are functions of momenta and polarization tensors. The term
0
fA x t +KA in the integrand of the above equation yields the t-channel pole at 0 t =
R
KA + 1, when it is integrated near x = 0: 0 dx . . . . From explicit computation we find
that the t-channel poles exist at
X
X
MI 0 (n + 1 I 0 ) +
ML0 0 (n0 + L0 ) + N,
(5.5)
0 t = 0 m2T + W +
I0

L0

with
W = 0, 1, 1 I , I , 1 + I , 2 I , 1 I L , I + L , 1 + I + L ,
or 1 I + L ,
where n, n0 , MI 0 , ML0 0 and N are non-negative integers. The terms proportional to MI 0 and
ML0 0 come from the exponential of the hypergeometric function H,



X


1
k3  0 k4 + k3 0 k4 I H I :
exp 0
(p,p )
(p,p )
x
I



+ k3  0 k4 k3 0 k4 I H(I : x)
(p,p )
(p,p )

 X


bI k3 0 k4 I
= exp 0
(p,p )

X
X

X
1
x n+1I 0 M
k3  0 k4 + k3 0 k4 I 0
0
(p,p )
(p,p )
M!
n + 1 I 0
M=0
n=0
I0
!M 0

n0 +L0
X
X
X


1
x

.
k3  0 k4 k3 0 k4 L0
0
(p,p )
(p,p )
M 0!
n0 + L0
0
0
0

(5.6)

n =0

The t-channel poles Eq. (5.5) correspond to the spectrum of the pp0 open string [7].
In view of the analysis in [7], we expect that a large number of light states should
be exchanged in the zero slope limit in which one of I goes to unity and the others
approach zero. In order to specify the situation, we assume, without loss of generality,
e
that (p+2)/2 goes to 1 and e
I (I 6= (p + 2)/2) go to 0 in the zero slope limit.
In this zero slope limit many light states are realized by the poles in Eq. (5.5) with
0
(n, n0 , Me
I 0 , M(p+2)/2 , N) = 0,
W = 0, 1 , e
I , 1 e
I , 1 + e
I,

(5.7)

0 being arbitrary non-negative integers. Aside from the multiplicative

and M(p+2)/2 and Me
I
factors and the momentum conserving delta functions, the massless pole obtained in the
zero slope limit turns out to be

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

1
0 ) J 3
3 ik3 (p,p
k2 (k1 + k4 ) (p)
2
t





0 ) J 4
4 ik4 (p,p
(k2 + k3 ) k1 (p)
 

1
 k4 + k3 k4 p+2 3 4

k
+
3
0
0
1
(p)
(p,p )
(p,p )
t (m2T + || )
2



1
k4 3  0 4 + 3 0 4 p+2
k3 (p)
+
(p,p )
(p,p )
2||
2
o

1n
3 ik3 (p,p
0 ) J 3 k3  0 4 + k3 0 4 p+2
+ (k2 k1 + k4 ) (p)
(p,p )
(p,p )
2
2
o
 
1n
0 ) J 4 k4  0 3 k4 0 3 p+2
4 ik4 (p,p
(k2 k1 k3 ) (p)
(p,p )
(p,p )
2
2


X
1 
1
3  0 4 3 0 4
+
e
I
(p,p )
(p,p )
t (m2T + |1 e| ) 2|e
I|
e
I
I
o



1n
3 ik3 (p,p
0 ) J 3 k 3  0 4 k 3 0 4
k2 (k1 + k4 ) (p)
+
e
I
(p,p )
(p,p )
2
o
 


1n
0 ) J 4 k 4  0 3 + k 4 0 3
4 ik4 (p,p

(k2 + k3 ) k1 (p)
e
I
(p,p )
(p,p )
2
X
1

t (m2T + |1 e| + |1 e| )
e
Ie,L
I
L

 

1
k 3  0 4 k 3 0 4
k 4  0 3 + k 4 0 3
e
e
I
L
(p,p )
(p,p )
(p,p )
(p,p )
2
X
1
+
1
t (m2T + ||
|1 e| )
Ie
I

 

1
k3  0 k4 + k3 0 k4 p+2 3  0 4 + 3 0 4
e
I
(p,p )
(p,p )
(p,p )
(p,p )
2
2




1
k3  0 k4 + k3 0 k4
3  0 4 + 3 0 4 p+2
Ie
(p,p )
(p,p )
(p,p )
(p,p )
2
2
 

1
+ k4  0 3 k4 0 3 p+2 k3  0 4 + k3 0 4
Ie
(p,p )
(p,p )
(p,p )
(p,p )
2
2





1


+ k4
3 k 4
3
4 + k 3
4 p+2
k3
e
I
(p,p 0 )
(p,p 0 )
(p,p 0 )
(p,p 0 )
2
2
X
1

1
t (m2T + ||
+ |1 e| )
Ie
I
 
 

1
k3  0 k4 + k3 0 k4 p+2 3  0 4 3 0 4

e
I
(p,p )
(p,p )
(p,p )
(p,p )
2
2





1
+ k3  0 k4 k3 0 k4
3  0 4 + 3 0 4 p+2
Ie
(p,p )
(p,p )
(p,p )
(p,p )
2
2
1

m2T

539

540

B. Chen et al. / Nuclear Physics B 593 (2001) 505544



 
X
exp
|I | k3  0 k4
,
(p,p )

where we have used

' 1 b
(p+2)/2

e
I '
be
I

H
H

(5.8)

0
(p+2)/2
1
0
e
I
e
I

(> 0),
(5.9)
(> 0)

in the zero slope limit. The first term in Eq. (5.8) comes from the tachyon state exchange.
Here the vertices derived from three point amplitude emerge. From the t-channel diagram
in Fig. 4, we find that these vertices depend on momenta in a proper way. It is worth
noting that combining the exponential factor in Eq. (5.8) with the multiplicative factor
0
exp(C3 (I ) + C4(I )), we obtain a Gaussian damping factor in the x p+1 , . . . , x p directions
in the zero slope limit,




X
|I | (k3 + k4 )  0 (k3 + k4 ) I .
(5.10)
exp
(p,p )
2
I

Next we focus on the behavior of the integrand near x = 1 from which we can read off
the s-channel poles. From Fig. 3, the s-channel poles come from the propagation of the
p0 p0 open string. In a similar way to
R 1 the t-channel, by expanding the integrand around
x = 1 and integrating it near x = 1: 1 dx . . . , we find that s-channel poles correspond
to the p0 p0 open string spectrum. In particular, aside from the multiplicative factor and
momentum conserving delta functions, the massless pole turns out to be

o
1 n
3 i(k3 + k4 ) (p,p
0 ) J 3 k3 (p0 ) 4
(k2 k1 ) (p)

s
n
o
4 i(k3 + k4 ) (p,p
0 ) J 4
k4 (p0 ) 3 (k2 k1 ) (p)



1
0 ) J k4 3 (p0 ) 4
(k1 k2 ) ik3 (p,p
(k3 k4 ) (p)
+
2
"
#
 
X


1
0
k3  0 k4
+ (I ) + (1 I )
,
(5.11)
exp 2
(p,p )
I
2
I

and terms with (k3 k4 ; 3 4 ), where

s (k3 + k4 ) (p0 ) (k3 + k4 ) = 2k3 (p0 ) k4

(5.12)

(k1 k2 ) s.
0 ) k4 ) = (k3 k4 ) (p)
and we have used 2(t mT + k3 (p,p
Here by using Eqs. (2.41), (A.1) and (A.5) we have expanded the hypergeometric
function H around x = 1 as



(1 + )m
1
= b ln(1 x) +
(1 x)m
H ;
x
2
m!
m=0

X

(1 )n 

(n + 1) (n + 1 ) (1 x)n ,
n!
n=0

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

541

Fig. 4. Field theory diagrams.

H(; x) =

X ()m

b ln(1 x) +
(1 x)m
2
m!
m=0

X
n=0

()n 
(n + 1) (n + ) (1 x)n .
n!

(5.13)

We find that in our result (5.11) the first two terms are in accordance with the vertex seen
at the three point amplitude (4.25). This vertex shows up again with proper momentum
dependence (see Fig. 4) as well as in the t-channel pole corresponding to the tachyon
exchange. Combined with the multiplicative factor exp(C3 (I ) + C4 ()), the exponential
factor in Eq. (5.11) gives us a Gaussian damping factor,


 X




1
+ (I ) + (1 I ) (k3 + k4 )  0 (k3 + k4 ) I .
(5.14)
exp 0
(p,p )
2
I

It is noteworthy that in the zero slope limit this Gaussian damping factor turns out to be the
same as that of t-channel (Eq. (5.10)).
While we do not try to derive here the complete action of the low energy noncommutative field theory in p0 + 1 dimensions, it is still possible to exhibit the interactions which
reproduce the parts of the amplitudes in the zero slope limit which are expressible in terms
of the inner product with respect to the open string metric. We find that this part of the
action is
S = S0 + S1 , with


Z

1
1
0
S0 = 2
d p +1 x G (D 8) D 8 m2 8 8 FMN F MN ,
4
gYM
Z

1
0
(5.15)
d p +1 x G8 FMN J MN 8,
S1 =
2
2gYM
where
D 8 = 8 iA 8,

(D 8) = 8 + i8 A ,

FMN = M AN N AM i[AM , AN ] ,

[AM , AN ] = AM AN AN AM ,
(5.16)

542

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

the product of two functions f and g is given by





i
f (x) g(x) = exp f (y)g(z)y,zx ,
2
y z

(5.17)

and gYM is the effective YangMills coupling defined by using the open sting coupling Gs
and that of the closed string gs as [6]


0
0
( 0 )(3p )/2
( 0 )(3p )/2 det(g + 2 0 B) 1/2
1
=
=
.
(5.18)
0
0
2
det G
(2)p 2 Gs
(2)p 2 gs
gYM
In Eq. (5.15) we have determined the seagull interaction corresponding to the non-pole
term by invoking the noncommutative U (1) invariance.
Let us, finally, discuss the Gaussian damping factor Eq. (5.10) which have originated
from the exponential multiplicative factor Eq. (4.8) and the lowest modes in the
hypergeometric function H. Recall that there is no momentum conservation for the
0
x p+1 , . . . , x p -directions and that the tachyon momenta k1 and k2 are constrained to lie
on the x 0 , . . . , x p -directions. Without the Gaussian damping factor, our picture would be
0
that (N 2) incident noncommutative U (1) photons travel freely in the x p+1 , . . . , x p directions until they get stopped by the Dp-brane. The actual spacetime picture which we
have exhibited here is that the lowest mode of the pp0 open string develops a physical

scale |I | and that this mode creates a cloud around the Dp-brane in the zero slope limit.
The noncommutative U (1) photons get decelerated by the presence of this cloud, which is

reflected in our damping factor (5.10). The mean free paths will be measured by |I |.
In this situation, the tachyon field in these directions should be expanded by the coherent
0
I
(or II ).
states {hx p+1 , . . . , x p |I i} associated with the would-be zero modes 1
I
On this basis, the complete analysis of low-lying states and I dependent interactions
obtained from the residual parts of the amplitudes will lead to the full-fledged form of the
tachyon-vector interactions in these directions. The appearance of the coherent states here
suggests that the fields which have originated from the pp0 open string should support
noncommutative solitons on the Dp0 -brane worldvolume which has recently been found
in [19].

Acknowledgements
We are grateful to Professor E. Date for helpful discussions on this subject.

Appendix A. More on the two-point function at the worldsheet boundary

The radius of convergence of the hypergeometric series F (a, b; c; z) is unity. Thus we
have evaluated the hypergeometric series on its convergent circle in Eq. (2.43) in deriving
the noncommutativity term (2.44). In this appendix we will give another derivation of
Eq. (2.43) to verify the noncommutativity term (2.44) and the two-point function (2.45).

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

543

Let us focus on the relation

F (; z) = ln(1 z) + z

X
()n 
n=0

n!

(n + 1) (n + ) (1 z)n .

(A.1)

One can obtain this relation by using Eq. (2.41) and a formula for the hypergeometric
function [20],

F (a, b; a + b; z) =

(a + b) X (a)n (b)n

2
(a) (b)
n=0 (n!


2(n + 1) (a + n) (b + n) ln(1 z) (1 z)n ,
(A.2)

which is derived from the following relation by putting c = a + b + and by taking the
limit of 0:
(c) (c a b)
F (a, b; a + b c + 1; 1 z)
F (a, b; c; z) =
(c a) (c b)
(c) (a + b c)
+
(a) (b)
(A.3)
(1 z)cab F (c a, c b; c a b + 1; 1 z).
From Eq. (A.1), one can find that


 


1
F (I ; z
lim F 1 I ;
z1
z
is sensitive to the way of taking the limit. In the original expression (2.39) of the two
point function, however, the way of sending z 1 is fixed in a definite way on the real
axis because of the step function in front of each hypergeometric function. The constant
noncommutativity term (2.44) should be more precisely described as





e2
e1
4 I J
lim F 1 I ; lim F I ;
1 2 0
1 + bI2 1 2 +0
e1
e2









e2 1I
e2 
4 I J

lim
(1) (1 I )
ln1  +
=
1 + bI2 1 2
e1
e1
  I






e1
e1 
(1) (I )
ln1  +
e2
e2
=

4 I J
bI ,
1 + bI2

(A.4)

where we have used a relation for the digamma function,

bI = cot(I ) = (I ) (1 I ).

(A.5)

Thus we obtained the same noncommutativity term as Eq. (2.44) through more careful
treatment of the hypergeometric functions and this provides another verification to the
two-point function (2.45).

544

B. Chen et al. / Nuclear Physics B 593 (2001) 505544

References
[1] A. Connes, M.R. Douglas, A. Schwarz, Noncommutative geometry and matrix theory: compactification on tori, JHEP 9802 (1998) 003, hep-th/9711162.
[2] M.R. Douglas, C. Hull, D-branes and the noncommutative torus, JHEP 9802 (1998) 008, hepth/9711165.
[3] C.S. Chu, P.M. Ho, Noncommutative open string and D-brane, Nucl. Phys. B 550 (1999) 151,
hep-th/9812219;
C.S. Chu, P.M. Ho, Constrained quantization of open string in background B field and noncommutative D-brane, Nucl. Phys. 568 (2000) 447, hep-th/9906192.
[4] F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, Noncommutative geometry from strings and
branes, JHEP 9902 (1999) 016, hep-th/9810072;
F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, Dirac quantization of open strings and noncommutativity in branes, hep-th/9906161.
[5] M.M. Sheikh-Jabbari, Super YangMills theory on noncommutative torus from open strings
interactions, Phys. Lett. B 450 (1999) 119, hep-th/9810179.
[6] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032,
hep-th/9908142.
[7] B. Chen, H. Itoyama, T. Matsuo, K. Murakami, pp0 system with B field, branes at angles and
noncommutative geometry, Nucl. Phys. B 576 (2000) 177, hep-th/9910263.
[8] A. Hashimoto, Dynamics of DirichletNeumann open strings on D-branes, Nucl. Phys. B 496
(1997) 243, hep-th/9608127.
[9] J. Frhlich, O. Grandjean, A. Recknagel, V. Schmerus, Fundamental strings in DpDq brane
systems, hep-th/9912079.
[10] E. Gava, K.S. Narain, M.H. Sarmadi, On the bound states of p- and (p + 2)-branes, Nucl. Phys.
B 504 (1997) 214, hep-th/9704006.
[11] N. Nekrasov, A. Schwarz, Instantons on noncommutative R 4 and (2,0) superconformal field
theory, Commun. Math. Phys. 198 (1998) 689, hep-th/9802068.
[12] M.R. Garousi, Superstring scattering from D-branes bound states, JHEP 9812 (1998) 008, hepth/9805078;
M.R. Garousi, R.C. Myers, World-volume interactions on D-branes, Nucl. Phys. B 542 (1999)
73, hep-th/9809100.
[13] A. Hashimoto, I.R. Klebanov, Scattering of strings from D-branes, Nucl. Phys. Proc. Suppl.
B 55 (1997) 118, hep-th/9611214.
[14] Y. Arakane, H. Itoyama, H. Kunitomo, A. Tokura, Infinity cancellation, type I0 compactification
and string S-matrix functional, Nucl. Phys. B 486 (1997) 149, hep-th/9609151.
[15] H. Itoyama, P. Moxhay, Multiparticle superstring tree amplitudes, Nucl. Phys. B 293 (1987)
685.
[16] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, String loop corrections to beta functions,
Nucl. Phys. B 288 (1987) 525;
A. Abouelsaood, C.G. Callan, C.R. Nappi, S.A. Yost, Open strings in background gauge fields,
Nucl. Phys. B 280 (1987) 599;
E.S. Fradkin, A.A. Tseytlin, Nonlinear electrodynamics from quantized strings, Phys. Lett.
B 163 (1985) 123.
[17] L. Dixon, D. Friedan, E. Martinec, S. Shenker, The conformal field theory of orbifolds, Nucl.
Phys. B 282 (1987) 13.
[18] J. Polchinski, String Theory, Vol. I, Cambridge University Press, Cambridge, 1998.
[19] R. Gopakumar, S. Minwalla, A. Strominger, Noncommutative solitons, JHEP 0005 (2000) 020,
hep-th/0003160.
[20] A. Erdlyi, W. Magnus, F. Oberhettinger, F.G. Tricomi, Higher Transcendental Functions,
Vol. 1, McGraw-Hill, New York, 1953.

Nuclear Physics B 593 (2001) 545561

www.elsevier.nl/locate/npe

Magnetic monopoles vs. Hopf defects in

the Laplacian (Abelian) gauge
F. Bruckmann , T. Heinzl 1 , T. Vekua 2 , A. Wipf
Friedrich-Schiller-Universitt Jena, Theoretisch-Physikalisches Institut, Max-Wien-Platz 1,
D-07743 Jena, Germany
Received 14 August 2000; accepted 25 October 2000

Abstract
We investigate the Laplacian Abelian gauge on the sphere S 4 in the background of a single t Hooft
instanton. To this end we solve the eigenvalue problem of the covariant Laplace operator in the
adjoint representation. The ground state wave function serves as an auxiliary Higgs field. We find
that the ground state is always degenerate and has nodes. Upon diagonalization, these zeros induce
topological defects in the gauge potentials. The nature of the defects crucially depends on the order
of the zeros. For first-order zeros one obtains magnetic monopoles. The generic defects, however,
arise from zeros of second order and are pointlike. Their topological invariant is the Hopf index
S 3 S 2 . These findings are corroborated by an analysis of the Laplacian gauge in the fundamental
representation where similar defects occur. Possible implications for the confinement scenario are
discussed. 2001 Elsevier Science B.V. All rights reserved.

1. Introduction
Although not derived from first principles, the dual superconductor scenario [13] is
widely believed to explain color confinement in QCD. To realize this idea, t Hooft
suggested to use Abelian projections [4] which allow for a straightforward identification
of magnetic monopoles in pure YangMills theories. In this approach one fixes the gauge
group up to its maximal Abelian subgroup. This partial gauge fixing can be characterized
by a Higgs field in the adjoint representation, which becomes diagonal in the Abelian
* Corresponding author.

E-mail address: [email protected] (F. Bruckmann).

1 Supported by DFG.
2 On leave from Department of Physics, Tbilisi State University, Charchavadze Avenue 3, 380028 Tbilisi,

Georgia. Address after July 15th: Institut fr Theoretische Physik, Universitt Hannover, Appelstrae 2, D-30167
Hannover, Germany.
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 3 5 - 0

546

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

gauge (AG) 3 . Magnetic monopoles arise as gauge fixing defects whenever vanishes.
At these points, the gauge transformation diagonalizing becomes ambiguous. In the low
temperature phase of QCD these defects should condense and play the role of Cooper pairs.
This picture is strongly supported by lattice calculations (for recent reviews, see [5,6]).
In the continuum, however, Abelian gauges are not that well understood. Considerable
progress has only been made for the Polyakov Abelian gauge (PAG) [712]. The defects
occurring in the PAG are characterized by a winding number S 2 S 2 of the (normalized)
Higgs field, n /||, or, equivalently, by the magnetic charge q of the Abelian gauge
field. A relation between monopole charge q and instanton number [A] has been
established which enforces the presence of monopoles in any non-trivial instanton sector
( 6= 0).
For the maximally Abelian gauge (MAG) [4,13,14], there are only few analytical results.
It is known that configurations with monopole lines [15] and monopole loops [16] are in
this gauge. They are, however, strongly suppressed by the gauge fixing functional, at least
in the background of single instantons. Recently, it has been explicitly shown that the
continuum MAG suffers from a Gribov problem [17] as expected from Singers theorem
[18]: the t Hooft instanton in the singular gauge is located on the Gribov horizon of the
MAG [19].
In order to circumvent the Gribov (spin glass) problem of the MAG on the lattice,
the Laplacian Abelian gauge (LAG) has been proposed as a superior alternative [2022].
Some first applications of this idea in the context of lattice gauge theory have appeared only
recently [23,24]. Analytically, however, it seems that only one result has been obtained so
far: by comparing the behaviour of the gauge fixing functionals one finds [20] that in the
LAG magnetic degrees of freedom are less suppressed than in the MAG. Some deeper
understanding of this Abelian gauge is obviously desirable.
This paper presents our first investigation of the continuum Laplacian Abelian gauge.
In order to have a large amount of symmetry, we consider (the orbit of) a single t Hooft
instanton. As we shall see, the LAG is somewhat ill-defined on infinite-volume manifolds,
and thus we compactify space to a sphere S 4 . For a special choice of the compactification
radius, the symmetry is enhanced to SO(5) so that the (gauge fixing) problem can be
exactly solved. For other radii, symmetry arguments still provide some insights. Going
back to infinite volume, we find that the singular gauge instanton and global SU(2)
rotations thereof lie in the LAG. Accordingly, the instanton in the singular gauge is a
horizon configuration, as was the case for the MAG.
Of particular interest is the question how the submanifolds of vanishing Higgs field look
like. It has been argued that, generically, these are loops, i.e., closed monopole worldlines,
having codimension three. 4 The instanton number can be recovered from these loops for
general Higgs fields [25,26]. It is related to the winding number S 2 S 2 . While the same
is true for the LAG, we find that the Higgs field associated with a single t Hooft instanton in
3 In the following we will distinguish between the Abelian gauge, which is a partial gauge fixing, and the
Abelian projections, where one neglects the off-diagonal part of the gauge field after gauge fixing.
4 As one has to solve three equations on a four-dimensional manifold.

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

547

addition induces pointlike defects, i.e., events localized in spacetime. The corresponding
topological invariant is the Hopf index S 3 S 2 .
This paper is organized as follows. First, in Section 2, we define the LAG and discuss
its properties. Single t Hooft instantons on S 4 are introduced in Section 3. In Section 4
we diagonalize the covariant Laplacian in the adjoint representation. A classification of its
ground state wave functions, which serve as auxiliary Higgs fields, is given in Section 5. A
brief discussion of the Laplacian gauge (corresponding to the fundamental representation)
is added in Section 6. Finally, we conclude with some remarks on the physical implications
of our findings.

2. The Laplacian Abelian gauge

The Laplacian Abelian gauge on R4 is defined by diagonalizing the adjoint Higgs field
which minimizes the kinetic energy [21,22],
Z


1
(1)
D a D a E a a d4 x, D = i[A , . . .].
FLAG [A, ] =
2
Throughout this paper we will only consider gauge group SU(2), so a = 1, 2, 3. The energy
R
variable E is a Lagrange multiplier demanding that is square integrable, a a d4 x <
. The field configuration minimizing FLAG can be viewed as the ground state of the
covariant Laplacian
D2 [A] = E,

(2)

where E is the ground state energy. Obviously, (2) represents a four-dimensional timeindependent Schrdinger problem with a potential essentially given by A2 .
The gauge transformation diagonalizing puts the gauge field A into the LAG,
ALAG

A,

where

1 3 .

(3)

may be ambiguous for two reasons. First, if has zeros and second, if the groundstate
is degenerate. One can use node and uniqueness theorems to analyze these issues [27].
On a spacetime with infinite volume, the LAG is not straightforwardly defined for the
following reasons. Since D2 [A] is a non-negative operator we have E > 0. Moreover,
whenever the gauge field A tends to zero at infinity, there are scattering states and
the continuous spectrum always starts from zero. Scattering states, however, are not
normalisable. Thus, for a generic background (including the instantons to be studied),
one does not expect that the covariant Laplacian D2 [A] will have a normalisable
ground state. The situation is quite analogous to the quantum mechanics of the ordinary
Laplacian d 2 /dx 2 on the real line. We avoid this problem by considering gauge fields on
the four-sphere S 4 which leads to a purely discrete spectrum of the associated covariant
Laplacian.

548

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

3. Single instanton on the sphere

In the following we consider the single t Hooft instanton (in singular and regular gauge)
on a sphere S 4 of radius R. On Euclidean R4 the configurations read,
sg

a
x
A =

2
a ,
+ 2)

r 2 (r 2

reg

a
A =
x

1
a ,
(r 2 + 2 )

r 2 = x x ,

(4)

using the conventions of [28]. The configurations (4) are related by the gauge transformation
h = x4 12 + i x a a ,

x x /r,

(5)

which also relates the solutions of (2) in these two backgrounds, reg = h sg h .
We benefit from the facts that classical YangMills theories are conformally invariant
4 . If we
and that the sphere S 4 is conformally equivalent to compactified Euclidean space R
use conformal coordinates x on the sphere, which are simply the Cartesian coordinates of
the point stereographically projected onto R4 , the metric is conformally flat
g (x) = eR (r)

4R 4
.
(r 2 + R 2 )2

(6)

Field configurations minimizing the YangMills action on R4 are also minimizing

configurations on the sphere, if the Cartesian coordinates are substituted by conformal
coordinates. Thus, we can simply use expressions (4) for the instantons on the sphere.
What about the symmetry of these configurations? It is known [29] that on R4 they are
invariant under SO(4) rotations and a combination of translations and special conformal
transformations
x = x + 2 c x x / c (x 2 + 2 )/

(7)

up to a compensating gauge transformation, A = D , with

1
a
a
2c
x
a =
2

(reg. gauge).

(8)

Together these transformations form (a non-linear representation of) the group SO(5). This
symmetry is preserved on S 4 when the radius R of S 4 coincides with the instanton size
[30]. To illustrate this point, we note that the gauge invariant Lagrangian density
a
a
F
=
L g g F

12 4 (r 2 + R 2 )4
R 8 (r 2 + 2 )4

(9)

is constant on S 4 (and thus SO(5)-invariant) only if R = . For R 6= the explicit

appearance of r, which is only SO(4)-invariant, breaks SO(5) down to SO(4).
Since compensating gauge transformations do not spoil Eq. (2), the eigenfunctions
furnish representations of SO(5) and SO(4), respectively.

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

549

4. Solutions of the covariant Laplacian

Generalizing Eqs. (1) and (2) to curvilinear coordinates, we define the LAG on the sphere
S 4 via the functional
Z


1
(10)
D a D a g E a a g d4 x,
FLAG [A, ] =
2
S4

where g = exp(2R ) denotes the determinant of the metric (6). The equation of motion is
given in terms of the (gauge covariant) LaplaceBeltrami operator,
1

D gg D = E.
g

(11)

To proceed we make use of the symmetry and separate into angular and radial equations.
The angular part is expressed in terms of angular momenta derived from the decomposition
so(4)
= su(2) su(2),
i a
x ,
Ma =
2
i a
x ,
Na =
2

E 2 m(m + 1),
M
NE 2 n(n + 1).

(12)

E 2 = NE 2 . Their eigenvalues are halfIn this representation, the two Casimirs coincide, M
integer, m = n {0, 1/2, 1, 3/2, . . .}. The generators for isospin t = 1 are
(Ta )bc = ibac ,

TE 2 t (t + 1) = 2.

(13)

The radial equations on the sphere differ from those in Euclidean space by a metric factor,
exp(R ), and a dilatation term, rr ,

E 2 4 2 (JE2 M
E 2)
3
4M
R (r)
r2 r + 2 + 2 2
e
r
r
r (r + 2 )

2
2
4r
4TE
+

(14)
sg = E sg ,
2
r
(r + 2 )2 r 2 + R 2

3
4NE 2 4(JE2 NE 2 )
eR (r) r2 r + 2 +
r
r
(r 2 + 2 )

2
2
4r
4TE
+
r reg = E reg .
(15)
2
(r + 2 )2 r 2 + R 2
In the above, we have introduced the conserved angular momentum JE (spin from isospin,
[3133]),
E + TE ,
JE L

JE 2 j (j + 1),

j {l 1, l, l + 1},

(16)

E denotes M
E or NE , respectively. Replacing angular momenta by their eigenvalues
where L
and exchanging j n, m j , Eq. (14) turns into (15). This amounts exactly to the action
of the gauge transformations h from (5).
The symmetry considerations above suggest the following form of the ground state

550

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

(x) = Y(j,l) (x)(r),

(17)

where the Y s denote the spherical harmonics on S 3 (see Appendix A). Note that there
are two competing angular momentum terms in (14) and (15), so that it is not obvious in
which angular momentum sector the groundstate will be. By simply looking at the radial
potentials in the different sectors, we can only state the following bound on the energy in
an arbitrary sector


(18)
E(j,l) > min E(0,1), E(1/2,1/2), E(1,0) .
The quantum numbers of the ground state candidates on the r.h.s. correspond to the
representations (0,1), (1/2, 1/2) and (1,0) of su(2)j su(2)l and thus have degeneracies
3, 4 and 3, respectively. Note that the singlet (0, 0) is excluded by the selection rules for
t = 1, see (16). Accordingly, for any of the possible choices in (18), the groundstate will
be degenerate. The spherical harmonics for the three cases are listed in Appendix A.
At this point two further remarks are in order: First, the radial part shows power law
behaviour in r, both for small and large r, independent of R and ,
sg (r 0) r 2j ,

reg

(r 0) r ,
2n

sg (r ) r 2m ,

reg

(r ) r

(r 2 +R 2 )

2j

(19)
.

(20)

and ER 2 +2, one can absorb the dilatation

Second, upon substituting

term,
#
"
E 2 4 2 (JE 2 M
E 2)
4TE 2 2
3
4M
4R 2
2
2

sg = 0, (21)
r r + 2 + 2 2
r
r
r (r + 2 )
(r + 2 )2 (r 2 + R 2 )2
#
"
4TE 2 2
3
4NE 2 4(JE 2 NE 2 )
4R 2
2
2

reg = 0. (22)
r r + 2 +
r
r
(r 2 + 2 )
(r + 2 )2 (r 2 + R 2 )2
Setting R = , the differential equation (21) coincides with the one considered by t Hooft
in his analysis of the fluctuations around instantons [34]. The eigenvalues are k = (k +
j + l + 1 t)(k + j + l + t + 2). The lowest energy corresponds to k = 0 and j + l =
1, consistent with the three possible groundstates of (18). Together they form the 10dimensional adjoint representation 5 of SO(5) [35]. The value of the ground state energy is
E = 2/R 2 .
The radial eigenfunctions are rational
(r/R)2j
(r/R)22j
,
reg (r) = R 2
(23)
2
2
r +R
r + R2
and obey the asymptotics (19) and (20), respectively. In accordance with the node theorem
for the one-dimensional radial equation, the lowest-lying states have no zeros apart from
r = 0 and r = .
For the cases R > and R < we cannot solve the radial equation analytically.
However, we are able to prove the following statements:
sg (r) = R

5 Using the conventions of [35], this representation is labelled by the integers {n , n } = {0, 2} which are the
1 2
coefficients of the highest weight when expanded in terms of the fundamental weights.

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

551

Fig. 1. Energy of the lowest-lying states in the relevant angular momentum sectors as functions of the
compactification radius R (singular gauge). At the point R = the two triplets and the quadruplet
meet, while for R the triplet (1, 0) has lowest energy. For symmetry reasons we expect the
dashed line to stay inbetween the other two for R 6= .

For the singular gauge and R > , the triplet (1, 0) has lower energy than the
triplet (0, 1). For R < , the situation is vice versa with the triplet (0, 1) having lower
energy. Analogous statements hold for the regular gauge (see Fig. 1). These results are a
straightforward consequence of the FeynmanHellmann theorem (cf. Appendix B).
For the quadruplet (1/2, 1/2), the situation is somewhat more complicated. Using
perturbation theory in = 2 R 2 (see Appendix B), one finds that these states have
energy inbetween the two disjoint triplet states. For symmetry reasons we do not expect
the spectral flow E(1/2,1/2)(R) to intersect the others for some R 6= (see Fig. 1).
Finally, the node theorem again guarantees that vanishes only at r = 0 and r = , in
accordance with the asymptotics (19,20).

5. Properties of the solutions

Before characterizing the zeros of the solutions , let us point out the following subtlety:
Near the origin, the (0,1) wave functions (Higgs fields) in the singular gauge are bilinear
in x and thus discontinuous there. They inherit this singularity from the instanton field. 6
Nevertheless, the wave functions are square integrable on S 4 due to the measure factor
r 3 . The same, of course, is true for the regular gauge states near infinity. In order to
work with smooth Higgs fields, it is appropriate to use the principal fibre bundle picture.
This can be viewed as a non-Abelian S 4 -analogue of the WuYang construction for the
Dirac monopole on S 2 [36]. The A-field in the regular gauge represents the connection
smoothly 7 on the southern hemisphere (the chart containing the origin), while the A-field
in the singular gauge does the same on the northern hemisphere (the chart containing
6 Which results in the asymptotics (r) r 0 , see Eq. (19).
7 To be precise: The A-fields are the pullbacks of the connection under smooth sections of the bundle.

552

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

Fig. 2. In the bundle picture on S 4 there are two gauge and Higgs fields, which are smoothly defined
on their domains (hemispheres). In the transition region, they are related by the gauge transformation
(transition function) h. Note that the four-dimensional radius can be expressed in terms of the
azimuthal angle , r = R cot(/2).

infinity). In the transition region formed by the equatorial strip displayed in Fig. 2, the
gauge transformation h from (5) interpolates between the two. For simplicity we will
retract the transition region to a single three-sphere Sr3 of fixed four-dimensional radius
r (fixed azimuthal angle ).
The Higgs field is a section in an associated fibre bundle: on each of the two charts there
is a Higgs field. In the transition region, the same transition function h relates the two (see
Fig. 2). Our results obtained so far can immediately be carried over to the bundle picture,
since, for every solution in the singular gauge, there is a corresponding gauge transformed
mirror solution in the regular gauge with the same energy (and vice versa). Moreover, the
angular momenta are interchanged by h in such a way that the radial wavefunctions (23)
are smoothly defined on the whole of S 4 . The complete eigenfunctions are continuous
in their respective charts but jump (in their isospin direction) due to the action of the
transition function h in the transition region.
Along these lines, let us discuss the ground state in the (1/2, 1/2) sector, which has
zeros localized on loops. To simplify the discussion, we choose the fourth of the spherical
harmonics in (A.2) or (A.3), for which

x1
1
= reg (x),
(24)
sg (x) = x2 2
r + R2
x3
since h from (5) commutes with sg . This Higgs field is of hedgehog type. Thus, its
diagonalization induces a Dirac monopole at xE = 0 and a Dirac string along the negative 3axis. The world-line of the monopole is the great circle in 4-direction (which degenerates
to the 4-axis in the infinite-volume limit). For the other three states in the multiplet the
same holds true upon permutation of the coordinates.
It is well-known [37] that the monopole charge is characterized by the winding number
of the normalized Higgs field n = /||, explicitly given by

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

553

x1
nsg (x) = x2 /|E
x | = nreg (x).
x3

(25)

The n-field is singular at xE = 0, where vanishes. This singularity has the following
topological characterization. Consider the two-sphere, S 2 : xE = const, surrounding the
singularity. There, the n-field provides a smooth mapping, S 2 S 2
= SU(2)/U (1),
labelled by an integer, the winding number, which in our case is just one.
The LAG Higgs field also serves as an illustration of the relation between instanton
number and monopole charge recently proposed in [26]. We note that in the (1/2, 1/2)
sector the two Higgs fields nsg and nreg coincide on the whole of S 4 , their singularities
E r), p2 = (0,
E r) in the retracted transition region Sr3
being located at two points p1 = (0,
(see Fig. 3). Consider the submanifold Sr3 \ {p1 , p2 }
= S 2 I12 . In terms of the polar angle
(0, ) on the three-sphere, the interval I12 is parametrized by x4 = r cos (r, r),
while the two-spheres are given by |E
x | = r sin .
We already know that the magnetic charge measured on the two-sphere is q = 1. If we
express the transition function h in terms of n and ,
h = exp(i na a ),
R
the flux 8 = I12 d(2) is easily computed as

(26)

Z
d = 2.

=2

(27)

We thus recover the instanton number,

= 1.
(28)
2
Note that for linear combinations of states from different multiplets the monopole loops
become tilted. As an example, take a combination of the sectors (1,0) and (1/2, 1/2),
[A] = q

Fig. 3. Submanifolds of S 4 on which the ground state wave function vanishes so that the normalized
Higgs field becomes singular. The sector (1/2,1/2) gives rise to monopole loops C, while the generic
sectors (0,1) and (1,0) lead to pointlike singularities with Hopf index as topological invariant.
8 Notice the difference vs. /2 as compared to [26].
a
a

554

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

x1


1
1
reg
reg
x2 .
reg = (1,0) (1/2,1/2) =
2
2(r 2 + R 2 ) R x

(29)

This Higgs field vanishes for x = (0, 0, R, x4 ), a set of zeros which is still a great circle
but does no longer include the poles.
As we have already argued, the quadruplet states (1/2, 1/2) will occur as ground states
only for R = . In the general case, R 6= , the ground states will be the triplet states (0, 1)
and (1, 0), which have isolated, pointlike zeros. Let us specialize to the physical region R >
which includes the infinite-volume limit, R . (For R < one has to perform the
appropriate mirror transformation.) We know that for the southern hemisphere (regular
gauge) and for the northern hemisphere (singular gauge) we have to take (j, l) = (0, 1) and
(j, l) = (1, 0), respectively, since these multiplets consist of the lowest-lying states. If we
choose the third of the spherical harmonics in (A.1) and (A.5), the normalized Higgs field
is given by

2(x1 x3 x2 x4 )

nreg (x) = 2(x x + x x ) southern hem. 3 0,

2 3
1 4

x12 x22 + x 32 + x42

(30)
n(x) =

sg

northern hem. 3 .
n (x) = 0

1
nreg is singular at the origin r = 0 and closely resembles the standard Hopf map [38,39].
3
S2
For any finite radius r 6= 0, it provides a smooth mapping Sr=fixed
= SU(2)/U (1)
with Hopf index one [40].
As a result, we have obtained the simplest realization of the connection between
instanton number and Hopf indices derived in [25]: The (signed) sum of all Hopf indices
of n around its singularities equals the instanton number . This statement is analogous to
results from residue calculus where the singularities of n (the zeros of ) are replaced by
the poles of a meromorphic function: The (signed) sum of all residues equals the residue
at infinity. Like the magnetic monopoles in the PAG, n must possess singularities in any
non-trivial instanton sector ( 6= 0). In addition, pairs of singularities may occur which do
not contribute to the instanton number.
Coming back to the LAG, the remaining task is to diagonalize the ground-state Higgs
field (x). On the northern hemisphere, where is already diagonal, there is nothing to
be done. No gauge transformation is needed and the gauge field remains in the singular
gauge. On the southern hemisphere, we basically have to diagonalize the standard Hopf
map. This is achieved by the gauge transformation h, which transforms the gauge field
A from regular to singular gauge. Independent of where we choose the transition region,
the LAG-fixed configuration on the orbit of the single t Hooft instanton is in the singular
gauge (for R > ). Notice that the gauge fixed configuration inherits a singularity only at
the point where n is singular; there are no further Dirac strings.
If we choose an arbitrary linear combination of the triplet spherical harmonics, the
diagonalizing gauge transformation includes an additional global SU(2) rotation. Together

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

555

with Asg , all its global rotations are located on the gauge fixing hypersurface defined by
the LAG. We thus find a whole S 3 of gauge-equivalent configurations (Gribov copies).

6. The Laplacian gauge

The importance of pointlike defects (as compared to loops) is corroborated by their
occurrence in a closely related gauge, the Laplacian gauge (LG). 9 The Laplacian gauge
[4144] is defined via a Higgs field q in the fundamental representation, being the ground
state of the covariant Laplacian,
D2 [A]q = Eq,

D = iA .

(31)

It is a complete gauge fixing (up to defects) if the two-component complex vector q is

rotated into a fixed isospin direction and made real,
 
|q|

q 1 q =
,
ALG A.
(32)
0
Our formalism is easily adapted to this gauge by choosing the isospin t = 1/2
representation in terms of the Pauli matrices Ta = a /2. For R = one again has to
minimize j + l, whence (j, l) = (0, 1/2) or (j, l) = (1/2, 0). As before, these states form
an irreducible representation 10 of SO(5). For R > and the singular gauge, the state
(1/2, 0) has lowest energy (by the same FeynmanHellmann argument) so that the singular
gauge instanton again satisfies the gauge condition. The relevant spherical harmonics are
   
1
0
reg
,
,
(33)
Y(1/2,0) =
0
1
   

  

1
x4 + i x3
0
x2 + i x 1
reg
=
, h
=
,
(34)
Y(0,1/2) = h
x2 + i x 1
x4 i x 3
0
1
which are nonzero throughout S 4 . In analogy with (20), we have the following behaviour
near the origin, q(r) r 2n = r for (j , n) = (0, 1/2). Thus, the modulus of the Higgs field
is proportional to the four-dimensional distance r from the origin (where the topological
charge of the instanton is concentrated). This perfectly agrees with latest results from lattice
simulations [45].
Again, a topological description is possible. On a three-sphere surrounding the origin,
one can define n q/|q| : S 3 S 3 with integer winding number. In the case above, the
n-field simply reduces to the identity map


x4 + i x3
,
(35)
nY =
x2 + i x 1
the winding number k of which coincides with the instanton number, k = = 1.
9 The authors thank P. de Forcrand for drawing their attention to this issue.
10 The four-dimensional spinor representation labelled by {n , n } = {0, 1} in [35].
1 2

556

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

7. Conclusions
We have investigated the Laplacian Abelian gauge on the sphere S 4 in the background
of a single t Hooft instanton. This amounts to solving the eigenvalue problem for the
covariant Laplacian in the adjoint representation. For any sphere radius R we have
determined the angular dependence and isospin structure of the ground state wave
functions (Higgs fields). Diagonalization of the latter shows that the instanton in the
singular gauge is in the LAG if R is larger than the instanton size ; for the regular gauge
the same is true for R < . The gauge fixing procedure thus selects one of the two instanton
configurations, although, in a bundle picture, they represent the same connection.
It is interesting to note that the situation for the MAG on the sphere is similar: Singular
and regular gauge instantons both satisfy the differential MAG condition, but the MAG
functional FMAG picks out one of them in the very same way as FLAG : for R > (R < )
the singular (regular) gauge instanton minimizes FMAG (see Appendix C). It is, however,
a highly nontrivial task to check whether a given configuration, say the t Hooft instanton,
really corresponds to the absolute minimum along its orbit. In general, one can never be
sure that there is no other gauge equivalent configuration that lowers the functional even
further.
The LAG, on the other hand, has the big advantage that the ground state (and thus the
absolute minimum of FLAG ) can be found explicitly. We have done so for R = and have
given qualitative arguments concerning the angular and radial dependence for the case
R 6= . Apart from the degeneracies and the zeros (which we have under control), there are
no further ambiguities.
We have found a whole S 3 of gauge equivalent configurations (obtained by global
SU(2) rotations of Asg similar to what has been observed in [46]) located on the gauge
fixing hypersurface. These are Gribov copies of each other, generated by both finite and
infinitesimal gauge transformations. The latter give rise to three flat directions in the
configuration space along which the gauge fixing functional does not change. Only one
of these directions is covered by the residual U (1) freedom. The other two are related to
zero modes of the (coset part of the) FaddeevPopov operator. We do see no reason why
these Gribov ambiguities should not be present on the lattice. In contrast to the MAG (and
related gauges), however, where gauge fixing is a numerical problem of non-polynomial
complexity [42], there are no additional lattice Gribov copies beyond the denumerable
ones we have encountered in the continuum. This clearly makes the LAG a superior gauge.
Once a ground states is chosen for diagonalization, additional obstructions occur in
terms of gauge fixing defects caused by the nodes of the possible ground states. These
are the well-known source for magnetic monopoles in Abelian gauges. We have shown
that these defects must be present whenever the LAG background is in a non-trivial
instanton sector. Monopoles, however, only arise for a particular sphere radius R = and
for a particular choice of ground states. The generic defects are localized in spacetime
(with codimension 4). Their topological invariant is the Hopf index S 3 S 2 . Contrary
to monopoles they have finite action even in the infinite volume limit. One may speculate
that these defects condense in the low temperature phase of QCD, possibly giving rise to

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

557

a new confinement mechanism. In view of the results presented in [26], they may as well
be related to the solitonic excitations observed in recent effective theories for confinement
[4750].
As we have calculated the LAG Higgs field only for a highly symmetric background,
the question arises which features are generic also for other backgrounds. The degeneracy
of the ground state is mainly due to the matrix structure of the Hamiltonian. For a
single instanton background, this was induced by nonvanishing angular momentum (like
in quantum mechanical problems with spin). This should be contrasted with the case of
a trivial background. For the vacuum, A = 0, the ground state obviously has a threefold
degeneracy given by the canonical dreibein ea in isospace. The associated constant wave
functions do not have any zeros. We therefore conjecture that Singers obstruction [18]
against complete gauge fixing is reflected in the nodes rather than in the degeneracy of the
ground state. To completely settle this issue, a full topological classification of Higgs field
zeros would clearly be helpful.
A natural next step will be to analyze higher instanton sectors and instanton-antiinstanton pairs. The existence of fermionic zero modes in the background of Hopf defects
is currently being investigated (for related work see [51] and references therein). Such
zero modes may in the end lead to a relation between confinement and chiral symmetry
breaking. Finally, the dynamical role of Hopf defects in QCD has to be analyzed.

Acknowledgements
The authors thank S. Shabanov and T. Strobl for enlightening discussions, P. de
Forcrand for making his lattice results available prior to publication, and D. Hansen for
a careful reading of the manuscript. T.V. gratefully acknowledges the hospitality at the
TPI, University of Jena, where this work was performed. T.H. thanks C. Alexandrou and
G. Burgio for discussions on Laplacian gauges and acknowledges support under DFG grant
WI-777/5-1.

Appendix A. Spherical harmonics

E 2 for the three cases of interest
In the following we list the eigenfunctions of JE 2 and L
(suppressing the two magnetic quantum numbers labelling the vectors in each multiplet).
(i) For (j, l) = (1, 0) the spherical harmonics are given by the canonical dreibein ea of
constant unit vectors

0
0
1
sg
reg
(A.1)
Y(1,0) = Y(1,0) = 0 , 1 , 0 .

0
0
1
(ii) For (j, l) = (1/2, 1/2) there are four eigenfunctions, all linear in x ,

558

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

sg

Y(1/2,1/2)

reg

Y(1/2,1/2)

x3
x2
x1
x4
= x3 , x4 , x1 , x2 ,

x2
x1
x4
x3

x3
x2
x1
x4

=
x3 , x4 ,
x1 , x2 .

x2
x1
x4
x3

(A.2)

(A.3)

The following remarks are in order. Obviously, Y reg is obtained from Y sg upon exchanging
sg
reg
x4 x4 . This is achieved via conjugation with h, Y(j =1/2, m=1/2) h Y(j =1/2, n=1/2)h.
Note that the intertwining gauge transformation h is only defined up to rotations around
the direction of the Higgs field in isospace. It is convenient to combine the members
of each (1/2, 1/2) quadruplet into a four-vector Y . Introducing the basis matrices
reg
sg
(i a , 1), one finds the relation Y = Y for any = 1, . . . , 4. Any component
vanishes, if x = e , the e denoting the canonical basis of R4 . This means that
Y (x)
the zeros of the quadruplet eigenfunctions are given by two points located on a three-sphere
with fixed radius r (see Fig. 3).
(iii) For the case (j, l) = (0, 1) one has three basic eigenfunctions, now bilinear in x ,
( 2 2 2 2!
!
!)
x1 x2 x3 + x4
2(x1 x2 + x3 x4 )
2(x1 x3 x2 x4 )

sg

Y(0,1) =
(
reg
Y(0,1)

x12 x22 x32 + x42

2(x1 x2 x3 x4 )
2(x1 x3 + x2 x4 )

2(x1 x2 x3 x4 )
x12 + x22 x32 + x42
2(x2 x3 + x1 x4 )

!
,

2(x1 x2 + x3 x4 )
x12 + x22 x32 + x42
2(x2 x3 x1 x4 )

2(x1 x3 + x2 x4 )
2(x2 x3 x1 x4 )
x12 x22 + x32 + x42

,
!

2(x1 x3 x2 x4 )
2(x2 x3 + x1 x4 )
x12 x22 + x32 + x42

, (A.4)
!)
. (A.5)

Again, the two sets of eigenfunctions are related via x 4 x4 and can most easily be
obtained from case (i) by conjugation with h
sg

reg

Y(j =0, m=1) = h Y(j =1, n=0) h,

reg

sg

Y(j =0, n=1) = h Y(j =1, m=0) h ,

(A.6)

which, in particular, implies that they never vanish.

Appendix B. FeynmanHellmann theorem and perturbation theory

In order to obtain information when R 6= , we keep R fixed and vary . We restrict
ourselves to the singular gauge. The -dependent part of (14) contains two terms
 2 E2

E2
TE 2 2
sg
R (r) (J M )
2
.
(B.1)
V (j,m) (r) 4e
r 2 (r 2 + 2 )
(r + 2 )2
The 2 -dependence of the ground state energy is determined by the FeynmanHellmann
theorem

H
V

E = 2 h|H |i = h| 2 |i h| 2 |i.
2

For the three angular momentum sectors of interest (t = 1) we have

(B.2)

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

sg

V (0,1) (r)
2

(r 2 + R 2 )2 4r 2
< 0,
R4
(r 2 + 2 )3

sg

V (1/2,1/2)(r)
2
sg

V (1,0) (r)
2

559

(r 2 + R 2 )2 2( 2 r 2 )
,
R4
(r 2 + 2 )3

(r 2 + R 2 )2
4 2
> 0.
4
2
R
(r + 2 )3

(B.3)

According to (B.2), these functions have to be integrated with the positive factor ||2 g.
Therefore, the ground state energies in the first and the third sector are monotonic in 2 ,
their slopes satisfying
sg
sg
E(0,1) < 0,
E
> 0.
2

2 (1,0)
As the energies meet at R = (level crossing) we conclude,
sg

sg

E(0,1) < E(1,0)

for R < ,

sg

sg

E(0,1) > E(1,0)

(B.4)

for R > .

(B.5)

This explains the behaviour of the full lines in Fig. 1.

For the sector (1/2, 1/2) there is no such simple argument. Still, we can compute the
slope of E( 2 ) at the point = R by simply inserting the known function . This amounts
to ordinary perturbation theory in 2 R 2

H 
2
H ( ) = H ( = 0) + 2 
+ O( 2 ) = H0 + Hpert .
(B.6)
=0
In this way we find a vanishing slope for the sector (1/2, 1/2),

Z

(1 r 2 ) 5
sg
E(1/2,1/2)

r dr = 0.
2

(r 2 + 1)7
2 =R 2

(B.7)

The lowest-lying state of this sector is thus pinched between the other two, at least for
R (cf. Fig. 1).
Appendix C. The MAG on the sphere
In [16] it has been shown that, due to their particular Lorentz and isospin structure, both
Asg and Areg are in the MAG when defined on R4 . This still holds true on S 4 , where the
gauge fixing functional has the values
Z X
2

Aa Aa g g d4 x
FMAG [A] =
a=1


16 2 R 4 [R 4 2R 2 2 ln(R 2 / 2 ) 4 ]
1

=
2 / 2
2
2
2
3
R
(R )

for Asg ,
for Areg .

(C.1)

Obviously, FMAG [Areg ] = (R 2 / 2 ) FMAG [Asg], so that for R > (R < ) the singular
(regular) gauge is singled out.

560

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

References
[1] G. t Hooft, in: A. Zichichi (Ed.), High Energy Physics, Proceedings of the EPS International
Conference, Palermo, 1975, Editrice Compositori, Bologna, 1976.
[2] S. Mandelstam, Phys. Rept. 23 (1976) 245.
[3] G. Parisi, Phys. Rev. D 11 (1975) 970.
[4] G. t Hooft, Nucl. Phys. B 190 (1981) 455.
[5] M.N. Chernodub, M.I. Polikarpov, in: P. van Baal (Ed.), Confinement, Duality, and Nonperturbative Aspects of QCD, NATO ASI Series, Plenum Press, New York, 1998.
[6] R.W. Haymaker, Phys. Rept. 315 (1999) 153, hep-lat/9809004.
[7] N. Weiss, Phys. Rev. D 24 (1981) 475.
[8] H. Reinhardt, Nucl. Phys. B 503 (1997) 505, hep-th/9702049.
[9] O. Jahn, F. Lenz, Phys. Rev. D 58 (1998) 085006, hep-th/9803177.
[10] C. Ford, U.G. Mitreuter, T. Tok, A. Wipf, J.M. Pawlowski, Ann. Phys. (N.Y.) 269 (1998) 26,
hep-th/9802191.
[11] C. Ford, T. Tok, A. Wipf, Nucl. Phys. B 548 (1999) 585, hep-th/9809209.
[12] C. Ford, T. Tok, A. Wipf, Phys. Lett. B 456 (1999) 155, hep-th/9811248.
[13] A.S. Kronfeld, M.L. Laursen, G. Schierholz, U.-J. Wiese, Phys. Lett. B 198 (1987) 516.
[14] A.S. Kronfeld, G. Schierholz, U.-J. Wiese, Nucl. Phys. B 293 (1987) 461.
[15] M.N. Chernodub, F.V. Gubarev, JETP Lett. 62 (1995) 100, hep-th/9506026.
[16] R.C. Brower, K.N. Orginos, C.-I. Tan, Phys. Rev. D 55 (1997) 6313, hep-th/9610101.
[17] V. Gribov, Nucl. Phys. B 139 (1978) 1.
[18] I.M. Singer, Commun. Math. Phys. 60 (1978) 7.
[19] F. Bruckmann, T. Heinzl, T. Tok, A. Wipf, Nucl. Phys. B 584 (2000) 589, hep-th/0001175.
[20] A.J. van der Sijs, Nucl. Phys. (Proc. Suppl.) 53 (1997) 35, hep-lat/9608041.
[21] A.J. van der Sijs, Prog. Theor. Phys. Suppl. 131 (1998) 149, hep-lat/9803001.
[22] A.J. van der Sijs, Nucl. Phys. (Proc. Suppl.) 73 (1999) 548, hep-lat/9809126.
[23] C. Alexandrou, M. DElia, P. de Forcrand, The relevance of center vortices, hep-lat/9907028.
[24] C. Alexandrou, P. de Forcrand, M. DElia, The role of center vortices in QCD, hep-lat/9909005.
[25] O. Jahn, J. Phys. A 33 (2000) 2997, hep-th/9909004.
[26] T. Tsurumaru, I. Tsutsui, A. Fujii, Instantons, monopoles and the flux quantization in the
Faddeev-Niemi decomposition, hep-th/0005064.
[27] M. Reed, B. Simon, Analysis of Operators, Methods of Modern Mathematical Physics, Vol. 4,
Academic Press, Boston, 1978.
[28] T. Schfer, E.V. Shuryak, Rev. Mod. Phys. 70 (1998) 323, hep-ph/9610451.
[29] R. Jackiw, Acta Phys. Austr. Suppl. 22 (1980) 383.
[30] R. Jackiw, C. Rebbi, Phys. Rev. D 14 (1976) 517.
[31] R. Jackiw, C. Rebbi, Phys. Rev. Lett. 36 (1976) 1116.
[32] P. Hasenfratz, G. t Hooft, Phys. Rev. Lett. 36 (1976) 1119.
[33] A. Goldhaber, Phys. Rev. Lett. 36 (1976) 1122.
[34] G. t Hooft, Phys. Rev. D 14 (1976) 3432.
[35] J.F. Cornwell, Group Theory in Physics, Vol. 2, Academic Press, London, 1984.
[36] T.T. Wu, C.N. Yang, Phys. Rev. 12 (1975) 3854.
[37] J. Arafune, P.G.O. Freund, C.J. Goebel, J. Math. Phys. 16 (1975) 433.
[38] H. Hopf, Math. Ann. 104 (1931) 637.
[39] M. Nakahara, Geometry, Topology and Physics, Adam Hilger, Bristol, 1990.
[40] L.H. Ryder, J. Phys. A 13 (1980) 437.
[41] G. Schierholz, J. Seixas, M. Teper, Phys. Lett. B 157 (1985) 209.
[42] J.C. Vink, U.-J. Wiese, Phys. Lett. B 289 (1992) 122, hep-lat/9206006.
[43] J.C. Vink, Phys. Rev. D 51 (1995) 1292, hep-lat/9407007.
[44] P. van Baal, Nucl. Phys. Proc. Suppl. 42 (1995) 843, hep-lat/9411047.

F. Bruckmann et al. / Nuclear Physics B 593 (2001) 545561

[45]
[46]
[47]
[48]
[49]
[50]
[51]

561

P. de Forcrand, M. Pepe, Laplacian gauge and instantons, hep-lat/0010093.

L. Baulieu, A. Rozenberg, M. Schaden, Phys. Rev. D 54 (1996) 7825, hep-th/9607147.
Y.M. Cho, H. Lee, D.G. Pak, Effective theory of QCD, hep-th/9905215.
L. Faddeev, A. Niemi, Nature 387 (1997) 58.
L. Faddeev, A. Niemi, Phys. Rev. Lett. 82 (1998) 1624, hep-th/9807069.
S.V. Shabanov, Phys. Lett. B 458 (1999) 322, hep-th/9903223.
C. Adam, B. Muratori, C. Nash, Particle creation via relaxing hypermagnetic knots, hepth/0006230.

Nuclear Physics B 593 (2001) 562574

www.elsevier.nl/locate/npe

One-loop shift in noncommutative ChernSimons

coupling
Guang-Hong Chen a,b, , Yong-Shi Wu a
a Department of Physics, University of Utah, Salt Lake City, UT 84112, USA
b Institute for Theoretical Physics, University of California, Santa Barbara, CA 93106, USA

Abstract
In this paper we study the one-loop shift in the coupling constant in a noncommutative pure
U (N) ChernSimons gauge theory in three dimensions. The one-loop shift is shown to be a constant
proportional to N, independent of noncommutativity parameters, and non-vanishing for U (1) theory.
Possible physical and mathematical implications of this result are discussed. 2001 Elsevier Science
B.V. All rights reserved.

1. Introduction
Field theory (especially gauge field theory) on a noncommutative space (or spacetime)
has attracted much interest recently [111]. That such theory arises from string/M(atrix)
theory [12,13] suggests that space or spacetime noncommutativity should be a general
feature of quantum gravity for generic points deep inside the moduli space of M-theory.
Moreover, being a natural deformation of usual quantum field theory, noncommutative
field theory is of interests in its own right. A charge in the lowest Landau level in a
strong magnetic field can be viewed as living in a noncommutative space, because the
guiding-center coordinates of the charge are known not to commute. In this paper, we
study noncommutative ChernSimons (NCCS) field theory in 3 dimensions, which is a
deformation of ordinary ChernSimons (CS) theory and may have applications in planar
condensed matter systems, especially in the quantum Hall systems. For simplicity, in this
paper we mainly consider pure NCCS theory, with gauge group U (N) and with no matter
fields coupled to it.
Three-dimensional noncommutative spacetime has coordinates satisfying
[x , x ] = i ,

, = 0, 1, 2,

* Corresponding author.

E-mail addresses: [email protected] (G.-H. Chen), [email protected] (Y.-S. Wu).

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 3 0 - 1

(1)

G.-H. Chen, Y.-S. Wu / Nuclear Physics B 593 (2001) 562574

563

where are antisymmetric and real parameters of dimension length squared. The action
for a pure U (N) CS theory on this space reads


Z
i
2
d 3 x Tr A A + A A A .
(2)
ICS =
4
3
Here the dynamical field is the gauge potential A = Aa T a , T a the generators of the

gauge group G = U (N), normalized to Tr(T a T b ) = ab /2 with T 0 = i/ 2N for the

U (1) sector. is the CS coupling, the totally antisymmetric tensor with 012 = 1. In
the action (2) we are using a representation, in which the coordinates x are the same as
usual, but the product of any two functions of x is deformed to the Moyal star-product:

x y
(3)
f g(x) = e(i/2) f (x)g(y)y=x .
The commutator in Eq. (1) is understood as the Moyal bracket with respect to the star
product:
[x , x ] x x x x .

(4)

For applications to a system in the lowest Landau level, one considers only the spatial
noncommutativity: 01 = 02 = 0 and [x 1 , x 2 ] = i .
It is obvious that if = 0, the action (2) reduces to that of ordinary pure CS theory in
3 dimensions [14], which is known to be a topological quantum field theory [15], with
the partition function and the correlation functions of Wilson loops being topological
invariants, independent of spacetime metric. Diagrammatically the ordinary CS theory is
renormalizable [16]. Many topological features can be probed in perturbation theory [17].
One interesting result is the one-loop quantum shift of the non-Abelian CS coupling [17].
However, ordinary pure Abelian CS theory has no such shift, while theory with additional
matter coupling does at the two-loop level [1820]. In this paper we will show that there
is a non-vanishing one-loop shift in noncommutative CS coupling even if the gauge group
is U (1). This shift turns out to be a constant proportional to the integer N , independent
of the noncommutativity parameters , and identical to the one-loop shift in ordinary
SU(N) CS theory when N > 2. Possible physical and mathematical implications of our
results will be discussed.

2. Regularized Feynman rules

The action (2) is invariant under the following infinitesimal gauge transformations:
A = D + [A , ].

(5)

To do perturbation theory, we follow the standard procedure of path integral quantization

to establish the Feynman rules. The full, regularized action after gauge fixing in Euclidean
spacetime reads
Itot = ICS + IYM + Igf + Igh .

(6)

564

G.-H. Chen, Y.-S. Wu / Nuclear Physics B 593 (2001) 562574

Here we have added the noncommutative YangMills (YM) term

Z


1
IYM = 2 d 3 x Tr F F ,
2e

(7)

with the field strength F defined by

F = A A + [A , A ].

(8)

This gauge invariant term in the action provides a higher-derivative regularization for the
CS theory, since the YM coupling e2 is of dimension of mass, which is used as a cut-off
that is sent to infinity at the end of calculations. The third term in Eq. (6) is the gauge fixing
term:
Z

2
1
(9)
Igf = 2 d 3 x Tr A ,
e
a linear, covariant gauge condition convenient for perturbation theory. In the following
we are going to take the Landau gauge = 0, which was known to have computational
advantages in the infrared in ordinary CS theory [16,20]. The last term Igh is the ghost
action corresponding to the above gauge fixing:
Z


(10)
Igh = d 3 x Tr cD c ,
where c and c are the ghost and anti-ghost, respectively.
In ordinary YangMills theory, to write down the Feynman rules, in addition to full
action (6), one more thing one needs to know is the representation of the gauge group,
since it determines the normalization of the group factors. In the following, we will mainly
concentrate on U (1) theory, and we will come to U (N) case naturally after the explicit
calculation for U (1) case. In noncommutative U (1) gauge theory, though the group factor
is trivial, what is nontrivial is the noncommutativity of the kernel in Fourier transform.
Suppose we have two kernels of Fourier transform F k = eikx and Fp = eipx , by using
Eq. (3), one can easily check the following commutator

 ij



pi kj Fk+p = 2i sin k p Fk+p .

(11)
[Fk , Fp ] = 2i sin
2
2
where k p ki ij pj , and we have used Eq. (1). After making Fourier transform of the
action, one can immediately find out that this commutator plays exactly the same role
as that of Lie commutators of the gauge group in ordinary non-Abelian gauge theory.
Therefore, we can establish the Feynman rules by following the same procedure as that
of ordinary YangMills theory, with the group structure constants, f abc , being replaced by
a momentum-dependent factor [2,3], namely,



abc
k,p,k+p
(12)
f
= 2i sin k p ,
f
2

where 2 is due to the normalization of T 0 . With the help of this correspondence, we

establish the following Feynman rules for U (1) NCCS theory:

G.-H. Chen, Y.-S. Wu / Nuclear Physics B 593 (2001) 562574

565

Fig. 1. Feynman rules.

(i) The gluon propagator:

(p) =

m
4
m p + p2 p p ,
2
2
2
p (p + m )

(13)

where m = e2 /4 . At the end of computations, we remove the cut-off by taking e2

or m .
(ii) The ghost propagator:
1
.
p2

(14)

(iii) The ghostghostgluon vertex:




2q sin q p .
2
(iv) The three gluon vertex:






sin p q m (r q) (q p) (p r) .
4 m
2

(15)

(16)

(v) The four-gluon vertex:

1  p,q,t r,s,t
f
f
( )
4 m


+ f r,q,t f p,s,t ( ) + f s,q,t f r,p,t ( ) ,

where p, q, r, s, t are incoming momenta and f x,y,z is given by Eq. (12).

(17)

566

G.-H. Chen, Y.-S. Wu / Nuclear Physics B 593 (2001) 562574

3. WardSlavnovTaylor identities
In ordinary gauge theories, WardSlavnovTaylor (WST) identities play a very
important role in renormalized perturbation theory. For renormalizable gauge theories,
they are essentially manifestation of gauge invariance for the regularized and renormalized
action (with counter terms included). Conversely, checking WST identities is essentially
checking renormalizability and gauge invariance of the renormalized gauge theory. The
same is true for noncommutative gauge theories. In the following we are going to check
part of the Ward identities to assure gauge invariance, and to use part of them to simplify
the calculations.
Renormalizability of the theory requires that the full inverse A-propagator and the full
AAA-vertex are of the following form as the external momenta tend to zero

0
ZA  k + ZA
k 2 k k (k 0),
(18)
1
(k)
4



(19)
2 sin p q Zg  .
2
0 and Z . In next
These equations define the relevant renormalization constants ZA , ZA
g
section, we will confirm the validity of these equations for one-loop two-point and threepoint functions, to verify renormalizability of the theory at the one-loop level. Similarly
one can define Zgh and Zg , the renormalization constants 1 for the ghost wave function
and the cAc-vertex respectively, through the full ghost propagator and the full cAc-vertex

(p)

1
Zgh p2

(p 0),




i (p, q, r) i 2 sin p q Z g p .
2

(20)
(21)

In the next section we will see that at least at one loop, these renormalization constants are
in fact finite, i.e., they are independent of the cut-off.
Assuming renormalizability and introducing the renormalized fields and renormalization
constants, one can write the renormalized action as
0
0
0
+ Igf
+ Igh
Iren = ICS

with



Z
ir
2 1 (r)
(r)
(r)
1 (r)
3

(r)
=
Tr ZA A A + Zg A A A ,
d x
4
3
Z


1
2
0
=
,
d 3 x Tr A(r)
Igf

r
Z



1 (r)
0
(r)
= d 3 x Tr Zgh
c c(r) + Z g1 c(r) A(r)
.
Igh
,c

0
ICS

(22)

(23)
(24)
(25)

1 Here we would like to remind that all the renormalization constants we defined here are consistent with the
conventions used in the Refs. [16,17,19,20], while being the inverse of the standard ones used in many textbooks
on quantum field theory.

G.-H. Chen, Y.-S. Wu / Nuclear Physics B 593 (2001) 562574

567

The terms (23), (24) and (25) in the renormalized action (22) should be equal to,
respectively, the corresponding terms (2), (9) and (10) in the original action (6), if they
are expressed in terms of the bare fields through 2
A =

ZA (r)
A ,
Zg

1/2 (r)

c = Zgh

c ,

1/2 (r)

c = Zgh

c .

(26)

This requires that the renormalized CS coupling be related to the bare one by
r =

3
ZA
,
Zg2

(27)

and that the following WST identity be true for the renormalization constants:
Zg
ZA
=
.
Zgh Z g

(28)

As easy to check, the WST identities guarantee that the renormalized actions (23) and (25)
are gauge invariant.

4. One-loop renormalization in U (1) theory

In this section, we first study at the one-loop level the renormalization of U (1) NCCS
theory, in particular, the shift in the ChernSimons coupling .
Let us start with the ghost self-energy (see Fig. 2(d)). It contains a planar diagram
contribution
Z
4 m
k 2 p2 (k p)2
d 3k
(1)

(29)
p =
p2
(2)3 k 2 (k 2 + m2 )(k + p)2
and a non-planar diagram contribution
Z
4 m
k 2 p2 (k p)2
d 3k
(1)
=
eikp .
np
2
3
2
p
(2) k (k 2 + m2 )(k + p)2

(30)

The integral (29) is finite, with the leading term proportional to 1/|m|. Therefore, in the
large |m| limit, we get the contribution to the ghost self-energy as
2
p(1) = sgn().
3
On the other hand, the non-planar diagram contribution is evaluated to be
2
sgn()f (p |m|),
3
where the function f (x) is defined by
(1)
=
np

f (x) =

1
x

Zx

(1 + x)ex 1


.
dy 3y 1/2 y 3/2 K1/2 (y) = 2
2x

(31)

(32)

(33)

0
2 The relation between A and A(r) , though looks unusual, is appropriate for the CS theory. See, e.g., Refs.

[16,17,20].

568

G.-H. Chen, Y.-S. Wu / Nuclear Physics B 593 (2001) 562574

Fig. 2. One-loop Feynman diagrams in pure ChernSimons theory (solid line gluon; dashed line
ghost: (a)(c) gluon self-energy, (d) ghost self-energy, (e)(g) three-gluon vertex, (h), (i) ghostgluon
vertex.

Since f (x) 0 as x , the non-planar diagram does not contribute to the ghost selfenergy in the limit |m| . Therefore, the ghost self-energy correction at the one loop
level is finite and independent of external momentum p. Correspondingly, we get the oneloop ghost wave function renormalization constant
(1)
=1
Zgh

2
sgn().
3

(34)

(1)
(p), we decompose it into the following structure
To calculate the gluon self energy
(1)
(p) =

1 (1)
(1)
e p2 p p +
(p) p .
m
4 o

(35)

Since only gluon loop diagram allows an odd number of tensors, o(1) (p) receives a
(1)
nonzero contribution only from the gluon loop diagram Fig. 2(a). In contrast, e picks
up contributions from the tadpole diagram (Fig. 2(c)), the ghost loop diagram Fig. 2(b), as
well as from the gluon diagram Fig. 2(a) with an even number of tensors. Contracting
(1)
with /4(p /2p2 ) and m /2p2 , we obtain o(1) (p) and e(1) (p):

G.-H. Chen, Y.-S. Wu / Nuclear Physics B 593 (2001) 562574

o(1) =

4 2m
p2

and
e(1)

m
= 2
2p

"Z


d 3k  2
sin (/2k p)
3
(2)
[k 2 p2 (k p)2 ][5k 2 + 5(k p) + 4p2 + 2m2 ]

k 2 (k + p)2 (k 2 + m2 )[(k + p)2 + m2 ]

569

(36)


d 3k  2
sin (/2k p)
3
(2)

#
5m
Ne (p, k)
,
+
2
k (k + p)2 (k 2 + m2 )[(k + p)2 + m2 ] 3

(37)

where
Ne (p, k) = 6k 6 + 18k 4 (k p) + 20k 4p2 + 22k 2(k p)2 p2 12(k p)3 + 9k 2 p4


7(k p)2 p2 + m2 2k 4 + 4k 2 (k p) + k 2 p2 + (k p)2 .
(38)
At this point we would like to comment that the structures shown in the above results are
similar to those in ordinary non-Abelian ChernSimons gauge theory in 2 + 1 dimensions,
although here we are dealing with the U (1) case. Still they are different in the following
two aspects. The first is that we have non-planar diagram contributions due to the
oscillating factor 4 sin2 (/2k p). The second is that the value of the tadpole contribution
changes (see the second term in (37)), the reason being that one of the terms in the fourgluon vertex vanishes due to the fact sin(/2p p) = 0 by using the Feynman rule (17).
The integral in Eq. (36) is finite. To calculate it, again we separate it into planar and
non-planar contributions. The calculation of the planar contribution is standard. Taking
|m| , we obtain
7
sgn().
(39)
3
By using Feynman parameterization, we can rewrite the corresponding non-planar
contribution as follows:
(1)
=
o,p

Z1
(1)
o,np

= 2p

Zx
dx

Zy
dy

dz
0





5I2 () + 5p2 (1 + y x z)(y x z) + 4p2 + 2m2 I1 () , (40)
where the argument in the functions I1 and I2 is defined as
q
= p2 (1 + y x z) + m2 (1 y).
The functions I1 and I2 are defined as
  

3/2 p 3/2 3/2
p 1/2

K3/2 (p)
K1/2 (p)

I1 =
2
2
3
and

(41)

(42)

570

G.-H. Chen, Y.-S. Wu / Nuclear Physics B 593 (2001) 562574

  
 3/2
3/2 p 1/2
p
1/2
I2 =
K1/2 (p) 20
1/2 K1/2 (p)
15
12
2
2

 5/2
p
3/2 K3/2 (p) ,
+4
2

(43)

where K (x) is the modified Bessel function. It has an exponentially decay profile. In the
limit |m| , we see that the integrand in Eq. (40) vanishes, therefore the non-planar
(1)
diagram does not contribute to o . Namely,
(1)
= 0.
o,np

(44)

Thus, we get
7
sgn().
(45)
3
Similar analysis can be applied to the integral (37). It turns out that the integral is finite
as |m| . Therefore, the photon wave function renormalization constant is
o(1) =

7
sgn().
(46)
3
Furthermore, we study the one loop corrections to the vertex cAc, we show that the one
loop correction vanishes as |m| . The reason is that the one- term of Fig. 2(i) cancels
against the three- term of Fig. 2(h); the one- term of Fig. 2(h) goes to zero; the non-
terms in the two diagrams cancel each other as well. Finally, we get the renormalization
for the cAc vertex
(1)

ZA = 1 + o(1) = 1 +

Z g(1) = 1.

(47)

(1)
(1)
, ZA
, and Z g(1) , we can employ the WST identity (28) established
After we extract Zgh
in the previous section to get the three-gluon vertex renormalization constant:
(1)

Zg(1) =

ZA

(1)
Zgh

3
Z g(1) = 1 + sgn(),

(48)

where we have worked up to the first order in 1/, consistent with one-loop perturbative
theory.
Now we have shown that all renormalization constants at the one loop level are
finite, so the one-loop beta function vanishes, as in ordinary CS theory. Substituting the
(1)
(1)
renormalization constants ZA and Zg in the definition of the renormalized CS coupling
r(1) , we have
r(1) = + sgn().

(49)

This is the main result of the present paper. The second term is the desired one-loop shift in
the U (1) ChernSimons coupling. Note that the shift is just the unity in our normalization
for the coupling. Note that it is independent both of the noncommutativity parameters
and of the value of the bare coupling except for its sign. Also recall that for ordinary
U (1) CS theory, the one-loop shift vanishes in the same F 2 regularization.

G.-H. Chen, Y.-S. Wu / Nuclear Physics B 593 (2001) 562574

571

5. Generalization to U (N )
In this section we generalize the expression (49) we obtained in last section for the oneloop shift of the CS coupling from U (1) to U (N).
It is known that for a noncommutative gauge theory, the gauge group is restricted to be
only U (N); even SU(N) is not allowed, because the closure of the Moyal commutator is
violated [21]. Another way to see this is that the three-gluon coupling in the action (2)
mixes the U (1) gluon with the SU(N) gluons. Thus, unlike ordinary gauge theory
which allows the U (1) and SU(N) sectors to have independent coupling constants, in
noncommutative theory U (N) gauge invariance enforces the U (1) and the SU(N) gluons
to share the same coupling constant.
For U (N) noncommutative YangMills (NCYM) theory, it is this distinct feature that
makes the coupling in the U (1) sector runs in the same way as that in the SU(N) sector, as
verified in a recent explicit calculation [11]. Before this calculation was done, a beautiful
proof without doing any new calculation had been given in Ref. [5] for the statement that
the one-loop beta-function of U (N) NCYM can be simply read off from the known value
of the ordinary SU(N) YangMills theory. This is because in NCYM the nonplanar oneloop U (N) diagram contributes only to the U (1) part of the theory. In the following we
will apply the same trick to NCCS, and derive the one-loop shift in the CS coupling in the
U (N) theory without doing any new calculations.
Following Ref. [5], let us consider the quadratic one-loop 1PI effective action of the
ordinary U (N) CS theory, in which the U (1) and SU(N) sectors share the same coupling
constant, in the F 2 regularization. From the results in Refs. [17,20] one infers that after
removing the cut-off,
Z

i
(1)
d 3 x ( + Nsgn()) Tr(A A )
2 ( = 0) =
4

(50)
sgn()(Tr A ) (Tr A ) .
The coefficient + Nsgn() in the first term is read off from the known one-loop shift in
ordinary SU(N) CS theory [17,20], while the existence of the second term is due to the
necessity for cancelling the U (1) part in the first term, since we know there is no oneloop shift in the U (1) coupling constant. Also it is easy to check that the second term is
the only nonplanar contribution to 2(1) , coming from the nonplanar part of diagrams like
Fig. 2(a)(c).
Now let us turn on nonzero . The planar contributions to 2(1) are known to be the
same as in ordinary theory [4,5], while the nonplanar diagrams are suppressed to zero by
an extra rapidly oscillating phase factor. Our result (44) in last section verifies the latter by
explicit computation. Thus, with the second term put to zero, one reads from Eq. (50) that
in U (N) NCCS,
Z


i
(1)
(51)
d 3 x + Nsgn() Tr(A A ).
2 ( 6= 0) =
4
Therefore, without any new calculation, we infer that the one-loop shift in the coupling for
U (N) NCCS is

572

G.-H. Chen, Y.-S. Wu / Nuclear Physics B 593 (2001) 562574

r = + Nsgn().

(52)

For the U (1) case, we plug N = 1 in Eq. (52), reproducing exactly Eq. (49) that we have
obtained by explicit calculation in last section.
In summary, the one-loop shift of the CS coupling in U (N) NCCS is the same as that
in ordinary SU(N) CS theory for N > 2, while for the U (1) case it gives a non-vanishing
value in contrast to ordinary CS theory.

6. Conclusions and discussions

The induced CS coupling by fermionic fields in noncommutative quantum electrodynamics in 3 dimensions has been studied in Ref. [9]. In this paper we have considered
instead a pure CS theory without matter, and have studied the one-loop quantum correction to the coupling constant due to self-interactions of the gauge bosons that arise from
spacetime coordinate noncommutativity.
First of all, the renormalization constants we have calculated at the one loop level are
finite, showing that the beta function at this level vanishes. Moreover, all renormalization
constants, including the one-loop shift in the CS coupling are shown to be independent
of the noncommutativity parameters. This is a bit surprising, since the spacetime
noncommutativity parameters appear in the Lagrangian of pure CS theory explicitly.
This adds explicit evidence to a theorem proved in Ref. [10] that one-loop results in
noncommutative CS theory are all independent of spacetime noncommutativity. It would
be interesting to see whether the same independence remains true at higher orders.
Our conjectured answer is yes. In other words, we conjecture that noncommutative
CS theory is a deformed topological field theory, in the sense that the partition function
and correlation functions are topological invariants, independent of both metric and
noncommutativity parameters.
Furthermore, we have shown that the one-loop shift in the coupling constant does not
vanish in a pure U (1) NCCS theory. This arises as a consequence of spacetime coordinate
noncommutativity, since in ordinary U (1) CS theory there is no quantum shift in the
coupling constant at all. We notice two features of our result (49): (1) it is independent of
the bare coupling , except for its sign; (2) it is the simplest integer, the unity, independent
of the noncommutativity parameters. Therefore, this result is not smooth in the limit
0.
Finally, we have shown that the one-loop shift of the U (N) NCCS coupling is the
integer N that characterizes the gauge group U (N), exactly the same as that in ordinary
SU(N) CS theory for N > 2. In ordinary CS theory the quantization of the shift was
interpreted [17,20] as being consistent with the topological quantization of the non-abelian
CS coupling. The latter is known [14] to result from the topological fact that a large
gauge transformation changes the CS action by a value proportional to the integer winding
number of the gauge transformation, viewed as a map from the (compactified) spacetime
to the gauge group. Whether the topological quantization of the CS coupling remains true
in the noncommutative theory is not clear at all at this moment. However, our result (52)

G.-H. Chen, Y.-S. Wu / Nuclear Physics B 593 (2001) 562574

573

shows that the one-loop shift is still quantized in the noncommutative case. If one reverses
the logic in the above reasoning for the ordinary CS theory, this result seems to indicate
that possibly in noncommutative geometry there should be a counterpart of the concept of
the usual winding number that remains integer-valued, and that the NCCS coupling should
satisfy a similar topological quantization.
Here we would like to mention that the one-loop shift in the U (1) NCCS coupling
depends on the regularization used. Our result (49) was obtained in the F 2 regularization,
in which the YangMills term was added to the action and its coefficient was taken as cutoff. If we had used dimensional regularization, the shift would be zero. This situation is
not surprising, completely similar to the well-known situations for ordinary non-Abelian
CS theory; see, e.g., Refs. [17,20]. In our opinion, the F 2 regularization is more physical,
in the sense that the F 2 term may naturally appear in realistic planar systems.
In field theory on ordinary spacetime, the U (1) CS coupling (CS coefficient) is known
to have several interesting physical meanings, when the CS gauge field couples to
various fields. For example, when there is a Maxwell term in the action, is related
to the topological mass of the CS photon [14]. When there are matter fields coupled
to the CS field, the CS coefficient will give rise to fractional (exchange) statistics
for matter field quanta [22]. Finally a well-known folklore in the community is that
an effective CS coupling for the electromagnetic field indicates the Hall effect, with
the effective CS coefficient directly related to the Hall conductance. We expect all
these physical interpretations should be generalizable to noncommutative spacetime. A
systematic inverstigation of NCCS coupled to matter fields (both fermionic and bosonic)
will be published elsewhere [23].

Note added
After the paper was put on the e-print archive, we are informed by e-mail from Soo-jong
Rey that he has obtained the same result (in agreement with ours) on U (1) NCCS theory.

Acknowledgements
One of us, G.H.C., thanks the Institute for Theoretical Physics, University of California
at Santa Barbara, for an ITP Graduate Fellowship, and for the warm hospitality he receives
during his stay. This research was supported in part by the National Science Foundation
under Grants No. PHY-9907194 and PHY-9970701.

References
[1] T. Filk, Divergences in a field theory on quantum space, Phys. Lett. B 376 (1996).
[2] T. Krajewski, R. Wulkenhaar, Perturbative quantum gauge fields on the noncommutative torus,
hep-th/9903187.

574

G.-H. Chen, Y.-S. Wu / Nuclear Physics B 593 (2001) 562574

[3] C.P. Martin, D. Sanchez-Ruiz, The one-loop UV divergent structure of U (1) YangMills theory
on noncommutative R 4 , Phys. Rev. Lett. 83 (1999) 476479, hep-th/9903077.
[4] D. Bigatti, L. Susskind, Magnetic fields, branes, and noncommutative geometry, hepth/9908056.
[5] S. Minwalla, M.V. Raamsdonk, N. Seiberg, Noncommutative perturbative dynamics, hepth/9912072;
M.V. Raamsdonk, N. Seiberg, Comments on noncommutative perturbative dynamics, hepth/0002186.
[6] M. Hayakawa, Perturbative analysis on infrared and ultraviolet aspects of noncommutative
QED on R 4 , hep-th/9912167, hep-th/9912094.
[7] A. Matusis, L. Susskind, N. Toumbas, The UV/IR connection in noncommutative gauge
theories, hep-th/0002075.
[8] J. Gomis, K. Landsteiner, E. Lopez, Nonrelativistic and noncommutative field theory and
UV/IR mixing, hep-th/0004115.
[9] Chong-Sun Chu, Induced ChernSimons and WZW action in noncommutative spacetime,
hep-th/0003007.
[10] A.A. Bichl, J.M. Grimstrup, V. Putz, M. Schweda, Perturbative ChernSimons theory on
noncommutative R 3 , hep-th/0004071.
[11] A. Armoni, Comments on perturbative dynamics of non-commutative YangMill theory, hepth/0005208.
[12] P.M.Ho, Y.S. Wu, Noncommutative geometry and D-branes, Phys. Lett. B 398 (1997) 52, hepth/9611233;
M. Li, Strings from IIB matrices, Nucl. Phys. B 499 (1997) 149158, hep-th/9612222;
A. Connes, M.R. Douglas, A. Schwarz, Noncommutative geometry and matrix theory:
compactification on tori, JHEP 9802 (1998) 003, hep-th/9711162;
M.R. Douglas, C. Hull, D-branes and the noncommutative torus, JHEP 9802 (1998) 008, hepth/9711165;
P.M. Ho, Y.S. Wu, Noncommutative gauge theories in matrix theory, Phys. Rev. D 58 (1998)
026006, hep-th/9712201.
[13] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 09 (1999) 032, hepth/9908142, and references therein.
[14] W. Siegel, Nucl. Phys. B 156 (1979) 135;
J.F. Schonfled, Nucl. Phys. B 185 (1981) 157;
R. Jackiw, S. Templeton, Phys. Rev. D 23 (1981) 2291.
[15] E. Witten, Commun. Math. Phys. 121 (1989) 351.
[16] S. Deser, R. Jackiw, S. Templeton, Phys. Rev. Lett. 48 (1982) 975;
S. Deser, R. Jackiw, S. Templeton, Ann. Phys. (N.Y.) 140 (1982) 372;
R.D. Pisarski, S. Rao, Phys. Rev. D 32 (1985) 2081.
[17] W. Chen, G.W. Semenoff, Y.S. Wu, Mod. Phys. Lett. A 5 (1990) 1833.
[18] G.W. Semenoff, P. Sodano, Y.S. Wu, Phys. Rev. Lett. 62 (1989) 715.
[19] W. Chen, G.W. Semenoff, Y.S. Wu, Phys. Rev. D 44 (1991) R1625.
[20] W. Chen, G.W. Semenoff, Y.S. Wu, Phys. Rev. D 46 (1992) 5521.
[21] K. Matsubara, Restrictions on gauge groups in noncommutative gauge theories, hepth/0003294.
[22] See papers collected in the reprint volume, in: F. Wilczek (Ed.), Fractional Statistics and Anyon
Superconductivity, World Scientific, 1990.
[23] G.-H. Chen, Y.-S. Wu, Renormalization of the statistics parameters in three-dimensional
noncommutative ChernSimons electrodynamics, in preparation.

Nuclear Physics B 593 (2001) 577595

www.elsevier.nl/locate/npe

The anomalous process to two loops

Torben Hannah
Nordita, Blegdamsvej 17, DK-2100 Copenhagen , Denmark
Received 18 April 2000; accepted 25 September 2000

Abstract
The amplitude for the anomalous process is evaluated to two loops in the chiral
expansion by means of a dispersive method. The two new coupling constants that enter at this order
are estimated via sum rules derived from a non-perturbative chiral approach. With these coupling
constants fixed, the numerical results are given and compared with the available experimental
information. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.30.Rd; 11.55.Fv; 11.55.Hx; 13.60.Le
Keywords: Chiral perturbation theory; Dispersion relations; Sum rules; Chiral anomaly; Pion production

1. Introduction
At low energies, the strong interaction between pions can be described by the effective
field theory of QCD called chiral perturbation theory (ChPT) [14]. In this effective field
theory the interaction between pions is analyzed in terms of a systematic expansion in
powers of the low external momenta and the small pion mass. This chiral expansion is
equivalent to an expansion in the number of loops and works very well for processes
involving pions [5].
In the normal intrinsic parity sector, the chiral expansion starts at O(p2 ) and is obtained
from the leading order chiral Lagrangian. In this sector the one-loop or O(p4 ) corrections
were extensively treated in the works by Gasser and Leutwyler [2,3], and many processes
have now been calculated to this order. The further extension to two loops of O(p6 ) has
also been undertaken with several two-flavour calculations already published in the normal
sector [610].
For the abnormal intrinsic parity sector, the chiral expansion starts at O(p4 ) and is
obtained from the WessZuminoWitten effective action [11]. In this sector the one-loop
corrections of O(p6 ) have also been analyzed [12], and some anomalous processes have
E-mail address: [email protected] (T. Hannah).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 4 1 - 6

578

T. Hannah / Nuclear Physics B 593 (2001) 577595

been calculated to this order [13]. However, contrary to the case in the normal sector, no
two-loop calculations of O(p8 ) have been published for processes in the anomalous sector.
In this paper the anomalous process is calculated to two loops using a
dispersive method. This process is important for the theory of chiral anomalies and has
previously been calculated to both leading [14] and one-loop [15] order in the chiral
expansion. However, both of these results are somewhat below the present experimental
data [16]. This has motivated new and more precise experiments, which will be performed
at different facilities such as CERN [17], FNAL [18], and CEBAF [19].
It is therefore important to evaluate the two-loop corrections and compare the result with
the experimental information. This is the purpose of the present paper, which is organized
as follows. In Section 2, the notation and kinematics of the process are given,
whereas previous ChPT results are reviewed in Section 3. The general dispersive formula
is derived in Section 4, together with the explicit calculation of the two-loop corrections.
In Section 5, this two-loop result is used in a non-perturbative chiral approach in order to
estimate the two new coupling constants from sum rules. The numerical results are given
in Section 6, including the comparison with experiments, and Section 7 consists of a short
conclusion.

2. Kinematics
The amplitude for the anomalous process
(q) (p1 ) (p2 ) 0 (p0 )

(1)

is given in terms of the scalar function F3 (s, t, u) as

A = iF3 (s, t, u)  p1 p2 p0 ,

(2)

where s = (p1 + q)2 , t = (p1 p2 )2 , and u = (p1 p0 )2 are the Mandelstam variables.
In the following it is assumed that all particles are real. In this case s + t + u = 3M2 , so
that in the center of mass system one has
i

q
1h 2
3M s + s M2 1 4M2 /s cos ,
t =
2
i

q
1h 2
(3)
3M s s M2 1 4M2 /s cos ,
u=
2
with the center of mass scattering angle. Because of isospin symmetry, the other
reactions are given by the same scalar function F3 (s, t, u), which is fully
symmetrical in its arguments. The total cross section is obtained from the expression


1 (s 4M2 )3/2
s M2
(s) =
1/2
1024
s


2
d sin3 F3 (s, t, u) ,

(4)

and the partial wave expansion of the scalar function F3 (s, t, u) is of the form
X
fl (s)Pl0 (cos ).
F3 (s, t, u) =
odd l

(5)

T. Hannah / Nuclear Physics B 593 (2001) 577595

The projection onto the lowest P partial wave is given by

Z
3
d sin2 F3 (s, t, u),
f1 (s) =
8

579

(6)

and the higher partial waves can be projected out in similar ways. This partial wave
expansion contains only odd l since the pions in the final state have the total isospin
I = 1. In the unitarity relation, the complete set of intermediate states are the same as for
scattering. Within the elastic approximation containing only the two-pion intermediate
state, the unitarity relation is
(7)
Im fl (s) = (s)fl (s)tl1 (s),
p
with (s) = (s 4M2 )/s the phase-space factor and tl1 the partial wave with isospin
I = 1 and angular momentum l. This unitarity relation implies that the phase of fl will
coincide with the phase shift l1 in accordance with Watsons final-state theorem [20].
S states remain
Inelasticities due to other intermediate states such as the four-pion or the K K
very small below 1 GeV and are completely negligible at low energies.

3. Chiral expansion
Anomalous processes start at O(p4 ) in the chiral expansion and are obtained from
the WessZuminoWitten effective action at tree level. With only the electromagnetic
interaction as external fields, these anomalous processes contain exclusively an odd
number of pseudoscalars, i.e., they have abnormal or odd intrinsic parity. For the process
, the leading order chiral result for the scalar function F3 (s, t, u) is given
by [11,14]
(0)
=
F3 (s, t, u) = F3

eNc
,
12 2 F3

(8)

where Nc = 3 is the number of colors and F = 92.4 MeV is the pion decay constant.
Another related anomalous process is the decay 0 , where the F coupling
obtained at leading order is of the form [11,21]
(0)
F
=

Nc
.
3F

(9)

These two results are exact in the soft pion and soft photon limit where they are related
to each other by the low-energy theorem F3 = F /eF2 . For the decay 0 , the
prediction from Eq. (9) is in excellent agreement with the experimental information, thus
providing important evidence for the three-color nature of the strong interaction. However,
for the process , the corresponding prediction F3 = 9.7 GeV3 is somewhat
lower than the experimental measurement F3 = 12.9 0.9 0.5 GeV3 [16], which
could even be said to favor the value Nc = 4.
It is therefore of importance to calculate the corrections to the soft pion and soft photon
limit. Within ChPT these corrections can indeed be calculated in a systematic manner. The
first correction is the one-loop contribution of O(p6 ), which has previously been calculated

580

T. Hannah / Nuclear Physics B 593 (2001) 577595

for both processes. In the case of the decay 0 , these one-loop corrections turn out
to be very small [22], i.e., they do not spoil the excellent agreement with the experimental
information. On the other hand, for the process , the one-loop contributions are
larger and they will therefore be of importance when comparing with the experimental
measurement. The expression for the scalar function F3 (s, t, u) to this order in the chiral
expansion is [15]


64 2 r
(0)
C2 ()(s + t + u) + Cl ,
1
(10)
F3 (s, t, u) = F3
3e
where the term Cl contains the contributions from the loops and is given by
Cl =



1
5
M2
(s
+
t
+
u)
+
(s
+
t
+
u)
log
3
96 2 F2
2


1
F (s) + F (t) + F (u) ,
+
24 2 F2

(11)

with F (x) expressed in terms of the standard one-loop function J(x) as

1
F (x) = 4 2 x 4M2 J(x) x.
2

(12)
(0)

The expression (10) converges to the chiral anomaly F3 in the soft pion and soft
photon limit and is given in terms of the renormalized low-energy constant C2r from
the O(p6 ) anomalous chiral Lagrangian [12]. This low-energy constant depends on
the renormalization scale and is needed to absorb divergences in the one-loop
calculation. In principle, C2r should be determined phenomenologically, preferably from
other observables, but at present this appears to be rather out of reach. Therefore, this lowenergy constant has been estimated using the assumption of vector resonance saturation,
which is known to work well for the O(p4 ) non-anomalous low-energy constants [23].
Assuming that the same is the case for the O(p6 ) anomalous low-energy constants, one
has [15]
C2r () =

3e
,
128 2M2

(13)

with the renormalization scale typical chosen at the resonance scale = M = 770 MeV.
Having fixed the value of this low-energy constant, it is possible to obtain the prediction
for the one-loop expression (10). This improves the agreement with the experimental
measurement [16]. However, the prediction is still on the lower side of the data [15,17,18].
Therefore, it is important with both theoretical and experimental improvements.
Indeed, significant improvements on the experimental side are expected from several new
experiments [1719]. A similar improvement on the theoretical side would involve the
calculation of the two-loop corrections. This could be done by a full field theory calculation
to two loops. However, these two-loop corrections can also be calculated by a dispersive
method, which will be the subject of the next section.

T. Hannah / Nuclear Physics B 593 (2001) 577595

581

4. Dispersive representation
4.1. Derivation of the dispersive formula
The anomalous process is in many ways similar to the 0 0 0 0
scattering process. They are both described in terms of a single scalar function, which is
fully symmetrical in the s, t, and u variables. The scattering process can be described
by Roy equations [24] derived from the fundamental principles of analyticity, crossing, and
unitarity. When these Roy equations are combined with the chiral expansion, the general
structure of the scattering amplitude can be obtained from a dispersive representation
to two loops in the chiral expansion [8,25].
The same is also the case for the process . In order to show this, one starts
with a fixed-t dispersion relation for the scalar function F3 (s, t, u). Fixed-t dispersion
relations with two subtractions are known to exist for scattering due to the Froissart
bound. Assuming that the same is also the case for the process , one has the
fixed-t dispersion relation

 2
Z
u2
1
Im F3 (s 0 , t)
s
+
,
(14)
ds 0
F3 (s, t) = C(t) +

s0 s s0 u
s 02
4M2

where u = 3M2 s t. The s u symmetry of F3 (s, t, u) implies that there is no

subtraction constant linear in s. Therefore, the only subtraction constant is C(t), where
the t dependence can be obtained from the s t symmetry F3 (0, t) = F3 (t, 0). This
implies that C(t) can be written as
1
C(t) = C(0) +

Z
4M2

Z
4M2

ds 0

ds 0


 2
(3M2 t)2
Im F3 (s 0 , 0)
t
+
s 02
s 0 t s 0 3M2 + t

Im F3 (s 0 , t) (3M2 t)2
,
s 02
s 0 3M2 + t

(15)

which can be inserted in Eq. (14). By construction, the resulting dispersion relation exhibits
s u symmetry for fixed t, whereas s t symmetry for fixed u is not manifest. In order
to impose this latter symmetry one can expand the absorptive part of F3 (s 0 , t) in partial
waves using Eq. (5). Writing this expansion in the form
Im F3 (s 0 , t) = Im f1 (s 0 ) + Im (s 0 , t),

(16)

> 3 are contained in the function Im (s 0 , t), which is given

the higher partial waves with l

by
X
Im fl (s 0 )Pl0 (cos ).
Im (s 0 , t) =

(17)

l>3

With this decomposition of the partial waves, it is possible to rewrite Eq. (14) with C(t)
given by Eq. (15) as

582

T. Hannah / Nuclear Physics B 593 (2001) 577595

b3 (s, t, u) + 3 (s, t, u),

F3 (s, t, u) = F

(18)

b3 (s, t, u) and 3 (s, t, u) are given by

where F
Z

b3 (s, t, u) = C(0) + 1
F

4M2

3 (s, t, u) =

ds 0

4M2

1
+

Z
4M2

ds 0


 2
t2
u2
Im f1 (s 0 )
s
+
+
,
s 02
s0 s s0 t s0 u


 2
u2
(3M2 t)2
Im (s 0 , t)
s
+

s 02
s 0 s s 0 u s 0 3M2 + t

ds 0


 2
(3M2 t)2
Im (s 0 , 0)
t
+
.
s 0 t s 0 3M2 + t
s 02

(19)

This is the Roy equation for the anomalous process . One observes that
b3 (s, t, u) is now fully symmetrical in s, t, and u, whereas this symmetry is not manifest
F
in 3 (s, t, u). However, at low energies, the absorptive part of the higher partial waves
with l > 3 is negligible. This implies that in practice 3 (s, t, u) can be treated in a simple
way as a small, real correction at low energies. This fact makes the corresponding Roy
equations for scattering very useful.
In ChPT the absorptive part of the higher partial waves is indeed suppressed. Within this
methodology, these higher partial waves only start at O(p6 ) in the chiral expansion. Since
the corresponding partial waves start at O(p4 ), perturbative unitarity implies that the
absorptive part of the higher partial waves with l > 3 is of O(p10 ) or higher in the chiral
expansion. Thus, up to this order, the term 3 (s, t, u) in Eq. (18) vanishes, and the full
b3 (s, t, u). In order for the full amplitude to
amplitude is given entirely by the function F
formally satisfy the chiral anomaly in the soft pion and soft photon limit, one may write
(0)
S + t + u)], where s + t + u = 3M2 . Thus,
the subtraction constant as C(0) = F3 [1 + C(s
assuming for the moment that two subtractions are indeed sufficient, the full amplitude to
O(p8 ) in the chiral expansion may be written as

(0) 
S + t + u)
1 + C(s
F3 (s, t, u) = F3

 2
Z
t2
u2
Im f1 (s 0 )
s
1
+
+
ds 0
.
(20)
+

s 02
s0 s s0 t s0 u
4M2

In this dispersive representation, the absorptive part of the lowest partial wave can be
determined from unitarity. In the general unitarity relation, the 2n-pion invariant phase
space is of O(p4n4 ), the amplitude for multi-pion scattering is dominantly of O(p2 ),
and the amplitude for multi-pion photo-production is at least of O(p4 ). Consequently,
intermediate states containing more than two pions are suppressed at least up to O(p10 )
in the chiral expansion. Therefore, within SU(2) ChPT, the absorptive part of the lowest
partial wave is given by elastic unitarity to O(p8 ) as
Im f1(0)(s) = 0,

T. Hannah / Nuclear Physics B 593 (2001) 577595

Im f1(1)(s) = (s)f1(0)(s)t11(0) (s),

 (0)

(2)
1(1)
(1)
1(0)
Im f1 (s) = (s) f1 (s) Re t1 (s) + Re f1 (s)t1 (s) ,
(n)

583

(21)

1(n)

the corresponding partial

where f1 is the P partial wave of O(p2n+4 ) and t1
2n+2
S
) [2]. In SU(3) ChPT, one must also include the inelasticity from the K K
wave of O(p
intermediate state. However, since this inelasticity is completely negligible at low energies,
only SU(2) ChPT will be considered in the following.
4.2. The amplitude to two loops
From the dispersive representation (20), one obtains straightforwardly the one-loop
(1)
S
ChPT formula (10) by setting Im f1 = Im f1 . In this case the subtraction constant C
r
can be expressed in terms of the low-energy constant C2 and chiral logarithms. However,
in order to calculate the amplitude to two loops, it is necessary to use one subtraction more
(2)
in the dispersive representation. This is due to the fact that the absorptive part of f1
2
behaves as s modulo log factors. Thus, the amplitude to two loops can be obtained from
the following dispersion relation:


(0) 
S + t + u) + D
S s 2 + t 2 + u2
F3 (s, t, u) = F3 1 + C(s

 3
Z
t3
u3
Im f1 (s 0 )
s
1
+
+
,
(22)
ds 0
+

s 03
s0 s s0 t s0 u
4M2

S and D
S are the two subtraction constants and Im f1 is given by Im f1 = Im f (1) +
where C
1
(2)
(1)
(2)
Im f1 with Im f1 and Im f1 determined from Eq. (21). This dispersion relation can
be evaluated with the same methodology as has been applied in the calculation of the
scattering amplitude [8] and the pion form factors [6] to two loops. With the use of this
methodology, the result can be written as


(0) 
S + t + u) + D
S s 2 + t 2 + u2 + U A (s) + U A (t)
F3 (s, t, u) = F3 1 + C(s

(23)
+ U A (u) + U (s) + U (t) + U (u) ,
where the U A term can be expressed in the compact analytic form:


2
2 

M2
x 
M2
2 2
(x) x

(x)
J
1
+
24
+
U A (x) =
16 2 F2 9M2
60M4
16 2 F2



x2 
6M2 l4 9M2 c1
2 2
(x)
+
l2 l1 +

(x)
J
1
+
24
x
x
27M4
x2
3191 x 2
223 x
16
x2

+
+

l
4
1
4
4
4
2
6480 M
216 M
9
30M
20M
 2



2
2
x
4
x
x
x
7
+
15
+

151
+
99
37
J(x)

27
540M2
M2
M4
M2


x2
x
2 2 M2 x 3
S1 (x)
30 4 + 78 2 128 K
+
9x
M6
M
M

584

T. Hannah / Nuclear Physics B 593 (2001) 577595



13 x
x2
S4 (x) .

2
(24)
K
3 M2
M4
S4 (x) are given in Ref. [8], where
S1 (x) and K
The explicit expressions for the functions K
these functions were introduced in the evaluation of the scattering amplitude to two
loops. They are analytic functions with cuts starting at the threshold and can be
expressed in terms of the standard one-loop function J(x). As for the other term U ,
this part does not have a compact analytic representation in terms of elementary functions.
However, it can be obtained numerically from the dispersive representation:
Z
0
0 1(0) 0
x3

0 (s ) Re f1 (s )t1 (s )
.
(25)
ds
U (x) =

s 03 (s 0 x)


+ 8 2

4M2

1(0)

In this dispersive formula t1 is the lowest order P partial wave and f1 is given by


Z


3
1
5
(t
+
u)
+
4
F
(t)
+
F
(u)
,
(26)
d sin2
f1 (s) =
8
96 2 F2 3
with t and u determined from Eq. (3) and F (x) given by Eq. (12). The size of the
corrections from U are very small at low energies and they can therefore in practice
be neglected at these energies.
In the dispersive representation, Eq. (23), the full amplitude to two loops is determined
S and D.
S These subtraction constants may be parameterized
up to the subtraction constants C
in terms of the coupling constants c1 , c2 , and d2 as



1
M2
1
S
c2 ,
+
c1
C =
16 2 F2
6
16 2 F2


1
1
1
2 ,
S=
d
+
(27)
D
16 2 F2 60M2
16 2 F2
where the coupling constant c1 can be expressed in terms of the O(p6 ) anomalous lowenergy constant C2r and chiral logarithms:


1
M2
64 2 r
2 2
C ()
log 2 .
c1 = 16 F
(28)
3e 2
96 2 F2

The coupling constants c2 and d2 enter at two-loop order in the chiral expansion and
contain contributions both from two-loop diagrams and from the unknown renormalized
low-energy constants, which parameterize the O(p8 ) anomalous chiral Lagrangian.

5. Non-perturbative chiral approach

5.1. The lowest partial wave
From the two-loop expression for the amplitude, Eq. (23), one may project out the lowest
P partial wave with the use of Eq. (6). This gives the expansion
f1 (s) = f1(0)(s) + f1(1) (s) + f1(2) (s),

(29)

T. Hannah / Nuclear Physics B 593 (2001) 577595

585

which will satisfy the perturbative unitarity relations (21). These relations work very
well at low energies, whereas the deviation from exact unitarity, Eq. (7), becomes more
pronounced as the energy is increased. At these energies, still higher order unitarity
corrections will start to be of importance in the chiral expansion. This is particularly the
case in the (770) resonance region, where unitarity corrections are essential.
In the presence of the (770) resonance, unitarity will therefore be of the utmost
importance. There are different ways to combine exact unitarity and the chiral expansion
in order to try to account for this resonance. One such method, which has been
successfully applied to many different processes, is the so-called non-perturbative inverse
amplitude method (IAM) [2629]. Since this is a rather general method, it can also be
straightforwardly applied to the present case. The starting point for the IAM is to write
down a dispersion relation for the inverse of the partial wave f1 . In this dispersion relation,
exact unitarity and the chiral expansion are used to compute the important right cut,
whereas the left cut and the subtraction constants are approximated by ChPT. If one writes
down a similar dispersion relation for the chiral expansion using perturbative unitarity on
the right cut, it is possible to express the result of the IAM in a simple way in terms of the
chiral partial waves. With the two-loop ChPT expansion (29), the result of the IAM can be
written as
2

f1 (s) =

f1(0) (s)
(0)

(1)

(1) 2

f1 (s) f1 (s) + f1

(0)

(2)

(s)/f1 (s) f1 (s)

(30)

This is formally equivalent to the [0, 2] Pad approximant applied to ChPT and will
therefore coincide with the chiral expansion up to two loops. However, since exact unitarity
was used in the derivation of the IAM, it is expected that this result will improve ChPT at
higher energies. The IAM applied to two-loop ChPT has been extensively discussed in
Ref. [28], where the detailed derivation of the general result can also be found.
The result of the IAM, Eq. (30), depends on the pion mass and pion decay constant,
which will be set equal to M = 139.6 MeV and F = 92.4 MeV, respectively.
Furthermore, the IAM is also given in terms of the low-energy constants l2 l1 and l4 .
These low-energy constants appear in the ChPT P partial wave to one loop order and
can been determined phenomenologically from other sources. However, since the IAM
contains higher order unitarity corrections, the phenomenological values of these lowenergy constants in the IAM do not necessarily coincide precisely with the values obtained
in ChPT. For the combination l2 l1 , this has been determined by applying the IAM to oneloop ChPT in the case of scattering with the result l2 l1 = 5.8 [28]. The other lowenergy constant l4 has been determined in ChPT from the pion scalar form factor with the
central value l4 = 4.4 [7]. Since the IAM (30) does not depend much on the precise value
of this low-energy constant, the same value can also be applied in the IAM. Finally, this
result also depends on the one-loop coupling constant c1 . This coupling constant is related
to the low-energy constant C2r , which has been estimated from the assumption of vector
resonance saturation, Eq. (13) [15]. With the value of C2r chosen at the renormalization
scale = M , one finds c1 = 1.71. Therefore, in the IAM (30), the values
l4 = 4.4,
c1 = 1.71,
(31)
l2 l1 = 5.8,

586

T. Hannah / Nuclear Physics B 593 (2001) 577595

Fig. 1. The phase of the lowest partial wave f1 obtained from the IAM, Eq. (30), compared to
the experimental phase shift 11 from Ref. [31] (circles), Ref. [32] (squares), and Ref. [33]
(diamonds).

will be used throughout. Having fixed these constants, it is possible to determine the
remaining two-loop coupling constants c2 and d2 in the IAM from the mass and width
of the (770) resonance. The mass of this resonance can be defined as the energy where
the phase passed 90 and the width can be determined from the slope of the phase at the
resonance. With M = 770.0 MeV and = 150.7 MeV [30], this gives the values
c2 = 27,

d2 = 3.477,

(32)

where the IAM depends rather strongly on d2 and to a lesser extent on c2 . With these values
the IAM gives the phase shown in Fig. 1. It is observed that the result agrees very well
with the experimental phase shifts all the way up to 1 GeV. Therefore, the IAM satisfies
unitarity at least up to this energy. In Fig. 2, the normalized absolute square of the IAM
partial wave is shown. In this case there is no experimental data to be compared with.
However, assuming that the full amplitude is completely dominated by the lowest partial
wave in the resonance region, the cross section (4) can be expressed in terms of |f1 |2 . This
cross section is usually parameterized in the resonance region in terms of the
partial width by using the BreitWigner form
(s) =

M2
24s
.
(s M2 )2 (s M2 )2 + M2 2

(33)

Equating the two expressions for the cross section at the resonance energy, the value
= 96 keV is obtained from the IAM. Experimentally, the value of has been
extracted from the production of via the Primakoff effect by incident pion on
nuclear targets. The present experimental data are, however, not very consistent with
each other [30]. The latest experiment gave the value = 81 4 4 keV [34],

T. Hannah / Nuclear Physics B 593 (2001) 577595

587

(0)

Fig. 2. The lowest partial wave |f1 /f1 |2 obtained from the IAM, Eq. (30).

whereas two earlier experiments gave somewhat lower values for [35]. From
the process e+ e , this partial width has also been obtained giving the value
= 121 31 keV [36]. Therefore, at present it can only be concluded that the value
for obtained from the IAM is not in conflict with the experimental situation, but new
experiments in the resonance region are definitely needed.
5.2. The full amplitude
If the absorptive part of the higher partial waves with l > 3 is neglected, it was shown
in Section 4.1 that the full amplitude could be constructed from the absorptive part of the
lowest partial wave. Since the IAM gives a good description of this partial wave up to
a least 1 GeV, one may use the absorptive part of this partial wave in the Roy equation
for , Eq. (20). With the lowest partial wave approximated by the IAM, the Roy
equation may be written with only one subtraction, which is fully determined by the chiral
anomaly result. Therefore, the non-perturbative chiral result for the full amplitude can be
written as

Z
0  s
t
u
1
(0)
0 Im f1 (s )
+
+ 0
,
(34)
ds
F3 (s, t, u) = F3 +

s0
s0 s s0 t
s u
4M2

where Im f1 is given by the absorptive part of the IAM result (30). The full amplitude is
by construction fully symmetrical in s, t, and u. However, if the lowest partial wave f1
is projected out from Eq. (34), the result does not agree exactly with the IAM result (30).
This is due to the fact that the left cut is now fully determined from crossing symmetry
instead of being approximated by the chiral expansion. In the elastic region below 1 GeV,
the difference is, however, negligible. From the full amplitude, Eq. (34), it is also found

588

T. Hannah / Nuclear Physics B 593 (2001) 577595

that the higher partial waves are very small below 1 GeV. Therefore, the cross section
can indeed be expressed in terms of |f1 |2 to a very good approximation. For a somewhat
different evaluation of the Roy equation for , see Ref. [37].
As the IAM was derived using elastic unitarity, it is expected that this result is only
applicable in the elastic region. The high energy part of the dispersion relation (34)
is therefore not expected to be very well approximated by the IAM. Still, at low and
moderate energies the dispersion relation should be almost completely saturated by the
(770) contribution. Therefore, at these energies the high energy part of the dispersion
relation should not be very important. Furthermore, at these energies it is also expected
that the contribution from the higher partial waves to the Roy equation should be negligible.
Therefore, the non-perturbative chiral result (34) should give a rather accurate description
of the full amplitude at low and moderate energies. From this result, it is possible to
S and D
S in the two-loop ChPT result (23) via the
determine the subtraction constants C
sum rules
Z
Z
0)
0
1
1
Im
f
(s
1
1
1
0
0 Im f1 (s )
S=
S=
ds
,
D
ds
.
(35)
C
(0)
(0)
s 02
s 03
F
F
3

4M2

4M2

S = 2.12 GeV4 ,
S = 0.93 GeV2 and D
Evaluating these sum rules, one obtains the values C
respectively. These sum rules are indeed almost completely saturated by the dispersion
S
relation up to 1 GeV, as the higher energy part only contributes by approximately 2% for C
S
and approximately 1% for D. With the one-loop coupling constant c1 determined from the
assumption of vector resonance saturation, Eq. (31), this gives the following values for the
two-loop coupling constants
c2 = 20,

d2 = 2.7.

(36)

These values are slightly different from the values determined directly from the IAM (32).
As it has already been stated, this is indeed to be expected since the IAM contains higher
order unitarity effects, which will effect the determination of these coupling constants.
Another way to combine unitarity and the chiral expansion has been proposed by
Holstein [38]. In this model, vector meson dominance is combined with one-loop ChPT
in order to account for rescattering effects. Projecting out the lowest partial wave, it is
found that unitarity is approximately satisfied and that the partial width is given
by = 84 keV. Within this model, one obtains the values
c2 = 19,

d2 = 2.9,

(37)

which agree very well with the determination given in (36). However, from the model
proposed by Holstein, all higher partial waves have the same phase as the lowest partial
wave, which is in contradiction with the experimental facts. In order to remedy this, the
lowest partial wave obtained in this model may be inserted in the dispersion relation
(34). The resulting amplitude still agrees well with the experimental information, but now
implies that all the higher partial waves are real. This gives the values c2 = 16 and
d2 = 2.8 from the sum rules (35), which is also fully consistent with the values given

T. Hannah / Nuclear Physics B 593 (2001) 577595

589

in (36). Therefore, it seems that the solution of the sum rules is rather robust against small
variations of the input.
In the numerical results for two-loop ChPT (23), the values of c2 and d2 determined in
(36) will be used together with the value of c1 given in (31). For the low-energy constants
l1 and l2 , these have been determined in ChPT. With the use of the two-loop result for
the scattering amplitude and fitting the D partial wave scattering lengths, the value
l2 l1 = 6.0 has been obtained [9]. Similar values have also recently been obtained from
a two-loop calculation of the Kl4 form factors [39]. These values are rather consistent
with the value given in (31) but are somewhat lower than the value l2 l1 = 7.8 obtained
from a dispersive improved one-loop calculation of the Kl4 form factors [40]. However,
the numerical result only depends very slightly on the exact value of l2 l1 . Therefore, the
values given in Eq. (31) for both l2 l1 and l4 will also be used for ChPT.

6. Numerical results
6.1. Amplitude and cross section
Fig. 3 shows the absolute square of the amplitude normalized to the chiral anomaly. In
the sub-threshold region, the two-loop result is rather close to the one-loop result, whereas
the corrections start to be of importance slightly above threshold. As for the one-loop
result, this gives an almost constant correction to the leading order chiral anomaly result.
This is due to the fact that the only variation comes from the s, t, and u dependency of the

(0)

Fig. 3. The amplitude |F3 /F3 |2 evaluated at t = u as a function of s/M2 . The solid line is the
two-loop result, Eq. (23), the dashed line the one-loop result, Eq. (10), and the dotted line the leading
order chiral anomaly result, Eq. (8). Finally, the long-dashed line is the non-perturbative chiral result,
Eq. (34), and the dash-dotted line is the contribution from the subtraction constants.

590

T. Hannah / Nuclear Physics B 593 (2001) 577595

function F in the one-loop expression (10). The contribution from the subtraction constants
is also shown in the figure, where this contribution is obtained from the expression
(0)
S + t + u) + D(s
S 2 + t 2 + u2 )]. It is observed that this part gives
[1 + C(s
F3 (s, t, u) = F3
the main contribution to the two-loop result, which is due to the presence of the (770)
resonance. However, the unitarity corrections coming from the U (x) terms are also of
some importance in the numerical result. In fact, these unitarity corrections are essential in
the non-perturbative chiral result, Eq. (34), which is also shown in the figure. The deviation
between this result and the two-loop chiral result gives an estimate of the range of validity
of the truncated chiral expansion. From this deviation, it is observed that still higher order
chiral corrections should begin to be of some importance somewhat around s = 10M2 ,
which is also observed in other two-loop calculations.
In Fig. 4, the total cross section is shown as a function of s/M2 , where the total cross
section is given by Eq. (4). Since the one-loop expression for F3 (s, t, u) is more or less
constant in the whole phase space, the one-loop result for the cross section gives an almost
constant correction to the leading order result of approximately 20%. Close to threshold,
the additional two-loop correction is rather small compared to the one-loop correction,
whereas this is not the case for higher energies. The additional two-loop correction amounts
to approximately 75% of the one-loop correction at s = 10M2 , and even more at higher
energies. This indicates that even higher order terms in the chiral expansion should begin
to be of importance at this energy. This is also observed in the figure, where the nonperturbative chiral result begins to deviate from the two-loop result around this energy.

Fig. 4. The total cross section as a function of s/M2 . The solid line is the two-loop result, the
dashed line the one-loop result, the dotted line the leading order result and the long-dashed line the
non-perturbative chiral result.

T. Hannah / Nuclear Physics B 593 (2001) 577595

591

6.2. Comparison with experiments

The process has been investigated at low energies by the Serpukhov
experiment [16]. This experiment used the Primakoff reaction of pion pair production by
pions in the nuclear Coulomb field
+ (Z, A) + 0 + (Z, A).

(38)

The cross section for this process is related to the cross section through the
equivalent photon method

2 
d
Z 2 q 2 qmin
1
d
,
(39)
=

dt
ds dt dq 2
q4
s M2
where


d
|F3 (s, t, u)|2
=
s 4M2 sin2 ,
dt
512
and


2
qmin

s M2
2E

(40)

2
,

(41)

and E is the energy of the incident pion beam. Neglecting the q 2 dependency in
F3 (s, t, u), i.e., setting q 2 0 so that s + t + u = 3M2 , the total cross section is given by
Z2
=
1024 2

smax
Z

4M2



2
2
qmin
(s 4M2 )3/2
qmax
ds
ln 2 + 2 1
s 1/2
qmax
qmin


2
d sin3 F3 (s, t, u) .

(42)

The Serpukhov experiment was carried out with a 40 GeV pion beam and the maximum
2
= 2 103 GeV2  M2 . Thus, the q 2 dependency in
momentum transfer was qmax
F3 (s, t, u) can indeed be neglected when one compares with this experiment.
The kinematical region studied in the Serpukhov experiment was restricted to smax =
10M2 . Three different targets (C, Al, and Fe) were used and the measured total cross
sections were found to agree well with the theoretical Z 2 dependence. Averaging the values
of /Z 2 obtained from the three different targets, the experimental result given in Table 1
was obtained. This experimental result has to be compared with the theoretical predictions,
which are also given in Table 1. It is observed that the two-loop contribution gives a sizable
correction to the one-loop result, whereas the further correction from the non-perturbative
chiral result is rather small. Nevertheless, all results are still somewhat below the central
experimental value, which, however, also has rather large error bars.
From the experimental result of /Z 2 , it is possible to extract the value of the chiral
(0)
, if this is regarded as a free parameter. These extracted values are shown
anomaly F3
in Table 2, where the statistical and systematic errors have been added. The theoretical

592

T. Hannah / Nuclear Physics B 593 (2001) 577595

Table 1
The total cross section for the Primakoff reaction (38) divided by Z 2 . The theoretical predictions are
obtained from Eq. (42) with F3 (s, t, u) given by the leading order chiral anomaly result (O(p4 )),
the one-loop result (O(p6 )), the two-loop result (O(p8 )), and the non-perturbative chiral result
(NPCR). The experimental result is the average value obtained in Ref. [16], where the first (second)
error is statistical (systematic)

/Z 2 (nb)

O(p4 )

O(p6 )

O(p8 )

NPCR

Experiment

0.92

1.09

1.18

1.21

1.63 0.23 0.13

Table 2
(0)
The chiral anomaly F3 extracted from the experimental data. The different values correspond to
setting F3 (s, t, u) in Eq. (42) equal to the leading order chiral anomaly result (O(p4 )), the oneloop result (O(p6 )), the two-loop result (O(p8 )), and the non-perturbative chiral result (NPCR).
The theoretical result is from Eq. (8) with F = 92.4 0.3 MeV [30]

F3 (GeV3 )
(0)

O(p4 )

O(p6 )

O(p8 )

NPCR

Theory

12.9 1.4

11.9 1.3

11.4 1.3

11.3 1.3

9.7 0.1

prediction obtained from Eq. (8) is also shown in this table. With F3 (s, t, u) given by the
(0)
which is 2.3 too high
leading order chiral anomaly result, one obtains a value of F3
(0)
compared with the theoretical prediction. This result for F3 is the value generally quoted
from the analysis in Ref. [16]. However, with the one-loop expression for F3 (s, t, u),
(0)
which is only 1.7 too high, and with the two-loop result the
one obtains a value of F3
disagreement is only at the 1.3 level.
Since the non-perturbative chiral result is close to the two-loop result, the uncertainty
coming from yet higher orders in the chiral expansion should be rather small, at least in
the kinematical region probed by the Serpukhov experiment. There is of course also some
uncertainty due to the uncertainty in the determination of the coupling constants c2 and d2 .
However, as already discussed, this uncertainty should also be rather small. Therefore,
in view of the disagreement between the theoretical predictions and the Serpukhov
experiment, new improved Primakoff experiments are definitely needed.
The process has also been investigated at low energies at CERN [41]. This
experiment used the reaction e 0 e with 300 GeV pions and measured the total
cross section for this process. It was found that the cross section obtained agreed with the
chiral anomaly prediction with three number of colors. However, since the error bars were
rather large, it was not possible to observe any systematic deviations from the soft pion and
soft photon limit in this experiment.
Therefore, new precision experiments are necessary in order to investigate the process
and thus the theory of chiral anomalies in greater detail. Indeed, such
new experiments are under way at different facilities. In the COMPASS experiment
at CERN [17] and in the SELEX experiment at FNAL (E781) [18], the Primakoff

T. Hannah / Nuclear Physics B 593 (2001) 577595

593

reaction (38) will be measured with 600 and 50280 GeV pion beams, respectively. In
these two experiments, the expected number of near threshold two-pion events is several
orders of magnitude higher than previously obtained. This will allow analysis of the data
separately in different intervals of s with small statistical errors.
The SELEX experiment will also measure the reaction e 0 e in order to obtain
independent information on the amplitude. For this reaction, the expected
number of events is also significantly larger than previously obtained, which would give
excellent complementary information on F3 (s, t, u).
Finally, at CEBAF the process is investigated by measuring p + 0 n
cross sections using tagged photons [19]. Since the resonance region will also be measured
in this experiment, this could give new information on the partial width .
However, as the incident pion is virtual in the CEBAF experiment, this has to be taken into
account when comparing with theory.

7. Conclusion
The anomalous process plays an important role in the theory of chiral
anomalies. At leading order in the chiral expansion, which corresponds to the soft pion
and soft photon limit, the amplitude for this process is given in terms of the number of
colors Nc of the strong interaction. Comparing the leading order result with the present
experimental information [16], one finds that the value Nc = 4 is favored.
In order to test this important conclusion more precisely, it is necessary with both
experimental and theoretical improvements. Indeed, significant improvements on the
experimental side are expected from several new experiments [1719]. On the theoretical
side, the one-loop correction to the leading order result has previously been calculated [15].
However, in view of the new precision experiments, it is important also to calculate the
additional two-loop correction to the amplitude.
This has been the purpose of the present paper, where the amplitude for the anomalous
process has been evaluated to two loops in the chiral expansion by means
of a dispersive method. The two new coupling constants that enter at two-loop order
were determined from sum rules with the use of a non-perturbative chiral approach. The
uncertainty in the numerical results due to this determination was estimated to be rather
small. Moreover, the still higher order terms in the chiral expansion were also estimated to
be small at low energies.
The two-loop result improves the agreement with the present experimental information [16] compared to both the leading order and the one-loop results. However, the twoloop prediction is still significantly below the central experimental data for Nc = 3. This
fact is not likely to be due to theoretical uncertainties. Therefore, should the new experiments [1719] confirm the present central experimental value with better precision, it
would be a serious problem for QCD.

594

T. Hannah / Nuclear Physics B 593 (2001) 577595

Acknowledgements
The author thanks T.N. Truong for useful discussions. Part of this work was done
while the author was visiting Ecole Polytechnique, Paris, which is acknowledged for its
hospitality and financial support.

References
[1]
[2]
[3]
[4]
[5]

[6]
[7]
[8]
[9]

[10]

[11]
[12]

[13]
[14]

[15]
[16]
[17]

[18]

S. Weinberg, Physica A 96 (1979) 327.

J. Gasser, H. Leutwyler, Ann. Phys. (N.Y.) 158 (1984) 142.
J. Gasser, H. Leutwyler, Nucl. Phys. B 250 (1985) 465.
H. Leutwyler, Ann. Phys. (N.Y.) 235 (1994) 165.
For some reviews on ChPT see, e.g.:
U.G. Meiner, Rep. Prog. Phys. 56 (1993) 903;
A. Pich, Rep. Prog. Phys. 58 (1995) 563;
G. Ecker, Prog. Part. Nucl. Phys. 35 (1995) 1;
J. Bijnens, U.G. Meiner, Proc. Workshop on Chiral Effective Theories, Bad Honnef, Germany,
November 30December 4, 1998, hep-ph/9901381.
J. Gasser, U.G. Meiner, Nucl. Phys. B 357 (1991) 90;
G. Colangelo, M. Finkemeier, R. Urech, Phys. Rev. D 54 (1996) 4403.
J. Bijnens, G. Colangelo, P. Talavera, JHEP 05 (1998) 014.
M. Knecht et al., Nucl. Phys. B 457 (1995) 513.
J. Bijnens et al., Phys. Lett. B 374 (1996) 210;
J. Bijnens et al., Nucl. Phys. B 508 (1997) 263;
J. Bijnens et al., Nucl. Phys. B 517 (1998) 639, Erratum.
S. Bellucci, J. Gasser, M.E. Sainio, Nucl. Phys. B 423 (1994) 80;
S. Bellucci, J. Gasser, M.E. Sainio, Nucl. Phys. B 431 (1994) 413, Erratum;
U. Brgi, Phys. Lett. B 377 (1996) 147;
U. Brgi, Nucl. Phys. B 479 (1996) 392;
J. Bijnens, P. Talavera, Nucl. Phys. B 489 (1997) 387.
J. Wess, B. Zumino, Phys. Lett. B 37 (1971) 95;
E. Witten, Nucl. Phys. B 223 (1983) 422.
J. Bijnens, A. Bramon, F. Cornet, Z. Phys. C 46 (1990) 599;
D. Issler, preprint SLAC-PUB-4943;
R. Akhoury, A. Alfakih, Ann. Phys. (N.Y.) 210 (1991) 81;
H.W. Fearing, S. Scherer, Phys. Rev. D 53 (1996) 315.
For a review on ChPT in the anomalous sector see: J. Bijnens, Int. J. Mod. Phys. A 8 (1993)
3045.
S.L. Adler et al., Phys. Rev. D 4 (1971) 3497;
M.V. Terentev, Phys. Lett. B 38 (1972) 419;
R. Aviv, A. Zee, Phys. Rev. D 5 (1972) 2372.
J. Bijnens, A. Bramon, F. Cornet, Phys. Lett. B 237 (1990) 488.
Yu.M. Antipov et al., Phys. Rev. D 36 (1978) 21.
M.A. Moinester, V. Steiner, S. Prakhov, in: I. Iori (Ed.), Proc. XXXVII Int. Winter Meeting
on Nuclear Physics, Bormio, Italy, January 2529, 1999, Ricerca Scientifica ed Educazione
Permanente Supplemento, Vol. 114, 1999, preprint TAUP-2562-99, hep-ex/9903017.
SELEX Collaboration, M.A. Moinester et al., in: D.W. Menze, B. Metsch (Eds.), Proc. 8th Int.
Conf. on the Structure of Baryons, Bonn, Germany, September 2226, 1998, BARYONS 98,
World Scientific, Singapore, 1999, preprint TAUP-2568-99, hep-ex/9903039;

T. Hannah / Nuclear Physics B 593 (2001) 577595

[19]
[20]
[21]

[22]
[23]
[24]

[25]
[26]
[27]

[28]
[29]

[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]

595

M.A. Moinester, in: H. Machner, K. Sistemich (Eds.), Proc. Int. Conf. on Physics with GeV
Particle Beams, Jlich, Germany, August 2225, 1994, World Scientific, N.J., 1994, preprint
TAUP-2176-94, hep-ph/9409307.
R.A. Miskimen, K. Wang, A. Yegneswaran, Spokesmen, CEBAF Proposal PR-94-015, 1994.
K.M. Watson, Phys. Rev. 95 (1954) 228.
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.
J.F. Donoghue, B.R. Holstein, Y.C.R. Lin, Phys. Rev. Lett. 55 (1985) 2766;
J. Bijnens, A. Bramon, F. Cornet, Phys. Rev. Lett. 61 (1988) 1453.
G. Ecker et al., Nucl. Phys. B 321 (1989) 311;
G. Ecker et al., Phys. Lett. B 223 (1989) 425.
S.M. Roy, Phys. Lett. B 36 (1971) 353;
J.L. Basdevant, J.C. Le Guillou, H. Navelet, Nuovo Cimento A 7 (1972) 363;
J.L. Petersen, CERN Yellow Report, CERN 77-04.
J. Stern, H. Sazdjian, N.H. Fuchs, Phys. Rev. D 47 (1993) 3814.
T.N. Truong, Phys. Rev. Lett. 61 (1988) 2526;
T.N. Truong, Phys. Rev. Lett. 67 (1991) 2260.
A. Dobado, J.R. Pelez, Phys. Rev. D 47 (1993) 4883;
A. Dobado, J.R. Pelez, Phys. Rev. D 56 (1997) 3057;
T. Hannah, Phys. Rev. D 51 (1995) 103;
T. Hannah, Phys. Rev. D 52 (1995) 4971;
T. Hannah, Phys. Rev. D 54 (1996) 4648.
T. Hannah, Phys. Rev. D 55 (1997) 5613.
J.A. Oller, E. Oset, J.R. Pelez, Phys. Rev. Lett. 80 (1998) 3452;
J.A. Oller, E. Oset, J.R. Pelez, Phys. Rev. D 59 (1999) 074001;
J.A. Oller, E. Oset, J.R. Pelez, Phys. Rev. D 60 (1999) 099906, Erratum;
F. Guerrero, J.A. Oller, Nucl. Phys. B 537 (1999) 459.
Particle Data Group, C. Caso et al., Eur. Phys. J. C 3 (1998) 1.
P. Estabrooks, A.D. Martin, Nucl. Phys. B 79 (1974) 301.
S.D. Protopopescu et al., Phys. Rev. D 7 (1973) 1279.
B. Hyams et al., Nucl. Phys. B 64 (1973) 134;
W. Ochs, Ph.D. Thesis, Ludwig-Maximilians-Universitt, 1973.
L. Capraro et al., Nucl. Phys. B 288 (1987) 659.
J. Huston et al., Phys. Rev. D 33 (1986) 3199;
T. Jensen et al., Phys. Rev. D 27 (1983) 26.
S.I. Dolinsky et al., Z. Phys. C 42 (1989) 511.
T.N. Truong, hep-ph/9903378.
B.R. Holstein, Phys. Rev. D 53 (1996) 4099.
G. Amors, J. Bijnens, P. Talavera, Phys. Lett. B 480 (2000) 71;
G. Amors, J. Bijnens, P. Talavera, Nucl. Phys. B 585 (2000) 293.
J. Bijnens, G. Colangelo, J. Gasser, Nucl. Phys. B 427 (1994) 427.
S.R. Amendolia et al., Phys. Lett. B 155 (1985) 457.

Nuclear Physics B 593 (2001) 599633

www.elsevier.nl/locate/npe

Implications of N = 1 superconformal symmetry

for chiral fields
F.A. Dolan, H. Osborn
Department of Applied Mathematics and Theoretical Physics, Silver Street, Cambridge, CB3 9EW, UK
Received 28 June 2000; accepted 8 September 2000

Abstract
The requirements of N = 1 superconformal invariance for the correlation functions of chiral
superfields are analysed. Complete expressions are found for the three-point
P function for the
general spin case and for the four-point function for scalar superfields for
qi = 3 where qi
is the scale dimension for the ith superfield and is related to the U (1) R-charge. In the latter
case the relevant Ward identities reduce to eight differential equations for four functions of u, v
which are invariants when the superconformal symmetry is reduced to the usual conformal group.
The differential equations have a general solution given by four linearly independent expressions
involving a two-variable generalisation of the hypergeometric function. By considering the behaviour
under permutations, or crossing symmetry, the chiral four-point function is shown to be determined
up to a single overall constant. The results are in accord with the supersymmetric operator product
expansion. 2001 Elsevier Science B.V. All rights reserved.

1. Introduction
It is now clear that there exist in four, and also five and six, dimensions a plethora of
nontrivial quantum field theories, without any coupling to gravity, which possess fixed
points realising conformal symmetry. The evidence for such conformally invariant theories
is strongest in the case where the field theory is supersymmetric and the conformal
group is extended to the superconformal group. In four dimensions this may be identified
as SU(2, 2|N ) where the cases of N = 1, 2, 4 are relevant for renormalisable field
theories, although for N = 4 the group contains an ideal so the group may be reduced
to the projective group PSU(2, 2|4). Besides the usual 15 parameter conformal group the
superconformal group also contains an R-symmetry, U (1), U (2), and SU(4) for N = 1, 2
and 4. The case of N = 4 gauge theories has been known for a long time since these
* Corresponding author. Address for correspondence: Trinity College, Cambridge, CB2 1TQ, England.

E-mail addresses: [email protected] (F.A. Dolan), [email protected] (H. Osborn).

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 5 5 3 - 8

600

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

have vanishing -functions and recently the strong coupling limit of such theories has
been explored through the ADS/CFT correspondence. Nevertheless N = 1 theories may
possess, for appropriate matter content, a conformal window with superconformal invariant
fixed points.
Such theories are naturally described in terms of superfields defined over a superspace
involving Grassman variables. In general the action of conformal transformations on the
coordinates is nonlinear and the construction of covariant correlation functions is not
as straightforward as realising the consequences of symmetry under the usual Poincar
group transformations which act linearly on the coordinates. A formalism which allows the
construction of two- and three-point, and in principle n > 3 as well, correlation functions
was described in [1] for the N = 1 superconformal group in four dimensions, extending
a similar discussion for the ordinary conformal group in [2,3]. This has been extended to
six dimensions and also to N > 1 in four dimensions by [4] and this approach has been
used for an analysis of particular two- and three-point functions for N = 2 in [5]. In four
dimensions the superconformal group is compatible with acting on superfields restricted
to the chiral projections, z+ = (x+ , ) where is a chiral spinor, or z = (x , ), where
is an antichiral spinor. For two points z1 = (x1 , 1 , 1 ) and z2 = (x2 , 2 , 2 ) then the
a
a which is constructed
constructions described in [1] are expressed in terms of x 12
= x12

from z1 and z2+ and transforms homogeneously under superconformal transformations
with a local scale transformation and rotation at z1 and also at z2 , in particular,
2
x12

2
x12

S 1 )(z2+ )
(z

(1.1)

which is a direct generalisation of the transformation of (x1 x2 )2 under the usual

conformal group. For three points in superspace z1 , z2 , z3 it is also possible to generalise
the treatment for forming conformally covariant expressions [2,3] by introducing the
S1 ,
S1 , which are related in the same fashion
variables X1 , 1 and their conjugates X

as x+ , and x , , that transform homogeneously under local rotations and scale

2 /(x 2 x 2 ) and its transformation follows from
transformations at z1 . In particular X12 = x23

13

21
(1.1). These variables play an essential role in the general construction of three-point
functions described in [1].
However X1 depends on z1 , z2 , z3+ while 1 is formed from z1 , z2 , z3 and
correspondingly for their conjugates. In consequence the construction of correlation
functions for chiral fields alone, which depend only on the z+ s, requires cancellations
of unwanted chiral components in the formalism of [1]. This was achieved quite easily for
the three-point function for chiral scalar superfields h1 (z1+ )2 (z2+ )3 (z3+ )i but efforts
to generalise such a treatment to the four-point function became very involved and were not
carried through to a conclusion. In a different approach Pickering and West [6] obtained
identities expressing superconformal invariance directly for three- and four-point functions
which from the start depend solely on the z+ s. 1 The three-point case was again rather
simple but the invariance for the four-point function led to a system of eight linear first1 The three-point function was considered earlier by Conlong and West [7].

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

601

order partial differential equations for four functions of two variables u, v which are the
two independent cross ratios formed from (xi+ xj + )2 . Without supersymmetry u, v are
conformal invariants and the general conformally covariant four-point function involves an
arbitrary function of u, v.
In this paper we follow a similar path to finding the conditions following from N = 1
superconformal invariance for chiral-superfield correlation functions. For such superfields
the scale dimension q is trivially related to the R-charge 3q and they may belong only
to (j, 0) spinor representations of the four-dimensional Lorentz group. For a subgroup
of Sl(4|1), the complexification of SU(2, 2|1), which contains the conformal group, the
constraints are simply realised where by introducing, for three points z1+ , z2+ , z3+ , spinor
variables which transform homogeneously. Extending this to the full superconformal
P
group involves further constraints. For the three-point function we must require qi = 3
and for the four-point function, with the same condition on the qi s, we obtain eight
differential equations similar to those found by Pickering and West. These are here solved
in terms of a certain two-variable generalisation of the hypergeometric function F4 . The
superconformal symmetry identities for the correlation function for four scalar chiral
superfields have in general four linearly independent solutions. A careful analysis of the
behaviour under permutations of the superfields or imposing crossing symmetry, which
requires taking into account nontrivial identities for the F4 functions under analytic
continuation, leads to a unique expression up to an overall constant.
In this paper in the next section we describe how the the superconformal group acts on
the chiral coordinates z+ and obtain the essential transformation formulae used to construct
superconformal correlation functions later. In Section 3 we derive the superconformal
identities for the three- and four-point functions for chiral fields. Results for the threepoint function are obtained for arbirary spin. For the four-point function we restrict to
scalar fields and the eight equations for four functions of u, v are obtained. These equations
have four linearly independent solutions which are expressed in terms of the functions F4
mentioned above. The conditions necessary for Bose symmetry or crossing symmetry are
analysed in Section 4. By requiring independence under the path of analytic continuation,
and using the transformation formulae for the F4 function, a unique result, up to a single
overall constant, is obtained. In Section 5 an alternative expression in terms of a function G,
defined in terms of two F4 functions, is obtained. This allows the short distance limits of the
four-point function to become manifest. In Section 6 we endeavour to count the constraints
for higher-point functions of chiral scalar superfields. At least for six- and higher-point
functions the differential equations arising from superconformal invariance do not fully
determine the functions arising in the general expansion. In Section 7 we consider the
application of the operator product expansion for two chiral superfields to the four-point
function. The contribution of a chiral scalar superfield in the operator product expansion to
the four-point function is obtained in a form which satisfies all the superconformal Ward
identities but which does not satisfy the crossing symmetry relations by itself. In Section 8
we show how an integral over superspace defines a superconformal N -point function for
any N and this is shown to be in accord with the results obtained in Section 3 for N =
3, 4. In Appendix A we discuss an alternative derivation of the essential equations for the

602

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

four-point function based on using superconformal transformations to fix two points. In

Appendix B we describe some results relevant in Section 8 for the evaluation of conformal
integrals while in Appendix C some results which allow for the simplification of the twovariable function G used here are described. For the cases of relevance here it is shown
how they may be reduced to sums of products of ordinary hypergeometric functions. In
Appendix D we recapitulate some results for the operator product expansion which are
applied to the chiral-superfield four-point function in Appendix E.

2. Superconformal transformations on chiral superspace

We use the standard identification of 4-vectors with 2 2 matrices 2 so that the action
of infinitesimal superconformal transformations on z+ is given by [1,9]
x+ + x + bx+ + 4i 4x+ ,
x + = a + x + x + + ( + )
=  + + bx+ i x + + 2 2 ,
= 14 ab (a b ) ,

= 14 ab ( a b ) ,

(2.1)

where a a corresponds to a translation, ab = ba an infinitesimal rotation, ba a special

conformal transformation, + a rescaling and ( )/i
a U (1)R phase while  ,  are
supertranslations with , their superconformal extensions. These parameters may be
written as elements of a supermatrix

ib
2
13 ( + 2)1
,
(2.2)
M =

2
ia
+ 13 (2 + )1
2
(

2
2
3
so that
2 1 1 2 = 3

[M1 , M2 ] = M3 .

(2.3)

Since str M = 0, M is a generator of Sl(4|1) which is reduced to SU(2, 2|1) by imposing

the appropriate reality condition.
For two points zi+ , zj + we may define
x ij = x i+ x j + ,

`ij = (i j )x1
ij ,

2
x 1
ij = xj i /xij ,

(2.4)

which, from (2.1), transform as

x ij = ( + + x i+ b + 4i i `ij )xij + x ij ( + + bxj + 4j ),
`ij = `ij ( + + x i+ b 4i ) i + i b + 2i`2ij  i ,

(2.5)

where we also define

2 The notation is identical with [1] and is essentially that of Wess and Bagger [8]. Thus , are regarded as

2 =
row, column vectors and we let =  , =  form associated column, row vectors, 2 = ,

. The basis of 2 2-matrices is given by the hermitian -matrices a , a , (a b) = ab 1, and for a 4-vector
= x a ( )
=   x , with inverse x a = 1 tr( a x ).
x a then x = x a (a ) , x
a

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

 (x+ ) =  + ix+ ,

i =  (xi+ ).

603

(2.6)

Furthermore if
rij = (xi+ xj + )2 = det(xij ),

(2.7)

defining also the spinor i(j k) and 4-vector Yi(j k) by

i(j k) = `ij `ik ,
1
1
j k x 1
x1
kj x 1
Yi(j k) = x1
ij + x
ik = x
ji x
ik =
ki x
ij ,

(2.8)

we then have
rij = 2( + b xi b xj + 2j 2i `ij i )rij ,
and

(2.9)

i(j k) = i(j k) + + x i+ b + 4i 2i (`ij + `ik ) i ,
Yi(j k) = ( bxi+ + 4i )Yi(j k)

Yi(j k) + + x i+ b + 4i i `ij 4i x ij x 1
kj k i(j k) .

(2.10)

If we set  , = 0 then it is evident from (2.5) that x ij varies homogeneously with local
scale transformations and rotations at zi+ and zj + . Similarly the variations of i(j k) and
Yi(j k) given by (2.10) then corresponds to a local scale transformation and rotation at
= 0 contains
zi+ . The group G0 Sl(4|1) generated by matrices M as in (2.2) with ,
the usual bosonic conformal group and it is straightforward to construct covariantly
transforming correlation functions and also invariants under G0 . Thus for four points
z1+ , z2+ , z3+ , z4+ we may define the G0 invariant cross ratios
r14 r23
r12 r34
,
v=
.
(2.11)
u=
r13 r24
r13 r24
For later use we also define unit 4-vectors by


rj k 1/2
1/2
1/2
bi(j k) = Yi(j k)
= x 1
r
,
Y
,
(2.12)
x j i = xj i rij
ij ij
rij rik
so that, from (2.5) and (2.10), we have

i ) + 4i i `ij + 1 `ij i 1 ,
j x j i x j i + 12 (xi b bx
x j i = b
2


i ) + 4i i `ij + 1 `ij i 1
bi(j k) + 1 (xi b bx
bi(j k) Y
bi(j k) = b
i Y
Y
2
2

1
ij x 1
4i x ij x 1
kj k i(j k) + 2 i(j k) x
kj k 1 ,

(2.13)

= 0, is given by
where b
(z+ ), b

+ 4 + 1 1 ,
b
(z+ ) = 12 (bx+ x+ b)
2

b
i = b
(zi+ ).

(2.14)

In obtaining (2.13) it is useful to note, with the definitions in (2.6),

ik x 1
i = x ij x 1
kj k + x
j k j .

(2.15)

To achieve symmetry under the full superconformal group requires cancellation of all
terms involving i . This requires further constraints. To obtain the necessary conditions we
first obtain from (2.10),

604

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

2i(j k) = 2 b xi+ + 4 i + i(`ij + `ik )i 2i(j k) ,

(2.16)

so that, using (2.9), we have

2i(j k)
rij

= 2(2i + j )

2i(j k)
rij

(2.17)

with
b x+ + 2 ,
(z+ ) = 13 ( + 2)

i = (zi+ ).

(2.18)

In (2.17) the terms involving i have therefore cancelled. For subsequent use we also note
that, from (2.9) and (2.18),

rij
rij
= 2(j k 2ii(j k) i ) .
rik
rik

(2.19)

From the definition in (2.7) we have

i(j k) = i(j l) + i(lk) ,

i(j k) = i(kj ) ,

(2.20)

as well as
i(j k) = j (ik) x j k x 1
ik .

(2.21)

This leads to
2i(j k)
rj k

2j (ki)
rik

2k(ij )
rij

(2.22)

so that 2i(j k) /rj k is completely symmetric. This is manifest in the transformation

properties, obtained from (2.17) and (2.19),

2i(j k)
rj k

= 2(i + j + k )

2i(j k)
rj k

(2.23)

These results allow the construction of superconformal covariant expressions for chiral
superfield correlation functions. For a quasi-primary spin-j chiral field belonging to the
(j, 0) representation I = 1 ...2j , totally symmetric in 1 . . . 2j so that I takes 2j + 1
values, we have, with b
and defined in (2.14) and (2.18):
(z+ )s)I J J (z+ ),
I (z+ ) = Lz+ I (z+ ) 2q (z+ )I (z+ ) + (b

,
(b
s)I J J = 2j b
(1 2 ...2j ) ,
Lz+ = z+
z+

(2.24)

where b
s = b
s with s , [s , s ] = s s , generators for spin j . The
representations are specified by (j, q) and for these to be unitary with positive energy
either j = q = 0, which is the trivial singlet case, or q > j + 1, with q determining both
the R-charge and scale dimension.

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

605

3. Chiral three- and four-point functions

The superconformal Ward identities for the n-point chiral correlation function of chiral
superfields belonging to representations (ji , qi ), i = 1, . . . , n, requires
X


(Li + 2qi i ) I1 (z1+ ) . . . In (zn+ )
i

(b
i si )Ii J I1 (z1+ ) . . . J (zi+ ) . . . In (zn+ ) = 0,

(3.1)

with si the generators for spin ji . We solve these identities explicitly for general spins
when n = 3 and for scalar chiral superfields when n = 4 so long as
X
qi = 3.
(3.2)
i

For the three-point function we may write



I1 (z1+ )I2 (z2+ )I3 (z3+ )

21(23)  r12 1q2  r12 1q1
FI1 I2 I3 (x1+ , x2+ , x3+ ).
=
r23
r13
r23

(3.3)

For scalar fields FI1 I2 I3 C123 , a constant, and it is easy to see from (2.23) and (2.19),
since 31(23) = 0, that this has the correct transformation properties in accord with (3.1) and
also from (2.22), with (3.2), it is completely symmetric under simultaneous permutations
of z1+ , z2+ , z3+ and q1 , q2 , q3 . To construct FI1 I2 I3 for nonzero spins we introduce
I I = x (1 | 1 . . . x 2j ) 2j ,
I (j ) (x)

(3.4)

which is a bispinor, transforming as (j, 0) (0, j ). The appropriate solution is then

b1(23) I (j2 ) (x21) I (j3 ) (x31 ) t I1 I2 I3 , (3.5)
FI1 I2 I3 (x1+ , x2+ , x3+ ) = I (j1 ) Y
I2 I2
I3 I3 j1 j2 j3
I I
1 1

b1(23) are given by (2.13) and t I1 I2 I3 is an invariant tensor satisfying,

where x 21, x 31 and Y
j1 j2 j3
for any rotation r,

D (j1 ) (r)I1J1 D (j2 ) (r)I2J2 D (j3 ) (r)I3J3 t Jj11jJ22jJ33 = t Ij11Ij22Ij33 .

(3.6)

Such tensors are given by the usual ClebschGordan coefficients if |j2 j3 | 6 j1 6 j2 + j3 .

s 2 )I2J t Ij11Jj2Ij33 + (
s 3 )I3J t Ij11Ij22Jj3 = 0,
Infinitesimally (3.6) requires (
s 1 )I1J t Jj1Ij22Ij33 + (
for any , = 0, and using this with (2.13), and since terms 1(23) in (2.13)
vanish, it is straightforward to see that this ensures that (3.5) has the correct spinorial
transformation properties according to (3.1). Although the construction in (3.5) is apparently asymmetric we may reexpress the result in the equivalent form expected by permutation symmetry. Using (3.6) first with D (j ) (r) = I (j ) (x32 )1 I (j ) (x31 ) then we have
b1(23)) = I (j1 ) (x12 )D (j1 ) (r), since Y
b1(23) = x 12 x 1 x 31 , and similarly I (j2 ) (Y
b2(13)) =
I (j1 ) (Y
32
(j
)
(j
)
1
(j
)
(j
)
1
(j
)
2
2
I (x21)D (r) , or secondly with D (r) = I (x23 ) I (x21 ) when we have
b1(23)) = (1)2j1 I (j1 ) (x13 )D (j1 ) (r), allows the expression (3.5) also to be written as
I (j1 ) (Y

606

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

b2(13)
FI1 I2 I3 (x1+ , x2+ , x3+ ) = I (j1 ) (x12 )I1 I1 I (j2 ) Y

I2 I2

b3(12)
= I (j1 ) (x13)I1 I1 I (j2 ) (x23)I2 I2 I (j3 ) Y

I (j3 ) (x32 )I3 I3 t Ij11Ij22Ij33

I3 I3

I I I

(1)2j1 t j11 j22 j33 .

(3.7)

Fermi/Bose symmetry following from 1 (z1+ )2 (z2+ ) = P12 2 (z2+ )1 (z1+ ), where
P12 = 1 for both 2j1 , 2j2 odd, and with similar definitions of P23 , P13 , requires then
I I I

I I I

P23 t j11 j33 j22 = (1)2j1 t j11 j22 j33 ,

I I I

I I I

I I I

P12 t j22 j11 j33 = t j11 j22 j33 ,

I I I

P13 P23 t j33 j11 j22 = (1)2j1 t j11 j22 j33 .

(3.8)

If (j1 , q1 ) = (j2 , q2 ) this provides constraints on possible values of j3 , and similarly for
any other identical pair.
The discussion of the four-point function is more involved and so is restricted here to
chiral scalar superfields i , i = 1, 2, 3, 4. As shown in Section 6, and in accord with the
analysis of Pickering and West [6], we may restrict to an expansion in a nilpotent basis
formed by 2i(j k) , 1 6 i < j < k 6 4, with coefficients involving functions of the cross
ratios u, v defined in (2.11). Thus we write,


1 (z1+ )2 (z2+ )3 (z3+ )4 (z4+ )

21(23)  r12 q3 1 r23 q1 +q4 1 r12 q4
f123 (u, v)
=
r23
r13
r13
r24
21(24)  r14 q2 1 r12 q3 +q4 1 r14 q3
f124 (u, v)
+
r24
r24
r24
r13
21(34)  r34 q1 1 r14 q2 +q3 1 r34 q2
f134 (u, v)
+
r34
r13
r13
r24
22(34)  r23 q4 1 r34 q1 +q2 1 r23 q1
f234 (u, v).
(3.9)
+
r34
r24
r24
r13
It is easy to see that this has the correct form to satisfy the superconformal Ward identities
up to terms proportional to i which are generated by the variation of u, v and also by the
variation of the last factor of rij /rik in each of the four terms appearing in the expression
for the four-point function given by (3.9). Such terms must cancel. To achieve a linearly
independent basis we restrict all contributions to only 1 , 2 , by using (2.15), and also to
involve just 21(23)1(24) and 21(24)1(23) . Using (2.19) and (2.21) we may find
v = 4iv(1(43)1 + 2(34)2 )

14 x 1
= 4iv 1(23) 1 x 13 x 1
23 2 1(24) 1 x
24 2 ,

(3.10)

and also
u = 4iu(1(23)1 + 4(32)4 )
14 x 1
x13 x 1
42 x 1
= 4i(1(23) 1(24)) u1 v x 13 x 1
23 2 + x
24 2 u
43 x
12 1
4iu 1(24)x 14x 1
24 2 .
It is easy to see that

(3.11)

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

21(34)1(23) = 21(34)1(24) = 21(23)1(24) + 21(24)1(23),

607

(3.12)

and, from (3.10) and (3.11), we have

r13
23 x 1
22(34)v = 4iv 21(23)1(24) 1 x 14 x 1
24 x
13 1
r23


r14
x 24x 1
1 ,
(3.13)
4iv 21(24)1(23) 1 x 13x 1
23
14
r24
and
r13 2
r14 2
1(23)1(24)x 14x 1
x 23x 1
1 4iu

1(23)1 . (3.14)
22(34)u = 4iu
24
13
r23
r24 1(24)
To complete the calculation we need the contributions from the variations of the relevant
rij /rik factors in the four terms appearing in (3.9). These are given respectively by
4iq4 21(23)2(14)2 = 4iq421(23)1(24)x 14 x 1
24 2 ,
4iq3 21(24)1(43)1 = 4iq321(24)1(23)1 ,

4iq2 21(34)4(32)4 = 4iq2 21(23)1(24) + 21(24)1(23) 1 x 14 x 1
24 2 ,


r13
23 x 1
14x 1
4iq1 22(34)3(21)3 = 4iq1 21(23)1(24) x 14 x 1
24 x
13 1 x
24 2
r23


r14
(3.15)
4iq1 21(24)1(23) 1 x 13x 1
23 2 .
r24
Using in addition the identities
23 x 1
x13 x 1
42x 1
x 14 x 1
24 x
13 = u
43 x
12 + v u,
24 x 1
x13 x 1
42 x 1
v x 13 x 1
23 x
14 = u
43 x
12 + 1,

(3.16)

we find the following relations necessary to satisfy the superconformal Ward identity, to
cancel terms involving 21(23)1(24)x 14x 1
24 2 ,


q4 + vv + (u 1)u f123 + uq1 +q2 2 v q2 +q3 1 (q2 + vv + uu )f134
uq1 +q2 2 q1 f234 = 0,

(3.17)

to cancel 21(23)1(24)x 13 x 1
23 2 ,
u f123 uq1 +q2 2 v q2 +q3 1 v f134 = 0,

(3.18)

to cancel 21(24)1(23)x 14 x 1
24 2 ,
u f124 + uq1 +q2 2 (q2 + vv + uu )f134 = 0,

(3.19)

to cancel 21(24)1(23)x 13 x 1
23 2 ,
(v + u )f124 + uq1 +q2 2 v f134 + uq1 +q2 2 v q1 +q4 2 q1 f234 = 0,

(3.20)

42 x 1
to cancel 21(23)1(24)x 13 x 1
43 x
12 1 ,
u f123 + uq1 +q2 2 (q1 + vv + uu )f234 = 0,
42 x 1
to cancel 21(24)1(23)x 13 x 1
43 x
12 1 ,

(3.21)

608

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

u f124 + uq1 +q2 2 v q1 +q4 1 v f234 = 0,

(3.22)

to cancel 21(23)1(24)1 ,
(vv + uu )f123 + uq1 +q2 2 v q2 +q3 1 q2 f134

uq1 +q2 2 (v u)(q1 + vv + uu ) vv f234 = 0,

(3.23)

and finally to remove 21(24)1(23)1 ,

(q3 + vv + uu )f124 uq1 +q2 2 q2 f134
+ uq1 +q2 2 v q1 +q4 1 (q1 + (v 1)v + uu )f234 = 0.

(3.24)

By taking linear combinations the above eight conditions, (3.17)(3.24), may be

expressed more succinctly as
u f123 = uq1 +q2 2 v q2 +q3 1 v f134 ,

(3.25a)

q1 +q2 2

(3.25b)

(q1 + vv + uu )f234 ,
q1 +q2 1 q2 +q3 2
u
v
u f134 ,
q1 +q2 2 q1 +q4 1
u
v
v f234 ,
q1 +q2 2
(q2 + vv + uu )f134 ,
u
q1 +q2 1 q1 +q4 2
u
v
u f234 ,
q1 +q2 1
u
u f234 ,
q1 +q2 1
u
u f134 .

= u
v f123 =
u f124 =
=
v f124 =
(q4 + vv + uu )f123 =
(q3 + vv + uu )f124 =

(3.25c)
(3.25d)
(3.25e)
(3.25f)
(3.25g)
(3.25h)

An important consistency check is that this set of equations are invariant under simultaneous permutations of zi+ and qi . For the cyclic permutation z1+ z2+ z3+
z4+ z1+ , which implies u v, and also letting q1 q2 q3 q4 q1 then the
equations are invariant for f123 f234 f134 f124 f123 as expected from the
form of (3.9). Similarly under z1+ z2+ , when u u0 = u/v, v v 0 = 1/v so that
uu = u0 u0 , vv = v 0 v 0 u0 u0 , then also taking q1 q2 and letting f123 (u, v)
v 0q4 f123 (u0 , v 0 ), f124 (u, v) v 0q3 f124 (u0 , v 0 ), f134 (u, v) v 0q1 f234 (u0 , v 0 ), f234 (u, v)
v 0q2 f134 (u0 , v 0 ), we have (3.25a) (3.25b), (3.25c) (3.25g), (3.25d) (3.25e) and
(3.25f) (3.25h). It is straightfoward to obtain equations for f123 , etc., alone. Eliminating
f234 from (3.25b) and (3.25g) gives


 2
u u u2 + 2uvu v + v 2 v2 + (3 q2 + q4 )(vv + uu ) (q3 + q4 1)u

+ (2 q2 )q4 f123 = 0,
(3.26)
while from (3.25a) and (3.25c) we have

 2
uu vv2 + (q3 + q4 1)u (q1 + q4 1)v f123 = 0.

(3.27)

Similar equations are easily found for f124 , f134 , f234 . 3

3 To compare the results here with those of Pickering and West in [6] which are expressed in terms of functions
f, g, k, l then uq3 +q4 v q1 +q4 1 f123 = f , uq3 +q4 f124 = k, f134 = l and v q1 +q4 1 f234 = g and we should let

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

609

Eqs. (3.26) and (3.27) are identical with a particular generalisation of the differential
equation defining the hypergeometric function to two variables [10]:


x(1 x)fxx y 2 fyy 2xyfxy + ( + + 1)x fx ( + + 1)yfy f
= 0,

(3.28a)


y(1 y)fyy x fxx 2xyfxy + ( + + 1)y fy ( + + 1)xfx f
2

= 0,

(3.28b)

since (3.26) has the same form as (3.28a) and (3.27) corresponds exactly to the difference
of (3.28a) and (3.28b). These equations have four independent solutions:
F4 (, , , 0 ; x, y),
y

1 0

x 1 F4 ( + 1 , + 1 , 2 , 0 ; x, y),

F4 ( + 1 0 , + 1 0 , , 2 0 ; x, y),
0

x 1 y 1 F4 ( + 2 0 , + 2 0 , 2 , 2 0 ; x, y),

(3.29)

with F4 , introduced by Appell in 1880, 4 defined by

F4 (, , , 0 ; x, y) = F4 (, , , 0 ; x, y) =

X ()m+n ()m+n
x my n,
m!n!( )m ( 0 )n

(3.30)

m,n=0

where
( )m =

0( + m)
.
0( )

(3.31)

The series in (3.30) is convergent for |x|1/2 +|y|1/2 < 1. The functions F4 have two crucial,
for later use, symmetry properties:
F4 (, , , 0 ; x, y) = F4 (, , 0 , ; y, x),

(3.32)

which follows trivially from (3.30), and also, under analytic continuation,
F4 (, , , 0 ; x, y)
0( 0 )0( )
(y) F4 (, + 1 0 , , + 1 ; x/y, 1/y)
=
0( 0 )0()
0( 0 )0( )
(y) F4 ( + 1 0 , , , + 1 ; x/y, 1/y). (3.33)
+
0( 0 )0()
With the aid of the solutions of (3.28a,b) given by (3.29) the general solution of (3.26)
and (3.27) may then be written as
f123 (u, v) = aF4 (q4 , 2 q2 , q3 + q4 1, q1 + q4 1; u, v)
+ buq1+q2 1 F4 (2 q3 , q1 + 1, q1 + q2 , q1 + q4 1; u, v)
+ cv q2 +q3 1 F4 (2 q1 , q3 + 1, q3 + q4 1, q2 + q3 ; u, v)
+ duq1 +q2 1 v q2 +q3 1 F4 (q2 + 1, 3 q4 , q1 + q2 , q2 + q3 ; u, v), (3.34)
qi 13 qi . It is difficult to compare our set of eight equations with theirs although their Eqs. (100) and (98) are
identical to the equations corresponding to (3.26) and (3.27) for f134 , and we have verified that their equations
(89)(92) transform appropriately under cyclic permutations using the above relations for f, g, k, l.
4 For historical references, see [11].

610

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

where we have used (3.2) extensively. The complete solution of (3.25ah) is then given by
(3.34) along with
q4
v q1 +q4 1
q1 + q4 1
F4 (q4 + 1, 2 q2 , q3 + q4 1, q1 + q4 ; u, v)
2 q3
uq1 +q2 1 v q1 +q4 1
b
q1 + q4 1
F4 (3 q3 , q1 + 1, q1 + q2 , q1 + q4 ; u, v)
1
c (q2 + q3 1)F4 (2 q1 , q3 , q3 + q4 1, q2 + q3 1; u, v)
q3
q2 + q3 1 q1 +q2 1
u
d
2 q4
F4 (q2 + 1, 2 q4 , q1 + q2 , q2 + q3 1; u, v),
(3.35)
q4 (2 q2 )
uq3 +q4 1 v q1 +q4 1
f134 (u, v) = a
(q3 + q4 1)(q1 + q4 1)
F4 (q4 + 1, 3 q2 , q3 + q4 , q1 + q4 ; u, v)
q1 + q2 1 q1 +q4 1
v
+b
q1 + q4 1
F4 (2 q3 , q1 + 1, q1 + q2 1, q1 + q4 ; u, v)
q2 + q3 1 q3 +q4 1
u
+c
q3 + q4 1
F4 (2 q1 , q3 + 1, q3 + q4 , q2 + q3 1; u, v)
(q1 + q2 1)(q2 + q3 1)
+d
q2 (2 q4 )
(3.36)
F4 (q2 , 2 q4 , q1 + q2 1, q2 + q3 1; u, v),
q4
q3 +q4 1
u
f234 (u, v) = a
q3 + q4 1
F4 (q4 + 1, 2 q2 , q3 + q4 , q1 + q4 1; u, v)
1
b (q1 + q2 1) F4 (2 q3 , q1 , q1 + q2 1, q1 + q4 1; u, v)
q1
2 q1
uq3 +q4 1 v q2 +q3 1
c
q3 + q4 1
F4 (3 q1 , q3 + 1, q3 + q4 , q2 + q3 ; u, v)
q1 + q2 1 q2 +q3 1
v
d
2 q4
F4 (q2 + 1, 2 q4 , q1 + q2 1, q2 + q3 ; u, v).
(3.37)
f124 (u, v) = a

In each case the F4 functions, defined by (3.30), satisfy + = + 0 + 1. 5

5 For particular values of q the infinite series for F truncates. Thus for q = 0 from the terms d we have
i
4
2
f134 =const: while f123 = f124 = f234 = 0, which corresponds to the situation when the four-point function
q 2
reduces to a three-point function. If q1 = 1 there is a solution f123 = buq2 2 , f124 = b q3 2 uq2 2 v q4 2 ,
4

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

611

4. Crossing symmetry relations

Although the above results provide a complete solution of the superconformal Ward
identities, the solutions of the differential equations are further constrained by considering
permutations of zi+ and simultaneously qi . Such permutations act on the invariants u, v so
that in general they are not restricted to the domain of convergence of the F4 functions and
thus analytic continuation is necessary. It is essential that the results be independent of the
path of analytic continuation or that they be monodromy invariant.
Firstly for cyclic permutations, when u v, in (3.9) we require
f123 (q1 , q2 , q3 , q4 ; u, v) = f124 (q2 , q3 , q4 , q1 ; v, u) = f134 (q3 , q4 , q1 , q2 ; u, v)
= f234 (q4 , q1 , q2 , q3 ; v, u),

(4.1)

which, with 3.32, leads to

1
(q3 + q4 1)c(q2, q3 , q4 , q1 ),
q4
2 q4
b(q2, q3 , q4 , q1 ),
d(q1 , q2 , q3 , q4 ) =
q1 + q2 1
q1
a(q2 , q3 , q4 , q1 ),
b(q1, q2 , q3 , q4 ) =
q1 + q2 1
q3 + q4 1
d(q2 , q3 , q4 , q1 ).
c(q1 , q2 , q3 , q4 ) =
2 q1
a(q1, q2 , q3 , q4 ) =

(4.2a)
(4.2b)
(4.2c)
(4.2d)

The critical conditions arise from considering z1+ , q1 z2+ , q2 so that u u0 = u/v,
v v 0 = 1/v. In this case we require
f123 ( q1 , q2 , q3 , q4 ; u, v) = v 0q4 f123 (q2 , q1 , q3 , q4 ; u0 , v 0 ),
f124 ( q1 , q2 , q3 , q4 ; u, v) = v

0q3

f134 ( q1 , q2 , q3 , q4 ; u, v) = v

(4.3b)

(4.3c)

f124 (q2 , q1 , q3 , q4 ; u , v ),

0q1

(4.3a)

f234 (q2 , q1 , q3 , q4 ; u , v ).

Using now (3.33), we get from (4.3a), for one path of analytic continuation:


eiq4 0(q1 + q4 1)
ei(q12) 0(q2 + q3 )
+ c(q1, q2 , q3 , q4 )
a(q1 , q2 , q3 , q4 )
0(q1 1)0(2 q2 )
0(1 q4 )0(q3 + 1)
(4.4a)
0(q1 + q3 1) = a(q2, q1 , q3 , q4 ),


i(q
1)
i(q
+1)
2
3
e
e
0(q1 + q4 1)
0(q2 + q3 )
+ c(q1 , q2 , q3 , q4 )
a(q1 , q2 , q3 , q4 )
0(q3 )0(q4 )
0(q2 1)0(2 q1 )
(4.4b)
0(q2 + q4 2) = c(q2, q1 , q3 , q4 ),


ei(q32) 0(q1 + q4 1)
ei(q2+1) 0(q2 + q3 )
+ d(q1 , q2 , q3 , q4 )
b(q1 , q2 , q3 , q4 )
0(q2 )0(q1 + 1)
0(q3 1)0(3 q4 )
(4.4c)
0(q1 + q3 1) = b(q2, q1 , q3 , q4 ),
q 2

q 2

q 2

f134 = b q2 2 v q4 2 , f234 = b(q2 2)(1 + q3 2 u + q3 2 v). For q2 = 0 as well this coincides, up to misprints,
4
2
4
with the solution (104) in [6].

612

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633



ei(q1 +1) 0(q1 + q4 1)
ei(q43) 0(q2 + q3 )
+ d(q1 , q2 , q3 , q4 )
b(q1 , q2 , q3 , q4 )
0(q4 2)0(2 q3 )
0(q1 )0(q2 + 1)
(4.4d)
0(q2 + q4 2) = d(q2 , q1 , q3 , q4 ).
For consistency the imaginary parts of the left hand sides of (4.4ad) must vanish. In this
case the same result is obtained for the other possible analytic continuation since their
difference is just in the imaginary part. 6 From (4.4a) or (4.4c), using 0(x)0(1 x) =
0(x + 1)0(x) = / sin x, we get
c(q1 , q2 , q3 , q4 ) =

0(q1 + q4 1)0(2 q1 )0(q3 + 1)

a(q1 , q2 , q3 , q4 ),
0(q2 + q3 )0(2 q2 )0(q4 )

(4.5)

while from (4.4c) or (4.4d) we also obtain

d(q1 , q2 , q3 , q4 ) =

0(q1 + q4 1)0(3 q4 )0(q1 + 1)

b(q1 , q2 , q3 , q4 ).
0(q2 + q3 )0(2 q3 )0(q1 + 1)

(4.6)

Using (4.5), (4.4a) reduces to

a(q1 , q2 , q3 , q4 )

0(q1 + q3 1)0(2 q1 )
= a(q2, q1 , q3 , q4 ),
0(q2 + q3 1)0(2 q2 )

(4.7)

while (4.4b) gives an equivalent, by virtue of (4.5), relation for c. Similarly (4.4c) or (4.4d)
give
b(q1 , q2 , q3 , q4 )

0(q1 + q3 1)0(q2 + 1)
= b(q2, q1 , q3 , q4 ).
0(q2 + q3 1)0(q1 + 1)

(4.8)

Identical results may also be obtained from (4.3b) or (4.3c).

To simplify further, it is convenient to define
a(q1 , q2 , q3 , q4 ) = 0(q2 + q3 1)0(q1 + q2 1)0(2 q2 )0(q4 ) A(q1, q2 , q3 , q4 ),
c(q1 , q2 , q3 , q4 ) = 0(q1 + q4 2)0(q1 + q2 1)0(2 q1 )0(q3 + 1)
A(q1 , q2 , q3 , q4 ),

(4.9)

and also
b(q1, q2 , q3 , q4 ) = 0(q2 + q3 1)0(q3 + q4 2)0(2 q3 )0(q1 + 1)
B(q1 , q2 , q3 , q4 ),
d(q1 , q2 , q3 , q4 ) = 0(q1 + q4 2)0(q3 + q4 2)0(3 q4 )0(q2 + 1)
B(q1 , q2 , q3 , q4 ),

(4.10)

where the second lines follow from (4.5) and (4.6), respectively. Now (4.7) and (4.8)
become
A(q1 , q2 , q3 , q4 ) = A(q2 , q1 , q3 , q4 ),
B(q1 , q2 , q3 , q4 ) = B(q2 , q1 , q3 , q4 ).
Inserting (4.9) into (4.2a) and (4.10) into (4.2b) gives
6 For a related discussion in two dimensions, see [12].

(4.11)

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

613

A(q1 , q2 , q3 , q4 ) = A(q2 , q3 , q4 , q1 ),
B(q1 , q2 , q3 , q4 ) = B(q2 , q3 , q4 , q1 ).

(4.12)

Similarly (4.2c) and (4.2d) lead to

B(q1 , q2 , q3 , q4 ) = A(q2 , q3 , q4 , q1 ),
A(q1 , q2 , q3 , q4 ) = B(q2 , q3 , q4 , q1 ).

(4.13)

In consequence we must have B = A so that the chiral four-point function is determined up

to a single overall constant. f123 (u, v) is given by substituting (4.9) and (4.10), with A = B,
into (3.34) while f124 (u, v), f134 (u, v), f234 (u, v) are given by similar expressions which
may easily be found from (4.1).
For completeness we may also consider the permutation z1+ , q1 z3+ , q3 and
z2+ , q2 z4+ , q4 , when u, v are invariant. This requires f123 (q3 , q4 , q1 , q2 ; u, v) =
f134 (q1 , q2 , q3 , q4 ; u, v) and f124 (q3 , q4 , q1 , q2 ; u, v) = f234 (q1 , q2 , q3 , q4 ; u, v). This relates a to d and b to d and with (4.9) and (4.10) would require A(q3 , q4 , q1 , q2 ) =
B(q1 , q2 , q3 , q4 ) and B(q3 , q4 , q1 , q2 ) = A(q1 , q2 , q3 , q4 ).

5. Short-distance expansions
In the limits x1+ x2+ or x3+ x4+ we have, with the definitions in (2.11), u 0,
v 1. Similarly for x1+ x4+ or x2+ x3+ , u 1, v 0 while for x1+ x3+
or x2+ x4+ , 1/v 0, u/v 1. In order to understand the behaviour in these shortdistance limits a different expansion, which reveals the form of the F4 function for one of
the arguments near the singular point 1, than that given by (3.30) is necessary.
In the simpler case of standard hypergeometric functions the relevant results are easily
obtained. The associated second-order ordinary differential equation has three singular
points at 0, 1, . The two independent solutions may be taken as F (, ; ; x) and
x 1 F ( + 1 , + 1 ; 2 ; x), which are thus given as an expansion in powers
of x, but they can be equally expressed in terms of F (, ; + + 1 ; 1 x) and
(1 x) F ( , ; + 1; 1 x), determining the form as x 1, or
(x) F (, + 1 ; + 1 ; 1/x) and (x) F ( + 1 , ; + 1 ; 1/x),
which gives an equivalent expression in terms of 1/x, revealing the behaviour at . Any
of these functions may be written in terms of a linear combination of the two functions
defined by a series expansion at either of the other singular points. For the F4 function,
given by (3.30), (3.33) gives an equivalent result involving the behaviour for one variable
at . To determine an analogous result for the form of F4 , as used here, for one argument
approaching 1 we define the new function
G(, , , ; x, y) =

X ( )m ( )m ()m+n ()m+n
x my n.
m!( )m
n!()2m+n

m,n=0

(5.1)

614

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

Writing the sum over n in terms of a hypergeometric function and using the identity relating
hypergeometric functions of arguments y and 1 y we may obtain 7
G(, , , ; x, 1 y)
0()0( )
F4 (, , , + + 1 ; x, y)
=
0( )0( )
0()0( + )
y
F4 ( , , , + 1; x, y).
+
0()0()

(5.2)

From this it is evident that

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

(5.3)

and using (3.33) and standard 0-function identities we find, from (5.2),
G(, , , ; x, 1 y) = y G(, , , ; x/y, 1 1/y).

(5.4)

Using (5.2) with (3.34) where a, b, c, d are given by (4.9) and (4.10) we may now find
in terms of the function G:
f123 (u, v) =

0(q1 + q2 1)0(2 q1 )0(2 q2 )0(q3 + 1)0(q4 )

A
0(q3 + q4 + 1)
G(q4 , 2 q2 , q3 + q4 1, q3 + q4 + 1; u, 1 v)
0(q3 + q4 2)0(q1 + 1)0(q2 + 1)0(2 q3 )0(3 q4 )
B
+
0(q1 + q2 + 2)

uq1 +q2 1 G(2 q3 , q1 + 1, q1 + q2 , q1 + q2 + 2; u, 1 v), (5.5)

0(q1 + q2 1)0(2 q1 )0(2 q2 )0(q3 )0(q4 + 1)
A
f124 (u, v) =
0(q3 + q4 + 1)
G(q3 , 2 q1 , q3 + q4 1, q3 + q4 + 1; u, 1 v)
0(q3 + q4 2)0(q1 + 1)0(q2 + 1)0(3 q3 )0(2 q4 )
B
+
0(q1 + q2 + 2)
uq1 +q2 1 G(2 q4 , q2 + 1, q1 + q2 , q1 + q2 + 2; u, 1 v), (5.6)
0(q3 + q4 1)0(q1 + 1)0(q2 )0(2 q3 )0(2 q4 )
B
f134 (u, v) =
0(q1 + q2 + 1)
G(q2 , 2 q4 , q1 + q2 1, q1 + q2 + 1; u, 1 v)
0(q1 + q2 2)0(2 q1 )0(3 q2 )0(q3 + 1)0(q4 + 1)
A
+
0(q3 + q4 + 2)
uq3 +q4 1 G(2 q1 , q3 + 1, q3 + q4 , q3 + q4 + 2; u, 1 v), (5.7)
0(q3 + q4 1)0(q1 )0(q2 + 1)0(2 q3 )0(2 q4 )
B
f234 (u, v) =
0(q1 + q2 + 1)
G(q1 , 2 q3 , q1 + q2 1, q1 + q2 + 1; u, 1 v)
7 The behaviour of F functions near y = 1 was investigated quite recently by Exton [13]. The function G is
4
essentially that defined by Exton and the relation obtained here is a special case of his formulae. The series for G
in (5.1) also features in a related discussion in [14].

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

615

0(q1 + q2 2)0(3 q1 )0(2 q2 )0(q3 + 1)0(q4 + 1)

A
0(q3 + q4 + 2)
uq3 +q4 1 G(2 q2 , q4 + 1, q3 + q4 , q3 + q4 + 2; u, 1 v). (5.8)

With these expressions the relations (4.3ac) are easy to verify using (5.3) and (5.4).
Clearly the results are manifestly analytic at v = 1.
In general, if we define
H (, , , ; u, v)
0(1 )
0()0()0( )0( ) G(, , , ; u, 1 v)
=
0()
0( 1)0( + 1)0( + 1)0( + 1)0( + 1)
+
0( 2 + 2)
u1 G( + 1, + 1, 2 , 2 + 2; u, 1 v),

(5.9)

then, from (5.2) and (3.32), we have

H (, , , ; u, v) = H (, , + + 1 , + + 1 ; v, u).

(5.10)

The expression given by 5.9 coincides, for appropriate choices of , , , , with the forms
in (5.5)(5.8) if A = B. From (5.4) we have
H (, , , ; u, v) = v H (, , , ; u/v, 1/v),

(5.11)

and from (5.3) together with (5.10) we may obtain

H (, , , ; u, v)
= v H ( , , , ; u, v)
= u1 v H ( + 1 , + 1 , 2 , 2 + 2; u, v). (5.12)
Together, (5.10)(5.12) are sufficient to obtain the necessary symmetry relations.
If = 1 the definition (5.9) reduces to
H (, , 1, ; u, v)
1
0()0()0( )0( )
=
0()

ln u G(, , 1, ; u, 1 v)
+


X ( )m ( )m ()m+n ()m+n
m
n
f
u
(1

v)
,
mn
(m!)2
n!()2m+n

m,n=0

fmn = 2(1 + m) + 2( + 2m + n) ( + m) ( + m)
( + m + n) ( + m + n),
involving ln u although (5.10)(5.12) remain valid.

(5.13)

616

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

6. Counting for higher N -point functions

The constraints arising from superconformal invariance uniquely determine the form of
the scalar chiral superfield four-point function, as shown in Section 3. We attempt here
to count the number of independent functions of the appropriate generalisations of u, v
and to determine the number of independent differential equations constraining them for
higher-point functions of scalar chiral superfields.
In order to undertake such an analysis the superconformal group is restricted by first
using superconformal transformations to set
x1+ = 0,

x2+ = ,

1 = 0,

2 = 0.

(6.1)

The residual symmetry group is then that generated by matrices of the form (2.2) with
a, b, , zero. The associated subgroup of G0 is given by rotations and scale, U (1)R ,
With (6.1) we have
transformations for which the infinitesimal parameters are ab , , .
1(2i) = i x 1
i .

(6.2)

After imposing (6.1) an N -point function for chiral superfields depends only on zi+ =
(xi+ , i ), i = 3, . . . , N . The rotational and scale invariant variables constructed from zi+
are then easily seen to be
uij =

xj + xi+
2
x3+

j > i > 3,

ui =

2xi+ x3+
,
2
x3+

i > 3.

(6.3)

Evidently there are 12 (N 3)(N 2) such uij and N 3 ui giving 12 N(N 3) in total.
However, since we are concerned only with four-dimensional space, if N > 7 we may
express xi+ , i > 7, in terms of xj + , j = 3, 4, 5, 6, with coefficients involving xi+ xj + .
This ensures that for the uij in (6.3) we may therefore restrict to j = 3, 4, 5, 6 and the
number of such independent invariants becomes N 3 + N 4 + N 5 giving, with
the ui as well, altogether 4N 15. Clearly for N > 7 the 15 parameter conformal group
SU(2, 2) is acting transitively. In fact this result gives the correct number also for N = 5, 6
whereas for N = 4 we have just two invariants u44 , u4 .
P
For application to N -point functions for chiral superfields with i qi = 3 we need a set
of nilpotent monomials O( 2 ), generalising 2i(j k) , i < j < k, used for N = 4. With the
choice in (6.1) we may take as a linearly independent basis
i = i2 ,

i > 3,

ij,r = j Mr i ,

j > i > 3,

(6.4)

where (Mr ) are linearly independent 2 2 matrices constructed from xi+ , which
we assume to have dimension zero. To achieve the form in (6.4) we may note that
e =   M . For N > 5 it is sufficient to take just r =
e i where M
i M j = j M
1, 2, 3, 4 since any such matrix may be expressed in terms of the basis formed by
2 ,x x
2
2 . Clearly there are N 2 independent and
4+ /x3+
1, x4+ x 3+ /x3+
5+ 3+ /x3+ , x5+ x
i
2(N 3)(N 2) independent ij,r giving (2N 5)(N 2) independent such I in
total. The general N -point function can then be written in terms of scalar functions of the
invariants u as

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

2
x2+

q2


h1 (z1+ )2 (z2+ ) . . . N (zN+ )i

x1+ =0, 1 =0
x2+ , 2 =0

2
= x3+

617

q2 2 X

I fI (u).

(6.5)
For N = 4 there are only two independent matrices Mr since they can be restricted to
2 . There are therefore just four independent , which
the basis formed by 1, x4+ x 3+ /x3+
I
can be alternatively expressed in terms 21(23), 21(24), 21(34), 22(34). Thus these results
are in accord with the treatment in Section 3 and the expansion of the four-point function
exhibited in (3.9). For N > 4 it is easy to see that the I cannot all be expressed solely
in terms of 2i(j k) , for N = 5 there are 10 2i(j k) , i < j < k, whereas the basis in (6.4)
gives 15.
The nontrivial constraints in the superconformal Ward identities (3.1) arise from the
terms involving  or . From (2.1) such terms are O( 3 ), and the number of independent
conditions, which involve linear first-order partial differential equations for the fI (u), is
equal to the number of independent monomials of the form 3  or 3 . For N > 4 we have

j2 i Ns ,

j2 i Mr ,

i 6= j, i, j > 3,

(6.6)

where (Ns ) is a linearly independent set of 2 2 matrices. For N > 5, Mr may be

reduced to the basis described above and for N > 6, Ns may be given in terms of the basis
x3 , x4 , x5 , x6 . Thus if N > 6 there are four possible Mr and also four Ns so that the number
of independent monomials of the form (6.6) is 8(N 3)(N 2). In addition to (6.6), for
N > 5, we may also construct
k Mr j i Ns  ,

k Mr j i Ms ,

k > j > i > 3.

(6.7)

The ordering k > j > i is achieved by Fierz type identities, since if i > j we may write
P
j i = r i Mr j M 0r for {Mr } forming a basis for 2 2 matrices and where M 0r
satisfy tr(M 0r Ms ) = rs . The expressions (6.6) and (6.7) are a linearly independent basis.
Assuming both four linearly independent Mr and Ns we have 16
3 (N 4)(N 3)(N 2)
monomials of the form (6.7). Combining these with those exhibited in (6.6) we have
8
3 (N 3)(N 2)(2N 5) in total giving this number of constraints on the fI . Since
there are (N 2)(2N 5) fI depending on 4N 15 variables the number of functional
degrees of freedom remaining after imposing the differential constraints is 13 (N 2)
(2N 5)(4N 21), which for N = 6 is 28.
When N = 4 the relevant monomials are only of the form in (6.6), with i, j = 3, 4, and
2 , and also s = 1, 2, since N
we further restrict to r = 1, 2, corresponding to 1, x4+ x 3+ /x3+
s
is also restricted to the basis x3 , x4 . This gives 8 independent monomials so that there are
8 conditions on the 4 functions of 2 variables fI . In Appendix A we explicitly obtain the
8 relations for f1 , f2 , f3 , f4 which are equivalent to those found earlier in (3.17)(3.24) or
(3.25ah). For N = 4 we therefore expect a unique functional form for the solution, up to
choices of integration constants, which of course is in accord with the results of Section 3.
For N = 5 there are apparently 80 constraints on 15 functions fI of 5 variables leading
to an overdetermined system. However if any qi = 0 the equations should reduce to those
determining the four-point function so that there are possible linear dependencies in the
constraints but this case is too complicated to investigate in detail.

618

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

7. Operator product expansion

In conformal field theories there are further conditions to be obtained by imposing the
operator product expansion. A four-point function involving fields at x1 , x2 , x3 , x4 may
be expanded as a convergent series for x1 x2 , x3 x4 in terms of an infinite set of
quasi-primary fields and also equivalently for x1 x3 , x2 x4 . The equality of the two
expansions provides a constraint on the operator content of the theory. In this section we
consider the contribution of a single chiral quasi-primary superfield and its descendents to
the four-point function of chiral superfields.
The operator product coefficients are determined by the two- and three-point functions.
For scalar fields i , with dimension i in dimension d, then the two-point functions in
a conformal theory may be chosen as

 ij
i (x1 )j (x2 ) = ,
r12i

(7.1)

and the three-point functions are then


i (x1 )j (x2 )k (x3 ) =

Cij k
1
2 (i +j k )

r12

1
2 (j +k i )

r23

(k +i j )

(7.2)

2
r31

The contribution of the field k to the operator product of i and j is then determined
from (1.1) and (7.2) to be
1 (i +j k )

i (x1 )j (x2 ) Cij k r12 2

C2

(k +i j ), 21 (k i +j )

k +1 12 d

(x12, x2 )k (x2 ),

(7.3)

where
C a,b

S+1 12 d

(x12 , x2 )

1
1
= a b ,
S
r23 r13 r23

S = a + b.

(7.4)

Ca,b (s, ) is given as an infinite series in s and s 2 2 in Appendix D, Ca,b (0, ) = 1. For
application to four-point functions we have


1
1  r14 e
=
G b, e, S + 1 12 d, S; u, 1 v
C a,b 1 (x12 , x2 )
f
a
b
S+1 2 d
e
r
r14 r24 13
r23 r24


1  r13 f
G a, f, S + 1 12 d, S; u, 1 v , (7.5)
= a b
r13 r23 r14
with S = a + b = e + f and using (5.3).
Similar results may be obtained for chiral scalar superfields, restricting now to d = 4.
The basic two-point function involves a chiral superfield and its antichiral conjugate [1]:


i (z1+ ) j (z2 ) =

ij
,
(x1+ 2i1 2 x2 )2qi

and the corresponding three-point function has the form

(7.6)

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633


i (z1+ )j (z2+ ) k (z3 ) =

619

Cij k

,
(x1+ 2i1 3 x3 )2qi (x2+ 2i2 3 x3 )2qj
(7.7)

and is only possible if qk = qi +qj . As a consequence of (7.6) and (7.7) the chiral superfield
contributes to the operator product expansion of i and j :
i (z1+ )j (z2+ ) Cij k C q1 ,q2 (z12+ , z2+ )k (z2+ ),

where

1
(x2+ 2i2 x )2q
1
=
,
(x1+ 2i1 x )2q1 (x2+ 2i2 x )2q2

(7.8)

C q1 ,q2 (z12+ , z2+ )

q = q1 + q2 .

As shown in Appendix E we may express C q1 ,q2 as

q1 q1 +1,q2
q1 ,q2
C
(x12+ , x2+ ) +
(x12+ , x2+ )122
C q1 ,q2 (z12+ , z2+ ) = Cq1
q q1
q1 q2
q +1,q2 +1
Cq 1
+
(x12+ , x2+ )12x12+ x2+ 2
2q(q 2 1)
q1 (q1 1) 2 q1 +1,q2
Cq

(x12+ , x2+ )22 ,

4q(q 1) 12

(7.9)

(7.10)

where x = x . The result (7.8) has no singular terms for z1+ z2+ so that the chiral
superfields form a closed algebraic ring with the scale dimension/R-charge additive.
We may use (7.8) to determine the contribution arising from a chiral scalar superfield
i to the four-point function h1 (z1+ )2 (z2+ )3 (z3+ )4 (z4+ )i arising from the operator
product expansion for 1 (z1+ )2 (z2+ ) where qi = q = q1 + q2 = 3 q3 q4 . From (3.3),


i (z2+ )3 (z3+ )4 (z4+ ) = Ci34 22(34)

q2

r34

1q4 1q3
r24

(7.11)

r23

and using (7.8) gives, after lengthy calculations extending those which led to (7.5) which
are described in Appendix E, exactly the form shown in (3.9) with
q1 q2
(2 q4 )uq1
f123 (u, v) = C12i Ci34
q(q 2 1)
G(2 q3 , q1 + 1, q, q + 2; u, 1 v),
q1 q2
(2 q3 )uq1
f124 (u, v) = C12i Ci34
q(q 2 1)
G(2 q4 , q2 + 1, q, q + 2; u, 1 v),
q1
f134 (u, v) = C12 Ci34 G(q2 , 2 q4 , q 1, q + 1; u, 1 v),
q
q2
(7.12)
f234 (u, v) = C12 Ci34 G(q1 , 2 q3 , q 1, q + 1; u, 1 v).
q
This result is identical to that given by (5.5)(5.8) for a suitable choice of the coefficient B.
These terms may therefore be given solely by a chiral superfield in the operator product
expansion of 1 , 2 and, conversely, the A terms may be related to the contribution of
a chiral superfield in the operator product of 3 , 4 .

620

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

8. Superconformal integrals
The results obtained above are illustrated by application to integrals which automatically
define superconformally covariant N -point functions. Partial results were obtained by one
of us earlier [1] and a more complete analysis is undertaken here. As explained in the
introduction, and as exhibited in (1.1), we may straightforwardly define a superconformally
covariant scalar function for two points which is chiral at zi and antichiral at z. By
integration over products of such factors for i = 1, . . . , N , with the appropriate weighting,
and integrating over z we may then obtain a chiral N -point function. Explicitly this gives
Z
N
Y
i
0(qi )
,
(8.1)
IN (z1+ , . . . , zN+ ) = 2 d4 x d2

(xi+ 2i i x )2
i=1
and the qi are constrained by (3.2), since this is necessary to ensure that under
S ) from (1.1) cancel the associated
a superconformal transformation the factors (z
transformation of the measure in (8.1). The technique for dealing with such integrals is
described in Appendix B and, with due account of analytic continuation to Minkowski
P
space, the procedure described there gives, for = i i :
IN (z1+ , . . . , zN+ )


ZY
Z

2
1 X
qi 1 1
2

di i
i j xi+ xj + 2i(i j )
d exp
.
=
2

i<j

(8.2)
The integration is as usual performed by expanding the exponential to order 2 . Since
only appears in xi+ 2ii we may write
IN (z1+ , . . . , zN+ ) =

ZY

q 1

di i i

1 P
1 X
j j k k e i<j i j rij , (8.3)
2

jk

for rij as in (2.7). By carrying out the differentiation (8.3) becomes

IN (z1+ , . . . , zN+ ) = 4

ZY
0

q 1

di i i

1 1 Pi<j i j rij
e
3

X

X
1X
2

i j k i xik+ x kj + j
i j rij
k k ,
2
ij k

ij

(8.4)
and with further rearrangement we find
IN (z1+ , . . . , zN+ ) = 4

ZY
0

q 1

di i i
X

i<j <k

1 1 Pi<j i j rij
e
3

i j k rij rik 2i(j k) .

(8.5)

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

621

The integrals appearing in (8.4) and (8.5) are of the particular form considered by
Symanzik [15] for conformally invariant theories. His discussion is briefly reviewed in
Appendix B and the essential integral, exhibited in (B.5) with (B.2), is applicable in (8.4)
and (8.5) for the particular case of = 3. For such integrals radical simplifications are
possible so that for N > 3 they may be expressed in terms of functions of the harmonic
cross ratios of the rij . For N = 3 we have from (B.6):
I3 (z1+ , z2+ , z3+ ) = 4

21(23) 0(2 q1 )0(2 q2 )0(2 q3 )

,
1q 1q 1q
r23
r 1r 2r 3
23

31

(8.6)

12

which is in accord with the expected form in (3.3). For N = 4 using (B.11) and (5.12)
where appropriate, we have
 2  
1(23) r13 q2 1  r12 q3 1  r12 q4
I4 (z1+ , z2+ , z3+ , z4+ ) = 4
r23
r23
r23
r24
H (q4, 2 q2 , q3 + q4 1, q3 + q4 + 1; u, v)
21(24)  r24 q1 1  r12 q3  r12 q4 1
+
r24
r14
r13
r14
H (q3, 2 q1 , q3 + q4 1, q3 + q4 + 1; u, v)
21(34)  r14 q2  r14 q3 1  r13 q4 1
+
r34
r24
r34
r34
H (q2, 2 q4 , q1 + q2 1, q1 + q2 + 1; u, v)
22(34)  r23 q1  r24 q3 1  r23 q4 1
+
r34
r13
r34
r34

H (q1, 2 q3 , q1 + q2 1, q1 + q2 + 1; u, v) .
(8.7)
This result is identical, up to an overall constant, with the chiral four-point function
constructed in Sections 3 and 4.

9. Conclusion
The results of this paper demonstrate that in some cases conformal invariance, in
conjunction with supersymmetry, is sufficient to determine four-point functions in four
dimensions. Nevertheless the present results are of limited direct relevance in that the
P
unitarity bound for scalar chiral superfields q > 1 is incompatible with i qi = 3 which
is the case when our nontrivial results are obtained. Even the three-point function is
applicable only if qi = 1 which corresponds to free fields. However in some cases it
is possible that the chiral superfields play the role of a potential for physical fields and
the unitarity bound might not then apply. For N = 4 supersymmetry there are various
results which require that four- and higher-point functions of scalar primary superfields are
just products of free fields in some extremal cases [16]. Perhaps for other cases as well

622

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

superconformal symmetry may constrain the functions of the conformal cross ratios u, v
while not requiring just a solution corresponding to free fields.
In this context the two-variable Appell functions F4 together with the related G and H
form a natural set of functions in terms of which four-point correlation functions may be
expressed. The function G clearly arises in the operator product expansion and includes
the contributions of all derivatives of quasi-primary fields. The properties (5.3), (5.4)
and (5.10)(5.12) are clearly what is required to ensure the essential crossing symmetry
conditions on four-point functions.

Acknowledgements
One of us (F.A.D.) would like to thank the EPSRC, the National University of Ireland
and Trinity College, Cambridge for support.

Appendix A. Alternative analysis of four-point functions

In terms of the discussion of Section 5 we may derive an equivalent set of equations
expressing superconformal invariance. We start from



2 q2
h1 (z1+ )2 (z2+ )3 (z3+ )4 (z4+ )i x1+ =0, 1 =0
x2+
x

2
= x3+

q2 2

, =0

2+
2

2
2
3 f1 (v, w) + 4 f2 (v, w) + 24 3 f3 (v, w)

1
+ 24x4 x 3 3 2 f4 (v, w) ,
x3+

(A.1)

for
v=

2
x4+
2
x3+

w=

2x4+ x3+
.
2
x3+

(A.2)

In this particular limit we need consider only the variations in z3+ , z4+ for the terms
involving  , in (2.1). From terms proportional to i32 4 x4  we have
v f1 w f3 (vv q2 + 1)f4 = 0,
from

(A.3)

i32 4 x3 
w f1 + (vv + ww q2 + 2)f3 vw f4 = 0,

(A.4)

from i42 3 x4 
w f2 v f3 w f4 = 0,

(A.5)

from i42 3 x3 
(vv + ww q2 + 2)f2 w f3 + (vv + ww + 2)f4 = 0,
from 32 4 x4 x 3

(A.6)

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

w f1 + w f3 + (vv q2 q3 + 2)f4 = 0,

623

(A.7)

from 42 3 x3 x 4
w f2 + w f3 (vv + ww + q4 + 1)f4 = 0,

(A.8)

from 32 4
(vv + ww + q4 )f1 + (vv + ww q2 q3 + 3)f3 vw f4 = 0,
from

(A.9)

42 3
(vv q2 q3 + 2)f2 (vv + q4 1)f3 vw f4 = 0.

(A.10)

For the results involving it is necessary to include the terms involving q3 3 and q4 4
arising from in the superconformal Ward identity (3.1).
The coefficient functions appearing in (3.9) may be related to those in (3.9) by
v q2 +q3 2 f124 = f2 + f3 f4 ,

f123 = f1 + f3 vf4 ,

uq1 +q2 2 v q2 +q3 2 f134 = f4 ,

uq1 +q2 2 f234 = f3 ,

u = 1 w + v. (A.11)

With the basis in (3.9) the symmetry properties are less evident but the equations obtained
here are equivalent, with (A.11), to those obtained in Section 3.

Appendix B. Conformal integrals

For completeness we recapitulate some of the results of Symanzik [15] concerning
integrals defining conformally invariant N -point functions. For the purposes of this
appendix we assume a Euclidean metric in general d dimensions and define
IN (x1 , . . . , xN ) =

Z
dd x

N
Y
i=1

0(i )
,
(x xi )2i

1
= d,
2

(B.1)

and we require
N
X

i = d = 2,

(B.2)

i=1

which ensures conformal invariance. Using

0(i )
=
(x xi )2i
P

1 i (xxi )2

di i i

and writing i i (x xi )2 = (x
X
i , rij = (xi xj )2 ,
=

we may evaluate the x integral to give

i xi /)2 +

(B.3)
P
i<j

i j rij / for
(B.4)

624

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

ZY
1 P
1
IN (x1 , . . . , xN ) =
di ii 1 e i<j i j rij .

(B.5)

The crucial observation of Symanzik is that when (B.2) holds, the integral (B.5) is
P
unchanged if we take, instead of (B.4), = i i i for any i > 0, not all zero. To verify
P
this we may make the change of variables i = i , with i constrained by i i i = 1.
Q
P
Q
The integration measure in (B.5), i di i 1 = i di ii 1 (1 i i i ) d d1 and
then, carrying out the integration over , the explicit dependence on disappears. This
result is then equivalent to making the above change in .
For N = 3 if we choose = 3 in (B.5) then we may easily carry out the 3 integration
and subsequently integrate 1 , 2 to give [17]
I3 (x1 , x2 , x3 ) =

0( 1 )0( 2 )0( 3 )
r231 r312 r123

(B.6)

For N = 4 we choose = 4 and, following Symanzik [15], write

e

1 2 r12 /4

1
=
2i

c+i
Z

1 2
ds 0(s)
r12
4

s
,

ci

e2 3 r23 /4 =

1
2i

c+i
Z

dt 0(t)

2 3
r23
4

t
,

c < 0,

(B.7)

ci

and then we can evaluate the 4 and subsequently the 1 , 2 , 3 integrations to obtain
1 4 3 4 +4 2
r34
r13
r24

I4 (x1 , x2 , x3 , x4 ) = r14

F4 (u, v),

where u, v are as in (2.11) and

Z
1
ds dt 0(s)0(t)0(s + t + 2 )0(s + t + 4 )
F4 (u, v) =
(2i)2
0(3 + 4 s)0(1 + 4 t) us v t .

(B.8)

(B.9)

By closing the contours in Re s, Re t > 0, using 0(x m) = 0(x)(1)m /(1 x)m , the
integrals may be obtained in terms of the F4 functions defined by (3.30): 8
F4 (u, v) = 0(2 )0( 4 )0(3 + 4 )0(1 + 4 )
F4 (2 , 4 , + 1 3 4 , + 1 1 4 ; u, v)
+ 0( 3 )0(1 )0(3 + 4 )0(2 + 3 )v 1 +4
F4 ( 3 , 1 , + 1 3 4 , + 1 2 3 ; u, v)
+ 0( 1 )0(3 )0(1 + 2 )0(1 + 4 )u3 +4
F4 ( 1 , 3 , + 1 1 2 , + 1 1 4 ; u, v)
8 Such functions were introduced also in a related context by Ferrara et al. in [18] and the essential integral
(B.9) was considered by Davydychev and Tausk [19] who also found its expression in terms of F4 functions.

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

625

+ 0(4 )0( 2 )0(1 + 2 )0(2 + 3 )u3 +4 v 1 +4

F4 (4 , 2 , + 1 1 2 , + 1 2 3 ; u, v).

(B.10)

With the aid of (5.1) and (5.9) this may be reduced just to the function H so that
1  r14 2  r14 3  r13 4
I4 (x1 , x2 , x3 , x4 ) =
r34
r34
r13 r24
H (2 , 4 , 1 + 2 + 1 , 1 + 2 ; u, v)
1  r13 2  r12 3  r12 4
=
r23
r24
r13 r23
H (4 , 2 , 3 + 4 + 1 , 3 + 4 ; u, v),

(B.11)

where the two expressions are related by (5.12). 9

Appendix C. Particular results for G and H

The two variable functions F4 , and also G, H given by (5.1), (5.9), are relatively
unfamiliar although H occurs in various contexts in the evaluation of Feynman integrals
[19,20]. In this appendix, we list some results for G which are relevant and give formulae
for special cases, when they reduce to well-known single-variable functions.
From the definition (5.1) we may directly obtain
u G(, , , ; u, 1 v)
( )( )
G( + 1, + 1, + 1, + 2; u, 1 v),
=
( + 1)

G( + 1, + 1, , + 1; u, 1 v),
v G(, , , ; u, 1 v) =

( + uu + vv )G(, , , ; u, 1 v)
( )
G(, + 1, , + 1; u, 1 v).
=

There are also various recurrence relations:

(C.1)

( + 1)G(, , , ; u, 1 v)
= ( 1)G(, , 1, ; u, 1 v)
+ ( )G( + 1, , , + 1; u, 1 v)
+ ( )( )G(, , , + 1; u, 1 v),

(C.2a)

G(, , , ; u, 1 v)
= G(, + 1, , + 1; u, 1 v) + ( )G(, , , + 1; u, 1 v)
( )
uG( + 1, + 1, + 1, + 2; u, 1 v),

( + 1)

(C.2b)

9 By considering the limit x 2 , when u = r /r , v = r /r , this also gives an expression for the
12 13
23 13
4
integral I3 in (B.1) without the condition (B.2) which then determines 4 in (B.11).

626

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

( )G(, , , ; u, 1 v)
= ( )G(, + 1, , + 1; u, 1 v)
( )G( + 1, , , + 1; u, 1 v),

(C.2c)

( )G(, , , ; u, 1 v)
= ( )( )G(, , , + 1; u, 1 v)
vG( + 1, + 1, , + 1; u, 1 v).

(C.2d)

For the particular case when = + n, n = 0, 1, . . . , the G-function may be reduced to

products of ordinary hypergeometric functions. Defining 10
u = x(1 y),

v = y(1 x),

(C.3)

these originate from the reduction formula for F4 :

F4 (, , , 0 ; u, v) = F (, ; ; x)F (, ; 0; y),

+ = + 0 1.

(C.4)

Applying this in (5.2) and using standard hypergeometric identities gives

G(, , , ; u, 1 v) = F (, ; ; x)F (, ; ; 1 y).

(C.5)

Using (C.1) for v G then gives

G(, , , + 1; u, 1 v)
1
(1 y)F ( 1, 1; ; x)F (, ; + 1; 1 y)
=
1xy


xF (, ; + 1; x)F ( 1, 1; ; 1 y)

1
F ( 1, 1; ; x)F ( 1, ; ; 1 y)
=
1x y 1


F (, 1; ; x)F ( 1, 1; ; 1 y)
( 1)
1
=
1 x y ( 1)( 1)
F ( 1, 1; ; x)F ( 1, 1; 1; 1 y)


F ( 1, 1; 1; x)F ( 1, 1; ; 1 y) .

(C.6)

The results for the four-point functions obtained here are given by the G-function for =
+ 2. Following a similar route as which led to (C.6) we may obtain
G( + 1, + 1, , + 2; u, 1 v)
1
(1 y)2 F ( 1, 1; ; x)F ( + 1, + 1; + 2; 1 y)
=
(1 x y)2


+ x 2 F ( + 1, + 1; + 2; x)F ( 1, 1; ; 1 y)
1

10 To invert we may define = ((1 u v)2 4uv) 2 = 1 x y and = 2/(1 u v + )

= 1/((1 x)(1 y)) and then x = u/(1 + u), y = v/(1 + v).

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

627

+ 1 x(1 y)
(1 x y)3


F (1, 1; ; x) ( 1)F (1, 1; 1; x)
2

F (, ; +1; 1 y)
F (, ; +1; x) F (1, 1; ; 1 y)

( 1)F (1, 1; 1; 1 y) .

(C.7)

The result (C.5) is relevant, according to (B.10), in two dimensions. Using complex
coordinates for this case and defining
=

(z1 z2 )(z3 z4 )
,
(z1 z3 )(z2 z4 )

(C.8)

we have
u = ,

v = (1 )(1 ),

(C.9)

and then from (C.5) we have

G(, , , ; u, 1 v) = F (, ; ; )F (, ; ; ),

(C.10)

exhibiting a holomorphic factorisation. For the crossing symmetric function H given by

(5.9) we have
2
 
N1 F (, ; ; )
H (, , , ; u, v) =
sin
2

+ N2 1 F ( + 1, + 1; 2 ; ) ,
0()0()0( )0( )
,
0( )2
0(1 )0(1 )0( + 1)0( + 1)
,
N2 =
0(2 )2

N1 =

(C.11)

which coincides with expressions previously obtained for two-dimensional conformal fourpoint functions [21].
We may also apply (C.6) to cases of relevance in four dimensions. It is easy to see that
it gives
G(1, 1, 1, 2; u, 1 v) =

y
1
ln
.
1x y 1x

The definition (5.9) for H gives

H (1, 1, 1 + , 2 + ; u, v) =

and using (C.6) with

(C.12)


1

G(1, 1, 1 + , 2 + ; u, 1 v)
sin 
1+

1  v 
G(1, 1, 1 , 2 ; u, 1 v) ,
+
1 u
(C.13)

628

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

wF (1, 1; 2 + ; w)
= (1 + ) ln(1 w) 

2
1
2 (ln(1 w))

+ Li2 (w) + O  2 ,

(C.14)

where Li2 is the dilogarithm function, we have

H (1, 1, 1, 2; u, v)


y
1
ln x(1 y) ln
2Li2 (x) + 2Li2 (1 y) .
(C.15)
=
1xy
1x
This result is relevant acording to (B.11) if N, d = 4 in (B.1) and we assume 1 = 2 =
3 = 4 = 1 and is equivalent with the the results given in [19,20] for this integral.
It is of interest to verify that (C.15) satisfies the symmetry relations (5.10) and (5.11).
For u v when x y we may use Li2 (x) + Li2 (1 x) = 16 2 ln x ln(1 x) whereas
when u u0 = u/v, v v 0 = 1/v we have x 0 = 1 1/y, y 0 = 1/(1 x) we may use the
identity Li2 (x) + Li2 (x/(x 1)) = 12 (ln(1 x))2 .
Appendix D. Operator product expansion calculations
We describe here some details of the calculations involved in applying the operator
product expansion to the four-point function. The essential derivative operator defined by
(7.4) is given explicitly by
Ca,b (s, )

X 1

n
1
=
14 s 2 2
B(a, b)
n!()n
n=0

Z1
d a+n1 (1 )b+n1 es
0

X

n
1
(a)m+n (b)n
=
(s)m 14 s 2 2 .
B(a, b)
m!n!(a + b)m+2n ()n

(D.1)

m,n=0

To verify (7.4) we may use

1
1 2 n 1
= (S)n S + 1 12 d n S+n ,
4 x2
S
r23
r23
so that
C a,b 1 (x12 , x2 )
S+1 2 d

1
r23S

1
=
B(a, b)

(D.2)

Z1
d a1 (1 )b1
0

X (S)n ((1 )r12 )n

.
n! (x23 + x12 )2(S+n)

(D.3)

n=0

With (x23 + x12 )2 = (1 )r23 + r13 (1 )r12 the sum over n is straightforward,
giving ((1 )r23 + r13 )S , and then the integration, with S = a + b, is of the form
Z1
d a1 (1 )b1
0

giving the desired result.

1
1
= B(a, b) a b ,
(s + (1 )t)a+b
s t

(D.4)

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

629

To sketch the calculation leading to (7.5) we follow the method described in [22]
(although related results were obtained long ago in [23]) and first write, by applying similar
methods as which led to (D.3),
C a,b

S+1 12 d

(x12 , x2 )

1
e f
r23 r24
Z1

1
=
B(a, b)B(e, f )

Z1

a1

(1 )

b1

d e1 (1 )f 1

(S)m+n (S + 1 12 d)m+n




m!n! S + 1 12 d m S + 1 12 d n
m,n=0

Am B n
,
(x12 x34 + x24)2(S+m+n)
B = (1 )r34 .

A = (1 )r12 ,

(D.5)

By using (x12 x34 + x24 )2 = A + B + C where C = (1 )r14 + (1 )r23 +

r13 + (1 )(1 )r24 , we may express the second line on the r.h.s. of (D.5) in the
form


(S)2m
AB m
1 X
.
(D.6)
CS
m!(S + 1 12 d)m C 2
m=0

The integration is just as in (D.4) and the integration may be carried out using
Z1

d e1 (1 )f 1

= B(e, f )

1
(1 x)a (1 y)b

X (e)n (b)n  y x n
1
,
(1 x)e
n!(S)n 1 x

(D.7)

n=0

where S = a + b = e + f and we apply this result for x = 1 r13 /r14 , y = 1 r23 /r24
so that (y x)/(1 x) = 1 v. Combining these expressions gives (7.5) with G given by
the series in (5.1).
These results simplify in two dimensions since, using complex coordinates, (D.1)
factorises:
CSa,b (s, ) = 1 F1 (a; S; sz z ) 1 F1 (a; S; sz z ),

S = a + b,

(D.8)

since
1

1 F1 (a; S; z12z2 ) S
z23

1
a zb
z12
23

For this case

1 F1 (a; S; z12z2 )

1
f

e z
z23
24

1
a b
z14
z24

(D.9)


z14
z13

e
F (b, e; S; ),

(D.10)

630

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

with defined by (C.8). This result is in accord with the holomorphic factorisation required
by (C.5).
For application to the supersymmetric case we also consider, when d = 4, the extension
to spinor fields. For this we require
C a,b (x12, x2 )

(x23 )
S+1
r23

(x13)
a+1 b

(D.11)

r13 r23

This has the solution

a+1,b
(x12 , x2 )
C a,b (x12, x2 ) = CS1
b
C a+1,b+1 (x12 , x2 ) 12 (x12 x2 )
+ 2
S 1 S
and, in a similar fashion to the derivation of (7.5),

C a,b (x12 , x2 )

(D.12)

(x23)
e+1 f

r23 r24
 e+1 
1
r14
= a+1 b
(x13 ) G(b, e + 1, S 1, S + 1; u, 1 v)
r14 r24 r13



bf x12 x1
42 x43
G(b + 1, e + 1, S, S + 2; u, 1 v) .
+
S2 1
(D.13)

Appendix E. Supersymmetric calculations

The application of the operator product expansion to the supersymmetric case requires
an extension of the previous results. We first derive the expression (7.10) which is required
to satisfy (7.9). To achieve this we use
1
12 2 2
12 x
1
=
+
4iq
+
4q
(q

1)
,
(E.1)
1
1
1
x 2q1
x 2(q1+1)
x 2(q1+1)
(x 2i12 )2q1
in (7.9) with x = x1+ 2i2 x . The results (7.4) and (D.11) then show, since
14 22 22 = 1, that (7.9) is satisfied if
q ,q

1 2
(x12+ , x2+ ) +
C q1 ,q2 (z12+ , z2+ ) = Cq1

q1
12C q1 ,q2 (x12+ , x2+ )2
q

q1 (q1 1) 2 q1 +1,q2
Cq
(x12+ , x2+ )22 ,
4q(q 1) 12

(E.2)

which is identical with (7.10).

The results in (7.12) are obtained by calculating
C q1 ,q2 (z12+ , z2+ )

22(34)

q2
r .
1q4 1q3 34
r23 r24

2 ) are easily seen, using (7.5), to be

The terms O(12

(E.3)

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

631

q1

r
q1 (q1 1) 2 q1 +1,q2
12 Cq
(x12+ , x2+ ) 1q34 1q
q(q 1)
r23 4 r24 3

q1 
q1 (q1 1) 2 r34
r14 2q4

=
G(2 q4 , q2 , q, q + 1; u, 1 v),
q(q 1) 12 rq1 +1 rq2 r13
14

(E.4)

24

where, using (5.3) and (C.2a),

1 q1
G(2 q4 , q2 , q, q + 1; u, 1 v)
q2
1 q1 q1 +q4 1
v
G(2 q3 , q1 + 1, q, q + 1; u, 1 v)
=
q2
2 q4 q1 +q4 1
v
G(2 q3 , q1 + 1, q, q + 2; u, 1 v)
=
q +1
2 q3
G(2 q4 , q2 + 1, q, q + 2; u, 1 v)
+
q +1
1
+ (1 q) G(2 q4 , q2 , q 1, q + 1; u, 1 v).
q2

(E.5)

The O(12) terms arise from

2

23 + x23+ `34 q1
q1
12 C q1 ,q2 (x12+ , x2+ ) 2q 2q r34 ,
q
r 4r 3
23

(E.6)

24

with C q1 ,q2 (x12+ , x2+ ) given by (D.12). For the terms not involving 23 (D.13) gives

q1 
q1 r34
r14 2q4
12 x13+ `34 G(q2 , 2 q4 , q 1, q + 1; u, 1 v)
2
q rq1 +1 rq2 r13
24
14
q2 (2 q3 )

12 x12+ x1
+
42+ x43+ `34
q2 1

G(q2 + 1, 2 q4 , q, q + 2; u, 1 v) .

(E.7)

The remaining terms involving 23 may be calculated from this using

(x23+ )
2q
r23 4

x3+

= 2q4

1
2q
r23 4

Using both the derivative relations (C.1) and (C.2b) we get



q1 
r14 2q4
q1 r34
12 23 G(q2 , 2 q4 , q 1, q + 1; u, 1 v)
2
q r q1 +1 r q2 r13
24
14
q2 (2 q4 )
q1 +q4 1
12 x12+x1
+
32+ 23 v
q2 1
G(q1 + 1, 2 q3 , q, q + 2; u, 1 v)
q2 (2 q3 )

12 x12+x1
+
42+ 23
q2 1

(E.8)

632

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633


G(q2 + 1, 2 q4 , q, q + 2; u, 1 v) .

(E.9)

The remaining terms arise from

q ,q

1 2
(x12+ , x2+ )
Cq1

22(34)

q2
r
,
1q4 1q3 34
r24

(E.10)

r23

where it is convenient to write

22(34) =

1
2
2
r24 + 24
22324 r23 + 223x23+ x 34+ 24 .
23
r23 r24

(E.11)

2 and 2 terms are found from (7.5) to be

The 23
24
q2

r34

 r 1q4  r
14

q q
r141 r242 r13

14 2
23 G(2 q4 , q2 , q

r13

2
G(1 q4 , q2 , q
+ 24

1, q; u, 1 v)


1, q; u, 1 v) .

(E.12)

Using (5.3) and (C.2b) we may rewrite the G-functions as

G(2 q4 , q2 , q 1, q; u, 1 v)
q1
= G(2 q4 , q2 , q 1, q + 1; u, 1 v)
q
q2
+ v q1 +q4 2 G(2 q3 , q1 , q 1, q + 1; u, 1 v)
q
q1 q2 (2 q4 ) q1 +q4 2
uv
G(2 q3 , q1 + 1, q, q + 2; u, 1 v),

q(q 2 1)
G(1 q4 , q2 , q 1, q; u, 1 v)
q1
= G(2 q4 , q2 , q 1, q + 1; u, 1 v)
q
q2
+ v q1 +q4 1 G(2 q3 , q1 , q 1, q + 1; u, 1 v)
q
q1 q2 (2 q3 )
uG(2 q4 , q2 + 1, q, q + 2; u, 1 v).

q(q 2 1)

(E.13)

For the other terms present in (E.11), we may use

x23
2q
r23 4

1
1

3.
1q
4
2(1 q4 ) r
23

(E.14)

2 in (E.12) we then obtain

From the derivative of the terms 24

 r 1q4 
q1
14

23 x13+ x1
q q
41+ 24 q G(2 q4 , q2 , q 1, q + 1; u, 1 v)
r141 r242 r13

1 q2
+ 23x23+ x42+ 24 G(2 q4 , q2 + 1, q 1, q + 1; u, 1 v) .
q
(E.15)
q2

r34

F.A. Dolan, H. Osborn / Nuclear Physics B 593 (2001) 599633

633

The results (E.4), with (E.5), (E.7), (E.9), (E.12), with (E.13), and (E.15) give exactly the
superconformal form with (7.12).

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]

[17]
[18]
[19]
[20]
[21]
[22]
[23]

H. Osborn, Ann. Phys. (NY) 272 (1999) 243, hep-th/9808041.

H. Osborn, A. Petkou, Ann. Phys. (NY) 231 (1994) 311, hep-th/9307010.
J. Erdmenger, H. Osborn, Nucl. Phys. B 483 (1997) 431, hep-th/9605009.
J-H. Park, Nucl. Phys. B 539 (1999) 599, hep-th/9807176;
J-H. Park, Nucl. Phys. B 559 (1999) 455, hep-th/9903230.
S.M. Kuzenko, S. Theisen, Class. Quantum. Gravity 17 (2000) 665, hep-th/9907107.
A. Pickering, P. West, Nucl. Phys. B 569 (2000) 303, hep-th/9904076.
B.P. Conlong, P.C. West, J. Phys. A 26 (1993) 3325.
J. Wess, J. Bagger, Supersymmetry and Supergravity, Princeton University Press, Princeton,
1983.
I.L. Buchbinder, S.M. Kuzenko, Ideas and Methods of Supersymmetry and Supergravity, IOP
Publishing, Bristol, 1995.
A. Erdlyi (Ed.), Higher Transcendental Functions, Vol. I, McGraw-Hill, New York, 1953,
pp. 222242.
P. Appell, J. Kamp de Friet, Fonctions hypergometriques et hypersphriques: polynomes
dHermite, Gauthier-Villars, Paris, 1926.
P. Di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Springer, New York, 1996.
H. Exton, J. Physics A 28 (1995) 631.
K. Lang, W. Rhl, Nucl. Phys. B 377 (1992) 371.
K. Symanzik, Lett. Nuovo Cimento 3 (1972) 734.
E. DHoker, D.Z. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, in: M.A. Shifman (Ed.), The
Many Faces of the Superworld, World Scientific, Singapore, 2000, hep-th/9908160;
B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Phys. Lett. B 472 (2000) 323, hepth/9910150;
B. Eden, P.S. Howe, E. Sokatchev, P.C. West, hep-th/0004102;
M. Bianchi, S. Kovacs, Phys. Lett. B 468 (1999) 102, hep-th/9910016;
J. Erdmenger, M. Prez-Victoria, Phys. Rev. D 62 (2000) 045008, hep-th/9912250;
B. Eden, C. Schubert, E. Sokatchev, Phys. Lett. B 482 (2000) 309, hep-th/0003096;
E. DHoker, J. Erdmenger, D.Z. Freedman, M. Prez-Victoria, Nucl. Phys. B 589 (2000) 3,
hep-th/0003218.
M. DEramo, G. Parisi, L. Peliti, Lett. Nuovo Cimento 2 (1971) 878.
S. Ferrara, A.F. Grillo, R. Gatto, G. Parisi, Nuovo Cimento 19A (1974) 667.
A.I. Davydychev, J.B. Tausk, Nucl. Phys. B 397 (1993) 133.
N.I. Ussyukina, A.I. Davydychev, Phys. Lett. B 298 (1993) 363, Phys. Lett. B 305 (1993) 136.
Vl.S. Dotsenko, Nucl. Phys. B 235 [FS11] (1984) 54.
A.C. Petkou, Ann. Phys. (NY) 249 (1996) 180, hep-th/9410093.
S. Ferrara, A.F. Grillo, R. Gatto, G. Parisi, Nucl. Phys. B 49 (1972) 77.

Nuclear Physics B 593 (2001) 634650

www.elsevier.nl/locate/npe

Comments on N = 4 superconformal algebras

Jrgen Rasmussen
Physics Department, University of Lethbridge, Lethbridge, Alberta, T1K 3M4, Canada
Received 20 March 2000; accepted 25 September 2000

Abstract
We present a new and asymmetric N = 4 superconformal algebra for arbitrary central charge,
thus completing our recent work on its classical analogue with vanishing central charge. Besides the
Virasoro generator and 4 supercurrents, the algebra consists of an internal SU(2) U (1) KacMoody
algebra in addition to two spin 1/2 fermions and a bosonic scalar. The algebra is shown to be invariant
under a linear twist of the generators, except for a unique value of the continuous twist parameter.
At this value, the invariance is broken and the algebra collapses to the small N = 4 superconformal
algebra. The asymmetric N = 4 superconformal algebra may be seen as induced by an affine SL(2|2)
current superalgebra. Replacing SL(2|2) with the coset SL(2|2)/U (1), results directly in the small
N = 4 superconformal algebra. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Hf
Keywords: Superconformal field theory; Free field realization

1. Introduction
We have recently outlined an explicit construction of an infinite-dimensional class
of superconformal algebras (SCAs) [1]. These N -extended SCAs are induced by affine
SL(2|N/2) current superalgebras (N is even). Our construction generalizes a work
by Ito [2] in which N = 1, 2 and 4 SCAs are studied. Our constructions are both
supersymmetric extensions of the GiveonKutasovSeiberg construction of the Virasoro
algebra [3]. A related approach to building N = 1, 2 and 4 SCAs for fixed central charges
may be found in Ref. [4].
In the present paper we shall complete our construction of the new and asymmetric
N = 4 SCA discovered in Ref. [1], where the case of vanishing central charge was treated.
Here we shall provide the algebra for generic central charge. Besides the Virasoro generator
and 4 supercurrents, the algebra consists of an internal SU(2) U (1) KacMoody algebra
E-mail address: [email protected] (J. Rasmussen).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 3 7 - 4

J. Rasmussen / Nuclear Physics B 593 (2001) 634650

635

in addition to two spin 1/2 fermions and a bosonic scalar. Hence, it is not included in the
standard classification of N = 4 SCAs [59].
Among N -extended SCAs, the case N = 4 is particularly interesting. As already
mentioned, the new and asymmetric N = 4 SCA may be induced by an SL(2|2) current
superalgebra. We shall also show that using the coset SL(2|2)/U (1) instead of SL(2|2),
induces the well-known small N = 4 SCA rather than the bigger, asymmetric N = 4 SCA.
We hope this will clarify the incompatibility between our work and the result for N = 4 in
Ref. [2]; there a construction (similar to ours) based on SL(2|2) was announced to result in
the small N = 4 SCA.
The new and asymmetric N = 4 SCA is invariant under a one-parameter twist. The
starting point for this observation is a linear modification of the GiveonKutasovSeiberg
Virasoro generator. The remaining twisted generators are obtained by requiring the SCA
to be invariant. All twisted generators are given in terms of linear combinations of the
original untwisted generators. It is an interesting observation that for precisely one value
of the continuous twist parameter, the invariance is broken and the twisting results in the
small N = 4 SCA. It is stressed that the invariance and this collapse of the asymmetric
SCA to the small N = 4 SCA, are both independent of our construction and rely solely on
the defining (anti-)commutators of the two SCAs.
The rest of this paper is organized as follows. In Section 2 we summarize briefly some
of our results [1] on the general construction of SCAs induced by affine SL(2|N/2) current
superalgebras.
Section 3 is devoted to N = 4 SCAs, where the new and asymmetric algebra is provided.
The small N = 4 is shown to be induced by the coset SL(2|2)/U (1) current superalgebra,
and the invariance of the asymmetric SCA is discussed.
Section 4 contains concluding remarks, while details on the Lie superalgebra sl(2|M)
along with explicit free field realizations of the currents, are given in Appendices A, B
and C.

2. N -extended superconformal algebras

2.1. Free field realizations of affine current superalgebras
Associated to a Lie superalgebra g is an affine Lie superalgebra characterized by
the central extension k. Associated to an affine Lie superalgebra is an affine current
superalgebra whose generators, Ja , are conformal spin one primary fields (with respect
to the Sugawara construction) and have the mutual operator product expansions (OPEs)
Ja (z)Jb (w) =

a,b k
fa,b c Jc (w)
,
+
(z w)2
zw

(1)

a,b and fa,b c are the CartanKilling form and structure constants, respectively, of the
underlying Lie superalgebra. Regular terms have been omitted. The general free field
realization obtained in Ref. [10] is based on a pair of free ghost fields ( , ) of conformal
weights (1, 0) for every positive root + (written > 0), and on a free scalar boson i

636

J. Rasmussen / Nuclear Physics B 593 (2001) 634650

for every Cartan index i = 1, . . . , r, where r is the rank of the underlying Lie superalgebra.
They satisfy the OPEs
0

(z) (w) =

,
zw

i (z)j (w) = i,j ln(z w).

(2)

Note that in this notation, the ghost fields ( , ) associated to odd roots 1+ are
fermionic. (1 ) 0 denotes the space of positive or negative (odd) even roots of the
underlying Lie superalgebra, and = 0 1 . The corresponding energymomentum
tensor is
X
1
1
+
2 ,
(3)
T=
2
k + h
>0

are the Weyl vector and the dual Coxeter number, respectively, of the underlying
and
Lie superalgebra. Normal ordering is implicit. The generators of the affine current
superalgebra are realized according to [10]
Ja (z) =

r
X

j
Va (z) (z) + k + h
Pa (z) j (z) + Jaanom (z), (z) ,
j =1

>0

(4)
where
Jaanom

0 for a = i, > 0,

X
0
(z), (z) =
(z)F, 0 ( (z)) for a = < 0.

(5)

0 >0

The explicit form of F, 0 is not needed here but may be found in Refs. [1,10]. For = i
a simple root Fi , 0 is a constant independent of the ghost fields :


1
(6)
Fi , 0 ( ) = i , 0 (2k + h )i ,i Aii .
2
Aij is the Cartan matrix of the underlying Lie superalgebra, and is related to the Cartan
Killing form as i,j = Aij j ,j . V and P are given by

 0

 0
0
0
Vi ( ) = C( ) i ,
V ( ) = B(C( )) ,
X
 00 
 0
0
( ) =
eC( ) B(C( )) 00 ,
V
00 >0

Pj ( ) = 0,

Pi ( ) = i ,


j
j
P ( ) = eC( ) .

B(u) is the generating function for the Bernoulli numbers Bn

X Bn
u
=
un ,
B(u) = u
e 1
n!

(7)

(8)

n>0

whereas the matrix C is defined by

X
f,a b .
Cab ( ) =
>0

The formal power series (7) all truncate and are thus polynomials.

(9)

J. Rasmussen / Nuclear Physics B 593 (2001) 634650

637

2.2. Superconformal algebra generators

Most Lie superalgebras with even subalgebra g0 = sl(2) g0 have the property that the
embedding of sl(2) in g carried by the odd generators (the set of which is denoted g1 ) is
a fundamental (spin 1/2) representation. 1 This means that the space of odd roots may be
divided into two parts
1 = 1 1+ .

(10)

The roots 1 are characterized by

sl(2)
2
sl(2)

1
= ,
2

(11)

and we have the correspondence

1+ = sl(2) + 1 .

(12)

Here sl(2) is the positive root associated to the embedded sl(2). In particular, the Lie
superalgebra sl(2|N/2) allows such a decomposition of the root space. In the distinguished
representation of sl(2|N/2) discussed in Appendix A, the embedding is associated to the
simple root 1 , while the only odd simple root is 2 . Furthermore, the root space enjoys
the refined decomposition
1+
1+ = 1
+ + .

(13)

This refinement is not valid for all Lie superalgebras respecting (10), as osp(1|2) illustrates.
In the following we shall concentrate on SCAs induced by affine SL(2|N/2) current
superalgebras.
Now, the Virasoro algebra is induced by the embedded SL(2) and is generated by
I
dz
Ln (z),
Ln =
2i

n+1

n

n1
E1 + a3 (n) 1 H1 + a (n) 1
F1 .
(14)
Ln = a+ (n) 1
The central charge is
c = 6k1 p1 ,
p1 is the winding number
I
dz 1
,
p1 =
2i 1

(15)

(16)

while k1 = 1 ,1 k is the level of the embedded sl(2) or the level in the direction 1 . The
constants a+ , a3 and a are defined by






1
1
1
a3 (n) = 1 n2 ,
a (n) = n + n2 .
(17)
a+ (n) = n n2 ,
2
2
2
1 This is true for all basic Lie superalgebras with even subalgebra g0 = sl(2) g0 except osp(3|2M) where the
embedding is a spin 1 representation, see, e.g., [11].

638

J. Rasmussen / Nuclear Physics B 593 (2001) 634650

In Ref. [1] it was found that for each pair of roots ( , + ) we have a pair of supercurrents
of spin 3/2 with respect to (14):
I
dz
G ;n+1/2 (z),
G ;n+1/2 =
2i

n

n+1
E +
(18)
G ;n+1/2 = (n + 1) 1 E n 1
and

S ;n1/2
G

dz S
G ;n1/2 (z),
2i

n

n1
= (n 1) 1 F + n 1
F +






1 n2 1
1 2
n(n 1)
V
E1 + 1 H1 F1

n2 X

2

1 V1 + 1 V1 V
+ n(n 1) 1
1

S ;n1/2 =
G

, >0
1

V
.

(19)

Their anti-commutators are

{G ;n+1/2 , G ;m+1/2 } = 0,
and

S ;m1/2 = , Ln+m + (n m + 1)K ; ;n+m
G ;n+1/2 , G
1
+ cn(n + 1)n+m,0 , ,
6
where the primary spin 1 (with respect to (14)) current K is defined by
I
dz
K ; ;n (z)
K ; ;n =
2i

n1 1

n
V ( 1 E + E ) 1 f , c Jc
K ; ;n = n 1

(20)

(21)

n1

2




1
+ , 1
n 1 E1 + 1 H1 F1 1 H1
2

n1 X


1

(22)
1 V + V V
+ n 1

, >0

It was also shown in Ref. [1], that the remaining anti-commutators are generally nonvanishing:


S ;m1/2 6= 0, for 6= , n 6= m, n + m 6= 1.
S ;n1/2 , G
(23)
G
This asymmetric property of the SCAs induced by SL(2|N/2) will be illustrated in the
following where we consider the case N = 4.
3. N = 4 superconformal algebras
Here we shall specialize the general considerations on SL(2|N/2) in Section 2 to the case
N = 4. The resulting SCA is of a new and asymmetric form. Its classical and centerless

J. Rasmussen / Nuclear Physics B 593 (2001) 634650

639

analogue has recently been obtained in Ref. [1]. Below we shall provide the full SCA
for arbitrary central charge. We shall furthermore show that replacing the original SL(2|2)
with the coset SL(2|2)/U (1) reduces the asymmetric N = 4 SCA to the well known small
N = 4 SCA. An invariance of the asymmetric N = 4 SCA is also presented.
3.1. Asymmetric N = 4 SCA from SL(2|2)
In the distinguished representation discussed in Appendix A, the Cartan matrix and the
CartanKilling form of the Lie superalgebra SL(2|2) are

2 1 0
2 1
0
i,j = 1
(24)
Aij = 1
0 1 ,
0 1 .
0 1 2
0 1 2
The dual Coxeter number is h = 0 while the number of supercurrents is 2|1
+ | = 4.
Accordingly, there are a priori 4 generators, K, of the internal KacMoody algebra. Using
these facts, with the explicit realizations of the Virasoro generators (14), the supercurrents
(18) and (19), and the affine currents (22) (given in Appendix C), one may work out the
entire SCA. We find that closure is ensured by the following set of generators
Virasoro generator L
S2 , G
S(2 +3 )
supercurrents G2 , G2 +3 , G
e
affine SU(2) E = K2 +3 ;2 ,
e = K2 +3 ;(2 +3 ) K2 ;2 ,
H
e = K2 ;(2 +3 )
F
affine U (1) U = K2 +3 ;(2 +3 ) + K2 ;2
fermions 2 , (2 +3 )
scalar S

h = 2,
h = 3/2,

h = 1,
h = 1,
h = 1/2,
h = 0,

and that the non-trivial (anti-)commutators are

c 3
n n n+m,0 ,
[Ln , Lm ] = (n m)Ln+m +
12

[Ln , Am ] = (h(A) 1)n m An+m ,



S ;m1/2 = 0,
S ;n1/2 , G
G ;n+1/2 , G ;m+1/2 = G


S ;m1/2 = , Ln+m + (n m + 1)K ; ;n+m
G ;n+1/2 , G
1
+ cn(n + 1)n+m,0 , ,
6

S
S
G2 ;n1/2 , G(2 +3 );m1/2 = (n m)(n + m 1)Sn+m1 ,




em = 2E
en , F
em = 2F
en+m ,
en+m ,
en , E
H
H



em = H
en , H
em = 1 cnn+m,0 ,
en+m + 1 cnn+m,0 ,
en , F
H
E
6
3




en , G2 +3 ;m+1/2 = G2 ;n+m+1/2 ,
en , G2 ;m+1/2 = G2 +3 ;n+m+1/2 ,
F
E


en , G2 ;m+1/2 = G2 ;n+m+1/2 ,
H


(25)

640

J. Rasmussen / Nuclear Physics B 593 (2001) 634650


en , G2 +3 ;m+1/2 = G2 +3 ;n+m+1/2 ,
H


S(2 +3 );m1/2 = G
S2 ;n+m1/2 n2 ;n+m1/2 ,
en , G
E


S2 ;m1/2 = G
S2 ;n+m1/2 + n2 ;n+m1/2 ,
en , G
H


S( + );m1/2 = G
S( + );n+m1/2 n( + );n+m1/2 ,
en , G
H
2
3
2
3
2
3


S ;m1/2 = G
S( + );n+m1/2 n( + );n+m1/2 ,
en , G
F
2
2
3
2
3


S ;m1/2 = n ;n+m1/2 ,
Un , G


Sn , G2 ;m+1/2 = (2 +3 );n+m+1/2 ,


Sn , G2 +3 ;m+1/2 = 2 ;n+m+1/2 ,




G2 +3 ;n+1/2 , (2 +3 );m1/2 = Un+m ,
G2 ;n+1/2 , 2 ;m1/2 = Un+m ,


S ;n1/2 , ( + );m1/2 = (n + m 1)Sn+m1 ,
G
2
2
3


S( + );n1/2, ;m1/2 = (n + m 1)Sn+m1 ,
G
2
3
2


en , (2 +3 );m1/2 = 2 ;n+m1/2 ,
E




en , 2 ;m1/2 = 2 ;n+m1/2 ,
en , 2 ;m1/2 = (2 +3 );n+m1/2 ,
H
F


en , ( + );m1/2 = ( + );n+m1/2 ,
H
2
3
2
3


(26)
[Un , Um ] = Sn , Sm = 0.
Am denotes any of the 11 generators listed in (25) different from the Virasoro generator. In
the derivation we have used that integrating a total derivative gives zero. In particular, we
find


S(2 +3 );m1/2
S2 ;n1/2 , G
G
= (n m)(n + m 1)


I

1

i
dz h n+m2
1

(z)V

(z)
V

(z)
Sn+m1 +
(2 +3 )
2
2i z
= (n m)(n + m 1)Sn+m1 .

(27)

1
1
and V(
are given (C.9) in Appendix C.
The polynomials V
2
2 +3 )
It is observed that this N = 4 SCA (26) is a new and asymmetric one not contained in
the standard classification of N = 4 SCAs [59]. Besides being asymmetric in the way
S supercurrents are treated, it involves the unfamiliar number two of spin 1/2
the G and G
fermions. We recall that the small N = 4 SCA does not contain any spin 1/2 fermions,
whereas the big N = 4 SCAs are characterized by containing 4 such generators.

3.2. Small N = 4 SCA from coset SL(2|2)/U (1)

Among the Lie superalgebras sl(2|M), M > 1, sl(2|2) is the only one having a nontrivial center. This may be seen easily by considering the associated Cartan matrices; they
are all invertible except when M = 2. By simple inspection of the Cartan matrix for sl(2|2)
(24) it follows that the Lie superalgebra element
Hu(1) = H1 + 2H2 H3 ,

(28)

J. Rasmussen / Nuclear Physics B 593 (2001) 634650

641

generates the center u(1) of sl(2|2). The coset algebra sl(2|2)/u(1) may therefore be
realized straightforwardly in terms of the generators of the original sl(2|2). All that one
has to do is to invoke the vanishing of the generator of the center (28), and otherwise make
no changes. It should be noted that the root space of the coset algebra is identified with the
root space of sl(2|2). In particular, the sets of simple roots are identical, despite the fact
that the rank of the coset algebra is one smaller than the rank of sl(2|2), the latter being
r = 3.
The associated affine SL(2|2)/U (1) current superalgebra is equally simple to realize.
By construction, its Virasoro generator TSL(2|2)/U (1) = TSL(2|2) TU (1) has vanishing OPE
with the current
HU (1)(z) = H1 (z) + 2H2 (z) H3 (z)

= k 1 (z) + 22(z) 3 (z) ,

(29)

while TU (1) has vanishing OPEs with all other affine currents. Thus, it makes sense from
the conformal field theory point of view to put the central current (29) equal to zero
without modifying the Virasoro generator of the original SL(2|2) current superalgebra,
but by imposing the (coset) condition
1 + 22 3 0.

(30)

In this way the coset current superalgebra has the same free field realization as the original
current superalgebra, though subject to the coset condition (30). Thus, at the level of the
free field realization the coset condition (30) reflects modding out the U (1) center of
SL(2|2).
From the explicit realization of the generators of the asymmetric N = 4 SCA induced by
SL(2|2) (see Appendix C) it follows that imposing the coset condition (30) has fundamental
consequences for the resulting N = 4 SCA. Even the number of generators is reduced as
the affine U (1) subalgebra generated by U (C.6), the 2 spin 1/2 fermions (C.7), and the
bosonic scalar S (C.8) all vanish identically
Un 2 ;n1/2 (2 +3 );n1/2 Sn 0.

(31)

We conclude that the N = 4 SCA induced by the affine SL(2|2)/U (1) current superalgebra
is generated by
Virasoro generator L
S2 , G
S( + )
supercurrents G2 , G2 +3 , G
2
3
e = K + ; ,
affine SU(2) E
2
3
2
e = K + ;( + ) K ; ,
H
2
3
2
3
2
2
e
F = K2 ;(2 +3 )

h = 2,
h = 3/2,

h = 1,

where U = K2 +3 ;(2 +3 ) + K2 ;2 = 0. The non-trivial (anti-)commutators are

c 3
n n n+m,0 ,
12

[Ln , Am ] = (h(A) 1)n m An+m ,
[Ln , Lm ] = (n m)Ln+m +

(32)

642

J. Rasmussen / Nuclear Physics B 593 (2001) 634650



S ;m1/2 = 0,
S ;n1/2 , G
G ;n+1/2 , G ;m+1/2 = G


S ;m1/2 = , Ln+m + (n m + 1)K ; ;n+m
G ;n+1/2 , G
1
+ cn(n + 1)n+m,0 , ,
6




en , G2 +3 ;m+1/2 = G2 ;n+m+1/2 ,
en , G2 ;m+1/2 = G2 +3 ;n+m+1/2 ,
F
E




en , G + ;m+1/2 = G + ;n+m+1/2 ,
en , G ;m+1/2 = G ;n+m+1/2 ,
H
H
2
2
2
3
2
3


S( + );m1/2 = G
S ;n+m1/2 ,
en , G
E
2
3
2


S ;m1/2 = G
S( + );n+m1/2 ,
en , G
F
2
2
3


S ;m1/2 = G
S ;n+m1/2 ,
en , G
H
2
2


S
S( + );n+m1/2 ,
e
Hn , G(2 +3 );m1/2 = G
2
3




e
e
e
e
en+m ,
e
Hn , Fm = 2F
Hn , Em = 2En+m ,




em = H
en , H
em = 1 cnn+m,0 .
en+m + 1 cnn+m,0 ,
en , F
H
(33)
E
6
3
Am denotes any of the 7 generators listed in (32) different from the Virasoro generator. The
SCA (33) is recognized as the well known small N = 4 SCA thus proving our assertion.
3.3. An invariance and a reduction
It turns out that the new and asymmetric N = 4 SCA possesses an invariance which
may be revealed by considering the consequences of modifying the Virasoro generators
according to
Ln = Ln + (n + 1)Un ,

(34)

with arbitrary. The Virasoro algebra is easily seen to be generated with unchanged central
charge
c = c.

(35)

We find that the asymmetric N = 4 SCA (26) is invariant under the following oneparameter twist of its generators
Ln = Ln + (n + 1)Un ,

G ;n+1/2 = G ;n+1/2 ,

S
S
G
;n1/2 = G ;n1/2 + 2n ;n1/2 ,
K ; ;n = K ; ;n , Un ,

;n1/2 = (1 2) ;n1/2 ,

Sn = (1 2)Sn .

In particular, the 11 generators besides L are all primary of unchanged weights

 
Ln , Am = (h(A) 1)n m An+m .

(36)

(37)

It should be mentioned that one may obtain the twisted supercurrents by following the
exact same procedure which leads to the construction of the untwisted supercurrents, see
Ref. [1]. In particular, we have the relations

J. Rasmussen / Nuclear Physics B 593 (2001) 634650



I
dz
1
E = (n 1)G ;n+1/2 ,
Ln ,
2i
2


I
dz
1
S
F = (n + 1)G
Ln ,
;n1/2 .
2i
2

643

(38)

The remaining twisted generators may of course be found by working out the relevant
(anti-)commutators of twisted generators already obtained, and requiring the algebra to be
invariant. For the K generators, let us write out the result of the twisting (36)
en ,
en = E
E

en = H
en ,
H

Un = (1 2)Un .

en = F
en ,
F
(39)

We observe that for the unique value

= 1/2,

(40)
=1/2

twisting (36) is not an invariance but rather a reduction. The 4 generators U =1/2 ,
and S =1/2 all vanish identically, and the resulting SCA is instead the small N = 4 SCA.
It is remarked that all the generators of the two induced N = 4 SCAs discussed above
are integrated spin one primary fields with respect to the Virasoro algebras associated
to the original affine current superalgebras SL(2|2) and SL(2|2)/U (1), respectively. In
Ref. [1] we have shown that all the generators of the SL(2|2) induced SCA are of that kind.
Since the twisted generators (36) are linear combinations of those fields, they themselves
satisfy this condition. The generators of the small N = 4 SCA induced by the coset
SL(2|2)/U (1) in Section 3.2, are also integrated primary fields. This follows immediately,
as the construction is based on the same free field realization as the original SL(2|2), the
only (and in this respect trivial) difference being the coset condition (30).

4. Conclusion
We believe that the general construction of SCAs outlined in Ref. [1] (and indicated in
Ref. [2]) is interesting from a mathematical as well as from a string-theoretical point of
view. A mathematical or conformal field theoretical interest lies in the fact that besides
providing new realizations of well-known SCAs, the construction also produces whole
new classes of SCAs. An additional virtue is that the SCAs are realized explicitly. The
asymmetric N = 4 SCA discussed in the present paper is an example of such a new SCA.
There are also convincing indications that new bosonic extensions of the Virasoro algebra
may be obtained using a modification of the construction, and we hope to return to a
discussion on that.
The construction is interesting from the string-theoretical point of view, as it presumably
produces the unique boundary SCA associated to a string theory on AdS3 with a certain
affine Lie supergroup symmetry. As already pointed out, this pertains to Lie supergroups
with SL(2) G0 decomposable bosonic subgroups. Due to the recently discovered
AdS/CFT correspondence [1214], the question of determining which superconformal
field theory is associated to which string theory has become increasingly relevant. We hope

644

J. Rasmussen / Nuclear Physics B 593 (2001) 634650

that our work will add to the understanding of this, and intend to study the role of the new
and asymmetric N = 4 SCA in string theory.

Acknowledgement
The author thanks Spenta Wadia and Mark Walton for comments. The author is also
grateful to The Niels Bohr Institute, where part of this work was done, for its kind
hospitality. He is supported in part by a PIMS Postdoctoral Fellowship and NSERC of
Canada.

Appendix A. Lie superalgebra sl(2|M)

The root space of the Lie superalgebra sl(2|M) in the distinguished representation may
be realized in terms of an orthonormal two-dimensional basis {1 , 2 } and an orthonormal
M-dimensional basis {u }u=1,...,M with metrics
 0 = ,0 ,

u u0 = u,u0 ,

 u = 0.

(A.1)

The 12 (M + 1)(M + 2) positive roots are then represented as

0+ = {1 2 } {u v | u < v},
1+
+ = {1 u | u = 1, . . . , M},
1
+ = {2 u | u = 1, . . . , M},

(A.2)

where the M + 1 simple roots i are

1 = 1 2 ,

2 = 2 1 ,

u+2 = u u+1 .

(A.3)

The associated ladder operators E and F , and the Cartan generators Hi admit a standard
oscillator realization (see, e.g., [15])
E1 2 = a1a2 ,
F1 2 = a2 a1 ,
H1 = a1 a1 a2a2 ,
a()

where


E u = a bu ,
F u = bu a ,
H2 = a2 a2 + b1 b1 ,

Eu v = bu bv ,
Fu v = bv bu ,

Hu+2 = bu bu bu+1
bu+1 ,

(A.4)

bu()

and
are fermionic and bosonic oscillators, respectively, satisfying

 () ()


bu , bv = u,v ,
bu , a = 0.
a , a0 = ,0 ,

(A.5)

Appendix B. Free field realization of affine SL(2|2) current superalgebra

In this appendix we shall provide the explicit free field realization of the affine SL(2|2)
current superalgebra that is discussed in Section 2. Recall that the only bosonic ghost fields

J. Rasmussen / Nuclear Physics B 593 (2001) 634650

645

are 1 , 3 , 1 , 3 . Let us introduce the abbreviations 1 , 23 , 123, . . . for the ghost

fields 1 , 2 +3 , 1 +2 +3 , . . . . The free field realization is


1
1 1 2 3
23 123,
E1 = 1 2 12 +
2
2 6
1
E2 = 2 + 1 12
2
1
E3 = 3 + 2 23 +
2

1 3
1
23 1 3 123 ,
2
6


1 1 1 2
12
+
123,
2 6

1
E1 +2 = 12 3 123 ,
2

1
E2 +3 = 23 + 1 123,
E1 +2 +3 = 123 ,
2

H1 = 2 1 1 + 2 2 12 12 + 23 23 123 123 + k 1 ,

H2 = 1 1 3 3 + 12 12 23 23 + k 2 ,

H3 = 2 2 2 3 3 + 12 12 23 23 123 123 + k 3 ,





2
1 1 2
1
1
12 2 1 1 2 + 12 12
F1 = 1 1 +
2
2
2


1 1 2 3 1 1 23
+ 123 23
+
12
2



1
1
1
1 1 23 + 12 3 + 123 123 + 1 k 1 + (k 1) 1,
2
2
6


F2

F3




1 1 2
1 2 3
1
12
23
+

=
1
3 + 2 12 12
2
2
2

1 2 23
1 2 1 23
23 + + 12 3 123 + 2 k 2 + k 2 ,
2
6






1 2 3
1 1 2 3 1 12 3
23
3 2
123
+

=
2 3
12
2
12
2




1
1
1
1
1
+ 3 2 3 23 23 + 3 1 23 + 12 3 123 123
2
2
2
6
2

+ 3 k 3 (k + 1) 3 ,


F1 +2


1 1 2
12
+
=
1 2 12 2
2


1 1 2 3 1 1 23 1 12 3
123
+
+
3
6
2
2


1
1
1
+ 2 1 23 12 3 123 23
2
2
2

1 5 1
2 2 23 1 1 2 12 3 1 1 2 123
+ +

2 12
4
6
1

646

J. Rasmussen / Nuclear Physics B 593 (2001) 634650


1
+ 1 12 23 + 12 123 123
2





1 1 2
1 1 2
+ 12
12
+
k 1
k 2
2
2
6k 11 2 1 3k + 1 1 2

+ (k 1/2) 12,
+
12
6


1 1 2 3 1 1 23 1 12 3
123
+ + +
1 + 2 23 2
F2 +3 =
6
2
2




1 2 1 1 23 1 12 3
3 1 2 3
23
123

3
12
2
2
2
2

1 1 1 2 23 3
5
1
+ 2 12 ( 3 )2 2 123 3
+
2 4
12
6

1 12 23 3
23 123
+ +
123
2





1 2 3
1 2 3
23
23
+

+
k 2 +
k 3
2
2
3k 1 3 2 6k + 11 2 3

+ (k + 1/2) 23,
+
6
12


1
1
1
F1 +2 +3 = 1 1 2 3 + 1 23 + 12 3 + 123 1
6
2
2


1 1 2 23 1 2 12 3
+ 12 23 2

2
2


1
1
1
+ 3 1 2 3 1 23 12 3 + 123 3
6
2
2


1 1 2 2 23
1 1 2 12 3
( ) + + 12 123 12
+
4
12


1 2 12 3
2
1

+ 1 2 23 3 + 23 123 23
4
12


1 1 3 1 1 2 23 1 2 12 3
+ + 12 23 123

6
2
2



1 1 2 3 1 1 23 1 12 3
+ + + 123
+
k 1
6
2
2



1 1 2 3 1 1 23 1 12 3
+ 123

k 2
3
2
2



1 1 2 3 1 1 23 1 12 3
+ 123

k 3
6
2
2
2k 5 2 3 1 k 2 23 1 k 1 3 2 2k + 5 1 2 3
+
+

+
12
2
3
12
k + 2 12 3 k 1 3 12 k + 1 1 23
+

+ k 123.

(B.1)
2
2
2

J. Rasmussen / Nuclear Physics B 593 (2001) 634650

647

Appendix C. Generators of asymmetric N = 4 SCA

For completeness and reference, below are listed explicit free field realizations of the
integrands An of the generators An (25) of the asymmetric N = 4 SCA
I
dz
An (z).
(C.1)
A=
2i
The realizations are obtained by inserting the results of Appendix B in the general
expressions for the generators provided in Section 3. The realizations of 2 , (2 +3 )
and S may be obtained by working out explicitly the relevant (anti-)commutators of the
asymmetric N = 4 SCA (26).
The Virasoro generator is realized as



n+1
n + 1 1
n1 n 2 1 2

+ n 12 2
1
Ln = 1
2
2


2


n n 2 12
1 n n(n 3) 1 2
+

+
12
8
4


n + 1 1
n1 n 1 2 3 n 2 1 23
123

n
+
23
2
12
2


n n(n 1) 1 2 3 n(n 3) 1 23 n(n + 1) 12 3

+

1
24
8
24


n
(n + 1)(n 2) 123
1

k 1 .
(C.2)
123 + (n + 1) 1
4
2
Here and throughout Appendix C we have taken advantage of the fact that terms
proportional to n( 1 )n1 1 vanish upon integration. They are not included in the
expressions for the integrands. The supercurrents are realized as



n
n1 1
n+1 3
2n 1 1 3
12
23 +
123 ,
(n + 1)2
G2 ;n+1/2 = 1
2
2
6




n1 1
n
123 ,
(n + 1)23
(C.3)
G2 +3 ;n+1/2 = 1
2
and
S ;n1/2
G
2




n 1 1 2

n1 2 12
+ 12 1 n 1
2
2



n1 2n 3 1 2 3 n 2 1 23 1 12 3

+ n n 123 3
1
6
2
2
n 1 1
n 2 12
+
12
2
 2



2
n(n + 2) 1 2 3 12
1 n2 n 2
1 2 23 +


4
12

= 1


n(n 1) 1 12 23 n2 1 2 123
12 123
+
+
+ n(n 1)
23
2
2

648

J. Rasmussen / Nuclear Physics B 593 (2001) 634650



n1 n2 4 1 2 12 3 3n2 4n 4 1
2 2 23 n(n 2) 1 12 23
+

+
24
24
4

(3n 4)n 1 2 123 n(n 2) 12 123


+
+
123
12
2





n1 1 1 2

n1 n 2 1 2
+ 12
+ n 12
+n 1
k 1 + 1
k 2
2
2

n2 12 1
(6kn 3n 8)n 1
n1 2 1

+
+ (k 1/2)(n2 n) 1

12

n1 12
(3n 6)k n 1
n 2
+ (k 1/2)n 1
+
,
6
+ 1

S( + );n1/2
G
2
3



n 1 1 2 3 1 1 23 1 12 3
+ + + 123 1
= 1
6
2
2


n2 n(n 7) 1 2 12 3 n2 + n 4 1
2 2 23 n(n + 1) 1 12 23
+

+ 1
+
12
4
2

n(n 1) 1 2 123

+
+ n(n 1) 12 123 2
2




n 2 1 23 n 12 3
1 n1 3 2n 3 1 2 3
123

+ n

3
6
2
2


n1 n2 9n + 6 1 2 12 3 n2 n 2 1
2 2 23
+

1
24
8
n(n 1) 1 12 23 n2 3n + 2 1 2 123
+

4
4

n(n 3) 12 123

+
12
2


n2 n(n + 5) 1 2 12 3 2 n(n 1) 1 12 23 3
( ) +

1
24
4
n(n 1) 1 2 3 123 (3n 1)n 1
2 2 23 3


+
+
4
24

n(n 1) 12 3 123

+
+ n 1 23 123 23
2


n1 2n2 + 7n 15 1 2 12 3 2 2n2 n 3 1 12 23 3
( ) +

+ 1
72
12
2n2 3n + 1 1 2 3 123 2n2 n 3 1
2 2 23 3


+
+
12
24 
n(n 1) 12 3 123 n 1 1 23 123


+
+
123
3
2



n1 n 1 2 3 n 1 23 n 12 3
+ + + n 123
+ 1
k 1
6
2
2
+

J. Rasmussen / Nuclear Physics B 593 (2001) 634650




n1 n 3 1 2 3 n 2 1 23 n 12 3
+
+ + n 123
k 2
6
2
2



n1 2n 3 1 2 3 n 2 1 23 n 12 3

+ n 123
+ 1
k 3
6
2
2


n2 (4kn 3n 7)n 1 2 3 (3k 2)n(n 1) 12 3
+ 1
+

24
6

(2kn n 3)n 1 23
123
+ (k 1/2)n(n 1)
+
1
4
kn n + 1 1
n 3 2 (k 1)n 1
n1 3 12
+

+

6
2


2


(6k + n + 11)n 12
1 n1 8kn 12k + n + 11n 22 1 2
+

3
24
12

n1 123
kn 2k 1 1
n 23
+
.
+ kn 1
2
The internal SU(2) KacMoody algebra is realized as






1 n
1 n1 n + 1 1 2
12
e
+ n
23
E n = 3
2



n 3n 1 1 2 n 1 12
+

+ 1
123,
12
2





n
1 n1 n + 2 1 2
12
e
+ n
2 2 1 3 3
Hn =
2




n 2 12
n
n
1
1 2
+

12
4
2




n + 2 1 23
1 n1 5n 1 2 3
12 3
123
+
+ n + n

23
12
2



n n 1 2 3 n 1 23 7n 12 3 n 2 123
+ + +

+ 1
123
4
4
12
2

n
+ 1
k 3 ,

649

+ 1




n1 n + 3 1 2 3 n + 2 1 23 n 12 3
123
+
+ + n
2
6
2
2


n
1 ( 3 )2 3




n 1 23 n 2 12 3 n 2 123
1 n n+1 1 2 3
+ +
+

12
12
4
4
2



n1 3 n 3 1 2 3 n + 2 1 23 n 12 3 n 123
+
+ +
1

23
12
4
4
2




2n + 1 1 23 2n + 3 12 3 2n 3 123
1 n 3 n 1 2 3
+
+
+

123
18
12
12
6



n

n
n
+ 1 1 3 ,
1 3 k 3 + k +
(C.5)
12

en = 1
F
+
+
+

(C.4)

650

J. Rasmussen / Nuclear Physics B 593 (2001) 634650

whereas the U (1) generator is

I
dz 1
n
k (1 22 + 3 ).

Un =
2i
The remaining 2 fermionic spin 1/2 generators and the bosonic scalar are
I
dz 1
n1 1
V2 k (1 22 + 3 ),

2 ;n1/2 =
2i
I

dz 1
n1 1

V(2 +3 ) k (1 22 + 3 ),
(2 +3 );n1/2 =
2i
and

I
Sn =

dz 1
n1 1
1

V(2 +3 ) V
k (1 22 + 3 ).
2
2i

(C.6)

(C.7)

(C.8)

1
1
and V(
are
According to (B.1), the polynomials V
2
2 +3 )

1
1
= 1 2 + 12 ,
V
2
2
1
1
1
1
V(2 +3 ) = 1 2 3 + 1 23 + 12 3 + 123 .
6
2
2

(C.9)

References
[1]
[2]
[3]
[4]
[5]

[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]

J. Rasmussen, Nucl. Phys. B 582 (2000) 649.

K. Ito, Phys. Lett. B 449 (1999) 48.
A. Giveon, D. Kutasov, N. Seiberg, Adv. Theor. Math. Phys. 2 (1998) 733.
O. Andreev, Nucl. Phys. B 552 (1999) 169;
O. Andreev, Nucl. Phys. B 561 (1999) 413.
M. Ademollo, L. Brink, A. DAdda, R. DAuria, E. Napolitano, S. Sciuto, E. Del Guidice, P.
Di Vecchia, S. Ferrara, F. Gliozzi, R. Musto, R. Pettorino, Phys. Lett. B 62 (1976) 105;
M. Ademollo, L. Brink, A. DAdda, R. DAuria, E. Napolitano, S. Sciuto, E. Del Guidice, P.
Di Vecchia, S. Ferrara, F. Gliozzi, R. Musto, R. Pettorino, Nucl. Phys. B 114 (1976) 297.
K. Schoutens, Phys. Lett. B 194 (1987) 75;
K. Schoutens, Nucl. Phys. B 295 (1988) 634.
A. Sevrin, W. Troost, A. Van Proyen, Phys. Lett. B 208 (1998) 447.
A. Ali, A. Kumar, Mod. Phys. Lett. A 8 (1993) 1527.
A. Ali, Classification of two dimensional N = 4 superconformal symmetries, hep-th/9906096.
J. Rasmussen, Nucl. Phys. B 510 (1998) 688.
K. Ito, J.O. Madsen, J.L. Petersen, Nucl. Phys. B 398 (1993) 425.
J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231.
S. Gubser, I. Klebanov, A. Polyakov, Phys. Lett. B 428 (1998) 105.
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253.
D.-S. Tang, J. Math. Phys. 25 (1984) 2966.

Nuclear Physics B 593 (2001) 651670

www.elsevier.nl/locate/npe

More on correlators and contact terms in N = 4

SYM at order g 4
Silvia Penati a , Alberto Santambrogio b, , Daniela Zanon c
a Dipartimento di Fisica dellUniversit di Milano-Bicocca and INFN, Sezione di Milano, Via Celoria 16,

20133 Milano, Italy

b Instituut voor Theoretische Fysica Katholieke Universiteit Leuven, Celestijnenlaan 200D,

B-3001 Leuven, Belgium

c Dipartimento di Fisica dellUniversit di Milano and INFN, Sezione di Milano, Via Celoria 16,

20133 Milano, Italy

Received 13 June 2000; accepted 15 October 2000

Abstract
We compute two-point functions of chiral operators Tr k for any k, in N = 4 supersymmetric
SU(N) YangMills theory. We find that up to the order g 4 the perturbative corrections to the
correlators vanish for all N. The cancellation occurs in a highly non trivial way, due to a complicated
interplay between planar and non planar diagrams. In complete generality we show that this
same result is valid for any simple gauge group. Contact term contributions signal the presence
of ultraviolet divergences. They are arbitrary at the tree level, but the absence of perturbative
renormalization in the non singular part of the correlators allows to compute them unambiguously at
higher orders. In the spirit of the AdS/CFT correspondence we comment on their relation to infrared
singularities in the supergravity sector. 2001 Elsevier Science B.V. All rights reserved.
PACS: 11.30.Pb; 11.10.Gh; 11.15.Pg; 04.65.+e
Keywords: AdS/CFT correspondence; Supersymmetric YangMills theory; Renormalization; Superfields

1. Introduction
In this paper we continue the work presented in Ref. [1]. There we computed two-point
functions of chiral operators Tr 3 in N = 4 SU(N) supersymmetric YangMills theory to
the order g 4 in perturbation theory and proved that corrections vanish for all values of N . In
our perturbative approach, the fact that the non renormalization occurs for any value of N ,
not only at leading order in the N limit, was strictly connected to the vanishing of
* Corresponding author.

E-mail addresses: [email protected] (S. Penati), [email protected]

(A. Santambrogio), [email protected] (D. Zanon).
0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 3 3 - 7

652

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

the colour combinatoric factor of the nonplanar diagrams. The open question was to see if
the same pattern were true also for correlators of operators Tr k , with general k.
We have found that for k > 3 the colour structure of nonplanar diagrams at order g 4 does
not vanish. It is only a complicated cancellation between 1/N subleading contributions
from planar diagrams and nonplanar ones which allows to obtain a complete all N
nonrenormalization of the two-point correlators.
The colour structure identities we have used in our calculations can be generalized to
arbitrary gauge groups. In so doing we have been able to prove that the nonrenormalization
of the correlators up to order g 4 is valid for any group.
The other issue we focus on in this paper, is related to the presence of contact term
contributions [2]. They arise in the computation of the correlators because the theory is
affected by ultraviolet divergences. In order to control such infinities one has to choose
a regularization scheme: at order g 0 , i.e., at the tree level, the singularity of the twopoint function needs to be subtracted and this leads to the introduction of arbitrary contact
terms. At higher orders in g (we have computed the two-point functions explicitly up to
the order g 4 ) divergences cancel out and as a consequence, the local contact terms are
determined unambiguously. Thus the contact terms in the two-point correlators are of the
form


(1.1)
a + f (g 2 , N) p (4) (x y)
with a the arbitrary finite constant of the subtraction and p an integer depending on the
dimensions of the operators. The function f (g 2 , N) is in principle exactly computable
order by order in g 2 for any finite N , such that f (g 2 = 0, N) = 0.
According to the AdS/CFT prescription one should establish a correspondence between
these terms and corresponding ones in the supergravity sector. There ambiguities arise due
to the fact that the theory suffers from infrared divergences when approaching the boundary
of the AdS space. Within the holographic viewpoint proposed in [3], the supergravity
effective action, evaluated on a solution of the equations of motion with prescribed
boundary conditions, becomes the generating functional for the conformal field theory
in the large N limit. The bulk fields evaluated at the boundary act as source terms for
composite operators O of the YangMills theory

R d 4 x O 
0
= eS[0 ]
(1.2)
e
CFT
with the boundary action given by


Z
1
2 2 (4)
+
b(
)

(x

y)
0 (y).
S[0 ] = d 4 x d 4 y 0 (x)
(x y)2

(1.3)

The local term, which acts as a regulator at short distances in (1.3), has an infrared origin in
the 5d AdS supergravity expanded near the boundary [46]. The coefficient b is a function
of a mass scale parameter identified with the inverse of the IR 5d cutoff [7].
Now, comparing Eq. (1.1) at small coupling with Eq. (1.3) one should find that in the
large N limit, with the t Hooft coupling g 2 N fixed but large
a + f (g 2 , N) bf ,

(1.4)

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

653

where bf is the arbitrary finite part of the subtraction in (1.3). In the context of our specific
calculation we will show how this identification can be realized.
We present this paper as a sequel to the one referred in [1], since the techniques used
here to perform the perturbative calculations of the correlators are the same as the ones
used in [1]. Therefore in order to avoid lengthy repetitions in the main text, and at the same
time to make this paper self-contained, we have recollected the main formulas and the
rules of the game in the appendices. In the next section we simply remind the reader which
are the quantities we have to focus on. Then in Section 3 we enter directly in medias res:
S1 )k i correlators. In
we present the tree-level and the order g 2 result for the hTr( 1 )k Tr(
4
Section 4 the order g contributions are considered. For the relevant diagrams we compute
the colour structure factors and the momentum integrals that one obtains after completion
of the D-algebra. The various, complicated contributions give rise in the end to a complete
cancellation for any finite N [8]. This is in agreement with previous results [9,10]. In
Section 5 we show that the nonrenormalization properties we have found for the SU(N)
gauge group, are actually valid for general groups. Finally in Section 6 we concentrate on
the evaluation of the contact terms and study what they correspond to in the supergravity
sector.

2. Two-point functions of chiral operators

Our goal is to compute two-point correlators for N = 4 supersymmetric SU(N) Yang
Mills theory, perturbatively in N = 1 superspace. The operators under consideration are the
chiral primary operators in the (0, k, 0) representation of the SU(4) R-symmetry group. In
a N = 1 superspace description of the theory (see Appendix A for details) they are given by
O = Tr( {i1 i2 ik } ), with flavor indices on the superfields symmetrized and traceless.
In fact, in order to simplify matters as in Refs. [1,9], we consider the SU(3) highest weight
S1 )k i. In this way the flavor combinatorics is
superfield 1 and compute hTr( 1 )k Tr(
avoided. At the same time we do not lose in generality since the SU(3) transformations,
which are invariances of the theory, allow to reconstruct all the other primary chiral
correlators from the one above.
The general strategy we have adopted for performing the actual calculation is outlined in
Appendix A. Quite generally, at non-coincident points we can write the two-point function
as



k


F (g 2 , N) (4)
S1 k (z2 ) =
(1 2 ),
Tr 1 (z1 ) Tr
(x1 x2 )2k

(2.1)

Away from short distance singularities, the x-dependence of the result

where z (x, , ).
is fixed by the conformal invariance of the theory, and F (g 2 , N) is the function that we
want to determine perturbatively in g 2 . As we have done in Ref. [1] we compute loop
integrals in momentum space and use dimensional regularization and minimal subtraction
scheme to treat ultraviolet divergences. In n dimensions, with n = 4 2, the Fourier
transform of a power factor (x1 x2 )2 is given by

654

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

d n p eipx
,
n
(2)n (p2 ) 2
n n
(2.2)
6= 0, 1, 2, . . ., 6= , + 1, . . . .
2 2
The main advantage of such an approach is due to the fact that the x-space structure in (2.1),
with a nonvanishing contribution to F (g 2 , N), can be obtained simply by looking at the
contributions that behave like 1/ from the singular factor ( n2 ) = ( + 2 ),
> 2 in (2.2). By analytic continuation one can write the general identity
Z
eipx
d np
(2)n (p2 )2k+
22k4

=
(1)k (k 1)!(k 2)! 2 k(+1) [1 + O()].
(2.3)
2

(x )
n
1
n2 n2 ( 2 )
=
2

(x 2 )
()

Once the UV divergent terms are determined at a given order in g, one can reconstruct the
complete answer using (2.3).
The UV divergent terms have been computed using the method proposed in [12] and
various techniques presented in [13,14].
Finally we emphasize that finite momentum space contributions to the correlators
correspond in x-space to terms proportional to . These are the terms which give rise
to contact terms [13]. We will come back to this point in Section 5.

3. At tree-level and order g 2

S1 )k (z2 )i is given by the
At tree-level the 2-point correlation function hTr( 1 )k (z1 ) Tr(
diagram in Fig. 1. The colour structure

X


Tr Ta (1) Ta (k)
(3.1)
Tr Ta1 Tak

which includes all possible permutations in the contractions of the scalar lines, is a
k-degree polynomial in N . In a double line representation for the colour indices, the
N -leading power term corresponds to the planar double line graph associated to the
diagram in Fig. 1, whereas nonplanar graphs give rise to subleading contributions.

S1 )k i.
Fig. 1. Tree-level contribution to hTr(1 )k Tr(

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

655

S1 )k i.
Fig. 2. g 2 -order contribution to hTr(1 )k Tr(

Now using the result (B.5) for the momentum integrals to leading order in the 
expansion, we obtain

k1
1
(1)k
1
(p2 )k2(k1)
2
 (4)
[(k 1)!]2

X


Tr Ta1 Tak
Tr Ta (1) Ta (k) (4) (1 2 ).
(3.2)

In x-space, using Eq. (2.3), the result can be rewritten as



k

S1 k (z2 )
Tr 1 (z1 ) Tr
0

k

X


1
1
=
Tr
T

T
Tr Ta (1) Ta (k)
(4) (1 2 ). (3.3)
a
a
k
1
4 2
(x

x2 )2k
1

At order g 2 the only contribution to the correlator is given by the diagram in Fig. 2, with
the insertion of a vector line.
From the two internal vertices V1 in (A.6) we obtain the colour structure famb fa 0 mb0
which contracted with the colour matrices associated to the rest of the diagram gives


XX
 



Tr Ta (1) Ta (i) , Tm Ta (j) , Tm Ta (k) .
Tr Ta1 Tak
(3.4)

i6=j

Here the sum is over all possible permutations of the external lines and Eq. (C.1) has been
used. The previous expression can be simplified [9] by noticing that for any set of matrices
Mj , j = 1, . . . , n, and any matrix P the following identity holds
n
X

Tr M1 [Mi , P ] Mn = 0.

(3.5)

i=1

Thus (3.4) can be written as

Tr Ta1 Tak

k

XX






Tr Ta (1) Ta (i) , Tm , Tm Ta (k)

(3.6)

i=1

which, using the identity (C.6), can be reduced further to the expression in (3.1) up to a
factor 2N . Including the various factors from vertices and propagators, we finally have

X


Tr Ta (1) Ta (k) .
(3.7)
(g 2 N)k Tr Ta1 Tak

The momentum integral associated to the graph after completion of the D-algebra, is
evaluated in (B.6). The final result (here we reinstate a factor 1/(4)2 for each loop) is

656

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

k
1
(1)k1 (k 1) 2 k2k
12 (3)(g N)
(p )
(4)2
[(k 1)!]2
X




Tr Ta1 Tak
Tr Ta (1) Ta (k) .
2

(3.8)

In the limit  0 this expression is finite, therefore it does not contribute to the correlation
function at separate points. It gives rise to a contact term which will be discussed in detail
in Section 5.

4. At order g 4
The relevant diagrams, drawn in single line representation, are shown in Figs. 3 and 4.
They group themselves into planar (A) and nonplanar (B) ones. If one were to use a
double line representation for the colour indices from Tija , then the single-line planar graphs
(A) would give rise to two distinct types of graphs, double-line planar graphs (A1 ) and
nonplanar double-line graphs (A2 ). Now from the A1 graphs the colour structure would
produce leading N contributions, while from the A2 graphs subleading contributions would
arise. Obviously the single-line type (B) diagrams could only give structures subleading
in N since their double-line representation would be necessarily nonplanar.
For the correlators computed in [1], i.e., for the k = 3 case, the colour factor of the
diagrams in the class (B) turned out to be identically zero. All the graphs in the class (A)
gave rise to the same overall combinatorics, so that the final cancellation occurred as a
cancellation among the various terms produced by the momentum integrations.

S1 )k i.
Fig. 3. Planar g 4 -order contribution to hTr(1 )k Tr(

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

657

S1 )k i.
Fig. 4. Nonplanar g 4 -order contribution to hTr(1 )k Tr(

In the present case, k > 3, we will show that the situation is much more complicated
and highly nontrivial. All types of diagrams mentioned above have nonvanishing colour
structures. The diagrams (A) give contributions proportional to the colour structure (3.1)
which, after multiplication by the momentum integrals, sum up to zero. However in
addition, from the diagrams in Figs. 3c and 3d a new subleading structure arises which
is of the same kind as the one from diagrams (B). Again, a nontrivial cancellation occurs
among the diagrams in Figs. 3c, 3d and (B) due to the special structure of their momentum
integrals.
Now for each of diagram we present the evaluation of the colour factor, while we list the
corresponding momentum integrals in Appendix B. The final results are summarized at the
end of the section where the actual cancellation is discussed.
In Fig. 3a we have the insertion of a two-loop propagator correction [1,15]
Z
d nq d nk
Sai (p, )ai (p, )p2
2g 4 N 2
2
2
k q (k q)2 (k p)2 (p q)2

1 
Sai (p, )ai (p, )
= 2g 4 N 2
6 (3) + O() .
(4.1)
2
2
(p )
In this case the colour structure is easy to compute. Inserting all the coefficients from
vertices, propagators and combinatorics we obtain

X


Tr Ta (1) Ta (k) .
(4.2)
12 (3)(g 2 N)2 k Tr Ta1 Tak

The momentum integral emerging after the D-algebra is given in Eq. (B.7).
We now consider the diagram 3b where the O(g 2 ) effective vertex [15] appears
g3
Sai (q, )bi (p, )
Nif abc
4
Z



S2

S
4D D D + (p + q) D , D Vc (p q, )

d nk
. (4.3)
k 2 (k p)2 (k q)2
Performing the D-algebra one easily realizes that only the first term in (4.3) gives rise to
potentially divergent loop integrals. The second term produces finite contributions which
S 2 D term, since we
might generate order g 4 contact terms. We concentrate on the D D
4
will not compute contact contributions to g order. The colour structure for this diagram is
computed following the same procedure as in the case of the diagram in Fig. 2 (see Eqs.
(3.4)(3.7)). Inserting all the various factors we obtain

2

X


Tr Ta (1) Ta (k) .
(4.4)
4 g 2 N k Tr Ta1 Tak

The corresponding momentum integral is given in Eq. (B.8).

658

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

We now turn to the discussion of the colour structure for the diagram in Fig. 3c. The
insertion of the two vector lines gives rise to the structure fa1 mb1 fa2 mn fnpb2 fa3 pb3 . The
contraction with scalar lines from the external vertices takes into account all possible
permutations. Using Eq. (C.1) we can write


X X
 
 
Tr Ta (1) Ta (i) , Tm Ta (j) , Tm , Tn
Tr Ta1 Tak
i6=j 6=l





Ta (l) , Tn Ta (k) .

(4.5)

This expression can be manipulated by making use of the identity (3.5) and it can be written
as the sum of two terms


X X
 
 



Tr Ta (1) Ta (j) , Tm , Tn , Tm Ta (l) , Tn Ta (k)
Tr Ta1 Tak

j 6=l


  




+ Tr Ta (1) Ta (j) , Tm , Tn Ta (l) , Tn , Tm Ta (k) .
(4.6)

The first term can be further reduced by using the identities (C.6), (C.7) and (3.5) again.
Inserting the factors from combinatorics, vertices and propagators the final expression for
the colour structure can be written as the sum of a term leading in N plus a subleading
contribution

X


2kN 2 Tr Ta (1) Ta (k)
g 4 Tr Ta1 Tak


  




Tr Ta (1) Ta (j) , Tm , Tn Ta (l) , Tn , Tm Ta (k) .

(4.7)

j 6=l

For k = 3 the second term vanishes and the above expression reduces to the one computed
in Ref. [1]. However for k > 3 the subleading term is nonzero, as shown in Appendix
C explicitly for the k = 4 case. For this diagram the loop-integral result can be read in
Eq. (B.9).
The colour factor for the diagram in Fig. 3d can be computed by exploiting the previous
results. In fact, from the internal vertices V1 and V3 (see Eq. (A.6)) we have the structure
fa1 mb1 fpb2 n fma2 n fa3 pb3 which is identical to the one from the diagram 3c. Performing all
the contractions and taking into account the coefficients from combinatorics, vertices and
propagators we finally obtain

X


g4
Tr Ta1 Tak
2kN 2 Tr Ta (1) Ta (k)
2

X

  




Tr Ta (1) Ta (j) , Tm , Tn Ta (l) , Tn , Tm Ta (k) .

(4.8)

j 6=l

The momentum integral for this graph is given in Eq. (B.10).

The last potential contributions from diagrams of type (A) are shown in Figs. 3e3h.
The loop integral resulting from the D-algebra on the diagram 3e given in Eq. (B.11), is
O() and then it does not contribute to the correlation function. The diagrams in Figs. 3f,
3g and 3h, exactly as for k = 3, only contribute to finite terms.

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

659

We now study the nonplanar diagrams (B) in Fig. 4. The combinatorial and colour
factors, and the loop integrals for the two diagrams 4a and 4b are identical. Thus we
concentrate on one of them, e.g., 4b. From the internal V1 vertices with vector lines
contracted as in figure, the colour factor which arises is fa1 mn fnpb1 fa2 pq fqmb2 . When
connected with the rest of the diagram it gives

  




g4 X X
Tr Ta (1) Ta (j) , Tm , Tn Ta (l) , Tn , Tm Ta (k) ,
(4.9)
2
j 6=l

where combinatorial factors have been included. We note that the previous expression has
the same form as the subleading contribution already present in the diagrams 3c and 3d
(see Eqs. (4.7), (4.8)). As previously mentioned and proven in Appendix C, this term is
zero for k = 3 but in general nonvanishing when k > 3. The corresponding loop diagram
arising after the D-algebra is given in Eq. (B.8).
The last nonplanar graph to be considered is the one in Fig. 4c. As in the k = 3 case [1],
its colour coefficient vanishes. A simple way to prove this is to notice that from the internal
vertices one obtains
fa1 mb1 fa2 nb2 fa3 pb3 fmnp

(4.10)

which is antisymmetric under the exchange a1 a2 and b1 b2 . When multiplied by all

possible permutations of colour matrices from the external vertices it gives a zero result.
We now collect all the divergent contributions we have computed at order g 4 . For each
diagram we list the final result obtained as a product of the colour and combinatorial factors
times the results from momentum integrations. We define

X


Tr Ta (1) Ta (k) ,
Pk Tr Ta1 Tak

Qk Tr Ta1 Tak
XX

  




Tr Ta (1) Ta (j) , Tm , Tn Ta (l) , Tn , Tm Ta (k) .

j 6=l

(4.11)
Factorizing an overall common coefficient

k+1
(1)k (k 1)
1
1 2 2
(g N)
(p2 )k2(k+1)
2

(4)
[(k 1)!]2 (k + 1)
the various contributions are
diagram 3a:
12k (3)Pk ;

(4.12)

(4.13)

diagram 3b:
24k (3)Pk ;
diagram 3c:



1 
12k (3) 40k (5) Pk + 2 20 (5) 6 (3) Qk ;
N

(4.14)

(4.15)

660

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

diagram 3d:
40k (5)Pk

1
20 (5)Qk ;
N2

(4.16)

diagrams 4a + 4b:
1
6 (3)Qk .
(4.17)
N2
The leading structure Pk appears in the diagrams 3a and 3b with a coefficient proportional
to the (3) Riemann function, whereas in the diagram 3d it is proportional to (5). The
same pattern repeats itself for the non-leading structure Qk : in the diagrams 4a and 4b it
is multiplied by (3), whereas in the diagram 3d it is proportional to (5). Both structures
appear in the diagram 3c and they contribute with such a coefficient to cancel the rest of
the divergent terms. In conclusion, only finite contributions survive. As emphasized at the
beginning this amounts to say that the correlation function is not renormalized at order g 4 ,
up to contact terms.
We stress that the complete cancellation of divergent contributions for general k occurs
for any finite N .
The results in (4.13)(4.17) if restricted to k = 3, reproduce the results discussed
in Ref. [1]. In this case the subleading structure Qk vanishes, so that the proof of
nonrenormalization for the correlator with general k cannot be implemented from the k = 3
example.

5. General gauge groups

Here we want to show that the nonrenormalization properties of the correlators proven in
the two previous sections for the SU(N) super YangMills theory, actually hold for general
simple gauge groups. To this end it is sufficient to realize that, using (C.3) and (C.5) which
are valid for any group, we can generalize the identities in (C.6) and (C.7) to the following
ones
[[Ta , Tm ], Tm ] = k1 Ta

(5.1)

and
[[[Ta , Tm ], Tn ], Tm ] =

k1
[Ta , Tn ].
2

(5.2)

At the order g 2 this enables us to replace 2N with k1 in (3.7) and (3.8), thus obtaining a
result which depends only on the Casimir of the adjoint representation of the gauge group.
In any event to this order the momentum integral gives a finite result which does not affect
the correlator at non coincident points.
In the same way one can analyze the general situation at the next perturbative g 4 order.
For the propagator and vertex insertions which appear in the graphs of Figs. 3a, 3b we use
the result as in Ref. [15], i.e., we set 2N k1 in (4.1) and in (4.3). For the rest of the
diagrams we substitute (C.6), (C.7) with (5.1), (5.2), respectively. Once this operation is

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

661

performed consistently everywhere in the various formulas of Section 4 the final, complete
cancellation of all the corrections is achieved following exactly the same pattern and the
same steps as in the SU(N) case.

6. Contact terms in the AdS/CFT correspondence

Now we wish to focus our attention on the two-point correlators for SU(N) when the
two points approach each other. In the limit x1 x2 the expression in (2.1) becomes
singular and needs to be regulated. Within the dimensional regularization approach we
have adopted, this short-distance singularity is signaled by 1/ poles, according to the
general identity
1 2 k2 (n)
2
1
( ) (x) for  0.

2
kk
2k4
(x )
2
(k 1)!(k 2)! 

(6.1)

For the two-point function of the operator Tr( 1 )k these short-distance UV divergences
manifest themselves already at tree-level (see (3.3)). In order to obtain a well defined
function at coincident points we perform a subtraction and define in configuration space


k

S1 k (z2 ) (4) (1 2 )
Tr 1 (z1 ) Tr
reg



42k 2  1
Pk
1
k 2 (1k) 2
2 k2 (n)
+

(
+
(1)
(
)
)

(x)
,
lim
0 (4)2k (x 2 )kk
[(k 1)!]2 
(6.2)
where x x1 x2 , Pk has been defined in (4.11) and is the mass scale of dimensional
regularization. The coefficient corresponds to an arbitrary finite subtraction and
generates a scheme dependent finite contact term.
Performing explicitly the  0 limit in (6.2) a residual dependence on the mass
scale survives from the (divergent) counter-contact term. Therefore, the subtraction of
the infinity at tree-level necessarily introduces a mass scale in the regularized correlation
function which breaks conformal invariance [16]. On the other hand, the scheme dependent
finite term proportional to is independent of and it does not affect conformal
invariance.
Now we discuss the appearance of this type of terms in the perturbative computation
of the two-point correlator. We have performed our loop calculations in momentum space,
with the Fourier transformation given in the basic formula (2.3). Using (2.3) and the general
identity in (6.1) we can write
Z
eipx
d np
= (1)k+1 ( 2 )k2 (4) (x),
(6.3)
lim
n
0
(2)n (p2 ) 2 k+
i.e., any finite contribution in momentum space gives rise to finite contact terms in the
correlation functions.
In general, the presence of contact terms at any loop order is related to the UV
regularization and renormalization procedure. Different subtraction conditions correspond

662

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

to different finite counterterms which eventually contribute to contact terms in the

correlation functions of the theory. In n = 4 2 dimensions, one has to choose a particular
regularization for evaluating the integrals
Z
Z
(6.4)
d 4 p f (p) G() d n p f (p),
where G() is a regular function 1 near  = 0, with G(0) = 1. A given prescription has
to be used in the computation of both the effective action and physical quantities like
correlation functions or scattering amplitudes. By expanding G() in powers of  one can
easily realize that in multiloop integrals with only simple pole divergences the coefficients
of the divergent terms do not depend on the particular choice of the G-function, whereas
they might depend on the regularization scheme in multiloop integrals with higher order
divergences. In any case, a scheme dependence is always present in the finite part of
any divergent diagram. It follows that in the evaluation of correlation functions, different
choices of the G-function, i.e., different regularization prescriptions, give rise in general to
different finite quantum contact terms. However, if a nonrenormalization theorem holds (in
the sense of absence of quantum corrections to the correlation functions), then the contact
terms become independent of the regularization prescription and they are unambiguously
computable from order g 2 on.
Indeed let us exemplify for simplicity the case in which at a given loop order the
perturbative contribution to a correlation function contains at most a simple pole 1/
divergence


a
+ b + O() G().
(6.5)

By expanding G() = 1 + c + , in the limit  0 we obtain the divergent scheme
independent contribution a/ and the finite term (b + ca) which contains a scheme
dependence through the coefficient c from G(). However, if there is no perturbative
divergent contribution, i.e., a = 0, the finite term is uniquely determined.
In our specific case, we have shown that up to the g 4 -order the two-point function
S1 )k (z2 )i is not perturbatively corrected and we are in the situation where
hTr( 1 )k (z1 ) Tr(
the finite contact terms are uniquely determined. The term at order g 2 can be easily inferred
from (3.8) by using the identity (6.3). In the limit  0 we find
12 (3)(g 2 N)

k2 1
Pk
( 2 )k2 (4) (x1 x2 ) (4) (1 2 ).
[(k 1)!]2 (4)2k

(6.6)

The same procedure can be applied in order to compute the contact term at order g 4 . One
should keep track of all finite contributions from the momentum integrals and then use (6.3)
to obtain the result in configuration space. We have not performed the explicit calculation
but in general we expect to find a nonvanishing result.
1 A convenient choice is G() = (4 ) (1 ) (G-scheme [14]) which cancels irrelevant terms proportional
to the Euler constant, log 4 and (2) Riemann function in the -expansion of functions which appear in the
calculation of multiloop integrals.

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

663

We note that, since these loop contact terms come from finite contributions in the
-expansion, in the four-dimensional limit no dependence on the scale survives.
Therefore these terms do not affect the conformal invariance properties of the theory. As
discussed in Ref. [16] the breaking of conformal invariance can only be ascribed to the
UV divergences in the correlation function.
If the absence of quantum corrections holds for any finite N at all orders in perturbation
theory, one could determine unambiguously the finite contact terms loop by loop, and
generate a contribution of the form


(6.7)
a + f (g 2 , N) ( 2 )k2 (4) (x1 x2 ) (4)(1 2 ),
where a is the arbitrary finite part of the contact term at tree-level. It is given by


a + (1 k) log 2 (1)k

Pk
242k 2
.
(4)2k [(k 1)!]2

(6.8)

Our result shows that the complete answer for the two-point functions of the theory
necessarily contains contact terms, unless we were to choose a non-minimal subtraction
at short distances. This would amount to the subtraction of a finite term with a coefficient
a = f (g 2 , N).
The SL(2, Z) invariance of the theory would require f (g 2 , N) to be a modular function.
However, following the arguments in [17], if loop contact terms were present in the final
result they would necessarily break the U (1)Y invariance that the theory inherits from the
5d supergravity. It might be possible to establish a correspondence with U (1)-breaking
terms in the supergravity action [18].
According to the AdS/CFT conjecture, the generating functional of the regularized
correlation functions in the large N limit is the semiclassical 5d supergravity action with IR
divergences suitably subtracted [5,16]. At the boundary, i.e., with the IR cutoff removed, it
has the form (1.3). In particular, the arbitrary finite contact term in (6.7) corresponds to an
arbitrary finite subtraction in 5d
Z
(6.9)
Sc [0 ] = bf d 4 x 0 (x)( 2)k2 0 (x)
related to the particular IR regularization scheme chosen for the supergravity action. In
other words, in the large N limit with g 2 N  1 we should find


(6.10)
a + f (g 2 , N) bf .
If in this limit f (g 2 , N) is not vanishing, we might conclude that either coupling dependent
local terms appear in supergravity, or one is forced to choose a particular subtraction
scheme in the YangMills sector.
Finally we notice that perturbative contact terms in the two-point functions give rise
in general to coupling dependent contact-type contributions in higher-point correlators, as
well as enter the definition of multitrace operators through the point-splitting regularization [5].

664

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

Acknowledgements
We wish to thank Misha Bershadsky, Dan Freedman and Michael Green for discussions
and suggestions. This work has been supported by the European Commission TMR
programme ERBFMRX-CT96-0045, in which S.P. and D.Z. are associated to the
University of Torino.

Appendix A. The basic rules for the computation of the two-point correlators
In N = 1 superspace the action of N = 4 supersymmetric YangMills theory can be
written in terms of one real vector superfield V and three chiral superfields i (we follow
the notations in [11])
Z
Z




1
8
gV S gV i

i e + 2 d 6 z Tr W W
S J, J =
d z Tr e
2g
Z
Z

 ig
 j k
ig
Si
S
S ,
+ Tr d 6 z ij k i j , k + Tr d 6 z ij k
3!
3!
Z
Z

(A.1)
+ d 6 z J O + d 6 z JO,
S 2 (egV D egV ), and V = V a T a , i = a T a , T a being SU(N) matrices
where W = i D
i
in the fundamental representation. We have added to the classical action source terms for
the chiral primary operators generically denoted by O.
We define the generating functional in Euclidean space
Z


S DV eS[J,J]
(A.2)
W J, J = D D
so that for O = Tr( 1 )k the two-point function is given by




2W
1 k
1 k

S
,
Tr (z1 ) Tr (z2 ) =

J (z1 ) J (z2 ) J =J=0

(A.3)

We use perturbation theory to evaluate the contributions to W [J, J]

where z (x, , ).
which are quadratic in the sources, i.e.,
Z


F (g 2 , N)

J (x2 , , ).
(A.4)
W J, J d 4 x1 d 4 x2 d 4 J (x1 , , )
(x1 x2 )2k
The x-dependence of the result is fixed by the conformal invariance of the theory, and
F (g 2 , N) is the function to be determined.
In order to obtain the result in (A.4) one has to consider all the two-point diagrams from
W [J, J] with J and J on the external legs. First one evaluates all factors coming from
combinatorics and colour structures of a given diagram. Then one performs the superspace
D-algebra following standard techniques (see, for example, [11]), and reduces the result to
a multi-loop integral.
The quantization procedure of the classical action in (A.1) requires the introduction of
a gauge fixing (we work in Feynman gauge) and corresponding ghost terms. The ghost

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

665

superfields only couple to the vector multiplet and are not interesting for our calculation.
In momentum space we have the superfield propagators


ab
V aV b = 2 ,
p


ab
Sjb = ij
ia
.
p2

(A.5)

S 2 , D2
The vertices are read directly from the interaction terms in (A.1), with additional D
factors for chiral, antichiral lines respectively. The ones that we need are the following
i
S2 D V b D V c ,
V2 = gfabc V a D
2
g 2 ij
g
Sic jd ,
fadm fbcm V a V b
V4 =  ij k fabc ia jb kc ,
V3 =
2
3!
S4 = g  ij k fabc
Sia
Sjb
Skc .
(A.6)
V
3!
All the calculations are performed in n dimensions with n = 4 2 and in momentum
space. We have used the method of uniqueness [12] which is particularly efficient for the
computation of massless Feynman integrals of a single variable.
Sia V b jc ,
V1 = igfabc ij

Appendix B. Relevant integrals in momentum space

In this appendix we list the relevant multiloop integrals which have been used in the
course of our calculation.
As described in Section 5, in dimensional regularization n = 4 2, one has to choose
a particular prescription for the regularized integrals. In our case the integrals have at the
most 1/ divergences so that they do not depend on the particular choice of the G-function,
as far as divergent contributions are concerned. In order to simplify the calculation, we
forget about (2) factors at intermediate stages and reinsert at the end a 1/(4)2 factor for
each loop. Having this in mind we now list the relevant integrals and their  expansion.
The basic integrals from which all our results can be deduced are the following:
at one loop:
Z
( + n2 ) ( n2 ) ( n2 )
1
d nk
=
I1 =
n , (B.1)
() ()
(n ) (p2 )+ 2
(k 2 ) [(p k)2 ]
at two loops:
Z
I2 =
at four loops:
Z
I4 =



d nk d nl
1
=
6 (3) + O() ,
k 2 l 2 (k l)2 (p k)2 (p l)2 (p2 )1+2

d nk d nl d nr d ns
k)2 (s l)2 (r s)2 (p r)2 (p s)2

1 1 
=
5 (5) + O() ,
2
4
 (p )

and

k 2 l 2 (k

(B.2)

l)2 (r

(B.3)

666

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

d n k d n l d n r d n s r 2 (p l)2
k)2 (r l)2 (s l)2 (r s)2 (p r)2 (p s)2


1
3
5
= (p2 )14 (5) (3) + O()

2
4

I4 =

k 2 l 2 (k

l)2 (r

(B.4)

(the explicit evaluation of the last integral was reported in Ref. [1]).
From the previous integrals we can derive all the results needed for our calculation. By
repeated use of (B.1) we obtain for the g 0 -order diagram
Z
1 (1)k
d n q1 d n qk1
=
(p2 )k2(k1) + O(1).
q12 (q2 q1 )2 (q3 q2 )2 (p qk1 )2  [(k 1)!]2
(B.5)
For the g 2 -order (finite) diagram we have
Z
Z
q22
d nk d nl
d n q2 d n qk1
(q3 q2 )2 (p qk1 )2
k 2 l 2 (k l)2 (q2 k)2 (q2 l)2
(1)k1 (k 1)
12 (3)(p2 )k2k + O().
=
(B.6)
[(k 1)!]2 k
In order to obtain this result, one first performs the k and l integrations with the help of
Eq. (B.2), then the other integrals are computed with the use of Eq. (B.1).
At g 4 -order, the integrals emerging after having performed the D-algebra are the
following.
For the graph 3a, by using Eq. (B.1) one obtains:
Z
d n q1 d n qk1
(q12)1+2 (q2 q1 )2 (q3 q2 )2 (p qk1 )2
=

1 (1)k (k 1)
(p2 )k2(k+1) + O(1).
 [(k 1)!]2 (k + 1)

For the graphs 3b, 4a and 4b, it is easy to deduce

Z
1
d n r d n q2 d n qk1
(q2 r)2 (q3 q2 )2 (p qk1 )2
Z
d nk d nl

2
2
k l (k l)2 (r k)2 (r l)2
1 (1)k (k 1)
6 (3)(p2 )k2(k+1) + O(1)
=
 [(k 1)!]2 (k + 1)

(B.7)

(B.8)

by first evaluating the two-loop integrals in k and l with the help of Eq. (B.2), and then
applying Eq. (B.1).
The integral relevant for the graph 3c is
Z
1
d n q3 d n qk1
(q4 q3 )2 (p qk1 )2
Z
d n k d n l d n r d n s r 2 (q3 l)2

k 2 l 2 (k l)2 (r k)2 (r l)2 (s l)2 (r s)2 (q3 r)2 (q3 s)2

S. Penati et al. / Nuclear Physics B 593 (2001) 651670


1 (1)k (k 1) 
6 (3) 20 (5) (p2 )k2(k+1) + O(1)
2
 [(k 1)!] (k + 1)

667

(B.9)

by first evaluating the four-loop integral with momenta k, l, r, s with use of Eq. (B.4), and
then performing the rest of the integrations using Eq. (B.1).
Finally, the momentum integral for the graph 3d is given by
Z
q32
d n q3 d n qk1
(q4 q3 )2 (p qk1 )2
Z
d nk d nl d nr d ns

k 2 l 2 (k l)2 (r k)2 (s l)2 (r s)2 (q3 r)2 (q3 s)2

1 (1)k (k 1)
40 (5)(p2 )k2(k+1) + O(1).
=
(B.10)
 [(k 1)!]2 (k + 1)
Here, one first performs the k, l, r, s integrations with Eq. (B.3), and then uses Eq. (B.1).
For the graph 3e the momentum integral produced after completion of the D-algebra is
of order . Indeed, it is given by
Z
Z
1
d n r r 2 (q4 r)2
d n q4 d n qk1
(q5 q4 )2 (p qk1 )2
Z
Z
d nk d nl
d ns d nt

k 2 l 2 (k l)2 (r k)2 (r l)2

(s r)2 (t r)2 (s t)2 (q4 s)2 (q4 t)2
(1)k (k 1)
144 (3)2(p2 )k2(k+1) + O( 2 ).
=
(B.11)
[(k 1)!]2(k + 1)
This result can be obtained by using first Eq. (B.2) for the k, l and s, t two-loop integrals,
and then Eq. (B.1) for the remaining integrations.

Appendix C. Colour structures

In this appendix we give our conventions and a series of useful identities involving
the group generators. Moreover we show that the colour structure (4.9) of the nonplanar
graphs 4a and 4b is nonvanishing, by evaluating it explicitly in the k = 4 case.
For a general simple Lie algebra we have
[Ta , Tb ] = ifabc Tc ,

(C.1)

where Ta are the generators and fabc the structure constants. The matrices Ta s are
normalized as
Tr(Ta Tb ) = k2 ab .

(C.2)

We have also
famn fbmn = k1 ab .

(C.3)

From the Jacobi identity one obtains

fabm fcdm + fcbm fdam + fdbm facm = 0

(C.4)

668

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

which in turn allows to write

1
falm fbmn fcnl = k1 fabc .
2

(C.5)

Now we specialize our formulas to the gauge group SU(N): we have k1 = 2k2N and we
choose a normalization in (C.2) such that k2 = 1. The generators Ta , a = 1, . . . , N 2 1,
in the fundamental representation of SU(N) are N N traceless matrices. For SU(N) the
relations in (C.3) and (C.5) can be written as
[[Ta , Tm ], Tm ] = 2NTa ,

(C.6)

[[[Ta , Tm ], Tn ], Tm ] = N[Ta , Tn ].

(C.7)

Now we concentrate on the explicit evaluation of some colour factors in the case k = 4.
In particular we want to show that the nonplanar colour structure of graphs 4a and 4b is
nonvanishing, in contradistinction to the case k = 3 (see formula (A.18) in Appendix A
of [1]). The relation which allows to deal with products of Ta s is the following


1
a a
(C.8)
Tij Tkl = il j k ij kl .
N
From (C.8) one can obtain several useful formulas




1
N2 1 N2 + 3 ,
N2




1
Tr(Ta Tb Tc Td ) Tr(Ta Tb Td Tc ) = 2 N 2 1 N 2 3 ,
N




1
Tr(Ta Tb Tc Td ) Tr(Td Tc Tb Ta ) = k24 2 N 2 1 N 4 3N 2 + 3 ,
N
Tr(Ta Tb Tc Td ) Tr(Ta Tb Tc Td ) =

the last one being a planar type contribution.

Moreover, by noticing that


fabc = i Tr [Ta , Tb ]Tc

(C.9)
(C.10)
(C.11)

(C.12)

one can compute

Tr(Tc1 Ta1 Tc2 Ta2 )fc1 mb1 fc2 mb2 = (b1 a1 b2 a2 + b2 a1 b1 a2 ),

(C.13)

Tr(Tc1 Tc2 Ta1 Ta2 )fc1 mb1 fc2 mb2 = b1 b2 a1 a2 + Tr(Tb1 Tb2 Ta1 Ta2 ).

(C.14)

Consider now the nonplanar colour structure for the graphs 4a and 4b


XX
 
Tr Ta (1) Ta (i) , Tm , Tn
Qk Tr Ta1 Tak
i6=j






Ta (j) , Tn , Tm Ta (k) .

(C.15)

As already mentioned, Q3 = 0. In general, however, it does not vanish. In the case k = 4

by exploiting the various symmetries of this structure, it can be brought to the more
manageable expression

S. Penati et al. / Nuclear Physics B 593 (2001) 651670






Q4 = 32 2 Tr Ta1 Ta2 Tb3 Tb4 + Tr Ta1 Tb3 Ta2 Tb4 fa1 nm fmpc1 fa2 pr frnc2







Tr Tc1 Tc2 Tb3 Tb4 + Tr Tc1 Tb3 Tc2 Tb4 + Tr Tc1 Tc2 Tb4 Tb3 .

669

(C.16)

By using the Jacobi identity (C.4) for the four f structure

fa1 nm fmpc1 fa2 pr frnc2 = Nfa1 mc1 fa2 mc2 + fa1 pm fa2 pr fc1 nm fc2 nr

(C.17)

and using Eqs. (C.13), (C.14) and (C.9), (C.10), it is straightforward to obtain




Q4 = 192 N 2 1 2N 2 3 .

(C.18)

As claimed, the result is nonvanishing.

Now, as a check, we want to verify in the k = 4 case that the colour structure of the
diagrams 3c and 3d is given by the sum of a term proportional to the tree level structure Pk
and a term proportional to the nonplanar structure Qk .
The tree level colour structure

X


Tr Ta (1) Ta (k)
(C.19)
Pk Tr Ta1 Tak

for k = 4 can be reduced to



P4 = 4 Tr(Ta Tb Tc Td ) Tr(Td Tc Tb Ta ) + 4 Tr(Ta Tb Td Tc ) + Tr(Ta Tb Tc Td )

and then, with (C.9)(C.11), to

1
P4 = 4 2 N 2 1 N 4 6N 2 + 18 .
N
Now consider the structure from the diagrams 3c and 3d


Rk Tr Ta1 Tak
X X

 
 
Tr Ta (1) Ta (i) , Tm Ta (j) , Tm , Tn

i6=j 6=l





Ta (l) , Tn Ta (k) .

(C.20)

(C.21)

(C.22)

Setting k = 4 it can be written as








R4 = 16 Tr Ta1 Ta2 Ta3 Tb + Tr Ta1 Tb Ta2 Ta3 + Tr Ta1 Ta2 Tb Ta3






+ Tr Ta1 Ta3 Ta2 Tb + Tr Ta1 Tb Ta3 Ta2 + Tr Ta1 Ta3 Tb Ta2
fa1 mc1 fa2 rn fnmc2 fa3 rc3







Tr(Tc1 Tc2 Tc3 Tb + Tr Tc1 Tb Tc2 Tc3 + Tr Tc1 Tc2 Tb Tc3






+ Tr Tc1 Tc3 Tc2 Tb + Tr Tc1 Tb Tc3 Tc2 + Tr Tc1 Tc3 Tb Tc2 .

(C.23)

Making use of the Eqs. (C.9)(C.14), it is a lengthy but straightforward calculation to show
that




(C.24)
R4 = 32 N 2 1 N 4 18N 2 + 36 .
At this point it is immediate, from Eqs. (C.18), (C.21) and (C.24), to see that
R4 = 8N 2 P4 Q4
in accordance with the manipulations leading to Eq. (4.7).

(C.25)

670

S. Penati et al. / Nuclear Physics B 593 (2001) 651670

References
[1] S. Penati, A. Santambrogio, D. Zanon, JHEP 9912 (1999) 006, hep-th/9910197.
[2] F. Gonzalez-Rey, B. Kulik, I.Y. Park, Phys. Lett. B 455 (1999) 164, hep-th/9903094;
P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Nucl. Phys. B 571 (2000) 71, hep-th/9910011.
[3] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200;
S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109;
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[4] W. Mueck, K.S. Viswanathan, JHEP 9907 (1999) 018, hep-th/9904039;
I.R. Klebanov, E. Witten, Nucl. Phys. B 556 (1999) 89, hep-th/9905104.
[5] G. Chalmers, K. Schalm, Phys. Rev. D 61 (2000) 046001, hep-th/9901144.
[6] S. de Haro, K. Skenderis, S.N. Solodukhin, Holographic reconstruction of spacetime and
renormalization in the AdS/CFT correspondence, hep-th/0002230.
[7] L. Susskind, E. Witten, The holographic bound in anti-de-Sitter space, hep-th/9805114;
A. Peet, J. Polchinski, Phys. Rev. D 59 (1999) 065011, hep-th/9809022.
[8] S. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Adv. Theor. Math. Phys. 2 (1998) 697, hepth/9806074.
[9] E. DHoker, D.Z. Freedman, W. Skiba, Phys. Rev. D 59 (1999) 045008, hep-th/9807098.
[10] P.S. Howe, E. Sokatchev, P.C. West, Phys. Lett. B 444 (1998) 341, hep-th/9808162;
B. Eden, P.S. Howe, P.C. West, Phys. Lett. B 463 (1999) 19, hep-th/9905085.
[11] S.J. Gates, M.T. Grisaru, M. Rocek, W. Siegel, Superspace, BenjaminCummings, Reading,
MA, 1983.
[12] D.I. Kazakov, Phys. Lett. B 133 (1983) 406;
D.I. Kazakov, Dubna preprint JINR-E2-84-410 (1984).
[13] K.G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 (1981) 159.
[14] K.G. Chetyrkin, A.L. Kataev, F.V. Tkachov, Nucl. Phys. B 174 (1980) 345.
[15] M.T. Grisaru, W. Siegel, M. Rocek, Nucl. Phys. B 159 (1979) 429.
[16] A. Petkou, K. Skenderis, Nucl. Phys. B 561 (1999) 100, hep-th/9906030.
[17] K. Intriligator, Nucl. Phys. B 551 (1999) 575, hep-th/9811047.
[18] M.B. Green, S. Sethi, Phys. Rev. D 59 (1999) 046006, hep-th/9808061.

Nuclear Physics B 593 (2001) 671725

www.elsevier.nl/locate/npe

Fermionic quantum gravity

Lori D. Paniak a, , Richard J. Szabo b
a Joseph Henry Laboratories, Princeton University, Princeton, NJ 08544, USA
b The Niels Bohr Institute, Blegdamsvej 17, DK-2100 Copenhagen , Denmark

Received 25 May 2000; accepted 19 September 2000

Abstract
We study the statistical mechanics of random surfaces generated by N N one-matrix integrals
over anti-commuting variables. These Grassmann-valued matrix models are shown to be equivalent
to N N unitary versions of generalized Penner matrix models. We explicitly solve for the
combinatorics of t Hooft diagrams of the matrix integral and develop an orthogonal polynomial
formulation of the statistical theory. An examination of the large N and double scaling limits of
the theory shows that the genus expansion is a Borel summable alternating series which otherwise
coincides with two-dimensional quantum gravity in the continuum limit. We demonstrate that the
partition functions of these matrix models belong to the relativistic Toda chain integrable hierarchy.
The corresponding string equations and Virasoro constraints are derived and used to analyse the
generalized KdV flow structure of the continuum limit. 2001 Elsevier Science B.V. All rights
reserved.

1. Introduction and summary

Matrix models [1] provide a useful framework in which to extract quantitative information about complex statistical systems given only their general symmetry properties. The
key feature which enables this description is the existence of universal behaviour in these
models. Over the years random matrix models have been utilized in a wide range of studies in condensed matter systems [2], quantum chromodynamics [3], and low-dimensional
string theory [4,5]. They are also useful for solving various statistical mechanical models
and combinatorial problems [6]. One interpretation of random matrix ensembles is in terms
of generators of t Hooft diagram expansions which correspond to discretizations of compact Riemann surfaces. The continuum limits of such discretizations are realized as phase
transitions in the statistical mechanics of the matrix model and characterized by universal
data at the critical points.
* Corresponding author.

E-mail addresses: [email protected] (L.D. Paniak), [email protected] (R.J. Szabo).

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 3 1 - 3

672

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

Despite the success of matrix model technology over the years, there remain several
issues in random surface theory which are not adequately addressed by standard
approaches. The simplest, and one of the most fundamental issues regards a self-consistent
definition of the vacuum of pure two-dimensional quantum gravity. The conventional
matrix model methods for investigating this theory lead to inconsistencies. For instance,
bosonic matrix integrals are typically ill-defined in the physically interesting region of
parameter space corresponding to the non-perturbative continuum limit. Furthermore, they
are unable to describe the coupling of gravity to conformal matter fields of central charge
c > 1. It has been suggested [7] that above this c = 1 barrier the two-dimensional geometry
degenerates into a tree-like or branched polymer phase.
In this paper we will study a model of discretized random surfaces which is in part
motivated by such issues. It is parameterized by a zero-dimensional action involving
matrices with Grassmann-valued elements [8]. This model is known as the adjoint fermion
one-matrix model and it is defined by the partition function
Z

d d eN tr V () ,
(1.1)
ZN =
Gr(N)c

where

V ()
=

K
X
gk
k=1

k
()

(1.2)

Q
is a polynomial potential of order K 6 , and d d = i,j dij d ij is the standard
Berezin measure on the complexified Grassmann algebra Gr(N)c = Gr(N) R C. For any
P
matrix X the trace is normalized as tr X = i Xii . The matrix elements ij and ij , i, j =
1, . . . , N , are independent, complex-valued Grassmann numbers with the usual rules for

(this ensures that

complex conjugation, ij = ij , ij = ij , and ( ij kl ) = kl
ij
the generating function (1.1) is real-valued). The large N limit of this model has been studied in [8,9] and shown to exhibit the range of critical behaviour seen in the usual bosonic
one-matrix models. The topological expansion of the matrix integral (1.1) was considered in [10] and some novel critical behaviour was observed in [11]. The fermionic matrix
model can also be coupled to an ordinary complex bosonic one to produce a supersymmetric matrix model, which has been used to describe solutions of certain combinatorial
problems [12] and also to generate statistical models of branched polymers [13].
The most attractive feature of the fermionic matrix model (1.1) is that the integration
over Grassmann variables is always well-defined and the theory yields finite, computable
observables. The dimension N of the Grassmann matrices acts as a cutoff on the number
of terms in the partition function (1.1) which counts all fat-graphs generated. The fact that
the partition function is a polynomial in the coupling constants gk for finite N , coupled
with the fact that inner products in the Grassmann integration measure are effectively
calculable, leads to explicit expressions for statistical quantities in this class of models.
In the following we will use this property to solve analytically the combinatorial problem
of counting the dynamical triangulations generated by the fermionic partition function via
an explicit determinant representation of (1.1).

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

673

Since the generating function (1.1) is always a well-defined convergent object at finite
N , one might expect that its large N limit is also well-defined. In perturbation theory, the
Feynman rules for fermionic matrices include a factor of 1 for each fermion loop thereby
resulting in cancellations between large numbers of Feynman diagrams. This observation
has led to the conjecture that the genus expansion of the matrix integral (1.1) in the large N
limit is an alternating series which may be Borel summable. The evidence for this striking
property of the adjoint fermion one-matrix model is its intimate relationship with generalized Penner matrix models [810] which are defined by the Hermitian matrix integrals
Z
H
=
d eN tr(V ()+ log ) ,
(1.3)
ZN,
u(N)

where d = i,j dij is the Haar measure on the Lie algebra u(N) of the N N unitary group U (N). It has been argued that the matrix model (1.3) is equivalent to (1.1) for
This has been established in the
= 2, with the Hermitian matrix identified with .
[9], and by showing that their
large N limit by a saddle-point computation on
SchwingerDyson (loop) equations are identical order by order in the large N expansions
of the matrix integrals [8,10]. However, for < 0 the integration over Hermitian matrices
in (1.3) is not well-defined because of the logarithmic divergence at = 0. In this region the
integral can only be defined by analytical continuation and the matrix model only makes
sense at N = . This continuation produces complex-valued endpoints for the support
of the corresponding distribution of eigenvalues, analogous to what naturally occurs in
fermionic matrix models [8,9]. It is this complexification of the support of the distribution
of eigenvalues that is asserted to result in an alternating genus expansion. Moreover, it can
be argued [10] that the topological expansion coincides, modulo these sign factors, with
the usual Painlev expansions of Hermitian matrix integrals with polynomial potentials [4].
The near equivalence of the genus expansions strongly suggests a correspondence
between the Feynman graph expansions in the bosonic and fermionic models. One way of
understanding these issues is to appeal to a simplified version of the matrix integral (1.1),
namely, a fermionic vector model defined by anticommuting vector elements i and i ,
i = 1, . . . , N [14]. A fermionic vector model is related to a bosonic vector model, defined
with the same polynomial potential V , by a simple mapping of its Feynman diagrams
and an analytical continuation N 7 N/2 of the vector dimension which results in a
bosonic perturbation series with a cutoff on the number of terms generated. This analytical
continuation can be understood in terms of a complex-valued saddle-point of the bosonic
model in the large N limit, leading to an alternating genus expansion which produces
a well-defined, unique function [14]. As bosonic vector models generate random models
of branched polymers, the fermionic vector model thereby produces a statistical theory of
branched polymers whose continuum characteristics are equivalent order by order in large
N to the conventional models, but whose double-scaling expansion is a non-perturbatively
well-defined function.
The arguments given in [10] are based on the class of models with odd polynomial
potential, i.e., V () = V (). In these cases, the endpoints of the support contour
of the spectral density are located symmetrically about the origin on the imaginary axis

674

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

in the complex plane [8,9]. Generic polynomial potentials are difficult to treat because
the corresponding support contours are located asymmetrically in the complex plane and
the loop equations become cumbersome to deal with. Moreover, although the equations
of motion for the matrix models (1.1) and (1.3) are identical at each order of the large
N expansion, beyond the leading order they must be solved with different boundary
conditions and the non-perturbative solutions are not the same in the two cases [10]. One
of the main results of this paper will be the clarification of the role played by generalized
Penner models in fermionic matrix integrals. The analytical expressions for the partition
function (1.1) at finite N will be interpreted in terms of Toeplitz determinants, which will
naturally lead to a proof of the equivalence of the fermionic one-matrix model and a unitary
version of the Penner one-matrix model,
Z
U
[dU ] eN tr(V (U )log U ) ,
(1.4)
ZN =
U (N)

where [dU ] is the invariant Haar measure on the unitary group U (N). The important
feature of this equivalence is that it holds for any matrix dimension N . The finiteness
of the fermionic matrix model at finite N is captured by the compactness of the field
variable in (1.4). As we will see, the logarithmic term in the effective potential of (1.4) gives
certain restrictions on the configurations which contribute to the partition function and its
observables when the unitary matrix model is treated from a group theoretic perspective.
These restrictions naturally reproduce the basic characteristics of the matrix integral (1.1),
and moreover show precisely how the analytic, functional forms of the models (1.1) and
(1.4) are equivalent for any value of N . We will also prove directly that the loop equations
of the fermionic and unitary matrix models are identical, and that they admit the same,
perturbative Gaussian boundary conditions. It is a remarkable fact that a unitary matrix
model such as (1.4) admits an interpretation in terms of random surfaces.
The biggest advantage of the mapping of the fermionic matrix model to a unitary one
is that we now have an eigenvalue model to analyse. Grassmann-valued matrices are not
diagonalizable, and a direct analysis of fermionic matrix models is only possible using the
method of loop equations [8,9]. In this paper we shall exploit this eigenvalue representation
to give a systematic and detailed account of the exotic properties possessed by fermionic
matrix models. In particular we will develop a generalization of the orthogonal polynomial
technique on the circle which is necessary to study the unitary matrix model defined
by (1.4), and describe some of their functional analytic properties. These methods will
naturally lead, irrespective of the parity of the potential V , to a precise investigation
of the double-scaling limit of the fermionic matrix model which determines a nonperturbative, all-genus continuum theory of two-dimensional quantum gravity coupled to
matter. We will show that the fermionic matrix model leads to a Borel summable genus
expansion, in contradistinction to Hermitian one-matrix models. Unlike the vector models,
the diagrammatic, finite N situation is not quite so simple in the case of fermionic matrices,
but the final qualitative result will be the same. In particular, we shall obtain a nonperturbative definition of pure two-dimensional quantum gravity which is equivalent to the
usual models order by order in the genus expansion, but which is given by a well-defined

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

675

unique generating function. This non-perturbative model is what we shall call fermionic
quantum gravity in the following.
Another advantage of the eigenvalue representation is that it will allow us describe
the fermionic one-matrix model (1.1) as an integrable system. Generally, matrix models,
being examples of exactly solvable systems, are intimately related to certain reductions
of well-known integrable hierarchies of differential equations [15]. In the usual bosonic
Hermitian one-matrix models, describing two-dimensional quantum gravity, the partition
functions are related to the -functions of the integrable Toda chain hierarchy. In the
following we will see that the fermionic gravity theory is related to the -function of
a particular deformation of this hierarchy known as the relativistic Toda chain. We can
therefore understand many of the novel aspects of the fermionic theory, such as what
sort of matter coupling it involves, in terms of this deformation. Furthermore, we will
develop an operator theoretic approach to the equations of motion of the matrix integral
(1.1) and gain in this setting a precise picture of how this deformation leads to a welldefined Borel summable partition function of the statistical model. This sets the stage for
a description of the integrable differential hierarchies satisfied by the continuum theory
which generalizes the usual KdV flow structure of gravity [4,5]. It will also indicate that
the partition function (1.1) in the continuum limit serves as a concrete, non-perturbative
definition of two-dimensional quantum gravity.
The organization of the remainder of this paper is as follows. In Section 2 we will
derive an analytical expression for the partition function (1.1) and use it to describe the
solution to the problem of counting fermionic fat-graphs. We shall see that the matrix
integral generates a novel class of discretizations of Riemann surfaces whose polygons are
two-colourable. We will also briefly describe what sort of matter-coupled gravity theory
is represented by the fermionic one-matrix model, and prove its equivalence to the unitary
one-matrix model (1.4). In Section 3 we will present the formal orthogonal polynomial
solution of the fermionic matrix model. In Section 4 we will describe the large N limit
of the model from the perspective of these orthogonal polynomials. We will find that the
adjoint fermion one-matrix model actually possesses a sort of internal branched polymer
critical behaviour which can be understood in terms of the fermionic characteristics of
its fat-graph combinatorics. We shall also demonstrate that the usual universality classes of
two-dimensional gravity arise, and that they cannot be smoothly connected to the branched
polymer phases of the theory. Of course, the polymer effect is washed out at N =
because the surface effect is of higher order in N . We also construct the genus expansion
of the theory and establish its Borel summability explicitly. In Section 5 we present the
operator approach to the fermionic matrix model and use it to interpret the partition
function as a -function. We derive the Virasoro constraints satisfied by this -function,
which at the same time establishes explicitly the equivalence of the loop equations of
the matrix models (1.1), (1.3) with = 2, and (1.4), in the large N expansion. We
then use these descriptions to give an functional analytic explanation for the nature of
the topological expansion in the fermionic one-matrix model. Finally, in Appendix A we
describe properties of observables of the statistical model (1.1) using the unitary matrix
model and its orthogonal polynomials, while in Appendix B we show that the orthogonal

676

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

polynomials of the Gaussian fermionic one-matrix model are given in terms of confluent
hypergeometric polynomials.

2. Determinant representations of the partition function

In this section we will present an analytic solution to the problem of counting
fermionic fat-graphs. We will obtain a closed expression for the generating function of
such discretized random surfaces, and at the same time we will be naturally led to the
equivalence of the fermionic matrix model and the unitary matrix model which will occupy
the bulk of our analysis in subsequent sections. We will also obtain in this way a geometric
picture of what sort of theory of random surfaces is represented by the adjoint fermion
one-matrix model and also of the precise role of the corresponding Penner interaction.
2.1. Hermitian representation
One intriguing feature of the fermionic one-matrix model (1.1) is the extent to which it
can be solved at finite-N . This is a consequence of the convergence of the integration over
Grassmann variables, in contrast to the case of bosonic matrix models. The combinatorics
of the fat-graph expansion of the fermionic partition function ZN can be determined by
using the observation of [11] that the partition function and observables of the matrix
model (1.1) can be evaluated explicitly by introducing two N N Hermitian matrices
and defined by
Z
Z
d

d ( ), () =
ei tr .
(2.1)
1=
2
(2)N
u(N)

u(N)

By inserting these definitions into (1.1), the fermionic partition function can be written as
Z
Z
Z
d

d eN tr V ()
d d
ei tr ()
ZN =
2
(2)N
u(N)
Gr(N)c
u(N)
ZZ
1
=
d d eN tr V ()+i tr detN (i)
2
(2)N
u(N)


Z

d eN tr V () detN
(),
(2.2)
=

u(N)

so that

ZN = det

N tr V () 
e
 .

=0

(2.3)

A similar expression can also be derived for the correlators of the fermionic one-matrix
model [9,11]. However, it is difficult to write down an explicit expression for (2.3) which
is informative and amenable to analysis.

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

677

To evaluate the determinant in (2.3) explicitly, we shall instead write the fermionic
matrix model as a random theory of the two Hermitian matrices and introduced above,
keeping only the second line of the identity (2.2). In this way, the fermionic one-matrix
model can be written as the Hermitian two-matrix model
ZZ
N N(N+1)
d d eN tr(V ()+log(i)+i) ,
(2.4)
ZN =
2
(2)N
u(N)

where we have rescaled ij Nij . The doubling of degrees of freedom is required to

compensate for both the matrix and its adjoint . The Hermitian two-matrix model (2.4)
describes two-dimensional discretized gravity with a single type of matter state at each
of the vertices. It does not admit the usual Z2 Ising symmetry of a classical spin lattice
that characterizes the unitary minimal models of rational conformal field theory [4]. Thus
the matter states in the worldsheet interpretation of the fermionic matrix model are more
complicated degrees of freedom than just simple Ising spins or other conventional types of
conformal matter. The logarithmic nature of the effective two-matrix potential in (2.4) is
just another reflection of the connection with Penner matrix models. The original Penner
model [16] (gk = 0 for k > 1, g1 = 1 and = 1 in (1.3)) we used to calculate the virtual
Euler characteristics of the moduli spaces of compact Riemann surfaces. For > 0 the
generalized models are intimately related to the coupling of gravity to matter fields of
central charge c = 1 [10]. Note that the Penner interaction in (2.4) is well-defined. We
may refer to the fermionic matrix model as a model of two-dimensional quantum gravity
interacting with Penner matter. As we will see later on, it is this worldsheet model which
leads to a well-defined nonperturbative theory of quantum gravity in two-dimensions which
we will call fermionic gravity.
The partition function (2.4) can be written in terms of a double eigenvalue distribution
by diagonalizing the Hermitian matrices = U diag(1 , . . . , N )U and = V diag(1 ,
. . . , N )V by unitary transformations, where i , i R are the eigenvalues of , .
Computing the Jacobian for the change of integration measure and using the Itzykson
Zuber formula [17] to integrate out the unitary degrees of freedom, we arrive at the
eigenvalue model
N Z Z
Y
NV (i )+iNi i
di di N
(1 , . . . , N )(1 , . . . , N ), (2.5)
Z N = cN
i e
i=1

where
cN =
and

(iN 2 )N(N+1)/2
,
Q
(2)N N
n=1 n!

 j 1  Y
= (xi xj )
(x1 , . . . , xN ) = det xi
i,j

(2.6)

i<j

is the Vandermonde determinant. Using (2.6) we can then write the partition function as



(2.7)
ZN = cN N! det i1 , j 1 ,
i,j

678

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

where we have introduced the inner product

F (), G()

Z Z
d d N eNV ()+iN F ()G(),

(2.8)

on the vector space C[] C[] of complex coefficient polynomials in (, ). A remarkable feature of the fermionic matrix model is the extent to which its inner product (2.8) can
be computed. For arbitrary polynomial potential, integrating over first in (2.8) leads to
the result




2 i N+k 

k
NV () 
,
(2.9)
F () e
F (), =

N N
=0
which is valid for any analytic function F () and any integer k > 0. Using (2.9), the
representation (2.7) of the partition function as the determinant of the moment matrix of
the inner product (2.8) thereby yields
"
N+j i  #


[N]2
0
+ NV ()
det
1
(2.10)
ZN = (1)
i,j

=0
where, for any integer K > 1, [N]K denotes the largest integer less than or equal to N/K.
Here and in the following a prime denotes differentiation. The expression (2.10) is the
desired explicit expansion of the determinant operator in (2.3).
2.2. Combinatorics of fermionic ribbon graphs
In contrast to the determinant form (2.3), the representation (2.10) leads to an explicit
expression for the perturbation series of the fermionic one-matrix model. For a polynomial
potential (1.2) of order K, the determinant in (2.10) may be evaluated explicitly by
expanding the exponential function in its Taylor series and applying the multinomial
theorem to each term in the expansion to get

[L]2
[L]
X
XK
L!
L NV () 
L
e
=
(Ng
)

1

L

(L 2k2 KkK )! k2! kK !

=0

K 
Y
`=2

k2 =0

kK =0

g`
`N `1 (g1 )`

k`
,

(2.11)

for any integer L > 0. Normalizing by the Gaussian model for which V () = g1 and
Gauss = (1)[N]2 (Ng )N 2 , the partition function (2.10) may be written as
ZN
1
ZN
Gauss
ZN

 [N+j
i]K
Xi]2 [N+j
X
= det

i,j

k2 =0

K 
Y

`=2

kK =0

g`
`1
`N (g1 )`

(N + j i)!
P
(N + j i ` `k` )!k2 ! kK !

k` 
.

(2.12)

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

679

We can extract the sums in (2.12) out of the determinant by using the multilinearity of the
determinant as a function of its N rows to get
" K N
km(n) #
X
X
YY 1 
gm
ZN
=

Gauss
(n)
m1 (g )m
ZN
1
km ! mN
(1)
(N)
(1)
(N)
k2 ,...,k2 >0

"

kK ,...,kK >0

m=2 n=1

K
Y
(N + j i)!
(i)

[N + j i]` k`
det
P
(i)
i,j (N + j i
` `k` )! `=2

#
(2.13)

where denotes the step function with the convention (0) = 1.

The expression (2.13) leads to a relatively straightforward expansion of the partition
function in powers of the coupling constants g` , ` = 2, . . . , K. It represents the formal
solution to the problem of counting the (connected and disconnected) Feynmant Hooft
diagrams of the adjoint fermion one-matrix model. It is a closed formula, in terms of a sum
over a large set of integers, for the generating function of ribbon graphs of the fermionic
matrix model. A fermionic ribbon graph has much more structure to it than a conventional
fat-graph, and so we shall now formulate the rules for generating them from the matrix
integral. The fermionic propagator is h ij kl iGauss = il j k /Ng1 which we represent in
the usual way as a double line, along with an orientation defined by an arrow pointing
towards the matrix. Vertices are likewise given an orientation by assigning an outgoing
arrow for a line and an incoming arrow for a line. Fat-graphs are now drawn with the
rule that only and lines can contract. Each such graph is dual to some discretization
of a Riemann surface in the standard way. The surface may be two-coloured by shading
the triangles in a polygon if and only if their outer edge is dual to a propagator with
an incoming arrow. In this way black edges are always associated with matrices and
with s, it follows
white ones with . Since the fermionic propagator only connects s
that every such discretization can be two-coloured. Thus the fat-graphs generated by the
fermionic matrix integral can be obtained by drawing all discretizations associated with
fermionic k-point vertices in terms of the corresponding Hermitian 2k-point vertices, and
keeping only those graphs which are two-colourable. The even parity of the Hermitian
potential V ( 2 ) is required because only discretizations with even-sided polygons can be
two-coloured.
However, the Grassmann nature of the generating matrices yields important changes to
the rules for counting such triangulations. For a given vertex in the Wick expansion of
the matrix integral, we fix a chosen line. Lines are joined into propagators with a lefthanded orientation. Every time a line is joined into a line (rather than the other way
around), we twist the ribbon and thereby reverse its orientation. To each such twisted
line, we associate a factor of 1. This standard sign factor for fermionic fields yields
significant reductions in the overall number of ribbon graphs which are actually generated
by the fermionic matrix model, because of the cancellations which occur, for example,
between fermionic diagrams that are topologically the same but which are twisted relative
to one another. Thus topologically equivalent diagrams do not necessarily add up, but
may have the effect of cancelling each other. As a simple example, consider the quadratic

680

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

potential, K = 2. There are two graphs which contain only a single four-point vertex,
but they contribute with equal magnitude and opposite sign to the generating function.
This vanishing contribution is the coefficient of the g2 /2N(g1 )2 term in the perturbative
expansion of the partition function. Symbolically, it is given by the Wick expansion of the
Gaussian correlator

>

< 
>



2

=
tr()

= 0.
Gauss


(2.14)

<
Note the twist in the second propagator of the second graph in (2.14), which induces the
relative minus sign. Notice also that the usual toroidal four-point graph does not appear
in (2.14), because it cannot be two-coloured. This vanishing contribution is also easily
found from the analytical formula (2.13). The leading term is 1 and it comes from the
configuration whereby all ks are 0. The terms of order g2 come from the configuration
whereby only one k is non-vanishing and equal to unity. In the determinant in (2.13),
the only s which are not equal to unity are those which appear in row i with k (i) = 1.
The determinant therefore vanishes, reproducing the graphical result (2.14). The remaining
fat-graph combinatorics are now readily determined from the graphs of the Hermitian onematrix model with even potential V ( 2 ) by keeping only those which are two-colourable
and incorporating the appropriate twists. The determinant formula (2.13) is readily seen
to reproduce the correct combinatorics, with appropriate minus signs coming from the
determinant.
These arguments readily generalize to arbitrary polynomial potentials. Generally, the
total number of 2`-valent vertices in a given ribbon graph as determined by the
formula (2.13) is the power of g` which is given by
n2` ( ) =

N
X

k`(n) ,

(2.15)

n=1

and, because of Fermi statistics, this number is bounded as

0 6 n2` ( ) 6

N
X
[2N n]` .

(2.16)

n=1

It follows that there are only a finite number of fermionic fat-graphs generated. The
twisting of ribbons and also the finiteness of the perturbation series may be attributed
to the inclusion of a logarithmic interaction 2 log in the corresponding Hermitian onematrix model with potential V ( 2 ). This role of the Penner potential will be demonstrated
explicitly in the next subsection.
The determinant expansion (2.13) is therefore an analytic expression of the combinatorial formula

K 
X (1)t ( ) (Ng1 )( ) Y
g` n2` ( )
ZN
=
,
(2.17)
Gauss
|Aut( )|
g1
ZN

`=2

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

681

where the sum runs through all (connected and disconnected) two-colourable fat-graphs
with n2` ( ) vertices of order 2` bounded as in (2.16), and ( ) is the Euler characteristic
of . The alternating sign factor in (2.17), with t ( ) the total number of twisted ribbons
of , is due to Fermi statistics, while Aut( ) is the automorphism group of the unmarked
graph . The sum over fat-graphs in (2.17) is finite, since the maximum number of vertices
that a given diagram can have is
vmax =

N
K X
X

[2N n]` .

(2.18)

`=2 n=1

Each such ribbon graph is dual to a tessellation of a Riemann surface of Euler

characteristic ( ) by n2` ( ) 2`-valent tiles for ` = 2, . . . , K. Because of the twist
factors, the overall combinatorial numbers generated by the fermionic matrix integral will
be drastically reduced. Moreover, depending on the precise details of the potential V , the
perturbative or topological expansions of the matrix model may be alternating series. The
above considerations show that there is clearly no simple mapping between the Hermitian
and fermionic matrix models. These features will have remarkable implications later on for
the topological expansion of the matrix integral (1.1).
2.3. Unitary representation
We will now derive an alternative dual representation of the perturbative expansion
of the fermionic partition function. For this, we note that the derivatives in (2.9) can be
evaluated in terms of a contour integration about the origin of the complex plane as
 N+k+1
I


F (z) eNV (z)
i
(N + k)!
dz
.
(2.19)
F (), k =
N
zN+k+1
z=0

The partition function (2.10) may then be written as


N   N+k
Y
i
(N + k 1)! det A,

ZN = cN N!
N

(2.20)

k=1

where we have defined the N N matrix

I
eNV (z)
dz N+j i+1 .
Aij =
z

(2.21)

z=0

The crucial observation now is that the matrix elements (2.21) depend only on the
differences j i of row and column positions, i.e., Aij = Aj i . Matrices with such a
symmetry are known as Toeplitz matrices and their determinants can be evaluated in terms
of averages over the unitary group [19]. The Toeplitz determinant deti,j [Aj i ] appearing
in (2.20) is associated with the Laurent series of a function A through the definition
A() =

X
n=

An n .

(2.22)

682

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

The matrix elements An are then interpreted as the Fourier coefficients of A on the unit
circle || = 1,
I
d n
A().
(2.23)
An =
2i
Using the fact that a determinant is a linear function of each of its rows, it follows that
det A =

N I
Y
k=1

dk
A(k )kk (1 , . . . , N ).
2ik

(2.24)

Replacing the right-hand side of (2.24) by its average over all permutations of the
integration variables k yields
det A =

N I

2
1 Y
dk
A(k ) (1 , . . . , N ) ,
N!
2ik

(2.25)

k=1

which shows that the Toeplitz determinant is given by the unitary matrix integral
QN1
Z
n=1 n!
[dU ] det A(U ).
det A =
(2)N(N1)/2

(2.26)

U (N)

Upon substituting (2.21) into (2.22), we can interchange the sum and contour integration
provided we take into account that the latter vanishes for N + n < 0. We then have
I
eNV (z)
= 2i eN(V ()log ) .
dz N
(2.27)
A() =
(z )
z=0

From (2.20), (2.26) and (2.27) it follows that the partition function of the adjoint fermion
one-matrix model (1.1) is completely equivalent, for any dimension N , to the N N
unitary matrix model
Z
[dU ] eN tr(V (U )log U ) ,
(2.28)
ZN = kN
U (N)

Q
where kN = N N(N+1)/2 /(2)N(N1)/2 N
n=1 (N + n 1)!. We see that the fermionic
one-matrix model for any N is completely equivalent to both a Hermitian two-matrix
model, and a unitary one-matrix model, both of which involve Penner-type interactions.
In particular, the unitary representation clearly demonstrates that the equivalence with the
Hermitian Penner one-matrix model is not true at finite N , but rather only at N = .
Note that the matrix integral (2.28) is perfectly well-defined at finite N , but it involves an
integrand which is not a real-valued potential.
With the partition function given as an integral over the unitary group, we can clarify
the geometric role of the Penner-type potential that characterizes fermionic matrix models,
and also give an alternative method for a perturbative expansion in terms of the coupling
constants in the potential V . For this, we integrate over the U (1) factor of the unitary group
U (N) = U (1) SU(N)/ZN and leave an SU(N) integral,

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

i U

ZN = kN

d [dU0 ] eN(tr V (e

683

0 )iN)

U (N)

kN
=
N

Z2
d e
0

iN 2

i U

[dU0 ] eN tr V (e

0)

(2.29)

SU(N)

The role of the U (1) integration over is to restrict the terms in the SU(N) integration to
those involving only N 2 factors of U0 . Equivalently, we may expand the invariant function
which is the integrand of (2.29) in U (N) characters nE as
X
cnE nE (U ), U = ei U0 ,
(2.30)
eN tr V (U ) =
nE

where cnE C are functions of the coupling constants of the potential V , and the sum goes
over all irreducible polynomial representations of U (N) which may be parameterized by
their highest weight components n1 > > nN > 0 associated with the lengths of the lines
in the corresponding Young tableaux. Since
nE (U ) = eiC1 (En) nE (U0 ),

(2.31)

where
n) =
C1 (E

N
X

nk ,

(2.32)

k=1

is the linear Casimir of the U (N) representation nE (the total number of boxes in the Young
tableau), it follows that the constraint on the SU(N) integration in (2.29) is a restriction to
terms with linear U (N) Casimir C1 equal to N 2 . We may therefore write
" Z
#
kN
N tr V (U0 )
[dU0 ] e
,
(2.33)
ZN =
N
2
SU(N)

C1 =N

where the bracket means to restrict the character expansion of eN tr V (U0 ) to those nE
n) = N 2 . The role of the logarithmic interaction in (2.28) may therefore be
with C1 (E
characterized as restricting the matrix integral to completely filled Young diagrams.
The SU(N) integral in (2.33) can be evaluated using standard techniques from lattice
gauge theory [20]. For this, we introduce the generating function
Z
[dU0 ] etr J U0 ,
(2.34)
W (J ) =
SU(N)

with the property


 
Z


[dU0 ]f (U0 ) = f
,
W (J )
J
J =0

(2.35)

SU(N)

where J is an arbitrary N N matrix and f an arbitrary function on SU(N). Using the

invariance of the Haar measure [dU0 ] under SU(N) rotations, we have W (J ) = W (U J V ),

684

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

U, V SU(N), from which it can be shown that the generating function may be expanded
in powers of determinants of J as [20]
W (J ) =

(2)

N(N+1)
1
2

QN

k=0

(det J )k

n=1 (n + k

1)!

(2.36)

n) = N 2 implies that
Applying this result to the partition function (2.33), the restriction C1 (E
only the k = N term of the expansion (2.36) contributes, giving



.
(2.37)
ZN = eN tr V ( J ) detN J 
J =0

Comparing (2.3) and (2.37), we see that the finite N expressions for the fermionic partition
function in the Hermitian two-matrix and unitary one-matrix representations are dual to
each other. We can make this duality precise by writing (2.37) as
ZZ

1
dJ d ei tr J eN tr V ( J ) detN J
ZN =
2
N
(2)
gl(N)


ZZ

1
N tr V (i)
N
ei tr J
dJ d e
det i
=
2

(2)N
Z
=

gl(N)




 


N tr V (i)
N
N tr V () 
d () det i
= det
e
e
 .

=0
N

gl(N)

(2.38)

Some further aspects of the unitary representation of the fermionic matrix model are
discussed in Appendix A. In Section 5 we will show explicitly that the equations of motion
of the unitary and fermionic one-matrix models are identical.

3. Orthogonal polynomial solution

In this section we shall demonstrate how to generalize the orthogonal polynomial
technique [21] to the adjoint fermion one-matrix model (1.1), with the goal of solving the
matrix model in the large N limit. Although the fermionic matrix model is not naturally
an eigenvalue model, since it is not possible to diagonalize a Grassmann valued matrix,
we can exploit its equivalence with the unitary one-matrix model (2.28). We will thereby
define the orthogonal polynomials of the fermionic matrix model (1.1) to be the orthogonal
polynomials associated with the corresponding unitary eigenvalue model. This definition
will lead, as we shall see, to a well-defined real-valued solution for the fermionic matrix
model.
3.1. General properties
Let us consider the fermionic partition function in the representation (2.20), (2.21),
which involves the measure
eNV (z)
(3.1)
d(z) = N+1 dz,
z

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

685

on the unit circle zz = 1. The partition function (2.20) may then be written as the
determinant of the moment matrix of the measure (3.1),




(3.2)
ZN = dN det zi1 z j 1 ,
i,j

where dN is an irrelevant numerical constant and we have defined the inner product
I



d(z) F (z) G(z),

(3.3)
F (z)G(z) =
z=0

on the space C[z] C[z] of Laurent polynomials on the unit circle. The construction of
orthogonal polynomials to evaluate inner products on the circle such as (3.3) with real,
positive definite measure d was discussed long ago in the mathematics literature [22]
and subsequently in the physics literature [23]. Some aspects of generic unitary matrix
models are also dealt with in [24,25]. In the present case, the measure (3.1) is complexvalued, and we must deal accordingly with defining the appropriate system of orthogonal
polynomials.
These are the monic polynomials (n (z), m (z)) of order (n, m) which form a complete
set in the space C[z] C[z] and which are bi-orthogonal in the measure (3.1),



(3.4)
n (z)m (z) = hn nm ,
where hn are some constants. They are normalized as
n (z) = zn

n1
X

pn,k zk ,

m (z) = zm

k=0

m1
X

lm,k zk ,

(3.5)

k=0

where the coefficients pn,k and lm,k can be formally obtained from the usual Gram
Schmidt orthogonalization procedure applied to the monomials (zn , zm ) and the inner
product (3.3). In fact, by iterating (3.5) we find that the monomials (zn , zm ) can be
expanded as
!
kj1 1
n
n1 kX
1 1
X
X
X
n

pn,k1 pk1 ,k2 pkj1 ,kj kj (z) ,

z = n (z) +
j =0

= m (z) +

k1 =0 k2 =0

m
m1
1 1
X
X kX
j =0

k1 =0 k2 =0

kj =0

kj1 1

lm,k1 lk1 ,k2 lkj1 ,kj kj (z) .

(3.6)

kj =0

It follows that any polynomial in the space C[z] C[z] of degree (n, m) can be expressed
as a linear combination of (k (z), ` (z)) with k 6 n and ` 6 m. From (3.5) it follows they
define a change of basis (zk1 , z`+1 ) 7 (k1 (z), `1 (z)) of the vector space C[z]
C[z] that leaves the partition function (3.2) invariant and at the same time diagonalizes the
inner product (3.3). In particular, from (3.4) we have
N1
Y




RnNn ,
ZN = dN det i1 (z)j 1 (z) = dN hN
0
i,j

n=1

where we have introduced the recursion coefficients

(3.7)

686

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

hn
Rn =
,
(3.8)
hn1
H
and h0 = z=0 d(z) is the normalization of the measure (3.1).
The reason why two independent sets of functions corresponding to clockwise and anticlockwise rotations on the circle are required here is the complexity of the measure (3.1).
In the classical case whereby the measure d is real and positive [22], the polynomials
n (z) and n (z) are related to each other by complex conjugation. The present system of
polynomials may be thought of as a deformation of those for positive definite measures
on the unit circle. This deformation is essentially encoded in the coefficients pn+1,0 and
ln+1,0 . In order for the GramSchmidt process to work for the complex measure (3.1),
one needs polynomials in both z and 1/z and a priori these polynomials are unrelated.
Alternatively, we can think of this doubling as the usual doubling of degrees of freedom
due to the fermionic nature of the original matrix model. As the orthogonal polynomials
define a complete bi-orthogonal system of functions, they satisfy the completeness relation

X
n (z)n (z)
n=0

hn

= () (z) = zN+1 eNV (z) (z),

(3.9)

on the vector space C[z] C[z], where the Dirac delta-function corresponding to the
measure d is defined so that
I
d(z)F (z) ()(z z0 ) = F (z0 ),
(3.10)
z=0

for any function F (z) on the unit circle. The relation (3.9) is understood as an analytical
continuation (to the distribution space that is the completion of C[z] C[z]) which can

be thought of as a representation of the potential V ()

in terms of the orthogonal
polynomials.
The problem of evaluating the partition function of the fermionic matrix model therefore
boils down to evaluating the coefficients Rn appearing in (3.7). For this, we will first derive
a few properties of the system of orthogonal polynomials introduced above. Completeness
implies the relations
zn (z) = n+1 (z) +

n
X

(n)

Pk k (z),

k=0
m

X (m)
1
m (z) = m+1 (z) +
Lk k (z).
z

(3.11)

k=0

completely determine the form of the orthogonal

The coefficients Pk(n) and L(m)
k
polynomials and hence the partition function. To see this, we use (3.5) and (3.11) to write
n (z) = zn1 (z)
(n)

= z n P0

n1
X
k=0
n1
X
k=1

Pk(n1) k (z)
(n1)
k

pn1,k1 + Pk

n2
X
k=0

(n1)

Pk

k1
X
j =0

pk,j zj .

(3.12)

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

687

Comparing the various polynomial coefficients in (3.12) with those of the definition (3.5),
we arrive at the iterative relations
(n1)
,
pn,n1 = pn1,n2 + Pn1

(3.13)
n1
X

pn,k = Pk(n1) + pn1,k1 +

pj,k Pj(n1) ,

1 6 k 6 n 2,

(3.14)

j =k+1

pn,0 = P0(n) +

n1
X

pj,0 Pj(n1) .

(3.15)

j =1

The relations (3.13) and (3.15) can be iterated straightforwardly and by induction we have
pn,n1 =

n1
X

(j )

Pj ,

(3.16)

j =0
(n)
pn,0 = P0

n3 X
n1 jX
1 1
X

j1 =1 j2 =1

k=1
(1)

jk1 1

n1 jX
1 1
X

(1)

+ P 0 P1

j1 =1 j2 =1

jk =1

!
(j 1) (j )
(n1) (j 1)
Pj1 Pj2 1 Pjk k1 P0 k

jn3 1

jn2 =1

(n1)

Pj1

(j 1)

Pj2 1

(j

n3
Pjn2

1)

(3.17)

which can be substituted into (3.14) to iteratively determine the remaining pn,k . Analogous
relations exist between the coefficients lm,k and L(m)
k .
We will also need higher degree versions of the completeness relations (3.11). For this,
[`]
by
we define constants Pn,k
z` n (z) =

n+`
X

[`]
[`]
Pn,k
k (z) with Pn,n+`
= 1,

(3.18)

k=0

and use (3.11) to write

z

`+1

n (z) =

n+`+1
X

[`+1]
Pn,k
k (z) =

k=0

n+`
X

[`]
Pn,k

k+1 (z) +

k=0

k
X

!
(k)
Pj j (z)

(3.19)

j =0

Using completeness of the orthogonal polynomials we arrive at a recursive relation for the
[`]
,
coefficients Pn,k
[`+1]
[`]
= Pn,k1
+
Pn,k

n+`
X

[`]
Pn,j
Pk ,
(j )

0 6 k 6 n + ` + 1,

(3.20)

j =k
[`]
[0]
0, Pk 0 for k > n, and the initial condition Pn,k
= nk .
with Pn,1
(n)

3.2. Recursion relations

We will now determine the coefficients appearing in the completeness relations using
the above properties of the orthogonal polynomials. This will lead to a set of recursion

688

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

equations for the coefficients Rn which thereby completely determines the solution of
adjoint fermion one-matrix model. Consider the bi-orthogonality relation



(3.21)
n+1 (z)m (z) = 0, 0 6 m 6 n.
Using the completeness relations (3.11) we can write (3.21) as



(n)
hm .
zn (z)m (z) = Pm

(3.22)

The inner product in (3.22) can be evaluated by grouping the factor of z with the
polynomial m (z) with the result




zn (z)m (z) = n (z)m1 (z) + O zm+2 lm,0 z

 

(3.23)
= lm,0 zn (z)1 ,
where we have used the fact that, aside from the term lm,0 z which comes from the constant
part of the polynomials in (3.5), the function z m (z) is a sum of polynomials of degree
6 m 1. The final inner product in (3.23) can be evaluated in a straightforward fashion
using the relation



(3.24)
zn (z)n+1 (z) = hn+1 ,
which follows from (3.11). Again, by grouping the factor of z with n+1 (z) one finds

  hn hn+1
.
zn (z)1 =
ln+1,0

(3.25)

Substituting (3.25) and (3.23) into (3.22), we find a concise relation between the constant
(n)
terms of the polynomials and the completeness coefficients Pm ,
lm,0 hn+1 hn
ln+1,0
hm

n
Y

Q (R
Qk Rk
m n+1 1)
=
k=m+1

Qn (Rn+1 1)

(n)
=
Pm

for 0 6 m < n,

(3.26)

for m = n,

where we have defined

ln,0
.
Qn =
ln+1,0

(3.27)

The completeness relation (3.11) for the polynomials can therefore be written as
zn (z) = n+1 (z) +

n
hn+1 hn X lk,0
k (z).
ln+1,0
hk

(3.28)

k=0

Subtracting this relation from itself under the shift of index n n 1 leads to the threeterm recursion relation for the polynomials
n+1 (z) = zn (z) Qn

Rn+1 1
n (z) zRn n1 (z) .
Rn 1

(3.29)

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

689

Analogous relations for the polynomials may also be derived. Equating the constant
terms on both sides of (3.29) yields
Rn+1 1
pn+1,0
,
= Qn
pn,0
Rn 1

(3.30)

so that
pm,0 hn+1 hn
pn+1,0
hm

n
Y
Rk Rk 1

(1)nm 1 Rm Rn+1 1

Qm Rm+1 1
Qk Rk+1 1
k=m+1
=
1R

Qn

L(n)
m =

for 0 6 m < n,
(3.31)
for m = n.

Substituting (3.31) into the completeness relation (3.11) for the polynomials and
subtracting the resulting expression from itself under the index shift n n 1, we arrive
at the three-term recursion relation


1
1
(3.32)
n (z) = Rn n1 (z) Qn n+1 (z) n (z) .
z
z
The recursion relations (3.29) and (3.32) have been derived without any reference to
the particular details of the model, i.e., they hold independently of the precise form of
the potential V of the fermionic matrix model. Consequently, these relations are merely
kinematical in origin. In order to solve for the dynamics of the matrix model we need
more information. From a more pragmatic point of view, in order to solve for the partition
function we need to find a solution for the coefficients Rn , but the recursion relations also
involve the quantities Qn . More information is needed to link these two objects. As we will
see, one only needs two dynamical relations to obtain a closed set of recursion relations
for any polynomial potential. The first one is given from (3.4) and (3.5) as



(3.33)
n (z)0n1 (z) = (1 n)hn .
Integrating by parts on the left-hand side of (3.33) gives



(N + 1) n (z) 1z n1 (z)




N V 0 (z)n (z)n1 (z) 0 (z)n1 (z) = (1 n)hn

n

(3.34)

Using the completeness relation (3.11) for the polynomials and the fact that n0 (z) =
nn1 (z) + O(zn2 ), we arrive at



(3.35)
N V 0 (z)n (z)n1 (z) = (N + n)hn nhn1 .
For a polynomial potential (1.2) of degree K, we may use (3.18) to write (3.35) as an
equation for the recursion coefficients Rn ,
Rn =

K1
N X
n
[k]
+
gk+1 Pn,n1
.
n+N
n+N
k=1

(3.36)

690

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

Closely related to the first one, the second dynamical relation is given by the biorthogonality relation



(3.37)
n (z)0n2 (z) = 0.
Integrating this by parts gives




N V 0 (z)n (z)n2 (z) = n0 (z)n2 (z) .

(3.38)

The right-hand side of (3.38) can be evaluated using (3.5), (3.6) and orthogonality, and
keeping only those terms in the polynomials which have a non-vanishing overlap with
n2 (z) in (3.38). This gives


0 
n (z)n2 (z) = nzn1 (n 1)pn,n1 zn2 + O zn3 n2 (z)


= (n 1)pn,n1 npn1,n2 hn2 .
(3.39)
By defining
pn,n1
,
N
we thus find that the second dynamical relation reads
Pn =

nPn1 (n 1)Pn =

K1
X

[k]
gk+1 Pn,n2
.

(3.40)

(3.41)

k=1

The coefficients Pn can be related to the coefficients Qn and Rn by using (3.16) and (3.26)
to get
N(Pn+1 Pn ) = Qn (Rn+1 1).

(3.42)

To summarize the results of this subsection, we note that the right-hand sides
of (3.36) and (3.41) can be evaluated, using the recursion formula (3.20), in terms of
(n)
the completeness coefficients Pm . Using (3.26), it follows that, for any polynomial
potential, (3.36), (3.41) and (3.42) form a closed set of recursion relations involving only
the coefficients Rn , Qn and Pn . An explicit system of fermionic orthogonal polynomials
is constructed in Appendix B.
3.3. Free energy
The final quantity we need for the solution of the fermionic matrix model is the
normalization h0 of the measure (3.1). This integral can be evaluated explicitly to give

I
eNV (z) 2i N NV (z) 
dz N+1 =
e
(3.43)
h0 =
 .
N! zN
z
z=0
z=0

We substitute these quantities into (3.7) and normalize by the Gaussian partition
2
Gauss
= (1)[N]2 (Ng1 )N for which V (z) = g1 z, Rn = n/(n + N) and
function ZN
h0 = 2i(Ng1 )N /N!. For a polynomial potential, we may evaluate (3.43) explicitly
using (2.11), and using the multinomial theorem we thus find that the free energy FN =
1
N
log ZGauss
is given by
N2
ZN

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

FN

X
=
(1)n (n 1)!
n=1

X
(2)

(2)

(K)

06k0 ,...,k[N] 6n

j =2 `j =1

(K)
K

06k0 ,...,k[N] 6n

K [N]
Y
Yj 

691
(m)
`,m k` n)
(m)
`,m k` !

`m k (j)
K 
Y
`j
N!
gm
P
m1
m
(N i i`i )!`2 ! `K !
mN
(g1 )
m=2


N1
(n + N)Rn
1 X
n
log
,
+
1
N
N
n

(3.44)

n=1

where the coefficients Rn are determined from (3.36).

Generally, in the large-N limit, the variable x n/N becomes a continuous parameter
in the interval [0, 1]. We assume that in this limit the recursion coefficients Rn (any of the
quantities Rn , Qn or Pn ), determined as the solutions of the recursion equations derived
in the previous subsection in the large-N limit, become continuous functions R(x) of x
[0, 1]. This is justified by the form of the recursion relations which suggest the replacement,
Rn R(n/N) = R(x). It follows that shifts in the index n are related to sub-leading
contributions in the large N limit,
Rn+a

 

a X 1 a k1 k R(x)
R(x) +
,
N
k! N
x k

(3.45)

k=1

and the second, finite sum in (3.44) can be approximated by a simple one-dimensional
integral over x [0, 1] at N = . To deal with the first term in (3.44), we need to
determine the large N limit of the normalization constant h0 . For this, we note that there
is nothing particularly special about the choice of integration contour in (3.43), as it was
merely introduced in (2.19) as a means of evaluating the derivatives in (2.9). In particular,
since its integrand is an analytic function, we can deform the contour arbitrarily in the
complex plane, and in particular to one out at infinity. In the large N limit, the onedimensional integral h0 may then be evaluated using the saddle-point approximation. In
this way we find that the genus 0 free energy is given by
Z1
F0 lim FN = f0 +

dx (1 x) log

(1 + x)R(x)
,
x

(3.46)

where
f0 = V ( ) log g1 ,

(3.47)

with the solution of the order K algebraic saddle-point equation

V 0 ( ) = 1,

(3.48)

for the branch which has the perturbative Gaussian limit |g2 ==gK =0 = 1/g1 and for
which the function (3.47) is real-valued.

692

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

4. Topological expansion
In this section we will solve for the continuum limit of the discretized random surface
theory represented by the adjoint fermion one-matrix model. We will consider two limits of
the matrix model. The first one is the naive large-N limit which captures the leading critical
behaviour and gives a non-perturbative solution to the problem of counting fermionic
ribbon graphs of spherical topology. The second one is the double-scaling limit in which an
appropriate coupling constant is tuned in the limit N in order to obtain contributions
from all orders of the 1/N expansion of the free energy and which yields the complete
topological expansion of the fermionic matrix model.
4.1. Critical behaviour
We will start by making some general remarks concerning the large N limit of the
fermionic matrix model. From (3.46) we see that there are two contributions to the N =
free energy. These two quantities represent very different phases of the system. The first
one is given by (3.47) which is determined by the saddle-point Eq. (3.48). This equation
is immediately recognized as the critical equation describing the continuum limit of an
abstract branched polymer [5,7]. In fact, the explicit derivative expression for the measure
normalization (3.43) coincides exactly with the partition function of the N -dimensional
fermionic vector model with the same polynomial potential V [14]. The latter quantity is
related to the generating function for branched polymer networks by a simple mapping of
its Feynman diagrams and an analytical continuation N 7 N/2 of the fermionic vector
dimension. The analytic reason for the appearance of branched polymer behaviour in the
fermionic matrix model is that the normalization of its measure is not of sub-leading
order in N . From a graphical point of view, its appearance is clear from the analysis
of Section 2.2. The entropy factors associated with the number of Feynman graphs is
drastically reduced in the fermionic case because of the cancellations which occur between
twisted diagrams. There is a class of diagrams with no twists in the fermionic matrix
model, and these have a tree-like growth in the large N limit. The reduced entropy of
the remaining twisted graphs is then comparable to that of the graphs which produce a
polymer-like behaviour. This feature is unique to fermionic matrices, and it is the property
which enables the construction of a model of branched polymers using supersymmetric
matrix models [13] via the coupling of the fermionic matrix model to an ordinary, complex
matrix model which has the effect of cancelling the set of twisted diagrams. In fact, it
is precisely the twisting mechanism described in Section 2.2 that enables one to isolate
graphs with tree-like growth in the fermionic case. The untwisted diagrams may then be
mapped onto branched polymers similarly to the case of the cactus diagrams which appear
in supersymmetric matrix models [13].
The twisted graphs are associated with the second contribution in (3.46) and, as we
will demonstrate, they lead to the usual surface effects in the theory. Thus the orthogonal
polynomial formalism that we have developed naturally splits the generating function into
a piece corresponding to the tree-like graphs and a piece corresponding to the twisted

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

693

diagrams. The continuum limit of the latter ensemble of graphs describes a surface theory
and describes an effect which is of order N 2 , while the former ensemble which has
a polymer growth produces an effect of order N . One typical feature of the polymer
generating function is that it becomes non-analytic in the large N limit at the Gaussian
point gk = 0, k > 2. However, this effect is of sub-leading order in N , so that a large N
analysis with arbitrary coupling constants will miss the branched polymer phase transition.
There is a barrier at the Gaussian point which separates the surface-like continuum limit
from the branched polymer behaviour. From a geometrical point of view, the fermionic
matrix model probes an intermediate lattice phase where the number of Feynman diagrams
contributing to a given process is greater than that expected of a vector model but less than
that of a matrix model. Thus the matrix integral generates a random surface theory which
contains an internal branched polymer phase.
This decrease in the number of graphs as compared to the usual matrix models will
be responsible for the Borel summability of the all genus expansion of the free energy
of the fermionic matrix model that we will prove in this section. As discussed above, the
essential critical behaviour of the fermionic matrix model is encoded in the large-N limit
of the recursion coefficients Rn . If we introduce the quantities n defined by
n+1
,
(4.1)
hn =
n
with 0 = dN as in (3.2), then the recursion coefficients are given by Rn = n+1 n1 /n2
and the free energy by FN = FN , where Fn = log(n /nGauss). Replacing sequences by
functions of x [0, 1] as prescribed by (3.45), we then have to leading order in N ,
(1 + x)R(x) 2 F (x)
=
,
F0 = F (1).
(4.2)
x
x 2
The relations (4.2) show that the function R(x) is related to the specific heat u0 of
the matrix model, and also that any singularity in the free energy will occur at the
boundary x = 1 of the domain of R(x). These facts will enable us to construct the
topological expansion of the fermionic matrix model from the recursion coefficients Rn
of the orthogonal polynomials. We remark that it is possible to demonstrate from the
constrained sum over SU(N) group characters of Section 2.3 that the partition function is
real and has a definite value for any N . Combining this with our knowledge from large N
loop equation calculations, wherein the system goes to a definite limit, one can argue that
there is a definitive 1/N expansion of the free energy. The phase transition at large N can
thereby be understood as a sort of percolation transition in the Young tableaux, whereby
the system evolves into a phase in which the number of filled SU(N) Young tableaux grows
asymptotically.
log

4.2. Main recursion relation

We will now begin quantifying the above discussion. For definiteness, in the remainder
of this section we will deal primarily with the simplest non-Gaussian model which is given
by the quadratic potential

694

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

g 2
z ,
(4.3)
2
although, as we will discuss, the results we obtain are universal. The potential (4.3) is
difficult to treat within the loop equation approach because the endpoints of the support of
the corresponding spectral density are located asymmetrically in the complex plane [10].
Indeed, one power of the technique that we develop is that it is insensitive to the parity
symmetry (or lack thereof) of the potential. This will account for the universality properties
of the solution that we shall find. In this subsection we will derive the main recursion
relation corresponding to the potential (4.3) that will be used to construct the topological
expansion.
To solve the recursion relations (3.36), (3.41) and (3.42) corresponding to (4.3), we first
use (3.20) and (3.26) to determine the coefficients
V (z) = z +

(n)
[1]
= Pn1
= Qn1 Qn Rn (Rn+1 1),
Pn,n1
[1]
= Pn2 = Qn2 Qn1 Qn Rn1 Rn (Rn+1 1).
Pn,n2
(n)

(4.4)

This leads to the closed set of recursion relations

(N + n)Rn n = gNQn1 Qn Rn (Rn+1 1),

(4.5)

nPn1 (n 1)Pn = gQn2 Qn1 Qn Rn1 Rn (Rn+1 1),

(4.6)

N(Pn+1 Pn ) = Qn (Rn+1 1).

(4.7)

From the relations (4.6) and (4.7) one can solve for the coefficients Pn as
NPn = gNQn2 Qn1 Qn Rn1 Rn (Rn+1 1) + nQn1 (Rn 1).

(4.8)

Substituting (4.8) into (4.7) and using (4.5) to reduce the degree of the Qn s in the resulting
equation yields




0 = Qn1 Rn (N + n + 1)Rn+1 n 1 Qn2 Rn1 (N + n)Rn n
+ nQn (Rn+1 1) nQn1 (Rn 1).

(4.9)

We now use (4.5) to define the quantity

n {R} Qn Qn1 =

1 (N + n)Rn n
,
g NRn (Rn+1 1)

(4.10)

and multiply the relation (4.9) through by Qn . Using (4.6) along with (4.8), we can then
solve for Qn in terms of the Rn s alone as
Qn {R}2 = n {R}2

n(Rn 1) Rn [(N + n + 1)Rn+1 n 1]

.
nn {R}(Rn+1 1) n1 {R}Rn1 [(N + n)Rn n]

(4.11)

We may therefore write down a recursion relation involving only the coefficients Rn in
the form
Qn {R}2 Qn1 {R}2 = n {R}2 ,

(4.12)

where the function Qn {R}2 is given by (4.11) and n {R} by (4.10). Notice that the
coupling constant g completely cancels out in this final equation which is an involved

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

695

non-linear recursion relation for the Rn s solely in terms of n and N . The dependence
of Rn on g comes about purely as a boundary condition and can only be recovered after
solving for the dynamics of Rn . However, we will see that the entire topological expansion
can be constructed without ever knowing this dependence explicitly. All information about
the continuum limit will come from the details of the main recursion relation (4.12).
4.3. Planar limit
The spherical continuum limit of the matrix model is found by taking the limit N .
Let us first describe the polymer contribution f0 in (3.46). For the potential (4.3), the
solution of the quadratic saddle-point Eq. (3.48) which is regular at g = 0 is given by


1 p
1 + 4g 1 ,
(4.13)
=
2g
so that the corresponding fermionic vector free energy (3.47) is




1 p
1
1 p
1 + 4g 1 log
1 + 4g 1 .
f0 (g) = +
2 4g
2g

(4.14)

There is a phase transition at g = g (0) = 1/4 below which the free energy becomes
complex-valued. Near this critical point, (4.14) behaves as f0 (g) (g g (0) )3/2 , which
(0)
= +1/2. The phase
identifies the corresponding string susceptibility exponent as str
transition is therefore associated with the evolution of the system into a pure branched
polymer phase of two-dimensional quantum gravity. In fact, the function (4.14) is related
to the usual bosonic vector model free energy Fvec , which is the generating function for
pure, connected branched polymer graphs, by f0 = 1 2Fvec [14]. The N dimensional
Euclidean radial coordinate r of the bosonic vector model is related to the saddlepoint (4.13) by r 2 = 12 . One can also compute the one-loop fluctuations around the
saddle-point (4.13) and show that the resulting free energy is logarithmically divergent
at g = 0 [14]. Therefore, in the regime g > 0 there is no polymer behaviour and one may
imagine a barrier in the large N limit of the theory at the Gaussian point g = 0. It is
further possible to carry out a double scaling expansion of the fermionic vector model
partition function (3.43) in the limit N , g g (0) with the parameter N(g g (0) )3/2
held fixed. The resulting expression is an alternating, Borel summable series which can be
expressed as a Bessel function [14].
We now consider the large N behaviour associated with the second term in the
free energy (3.46). Replacing the discrete Rn coefficients with the large-N continuous
function R(x) as prescribed by (3.45), after some algebra we obtain from the recursion
relation (4.12), at leading order in N , the first order non-linear ordinary differential
equation
R(R 1)2 (x + 2xR 3(1 + x)R 2 )
dR
= 2
.
dx
x 2x 2 R + 4x(1 + x)R 2 2(1 + x)(2 + 3x)R 3 + 3(1 + x)2 R 4

(4.15)

The dependence of R on the coupling constant g arises only from the constant of
integration of (4.15). It is known from the general theory of ordinary differential

696

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

equations [26] that the solutions R(x) to equations such as (4.15) possess algebraic nonanalytic behaviour. Since dR/dx in (4.15) is given as the quotient of a quintic polynomial
in R by a quartic one, it follows that the function R(x) is finite for all finite values of x [26].
The only singularities which can arise from this differential equation occur when the
denominator on the right-hand side vanishes and the numerator is non-zero. In light of
the discussion of the previous subsection, we will demand that these singular points occur
at the boundary x = 1 of the domain of the function R(x). At x = 1, the denominator of
the right-hand side of (4.15) has four distinct zeros Rc determined as the solutions of the
quartic equation
1 2Rc + 8Rc2 20Rc3 + 12Rc4 = 0.

(4.16)

Denoting the corresponding critical value of the coupling constant g by gc , we make an

ansatz for the form of the function R(x; g) near the critical point,
R(x; g) = Rc + a(gc gx)str + ,

(4.17)

where a is some constant and the ellipsis denotes terms which are less singular at the
critical point. This ansatz ensures that for g gc , the singularities occur at x = 1,
or alternatively that at x = 1 the free energy F (x) becomes non-analytic at g = gc .
Substituting (4.17) into the differential Eq. (4.15), we find for each branch Rc of the
Eq. (4.16) the leading behaviour
1
dR

.
(4.18)
dx
Rc R(x)
Comparing with (4.17) fixes the exponent str = 1/2, and so the critical point g = gc of
the fermionic matrix model describes a continuum limit that lies in the same universality
class as pure two-dimensional quantum gravity in the planar limit. Note that this criticality
argument requires no knowledge of the precise value of the critical coupling gc . The same
will be true of the double scaling limit which will be analysed in the next subsection.
It is possible to show using loop equations [9] that gc > 0. Thus the continuum surface
behaviour occurs on the opposite side of the Gaussian point relative to the branched
polymer phase transition, and the two phases cannot be connected together in a smooth
way.
We conclude our analysis of the planar limit of the adjoint fermion one-matrix model
by briefly describing how it extends to more general potentials. For example, consider a
potential of the generic form V (z) = z + Kg zK , K > 2. The recursion equations (3.36)
[K]
[K]
and Pn,n2
determined
and (3.41) in this case require knowledge of the coefficients Pn,n1
[K]
is of
from (3.20) and (3.26). It is straightforward to see from these relations that Pn,n1

[K]
is of degree K + 1. In the large N
degree K in both the Qn s and the Rn s, while Pn,n2
limit, the recursion equations will therefore assume the form

(1 + x)R x = gQK W1 [R],

P
= gQK+1 W2 [R],
P x
x
P
= Q(R 1),
x

(4.19)
(4.20)
(4.21)

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

697

where W1 [R] and W2 [R] are polynomials of degree K and K + 1 in the function R,
respectively, which each contain the factor R 1. Using (4.19) we can solve for QK and
write (4.20) as an equation determining the function Q. Comparing with the solution for Q
obtained using (4.21), the relation (4.20) then becomes a first order, linear, homogeneous
differential equation for the function P . Substituting the resulting solution for P back
into (4.19) yields an integral equation for the function R which, upon differentiation, can
be transformed into a first order non-linear ordinary differential equation of the form
dR J2 [R; x]
=
,
dx
J1 [R; x]

(4.22)

where J1 [R; x] and J2 [R; x] are polynomial functions of their arguments of respective
degrees 2K and 2K + 1 in R which are independent of the coupling constant g. Arguing
as we did above, the differential equation (4.22) will admit critical behaviour for the
function R of the square root type, i.e., that of pure gravity. Provided that the zeros of
the function J1 [R; x] are non-degenerate at x = 1, this will be the only type of critical
behaviour that the model admits. It is of course natural that pure gravity exists for a generic
matrix potential. Multicritical points require more complicated potentials with two or more
coupling constants and an appropriate fine-tuning of them. In this case the differential
equations derived analogously to those above become highly non-linear and quite involved.
But the general picture will be the same, namely the function R will be determined by some
differential equation whose polynomial coefficients (in R and x) can be tuned to obtain
higher order critical points. In this way we can recover, in the planar limit, the standard
universality classes of conformal matter coupled to two-dimensional quantum gravity [4].
4.4. Double scaling limit
In this subsection we will study the all genus expansion of the specific heat
u(g, N) =

N 2h uh (g),

uh (g) = Fh00 (g),

(4.23)

h=0

where Fh (g) is the genus h contribution to the free energy FN (g) and the genus h
susceptibility is a homogeneous function of degree 2h, uh (a g) = a 2h uh (g). From
the analysis of the previous subsection, we know that the leading singular behaviour of

(4.23) at genus zero is u0 (g) g gc . In order to keep the partition function Z0 (g)
2
eN F0 (g) finite in the large N limit, we should therefore perform the double scaling limit
g gc , N by keeping fixed the parameter
= N 4/5 (gc g).

(4.24)

We now define a scaling function u() which captures the contributions to the string
susceptibility from all genera in the double scaling limit g gc , N with the
variable (4.24) fixed. For this, we introduce the large-N expansion parameter  = 1/N
and write the total susceptibility (4.23) in the vicinity of the critical point as

4/5

6/5

7/5 2

a0 +  2 u() +  2
a1 + O  2
, ,
(4.25)
u(g, N) =  2

698

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

where an are constants. By defining the new effective variable

=  4/5 (gc gx),

(4.26)

in the large N limit, this scaling function may be determined from the = limit of a
scaling ansatz for the function R(x) in terms of an unknown function u( ),
R(x) = Rc +  2/5 u( ),

(4.27)

where Rc is a solution of (4.16). By substituting (4.27) into (3.46) and changing variables
from x to in the integral, we find that, up to irrelevant constants (which may be absorbed
by suitable rescaling using the homogeneity of the specific heat) and terms which vanish
as  0, the free energy in the double scaling limit is determined as
Z
F () = f0 +

d ( )u( ).

(4.28)

 4/5 gc

It follows that
u() = F 00 (),

(4.29)

and so the problem of obtaining the topological expansion of the fermionic matrix model is
thereby reduced to the task of finding the solution of the main recursion relation (4.12) for
the function R(x) in this special limit. Note that the polymer free energy (4.14) contributes
only an irrelevant constant in the double scaling limit about the positive-valued critical
point gc .
We now rewrite (4.12) using (3.45) and take the limit of large-N while holding the
quantity fixed through the relation g = gc N 4/5 . We then substitute in the scaling
ansatz (4.27) and rewrite derivatives according to the change of variables (4.26), i.e., x
 4/5 . After some algebra, we find, as the coefficient of the leading order  4/5 term, a
non-linear differential equation for the specific heat u( ),


0 = 6gc 3 22Rc + 80Rc2 192Rc3 + 376Rc4 632Rc5 + 752Rc6 512Rc7 + 144Rc8


u( )u0 ( ) + Rc 1 5Rc + 10Rc2 10Rc3 + 4Rc4
h


6Rc (Rc 1)2 6Rc2 2Rc 1
i


(4.30)
gc3 1 2Rc + 20Rc2 56Rc3 + 36Rc4 u000 ( ) .
We now integrate (4.30) up once and use an inessential shift of the independent variable
to eliminate the constant term. After applying (4.16) and a rescaling using the homogeneity
of the function u( ) (which preserves the relationship (4.29)), we arrive at the parameterfree equation
u00 ( ) + u( )2 = .

(4.31)

The non-linear differential equation (4.31) governs the behaviour of the partition function
of the fermionic random surface model to all orders in the genus expansion, with the
parameter 5/4 identified as the renormalized string coupling constant. We note again

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

699

the remarkable feature that one never needs to know the precise location of the critical
point to arrive at this equation. In the present case the critical coupling constant gc can
be consistently rescaled out of the pertinent equations. This parametric independence is
indicative of the universality of the double scaling equation (4.31) for the given class of
generic polynomial potentials.
The differential equation (4.31) is known as the Painlev I equation and it is solved
by the first Painlev transcendent. This equation differs from the versions which usually
appear in matrix models [4] by a change in sign of the specific heat u u. The boundary
conditions which u( ) satisfies follow from the analysis of the planar limit of the previous
subsection. The genus zero contribution arises in the limit of large, positive where the
leading behaviour is u( )2 = . As a consequence, we are led to postulate an asymptotic
expansion for u( ) of the form
!

X
p
5k/2
uk
.
(4.32)
u( ) = 1 +
k=1

By substituting (4.32) into (4.31), the first few coefficients are easily calculated to be
u1 =

1
,
8
49
,
128
1225
,
256
4412401
,

32768
220680075
,
32768
2207064977649
,

4194304

u2 =
u3 =
u4 =
u5 =
u6 =

(4.33)

..
.
and in general they are determined by the recursive equation
(5k 6)(5k 4)
1X
uk1
uk ukn ,
8
2
k1

uk =

k > 2.

(4.34)

n=1

With the normalization of the spherical specific heat taken as u0 = 1, we see from (4.33)
that the coefficients of the genus expansion have precisely the same absolute numerical
values as those obtained for pure two-dimensional quantum gravity, i.e., from the usual
Painlev expansion [4,5]. In particular, they have the same high rate asymptotic growth in
magnitude. However, the most notable feature of the coefficients (4.33) is the fact that they
oscillate in sign (in the Hermitian cases, all uk are negative). This raises the possibility that
the asymptotic series solution (4.32) may in fact be Borel summable. If so, then the free
energy would give an unambiguous definition of the genus expansion of pure fermionic
gravity in powers of N 2 5/2 . A numerical solution of Eq. (4.31) is possible and

700

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

Fig. 1. Plot of the solution u( ) versus of the Painlev equation with the asymptotic boundary
condition u( )2 = for +. The first double pole singularity (corresponding to a zero of the
partition function) occurs on the negative axis at ' 3.1477.

the result is shown in Fig. 1. This confirms numerically that the free energy exists as a
well-defined function and is real-valued on the positive real -axis. In the next subsection
we shall prove that the coefficients uk of the asymptotic series (4.32) alternate in sign to
arbitrarily large orders of the genus expansion, and, moreover, that there is a well-defined
Borel resummation of u( ) for > 0.
4.5. Borel summability
In this subsection we will argue analytically, along the lines of [4], that the asymptotic
series (4.32) defines a unique function u( ). For this, we consider the Borel transform
B(s) =

X
uk k
s ,
(k)!

(4.35)

k=1

where the constant will be self-consistently determined by the condition that this series
has a finite radius of convergence. A solution of the original Painlev equation (4.31) is
then given by
!
Z
p


(4.36)
u( ) = 1 dt et B t 5/2 .
0

The crucial issue now is whether or not the integral in (4.36) actually exists. We will show
that there is a contour running through the Re t > 0 region of the complex t-plane from
t = 0 to t = along which the integral transform (4.36) converges. We will thereby argue
that the Borel resummation (4.36) of the specific heat is well-defined and real-valued.
The function (4.35) has singularities in the complex s-plane with branch cuts running
between them and infinity. If some of these singularities lie on the positive s-axis then the
integral definition (4.36) is ambiguous. We will now extract the large order behaviour of
the coefficients uk and argue that the singularities of the Borel transform B(s) all lie on the
negative s-axis. From (4.35) and (4.36) it follows that the coefficients of the asymptotic
expansion (4.32) are given by

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

1
uk =
2i

Z
dt e

ds
B t s .
s k+1

701

(4.37)

s=0

The contour of integration in (4.37) can be extended out to infinity around the cuts of the
Borel transform in the complex s-plane. In the limit k , the contour integral in (4.37)
is dominated by contributions along the branch cut of B(s) in the complex s-plane which
begins at the point s = s0 that is closest to the origin. We then have, for large k,
Z
uk

dt e

ds
s k+1

s0 /t

Disc B t s

ds
s k+1

dt et Disc B t s , (4.38)

(s0 /s)1/

where we have defined the discontinuity of the Borel transform

Disc B(s) = B+ (s) B (s),

B (s) = B(s i0),

(4.39)

for s a point on its cut.

Going back to (4.36), we see from (4.38) that we should study the functions u ( )
corresponding to the Borel transforms on either side of the dominant cut,
Z
u ( ) =

dt et B t 5/2 .

(4.40)

(s0 5/2 )1/

Since each of the functions (4.40) solves the Painlev equation (4.31), it is convenient to
define their sum and difference


1
(4.41)
ud ( ) = u+ ( ) u ( ) = Disc B( ), us ( ) = u+ ( ) + u ( ) ,
2
which on using (4.31) are seen to obey the coupled system of differential equations
(4.42)
u00d ( ) + 2ud ( )us ( ) = 0,
1
(4.43)
u00d ( ) + us ( )2 + ud ( )2 = .
4
These equations can be solved in the WKB approximation. In the planar limit , we
self-consistently assume that ud ( ) is of sub-leading order in (4.43) and thereby find, to

leading order, the solution us ( ) . By substituting this into (4.42), to leading order
we have the first order linear differential equation for the function ud ( ),
p
(4.44)
u00d ( ) + 2 ud ( ) = 0,

which can be solved in terms of Bessel functions as ud ( ) = Z 2 4 5 2 5/4 . The

5
solution for is therefore given by



4 2 5/4 5
ud ( )
5/8

cos
.
(4.45)
5
4
4

We are finally ready to estimate the large order behaviour of the coefficients uk . For this,
it is convenient to change variables in order to bring the asymptotic expansion for

702

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

to 0 through the definition = 5/4 . Then, from (4.38)(4.41) and (4.45), we find
as k ,


Z
Z


d 2/5
4 2 5
d
4/5

.
(4.46)
ud
cos

uk
2k+1
2k+1
5
4
4
For each choice of sign the integral (4.46) gives, up to constant factors, the result




32 k
2k 12
for k .
uk
25

(4.47)

The asymptotic estimate (4.47) contains a large amount of information. Going back to the
Borel transform (4.35), the large order behaviour of uk fixes = 2 and gives
X 32 k
sk .
(4.48)
B(s)

25
k

We see therefore that the Borel transform B(s) is well-defined in an open region of the
complex s-plane and, moreover, that its first singularity appears on the negative real s-axis.
Assuming that all of its singularities are so restricted, we can expect that the asymptotic
series (4.32) can be Borel resummed through the integration in (4.36) to define a unique
real function u( ) on the positive real -axis. The results of a numerical analysis (Fig. 1)
support these arguments.
From this analysis it follows that the double-scaled partition function of the adjoint
fermion one-matrix model leads to a completely well-defined genus expansion of the
corresponding random surface theory. The coefficients (4.47) have precisely the same large
order behaviour (2k 1/2) as those of pure two-dimensional quantum gravity. In fact, the
topological expansion of the fermionic matrix model is identical term by term to that of the
usual Hermitian matrix models. However, the alternating nature of the asymptotic series
expansion allows one to define the string susceptibility in an unambiguous way through
its Borel resummation (4.36), at least for positive values of the string coupling . In this
way we may think of the fermionic matrix model as lending a well-defined version of the
generating functions provided by Hermitian matrix models. It gives an analytic solution
to the problem of counting fermionic random triangulations on arbitrary genus Riemann
surfaces, whose asymptotic expansion is Borel summable but otherwise coincides with the
usual Painlev expansion of two-dimensional quantum gravity.
The usual movable, double pole singularities of the Painlev I equation appear as well
in the present case, but this time they appear for negative values of the cosmological
constant . The existence of such poles is inconsistent with the loop equations of twodimensional quantum gravity [27]. Pole-free solutions of Painlev equations are provided
by the triply-truncated Boutroux solutions, but these are complex-valued and do not lead
to physically acceptable free energy functions. However, in the present situation there is
a unique, well-defined real-valued specific heat on the positive -axis which should have
a meaningful analytical continuation to < 0. It provides an unambiguous definition of
a whole branch of the random surface theory, and, given the relation between the double
scaling variable and the continuum cosmological constant, the continuum limit of the

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

703

discretized model is indeed well-defined. This suggests that by restricting our fermionic
gravity model to > 0, we could define a model of two-dimensional quantum gravity
with the canonical characteristics order by order in the genus expansion, but whose
full asymptotic series makes perfect sense and provides a well-defined non-perturbative
definition of the theory.
In the next and final section we will argue that this is indeed the case. In the usual
Hermitian one-matrix models, the problem with the double scaling limit, in which the
original matrix integral can only be defined by an analytical continuation, can be traced
back to an instability in the corresponding double-scaled eigenvalue model. The noncompactness of the double-scaled eigenvalue space R and the form of the effective
potential for the eigenvalues demonstrates that the definition of the critical points of the
model is unstable to the tunneling of eigenvalues into a different configuration [29]. This
leads to instanton solutions in the single-well eigenvalue model, and thereby explains the
complexity of the free energy. However, in the present situation the Grassmann matrix
integral is completely well-defined at finite N , in contrast to the Hermitian cases, and there
is no reason a priori to expect that the model becomes unstable in any way at large N .
Indeed, with the mapping of the fermionic matrix model onto a unitary matrix model,
we see that the matrix integral is determined by an eigenvalue space whose topology is
that of a circle. Being a compact space, there is no asymptotic behaviour in the effective
eigenvalue action, and thus the compactness could have the effect of eliminating the
eigenvalue tunneling problem. We see then that the unitary representation of the fermionic
matrix model provides, at least naively, an eigenvalue description of the random surface
model which reflects the reasons why the matrix integral is superior to its Hermitian
counterparts. We shall tackle this problem using an operator theoretic approach which
gives the fermionic analog of the (generalized) KdV hierarchical structure of the Hermitian
one-matrix models [4,30]. The integrable flows that we shall find are similar to those of
the bosonic models, but with some very important changes. The most important difference
will be the absence of a translational symmetry in eigenvalue space. This feature can be
interpreted as allowing one to restrict the double-scaled specific heat to positive values of
the cosmological constant . It may then follow that the partition function of the fermionic
matrix model serves as an unambiguous definition of that for two-dimensional quantum
gravity.
We stress that the resulting Borel summability of the fermionic case is very different
from that of Borel summable bosonic models, such as the conformal field theory of the
YangLee edge singularity in which the lattice system couples to an imaginary magnetic
field and is therefore non-unitary [4]. The coefficients of the genus expansion of the
adjoint fermion one-matrix model are all real-valued. 1 The stability of the fermionic model
is also quite distinct from the stabilizations provided by supersymmetric and stochastic
quantization methods [32], because these latter models violate the KdV flow structure of
two-dimensional quantum gravity at a non-perturbative level [33]. In the present case, we
1 In [31] a similar, but non-unitary, alternating genus expansion was used to formulate a model of twodimensional quantum gravity in terms of integrals over the moduli spaces of punctured spheres.

704

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

will see that the appropriate integrable hierarchical structure is a deformation of that for
gravity, and so the KdV flow structure continues to hold in a certain sense. The precise
meaning of this deformation comes from the worldsheet interpretation of the fermionic
one-matrix model. This is the model of fermionic gravity that we alluded to in Section 2.1,
wherein Penner matter states are located at the vertices of the corresponding surface
discretizations.

5. Operator formalism
The discussion at the end of the previous section motivates the development of an
operator approach to analyse more carefully the properties of the adjoint fermion onematrix model within the orthogonal polynomial formalism. The main purpose is twofold. First of all, it will allow us to relate the partition function of the fermionic matrix
model to the -function of an integrable hierarchy [15,30,34]. The usefulness of this
association is that it partitions the solutions of the fermionic matrix model into universality
classes which are related to well-known integrable hierarchies, and thereby formulates the
partition function in an invariant way as the solution to a set of differential equations,
in particular in the double scaling limit. Secondly, given this relationship, we will obtain a
clear geometric picture of the origin of the alternating, Borel summable genus expansion of
the model. This will allow us to clarify the description of the string susceptibility as a welldefined, non-perturbative generating function for the fermionic gravity model, as alluded
to at the end of the previous section, and at the same time it will present the appropriate
generalization of the KdV flow structure that characterizes the gravity model, lending a
complementary characterization to the worldsheet description. In the following we will
begin by introducing the formalism and deriving the main equations that will be required.
We will then identify the integrable hierarchy of which the fermionic partition function is
a -function, and derive the Virasoro constraints which must be satisfied by the flows.
5.1. String equations
S defined on the space C[z]
We begin by introducing multiplication operators Q and Q
C[z] by




S m (z) 1 m (z),
(5.1)
Q
Qm (z) zm (z),
z
and differentiation operators P and S
P by




0
S
(5.2)
Pm (z) m (z),
Pm (z) z2 0m (z).
S are related to each other in a very simple way. From (5.1) it follows
The operators Q and Q
that




Qm (z) = n (z)m (z) ,

(5.3)
Qn (z)S
and so Q is an invertible operator with inverse
S ,
Q1 = Q

(5.4)

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

where the adjoint O of any operator O on C[z] C[z] is defined by






F (z)(OG)(z) = O F (z)G(z) .

705

(5.5)

0 (z), we
By writing [z m (z)]0 in two different ways as [(Qm )(z)]0 and m (z) + zm
can infer the canonical commutation relation


P, Q = 1.
(5.6)

S S
P] = 1, which with the unitarity
Similarly, by considering z2 [ 1z m (z)]0 we arrive at [Q,
condition (5.4) implies the canonical commutator
 1 
S
= 1.
(5.7)
P ,Q
Geometrically, the operator P is the canonical conjugate of the operator generating
clockwise unit shifts on the circle, while S
P serves as the canonical conjugate of that which
generates counterclockwise shifts. The operators introduced above are not all independent,
but are related to each other through a SchwingerDyson equation. This follows from the
inner product




(Pn )(z)m (z) = n0 (z)m (z)





= (N + 1) n (z) 1z m (z) N V 0 (z)n (z)m (z)




(5.8)
n (z)0m (z) ,
which implies the operator identity
P +S
P Q2 = (N + 1)Q1 NV 0 (Q).

(5.9)

The Eqs. (5.6), (5.7) and (5.9) define the string equations of the adjoint fermion one-matrix
model.
Let us now describe some basic properties of the operators introduced above. For this,
we first rewrite their actions on the same basis n of the space of polynomials in a single
variable on the circle. We have
X



Q1 nm m (z),
Q1 n (z) =

Pn (z) =

m>0

[P]nm m (z),

m>0

X 


S
S
P nm m (z).
P n (z) =

(5.10)

m>0

We denote by O+ the upper triangular part, including the main diagonal, of the discrete
operator O when represented in the basis of polynomials in (5.10). Then O = O O+ is
the lower triangular part of O.
From the definition (5.1) it follows that the operator Q defines a Jacobi matrix, because
 
(5.11)
Q nm = 0 for m n > 1.
In particular, its pure upper triangular part is given by
X 
X
En,n+1 ,
Q nn En,n =
Q+
n>0

n>0

(5.12)

706

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

where En,m are the step operators with matrix elements [En,m ]k` = kn `m . The remaining
(n)
matrix elements of Q are given by [Q]nm = Pm in (3.26), along with the recurrence
relations derived in Section 3. Similarly, the operator Q1 defines a Jacobi matrix in the
basis (5.10) because
 1 
(5.13)
Q nm = 0 for n m > 1.
Since




S n (z)n1 (z) = n (z) 1 n1 (z) = hn ,
hn1 Q1 n,n1 = Q
z

it follows that its lower triangular part is given by

X
Rn En,n1 .
Q1
=

(5.14)

(5.15)

n>1

The remaining matrix elements of Q1 are given by [Q1 ]nm = Ln in (3.31).

Furthermore, from the definition (5.2) we see that P = P is purely lower triangular with
 
(5.16)
P n,n1 = n,
(m)

and since


 
P n (z)n+1 (z) = n (z)z2 0n+1 (z) = (n + 1)hn , (5.17)
P n,n+1 = S
hn+1 S

P+ is purely upper triangular, [S

P]nn = 0, with
it follows that S
P = S
 
n+1
S
.
(5.18)
P n,n+1 =
Rn+1
Note that, in this basis, the (n, n 1) matrix element of the SchwingerDyson equation (5.9) coincides with (3.36). This follows from the structure of the matrix elements



 2 
(5.19)
P Q n,n1 = n (z)0n1 (z) = (n 1)hn ,
hn1 S
which implies that [S
P Q2 ]n,n1 = (n 1)Rn .
5.2. Flow equations
We will now derive a system of flow equations for the operators above and the partition
function (3.7), which will enable us to identify the particular integrable hierarchy at work
here. We shall describe the evolutions with respect to the discrete set of time variables
tk Ngk /k of the generic potential (1.2). Consider the relation





(5.20)
n (z)nm (z) = 0 = zk n (z)nm (z) + tk n (z)nm (z) ,
tk
where 1 6 m 6 n and we have used the fact that nm (z)/tk is a polynomial of degree
at most n m 1. This relation leads immediately to the evolution equation
X 


n (z)
=
Qk nm m (z) = Qk n (z).
tk
n1

(5.21)

m=0

By taking time derivatives of Eqs. (5.10), we then arrive at the discrete operator flow
equations

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

707


Q 
= Q, Qk ,
(5.22)
tk

P 
= P, Qk ,
(5.23)
tk
Q1  1 k 
= Q , Q ,
(5.24)
tk
S
P  S k 
= P , Q .
(5.25)
tk
The flow equations (5.22) constitute a discrete KP hierarchy [34]. It is straightforward to
show that these flows are mutually commutative, so that this hierarchy is in fact integrable.
This follows from taking time derivatives of (5.10) and using (5.21) to get
n1
n+k
X
X
 k  r 
 r   k
[Qk ]mn
=
Q m` Q `n
Q m` Q `n ,
tr
`=mk

(5.26)

`=m+1

which leads to the ZakharovShabat equations

Qn  m n 
Qm

+ Q , Q = 0.
(5.27)
tn
tm
The matrix model actually defines a reduced, generalized KP hierarchy, because the flow
equations are to be supplemented with the constraints imposed by the string equations.
The knowledge of the solutions Q to (5.22) completely determines the partition function
of the fermionic matrix model. To see this, we consider the (n, n 1) matrix element of
the flow equation (5.24), which using (5.13) and (5.15) leads to the differential equation
 
log Rn  k 
= Q nn Qk n1,n1 .
(5.28)
tk
We can use (3.7) and (5.28) to obtain the flow equation for the partition function


log ZN
= Tr(N) Qk ,
(5.29)
tk
where we have defined the N dimensional trace
X 

N1
O nn .
Tr(N) O =

(5.30)

n=0

Using the KP equation (5.22) and the Jacobi property (5.11) of the Q operator, we have
 
[Q]nn  k 
= Q n+1,n Qk n,n1 ,
(5.31)
tk
which, on using (5.28) with k = 1, leads to
2 log ZN  k 
= Q N,N1 .
(5.32)
t1 tk
This procedure can be iterated to determine any number of time derivatives of log ZN in
terms of the Q operators. Knowing Q therefore determines the free energy of the matrix
model up to an overall integration constant.

708

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

5.3. Fermionic -function

It is clear from (5.32) that the partition function ZN is a -function for the discrete KP
hierarchy (5.22) [34]. Let us introduce the BakerAkhiezer functions
N

  e 2 V (z) n (z)
n tE; z = N+1 ,
hn
z 2

(5.33)

and their conjugates

N

  e 2 V (z) n (z)
Sn tE; z =
,

N+1
hn
z 2

(5.34)

which are bi-orthonormal in the standard line measure of the complex plane,
I
   
Sm tE; z = nm .
dz n tE; z

(5.35)

z=0

The operator Q is a discrete Lax type operator which has eigenfunctions n [tE; z] with
eigenvalue z,
 
 
(5.36)
Qn tE; z = zn tE; z .
The function ZN [tE ] is then the generating function for the Lax operator Q, in the sense
that Q can be reconstructed from a knowledge of ZN . In fact, if we introduce the discrete
-function n [tE] for this hierarchy by the formula (4.1) [15], then from (3.7) we see that
the partition function of the fermionic matrix model is given by
 
 
(5.37)
ZN tE = N tE .
Formally, the -function is a section of the determinant line bundle over a Sato
Grassmannian associated with Riemann surfaces of the spectral parameters z [34]. It may
be characterized as the unique function depending on the times tk and satisfying the given
hierarchy of constrained differential equations. It is the generating function for solutions of
the generalized KdV equations. Thus once the -function is known, everything about the
matrix model is likewise known. The usefulness of the relationship (5.37) is that there are
a large number of identities satisfied by N [tE] that can be used to characterize the partition
function of the fermionic matrix model.
The flow Eqs. (5.22) or (5.26) constitute the Lax representation of a discrete integrable
hierarchy of differential equations of Toda type [35]. It is tempting therefore to characterize
the fermionic partition function as the -function of a certain reduction of the integrable
Toda chain. However, as we will now demonstrate, this is not the case. By equating the
constant terms on both sides of the flow Eq. (5.21) for k = 1 and using (3.26) and (3.27),
we have
n1
n
Y
pn,0 1 Rn+1 X
=
pm,0 lm,0
Rk .
t1
ln+1,0
m=0

k=m+1

(5.38)

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

709

By iterating the relation (3.30) using (3.27) it follows that

pn,0 ln,0 = (1)n+1 (Rn 1),

(5.39)

where we have defined R0 0. Substituting (5.39) into (5.38) we obtain

pn,0
= pn+1,0 Rn = pn+1,0 1 + (1)n+1 pn,0 ln,0 .
(5.40)
t1
Similarly, one can derive the evolution equation


ln,0
= ln1,0 1 + (1)n+1 pn,0 ln,0 .
(5.41)
t1
The flow equations (5.40) and (5.41) are the first members of an integrable Hamiltonian
system known as the Toeplitz chain hierarchy [25]. The general evolution equations may
be similarly derived using the flow equations of the previous subsection and are given by
the Hamilton equations of motion

Hk
pn,0
= 1 + (1)n+1 pn,0 ln,0
,
tk
ln,0

Hk
ln,0
= 1 + (1)n+1 pn,0 ln,0
,
(5.42)
tk
pn,0
for the system of Hamiltonians


1
1 log ZN
,
Hk = Tr(N) Qk =
k
k
tk
and the symplectic structure
X
dpn,0 dln,0
.
=
1 + (1)n+1 pn,0 ln,0

k > 1,

(5.43)

(5.44)

n>1

The Toeplitz chain is a particular reduction of the two-dimensional Toda lattice hierarchy,
the latter of which characterizes generic matrix integrals [15,30]. Thus the integrable
hierarchy to which the adjoint fermion one-matrix model belongs is not the reduction of a
one-dimensional (Toda) hierarchy, but rather of a two-dimensional one. This is of course
anticipated from the lattice interpretation of the fermionic matrix model and its associated
doubling of degrees of freedom due to the Penner matter states.
However, the Toeplitz chain hierarchy can be viewed as a deformation of the standard
Toda chain hierarchy. To see this, we use (5.28) for k = 1 and (3.26) for m = n to obtain
the flow equation
log hn  
= Q nn = Qn (Rn+1 1).
(5.45)
t1
On the other hand, by using (5.31) for k = 1 and (3.26) for m = n 1 we have


[Q]nn
= Qn Qn+1 Rn+1 (Rn+2 1) Qn1 Rn (Rn+1 1) .
(5.46)
t1
The two flow equations (5.45) and (5.46) may be combined together to give


Rn log hn1
Rn+1 log hn+1
2 log hn log hn
=

.
(5.47)
t1
Rn+1 1
t1
Rn 1
t1
t12

710

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

By introducing time-dependent functions qn [tE ] through

 
hn tE = (1)n 2n et1 / eqn [tE] ,

(5.48)

where is a time-independent parameter, we can write (5.47) as




eqn+1 qn
qn
qn+1
2 qn
=
1
+

1
+

t1
t1
1 + 2 eqn+1 qn
t12



qn1
qn
eqn qn1
1+
.
(5.49)
1+
t1
t1 1 + 2 eqn qn1
The differential Eq. (5.49) is the first member of an integrable hierarchy known as the
relativistic Toda chain [36]. In the non-relativistic limit 0, it reduces to the Toda
chain hierarchy which characterizes generic Hermitian one-matrix models. Thus, the
partition function of the adjoint fermion one-matrix model is a -function of not the Toda
chain hierarchy, but rather of its deformation to the relativistic Toda chain. The latter
chain is itself a reduction of the two-dimensional Toda lattice. From this point of view,
the fermionic matrix model may be thought of as a deformation of the Hermitian onematrix model, the role of the deformation being played by the Penner interaction.
The remaining differential equations of the relativistic Toda chain hierarchy are given
by the Lax equations (5.26), by (5.28), and by exploiting the Jacobi properties (5.11)
and (5.12) of the operator Q. It is important to realize that, because the present model
contains only a single set tE of times, the ordinary Toda lattice does not itself appear in
the hierarchy satisfied by the partition function. Thus the standard Toda lattice structure
which underlies all integrable models (and in particular matrix integrals) only appears in
a very subtle way through the reductions obtained above by eliminating one set of its
time variables. This reduced structure leads to a certain degeneracy in the -function as
compared to the usual two-dimensional Toda lattice structure. Of course, the relations to
integrable hierarchies described in this subsection are only kinematical, as they merely
rely on the very basic recursion properties satisfied by the orthogonal polynomials. To
incorporate the dynamical aspects, and in particular the effects of the Penner potential,
we need to impose the constraints implied by the string equations. Indeed, there are many
-function solutions of the above hierarchies, and the string equations select the one
appropriate to describe the nonperturbative dynamics of the adjoint fermion one-matrix
model. This is the topic of the next subsection.
5.4. Virasoro constraints
We will now begin examining the constraints imposed on the fermionic -function ZN [tE]
as dictated by the string equations. Proceeding as in (5.8) for the operators PQn+1 , n > 1,
we can express the string equation (5.9) as an infinite system of equations



X
2
1
k1
n+1
S
ktk Q
Q
= 0, n > 1. (5.50)
Tr(N) P + P Q (N + 1)Q +
k>1

The strategy now is to rewrite the equations of motion (5.50) using the flow equations (5.29) and (5.32). In this way, the string equations will be represented as the an-

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

711

nihilation of the partition function by a system of differential operators in the coupling

constants of the potential V . In this subsection we will deal with the cases n > 0.
Let us start with the n = 0 string equation of (5.50). Using the canonical commutation
relations (5.6) and the matrix elements



 
(5.51)
hn QP nn = zn0 (z)n (z) = nhn ,
we may compute the first trace in (5.50) for n = 0 as
N1
X


N(N + 1)
.
n=
Tr(N) PQ = N +
2

(5.52)

n=0

The second trace may be similarly computed by using




 1 
hn S
P Q nn = n (z)z0n (z) = nhn ,

(5.53)

to get


N(N 1)
.
(5.54)
P Q1 =
Tr(N) S
2
By substituting (5.52) and (5.54), and using (5.29), we can represent the n = 0 constraint
of the system (5.50) as the flow equation
!
X
 

2
ktk
N ZN tE = 0.
(5.55)
tk
k>1

Next, let us consider the n = 1 constraint of (5.50). For the first trace, we use (3.5)
and (3.6) to compute the matrix elements





hn Q2 P nn = z2 n0 (z)n (z)



= nzn+1 (n 1)pn,n1 zn + O zn1 n (z)


(5.56)
= npn+1,n (n 1)pn,n1 hn .
Using (3.16) and the canonical commutator (5.6) we thereby arrive after a little algebra at
X  



N1



n Q nn + Tr(n) Q
Tr(N) PQ2 = 2 Tr(N) Q +
n=1

= (N + 1) Tr(N) Q .

(5.57)

Since [S
P ]nn = 0, upon substituting (5.57) into (5.50) we find that the terms linear in Q
cancel out and applying the flow equation (5.29) we see that the n = 1 constraint can be
written as
X
 

ktk
ZN tE = 0.
(5.58)
tk+1
k>1

Now we move on to the n = 2 string equation of (5.50). Again using (3.5) and (3.6) we
may compute

712

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725



hn Q3 P nn = z3 n0 (z)n (z)




= nzn+2 (n 1)pn,n1 zn+1 (n 2)pn,n2 zn + O zn1 n (z)


= n(pn+2,n + pn+2,n+1 pn+1,n ) (n 2)pn,n2 (n 1)pn+1,n pn,n1 hn .
(5.59)
Using (3.16) we can iterate the relation (3.14) for k = n 2 to get
pn,n2 =

n1
X
 
 



Q k,k1 + Q kk Tr(k) Q .

(5.60)

k=1

Substituting (5.60) and (3.16) into (5.59) and using (5.6), we arrive after some algebra at




 


2
Tr(N) PQ3 = 3 Tr(N) Q2 + (N 1) Q N,N1 + (N + 1) Tr(N) Q
+ (2N 3)

N1
X

 
 



Q n,n1 Q nn Tr(n) Q .

(5.61)

n=1

The sums in (5.61) can be simplified by using the Jacobi property (5.11) of the operator Q
to get
N1
N1
X 
X  
2

 
Q n,n1 +
Q nn ,
Tr(N) Q2 = Q N,N1 + 2
n=1

(5.62)

n=0

and in this way we arrive at an expression for the first trace in the n = 2 equation of (5.50),





1 


2
1
3
3
(5.63)
Tr(N) Q2 + Q N,N1 + Tr(N) Q .
Tr(N) PQ = N +
2
2
2
For the second trace in (5.50), we use (3.5) to compute the matrix elements



 
P Q nn = n (z)z3 0n (z)
hn S



= n (z) nz(n2) + O z(n3) + ln,1 z

 

(5.64)
= ln,1 zn (z)1 .
Using (3.25) we obtain
 
S
P Q nn = Sn Qn (1 Rn+1 ),
where we have defined
ln,1
.
Sn =
ln,0

(5.65)

(5.66)

By equating the coefficients of the z1 terms on both sides of the three-term recursion
relation (3.32) for the polynomials, we can write down an iterative equation for the
coefficients (5.66),
Sn = Rn Qn1 + Sn+1 Qn ,

(5.67)

which has solution

Sn = Qn1 Rn +

n1
X
k=0

Qk (1 Rk+1 ).

(5.68)

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

713

Substituting (5.68) into (5.65) and using (3.26) for m = n 1 and m = n, we therefore find
 

 
 
S
(5.69)
P Q nn = Q nn Tr(n) Q Q n,n1 .
Using the identity (5.62) we then arrive at an expression for the second trace in the n = 2
equation of (5.50),

1 


2 1


1
P Q = Q N,N1 + Tr(N) Q Tr(N) Q2 .
(5.70)
Tr(N) S
2
2
2
Upon substitution of (5.63) and (5.70) into (5.50) for n = 2, we see that the terms
quadratic in Q cancel out. The remaining terms can be simplified using the flow
equations (5.29) and (5.32), which leads to the n = 2 constraint equation

X
 
2

ktk
+ 2 ZN tE = 0.
(5.71)
tk+2 t1
k>1
This procedure can be easily generalized to higher orders n > 3. However, it is possible
to determine the general relation from the first three equations (5.55), (5.58) and (5.71) by
using the fact that the commutator bracket of the pertinent differential operators must also
annihilate the partition function and thereby using commutators of the above operators to
generate the higher order ones. The final result can be expressed in a more familiar form
by introducing the zeroth time t0 through the flow equation
ZN
= NZN ,
t0

(5.72)

for the partition function.

In this way we find that the string equations (5.50) for n > 0 can be written as the set of
discrete Virasoro constraints
 
(5.73)
Ln ZN tE = 0, n > 0,
where the second order linear differential operators
Ln =

X
k>1

ktk

tk+n

n
X

2N
,
tk tnk
tn

(5.74)

k=0

satisfy the Witt algebra

[Ln , Lm ] = (n m)Ln+m ,

(5.75)

for n, m Z+ . These Virasoro constraints are the usual ones for the fermionic matrix
model [8,9] and they are identical to those of the Hermitian Penner matrix model [37]
as defined in (1.3) with = 2. They represent the full system of equations of motion
of the matrix model and can be equivalently derived by demanding the invariance of
the matrix integral under arbitrary changes of the matrix variables [8,9]. The first set of
operators in (5.74) comes from the variation of the potential V in the action, while the
second one comes from the Jacobian of the change of matrix integration measure. The first
two terms in (5.74) therefore coincide with the standard Virasoro generators of generic

714

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

Hermitian one-matrix models [30]. The last operator in (5.74) comes from the variation of
the logarithmic Penner interaction.
It is a remarkable fact that the loop equations of the adjoint fermion one-matrix
model (1.1), the Hermitian Penner one-matrix model ((1.3) with = 2), and the unitary
Penner matrix model (1.4) are all equivalent. However, the equivalence between the
Hermitian model and the other two matrix models only holds order by order in the 1/N
expansions of the matrix integrals. Although the models are equivalent at N = , beyond
leading order in the large N expansion the loop equations should be solved with different
boundary conditions and the solutions are different. The fermionic and unitary equations
should be solved with boundary conditions appropriate to a perturbative Gaussian limit,
in order that the models admit interpretations in terms of random surfaces, while the
Hermitian ones should be solved with boundary conditions appropriate to recover the
Penner model of the discretized moduli space of Riemann surfaces [10,16]. The distinction
of the fermionic and unitary matrix integrals from the Hermitian one is evident from the
structure of the finite N correlators (see Appendix A). In fact, it is only in the large N
limit that a full Virasoro symmetry is realized in these models, as well as an infinite
hierarchy of generalized KP differential equations. This makes the large N limit a very
natural ingredient of the fermionic matrix model, as is also apparent from its fat-graph
interpretation of Section 2.2.
5.5. Origin of the Painlev expansion
In the previous subsection we have shown that the SchwingerDyson equations of the
fermionic matrix model are generated by the positive Borel subalgebra of a Virasoro
algebra of vanishing central charge. There is a nice algebraic way to characterize the effect
of the Penner interaction potential in the Virasoro generators (5.74). Let us write them as
H
Ln = LH
n + Tn , where Ln are the standard Virasoro generators of a Hermitian one-matrix
model with potential V , and
Tn = 2N

,
tn

(5.76)

are the generators of translations tn 7 tn 2N in coupling constant space. Together these

operators generate the symmetry algebra
[Tn , Tm ] = 0,
 H H
Ln , Lm = (n m)LH
n+m ,


H
Tn , Lm = 2NnTn+m .

(5.77)

The commutation relations (5.77) characterize a master symmetry of the generalized

KdV hierarchy [34]. We see then that the Virasoro constraints of the fermionic matrix
model constitute a deformation of those for Hermitian one-matrix models by the master
symmetry algebra generators of the KdV flow equations. Note that the particular symmetry
generated by the operators (5.76) is restricted to translations in units of a discrete lattice
spacing 2N , indicating what sort of reduction of this master symmetry is taken. These

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

715

Virasoro constraints describe the desired reduction of the fermionic -function, and they
lead immediately to the desired geometrical interpretation of the Painlev expansion of
the adjoint fermion one-matrix model. The Painlev equations may be characterized as
following from a reduction of the -function of the integrable Toeplitz (or relativistic Toda)
chain hierarchy, subject to the Virasoro symmetry obtained in the previous subsection.
They yield a stability condition Ln N = 0 on the points of the model Grassmannian, which
are associated with the BakerAkhiezer functions (5.33), that select a particular class of
transcendental solutions to the KP equations.
It is a standard fact that a set of operators of the form (5.74) can be embedded into
a Virasoro algebra of non-vanishing central charge. However, an important aspect of the
present approach to the fermionic matrix model is that there are in fact extra dynamical
constraints imposed on the partition function which can be attributed to the invertibility of
the Lax operator Q. For instance, the constraint (5.50) makes perfect sense for n = 1,
and the calculation of the trace of the operator S
P Q2 proceeds in an analogous manner to
that of (5.56) with the result




(5.78)
P Q2 = (N 1) Tr(N) Q1 .
Tr(N) S
Since [P]nn = 0, we see that on substitution of (5.78) into (5.50) for n = 1 we produce an
extra non-vanishing term 2N Tr(N) (Q1 ). This trace can be written in an operator form
as follows. From (3.31) with m = n and (5.45) we have


 1 
log hn 1
.
(5.79)
Q nn = (Rn 1)(Rn+1 1)
t1
This quantity is independent of the first time t1 , since on using the flow equation (5.24) for
S we find [Q1 ]nn /t1 = 0. This
k = 1 and the Jacobi properties of the operators Q and Q
suggests defining a negative time t1 such that Tr(N) (Q1 ) is represented as an operator
acting on the partition function as
N1
2 )
X 1
(n+1 n1 n2 )(n+2 n n+1
DZN
=
,
Dt1
n n+1
n n+1 n+1 n
n=1

DZN
= 0.
t1 Dt1

t1

t1

(5.80)

where we have used (4.1). Then the constraint (5.50) for n = 1 can be written as

X
 
D

ktk
+ Nt1 2N
ZN tE = 0.
(5.81)
tk1
Dt1
k>2

In fact, there are infinitely many negatively moded constraints such as this, associated
with higher powers Qm , m > 1. However, the operators such as that in (5.81) which
are thereby generated do not form a closed algebra among themselves or with the
operators (5.74). Indeed, this set of constraints arises from the previously mentioned
reduction from a two-dimensional lattice (containing both negative and positive time
parameters) to a (deformed) chain by eliminating all negative couplings. As usual, such
reductions break the full Virasoro symmetry of the original integrable model and only

716

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

the positive Borel subalgebra survives here as a symmetry of the matrix model. The
negatively moded constraints are therefore merely an artifact of the lattice reduction or
equivalently the chain deformation. In practical terms, all observables in the fermionic
model are described by observables conjugate to positive times in the unitary matrix model.
Observables conjugate to negative times exist but do not appear to play a role in the
equivalence.
We can understand all of these matters more clearly by rewriting the Virasoro
constraints (5.73) in terms of the BakerAkhiezer functions (5.33). For this, we note first
of all that (5.21) and (5.28) imply that they obey the flow equations
 
 
n [tE; z] 1 k
= Q+ Qk Qk nn n tE; z .
tk
2

(5.82)

in eigenvalue
Now let us compute the action of the scale transformation generator z z
space on the BakerAkhiezer functions. We have


N
  e 2 V (z)
N +3 N
n [tE; z]
=
+ zV 0 (z) n tE; z + N+1 PQn (z).
(5.83)
z
z
2
2
z 2
To evaluate the last term in (5.83), we have to take into account the triangularity of the
operator PQ. From P = P and the Jacobi property (5.11), we have
X 


(5.84)
PQ nn En,n .
PQ = PQ +
n>0

We can therefore evaluate the action of PQ on n (z) by using the string equation (5.9)
multiplied on the right by Q, but eliminating the pure upper triangular part of both sides of
the equation. Using (5.53) and the fact that


S
(5.85)
P Q1 = 0
which can be derived similarly to the other matrix elements of this section, we can
write (5.83) as
 
1
n [tE; z]
= (2n + N 1)n tE; z
z
2





   
N
QV 0 (Q) + QV 0 (Q) QV 0 (Q) nn n tE; z .
(5.86)
+
2
The last term in (5.86) can be rewritten using the flow equation (5.82), leading to the
eigenvalue equation
   
 
(5.87)
L0 tE; z n tE; z = nn tE; z ,
z

where we have defined the Virasoro operator

  X
N 1

+
.
ktk
z
L0 tE; z =
tk
z
2

(5.88)

k>1

The eigenvalue equation (5.87) shows that the lowest Virasoro constraint is indeed a
true symmetry of the system, in that the operators Q and L0 [tE; z] are simultaneously

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

717

diagonalizable in the basis of bi-orthonormal BakerAkhiezer functions. The same is

true of all higher Virasoro constraints. This shows precisely what sort of constraints the
auxiliary eigenvalue problem (5.36) must possess in order to correctly reconstruct the
fermionic -function. However, an analogous computation using the translation generator
/z on eigenvalue space leads to the result
!
X
 
 
N +1

ktk

n tE; z = (n + N)Rn n1 tE; z ,

(5.89)
tk1 z
2z
k>1

which shows that there is no corresponding L1 operator which commutes with the Lax
operator Q. This means that there is no translational symmetry in the system, and the
leading equations of motion at large N are determined instead by the generator L0 of
scale transformations of the system. These equations can be written in a form comparable
to those of Sections 3 and 4 by writing a consistency condition between the eigenvalue
equations (5.36) and (5.87) in the form
 
 


(5.90)
L0 tE; z + n (Q z)n tE; z = 0.
Expanding (5.90) using (5.87), (5.88) and completeness of the BakerAkhiezer functions
then leads to the system of equations
X
 
[Q]nm
ktk
= 0.
(5.91)
(1 + n m) Q nm +
tk
k>1

In particular, setting n = m in (5.91) and using the flow equation (5.31), we have
X
 
 


 
ktk Qk n,n1 Qk n+1,n .
Q nn =

(5.92)

k>1

What this means is that it is the operator PQ (along with its conjugate S
P Q1 ) which
plays the fundamental role in the dynamics of the adjoint fermion one-matrix model.
Thus it is the scale invariance of the system which leads to the fundamental double
scaling equations in the large N limit. The translational symmetry generated by the
operator P itself (and its conjugate S
P Q2 ) is an extra constraint on the theory which
plays no immediate role in the continuum limit of the matrix model. By considering
the matrix elements hn (z)|z[zm (z)]0 i, it is possible to infer the commutation relation
[Q, S
P Q1 ] = Q. Using the canonical commutator (5.6), it follows that the fundamental
symmetry operator of the system is given by


1
PQ S
P Q1 ,
(5.93)
2
which together with the Lax operator Q obeys the non-canonical commutation relation
=

[, Q] = Q.

(5.94)

The operator expression (5.94) defines the appropriate string equations of the fermionic
matrix model. The upper and lower triangular parts of the scaling operator (5.93) may be
computed by using (5.9), (5.51), (5.84) and (5.85) to get

718

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

+ =

1X
1X
(n N)En,n +
ktk Qk+ ,
2
2
n>0

1X
=
ktk Qk .
2

k>1

(5.95)

k>1

Substituting (5.95) into the string equation (5.94) and using the KP equation (5.22), we
arrive at (5.91).
It is the equation (5.92) which produces the novel Painlev expansions of the fermionic
matrix model. For instance, in the case of the quadratic potential studied in Section 4, it is
implicit in the manipulations carried out in Section 4.2. In fact, the structural form of (5.91)
is identical to the so-called automodel constraints which arise in the usual Hermitian onematrix models [30]. On the other hand, in the continuum limit Q becomes a differential
operator of a certain finite degree and the string equations (5.94) can be translated into an
equation for pseudo-differential operators which satisfy a generalized KdV hierarchy of
flow equations [4]. General solutions of string equations such as (5.94) have been studied
before, in the case that the double scaling limit of Q is described by a Schrdinger operator,
with stable, pole-free solutions for the corresponding string susceptibility [38]. We see
therefore that the string equations of the fermionic one-matrix model lead to a Painlev
differential equation which has the same structure as that in the usual Hermitian one-matrix
models, except that now the automodel form of the equations produce an alternating genus
expansion. Moreover, the structural form (5.94) of the string equations allow us to restrict
to positive values of the double-scaling variable and thereby obtain pole-free solutions
for the free energy of the fermionic matrix model.
In the present case, the details of the continuum limit of the operator Q are somewhat
involved. However, the discrete equations (5.91), (5.92) and (5.94) completely characterize
the double scaled equations for the partition function in the large N limit and the ensuing
topological expansion. They serve as the starting point for a characterization of the
differential hierarchies of the fermionic one-matrix model. From the generic structure of
the flow equations obtained in this section, we can expect that the continuum equations
satisfied by the partition function of fermionic quantum gravity are closely related to the
usual KdV flow structure of two-dimensional quantum gravity in terms of Gelfand-Dikii
differential polynomials [4,5].

Acknowledgements
We thank J. Ambjrn, C. Kristjansen, Y. Makeenko, G. Semenoff and J. Wheater
for helpful discussions. We also thank A.A. Kapaev, M. Matone and P. Zinn-Justin for
comments on the manuscript. The work of L.D.P. was supported in part by the Natural
Sciences and Engineering Research Council of Canada and NSF grant PHY98-02484. The
work of R.J.S. was supported in part by the Danish Natural Science Research Council.

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

719

Appendix A. Spectral correlation functions

In this appendix we will briefly describe some properties of correlators of the fermionic
one-matrix model within the formalism of this paper. The symmetries of the matrix
[9].
integral (1.1) restrict its observables to those which are invariant functions of
Connected correlation functions may be generated by taking derivatives of the free energy
with respect to the coupling constants of the potential V ,
+
* L
L
Y
1 Y

pj

tr()
= L
pj
log ZN ,
(A.1)
N
gpj
j =1

j =1

F,conn

where the normalized fermionic correlation functions are defined by

Z

eN tr V ()

d d f ()
,
f () F
ZN

(A.2)

Gr(N)c

and in (A.1) it is understood that, if necessary, a set of auxiliary coupling constants

gk are introduced into the potential V and then set to zero after differentiation. Given
the equivalence of the generating function log ZN for the connected correlators of the
fermionic and unitary matrix models, we therefore also have complete equivalence of their
observables,
* L
+
+
* L
Y
Y
p
p
j

tr()
=
tr U j
,
(A.3)
j =1

F,conn

j =1

U,conn

where the normalized unitary correlation functions are defined by

Z


kN
[dU ]f (U ) eN tr(V (U )log U ) .
f (U ) U
ZN

(A.4)

U (N)

The correlation functions in (A.3) may be evaluated as in (2.29)(2.33). For example,

one may readily compute

p = eip tr U p
tr()
0 U
F
" Z
#
kN
p N tr V (U0 )
=
[dU0 ] tr U0 e
.
(A.5)
NZN
2
SU(N)

C1 =N p

The restriction in (A.5) of the character expansion of eN tr V (U0 ) to Young tableaux with
N 2 p filled boxes implies that single trace correlators are non-vanishing only for p 6 N 2 .
The finiteness of the set of non-vanishing correlators is of course natural as a consequence
of the anticommuting property of the matrices and , but it is a dramatic result in the
unitary matrix model. Thus not only does the Penner interaction term in the unitary matrix
model (2.28) capture the finite nature of the perturbation expansion of ZN , but it also
reduces the correlation functions appropriately to reproduce the correct properties of the
p inserts,
fermionic correlators at finite N . Note that, geometrically, the operator tr()
on the dual triangulated surface, a hole with p boundary lengths. The lattice expansion of

720

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

such operators thereby generates fermionic ribbon graphs which are dual to tesselations of
Riemann surfaces with a given number of boundaries. Again there are only finitely many
such fat-graphs, because the maximum number of boundaries that a given discretization
can have is N 2 .
The complete set of observables of the fermionic matrix model can be generated by the
joint probability distributions
+
* n


(N n)! Y

tr zk
,
(A.6)
n (z1 , . . . , zn ) =
N!
k=1

where zk are points in the complex plane and 1 6 n 6 N . When the correlation function
in (A.6) is mapped to the unitary matrix model as described above, the points zk can
be interpreted as eigenvalues of unitary matrices. By diagonalizing the unitary matrices
in (A.4), using the identities




(A.7)
(z1 , . . . , zN ) = det j 1 (zi ) , (z1 , . . . , zN ) = det j 1 (zi ) ,
i,j

i,j

which follow from (2.6) and (3.5), and by using (3.7), it is straightforward to derive the
determinant representation
* n
+

(N n)! Y
(N n)! 
det K(zi , zj ) , (A.8)
tr (zk U ) =
n (z1 , . . . , zn ) =
N!
N!
i,j
k=1

K(z, z0 )

is the spectral kernel which is defined in terms of the orthogonal

where
polynomials as
N
0
N1
e 2 (V (z)+V (z )) X n (z)n (z0 )
.
K(z, z ) =
N+1
hn
(z z0 ) 2

(A.9)

n=0

We see therefore that the problem of evaluating correlation functions of the fermionic
matrix model reduces to that of determining the spectral kernel (A.9). It is possible
to express it in a much simpler form by deriving the appropriate generalization of the
ChristoffelDarboux formula [1]. For this, we first need to derive a mixed recursion
relation between the and polynomials. Consider the bi-orthogonality relations





n+1 (z) zn (z)zm = n+1 (z)zm n (z)z(m1) = 0,






n
(A.10)
z n (z)zm = znm n (z) = 0,
which are valid for 1 6 m 6 n. From (A.10) it follows that the two polynomials n+1 (z)
zn (z) and zn n (z) of degree n are equal up to some constant cn . Equating the constant
terms of these polynomials gives cn = pn+1,0 , and we arrive at the mixed three-term
recursion relation
n+1 (z) = zn (z) + pn+1,0 zn n (z).

(A.11)

In an analogous way, it is possible to derive the recursion relation

n+1 (z) =

1
1
n (z) + ln+1,0 n n (z).
z
z

(A.12)

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

721

We now multiply (A.11) through by n (z0 )/ hn and (A.12) with the index shift n
n 1 and z z0 through by z0 n (z)/ hn , subtract the resulting two equations, and then
sum over n = 1, . . . , N 1. This yields
N1
X

n (z)n (z0 ) N (z)N1 (z0 ) 1 (z)

=

hn
hN1
h1
n=1


N1
X 1
ln,0
n
0
0

pn+1,0 z n (z)n (z ) 0n1 n (z)n1 (z ) .

hk
z

(z z0 )

(A.13)

n=1

We can iterate the recursion equations (A.11) and (A.12) to get

zk k (z) = 1 + z

k
X

lj,0 j 1 (z),

j =1

1
z0k1

k1 (z0 ) = 1 +

k1
1X
pj,0 j 1 (z0 ).
z0

(A.14)

j =1

Substituting (A.14) into (A.13) and comparing the resulting equation with itself under the
interchange of arguments z z0 , we arrive after some algebra at
0

(z z )

N1
X
n=0

z0N1 N1 (z0 )N (z) zN1 N1 (z)N (z0 )

n (z)n (z0 )
=
hn
hN1 z0N1
1 h0
+
(z z0 ).
(A.15)
h0

The expression (A.15) is valid for any pair of complex numbers z 6= z0 . We can take the
z z0 limit of (A.15) using lHospitals rule to get
N1
X
n=0

zN1 N1 (z)N0 (z) N (z)(zN1 N1 (z))0

n (z)n (z)
=
hn
hN1 zN1
1 h0
+
.
h0

(A.16)

The identities (A.15) and (A.16) are the fermionic analogs of the ChristoffelDarboux
formula for the usual Hermitian one-matrix model orthogonal polynomials.
The ChristoffelDarboux formula allows us to express the spectral kernel (A.9) in terms
of orthogonal polynomials with indices n close to the matrix dimension N . Using the
mixed recursion relation (A.11), we obtain
0

K(z, z0 ) =

e 2 (V (z)+V (z ))
N+1

(z z0 ) 2

 1 h0

z0N+1
0
0
0

zN1 (z)N (z ) z N1 (z )N (z) +

.
hN1 pN,0 (z z0 )
h0
(A.17)


722

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

It follows that the fermionic matrix model is completely determined by the single set of
polynomials, as expected because of the one-matrix nature of the model and the heuristic
. In particular, for the spectral density we find
identification U




eNV (z)
1
1
0
(z) N0 (z)N1 (z)
z N (z)N1
(z) K(z, z) =
N
NzN hN1 pN,0 zN

 1 h0
+ N (z)N1 (z) +
.
(A.18)
h0
The formula (A.18) is particularly useful for determining the spectral density, and hence
all correlators, of the fermionic matrix model in the large N limit. In that case, factorization
and symmetry imply that all connected correlation functions vanish and the large N limit
of the model is completely characterized by the set of correlators


Z
1
p = dz (z)zp ,
tr()
(A.19)
N
F
where the integral goes over the support of the function (z) in the complex plane.
Note that the spectral density is defined a priori in the adjoint fermion matrix model
by the formula (A.6) [9,11]. In the present formalism it has a natural interpretation as
the probability density of eigenvalues in the corresponding unitary one-matrix model. As
such, it is generally supported on the unit circle. However, as discussed in the previous
subsection, the restriction of the integration contour to the unit circle in the large N limit is
not necessary, and the spectral density can be supported generically in the complex plane.
These facts explain the general properties of the spectral densities of fermionic matrix
models [911], for instance how the pole generated by the Penner interaction requires the
support contour of (z) to be adjusted so as to avoid the origin of the complex plane and
why the support endpoints are complex-valued.

Appendix B. Solution of the Gaussian model

In this appendix we will illustrate how the formalism of Section 3 works in a specific
example. We shall consider the example of the Gaussian potential
V (z) = z,

(B.1)

for which everything can be obtained explicitly. In this case the recursion relations of
section 3.2 are easily solved to give the coefficients


n
n
n+1
,
Pn = ,
Qn = 1 +
,
(B.2)
Rn =
n+N
N
N
and the normalization constants are
n!
.
(B.3)
hn = 2iN N
(n + N)!
Substituting (B.2) into the three-term recursion relation (3.29), we obtain


n1+N
(n 1)z
n2 (z).
(B.4)
n1 (z)
n (z) = z +
N
N

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

723

To solve the recurrence relation (B.4), we introduce the generating function which is
defined as the formal power series
N (z; s) =

n (z)s n ,

(B.5)

n=0

in a variable s, with the boundary condition N (z; 0) = 1. The recursion relation (B.4)
is then equivalent to a first order inhomogeneous linear differential equation for the
function N ,



 2

sz
s
(1 sz) + s z + 1
1 N (z; s) = 1.
(B.6)
N
s
N
Integrating (B.6) we find that the generating function (B.5) is given by
  
k+1

1 X (N + k 1)! 1 k
s
.
N (z; s) =
s
(N 1)!
N
1 sz

(B.7)

k=0

Expanding the function (B.7) as a power series in s and equating its coefficient of s n with
that of the definition (B.5), we arrive at an explicit form for the polynomials n (z),
 
n  
X
n (N + k 1)! 1 k nk
z ,
(B.8)
n (z) =
(N 1)!
N
k
k=0

which can be expressed as

n (z) = N N zN+n U [N, n + 1 + N; Nz],
where
U [a, c; x] =

(B.9)



x 1c 1 F1 [a c + 1, 2 c; x]

1 F1 [a, c; x]
+
, (B.10)
sin c (a c + 1) (c)
(a) (2 c)

is a confluent hypergeometric function. The partition function and observables of the

Gaussian fermionic one-matrix model are therefore completely determined by a system
of confluent hypergeometric polynomials.

References
[1] M.L. Mehta, Random Matrices, Academic Press, New York, 1991;
C. Itzykson, J.M. Drouffe, Statistical Field Theory, Vol. 2, Strong Coupling, Monte Carlo
Methods, Conformal Field Theory, And Random Systems, Cambridge University Press, 1989.
[2] T. Guhr, A. Mller-Groeling, H.A. Weidenmller, Random matrix theories in quantum physics:
common concepts, Phys. Rep. 299 (1998) 189, cond-mat/9707301.
[3] J.J.M. Verbaarschot, T. Wettig, Random matrix theory and chiral symmetry in QCD, hepph/0003017.
[4] P. Di Francesco, P. Ginsparg, J. Zinn-Justin, 2D gravity and random matrices, Phys. Rep. 254
(1995) 1, hep-th/9306153.
[5] J. Ambjrn, B. Durhuus, T. Jonsson, Quantum Geometry: A Statistical Field Theory Approach,
Cambridge University Press, 1997.

724

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

[6] P. Di Francesco, Matrix model combinatorics: applications to folding and colouring, mathph/9911002.
[7] S.R. Das, A. Dhar, A.M. Sengupta, S.R. Wadia, New critical behaviour in D = 0 large N matrix
models, Mod. Phys. Lett. A 5 (1990) 1041;
B. Durhuus, Multi-spin systems on a randomly triangulated surface, Nucl. Phys. B 426 (1994)
203, hep-th/9402052;
J. Ambjrn, Quantum gravity represented as dynamical triangulations, Class. Quantum Grav. 12
(1995) 2079;
I.R. Klebanov, Touching random surfaces and liouville gravity, Phys. Rev. D 51 (1995) 1836,
hep-th/9407167.
[8] Y. Makeenko, K. Zarembo, Adjoint fermion matrix models, Nucl. Phys. B 422 (1994) 237,
hep-th/9309012.
[9] G.W. Semenoff, R.J. Szabo, Fermionic matrix models, Int. J. Mod. Phys. A 12 (1997) 2135,
hep-th/9605140.
[10] J. Ambjrn, C.F. Kristjansen, Y. Makeenko, Generalized Penner models to all genera, Phys.
Rev. D 50 (1994) 5193, hep-th/9403024.
[11] N. Marshall, G.W. Semenoff, R.J. Szabo, Critical behavior of a fermionic random matrix model
at large N , Phys. Lett. B 351 (1995) 153, hep-th/9410214.
[12] Y. Makeenko, I. Chepelev, Supersymmetric matrix models and the meander problem, hepth/9601139.
[13] J. Ambjrn, Y. Makeenko, K. Zarembo, Supersymmetric matrix models and branched polymers,
Nucl. Phys. B 482 (1996) 660, cond-mat/9606041.
[14] G.W. Semenoff, R.J. Szabo, Polymer statistics and fermionic vector models, Mod. Phys. Lett.
A 11 (1996) 1185, hep-th/9602007.
[15] A. Morozov, Integrability and matrix models, Phys. Usp. 37 (1994) 1, hep-th/9303139;
A. Mironov, 2D gravity and matrix models, Int. J. Mod. Phys. A 9 (1994) 4355, hep-th/9312212;
M. Adler, P. van Moerbeke, String-orthogonal polynomials, string equations, and two-Toda
symmetries, Commun. Pure Appl. Math. 50 (1997) 241, hep-th/9706182.
[16] R.C. Penner, Perturbative series and the moduli space of Riemann surfaces, J. Diff. Geom. 27
(1988) 35.
[17] C. Itzykson, J.B. Zuber, The planar approximation. 2, J. Math. Phys. 21 (1980) 411.
[18] P. Zinn-Justin, Adding and multiplying random matrices: a generalization of Voiculescus
formulae, Phys. Rev. E 59 (1999) 4884.
[19] C. Itzykson, Toeplitz determinants as group averages, in: Proc. Symp. Analytic Methods in
Mathematical Physics, Bloomington, IN, June 26, 1968, Gordon and Breach, New York, 1970,
p. 469.
[20] M. Creutz, On invariant integration over SU(N), J. Math. Phys. 19 (1978) 2043.
[21] D. Bessis, C. Itzykson, J.B. Zuber, Quantum field theory techniques in graphical enumeration,
Adv. Appl. Math. 1 (1980) 109;
M.L. Mehta, A method of integration over matrix variables, Commun. Math. Phys. 79 (1981)
327.
[22] U. Grenander, G. Szeg, Toeplitz Forms and their Applications, University of California Press,
Berkeley, 1958.
[23] V. Periwal, D. Shevitz, Unitary matrix models as exactly solvable string theories, Phys. Rev.
Lett. 64 (1990) 1326;
V. Periwal, D. Shevitz, Exactly solvable unitary matrix models: multicritical potentials and
correlations, Nucl. Phys. B 344 (1990) 731.
[24] M.J. Bowick, A. Morozov, D. Shevitz, Reduced unitary matrix models and the hierarchy of
-functions, Nucl. Phys. B 354 (1991) 496;
M. Hisakado, Unitary matrix models and Painlev III, Mod. Phys. Lett. A 11 (1996) 3001,
hep-th/9609214;

L.D. Paniak, R.J. Szabo / Nuclear Physics B 593 (2001) 671725

[25]
[26]
[27]
[28]

[29]
[30]

[31]
[32]

[33]
[34]
[35]
[36]
[37]
[38]

725

S. Kharchev, A. Mironov, A. Zhedanov, Faces of relativistic Toda chain, Int. J. Mod. Phys. A 12
(1997) 2675, hep-th/9606144.
M. Adler, P. van Moerbeke, Integrals over classical groups, random permutations, Toda and
Toeplitz lattices, math.CO/9912143.
E.L. Ince, Ordinary Differential Equations, Dover, 1956;
E. Hille, Lectures on Ordinary Differential Equations, AddisonWesley, Reading, MA, 1969.
F. David, Loop equations and nonperturbative effects in two-dimensional quantum gravity,
Mod. Phys. Lett. A 5 (1990) 1019.
A.A. Kapaev, Asymptotics of solutions of the Painlev equation of the first kind, Diff. Eqns. 24
(1989) 1107;
A.A. Kapaev, A.V. Kitaev, Connection formulae for the first Painlev transcendent in the
complex domain, Lett. Math. Phys. 27 (1993) 243.
F. David, Phases of the large N matrix model and nonperturbative effects in 2D gravity, Nucl.
Phys. B 348 (1991) 507.
A. Gerasimov, A. Marshakov, A. Mironov, A. Morozov, A. Orlov, Matrix models of 2D gravity
and Toda theory, Nucl. Phys. B 357 (1991) 565;
L. Bonora, C.S. Xiong, Multi-matrix models without continuum limit, Nucl. Phys. B 405 (1993)
329, hep-th/9212070.
G. Bonelli, P.A. Marchetti, M. Matone, Nonperturbative 2-D gravity, punctured spheres and
theta vacua in string theories, Phys. Lett. B 339 (1994) 49, hep-th/9407091.
E. Marinari, G. Parisi, The supersymmetric one-dimensional string, Phys. Lett. B 240 (1990)
375;
M. Karliner, A.A. Migdal, Nonperturbative 2D quantum gravity via supersymmetric string,
Mod. Phys. Lett. A 5 (1990) 2565;
J. Ambjrn, J. Greensite, S. Varsted, A nonperturbative definition of 2D quantum gravity by the
fifth time action, Phys. Lett. B 249 (1990) 411.
J. Ambjrn, C.V. Johnson, T.R. Morris, Stochastic quantization vs. KdV flows in 2D quantum
gravity, Nucl. Phys. B 374 (1992) 496, hep-th/9109018.
G. Segal, G. Wilson, Loop groups and equations of KdV type, Publ. Math. IHES 61 (1985) 1.
K. Ueno, K. Takasaki, Toda lattice hierarchy, Adv. Studies Pure Math. 4 (1984) 1.
S.N. Ruijsenaars, Relativistic Toda systems, Commun. Math. Phys. 133 (1990) 217.
M. Hisakado, M. Wadati, Matrix models of two-dimensional gravity and discrete Toda theory,
Mod. Phys. Lett. A 11 (1996) 1797, hep-th/9605175.
S. Dalley, C.V. Johnson, T.R. Morris, Nonperturbative two-dimensional quantum gravity, Nucl.
Phys. B 368 (1992) 655.

Nuclear Physics B 593 (2001) 729730

www.elsevier.nl/locate/npe

Erratum

Erratum to Universality of 1/Q corrections to

jet-shape observables rescued
[Nucl. Phys. B 511 (1998) 396418]
Yu.L. Dokshitzer a , A. Lucenti, G. Marchesini b , G.P. Salam c,
a LPT, Universit de Parix XI, Centre dOrsay, Paris, France
b Universit di Milano Bicocca and INFN, Sezione di Milano, Milano, Italy
c TH Division, CERN, 1211 Geneva 23, Switzerland

Abstract
We correct a mistake in the calculation of the two-loop correction factor (known as the Milan
factor) to 1/Q corrections in event shapes. 2001 Elsevier Science B.V. All rights reserved.

This erratum corrects a mistake in the calculation [1] of the two-loop correction factor to
the coefficient of 1/Q power corrections to the thrust in e+ e , discovered subsequent to an
independent calculation of the nf -dependent piece of the 1/Q suppressed contribution to
the C-parameter in [2]. While the mistake has already been briefly discussed and corrected
in [3], this erratum gives the details of the modifications that should be made to [1].
In [1] the last line of the equation preceding Eq. (4.12) is missing a factor of two. The
correct form reads
d 2 k d 4q 2 u1 du1 u2 du2
=
.
2
2u1 u2 | sin |
This correction propagates through Eqs. (4.12), (4.14) and (4.16) to (4.18). In particular
these last three equations, which embody the final result, become:
(n)

rT =

2
(1.227 CA + 0.365 CA 0.052 nf ),
0

rT(n) = 0.710 (0.470),

for nf = 3 (0),

PII of original article: S0550-3213(97)00650-0.

* Corresponding author.

E-mail address: [email protected] (G.P. Salam).

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 4 6 - 5

(4.16)
(4.17)

730

Yu.L. Dokshitzer et al. / Nuclear Physics B 593 (2001) 729730

(i)

(n)

rT = rT + rT =

1
[1.575 CA 0.104 nf ] = 0.490(0.430),
0

for nf = 3 (0).
(4.18)

The correction of the error does not modify any of the conclusions of the paper. Identical
corrections apply to calculations of the two corrections for the 1/Q power correction in
other event-shape observables in e+ e [4,5] (and ensures agreement with a pure nf result
of [6]) and in deep inelastic scattering [7].

Acknowledgements
We wish to thank Mrinal Dasgupta, Lorenzo Magnea and Graham Smye (the authors
of [2]) for discussions about their results prior to publication, and Bryan Webber for
discussions which helped locate the error.

References
[1] Y.L. Dokshitzer, A. Lucenti, G. Marchesini, G.P. Salam, Nucl. Phys. B 511 (1998) 396, hepph/9707532.
[2] M. Dasgupta, L. Magnea, G. Smye, JHEP 9911 (1999) 025, hep-ph/9911316.
[3] Y.L. Dokshitzer, Contribution to the proceedings of the 11th Rencontres de Blois Frontiers of
Matter, hep-ph/9911299.
[4] Y.L. Dokshitzer, A. Lucenti, G. Marchesini, G.P. Salam, JHEP 9805 (1998) 003, hepph/9802381.
[5] Y.L. Dokshitzer, G. Marchesini, G.P. Salam, Eur. Phys. J. C 3 (1999) 1, hep-ph/9812487.
[6] M. Beneke, V.M. Braun, L. Magnea, Nucl. Phys. B 497 (1997) 297, hep-ph/9701309.
[7] M. Dasgupta, B.R. Webber, JHEP 9810 (1998) 001, hep-ph/9809247.

Nuclear Physics B 593 (2001) 731732

www.elsevier.nl/locate/npe

Erratum

Erratum to Golden measurements at

a neutrino factory
[Nucl. Phys. B 579 (2000) 17]
A. Cervera, A. Donini, M.B. Gavela, J.J. Gomez Cdenas,
P. Hernndez, O. Mena, S. Rigolin
P. Lipari has pointed out an incorrect statement in the article. In Section 2.2 we stated
that the number of neutrinos produced with a given E < E is independent of E . This
does not follow from Eq. (3) as we claimed. The corresponding paragraph should have
been instead: Unlike traditional neutrino beams obtained from and K decays, the fluxes
in Eq. (3), in the forward direction, present a leading quadratic dependence on E . As a
consequence, the oscillation signal does not decrease with increasing E . In Appendix...
Furthermore, a few misprints have been detected in Tables 2, 5, 6 and 7. The corrected
tables are shown below.
Table 2
Data samples expected in a 40 kT detector for 1021 useful + decays. e oscillations with
parameters as in Appendix B
Baseline (km)

CC

e CC

+ e NC

(signal)

732
3500
7332

3.5 107
1.5 106
3.5 105

5.9 107
2.6 106
5.9 105

3.1 107
1.3 106
3.0 105

1.1 105
1.0 105
3.8 104

Table 5
Data samples expected in a 40 kT detector for 1021 useful decays
Baseline (km)

CC

e CC

+ e NC

(signal)

732
3500
7332

6.9 107
3 106
6.9 105

3.0 107
1.3 106
3.0 105

3.1 107
1.4 106
3.1 105

5 104
1.6 104
0.2 104

PII of original article: S0550-3213(00)00221-2.

E-mail address: [email protected] (A. Donini).

0550-3213/01/$ see front matter 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 6 0 6 - 4

732

A. Cervera et al. / Nuclear Physics B 593 (2001) 731732

Table 6
Fractional backgrounds and signal selection efficiency for the wrong-sign muon search with
decays
CC

e CC

+ e NC

(signal)

3.0 107

6.0 107

2.0 106

4 101

Table 7
Events surviving the cuts in a 40 kT detector for 1021 useful decays
Baseline (km)

CC

e CC

+ e NC

(signal)

732
3500
7332

21
0.9
0.2

18
0.8
0.2

66
2.4
0.6

2 104
6.4 103
8 102

None of our results or conclusions are affected by these corrections.

Nuclear Physics B 593 (2001) 733735

www.elsevier.nl/locate/npe

CUMULATIVE AUTHOR INDEX B591B593

Akhmedov, E.T.
Armoni, A.
Arnowitt, R.
Asakawa, T.

B592 (2000) 234

B593 (2001) 229
B592 (2000) 143
B591 (2000) 611

Bakas, I.
Belitsky, A.V.
Beneke, M.
Beneke, M.
Bialas, A.
Bialas, P.
Bigi, I.I.
Bilal, A.
Bill, M.
Bonneau, G.
Boos, H.E.
Borlaf, J.
Broadhurst, D.J.
Brodsky, S.J.
Bruckmann, F.
Buchalla, G.
Bunder, J.E.
Buras, A.J.
Burda, Z.

B593 (2001) 31
B593 (2001) 289
B591 (2000) 313
B592 (2000) 3
B593 (2001) 438
B592 (2000) 391
B592 (2000) 92
B593 (2001) 31
B591 (2000) 139
B593 (2001) 398
B592 (2001) 597
B593 (2001) 243
B592 (2000) 247
B593 (2001) 311
B593 (2001) 545
B591 (2000) 313
B592 (2001) 445
B592 (2000) 55
B592 (2000) 391

Capella, A.
Capitani, S.
Capitani, S.
Carena, M.
Casteill, P.-Y.
Catani, S.
Cervera, A.
Chalmers, G.
Chaudhuri, S.
Chen, B.
Chen, G.-H.
Chernodub, M.N.
Chizhov, M.V.
Choudhury, D.
Cleaver, G.B.
Costa, M.S.

B593 (2001) 336

B592 (2000) 183
B593 (2001) 183
B592 (2000) 164
B591 (2000) 491
B591 (2000) 435
B593 (2001) 731
B591 (2000) 39
B591 (2000) 243
B593 (2001) 505
B593 (2001) 562
B592 (2000) 107
B591 (2000) 457
B592 (2000) 35
B593 (2001) 471
B591 (2000) 469

Dalmazi, D.
Datta, A.
Derendinger, J.-P.
Dobado, A.
Dokshitzer, Yu.L.
Dolan, F.A.
Dolcini, F.
Donini, A.
Dotti, G.
Duan, Z.
Dutta, B.

B592 (2001) 419

B592 (2000) 35
B593 (2001) 31
B592 (2000) 203
B593 (2001) 729
B593 (2001) 599
B592 (2001) 563
B593 (2001) 731
B591 (2000) 636
B592 (2000) 371
B592 (2000) 143

Ecker, G.
Ennes, I.P.
Evans, N.

B591 (2000) 419

B591 (2000) 195
B592 (2000) 129

Fabbri, D.
Faraggi, A.E.
Feldmann, Th.
Ferreiro, E.G.
Fioravanti, D.
Frste, S.
Fr, P.

B591 (2000) 139

B593 (2001) 471
B592 (2000) 3
B593 (2001) 336
B591 (2000) 685
B593 (2001) 127
B591 (2000) 139

Gambino, P.
Ganor, O.J.
Gavela, M.B.
Ghezelbash, A.M.
Gckeler, M.
Gomez Cdenas, J.J.
Gomis, J.
Gorbahn, M.
Govindarajan, S.
Grazzini, M.
Gubarev, F.V.
Guhr, T.
Gunion, J.F.
Gutierrez, T.D.

B592 (2000) 55
B591 (2000) 547
B593 (2001) 731
B592 (2000) 408
B593 (2001) 183
B593 (2001) 731
B591 (2000) 265
B592 (2000) 55
B593 (2001) 155
B591 (2000) 435
B592 (2000) 107
B593 (2001) 361
B591 (2000) 277
B591 (2000) 277

Han, T.
Hannah, T.
Harmark, T.
Heinzl, T.

B593 (2001) 415

B593 (2001) 577
B593 (2001) 76
B593 (2001) 545

734

Nuclear Physics B 593 (2001) 733735

Hernndez, P.
Honecker, G.
Horsley, R.
Huitu, K.
Hwang, D.S.

B593 (2001) 731

B593 (2001) 127
B593 (2001) 183
B592 (2000) 164
B593 (2001) 311

Isidori, G.
Itoyama, H.

B591 (2000) 419

B593 (2001) 505

Jger, S.
Jayaraman, T.

B592 (2000) 55
B593 (2001) 155

Kaidalov, A.B.
Kataev, A.L.
Kehrein, S.
Kharraziha, H.
Kim, J.E.
Kim, Y.J.
King, S.F.
Kirilova, D.P.
Kishimoto, I.
Klebanov, I.R.
Kniehl, B.A.
Kobayashi, T.
Konechny, A.
Kyae, B.

B593 (2001) 336

B592 (2000) 247
B592 (2001) 512
B592 (2000) 321
B591 (2000) 587
B593 (2001) 415
B591 (2000) 3
B591 (2000) 457
B591 (2000) 611
B591 (2000) 26
B591 (2000) 296
B592 (2000) 164
B591 (2000) 667
B591 (2000) 587

Lechtenfeld, O.
Lee, H.M.
Linart, S.
Likhoded, A.
Lozano, C.
Lucenti, A.

B591 (2000) 39
B591 (2000) 587
B592 (2001) 479
B593 (2001) 415
B591 (2000) 195
B593 (2001) 729

Ma, B.-Q.
Magnea, L.
Mangazeev, V.V.
Marchesini, G.
Maroto, A.L.
Matsuo, T.
Maxwell, C.J.
Mayr, P.
McKenzie, R.H.
McKeon, D.G.C.
Mehen, T.
Melnikov, K.
Mena, O.
Merlatti, P.
Mikhailov, A.Yu.
Mintchev, M.
Mohapatra, R.N.
Montorsi, A.
Mller, D.
Mller, G.
Murakami, K.

B593 (2001) 311

B593 (2001) 269
B592 (2001) 597
B593 (2001) 729
B592 (2000) 203
B593 (2001) 505
B592 (2000) 247
B593 (2001) 99
B592 (2001) 445
B591 (2000) 591
B591 (2000) 265
B591 (2000) 515
B593 (2001) 731
B591 (2000) 139
B591 (2000) 547
B592 (2000) 219
B593 (2001) 451
B592 (2001) 563
B593 (2001) 289
B591 (2000) 419
B593 (2001) 505

Naculich, S.G.
Nanopoulos, D.V.
Narison, S.
Navelet, H.
Nekrasov, N.A.
Neubert, M.
Neufeld, H.
Niedermeier, L.
Niemeyer, B.
Nimai Singh, N.

B591 (2000) 195

B593 (2001) 471
B593 (2001) 3
B593 (2001) 438
B591 (2000) 26
B591 (2000) 313
B591 (2000) 419
B593 (2001) 289
B591 (2000) 39
B591 (2000) 3

Osborn, H.

B593 (2001) 599

Palisoc, C.P.
Pallante, E.
Paniak, L.D.
Parvizi, S.
Pasti, P.
Penati, S.
Prez-Lorenzana, A.
Perlt, H.
Perry, M.J.
Peschanski, R.
Petersson, B.
Petrini, M.
Pich, A.
Pich, A.
Pilo, L.
Polikarpov, M.I.

B591 (2000) 296

B592 (2000) 294
B593 (2001) 671
B592 (2000) 408
B591 (2000) 109
B593 (2001) 651
B593 (2001) 451
B593 (2001) 183
B591 (2000) 469
B593 (2001) 438
B592 (2000) 391
B592 (2000) 129
B591 (2000) 419
B592 (2000) 294
B592 (2000) 219
B592 (2000) 107

Rakow, P.E.L.
Rasmussen, J.
Rigolin, S.
Rodrigues da Silva, P.S.
Ruelle, P.

B593 (2001) 183

B593 (2001) 634
B593 (2001) 731
B592 (2000) 371
B592 (2001) 479

Sachrajda, C.T.
Salam, G.P.
Salgado, C.A.
Sannino, F.
Santambrogio, A.
Sarkar, T.
Saulina, N.
Schfer, A.
Schierholz, G.
Schiller, A.
Schmidt, I.
Schnitzer, H.J.
Schreyer, R.
Schwarz, A.
Sfetsos, K.
Shatashvili, S.L.
Silvestrini, L.
Simonov, Yu.A.
Sirlin, A.
Smolin, L.

B591 (2000) 313

B593 (2001) 729
B593 (2001) 336
B592 (2000) 371
B593 (2001) 651
B593 (2001) 155
B591 (2000) 547
B593 (2001) 289
B593 (2001) 183
B593 (2001) 183
B593 (2001) 311
B591 (2000) 195
B593 (2001) 127
B591 (2000) 667
B593 (2001) 31
B591 (2000) 26
B592 (2000) 55
B592 (2000) 350
B591 (2000) 296
B591 (2000) 227

Nuclear Physics B 593 (2001) 733735

Sorokin, D.
Stanishkov, M.
Sugiyama, K.
Szabo, R.J.

B591 (2000) 109

B591 (2000) 685
B591 (2000) 701
B593 (2001) 671

Tabaczek, J.
Tonin, M.

B592 (2000) 391

B591 (2000) 109

Uehara, S.
Uraltsev, N.G.

B591 (2000) 77
B592 (2000) 92

Valencia, G.
Valent, G.
van Ritbergen, T.
Vekua, T.

B593 (2001) 415

B591 (2000) 491
B591 (2000) 515
B593 (2001) 545

735

Verbaarschot, J.J.M.
Verhoeven, O.
Vogt, R.

B592 (2001) 419

B592 (2001) 479
B591 (2000) 277

Walker, J.W.
Wilke, T.
Wipf, A.
Wu, Y.-S.

B593 (2001) 471

B593 (2001) 361
B593 (2001) 545
B593 (2001) 562

Yamada, S.

B591 (2000) 77

Zaffaroni, A.
Zakharov, V.I.
Zanon, D.

B591 (2000) 139

B592 (2000) 107
B593 (2001) 651